********> bugfix.85
Author:   Jean-Marie Teuler, CNRS-IDRIS, France
Date:     5/13/96, 'officialized' 2/18/97

Programs: Gibbs

Severity: MPI parallel only: severe.

Problem:  Divide by 0 error due to inadequate sharing of data between
          processors when NTR=1 and INTR=2 in the CONTROL namelist: a 
          PMF calculation. 

Affects:  PMF calculations with MPI.

Cause:    The constraints are read on processor 0 but are not dispatched 
          to the other processors.

Fix:  Make the following change to gibbs/parallel.f:
----------------------------------------------------------------------------
*** OLD parallel.f
--- NEW parallel.f
***************
*** 236,241 ****
--- 236,245 ----
       +             RDLMB(3,100),WELMB(3,100),NTYPS
        COMMON/PARMS/RK(900),REQ(900),TK(900),TEQ(900),PK(900),PN(900),
       +             PHASE(900),CN1(1830),CN2(1830),SOLTY(60)
+       COMMON/MEMCON/LCR04,LCR08,LCR10,LCR12,LCR13,LCR14,LCR15,LCR16,
+      *              LCR17,
+      *              LCR18,ICR20,ICR24,ICR28,ICR32,ICR36,ICR40,ICR44,
+      *              ICR48,ICR52,ICR56,ICR60,ICR64,ICR68,ICR72,MXRST
          
          dimension xx(*),vir(*)
          
***************
*** 275,280 ****
--- 279,286 ----
         call mpi_bcast(XX(L45),3*natom,MPI_DOUBLE_PRECISION,0,
       .        MPI_COMM_WORLD,ierr)
        endif 
+       call mpi_bcast(XX(LCR16),LCR17-LCR16,MPI_DOUBLE_PRECISION,0,
+      .     MPI_COMM_WORLD,ierr)
        return
        end
        
----------------------------------------------------------------------------

Temporary workarounds: Do not run with > 1 processor.

--