[gmx-users] GMX 2018 and 'distance' pull

[Alex]

Hi all,

There seems to be a problem with the new pull code when it comes to 
using it for umbrella sampling with 'distance' directive for WHAM (the 
original thread is my earlier exchange with Justin in '[gmx-users] 
seeming paradox with gmx wham'). In order to subdue unknown statement 
errors, the code was modified to:

pull                 = yes
pull-coord1-type     = umbrella
pull_ngroups            = 2
pull_ncoords            = 1
pull_group1_name        = K
pull_group2_name        = CNT
pull_coord1_geometry    = distance
pull_coord1_groups      = 1 2
pull-coord1-dim         = Y Y Y
pull_coord1_rate        = 0.0
pull_coord1_k           = 1000
pull_coord1_start       = yes

But it is a mess anyway. The box is 5.7 nm tall (6 nm by 5.5 nm wide) 
and the membrane's COM is definitely close to the geometric center of 
the box and it is definitely not moving beyond a few angstroms from 
thermal fluctuations (the membrane is very strongly restrained at its 
perimeter). The initial coordinate configurations for the other group 
(an ion) are a sweep of heights between -1.5 nm and 1.5 nm right 
relative to the membrane's COM along the height -- initial x and y are 
those of the membrane's COM -- see my very angry drawing here: 
https://www.dropbox.com/s/bwugqp5v9qt7iaq/pull.jpg?dl=0   :) The 
distance simply cannot exceed 1.5 nm with spherical symmetry, it just 
can't. And yet, i get errors like

"Distance between pull groups 1 and 2 (3.035939 nm) is larger than 0.49 
times the box size (2.728418)." -- this is from grompp

Sometimes grompp is actually satisfied, but then mdrun throws this after 
a few thousand steps:

"Program:     gmx mdrun, version 2018
Source file: src/gromacs/pulling/pull.cpp (line 707)
MPI rank:    0 (out of 2)

Fatal error:
Distance between pull groups 1 and 2 (2.732163 nm) is larger than 0.49 times
the box size (2.728418)."

This does not seem possible, because at this point the ion is radially 
restrained (1000 kJ/mol/nm2). This was /never/ an issue with 5.0.4 with 
a considerably smaller box (4 nm) -- the entire mdp file is a copy of 
what worked in 5.0.4, except the pull code is as above. Do I just 
continue increasing the box size?

Thanks,

Alex