I had a conversation with Taisung, which I thank warmly for the detailed
explanation of the situation.
I have a further request "to whom it may concern": is it possible to have
AMBER output DV/DL values even if they are exactly 0.0 when doing TI?
I'll sum up here the conversation with Taisung, as a reference to other
users of AMBER 2022, and also to better clarify the request I just made!
- Currently, AMBER does not report the DV/DL value when precisely 0.
- Enabling the default lambda-scheduling in AMBER 2022 (gti_lam_sch=1),
DV/DL will be precisely 0 at lambda=0 and lambda=1 and hence no DV/DL
values are written in the output file at this lambda values.
This is an under-the-hood changed behavior that is not
explicitly stated in the manual. Prior to AMBER 2022, by default, the
lambda scheduling was applied only to the direct-space non-bonded terms,
which almost account for all problematic TI singularity problems. To be
consistent, in AMBER 2022 this behavior was extended by default to all
interactions. Hence the same simulation with the same input files will
give different results (numerically, not substantially considering all
lambdas) with AMBER 2020 vs 2022. And in particular, the same simulation
with default lambda scheduling will not print any DV/DL valuers at lambda=0
and lambda=1 in AMBER 2022, while it printed DV/DL in AMBER 2020.
- To perform the analysis of DV/DL / MBAR data, users often rely on
external (or in-house developed) libraries. I can speak mainly for the
alchemlyb library, where I'm contributing to the updating of the AMBER
output parser.
When parsing AMBER output for different lambda values, we rely on reading
DV/DL data points along with MBAR information (if present). Simulations at
lambda=0 and 1 are still needed to collect MBAR data, and also data points
at these lambda values are needed, even if they are 0.0, to compute the
integral if we use the trapezoidal rule for example. It would be
particularly difficult to implement a proper check of the kind of
simulation that was run, as only checking if gti_lam_sch=1 is not
sufficient. Different custom-made scheduling can be developed by the user.
Given the flexibility of lambda-scheduling, it is possible to have DV/DL
that are exactly 0 at different lambda values, and a proper check for that
would be prone to bugs and errors.
Hence my request, is it possible to have AMBER always print DV/DL values
when TI is run, by default? I think codewise the change will be minimal,
but it will maintain compatibility with current workflows for the analysis
of TI simulations.
Thank you all and have a nice day!
*Domenico Marson*, Ph.D.
Ricercatore T.D. (A) | Research Fellow
Università degli Studi di Trieste | University of Trieste
Department of Engineering and Architecture (DEA)
Molecular Biology and Nanotechnology Laboratory @ UniTS (MolBNL.UniTs)
Via Valerio, 10 - 34127 Trieste (Italy)
domenico.marson.dia.units.it
+39 040 5583437
Skype: domenicomars
On Fri, Sep 23, 2022 at 8:10 PM Domenico Marson <domenico87.gmail.com>
wrote:
> Dear all,
> I just installed and tested Amber22 in two different systems (Kubuntu
> 22.04.1 and CentOS 7), observing the same behavior.
>
> I restarted some TI simulations that I previously run with Amber20 because
> I wanted to try the new gti_scale_beta and gti_cut_sc options.
>
> The simulations were previously run with Amber20 with just gti_lam_sch=1
> and the default scheduling, without any apparent problem.
>
> Now, if I restart the simulation with Amber22 keeping gti_lam_sch=1 (with
> or without the new options, it doesn't change), the simulation runs but
> there are no DV/DL values reported at every ntpr. The MBAR calculation
> seems fine, it's just the DV/DL that's missing. If I switch to
> gti_lam_sch=0, then the DV/DL values are printed.
>
> I don't know what I'm doing wrong, and would really appreciate any
> suggestion here.
>
>
> PS: also, I noticed that the SC_DERIV value is always 0 in the simulations
> ran with amber22.
>
> PPS: I tried to change the gti_cpu_output variable without success.
>
> Thank you for your time,
>
> Domenico
>
>
>
> *Domenico Marson*, Ph.D.
> Ricercatore T.D. (A) | Research Fellow
> Università degli Studi di Trieste | University of Trieste
> Department of Engineering and Architecture (DEA)
> Molecular Biology and Nanotechnology Laboratory @ UniTS (MolBNL.UniTs)
> Via Valerio, 10 - 34127 Trieste (Italy)
> domenico.marson.dia.units.it
> +39 040 5583437
> Skype: domenicomars
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Sep 27 2022 - 08:00:06 PDT
