[MITgcm-devel] line search for large ctrl vectors

[Matthew Mazloff]

aha -- DYNAMIC -- thanks!

on a side note -- thanks for contributing optim_m1qn3 -- if I give it a try I will let you know performance

Matt

 
On Dec 16, 2015, at 2:32 AM, Martin Losch <Martin.Losch at awi.de> wrote:

> Sorry Matt, was too hasty,
> 
> I still use the MAX_INDEPEND variable static allocation, but there’s a cpp flag DYNAMIC that allocates dynamically and does not need MAX_INDEPEND.
> 
> M.
> 
>> On 16 Dec 2015, at 11:27, Martin Losch <Martin.Losch at awi.de> wrote:
>> 
>> Hi Matt,
>> 
>> did you try this:
>> 
>> <http://mitgcm.org/viewvc/MITgcm/MITgcm_contrib/mlosch/optim_m1qn3/>
>> 
>> ? 
>> 
>> I am not claiming that it is better than the optim/lsopt combination (in fact I saw cases where convergence is slower/worse), but it uses m1qn3 in a way it is supposed to be used and it is a more transparent compilation process (at least for me, since I wrote it … (o:)
>> 
>> Martin
>> 
>>> On 15 Dec 2015, at 19:28, Matthew Mazloff <mmazloff at ucsd.edu> wrote:
>>> 
>>> Hello
>>> 
>>> I am once again running into issues from my compiler not liking the size of the optimization line search variables
>>> (e.g.   xx(nmax), adxx(nmax), dd(nmax),  gold(nmax),   xdiff(nmax) where nmax is set in makefile as MAX_INDEPEND).
>>> 
>>> In the past I have hacked the code to make it work, though I am not really satisfied with my solution. I have the feeling that others in the community have dealt with similar issues and would love to know what others have done. 
>>> 
>>> Input much appreciated!
>>> 
>>> Thanks!
>>> Matt
>>> 
>>> 
>>> 
>>> 
>>> 
>>> _______________________________________________
>>> MITgcm-devel mailing list
>>> MITgcm-devel at mitgcm.org
>>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>> 
>> 
>> _______________________________________________
>> MITgcm-devel mailing list
>> MITgcm-devel at mitgcm.org
>> http://mitgcm.org/mailman/listinfo/mitgcm-devel
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel