[Trilinos-Users] Powell's Hybrid Method

[Roger Pawlowski]

Hi Donald,

NOX does support Powell's method, extended for inexact Newton solves.  
For NOX, you would use the Inexact Trust Region solver with the default 
directions (the Cauchy and Netwon directions).  The algorithm is very 
similar to the minpack code, but extra safeguards are in place for 
situations caused by inexact Newton solves.

An example can be found in:

Trilinos/packages/nox/test/epetra/1Dfem/1DfemScaledInexactTrustRegion.C

The reference for our implementation is:

Inexact Newton dogleg methods 
<https://oppie.lanl.gov/oppie/service?url_ver=Z39.88-2004&rft_id=info:lanl-repo/oppie&svc_val_fmt=http://oppie.lanl.gov/openurl/oppie.html&svc_id=info:lanl-repo/svc/oppie/full-rec&svc.oparam1=info:lanl-repo/isi/000256453400021&svc.oparam2=info:lanl-repo/svc/getDIDL&svc.oparam3=pawlowski%20inexact%20newton%20dogleg&svc.oparam4=0&svc.oparam5=5&svc.oparam6=score%20desc&svc.oparam7=SNL&svc.oparam8=&svc.oparam9=> 

Pawlowski, RP 
<https://oppie.lanl.gov/oppie/service?url_ver=Z39.88-2004&rft_id=info:lanl-repo/oppie&svc_val_fmt=http://oppie.lanl.gov/openurl/oppie.html&svc_id=info:lanl-repo/svc/oppie/solr-bib-search&svc.oparam2=0&svc.oparam3=25&svc.oparam4=score%20desc&svc.oparam1=name:%22Pawlowski,%20RP%20%22> 
; Simonis, JP 
<https://oppie.lanl.gov/oppie/service?url_ver=Z39.88-2004&rft_id=info:lanl-repo/oppie&svc_val_fmt=http://oppie.lanl.gov/openurl/oppie.html&svc_id=info:lanl-repo/svc/oppie/solr-bib-search&svc.oparam2=0&svc.oparam3=25&svc.oparam4=score%20desc&svc.oparam1=name:%22Simonis,%20JP%20%22> 
; Walker, HF 
<https://oppie.lanl.gov/oppie/service?url_ver=Z39.88-2004&rft_id=info:lanl-repo/oppie&svc_val_fmt=http://oppie.lanl.gov/openurl/oppie.html&svc_id=info:lanl-repo/svc/oppie/solr-bib-search&svc.oparam2=0&svc.oparam3=25&svc.oparam4=score%20desc&svc.oparam1=name:%22Walker,%20HF%20%22> 
; Shadid, JN 
<https://oppie.lanl.gov/oppie/service?url_ver=Z39.88-2004&rft_id=info:lanl-repo/oppie&svc_val_fmt=http://oppie.lanl.gov/openurl/oppie.html&svc_id=info:lanl-repo/svc/oppie/solr-bib-search&svc.oparam2=0&svc.oparam3=25&svc.oparam4=score%20desc&svc.oparam1=name:%22Shadid,%20JN%20%22> 

SIAM JOURNAL ON NUMERICAL ANALYSIS (2007) Vol.46, iss.4, p.2112-2132

Best regards,
Roger Pawlowski


Kunz Donald L Civ AFIT/ENY wrote:
>
> A couple of years ago, I asked if there was a NOX solver that could be 
> made to work like Powell’s hybrid method (HYBRD in MINPACK).  The 
> response has long since been lost, so I am asking again.  Your 
> patience is appreciated.
>
> Don
>
> // SIGNED //
>
> Donald L. Kunz, Ph.D., P.E.
>
> Associate Professor of Aerospace Engineering, AD-23
>
> Department of Aeronautics & Astronautics
>
> Air Force Institute of Technology
>
> 2950 Hobson Way, AFIT/ENY
>
> Wright-Patterson AFB, OH  45458-7765
>
> Phone: 937-255-3636 x4548 (DSN 785)
>
> Fax: 937-656-7053 (DSN 986)
>