mdrun minimization crashes with GBSA / HCT and virtual site
Created an attachment (id=559)
i minimize protein-ligand complex with gbsa/htc solvent and virtual site. when cut-offs > 0 used, mdrun crashes with f-norm = -nan with both cg and l-bfgs optimizers. topology works fine with still algorithm as well as without any implicit solvent. topology without virtual site also works.
mdrun 4.5.1 is build with gcc-4.4.3, fftw-3.2.2 disabling threads. the error occurs with both single and double precision.
#6 Updated by Per Larsson over 9 years ago
The reason for this crash is your ET-vsite.
You have set the gb-parameters all to zero, while ET is also having a charge of 0.15325.
This causes problems since the particle will be added to the pairlist for inclusion in the GB-radii calculations, but the all parameters are zero.
I could fix this but introducing a check for this case, which will prevent this situation from happening.
It is not clear to me that vsites should have charges, what is the reason here?
#8 Updated by michal kolar about 9 years ago
thank you, apparently the error was on my side. i didn't think over the gb parameters for charged vsites.
i'm trying to use charged vsite for modelling a bromine atom with a region with a positive electrostatic potential.
but still don't know why it worked for infinite cut-offs...