SUBROUTINE DGELG

PURPOSE
   TO SOLVE A GENERAL SYSTEM OF SIMULTANEOUS LINEAR EQUATIONS.

USAGE
   CALL DGELG(R,A,M,N,EPS,IER)

DESCRIPTION OF PARAMETERS
   R	  - DOUBLE PRECISION M BY N RIGHT HAND SIDE MATRIX
	    (DESTROYED). ON RETURN R CONTAINS THE SOLUTIONS
	    OF THE EQUATIONS.
   A	  - DOUBLE PRECISION M BY M COEFFICIENT MATRIX
	    (DESTROYED).
   M	  - THE NUMBER OF EQUATIONS IN THE SYSTEM.
   N	  - THE NUMBER OF RIGHT HAND SIDE VECTORS.
   EPS	  - SINGLE PRECISION INPUT CONSTANT WHICH IS USED AS
	    RELATIVE TOLERANCE FOR TEST ON LOSS OF
	    SIGNIFICANCE.
   IER	  - RESULTING ERROR PARAMETER CODED AS FOLLOWS
	    IER=0  - NO ERROR,
	    IER=-1 - NO RESULT BECAUSE OF M LESS THAN 1 OR
		     PIVOT ELEMENT AT ANY ELIMINATION STEP
		     EQUAL TO 0,
	    IER=K  - WARNING DUE TO POSSIBLE LOSS OF SIGNIFI-
		     CANCE INDICATED AT ELIMINATION STEP K+1,
		     WHERE PIVOT ELEMENT WAS LESS THAN OR
		     EQUAL TO THE INTERNAL TOLERANCE EPS TIMES
		     ABSOLUTELY GREATEST ELEMENT OF MATRIX A.

REMARKS
   INPUT MATRICES R AND A ARE ASSUMED TO BE STORED COLUMNWISE
   IN M*N RESP. M*M SUCCESSIVE STORAGE LOCATIONS. ON RETURN
   SOLUTION MATRIX R IS STORED COLUMNWISE TOO.
   THE PROCEDURE GIVES RESULTS IF THE NUMBER OF EQUATIONS M IS
   GREATER THAN 0 AND PIVOT ELEMENTS AT ALL ELIMINATION STEPS
   ARE DIFFERENT FROM 0. HOWEVER WARNING IER=K - IF GIVEN -
   INDICATES POSSIBLE LOSS OF SIGNIFICANCE. IN CASE OF A WELL
   SCALED MATRIX A AND APPROPRIATE TOLERANCE EPS, IER=K MAY BE
   INTERPRETED THAT MATRIX A HAS THE RANK K. NO WARNING IS
   GIVEN IN CASE M=1.

SUBROUTINES AND FUNCTION SUBPROGRAMS REQUIRED
   NONE

METHOD
   SOLUTION IS DONE BY MEANS OF GAUSS-ELIMINATION WITH
   COMPLETE PIVOTING.