Hi Qin,
my main problem was that the opimization sometimes would cycle between
maximization and minimization (chaotic behavior)
As it turns out, it was a problem with the maximal step size restriction.
PB seems to be more sensitive to too large steps than GB.
I calculated the gradient of PB numerically and compared it with the
forcedump.dat gradient, and they are indeed identical.
So the problem was all on my side.
Cheers,
Holger
2012/12/4 <amber-request.ambermd.org>
> ---------------
>
> Message: 6
> Date: Mon, 3 Dec 2012 17:01:13 -0600
> From: Qin Cai <qcai.uci.edu>
> Subject: Re: [AMBER] pbsa forces printout from sander (amber12)
> To: "Ray Luo, Ph.D." <ray.luo.uci.edu>
> Cc: AMBER Mailing List <amber.ambermd.org>
> Message-ID: <6566148F-3729-4234-81AF-99199F5BE3AB.uci.edu>
> Content-Type: text/plain; charset=us-ascii
>
> Hi Holger,
>
> I used your input pasted here and was able to get forcedump.dat for both
> pb and gb and they are in the same format.
>
> The "Total Forces" sections gives the force on every atom. The numbers are
> different but still consistent between pb and gb.
>
> So, to better understand what your problem is, could you email us your
> topology and coordinate files as well as your output files? See if we can
> reproduce.
>
> All the best,
> Qin
>
> On Dec 3, 2012, at 11:53 AM, Ray Luo, Ph.D. wrote:
>
> > I'm forwarding this to Qin who is in charge of force computation in pbsa
> ...
> >
> > Ray
> >
> > ---------- Forwarded message ----------
> > From: Dr. Holger Kruse <kruse.ceitec.muni.cz>
> > Date: Mon, Dec 3, 2012 at 9:08 AM
> > Subject: [AMBER] pbsa forces printout from sander (amber12)
> > To: amber.ambermd.org
> >
> >
> > Hi everyone,
> >
> > I want to combine sander (amder12) with an external (self-written)
> > optimizer. For gbsa computations it works as expected and
> > I get basically the same result as the optimization with sander itself.
> > However, for pbsa computations my approach does not work.
> > (see below for inputs)
> >
> > For gbsa I generate the file "forcedump.dat", from which I read the
> "Total
> > Forces" section to get the gradient.
> >
> > For pbsa, reading simply the "forcedump.dat" file leads to a very chaotic
> > optimization behavior.
> > It seems the gradient is incomplete.
> > I know I can generate a "force.dat" file from the &pb section
> > with frcopt=2, but I dont know if I need it.
> > Sander also computes the forces 4 times with my input ("qE forces" are to
> > be found 4 times in force.dat)
> >
> > Any suggestions or hints?
> >
> >
> > Cheers,
> > Holger
> >
> > --------gbsa input
> > &cntrl
> > imin=1, ntmin=1, ntx=1, ntpr=10,
> > maxcyc=1, ncyc=1, igb=1, ntf=1, ntc=1,
> > ntb=0, irest=0
> > /
> > &debugf
> > do_debugf=1,dumpfrc=1
> > /
> > &end
> > ---------
> >
> > --------------pbsa input
> > Minimization PB calculation
> > &cntrl
> > imin=1, ntmin=1, ntx=1, ntpr=1,
> > maxcyc=1, ncyc=1, igb=10, ntf=1, ntc=1,
> > ntb=0, irest=0
> > /
> > &pb
> > npbverb=0, istrng=0, epsout=80.0,
> > epsin=1.0, sprob=1.6, radiopt=1,
> > space=0.5, nbuffer=0, accept=0.001,
> > cutnb=0, dbfopt=1, npopt=2,
> > maxitn=100,frcopt=2
> > /
> > &debugf
> > do_debugf=1,dumpfrc=1
> > /
> > &end
> > --------------
> > _______________________________________________
> > 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 Fri Dec 07 2012 - 01:30:02 PST
