Re: [Libmesh-users] Accessing DiffSolver::_solve_result

[Roy Stogner <roy@st...>]

On Thu, 4 Sep 2008, David Knezevic wrote:

> The patch is attached (I must've checked out libmesh anonomously cos I can't 
> seem to commit myself, I'll look into that).

There've been some server side SVN problems just recently.  But try it
again now; I just double-checked and got Tim's logging typo fix
through.
---
Roy

[Libmesh-users] Accessing DiffSolver::_solve_result

[David Knezevic <dave.knez@gm...>]

I'm using (a subclass) of FEMSystem, with NewtonSolver, and I'd like to 
detect whether or not a quasi-Newton solve converged, i.e. I'd like to 
detect whether NewtonSolver::solve returns the enum 
DiffSolver::DIVERGED_MAX_NONLINEAR_ITERATIONS. Maybe I'm missing 
something, but I didn't see any easy way to access this return value? 
Also, the _solve_result member in DiffSolver is protected, so I just 
wanted to ask what the easiest way to access that value would be? Maybe 
a getter method in DiffSolver for that member would be useful?

Thanks for the help.

- Dave

Re: [Libmesh-users] Accessing DiffSolver::_solve_result

[Roy Stogner <roy@st...>]

On Thu, 4 Sep 2008, David Knezevic wrote:

> I'm using (a subclass) of FEMSystem, with NewtonSolver, and I'd like to
> detect whether or not a quasi-Newton solve converged, i.e. I'd like to
> detect whether NewtonSolver::solve returns the enum
> DiffSolver::DIVERGED_MAX_NONLINEAR_ITERATIONS. Maybe I'm missing
> something, but I didn't see any easy way to access this return value?
> Also, the _solve_result member in DiffSolver is protected, so I just
> wanted to ask what the easiest way to access that value would be? Maybe
> a getter method in DiffSolver for that member would be useful?

So far IIRC we've only been using that value internally to let the
transient solvers and continuation solvers which call
DiffSolver::solve back off their time step or continuation step size
on failure.  My thinking was that whether such reduction succeeded in
making the problem well-behaved enough to succeed on the retry or not,
the original solver divergence would have become moot.  Satisfy my
curiosity: What's the usage case where you have a function which isn't
responsible for calling DiffSolver::solve() but wants to check it's
result?  Whatever the reason, though, a getter method would be great,
if you want to send a patch or commit one.
---
Roy

Re: [Libmesh-users] Accessing DiffSolver::_solve_result

[David Knezevic <dave.knez@gm...>]

Hi Roy,

> So far IIRC we've only been using that value internally to let the
> transient solvers and continuation solvers which call
> DiffSolver::solve back off their time step or continuation step size
> on failure.  My thinking was that whether such reduction succeeded in
> making the problem well-behaved enough to succeed on the retry or not,
> the original solver divergence would have become moot.  Satisfy my
> curiosity: What's the usage case where you have a function which isn't
> responsible for calling DiffSolver::solve() but wants to check it's
> result?  Whatever the reason, though, a getter method would be great,
> if you want to send a patch or commit one.

It's nothing exciting, I've just implemented a simple continuation 
scheme where I'm doing the continuation parameter adjustments in the 
main function, so I needed to access the solver info there... It'd 
probly be better if I did something in the spirit of ContinuationSystem, 
but the approach I'm using is working for what I need at the moment.

The patch is attached (I must've checked out libmesh anonomously cos I 
can't seem to commit myself, I'll look into that).

- Dave

Re: [Libmesh-users] Accessing DiffSolver::_solve_result

[Roy Stogner <roy@st...>]

On Thu, 4 Sep 2008, David Knezevic wrote:

> The patch is attached (I must've checked out libmesh anonomously cos I can't 
> seem to commit myself, I'll look into that).

There've been some server side SVN problems just recently.  But try it
again now; I just double-checked and got Tim's logging typo fix
through.
---
Roy

Re: [Libmesh-users] Accessing DiffSolver::_solve_result

[John Peterson <jwpeterson@gm...>]

On Thu, Sep 4, 2008 at 10:46 AM, Roy Stogner <roy@...> wrote:
>
> On Thu, 4 Sep 2008, David Knezevic wrote:
>
>> The patch is attached (I must've checked out libmesh anonomously cos I can't
>> seem to commit myself, I'll look into that).
>
> There've been some server side SVN problems just recently.  But try it
> again now; I just double-checked and got Tim's logging typo fix
> through.

I'm not sure why I left out an accessor when I added this, probably
just an oversight.  I remember thinking that the need to AND the
result against the different possible SolveResult flags was something
a little ugly for the public interface...

-- 
John