SUBROUTINE SEARCH(M, DMD, D, K, ITER, ALPHA, DMWORK, DMWRK1,
     *                  WORK1, DMWRK2, WORK2, IWORK, OUNIT)
C
C<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
C
C   PURPOSE
C   -------
C
C      PRODUCES AND OUTPUTS TREE WHICH MAXIMIZES ALPHA*TRUE -
C      (1.0-ALPHA)*FALSE WHERE TRUE IS THE NUMBER OF TRIADS CORRECTLY
C      PREDICTED BY THE TREE AND FALSE IS THE NUMBER OF TRIADS
C      INCORRECTLY PREDICTED BY THE TREE AND ALPHA IS A USER-DEFINED
C      PROBABILITY
C
C   DESCRIPTION
C   -----------
C
C   1.  A TRIAD IS A SEQUENCE OF THREE OBSERVATIONS.  (IJ)K IS A LEGAL
C       TRIAD IF THE DISTANCE BETWEEN OBJECTS I AND J IS SMALLER THAN
C       THE DISTANCE BETWEEN OBJECTS I AND K AND THE DISTANCE BETWEEN
C       OBJECTS J AND K.
C
C   2.  THE ROUTINE LOOPS FOR A USER-DEFINED NUMBER OF ITERATIONS.  FOR
C       THE FIRST LOOP, THE FIRST TREE CONSISTS OF THE FIRST TWO
C       OBJECTS.  THE THIRD OBJECT IS ADDED IN ALL PLACES IN THE TREE
C       AND THE TRIADS DICTATED BY THE TREE ARE FORMED.  THE LOCATION
C       IN THE TREE WHICH PREDICTS THE MOST TRUE TRIADS IS RETAINED.
C       ALL OTHER OBJECTS ARE ADDED SIMILARLY TO COMPLETE THE TREE.
C       ALL SUBSEQUENT LOOPS START WITH THE TREE FROM THE PREVIOUS LOOP
C       AND MOVE EACH OBJECT IN TURN TO EACH LOCATION IN THE TREE
C       PLACING THE OBJECT IN THE LOCATION THAT PREDICTS THE MOST TRUE
C       TRIADS.
C
C   3.  AFTER EACH ITERATION, EACH NODE IN THE TREE IS PRINTED ALONG
C       WITH ITS ANCESTOR AND THE NUMBER OF TRUE AND FALSE TRIADS
C       PREDICTED WHICH CONTAIN THAT NODE.  THEN THE TREE IS PRINTED
C       OUT ON FORTRAN UNIT OUNIT AS MATRIX OF INTEGERS.  EACH COLUMN
C       IN THE MATRIX CORRESPONDS TO THE OBJECT WHOSE NUMBER IS THE
C       FIRST ENTRY IN THE COLUMN.  THE SECOND NUMBER IN EACH COLUMN IS
C       THE ANCESTOR NODE OF THE OBJECT, THE THIRD NUMBER IS THE
C       ANCESTOR NODE OF THE SECOND NUMBER, ETC.  THE FINAL TREE CAN BE
C       RECOVERED BY CONNECTING EACH NODE TO ITS ANCESTOR AND LINKING
C       THE ANCESTORS TO THEIR ANCESTORS UNTIL ONLY ONE NODE IS LEFT.
C
C   4.  EVALUATING ALL POSSIBLE TRIADS IS EXTREMELY EXPENSIVE AND THE
C       SUBROUTINE SHOULD ONLY BE RUN ON NO MORE THAN ABOUT 15 OBJECTS.
C       ALSO THE NUMBER OF ITERATIONS SHOULD BE NO MORE THAN 5.
C
C   INPUT PARAMETERS
C   ----------------
C
C   M     INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE NUMBER OF OBJECTS.
C
C   DMD   INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE LEADING DIMENSION OF THE MATRIX D.  MUST BE AT LEAST M.
C
C   D     REAL MATRIX WHOSE FIRST DIMENSION MUST BE DMD AND SECOND
C            DIMENSION MUST BE AT LEAST M (UNCHANGED ON OUTPUT).
C         D(I,J) IS THE DISTANCE FROM OBJECT I TO OBJECT J.
C
C   K     INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE NUMBER OF NODES OF THE TREE, MUST BE GREATER THAN M BUT
C            SHOULD BE NO GREATER THAN 2*M.
C
C   ITER  INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE NUMBER OF ITERATIONS.
C
C   ALPHA REAL SCALAR (UNCHANGED ON OUTPUT).
C         THE WEIGHT IN THE MAXIMIZATION EQUATION.
C
C              0. <= ALPHA <= 1.
C
C   DMWORK INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE FIRST DIMENSION OF THE MATRIX WORK1.  MUST BE AT LEAST M.
C
C   DMWRK1 INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE SECOND DIMENSION OF THE MATRIX WORK1.  MUST BE AT LEAST M.
C
C   WORK1 REAL MATRIX WHOSE FIRST DIMENSION MUST BE DMWORK, WHOSE SECOND
C            DIMENSION MUST BE AT DMWRK1, AND WHOSE THIRD DIMENSION MUST
C            BE AT LEAST M.
C         WORK MATRIX.
C
C   DMWRK2 INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         THE FIRST DIMENSION OF THE MATRIX WORK2.  MUST BE AT LEAST M.
C
C   WORK2 REAL MATRIX WHOSE FIRST DIMENSION MUST BE DMWRK2 AND SECOND
C            DIMENSION MUST BE AT LEAST M.
C         WORK MATRIX.
C
C   IWORK INTEGER VECTOR DIMENSIONED AT LEAST M+K.
C         WORK VECTOR.
C
C   OUNIT INTEGER SCALAR (UNCHANGED ON OUTPUT).
C         UNIT NUMBER FOR OUTPUT.
C
C   REFERENCE
C   ---------
C
C     HARTIGAN, J. A. (1975).  CLUSTERING ALGORITHMS, JOHN WILEY &
C        SONS, INC., NEW YORK.  PAGES 177-188.
C
C<><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><><>
C