[Ipopt] Status of MA86 Support?

[Jonathan Hogg]

You can use HSL_MA86 using the latest svn trunk, but a crucial bug fix 
just missed making the last stable release. My current performance 
results indicate it is only faster than MA57 on larger problems, and 
also suffers from not being bit-compatible so may find different local 
optima of a non-convex problem when run multiple times.

You can link MA57 against multi-threaded BLAS such as the Intel MKL. 
This often gives a limited (but free) speedup, especially on 
larger/denser problems.

I've also got a prototype interface to HSL_MA97 that *is* bit-compatible 
(so you get repeatable results), and faster on small problems than 
HSL_MA86 (though 20-30% slower and more memory hungry on very large 
problems). Drop me a direct email if you want a copy.

Jonathan.

On 27/01/12 05:28, Andreas Voegel wrote:
> Dear Ipopt Community,
>
> Looking through the Ipopt source, I noticed support code for the HSL 
> MA86 solver. This interface is not mentioned in the documentation, 
> however. A post from August seems to indicate that the use of MA86 was 
> not recommended, at the time. Is this still the case?
>
> Best regards,
> Andreas
>
>
> _______________________________________________
> Ipopt mailing list
> Ipopt at list.coin-or.org
> http://list.coin-or.org/mailman/listinfo/ipopt


-- 
Scanned by iCritical.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://list.coin-or.org/pipermail/ipopt/attachments/20120127/7ec1dfcf/attachment.html>