[gmx-developers] Segfault during EM in recent git version
Berk Hess
hess at cbr.su.se
Tue Oct 6 16:12:54 CEST 2009
Ah, the explanation is much simpler.
You should never use the vsite constructions in proteins without all
bonds constrained.
grompp does not check for this and it would be difficult to implement.
We could consider having pdb2gmx replace all bonds by constraints
when you ask for virtual sites.
The actual problem is only in tetrahedral hydrogens in CH2 groups,
which are constructed using a non-normalized linear combinations and
cross product.
Even with all bonds constrained, this makes the hydrogen position
fluctuate a few percent
from the intended position.
Berk
Jochen Hub wrote:
> Berk Hess wrote:
>> Hi,
>>
>> I think this is "normal".
>> It could be that with vsites you get such warnings easier, since the
>> distance between
>> the two masses is small and in EM displacement is directly proportional
>> to the force
>> (no mass factor), so the rotation will be larger than without vsites.
>>
>>
>
> Hi,
>
> Here's some more news on the lincs warnings. Please have a look at his
> shapshot series:
>
> http://xray.bmc.uu.se/~jochen/lyp.jpg
>
> Snapshot 2 (EM step 28) is right after the first one (EM step 27) and
> caused the first lincs warning. EM steps 1-27 look just fine.
> Apparently, the lincs warnings in the lysine tip are rather a side
> effect of wrong C-H bond lengths along the lysine side chain. A few
> steps later, the C-H bonds are starting to fluctuate, and finally the
> EM crashes (sometimes again with a Segfault), or it hangs.
>
> In addition, I am also getting weired lincs warnings with rotations of
> exactly (!) 90 degree:
>
> atom 1 atom 2 angle previous, current, constraint length
> 247 250 90.0 0.1635 8.3257 0.1635
> 247 251 90.0 0.1635 7.3325 0.1635
> 250 251 90.0 0.0922 6.1981 0.0922
> 886 889 90.0 0.1583 0.6672 0.1583
> 886 890 90.0 0.1582 0.6672 0.1583
> 889 890 90.0 0.0802 0.3093 0.0802
> 1291 1294 90.0 0.1583 0.4725 0.1583
> 1291 1295 90.0 0.1582 0.4725 0.1583
> 1294 1295 90.0 0.0802 0.2394 0.0802
> 1513 1516 40.2 0.1583 0.1586 0.1583
> 1513 1517 39.9 0.1583 0.1578 0.1583
> 1516 1517 38.7 0.0802 0.0802 0.0802
>
> That does not look like coincidence, I strongly feel there is
> something wrong with the construction of the vsites?
>
> Jochen
>
>
>> Berk
>>
>> Jochen Hub wrote:
>>
>>> Hi,
>>>
>>> this fixes the Segfault, thanks! However, there are still LINCS
>>> warnings in the lysines, and sometimes also at isoleucines and at the
>>> OH-group of Tyr. Typically after 20-40 EM steps which look fine
>>> according to the decreasing Fmax etc. They always appear if the the
>>> cases with C in one cg, and the 2 heavy vsites plus one vsite (such as
>>> N) together in one cg:
>>>
>>> 1291 amber99_11 80 LYP CE 1189 -0.0143 14.026 ;
>>> qtot -4.006
>>> 1292 amber99_28 80 LYP HE1 1190 0.1135 0 ;
>>> qtot -3.893
>>> 1293 amber99_28 80 LYP HE2 1191 0.1135 0 ;
>>> qtot -3.779
>>> 1294 MNH3 80 LYP MNZ1 1192 0 8.517 ;
>>> qtot -3.779
>>> 1295 MNH3 80 LYP MNZ2 1192 0 8.517 ;
>>> qtot -3.779
>>> 1296 amber99_39 80 LYP NZ 1192 -0.3854 0 ;
>>> qtot -4.165
>>> 1297 amber99_17 80 LYP HZ1 1193 0.34 0 ;
>>> qtot -3.825
>>> 1298 amber99_17 80 LYP HZ2 1194 0.34 0 ;
>>> qtot -3.485
>>> 1299 amber99_17 80 LYP HZ3 1195 0.34 0 ;
>>> qtot -3.145
>>>
>>> Where the warning looks like that:
>>>
>>> bonds that rotated more than 30 degrees:
>>> atom 1 atom 2 angle previous, current, constraint length
>>> 1291 1294 36.8 0.1583 0.1576 0.1583 -> C--MNH3
>>> 1291 1295 37.6 0.1583 0.1587 0.1583 -> C--MNH3
>>> 1294 1295 35.4 0.0802 0.0805 0.0802 -> MNH3-MNH3
>>>
>>> Cheers,
>>> Jochen
>>>
>>>
>>> hess at sbc.su.se wrote:
>>>
>>>> Ah, this is in the vsite fix I made for the bug you
>>>> reported previously.
>>>> This code is quite obviously incorrect.
>>>> But I realize now it can be done simpler.
>>>> if (atom[a].ptype != eptVSite || pbc_set[a]) {
>>>> can be replaced by
>>>> if (pbc_set[a]) {
>>>>
>>>> This will certainly solve your last crash.
>>>> Just to be sure, could you check that this still fixes
>>>> the bug you reported previously?
>>>>
>>>> Berk
>>>>
>>>>
>>>>
>>>>
>>>>> Sorry, here a few more details:
>>>>>
>>>>> (gdb) list
>>>>> 1498 }
>>>>> 1499 if (vsite_pbc_f[vsi] == -1) {
>>>>> 1500 /* Check if this is the first processed atom of a
>>>>> vsite only cg */
>>>>> 1501 bViteOnlyCG_and_FirstAtom = TRUE;
>>>>> 1502 for(a=cgs->index[cg_v]; a<cgs->index[cg_v+1];
>>>>> a++) {
>>>>> 1503 if (atom[a].ptype != eptVSite || pbc_set[a]) {
>>>>> 1504 bViteOnlyCG_and_FirstAtom = FALSE;
>>>>> 1505 break;
>>>>> 1506 }
>>>>> 1507 }
>>>>>
>>>>> The problem is that the atom[a=0] is not defined.
>>>>>
>>>>> Jochen
>>>>>
>>>>>
>>>>>
>>>>> Berk Hess wrote:
>>>>>
>>>>>> Jochen Hub wrote:
>>>>>>
>>>>>>
>>>>>>> Berk Hess wrote:
>>>>>>>
>>>>>>>
>>>>>>>> Did you use cut-off's?
>>>>>>>> New all-versus-all loops have been added for simulations without
>>>>>>>> cut-off's.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>> Yes, for both coulomb and vdw:
>>>>>>>
>>>>>>> coulombtype = Cut-off
>>>>>>> vdw-type = Cut-off
>>>>>>>
>>>>>>> Jochen
>>>>>>>
>>>>>>>
>>>>>> Sorry, maybe my question was a but unclear.
>>>>>> Do you have finite cut-off lengths, or do you have pbc=no and
>>>>>> cut-off's
>>>>>> set to 0 (infinity).
>>>>>>
>>>>>> Berk
>>>>>>
>>>>>>
>>>>>>>> Berk
>>>>>>>>
>>>>>>>> Jochen Hub wrote:
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>>>>> Hi Berk,
>>>>>>>>>
>>>>>>>>> yes, I have always reconfigured with distclean and everything.
>>>>>>>>> And
>>>>>>>>> checked that no CC variables or similar were messing up
>>>>>>>>> something. I
>>>>>>>>> could well imagine that the problem is again due to virtual sites
>>>>>>>>> which I have in my structure.
>>>>>>>>>
>>>>>>>>> In addition, I am getting LINCS warnings during EM with the 405
>>>>>>>>> from
>>>>>>>>> a
>>>>>>>>> few weeks ago, but *not* related to the bug you fixed recently
>>>>>>>>> (with
>>>>>>>>> CGs with only vsites). These lincs warnings appear typically in
>>>>>>>>> lysines, somewhere between the last carbon and the NH3 in the
>>>>>>>>> lysine
>>>>>>>>> chain. Noteworthy, the warnings did not appear with steepest
>>>>>>>>> descent,
>>>>>>>>> but with conjugate gradient. Higher lincs-order did not help, but
>>>>>>>>> even
>>>>>>>>> increased the large changes in angles. For example:
>>>>>>>>>
>>>>>>>>> atom 1 atom 2 angle previous, current, constraint length
>>>>>>>>> 886 889 34.3 0.1583 0.1582 0.1583
>>>>>>>>> 886 890 33.8 0.1583 0.1580 0.1583
>>>>>>>>> 889 890 34.0 0.0802 0.0805 0.0802
>>>>>>>>> 1291 1294 33.8 0.1583 0.1578 0.1583
>>>>>>>>> 1291 1295 34.3 0.1583 0.1586 0.1583
>>>>>>>>> 1294 1295 32.3 0.0802 0.0803 0.0802
>>>>>>>>>
>>>>>>>>> where the respective atoms are:
>>>>>>>>> 886 amber99_11 55 LYP CE 822 -0.0143
>>>>>>>>> 14.026
>>>>>>>>> ;
>>>>>>>>> qtot -3.006
>>>>>>>>> 887 amber99_28 55 LYP HE1 823
>>>>>>>>> 0.1135 0
>>>>>>>>> ;
>>>>>>>>> qtot -2.893
>>>>>>>>> 888 amber99_28 55 LYP HE2 824
>>>>>>>>> 0.1135 0
>>>>>>>>> ;
>>>>>>>>> qtot -2.779
>>>>>>>>> 889 MNH3 55 LYP MNZ1 825 0
>>>>>>>>> 8.517
>>>>>>>>> ;
>>>>>>>>> qtot -2.779
>>>>>>>>> 890 MNH3 55 LYP MNZ2 825 0
>>>>>>>>> 8.517
>>>>>>>>> ;
>>>>>>>>> qtot -2.779
>>>>>>>>> 891 amber99_39 55 LYP NZ 825
>>>>>>>>> -0.3854 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.165
>>>>>>>>> 892 amber99_17 55 LYP HZ1 826
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -2.825
>>>>>>>>> 893 amber99_17 55 LYP HZ2 827
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -2.485
>>>>>>>>> 894 amber99_17 55 LYP HZ3 828
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -2.145
>>>>>>>>>
>>>>>>>>> The atoms around nr 1291 are also the tip of a lysine side
>>>>>>>>> chain.
>>>>>>>>> 1291 amber99_11 80 LYP CE 1189 -0.0143
>>>>>>>>> 14.026
>>>>>>>>> ;
>>>>>>>>> qtot -4.006
>>>>>>>>> 1292 amber99_28 80 LYP HE1 1190
>>>>>>>>> 0.1135 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.893
>>>>>>>>> 1293 amber99_28 80 LYP HE2 1191
>>>>>>>>> 0.1135 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.779
>>>>>>>>> 1294 MNH3 80 LYP MNZ1 1192 0
>>>>>>>>> 8.517
>>>>>>>>> ;
>>>>>>>>> qtot -3.779
>>>>>>>>> 1295 MNH3 80 LYP MNZ2 1192 0
>>>>>>>>> 8.517
>>>>>>>>> ;
>>>>>>>>> qtot -3.779
>>>>>>>>> 1296 amber99_39 80 LYP NZ 1192
>>>>>>>>> -0.3854 0
>>>>>>>>> ;
>>>>>>>>> qtot -4.165
>>>>>>>>> 1297 amber99_17 80 LYP HZ1 1193
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.825
>>>>>>>>> 1298 amber99_17 80 LYP HZ2 1194
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.485
>>>>>>>>> 1299 amber99_17 80 LYP HZ3 1195
>>>>>>>>> 0.34 0
>>>>>>>>> ;
>>>>>>>>> qtot -3.145
>>>>>>>>>
>>>>>>>>> Jochne
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>> Berk Hess wrote:
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>
>>>>>>>>>> Hi,
>>>>>>>>>>
>>>>>>>>>> Did you try to rerun configure and recompile everything?
>>>>>>>>>> I just compared EM to an older version and got exactly the same
>>>>>>>>>> results.
>>>>>>>>>>
>>>>>>>>>> Berk
>>>>>>>>>>
>>>>>>>>>> Jochen Hub wrote:
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>> Hi Ran,
>>>>>>>>>>>
>>>>>>>>>>> did you also have these problems only during energy
>>>>>>>>>>> minimization? I
>>>>>>>>>>> asking since I want to use the present git for longer
>>>>>>>>>>> production
>>>>>>>>>>> runs?
>>>>>>>>>>>
>>>>>>>>>>> Best,
>>>>>>>>>>> Jochen
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>> Ran Friedman wrote:
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>>>> Hi,
>>>>>>>>>>>>
>>>>>>>>>>>> I'm fighting these issues as well.
>>>>>>>>>>>>
>>>>>>>>>>>> With the new git version I either get a segfault or really
>>>>>>>>>>>> unrealistic
>>>>>>>>>>>> structures. For example, I saw that a hydrogen bound to
>>>>>>>>>>>> nitrogen
>>>>>>>>>>>> near a
>>>>>>>>>>>> C-terminal carboxylate group was located at the same place
>>>>>>>>>>>> of a
>>>>>>>>>>>> carboxylate oxygen (!) following minimisation. As this
>>>>>>>>>>>> happened in
>>>>>>>>>>>> more
>>>>>>>>>>>> than one structure I guess something is broken. It's as if
>>>>>>>>>>>> minimisation
>>>>>>>>>>>> doesn't take the LJ parameters into account.
>>>>>>>>>>>>
>>>>>>>>>>>> Ran.
>>>>>>>>>>>>
>>>>>>>>>>>> Jochen Hub wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>> Hi,
>>>>>>>>>>>>>
>>>>>>>>>>>>> mdrun of the present git version stops during energy
>>>>>>>>>>>>> minimization in
>>>>>>>>>>>>> Segfaul/Bus error on my Mac. Gromacs 405 from a few weeks ago
>>>>>>>>>>>>> works
>>>>>>>>>>>>> fine with the same tpr and the same configure options. gdb
>>>>>>>>>>>>> reports
>>>>>>>>>>>>> some error in the assembly loops, I am not sure what the
>>>>>>>>>>>>> error
>>>>>>>>>>>>> means
>>>>>>>>>>>>> however:
>>>>>>>>>>>>>
>>>>>>>>>>>>> gcc version 4.0.1 (Apple Inc. build 5490)
>>>>>>>>>>>>> -------------------------------------------
>>>>>>>>>>>>> Program received signal EXC_BAD_INSTRUCTION, Illegal
>>>>>>>>>>>>> instruction/operand.
>>>>>>>>>>>>> _nb_kernel_ia32_3dnow_test_asm () at
>>>>>>>>>>>>> nb_kernel_ia32_3dnow_test_asm.s:23
>>>>>>>>>>>>> 23 nb_kernel_ia32_3dnow_test_asm.s: No such file or
>>>>>>>>>>>>> directory.
>>>>>>>>>>>>> in nb_kernel_ia32_3dnow_test_asm.s
>>>>>>>>>>>>> (gdb) bt
>>>>>>>>>>>>> #0 _nb_kernel_ia32_3dnow_test_asm () at
>>>>>>>>>>>>> nb_kernel_ia32_3dnow_test_asm.s:23
>>>>>>>>>>>>> #1 0x002682a0 in nb_kernel_ia32_3dnow_test
>>>>>>>>>>>>> (log=0xa08834c0) at
>>>>>>>>>>>>> nb_kernel_ia32_3dnow.c:186
>>>>>>>>>>>>> #2 0x0026831f in nb_kernel_setup_ia32_3dnow (log=0xa08834c0,
>>>>>>>>>>>>> list=0x804200) at nb_kernel_ia32_3dnow.c:212
>>>>>>>>>>>>> #3 0x00268bb2 in gmx_setup_kernels (fplog=0xa08834c0) at
>>>>>>>>>>>>> nonbonded.c:248
>>>>>>>>>>>>> #4 0x00057899 in init_forcerec (fp=0xa08834c0, fr=0x81d000,
>>>>>>>>>>>>> fcd=0x503130, ir=0x801a00, mtop=0x501860, cr=0x500330,
>>>>>>>>>>>>> box=0xbfffd76c,
>>>>>>>>>>>>> bMolEpot=0, tabfn=0x500590 "table.xvg", tabpfn=0x5005b0
>>>>>>>>>>>>> "tablep.xvg",
>>>>>>>>>>>>> tabbfn=0x5005d0 "table.xvg", bNoSolvOpt=0, print_force=-1) at
>>>>>>>>>>>>> force.c:1436
>>>>>>>>>>>>> #5 0x00012529 in mdrunner (fplog=0xa08834c0, cr=0x500330,
>>>>>>>>>>>>> nfile=27,
>>>>>>>>>>>>> fnm=0x2a06c0, bVerbose=1, bCompact=1, ddxyz=0xbfffe8e0,
>>>>>>>>>>>>> dd_node_order=1, rdd=0, rconstr=0, dddlb_opt=0x2801b1 "auto",
>>>>>>>>>>>>> dlb_scale=0.800000012, ddcsx=0x0, ddcsy=0x0, ddcsz=0x0,
>>>>>>>>>>>>> nstepout=100,
>>>>>>>>>>>>> ed=0x0, repl_ex_nst=0, repl_ex_seed=-1, pforce=-1,
>>>>>>>>>>>>> cpt_period=15,
>>>>>>>>>>>>> max_hours=-1, Flags=7168) at md.c:305
>>>>>>>>>>>>> #6 0x00018ce0 in main (argc=6, argv=0xbfffe94c) at
>>>>>>>>>>>>> mdrun.c:496
>>>>>>>>>>>>> Current language: auto; currently asm
>>>>>>>>>>>>>
>>>>>>>>>>>>> Newer gccs (4.2 or 4.4) did not help. Compiling in 64bit did
>>>>>>>>>>>>> not
>>>>>>>>>>>>> help
>>>>>>>>>>>>> either.
>>>>>>>>>>>>>
>>>>>>>>>>>>> Any suggestions?
>>>>>>>>>>>>>
>>>>>>>>>>>>> Thanks a lot,
>>>>>>>>>>>>>
>>>>>>>>>>>>> Jochen
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>>>
>>>>>>>>>>>> _______________________________________________
>>>>>>>>>>>> gmx-developers mailing list
>>>>>>>>>>>> gmx-developers at gromacs.org
>>>>>>>>>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>>>>>>>>> Please don't post (un)subscribe requests to the list. Use the
>>>>>>>>>>>> www
>>>>>>>>>>>> interface or send it to gmx-developers-request at gromacs.org.
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>>
>>>>>>>>>>>
>>>>>>>>>> _______________________________________________
>>>>>>>>>> gmx-developers mailing list
>>>>>>>>>> gmx-developers at gromacs.org
>>>>>>>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>>>>>>> Please don't post (un)subscribe requests to the list. Use the
>>>>>>>>>> www
>>>>>>>>>> interface or send it to gmx-developers-request at gromacs.org.
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>>
>>>>>>>>>
>>>>>>>> _______________________________________________
>>>>>>>> gmx-developers mailing list
>>>>>>>> gmx-developers at gromacs.org
>>>>>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>>>>> Please don't post (un)subscribe requests to the list. Use the www
>>>>>>>> interface or send it to gmx-developers-request at gromacs.org.
>>>>>>>>
>>>>>>>>
>>>>>>>>
>>>>>> _______________________________________________
>>>>>> gmx-developers mailing list
>>>>>> gmx-developers at gromacs.org
>>>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>>> Please don't post (un)subscribe requests to the list. Use the
>>>>>> www interface or send it to gmx-developers-request at gromacs.org.
>>>>>>
>>>>>>
>>>>> --
>>>>> ---------------------------------------------------
>>>>> Dr. Jochen Hub
>>>>> Molecular Biophysics group
>>>>> Dept. of Cell & Molecular Biology
>>>>> Uppsala University. Box 596, 75124 Uppsala, Sweden.
>>>>> Phone: +46-18-4714451 Fax: +46-18-511755
>>>>> ---------------------------------------------------
>>>>>
>>>>> _______________________________________________
>>>>> gmx-developers mailing list
>>>>> gmx-developers at gromacs.org
>>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>>> Please don't post (un)subscribe requests to the list. Use the
>>>>> www interface or send it to gmx-developers-request at gromacs.org.
>>>>>
>>>>>
>>>> _______________________________________________
>>>> gmx-developers mailing list
>>>> gmx-developers at gromacs.org
>>>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>>>> Please don't post (un)subscribe requests to the list. Use the www
>>>> interface or send it to gmx-developers-request at gromacs.org.
>>>>
>>>
>>
>> _______________________________________________
>> gmx-developers mailing list
>> gmx-developers at gromacs.org
>> http://lists.gromacs.org/mailman/listinfo/gmx-developers
>> Please don't post (un)subscribe requests to the list. Use the www
>> interface or send it to gmx-developers-request at gromacs.org.
>>
>
>
More information about the gromacs.org_gmx-developers
mailing list