Project

General

Profile

Bug #122

mdrun-mpi writes broken molecules

Added by Perttu Niemelä almost 13 years ago. Updated over 12 years ago.

Status:
Closed
Priority:
High
Assignee:
Erik Lindahl
Category:
mdrun
Target version:
Affected version - extra info:
Affected version:
Difficulty:
uncategorized
Close

Description

The .xtc trajectory written by mpi-version of the mdrun contains broken
molecules across the periodic boundary. This happens only for a few molecules at
some time-frames, most of the molecules remain whole during simulation time. The
program trjconv could not make the molecules whole again with any option for -pbc.

The problem happened only with the mpi-version of mdrun (tested with 4 CPUs on
two different platforms: IBM Power4 and DELL cluster), but not on single
processor job. To describe the problem, I included sections from coordinates for
one molecule at different stages of simulation (converted to .gro and checked
with gmxdump):

1) Initial configuration (from .tpr) ====================================
127DPPC NC3 1513 4.939 0.585 6.960
127DPPC PO4 1514 4.565 0.271 6.959
127DPPC GL1 1515 4.456 0.318 6.508
127DPPC GL2 1516 4.106 0.404 6.414
127DPPC C1A 1517 4.379 0.296 6.060
127DPPC C2A 1518 4.469 0.468 5.752
127DPPC C3A 1519 4.771 0.477 5.404
127DPPC C4A 1520 4.983 0.731 5.051
127DPPC C1B 1521 3.928 0.449 5.934
127DPPC C2B 1522 4.024 0.485 5.487
127DPPC C3B 1523 4.341 0.499 5.190
127DPPC C4B 1524 4.535 0.629 4.899

2) At timeframe N (from .xtc) =============================
127DPPC NC3 1513 4.099 0.003 6.747
127DPPC PO4 1514 4.011 0.320 6.473
127DPPC GL1 1515 4.341 0.200 6.056
127DPPC GL2 1516 4.136 0.434 5.918
127DPPC C1A 1517 4.584 0.090 5.635
127DPPC C2A 1518 4.554 0.154 5.241
127DPPC C3A 1519 4.530 0.190 4.756
127DPPC C4A 1520 4.611 0.251 4.352
127DPPC C1B 1521 4.073 0.173 5.607
127DPPC C2B 1522 3.945 -0.074 5.247
127DPPC C3B 1523 3.865 -0.358 4.900
127DPPC C4B 1524 3.613 -0.389 4.521

3) At timeframe N+1 (see y-coordinate) ======================================
127DPPC NC3 1513 3.936 6.210 7.009
127DPPC PO4 1514 4.060 6.450 6.650
127DPPC GL1 1515 4.123 6.550 6.164
127DPPC GL2 1516 4.312 6.845 6.172
127DPPC C1A 1517 4.130 6.532 5.584
127DPPC C2A 1518 4.305 6.438 5.189
127DPPC C3A 1519 4.598 0.194 4.846
127DPPC C4A 1520 4.956 0.019 4.513
127DPPC C1B 1521 4.045 0.707 5.823
127DPPC C2B 1522 3.893 0.673 5.363
127DPPC C3B 1523 3.833 0.417 5.130
127DPPC C4B 1524 3.793 -0.035 5.205

test_4cpu_00000.tpr (500 KB) test_4cpu_00000.tpr A 4cpu-run (writes broken molecules) Perttu Niemelä, 11/27/2006 09:08 AM
test_1cpu_00000.tpr (500 KB) test_1cpu_00000.tpr A 1cpu-run (noes NOT write broken molecules) Perttu Niemelä, 11/27/2006 09:09 AM
test.tar.gz (72.4 KB) test.tar.gz Files for grompp Perttu Niemelä, 11/27/2006 09:19 AM
test_4cpu_00000.trr (1.32 MB) test_4cpu_00000.trr A .trr file that contains broken molecules Perttu Niemelä, 12/07/2006 01:54 PM

History

#1 Updated by Perttu Niemelä almost 13 years ago

Created an attachment (id=100)
A 4cpu-run (writes broken molecules)

#2 Updated by Perttu Niemelä almost 13 years ago

Created an attachment (id=101)
A 1cpu-run (noes NOT write broken molecules)

#3 Updated by Perttu Niemelä almost 13 years ago

Created an attachment (id=102)
Files for grompp

#4 Updated by Perttu Niemelä almost 13 years ago

Created an attachment (id=103)
A .trr file that contains broken molecules

Attached is a .trr trajectory with 20 timeframes, written by the mdrun-mpi
program. If you look at molecule number 127 in the trajectory, you should see
how it is broken in some of the time frames (first time at t= 1200.000 ps).

The funny thing is that trjconv (version 3.3.1) was not able to make the
molecule whole, but if I use previous versions of trjconv (3.1.4 or 3.2.1) with
"-pbc whole" option, then that fixes the trajectory.

My main concern is, wether this is only related to how mdrun writes out
coordinates or is there something strange also in the pbc handling during
simulation?

#5 Updated by David van der Spoel over 12 years ago

Hi, I can not reproduce your problem with the current CVS version of GROMACS,
but I can with older versions, slightly pre 3.3.1. It seems therefore that this
bug has already been fixed both in CVS 3.3. and head.

Also available in: Atom PDF