Migrated suggested edits for thermostat section

[JIMonroe]

Comments on thermostat pull request from @davidlmobley migrated here:

OK, I'm going through your edits, and have some comments in a couple categories...

Big picture:

I like it!
I don't think you can make it much shorter without losing detail we need
Probably needs more explicit recommendations/warnings (see below)
Probably needs more references
Things which need some attention:

Velocity rescaling: One of the big drawbacks, it says, is that particles do not know their previous thermal history and this makes dynamical properties impossible to measure. It is not explained and not obvious why this is as stated; you probably want at least a reference to further reading. Also, people interested in equilibrium properties (free energies, etc.) don't care about this, so it may be that you want more stringent disclaimers for these people. I think it would be safe to EXPLICITLY give a recommendation at the end of this paragraph such as, "We recommend that velocity rescaling thermostats such as (whatever we're calling this one) never be used for production level simulations, except perhaps during an initial equilibration phase, because they do not sample from the correct ensemble and lead to incorrect estimates of various computed properties [refs]." (I wrote this thinking about just item 1 within 4.4.3, but it probably applies to items 1-3). Maybe also we should be more specific about what we mean by "velocity rescaling" since the subsequent "Hoover" section is also explained as "velocity rescaling".
Do you need to separately note the "massive collision" and "gradual buffeting" versions of this?
Maybe you want to say something more positive about Langevin; right now there's not really a warning against it nor anything that positive, except that the statement "Careful consideration must be taken when choosing the friction damping parameter" could be interpreted as a warning.
I wonder if we should come down a little more strongly at the end of this thermostat discussion in terms of recommending certain types to be used/avoided -- e.g. a lot of these really shouldn't be used for most purposes and we should perhaps say that clearly and concisely.
Up for me writing a comment to include on how "small system" issues apply to certain types of free energy calculations (e.g. if I decouple a ligand, the decoupled ligand can trigger problems with thermostat issues)
Should we be referencing the "block of ice" problem?
Minor details: I should edit a few things in detail when we are ready:

Right now it says in the motivation that, basically, thermostats are used to change the temperature during a simulation, which is not really what we mean
I should just adjust some grammar/clarity issues
Basically, for this last category, I can do it in a separate PR once we merge. :)

[avisekdas]

My take on the topics of thermostat/barostat is the following. Let's stick to the basic issues and warn users about common mistakes and known issues that are non-trivial. We should be clear about some old implementations that do not sample proper ensembles. But let's not go into the details, for example issues related to free energy calculations will be too much information for a new user. Those issues are better left for more specialized documents. Thanks, Avisek

…

On Thu, May 10, 2018 at 10:16 AM, Jacob Monroe ***@***.***> wrote: Comments on thermostat pull request from @davidlmobley <https://github.com/davidlmobley> migrated here: OK, I'm going through your edits, and have some comments in a couple categories... Big picture: I like it! I don't think you can make it much shorter without losing detail we need Probably needs more explicit recommendations/warnings (see below) Probably needs more references Things which need some attention: Velocity rescaling: One of the big drawbacks, it says, is that particles do not know their previous thermal history and this makes dynamical properties impossible to measure. It is not explained and not obvious why this is as stated; you probably want at least a reference to further reading. Also, people interested in equilibrium properties (free energies, etc.) don't care about this, so it may be that you want more stringent disclaimers for these people. I think it would be safe to EXPLICITLY give a recommendation at the end of this paragraph such as, "We recommend that velocity rescaling thermostats such as (whatever we're calling this one) never be used for production level simulations, except perhaps during an initial equilibration phase, because they do not sample from the correct ensemble and lead to incorrect estimates of various computed properties [refs]." (I wrote this thinking about just item 1 within 4.4.3, but it probably applies to items 1-3). Maybe also we should be more specific about what we mean by "velocity rescaling" since the subsequent "Hoover" section is also explained as "velocity rescaling". Do you need to separately note the "massive collision" and "gradual buffeting" versions of this? Maybe you want to say something more positive about Langevin; right now there's not really a warning against it nor anything that positive, except that the statement "Careful consideration must be taken when choosing the friction damping parameter" could be interpreted as a warning. I wonder if we should come down a little more strongly at the end of this thermostat discussion in terms of recommending certain types to be used/avoided -- e.g. a lot of these really shouldn't be used for most purposes and we should perhaps say that clearly and concisely. Up for me writing a comment to include on how "small system" issues apply to certain types of free energy calculations (e.g. if I decouple a ligand, the decoupled ligand can trigger problems with thermostat issues) Should we be referencing the "block of ice" problem? Minor details: I should edit a few things in detail when we are ready: Right now it says in the motivation that, basically, thermostats are used to change the temperature during a simulation, which is not really what we mean I should just adjust some grammar/clarity issues Basically, for this last category, I can do it in a separate PR once we merge. :) — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#13>, or mute the thread <https://github.com/notifications/unsubscribe-auth/Ad5Xu8gKBkTUU5bErSEK4FiH0Uo5jxb5ks5tw8YogaJpZM4T5WyV> .

[davidlmobley]

I'll go through the thermostats section with an eye on my comments above and see what I think needs addressing in view of @avisekdas 's remarks, make a PR, and close this. Probably next week.

[davidlmobley]

Closed via 26.