plumed / plumed2 Goto Github PK

This is because it is read in a way different from other Actions. Should be fixed!

Is anyone having trouble with analysis/rt43? Davide told me that it is failing but, for me at least, it is working fine on the v2.0 branch. @davidebr

Using domain decomposition and implicit solvent simulations in gromacs it could happen that a domain is temporarily empty thus resulting in null pointers. But null pointer make plumed2 crash when check_null is done. Do we need it?

It would be useful for both replica exchange and multiple walkers (see #12)

This is just to test the tracking system

+++ Internal PLUMED error
+++ file Grid.cpp, line 170
+++ message: assertion failed indices[i]>=0 && indices[i]<nbin_[i]

when a CV fall outside the GRID the error should be more explicative: which cv? and possible which replica?

This is probably done effectively allowing a Langevin thermostat in DAFED when it will be merged (#9).

(started in #19, I open a new thread for it)

The gradients of ENERGY are never stored, so that it is not possible to use the adaptive hills based on projections of gradients. It is possible to remove this constraint, but it takes a bit of work and, to me, it is not clear if it is useful.

(pointed out by Apratim Bhattacharya; I just put it here to test issues in github)

Replica exchange with gromacs and multiple processors per replica is stuck after first exchange if plumed.dat file is empty. Observed with gromacs 4.6 and plumed 2.0b0.

This was discussed in #25 and was closed adding a check that crashes the code. It would be nice to implement the possibility to do something inside plumed upon exchange (need to change the MD patch to pass this extra info).

This could be a simple way to emulate the -interval option of driver in plumed 1. Doing it in the input file would allow to do it on the fly during the simulation.

We should explain in the manual what is going to happen when one types "make install" (where are files copied etc). Also I think we can say something about how to use modules ("module load ...") to have concurrent plumed versions.

It can accelerate a lot restart when one has a lot of accumulated hills. It is already possible to do it using GRID_RFILE (see f623f3c). However, I think that it should be integrated with RESTART procedure and it should scale properly bias potential when restarting a well-tempered metadynamics with a different BIASFACTOR

It would be nice to import groups from ndx files in gromacs format

@gtribello I got this warning on master branch:
./MultiColvarFunction.h:58:8: warning: 'PLMD::multicolvar::MultiColvarFunction::addOrientationDerivatives' hides overloaded virtual function [-Woverloaded-virtual]
void addOrientationDerivatives( const unsigned& iatom, const std::vector& der );

This is possibly a bug (it is not clear which function should be called). Can you fix it?
Thanks!
Giovanni

if you try to run "plumed simplemd help" the code crashes, while with --help/-h it works

Since 68473e2 we can use regexps in argument selection.

This could be extended to atom selection also. This, in combination with atom selections by name (#32) would allow for a very powerful selection syntax, e.g.:
ATOMS=(ALA1:.) # the whole residue
ATOMS=(.:CA) # all the C-alpha
ATOMS=(.:(CA)|(CB)) # all the C-alpha and C-beta
ATOMS=(.:H.) # all hydrogens
ATOMS=(.:[^H].*) # all non-hydrogens

I experimented a bit on BlueGeneQ and it seems that we could gain performance in atom distribution. In particular, an useful trick I found useful is to use asynchronous distribution of atoms to a subset of the processors and then using an intra-group broadcast. This is not difficult to implement, but it is a bit difficult to choose the parameters.

Now there are two routines, one that is used when align==displace (faster) and another one used when align!=displace (slower). Difference in speed is large (x10-50). We should either optimize the slow routine or implement align!=displace in the fast one.

I had a look at autoconf some time ago. It is not really straightforward, but since the number of external optional libraries we are linking grows (matheval, almost, regexp, possibly molfile, what else...) it is probably a good idea to try to automatize the search and the setting of -D__PLUMED_HAS_WHATEVER macros.

I think we should take the Lausanne tutorial (sites.google.com/site/plumedtutorial2010/) and translate it in plumed 2 syntax. It could be incorporated in the manual (in the How To section).

I want to allow one to write
GROUP ATOMS=10-1:-1
instead of explicitly
GROUP ATOMS=10,9,8,7,6,5,4,3,2,1

I find this useful in some case with WHOLEMOLECULES, where order of the atoms needs to be correct and backward ordering is sometime necessary.

Since all the plumed output files are textual, one could gain a lot of space using gzip library to output directly compressed files. Then it is easy to restore them with gzcat (also from gnuplot)

I think we should be able to select atoms by residue:name syntax, e.g.
ALA1:CA

As of now, this command should be called before initialization. Since it should point to a variable that is possibly changed at every step, I think it makes more sense to call it after initialization.

I think this should be considered as a bug and I want to fix it in 2.0.x. In this way, in case someone integrates plumed 2.0.x with an external MD code, behavior will be consistent in the future.

I pushed a fix for this (392bac3) where I also changed accordingly the Driver.
I will merge it to v2.0 in the next days. @gtribello can you check if this makes sense?

Giovanni

It should be possible to speed up the coordination (and other variables based on switching functions) when using RATIONAL switching functions with even exponents (e.g. 6 and 12) and d0=0, by avoiding the square root calculation. Since this is a very very common case, I want to change the code to allow this optimization.

The current patch for Quantum Espresso does not correctly add virial contribution.

They should possibly be usable also for 5 membered rings

Parallelization with OpenMP. I started doing some test and could be good to achieve better scaling.

See https://groups.google.com/forum/#!topic/plumed-users/-SpZbVLLCyo

Possibility to use the -multi framework of GROMACS to run multiple walkers metadynamics.
I am working on it

Currently all variables involving neighbor lists are not working properly with replica exchange. They should all crash the code, and one can easily disable neighbor lists to have a working code. However, we should make an effort such that it is possible to reset the neighbor lists after an exchange has been done.

This is related to #26

There are some info that presently should be specified on both MD input and PLUMED input. I think for as many as possible of them we should allow setting them directly from the MD code (as I learned from a student's error...). Examples are:
RESTART
TEMP in METAD and in HISTOGRAM
MOLINFO # this is very difficult, so probably still it is better to provide it with a pdb

Feel free to add to this list!

This will obviously require changing the MD patches adding calls such as
plumed_cmd("setTemp",&temperature)
This calls should be optional, so as to allow other patches (or MD codes directly calling plumed) to work correctly. I will try to find a way to do it with a minimal impact on current MD patches. I mean: if a patch is not setting temperature, plumed should not complain and TEMP should be set in the usual way from plumed.dat (this is easy). Conversely, if a patch is setting the temperature, an older (2.0.x) version of plumed should just ignore it (this requires some work).

I currently foresee two approaches.

(A) Probe the PLUMED version from the MD code:
plumed_cmd("getApiVersion",&version);
if(version>1) plumed_cmd("setTemp",&temp);

(B) Allow optional commands. Something like
plumed_cmd("?setTemp",&temp);
// will not crash if "setTemp" is not currently implemented

The second I think could be cleaner in the long run. The negative side is that it requires updating also plumed 2.0.x to be compatible with future MD patches so as to ignore all the unknown commands beginning with "?". Moreover, an erroneous
plumed_cmd("?setTem",&temp); // notice setTem instead of setTemp
will be just ignored. This scares me a bit.

Any comment is welcome!

Giovanni

It would be useful for both replica exchange and multiple walkers (see #12)

We already improved a lot removing many numerical derivatives and by decreasing the precision of output file. However, I am still finding errors where "0.000" is different from "-0.000". Probably the only solution is to use a sort of numerical diff, but then it will not be easy to set the threshold... Ideas are welcome!

Should be ported from plumed1. Could be implemented together with TAMD.

Something like
DUMPATOMS FILE=out.gro PRECISION=5
could be useful

Using TORSION, METAD and GRID with alanine dipeptide I get a segfault

file Grid.cpp, line 170
+++ message: assertion failed indices[i]>=0 && indices[i]<nbin_[i]

in particular the crash is the result of indices[i]=nbin_[i].

This doesn't happen when adaptive is not used.

@gtribello do you think it is possible (and easy) to have short form for all the command line options, in such a way that this also appears in the manual? E.g.
--hills|-H ....

Maybe if we add keys using "add(--hills|-H)" we could just modify the mechanism that does the matching. I have no idea if this is something easy or difficult to do...

Probably simple conditions such as "lower than" or "higher than" are flexible enough - at least in combination with e.g. MATHEVAL. It would be nice to have a clean exit (also from MD simulation).

@gtribello claims he has reimplemented them in plumed 2

When using HISTOGRAM with REWEIGHT_BIAS, the bias does not seems to be used unless also the temperature is used. I will push a tentative fix for it. @gtribello can you check? Thanks!

This was noted by @carlocamilloni and hopefully I will have something working in short time.

for example

bias: METAD ARG=xx ....

give as a result

+++ Internal PLUMED error
+++ file ActionWithValue.cpp, line 135
+++ message: assertion failed name!=getLabel(), You should never be calling this routine to retrieve the value

probably we should either give a clearer explanation of the error or allow this possibility

@davidebr started working on this and pushed a branch with a tentative implementation (see 1e857ce).

GROMACS is checkpointing regularly. When restarting, trajectories are properly appended also if the checkpoint has been saved earlier than the last trajectory snapshot. I.e., trajectories are cut at the proper time (and a small portion of the trajectory is repeated). I think it would be great to have something such that if the first column of a file is time (this is true for most plumed output files) the latest part of the file (past the checkpoint time) is deleted upon restart.