There's no reason that the optimum cannot be at the bounds. Bounded problems really do sometimes have solutions on those bounds.
Compute the unconstrained gradient of your objective function at the bounds and see if the function is reduced when going across the bounds. The function here is assumed to be the neg. LL that one is MINIMIZING. There are many aspects of R, optimization in particular, where it would help if we required an operator's license. It is not a magical spell that solves problems automatically, but a tool much like a chainsaw that can as easily cut your arm off as saw a log. JN On 05/01/2012 06:00 AM, [email protected] wrote: > Message: 54 Date: Mon, 30 Apr 2012 08:30:24 -0700 (PDT) From: barb > <[email protected]> > To: [email protected] Subject: [R] Optim (fct): Parameters=LowerBounds!? > Message-ID: > <[email protected]> Content-Type: text/plain; > charset=UTF-8 Hey, i > am trying to do the MLE for Garch and have a problem with the optim function. > Initally i > tried optim with Method=BFGS. Reading trhough the forum i found out i would > neet bounds. > So i went on with Method=L-BFGS-B. But now my parameters equal the lower > bounds. >> > out <- optim(par=initial, fn=LogLik.GARCH,X=X, P=1, Q=1, method = >> > "L-BFGS-B",lower =c(0.01,0.01,0.01), upper =c(0.99,0.99,0.99)) >> > out > ----------------------------------- > *$par > [1] 0.01 0.01 0.01* > [...] > $convergence > [1] 0 > > *$message > [1] "CONVERGENCE: NORM OF PROJECTED GRADIENT <= > PGTOL"*----------------------------------- > ______________________________________________ [email protected] mailing list https://stat.ethz.ch/mailman/listinfo/r-help PLEASE do read the posting guide http://www.R-project.org/posting-guide.html and provide commented, minimal, self-contained, reproducible code.
