[gmx-users] AWH with two bias potentials
Cédric Leyrat
Cedric.Leyrat at igf.cnrs.fr
Fri Feb 15 16:36:20 CET 2019
Thanks a lot for all the helpful information, I will decrease the
initial error and increase the diffusion. Actually before seeing your
email I had modified my .mdp file like this, which made mdrun behave
better (although the .edr file is still very large - about 7GB after 5
hours):
awh = yes
awh-nstout = 10000
awh-nbias = 1
awh1-ndim = 2
awh1-dim1-coord-index = 1
awh1-dim1-start = 0.0
awh1-dim1-end = 3.8
awh1-dim1-diffusion = 1.0e-7
awh1-error-init = 100.0
awh1-dim1-force-constant = 100000
awh1-dim2-coord-index = 2
awh1-dim2-start = 0.3
awh1-dim2-end = 3.8
awh1-dim2-diffusion = 1.0e-7
awh1-dim2-force-constant = 100000
awh1-target-cutoff = 50
awh1-target = cutoff
I will now further adjust the parameters based on what you said. Also
with the 2D bias, gmx awh now generates a 2D PMF as a XYZ triplet which
I can plot easily, so thanks also for that.
Cedric
Le 2/15/2019 à 3:50 PM, Berk Hess a écrit :
> Your initial error parameters is orders of magnitude too large. You don't expect an error of 10000 kJ/mol, or do you? 10 is more reasonable.
> The initial update size depends on this error squared. Likely your diffusion parameters are one or two order of magnitude too small, so you should increase those. But that still means the initial update size is a factor 1000 to 10000 to large. This likely leads to instabilities, which explains mdrun behavior.
>
> Berk
>
>
> ________________________________
> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Cédric Leyrat <Cedric.Leyrat at igf.cnrs.fr>
> Sent: Thursday, February 14, 2019 9:00 PM
> To: gromacs.org_gmx-users at maillist.sys.kth.se
> Subject: Re: [gmx-users] AWH with two bias potentials
>
> awh = yes
> awh-nstout = 10000
> awh-nbias = 1
> awh1-ndim = 2
> awh1-dim1-coord-index = 1
> awh1-dim1-start = 0.001
> awh1-dim1-end = 3.8
> awh1-dim1-diffusion = 1.0e-6
> awh1-error-init = 10000.0
> awh1-dim1-force-constant = 1000000
> awh1-dim2-coord-index = 2
> awh1-dim2-start = 0.3
> awh1-dim2-end = 3.8
> awh1-dim2-diffusion = 1.0e-6
> awh1-dim2-force-constant = 1000000
>
> However something very strange is happening with my job when I try to
> run it on the server. The simulation seems to run very slow, doesn't
> manage to use the GPUs at all, doesn't use the CPUs properly either and
> is using more than 95% of the machine's memory (normally it uses about
> 0.5%). Importantly, it also generates huge energy and checkpoint files
> within half-an-hour. I've killed the previous job and started again but
> it's doing the same thing.
>
> -bash-4.2$ ll -trh
> total 21G
> drwxr-xr-x. 2 cleyrat granier 4.0K Feb 14 19:53 restraints
> drwxr-xr-x. 2 cleyrat granier 4.0K Feb 14 19:53 toppar
> -rw-r--r--. 1 cleyrat granier 5.3M Feb 14 19:54 index_awh.ndx
> -rw-r--r--. 1 cleyrat granier 12M Feb 14 19:56 step6.6_equilibration.gro
> -rw-r--r--. 1 cleyrat granier 2.3K Feb 14 19:56 step7_production2_awh1.mdp
> -rw-r--r--. 1 cleyrat granier 2.2K Feb 14 19:56 step7_production2_awh4.mdp
> -rw-r--r--. 1 cleyrat granier 1.6K Feb 14 19:56 topol.top
> -rw-r--r--. 1 cleyrat granier 8.2M Feb 14 19:57 step7_awh4.tpr
> -rw-r--r--. 1 cleyrat granier 5.9M Feb 14 20:08 step7_awh4.trr
> -rw-r--r--. 1 cleyrat granier 4.9M Feb 14 20:19 step7_awh4.xtc
> -rw-r--r--. 1 cleyrat granier 2.2K Feb 14 20:19 step7_production2_awh5.mdp
> -rw-r--r--. 1 cleyrat granier 9.0G Feb 14 20:21 step7_awh4.cpt
> -rw-r--r--. 1 cleyrat granier 1.7G Feb 14 20:21 step7_awh4.edr
> -rw-r--r--. 1 cleyrat granier 36K Feb 14 20:21 pullx_step7_awh4.xvg
> -rw-r--r--. 1 cleyrat granier 43K Feb 14 20:21 pullf_step7_awh4.xvg
> -rw-r--r--. 1 cleyrat granier 83K Feb 14 20:21 step7_awh4.log
> -rw-r--r--. 1 cleyrat granier 16K Feb 14 20:23 mdout.mdp
> -rw-r--r--. 1 cleyrat granier 8.2M Feb 14 20:23 step7_awh5.tpr
> -rw-r--r--. 1 cleyrat granier 5.9M Feb 14 20:35 step7_awh5.trr
> -rw-------. 1 cleyrat granier 7.5K Feb 14 20:41 nohup.out
> -rw-r--r--. 1 cleyrat granier 6.1M Feb 14 20:49 step7_awh5.xtc
> -rw-r--r--. 1 cleyrat granier 2.2G Feb 14 20:50 step7_awh5.edr
> -rw-r--r--. 1 cleyrat granier 90K Feb 14 20:50 step7_awh5.log
> -rw-r--r--. 1 cleyrat granier 47K Feb 14 20:50 pullx_step7_awh5.xvg
> -rw-r--r--. 1 cleyrat granier 55K Feb 14 20:50 pullf_step7_awh5.xvg
> -rw-r--r--. 1 cleyrat granier 7.5G Feb 14 20:51 step7_awh5_step96200.cpt
> -bash-4.2$
>
> Is this some kind of bug or am I doing something wrong with the parameters?
>
> Cedric
>
>
>
> Le 2/14/2019 à 8:09 PM, Berk Hess a écrit :
>> Note though, as also noted in the manual, you need to be careful what you wish for. 2D biasing leads to uniform sampling of the requested rectangular area. Often you want to suppress sampling of high free-energy (unphysical or uninteresting) regions in this rectangle by using a cutoff on the free-energy.
>>
>> ________________________________
>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Berk Hess <gmx3 at hotmail.com>
>> Sent: Thursday, February 14, 2019 5:41 PM
>> To: gmx-users at gromacs.org
>> Subject: Re: [gmx-users] AWH with two bias potentials
>>
>> I see now that you used awh-nbias=2 with awh-ndim=1. Thus you asked for 2 independent 1D AWH biases and PMFs. You might be able to get a 2D PMF from that, but that is complicated.
>> I think that what you want is one 2D bias: awh-nbias=1 with awh-ndim=2.
>>
>> Berk
>>
>> ________________________________
>> From: gromacs.org_gmx-users-bounces at maillist.sys.kth.se <gromacs.org_gmx-users-bounces at maillist.sys.kth.se> on behalf of Cédric Leyrat <Cedric.Leyrat at igf.cnrs.fr>
>> Sent: Thursday, February 14, 2019 4:32 PM
>> To: gromacs.org_gmx-users at maillist.sys.kth.se
>> Subject: Re: [gmx-users] AWH with two bias potentials
>>
>> Dear Berk,
>>
>> I do get a set of PMF files at different times for each reaction
>> coordinate from gmx awh. Each file has two columns (x=coordinate 1,
>> y=PMF or x=coordinate 2, y=PMF). My coding and maths skills are quite
>> limited so if you have a script that can convert these data to a 2D
>> format, I would be happy to use it.
>>
>> Cedric
>>
>> Le 2/14/2019 à 3:27 PM, Berk Hess a écrit :
>>> gmx awh writes out the 2D (or any-D) PMF to xvg files for different times. The 1D nature of xvg files is not convenient for visualization though, so you need to write a script to convert it to you favorite 2D data format. I probably have a script somewhere for some 2D format (python?) if you can't write it yourself.
>>>
>>> Berk
>>>
>> --
>> Cédric LEYRAT, Ph.D
>> Chargé de recherche - INSERM
>> PHARMACOLOGY AND STRUCTURAL BIOLOGY OF MEMBRANE PROTEINS
>> Institut de Génomique Fonctionnelle
>> U1191 INSERM
>> 141, rue de la cardonille
>> 34094 Montpellier Cedex 05 - France
>> tel: +33 (0)434359272
>> www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat>>>
>>
>> --
>> Gromacs Users mailing list
>>
>> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.
>> --
>> Gromacs Users mailing list
>>
>> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>>
>> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>>
>> * For (un)subscribe requests visit
>> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.
> --
> Cédric LEYRAT, Ph.D
> Chargé de recherche - INSERM
> PHARMACOLOGY AND STRUCTURAL BIOLOGY OF MEMBRANE PROTEINS
> Institut de Génomique Fonctionnelle
> U1191 INSERM
> 141, rue de la cardonille
> 34094 Montpellier Cedex 05 - France
> tel: +33 (0)434359272
> www.researchgate.net/profile/Cedric_Leyrat<http://www.researchgate.net/profile/Cedric_Leyrat>
>
> --
> Gromacs Users mailing list
>
> * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting!
>
> * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
>
> * For (un)subscribe requests visit
> https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-request at gromacs.org.
--
Cédric LEYRAT, Ph.D
Chargé de recherche - INSERM
PHARMACOLOGY AND STRUCTURAL BIOLOGY OF MEMBRANE PROTEINS
Institut de Génomique Fonctionnelle
U1191 INSERM
141, rue de la cardonille
34094 Montpellier Cedex 05 - France
tel: +33 (0)434359272
www.researchgate.net/profile/Cedric_Leyrat
More information about the gromacs.org_gmx-users
mailing list