\begin{page}{manpageXXd01}{NAG On-line Documentation: d01}
\beginscroll
\begin{verbatim}



     D01(3NAG)         Foundation Library (12/10/92)         D01(3NAG)



          D01 -- Quadrature                             Introduction -- D01
                                    Chapter D01
                                    Quadrature

          1. Scope of the Chapter

          This chapter provides routines for the numerical evaluation of
          definite integrals in one or more dimensions and for evaluating
          weights and abscissae of integration rules.

          2. Background to the Problems

          The routines in this chapter are designed to estimate:

          (a)   the value of a one-dimensional definite integral of the
                form:
                                        b
                                        /
                                        |f(x)dx                         (1)
                                        /
                                        a
                where f(x) is defined by the user, either at a set of
                points (x ,f(x )), for i=1,2,...,n where a=x <x <...<x =b,
                         i    i                             1  2      n
                or in the form of a function; and the limits of integration
                a,b may be finite or infinite.

                Some methods are specially designed for integrands of the
                form
                                    f(x)=w(x)g(x)                       (2)
                which contain a factor w(x), called the weight-function, of
                a specific form. These methods take full account of any
                peculiar behaviour attributable to the w(x) factor.

          (b)   the value of a multi-dimensional definite integral of the
                form:
                             /
                             | f(x ,x ,...,x )dx ...dx dx               (3)
                             /    1  2      n   n     2  1
                             R
                              n
                where f(x ,x ,...,x ) is a function defined by the user and
                         1  2      n
                R  is some region of n-dimensional space.
                 n

                The simplest form of R  is the n-rectangle defined by:
                                      n
                               a <=x <=b ,  i=1,2,...,n                 (4)
                                i   i   i
                where a  and b  are constants. When a  and b  are functions
                       i      i                      i      i
                of x  (j<i), the region can easily be transformed to the
                    j
                rectangular form (see Davis and Rabinowitz [1] page 266).
                Some of the methods described incorporate the
                transformation procedure.

          2.1. One-dimensional Integrals

          To estimate the value of a one-dimensional integral, a quadrature
          rule uses an approximation in the form of a weighted sum of
          integrand values, i.e.,

                              b         N
                              /         --
                              |f(x)dx~= >  w f(x ).                     (5)
                              /         --  i   i
                              a         i=1

          The points x  within the interval [a,b] are known as the
                      i
          abscissae, and the w  are known as the weights.
                              i

          More generally, if the integrand has the form (2), the
          corresponding formula is

                            b             N
                            /             --
                            |w(x)g(x)dx~= >  w g(x ).                   (6)
                            /             --  i   i
                            a             i=1

          If the integrand is known only at a fixed set of points, these
          points must be used as the abscissae, and the weighted sum is
          calculated using finite-difference methods. However, if the
          functional form of the integrand is known, so that its value at
          any abscissa is easily obtained, then a wide variety of
          quadrature rules are available, each characterised by its choice
          of abscissae and the corresponding weights.

          The appropriate rule to use will depend on the interval [a,b] -
          whether finite or otherwise - and on the form of any w(x) factor
          in the integrand. A suitable value of N depends on the general
          behaviour of f(x); or of g(x), if there is a w(x) factor present.

          Among possible rules, we mention particularly the Gaussian
          formulae, which employ a distribution of abscissae which is
          optimal for f(x) or g(x) of polynomial form.

          The choice of basic rules constitutes one of the principles on
          which methods for one-dimensional integrals may be classified.
          The other major basis of classification is the implementation
          strategy, of which some types are now presented.

          (a)   Single rule evaluation procedures

                A fixed number of abscissae, N, is used. This number and
                the particular rule chosen uniquely determine the weights
                and abscissae. No estimate is made of the accuracy of the
                result.

          (b)   Automatic procedures

                The number of abscissae, N, within [a,b] is gradually
                increased until consistency is achieved to within a level
                of accuracy (absolute or relative) requested by the user.
                There are essentially two ways of doing this; hybrid forms
                of these two methods are also possible:
                (i)   whole interval procedures (non-adaptive)

                      A series of rules using increasing values of N are
                      successively applied over the whole interval [a,b].
                      It is clearly more economical if abscissae already
                      used for a lower value of N can be used again as part
                      of a higher-order formula. This principle is known as
                       optimal extension. There is no overlap between the
                      abscissae used in Gaussian formulae of different
                      orders. However, the Kronrod formulae are designed to
                      give an optimal (2N+1)-point formula by adding (N+1)
                      points to an N-point Gauss formula. Further
                      extensions have been developed by Patterson.

                (ii)  adaptive procedures

                      The interval [a,b] is repeatedly divided into a
                      number of sub-intervals, and integration rules are
                      applied separately to each sub-interval. Typically,
                      the subdivision process will be carried further in
                      the neighbourhood of a sharp peak in the integrand,
                      than where the curve is smooth. Thus, the
                      distribution of abscissae is adapted to the shape of
                      the integrand.

                      Subdivision raises the problem of what constitutes an
                      acceptable accuracy in each sub-interval. The usual
                      global acceptability criterion  demands that the sum
                      of the absolute values of the error estimates in the
                      sub-intervals should meet the conditions required of
                      the error over the whole interval. Automatic
                      extrapolation over several levels of subdivision may
                      eliminate the effects of some types of singularities.

          An ideal general-purpose method would be an automatic method
          which could be used for a wide variety of integrands, was
          efficient (i.e., required the use of as few abscissae as
          possible), and was reliable (i.e., always gave results within the
          requested accuracy). Complete reliability is unobtainable, and
          generally higher reliability is obtained at the expense of
          efficiency, and vice versa.  It must therefore be emphasised that
          the automatic routines in this chapter cannot be assumed to be
          100% reliable. In general, however, the reliability is very high.

          2.2. Multi-dimensional Integrals

          A distinction must be made between cases of moderately low
          dimensionality (say, up to 4 or 5 dimensions), and those of
          higher dimensionality. Where the number of dimensions is limited,
          a one-dimensional method may be applied to each dimension,
          according to some suitable strategy, and high accuracy may be
          obtainable (using product rules). However, the number of
          integrand evaluations rises very rapidly with the number of
          dimensions, so that the accuracy obtainable with an acceptable
          amount of computational labour is limited; for example a product
                                                                      9
          of 3-point rules in 20 dimensions would require more than 10
          integrand evaluations. Special techniques such as the Monte Carlo
          methods can be used to deal with high dimensions.

          (a)   Products of one-dimensional rules

                Using a two-dimensional integral as an example, we have
                         b  b                   [ b          ]
                          1  2             N    [  2         ]
                         /  /              --   [ /          ]
                         |  | f(x,y)dydx~= >  w [ | f(x ,y)dy]          (7)
                         /  /              --  i[ /    i     ]
                         a  a              i=1  [ a          ]
                          1  2                  [  2         ]

                         b  b
                          1  2             N   N
                         /  /              --  --
                         |  | f(x,y)dydx~= >   >  w v f(x ,y )          (8)
                         /  /              --  --  i j   i  j
                         a  a              i=1 j=1
                          1  2
                where (w ,x ) and (v ,y ) are the weights and abscissae of
                        i  i        i  i
                the rules used in the respective dimensions.

                A different one-dimensional rule may be used for each
                dimension, as appropriate to the range and any weight
                function present, and a different strategy may be used, as
                appropriate to the integrand behaviour as a function of
                each independent variable.

                For a rule-evaluation strategy in all dimensions, the
                formula (8) is applied in a straightforward manner. For
                automatic strategies (i.e., attempting to attain a
                requested accuracy), there is a problem in deciding what
                accuracy must be requested in the inner integral(s).
                Reference to formula (7) shows that the presence of a
                limited but random error in the y-integration for different
                values of x  can produce a 'jagged' function of x, which
                           i
                may be difficult to integrate to the desired accuracy and
                for this reason products of automatic one-dimensional
                routines should be used with caution (see also Lyness [3]).

          (b)   Monte Carlo methods

                These are based on estimating the mean value of the
                integrand sampled at points chosen from an appropriate
                statistical distribution function. Usually a variance
                reducing procedure is incorporated to combat the
                fundamentally slow rate of convergence of the rudimentary
                form of the technique. These methods can be effective by
                comparison with alternative methods when the integrand
                contains singularities or is erratic in some way, but they
                are of quite limited accuracy.

          (c)   Number theoretic methods

                These are based on the work of Korobov and Conroy and
                operate by exploiting implicitly the properties of the
                Fourier expansion of the integrand. Special rules,
                constructed from so-called optimal coefficients, give a
                particularly uniform distribution of the points throughout
                n-dimensional space and from their number theoretic
                properties minimize the error on a prescribed class of
                integrals. The method can be combined with the Monte Carlo
                procedure.

          (d)   Sag-Szekeres method

                By transformation this method seeks to induce properties
                into the integrand which make it accurately integrable by
                the trapezoidal rule. The transformation also allows
                effective control over the number of integrand evaluations.

          (e)   Automatic adaptive procedures

                An automatic adaptive strategy in several dimensions
                normally involves division of the region into subregions,
                concentrating the divisions in those parts of the region
                where the integrand is worst behaved. It is difficult to
                arrange with any generality for variable limits in the
                inner integral(s). For this reason, some methods use a
                region where all the limits are constants; this is called a
                hyper-rectangle. Integrals over regions defined by variable
                or infinite limits may be handled by transformation to a
                hyper-rectangle. Integrals over regions so irregular that
                such a transformation is not feasible may be handled by
                surrounding the region by an appropriate hyper-rectangle
                and defining the integrand to be zero outside the desired
                region. Such a technique should always be followed by a
                Monte Carlo method for integration.

                The method used locally in each subregion produced by the
                adaptive subdivision process is usually one of three types:
                Monte Carlo, number theoretic or deterministic.
                Deterministic methods are usually the most rapidly
                convergent but are often expensive to use for high
                dimensionality and not as robust as the other techniques.

          2.3. References

          Comprehensive reference:

          [1]   Davis P J and Rabinowitz P (1975) Methods of Numerical
                Integration. Academic Press.

          Special topics:

          [2]   Gladwell I (1986) Vectorisation of one dimensional
                quadrature codes. Techincal Report. TR7/86  NAG.

          [3]   Lyness J N (1983) When not to use an automatic quadrature
                routine. SIAM Review. 25 63--87.

          [4]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [5]   Sobol I M (1974) The Monte Carlo Method. The University of
                Chicago Press.

          [6]   Stroud A H (1971) Approximate Calculation of Multiple
                Integrals. Prentice-Hall.

          3. Recommendations on Choice and Use of Routines

          The following three sub-sections consider in turn routines for:
          one-dimensional integrals over a finite interval, and over a
          semi-infinite or an infinite interval; and multi-dimensional
          integrals. Within each sub-section, routines are classified by
          the type of method, which ranges from simple rule evaluation to
          automatic adaptive algorithms. The recommendations apply
          particularly when the primary objective is simply to compute the
          value of one or more integrals, and in these cases the automatic
          adaptive routines are generally the most convenient and reliable,
          although also the most expensive in computing time.

          Note however that in some circumstances it may be counter-
          productive to use an automatic routine. If the results of the
          quadrature are to be used in turn as input to a further
          computation (e.g. an 'outer' quadrature or an optimization
          problem), then this further computation may be adversely affected
          by the 'jagged performance profile' of an automatic routine; a
          simple rule-evaluation routine may provide much better overall
          performance. For further guidance, the article Lyness [3] is
          recommended.

          3.1. One-dimensional Integrals over a Finite Interval

          (a)   Integrand defined as a set of points

                If f(x) is defined numerically at four or more points, then
                the Gill-Miller finite difference method (D01GAF) should be
                used. The interval of integration is taken to coincide with
                the range of x-values of the points supplied. It is in the
                nature of this problem that any routine may be unreliable.
                In order to check results independently and so as to
                provide an alternative technique the user may fit the
                integrand by Chebyshev series using E02ADF and then use
                routines E02AJF and E02AKF to evaluate its integral (which
                need not be restricted to the range of the integration
                points, as is the case for D01GAF). A further alternative
                is to fit a cubic spline to the data using E02BAF and then
                to evaluate its integral using E02BDF.

          (b)   Integrand defined as a function

                If the functional form of f(x) is known, then one of the
                following approaches should be taken. They are arranged in
                the order from most specific to most general, hence the
                first applicable procedure in the list will be the most
                efficient.  However, if the user does not wish to make any
                assumptions about the integrand, the most reliable routine
                to use will be D01AJF, although this will in general be
                less efficient for simple integrals.
                (i)   Rule-evaluation routines

                      If f(x) is known to be sufficiently well-behaved
                      (more precisely, can be closely approximated by a
                      polynomial of moderate degree), a Gaussian routine
                      with a suitable number of abscissae may be used.

                      D01BBF may be used if it is required to examine the
                      weights and abscissae. In this case, the user should
                      write the code for the evaluation of quadrature
                      summation (6).

                (ii)  Automatic adaptive routines

                      Firstly, several routines are available for
                      integrands of the form w(x)g(x) where g(x) is a '
                      smooth' function (i.e., has no singularities, sharp
                      peaks or violent oscillations in the interval of
                      integration) and w(x) is a weight function of one of
                      the following forms:

                                   (alpha)     (beta)          k          l
                      if w(x)=(b-x)       (x-a)      (log(b-x)) (log(x-a))

                      where k,l=0 or 1, (alpha),(beta)>-1: use D01APF;

                      if w(x)=1/(x-c): use D01AQF (this integral is called
                      the Hilbert transform of g);

                      if w(x)=cos((omega)x) or sin((omega)x): use D01ANF
                      (this routine can also handle certain types of
                      singularities in g(x)).

                      Secondly, there are some routines for general f(x).
                      If f(x) is known to be free of singularities, though
                      it may be oscillatory, D01AKF may be used.

                      The most powerful of the finite interval integration
                      routine is D01AJF (which can cope with singularities
                      of several types). It may be used if none of the more
                      specific situations described above applies. D01AJF
                      is very reliable, particularly where the integrand
                      has singularities other than at an end-point, or has
                      discontinuities or cusps, and is therefore
                      recommended where the integrand is known to be badly-
                      behaved, or where its nature is completely unknown.

                      Most of the routines in this chapter require the user
                      to supply a function or subroutine to evaluate the
                      integrand at a single point.

                      If f(x) has singularities of certain types,
                      discontinuities or sharp peaks  occurring at known
                      points, the integral should be evaluated separately
                      over each of the subranges or D01ALF may be used.

          3.2. One-dimensional Integrals over a Semi-infinite or Infinite
               Interval

          (a)   Integrand defined as a set of points

                If f(x) is defined numerically at four or more points, and
                the portion of the integral lying outside the range of the
                points supplied may be neglected, then the Gill-Miller
                finite difference method, D01GAF, should be used.

          (b)   Integrand defined as a function
                (i)   Rule evaluation routines

                      If f(x) behaves approximately like a polynomial in x,
                      apart from a weight function of the form
                       -(beta)x
                      e                    (beta)>0 (semi-infinite interval,
                                                     lower limit finite);

                       -(beta)x
                 or   e                    (beta)<0 (semi-infinite interval,
                                                     upper limit finite);

                                          2
                        -(beta)(x-(alpha))
                 or    e                   (beta)>0 (infinite interval);

                               or if f(x) behaves approximately like a
                                                  -1
                               polynomial in (x+B)   (semi-infinite range);
                               then the Gaussian routines may be used.

                               D01BBF may be used if it is required to
                               examine the weights and abscissae. In this
                               case, the user should write the code for the
                               evaluation of quadrature summation (6).

                (ii)  Automatic adaptive routines

                      D01AMF may be used, except for integrands which decay
                      slowly towards an infinite end-point, and oscillate
                      in sign over the entire range. For this class, it may
                      be possible to calculate the integral by integrating
                      between the zeros and invoking some extrapolation
                      process.

                      D01ASF may be used for integrals involving weight
                      functions of the form cos((omega)x) and sin((omega)x)
                      over a semi-infinite interval (lower limit finite).

                      The following alternative procedures are mentioned
                      for completeness, though their use will rarely be
                      necessary:
                      (1)   If the integrand decays rapidly towards an
                            infinite end-point, a finite cut-off may be
                            chosen, and the finite range methods applied.

                      (2)   If the only irregularities occur in the finite
                            part (apart from a singularity at the finite
                            limit, with which D01AMF can cope), the range
                            may be divided, with D01AMF used on the
                            infinite part.

                      (3)   A transformation to finite range may be
                            employed, e.g.
                                             1-t
                                          x= ---  or x=-log t
                                              t            e
                            will transform (0,infty) to (1,0) while for
                            infinite ranges we have
                                   +infty        infty
                                   /             /
                                   |     f(x)dx= |    [f(x)+f(-x)]dx.
                                   /             /
                                   -infty        0
                            If the integrand behaves badly on (-infty,0)
                            and well on (0,infty) or vice versa it is
                            better to compute it as
                                       0             infty
                                       /             /
                                       |     f(x)dx+ |    f(x)dx.
                                       /             /
                                       -infty        0
                            This saves computing unnecessary function
                            values in the semi-infinite range where the
                            function is well behaved.

          3.3. Multi-dimensional Integrals

          A number of techniques are available in this area and the choice
          depends to a large extent on the dimension and the required
          accuracy. It can be advantageous to use more than one technique
          as a confirmation of accuracy particularly for high dimensional
          integrations. Many of the routines incorporate the transformation
          procedure REGION which allows general product regions to be
          easily dealt with in terms of conversion to the standard n-cube
          region.

          (a)   Products of one-dimensional rules (suitable for up to about
                5 dimensions)

                If f(x ,x ,...,x ) is known to be a sufficiently well-
                      1  2      n
                behaved function of each variable x , apart possibly from
                                                   i
                weight functions of the types provided, a product of
                Gaussian rules may be used. These are provided by D01BBF.
                In this case, the user should write the code for the
                evaluation of quadrature summation (6). Rules for finite,
                semi-infinite and infinite ranges are included.

                The one-dimensional routines may also be used recursively.
                For example, the two-dimensional integral
                                       b  b
                                        1  2
                                       /  /
                                    I= |  | f(x,y)dydx
                                       /  /
                                       a  a
                                        1  2
                may be expressed as
                                          b
                                           1
                                          /
                                       I= | F(x)dx,
                                          /
                                          a
                                           1
                where
                                           b
                                            2
                                           /
                                     F(x)= | f(x,y)dy.
                                           /
                                           a
                                            2
                The user segment to evaluate F(x) will call the integration
                routine for the y-integration, which will call another user
                segment for f(x,y) as a function of y (x being effectively
                a constant). Note that, as Fortran is not a recursive
                language,  a different library integration routine must be
                used for each dimension. Apart from this restriction, the
                full range of one-dimensional routines are available, for
                finite/infinite intervals, constant/variable limits, rule
                evaluation/automatic strategies etc.

          (b)   Automatic routines (D01GBF and D01FCF)

                Both routines are for integrals of the form
                         b  b     b
                          1  2     n
                         /  /     /
                         |  | ... | f(x ,x ,...,x )dx dx   ...dx .
                         /  /     /    1  2      n   n  n-1     1
                         a  a     a
                          1  2     n
                D01GBF is an adaptive Monte Carlo routine. This routine is
                       usually slow and not recommended for high accuracy
                       work. It is a robust routine that can often be used
                       for low accuracy results with highly irregular
                       integrands or when n is large.

                D01FCF is an adaptive deterministic routine. Convergence is
                       fast for well-behaved integrands. Highly accurate
                       results can often be obtained for n between 2 and 5,
                       using significantly fewer integrand evaluations than
                       would be required by D01GBF. The routine will
                       usually work when the integrand is mildly singular
                       and for n<=10 should be used before D01GBF. If it is
                       known in advance that the integrand is highly
                       irregular, it is best to compare results from at
                       least two different routines.

          There are many problems for which one or both of the routines
          will require large amounts of computing time to obtain even
          moderately accurate results. The amount of computing time is
          controlled by the number of integrand evaluations allowed by the
          user, and users should set this parameter carefully, with
          reference to the time available and the accuracy desired.


          3.4. Decision Trees

          (i) One-dimensional integrals over a finite interval. (If in
          doubt, follow the downward branch.)



                   Please see figure in printed Reference Manual


          (ii) One-dimensional integrals over a semi-infinite or infinite
          interval. (If in doubt, follow the downward branch.)



                   Please see figure in printed Reference Manual


          D01 -- Quadrature                                 Contents -- D01
          Chapter D01

          Quadrature

          D01AJF  1-D quadrature, adaptive, finite interval, strategy due
                  to Piessens and de Doncker, allowing for badly-behaved
                  integrands

          D01AKF  1-D quadrature, adaptive, finite interval, method
                  suitable for oscillating functions

          D01ALF  1-D quadrature, adaptive, finite interval, allowing for
                  singularities at user-specified break-points

          D01AMF  1-D quadrature, adaptive, infinite or semi-infinite
                  interval

          D01ANF  1-D quadrature, adaptive, finite interval, weight
                  function cos((omega)x) or sin((omega)x)

          D01APF  1-D quadrature, adaptive, finite interval, weight
                  function with end-point singularities of algebraico-
                  logarithmic type

          D01AQF  1-D quadrature, adaptive, finite interval, weight
                  function 1/(x-c), Cauchy principal value (Hilbert
                  transform)

          D01ASF  1-D quadrature, adaptive, semi-infinite interval, weight
                  function cos((omega)x) or sin((omega)x)

          D01BBF  Weights and abscissae for Gaussian quadrature rules

          D01FCF  Multi-dimensional adaptive quadrature over hyper-
                  rectangle

          D01GAF  1-D quadrature, integration of function defined by data
                  values, Gill-Miller method

          D01GBF  Multi-dimensional quadrature over hyper-rectangle, Monte
                  Carlo method

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01ajf}{NAG On-line Documentation: d01ajf}
\beginscroll
\begin{verbatim}



     D01AJF(3NAG)      Foundation Library (12/10/92)      D01AJF(3NAG)



          D01 -- Quadrature                                          D01AJF
                  D01AJF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01AJF is a general-purpose integrator which calculates an
          approximation to the integral of a function f(x) over a finite
          interval [a,b]:

                                        b
                                        /
                                     I= |f(x)dx.
                                        /
                                        a

          2. Specification

                 SUBROUTINE D01AJF (F, A, B, EPSABS, EPSREL, RESULT,
                1                   ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION F, A, B, EPSABS, EPSREL, RESULT, ABSERR, W
                1                 (LW)
                 EXTERNAL         F

          3. Description

          D01AJF is based upon the QUADPACK routine QAGS (Piessens et al
          [3]). It is an adaptive routine, using the Gauss 10-point and
          Kronrod 21-point rules. The algorithm, described by de Doncker
          [1], incorporates a global acceptance criterion (as defined by
          Malcolm and Simpson [2]) together with the (epsilon)-algorithm
          (Wynn [4]) to perform extrapolation. The local error estimation
          is described by Piessens et al [3].

          The routine is suitable as a general purpose integrator, and can
          be used when the integrand has singularities, especially when
          these are of algebraic or logarithmic type.

          D01AJF requires the user to supply a function to evaluate the
          integrand at a single point.

          The routine D01ATF(*) uses an identical algorithm but requires
          the user to supply a subroutine to evaluate the integrand at an
          array of points. Therefore D01ATF(*) will be more efficient if
          the evaluation can be performed in vector mode on a vector-
          processing machine.

          4. References

          [1]   De Doncker E (1978) An Adaptive Extrapolation Algorithm for
                Automatic Integration. Signum Newsletter. 13 (2) 12--18.

          [2]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [4]   Wynn P (1956) On a Device for Computing the e (S )
                                                             m  n
                Transformation. Math. Tables Aids Comput. 10 91--96.

          5. Parameters

           1:  F -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               F must return the value of the integrand f at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION F (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the integrand f must be
                    evaluated.
               F must be declared as EXTERNAL in the (sub)program from
               which D01AJF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. It is not
               necessary that a<b.

           4:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           5:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           6:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           7:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

           8:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

           9:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01AJF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: a value
               in the range 800 to 2000 is adequate for most problems.
               Constraint: LW >= 4.

          10:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          11:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01AJF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW. Suggested
               value: LIW = LW/4. Constraint: LIW >= 1.

          12:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requirements
               being achieved. Look at the integrand in order to determine
               the integration difficulties. If the position of a local
               difficulty within the interval can be determined (e.g. a
               singularity of the integrand or its derivative, a peak, a
               discontinuity, etc) you will probably gain from splitting up
               the interval at this point and calling the integrator on the
               subranges. If necessary, another integrator, which is
               designed for handling the type of difficulty involved, must
               be used. Alternatively, consider relaxing the accuracy
               requirements specified by EPSABS and EPSREL, or increasing
               the amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. The error may be under-estimated. Consider
               requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.

          IFAIL= 4
               The requested tolerance cannot be achieved, because the
               extrapolation does not increase the accuracy satisfactorily;
               the returned result is the best which can be obtained. The
               same advice applies as in the case of IFAIL = 1.

          IFAIL= 5
               The integral is probably divergent, or slowly convergent.
               Please note that divergence can occur with any non-zero
               value of IFAIL.

          IFAIL= 6
               On entry LW < 4,

               or       LIW < 1.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerance. Moreover it returns the quantity ABSERR which,
          in normal circumstances, satisfies

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01AJF along with the integral
          contributions and error estimates over the sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
          estimate.

                 b
                  i                       n
                 /                        --
          Then,  | f(x)dx~=r  and RESULT= >  r , unless D01AJF terminates
                 /          i             --  i
                 a                        i=1
                  i
          while testing for divergence of the integral (see Piessens et al
          [3], Section 3.4.3). In this case, RESULT (and ABSERR) are taken
          to be the values returned from the extrapolation process. The
          value of n is returned in IW(1), and the values a , b , e  and r
                                                           i   i   i      i
          are stored consecutively in the array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i) and
                i

               r =W(3n+i).
                i

          9. Example

          To compute
                              2(pi)
                              /         x sin(30x)
                              |     ----------------dx.
                              /         
                              0        /(  (  x  )2)
                                      / (1-(-----) )
                                    \/  (  (2(pi)) )

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01akf}{NAG On-line Documentation: d01akf}
\beginscroll
\begin{verbatim}



     D01AKF(3NAG)      Foundation Library (12/10/92)      D01AKF(3NAG)



          D01 -- Quadrature                                          D01AKF
                  D01AKF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01AKF is an adaptive integrator, especially suited to
          oscillating, non-singular integrands, which calculates an
          approximation to the integral of a function f(x) over a finite
          interval [a,b]:

                                        b
                                        /
                                     I= |f(x)dx.
                                        /
                                        a

          2. Specification

                 SUBROUTINE D01AKF (F, A, B, EPSABS, EPSREL, RESULT,
                1                   ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION F, A, B, EPSABS, EPSREL, RESULT, ABSERR, W
                1                 (LW)
                 EXTERNAL         F

          3. Description

          D01AKF is based upon the QUADPACK routine QAG (Piessens et al [3]
          ). It is an adaptive routine, using the Gauss 30-point and
          Kronrod 61-point rules. A 'global' acceptance criterion (as
          defined by Malcolm and Simpson [1]) is used. The local error
          estimation is described in Piessens et al [3].

          Because this routine is based on integration rules of high order,
          it is especially suitable for non-singular oscillating
          integrands.

          D01AKF requires the user to supply a function to evaluate the
          integrand at a single point.

          The routine D01AUF(*) uses an identical algorithm but requires
          the user to supply a subroutine to evaluate the integrand at an
          array of points. Therefore D01AUF(*) will be more efficient if
          the evaluation can be performed in vector mode on a vector-
          processing machine.

          D01AUF(*) also has an additional parameter KEY which allows the
          user to select from six different Gauss-Kronrod rules.

          4. References

          [1]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [2]   Piessens R (1973) An Algorithm for Automatic Integration.
                Angewandte Informatik. 15 399--401.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          5. Parameters

           1:  F -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               F must return the value of the integrand f at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION F (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the integrand f must be
                    evaluated.
               F must be declared as EXTERNAL in the (sub)program from
               which D01AKF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. It is not
               necessary that a<b.

           4:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           5:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           6:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           7:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound |I-RESULT|.

           8:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

           9:  LW -- INTEGER                                          Input
               On entry: the dimension of W, as declared in the (sub)
               program from which D01AKF is called. The value of LW
               (together with that of LIW below) imposes a bound on the
               number of sub-intervals into which the interval of
               integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: a value
               in the range 800 to 2000 is adequate for most problems.
               Constraint: LW >= 4. See IW below.

          10:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          11:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01AKF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW. Suggested
               value: LIW = LW/4. Constraint: LIW >= 1.

          12:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requirements
               being achieved. Look at the integrand in order to determine
               the integration difficulties. Probably another integrator
               which is designed for handling the type of difficulty
               involved must be used. Alternatively, consider relaxing the
               accuracy requirements specified by EPSABS and EPSREL, or
               increasing the amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. Consider  requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               On entry LW < 4,

               or       LIW < 1.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerances. Moreover it returns the quantity ABSERR which,
          in normal circumstances satisfies

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01AKF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                       n
                           /                        --
          estimate. Then,  | f(x)dx~=r  and RESULT= >  r . The value of n
                           /          i             --  i
                           a                        i=1
                            i
          is returned in IW(1), and the values a , b , e  and r  are stored
                                                i   i   i      i
          consecutively in the array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i) and
                i

               r =W(3n+i).
                i

          9. Example

          To compute

                                2(pi)
                                /
                                |    x sin(30x) cosx dx.
                                /
                                0

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01alf}{NAG On-line Documentation: d01alf}
\beginscroll
\begin{verbatim}



     D01ALF(3NAG)      Foundation Library (12/10/92)      D01ALF(3NAG)



          D01 -- Quadrature                                          D01ALF
                  D01ALF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01ALF is a general purpose integrator which calculates an
          approximation to the integral of a function f(x) over a finite
          interval [a,b]:

                                        b
                                        /
                                     I= |f(x)dx
                                        /
                                        a

          where the integrand may have local singular behaviour at a finite
          number of points within the integration interval.

          2. Specification

                 SUBROUTINE D01ALF (F, A, B, NPTS, POINTS, EPSABS, EPSREL,
                1                   RESULT, ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          NPTS, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION F, A, B, POINTS(*), EPSABS, EPSREL,
                1                 RESULT, ABSERR, W(LW)
                 EXTERNAL         F

          3. Description

          D01ALF is based upon the QUADPACK routine QAGP (Piessens et al
          [3]). It is very similar to D01AJF, but allows the user to supply
          difficult. It is an adaptive routine, using the Gauss 10-point
          and Kronrod 21-point rules. The algorithm described by de Doncker
          [1], incorporates a global acceptance criterion (as defined by
          Malcolm and Simpson [2]) together with the (epsilon)-algorithm
          (Wynn [4]) to perform extrapolation. The user-supplied 'break-
          points'always occur as the end-points of some sub-interval during
          the adaptive process. The local error estimation is described by
          Piessens et al [3].

          4. References

          [1]   De Doncker E (1978) An Adaptive Extrapolation Algorithm for
                Automatic Integration. Signum Newsletter. 13 (2) 12--18.

          [2]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [4]   Wynn P (1956) On a Device for Computing the e (S )
                                                             m  n
                Transformation. Math. Tables Aids Comput. 10 91--96.

          5. Parameters

           1:  F -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                    External Procedure
               F must return the value of the integrand f at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION F (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the integrand f must be
                    evaluated.
               F must be declared as EXTERNAL in the (sub)program from
               which D01ALF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. It is not
               necessary that a<b.

           4:  NPTS -- INTEGER                                        Input
               On entry: the number of user-supplied break-points within
               the integration interval. Constraint: NPTS >= 0.

           5:  POINTS(NPTS) -- DOUBLE PRECISION array                 Input
               On entry: the user-specified break-points. Constraint: the
               break-points must all lie within the interval of integration
               (but may be supplied in any order).

           6:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           7:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           8:  RESULT -- DOUBLE PRECISION                             Input
               On entry: the approximation to the integral I.

           9:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

          10:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

          11:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01ALF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed (LW-2*NPTS-4)/4. The more
               difficult the integrand, the larger LW should be. Suggested
               value: a value in the range 800 to 2000 is adequate for most
               problems. Constraint: LW>=2*NPTS+8.

          12:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          13:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01ALF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed (LIW-NPTS-2)/2.
               Suggested value: LIW = LW/2. Constraint: LIW >= NPTS + 4.

          14:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached, without the accuracy
               requirements being achieved. Look at the integrand in order
               to determine the integration difficulties. If the position
               of a local difficulty within the interval can be determined
               (e.g. a singularity of the integrand or its derivative, a
               peak, a discontinuity, etc) it should be supplied to the
               routine as an element of the vector POINTS. If necessary,
               another integrator should be used, which is designed for
               handling the type of difficulty involved. Alternatively,
               consider relaxing the accuracy requirements specified by
               EPSABS and EPSREL, or increasing the amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. The error may be under-estimated. Consider
               requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               The requested tolerance cannot be achieved, because the
               extrapolation does not increase the accuracy satisfactorily;
               the result returned is the best which can be obtained.  The
               same advice applies as in the case IFAIL = 1.

          IFAIL= 5
               The integral is probably divergent, or slowly convergent.
               Please note that divergence can also occur with any other
               non-zero value of IFAIL.

          IFAIL= 6
               The input is invalid: break-points are specified outside the
               integration range, NPTS > LIMIT or NPTS < 0. RESULT and
               ABSERR are set to zero.

          IFAIL= 7
               On entry LW<2*NPTS+8,

               or       LIW<NPTS+4.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:
                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerances. Moreover it returns the quantity ABSERR which,
          in normal circumstances, satisfies

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and on the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01ALF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                       n
                           /                        --
          estimate. Then,  | f(x)dx~=r  and RESULT= >  r  unless D01ALF
                           /          i             --  i
                           a                        i=1
                            i
          terminates while testing for divergence of the integral (see
          Piessens et al [3] Section 3.4.3). In this case, RESULT (and
          ABSERR) are taken to be the values returned from the
          extrapolation process. The value of n is returned in IW(1), and
          the values a , b , e  and r  are stored consecutively in the
                      i   i   i      i
          array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i) and
                i

               r =W(3n+i).
                i

          9. Example

          To compute

                                    1
                                    /     1
                                    | ---------dx.
                                    /   
                                    0 \/|x-1/7|

          A break-point is specified at x=1/7, at which point the integrand
          is infinite. (For definiteness the function FST returns the value
          0.0 at this point.)

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01amf}{NAG On-line Documentation: d01amf}
\beginscroll
\begin{verbatim}



     D01AMF(3NAG)      Foundation Library (12/10/92)      D01AMF(3NAG)



          D01 -- Quadrature                                          D01AMF
                  D01AMF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01AMF calculates an approximation to the integral of a function
          f(x) over an infinite or semi-infinite interval [a,b]:

                                        b
                                        /
                                     I= |f(x)dx
                                        /
                                        a

          2. Specification

                 SUBROUTINE D01AMF (F, BOUND, INF, EPSABS, EPSREL, RESULT,
                1                   ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          INF, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION F, BOUND, EPSABS, EPSREL, RESULT, ABSERR,
                1                 W(LW)
                 EXTERNAL         F

          3. Description

          D01AMF is based on the QUADPACK routine QAGI (Piessens et al [3])
          [0,1] using one of the identities:

                            a             1
                            /             / (   1-t) 1
                            |     f(x)dx= |f(a- ---) --dt
                            /             / (    t )  2
                            -infty        0          t

                             infty        1
                             /            / (   1-t) 1
                             |    f(x)dx= |f(a+ ---) --dt
                             /            / (    t )  2
                             a            0          t

             infty         infty                1
             /             /                    /[ ( 1-t)  ( -1+t)] 1
             |     f(x)dx= |    (f(x)+f(-x))dx= |[f( ---)+f( ----)] --dt
             /             /                    /[ (  t )  (  t  )]  2
             -infty        0                    0                   t

          where a represents a finite integration limit. An adaptive
          procedure, based on the Gauss seven-point and Kronrod 15-point
          rules, is then employed on the transformed integral. The
          algorithm, described by de Doncker [1], incorporates a global
          acceptance criterion (as defined by Malcolm and Simpson [2])
          together with the (epsilon)-algorithm (Wynn [4]) to perform
          extrapolation. The local error estimation is described by
          Piessens et al [3].

          4. References

          [1]   De Doncker E (1978) An Adaptive Extrapolation Algorithm for
                Automatic Integration. Signum Newsletter. 13 (2) 12--18.

          [2]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [4]   Wynn P (1956) On a Device for Computing the e (S )
                                                             m  n
                Transformation. Math. Tables Aids Comput. 10 91--96.

          5. Parameters

           1:  F -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               On entry: the point at which the integrand f must be
               evaluated.

               Its specification is:

                      DOUBLE PRECISION FUNCTION F (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the integrand f must be
                    evaluated.
               F must be declared as EXTERNAL in the (sub)program from
               which D01AMF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  BOUND -- DOUBLE PRECISION                              Input
               On entry: the finite limit of the integration range (if
               present). BOUND is not used if the interval is doubly
               infinite.

           3:  INF -- INTEGER                                         Input
               On entry: indicates the kind of integration range:
                    if INF =1, the range is [BOUND, +infty)

                    if INF =-1, the range is (-infty, BOUND]

                    if INF =+2, the range is (-infty, +infty).
                Constraint: INF =-1, 1 or 2.

           4:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           5:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           6:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           7:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

           8:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

           9:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01AMF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: a value
               in the range 800 to 2000 is adequate for most problems.
               Constraint: LW >= 4.

          10:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          11:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01AMF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW. Suggested
               value: LIW = LW/4. Constraint: LIW >= 1.

          12:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.
               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the requested accuracy
               requirements being achieved. Look at the integrand in order
               to determine the integration difficulties. If the position
               of a local difficulty within the interval can be determined
               (e.g. a singularity of the integrand or its derivative, a
               peak, a discontinuity, etc) you will probably gain from
               splitting up the interval at this point and calling D01AMF
               on the infinite subrange and an appropriate integrator on
               the finite subrange. Alternatively, consider relaxing the
               accuracy requirements specified by EPSABS and EPSREL, or
               increasing the amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. The error may be underestimated. Consider
               requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               The requested tolerance cannot be achieved, because the
               extrapolation does not increase the accuracy satisfactorily;
               the returned result is the best which can be obtained.  The
               same advice applies as in the case of IFAIL = 1.

          IFAIL= 5
               The integral is probably divergent, or slowly convergent. It
               must be noted that divergence can also occur with any other
               non-zero value of IFAIL.

          IFAIL= 6
               On entry LW < 4,

               or       LIW < 1,

               or       INF /= -1, 1 or 2.
               Please note that divergence can occur with any non-zero
               value of IFAIL.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerances. Moreover it returns the quantity ABSERR, which,
          in normal circumstances, satisfies

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01AMF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                       n
                           /                        --
          estimate. Then,  | f(x)dx~=r  and RESULT= >  r  unless D01AMF
                           /          i             --  i
                           a                        i=1
                            i
          terminates while testing for divergence of the integral (see
          Piessens et al [3] Section 3.4.3). In this case, RESULT (and
          ABSERR) are taken to be the values returned from the
          extrapolation process. The value of n is returned in IW(1), and
          the values a , b , e  and r  are stored consecutively in the
                      i   i   i      i
          array W, that is:

                    a =W(i),b =W(n+i),e =W(2n+i) and r =W(3n+i).
                     i       i         i              i

          Note: that this information applies to the integral transformed
          to (0,1) as described in Section 3, not to the original integral.

          9. Example

          To compute

                                  infty
                                  /        1
                                  |     --------dx.
                                  /            
                                  0     (x+1)\/x

          The exact answer is (pi).

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01anf}{NAG On-line Documentation: d01anf}
\beginscroll
\begin{verbatim}



     D01ANF(3NAG)      Foundation Library (12/10/92)      D01ANF(3NAG)



          D01 -- Quadrature                                          D01ANF
                  D01ANF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01ANF calculates an approximation to the sine or the cosine
          transform of a function g over [a,b]:

             b                                  b
             /                                  /
          I= |g(x)||,sin((omega)x)||,dx  or  I= |g(x)||,cos((omega)x)||,dx
             /                                  /
             a                                  a

          (for a user-specified value of (omega)).

          2. Specification

                 SUBROUTINE D01ANF (G, A, B, OMEGA, KEY, EPSABS, EPSREL,
                1                   RESULT, ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          KEY, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION G, A, B, OMEGA, EPSABS, EPSREL, RESULT,
                1                 ABSERR, W(LW)
                 EXTERNAL         G

          3. Description

          D01ANF is based upon the QUADPACK routine QFOUR (Piessens et al
          [3]). It is an adaptive routine, designed to integrate a function
          of the form g(x)w(x), where w(x) is either sin((omega)x) or
          cos((omega)x). If a sub-interval has length

                                             -l
                                     L=|b-a|2

          then the integration over this sub-interval is performed by means
          of a modified Clenshaw-Curtis procedure (Piessens and Branders
          [2]) if L(omega)>4 and l<=20. In this case a Chebyshev-series
          approximation of degree 24 is used to approximate g(x), while an
          error estimate is computed from this approximation together with
          that obtained using Chebyshev-series of degree 12. If the above
          conditions do not hold then Gauss 7-point and Kronrod 15-point
          rules are used. The algorithm, described in [3], incorporates a
          global acceptance criterion (as defined in Malcolm and Simpson
          [1]) together with the (epsilon)-algorithm Wynn [4] to perform
          extrapolation. The local error estimation is described in [3].

          4. References

          [1]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [2]   Piessens R and Branders M (1975) Algorithm 002. Computation
                of Oscillating Integrals. J. Comput. Appl. Math. 1 153--164.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [4]   Wynn P (1956) On a Device for Computing the e (S )
                                                             m  n
                Transformation. Math. Tables Aids Comput. 10 91--96.

          5. Parameters

           1:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               G must return the value of the function g at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION G (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the function g must be
                    evaluated.
               G must be declared as EXTERNAL in the (sub)program from
               which D01ANF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. It is not
               necessary that a<b.

           4:  OMEGA -- DOUBLE PRECISION                              Input
               On entry: the parameter (omega) in the weight function of
               the transform.

           5:  KEY -- INTEGER                                         Input
               On entry: indicates which integral is to be computed:
                    if KEY = 1, w(x)=cos((omega)x);

                    if KEY = 2, w(x)=sin((omega)x).
                Constraint: KEY = 1 or 2.

           6:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           7:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           8:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           9:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

          10:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

          11:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01ANF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: a value
               in the range 800 to 2000 is adequate for most problems.
               Constraint: LW >= 4.

          12:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          13:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01ANF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW/2. Suggested
               value: LIW = LW/2. Constraint: LIW >= 2.

          14:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requested
               being achieved. Look at the integrand in order to determine
               the integration difficulties. If the position of a local
               difficulty within the interval can be determined (e.g. a
               singularity of the integrand or its derivative, a peak, a
               discontinuity, etc) you will probably gain from splitting up
               the interval at this point and calling the integrator on the
               subranges. If necessary, another integrator, which is
               designed for handling the type of difficulty involved, must
               be used. Alternatively consider relaxing the accuracy
               requirements specified by EPSABS and EPSREL, or increasing
               amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. The error may be underestimated. Consider
               requesting less accuracy.

          IFAIL= 3
               Extremely bad local behaviour of g(x) causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               The requested tolerance cannot be achieved because the
               extrapolation does not increase the accuracy satisfactorily;
               the returned result is the best which can be obtained.  The
               same advice applies as in the case of IFAIL = 1.

          IFAIL= 5
               The integral is probably divergent, or slowly convergent. It
               must be noted that divergence can occur with any non-zero
               value of IFAIL.

          IFAIL= 6
               On entry KEY < 1,

               or       KEY > 2.

          IFAIL= 7
               On entry LW < 4,

               or       LIW < 2.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          tolerances. Moreover it returns the quantity ABSERR, which, in
          normal circumstances, satisfies

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and on the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01ANF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                           n
                           /                            --
          estimate. Then,  | g(x)w(x)dx~=r  and RESULT= >  r  unless D01ANF
                           /              i             --  i
                           a                            i=1
                            i
          terminates while testing for divergence of the integral (see
          Piessens et al [3] Section 3.4.3). In this case, RESULT (and
          ABSERR) are taken to be the values returned from the
          extrapolation process. The value of n is returned in IW(1), and
          the values a , b , e  and r  are stored consecutively in the
                      i   i   i      i
          array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i) and
                i

               r =W(3n+i).
                i

          9. Example

          To compute

                                 1
                                 /
                                 |ln(x)sin(10(pi)x)dx.
                                 /
                                 0

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01apf}{NAG On-line Documentation: d01apf}
\beginscroll
\begin{verbatim}



     D01APF(3NAG)      Foundation Library (12/10/92)      D01APF(3NAG)



          D01 -- Quadrature                                          D01APF
                  D01APF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01APF is an adaptive integrator which calculates an
          approximation to the integral of a function g(x)w(x) over a
          finite interval [a,b]:

                                      b
                                      /
                                   I= |g(x)w(x)dx
                                      /
                                      a

          where the weight function w has end-point singularities of
          algebraico-logarithmic type.

          2. Specification

                 SUBROUTINE D01APF (G, A, B, ALFA, BETA, KEY, EPSABS,
                1                   EPSREL, RESULT, ABSERR, W, LW, IW, LIW,
                2                   IFAIL)
                 INTEGER          KEY, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION G, A, B, ALFA, BETA, EPSABS, EPSREL,
                1                 RESULT, ABSERR, W(LW)
                 EXTERNAL         G

          3. Description

          D01APF is based upon the QUADPACK routine QAWSE (Piessens et al
          [3]) and integrates a function of the form g(x)w(x), where the
          weight function w(x) may have algebraico-logarithmic
          singularities at the end-points a and/or b. The strategy is a
          modification of that in D01AKF. We start by bisecting the
          original interval and applying modified Clenshaw-Curtis
          integration of orders 12 and 24 to both halves. Clenshaw-Curtis
          integration is then used on all sub-intervals which have a or b
          as one of their end-points (Piessens et al [2]). On the other
          sub-intervals Gauss-Kronrod (7-15 point) integration is carried
          out.

          A 'global' acceptance criterion (as defined by Malcolm and
          Simpson [1]) is used. The local error estimation control is
          described by Piessens et al [3].

          4. References

          [1]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [2]   Piessens R, Mertens I and Branders M (1974) Integration of
                Functions having End-point Singularities. Angewandte
                Informatik. 16 65--68.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          5. Parameters

           1:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                        External Procedure
               G must return the value of the function g at a given point
               X.

               Its specification is:

                      DOUBLE PRECISION FUNCTION G (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the function g must be
                    evaluated.
               G must be declared as EXTERNAL in the (sub)program from
               which D01APF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. Constraint: B
               > A.

           4:  ALFA -- DOUBLE PRECISION                               Input
               On entry: the parameter (alpha) in the weight function.
               Constraint: ALFA>-1.

           5:  BETA -- DOUBLE PRECISION                               Input
               On entry: the parameter (beta) in the weight function.
               Constraint: BETA>-1.

           6:  KEY -- INTEGER                                         Input
               On entry: indicates which weight function is to be used:
                                          (alpha)     (beta)
                    if KEY = 1, w(x)=(x-a)       (b-x)

                                          (alpha)     (beta)
                    if KEY = 2, w(x)=(x-a)       (b-x)      ln(x-a)

                                          (alpha)     (beta)
                    if KEY = 3, w(x)=(x-a)       (b-x)      ln(b-x)

                                          (alpha)     (beta)
                    if KEY = 4, w(x)=(x-a)       (b-x)     ln(x-a)ln(b-x)

               Constraint: KEY = 1, 2, 3 or 4

           7:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           8:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           9:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

          10:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

          11:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

          12:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01APF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: LW =
               800 to 2000 is adequate for most problems. Constraint: LW >=
               8.

          13:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          14:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01APF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW. Suggested
               value: LIW = LW/4. Constraint: LIW >= 2.

          15:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requirements
               being achieved. Look at the integrand in order to determine
               the integration difficulties. If the position of a
               discontinuity or a singularity of algebraico-logarithmic
               type within the interval can be determined, the interval
               must be split up at this point and the integrator called on
               the subranges. If necessary, another integrator, which is
               designed for handling the difficulty involved, must be used.
               Alternatively consider relaxing the accuracy requirements
               specified by EPSABS and EPSREL, or increasing the amount of
               workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. Consider  requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               On entry B <= A,

               or       ALFA <= -1,

               or       BETA <= -1,

               or       KEY < 1,

               or       KEY > 4.

          IFAIL= 5
               On entry LW < 8,

               or       LIW < 2.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerances.

          Moreover it returns the quantity ABSERR which, in normal
          circumstances, satisfies:

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and on the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01APF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                           n
                           /                            --
          estimate. Then,  | f(x)w(x)dx~=r  and RESULT= >  r . The value of
                           /              i             --  i
                           a                            i=1
                            i
          n is returned in IW(1), and the values a , b , e  and r  are
                                                  i   i   i      i
          stored consecutively in the array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i),
                i

               r =W(3n+i).
                i

          9. Example

          To compute:

                               1
                               /
                               |ln(x)cos(10(pi)x)dx  and
                               /
                               0

                                    1
                                    / sin(10x)
                                    | --------dx.
                                    /   
                                    0 \/x(1-x)

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01aqf}{NAG On-line Documentation: d01aqf}
\beginscroll
\begin{verbatim}



     D01AQF(3NAG)      Foundation Library (12/10/92)      D01AQF(3NAG)



          D01 -- Quadrature                                          D01AQF
                  D01AQF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01AQF calculates an approximation to the Hilbert transform of a
          function g(x) over [a,b]:

                                        b
                                        / g(x)
                                     I= | ----dx
                                        / x-c
                                        a

          for user-specified values of a, b and c.

          2. Specification

                 SUBROUTINE D01AQF (G, A, B, C, EPSABS, EPSREL, RESULT,
                1                   ABSERR, W, LW, IW, LIW, IFAIL)
                 INTEGER          LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION G, A, B, C, EPSABS, EPSREL, RESULT,
                1                 ABSERR, W(LW)
                 EXTERNAL         G

          3. Description

          D01AQF is based upon the QUADPACK routine QAWC (Piessens et al
          [3]) and integrates a function of the form g(x)w(x), where the
          weight function

                                             1
                                      w(x)= ---
                                            x-c

          is that of the Hilbert transform. (If a<c<b the integral has to
          be interpreted in the sense of a Cauchy principal value.) It is
          an adaptive routine which employs a 'global' acceptance criterion
          (as defined by Malcolm and Simpson [1]). Special care is taken to
          ensure that c is never the end-point of a sub-interval (Piessens
          et al[2]). On each sub-interval (c ,c ) modified Clenshaw-Curtis
                                            1  2
          integration of orders 12 and 24 is performed if c -d<=c<=c +d
                                                           1        2
          where d=(c -c )/20. Otherwise the Gauss 7-point and Kronrod 15-
                    2  1
          point rules are used. The local error estimation is described by
          Piessens et al [3].

          4. References

          [1]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [2]   Piessens R, Van Roy-Branders M and Mertens I (1976) The
                Automatic Evaluation of Cauchy Principal Value Integrals.
                Angewandte Informatik. 18 31--35.

          [3]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          5. Parameters

           1:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               G must return the value of the function g at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION G (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the function g must be
                    evaluated.
               G must be declared as EXTERNAL in the (sub)program from
               which D01AQF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  B -- DOUBLE PRECISION                                  Input
               On entry: the upper limit of integration, b. It is not
               necessary that a<b.

           4:  C -- DOUBLE PRECISION                                  Input
               On entry: the parameter c in the weight function.
               Constraint: C must not equal A or B.

           5:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy required. If EPSABS is
               negative, the absolute value is used. See Section 7.

           6:  EPSREL -- DOUBLE PRECISION                             Input
               On entry: the relative accuracy required. If EPSREL is
               negative, the absolute value is used. See Section 7.

           7:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           8:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

           9:  W(LW) -- DOUBLE PRECISION array                       Output
               On exit: details of the computation, as described in
               Section 8.

          10:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01AQF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which the interval
               of integration may be divided by the routine. The number of
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: LW =
               800 to 2000 is adequate for most problems. Constraint: LW >=
               4.

          11:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the actual number of sub-intervals
               used. The rest of the array is used as workspace.

          12:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01AQF is called.
               The number of sub-intervals into which the interval of
               integration may be divided cannot exceed LIW. Suggested
               value: LIW = LW/4. Constraint: LIW >= 1.

          13:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requirements
               being achieved. Look at the integrand in order to determine
               the integration difficulties. Another integrator which is
               designed for handling the type of difficulty involved, must
               be used. Alternatively consider relaxing the accuracy
               requirements specified by EPSABS and EPSREL, or increasing
               the workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. Consider  requesting less accuracy.

          IFAIL= 3
               Extremely bad local behaviour of g(x) causes a very strong
               subdivision around one (or more) points of the interval. The
               same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               On entry C = A or C = B.

          IFAIL= 5
               On entry LW < 4,

               or       LIW < 1.

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                  |I-RESULT|<=tol,

          where

                           tol=max{|EPSABS|,|EPSREL|*|I|},

          and EPSABS and EPSREL are user-specified absolute and relative
          error tolerances. Moreover it returns the quantity ABSERR which,
          in normal circumstances satisfies:

                              |I-RESULT|<=ABSERR<=tol.

          8. Further Comments

          The time taken by the routine depends on the integrand and on the
          accuracy required.

          If IFAIL /= 0 on exit, then the user may wish to examine the
          contents of the array W, which contains the end-points of the
          sub-intervals used by D01AQF along with the integral
          contributions and error estimates over these sub-intervals.

          Specifically, for i=1,2,...,n, let r  denote the approximation to
                                              i
          the value of the integral over the sub-interval [a ,b ] in the
                                                            i  i
          partition of [a,b] and e  be the corresponding absolute error
                                  i
                           b
                            i                           n
                           /                            --
          estimate. Then,  | g(x)w(x)dx~=r  and RESULT= >  r . The value of
                           /              i             --  i
                           a                            i=1
                            i
          n is returned in IW(1), and the values a , b , e  and r  are
                                                  i   i   i      i
          stored consecutively in the array W, that is:

               a =W(i),
                i

               b =W(n+i),
                i

               e =W(2n+i) and
                i

               r =W(3n+i).
                i

          9. Example

          To compute the Cauchy principal value of

                                 1
                                 /         dx
                                 |  ----------------.
                                 /    2     2     1
                                 -1 (x +0.01 )(x- -)
                                                  2

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01asf}{NAG On-line Documentation: d01asf}
\beginscroll
\begin{verbatim}



     D01ASF(3NAG)      Foundation Library (12/10/92)      D01ASF(3NAG)



          D01 -- Quadrature                                          D01ASF
                  D01ASF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01ASF calculates an approximation to the sine or the cosine
          transform of a function g over [a,infty):


             infty
             /
          I= |    g(x)||,sin((omega)x)dx  or  I=
             /
             a
           infty
           /
           |    g(x)||,cos((omega)x)dx
           /
           a

          (for a user-specified value of (omega)).

          2. Specification

                 SUBROUTINE D01ASF (G, A, OMEGA, KEY, EPSABS, RESULT,
                1                   ABSERR, LIMLST, LST, ERLST, RSLST,
                2                   IERLST, W, LW, IW, LIW, IFAIL)
                 INTEGER          KEY, LIMLST, LST, IERLST(LIMLST), LW, IW
                1                 (LIW), LIW, IFAIL
                 DOUBLE PRECISION G, A, OMEGA, EPSABS, RESULT, ABSERR, ERLST
                1                 (LIMLST), RSLST(LIMLST), W(LW)
                 EXTERNAL         G

          3. Description

          D01ASF is based upon the QUADPACK routine QAWFE (Piessens et al
          [2]). It is an adaptive routine, designed to integrate a function
          of the form g(x)w(x) over a semi-infinite interval, where w(x) is
          either sin((omega)x) or cos((omega)x). Over successive intervals

                          C =[a+(k-1)c,a+kc],   k=1,2,...,LST
                           k

          integration is performed by the same algorithm as is used by
          D01ANF. The intervals C  are of constant length
                                 k

                    c={2[|(omega)|]+1}(pi)/|(omega)|,  (omega)/=0

          where [|(omega)|] represents the largest integer less than or
          equal to |(omega)|. Since c equals an odd number of half periods,
          the integral contributions over succeeding intervals will
          alternate in sign when the function g is positive and
          monotonically decreasing over [a,infty). The algorithm, described
          by [2], incorporates a global acceptance criterion (as defined by
          Malcolm and Simpson [1]) together with the (epsilon)-algorithm
          (Wynn [3]) to perform extrapolation. The local error estimation
          is described by Piessens et al [2].

          If (omega)=0 and KEY = 1, the routine uses the same algorithm as
          D01AMF (with EPSREL = 0.0).

          In contrast to the other routines in Chapter D01, D01ASF works
          only with a user-specified absolute error tolerance (EPSABS).
          Over the interval C  it attempts to satisfy the absolute accuracy
                             k
          requirement
                                   EPSA =U *EPSABS
                                       k  k
                         k-1
          where U =(1-p)p   , for k=1,2,... and p=0.9.
                 k

          However, when difficulties occur during the integration over the
          kth sub-interval C  such that the error flag IERLST(k) is non-
                            k
          zero, the accuracy requirement over subsequent intervals is
          relaxed. See Piessens et al [2] for more details.

          4. References

          [1]   Malcolm M A and Simpson R B (1976) Local Versus Global
                Strategies for Adaptive Quadrature. ACM Trans. Math. Softw.
                1 129--146.

          [2]   Piessens R, De Doncker-Kapenga E, Uberhuber C and Kahaner D
                (1983) QUADPACK, A Subroutine Package for Automatic
                Integration. Springer-Verlag.

          [3]   Wynn P (1956) On a Device for Computing the e (S )
                                                             m  n
                Transformation. Math. Tables Aids Comput. 10 91--96.

          5. Parameters

           1:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               G must return the value of the function g at a given point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION G (X)
                      DOUBLE PRECISION X

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the point at which the function g must be
                    evaluated.
               G must be declared as EXTERNAL in the (sub)program from
               which D01ASF is called. Parameters denoted as Input
               must not be changed by this procedure.

           2:  A -- DOUBLE PRECISION                                  Input
               On entry: the lower limit of integration, a.

           3:  OMEGA -- DOUBLE PRECISION                              Input
               On entry: the parameter (omega) in the weight function of
               the transform.

           4:  KEY -- INTEGER                                         Input
               On entry: indicates which integral is to be computed:
                    if KEY = 1, w(x)=cos((omega)x);

                    if KEY = 2, w(x)=sin((omega)x).
                Constraint: KEY = 1 or 2.

           5:  EPSABS -- DOUBLE PRECISION                             Input
               On entry: the absolute accuracy requirement. If EPSABS is
               negative, the absolute value is used. See Section 7.

           6:  RESULT -- DOUBLE PRECISION                            Output
               On exit: the approximation to the integral I.

           7:  ABSERR -- DOUBLE PRECISION                            Output
               On exit: an estimate of the modulus of the absolute error,
               which should be an upper bound for |I-RESULT|.

           8:  LIMLST -- INTEGER                                      Input
               On entry: an upper bound on the number of intervals C
                                                                     k
               needed for the integration. Suggested value: LIMLST = 50 is
               adequate for most problems. Constraint: LIMLST >= 3.

           9:  LST -- INTEGER                                        Output
               On exit: the number of intervals C  actually used for the
                                                  k
               integration.

          10:  ERLST(LIMLST) -- DOUBLE PRECISION array               Output
               On exit: ERLST(k) contains the error estimate corresponding
               to the integral contribution over the interval C , for
                                                               k
               k=1,2,...,LST.

          11:  RSLST(LIMLST) -- DOUBLE PRECISION array               Output
               On exit: RSLST(k) contains the integral contribution over
               the interval C  for k=1,2,...,LST.
                             k

          12:  IERLST(LIMLST) -- INTEGER array                       Output
               On exit: IERLST(k) contains the error flag corresponding to
               RSLST(k), for k=1,2,...,LST. See Section 6.

          13:  W(LW) -- DOUBLE PRECISION array                    Workspace

          14:  LW -- INTEGER                                          Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D01ASF is called.
               The value of LW (together with that of LIW below) imposes a
               bound on the number of sub-intervals into which each
               interval C  may be divided by the routine. The number of
                         k
               sub-intervals cannot exceed LW/4. The more difficult the
               integrand, the larger LW should be. Suggested value: a value
               in the range 800 to 2000 is adequate for most problems.
               Constraint: LW >= 4.

          15:  IW(LIW) -- INTEGER array                              Output
               On exit: IW(1) contains the maximum number of sub-intervals
               actually used for integrating over any of the intervals C .
                                                                        k
               The rest of the array is used as workspace.

          16:  LIW -- INTEGER                                         Input
               On entry:
               the dimension of the array IW as declared in the
               (sub)program from which D01ASF is called.
               The number of sub-intervals into which each interval C  may
                                                                     k
               be divided cannot exceed LIW/2. Suggested value: LIW = LW/2.
               Constraint: LIW >= 2.

          17:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               The maximum number of subdivisions allowed with the given
               workspace has been reached without the accuracy requirements
               being achieved. Look at the integrand in order to determine
               the integration difficulties. If the position of a local
               difficulty within the interval can be determined (e.g. a
               singularity of the integrand or its derivative, a peak, a
               discontinuity, etc) you will probably gain from splitting up
               the interval at this point and calling D01ASF on the
               infinite subrange and an appropriate integrator on the
               finite subrange. Alternatively, consider relaxing the
               accuracy requirements specified by EPSABS or increasing the
               amount of workspace.

          IFAIL= 2
               Round-off error prevents the requested tolerance from being
               achieved. The error may be underestimated. Consider
               requesting less accuracy.

          IFAIL= 3
               Extremely bad local integrand behaviour causes a very strong
               subdivision around one (or more) points of the interval.
               The same advice applies as in the case of IFAIL = 1.

          IFAIL= 4
               The requested tolerance cannot be achieved, because the
               extrapolation does not increase the accuracy satisfactorily;
               the returned result is the best which can be obtained.  The
               same advice applies as in the case of IFAIL = 1.

          IFAIL= 5
               The integral is probably divergent, or slowly convergent.
               Please note that divergence can occur with any non-zero
               value of IFAIL.

          IFAIL= 6
               On entry KEY < 1,

               or       KEY > 2,

               or       LIMLST < 3.

          IFAIL= 7
               Bad integration behaviour occurs within one or more of the
               intervals C . Location and type of the difficulty involved
                          k
               can be determined from the vector IERLST (see below).

          IFAIL= 8
               Maximum number of intervals C  (= LIMLST) allowed has been
                                            k
               achieved. Increase the value of LIMLST to allow more cycles.

          IFAIL= 9
               The extrapolation table constructed for convergence
               acceleration of the series formed by the integral
               contribution over the intervals C , does not converge to the
                                                k
               required accuracy.

          IFAIL= 10
               On entry LW < 4,

               or       LIW < 2.

          In the cases IFAIL = 7, 8 or 9, additional information about the
          cause of the error can be obtained from the array IERLST, as
          follows:

          IERLST(k)=1
                The maximum number of subdivisions = min(LW/4,LIW/2) has
                been achieved on the kth interval.

          IERLST(k)=2
                Occurrence of round-off error is detected and prevents the
                tolerance imposed on the kth interval from being achieved.

          IERLST(k)=3
                Extremely bad integrand behaviour occurs at some points of
                the kth interval.

          IERLST(k)=4
                The integration procedure over the kth interval does not
                converge (to within the required accuracy) due to round-off
                in the extrapolation procedure invoked on this interval. It
                is assumed that the result on this interval is the best
                which can be obtained.

          IERLST(k)=5
                The integral over the kth interval is probably divergent or
                slowly convergent. It must be noted that divergence can
                occur with any other value of IERLST(k).

          7. Accuracy

          The routine cannot guarantee, but in practice usually achieves,
          the following accuracy:

                                |I-RESULT|<=|EPSABS|,

          where EPSABS is the user-specified absolute error tolerance.
          Moreover, it returns the quantity ABSERR, which, in normal
          circumstances, satisfies

                            |I-RESULT|<=ABSERR<=|EPSABS|.

          8. Further Comments

          None.

          9. Example

          To compute

                               infty
                               /      1
                               |     ---cos((pi)x/2)dx.
                               /       
                               0     \/x

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01bbf}{NAG On-line Documentation: d01bbf}
\beginscroll
\begin{verbatim}



     D01BBF(3NAG)                 D01BBF                  D01BBF(3NAG)



     D01 -- Quadrature                                          D01BBF
             D01BBF -- NAG Foundation Library Routine Document

     Note: Before using this routine, please read the Users' Note for
     your implementation to check implementation-dependent details.
     The symbol (*) after a NAG routine name denotes a routine that is
     not included in the Foundation Library.

     Note for users via the AXIOM system: the interface to this routine
     has been enhanced for use with AXIOM and is slightly different to
     that offered in the standard version of the Foundation Library.

     1. Purpose

     D01BBF returns the weights and abscissae appropriate to a
     Gaussian quadrature formula with a specified number of abscissae.
     The formulae provided are Gauss-Legendre, Gauss-Rational, Gauss-
     Laguerre and Gauss-Hermite.

     2. Specification

            SUBROUTINE D01BBF (A, B, ITYPE, N, WEIGHT, ABSCIS, GTYPE,
           1                   IFAIL)
            INTEGER          ITYPE, N, GTYPE, IFAIL
            DOUBLE PRECISION A, B, WEIGHT(N), ABSCIS(N)

     3. Description

     This routine returns the weights and abscissae for use in the
     Gaussian quadrature of a function f(x). The quadrature takes the
     form

                                  n
                                  --
                               S= >  w f(x )
                                  --  i   i
                                  i=1

     where w  are the weights and x  are the abscissae (see Davis and
            i                      i
     Rabinowitz [1], Froberg [2], Ralston [3] or Stroud and Secrest
     [4]).

     Weights and abscissae are available for Gauss-Legendre, Gauss-
     Rational, Gauss-Laguerre and Gauss-Hermite quadrature, and for a
     selection of values of n (see Section 5).

     (a)   Gauss-Legendre Quadrature:
                                       b
                                       /
                                   S~= |f(x)dx
                                       /

                                       a
           where a and b are finite and it will be exact for any
           function of the form
                                       2n-1
                                       --     i
                                 f(x)= >   c x
                                       --   i
                                       i=0

     (b)   Gauss-Rational quadrature:
                infty                           a
                /                               /
            S~= |    f(x)dx  (a+b>0)   or   S~= |     f(x)dx  (a+b<0)
                /                               /
                a                               -infty
           and will be exact for any function of the form
                                          2n-1
                                          --              i
                                          >   c      (x+b)
                            2n+1   c      --   2n+1-i
                            --      i     i=0
                      f(x)= >    ------= ------------------
                            --        i           2n+1
                            i=2  (x+b)       (x+b)

     (c)   Gauss-Laguerre quadrature, adjusted weights option:
                 infty                          a
                 /                              /
             S~= |    f(x)dx  (b>0<)   or   S~= |     f(x)dx  (b<0)
                 /                              /
                 a                              -infty
           and will be exact for any function of the form
                                         2n-1
                                     -bx --     i
                               f(x)=e    >   c x
                                         --   i
                                         i=0

     (d)   Gauss-Hermite quadrature, adjusted weights option:
                                    +infty
                                    /
                                S~= |     f(x)dx
                                    /
                                    -infty
           and will be exact for any function of the form

                                      2 2n-1
                               -b(x-a)  --     i
                         f(x)=e         >   c x   (b>0)
                                        --   i
                                        i=0

     (e)   Gauss-Laguerre quadrature, normal weights option:

               infty                             a
               /     -bx                         /      -bx
           S~= |    e   f(x)dx  (b>0)   or   S~= |     e   f(x)dx
               /                                 /
               a                                 -infty


            (b<0)


           and will be exact for any function of the form
                                       2n-1
                                       --     i
                                 f(x)= >   c x
                                       --   i
                                       i=0

     (f)   Gauss-Hermite quadrature, normal weights option:

                                +infty        2
                                /      -b(x-a)
                            S~= |     e        f(x)dx
                                /
                                -infty
           and will be exact for any function of the form:
                                       2n-1
                                       --     i
                                 f(x)= >   c x
                                       --   i
                                       i=0

     Note: that the Gauss-Legendre abscissae, with a=-1, b=+1, are the
     zeros of the Legendre polynomials; the Gauss-Laguerre abscissae,
     with a=0, b=1, are the zeros of the Laguerre polynomials; and the
     Gauss-Hermite abscissae, with a=0, b=1, are the zeros of the
     Hermite polynomials.

     4. References

     [1]   Davis P J and Rabinowitz P (1967) Numerical Integration.
           Blaisdell Publishing Company. 33--52.

     [2]   Froberg C E (1965) Introduction to Numerical Analysis.
           Addison-Wesley. 181--187.

     [3]   Ralston A (1965) A First Course in Numerical Analysis.
           McGraw-Hill. 87--90.

     [4]   Stroud A H and Secrest D (1966) Gaussian Quadrature
           Formulas. Prentice-Hall.

     5. Parameters

      1:  A -- DOUBLE PRECISION                                  Input

      2:  B -- DOUBLE PRECISION                                  Input
          On entry: the quantities a and b as described in the
          appropriate subsection of Section 3.

      3:  ITYPE -- INTEGER                                       Input
          On entry: indicates the type of weights for Gauss-Laguerre
          or Gauss-Hermite quadrature (see Section 3):

          if ITYPE = 1, adjusted weights will be returned;

          if ITYPE = 0, normal weights will be returned.

          Constraint: ITYPE = 0 or 1.

          For Gauss-Legendre or Gauss-Rational quadrature, this
          parameter is not used.

      4:  N -- INTEGER                                           Input
          On entry: the number of weights and abscissae to be
          returned, n. Constraint: N =
          1,2,3,4,5,6,8,10,12,14,16,20,24,32,48 or 64.

      5:  WEIGHT(N) -- DOUBLE PRECISION array                   Output
          On exit: the N weights. For Gauss-Laguerre and Gauss-
          Hermite quadrature, these will be the adjusted weights if
          ITYPE = 1, and the normal weights if ITYPE = 0.

      6:  ABSCIS(N) -- DOUBLE PRECISION array                   Output
          On exit: the N abscissae.

      7:  GTYPE -- INTEGER                                       Input
          On entry: The value of GTYPE indicates which quadrature
          formula are to be used:
               GTYPE = 0 for Gauss-Legendre weights and abscissae;

               GTYPE = 1 for Gauss-Rational weights and abscissae;

               GTYPE = 2 for Gauss-Laguerre weights and abscissae;

               GTYPE = 3 for Gauss-Hermite weights and abscissae.

      8:  IFAIL -- INTEGER                                Input/Output
          On entry: IFAIL must be set to 0, -1 or 1. For users not
          familiar with this parameter (described in the Essential
          Introduction) the recommended value is 0.

          On exit: IFAIL = 0 unless the routine detects an error (see
          Section 6).

     6. Error Indicators and Warnings

     Errors detected by the routine:

     IFAIL= 1
          The N-point rule is not among those stored. If the soft fail
          option is used, the weights and abscissae returned will be
          those for the largest valid value of N less than the
          requested value, and the excess elements of WEIGHT and
          ABSCIS (i.e., up to the requested N) will be filled with
          zeros.

     IFAIL= 2
          The value of A and/or B is invalid.
               Gauss-Rational: A + B = 0

               Gauss-Laguerre: B = 0

               Gauss-Hermite: B <= 0
          If the soft fail option is used the weights and abscissae
          are returned as zero.

     IFAIL= 3
          Laguerre and Hermite normal weights only: underflow is
          occurring in evaluating one or more of the normal weights.
          If the soft fail option is used, the underflowing weights
          are returned as zero. A smaller value of N must be used; or
          adjusted weights should be used (ITYPE = 1). In the latter
          case, take care that underflow does not occur when
          evaluating the integrand appropriate for adjusted weights.

     IFAIL=4
         GTYPE < 0 or GTYPE > 3

     7. Accuracy

     The weights and abscissae are stored for standard values of A and
     B to full machine accuracy.

     8. Further Comments

     Timing is negligible.

     9. Example

     This example program returns the abscissae and (adjusted) weights
     for the six-point Gauss-Laguerre formula.

     The example program is not reproduced here. The source code for
     all example programs is distributed with the NAG Foundation
     Library software and should be available on-line.
\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01fcf}{NAG On-line Documentation: d01fcf}
\beginscroll
\begin{verbatim}



     D01FCF(3NAG)      Foundation Library (12/10/92)      D01FCF(3NAG)



          D01 -- Quadrature                                          D01FCF
                  D01FCF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01FCF attempts to evaluate a multi-dimensional integral (up to
          15 dimensions), with constant and finite limits, to a specified
          relative accuracy, using an adaptive subdivision strategy.

          2. Specification

                 SUBROUTINE D01FCF (NDIM, A, B, MINPTS, MAXPTS, FUNCTN,
                1                   EPS, ACC, LENWRK, WRKSTR, FINVAL, IFAIL)
                 INTEGER          NDIM, MINPTS, MAXPTS, LENWRK, IFAIL
                 DOUBLE PRECISION A(NDIM), B(NDIM), FUNCTN, EPS, ACC, WRKSTR
                1                 (LENWRK), FINVAL
                 EXTERNAL         FUNCTN

          3. Description

          The routine returns an estimate of a multi-dimensional integral
          over a hyper-rectangle (i.e., with constant limits), and also an
          estimate of the relative error. The user sets the relative
          accuracy required, supplies the integrand as a function
          subprogram (FUNCTN), and also sets the minimum and maximum
          acceptable number of calls to FUNCTN (in MINPTS and MAXPTS).

          The routine operates by repeated subdivision of the hyper-
          rectangular region into smaller hyper-rectangles. In each
          subregion, the integral is estimated using a seventh-degree rule,
          and an error estimate is obtained by comparison with a fifth-
          degree rule which uses a subset of the same points. The fourth
          differences of the integrand along each co-ordinate axis are
          evaluated, and the subregion is marked for possible future
          subdivision in half along that co-ordinate axis which has the
          largest absolute fourth difference.

          If the estimated errors, totalled over the subregions, exceed the
          requested relative error (or if fewer than MINPTS calls to FUNCTN
          have been made), further subdivision is necessary, and is
          performed on the subregion with the largest estimated error, that
          subregion being halved along the appropriate co-ordinate axis.

          The routine will fail if the requested relative error level has
          not been attained by the time MAXPTS calls to FUNCTN have been
          made; or, if the amount LENWRK of working storage is
          insufficient. A formula for the recommended value of LENWRK is
          given in Section 5. If a smaller value is used, and is exhausted
          in the course of execution, the routine switches to a less
          efficient mode of operation; only if this mode also breaks down
          is insufficient storage reported.

          D01FCF is based on the HALF subroutine developed by van Dooren
          and de Ridder [1]. It uses a different basic rule, described by
          Genz and Malik [2].

          4. References

          [1]   Van Dooren P and De Ridder L (1976) An Adaptive Algorithm
                for Numerical Integration over an N-dimensional Cube. J.
                Comput. Appl. Math. 2 207--217.

          [2]   Genz A C and Malik A A (1980) An Adaptive Algorithm for
                Numerical Integration over an N-dimensional Rectangular
                Region. J. Comput. Appl. Math. 6 295--302.

          5. Parameters

           1:  NDIM -- INTEGER                                        Input
               On entry: the number of dimensions of the integral, n.
               Constraint: 2 <= NDIM <= 15.

           2:  A(NDIM) -- DOUBLE PRECISION array                      Input
               On entry: the lower limits of integration, a , for
                                                            i
               i=1,2,...,n.

           3:  B(NDIM) -- DOUBLE PRECISION array                      Input
               On entry: the upper limits of integration, b , for
                                                            i
               i=1,2,...,n.

           4:  MINPTS -- INTEGER                               Input/Output
               On entry: MINPTS must be set to the minimum number of
               integrand evaluations to be allowed. On exit: MINPTS
               contains the actual number of integrand evaluations used by
               D01FCF.

           5:  MAXPTS -- INTEGER                                      Input
               On entry: the maximum number of integrand evaluations to be
               allowed.
               Constraints:
                        MAXPTS >= MINPTS

                        MAXPTS >= (alpha),

                                       NDIM       2
                        where (alpha)=2    +2*NDIM +2*NDIM+1.

           6:  FUNCTN -- DOUBLE PRECISION FUNCTION, supplied by the
          user.
                                                         External Procedure
               FUNCTN must return the value of the integrand f at a given
               point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION FUNCTN (NDIM,Z)
                      INTEGER          NDIM
                      DOUBLE PRECISION Z(NDIM)

                1:  NDIM -- INTEGER                                   Input
                    On entry: the number of dimensions of the integral, n.

                2:  Z(NDIM) -- DOUBLE PRECISION array                 Input
                    On entry: the co-ordinates of the point at which the
                    integrand must be evaluated.
               FUNCTN must be declared as EXTERNAL in the (sub)program
               from which D01FCF is called. Parameters denoted as
               Input must not be changed by this procedure.

           7:  EPS -- DOUBLE PRECISION                                Input
               On entry: the relative error acceptable to the user. When
               the solution is zero or very small relative accuracy may not
               be achievable but the user may still set EPS to a reasonable
               value and check for the error exit IFAIL = 2. Constraint:
               EPS > 0.0.

           8:  ACC -- DOUBLE PRECISION                               Output
               On exit: the estimated relative error in FINVAL.

           9:  LENWRK -- INTEGER                                      Input
               On entry:
               the dimension of the array WRKSTR as declared in the
               (sub)program from which D01FCF is called.
               Suggested value: for maximum efficiency, LENWRK >=
               (NDIM+2)*(1+MAXPTS/(alpha)) (see parameter MAXPTS for
               (alpha)).

               If LENWRK is less than this, the routine will usually run
               less efficiently and may fail. Constraint: LENWRK>=2*NDIM+4.

          10:  WRKSTR(LENWRK) -- DOUBLE PRECISION array           Workspace

          11:  FINVAL -- DOUBLE PRECISION                            Output
               On exit: the best estimate obtained for the integral.

          12:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.
               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit. To suppress
               the output of an error message when soft failure occurs, set
               IFAIL to 1.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          IFAIL= 1
               On entry NDIM < 2,

               or       NDIM > 15,

               or       MAXPTS is too small,

               or       LENWRK<2*NDIM+4,

               or       EPS <= 0.0.

          IFAIL= 2
               MAXPTS was too small to obtain the required relative
               accuracy EPS. On soft failure, FINVAL and ACC contain
               estimates of the integral and the relative error, but ACC
               will be greater than EPS.

          IFAIL= 3
               LENWRK was too small. On soft failure, FINVAL and ACC
               contain estimates of the integral and the relative error,
               but ACC will be greater than EPS.

          7. Accuracy

          A relative error estimate is output through the parameter ACC.

          8. Further Comments

          Execution time will usually be dominated by the time taken to
          evaluate the integrand FUNCTN, and hence the maximum time that
          could be taken will be proportional to MAXPTS.

          9. Example

          This example program estimates the integral

                                  2
                  1  1  1  1  4z z exp(2z z )
                  /  /  /  /    1 3      1 3
                  |  |  |  |  ---------------dz dz dz dz =0.575364.
                  /  /  /  /             2     4  3  2  1
                  0  0  0  0    (1+z +z )
                                    2  4

          The accuracy requested is one part in 10,000.

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01gaf}{NAG On-line Documentation: d01gaf}
\beginscroll
\begin{verbatim}



     D01GAF(3NAG)      Foundation Library (12/10/92)      D01GAF(3NAG)



          D01 -- Quadrature                                          D01GAF
                  D01GAF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01GAF integrates a function which is specified numerically at
          four or more points, over the whole of its specified range, using
          third-order finite-difference formulae with error estimates,
          according to a method due to Gill and Miller.

          2. Specification

                 SUBROUTINE D01GAF (X, Y, N, ANS, ER, IFAIL)
                 INTEGER          N, IFAIL
                 DOUBLE PRECISION X(N), Y(N), ANS, ER

          3. Description

          This routine evaluates the definite integral

                                       x
                                        n
                                       /
                                    I= | y(x)dx,
                                       /
                                       x
                                        1

          where the function y is specified at the n-points x ,x ,...,x ,
                                                             1  2      n
          which should be all distinct, and in either ascending or
          descending order. The integral between successive points is
          calculated by a four-point finite-difference formula centred on
          the interval concerned, except in the case of the first and last
          intervals, where four-point forward and backward difference
          formulae respectively are employed. If n is less than 4, the
          routine fails. An approximation to the truncation error is
          integrated and added to the result. It is also returned
          separately to give an estimate of the uncertainty in the result.
          The method is due to Gill and Miller.

          4. References

          [1]   Gill P E and Miller G F (1972) An Algorithm for the
                Integration of Unequally Spaced Data. Comput. J. 15 80--83.

          5. Parameters

           1:  X(N) -- DOUBLE PRECISION array                         Input
               On entry: the values of the independent variable, i.e., the
               x ,x ,...,x . Constraint: either X(1) < X(2) <... < X(N) or
                1  2      n
               X(1) > X(2) >... > X(N).

           2:  Y(N) -- DOUBLE PRECISION array                         Input
               On entry: the values of the dependent variable y  at the
                                                                i
               points x , for i=1,2,...,n.
                       i

           3:  N -- INTEGER                                           Input
               On entry: the number of points, n. Constraint: N >= 4.

           4:  ANS -- DOUBLE PRECISION                               Output
               On exit: the estimate of the integral.

           5:  ER -- DOUBLE PRECISION                                Output
               On exit: an estimate of the uncertainty in ANS.

           6:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. For users not
               familiar with this parameter (described in the Essential
               Introduction) the recommended value is 0.

               On exit: IFAIL = 0 unless the routine detects an error (see
               Section 6).

          6. Error Indicators and Warnings

          Errors detected by the routine:

          IFAIL= 1
               Indicates that fewer than four-points have been supplied to
               the routine.

          IFAIL= 2
               Values of X are neither strictly increasing nor strictly
               decreasing.

          IFAIL= 3
               Two points have the same X-value.

          No error is reported arising from the relative magnitudes of ANS
          and ER on return, due to the difficulty when the true answer is
          zero.

          7. Accuracy

          No accuracy level is specified by the user before calling the
          routine but on return ABS(ER) is an approximation to, but not
          necessarily a bound for, |I-ANS|. If on exit IFAIL > 0, both ANS
          and ER are returned as zero.

          8. Further Comments

          The time taken by the routine depends on the number of points
          supplied, n.

          In their paper, Gill and Miller [1] do not add the quantity ER to
          ANS before return. However, extensive tests have shown that a
          dramatic reduction in the error often results from such addition.
          In other cases, it does not make an improvement, but these tend
          to be cases of low accuracy in which the modified answer is not
          significantly inferior to the unmodified one. The user has the
          option of recovering the Gill-Miller answer by subtracting ER
          from ANS on return from the routine.

          9. Example

          The example program evaluates the integral

                                    1
                                    /  4
                                    | ----dx=(pi)
                                    /    2
                                    0 1+x

          reading in the function values at 21 unequally-spaced points.

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd01gbf}{NAG On-line Documentation: d01gbf}
\beginscroll
\begin{verbatim}



     D01GBF(3NAG)      Foundation Library (12/10/92)      D01GBF(3NAG)



          D01 -- Quadrature                                          D01GBF
                  D01GBF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D01GBF returns an approximation to the integral of a function
          over a hyper-rectangular region, using a Monte Carlo method. An
          approximate relative error estimate is also returned. This
          routine is suitable for low accuracy work.

          2. Specification

                 SUBROUTINE D01GBF (NDIM, A, B, MINCLS, MAXCLS, FUNCTN,
                1                   EPS, ACC, LENWRK, WRKSTR, FINEST, IFAIL)
                 INTEGER          NDIM, MINCLS, MAXCLS, LENWRK, IFAIL
                 DOUBLE PRECISION A(NDIM), B(NDIM), FUNCTN, EPS, ACC, WRKSTR
                1                 (LENWRK), FINEST
                 EXTERNAL         FUNCTN

          3. Description

          D01GBF uses an adaptive Monte Carlo method based on the algorithm
          described by Lautrup [1]. It is implemented for integrals of the
          form:

                       b  b     b
                        1  2     n
                       /  /     /
                       |  | ... | f(x ,x ,...,x )dx ...dx dx .
                       /  /     /    1  2      n   n     2  1
                       a  a     a
                        1  2     n

          Upon entry, unless LENWRK has been set to the minimum value
          10*NDIM, the routine subdivides the integration region into a
          number of equal volume subregions. Inside each subregion the
          integral and the variance are estimated by means of pseudo-random
          sampling. All contributions are added together to produce an
          estimate for the whole integral and total variance. The variance
          along each co-ordinate axis is determined and the routine uses
          this information to increase the density and change the widths of
          the sub-intervals along each axis, so as to reduce the total
          variance. The total number of subregions is then increased by a
          factor of two and the program recycles for another iteration. The
          program stops when a desired accuracy has been reached or too
          many integral evaluations are needed for the next cycle.

          4. References

          [1]   Lautrup B (1971) An Adaptive Multi-dimensional Integration
                Procedure. Proc. 2nd Coll. on Advanced Methods in
                Theoretical Physics, Marseille.

          5. Parameters

           1:  NDIM -- INTEGER                                        Input
               On entry: the number of dimensions of the integral, n.
               Constraint: NDIM >= 1.

           2:  A(NDIM) -- DOUBLE PRECISION array                      Input
               On entry: the lower limits of integration, a , for
                                                           i
               i=1,2,...,n.

           3:  B(NDIM) -- DOUBLE PRECISION array                      Input
               On entry: the upper limits of integration, b , for
                                                           i
               i=1,2,...,n.

           4:  MINCLS -- INTEGER                               Input/Output
               On entry: MINCLS must be set:

               either to the minimum number of integrand evaluations to be
               allowed, in which case MINCLS >= 0;

               or to a negative value. In this case the routine assumes
               that a previous call had been made with the same parameters
               NDIM, A and B and with either the same integrand (in which
               case D01GBF continues calculation) or a similar integrand
               (in which case D01GBF begins the calculation with the
               subdivision used in the last iteration of the previous call)
               . See also WRKSTR. On exit: MINCLS contains the number of
               integrand evaluations actually used by D01GBF.

           5:  MAXCLS -- INTEGER                                      Input
               On entry: the maximum number of integrand evaluations to be
               allowed. In the continuation case this is the number of new
               integrand evaluations to be allowed. These counts do not
               include zero integrand values.
               Constraints:
                        MAXCLS > MINCLS,

                        MAXCLS >= 4*(NDIM+1).

           6:  FUNCTN -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                         External Procedure
               FUNCTN must return the value of the integrand f at a given
               point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION FUNCTN (NDIM, X)
                      INTEGER          NDIM
                      DOUBLE PRECISION X(NDIM)

                1:  NDIM -- INTEGER                                   Input
                    On entry: the number of dimensions of the integral, n.

                2:  X(NDIM) -- DOUBLE PRECISION array                 Input
                    On entry: the co-ordinates of the point at which the
                    integrand must be evaluated.
               FUNCTN must be declared as EXTERNAL in the (sub)program
               from which D01GBF is called. Parameters denoted as
               Input must not be changed by this procedure.

           7:  EPS -- DOUBLE PRECISION                                Input
               On entry: the relative accuracy required. Constraint: EPS
               >= 0.0.

           8:  ACC -- DOUBLE PRECISION                               Output
               On exit: the estimated relative accuracy of FINEST.

           9:  LENWRK -- INTEGER                                      Input
               On entry:
               the dimension of the array WRKSTR as declared in the
               (sub)program from which D01GBF is called.
               For maximum efficiency, LENWRK should be about
                                              1/NDIM
                             3*NDIM*(MAXCLS/4)      +7*NDIM.
               If LENWRK is given the value 10*NDIM then the subroutine
               uses only one iteration of a crude Monte Carlo method with
               MAXCLS sample points. Constraint: LENWRK >= 10*NDIM.

          10:  WRKSTR(LENWRK) -- DOUBLE PRECISION array        Input/Output
               On entry: if MINCLS<0, WRKSTR must be unchanged from the
               previous call of D01GBF - except that for a new integrand
               WRKSTR(LENWRK) must be set to 0.0. See also MINCLS. On
               exit: WRKSTR contains information about the current sub-
               interval structure which could be used in later calls of
               D01GBF. In particular, WRKSTR(j) gives the number of sub-
               intervals used along the jth co-ordinate axis.

          11:  FINEST -- DOUBLE PRECISION                            Output
               On exit: the best estimate obtained for the integral.

          12:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit. To suppress
               the output of an error message when soft failure occurs, set
               IFAIL to 1.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          IFAIL= 1
               On entry NDIM < 1,

               or       MINCLS >= MAXCLS,

               or       LENWRK < 10*NDIM,

               or       MAXCLS < 4*(NDIM+1),

               or       EPS < 0.0.

          IFAIL= 2
               MAXCLS was too small for D01GBF to obtain the required
               relative accuracy EPS. In this case D01GBF returns a value
               of FINEST with estimated relative error ACC, but ACC will be
               greater than EPS. This error exit may be taken before MAXCLS
               non-zero integrand evaluations have actually occurred, if
               the routine calculates that the current estimates could not
               be improved before MAXCLS was exceeded.

          7. Accuracy

          A relative error estimate is output through the parameter ACC.
          The confidence factor is set so that the actual error should be
          less than ACC 90% of the time. If a user desires a higher
          confidence level then a smaller value of EPS should be used.

          8. Further Comments

          The running time for D01GBF will usually be dominated by the time
          used to evaluate the integrand FUNCTN, so the maximum time that
          could be used is approximately proportional to MAXCLS.

          For some integrands, particularly those that are poorly behaved
          in a small part of the integration region, D01GBF may terminate
          with a value of ACC which is significantly smaller than the
          actual relative error. This should be suspected if the returned
          value of MINCLS is small relative to the expected difficulty of
          the integral. Where this occurs, D01GBF should be called again,
          but with a higher entry value of MINCLS (e.g. twice the returned
          value) and the results compared with those from the previous
          call.

          The exact values of FINEST and ACC on return will depend (within
          statistical limits) on the sequence of random numbers generated
          within D01GBF by calls to G05CAF. Separate runs will produce
          identical answers unless the part of the program executed prior
          to calling D01GBF also calls (directly or indirectly) routines
          from Chapter G05, and the series of such calls differs between
          runs. If desired, the user may ensure the identity or difference
          between runs of the results returned by D01GBF, by calling G05CBF
          or G05CCF respectively, immediately before calling D01GBF.

          9. Example

          This example program calculates the integral

                  1  1  1  1  4x x exp(2x x )
                  /  /  /  /    1 3      1 3
                  |  |  |  |  ---------------dx dx dx dx =0.575364.
                  /  /  /  /             2     1  2  3  4
                  0  0  0  0    (1+x +x )
                                    2  4

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02}{NAG On-line Documentation: d02}
\beginscroll
\begin{verbatim}



     D02(3NAG)         Foundation Library (12/10/92)         D02(3NAG)



          D02 -- Ordinary Differential Equations        Introduction -- D02
                                    Chapter D02
                          Ordinary Differential Equations

          1. Scope of the Chapter

          This chapter is concerned with the numerical solution of ordinary
          differential equations. There are two main types of problem,
          those in which all boundary conditions are specified at one point
          (initial-value problems), and those in which the boundary
          conditions are distributed between two or more points (boundary-
          value problems and eigenvalue problems). Routines are available
          for initial-value problems, two-point boundary-value problems and
          Sturm-Liouville eigenvalue problems.

          2. Background to the Problems

          For most of the routines in this chapter a system of ordinary
          differential equations must be written in the form

                              y'  = f (x,y ,y ,...,y ),
                                1    1    1  2      n

                              y'  = f (x,y ,y ,...,y ),
                                2    2    1  2      n

                                .. ...

                              y'  = f (x,y ,y ,...y ),
                                n    n    1  2     n

          that is the system must be given in first-order form. The n
          dependent variables (also, the solution) y ,y ,...,y  are
                                                    1  2      n
          functions of the independent variable x, and the differential
          equations give expressions for the first derivatives y' =dy /dx
                                                                 i   i
          in terms of x and y ,y ,...,y . For a system of n first-order
                             1  2      n
          equations, n associated boundary conditions are usually required
          to define the solution.

          A more general system may contain derivatives of higher order,
          but such systems can almost always be reduced to the first-order
          form by introducing new variables. For example, suppose we have
          the third-order equation


                                                2
                                z'''+zz''+k(l-z' )=0.

          We write y =z, y =z', y =z'', and the third order equation may
                    1     2      3
          then be written as the system of first-order equations

                                y'  = y
                                  1    2

                                y'  = y
                                  2    3
                                                 2
                                y'  = -y y -k(l-y ).
                                  3     1 3      2

          For this system n = 3 and we require 3 boundary conditions in
          order to define the solution. These conditions must specify
          values of the dependent variables at certain points. For example,
          we have an initial-value problem if the conditions are:

              y   =  0        at x=0
               1

              y   =  0        at x=0
               2

              y   =  0.1      at x=0.
               3

          These conditions would enable us to integrate the equations
          numerically from the point x=0 to some specified end-point. We
          have a boundary-value problem if the conditions are:

              y   =  0        at x=0
               1

              y   =  0        at x=0
               2

              y   =  1        at x=10.
               2

          These conditions would be sufficient to define a solution in the
          range 0<=x<=10, but the problem could not be solved by direct
          integration (see Section 2.2 ). More general boundary conditions
          are permitted in the boundary-value case.

          2.1. Initial-value Problems

          To solve first-order systems, initial values of the dependent
          variables y , for i=1,2,...,n must be supplied at a given point,
                     i
          a. Also a point, b, at which the values of the dependent
          variables are required, must be specified. The numerical solution
          is then obtained by a step-by-step calculation which approximates
          values of the variables y , for i=1,2,...,n at finite intervals
                                   i
          over the required range [a,b]. The routines in this chapter
          adjust the step length automatically to meet specified accuracy
          tolerances. Although the accuracy tests used are reliable over
          each step individually, in general an accuracy requirement cannot
          be guaranteed over a long range. For many problems there may be
          no serious accumulation of error, but for unstable systems small
          perturbations of the solution will often lead to rapid divergence
          of the calculated values from the true values. A simple check for
          stability is to carry out trial calculations with different
          tolerances; if the results differ appreciably the system is
          probably unstable. Over a short range, the difficulty may
          possibly be overcome by taking sufficiently small tolerances, but
          over a long range it may be better to try to reformulate the
          problem.

          A special class of initial-value problems are those for which the
          solutions contain rapidly decaying transient terms. Such problems
          are called stiff; an alternative way of describing them is to say
          that certain eigenvalues of the Jacobian matrix (ddf /ddy ) have
                                                              i    j
          large negative real parts when compared to others. These problems
          require special methods for efficient numerical solution; the
          methods designed for non-stiff problems when applied to stiff
          problems tend to be very slow, because they need small step
          lengths to avoid numerical instability. A full discussion is
          given in Hall and Watt [6] and a discussion of the methods for
          stiff problems is given in Berzins, Brankin and Gladwell [1].

          2.2. Boundary-value Problems

          A full discussion of the design of the methods and codes for
          boundary-value problems is given in Gladwell [4]. In general, a
          system of nonlinear differential equations with boundary
          conditions given at two or more points cannot be guaranteed to
          have a solution. The solution has to be determined iteratively
          (if it exists). Finite-difference equations are set up on a mesh
          of points and estimated values for the solution at the grid
          points are chosen. Using these estimated values as starting
          values a Newton iteration is used to solve the finite-difference
          equations. The accuracy of the solution is then improved by
          deferred corrections or the addition of points to the mesh or a
          combination of both. Good initial estimates of the solution may
          be required in some cases but results may be difficult to compute
          when the solution varies very rapidly over short ranges. A
          discussion is given in Chapters 9 and 11 of Gladwell and Sayers
          [5] and Chapter 4 of Childs et al [2].

          2.3. Eigenvalue Problems

          Sturm-Liouville problems of the form

                             (p(x)y')'+q(x,(lambda))y=0

          with appropriate boundary conditions given at two points, can be
          solved by a Scaled Pruefer method. In this method the
          differential equation is transformed to another which can be
          solved for a specified eigenvalue by a shooting method. A
          discussion is given in Chapter 11 of Gladwell and Sayers [5] and
          a complete description is given in Pryce [7].

          2.6. References

          [1]   Berzins M, Brankin R W and Gladwell I (1987) Design of the
                Stiff Integrators in the NAG Library. Technical Report.
                TR14/87 NAG.

          [2]   (1979) Codes for Boundary-value Problems in Ordinary
                Differential Equations. Lecture Notes in Computer Science.
                (ed Childs B, Scott M, Daniel J W, Denman E and Nelson P) 76
                 Springer-Verlag.

          [3]   Gladwell I (1979) Initial Value Routines in the NAG Library.
                ACM Trans Math Softw. 5 386--400.

          [4]   Gladwell I (1987) The NAG Library Boundary Value Codes.
                Numerical Analysis Report. 134 Manchester University.

          [5]   (1980) Computational Techniques for Ordinary Differential
                Equations. (Gladwell I and Sayers D K) Academic Press.

          [6]   Hall G and Watt J M (eds) (1976) Modern Numerical Methods
                for Ordinary Differential Equations. Clarendon Press.

          [7]   Pryce J D (1986) Error Estimation for Phase-function
                Shooting Methods for Sturm-Liouville Problems. J. Num. Anal.
                6 103--123.

          3. Recommendations on Choice and Use of Routines

          There are no routines which deal directly with COMPLEX equations.
          These may however be transformed to larger systems of real
          equations of the required form. Split each equation into its real
          and imaginary parts and solve for the real and imaginary parts of
          each component of the solution. Whilst this process doubles the
          size of the system and may not always be appropriate it does make
          available for use the full range of routines provided presently.

          3.1. Initial-value Problems

          For simple first-order problems with low accuracy requirements,
          that is problems on a short range of integration, with derivative
          functions f  which are inexpensive to calculate and where only a
                     i
          few correct figures are required, the best routines to use are
          likely to be the Runge-Kutta-Merson (RK) routines, D02BBF and
          D02BHF. For larger problems, over long ranges or with high
          accuracy requirements the variable-order, variable-step Adams
          routine D02CJF should usually be preferred. For stiff equations,
          that is those with rapidly decaying transient solutions, the
          Backward Differentiation Formula (BDF) variable-order, variable-
          step routine D02EJF should be used.

          There are four routines for initial-value problems, two of which
          use the Runge-Kutta-Merson method:

          D02BBF     integrates a system of first order ordinary
                     differential equations over a range with intermediate
                     output and a choice of error control

          D02BHF     integrates a system of first order ordinary
                     differential equations with a choice of error control
                     until a position is determined where a function of the
                     solution is zero.

          one uses an Adams method:

          D02CJF     combines the functionality of D02BBF and D02BHF

          and one uses a BDF method:

          D02EJF     combines the functionality of D02BBF and D02BHF.

          3.2. Boundary-value Problems

          D02GAF may be used for simple boundary-value problems with
          assigned boundary values. The user may find that convergence is
          difficult to achieve using D02GAF since only specifying the
          unknown boundary values and the position of the finite-difference
          mesh is permitted. In such cases the user may use D02RAF which
          permits specification of an initial estimate for the solution at
          all mesh points and allows the calculation to be influenced in
          other ways too. D02RAF is designed to solve a general nonlinear
          two-point boundary value problem with nonlinear boundary
          conditions.

          A routine, D02GBF, is also supplied specifically for the general
          linear two-point boundary-value problem written in a standard

          The user is advised to use interpolation routines from the E01
          Chapter to obtain solution values at points not on the final
          mesh.

          3.3. Eigenvalue Problems

          There is one general purpose routine for eigenvalue problems,
          D02KEF. It may be used to solve regular or singular second-order
          Sturm-Liouville problems on a finite or infinite range.
          Discontinous coefficient functions can be treated and
          eigenfunctions can be computed.


          D02 -- Ordinary Differential Equations            Contents -- D02
          Chapter D02

          Ordinary Differential Equations

          D02BBF  ODEs, IVP, Runge-Kutta-Merson method, over a range,
                  intermediate output

          D02BHF  ODEs, IVP, Runge-Kutta-Merson method, until function of
                  solution is zero

          D02CJF  ODEs, IVP, Adams method, until function of solution is
                  zero, intermediate output

          D02EJF  ODEs, stiff IVP, BDF method, until function of solution
                  is zero, intermediate output

          D02GAF  ODEs, boundary value problem, finite difference technique
                  with deferred correction, simple nonlinear problem

          D02GBF  ODEs, boundary value problem, finite difference technique
                  with deferred correction, general linear problem

          D02KEF  2nd order Sturm-Liouville problem, regular/singular
                  system, finite/infinite range, eigenvalue and
                  eigenfunction, user-specified break-points

          D02RAF  ODEs, general nonlinear boundary value problem, finite
                  difference technique with deferred correction,
                  continuation facility

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02bbf}{NAG On-line Documentation: d02bbf}
\beginscroll
\begin{verbatim}



     D02BBF(3NAG)                 D02BBF                  D02BBF(3NAG)



     D02 -- Ordinary Differential Equations                     D02BBF
             D02BBF -- NAG Foundation Library Routine Document

     Note: Before using this routine, please read the Users' Note for
     your implementation to check implementation-dependent details.
     The symbol (*) after a NAG routine name denotes a routine that is
     not included in the Foundation Library.

     Note for users via the AXIOM system: the interface to this routine
     has been enhanced for use with AXIOM and is slightly different to
     that offered in the standard version of the Foundation Library.

     1. Purpose

     D02BBF integrates a system of first-order ordinary differential
     equations over an interval with suitable initial conditions,
     using a Runge-Kutta-Merson method, and returns the solution at
     points specified by the user.

     2. Specification

            SUBROUTINE D02BBF (X, XEND, M, N, Y, TOL, IRELAB, RESULT,
           1                   FCN, OUTPUT, W, IFAIL)
            INTEGER          M, N, IRELAB, IFAIL
            DOUBLE PRECISION X, XEND, Y(N), TOL, W(N,7), RESULT(M,N)
            EXTERNAL         FCN, OUTPUT

     3. Description

     The routine integrates a system of ordinary differential
     equations

                    y' =f (x,y ,y ,...,y )  i=1,2,...,n
                      i  i    1  2      n

     from x = X to x = XEND using a Merson form of the Runge-Kutta
     method. The system is defined by a subroutine FCN supplied by the
     user, which evaluates f  in terms of x and y ,y ,...,y , and the
                            i                    1  2      n
     values of y ,y ,...,y  must be given at x = X.
                1  2      n

     The solution is returned via the user-supplied routine OUTPUT at
     a set of points specified by the user. This solution is obtained
     by quintic Hermite interpolation on solution values produced by
     the Runge-Kutta method.

     The accuracy of the integration and, indirectly, the
     interpolation is controlled by the parameters TOL and IRELAB.

     For a description of Runge-Kutta methods and their practical
     implementation see Hall and Watt [1].

     4. References

     [1]   Hall G and Watt J M (eds) (1976) Modern Numerical Methods
           for Ordinary Differential Equations. Clarendon Press.

     5. Parameters

      1:  X -- DOUBLE PRECISION                           Input/Output
          On entry: X must be set to the initial value of the
          independent variable x. On exit: XEND, unless an error has
          occurred, when it contains the value of x at the error.

      2:  XEND -- DOUBLE PRECISION                               Input
          On entry: the final value of the independent variable. If
          XEND < X on entry, integration will proceed in a negative
          direction.

      3:  M -- INTEGER                                           Input
          On entry: the first dimension of the array RESULT.  This
          will usually be equal to the number of points at which the
          solution is required.
          Constraint: M > 0.

      4:  N -- INTEGER                                           Input
          On entry: the number of differential equations.
          Constraint: N > 0.

      5:  Y(N) -- DOUBLE PRECISION array                  Input/Output
          On entry: the initial values of the solution y ,y ,...,y .
                                                         1  2      n
          On exit: the computed values of the solution at the final
          value of X.

      6:  TOL -- DOUBLE PRECISION                         Input/Output
          On entry: TOL must be set to a positive tolerance for
          controlling the error in the integration.

          D02BBF has been designed so that, for most problems, a
          reduction in TOL leads to an approximately proportional
          reduction in the error in the solution at XEND. The relation
          between changes in TOL and the error at intermediate output
          points is less clear, but for TOL small enough the error at
          intermediate output points should also be approximately
          proportional to TOL. However, the actual relation between
          TOL and the accuracy achieved cannot be guaranteed. The user
          is strongly recommended to call D02BBF with more than one
          value for TOL and to compare the results obtained to
          estimate their accuracy. In the absence of any prior
          knowledge, the user might compare the results obtained by
                                      -p             -p-1
          calling D02BBF with TOL=10.0   and TOL=10.0     if p correct
          decimal digits in the solution are required. Constraint: TOL
          > 0.0. On exit: normally unchanged. However if the range X
          to XEND is so short that a small change in TOL is unlikely
          to make any change in the computed solution then, on return,
          TOL has its sign changed. This should be treated as a
          warning that the computed solution is likely to be more
          accurate than would be produced by using the same value of
          TOL on a longer range of integration.

      7:  IRELAB -- INTEGER                                      Input
          On entry: IRELAB determines the type of error control. At
          each step in the numerical solution an estimate of the local
          error, EST, is made. For the current step to be accepted the
          following condition must be satisfied:
          IRELAB = 0
                EST=10.0<=TOL*max{1.0,|y |,|y |,...,|y |};
                                        1    2        n

          IRELAB = 1
                EST <= TOL;

          IRELAB = 2
                EST<=TOL*max{(epsilon),|y |,|y |,...,|y |}, where
                                         1    2        n
                (epsilon) is machine precision.
          If the appropriate condition is not satisfied, the step size
          is reduced and the solution is recomputed on the current
          step.

          If the user wishes to measure the error in the computed
          solution in terms of the number of correct decimal places,
          then IRELAB should be given the value 1 on entry, whereas if
          the error requirement is in terms of the number of correct
          significant digits, then IRELAB should be given the value 2.
          Where there is no preference in the choice of error test
          IRELAB = 0 will result in a mixed error test. Constraint: 0
          <= IRELAB <= 2.

      8:  RESULT(M,N) -- DOUBLE PRECISION array                 Output
          On exit: the computed values of the solution at the points
          given by OUTPUT.

      9:  FCN -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          FCN must evaluate the functions f  (i.e., the derivatives
                                           i
          y' ) for given values of its arguments x,y ,...,y .
            i                                       1      n

          Its specification is:

                 SUBROUTINE FCN (X, Y, F)
                 DOUBLE PRECISION X, Y(n), F(n)

          where n is the actual value of N in the call of D02BBF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the argument x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the argument y , for
                                                     i
               i=1,2,...,n.

           3:  F(*) -- DOUBLE PRECISION array                   Output
               On exit: the value of f , for i=1,2,...,n.
                                       i
          FCN must be declared as EXTERNAL in the (sub)program
          from which D02BBF is called. Parameters denoted as
          Input must not be changed by this procedure.

     10:  OUTPUT -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          OUTPUT allows the user to have access to intermediate values
          of the computed solution at successive points specified by the
          user. These solution values may be returned to the user via
          the array RESULT if desired (this is a non-standard feature
          added for use with the AXIOM system).  OUTPUT is initially
          called by D02BBF with XSOL = X (the initial value of x). The
          user must reset XSOL to the next point where OUTPUT is to be
          called, and so on at each call to OUTPUT. If, after a call
          to OUTPUT, the reset point XSOL is beyond XEND, D02BBF will
          integrate to XEND with no further calls to OUTPUT; if a call
          to OUTPUT is required at the point XSOL = XEND, then XSOL
          must be given precisely the value XEND.

          Its specification is:

               SUBROUTINE OUTPUT(XSOL,Y,COUNT,M,N,RESULT)
               DOUBLE PRECISION Y(N),RESULT(M,N),XSOL
               INTEGER M,N,COUNT

           1:  XSOL -- DOUBLE PRECISION                   Input/Output
               On entry: the current value of the independent
               variable x. On exit: the next value of x at which
               OUTPUT is to be called.

           2:  Y(N) -- DOUBLE PRECISION array                    Input
               On entry: the computed solution at the point XSOL.

           3:  COUNT -- INTEGER                           Input/Output
               On entry: Zero if OUTPUT has not been called before, or
               the previous value of COUNT.
               On exit: A new value of COUNT: this can be used to keep
               track of the number of times OUTPUT has been called.

           4:  M -- INTEGER                                      Input
               On entry: The first dimension of RESULT.

           5:  N -- INTEGER                                      Input
               On entry: The dimension of Y.

           6:  RESULT(M,N) -- DOUBLE PRECISION array      Input/Output
               On entry: the previous contents of RESULT.
               On exit: RESULT may be used to return the values of the
               intermediate solutions to the user.

          OUTPUT must be declared as EXTERNAL in the (sub)program
          from which D02BBF is called. Parameters denoted as
          Input must not be changed by this procedure.

     11:  W(N,7) -- DOUBLE PRECISION array                   Workspace

     12:  IFAIL -- INTEGER                                Input/Output
          On entry: IFAIL must be set to 0, -1 or 1. For users not
          familiar with this parameter (described in the Essential
          Introduction) the recommended value is 0.

          On exit: IFAIL = 0 unless the routine detects an error (see
          Section 6).

     6. Error Indicators and Warnings

     Errors detected by the routine:

     IFAIL= 1
          On entry TOL <= 0,

          or       N <= 0,

          or       IRELAB /= 0, 1 or 2.

     IFAIL= 2
          With the given value of TOL, no further progress can be made
          across the integration range from the current point x = X,
          or the dependence of the error on TOL would be lost if
          further progress across the integration range were attempted
          (see Section 8 for a discussion of this error exit). The
          components Y(1),Y(2),...,Y(n) contain the computed values of
          the solution at the current point x = X.

     IFAIL= 3
          TOL is too small for the routine to take an initial step
          (see Section 8). X and Y(1),Y(2),...,Y(n) retain their
          initial values.

     IFAIL= 4
          X = XEND and XSOL /= X after the initial call to OUTPUT.

     IFAIL= 5
          A value of XSOL returned by OUTPUT lies behind the previous
          value of XSOL in the direction of integration.

     IFAIL= 6
          A serious error has occurred in an internal call to
          D02PAF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

     IFAIL= 7
          A serious error has occurred in an internal call to
          D02XAF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

     7. Accuracy

     The accuracy depends on TOL, on the mathematical properties of
     the differential system, on the length of the range of
     integration and on the method. It can be controlled by varying
     TOL but the approximate proportionality of the error to TOL holds
     only for a restricted range of values of TOL. For TOL too large,
     the underlying theory may break down and the result of varying
     TOL may be unpredictable. For TOL too small, rounding errors may
     affect the solution significantly and an error exit with IFAIL =
     2 or IFAIL = 3 is possible.

     At the intermediate output points the same remarks apply. For
     large values of TOL it is possible that the errors at some
     intermediate output points may be much larger than at XEND. In
     any case, it must not be expected that the error will have the
     same size at all output points. At any point, it is a combination
     of the errors arising from the integration of the differential
     equation and the interpolation. The effect of combining these
     errors will vary, though in most cases the integration error will
     dominate.

     The user who requires a more reliable estimate of the accuracy
     achieved than can be obtained by varying TOL, is recommended to
     call D02BDF(*) where both the solution and a global error
     estimate are computed.

     8. Further Comments

     The time taken by the routine depends on the complexity and
     mathematical properties of the system of differential equations
     defined by FCN, on the range, the tolerance and the number of
     calls to OUTPUT. There is also an overhead of the form a+b*n
     where a and b are machine-dependent computing times.

     If the routine fails with IFAIL = 3, then it can be called again
     with a larger value of TOL (if this has not already been tried).
     If the accuracy requested is really needed and cannot be obtained
     with this routine, the system may be very stiff (see below) or so
     badly scaled that it cannot be solved to the required accuracy.

     If the routine fails with IFAIL = 2, it is probable that it has
     been called with a value of TOL which is so small that the
     solution cannot be obtained on the range X to XEND. This can
     happen for well-behaved systems and very small values of TOL. The
     user should, however, consider whether there is a more
     fundamental difficulty. For example:

     (a)   in the region of a singularity (infinite value) of the
           solution, the routine will usually stop with IFAIL = 2,
           unless overflow occurs first. If overflow occurs using
           D02BBF, D02PAF(*) can be used instead to trap the
           increasing solution before overflow occurs. In any case,
           numerical integration cannot be continued through a
           singularity, and analytic treatment should be considered;

     (b)   for 'stiff' equations, where the solution contains rapidly
           decaying components, the routine will use very small steps
           in x (internally to D02BBF) to preserve stability. This
           will usually exhibit itself by making the computing time
           excessively long, or occasionally by an exit with IFAIL =
           2. Merson's method is not efficient in such cases, and the
           user should try using D02EBF(*) (Backward Differentiation
           Formula). To determine whether a problem is stiff,
           D02BDF(*) may be used.

     For well-behaved systems with no difficulties such as stiffness
     or singularities, the Merson method should work well for low
     accuracy calculations (three or four figures). For higher
     accuracy calculations or where FCN is costly to evaluate,
     Merson's method may not be appropriate and a computationally less
     expensive method may be D02CBF(*) which uses an Adams method.

     Users with problems for which D02BBF is not sufficiently general
     should consider using D02PAF(*) with D02XAF(*). D02PAF(*) is a
     more general Merson routine with many facilities including more
     general error control options and several criteria for
     interrupting the calculations. D02XAF(*) interpolates on values
     produced by D02PAF(*).

     9. Example

     To integrate the following equations (for a projectile)

                                y'=tan(phi)

                           -0.032tan(phi)   0.02v
                       v'= --------------- --------
                                 v         cos(phi)


                                      -0.032
                              (phi)'= ------
                                         2
                                        v

     over an interval X = 0.0 to XEND = 8.0, starting with values
     y=0.0, v=0.5 and (phi)=(pi)/5 and printing the solution at steps
     of 1.0. We write y=Y(1), v=Y(2) and (phi)=Y(3), and we set
     TOL=1.0E-4 and TOL=1.0E-5 in turn so that we may compare the
     solutions. The value of (pi) is obtained by using X01AAF(*).

     Note the careful construction of routine OUT to ensure that the
     value of XEND is printed.

     The example program is not reproduced here. The source code for
     all example programs is distributed with the NAG Foundation
     Library software and should be available on-line.
\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02bhf}{NAG On-line Documentation: d02bhf}
\beginscroll
\begin{verbatim}



     D02BHF(3NAG)      Foundation Library (12/10/92)      D02BHF(3NAG)



          D02 -- Ordinary Differential Equations                     D02BHF
                  D02BHF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D02BHF integrates a system of first-order ordinary differential
          equations over an interval with suitable initial conditions,
          using a Runge-Kutta-Merson method, until a user-specified
          function of the solution is zero.

          2. Specification

                 SUBROUTINE D02BHF (X, XEND, N, Y, TOL, IRELAB, HMAX, FCN,
                1                   G, W, IFAIL)
                 INTEGER          N, IRELAB, IFAIL
                 DOUBLE PRECISION X, XEND, Y(N), TOL, HMAX, G, W(N,7)
                 EXTERNAL         FCN, G

          3. Description

          The routine advances the solution of a system of ordinary
          differential equations

                        y' =f (x,y ,y ,...,y ),  i=1,2,...,n,
                          i  i    1  2      n

          from x = X towards x = XEND using a Merson form of the Runge-
          Kutta method. The system is defined by a subroutine FCN supplied
          by the user, which evaluates f  in terms of x and y ,y ,...,y
                                        i                    1  2      n
          (see Section 5), and the values of y ,y ,...,y  must be given at
                                              1  2      n
          x = X.

          As the integration proceeds, a check is made on the function
          g(x,y) specified by the user, to determine an interval where it
          changes sign. The position of this sign change is then determined
          accurately by interpolating for the solution and its derivative.
          It is assumed that g(x,y) is a continuous function of the
          variables, so that a solution of g(x,y) = 0 can be determined by
          searching for a change in sign in g(x,y).

          The accuracy of the integration and, indirectly, of the
          determination of the position where g(x,y) = 0, is controlled by
          the parameter TOL.

          For a description of Runge-Kutta methods and their practical
          implementation see Hall and Watt [1].

          4. References

          [1]   Hall G and Watt J M (eds) (1976) Modern Numerical Methods
                for Ordinary Differential Equations. Clarendon Press.

          5. Parameters

           1:  X -- DOUBLE PRECISION                           Input/Output
               On entry: X must be set to the initial value of the
               independent variable x. On exit: the point where g(x,y) = 0.
               0 unless an error has occurred, when it contains the value
               of x at the error. In particular, if g(x,y)/=0.0 anywhere on
               the range X to XEND, it will contain XEND on exit.

           2:  XEND -- DOUBLE PRECISION                               Input
               On entry: the final value of the independent variable x.

               If XEND < X on entry, integration proceeds in a negative
               direction.

           3:  N -- INTEGER                                           Input
               On entry: the number of differential equations, n.
               Constraint: N > 0.

           4:  Y(N) -- DOUBLE PRECISION array                  Input/Output
               On entry: the initial values of the solution y ,y ,...,y .
                                                             1  2      n
               On exit: the computed values of the solution at the final
               point x = X.

           5:  TOL -- DOUBLE PRECISION                         Input/Output
               On entry: TOL must be set to a positive tolerance for
               controlling the error in the integration and in the
               determination of the position where g(x,y) = 0.0.

               D02BHF has been designed so that, for most problems, a
               reduction in TOL leads to an approximately proportional
               reduction in the error in the solution obtained in the
               integration. The relation between changes in TOL and the
               error in the determination of the position where g(x,y) = 0.
               0 is less clear, but for TOL small enough the error should
               be approximately proportional to TOL. However, the actual
               relation between TOL and the accuracy cannot be guaranteed.
               The user is strongly recommended to call D02BHF with more
               than one value for TOL and to compare the results obtained
               to estimate their accuracy. In the absence of any prior
               knowledge the user might compare results obtained by calling
                                   -p             -p-1
               D02BHF with TOL=10.0   and TOL=10.0     if p correct decimal
               digits in the solution are required. Constraint: TOL > 0.0.

               On exit: normally unchanged. However if the range from x = X
               to the position where g(x,y) = 0.0 (or to the final value of
               x if an error occurs) is so short that a small change in TOL
               is unlikely to make any change in the computed solution,
               then TOL is returned with its sign changed. To check results
               returned with TOL < 0.0, D02BHF should be called again with
               a positive value of TOL whose magnitude is considerably
               smaller than that of the previous call.

           6:  IRELAB -- INTEGER                                      Input
               On entry: IRELAB determines the type of error control. At
               each step in the numerical solution an estimate of the local
               error, EST, is made. For the current step to be accepted the
               following condition must be satisfied:
               IRELAB = 0
                     EST<=TOL*max{1.0,|y |,|y |,...,|y |};
                                        1    2        n

               IRELAB = 1
                     EST <= TOL;

               IRELAB = 2
                     EST<=TOL*max{(epsilon),|y |,|y |,...,|y |},
                                              1    2        n

                     where (epsilon) is machine precision.
               If the appropriate condition is not satisfied, the step size
               is reduced and the solution recomputed on the current step.

               If the user wishes to measure the error in the computed
               solution in terms of the number of correct decimal places,
               then IRELAB should be given the value 1 on entry, whereas if
               the error requirement is in terms of the number of correct
               significant digits, then IRELAB should be given the value 2.
               Where there is no preference in the choice of error test,
               IRELAB = 0 will result in a mixed error test. It should be
               borne in mind that the computed solution will be used in
               evaluating g(x,y). Constraint: 0 <= IRELAB <= 2.

           7:  HMAX -- DOUBLE PRECISION                               Input
               On entry: if HMAX = 0.0, no special action is taken.

               If HMAX /= 0.0, a check is made for a change in sign of
               g(x,y) at steps not greater than |HMAX|. This facility
               should be used if there is any chance of 'missing' the
               change in sign by checking too infrequently. For example, if
               two changes of sign of g(x,y) are expected within a distance
               h, say, of each other, then a suitable value for HMAX might
               be HMAX = h/2. If only one change of sign in g(x,y) is
               expected on the range X to XEND, then the choice HMAX = 0.0
               is most appropriate.

           8:  FCN -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               FCN must evaluate the functions f  (i.e., the derivatives
                                                i
               y' ) for given values of its arguments x,y ,...,y .
                 i                                       1      n

               Its specification is:

                      SUBROUTINE FCN (X, Y, F)
                      DOUBLE PRECISION X, Y(n), F(n)

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the argument x.

                2:  Y(*) -- DOUBLE PRECISION array                    Input
                    On entry: the value of the argument y , for
                                                         i
                    i=1,2,...,n.

                3:  F(*) -- DOUBLE PRECISION array                   Output
                    On exit: the value of f , for i=1,2,...,n.
                                           i
               FCN must be declared as EXTERNAL in the (sub)program
               from which D02BHF is called. Parameters denoted as
               Input must not be changed by this procedure.

           9:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                    External Procedure
               G must evaluate the function g(x,y) at a specified point.

               Its specification is:

                      DOUBLE PRECISION FUNCTION G (X, Y)
                      DOUBLE PRECISION X, Y(n)
               where n is the actual value of N in the call of D02BHF.

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the independent variable x.

                2:  Y(*) -- DOUBLE PRECISION array                    Input
                    On entry: the value of y , for i=1,2,...,n.
                                             i
               G must be declared as EXTERNAL in the (sub)program from
               which D02BHF is called. Parameters denoted as Input
               must not be changed by this procedure.

          10:  W(N,7) -- DOUBLE PRECISION array                   Workspace

          11:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. For users not
               familiar with this parameter (described in the Essential
               Introduction) the recommended value is 0.

               On exit: IFAIL = 0 unless the routine detects an error (see
               Section 6).

          6. Error Indicators and Warnings

          Errors detected by the routine:

          IFAIL= 1
               On entry TOL <= 0.0,

               or       N <= 0,

               or       IRELAB /= 0, 1 or 2.

          IFAIL= 2
               With the given value of TOL, no further progress can be made
               across the integration range from the current point x = X,
               or dependence of the error on TOL would be lost if further
               progress across the integration range were attempted (see
               Section 8 for a discussion of this error exit). The
               components Y(1),Y(2),...,Y(n) contain the computed values of
               the solution at the current point x = X. No point at which
               g(x,y) changes sign has been located up to the point x = X.

          IFAIL= 3
               TOL is too small for the routine to take an initial step
               (see Section 8). X and Y(1),Y(2),...,Y(n) retain their
               initial values.

          IFAIL= 4
               At no point in the range X to XEND did the function g(x,y)
               change sign. It is assumed that g(x,y) = 0.0 has no
               solution.

          IFAIL= 5
               A serious error has occurred in an internal call to
               C05AZF(*). Check all subroutine calls and array dimensions.
               Seek expert help.

          IFAIL= 6
               A serious error has occurred in an internal call to
               D02PAF(*). Check all subroutine calls and array dimensions.
               Seek expert help.

          IFAIL= 7
               A serious error has occurred in an internal call to
               D02XAF(*). Check all subroutine calls and array dimensions.
               Seek expert help.

          7. Accuracy

          The accuracy depends on TOL, on the mathematical properties of
          the differential system, on the position where g(x,y) = 0.0 and
          on the method. It can be controlled by varying TOL but the
          approximate proportionality of the error to TOL holds only for a
          restricted range of values of TOL. For TOL too large, the
          underlying theory may break down and the result of varying TOL
          may be unpredictable. For TOL too small, rounding error may
          affect the solution significantly and an error exit with IFAIL =
          2 or IFAIL = 3 is possible.

          The accuracy may also be restricted by the properties of g(x,y).
          The user should try to code G without introducing any unnecessary
          cancellation errors.

          8. Further Comments

          The time taken by the routine depends on the complexity and
          mathematical properties of the system of differential equations
          defined by FCN, the complexity of G, on the range, the position
          of the solution and the tolerance. There is also an overhead of
          the form a+b*n where a and b are machine-dependent computing
          times.

          For some problems it is possible that D02BHF will return IFAIL =
          4 because of inaccuracy of the computed values Y, leading to
          inaccuracy in the computed values of g(x,y) used in the search
          for the solution of g(x,y) = 0.0. This difficulty can be overcome
          by reducing TOL sufficiently, and if necessary, by choosing HMAX
          sufficiently small. If possible, the user should choose XEND well
          beyond the expected point where g(x,y) = 0.0; for example make
          |XEND-X| about 50 larger than the expected range. As a simple
          check, if, with XEND fixed, a change in TOL does not lead to a
          significant change in Y at XEND, then inaccuracy is not a likely
          source of error.

          If the routine fails with IFAIL = 3, then it could be called
          again with a larger value of TOL if this has not already been
          tried. If the accuracy requested is really needed and cannot be
          obtained with this routine, the system may be very stiff (see
          below) or so badly scaled that it cannot be solved to the
          required accuracy.

          If the routine fails with IFAIL = 2, it is likely that it has
          been called with a value of TOL which is so small that a solution
          cannot be obtained on the range X to XEND. This can happen for
          well-behaved systems and very small values of TOL. The user
          should, however, consider whether there is a more fundamental
          difficulty. For example:

          (a)   in the region of a singularity (infinite value) of the
                solution, the routine will usually stop with IFAIL = 2,
                unless overflow occurs first. If overflow occurs using
                D02BHF, D02PAF(*) can be used instead to trap the
                increasing solution, before overflow occurs. In any case,
                numerical integration cannot be continued through a
                singularity, and analytical treatment should be considered;

          (b)   for 'stiff' equations, where the solution contains rapidly
                decaying components, the routine will compute in very small
                steps in x (internally to D02BHF) to preserve stability.
                This will usually exhibit itself by making the computing
                time excessively long, or occasionally by an exit with
                IFAIL = 2. Merson's method is not efficient in such cases,
                and the user should try D02EHF(*) which uses a Backward
                Differentiation Formula method. To determine whether a
                problem is stiff, D02BDF(*) may be used.

          For well-behaved systems with no difficulties such as stiffness
          or singularities, the Merson method should work well for low
          accuracy calculations (three or four figures). For high accuracy
          calculations or where FCN is costly to evaluate, Merson's method
          may not be appropriate and a computationally less expensive
          method may be D02CHF(*) which uses an Adams method.

          For problems for which D02BHF is not sufficiently general, the
          user should consider D02PAF(*). D02PAF(*) is a more general
          Merson routine with many facilities including more general error
          control options and several criteria for interrupting the
          calculations. D02PAF(*) can be combined with the rootfinder
          C05AZF(*) and the interpolation routine D02XAF(*) to solve
          equations involving y ,y ,...,y  and their derivatives.
                               1  2      n

          D02BHF can also be used to solve an equation involving x,
          y ,y ,...,y  and the derivatives of y ,y ,...,y . For example in
           1  2      n                         1  2      n
          Section 9, D02BHF is used to find a value of X > 0.0 where Y(1) =
          0.0. It could instead be used to find a turning-point of y  by
                                                                    1
          replacing the function g(x,y) in the program by:

                DOUBLE PRECISION FUNCTION G(X,Y)
                DOUBLE PRECISION X,Y(3),F(3)
                CALL FCN(X,Y,F)
                G = F(1)
                RETURN
                END

          This routine is only intended to locate the first zero of g(x,y).
          If later zeros are required, users are strongly advised to
          construct their own more general root finding routines as
          discussed above.

          9. Example

          To find the value X > 0.0 at which y=0.0, where y, v, (phi) are
          defined by

                                     y'=tan(phi)

                                -0.032tan(phi)   0.02v
                            v'= --------------- --------
                                      v         cos(phi)

                                            0.032
                                   (phi)'=- -----
                                              2
                                             v

          and where at X = 0.0 we are given y=0.5, v=0.5 and (phi)=(pi)/5.
          We write y=Y(1), v=Y(2) and (phi)=Y(3) and we set TOL=1.0E-4 and
          TOL=1.0E-5 in turn so that we can compare the solutions. We
          expect the solution X ~= 7.3 and so we set XEND = 10.0 to avoid
          determining the solution of y=0.0 too near the end of the range
          of integration. The value of (pi) is obtained by using X01AAF(*).


          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02cjf}{NAG On-line Documentation: d02cjf}
\beginscroll
\begin{verbatim}



     D02CJF(3NAG)                 D02CJF                  D02CJF(3NAG)



     D02 -- Ordinary Differential Equations                     D02CJF
             D02CJF -- NAG Foundation Library Routine Document

     Note: Before using this routine, please read the Users' Note for
     your implementation to check implementation-dependent details.
     The symbol (*) after a NAG routine name denotes a routine that is
     not included in the Foundation Library.

     Note for users via the AXIOM system: the interface to this routine
     has been enhanced for use with AXIOM and is slightly different to
     that offered in the standard version of the Foundation Library.

     1. Purpose

     D02CJF integrates a system of first-order ordinary differential
     equations over a range with suitable initial conditions, using a
     variable-order, variable-step Adams method until a user-specified
     function, if supplied, of the solution is zero, and returns the
     solution at points specified by the user, if desired.

     2. Specification

            SUBROUTINE D02CJF (X, XEND, M, N, Y, FCN, TOL, RELABS,
           1                   RESULT, OUTPUT, G, W, IFAIL)
            INTEGER          M, N, IFAIL
            DOUBLE PRECISION X, XEND, Y(N), TOL, G, W(28+21*N), RESULT(M,N)
            CHARACTER*1      RELABS
            EXTERNAL         FCN, OUTPUT, G

     3. Description

     The routine advances the solution of a system of ordinary
     differential equations

                   y' =f (x,y ,y ,...,y ),  i=1,2,...,n,
                     i  i    1  2      n

     from x = X to x = XEND using a variable-order, variable-step
     Adams method. The system is defined by a subroutine FCN supplied
     by the user, which evaluates f  in terms of x and y ,y ,...,y .
                                   i                    1  2      n
     The initial values of y ,y ,...,y  must be given at x = X.
                            1  2      n

     The solution is returned via the user-supplied routine OUTPUT at
     points specified by the user, if desired: this solution is
                  1
     obtained by C  interpolation on solution values produced by the
     method. As the integration proceeds a check can be made on the
     user-specified function g(x,y) to determine an interval where it
     changes sign. The position of this sign change is then determined
                    1
     accurately by C  interpolation to the solution. It is assumed
     that g(x,y) is a continuous function of the variables, so that a
     solution of g(x,y)=0.0 can be determined by searching for a
     change in sign in g(x,y). The accuracy of the integration, the
     interpolation and, indirectly, of the determination of the
     position where g(x,y)=0.0, is controlled by the parameters TOL
     and RELABS.

     For a description of Adams methods and their practical
     implementation see Hall and Watt [1].

     4. References

     [1]   Hall G and Watt J M (eds) (1976) Modern Numerical Methods
           for Ordinary Differential Equations. Clarendon Press.

     5. Parameters

      1:  X -- DOUBLE PRECISION                           Input/Output
          On entry: the initial value of the independent variable x.
          Constraint: X /= XEND. On exit: if g is supplied by the
          user, it contains the point where g(x,y)=0.0, unless
          g(x,y)/=0.0 anywhere on the range X to XEND, in which case,
          X will contain XEND. If g is not supplied by the user it
          contains XEND, unless an error has occurred, when it
          contains the value of x at the error.

      2:  XEND -- DOUBLE PRECISION                               Input
          On entry: the final value of the independent variable. If
          XEND < X, integration proceeds in the negative direction.
          Constraint: XEND /= X.

      3:  M -- INTEGER                                           Input
          On entry: the first dimension of the array RESULT.  This
          will usually be equal to the number of points at which the
          solution is required.
          Constraint: M > 0.

      4:  N -- INTEGER                                           Input
          On entry: the number of differential equations.
          Constraint: N >= 1.

      5:  Y(N) -- DOUBLE PRECISION array                  Input/Output
          On entry: the initial values of the solution y ,y ,...,y
                                                         1  2      n
          at x = X. On exit: the computed values of the solution at
          the final point x = X.

      6:  FCN -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          FCN must evaluate the functions f  (i.e., the derivatives
                                           i
          y' ) for given values of their arguments x,y ,y ,...,y .
            i                                         1  2      n

          Its specification is:

                 SUBROUTINE FCN (X, Y, F)
                 DOUBLE PRECISION X, Y(n), F(n)
          where n is the actual value of N in the call of D02CJF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the independent variable x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the variable y , for
                                                     i
               i=1,2,...,n.

           3:  F(*) -- DOUBLE PRECISION array                   Output
               On exit: the value of f , for i=1,2,...,n.
                                       i
          FCN must be declared as EXTERNAL in the (sub)program
          from which D02CJF is called. Parameters denoted as
          Input must not be changed by this procedure.

      7:  TOL -- DOUBLE PRECISION                                Input
          On entry: a positive tolerance for controlling the error in
          the integration. Hence TOL affects the determination of the
          position where g(x,y)=0.0, if g is supplied.

          D02CJF has been designed so that, for most problems, a
          reduction in TOL leads to an approximately proportional
          reduction in the error in the solution. However, the actual
          relation between TOL and the accuracy achieved cannot be
          guaranteed. The user is strongly recommended to call D02CJF
          with more than one value for TOL and to compare the results
          obtained to estimate their accuracy. In the absence of any
          prior knowledge, the user might compare the results obtained
                                         -p             -p-1
          by calling D02CJF with TOL=10.0   and TOL=10.0     where p
          correct decimal digits are required in the solution.
          Constraint: TOL > 0.0.

      8:  RELABS -- CHARACTER*1                                  Input
          On entry: the type of error control. At each step in the
          numerical solution an estimate of the local error, EST, is
          made. For the current step to be accepted the following
          condition must be satisfied:
                           ______________________________
                          / n
                         /  --                          2
                 EST=   /   >  (e /((tau) *|y |+(tau) )) <=1.0
                       /    --   i       r   i       a
                     \/     i=1
          where (tau)  and (tau)  are defined by
                     r          a
               RELABS   (tau)    (tau)
                             r        a

               'M'      TOL      TOL

               'A'      0.0      TOL

               'R'      TOL      (epsilon)

               'D'      TOL      TOL
          where (epsilon) is a small machine-dependent number and e
                                                                   i
          is an estimate of the local error at y , computed
                                                i
          internally. If the appropriate condition is not satisfied,
          the step size is reduced and the solution is recomputed on
          the current step. If the user wishes to measure the error in
          the computed solution in terms of the number of correct
          decimal places, then RELABS should be set to 'A' on entry,
          whereas if the error requirement is in terms of the number
          of correct significant digits, then RELABS should be set to
          'R'. If the user prefers a mixed error test, then RELABS
          should be set to 'M', otherwise if the user has no
          preference, RELABS should be set to the default 'D'. Note
          that in this case 'D' is taken to be 'M'. Constraint: RELABS
          = 'M', 'A', 'R', 'D'.

      9:  RESULT(M,N) -- DOUBLE PRECISION array                 Output
          On exit: the computed values of the solution at the points
          given by OUTPUT.

     10:  OUTPUT -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          OUTPUT allows the user to have access to intermediate values
          of the computed solution at successive points specified by the
          user. These solution values may be returned to the user via
          the array RESULT if desired (this is a non-standard feature
          added for use with the AXIOM system).  OUTPUT is initially
          called by D02CJF with XSOL = X (the initial value of x). The
          user must reset XSOL to the next point where OUTPUT is to be
          called, and so on at each call to OUTPUT. If, after a call
          to OUTPUT, the reset point XSOL is beyond XEND, D02CJF will
          integrate to XEND with no further calls to OUTPUT; if a call
          to OUTPUT is required at the point XSOL = XEND, then XSOL
          must be given precisely the value XEND.

          Its specification is:

               SUBROUTINE OUTPUT(XSOL,Y,COUNT,M,N,RESULT)
               DOUBLE PRECISION Y(N),RESULT(M,N),XSOL
               INTEGER M,N,COUNT

           1:  XSOL -- DOUBLE PRECISION                   Input/Output
               On entry: the current value of the independent
               variable x. On exit: the next value of x at which
               OUTPUT is to be called.

           2:  Y(N) -- DOUBLE PRECISION array                    Input
               On entry: the computed solution at the point XSOL.

           3:  COUNT -- INTEGER                           Input/Output
               On entry: Zero if OUTPUT has not been called before, or
               the previous value of COUNT.
               On exit: A new value of COUNT: this can be used to keep
               track of the number of times OUTPUT has been called.

           4:  M -- INTEGER                                      Input
               On entry: The first dimension of RESULT.

           5:  N -- INTEGER                                      Input
               On entry: The dimension of Y.

           6:  RESULT(M,N) -- DOUBLE PRECISION array      Input/Output
               On entry: the previous contents of RESULT.
               On exit: RESULT may be used to return the values of the
               intermediate solutions to the user.

          OUTPUT must be declared as EXTERNAL in the (sub)program
          from which D02CJF is called. Parameters denoted as
          Input must not be changed by this procedure.

     11:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                               External Procedure
          G must evaluate the function g(x,y) for specified values x,y
          . It specifies the function g for which the first position x
          where g(x,y)=0 is to be found.

          If the user does not require the root finding option, the
          actual argument G  mustbe the dummy routine D02CJW. (D02CJW
          is included in the NAG Foundation Library and so need not be
          supplied by the user).

          Its specification is:

                 DOUBLE PRECISION FUNCTION G (X, Y)
                 DOUBLE PRECISION X, Y(n)
          where n is the actual value of N in the call of D02CJF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the independent variable x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the variable y , for
                                                    i
               i=1,2,...,n.
          G must be declared as EXTERNAL in the (sub)program from
          which D02CJF is called. Parameters denoted as Input
          must not be changed by this procedure.

     12:  W(28+21*N) -- DOUBLE PRECISION array               Workspace

     13:  IFAIL -- INTEGER                                Input/Output
          On entry: IFAIL must be set to 0, -1 or 1. For users not
          familiar with this parameter (described in the Essential
          Introduction) the recommended value is 0.

          On exit: IFAIL = 0 unless the routine detects an error (see
          Section 6).

     6. Error Indicators and Warnings

     Errors detected by the routine:

     If on entry IFAIL = 0 or -1, explanatory error messages are
     output on the current error message unit (as defined by X04AAF).

     IFAIL= 1
          On entry TOL <= 0.0,

          or       N <= 0,

          or       RELABS /= 'M', 'A', 'R' or 'D'.

          or       X = XEND.

     IFAIL= 2
          With the given value of TOL, no further progress can be made
          across the integration range from the current point x = X.
          (See Section 8 for a discussion of this error exit.) The
          components Y(1),Y(2),...,Y(N) contain the computed values of
          the solution at the current point x = X. If the user has
          supplied g, then no point at which g(x,y) changes sign has
          been located up to the point x = X.

     IFAIL= 3
          TOL is too small for D02CJF to take an initial step. X and Y
          (1),Y(2),...,Y(N) retain their initial values.

     IFAIL= 4
          XSOL has not been reset or XSOL lies behind X in the
          direction of integration, after the initial call to OUTPUT,
          if the OUTPUT option was selected.

     IFAIL= 5
          A value of XSOL returned by OUTPUT has not been reset or
          lies behind the last value of XSOL in the direction of
          integration, if the OUTPUT option was selected.

     IFAIL= 6
          At no point in the range X to XEND did the function g(x,y)
          change sign, if g was supplied. It is assumed that g(x,y)=0
          has no solution.

     IFAIL= 7
          A serious error has occurred in an internal call. Check all
          subroutine calls and array sizes. Seek expert help.

     7. Accuracy

     The accuracy of the computation of the solution vector Y may be
     controlled by varying the local error tolerance TOL. In general,
     a decrease in local error tolerance should lead to an increase in
     accuracy. Users are advised to choose RELABS = 'M' unless they
     have a good reason for a different choice.

     If the problem is a root-finding one, then the accuracy of the
     root determined will depend on the properties of g(x,y). The user
     should try to code G without introducing any unnecessary
     cancellation errors.

     8. Further Comments

     If more than one root is required then D02QFF(*) should be used.

     If the routine fails with IFAIL = 3, then it can be called again
     with a larger value of TOL if this has not already been tried. If
     the accuracy requested is really needed and cannot be obtained
     with this routine, the system may be very stiff (see below) or so
     badly scaled that it cannot be solved to the required accuracy.

     If the routine fails with IFAIL = 2, it is probable that it has
     been called with a value of TOL which is so small that a solution
     cannot be obtained on the range X to XEND. This can happen for
     well-behaved systems and very small values of TOL. The user
     should, however, consider whether there is a more fundamental
     difficulty. For example:

     (a)   in the region of a singularity (infinite value) of the
           solution, the routine will usually stop with IFAIL = 2,
           unless overflow occurs first. Numerical integration cannot
           be continued through a singularity, and analytic treatment
           should be considered;

     (b)   for 'stiff' equations where the solution contains rapidly
           decaying components, the routine will use very small steps
           in x (internally to D02CJF) to preserve stability. This
           will exhibit itself by making the computing time
           excessively long, or occasionally by an exit with IFAIL =
           2. Adams methods are not efficient in such cases, and the
           user should try D02EJF.

     9. Example

     We illustrate the solution of four different problems. In each
     case the differential system (for a projectile) is

                                y'=tan(phi)

                           -0.032tan(phi)   0.02v
                       v'= --------------- --------
                                 v         cos(phi)

                                      -0.032
                              (phi)'= ------
                                         2
                                        v

     over an interval X = 0.0 to XEND = 10.0 starting with values
     y=0.5, v=0.5 and (phi)=(pi)/5. We solve each of the following
     problems with local error tolerances 1.0E-4 and 1.0E-5.

     (i)   To integrate to x=10.0 producing output at intervals of 2.0
           until a point is encountered where y=0.0.

     (ii)  As (i) but with no intermediate output.

     (iii) As (i) but with no termination on a root-finding condition.

     (iv)  As (i) but with no intermediate output and no root-finding
           termination condition.

     The example program is not reproduced here. The source code for
     all example programs is distributed with the NAG Foundation
     Library software and should be available on-line.
\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02ejf}{NAG On-line Documentation: d02ejf}
\beginscroll
\begin{verbatim}



     D02EJF(3NAG)                 D02EJF                  D02EJF(3NAG)



     D02 -- Ordinary Differential Equations                     D02EJF
             D02EJF -- NAG Foundation Library Routine Document

     Note: Before using this routine, please read the Users' Note for
     your implementation to check implementation-dependent details.
     The symbol (*) after a NAG routine name denotes a routine that is
     not included in the Foundation Library.

     Note for users via the AXIOM system: the interface to this routine
     has been enhanced for use with AXIOM and is slightly different to
     that offered in the standard version of the Foundation Library.

     1. Purpose

     D02EJF integrates a stiff system of first-order ordinary
     differential equations over an interval with suitable initial
     conditions, using a variable-order, variable-step method
     implementing the Backward Differentiation Formulae (BDF), until a
     user-specified function, if supplied, of the solution is zero,
     and returns the solution at points specified by the user, if
     desired.

     2. Specification

            SUBROUTINE D02EJF (X, XEND, M, N, Y, FCN, PEDERV, TOL,
           1                   RELABS, OUTPUT, G, W, IW, RESULT, IFAIL)
            INTEGER          M, N, IW, IFAIL
            DOUBLE PRECISION X, XEND, Y(N), TOL, G, W(IW), RESULT(M,N)
            CHARACTER*1      RELABS
            EXTERNAL         FCN, PEDERV, OUTPUT, G

     3. Description

     The routine advances the solution of a system of ordinary
     differential equations

                   y' =f (x,y ,y ,...,y ),  i=1,2,...,n,
                     i  i    1  2      n

     from x = X to x = XEND using a variable-order, variable-step
     method implementing the BDF. The system is defined by a
     subroutine FCN supplied by the user, which evaluates f  in terms
                                                           i
     of x and y ,y ,...,y  (see Section 5). The initial values of
               1  2      n
     y ,y ,...,y  must be given at x = X.
      1  2      n

     The solution is returned via the user-supplied routine OUTPUT at
     points specified by the user, if desired: this solution is
                  1
     obtained by C  interpolation on solution values produced by the
     method. As the integration proceeds a check can be made on the
     user-specified function g(x,y) to determine an interval where it
     changes sign. The position of this sign change is then determined
                    1
     accurately by C  interpolation to the solution. It is assumed
     that g(x,y) is a continuous function of the variables, so that a
     solution of g(x,y) = 0.0 can be determined by searching for a
     change in sign in g(x,y). The accuracy of the integration, the
     interpolation and, indirectly, of the determination of the
     position where g(x,y) = 0.0, is controlled by the parameters TOL
     and RELABS. The Jacobian of the system y'=f(x,y) may be supplied
     in routine PEDERV, if it is available.

     For a description of BDF and their practical implementation see
     Hall and Watt [1].

     4. References

     [1]   Hall G and Watt J M (eds) (1976) Modern Numerical Methods
           for Ordinary Differential Equations. Clarendon Press.

     5. Parameters

      1:  X -- DOUBLE PRECISION                           Input/Output
          On entry: the initial value of the independent variable x.
          Constraint: X /= XEND On exit: if G is supplied by the user,
          X contains the point where g(x,y) = 0.0, unless g(x,y) /= 0.
          0 anywhere on the range X to XEND, in which case, X will
          contain XEND. If G is not supplied X contains XEND, unless
          an error has occured, when it contains the value of x at the
          error.

      2:  XEND -- DOUBLE PRECISION                               Input
          On entry: the final value of the independent variable. If
          XEND < X, integration proceeds in the negative direction.
          Constraint: XEND /= X.

      3:  M -- INTEGER                                           Input
          On entry: the first dimension of the array RESULT.  This
          will usually be equal to the number of points at which the
          solution is required.
          Constraint: M > 0.

      4:  N -- INTEGER                                           Input
          On entry: the number of differential equations, n.
          Constraint: N >= 1.

      5:  Y(N) -- DOUBLE PRECISION array                  Input/Output
          On entry: the initial values of the solution y ,y ,...,y
                                                        1  2      n
          at x = X. On exit: the computed values of the solution at
          the final point x = X.

      6:  FCN -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          FCN must evaluate the functions f  (i.e., the derivatives
                                           i
          y' ) for given values of their arguments x,y ,y ,...,y .
            i                                         1  2      n

          Its specification is:

                 SUBROUTINE FCN (X, Y, F)
                 DOUBLE PRECISION X, Y(n), F(n)
          where n is the actual value of N in the call of D02EJF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the independent variable x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the variable y , for
                                                    i
               i=1,2,...,n.

           3:  F(*) -- DOUBLE PRECISION array                   Output
               On exit: the value of f , for i=1,2,...,n.
                                      i
          FCN must be declared as EXTERNAL in the (sub)program
          from which D02EJF is called. Parameters denoted as
          Input must not be changed by this procedure.

      7:  PEDERV -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          PEDERV must evaluate the Jacobian of the system (that is,
                                   ddf
                                      i
          the partial derivatives  ----) for given values of the
                                   ddy
                                      j
          variables x,y ,y ,...,y .
                       1  2      n

          Its specification is:

                 SUBROUTINE PEDERV (X, Y, PW)
                 DOUBLE PRECISION X, Y(n), PW(n,n)
          where n is the actual value of N in the call of D02EJF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the independent variable x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the variable y , for
                                                    i
               i=1,2,...,n.

           3:  PW(n,*) -- DOUBLE PRECISION array                Output
                                       ddf
                                          i
               On exit: the value of  ----, for i,j=1,2,...,n.
                                       ddy
                                          j
          If the user does not wish to supply the Jacobian, the
          actual argument PEDERV must be the dummy routine D02EJY
          . (D02EJY is included in the NAG Foundation Library and
          so need not be supplied by the user. The name may be
          implementation dependent: see the User's Note for your
          implementation for details).
          PEDERV must be declared as EXTERNAL in the (sub)program
          from which D02EJF is called. Parameters denoted as
          Input must not be changed by this procedure.

      8:  TOL -- DOUBLE PRECISION                         Input/Output
          On entry: TOL must be set to a positive tolerance for
          controlling the error in the integration. Hence TOL affects
          the determination of the position where g(x,y) = 0.0, if G
          is supplied.

          D02EJF has been designed so that, for most problems, a
          reduction in TOL leads to an approximately proportional
          reduction in the error in the solution. However, the actual
          relation between TOL and the accuracy achieved cannot be
          guaranteed. The user is strongly recommended to call D02EJF
          with more than one value for TOL and to compare the results
          obtained to estimate their accuracy. In the absence of any
          prior knowledge, the user might compare the results obtained
                                       -p           -p-1
          by calling D02EJF with TOL=10   and TOL=10     if p correct
          decimal digits are required in the solution. Constraint: TOL
          > 0.0. On exit: normally unchanged. However if the range X
          to XEND is so short that a small change in TOL is unlikely
          to make any change in the computed solution, then, on
          return, TOL has its sign changed.

      9:  RELABS -- CHARACTER*1                                  Input
          On entry: the type of error control. At each step in the
          numerical solution an estimate of the local error, EST, is
          made. For the current step to be accepted the following
          condition must be satisfied:
                          ________________________________
                         /   n
                        /  1 --                          2
                EST=   /   - >  (e /((tau) *|y |+(tau) )) <=1.0
                      /    n --   i       r   i       a
                    \/       i=1
          where (tau)  and (tau)  are defined by
                     r          a
               RELABS   (tau)    (tau)
                             r        a

               'M'      TOL      TOL

               'A'      0.0      TOL

               'R'      TOL      (epsilon)

               'D'      TOL      (epsilon)
          where (epsilon) is a small machine-dependent number and e
                                                                   i
          is an estimate of the local error at y , computed
                                                i
          internally. If the appropriate condition is not satisfied,
          the step size is reduced and the solution is recomputed on
          the current step. If the user wishes to measure the error in
          the computed solution in terms of the number of correct
          decimal places, then RELABS should be set to 'A' on entry,
          whereas if the error requirement is in terms of the number
          of correct significant digits, then RELABS should be set to
          'R'. If the user prefers a mixed error test, then RELABS
          should be set to 'M', otherwise if the user has no
          preference, RELABS should be set to the default 'D'. Note
          that in this case 'D' is taken to be 'R'. Constraint: RELABS
          = 'A', 'M', 'R' or 'D'.

     10:  OUTPUT -- SUBROUTINE, supplied by the user.
                                                    External Procedure
          OUTPUT allows the user to have access to intermediate values
          of the computed solution at successive points specified by the
          user. These solution values may be returned to the user via
          the array RESULT if desired (this is a non-standard feature
          added for use with the AXIOM system).  OUTPUT is initially
          called by D02EJF with XSOL = X (the initial value of x). The
          user must reset XSOL to the next point where OUTPUT is to be
          called, and so on at each call to OUTPUT. If, after a call
          to OUTPUT, the reset point XSOL is beyond XEND, D02EJF will
          integrate to XEND with no further calls to OUTPUT; if a call
          to OUTPUT is required at the point XSOL = XEND, then XSOL
          must be given precisely the value XEND.

          Its specification is:

               SUBROUTINE OUTPUT(XSOL,Y,COUNT,M,N,RESULT)
               DOUBLE PRECISION Y(N),RESULT(M,N),XSOL
               INTEGER M,N,COUNT

           1:  XSOL -- DOUBLE PRECISION                   Input/Output
               On entry: the current value of the independent
               variable x. On exit: the next value of x at which
               OUTPUT is to be called.

           2:  Y(N) -- DOUBLE PRECISION array                    Input
               On entry: the computed solution at the point XSOL.

           3:  COUNT -- INTEGER                           Input/Output
               On entry: Zero if OUTPUT has not been called before, or
               the previous value of COUNT.
               On exit: A new value of COUNT: this can be used to keep
               track of the number of times OUTPUT has been called.

           4:  M -- INTEGER                                      Input
               On entry: The first dimension of RESULT.

           5:  N -- INTEGER                                      Input
               On entry: The dimension of Y.

           6:  RESULT(M,N) -- DOUBLE PRECISION array      Input/Output
               On entry: the previous contents of RESULT.
               On exit: RESULT may be used to return the values of the
               intermediate solutions to the user.

          OUTPUT must be declared as EXTERNAL in the (sub)program
          from which D02EJF is called. Parameters denoted as
          Input must not be changed by this procedure.

     11:  G -- DOUBLE PRECISION FUNCTION, supplied by the user.
                                                    External Procedure
          G must evaluate the function g(x,y) for specified values x,y
          . It specifies the function g for which the first position x
          where g(x,y) = 0 is to be found.

          Its specification is:

                 DOUBLE PRECISION FUNCTION G (X, Y)
                 DOUBLE PRECISION X, Y(n)
          where n is the actual value of N in the call of D02EJF.

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the value of the independent variable x.

           2:  Y(*) -- DOUBLE PRECISION array                    Input
               On entry: the value of the variable y , for
                                                    i
               i=1,2,...,n.
          If the user does not require the root finding option,
          the actual argument G must be the dummy routine D02EJW.
          (D02EJW is included in the NAG Foundation Library and
          so need not be supplied by the user).
          G must be declared as EXTERNAL in the (sub)program from
          which D02EJF is called. Parameters denoted as Input
          must not be changed by this procedure.

     12:  W(IW) -- DOUBLE PRECISION array                    Workspace

     13:  IW -- INTEGER                                          Input
          On entry:
          the dimension of the array W as declared in the (sub)program
          from which D02EJF is called.
          Constraint: IW>=(12+N)*N+50.

     14:  RESULT(M,N) -- DOUBLE PRECISION array                 Output
          On exit: the computed values of the solution at the points
          given by OUTPUT.

     15:  IFAIL -- INTEGER                                Input/Output
          On entry: IFAIL must be set to 0, -1 or 1. For users not
          familiar with this parameter (described in the Essential
          Introduction) the recommended value is 0.

          On exit: IFAIL = 0 unless the routine detects an error (see
          Section 6).

     6. Error Indicators and Warnings

     Errors detected by the routine:

     If on entry IFAIL = 0 or -1, explanatory error messages are
     output on the current error message unit (as defined by X04AAF).

     IFAIL= 1
          On entry TOL <= 0.0,

          or       X = XEND,

          or       N <= 0,

          or       RELABS /= 'M', 'A', 'R', 'D'.

          or       IW<(12+N)*N+50.

     IFAIL= 2
          With the given value of TOL, no further progress can be made
          across the integration range from the current point x = X.
          (See Section 5 for a discussion of this error test.) The
          components Y(1),Y(2),...,Y(n) contain the computed values of
          the solution at the current point x = X. If the user has
          supplied G, then no point at which g(x,y) changes sign has
          been located up to the point x = X.

     IFAIL= 3
          TOL is too small for D02EJF to take an initial step. X and Y
          (1),Y(2),...,Y(n) retain their initial values.

     IFAIL= 4
          XSOL lies behind X in the direction of integration, after
          the initial call to OUTPUT, if the OUTPUT option was
          selected.

     IFAIL= 5
          A value of XSOL returned by OUTPUT lies behind the last
          value of XSOL in the direction of integration, if the OUTPUT
          option was selected.

     IFAIL= 6
          At no point in the range X to XEND did the function g(x,y)
          change sign, if G was supplied. It is assumed that g(x,y) =
          0 has no solution.

     IFAIL= 7
          A serious error has occurred in an internal call to
          C05AZF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

     IFAIL= 8
          A serious error has occurred in an internal call to
          D02XKF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

     IFAIL= 9
          A serious error has occurred in an internal call to
          D02NMF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

     7. Accuracy

     The accuracy of the computation of the solution vector Y may be
     controlled by varying the local error tolerance TOL. In general,
     a decrease in local error tolerance should lead to an increase in
     accuracy. Users are advised to choose RELABS = 'R' unless they
     have a good reason for a different choice. It is particularly
     appropriate if the solution decays.

     If the problem is a root-finding one, then the accuracy of the
                                              ddg      ddg
     root determined will depend strongly on  --- and  ----, for
                                              ddx      ddy
                                                          i
     i=1,2,...,n. Large values for these quantities may imply large
     errors in the root.

     8. Further Comments

     If more than one root is required, then to determine the second
     and later roots D02EJF may be called again starting a short
     distance past the previously determined roots. Alternatively the
     user may construct his own root finding code using D02QDF(*) (or
     the routines of the subchapter D02M-D02N), D02XKF(*) and
     C05AZF(*).

     If it is easy to code, the user should supply the routine PEDERV.
     However, it is important to be aware that if PEDERV is coded
     incorrectly, a very inefficient integration may result and
     possibly even a failure to complete the integration (IFAIL = 2).

     9. Example

     We illustrate the solution of five different problems. In each
     case the differential system is the well-known stiff Robertson
     problem.

                                      4
                         a'= -0.04a-10 bc
                                      4        7 2
                         b'=  0.04a-10 bc -3*10 b
                                               7 2
                         c'=               3*10 b

     with initial conditions a=1.0, b=c=0.0 at x=0.0. We solve each of
     the following problems with local error tolerances 1.0E-3 and 1.0
     E-4.

     (i)   To integrate to x=10.0 producing output at intervals of 2.0
           until a point is encountered where a=0.9. The Jacobian is
           calculated numerically.

     (ii)  As (i) but with the Jacobian calculated analytically.

     (iii) As (i) but with no intermediate output.

     (iv)  As (i) but with no termination on a root-finding condition.

     (v)   Integrating the equations as in (i) but with no
           intermediate output and no root-finding termination
           condition.

     The example program is not reproduced here. The source code for
     all example programs is distributed with the NAG Foundation
     Library software and should be available on-line.
\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02gaf}{NAG On-line Documentation: d02gaf}
\beginscroll
\begin{verbatim}



     D02GAF(3NAG)      Foundation Library (12/10/92)      D02GAF(3NAG)



          D02 -- Ordinary Differential Equations                     D02GAF
                  D02GAF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D02GAF solves the two-point boundary-value problem with assigned
          boundary values for a system of ordinary differential equations,
          using a deferred correction technique and a Newton iteration.

          2. Specification

                 SUBROUTINE D02GAF (U, V, N, A, B, TOL, FCN, MNP, X, Y, NP,
                1                   W, LW, IW, LIW, IFAIL)
                 INTEGER          N, MNP, NP, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION U(N,2), V(N,2), A, B, TOL, X(MNP), Y
                1                 (N,MNP), W(LW)
                 EXTERNAL         FCN

          3. Description

          D02GAF solves a two-point boundary-value problem for a system of
          n differential equations in the interval [a,b]. The system is
          written in the form

                      y' =f (x,y ,y ,...,y ) , i=1,2,...,n              (1)
                        i  i    1  2      n

          and the derivatives are evaluated by a subroutine FCN supplied by
          the user. Initially, n boundary values of the variables y  must
                                                                   i
          be specified (assigned), some at a and some at b. The user also
          supplies estimates of the remaining n boundary values and all the
          boundary values are used in constructing an initial approximation
          to the solution. This approximate solution is corrected by a
          finite-difference technique with deferred correction allied with
          a Newton iteration to solve the finite-difference equations. The
          technique used is described fully in Pereyra [1]. The Newton
                                                ddf
                                                   i
          iteration requires a Jacobian matrix  ---- and this is calculated
                                                ddy
                                                   j
          by numerical differentiation using an algorithm described in
          Curtis et al [2].

          The user supplies an absolute error tolerance and may also supply
          an initial mesh for the construction of the finite-difference
          equations (alternatively a default mesh is used). The algorithm
          constructs a solution on a mesh defined by adding points to the
          initial mesh. This solution is chosen so that the error is
          everywhere less than the user's tolerance and so that the error
          is approximately equidistributed on the final mesh. The solution
          is returned on this final mesh.

          If the solution is required at a few specific points then these
          should be included in the initial mesh. If on the other hand the
          solution is required at several specific points then the user
          should use the interpolation routines provided in Chapter E01 if
          these points do not themselves form a convenient mesh.

          4. References

          [1]   Pereyra V (1979) PASVA3: An Adaptive Finite-Difference
                Fortran Program for First Order Nonlinear, Ordinary Boundary
                Problems. Codes for Boundary Value Problems in Ordinary
                Differential Equations. Lecture Notes in Computer Science.
                (ed B Childs, M Scott, J W Daniel, E Denman and P Nelson) 76
                Springer-Verlag.

          [2]   Curtis A R, Powell M J D and Reid J K (1974) On the
                Estimation of Sparse Jacobian Matrices. J. Inst. Maths
                Applics. 13 117--119.

          5. Parameters

           1:  U(N,2) -- DOUBLE PRECISION array                       Input
               On entry: U(i,1) must be set to the known (assigned) or
               estimated values of y  at a and U(i,2) must be set to the
                                    i
               known or estimated values of y  at b, for i=1,2,...,n.
                                             i

           2:  V(N,2) -- DOUBLE PRECISION array                       Input
               On entry: V(i,j) must be set to 0.0 if U(i,j) is a known
               (assigned) value and to 1.0 if U(i,j) is an estimated value,
               i=1,2,...,n; j=1,2. Constraint: precisely N of the V(i,j)
               must be set to 0.0, i.e., precisely N of the U(i,j) must be
               known values, and these must not be all at a or all at b.

           3:  N -- INTEGER                                           Input
               On entry: the number of equations. Constraint: N >= 2.

           4:  A -- DOUBLE PRECISION                                  Input
               On entry: the left-hand boundary point, a.

           5:  B -- DOUBLE PRECISION                                  Input
               On entry: the right-hand boundary point, b. Constraint: B >
               A.

           6:  TOL -- DOUBLE PRECISION                                Input
               On entry: a positive absolute error tolerance. If
                                    a=x <x <...<x  =b
                                       1  2      NP
               is the final mesh, z (x ) is the jth component of the
                                   j  i
               approximate solution at x , and y (x) is the jth component
                                        i       j
               of the true solution of equation (1) (see Section 3) and the
               boundary conditions, then, except in extreme cases, it is
               expected that
                   |z (x )-y (x )|<=TOL , i=1,2,...,NP;j=1,2,...,n      (2)
                     j  i   j  i
                Constraint: TOL > 0.0.

           7:  FCN -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               FCN must evaluate the functions f  (i.e., the derivatives
                                                i
               y' ) at the general point x.
                 i

               Its specification is:

                      SUBROUTINE FCN (X, Y, F)
                      DOUBLE PRECISION X, Y(n), F(n)
               where n is the actual value of N in the call of D02GAF.

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the argument x.

                2:  Y(*) -- DOUBLE PRECISION array                    Input
                    On entry: the value of the argument y , for
                                                         i
                    i=1,2,...,n.

                3:  F(*) -- DOUBLE PRECISION array                   Output
                    On exit: the values of f , for i=1,2,...,n.
                                            i
               FCN must be declared as EXTERNAL in the (sub)program
               from which D02GAF is called. Parameters denoted as
               Input must not be changed by this procedure.

           8:  MNP -- INTEGER                                         Input
               On entry: the maximum permitted number of mesh points.
               Constraint: MNP >= 32.

           9:  X(MNP) -- DOUBLE PRECISION array                Input/Output
               On entry: if NP >= 4 (see NP below), the first NP elements
               must define an initial mesh. Otherwise the elements of X
               need not be set. Constraint:
                          A=X(1)<X(2)<...<X(NP)=B for NP>=4             (3)
               On exit: X(1),X(2),...,X(NP) define the final mesh (with
               the returned value of NP) satisfying the relation (3).

          10:  Y(N,MNP) -- DOUBLE PRECISION array                    Output
               On exit: the approximate solution z (x ) satisfying (2), on
                                                  j  i
               the final mesh, that is
                        Y(j,i)=z (x ) , i=1,2,...,NP;j=1,2,...,n,
                                j  i
               where NP is the number of points in the final mesh.

               The remaining columns of Y are not used.

          11:  NP -- INTEGER                                   Input/Output
               On entry: determines whether a default or user-supplied
               mesh is used. If NP = 0, a default value of 4 for NP and a
               corresponding equispaced mesh X(1),X(2),...,X(NP) are used.
               If NP >= 4, then the user must define an initial mesh using
               the array X as described. Constraint: NP = 0 or 4 <= NP <=
               MNP. On exit: the number of points in the final (returned)
               mesh.

          12:  W(LW) -- DOUBLE PRECISION array                    Workspace

          13:  LW -- INTEGER                                          Input
               On entry: the length of the array W as declared in the
                                                            2         2
               calling (sub)program. Constraint: LW>=MNP*(3N +6N+2)+4N +4N

          14:  IW(LIW) -- INTEGER array                           Workspace

          15:  LIW -- INTEGER                                         Input
               On entry: the length of the array IW as declared in the
                                                                  2
               calling (sub)program. Constraint: LIW>=MNP*(2N+1)+N +4N+2.

          16:  IFAIL -- INTEGER                                Input/Output
               For this routine, the normal use of IFAIL is extended to
               control the printing of error and warning messages as well
               as specifying hard or soft failure (see the Essential
               Introduction).

               Before entry, IFAIL must be set to a value with the decimal
               expansion cba, where each of the decimal digits c, b and a
               must have a value of 0 or 1.
               a=0   specifies hard failure, otherwise soft failure;

               b=0   suppresses error messages, otherwise error messages
                     will be printed (see Section 6);

               c=0   suppresses warning messages, otherwise warning
                     messages will be printed (see Section 6).
               The recommended value for inexperienced users is 110 (i.e.,
               hard failure with all messages printed).

               Unless the routine detects an error (see Section 6), IFAIL
               contains 0 on exit.

          6. Error Indicators and Warnings

          Errors detected by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               One or more of the parameters N, TOL, NP, MNP, LW or LIW has
               been incorrectly set, or B <= A, or the condition (3) on X
               is not satisfied, or the number of known boundary values
               (specified by V) is not N.

          IFAIL= 2
               The Newton iteration has failed to converge. This could be
               due to there being too few points in the initial mesh or to
               the initial approximate solution being too inaccurate. If
               this latter reason is suspected the user should use
               subroutine D02RAF instead. If the warning 'Jacobian matrix
               is singular' is printed this could be due to specifying zero
               estimated boundary values and these should be varied. This
               warning could also be printed in the unlikely event of the
               Jacobian matrix being calculated inaccurately. If the user
               cannot make changes to prevent the warning then subroutine
               D02RAF should be used.

          IFAIL= 3
               The Newton iteration has reached round-off level. It could
               be, however, that the answer returned is satisfactory. This
               error might occur if too much accuracy is requested.

          IFAIL= 4
               A finer mesh is required for the accuracy requested; that is
               MNP is not large enough.

          IFAIL= 5
               A serious error has occurred in a call to D02GAF. Check all
               array subscripts and subroutine parameter lists in calls to
               D02GAF. Seek expert help.

          7. Accuracy

          The solution returned by the routine will be accurate to the
          user's tolerance as defined by the relation (2) except in extreme
          circumstances. If too many points are specified in the initial
          mesh, the solution may be more accurate than requested and the
          error may not be approximately equidistributed.

          8. Further Comments

          The time taken by the routine depends on the difficulty of the
          problem, the number of mesh points used (and the number of
          different meshes used), the number of Newton iterations and the
          number of deferred corrections.

          The user is strongly recommended to set IFAIL to obtain self-
          explanatory error messages, and also monitoring information about
          the course of the computation. The user may select the channel
          numbers on which this output is to appear by calls of X04AAF (for
          error messages) or X04ABF (for monitoring information) - see
          Section 9 for an example. Otherwise the default channel numbers
          will be used, as specified in the implementation document.

          A common cause of convergence problems in the Newton iteration is
          the user specifying too few points in the initial mesh. Although
          the routine adds points to the mesh to improve accuracy it is
          unable to do so until the solution on the initial mesh has been
          calculated in the Newton iteration.

          If the user specifies zero known and estimated boundary values,
          the routine constructs a zero initial approximation and in many
          cases the Jacobian is singular when evaluated for this
          approximation, leading to the breakdown of the Newton iteration.

          The user may be unable to provide a sufficiently good choice of
          initial mesh and estimated boundary values, and hence the Newton
          iteration may never converge. In this case the continuation
          facility provided in D02RAF is recommended.

          In the case where the user wishes to solve a sequence of similar
          problems, the final mesh from solving one case is strongly
          recommended as the initial mesh for the next.

          9. Example

          We solve the differential equation

                                                    2
                              y'''=-yy''-(beta)(1-y' )

          with boundary conditions

                                    y(0)=y'(0)=0,

                                      y'(10)=1

          for (beta)=0.0 and (beta)=0.2 to an accuracy specified by TOL=1.0
          E-3. We solve first the simpler problem with (beta)=0.0 using an
          equispaced mesh of 26 points and then we solve the problem with
          (beta)=0.2 using the final mesh from the first problem.

          Note the call to X04ABF prior to the call to D02GAF.

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02gbf}{NAG On-line Documentation: d02gbf}
\beginscroll
\begin{verbatim}



     D02GBF(3NAG)      Foundation Library (12/10/92)      D02GBF(3NAG)



          D02 -- Ordinary Differential Equations                     D02GBF
                  D02GBF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D02GBF solves a general linear two-point boundary value problem
          for a system of ordinary differential equations using a deferred
          correction technique.

          2. Specification

                 SUBROUTINE D02GBF (A, B, N, TOL, FCNF, FCNG, C, D, GAM,
                1                   MNP, X, Y, NP, W, LW, IW, LIW, IFAIL)
                 INTEGER          N, MNP, NP, LW, IW(LIW), LIW, IFAIL
                 DOUBLE PRECISION A, B, TOL, C(N,N), D(N,N), GAM(N), X(MNP),
                1                 Y(N,MNP), W(LW)
                 EXTERNAL         FCNF, FCNG

          3. Description

          D02GBF solves the linear two-point boundary value problem for a
          system of n ordinary differential equations in the interval

          [a,b]. The system is written in the form

                                 y'=F(x)y+g(x)                          (1)

          and the boundary conditions are written in the form

                              Cy(a)+Dy(b)=(gamma)                       (2)

          Here F(x), C and D are n by n matrices, and g(x) and (gamma) are
          n-component vectors. The approximate solution to (1) and (2) is
          found using a finite-difference method with deferred correction.
          The algorithm is a specialisation of that used in subroutine
          D02RAF which solves a nonlinear version of (1) and (2). The
          nonlinear version of the algorithm is described fully in Pereyra
          [1].

          The user supplies an absolute error tolerance and may also supply
          an initial mesh for the construction of the finite-difference
          equations (alternatively a default mesh is used). The algorithm
          constructs a solution on a mesh defined by adding points to the
          initial mesh. This solution is chosen so that the error is
          everywhere less than the user's tolerance and so that the error
          is approximately equidistributed on the final mesh. The solution
          is returned on this final mesh.

          If the solution is required at a few specific points then these
          should be included in the initial mesh. If, on the other hand,
          the solution is required at several specific points, then the
          user should use the interpolation routines provided in Chapter
          E01 if these points do not themselves form a convenient mesh.

          4. References

          [1]   Pereyra V (1979) PASVA3: An Adaptive Finite-Difference
                Fortran Program for First Order Nonlinear, Ordinary Boundary
                Problems. Codes for Boundary Value Problems in Ordinary
                Differential Equations. Lecture Notes in Computer Science.
                (ed B Childs, M Scott, J W Daniel, E Denman and P Nelson) 76
                Springer-Verlag.

          5. Parameters

           1:  A -- DOUBLE PRECISION                                  Input
               On entry: the left-hand boundary point, a.

           2:  B -- DOUBLE PRECISION                                  Input
               On entry: the right-hand boundary point, b. Constraint: B >
               A.

           3:  N -- INTEGER                                           Input
               On entry: the number of equations; that is n is the order of
               system (1). Constraint: N >= 2.

           4:  TOL -- DOUBLE PRECISION                                Input
               On entry: a positive absolute error tolerance. If
                                    a=x <x <...<x  =b
                                       1  2      NP
               is the final mesh, z(x) is the approximate solution from
               D02GBF and y(x) is the true solution of equations (1) and
               (2) then, except in extreme cases, it is expected that
                                    ||z-y||<=TOL                        (3)
               where
                           ||u||=max       max        |u (x )|.
                                    1<=i<=N   1<=j<=NP  i  j
               Constraint: TOL > 0.0.

           5:  FCNF -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               FCNF must evaluate the matrix F(x) in (1) at a general point
               x.

               Its specification is:

                      SUBROUTINE FCN (X, F)
                      DOUBLE PRECISION X, F(n,n)
               where n is the actual value of N in the call of D02GBF.

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the independent variable x.

                2:  F(n,n) -- DOUBLE PRECISION array                 Output
                    On exit: the (i,j)th element of the matrix F(x), for
                    i,j=1,2,...,n. (See Section 9 for an example.)
               FCN must be declared as EXTERNAL in the (sub)program
               from which D02GBF is called. Parameters denoted as
               Input must not be changed by this procedure.

           6:  FCNG -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               FCNG must evaluate the vector g(x) in (1) at a general point
               x.

               Its specification is:

                      SUBROUTINE FCNG (X, G)
                      DOUBLE PRECISION X, G(n)
               where n is the actual value of N in the call of D02GBF.

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the independent variable x.

                2:  G(*) -- DOUBLE PRECISION array                   Output
                    On exit: the ith element of the vector g(x), for
                    i=1,2,...,n. (See Section Section 9 for an example.)
               FCNG must be declared as EXTERNAL in the (sub)program
               from which D02GBF is called. Parameters denoted as
               Input must not be changed by this procedure.

           7:  C(N,N) -- DOUBLE PRECISION array                Input/Output

           8:  D(N,N) -- DOUBLE PRECISION array                Input/Output

           9:  GAM(N) -- DOUBLE PRECISION array                Input/Output
               On entry: the arrays C and D must be set to the matrices C
               and D in (2). GAM must be set to the vector (gamma) in (2).
               On exit: the rows of C and D and the components of GAM are
               re-ordered so that the boundary conditions are in the order:
                 (i)   conditions on y(a) only;

                 (ii)  condition involving y(a) and y(b); and

                 (iii) conditions on y(b) only.

               The routine will be slightly more efficient if the arrays C,
               D and GAM are ordered in this way before entry, and in this
               event they will be unchanged on exit.

               Note that the problems (1) and (2) must be of boundary value
               type, that is neither C nor D may be identically zero. Note
               also that the rank of the matrix [C,D] must be n for the
               problem to be properly posed. Any violation of these
               conditions will lead to an error exit.

          10:  MNP -- INTEGER                                         Input
               On entry: the maximum permitted number of mesh points.
               Constraint: MNP >= 32.

          11:  X(MNP) -- DOUBLE PRECISION array                Input/Output
               On entry: if NP >= 4 (see NP below), the first NP elements
               must define an initial mesh. Otherwise the elements of x
               need not be set. Constraint:
                         A=X(1)<X(2)<...<X(NP)=B, for NP>=4.            (4)
               On exit: X(1),X(2),...,X(NP) define the final mesh (with the
               returned value of NP) satisfying the relation (4).

          12:  Y(N,MNP) -- DOUBLE PRECISION array                    Output
               On exit: the approximate solution z(x) satisfying (3), on
               the final mesh, that is
                         Y(j,i)=z (x ) , i=1,2,...,NP;j=1,2,...,n
                                 j  i
               where NP is the number of points in the final mesh.

               The remaining columns of Y are not used.

          13:  NP -- INTEGER                                   Input/Output
               On entry: determines whether a default mesh or user-supplied
               mesh is used. If NP = 0, a default value of 4 for NP and a
               corresponding equispaced mesh X(1),X(2),...,X(NP) are used.
               If NP >= 4, then the user must define an initial mesh X as
               in (4) above. On exit: the number of points in the final
               (returned) mesh.

          14:  W(LW) -- DOUBLE PRECISION array                    Workspace

          15:  LW -- INTEGER                                          Input
               On entry: the length of the array W, Constraint:
                          2         2
               LW>=MNP*(3N +5N+2)+3N +5N.

          16:  IW(LIW) -- INTEGER array                           Workspace

          17:  LIW -- INTEGER                                         Input
               On entry: the length of the array IW. Constraint:
               LIW>=MNP*(2N+1)+N.

          18:  IFAIL -- INTEGER                                Input/Output
               For this routine, the normal use of IFAIL is extended to
               control the printing of error and warning messages as well
               as specifying hard or soft failure (see the Essential
               Introduction).

               Before entry, IFAIL must be set to a value with the decimal
               expansion cba, where each of the decimal digits c, b and a
               must have a value of 0 or 1.
               a=0   specifies hard failure, otherwise soft failure;

               b=0   suppresses error messages, otherwise error messages
                     will be printed (see Section 6);

               c=0   suppresses warning messages, otherwise warning
                     messages will be printed (see Section 6).
               The recommended value for inexperienced users is 110 (i.e.,
               hard failure with all messages printed).

               Unless the routine detects an error (see Section 6), IFAIL
               contains 0 on exit.

          6. Error Indicators and Warnings

          Errors detected by the routine:

          For each error, an explanatory error message is output on the
          current error message unit (as defined by X04AAF), unless
          suppressed by the value of IFAIL on entry.

          IFAIL= 1
               One or more of the parameters N, TOL, NP, MNP, LW or LIW is
               incorrectly set, B <= A or the condition (4) on X is not
               satisfied.

          IFAIL= 2
               There are three possible reasons for this error exit to be
               taken:
                 (i)  one of the matrices C or D is identically zero (that
                      is the problem is of initial value and not boundary
                      value type). In this case, IW(1) = 0 on exit;

                 (ii) a row of C and the corresponding row of D are
                      identically zero (that is the boundary conditions are
                      rank deficient). In this case, on exit IW(1) contains
                      the index of the first such row encountered; and

                 (iii more than n of the columns of the n by 2n matrix [C,D
                 )    ] are identically zero (that is the boundary
                      conditions are rank deficient). In this case, on exit
                      IW(1) contains minus the number of non-identically
                      zero columns.

          IFAIL= 3
               The routine has failed to find a solution to the specified
               accuracy. There are a variety of possible reasons including:
                 (i)   the boundary conditions are rank deficient, which may
                       be indicated by the message that the Jacobian is
                       singular. However this is an unlikely explanation for
                       the error exit as all rank deficient boundary
                       conditions should lead instead to error exits with
                       either IFAIL = 2 or IFAIL = 5; see also (iv) below;

                 (ii)  not enough mesh points are permitted in order to
                       attain the required accuracy. This is indicated by NP
                       = MNP on return from a call to D02GBF. This
                       difficulty may be aggravated by a poor initial choice
                       of mesh points;

                 (iii) the accuracy requested cannot be attained on the
                       computer being used; and

                 (iv)  an unlikely combination of values of F(x) has led to
                       a singular Jacobian. The error should not persist if
                       more mesh points are allowed.

          IFAIL= 4
               A serious error has occurred in a call to D02GBF. Check all
               array subscripts and subroutine parameter lists in calls to
               D02GBF. Seek expert help.

          IFAIL= 5
               There are two possible reasons for this error exit which
               occurs when checking the rank of the boundary conditions by
               reduction to a row echelon form:
                    (i) at least one row of the n by 2n matrix [C,D] is a
                    linear combination of the other rows and hence the
                    boundary conditions are rank deficient. The index of
                    the first such row encountered is given by IW(1) on
                    exit; and

                    (ii) as (i) but the rank deficiency implied by this
                    error exit has only been determined up to a numerical
                    tolerance. Minus the index of the first such row
                    encountered is given by IW(1) on exit.
               In case (ii) above there is some doubt as to the rank
               deficiency of the boundary conditions. However even if the
               boundary conditions are not rank deficient they are not
               posed in a suitable form for use with this routine.

               For example, if
                                                          ((gamma) )
                        (1 0        )     (1 0)           (       1)
                      C=(1 (epsilon)) , D=(1 0) , (gamma)=((gamma) )
                                                          (       2)
               and (epsilon) is small enough, this error exit is likely to
               be taken. A better form for the boundary conditions in this
               case would be
                                           ((gamma)                       )
                                           (       1                      )
                 (1 0)     (1 0)           (         -1                   )
               C=(0 1) , D=(0 0) , (gamma)=((epsilon)  ((gamma) -(gamma) ))
                                           (                   2        1 )

          7. Accuracy

          The solution returned by the routine will be accurate to the
          user's tolerance as defined by the relation (3) except in extreme
          circumstances. If too many points are specified in the initial
          mesh, the solution may be more accurate than requested and the
          error may not be approximately equidistributed.

          8. Further Comments

          The time taken by the routine depends on the difficulty of the
          problem, the number of mesh points (and meshes) used and the
          number of deferred corrections.

          The user is strongly recommended to set IFAIL to obtain self-
          explanatory error messages, and also monitoring information about
          the course of the computation. The user may select the channel
          numbers on which this output is to appear by calls of X04AAF (for
          error messages) or X04ABF (for monitoring information) - see
          Section 9 for an example. Otherwise the default channel numbers
          will be used, as specified in the implementation document.

          In the case where the user wishes to solve a sequence of similar
          problems, the use of the final mesh from one case is strongly
          recommended as the initial mesh for the next.

          9. Example

          We solve the problem (written as a first order system)

                                  (epsilon)y''+y'=0

          with boundary conditions

                                   y(0)=0,  y(1)=1

                                    -1                 -2
          for the cases (epsilon)=10   and (epsilon)=10   using the default
          initial mesh in the first case, and the final mesh of the first
          case as initial mesh for the second (more difficult) case. We
          give the solution and the error at each mesh point to illustrate
          the accuracy of the method given the accuracy request TOL=1.0E-3.

          Note the call to X04ABF prior to the call to D02GBF.

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02kef}{NAG On-line Documentation: d02kef}
\beginscroll
\begin{verbatim}



D02KEF(3NAG)                 D02KEF                  D02KEF(3NAG)



     D02 -- Ordinary Differential Equations                     D02KEF
             D02KEF -- NAG Foundation Library Routine Document

     Note: Before using this routine, please read the Users' Note for
     your implementation to check implementation-dependent details.
     The symbol (*) after a NAG routine name denotes a routine that is
     not included in the Foundation Library.

     1. Purpose

     D02KEF finds a specified eigenvalue of a regular singular second-
     order Sturm-Liouville system on a finite or infinite range, using
     a Pruefer transformation and a shooting method. It also reports
     values of the eigenfunction and its derivatives. Provision is
     made for discontinuities in the coefficient functions or their
     derivatives.

     2. Specification

            SUBROUTINE D02KEF (XPOINT, M, MATCH, COEFFN, BDYVAL, K,
           1                   TOL, ELAM, DELAM, HMAX, MAXIT, MAXFUN,
           2                   MONIT, REPORT, IFAIL)
            INTEGER          M, MATCH, K, MAXIT, MAXFUN, IFAIL
            DOUBLE PRECISION XPOINT(M), TOL, ELAM, DELAM, HMAX(2,M)
            EXTERNAL         COEFFN, BDYVAL, MONIT, REPORT

     3. Description

     D02KEF has essentially the same purpose as D02KDF(*) with minor
     modifications to enable values of the eigenfunction to be
     obtained after convergence to the eigenvalue has been achieved.

                                           ~~~~~~~~
     It first finds a specified eigenvalue (lambda) of a Sturm-
     Liouville system defined by a self-adjoint differential equation
     of the second-order

                     (p(x)y')'+q(x;(lambda))y=0, a<x<b

     together with the appropriate boundary conditions at the two
     (finite or infinite) end-points a and b. The functions p and q,
     which are real-valued, must be defined by a subroutine COEFFN.
     The boundary conditions must be defined by a subroutine BDYVAL,
     and, in the case of a singularity at a or b, take the form of an
     asymptotic formula for the solution near the relevant end-point.

                                      ~~~~~~~~
     When the final estimate (lambda)=(lambda) of the eigenvalue has
     been found, the routine integrates the differential equation once
     more with that value of (lambda), and with initial conditions
     chosen so that the integral

                       b
                       /    2    ddq
                    S= |y(x)  ----------(x;(lambda))dx
                       /      dd(lambda)
                       a

     is approximately one. When q(x;(lambda)) is of the form
     (lambda)w(x)+q(x), which is the most common case, S represents
     the square of the norm of y induced by the inner product

                                 b
                                 /
                          <f,g>= |f(x)g(x)w(x)dx,
                                 /
                                 a

     with respect to which the eigenfunctions are mutually orthogonal.
     This normalisation of y is only approximate, but experience shows
     that S generally differs from unity by only one or two per cent.

     During this final integration the REPORT routine supplied by the
     user is called at each integration mesh point x. Sufficient
     information is returned to permit the user to compute y(x) and
     y'(x) for printing or plotting. For reasons described in Section
     8.2, D02KEF passes across to REPORT, not y and y', but the Pru
     efer variables (beta), (phi) and (rho) on which the numerical
     method is based. Their relationship to y and y' is given by the
     equations


              ______   ( (rho))   ( (phi))
     p(x)y'=\/(beta)exp( -----)cos( -----);
                       (   2  )   (   2  )

                1       ( (rho))   ( (phi))
          y= --------exp( -----)sin( -----).
               ______   (   2  )   (   2  )
             \/(beta)

     For the theoretical basis of the numerical method to be valid,
     the following conditions should hold on the coefficient
     functions:

     (a)   p(x) must be non-zero and of one sign throughout the
           interval (a,b); and,

               ddq
     (b)    ---------- must be of one sign throughout (a,b) for all
            dd(lambda)
           relevant values of (lambda), and must not be identically
           zero as x varies, for any (lambda).

     Points of discontinuity in the functions p and q or their
     derivatives are allowed, and should be included as 'break-points'
     in the array XPOINT.

     A good account of the theory of Sturm-Liouville systems, with
     some description of Pruefer transformations, is given in Birkhoff
     and Rota [4], Chapter X. An introduction for the user of Pruefer
     transformations for the numerical solution of eigenvalue problems
     arising from physics and chemistry is Bailey [2].

     The scaled Pruefer method is fairly recent, and is described in a
     short note by Pryce [6] and in some detail in the technical
     report [5].

     4. References

     [1]   Abramowitz M and Stegun I A (1968) Handbook of Mathematical
           Functions. Dover Publications.

     [2]   Bailey P B (1966) Sturm-Liouville Eigenvalues via a Phase
           Function. SIAM J. Appl. Math . 14 242--249.

     [3]   Banks D O and Kurowski I (1968) Computation of Eigenvalues
           of Singular Sturm-Liouville Systems. Math. Computing. 22
           304--310.

     [4]   Birkhoff G and Rota G C (1962) Ordinary Differential
           Equations. Ginn & Co., Boston and New York.

     [5]   Pryce J D (1981) Two codes for Sturm-Liouville problems.
           Technical Report CS-81-01. Dept of Computer Science, Bristol
           University .

     [6]   Pryce J D and Hargrave B A (1977) The Scale Pruefer Method
           for one-parameter and multi-parameter eigenvalue problems in
           ODEs. Inst. Math. Appl., Numerical Analysis Newsletter. 1(3)

     5. Parameters

      1:  XPOINT(M) -- DOUBLE PRECISION array                    Input
          On entry: the points where the boundary conditions computed
          by BDYVAL are to be imposed, and also any break-points,
          i.e., XPOINT(1) to XPOINT(m) must contain values x ,...,x
                                                            1      m
          such that
                             x <=x <x <...<x   <=x
                              1   2  3      m-1   m
          with the following meanings:
          (a)   x  and x  are the left and right end-points, a and b,
                 1      m
                of the domain of definition of the Sturm-Liouville
                system if these are finite. If either a or b is
                infinite, the corresponding value x  or x  may be a
                                                   1     m
                more-or-less arbitrarily 'large' number of appropriate
                sign.

          (b)   x  and x    are the Boundary Matching Points (BMP's),
                 2      m-1
                that is the points at which the left and right
                boundary conditions computed in BDYVAL are imposed.

                If the left-hand end-point is a regular point then the
                user should set x =x (=a), while if it is a singular
                                 2  1
                point the user must set x >x . Similarly x   =x (=b)
                                         2  1             m-1  m
                if the right-hand end-point is regular, and x   <x  if
                                                             m-1  m
                it is singular.

          (c)   The remaining m-4 points x ,...,x   , if any, define
                                          3      m-2
                `break-points' which divide the interval [x ,x   ]
                                                           2  m-1
                into m-3 sub-intervals
                           i =[x ,x ],...,i   =[x   ,x   ]
                            1   2  3       m-3   m-2  m-1
                Numerical integration of the differential equation is
                stopped and restarted at each break-point. In simple
                cases no break-points are needed. However if p(x) or
                q(x;(lambda)) are given by different formulae in
                different parts of the range, then integration is more
                efficient if the range is broken up by break-points in
                the appropriate way. Similarly points where any jumps
                occur in p(x) or q(x;(lambda)), or in their
                derivatives up to the fifth order, should appear as
                break-points.
          Constraint: X(1) <= X(2) < ... < X(M-1) <= X(M).

      2:  M -- INTEGER                                           Input
          On entry: the number of points in the array XPOINT.
          Constraint: M >= 4.

      3:  MATCH -- INTEGER                                Input/Output
          On entry: MATCH must be set to the index of the 'break-point
          ' to be used as the matching point (see Section 8.3). If
          MATCH is set to a value outside the range [2,m-1] then a
          default value is taken, corresponding to the break-point
          nearest the centre of the interval [XPOINT(2),XPOINT(m-1)].
          On exit: the index of the break-point actually used as the
          matching point.

      4:  COEFFN -- SUBROUTINE, supplied by the user.
                                               External Procedure
          COEFFN must compute the values of the coefficient functions
          p(x) and q(x;(lambda)) for given values of x and (lambda).
          Section 3 states conditions which p and q must satisfy.

          Its specification is:

                 SUBROUTINE COEFFN (P, Q, DQDL, X, ELAM, JINT)
                 DOUBLE PRECISION P, Q, DQDL, X, ELAM
                 INTEGER          JINT

           1:  P -- DOUBLE PRECISION                            Output
               On exit: the value of p(x) for the current value of x.

           2:  Q -- DOUBLE PRECISION                            Output
               On exit: the value of q(x;(lambda)) for the current
               value of x and the current trial value of (lambda).

           3:  DQDL -- DOUBLE PRECISION                         Output
                                         ddq
               On exit: the value of  ----------(x;(lambda)) for the
                                      dd(lambda)
               current value of x and the current trial value of
               (lambda). However DQDL is only used in error estimation
               and an approximation (say to within 20%) will suffice.

           4:  X -- DOUBLE PRECISION                             Input
               On entry: the current value of x.

           5:  ELAM -- DOUBLE PRECISION                          Input
               On entry: the current trial value of the eigenvalue
               parameter (lambda).

           6:  JINT -- INTEGER                                   Input
               On entry: the index j of the sub-interval i  (see
                                                          j
               specification of XPOINT) in which x lies.
          See Section 8.4 and Section 9 for examples.

          COEFFN must be declared as EXTERNAL in the (sub)program
          from which D02KEF is called. Parameters denoted as
          Input must not be changed by this procedure.

      5:  BDYVAL -- SUBROUTINE, supplied by the user.
                                               External Procedure
          BDYVAL must define the boundary conditions. For each end-
          point, BDYVAL must return (in YL or YR) values of y(x) and
          p(x)y'(x) which are consistent with the boundary conditions
          at the end-points; only the ratio of the values matters.
          Here x is a given point (XL or XR) equal to, or close to,
          the end-point.

          For a regular end-point (a, say), x=a; and a boundary
          condition of the form
                                c y(a)+c y'(a)=0
                                 1      2
          can be handled by returning constant values in YL, e.g.
          YL(1)=c  and YL(2)=-c p(a).
                 2             1

          For a singular end-point however, YL(1) and YL(2) will in
          general be functions of XL and ELAM, and YR(1) and YR(2)
          functions of XR and ELAM, usually derived analytically from
          a power-series or asymptotic expansion. Examples are given
          in Section 8.5 Section 9.

          Its specification is:

                 SUBROUTINE BDYVAL (XL, XR, ELAM, YL, YR))
                 DOUBLE PRECISION XL, XR, ELAM, YL(3), YR(3)

           1:  XL -- DOUBLE PRECISION                            Input
               On entry: if a is a regular end-point of the system (so
               that a=x =x ), then XL contains a. If a is a singular
                       1  2
               point (so that a<=x <x ), then XL contains a point x
                                  1  2
               such that x <x<=x ).
                          1     2

           2:  XR -- DOUBLE PRECISION                            Input
               On entry: if b is a regular end-point of the system (so
               that x   =x =b), then XR contains b. If b is a singular
                     m-1  m
               point (so that x   <x <=b), then XR contains a point x
                               m-1  m
               such that x   <=x<x .
                          m-1     m

           3:  ELAM -- DOUBLE PRECISION                          Input
               On entry: the current trial value of (lambda).

           4:  YL(3) -- DOUBLE PRECISION array                  Output
               On exit: YL(1) and YL(2) should contain values of y(x)
               and p(x)y'(x) respectively (not both zero) which are
               consistent with the boundary condition at the left-hand
               end-point, given by x = XL. YL(3) should not be set.

           5:  YR(3) -- DOUBLE PRECISION array                  Output
               On exit: YR(1) and YR(2) should contain values of y(x)
               and p(x)y'(x) respectively (not both zero) which are
               consistent with the boundary condition at the right-
               hand end-point, given by x = XR. YR(3) should not be
               set.

          BDYVAL must be declared as EXTERNAL in the (sub)program
          from which D02KEF is called. Parameters denoted as
          Input must not be changed by this procedure.

      6:  K -- INTEGER                                           Input
          On entry: the index k of the required eigenvalue when the
          eigenvalues are ordered
          (lambda) <(lambda) <(lambda) <...<(lambda) <....
                  0         1         2             k
          Constraint: K >= 0.

      7:  TOL -- DOUBLE PRECISION                                Input
          On entry: the tolerance parameter which determines the
          accuracy of the computed eigenvalue. The error estimate held
          in DELAM on exit satisfies the mixed absolute/relative error
          test
                        DELAM<=TOL*max(1.0,|ELAM|)                 (*)
          where ELAM is the final estimate of the eigenvalue. DELAM is
          usually somewhat smaller than the right-hand side of (*) but
          not several orders of magnitude smaller. Constraint: TOL >
          0.0.

      8:  ELAM -- DOUBLE PRECISION                        Input/Output
                                                          ~~~~~~~~
          On entry: an initial estimate of the eigenvalue (lambda).
          On exit: the final computed estimate, whether or not an
          error occurred.

      9:  DELAM -- DOUBLE PRECISION                       Input/Output
          On entry: an indication of the scale of the problem in the
          (lambda)-direction. DELAM holds the initial 'search step'
          (positive or negative). Its value is not critical but the
          first two trial evaluations are made at ELAM and ELAM +
          DELAM, so the routine will work most efficiently if the
          eigenvalue lies between these values. A reasonable choice
          (if a closer bound is not known) is half the distance
          between adjacent eigenvalues in the neighbourhood of the one
          sought. In practice, there will often be a problem, similar
          to the one in hand but with known eigenvalues, which will
          help one to choose initial values for ELAM and DELAM.

          If DELAM = 0.0 on entry, it is given the default value of
          0.25*max(1.0,|ELAM|). On exit: with IFAIL = 0, DELAM holds
          an estimate of the absolute error in the computed
                               ~~~~~~~~
          eigenvalue, that is |(lambda)-ELAM|~=DELAM. (In Section 8.2
          we discuss the assumptions under which this is true.) The
          true error is rarely more than twice, or less than a tenth,
          of the estimated error.

          With IFAIL /= 0, DELAM may hold an estimate of the error, or
          its initial value, depending on the value of IFAIL. See
          Section 6 for further details.

     10:  HMAX(2,M) -- DOUBLE PRECISION array             Input/Output
          On entry: HMAX(1,j) a maximum step size to be used by the
          differential equation code in the jth sub-interval i  (as
                                                              j
          described in the specification of parameter XPOINT), for
          j=1,2,...,m-3. If it is zero the routine generates a maximum
          step size internally.

          It is recommended that HMAX(1,j) be set to zero unless the
          coefficient functions p and q have features (such as a
          narrow peak) within the jth sub-interval that could be '
          missed' if a long step were taken. In such a case HMAX(1,j)
          should be set to about half the distance over which the
          feature should be observed. Too small a value will increase
          the computing time for the routine. See Section 8 for
          further suggestions.

          The rest of the array is used as workspace. On exit: HMAX(
          1,m-1) and HMAX(1,m) contain the sensitivity coefficients
          (sigma) ,(sigma) , described in Section 8.6. Other entries
                 l        r
          contain diagnostic output in case of an error (see Section 6
          ).

     11:  MAXIT -- INTEGER                                Input/Output
          On entry: a bound on n , the number of root-finding
                                r
          iterations allowed, that is the number of trial values of
          (lambda) that are used. If MAXIT <= 0, no such bound is
          assumed. (See also under MAXFUN.) Suggested value: MAXIT =
          0. On exit: MAXIT will have been decreased by the number of
          iterations actually performed, whether or not it was
          positive on entry.

     12:  MAXFUN -- INTEGER                                      Input
          On entry: a bound on n , the number of calls to COEFFN made
                                f
          in any one root-finding iteration. If MAXFUN <= 0, no such
          bound is assumed. Suggested value: MAXFUN = 0.

          MAXFUN and MAXIT may be used to limit the computational cost
          of a call to D02KEF, which is roughly proportional to n *n .
                                                                 r  f

     13:  MONIT -- SUBROUTINE, supplied by the user.
                                               External Procedure
          MONIT is called by D02KEF at the end of each root-finding
          iteration and allows the user to monitor the course of the
          computation by printing out the parameters (see Section 8
          for an example).
          If no monitoring is required, the dummy subroutine D02KAY
          may be used. (D02KAY is included in the NAG Foundation
          Library).

          Its specification is:

                 SUBROUTINE MONIT (MAXIT, IFLAG, ELAM, FINFO)
                 INTEGER          MAXIT, IFLAG
                 DOUBLE PRECISION ELAM, FINFO(15)

           1:  MAXIT -- INTEGER                                  Input
               On entry: the current value of the parameter MAXIT of
               D02KEF; this is decreased by one at each iteration.

           2:  IFLAG -- INTEGER                                  Input
               On entry: IFLAG describes what phase the computation is
               in, as follows:
               IFLAG < 0
                     an error occurred in the computation of the '
                     miss-distance' at this iteration;

                     an error exit from D02KEF with IFAIL =-IFLAG will
                     follow.

               IFLAG = 1
                     the routine is trying to bracket the eigenvalue
                     ~~~~~~~~
                     (lambda).

               IFLAG = 2
                     the routine is converging to the eigenvalue
                     ~~~~~~~~
                     (lambda) (having already bracketed it).

           3:  ELAM -- DOUBLE PRECISION                          Input
               On entry: the current trial value of (lambda).

           4:  FINFO(15) -- DOUBLE PRECISION array               Input
               On entry: information about the behaviour of the
               shooting method, and diagnostic information in the case
               of errors. It should not normally be printed in full if
               no error has occurred (that is, if IFLAG > 0), though
               the first few components may be of interest to the
               user. In case of an error (IFLAG < 0) all the
               components of FINFO should be printed. The contents of
               FINFO are as follows:

               FINFO(1): the current value of the 'miss-distance' or '
               residual' function f((lambda)) on which the shooting
               method is based. FINFO(1) is set to zero if IFLAG < 0.

               FINFO(2): an estimate of the quantity dd(lambda)
               defined as follows. Consider the perturbation in the
               miss-distance f((lambda)) that would result if the
               local error, in the solution of the differential
               equation, were always positive and equal to its maximum
               permitted value. Then dd(lambda) is the perturbation in
               (lambda) that would have the same effect on f((lambda))
               . Thus, at the zero of f((lambda)),|dd(lambda)| is an
               approximate bound on the perturbation of the zero (that
               is the eigenvalue) caused by errors in numerical
               solution. If dd(lambda) is very large then it is
               possible that there has been a programming error in
               COEFFN such that q is independent of (lambda). If this
               is the case, an error exit with IFAIL = 5 should
               follow. FINFO(2) is set to zero if IFLAG < 0.

               FINFO(3): the number of internal iterations, using the
               same value of (lambda) and tighter accuracy tolerances,
               needed to bring the accuracy (that is the value of
               dd(lambda)) to an acceptable value. Its value should
               normally be 1.0, and should almost never exceed 2.0.

               FINFO(4): the number of calls to COEFFN at this
               iteration.

               FINFO(5): the number of successful steps taken by the
               internal differential equation solver at this
               iteration. A step is successful if it is used to
               advance the integration (cf. COUT(8) in specification
               of D02PAF(*)).

               FINFO(6): the number of unsuccessful steps used by the
               internal integrator at this iteration (cf. COUT(9) in
               specification of D02PAF(*)).

               FINFO(7): the number of successful steps at the maximum
               step size taken by the internal integrator at this
               iteration (cf. COUT(3) in specification of D02PAF(*)).

               FINFO(8): is not used.

               FINFO(9) to FINFO(15): set to zero, unless IFLAG < 0 in
               which case they hold the following values describing
               the point of failure:

               FINFO(9): contains the index of the sub-interval where
               failure occurred, in the range 1 to m-3. In case of an
               error in BDYVAL, it is set to 0 or m-2 depending on
               whether the left or right boundary condition caused the
               error.

               FINFO(10): the value f the independent variable x, the
               point at which error occurred. In case of an error in
               BDYVAL, it is set to the value of XL or XR as
               appropriate (see the specification of BDYVAL).

               FINFO(11), FINFO(12), FINFO(13): the current values of
               the Pruefer dependent variables (beta), (phi) and (rho)
               respectively. These are set to zero in case of an error
               in BDYVAL.

               FINFO(14): the local-error tolerance being used by the
               internal integrator at the point of failure. This is
               set to zero in the case of an error in BDYVAL.

               FINFO(15): the last integration mesh point. This is set
               to zero in the case of an error in BDYVAL.
          MONIT must be declared as EXTERNAL in the (sub)program
          from which D02KEF is called. Parameters denoted as
          Input must not be changed by this procedure.

     14:  REPORT -- SUBROUTINE, supplied by the user.
                                               External Procedure
          This routine provides the means by which the user may
          compute the eigenfunction y(x) and its derivative at each
          integration mesh point x. (See Section 8 for an example).

          Its specification is:

                 SUBROUTINE REPORT (X, V, JINT)
                 INTEGER          JINT
                 DOUBLE PRECISION X, V(3)

           1:  X -- DOUBLE PRECISION                             Input
               On entry: the current value of the independent variable
               x. See Section 8.3 for the order in which values of x
               are supplied.

           2:  V(3) -- DOUBLE PRECISION array                    Input
               On entry: V(1), V(2), V(3) hold the current values of
               the Pruefer variables (beta), (phi), (rho)
               respectively.

           3:  JINT -- INTEGER                                   Input
               On entry: JINT indicates the sub-interval between
               break-points in which X lies exactly as for the routine
               COEFFN, except that at the extreme left end-point (when
               x = XPOINT(2)) JINT is set to 0 and at the extreme
               right end-point (when x=x =XPOINT(m-1)) JINT is set to
                                        r
               m-2.
          REPORT must be declared as EXTERNAL in the (sub)program
          from which D02KEF is called. Parameters denoted as
          Input must not be changed by this procedure.

     15:  IFAIL -- INTEGER                                Input/Output
          On entry: IFAIL must be set to 0, -1 or 1. For users not
          familiar with this parameter (described in the Essential
          Introduction) the recommended value is 0.

          On exit: IFAIL = 0 unless the routine detects an error (see
          Section 6).

     6. Error Indicators and Warnings

     Errors detected by the routine:

     IFAIL= 1
          A parameter error. All parameters (except IFAIL) are left
          unchanged. The reason for the error is shown by the value of
          HMAX(2,1) as follows:
          HMAX(2,1) = 1: M < 4;

          HMAX(2,1) = 2: K < 0;

          HMAX(2,1) = 3: TOL <= 0.0;

          HMAX(2,1) = 4: XPOINT(1) to XPOINT(m) are not in ascending
                         order.

                         HMAX(2,2) gives the position i in XPOINT
                         where this was detected.

     IFAIL= 2
          At some call to BDYVAL, invalid values were returned, that
          is, either YL(1) = YL(2) = 0.0, or YR(1) = YR(2) = 0.0 (a
          programming error in BDYVAL). See the last call of MONIT for
          details.

          This error exit will also occur if p(x) is zero at the point
          where the boundary condition is imposed. Probably BDYVAL was
          called with XL equal to a singular end-point a or with XR
          equal to a singular end-point b.

     IFAIL= 3
          At some point between XL and XR the value of p(x) computed
          by COEFFN became zero or changed sign. See the last call of
          MONIT for details.

     IFAIL= 4
          MAXIT > 0 on entry, and after MAXIT iterations the
          eigenvalue had not been found to the required accuracy.

     IFAIL= 5
          The 'bracketing' phase (with parameter IFLAG of MONIT equal
          to 1) failed to bracket the eigenvalue within ten
          iterations. This is caused by an error in formulating the
          problem (for example, q is independent of (lambda)), or by
          very poor initial estimates of ELAM, DELAM.

          On exit ELAM and ELAM + DELAM give the end-points of the
          interval within which no eigenvalue was located by the
          routine.

     IFAIL= 6
          MAXFUN > 0 on entry, and the last iteration was terminated
          because more than MAXFUN calls to COEFFN were used. See the
          last call of MONIT for details.

     IFAIL= 7
          To obtain the desired accuracy the local error tolerance was
          set so small at the start of some sub-interval that the
          differential equation solver could not choose an initial
          step size large enough to make significant progress. See the
          last call of MONIT for diagnostics.

     IFAIL= 8
          At some point inside a sub-interval the step size in the
          differenital equation solver was reduced to a value too
          small to make significant progress (for the same reasons as
          with IFAIL = 7). This could be due to pathological behaviour
          of p(x) and q(x;(lambda)) or to an unreasonable accuracy
          requirement or to the current value of (lambda) making the
          equations 'stiff'. See the last call of MONIT for details.

     IFAIL= 9
          TOL is too small for the problem being solved and the
          machine precision is being used. The final value of ELAM
          should be a very good approximation to the eigenvalue.

     IFAIL= 10
          C05AZF(*), called by D02KEF, has terminated with the error
          exit corresponding to a pole of the residual function
          f((lambda)). This error exit should not occur, but if it
          does, try solving the problem again with a smaller value for
          TOL.

     IFAIL= 11
          A serious error has occurred in an internal call to D02KDY.
          Check all subroutine calls and array dimensions. Seek expert
          help.

     IFAIL= 12
          A serious error has occurred in an internal call to
          C05AZF(*). Check all subroutine calls and array dimensions.
          Seek expert help.

          HMAX(2,1) holds the failure exit number from the routine
          where the failure occurred. In the case of a failure in
          C05AZF(*), HMAX(2,2) holds the value of parameter IND of
          C05AZF(*).

     Note: error exits with IFAIL = 2, 3, 6, 7, 8, 11 are caused by
     being unable to set up or solve the differential equation at some
     iteration, and will be immediately preceded by a call of MONIT
     giving diagnostic information. For other errors, diagnostic
     information is contained in HMAX(2,j), for j=1,2,...,m, where
     appropriate.

     7. Accuracy

     See the discussion in Section 8.2.

     8. Further Comments

     8.1. Timing

     The time taken by the routine depends on the complexity of the
     coefficient functions, whether they or their derivatives are
     rapidly changing, the tolerance demanded, and how many iterations
     are needed to obtain convergence. The amount of work per
     iteration is roughly doubled when TOL is divided by 16. To make
     the most economical use of the routine, one should try to obtain
     good initial values for ELAM and DELAM, and, where appropriate,
     good asymptotic formulae. The boundary matching points should not
     be set unnecessarily close to singular points. The extra time
     needed to compute the eigenfunction is principally the cost of
     one additional integration once the eigenvalue has been found.

     8.2. General Description of the Algorithm

     A shooting method, for differential equation problems containing
     unknown parameters, relies on the construction of a 'miss-
     distance function', which for given trial values of the
     parameters measures how far the conditions of the problem are
     from being met. The problem is then reduced to one of finding the
     values of the parameters for which the miss-distance function is
     zero, that is to a root-finding process. Shooting methods differ
     mainly in how the miss-distance is defined.

     This routine defines a miss-distance f((lambda)) based on the
     rotation around the origin of the point P(x)=(p(x)y'(x),y(x)) in
     the Phase Plane as the solution proceeds from a to b. The
     boundary-conditions define the ray (i.e., two-sided line through
     the origin) on which p(x) should start, and the ray on which it
     should finish. The eigenvalue index k defines the total number of
     half-turns it should make. Numerical solution is actually done by
     matching point x=c. Then f((lambda)) is taken as the angle
     between the rays to the two resulting points P (c) and P (c). A
                                                   a         b
     relative scaling of the py' and y axes, based on the behaviour of
     the coefficient functions p and q, is used to improve the
     numerical behaviour.



              Please see figure in printed Reference Manual

     The resulting function f((lambda)) is monotonic over -
                                             ddq
     infty<(lambda)<infty, increasing if  ---------->0 and decreasing
                                          dd(lambda)
            ddq
     if  ----------<0, with a unique zero at the desired eigenvalue
         dd(lambda)
     ~~~~~~~~
     (lambda). The routine measures f((lambda)) in units of a half-
     turn. This means that as (lambda) increases, f((lambda)) varies
     by about 1 as each eigenvalue is passed. (This feature implies
     that the values of f((lambda)) at successive iterations -
     especially in the early stages of the iterative process - can be
     used with suitable extrapolation or interpolation to help the
     choice of initial estimates for eigenvalues near to the one
     currently being found.)

     The routine actually computes a value for f((lambda)) with
     errors, arising from the local errors of the differential
     equation code and from the asymptotic formulae provided by the
     user if singular points are involved. However, the error estimate
     output in DELAM is usually fairly realistic, in that the actual
            ~~~~~~~~
     error |(lambda)-ELAM| is within an order of magnitude of DELAM.

     We pass the values of (beta), (phi), (rho) across through REPORT
     rather than converting them to values of y, y' inside D02KEF, for
     the following reasons. First, there may be cases where auxiliary
     quantities can be more accurately computed from the Pruefer
     variables than from y and y'. Second, in singular problems on an
     infinite interval y and y' may underflow towards the end of the
     range, whereas the Pruefer variables remain well-behaved. Third,
     with high-order eigenvalues (and therefore highly oscillatory
     eigenfunctions) the eigenfunction may have a complete oscillation
     (or more than one oscillation) between two mesh points, so that
     values of y and y' at mesh points give a very poor representation
     of the curve. The probable behaviour of the Pruefer variables in
     this case is that (beta) and (rho) vary slowly whilst (phi)
     increases quickly: for all three Pruefer variables linear
     interpolation between the values at adjacent mesh points is
     probably sufficiently accurate to yield acceptable intermediate
     values of (beta), (phi), (rho) (and hence of y,y') for graphical
     purposes.

     Similar considerations apply to the exponentially decaying 'tails
     Here (phi) has approximately constant value whilst (rho)
     increases rapidly in the direction of integration, though the
     step length is generally fairly small over such a range.

     If the solution is output through REPORT at x-values which are
     too widely spaced, the step length can be controlled by choosing
     HMAX suitably, or, preferably, by reducing TOL. Both these
     choices will lead to more accurate eigenvalues and eigenfunctions
     but at some computational cost.

     8.3. The Position of the Shooting Matching Point c

     This point is always one of the values x  in array XPOINT. It may
                                             i
     be specified using the parameter MATCH. The default value is
     chosen to be the value of that x ,2<=i<=m-1, that lies closest to
                                     i
     the middle of the interval [x ,x   ]. If there is a tie, the
                                  2  m-1
     rightmost candidate is chosen. In particular if there are no
     break-points then c=x   (=x ) - that is the shooting is from left
                          m-1   3
     to right in this case. A break-point may be inserted purely to
     move c to an interior point of the interval, even though the form
     of the equations does not require it. This often speeds up
     convergence especially with singular problems.

     Note that the shooting method used by the code integrates first
     from the left-hand end x , then from the right-hand end x , to
                             l                                r
     meet at the matching point c in the middle. This will of course
     be reflected in printed or graphical output. The diagram shows a
     possible sequence of nine mesh points (tau)  through (tau)  in
                                                1              9
     the order in which they appear, assuming there are just two sub-
     intervals (so m=5).


                                Figure 1
              Please see figure in printed Reference Manual

     Since the shooting method usually fails to match up the two 'legs
     p(x)y' or both, at the matching point c. The code in fact 'shares
     large jump does not imply an inaccurate eigenvalue, but implies
     either

     (a)   a badly chosen matching point: if q(x;(lambda)) has a '
           humped' shape, c should be chosen near the maximum value of
           q, especially if q is negative at the ends of the interval.

     (b)   An inherently ill-conditioned problem, typically one where
           another eigenvalue is pathologically close to the one being
           sought. In this case it is extremely difficult to obtain an
           accurate eigenfunction.

     In Section 9 below, we find the 11th eigenvalue and corresponding
     eigenfunction of the equation

                                     2
                  y''+((lambda)-x-2/x )y=0  on  0<x<infty

     the boundary conditions being that y should remain bounded as x
     tends to 0 and x tends to infty. The coding of this problem is
     discussed in detail in Section 8.5.

     The choice of matching point c is open. If we choose c=30.0 as in
     the D02KDF(*) example program we find that the exponentially
     increasing component of the solution dominates and we get
     extremely inaccurate values for the eigenfunction (though the
     eigenvalue is determined accurately). The values of the
     eigenfunction calculated with c=30.0 are given schematically in
     Figure 2.


                                Figure 2
              Please see figure in printed Reference Manual

     If we choose c as the maximum of the hump in q(x;(lambda)) (see
     (a) above) we instead obtain the accurate results given in
     Figure 3.


                                Figure 3
              Please see figure in printed Reference Manual

     8.4. Examples of Coding the COEFFN Routine

     Coding COEFFN is straightforward except when break-points are
     needed. The examples below show:

     (a)   a simple case,

     (b)   a case in which discontinuities in the coefficient
           functions or their derivatives necessitate break-points,
           and

     (c)   a case where break-points together with the HMAX parameter
           are an efficient way to deal with a coefficient function
           that is well-behaved except over one short interval.

     Example A

     The modified Bessel equation

                                         2     2
                       x(xy')'+((lambda)x -(nu) )y=0

     Assuming the interval of solution does not contain the origin,
     dividing through by x, we have p(x)=x,
                                 2
     q(x;(lambda))=(lambda)x-(nu) /x. The code could be


           SUBROUTINE COEFFN(P,Q,DQDL,X,ELAM,JINT)
           P = X
           Q = ELAM*X + NU*NU/X
           DQDL = X
           RETURN
           END


     where NU (standing for (nu)) is a real variable that might be
     defined in a DATA statement, or might be in user-declared COMMON
     so that its value could be set in the main program.

     Example B

     The Schroedinger equation

                          y''+((lambda)+q(x))y=0

                { 2
                {x -10 (|x|<=4),
     where q(x)={6/|x| (|x|>4),

     over some interval 'approximating to (-infty,infty)', say [-20,
     20]. Here we need break-points at +- 4, forming three sub-
     intervals i =[-20,-4], i =[-4,4],i =[4,20]. The code could be
                1            2         3


           SUBROUTINE COEFFN(P,Q,DQDL,X,ELAM,JINT)
           IF (JINT.EQ.2) THEN
             Q = ELAM + X*X - 10.0E0
           ELSE
             Q = ELAM + 6.0E0/ABS(X)
           ENDIF
           P = 1.0E0
           DQDL = 1.0E0
           RETURN
           END


     The array XPOINT would contain the values x , -20.0, -4.0, +4.0,
                                                1
     +20.0, x  and m would be 6. The choice of appropriate values for
             6
     x  and x  depends on the form of the asymptotic formula computed
      1      6
     by BDYVAL and the technique is discussed in the next subsection.

     Example C

                                     2
                                -100x

               y''+(lambda)(1-2e      )y=0, over -10<=x<=10

     Here q(x;(lambda)) is nearly constant over the range except for a
     sharp inverted spike over approximately -0.1<=x<=0.1. There is a
     danger that the routine will build up to a large step size and '
     step over' the spike without noticing it. By using break-points -
     say at +- 0.5 - one can restrict the step size near the spike
     without impairing the efficiency elsewhere.

     The code for COEFFN could be


           SUBROUTINE COEFFN(P,Q,DQDL,X,ELAM,JINT)
           P = 1.0E0
           DQDL = 1.0E0 - 2.0E0*EXP(-100.0E0*X*X)
           Q = ELAM*DQDL
           RETURN
           END


     XPOINT might contain -0.0, -10.0, -0.5, 0.5, 10.0, 10.0 (assuming
     +- 10 are regular points) and m would be 6. HMAX(1,j), j=1,2,3
     might contain 0.0, 0.1 and 0.0.

     8.5. Examples of Boundary Conditions at Singular Points

     Quoting from Bailey [2] page 243: 'Usually... the differential
     equation has two essentially different types of solution near a
     singular point, and the boundary condition there merely serves to
     distinguish one kind from the other. This is the case in all the
     standard examples of mathematical physics.'

     In most cases the behaviour of the ratio p(x)y'/y near the point
     is quite different for the two types of solution. Essentially
     what the user provides through his BDYVAL routine is an
     approximation to this ratio, valid as x tends to the singular
     point (SP).

     The user must decide (a) how accurate to make this approximation
     or asymptotic formula, for example how many terms of a series to
     use, and (b) where to place the boundary matching point (BMP) at
     which the numerical solution of the differential equation takes
     over from the asymptotic formula. Taking the BMP closer to the SP
     will generally improve the accuracy of the asymptotic formula,
     but will make the computation more expensive as the Pruefer
     differential equations generally become progressively more ill-
     behaved as the SP is approached. The user is strongly recommended
     to experiment with placing the BMPs. In many singular problems
     quite crude asymptotic formulae will do. To help the user avoid
     needlessly accurate formulae, D02KEF outputs two 'sensitivity
     coefficients' (sigma) ,(sigma)  which estimate how much the
                          l        r
     errors at the BMP's affect the computed eigenvalue. They are
     described in detail below, see Section 8.6.

     Example of coding BDYVAL:

     The example below illustrates typical situations:

                      (            2 )
                  y''+((lambda)-x- --)y=0, for 0<x<infty
                      (             2)
                      (            x )

     the boundary conditions being that y should remain bounded as x
     tends to 0 and x tends to infty.

                                                             2
     At the end x=0 there is one solution that behaves like x  and
                                -1
     another that behaves like x  . For the first of these solutions
     p(x)y'/y is asymptotically 2/x while for the second it is
     asymptotically -1/x. Thus the desired ratio is specified by
     setting

                          YL(1)=x and YL(2)=2.0.

     At the end x=infty the equation behaves like Airy's equation
     shifted through (lambda), i.e., like y''-ty=0 where t=x-(lambda),
     so again there are two types of solution. The solution we require
     behaves as

                                (  2 3/2) 4 _
                             exp(- -t   )/\/t
                                (  3    )

     and the other as

                                (  2 3/2) 4 _
                             exp(+ -t   )/\/t
                                (  3    )

                                                _
     once, the desired solution has p(x)y'/y~-\/t so that we could set
                              __________
     YR(1) = 1.0 and YR(2)=-\/x-(lambda). The complete subroutine
     might thus be


           SUBROUTINE BDYVAL(XL,XR,ELAM,YL,YR)
           real XL, XR, ELAM, YL(3), YR(3)
           YL(1) = XL
           YL(2) = 2.0E0
           YR(1) = 1.0E0
           YR(2) = -SQRT(XR - ELAM)
           RETURN
           END


     Clearly for this problem it is essential that any value given by
     D02KEF to XR is well to the right of the value of ELAM, so that
     the user must vary the right-hand BMP with the eigenvalue index k
                                k
     function Ai(x), so there is no problem in estimating ELAM.

     More accurate asymptotic formulae are easily found - near x=0 by
     the standard Frobenius method, and near x=infty by using standard
     asymptotics for Ai(x), Ai'(x) (see [1], p. 448). For example, by
     the Frobenius method the solution near x=0 has the expansion

                              2           2
                           y=x (c +c x+c x +...)
                                 0  1   2

     with

                                                c   -(lambda)c
                        (lambda)     1           n-3          n-2
         c =1,c =0,c =- --------,c = --,...,c = -----------------
          0    1    2      10     3  18      n       n(n+3)

     This yields

                                   2         2
                                2- -(lambda)x +...
                       p(x)y'      5
                       ------= --------------------.
                         y      (   (lambda) 2    )
                               x(1- --------x +...)
                                (      10         )

     8.6. The Sensitivity Parameters (sigma)  and (sigma)
                                            l            r

     The sensitivity parameters (sigma) ,(sigma)  (held in HMAX(1,m-1)
                                       l        r
     and HMAX(1,m) on output) estimate the effect of errors in the
     boundary conditions. For sufficiently small errors (Delta)y,
     (Delta)py' in y and py' respectively, the relations

           (Delta)(lambda)~=(y.(Delta)py'-py'.(Delta)y) (sigma)
                                                       l       l

           (Delta)(lambda)~=(y.(Delta)py'-py'.(Delta)y) (sigma)
                                                       r       r

     are satisfied where the subscripts l, r denote errors committed
     at left- and right-hand BMP's respectively, and (Delta)(lambda)
     denotes the consequent error in the computed eigenvalue.

     8.7. Missed Zeros

     This is a pitfall to beware of at a singular point. If the BMP is
     chosen so far from the SP that a zero of the desired
     eigenfunction lies in between them, then the routine will fail to
     number of zeros of its eigenfunction, the result will be that:

     (a)   The wrong eigenvalue will be computed for the given index k
           - in fact some (lambda)     will be found where k'>=1.
                                  k+k'

     (b)   The same index k can cause convergence to any of several
           eigenvalues depending on the initial values of ELAM and
           DELAM.

     It is up to the user to take suitable precautions - for instance
     by varying the position of the BMP's in the light of his
     knowledge of the asymptotic behaviour of the eigenfunction at
     different eigenvalues.

     9. Example

     To find the 11th eigenvalue and eigenfunction of the example of
     Section 8.5, using the simple asymptotic formulae for the
     boundary conditions.

     Comparison of the results from this example program with the
     corresponding results from D02KDF(*) example program shows that
     similar output is produced from the routine MONIT, followed by
     the eigenfunction values from REPORT, and then a further line of
     information from MONIT (corresponding to the integration to find
     the eigenfunction). Final information is printed within the
     example program exactly as with D02KDF(*).

                                                   3 _
     Note the discrepancy at the matching point c(=\/4 , the maximum
     of q(x;(lambda)), in this case) between the solutions obtained by
     integrations from left and right end-points.

     The example program is not reproduced here. The source code for
     all example programs is distributed with the NAG Foundation
     Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd02raf}{NAG On-line Documentation: d02raf}
\beginscroll
\begin{verbatim}



     D02RAF(3NAG)      Foundation Library (12/10/92)      D02RAF(3NAG)



          D02 -- Ordinary Differential Equations                     D02RAF
                  D02RAF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D02RAF solves the two-point boundary-value problem with general
          boundary conditions for a system of ordinary differential
          equations, using a deferred correction technique and Newton
          iteration.

          2. Specification

                 SUBROUTINE D02RAF (N, MNP, NP, NUMBEG, NUMMIX, TOL, INIT,
                1                   X, Y, IY, ABT, FCN, G, IJAC, JACOBF,
                2                   JACOBG, DELEPS, JACEPS, JACGEP, WORK,
                3                   LWORK, IWORK, LIWORK, IFAIL)
                 INTEGER          N, MNP, NP, NUMBEG, NUMMIX, INIT, IY,
                1                 IJAC, LWORK, IWORK(LIWORK), LIWORK, IFAIL
                 DOUBLE PRECISION TOL, X(MNP), Y(IY,MNP), ABT(N), DELEPS,
                1                 WORK(LWORK)
                 EXTERNAL         FCN, G, JACOBF, JACOBG, JACEPS, JACGEP

          3. Description

          D02RAF solves a two-point boundary-value problem for a system of
          n ordinary differential equations in the interval (a,b) with b>a.
          The system is written in the form

                      y '=f (x,y ,y ,...,y ) , i=1,2,...,n              (1)
                       i   i    1  2      n

          and the derivatives f  are evaluated by a subroutine FCN supplied
                               i
          by the user. With the differential equations (1) must be given a
          system of n (nonlinear) boundary conditions

                            g (y(a),y(b))=0 , i=1,2,...,n
                             i

          where

                                                     T
                         y(x)=[y (x),y (x),...,y (x)] .                 (2)
                                1     2         n

          The functions g  are evaluated by a subroutine G supplied by the
                         i
          user. The solution is computed using a finite-difference
          technique with deferred correction allied to a Newton iteration
          to solve the finite-difference equations. The technique used is
          described fully in Pereyra [1].

          The user must supply an absolute error tolerance and may also
          supply an initial mesh for the finite-difference equations and an
          initial approximate solution (alternatively a default mesh and
          approximation are used). The approximate solution is corrected
          using Newton iteration and deferred correction. Then, additional
          points are added to the mesh and the solution is recomputed with
          the aim of making the error everywhere less than the user's
          tolerance and of approximately equidistributing the error on the
          final mesh. The solution is returned on this final mesh.

          If the solution is required at a few specific points then these
          should be included in the initial mesh. If, on the other hand,
          the solution is required at several specific points then the user
          should use the interpolation routines provided in Chapter E01 if
          these points do not themselves form a convenient mesh.

          The Newton iteration requires Jacobian matrices

                          ( ddf )  (  ddg   )     (  ddg   )
                          (    i)  (     i  )     (     i  )
                          ( ----), ( -------) and ( -------).
                          ( ddy )  ( ddy (a))     ( ddy (b))
                          (    j)  (    j   )     (    j   )

          These may be supplied by the user through subroutines JACOBF for
          ( ddf )
          (    i)
          ( ----) and JACOBG for the others. Alternatively the Jacobians
          ( ddy )
          (    j)
          may be calculated by numerical differentiation using the
          algorithm described in Curtis et al [2].

          For problems of the type (1) and (2) for which it is difficult to
          determine an initial approximation from which the Newton
          iteration will converge, a continuation facility is provided. The
          user must set up a family of problems

                y'=f(x,y,(epsilon)),   g(y(a),y(b),(epsilon))=0,        (3)

                                T
          where f=[f ,f ,...,f ]  etc, and where (epsilon) is a
                    1  2      n
          continuation parameter. The choice (epsilon)=0 must give a
          problem (3) which is easy to solve and (epsilon)=1 must define
          the problem whose solution is actually required. The routine
          solves a sequence of problems with (epsilon) values

                   0=(epsilon) <(epsilon) <...<(epsilon) =1.            (4)
                              1          2              p

          The number p and the values (epsilon)  are chosen by the routine
                                               i
          so that each problem can be solved using the solution of its
                                                                  ddf
          predecessor as a starting approximation. Jacobians  -----------
                                                              dd(epsilon)
                   ddg
          and  ----------- are required and they may be supplied by the
               dd(epsilon)
          user via routines JACEPS and JACGEP respectively or may be
          computed by numerical differentiation.

          4. References

          [1]   Pereyra V (1979) PASVA3: An Adaptive Finite-Difference
                Fortran Program for First Order Nonlinear, Ordinary Boundary
                Problems. Codes for Boundary Value Problems in Ordinary
                Differential Equations. Lecture Notes in Computer Science.
                (ed B Childs, M Scott, J W Daniel, E Denman and P Nelson) 76
                Springer-Verlag.

          [2]   Curtis A R, Powell M J D and Reid J K (1974) On the
                Estimation of Sparse Jacobian Matrices. J. Inst. Maths
                Applics. 13 117--119.

          5. Parameters

           1:  N -- INTEGER                                           Input
               On entry: the number of differential equations, n.
               Constraint: N > 0.

           2:  MNP -- INTEGER                                         Input
               On entry: MNP must be set to the maximum permitted number
               of points in the finite-difference mesh. If LWORK or LIWORK
               (see below) is too small then internally MNP will be
               replaced by the maximum permitted by these values. (A
               warning message will be output if on entry IFAIL is set to
               obtain monitoring information.) Constraint: MNP >= 32.

           3:  NP -- INTEGER                                   Input/Output
               On entry: NP must be set to the number of points to be used
               in the initial mesh. Constraint: 4 <= NP <= MNP. On exit:
               the number of points in the final mesh.

           4:  NUMBEG -- INTEGER                                      Input
               On entry: the number of left-hand boundary conditions (that
               is the number involving y(a) only). Constraint: 0 <= NUMBEG
               < N.

           5:  NUMMIX -- INTEGER                                      Input
               On entry: the number of coupled boundary conditions (that
               is the number involving both y(a) and y(b)). Constraint: 0
               <= NUMMIX <= N - NUMBEG.

           6:  TOL -- DOUBLE PRECISION                                Input
               On entry: a positive absolute error tolerance. If
                                    a=x <x <...<x  =b
                                       1  2      NP
               is the final mesh, z (x ) is the jth component of the
                                   j  i
               approximate solution at x , and y (x) is the jth component
                                        i       j
               of the true solution of (1) and (2), then, except in extreme
               circumstances, it is expected that
                 |z (x )-y (x )|<=TOL , i=1,2,...,NP ; j=1,2,...,n.     (5)
                   j  i   j  i
                Constraint: TOL > 0.0.

           7:  INIT -- INTEGER                                        Input
               On entry: indicates whether the user wishes to supply an
               initial mesh and approximate solution (INIT /= 0) or whether
               default values are to be used, (INIT = 0).

           8:  X(MNP) -- DOUBLE PRECISION array                Input/Output
               On entry: the user must set X(1) = a and X(NP) = b. If INIT
               = 0 on entry a default equispaced mesh will be used,
               otherwise the user must specify a mesh by setting X(i)=x ,
                                                                       i
               for i = 2,3,...NP-1. Constraints:
                      X(1) < X(NP), if INIT = 0,

                      X(1) < X(2) <... < X(NP), if INIT /= 0.
               On exit: X(1),X(2),...,X(NP) define the final mesh (with
               the returned value of NP) and X(1) = a and X(NP) = b.

           9:  Y(IY,MNP) -- DOUBLE PRECISION array             Input/Output
               On entry: if INIT = 0, then Y need not be set.

               If INIT /= 0, then the array Y must contain an initial
               approximation to the solution such that Y(j,i) contains an
               approximation to
                           y (x ) , i=1,2,...,NP ; j=1,2,...,n.
                            j  i
               On exit: the approximate solution z (x ) satisfying (5) on
                                                  j  i
               the final mesh, that is
                       Y(j,i)=z (x ) , i=1,2,...,NP ; j=1,2,...,n,
                               j  i
               where NP is the number of points in the final mesh. If an
               error has occurred then Y contains the latest approximation
               to the solution. The remaining columns of Y are not used.

          10:  IY -- INTEGER                                          Input
               On entry:
               the first dimension of the array Y as declared in the
               (sub)program from which D02RAF is called.
               Constraint: IY >= N.

          11:  ABT(N) -- DOUBLE PRECISION array                      Output
               On exit: ABT(i), for i=1,2,...,n, holds the largest
               estimated error (in magnitude) of the ith component of the
               solution over all mesh points.

          12:  FCN -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               FCN must evaluate the functions f  (i.e., the derivatives
                                                i
               y' ) at a general point x for a given value of (epsilon),
                 i
               the continuation parameter (see Section 3).

               Its specification is:

                      SUBROUTINE FCN (X, EPS, Y, F, N)
                      INTEGER          N
                      DOUBLE PRECISION X, EPS, Y(N), F(N)

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the argument x.

                2:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter,
                    (epsilon). This is 1 if continuation is not being used.

                3:  Y(N) -- DOUBLE PRECISION array                    Input
                    On entry: the value of the argument y , for
                                                         i
                    i=1,2,...,n.

                4:  F(N) -- DOUBLE PRECISION array                   Output
                    On exit: the values of f , for i=1,2,...,n.
                                            i

                5:  N -- INTEGER                                      Input
                    On entry: the number of equations.
               FCN must be declared as EXTERNAL in the (sub)program
               from which D02RAF is called. Parameters denoted as
               Input must not be changed by this procedure.

          13:  G -- SUBROUTINE, supplied by the user.
                                                    External Procedure
               G must evaluate the boundary conditions in equation (3) and
               place them in the array BC.

               Its specification is:

                      SUBROUTINE G (EPS, YA, YB, BC, N)
                      INTEGER          N
                      DOUBLE PRECISION EPS, YA(N), YB(N), BC(N)

                1:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter,
                    (epsilon). This is 1 if continuation is not being used.

                2:  YA(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value y (a), for i=1,2,...,n.
                                         i

                3:  YB(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value y (b), for i=1,2,...,n.
                                         i

                4:  BC(N) -- DOUBLE PRECISION array                  Output
                    On exit: the values g (y(a),y(b),(epsilon)), for
                                         i
                    i=1,2,...,n. These must be ordered as follows:
                      (i)   first, the conditions involving only y(a) (see
                            NUMBEG description above);

                      (ii)  next, the NUMMIX coupled conditions involving
                            both y(a) and y(b) (see NUMMIX description
                            above); and,

                      (iii) finally, the conditions involving only y(b) (N-
                            NUMBEG-NUMMIX).

                5:  N -- INTEGER                                      Input
                    On entry: the number of equations, n.
               G must be declared as EXTERNAL in the (sub)program from
               which D02RAF is called. Parameters denoted as Input
               must not be changed by this procedure.

          14:  IJAC -- INTEGER                                   Input
               On entry: indicates whether or not the user is supplying
               Jacobian evaluation routines. If IJAC /= 0 then the user
               must supply routines JACOBF and JACOBG and also, when
               continuation is used, routines JACEPS and JACGEP. If IJAC =
               0 numerical differentiation is used to calculate the
               Jacobian and the routines D02GAZ, D02GAY, D02GAZ and D02GAX
                                                 ( ddf )
                                                 (    i)
               JACOBF must evaluate the Jacobian ( ----) for i,j=1,2,...,n,
                                                 ( ddy )
                                                 (    j)
               given x and y , for j=1,2,...,n.
                            j

               Its specification is:

                      SUBROUTINE JACOBF (X, EPS, Y, F, N)
                      INTEGER          N
                      DOUBLE PRECISION X, EPS, Y(N), F(N,N)

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the argument x.

                2:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter
                    (epsilon). This is 1 if continuation is not being used.

                3:  Y(N) -- DOUBLE PRECISION array                    Input
                    On entry: the value of the argument y , for
                                                         i
                    i=1,2,...,n.

                4:  F(N,N) -- DOUBLE PRECISION array                 Output
                                                                  ddf
                                                                     i
                    On exit: F(i,j) must be set to the value of   ----,
                                                                  ddy
                                                                     j
                    evaluated at the point (x,y), for i,j=1,2,...,n.

                5:  N -- INTEGER                                      Input
                    On entry: the number of equations, n.
               JACOBF must be declared as EXTERNAL in the (sub)program
               from which D02RAF is called. Parameters denoted as
               Input must not be changed by this procedure.

          16:  JACOBG -- SUBROUTINE, supplied by the user.
                                                         External Procedure
                                                  (  ddg   )     (  ddg   )
                                                  (     i  )     (     i  )
               JACOBG must evaluate the Jacobians ( -------) and ( -------)
                                                  ( ddy (a))     ( ddy (b))
                                                  (    j   )     (    j   )
               The ordering of the rows of AJ and BJ must correspond to
               the ordering of the boundary conditions described in the
               specification of subroutine G above.

               Its specification is:

                      SUBROUTINE JACOBG (EPS, YA, YB, AJ, BJ, N)
                      INTEGER          N
                      DOUBLE PRECISION EPS, YA(N), YB(N), AJ(N,N), BJ
                     1                 (N,N)

                1:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter,
                    (epsilon). This is 1 if continuation is not being used.

                2:  YA(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value y (a), for i=1,2,...,n.
                                          i

                3:  YB(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value y (b), for i=1,2,...,n.
                                         i

                4:  AJ(N,N) -- DOUBLE PRECISION array                Output
                                                                ddg
                                                                   i
                    On exit: AJ(i,j) must be set to the value  -------,
                                                               ddy (a)
                                                                  j
                    for i,j=1,2,...,n.

                5:  BJ(N,N) -- DOUBLE PRECISION array                Output
                                                                ddg
                                                                   i
                    On exit: BJ(i,j) must be set to the value  -------,
                                                               ddy (b)
                                                                  j
                    for i,j=1,2...,n.

                6:  N -- INTEGER                                      Input
                    On entry: the number of equations, n.
               JACOBG must be declared as EXTERNAL in the (sub)program
               from which D02RAF is called. Parameters denoted as
               Input must not be changed by this procedure.

          17:  DELEPS -- DOUBLE PRECISION                      Input/Output
               On entry: DELEPS must be given a value which specifies
               whether continuation is required. If DELEPS <= 0.0 or DELEPS
               >= 1.0 then it is assumed that continuation is not required.
               If 0.0 < DELEPS < 1.0 then it is assumed that continuation
               is required unless DELEPS < square root of machine precision
               when an error exit is taken. DELEPS is used as the increment
               (epsilon) -(epsilon)  (see (4)) and the choice DELEPS = 0.1
                        2          1
               is recommended. On exit: an overestimate of the increment
               (epsilon) -(epsilon)    (in fact the value of the increment
                        p          p-1
               which would have been tried if the restriction (epsilon) =1
                                                                       p
               had not been imposed). If continuation was not requested
               then DELEPS = 0.0.

               If continuation is not requested then the parameters JACEPS
               and JACGEP may be replaced by dummy actual parameters in the
               call to D02RAF. (D02GAZ and D02GAX respectively may be used
               as the dummy parameters.)

          18:  JACEPS -- SUBROUTINE, supplied by the user.
                                                         External Procedure
                                                       ddf
                                                          i
               JACEPS must evaluate the derivative  ----------- given x and
                                                    dd(epsilon)
               y if continuation is being used.

               Its specification is:

                      SUBROUTINE JACEPS (X, EPS, Y, F, N)
                      INTEGER          N
                      DOUBLE PRECISION X, EPS, Y(N), F(N)

                1:  X -- DOUBLE PRECISION                             Input
                    On entry: the value of the argument x.

                2:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter,
                    (epsilon).

                3:  Y(N) -- DOUBLE PRECISION array                    Input
                    On entry: the solution values y  at the point x, for
                                                   i
                    i=1,2,...,n.

                4:  F(N) -- DOUBLE PRECISION array                   Output
                                                             ddf
                                                                i
                    On exit: F(i) must contain the value  ----------- at
                                                          dd(epsilon)
                    the point (x,y), for i=1,2,...,n.

                5:  N -- INTEGER                                      Input
                    On entry: the number of equations, n.
               JACEPS must be declared as EXTERNAL in the (sub)program
               from which D02RAF is called. Parameters denoted as
               Input must not be changed by this procedure.

          19:  JACGEP -- SUBROUTINE, supplied by the user.
                                                         External Procedure
                                                        ddg
                                                           i
               JACGEP must evaluate the derivatives  ----------- if
                                                     dd(epsilon)
               continuation is being used.

               Its specification is:

                      SUBROUTINE JACGEP (EPS, YA, YB, BCEP, N)
                      INTEGER          N
                      DOUBLE PRECISION EPS, YA(N), YB(N), BCEP(N)

                1:  EPS -- DOUBLE PRECISION                           Input
                    On entry: the value of the continuation parameter,
                    (epsilon).

                2:  YA(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value of y (a), for i=1,2,...,n.
                                            i

                3:  YB(N) -- DOUBLE PRECISION array                   Input
                    On entry: the value of y (b), for i=1,2,...,n.
                                            i

                4:  BCEP(N) -- DOUBLE PRECISION array                Output
                    On exit: BCEP(i) must contain the value of
                        ddg
                           i
                     -----------, for i=1,2,...,n.
                     dd(epsilon)

                5:  N -- INTEGER                                      Input
                    On entry: the number of equations, n.
               JACGEP must be declared as EXTERNAL in the (sub)program
               from which D02RAF is called. Parameters denoted as
               Input must not be changed by this procedure.

          20:  WORK(LWORK) -- DOUBLE PRECISION array              Workspace

          21:  LWORK -- INTEGER                                       Input
               On entry:
               the dimension of the array WORK as declared in the
               (sub)program from which D02RAF is called.
                                         2         2
               Constraint: LWORK>=MNP*(3N +6N+2)+4N +3N.

          22:  IWORK(LIWORK) -- INTEGER array                     Workspace

          23:  LIWORK -- INTEGER                                      Input
               On entry:
               the dimension of the array IWORK as declared in the
               (sub)program from which D02RAF is called.
               Constraints:
                    LIWORK>=MNP*(2*N+1)+N, if IJAC /= 0,

                                         2
                    LIWORK>=MNP*(2*N+1)+N +4*N+2, if IJAC = 0.

          24:  IFAIL -- INTEGER                                Input/Output
               For this routine, the normal use of IFAIL is extended to
               control the printing of error and warning messages as well
               as specifying hard or soft failure (see the Essential
               Introduction).

               Before entry, IFAIL must be set to a value with the decimal
               expansion cba, where each of the decimal digits c, b and a
               must have a value of 0 or 1.
               a=0   specifies hard failure, otherwise soft failure;

               b=0   suppresses error messages, otherwise error messages
                     will be printed (see Section 6);

               c=0   suppresses warning messages, otherwise warning
                     messages will be printed (see Section 6).
               The recommended value for inexperienced users is 110 (i.e.,
               hard failure with all messages printed).

               Unless the routine detects an error (see Section 6), IFAIL
               contains 0 on exit.

          6. Error Indicators and Warnings

          Errors detected by the routine:

          For each error, an explanatory error message is output on the
          current error message unit (as defined by X04AAF), unless
          suppressed by the value of IFAIL on entry.

          IFAIL= 1
               One or more of the parameters N, MNP, NP, NUMBEG, NUMMIX,
               TOL, DELEPS, LWORK or LIWORK has been incorrectly set, or X
               (1) >= X(NP) or the mesh points X(i) are not in strictly
               ascending order.

          IFAIL= 2
               A finer mesh is required for the accuracy requested; that is
               MNP is not large enough. This error exit normally occurs
               when the problem being solved is difficult (for example,
               there is a boundary layer) and high accuracy is requested. A
               poor initial choice of mesh points will make this error exit
               more likely.

          IFAIL= 3
               The Newton iteration has failed to converge. There are
               several possible causes for this error:
               (i)   faulty coding in one of the Jacobian calculation
                     routines;

               (ii)  if IJAC = 0 then inaccurate Jacobians may have been
                     calculated numerically (this is a very unlikely
                     cause); or,

               (iii) a poor initial mesh or initial approximate solution
                     has been selected either by the user or by default or
                     there are not enough points in the initial mesh.
                     Possibly, the user should try the continuation
                     facility.

          IFAIL= 4
               The Newton iteration has reached round-off error level. It
               could be however that the answer returned is satisfactory.
               The error is likely to occur if too high an accuracy is
               requested.

          IFAIL= 5
               The Jacobian calculated by JACOBG (or the equivalent matrix
               calculated by numerical differentiation) is singular. This
               may occur due to faulty coding of JACOBG or, in some
               circumstances, to a zero initial choice of approximate
               solution (such as is chosen when INIT = 0).

          IFAIL= 6
               There is no dependence on (epsilon) when continuation is
               being used. This can be due to faulty coding of JACEPS or
               JACGEP or, in some circumstances, to a zero initial choice
               of approximate solution (such as is chosen when INIT = 0).

          IFAIL= 7
               DELEPS is required to be less than machine precision for
               continuation to proceed. It is likely that either the
               problem (3) has no solution for some value near the current
               value of (epsilon) (see the advisory print out from D02RAF)
               or that the problem is so difficult that even with
               continuation it is unlikely to be solved using this routine.
               If the latter cause is suspected then using more mesh points
               initially may help.

          IFAIL= 8
               Indicates that a serious error has occurred in a call to
               D02RAF. Check all array subscripts and subroutine parameter
               lists in calls to D02RAF. Seek expert help.

          IFAIL= 9
               Indicates that a serious error has occurred in a call to
               D02RAR. Check all array subscripts and subroutine parameter
               lists in calls to D02RAF. Seek expert help.

          7. Accuracy

          The solution returned by the routine will be accurate to the
          user's tolerance as defined by the relation (5) except in extreme
          circumstances. The final error estimate over the whole mesh for
          each component is given in the array ABT. If too many points are
          specified in the initial mesh, the solution may be more accurate
          than requested and the error may not be approximately
          equidistributed.

          8. Further Comments

          There are too many factors present to quantify the timing. The
          time taken by the routine is negligible only on very simple
          problems.

          The user is strongly recommended to set IFAIL to obtain self-
          explanatory error messages, and also monitoring information about
          the course of the computation.

          In the case where the user wishes to solve a sequence of similar
          problems, the use of the final mesh and solution from one case as
          the initial mesh is strongly recommended for the next.

          9. Example

          We solve the differential equation

                                                      2
                            y'''=-yy''-2(epsilon)(1-y' )

          with (epsilon)=1 and boundary conditions

                               y(0)=y'(0)=0,  y'(10)=1

          to an accuracy specified by TOL=1.0E-4. The continuation facility
          is used with the continuation parameter (epsilon) introduced as
          in the differential equation above and with DELEPS = 0.1
          initially. (The continuation facility is not needed for this
          problem and is used here for illustration.)

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd03}{NAG On-line Documentation: d03}
\beginscroll
\begin{verbatim}



     D03(3NAG)         Foundation Library (12/10/92)         D03(3NAG)



          D03 -- Partial Differential Equations         Introduction -- D03
                                    Chapter D03
                          Partial Differential Equations

          1. Scope of the Chapter

          This chapter is concerned with the solution of partial
          differential equations.

          2. Background to the Problems

          The definition of a partial differential equation problem
          includes not only the equation itself but also the domain of
          interest and appropriate subsidiary conditions. Indeed, partial
          differential equations are usually classified as elliptic,
          hyperbolic or parabolic according to the form of the equation and
          the form of the subsidiary conditions which must be assigned to
          produce a well-posed problem. Ultimately it is hoped that this
          chapter will contain routines for the solution of equations of
          each of these types together with automatic mesh generation
          routines and other utility routines particular to the solution of
          partial differential equations. The routines in this chapter will
          often call upon routines from the Linear Algebra Chapter F04 --
          Simultaneous Linear Equations.

          The classification of partial differential equations is easily
          described in the case of linear equations of the second order in
          two independent variables, i.e., equations of the form

                        au  +2bu  +cu  +du +eu +fu+g=0,                 (1)
                          xx    xy   yy   x   y

          where a, b, c, d, e, f and g are functions of x and y only.
          Equation (1) is called elliptic, hyperbolic or parabolic
                           2
          according as ac-b  is positive, negative or zero. Useful
          definitions of the concepts of elliptic, hyperbolic and parabolic
          character can also be given for differential equations in more
          than two independent variables, for systems and for nonlinear
          differential equations.

          For elliptic equations, of which Laplace's equation

                                   u  +u  =0                            (2)
                                    xx  yy

          is the simplest example of second order, the subsidiary
          conditions take the form of boundary conditions, i.e., conditions
          which provide information about the solution at all points of a
          closed boundary. For example, if equation (2) holds in a plane
          domain D bounded by a contour C, a solution u may be sought
          subject to the condition

                                   u=f  on C,                           (3)

          where f is a given function. The condition (3) is known as a
          Dirichlet boundary condition. Equally common is the Neumann
          boundary condition

                                  u'=g  on C,                           (4)

          which is one form of a more general condition

                                 u'+fu=g  on C,                         (5)

          where u' denotes the derivative of u normal to the contour C and
          f and g are given functions. Provided that f and g satisfy
          certain restrictions, condition (5) yields a well-posed  boundary
          value problem  for Laplace's equation. In the case of the Neumann
          problem, one further piece of information, e.g. the value of u at
          a particular point, is necessary for uniqueness of the solution.
          Boundary conditions similar to the above are applicable to more
          general second order elliptic equations, whilst two such
          conditions are required for equations of fourth order.

          For hyperbolic equations, the wave equation

                                   u  -u  =0                            (6)
                                    tt  xx

          is the simplest example of second order. It is equivalent to a
          first order system

                               u -v =0,  v -u =0.                       (7)
                                t  x      t  x

          The subsidiary conditions may take the form of initial
          conditions, i.e., conditions which provide information about the
          solution at points on a suitable open boundary. For example, if
          equation (6) is satisfied for t>0, a solution u may be sought
          such that

                          u(x,0)=f(x),  u (x,0)=g(x),                   (8)
                                         t

          where f and g are given functions. This is an example of an
          initial value problem, sometimes known as Cauchy's problem.

          For parabolic equations, of which the heat conduction equation

                                    u -u  =0                            (9)
                                     t  xx

          is the simplest example, the subsidiary conditions always include
          some of initial type and may also include some of boundary type.
          For example, if equation (9) is satisfied for t>0 and 0<x<1, a
          solution u may be sought such that

                             u(x,0)=f(x),  0<x<1,                      (10)

          and

                          u(0,t)=0,  u(1,t)=1,  t>0.                   (11)

          This is an example of a mixed  initial/boundary value problem.

          For all types of partial differential equations, finite
          difference methods (Mitchell and Griffiths [5]) and finite
          element methods (Wait and Mitchell [9]) are the most common means
          of solution and such methods obviously feature prominently either
          in this chapter or in the companion NAG Finite Element Library.
          Many of the utility routines in this chapter are concerned with
          the solution of the large sparse systems of equations which arise
          from the finite difference and finite element methods.

          Alternative methods of solution are often suitable for special
          classes of problems. For example, the method of characteristics
          is the most common for hyperbolic equations involving time and
          one space dimension (Smith [7]). The method of lines (see Mikhlin
          and Smolitsky [4]) may be used to reduce a parabolic equation to
          a (stiff) system of ordinary differential equations, which may be
          solved by means of routines from Chapter D02 -- Ordinary
          Differential Equations. Similarly, integral equation or boundary
          element methods (Jaswon and Symm [3]) are frequently used for
          elliptic equations. Typically, in the latter case, the solution
          of a boundary value problem is represented in terms of certain
          boundary functions by an integral expression which satisfies the
          differential equation throughout the relevant domain. The
          boundary functions are obtained by applying the given boundary
          conditions to this representation. Implementation of this method
          necessitates discretisation of only the boundary of the domain,
          the dimensionality of the problem thus being effectively reduced
          by one. The boundary conditions yield a full system of
          simultaneous equations, as opposed to the sparse systems yielded
          by the finite difference and finite element methods, but the full
          system is usually of much lower order. Solution of this system
          yields the boundary functions, from which the solution of the
          problem may be obtained, by quadrature, as and where required.

          2.1. References

          [1]   Ames W F (1977) Nonlinear Partial Differential Equations in
                Engineering. Academic Press (2nd Edition).

          [2]   Berzins M (1990) Developments in the NAG Library Software
                for Parabolic Equations. Scientific Software Systems. (ed J
                C Mason and M G Cox) Chapman and Hall. 59--72.

          [3]   Jaswon M A and Symm G T (1977) Integral Equation Methods in
                Potential Theory and Elastostatics. Academic Press.

          [4]   Mikhlin S G and Smolitsky K L (1967) Approximate Methods for
                the Solution of Differential and Integral Equations.
                Elsevier.

          [5]   Mitchell A R and Griffiths D F (1980) The Finite Difference
                Method in Partial Differential Equations. Wiley.

          [6]   Richtmyer R D and Morton K W (1967) Difference Methods for
                Initial-value Problems. Interscience (2nd Edition).

          [7]   Smith G D (1985) Numerical Solution of Partial Differential
                Equations: Finite Difference Methods. Oxford University
                Press (3rd Edition).

          [8]   Swarztrauber P N and Sweet R A (1979) Efficient Fortran
                Subprograms for the Solution of Separable Elliptic Partial
                Differential Equations. ACM Trans. Math. Softw. 5 352--364.

          [9]   Wait R and Mitchell A R (1985) Finite Element Analysis and
                Application. Wiley.

          3. Recommendations on Choice and Use of Routines

          The choice of routine will depend first of all upon the type of
          partial differential equation to be solved. At present no special
          allowances are made for problems with boundary singularities such
          as may arise at corners of domains or at points where boundary
          conditions change. For such problems results should be treated
          with caution.

          Users may wish to construct their own partial differential
          equation solution software for problems not solvable by the
          routines described in Sections 3.1 to 3.4 below. In such cases
          users can employ appropriate routines from the Linear Algebra
          Chapters to solve the resulting linear systems; see Section 3.5
          for further details.

          3.1. Elliptic Equations

          The routine D03EDF solves a system of seven-point difference
          equations in a rectangular grid (in two dimensions), using the
          multigrid iterative method. The equations are supplied by the
          user, and the seven-point form allows cross-derivative terms to
          be represented (see Mitchell and Griffiths [5]). The method is
          particularly efficient for large systems of equations with
          diagonal dominance.

          The routine D03EEF discretises a second-order equation on a two-
          dimensional rectangular region using finite differences and a
          seven-point molecule. The routine allows for cross-derivative
          terms, Dirichlet, Neumann or mixed boundary conditions, and
          either central or upwind differences. The resulting seven-
          diagonal difference equations are in a form suitable for passing
          directly to the multigrid routine D03EDF, although other solution
          methods could easily be used.

          The routine D03FAF, based on the routine HW3CRT from FISHPACK
          (Swarztrauber and Sweet [8]), solves the Helmholtz equation in a
          three-dimensional cuboidal region, with any combination of
          Dirichlet, Neumann or periodic boundary conditions. The method
          used is based on the fast Fourier transform algorithm, and is
          likely to be particularly efficient on vector-processing
          machines.

          3.2. Hyperbolic Equations

          There are no routines available yet for the solution of these
          equations.

          3.3. Parabolic Equations

          There are no routines available yet for the solution of these
          equations.

          But problems in two space dimensions plus time may be treated as
          a succession of elliptic equations [1], [6] using appropriate
          D03E- routines or one may use codes from the NAG Finite Element
          Library.

          3.4. Utility Routines

          There are no utility routines available yet, but routines are
          available in the Linear Algebra Chapters for the direct and
          iterative solution of linear equations. Here we point to some of
          the routines that may be of use in solving the linear systems
          that arise from finite difference or finite element
          approximations to partial differential equation solutions.
          Chapters F01 and F04 should be consulted for further information
          and for the routine documents. Decision trees for the solution of
          linear systems are given in Section 3.5 of the F04 Chapter
          Introduction.

          The following routines allow the direct solution of symmetric
          positive-definite systems:

              Band               F04ACF

              Variable band      F01MCF and F04MCF
              (skyline)

              Tridiagonal        F04FAF

              Sparse             F01MAF* and F04MAF

          (* the parameter DROPTL should be set to zero for F01MAF to give
          a direct solution)

          and the following routines allow the iterative solution of
          symmetric positive-definite systems:

              Sparse (incomplete F01MAF and F04MBF
              Cholesky)

              Sparse (conjugate  F04MBF
              gradient)

          The latter routine above allows the user to supply a pre-
          conditioner and also allows the solution of indefinite symmetric
          systems.

          The following routines allow the direct solution of unsymmetric
          systems:

              Band               F01LBF and F04LDF

              Almost block-      F01LHF and F04LHF
              diagonal

              Tridiagonal        F01LEF and F04LEF or F04EAF

              Sparse             F01BRF (and F01BSF) and F04AXF

          and the following routine allows the iterative solution of
          unsymmetric systems:

              Sparse             F04QAF

          The above routine allows the user to supply a pre-conditioner and
          also allows the solution of least-squares systems.

          3.5. Index

          Elliptic equations
               equations on rectangular grid (seven-point 2-D        D03EDF
               molecule)
               discretisation on rectangular grid (seven-point 2-D   D03EEF
               molecule)
               Helmholtz's equation in three dimensions              D03FAF



          D03 -- Partial Differential Equations             Contents -- D03
          Chapter D03

          Partial Differential Equations

          D03EDF  Elliptic PDE, solution of finite difference equations by
                  a multigrid technique

          D03EEF  Discretize a 2nd order elliptic PDE on a rectangle

          D03FAF  Elliptic PDE, Helmholtz equation, 3-D Cartesian co-
                  ordinates

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd03edf}{NAG On-line Documentation: d03edf}
\beginscroll
\begin{verbatim}



     D03EDF(3NAG)      Foundation Library (12/10/92)      D03EDF(3NAG)



          D03 -- Partial Differential Equations                      D03EDF
                  D03EDF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D03EDF solves seven-diagonal systems of linear equations which
          arise from the discretization of an elliptic partial differential
          equation on a rectangular region. This routine uses a multigrid
          technique.

          2. Specification

                 SUBROUTINE D03EDF (NGX, NGY, LDA, A, RHS, UB, MAXIT, ACC,
                1                   US, U, IOUT, NUMIT, IFAIL)
                 INTEGER          NGX, NGY, LDA, MAXIT, IOUT, NUMIT, IFAIL
                 DOUBLE PRECISION A(LDA,7), RHS(LDA), UB(NGX*NGY), ACC, US
                1                 (LDA), U(LDA)

          3. Description

          D03EDF solves, by multigrid iteration, the seven-point scheme

           6            7
          A   u       +A   u
           i,j i-1,j+1  i,j i,j+1

            3           4        5
          +A   u      +A   u   +A   u
            i,j i-1,j   i,j ij   i,j i+1,j

                        1          2
                      +A   u     +A   u        =f  ,
                        i,j i,j-1  i,j i+1,j-1  ij

                                               i=1,2,...,n ; j=1,2,...,n ,
                                                          x             y

          which arises from the discretization of an elliptic partial
          differential equation of the form


          (alpha)(x,y)U  +(beta)(x,y)U  +(gamma)(x,y)U  +(delta)(x,y)U
                       xx             xy              yy              x

          +(epsilon)(x,y)U +(phi)(x,y)U=(psi)(x,y)
                          y

          and its boundary conditions, defined on a rectangular region.
          This we write in matrix form as

                                        Au=f

          The algorithm is described in separate reports by Wesseling [2],
          [3] and McCarthy [1].

          Systems of linear equations, matching the seven-point stencil
          defined above, are solved by a multigrid iteration. An initial
          estimate of the solution must be provided by the user. A zero
          guess may be supplied if no better approximation is available.

          A 'smoother' based on incomplete Crout decomposition is used to
          eliminate the high frequency components of the error. A
          restriction operator is then used to map the system on to a
          sequence of coarser grids. The errors are then smoothed and
          prolongated (mapped onto successively finer grids). When the
          finest cycle is reached, the approximation to the solution is
          corrected. The cycle is repeated for MAXIT iterations or until
          the required accuracy, ACC, is reached.

          D03EDF will automatically determine the number l of possible
          coarse grids, 'levels' of the multigrid scheme, for a particular
          problem. In other words, D03EDF determines the maximum integer l
          so that n  and n  can be expressed in the form
                   x      y

                        l-1          l-1
                   n =m2   +1,  n =n2   +1,  with  m>=2 and n>=2.
                    x            y

          It should be noted that the rate of convergence improves
          significantly with the number of levels used (see McCarthy [1]),
          so that n  and n  should be carefully chosen so that n -1 and
                   x      y                                     x
                                         l
          n -1 have factors of the form 2 , with l as large as possible.
           y
          For good convergence the integer l should be at least 2.

          D03EDF has been found to be robust in application, but being an
          iterative method the problem of divergence can arise. For a
          strictly diagonally dominant matrix A

                                   4           k
                                    ij   --     ij
                                 |A   |> >   |A   |
                                         --
                                         k/=4

          no such problem is foreseen. The diagonal dominance of A is not a
          necessary condition, but should this condition be strongly
          violated then divergence may occur. The quickest test is to try
          the routine.

          4. References

          [1]   McCarthy G J (1983) Investigation into the Multigrid Code
                MGD1. Report AERE-R 10889. Harwell.

          [2]   Wesseling P (1982) MGD1 - A Robust and Efficient Multigrid
                Method. Multigrid Methods. Lecture Notes in Mathematics. 960
                Springer-Verlag. 614--630.

          [3]   Wesseling P (1982) Theoretical Aspects of a Multigrid
                Method. SIAM J. Sci. Statist. Comput. 3 387--407.

          5. Parameters

           1:  NGX -- INTEGER                                         Input
               On entry: the number of interior grid points in the x-
               direction, n . NGX-1 should preferably be divisible by as
                           x
               high a power of 2 as possible. Constraint: NGX >= 3.

           2:  NGY -- INTEGER                                         Input
               On entry: the number of interior grid points in the y-
               direction, n . NGY-1 should preferably be divisible by as
                           y
               high a power of 2 as possible. Constraint: NGY >= 3.

           3:  LDA -- INTEGER                                         Input
               On entry: the first dimension of the array A as declared in
               the (sub)program from which D03EDF is called, which must
               also be a lower bound for the dimensions of the arrays RHS,
               US and U. It is always sufficient to set
               LDA>=(4*(NGX+1)*(NGY+1))/3, but slightly smaller values may
               be permitted, depending on the values of NGX and NGY. If on
               entry, LDA is too small, an error message gives the minimum
               permitted value. (LDA must be large enough to allow space
               for the coarse-grid approximations).

           4:  A(LDA,7) -- DOUBLE PRECISION array              Input/Output
                                                           k
               On entry: A(i+(j-1)*NGX,k) must be set to  A  ,   for i =
                                                           ij
               1,2,...,NGX; j = 1,2,...,NGY and k = 1,2,...,7. On exit: A
               is overwritten.

           5:  RHS(LDA) -- DOUBLE PRECISION array              Input/Output
               On entry: RHS(i+(j-1)*NGX) must be set to f  , for i =
                                                           ij
               1,2,...,NGX; j = 1,2,...,NGY. On exit: the first NGX*NGY
               elements are unchanged and the rest of the array is used as
               workspace.

           6:  UB(NGX*NGY) -- DOUBLE PRECISION array           Input/Output
               On entry: UB(i+(j-1)*NGX) must be set to the initial
               estimate for the solution u  . On exit: the corresponding
                                          ij
               component of the residual r=f-Au.

           7:  MAXIT -- INTEGER                                       Input
               On entry: the maximum permitted number of multigrid
               iterations. If MAXIT = 0, no multigrid iterations are
               performed, but the coarse-grid approximations and incomplete
               Crout decompositions are computed, and may be output if IOUT
               is set accordingly. Constraint: MAXIT >= 0.

           8:  ACC -- DOUBLE PRECISION                                Input
               On entry: the required tolerance for convergence of the
               residual 2-norm:
                                              

                                             / NGX*NGY
                                            /  --         2
                                 ||r|| =   /   >      (r )
                                      2   /    --       k
                                        \/     k=1
               where r=f-Au and u is the computed solution. Note that the
               norm is not scaled by the number of equations. The routine
               will stop after fewer than MAXIT iterations if the residual
               2-norm is less than the specified tolerance. (If MAXIT > 0,
               at least one iteration is always performed.)

               If on entry ACC = 0.0, then the machine precision is used as
               a default value for the tolerance; if ACC > 0.0, but ACC is
               less than the machine precision, then the routine will stop
               when the residual 2-norm is less than the machine precision
               and IFAIL will be set to 4. Constraint: ACC >= 0.0.

           9:  US(LDA) -- DOUBLE PRECISION array                     Output
               On exit: the residual 2-norm, stored in element US(1).

          10:  U(LDA) -- DOUBLE PRECISION array                      Output
               On exit: the computed solution u   is returned in U(i+(j-
                                                ij
               1)*NGX), for i = 1,2,...,NGX; j = 1,2,...,NGY.

          11:  IOUT -- INTEGER                                        Input
               On entry: controls the output of printed information to the
               advisory message unit as returned by X04ABF:
               IOUT = 0
                     No output.

               IOUT = 1
                     The solution u  , for i = 1,2,...,NGX; j = 1,2,...
                                   ij
                     ,NGY.

               IOUT = 2
                     The residual 2-norm after each iteration, with the
                     reduction factor over the previous iteration.

               IOUT = 3
                     As for IOUT = 1 and IOUT = 2.

               IOUT = 4
                     As for IOUT = 3, plus the final residual (as returned
                     in UB).

               IOUT = 5
                     As for IOUT = 4, plus the initial elements of A and
                     RHS.

               IOUT = 6
                     As for IOUT = 5, plus the Galerkin coarse grid
                     approximations.

               IOUT = 7
                     As for IOUT = 6, plus the incomplete Crout
                     decompositions.

               IOUT = 8
                     As for IOUT = 7, plus the residual after each
                     iteration.
               The elements A(p,k), the Galerkin coarse grid approximations
               and the incomplete Crout decompositions are output in the
               format:
                    Y-index = j

                    X-index = i A(p,1) A(p,2) A(p,3) A(p,4) A(p,5) A(p,6)
                    A(p,7)

                    where p=i+(j-1)*NGX, i = 1,2,...,NGX and j = 1,2,...
                    ,NGY.
               The vectors U(p), UB(p), RHS(p) are output in matrix form
               with NGY rows and NGX columns. Where NGX > 10, the NGX
               values for a given j-value are produced in rows of 10.
               Values of IOUT > 4 may yield considerable amounts of output.
               Constraint: 0 <= IOUT <= 8.

          12:  NUMIT -- INTEGER                                      Output
               On exit: the number of iterations performed.

          13:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. For users not
               familiar with this parameter (described in the Essential
               Introduction) the recommended value is 0.

               On exit: IFAIL = 0 unless the routine detects an error (see
               Section 6).

          6. Error Indicators and Warnings

          Errors detected by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               On entry NGX < 3,

               or       NGY < 3,

               or       LDA is too small,

               or       ACC < 0.0,

               or       MAXIT < 0,

               or       IOUT < 0,

               or       IOUT > 8.

          IFAIL= 2
               MAXIT iterations have been performed with the residual 2-
               norm decreasing at each iteration but the residual 2-norm
               has not been reduced to less than the specified tolerance
               (see ACC). Examine the progress of the iteration by setting
               IOUT >= 2.

          IFAIL= 3
               As for IFAIL = 2, except that at one or more iterations the
               residual 2-norm did not decrease. It is likely that the
               method fails to converge for the given matrix A.

          IFAIL= 4
               On entry ACC is less than the machine precision. The routine
               terminated because the residual norm is less than the
               machine precision.

          7. Accuracy

          See ACC (Section 5).

          8. Further Comments

          The rate of convergence of this routine is strongly dependent
          upon the number of levels, l, in the multigrid scheme, and thus
          the choice of NGX and NGY is very important. The user is advised
          to experiment with different values of NGX and NGY to see the
          effect they have on the rate of convergence; for example, using a
                                    6
          value such as NGX = 65 (=2 +1) followed by NGX = 64 (for which l
          = 1).

          9. Example

          The program solves the elliptic partial differential equation

                        U  -(alpha)U  +U  =-4,   (alpha)=1.7
                         xx         xy  yy

          on the unit square 0<=x,y<=1, with boundary conditions

                           {x=0, (0<=y<=1)
                    U=0 on {y=0, (0<=x<=1) U=1 on x=1,  0<=y<=1.
                           {y=1, (0<=x<=1)

          For the equation to be elliptic, (alpha) must be less than 2.

          The equation is discretized on a square grid with mesh spacing h
          in both directions using the following approximations:



                   Please see figure in printed Reference Manual

                                       1
                                 U  ~= --(U -2U +U )
                                  xx    2  E   O  W
                                       h

                                       1
                                 U  ~= --(U -2U +U )
                                  yy    2  N   O  S
                                       h

                                1
                         U  ~= ---(U -U  +U -2U +U -U  +U ).
                          xy     2  N  NW  E   O  W  SE  S
                               2h

          Thus the following equations are solved:


              1                      1
              -(alpha)u        + (1- -(alpha))u
              2        i-1,j+1       2         i,j+1

               1                                           1

          +(1- -(alpha))u      + (-4+(alpha))u       + (1- -(alpha))u
               2         i-1,j                ij           2         i+1,j

                                     1                  1
                               + (1- -(alpha))u      +  -(alpha)u
                                     2         i,j-1    2        i+1,j-1

                    2
                =-4h

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd03eef}{NAG On-line Documentation: d03eef}
\beginscroll
\begin{verbatim}



     D03EEF(3NAG)      Foundation Library (12/10/92)      D03EEF(3NAG)



          D03 -- Partial Differential Equations                      D03EEF
                  D03EEF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D03EEF discretizes a second order elliptic partial differential
          equation (PDE) on a rectangular region.

          2. Specification

                 SUBROUTINE D03EEF (XMIN, XMAX, YMIN, YMAX, PDEF, BNDY,
                1                   NGX, NGY, LDA, A, RHS, SCHEME, IFAIL)
                 INTEGER          NGX, NGY, LDA, IFAIL
                 DOUBLE PRECISION XMIN, XMAX, YMIN, YMAX, A(LDA,7), RHS(LDA)
                 CHARACTER*1      SCHEME
                 EXTERNAL         PDEF, BNDY

          3. Description

          D03EEF discretizes a second order linear elliptic partial
          differential equation of the form


                         2                 2                  2
                       dd U              dd U               dd U
          (alpha)(x,y) ----+(beta)(x,y) ------+(gamma)(x,y) ----
                         2              ddxddy                2
                       dd x                                 dd y

                        ddU                ddU
          +(delta)(x,y) ---+(epsilon)(x,y) ---+(phi)(x,y)U=(psi)(x,y)   (1)
                        ddx                ddy


          on a rectangular region

                                      x <=x<=x
                                       A      B

                                      y <=y<=y
                                       A      B

          subject to boundary conditions of the form

                                             ddU
                              a(x,y)U+b(x,y) ---=c(x,y)
                                             ddn

                 ddU
          where  --- denotes the outward pointing normal derivative on the
                 ddn
          boundary. Equation (1) is said to be elliptic if

                                                              2
                      4(alpha)(x,y)(gamma)(x,y)>=((beta)(x,y))

          for all points in the rectangular region. The linear equations
          produced are in a form suitable for passing directly to the
          multigrid routine D03EDF.

          The equation is discretized on a rectangular grid, with n  grid
                                                                   x
          points in the x-direction and n  grid points in the y-direction.
                                         y
          The grid spacing used is therefore

                                  h =(x -x )/(n -1)
                                   x   B  A    x

                                  h =(y -y )/(n -1)
                                   y   B  A    y

          and the co-ordinates of the grid points (x ,y ) are
                                                    i  j

                            x =x +(i-1)h ,  i=1,2,...,n ,
                             i  A       x              x

                            y =y +(j-1)h ,  j=1,2,...,n .
                             j  A       y              y

          At each grid point (x ,y ) six neighbouring grid points are used
                               i  j
          to approximate the partial differential equation, so that the
          equation is discretized on the following seven-point stencil:



                   Please see figure in printed Reference Manual

          For convenience the approximation u   to the exact solution
                                             ij
          U(x ,y ) is denoted by u , and the neighbouring approximations
             i  j                 O
          are labelled according to points of the compass as shown. Where
          numerical labels for the seven points are required, these are
          also shown above.

          The following approximations are used for the second derivatives:

                 2
               dd U   1
               ----~= --(u -2u +u )
                 2     2  E   O  W
               dd x   h
                       x

                 2
               dd U   1
               ----~= --(u -2u +u )
                 2     2  N   O  S
               dd y   h
                       y

                  2
                dd U    1
               ------~= -----(u -u  +u -2u +u -u  +u ).
               ddxddy 2h h   N  NW  E   O  W  SE  S
                        x y

          Two possible schemes may be used to approximate the first
          derivatives:

          Central Differences

               ddU     1
               ---~=  ---(u -u )
               ddx    2h   W  E
                        x

               ddU     1
               ---~=  ---(u -u )
               ddy    2h   N  S
                        y

          Upwind Differences

               ddU    1
               ---~=  --(u -u )if (delta)(x,y)>0
               ddx    h   E  O
                       x

               ddU    1
               ---~=  --(u -u )if (delta)(x,y)<0
               ddx    h   O  W
                       x

               ddU    1
               ---~=  --(u -u )if (epsilon)(x,y)>0
               ddy    h   N  O
                       y

               ddU    1
               ---~=  --(u -u )if (epsilon)(x,y)<0.
               ddy    h   O  S
                       y

          Central differences are more accurate than upwind differences,
          but upwind differences may lead to a more diagonally dominant
          matrix for those problems where the coefficients of the first
          derivatives are significantly larger than the coefficients of the
          second derivatives.

          The approximations used for the first derivatives may be written
          in a more compact form as follows:

               ddU     1
               ---~=  ---((k -1)u -2k u +(k +1)u )
               ddx    2h (  x    W   x O   x    E)
                        x

               ddU     1
               ---~=  ---((k -1)u -2k u +(k +1)u )
               ddy    2h (  y    S   y O   y    N)
                        y

          where k =sign (delta) and k =sign (epsilon) for upwind
                 x                   y
          differences, and k =k =0 for central differences.
                            x  y

          At all points in the rectangular domain, including the boundary,
          the coefficients in the partial differential equation are
          evaluated by calling the user-supplied subroutine PDEF, and
          applying the approximations. This leads to a seven-diagonal
          system of linear equations of the form:

              6            7
             A  u        +A  u
              ij i-1,j+1   ij i,j+1
               3          4       5
             +A  u      +A  u   +A  u
               ij i-1,j   ij ij   ij i+1,j
                 1          2
               +A  u      +A  u        =f  , i=1,2,...,n ;j=1,2,...,n ,
                 ij i,j-1   ij i+1,j-1   ij             x            y

          where the coefficients are given by

           1                 1                   1                   1
          A  =(beta)(x ,y )----- +(gamma)(x ,y )-- +(epsilon)(x ,y )---(k -1)
           ij         i  j 2h h            i  j  2             i  j 2h   y
                             x y                h                     y
                                                 y
           2                   1
          A  =-(beta)(x ,y ) -----
           ij          i  j  2h h
                               x y

           3                 1                  1                   1
          A  =(alpha)(x ,y ) --+(beta)(x ,y ) -----+(delta)(x ,y ) ---(k -1)
           ij          i  j   2         i  j  2h h           i  j  2h   x
                             h                  x y                  x
                              x

           4                  2                  1                   2
          A  =-(alpha)(x ,y ) -- -(beta)(x ,y ) ---- -(gamma)(x ,y ) --
           ij           i  j   2          i  j  h h            i  j   2
                              h                  x y                 h
                               x                                      y

                              k                    k
                               y                    y
              -(delta)(x ,y ) -- -(epsilon)(x ,y ) -- -(phi)(x ,y )
                        i  j  h              i  j  h          i  j
                               x                    y

           5                 1                  1                   1
          A  =(alpha)(x ,y ) --+(beta)(x ,y ) -----+(delta)(x ,y ) ---(k +1)
           ij          i  j   2         i  j  2h h           i  j  2h   x
                             h                  x y                  x
                              x

           6                   1
          A  =-(beta)(x ,y ) -----
           ij          i  j  2h h
                               x y

           7                 1                  1                   1
          A  =(beta)(x ,y )-----+(gamma)(x ,y )--+(epsilon)(x ,y )---(k +1)
           ij         i  j 2h h           i  j  2             i j 2h   y
                             x y               h                    y
                                                y

          f  =(psi)(x ,y )
           ij        i  j

          These equations then have to be modified to take account of the
          boundary conditions. These may be Dirichlet (where the solution
          is given), Neumann (where the derivative of the solution is
          given), or mixed (where a linear combination of solution and
          derivative is given).

          If the boundary conditions are Dirichlet, there are an infinity
          of possible equations which may be applied:

                           (mu)u  =(mu)f   , (mu)/=0.                   (2)
                                ij      ij

          If D03EDF is used to solve the discretized equations, it turns
          out that the choice of (mu) can have a dramatic effect on the
          rate of convergence, and the obvious choice (mu)=1 is not the
          best. Some choices may even cause the multigrid method to fail
          altogether. In practice it has been found that a value of the
          same order as the other diagonal elements of the matrix is best,
          and the following value has been found to work well in practice:

                                       ( { 2   2 }  4 )
                             (mu)=min  (-{ --+ --},A  ).
                                     ij( {  2   2}  ij)
                                       ( { h   h }    )
                                       ( {  x   y}    )

          If the boundary conditions are either mixed or Neumann (i.e., B
          /= 0 on return from the user-supplied subroutine BNDY), then one
          of the points in the seven-point stencil lies outside the domain.
          In this case the normal derivative in the boundary conditions is
          used to eliminate the 'fictitious' point, u       :
                                                     outside

                           ddU   1
                           ---~= --(u       -u      ).                  (3)
                           ddn   2h  outside  inside

          It should be noted that if the boundary conditions are Neumann
          and (phi)(x,y)==0, then there is no unique solution. The routine
          returns with IFAIL = 5 in this case, and the seven-diagonal
          matrix is singular.

          The four corners are treated separately. The user-supplied
          subroutine BNDY is called twice, once along each of the edges
          meeting at the corner. If both boundary conditions at this point
          are Dirichlet and the prescribed solution values agree, then this
          value is used in an equation of the form (2). If the prescribed
          solution is discontinuous at the corner, then the average of the
          two values is used. If one boundary condition is Dirichlet and
          the other is mixed, then the value prescribed by the Dirichlet
          condition is used in an equation of the form given above.
          Finally, if both conditions are mixed or Neumann, then two '
          fictitious' points are eliminated using two equations of the form
          (3).

          It is possible that equations for which the solution is known at
          all points on the boundary, have coefficients which are not
          defined on the boundary. Since this routine calls the user-
          supplied subroutine PDEF at all points in the domain, including
          boundary points, arithmetic errors may occur in the user's
          routine PDEF which this routine cannot trap. If the user has an
          equation with Dirichlet boundary conditions (i.e., B = 0 at all
          points on the boundary), but with PDE coefficients which are
          singular on the boundary, then D03EDF could be called directly
          only using interior grid points with the user's own
          discretization.

          After the equations have been set up as described above, they are
          checked for diagonal dominance. That is to say,

                     4     --    k
                   |A  | > >   |A  |,  i=1,2,...,n ; j=1,2,...,n .
                     ij    --    ij               x             y
                           k/=4

          If this condition is not satisfied then the routine returns with
          IFAIL = 6. The multigrid routine D03EDF may still converge in
          this case, but if the coefficients of the first derivatives in
          the partial differential equation are large compared with the
          coefficients of the second derivative, the user should consider
          using upwind differences (SCHEME = 'U').

          Since this routine is designed primarily for use with D03EDF,
          this document should be read in conjunction with the document for
          that routine.

          4. References

          [1]   Wesseling P (1982) MGD1 - A Robust and Efficient Multigrid
                Method. Multigrid Methods. Lecture Notes in Mathematics. 960
                Springer-Verlag. 614--630.

          5. Parameters

           1:  XMIN -- DOUBLE PRECISION                               Input

           2:  XMAX -- DOUBLE PRECISION                               Input
               On entry: the lower and upper x co-ordinates of the
               rectangular region respectively, x  and x . Constraint: XMIN
                                                 A      B
               < XMAX.

           3:  YMIN -- DOUBLE PRECISION                               Input

           4:  YMAX -- DOUBLE PRECISION                               Input
               On entry: the lower and upper y co-ordinates of the
               rectangular region respectively, y  and y . Constraint: YMIN
                                                 A      B
               < YMAX.

           5:  PDEF -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               PDEF must evaluate the functions (alpha)(x,y), (beta)(x,y),
               (gamma)(x,y), (delta)(x,y), (epsilon)(x,y), (phi)(x,y) and
               (psi)(x,y) which define the equation at a general point
               (x,y).

               Its specification is:

                      SUBROUTINE PDEF (X, Y, ALPHA, BETA, GAMMA,
                     1                 DELTA, EPSLON, PHI, PSI)
                      DOUBLE PRECISION X, Y, ALPHA, BETA, GAMMA, DELTA,
                     1                 EPSLON, PHI, PSI

                1:  X -- DOUBLE PRECISION                             Input

                2:  Y -- DOUBLE PRECISION                             Input
                    On entry: the x and y co-ordinates of the point at
                    which the coefficients of the partial differential
                    equation are to be evaluated. 8

                3:  ALPHA -- DOUBLE PRECISION                        Output

                4:  BETA -- DOUBLE PRECISION                         Output

                5:  GAMMA -- DOUBLE PRECISION                        Output

                6:  DELTA -- DOUBLE PRECISION                        Output

                7:  EPSLON -- DOUBLE PRECISION                       Output

                8:  PHI -- DOUBLE PRECISION                          Output

                9:  PSI -- DOUBLE PRECISION                          Output
                    On exit: ALPHA, BETA, GAMMA, DELTA, EPSLON, PHI and PSI
                    must be set to the values of (alpha)(x,y), (beta)(x,y),
                    (gamma)(x,y), (delta)(x,y), (epsilon)(x,y), (phi)(x,y)
                    and (psi)(x,y) respectively at the point specified by X
                    and Y.
               PDEF must be declared as EXTERNAL in the (sub)program
               from which D03EEF is called. Parameters denoted as
               Input must not be changed by this procedure.

           6:  BNDY -- SUBROUTINE, supplied by the user.
                                                         External Procedure
               BNDY must evaluate the functions a(x,y), b(x,y), and c(x,y)
               involved in the boundary conditions.

               Its specification is:

                      SUBROUTINE BNDY (X, Y, A, B, C, IBND)
                      INTEGER          IBND
                      DOUBLE PRECISION X, Y, A, B, C

                1:  X -- DOUBLE PRECISION                             Input

                2:  Y -- DOUBLE PRECISION                             Input
                    On entry: the x and y co-ordinates of the point at
                    which the boundary conditions are to be evaluated.

                3:  A -- DOUBLE PRECISION                            Output

                4:  B -- DOUBLE PRECISION                            Output

                5:  C -- DOUBLE PRECISION                            Output
                    On exit: A, B and C must be set to the values of the
                    functions appearing in the boundary conditions.

                6:  IBND -- INTEGER                                   Input
                    On entry: specifies on which boundary the point (X,Y)
                    lies. IBND = 0, 1, 2 or 3 according as the point lies
                    on the bottom, right, top or left boundary.
               BNDY must be declared as EXTERNAL in the (sub)program
               from which D03EEF is called. Parameters denoted as
               Input must not be changed by this procedure.

           7:  NGX -- INTEGER                                         Input

           8:  NGY -- INTEGER                                         Input
               On entry: the number of interior grid points in the x- and y
               -directions respectively, n  and n . If the seven-diagonal
                                          x      y
               equations are to be solved by D03EDF, then NGX-1 and NGY-1
               should preferably be divisible by as high a power of 2 as
               possible. Constraint: NGX >= 3, NGY >= 3.

           9:  LDA -- INTEGER                                         Input
               On entry:
               the first dimension of the array A as declared in the
               (sub)program from which D03EEF is called.
               Constraint: if only the seven-diagonal equations are
               required, then LDA >= NGX*NGY. If a call to this routine is
               to be followed by a call to D03EDF to solve the seven-
               diagonal linear equations, LDA >= (4*(NGX+1)*(NGY+1))/3.

               Note: this routine only checks the former condition. D03EDF,
               if called, will check the latter condition.

          10:  A(LDA,7) -- DOUBLE PRECISION array                    Output
               On exit: A(i,j), for i=1,2,...,NGX*NGY; j = 1,2,...,7,
               contains the seven-diagonal linear equations produced by the
               discretization described above. If LDA > NGX*NGY, the
               remaining elements are not referenced by the routine, but if
               LDA >= (4*(NGX+1)*(NGY+1))/3 then the array A can be passed
               directly to D03EDF, where these elements are used as
               workspace.

          11:  RHS(LDA) -- DOUBLE PRECISION array                    Output
               On exit: the first NGX*NGY elements contain the right-hand
               sides of the seven-diagonal linear equations produced by the
               discretization described above. If LDA > NGX*NGY, the
               remaining elements are not referenced by the routine, but if
               LDA >= (4*(NGY+1)*(NGY+1))/3 then the array RHS can be
               passed directly to D03EDF, where these elements are used as
               workspace.

          12:  SCHEME -- CHARACTER*1                                  Input
               On entry: the type of approximation to be used for the first
               derivatives which occur in the partial differential
               equation.

               If SCHEME = 'C', then central differences are used.

               If SCHEME = 'U', then upwind differences are used.
               Constraint: SCHEME = 'C' or 'U'.

               Note: generally speaking, if at least one of the
               coefficients multiplying the first derivatives (DELTA or
               EPSLON as returned by PDEF) are large compared with the
               coefficients multiplying the second derivatives, then upwind
               differences may be more appropriate. Upwind differences are
               less accurate than central differences, but may result in
               more rapid convergence for strongly convective equations.
               The easiest test is to try both schemes.

          13:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               On entry XMIN >= XMAX,

               or       YMIN >= YMAX,

               or       NGX < 3,

               or       NGY < 3,

               or       LDA < NGX*NGY,

               or       SCHEME is not one of 'C' or 'U'.

          IFAIL= 2
               At some point on the boundary there is a derivative in the
               boundary conditions (B /= 0 on return from a BNDY) and there
                                                                     2
                                                                   dd U
               is a non-zero coefficient of the mixed derivative  ------
                                                                  ddxddy
               (BETA /= 0 on return from PDEF).

          IFAIL= 3
               A null boundary has been specified, i.e., at some point both
               A and B are zero on return from a call to BNDY.

          IFAIL= 4
                                                                    2
               The equation is not elliptic, i.e., 4*ALPHA*GAMMA<BETA  after
               a call to PDEF. The discretization has been completed, but
               the convergence of D03EDF cannot be guaranteed.

          IFAIL= 5
               The boundary conditions are purely Neumann (only the
               derivative is specified) and there is, in general, no unique
               solution.

          IFAIL= 6
               The equations were not diagonally dominant. (See Section 3).

          7. Accuracy

          Not applicable.

          8. Further Comments

          If this routine is used as a pre-processor to the multigrid
          routine D03EDF it should be noted that the rate of convergence of
          that routine is strongly dependent upon the number of levels in
          the multigrid scheme, and thus the choice of NGX and NGY is very
          important.

          9. Example

          The program solves the elliptic partial differential equation

                             2     2
                           dd U  dd U   { ddU  ddU}
                           ----+ ----+50{ ---+ ---}=f(x,y)
                              2     2   { ddx  ddy}
                           ddx   ddy

          on the unit square 0<=x, y<=1, with boundary conditions

                ddU
                --- given on x=0 and y=0,
                ddn

               U given on x=1 and y=1.

          The function f(x,y) and the exact form of the boundary conditions
          are derived from the exact solution U(x,y)=sinxsiny.

          The equation is first solved using central differences. Since the
          coefficients of the first derivatives are large, the linear
          equations are not diagonally dominated, and convergence is slow.
          The equation is solved a second time with upwind differences,
          showing that convergence is more rapid, but the solution is less
          accurate.

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXd03faf}{NAG On-line Documentation: d03faf}
\beginscroll
\begin{verbatim}



     D03FAF(3NAG)      Foundation Library (12/10/92)      D03FAF(3NAG)



          D03 -- Partial Differential Equations                      D03FAF
                  D03FAF -- NAG Foundation Library Routine Document

          Note: Before using this routine, please read the Users' Note for
          your implementation to check implementation-dependent details.
          The symbol (*) after a NAG routine name denotes a routine that is
          not included in the Foundation Library.

          1. Purpose

          D03FAF solves the Helmholtz equation in Cartesian co-ordinates in
          three dimensions using the standard seven-point finite difference
          approximation. This routine is designed to be particularly
          efficient on vector processors.

          2. Specification

                 SUBROUTINE D03FAF (XS, XF, L, LBDCND, BDXS, BDXF, YS, YF,
                1                   M, MBDCND, BDYS, BDYF, ZS, ZF, N,
                2                   NBDCND, BDZS, BDZF, LAMBDA, LDIMF,
                3                   MDIMF, F, PERTRB, W, LWRK, IFAIL)
                 INTEGER          L, LBDCND, M, MBDCND, N, NBDCND, LDIMF,
                1                 MDIMF, LWRK, IFAIL
                 DOUBLE PRECISION XS, XF, BDXS(MDIMF,N+1), BDXF(MDIMF,N+1),
                1                 YS, YF, BDYS(LDIMF,N+1), BDYF(LDIMF,N+1),
                2                 ZS, ZF, BDZS(LDIMF,M+1), BDZF(LDIMF,M+1),
                3                 LAMBDA, F(LDIMF,MDIMF,N+1), PERTRB, W
                4                 (LWRK)

          3. Description

          D03FAF solves the three-dimensional Helmholtz equation in
          cartesian co-ordinates:

                           2     2     2
                         dd u  dd u  dd u
                         ----+ ----+ ----+(lambda)u=f(x,y,z)
                           2     2     2
                         dd x  dd y  dd z

          This subroutine forms the system of linear equations resulting
          from the standard seven-point finite difference equations, and
          then solves the system using a method based on the fast Fourier
          transform (FFT) described by Swarztrauber [1]. This subroutine is
          based on the routine HW3CRT from FISHPACK (see Swarztrauber and
          Sweet [2]).

          More precisely, the routine replaces all the second derivatives
          by second-order central difference approximations, resulting in a
          block tridiagonal system of linear equations. The equations are
          modified to allow for the prescribed boundary conditions. Either
          the solution or the derivative of the solution may be specified
          on any of the boundaries, or the solution may be specified to be
          periodic in any of the three dimensions. By taking the discrete
          Fourier transform in the x- and y-directions, the equations are
          reduced to sets of tridiagonal systems of equations. The Fourier
          transforms required are computed using the multiple FFT routines
          found in Chapter C06 of the NAG Fortran Library.

          4. References

          [1]   Swarztrauber P N (1984) Fast Poisson Solvers. Studies in
                Numerical Analysis. (ed G H Golub) Mathematical Association
                of America.

          [2]   Swarztrauber P N and Sweet R A (1979) Efficient Fortran
                Subprograms for the Solution of Separable Elliptic Partial
                Differential Equations. ACM Trans. Math. Softw. 5 352--364.

          5. Parameters

           1:  XS -- DOUBLE PRECISION                                 Input
               On entry: the lower bound of the range of x, i.e., XS <= x
               <= XF. Constraint: XS < XF.

           2:  XF -- DOUBLE PRECISION                                 Input
               On entry: the upper bound of the range of x, i.e., XS <= x
               <= XF. Constraint: XS < XF.

           3:  L -- INTEGER                                           Input
               On entry: the number of panels into which the interval
               (XS,XF) is subdivided. Hence, there will be L+1 grid points
               in the x-direction given by x =XS+(i-1)*(delta)x, for
                                            i
               i=1,2,...,L+1, where (delta)x=(XF-XS)/L is the panel width.
               Constraint: L >= 5.

           4:  LBDCND -- INTEGER                                      Input
               On entry: indicates the type of boundary conditions at x =
               XS and x = XF.
               LBDCND = 0
                     if the solution is periodic in x, i.e.,
                     u(XS,y,z)=u(XF,y,z).

               LBDCND = 1
                     if the solution is specified at x = XS and x = XF.

               LBDCND = 2
                     if the solution is specified at x = XS and the
                     derivative of the solution with respect to x is
                     specified at x = XF.

               LBDCND = 3
                     if the derivative of the solution with respect to x is
                     specified at x = XS and x = XF.

               LBDCND = 4
                     if the derivative of the solution with respect to x is
                     specified at x = XS and the solution is specified at x
                     = XF.
                Constraint: 0 <= LBDCND <= 4.

           5:  BDXS(MDIMF,N+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to x at x = XS. When LBDCND = 3 or 4, BDXS
               (j,k)=u (XS,y ,z ), for j=1,2,...,M+1; k=1,2,...,N+1.
                      x     j  k

               When LBDCND has any other value, BDXS is not referenced.

           6:  BDXF(MDIMF,N+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to x at x = XF. When LBDCND = 2 or 3, BDXF(j,k) =
               u (XF,y ,z ), for j=1,2,...,M+1; k=1,2,...,N+1.
                x     i  k

               When LBDCND has any other value, BDXF is not referenced.

           7:  YS -- DOUBLE PRECISION                                 Input
               On entry: the lower bound of the range of y, i.e., YS <= y
               <= YF. Constraint: YS < YF.

           8:  YF -- DOUBLE PRECISION                                 Input
               On entry: the upper bound of the range of y, i.e., YS <= y
               <= YF. Constraint: YS < YF.

           9:  M -- INTEGER                                           Input
               On entry: the number of panels into which the interval
               (YS,YF) is subdivided. Hence, there will be M+1 grid points
               in the y-direction given by y  = YS + (j-1)*(delta)y for
                                            j
               j=1,2,...,M+1, where (delta)y=(YF-YS)/M is the panel width.
               Constraint: M >= 5.

          10:  MBDCND -- INTEGER                                      Input
               On entry: indicates the type of boundary conditions at y =
               YS and y = YF.
               MBDCND = 0
                     if the solution is periodic in y, i.e.,
                     u(x,YF,z)=u(x,YS,z).

               MBDCND = 1
                     if the solution is specified at y = YS and y = YF.

               MBDCND = 2
                     if the solution is specified at y = YS and the
                     derivative of the solution with respect to y is
                     specified at y = YF.

               MBDCND = 3
                     if the derivative of the solution with respect to y is
                     specified at y = YS and y = YF.

               MBDCND = 4
                     if the derivative of the solution with respect to y is
                     specified at y = YS and the solution is specified at y
                     = YF.
                Constraint: 0 <= MBDCND <= 4.

          11:  BDYS(LDIMF,N+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to y at y = YS. When MBDCND = 3 or 4, BDYS
               (i,k)=u (x ,y ,z ), for i=1,2,...,L+1; k=1,2,...,N+1.
                      y  i  s  k

               When MBDCND has any other value, BDYS is not referenced.

          12:  BDYF(LDIMF,N+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to y at y = YF. When MBDCND = 2 or 3, BDYF
               (i,k)=u (x ,YF,z ), for i=1,2,...,L+1; k=1,2,...,N+1.
                      y  i     k

               When MBDCND has any other value, BDYF is not referenced.

          13:  ZS -- DOUBLE PRECISION                                 Input
               On entry: the lower bound of the range of z, i.e., ZS <= z
               <= ZF. Constraint: ZS < ZF.

          14:  ZF -- DOUBLE PRECISION                                 Input
               On entry: the upper bound of the range of z, i.e., ZS <= z
               <= ZF. Constraint: ZS < ZF.

          15:  N -- INTEGER                                           Input
               On entry: the number of panels into which the interval
               (ZS,ZF) is subdivided. Hence, there will be N+1 grid points
               in the z-direction given by z =ZS+(k-1)*(delta)z, for
                                            k
               k=1,2,...,N+1, where (delta)z=(ZF-ZS)/N is the panel width.
               Constraint: N >= 5.

          16:  NBDCND -- INTEGER                                      Input
               On entry: specifies the type of boundary conditions at z =
               ZS and z = ZF.
               NBDCND = 0
                     if the solution is periodic in z, i.e.,
                     u(x,y,ZF)=u(x,y,ZS).

               NBDCND = 1
                     if the solution is specified at z = ZS and z = ZF.

               NBDCND = 2
                     if the solution is specified at z = ZS and the
                     derivative of the solution with respect to z is
                     specified at z = ZF.

               NBDCND = 3
                     if the derivative of the solution with respect to z is
                     specified at z = ZS and z = ZF.

               NBDCND = 4
                     if the derivative of the solution with respect to z is
                     specified at z = ZS and the solution is specified at z
                     = ZF.
                Constraint: 0 <= NBDCND <= 4.

          17:  BDZS(LDIMF,M+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to z at z = ZS. When NBDCND = 3 or 4, BDZS
               (i,j)=u (x ,y ,ZS)=u(x,y,z), for i=1,2,...,L+1;
                      z  i  j
               j=1,2,...,M+1.

               When NBDCND has any other value, BDZS is not referenced.

          18:  BDZF(LDIMF,M+1) -- DOUBLE PRECISION array              Input
               On entry: the values of the derivative of the solution with
               respect to z at z = ZF. When NBDCND = 2 or 3, BDZF
               (i,j)=u (x ,y ,ZF)=u(x,y,z), for i=1,2,...,L+1;
                      z  i  j
               j=1,2,...,M+1.

               When NBDCND has any other value, BDZF is not referenced.

          19:  LAMBDA -- DOUBLE PRECISION                             Input
               On entry: the constant (lambda) in the Helmholtz equation.
               For certain positive values of (lambda) a solution to the
               differential equation may not exist, and close to these
               values the solution of the discretized problem will be
               extremely ill-conditioned. If (lambda)>0, then D03FAF will
               set IFAIL to 3, but will still attempt to find a solution.
               However, since in general the values of (lambda) for which
               no solution exists cannot be predicted a priori, the user is
               advised to treat any results computed with (lambda)>0 with
               great caution.

          20:  LDIMF -- INTEGER                                       Input
               On entry:
               the first dimension of the arrays F, BDYS, BDYF, BDZS and
               BDZF as declared in the (sub)program from which D03FAF is
               called.
               Constraint: LDIMF >= L + 1.

          21:  MDIMF -- INTEGER                                       Input
               On entry: the second dimension of the array F and
               the first dimension of the arrays BDXS and BDXF as declared
               in the (sub)program from which D03FAF is called.
               Constraint: MDIMF >= M + 1.

          22:  F(LDIMF,MDIMF,N+1) -- DOUBLE PRECISION array    Input/Output
               On entry: the values of the right-side of the Helmholtz
               equation and boundary values (if any).

               F(i,j,k)=f(x ,y ,z ) i=2,3,...,L, j=2,3,...,M and k
                           i  j  k
               =2,3,...,N.

               On the boundaries F is defined by
               LBDCND    F(1,j,k)  F(L+1,j,k)

               0         f(XS,y ,z )f(XS,y ,z )
                               j  k      j  k

               1         u(XS,y ,z )u(XF,y ,z )
                               j  k      j  k

               2         u(XS,y ,z )f(XF,y ,z )  j=1,2,...,M+1
                               j  k      j  k

               3         f(XS,y ,z )f(XF,y ,z )  k=1,2,...,N+1
                               j  k      j  k

               4         f(XS,y ,z )u(XF,y ,z )
                               j  k      j  k



               MBDCND    F(i,1,k)  F(
                                   i,M+1,k)

               0         f(x ,YS,z )f(x ,YS,z )
                            i     k   i     k

               1         u(x ,YS,z )u(x ,YF,z )
                            i     k   i     k

               2         u(x ,YS,z )f(x ,YF,z )  i=1,2,...,L+1
                            i     k   i     k

               3         f(x ,YS,z )f(x ,YF,z )  k=1,2,...,N+1
                            i     k   i     k

               4         f(x ,YS,z )u(x ,YF,z )
                            i     k   i     k



               NBDCND    F(i,j,1)  F(i,j,N+1
                                   )

               0         f(x ,y ,ZS)f(x ,y ,ZS)
                            i  j      i  j

               1         u(x ,y ,ZS)u(x ,y ,ZF)
                            i  j      i  j

               2         u(x ,y ,ZS)f(x ,y ,ZF)  i=1,2,...,L+1
                            i  j      i  j

               3         f(x ,y ,ZS)f(x ,y ,ZF)  j=1,2,...,M+1
                            i  j      i  j

               4         f(x ,y ,ZS)u(x ,y ,ZF)
                            i  j      i  j
               Note: if the table calls for both the solution u and the
               right-hand side f on a boundary, then the solution must be
               specified. On exit: F contains the solution u(i,j,k) of the
               finite difference approximation for the grid point (
               x ,y ,z ) for i=1,2,...,L+1, j=1,2,...,M+1 and k=1,2,...,N+1.
                i  j  k

          23:  PERTRB -- DOUBLE PRECISION                            Output
               On exit: PERTRB = 0, unless a solution to Poisson's
               equation ((lambda)=0) is required with a combination of
               periodic or derivative boundary conditions (LBDCND, MBDCND
               and NBDCND = 0 or 3). In this case a solution may not exist.
               PERTRB is a constant, calculated and subtracted from the
               array F, which ensures that a solution exists. D03FAF then
               computes this solution, which is a least-squares solution to
               the original approximation. This solution is not unique and
               is unnormalised. The value of PERTRB should be small
               compared to the right-hand side F, otherwise a solution has
               been obtained to an essentially different problem. This
               comparison should always be made to insure that a meaningful
               solution has been obtained.

          24:  W(LWRK) -- DOUBLE PRECISION array                  Workspace

          25:  LWRK -- INTEGER                                        Input
               On entry:
               the dimension of the array W as declared in the (sub)program
               from which D03FAF is called.
               LRWK>=2*(N+1)*max(L,M)+3*L+3*M+4*N+6 is an upper bound on
               the required size of W. If LWRK is too small, the routine
               exits with IFAIL = 2, and if on entry IFAIL = 0 or IFAIL = -
               1, a message is output giving the exact value of LWRK
               required to solve the current problem.

          26:  IFAIL -- INTEGER                                Input/Output
               On entry: IFAIL must be set to 0, -1 or 1. Users who are
               unfamiliar with this parameter should refer to the Essential
               Introduction for details.

               On exit: IFAIL = 0 unless the routine detects an error or
               gives a warning (see Section 6).

               For this routine, because the values of output parameters
               may be useful even if IFAIL /=0 on exit, users are
               recommended to set IFAIL to -1 before entry. It is then
               essential to test the value of IFAIL on exit.

          6. Error Indicators and Warnings

          Errors or warnings specified by the routine:

          If on entry IFAIL = 0 or -1, explanatory error messages are
          output on the current error message unit (as defined by X04AAF).

          IFAIL= 1
               On entry XS >= XF,

               or       L < 5,

               or       LBDCND < 0,

               or       LBDCND > 4,

               or       YS >= YF,

               or       M < 5,

               or       MBDCND < 0,

               or       MBDCND > 4,

               or       ZS >= ZF,

               or       N < 5,

               or       NBDCND < 0,

               or       NBDCND > 4,

               or       LDIMF < L + 1 > 0,

               or       MDIMF < M + 1.

          IFAIL= 2
               On entry LWRK is too small.

          IFAIL= 3
               On entry (lambda) > 0.

          7. Accuracy

          None.

          8. Further Comments

          The execution time is roughly proportional to
          L*M*N*(log L+log M+5), but also depends on input parameters
                    2     2
          LBDCND and MBDCND.

          9. Example

          The example solves the Helmholz equation

                           2     2     2
                         dd u  dd u  dd u
                         ----+ ----+ ----+(lambda)u=f(x,y,z)
                            2     2     2
                         ddx   ddy   ddz

                                               (pi)
          for (x,y,z)is in [0,1]*[0,2(pi)]*[0, ----] where (lambda)=-2, and
                                                2
          f(x,y,z) is derived from the exact solution

                                          4
                                u(x,y,z)=x sin(y)cos(z).

          The equation is subject to the following boundary conditions,
          again derived from the exact solution given above.

               u(0,y,z) and u(1,y,z) are prescribed (i.e., LBDCND = 1).

               u(x,0,z)=u(x,2(pi),z) (i.e., MBDCND = 0).

                                    (pi)
               u(x,y,0) and u (x,y, ----) are prescribed (i.e. NBDCND = 2).
                             x       2

          The example program is not reproduced here. The source code for
          all example programs is distributed with the NAG Foundation
          Library software and should be available on-line.

\end{verbatim}
\endscroll
\end{page}