\begin{page}{manpageXXm01}{NAG On-line Documentation: m01}
\beginscroll
\begin{verbatim}



     M01(3NAG)         Foundation Library (12/10/92)         M01(3NAG)



          M01 -- Sorting                                Introduction -- M01
                                    Chapter M01
                                      Sorting

          1. Scope of the Chapter

          This chapter is concerned with sorting numeric data. It handles
          only the simplest types of data structure and it is concerned
          only with internal sorting -- that is, sorting a set of data
          which can all be stored within the program.

          Users with large files of data or complex data structures to be
          sorted should use a comprehensive sorting program or package.

          2. Background to the Problems

          The usefulness of sorting is obvious (perhaps a little too
          obvious, since sorting can be expensive and is sometimes done
          when not strictly necessary). Sorting may traditionally be
          associated with data-processing and non-numerical programming,
          but it has many uses within the realm of numerical analysis, for
          example, to arrange eigenvalues in ascending order of absolute
          value and in the ranking of observations for nonparametric
          statistics.

          The general problem may be defined as follows. We are given N
          items of data


                                    R ,R ,...,R .
                                     1  2      N

          Each item R  contains a key K  which can be ordered relative to
                     i                 i
          any other key according to some specified criterion (for example,
          ascending numeric value). The problem is to determine a
          permutation


                                 p(1),p(2),...,p(N)

          which puts the keys in order:

                              K    <=K    <=...<=K
                               p(1)   p(2)        p(N)

          Sometimes we may wish actually to rearrange the items so that
          their keys are in order; for other purposes we may simply require
          a table of indices so that the items can be referred to in sorted
          order; or yet again we may require a table of ranks, that is, the
          positions of each item in the sorted order.

          For example, given the single-character items, to be sorted into
          alphabetic order:

               E    B    A    D    C

          the indices of the items in sorted order are

               3    2    5    4    1

          and the ranks of the items are

               5    2    1    4    3

          Indices may be converted to ranks, and vice versa, by simply
          computing the inverse permutation.

          The items may consist solely of the key (each item may simply be
          a number). On the other hand, the items may contain additional
          information (for example, each item may be an eigenvalue of a
          matrix and its associated eigenvector, the eigenvalue being the
          key). In the latter case there may be many distinct items with
          equal keys, and it may be important to preserve the original
          order among them (if this is achieved, the sorting is called '
          stable').

          There are a number of ingenious algorithms for sorting. For a
          fascinating discussion of them, and of the whole subject, see
          Knuth [1].

          2.1. References

          [1]   Knuth D E (1973) The Art of Computer Programming, Vol 3.
                Addison-Wesley.

          3. Recommendations on Choice and Use of Routines

          Four categories of routines are provided:

          - routines which rearrange the data into sorted order (M01C-);

          - routines which determine the ranks of the data, leaving the
          data unchanged (M01D-);

          - routines which rearrange the data according to pre-determined
          ranks (M01E-);

          - service routines (M01Z-).

          Routines are provided for sorting double precision data only.

          If the task is simply to rearrange a one-dimensional array of
          data into sorted order, then M01CAF should be used, since this
          requires no extra workspace and is faster than any other method.

          For many applications it is in fact preferable to separate the
          task of determining the sorted order (ranking) from the task of
          rearranging data into a pre-determined order; the latter task may
          not need to be performed at all. Frequently it may be sufficient
          to refer to the data in sorted order via an index vector, without
          rearranging it. Frequently also one set of data (e.g. a column of
          a matrix) is used for determining a set of ranks, which are then
          applied to other data (e.g. the remaining columns of the matrix).

          To determine the ranks of a set of data, use an M01D- routine.
          Routines are provided for ranking one-dimensional arrays, and for
          ranking rows or columns of two-dimensional arrays.

          To create an index vector so that data can be referred to in
          sorted order, first call an M01D- routine to determine the ranks,
          and then call M01ZAF to convert the vector of ranks into an index
          vector.

          To rearrange data according to pre-determined ranks: use M01EAF
          if the data is stored in a one-dimensional array; or if the data
          is stored in a more complicated structure, use an index vector to
          generate a new copy of the data in the desired order.

          Examples of these operations can be found in the routine
          documents of the relevant routines.

          4. Index

           Ranking:
               columns of a matrix, double precision numbers         M01DJF
               rows of a matrix, double precision numbers            M01DEF
               vector, double precision numbers                      M01DAF
          Rearranging (according to pre-determined ranks):
               vector, double precision numbers                      M01EAF
          Service routines:
               invert a permutation (ranks to indices or vice        M01ZAF
               versa)
          Sorting (i.e., rearranging into sorted order):
               vector, double precision numbers                      M01CAF


          M01 -- Sorting                                    Contents -- M01
          Chapter M01

          Sorting

          M01CAF  Sort a vector, double precision numbers

          M01DAF  Rank a vector, double precision numbers

          M01DEF  Rank rows of a matrix, double precision numbers

          M01DJF  Rank columns of a matrix, double precision numbers

          M01EAF  Rearrange a vector according to given ranks, double
                  precision numbers

          M01ZAF  Invert a permutation

\end{verbatim}
\endscroll
\end{page}
\begin{page}{manpageXXm01caf}{NAG On-line Documentation: m01caf}
\beginscroll
\begin{verbatim}



     M01CAF(3NAG)      Foundation Library (12/10/92)      M01CAF(3NAG)



          M01 -- Sorting                                             M01CAF
                  M01CAF -- 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

          M01CAF rearranges a vector of double precision numbers into
          ascending or descending order.

          2. Specification

                 SUBROUTINE M01CAF (RV, M1, M2, ORDER, IFAIL)
                 INTEGER          M1, M2, IFAIL
                 DOUBLE PRECISION RV(M2)
                 CHARACTER*1      ORDER

          3. Description

          M01CAF is based on Singleton's implementation of the 'median-of-
          three' Quicksort algorithm [2], but with two additional
          modifications. First, small subfiles are sorted by an insertion
          sort on a separate final pass (Sedgewick [1]). Second, if a
          subfile is partitioned into two very unbalanced subfiles, the
          larger of them is flagged for special treatment: before it is
          partitioned, its end-points are swapped with two random points
          within it; this makes the worst case behaviour extremely
          unlikely.

          4. References

          [1]   Sedgewick R (1978) Implementing Quicksort Programs. Comm.
                ACM. 21 847--857.

          [2]   Singleton R C (1969) An Efficient Algorithm for Sorting with
                Minimal Storage: Algorithm 347. Comm. ACM. 12 185--187.

          5. Parameters

           1:  RV(M2) -- DOUBLE PRECISION array                Input/Output
               On entry: elements M1 to M2 of RV must contain double
               precision values to be sorted. On exit: these values are
               rearranged into sorted order.

           2:  M1 -- INTEGER                                          Input
               On entry: the index of the first element of RV to be sorted.
               Constraint: M1 > 0.

           3:  M2 -- INTEGER                                          Input

               On entry: the index of the last element of RV to be sorted.
               Constraint: M2 >= M1.

           4:  ORDER -- CHARACTER*1                                   Input
               On entry: if ORDER is 'A', the values will be sorted into
               ascending (i.e., non-decreasing) order; if ORDER is 'D',
               into descending order. Constraint: ORDER = 'A' or 'D'.

           5:  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 M2 < 1,

               or       M1 < 1,

               or       M1 > M2.

          IFAIL= 2
               On entry ORDER is not 'A' or 'D'.

          7. Accuracy

          Not applicable.

          8. Further Comments

          The average time taken by the routine is approximately
          proportional to n*logn, where n = M2 - M1 + 1. The worst case
                                   2
          time is proportional to n  but this is extremely unlikely to
          occur.

          9. Example

          The example program reads a list of double precision numbers and
          sorts them into ascending order.

          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}{manpageXXm01daf}{NAG On-line Documentation: m01daf}
\beginscroll
\begin{verbatim}



     M01DAF(3NAG)      Foundation Library (12/10/92)      M01DAF(3NAG)



          M01 -- Sorting                                             M01DAF
                  M01DAF -- 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

          M01DAF ranks a vector of double precision numbers in ascending or
          descending order.

          2. Specification

                 SUBROUTINE M01DAF (RV, M1, M2, ORDER, IRANK, IFAIL)
                 INTEGER          M1, M2, IRANK(M2), IFAIL
                 DOUBLE PRECISION RV(M2)
                 CHARACTER*1      ORDER

          3. Description

          M01DAF uses a variant of list-merging, as described by Knuth [1]
          pp 165-166. The routine takes advantage of natural ordering in
          the data, and uses a simple list insertion in a preparatory pass
          to generate ordered lists of length at least 10. The ranking is
          stable: equal elements preserve their ordering in the input data.

          4. References

          [1]   Knuth D E (1973) The Art of Computer Programming, Vol 3.
                Addison-Wesley.

          5. Parameters

           1:  RV(M2) -- DOUBLE PRECISION array                       Input
               On entry: elements M1 to M2 of RV must contain double
               precision values to be ranked.

           2:  M1 -- INTEGER                                          Input
               On entry: the index of the first element of RV to be ranked.
               Constraint: M1 > 0.

           3:  M2 -- INTEGER                                          Input
               On entry: the index of the last element of RV to be ranked.
               Constraint: M2 >= M1.

           4:  ORDER -- CHARACTER*1                                   Input
               On entry: if ORDER is 'A', the values will be ranked in
               ascending (i.e., non-decreasing) order; if ORDER is 'D',
               into descending order. Constraint: ORDER = 'A' or 'D'.

           5:  IRANK(M2) -- INTEGER array                            Output
               On exit: elements M1 to M2 of IRANK contain the ranks of the
               corresponding elements of RV. Note that the ranks are in the
               range M1 to M2: thus, if RV(i) is the first element in the
               rank order, IRANK(i) is set to M1.

           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:

          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 M2 < 1,

               or       M1 < 1,

               or       M1 > M2.

          IFAIL= 2
               On entry ORDER is not 'A' or 'D'.

          7. Accuracy

          Not applicable.

          8. Further Comments

          The average time taken by the routine is approximately
          proportional to n*logn, where n = M2 - M1 + 1.

          9. Example

          The example program reads a list of double precision numbers and
          ranks them in ascending order.

          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}{manpageXXm01def}{NAG On-line Documentation: m01def}
\beginscroll
\begin{verbatim}



     M01DEF(3NAG)      Foundation Library (12/10/92)      M01DEF(3NAG)



          M01 -- Sorting                                             M01DEF
                  M01DEF -- 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

          M01DEF ranks the rows of a matrix of double precision numbers in
          ascending or descending order.

          2. Specification

                 SUBROUTINE M01DEF (RM, LDM, M1, M2, N1, N2, ORDER, IRANK,
                1                   IFAIL)
                 INTEGER          LDM, M1, M2, N1, N2, IRANK(M2), IFAIL
                 DOUBLE PRECISION RM(LDM,N2)
                 CHARACTER*1      ORDER

          3. Description

          M01DEF ranks rows M1 to M2 of a matrix, using the data in columns
          N1 to N2 of those rows. The ordering is determined by first
          ranking the data in column N1, then ranking any tied rows
          according to the data in column N1 + 1, and so on up to column
          N2.

          M01DEF uses a variant of list-merging, as described by Knuth [1]
          pp 165-166. The routine takes advantage of natural ordering in
          the data, and uses a simple list insertion in a preparatory pass
          to generate ordered lists of length at least 10. The ranking is
          stable: equal rows preserve their ordering in the input data.

          4. References

          [1]   Knuth D E (1973) The Art of Computer Programming, Vol 3.
                Addison-Wesley.

          5. Parameters

           1:  RM(LDM,N2) -- DOUBLE PRECISION array                   Input
               On entry: columns N1 to N2 of rows M1 to M2 of RM must
               contain double precision data to be ranked.

           2:  LDM -- INTEGER                                         Input
               On entry:
               the first dimension of the array RM as declared in the
               (sub)program from which M01DEF is called.
               Constraint: LDM >= M2.

           3:  M1 -- INTEGER                                          Input
               On entry: the index of the first row of RM to be ranked.
               Constraint: M1 > 0.

           4:  M2 -- INTEGER                                          Input
               On entry: the index of the last row of RM to be ranked.
               Constraint: M2 >= M1.

           5:  N1 -- INTEGER                                          Input
               On entry: the index of the first column of RM to be used.
               Constraint: N1 > 0.

           6:  N2 -- INTEGER                                          Input
               On entry: the index of the last column of RM to be used.
               Constraint: N2 >= N1.

           7:  ORDER -- CHARACTER*1                                   Input
               On entry: if ORDER is 'A', the rows will be ranked in
               ascending (i.e., non-decreasing) order; if ORDER is 'D',
               into descending order. Constraint: ORDER = 'A' or 'D'.

           8:  IRANK(M2) -- INTEGER array                            Output
               On exit: elements M1 to M2 of IRANK contain the ranks of the
               corresponding rows of RM. Note that the ranks are in the
               range M1 to M2: thus, if the ith row of RM is the first in
               the rank order, IRANK(i) is set to M1.

           9:  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 M2 < 1,

               or       N2 < 1,

               or       M1 < 1,

               or       M1 > M2,

               or       N1 < 1,

               or       N1 > N2,

               or       LDM < M2.

          IFAIL= 2
               On entry ORDER is not 'A' or 'D'.

          7. Accuracy

          Not applicable.

          8. Further Comments

          The average time taken by the routine is approximately
          proportional to n*logn, where n = M2 - M1 + 1.

          9. Example

          The example program reads a matrix of double precision numbers
          and ranks the rows in ascending order.

          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}{manpageXXm01djf}{NAG On-line Documentation: m01djf}
\beginscroll
\begin{verbatim}



     M01DJF(3NAG)      Foundation Library (12/10/92)      M01DJF(3NAG)



          M01 -- Sorting                                             M01DJF
                  M01DJF -- 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

          M01DJF ranks the columns of a matrix of double precision numbers
          in ascending or descending order.

          2. Specification

                 SUBROUTINE M01DJF (RM, LDM, M1, M2, N1, N2, ORDER, IRANK,
                1                   IFAIL)
                 INTEGER          LDM, M1, M2, N1, N2, IRANK(N2), IFAIL
                 DOUBLE PRECISION RM(LDM,N2)
                 CHARACTER*1      ORDER

          3. Description

          M01DJF ranks columns N1 to N2 of a matrix, using the data in rows
          M1 to M2 of those columns. The ordering is determined by first
          ranking the data in row M1, then ranking any tied columns
          according to the data in row M1 + 1, and so on up to row M2.

          M01DJF uses a variant of list-merging, as described by Knuth [1]
          pp 165-166. The routine takes advantage of natural ordering in
          the data, and uses a simple list insertion in a preparatory pass
          to generate ordered lists of length at least 10. The ranking is
          stable: equal columns preserve their ordering in the input data.

          4. References

          [1]   Knuth D E (1973) The Art of Computer Programming, Vol 3.
                Addison-Wesley.

          5. Parameters

           1:  RM(LDM,N2) -- DOUBLE PRECISION array                   Input
               On entry: rows M1 to M2 of columns N1 to N2 of RM must
               contain double precision data to be ranked.

           2:  LDM -- INTEGER                                         Input
               On entry:
               the first dimension of the array RM as declared in the
               (sub)program from which M01DJF is called.
               Constraint: LDM >= M2.

           3:  M1 -- INTEGER                                          Input
               On entry: the index of the first row of RM to be used.
               Constraint: M1 > 0.

           4:  M2 -- INTEGER                                          Input
               On entry: the index of the last row of RM to be used.
               Constraint: M2 >= M1.

           5:  N1 -- INTEGER                                          Input
               On entry: the index of the first column of RM to be ranked.
               Constraint: N1 > 0.

           6:  N2 -- INTEGER                                          Input
               On entry: the index of the last column of RM to be ranked.
               Constraint: N2 >= N1.

           7:  ORDER -- CHARACTER*1                                   Input
               On entry: if ORDER is 'A', the columns will be ranked in
               ascending (i.e., non-decreasing) order; if ORDER is 'D',
               into descending order. Constraint: ORDER = 'A' or 'D'.

           8:  IRANK(N2) -- INTEGER array                            Output
               On exit: elements N1 to N2 of IRANK contain the ranks of the
               corresponding columns of RM. Note that the ranks are in the
               range N1 to N2: thus, if the ith column of RM is the first
               in the rank order, IRANK(i) is set to N1.

           9:  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 M2 < 1,

               or       N2 < 1,

               or       M1 < 1,

               or       M1 > M2,

               or       N1 < 1,

               or       N1 > N2,

               or       LDM < M2.

          IFAIL= 2
               On entry ORDER is not 'A' or 'D'.

          7. Accuracy

          Not applicable.

          8. Further Comments

          The average time taken by the routine is approximately
          proportional to n*logn, where n = N2 - N1 + 1.

          9. Example

          The example program reads a matrix of double precision numbers
          and ranks the columns in ascending order.

          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}{manpageXXm01eaf}{NAG On-line Documentation: m01eaf}
\beginscroll
\begin{verbatim}



     M01EAF(3NAG)      Foundation Library (12/10/92)      M01EAF(3NAG)



          M01 -- Sorting                                             M01EAF
                  M01EAF -- 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

          M01EAF rearranges a vector of double precision numbers into the
          order specified by a vector of ranks.

          2. Specification

                 SUBROUTINE M01EAF (RV, M1, M2, IRANK, IFAIL)
                 INTEGER          M1, M2, IRANK(M2), IFAIL
                 DOUBLE PRECISION RV(M2)

          3. Description

          M01EAF is designed to be used typically in conjunction with the
          M01D- ranking routines. After one of the M01D- routines has been
          called to determine a vector of ranks, M01EAF can be called to
          rearrange a vector of real numbers into the rank order. If the
          vector of ranks has been generated in some other way, then
          M01ZBF(*) should be called to check its validity before M01EAF is
          called.

          4. References

          None.

          5. Parameters

           1:  RV(M2) -- DOUBLE PRECISION array                Input/Output
               On entry: elements M1 to M2 of RV must contain double
               precision values to be rearranged. On exit: these values are
               rearranged into rank order. For example, if IRANK(i) = M1,
               then the initial value of RV(i) is moved to RV(M1).

           2:  M1 -- INTEGER                                          Input

           3:  M2 -- INTEGER                                          Input
               On entry: M1 and M2 must specify the range of the ranks
               supplied in IRANK and the elements of RV to be rearranged.
               Constraint: 0 < M1 <= M2.

           4:  IRANK(M2) -- INTEGER array                             Input
               On entry: elements M1 to M2 of IRANK must contain a
               permutation of the integers M1 to M2, which are interpreted
               as a vector of ranks.

           5:  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 M2 < 1,

               or       M1 < 1,

               or       M1 > M2.

          IFAIL= 2
               Elements M1 to M2 of IRANK contain a value outside the range
               M1 to M2.

          IFAIL= 3
               Elements M1 to M2 of IRANK contain a repeated value.

          If IFAIL = 2 or 3, elements M1 to M2 of IRANK do not contain a
          permutation of the integers M1 to M2. On exit, the contents of RV
          may be corrupted. To check the validity of IRANK without the risk
          of corrupting RV, use M01ZBF(*).

          7. Accuracy

          Not applicable.

          8. Further Comments

          The average time taken by the routine is approximately
          proportional to n, where n = M2 - M1 + 1.

          9. Example

          The example program reads a matrix of double precision numbers
          and rearranges its rows so that the elements of the kth column
          are in ascending order. To do this, the program first calls
          M01DAF to rank the elements of the kth column, and then calls
          M01EAF to rearrange each column into the order specified by the
          ranks. The value of k is read from the data-file.

          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}{manpageXXm01zaf}{NAG On-line Documentation: m01zaf}
\beginscroll
\begin{verbatim}



     M01ZAF(3NAG)      Foundation Library (12/10/92)      M01ZAF(3NAG)



          M01 -- Sorting                                             M01ZAF
                  M01ZAF -- 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

          M01ZAF inverts a permutation, and hence converts a rank vector to
          an index vector, or vice versa.

          2. Specification

                 SUBROUTINE M01ZAF (IPERM, M1, M2, IFAIL)
                 INTEGER          IPERM(M2), M1, M2, IFAIL

          3. Description

          There are two common ways of describing a permutation using an
          integer vector IPERM. The first uses ranks: IPERM(i) holds the
          position to which the ith data element should be moved in order
          to sort the data; in other words its rank in the sorted order.
          The second uses indices: IPERM(i) holds the current position of
          the data element which would occur in ith position in sorted
          order. For example, given the values

                 3.5    5.9    2.9    0.5

          to be sorted in ascending order, the ranks would be

                 3      4      2      1

          and the indices would be

                 4      3      1      2

          The M01D- routines generate ranks, and the M01E- routines require
          ranks to be supplied to specify the re-ordering. However if it is
          desired simply to refer to the data in sorted order without
          actually re-ordering them, indices are more convenient than ranks
          (see the example in Section 9).

          M01ZAF can be used to convert ranks to indices, or indices to
          ranks, as the two permutations are inverses of one another.

          4. References

          None.

          5. Parameters

           1:  IPERM(M2) -- INTEGER array                      Input/Output
               On entry: elements M1 to M2 of IPERM must contain a
               permutation of the integers M1 to M2. On exit: these
               elements contain the inverse permutation of the integers M1
               to M2.

           2:  M1 -- INTEGER                                          Input

           3:  M2 -- INTEGER                                          Input
               On entry: M1 and M2 must specify the range of elements used
               in the array IPERM and the range of values in the
               permutation, as specified under IPERM. Constraint: 0 < M1 <=
               M2.

           4:  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 M2 < 1,

               or       M1 < 1,
               or       M1 > M2.
          IFAIL= 2
               Elements M1 to M2 of IPERM contain a value outside the range
               M1 to M2.

          IFAIL= 3
               Elements M1 to M2 of IPERM contain a repeated value.

          If IFAIL = 2 or 3, elements M1 to M2 of IPERM do not contain a
          permutation of the integers M1 to M2; on exit these elements are
          usually corrupted. To check the validity of a permutation without
          the risk of corrupting it, use M01ZBF(*).

          7. Accuracy

          Not applicable.

          8. Further Comments

          None.

          9. Example

          The example program reads a matrix of double precision numbers
          and prints its rows in ascending order as ranked by M01DEF. The
          program first calls M01DEF to rank the rows, and then calls
          M01ZAF to convert the rank vector to an index vector, which is
          used to refer to the rows in sorted order.

          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}