Bug #2024
gmx check does not notice difference in .tpr files that differ only in LJ parameters
Description
I created a system with two magnesium ions. Used the -DMG_ALLNER option in the .mdp file (or not) to get different MG parameters with the charmm36 force field from MacKerell's site. The two .tpr files do indeed differ, but gmx check -s1 -s2 does not notice this.
All tested in gmx 5.1.2.
- There is a difference in these two .tpr files:
$ gmx dump -s regmg.tpr >z.reg
$ gmx dump -s allnermg.tpr >z.allner
$ diff z.reg z.allner
1c1
< regmg.tpr:
---
> allnermg.tpr:
12c12
< ld-seed = 3435682522
---
> ld-seed = 2505241084
190c190
< functype[0]=LJ_SR, c6= 2.22434992e-05, c12= 1.97089389e-09
---
> functype[0]=LJ_SR, c6= 2.23024417e-05, c12= 1.00746762e-08
3822c3822
< type[0]={name="MG",nameB="MG"}
---
> type[0]={name="MGA",nameB="MGA"}
##### But gmx check -s -2 doesn't pick it up:
$ gmx check -s1 regmg.tpr -s2 allnermg.tpr Reading file regmg.tpr, VERSION 5.1.2 (single precision) Reading file allnermg.tpr, VERSION 5.1.2 (single precision) comparing inputrec inputrec->ld_seed (3435682522 - 2505241084) comparing top comparing idef comparing ilist BONDS comparing ilist G96BONDS comparing ilist MORSE comparing ilist CUBICBONDS comparing ilist CONNBONDS comparing ilist HARMONIC comparing ilist FENEBONDS comparing ilist TABBONDS comparing ilist TABBONDSNC comparing ilist RESTRAINTPOT comparing ilist ANGLES comparing ilist G96ANGLES comparing ilist RESTRANGLES comparing ilist LINEAR_ANGLES comparing ilist CROSS_BOND_BOND comparing ilist CROSS_BOND_ANGLE comparing ilist UREY_BRADLEY comparing ilist QANGLES comparing ilist TABANGLES comparing ilist PDIHS comparing ilist RBDIHS comparing ilist RESTRDIHS comparing ilist CBTDIHS comparing ilist FOURDIHS comparing ilist IDIHS comparing ilist PIDIHS comparing ilist TABDIHS comparing ilist CMAP comparing ilist GB12 comparing ilist GB13 comparing ilist GB14 comparing ilist GBPOL comparing ilist NPSOLVATION comparing ilist LJ14 comparing ilist COUL14 comparing ilist LJC14_Q comparing ilist LJC_NB comparing ilist LJ_SR comparing ilist BHAM comparing ilist LJ_LR comparing ilist BHAM_LR comparing ilist DISPCORR comparing ilist COUL_SR comparing ilist COUL_LR comparing ilist RF_EXCL comparing ilist COUL_RECIP comparing ilist LJ_RECIP comparing ilist DPD comparing ilist POLARIZATION comparing ilist WATERPOL comparing ilist THOLE comparing ilist ANHARM_POL comparing ilist POSRES comparing ilist FBPOSRES comparing ilist DISRES comparing ilist DISRESVIOL comparing ilist ORIRES comparing ilist ORDEV comparing ilist ANGRES comparing ilist ANGRESZ comparing ilist DIHRES comparing ilist DIHRESVIOL comparing ilist CONSTR comparing ilist CONSTRNC comparing ilist SETTLE comparing ilist VSITE2 comparing ilist VSITE3 comparing ilist VSITE3FD comparing ilist VSITE3FAD comparing ilist VSITE3OUT comparing ilist VSITE4FD comparing ilist VSITE4FDN comparing ilist VSITEN comparing ilist COM_PULL comparing ilist EQM comparing ilist EPOT comparing ilist EKIN comparing ilist ETOT comparing ilist ECONS comparing ilist TEMP comparing ilist VTEMP comparing ilist PDISPCORR comparing ilist PRES comparing ilist DH/DL_CON comparing ilist DV/DL comparing ilist DK/DL comparing ilist DVC/DL comparing ilist DVV/DL comparing ilist DVB/DL comparing ilist DVR/DL comparing ilist DVT/DL comparing atoms comparing block cgs comparing block mols comparing blocka excls comparing groups comparing flags comparing box comparing box_rel comparing boxv comparing x comparing v gcq#401: "It's easy to remember: a half a kT is equal to five fourths of a kJ/mol." (Anders Gabrielsson)
Associated revisions
History
#1 Updated by Mark Abraham over 3 years ago
- Description updated (diff)
#2 Updated by Berk Hess over 3 years ago
- Status changed from New to Accepted
I assume your issue arises because your difference is smaller than the default absolute tolerance of 0.001. Try check -abstol 0.
We might want to make 0 the default value for -abstol.
#3 Updated by Chris Neale over 3 years ago
Good catch. If you want to avoid setting a default of zero to a real number comparison, perhaps set the tolerance based on the number of significant digits, not on an absolute real number?
#4 Updated by Berk Hess over 3 years ago
But that's effectively what the relative tolerance already does.
#5 Updated by Chris Neale over 3 years ago
I see. Thanks for pointing that out.
I don't understand then why the differences of 2.22434992e-05 vs. 2.23024417e-05 and 1.97089389e-09 vs. 1.00746762e-08 were not picked up (unless gmx check is applying both absolute and relative tolerances? -- if this is the case, is there a good reason to keep the absolute tolerance rather than using a relative tolerance also with zero?)
#6 Updated by Erik Lindahl almost 2 years ago
Sorry for not addressing a (very) old comment eariler.
The problem with relative tolerances is that if the reference value is zero, the tolerance is also zero. For tpr files that might actually be a good idea, but the way the options work it's difficult to have different defaults for different types of files. For a trajectory or energy file, some terms (e.g. forces) are sums of many components but the sum might still be very close to zero. Already for a medium-size trajectory, this is pretty much guaranteed to be the case already for step 1 which gives the user a ton of differences unless we allow for an absolute tolerance by default.
I will upload a fix that uses different defaults just for TPR files.
#7 Updated by Gerrit Code Review Bot almost 2 years ago
Gerrit received a related patchset '1' for Issue #2024.
Uploader: Erik Lindahl (erik.lindahl@gmail.com)
Change-Id: gromacs~release-2018~I55b882a194d931bf5c36541e25339b6e1eb0a1e4
Gerrit URL: https://gerrit.gromacs.org/7358
#8 Updated by Mark Abraham almost 2 years ago
Moral: even gmx check is too complex
#9 Updated by Erik Lindahl almost 2 years ago
- Status changed from Accepted to Resolved
Applied in changeset fc0b6a720137828e1071799526d918947494e9c2.
#10 Updated by Erik Lindahl almost 2 years ago
- Status changed from Resolved to Closed
Use reduced default tolerances for tpx comparison
The tolerances for gmx check are mainly intended
for handling slight statistical deviations, but they
can hide differences between tpr files, when the
user likely wants exact checks. This changes
changes the default relative tolerance to 0.000001
and the absolute tolerance to zero, so that we only
allow for minor differences due to compiler optimization.
Fixes #2024.
Change-Id: I55b882a194d931bf5c36541e25339b6e1eb0a1e4