Dear developers

I'm simulating membrane protein by using GBIM/ASPENRMB.
However, transmembrane region(alpha helical) is extended quickly...
I'm searching the reason, and I find that the aspenrmb responds different energy value(ASP) if using single version CHARMM or parallel version.

single:
MINI MIN: Cycle ENERgy Delta-E GRMS Step-size
MINI INTERN: BONDs ANGLes UREY-b DIHEdrals IMPRopers
MINI EXTERN: VDWaals ELEC HBONds ASP USER
---------- --------- --------- --------- --------- ---------
MINI> 0 -1443.88280 0.00000 224.75333 0.02000
MINI INTERN> 387.15308 727.78460 0.00000 204.69932 923.68287
MINI EXTERN> 1572.22663 -5528.18135 0.00000 268.75205 0.00000


parallel:
MINI MIN: Cycle ENERgy Delta-E GRMS Step-size
MINI INTERN: BONDs ANGLes UREY-b DIHEdrals IMPRopers
MINI EXTERN: VDWaals ELEC HBONds ASP USER
---------- --------- --------- --------- --------- ---------
MINI> 0 7710.00478 0.00000 224.75284 0.02000
MINI INTERN> 387.15308 727.78460 0.00000 204.69932 923.68287
MINI EXTERN> 1572.22663 -5528.18135 0.00000 9422.63963 0.00000

So, before minimazation, all energies(except ASP) show same value.


I also compare aspenr (same feature of aspenrmb, but not including membrane effect),
both versions respond same value(E_ASP=433.13174),
so, aspenr module has no problem.


In addition, I calculate aspenermb with 0.0 nm membrane (membrane thickness parameter in .surf file is set to 0.0).
The result should be same to result by aspenr.

The value by single CHARMM is same, E_ASP=433.13174,
but parallelized CHARMM replies E_ASP=10034.24990, quite different value.


The calculation is done in SGI Altix with MPICH,
but similar results are acquired in Xeon cluster with MPICH,
which may show our result has no machine dependency.

And, the version of CHARMM we use is c35b2.
From c33a1, Spatially distributed parallel method was introduced.
Thus, I guess parallelized aspenrmb contains bug...