[gmx-developers] Incorrectly large pressures at lambda=1 with perturbed masses

[Jochen Hub]

Hi Paul,

I started an issue with the test system here:

https://gitlab.com/gromacs/gromacs/-/issues/3959

Cheers,
Jochen

Am 02.03.21 um 10:17 schrieb Paul bauer:
> Hello Jochen,
> 
> I'll have a look at this now and see if I can bisect this to the 
> breaking change.
> Can you please open an issue on GitLab for this as well and include the 
> test system there?
> 
> Cheers
> 
> Paul
> 
> On 02/03/2021 10:14, Jochen Hub wrote:
>> Hi Bert,
>>
>> thanks for this quick suggestion. I just tried with the latest 
>> release-2021 from gitlab (2021.1-dev-20210219-d8cd16f885-unknown) and 
>> it did not help.
>>
>> Apparently, if I got this right, the fix you mentioned affects only 
>> the dH/dl calculations and should not change the simulation 
>> trajectory. Our issue affect the trajectory massively, as the box 
>> expands within picoseconds.
>>
>> Cheers,
>> Jochen
>>
>>
>> Am 02.03.21 um 09:41 schrieb Bert de Groot:
>>> Hi Jochen,
>>>
>>> sounds as if this could be related to 
>>> https://gitlab.com/gromacs/gromacs/-/issues/3943
>>>
>>> which was fixed only very recently:
>>>
>>> https://gitlab.com/gromacs/gromacs/-/merge_requests/1197
>>>
>>> could you perhaps test if that fixes your issue as well?
>>>
>>> cheers
>>>
>>> Bert
>>>
>>>
>>> On 3/2/21 9:35 AM, Jochen Hub wrote:
>>>> Dear Gromacs developers,
>>>>
>>>> we get incorrectly large pressures in FEP simulations in the 
>>>> lambda=1 state if we perturb the masses of all hydrogen atoms. 
>>>> Specifically, we simulate a pure-water box and perturb the masses of 
>>>> all H atoms from 1amu to 2amu. This happens with Gromacs 2019.6, 
>>>> 2020.5, and 2021, but NOT with 2018.6.
>>>>
>>>> The pressures are shown in a PDF here:
>>>>
>>>> https://cloud.hiz-saarland.de/s/NH6Mz8QKi39qBH7
>>>>
>>>> (The relaxation in the problematic cases is due to rapid box 
>>>> expansion).
>>>>
>>>> We have a test system here:
>>>>
>>>> https://cloud.hiz-saarland.de/s/o2j7K4GK7wojayi
>>>>
>>>> This does NOT happen if:
>>>>
>>>> - masses are not perturbed
>>>> - using Gromacs 2018.6
>>>> - the same state is defined to lambda=0 by modifying the topology
>>>>
>>>> To us, this looks like a potentially severe bug which was not 
>>>> detected because we typically perturb only few atoms and not 2/3 of 
>>>> the all atoms, as we do here.
>>>>
>>>> Is this issue known?
>>>>
>>>> Best wishes,
>>>> Johanna and Jochen
>>>>
>>>>
>>>
>>
> 

-- 
---------------------------------------------------
Prof. Dr. Jochen Hub
Theoretical Biophysics Group
Department of Physics, Saarland University
Campus E2 6, Zi. 4.11, 66123 Saarbruecken, Germany
Phone: +49 (0)681 302-2740
https://biophys.uni-saarland.de/
---------------------------------------------------