Bug #268
energy minimization differs between CG with and without parallel
Description
drastically different convergence between CG run in parallel with domain decomposition vs CG not run in parallel.
History
#1 Updated by Berk Hess about 12 years ago
Could you supply a tpr file and the command line for the parallel run?
Berk
#2 Updated by Berk Hess about 12 years ago
Created an attachment (id=331)
fixed? src/mdlib/minimize.c
#3 Updated by Berk Hess about 12 years ago
I just realized that I fixed a bug with crashes with CG and DD
a while ago.
I don't know if the system could not crash, but converge badly
due to this bug.
Could you try the fixed file I attached before and report back?
Berk
#4 Updated by Berk Hess about 12 years ago
I assume my changes to minimize.c (committed a while ago),
fixed this bug.
Can this be confirmed, so we can close this bug?
Berk
#5 Updated by Erik Lindahl about 12 years ago
Appears to work fine now. Fix will be in 4.0.3 release shortly.