On Thu, Jan 8, 2009 at 7:45 PM, Daniel Russel drussel@gmail.com wrote:
> On Jan 8, 2009, at 8:04 AM, Keren Lasker wrote: > > any thoughts on dealing with the derivatives at x=std? >> > Just setting it to 0 should do for 1 as it is 0 infinitely close to std and > we are doing floating point arithmetic. Or am I confused?
for f=const for x>std: i'd agree with daniel. anyways, the first derivative is not continuous, so i'd opt for 0, i.e. the value for x>std.
for f=f(x) for x>std: my suggestion (2-x^-2) or the morse potential would both be continuous for the 1st derivative. for the morse potential, one would probably need javi's dodgy exp() macro to make things faster (in embed). the loss in accuracy would probably be acceptable, as long as the same macro is taken for the derivatives.
cheers
frido
> > > >> >> On Thu, 8 Jan 2009, Friedrich Foerster wrote: >> >> hi all, >>> >>> when dealing with experimental restraints i often feel that harmonic >>> restraints are somewhat unsatisfactory: >>> experimental data can be wrong and thus there is not necessarily a >>> solution that fulfills all restraints (score=0). in those cases, a model >>> with a lower score is not necessarily any more correct than a higher scoring >>> one. thus, i think it'd be good to have an additional scoring function that >>> levels off if the feature exceeds a certain value. for example, i'd suggest >>> the following two functions alternatively to Harmonic: >>> 1. a function that is a Harmonic for x<std and constant for x>std >>> 2. a function that is a Harmonic for x<std and asymptotically approaches >>> a constant value for x-> inf: >>> e.g.: f=x^2 for x<1 and f=2-x^-2 for x>1 (would even be continuous in >>> 1st derivative) >>> >>> is anyone else also interested in something like that? >>> any better suggestions? >>> is any IMP core expert interested and willing to code that (including >>> analogous functions for Upper and Lower Harmonic)? probably i'd be one of >>> the slowest and clumsiest persons in doing that... >>> >>> thanks >>> >>> frido >>> >>> >>> -- >>> >>> Friedrich Foerster >>> Max-Planck Institut fuer Biochemie >>> Am Klopferspitz 18 >>> D-82152 Martinsried >>> >>> Tel: +49 89 8578 2651 >>> Fax: +49 89 8578 2641 >>> >>> foerster@biochem.mpg.de >>> >>> www.tomotronic.org >>> >>> >>> >>> >>> >>> >>> _______________________________________________ >>> IMP-dev mailing list >>> IMP-dev@salilab.org >>> https://salilab.org/mailman/listinfo/imp-dev >>> >> _______________________________________________ >> IMP-dev mailing list >> IMP-dev@salilab.org >> https://salilab.org/mailman/listinfo/imp-dev >> > > _______________________________________________ > IMP-dev mailing list > IMP-dev@salilab.org > https://salilab.org/mailman/listinfo/imp-dev > >