[AMBER-Developers] Problem with SHAKE on pmemd.MPI only

From: Cruzeiro,Vinicius Wilian D <vwcruzeiro.ufl.edu>
Date: Wed, 21 Jun 2017 18:22:46 +0000

Dear all,

I am having a problem in some of my simulations that only happens in pmemd.MPI. This is the error I get right before the simulation starts:


    Coordinate resetting cannot be accomplished,
    deviation is too large
    iter_cnt, my_bond_idx, i and j are : 1 50 111 117

This error does not appear in pmemd, pmemd.cuda, pmemd.cuda.MPI, sander and sander.MPI, and the simulation finishes fine. This error happens both in the current amber from git and on Amber16. This error does not depend on the number of processors I choose.

After an inspection in the PMEMD code, I found out the error was being printed from shake.i :

! Shake resetting of coordinate is done here:

      rrpr = zero

      do m = 1, 3
        rrpr = rrpr + xij(m) * xpij(m)
      end do

      if (rrpr .lt. toler * 1.0d-06) then ! Deviation too large. Kill PMEMD.
        if (master) then
          write(mdout, 321) iter_cnt, my_bond_idx, i, j
          if (imin .eq. 1) write(mdout, 331)
        end if
        call mexit(6, 1)
      end if

Here is where the 321 is defined:

321 format(/5x, 'Coordinate resetting cannot be accomplished,', &
           /5x, 'deviation is too large', &
           /5x, 'iter_cnt, my_bond_idx, i and j are :', 4i8)

I checked what values were being generated for the variables rrpr and toler for the atom pair 111 and 117 (for which the error happens on pmemd.MPI). For pmemd I get rrpr=0.9305 and toler=0.9216 and for pmemd.MPI I get rrpr=-28.1436 and toler=0.9216 . This negative value for rrpr is very strange (it should be positive by construction).

I am sending attached my parmtop and rst7 files so you may reproduce this error on your computers. Has anybody ever seen this problem before?

Best regards,

Vinícius Wilian D Cruzeiro

PhD Candidate
Department of Chemistry, Physical Chemistry Division
University of Florida, United States

Voice: +1(352)846-1633

AMBER-Developers mailing list

Received on Wed Jun 21 2017 - 11:30:02 PDT
Custom Search