Re: [AMBER] Amber 16.12 and 12-6-4 force field Problem

From: Ross Walker <ross.rosswalker.co.uk>
Date: Sat, 13 Jan 2018 20:19:32 -0800

Hi Chris,

I fully agree with you. In this case the group in question are recipients of a share of royalties from AMBER licensing.

All the best
Ross

> On Jan 13, 2018, at 19:01, Chris Neale <candrewn.gmail.com> wrote:
>
> Dear Ross:
>
> I certainly understand your argument, but when we purchased a licence
> for amber16, I personally assumed that it was the seller who was ultimately
> responsible for ensuring that things worked as expected. I understand that
> mistakes can be made, and therefore we test all code fairly extensively
> before putting it to any production use, but accepting risk is not the same
> as assigning responsibility. My own thought is that as soon as a liscencer
> starts accepting money then it's a lot more dubious to start putting
> responsibility on outside code contributors, unless of course those code
> contributors are getting a cut of the licensing fee. So I wonder if you've
> considered that possibly those code sellers who freely accept added
> features to scientific code run the risk of wasting people's time unless
> they add comprehensive support (or require the use of an "-unlicensed" flag
> to pmemd)? I don't even use the 12-6-4 potential so I'm just offering my
> personal thoughts on something that hasn't affected me yet.
>
> Thank you,
> Chris.
>
> On Sat, Jan 13, 2018 at 7:40 PM, Pengfei Li <ambermailpengfei.gmail.com>
> wrote:
>
>> Hi Guanglin (and others),
>>
>> First, thank you for interested in using 12-6-4. However, I have to
>> clarify the issue to you and also to other AMBER users or developers.
>>
>> The 12-6-4 potential was proposed in 2013 by me and my former advisor
>> Prof. Kennie Merz. And then AMBER began to support the CPU simulations
>> about 12-6-4 from Amber14. Unfortunately, the GPU version of AMBER -
>> pmemd.cuda has never supported 12-6-4 potential, even up to now. We have
>> specify this (that 12-6-4 is not supported by pmemd.cuda) in all the Amber
>> manuals from Amber14. Please see Page 314 of Amber14 manual:
>> http://ambermd.org/doc12/Amber14.pdf <http://ambermd.org/doc12/Amber14.pdf>,
>> Page 350 in the AMBER 2015 manual: http://ambermd.org/doc12/Amber16.pdf <
>> http://ambermd.org/doc12/Amber16.pdf>, Page 360 in the AMBER 2016 manual:
>> http://ambermd.org/doc12/Amber16.pdf <http://ambermd.org/doc12/Amber16.pdf>,
>> and the Page 369 in the AMBER 2017 manual: http://ambermd.org/doc12/
>> Amber16.pdf <http://ambermd.org/doc12/Amber16.pdf>.
>>
>> I understand that support 12-6-4 simulation in GPU is an significant step
>> for making it have more users, but since I am not familiar with CUDA codes,
>> I was not able to add this function into pmemd.cuda. Later on Jason Swails,
>> another AMBER developer, told me that it would be good to make the code
>> raise an error when users try to use the 12-6-4 potential in pmemd.cuda, in
>> order to prevent tragedy happening - like some people did not notice the
>> information in the manual (that 12-6-4 is not supported by pmemd.cuda) but
>> perform 12-6-4 simulations with pmemd.cuda. Then he helped me add a couple
>> of lines in the code about this issue. That is why the pmemd.cuda code in
>> recent versions will raise an error when you trying to use 12-6-4 in it but
>> in former versions it will not raise an error.
>>
>> There is usually not a flat road for scientific research, and so it is for
>> the software development. We really tried to prevent the tragedies happen
>> even it may not prevent all of them. I am sorry to hear that you may waste
>> a lot of time for doing 12-6-4 simulations with pmemd.cuda. If so, I should
>> be honest about these results that they are not 12-6-4 but 12-6 because the
>> C4 terms will be ignored in these simulations.
>>
>> However, I also have to clarify that the CPU versions of AMBER, not matter
>> sander or pmemd, no matter the serial or MPI versions, they well support
>> the 12-6-4 simulations. And we have a test about this issue (under the
>> $AMBERHOME/test/lj_12-6-4/ directory). If you perform the tests about Amber
>> or/and AmberTools, no matter serial or parallel CPU versions, there should
>> be “cd lj_12-6-4 && ./Run.12-6-4” contained in the test log files which are
>> under the subdirectories under the $AMBERHOME/logs/ directory. So for
>> people who performed 12-6-4 simulations with the CPU version of AMBER, you
>> *DO NOT* have to worry about this issue.
>>
>> To my knowledge, only OpenMM supports doing 12-6-4 simulations with GPU
>> now. Jason has added this function into OpenMM and I really appreciate his
>> work. And if you want to perform GPU simulations with 12-6-4 potentials, I
>> would like to suggest you to use OpenMM. And if you need any help about
>> this, please feel free to let me know.
>>
>> I hope this email clarifies the issue.
>>
>> Kind regards,
>> Pengfei
>>
>>> On Jan 13, 2018, at 2:10 PM, Ross Walker <ross.rosswalker.co.uk> wrote:
>>>
>>> Hi Guanglin,
>>>
>>> Yes, your simulations using Amber 16.05 are incorrect - the 12-6-4
>> potentials would have been ignored and so the results are not reliable and
>> should be published. The 12-6-4 potential has never been supported by the
>> GPU code and still isn't. It was the responsibility of the person who added
>> 12-6-4 support to pmemd to also add traps to make sure other parts of code
>> that did not support their potential would quit if one tried to use it.
>> Unfortunately they did not do that, nor was there a warning added to the
>> tutorial that GPU support was not present and so unfortunately a number of
>> people will have unwittingly run simulations believing they were using the
>> 12-6-4 parameters. The trap was added in the latest update when the
>> omission was spotted.
>>>
>>> Such failures should serve as a warning to anyone adding features to
>> scientific codes where they are not writing comprehensive support - you run
>> the risk of wasting lots of people's time and, in some cases causing people
>> to have to retract papers. It is firmly the responsibility of the person
>> adding the new methodology to properly deal with this.
>>>
>>> In terms of running the 12-6-4 potentials with the GPU code this
>> requires someone to write, test, optimize and validate the code to support
>> this. I have brought this up multiple times but nobody seems to want to
>> volunteer to do it. As such at present the 12-6-4 potential is only
>> available in the CPU code, and in my opinion this should be made clear in
>> anything that references the 12-6-4 potential until such a time that it is
>> supported.
>>>
>>> I would also note that any papers published that used the 12-6-4
>> potential and relied on the PMEMD GPU code should be retracted.
>>>
>>> All the best
>>> Ross
>>>
>>>> On Jan 11, 2018, at 22:01, Guanglin Kuang <guanglin.kth.se> wrote:
>>>>
>>>> Hi David,
>>>>
>>>> I'm afraid I don't fully get your points.
>>>>
>>>> When I follow the instructions on "http://ambermd.org/tutorials/
>> advanced/tutorial20/12_6_4.htm", the same simulation can run on Amber
>> 16.05, but it failed on Amber 16.12.
>>>>
>>>> If 12-6-4 potentials are not supported in the CUDA code, are the
>> results produced by Amber 16.05 with CUDA GPU for a system containing these
>> potentials problematic and not reasonable at all? If not, how can we make
>> it run on Amber 16.12?
>>>>
>>>> Thanks!
>>>> Best regards,
>>>> Guanglin
>>>>
>>>>
>>>> ________________________________________
>>>> From: David Cerutti <dscerutti.gmail.com>
>>>> Sent: Friday, January 12, 2018 6:11 AM
>>>> To: AMBER Mailing List
>>>> Subject: Re: [AMBER] Amber 16.12 and 12-6-4 force field Problem
>>>>
>>>> We've just had another discussion about this problem. The problem is
>> that
>>>> there was never any support for 12-6-4 potentials in the CUDA code, so
>> you
>>>> haven't lost any functionality. The bug was that the code was still
>>>> allowing you to run topologies containing those parameters without
>> actually
>>>> changing the potential at all, so you might have thought that they were
>>>> part of the simulation even though they weren't.
>>>>
>>>> On Thu, Jan 11, 2018 at 9:49 PM, Guanglin Kuang <guanglin.kth.se>
>> wrote:
>>>>
>>>>> ?
>>>>>
>>>>> Dear Amber Users,
>>>>>
>>>>>
>>>>> I just updated Amber 16.05 and AmberTools 16.13 to the latest Amber
>> 16.12
>>>>> and AmberTools 17.08.
>>>>>
>>>>>
>>>>> Now the 12-6-4 force field parameters for ions seem to be not
>> supported in
>>>>> Amber 16.12, with the following error in the output file:
>>>>>
>>>>> CUDA (GPU): Implementation does not support 12-6-4 LJ potential.
>>>>> The topology contains 12-6-4 coefficients and must be
>> changed.
>>>>>
>>>>> Input errors occurred. Terminating execution.
>>>>>
>>>>> Anyone encounter this problem and any solutions?
>>>>>
>>>>> Thank you!
>>>>> Best regards,
>>>>> Guanglin?
>>>>>
>>>>>
>>>>>
>>>>> _______________________________________________
>>>>> AMBER mailing list
>>>>> AMBER.ambermd.org
>>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>>
>>>> _______________________________________________
>>>> AMBER mailing list
>>>> AMBER.ambermd.org
>>>> http://lists.ambermd.org/mailman/listinfo/amber
>>>
>>>
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber


_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sat Jan 13 2018 - 20:30:02 PST
Custom Search