Account Options

  1. Sign in
The old Google Groups will be going away soon, but your browser is incompatible with the new version.
Google Groups Home
« Groups Home
sci.math.num-analysis
+ new post
About this group
Subscribe to this group
This is a Usenet group - learn more
Message from discussion Nonlinear solvers for expensive functions
The group you are posting to is a Usenet group. Messages posted to this group will make your email address visible to anyone on the Internet.
Your reply message has not been sent.
Your post was successful
 
From:
To:
Cc:
Followup To:
Add Cc | Add Followup-to | Edit Subject
Subject:
Validation:
For verification purposes please type the characters you see in the picture below or the numbers you hear by clicking the accessibility icon. Listen and type the numbers you hear
 
Peter Spellucci  
View profile  
 More options Mar 12 2002, 5:40 am
Newsgroups: comp.lang.fortran, sci.math.num-analysis
From: spellu...@mathematik.tu-darmstadt.de (Peter Spellucci)
Date: 12 Mar 2002 10:37:14 GMT
Local: Tues, Mar 12 2002 5:37 am
Subject: Re: Nonlinear solvers for expensive functions
Print | View thread | Show original | Report this message | Find messages by this author
In article <204e7041.0203110920.3e8f2...@posting.google.com>,
 adr...@bjahouston.com (Adrian Ferramosca) writes:
|> I have a multidimensional nonlinear problem to solve where the
|> function evaluations are very expensive and the number of variables
|> can be large.  Therefore derivative based solvers (user-supplied or
|> numerical) are not an option.  I am using a Wegstein type solver at
|> present, but are there any other solutions methods out there which can
|> handle these types of problems?  I tried implementing GDEM from the
|> original paper but couldn't get it working.
|>
|> Adrian

until now i believed to know something about optimization but must state that
neither I know was a wegstein solver is nor what GDEM.
It would be a good idea not to assume that others know your slang.
you did say nothing about the nature of your function evaluations. as others
correctly stated, the use of gradient (and even Hessian) information becomes
more important then the function evaluation is costly and dimension is large.
(what is "large" for you? n=50000?)
I assume that  your function evaluation come from a software (or even worse
hardware) blackbox, and in this case indeed automatic differentiation is not
possible. If you have an algorithm in the form of a subprogram, then automatic
differentiation is the way to go.
but here come some info's, which possibly might be of help
peter

Jones, Donald R.; Schonlau, Matthias; Welch, William J.
Efficient global optimization of expensive black-box functions. (English)
[J] J. Glob. Optim. 13, No.4, 455-492 (1998). [ISSN 0925-5001]

Toropov, V.V.
Multipoint approximation method in optimization problems with
expensive function values. (English)
[CA] Computational systems analysis, Proc. 4th Int. Symp. Syst. Anal.
Simulation, Berlin/Ger. 1992, 207-212 (1992).

Schagen, I.P.
Stochastic interpolation applied to the optimisation of expensive
objective functions. (English)
[CA] Computational statistics, Proc. 4th Symp., Edinburgh 1980, 362-367 (1980).


 
You must Sign in before you can post messages.
To post a message you must first join this group.
Please update your nickname on the subscription settings page before posting.
You do not have the permission required to post.
Create a group - Google Groups - Google Home - Terms of Service - Privacy Policy
©2013 Google