CHARMM Development Project Forums User Discussion & Questions Energy terms, Constraints, Restraints, and Solvation charmm minimization machine dependency?

Hi, all

I tested the same energy minimization .inp (mini sd, mini abnr) using Linux sever and Mac OSX. I found out energy results are different by different machines.

I pasted my charmm test input file and attached my Linux and Mac osx results. You could run the input file in charmm/test/.


*Linux vs osx test
*
stream datadir.def
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
! Setup Structure
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

! read topology
open read card unit 10 name @0/top_all22_prot.rtf
read rtf card unit 10
clos unit 10

! read parameter
open read card unit 12 name @0/par_all22_prot.prm
read para card unit 12
clos unit 12

!
open read card unit 12 name @0/1ypa_minim.pdb
read sequ pdb unit 12
clos unit 12

! generate structure with the sequences
generate prot setup

open read card unit 12 name @0/1ypa_minim.pdb
read coor pdb unit 12 resid
clos unit 12

ic para
ic build
hbuild

coor print sele .not. INIT end

ener

!minimization
mini sd nstep 500
mini abnr nstep 1000
ener

stop

Any idea what causes machine dependency?
Thanks in advance

Small differences in the way the computations are performed can accumulate and lead to slightly different optimized structures, which have different energies. Changing compiler, or compiler options, can give this behavior. More often with the more efficient minimization algorithms (ABNR).
You probably do not want to use the CHARMM22 forcefield.