search

cvanaret / Uno

A next-gen Lagrange-Newton solver for nonconvex optimization. It unifies barrier and SQP methods in a modern and generic way, and implements different globalization flavors (line search/trust region and merit function/filter method/funnel method). Competitive against filterSQP, IPOPT, SNOPT, MINOS and CONOPT.
MIT License
293 stars 22 forks source link

TR strategy: properly reset the bounds duals and set the warm-start information #83

Closed cvanaret closed 1 week ago

cvanaret commented 1 week ago

Properly update the WarmstartInformation object: handle phase switches, errors, etc. Two fixes in TrustRegionStrategy:

odow commented 1 week ago

Can we test the TR without BQPD?

cvanaret commented 1 week ago

Yes, we can assemble a trust-region SLP method with HiGHS as LP solver. So far, I've fixed lack of convergence on infeasible problems for SLP methods. I'm still working on properly setting the warmstart information.

leyffer commented 1 week ago

Excellent. It would be good to also test it with BQPD, just to make make sure the two solvers behave the same on smaller instances. Eventually, we should add an EQP phase and get a second order method. Sven

Sven Leyffer Senior Computational Mathematician and Deputy Director Mathematics and Computer Science Division Argonne National Laboratory

From: Charlie Vanaret @.> Sent: Tuesday, November 12, 2024 1:26 AM To: cvanaret/Uno @.> Cc: Subscribed @.***> Subject: Re: [cvanaret/Uno] TR strategy: properly reset the bounds duals and set the warm-start information (PR #83)

Yes, we can assemble a trust-region SLP method with HiGHS as LP solver. So far, I've fixed lack of convergence on infeasible problems for SLP methods. I'm still working on properly setting the warmstart information. — Reply to this email ZjQcmQRYFpfptBannerStart This Message Is From an External Sender This message came from outside your organization.

ZjQcmQRYFpfptBannerEnd

Yes, we can assemble a trust-region SLP method with HiGHS as LP solver. So far, I've fixed lack of convergence on infeasible problems for SLP methods. I'm still working on properly setting the warmstart information.

— Reply to this email directly, view it on GitHubhttps://urldefense.us/v3/__https://github.com/cvanaret/Uno/pull/83*issuecomment-2469782394__;Iw!!G_uCfscf7eWS!bZvRUVGsxxuiRXPSLQUQSdWAvvVFWYr90UTzEFfBtCwYjZ9t1F2MJ8me78JDdYtuhKPC-ysyHWvx0korKllKHZFntfo$, or unsubscribehttps://urldefense.us/v3/__https://github.com/notifications/unsubscribe-auth/AB52TLLHQTYEDGY75LPTYST2AGUUHAVCNFSM6AAAAABRR5UPBCVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMZDINRZG44DEMZZGQ__;!!G_uCfscf7eWS!bZvRUVGsxxuiRXPSLQUQSdWAvvVFWYr90UTzEFfBtCwYjZ9t1F2MJ8me78JDdYtuhKPC-ysyHWvx0korKllKBWGr1yE$. You are receiving this because you are subscribed to this thread.Message ID: @.***>

odow commented 1 week ago

It would be good to also test it with BQPD

Sure. But this requires us to be able to compile a version of Uno that includes BQPD on a public CI instance.