Okay, guys, on the issue of pme vs. force shifts. I actually have done a
ton of work on this which is on hold right now so I can do all this other
software work. This last summer I did a couple of things in this area.
First of all, I carefully looked at what had been done with cutoffs,
analyzing the flaws in the extant methods. I started this because Lee was
rather peeved about a paper that came out of Daggett's lab last year
blasting pme and claiming a force shift method is okay. Well, Daggett's
force shift method, which is really fsh out of the east coast Brooks lab way
back, is not at all okay. I, however, with a huge amount of work, managed
to develop some functions that produce a smooth cutoff method that seems to
actually be okay if you use at least a 13 angstrom cutoff (I prefer 16, but
that is getting really spendy on time). I actually have several related
methods, all basically have smooth derivatives for everything, and have not
been done before. I have not been able to find anything wrong with these
cutoff methods other than that they are slow compared to pme. That sort of
slowed me down; I was looking at this stuff hard for two reasons: 1) to show
that you can't be cavalier, ala Daggett and just use any old cutoff, and 2)
if I could get sort-of-solid results, I would consider a really well
validated smooth cutoff method when pme runs completely out of gas on
scaling. Well, chances are that pme is going to go just about as far as
anything. So the methods I have are now more interesting to me in terms of
being able to understand the importance of various distances in
electrostatics. So what do I have that looks okay? Well, I have run plain
tip3p, 0.1 NaCl, dna, small peptides, gla fragments, various larger
proteins - somewhere around 30 systems when you consider different box sizes
and forcefields (rough guess - I did not recount). I have hundreds upon
hundreds of nsec of data on this stuff, running pme, truncation cuts,
daggett fsh, and my methods. Looking at rdf's, radius of gyration,
secondary structure, whatever ptraj will do, my stuff basically looks okay
(also visualization), but I have a max of something like 30 nsec on any one
system (the rdf's look very good, including salt rdf's - I have further
plans with Kirkwood G factors but need to do some code). Now the rub in all
this was that I ran a small peptide that daggett uses as a test case of
helicity - 17 aa's, lots of lys, supposedly something like 80-90% alpha by
CD. Well, for ff03 it is helix under pme; for ff99sb it is not (I could
actually fold it from random coil under ff03 and ff94, which we know
overestimates helix). This stopped me, and this was going to be a sensitive
secondary structure test (I did not run my systems on it since I could not
get a good result from pme). At this point I decided I really had to start
looking at some better structure markers - nmr probably, and I just ran out
of time before other things became pressing. So I have a variety of plans
here, but am just one guy, and the bloody ff's not agreeing on something
simple has really got me screwed up. But basically the delta between my
smooth cutoff stuff and pme looks smaller than the delta between ff99sb and
ff03 for several systems. Carlos and I have talked; I gather he does not
trust the CD, but maybe I misunderstood. So I do have some plans in this
area, but I stay kind of busy with this other stuff too...
Best Regards - Bob
----- Original Message -----
From: "Ross Walker" <ross.rosswalker.co.uk>
To: <amber-developers.scripps.edu>
Sent: Wednesday, February 28, 2007 7:15 PM
Subject: RE: amber-developers: amber performance
> Hi All,
>
> Tom C and I spent some time talking about this on Sunday evening. We
> pretty
> much came to the conclusion that we want to do is design a series of
> calculations that attempt to address the various issues that people simply
> "believe" at present. I.e. find something that is sensitive to single vs
> double precision and see if we can address whether single precision is
> okay
> or not. In addition we thought about answering PME vs force switch
> simulations. Here a salt water solution simulation might be useful as this
> would be an extreme test of the electrostatics and we could address
> whether
> a force switch cutoff is actually okay or if you really need pme. There
> are
> some other options etc in here that we could consider as well, such as
> timesteps etc. Is 2fs really bleeding edge? Note: in the DeShaw paper they
> use a 2.5fs time step with shake on just hydrogens.
>
> This I think is a good plan since we can do these calculations and then
> put
> something along the lines of - oh btw here is the performance of the
> simulations on all the different codes etc.
>
> A straight benchmark paper is unlikely to get published so I think
> wrapping
> it up in a paper that attempts to address issues regarding PME, time step,
> single vs double precision etc is a good way to go.
>
> Comments?
>
> All the best
> Ross
>
> /\
> \/
> |\oss Walker
>
> | HPC Consultant and Staff Scientist |
> | San Diego Supercomputer Center |
> | Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
> | http://www.rosswalker.co.uk | PGP Key available on request |
>
> Note: Electronic Mail is not secure, has no guarantee of delivery, may not
> be read every day, and should not be used for urgent or sensitive issues.
>
>> -----Original Message-----
>> From: owner-amber-developers.scripps.edu
>> [mailto:owner-amber-developers.scripps.edu] On Behalf Of Scott Brozell
>> Sent: Wednesday, February 28, 2007 14:24
>> To: amber-developers.scripps.edu
>> Subject: amber-developers: amber performance
>>
>> Hi,
>>
>> Today Kent Milfeld from TACC asked me about amber performance because
>> he saw that there is work underway for charmm's performance.
>> I told him about pmemd (which he had never heard of), namd,
>> gromacs, etc.
>> He was very receptive to the idea that namd has great advertising
>> and might not be targeting the best science.
>>
>> This follows on the discussion we had at the amber meeting regarding
>> a publication on good vs dubious science, pmemd, single vs
>> double precision,
>> scaling vs throughput, etc. I hope that the ball will be
>> rolling soon.
>> And I'm happy to help out.
>>
>> Scott
>>
>>
>>
>
>
>
Received on Sun Mar 04 2007 - 06:07:29 PST
