[Date Index] [Thread Index] [Author Index]
Re: Re: Constraints to parameters in FindFit?
Jill et al., In general, you do need global optimization to solve constrained model calibration problems. In the monograph http://www.springeronline.com/sgw/cda/frontpage/0,0,4-0-22-33656158-0,0.html?referer=www.springeronline.com/isbn/0-7923-3757-3 I discuss several related case studies. The short tutorial book http://www.lionhrtpub.com/books/globaloptimization.html includes a demo example that is [sounds] very similar to Jill's model. Please also see e.g., Globally optimized calibration of nonlinear models: techniques, software, and applications. Optimization Methods and Software 18 (2003) (3) 335-355. Regards, Janos Pinter At 05:18 AM 10/9/2004, Peter Valko wrote: >For sign restiction an old trick is to use instead of paramold the >expression Exp[paramnew] or -Exp[paramnew] in your model, and search >for paramnew (instead of paramold). > >P. > > >Jill Foley <efoley at princeton.edu> wrote in message >news:<ck5e57$oio$1 at smc.vnet.net>... > > Hi All, > > > > I am using FindFit to fit a series of peaks to some data. I would like > > to be able to constrain some of the parameters of my fit to correspond > > to physical reality. For example, some peaks should have a negative > > amplitude, others positive, where the amplitudes are the parameters > > that Mathematica is finding in FindFit. The peaks are all very near > > each other, so without any constraint, it is making the wrong ones > > negative. I'd like to specify that a given parameter should always be > > negative. I am already giving an initial guess of the proper sign, but > > it doesn't fix the problem. > > > > Please advise - Thanks! > > > > Jill.