If I understand you correctly: I should either recheck my matrix assembly,
or test a smaller step when my solution stalls, even though that goes
against my understanding of the Newton method (step size increases)? I
noted that (when stalling) my residual value was approximately half the
l2-norm of the right hand side. Is that something I can use for debugging?
Thank you!
Am Freitag, 6. Oktober 2017 17:40:41 UTC+2 schrieb Daniel Arndt:
>
> Maxi,
>
> I was arguing that if the Newton update you are computing is not
> essentially zero, there should be a (sufficiently small) step size such
> that the residual decreases when updating the solution.
> Otherwise, there is clearly something wrong in the assmebly of your Newton
> matrix or the residual.
>
> Best,
> Daniel
>
> Am Freitag, 6. Oktober 2017 15:32:00 UTC+2 schrieb Maxi Miller:
>>
>> Step-15 uses a fixed alpha = 0.1, but in the suggestions it is proposed
>> to increase alpha to 1, when suitable in order to increase convergence. I
>> am also using a fixed alpha of 0.1 at the moment, thus it will never go to
>> zero.
>>
>> Am Freitag, 6. Oktober 2017 15:27:40 UTC+2 schrieb Wolfgang Bangerth:
>>>
>>> On 10/06/2017 07:20 AM, 'Maxi Miller' via deal.II User Group wrote:
>>> > Isn't that basically the method done in step-15 and step-33 (line
>>> search)? But
>>> > why should my step length (I assume it is alpha) go to zero? Or is it
>>> the
>>> > newton-update (which also should go to zero, when being close to the
>>> solution)?
>>>
>>> The step length is alpha in
>>>
>>> u_{n+1} = u_n + alpha delta u_n
>>>
>>> I don't think step-15 uses a line search (i.e., it *always* uses
>>> alpha=1). I
>>> don't recall about step-33.
>>>
>>> Your question "why should my step length (I assume it is alpha) go to
>>> zero?"
>>> Is the wrong one. Maybe it shouldn't. The question is whether it does.
>>> Debugging is all about understanding that what you *think is true* is
>>> apparently not true. So you have to test hypotheses.
>>>
>>> Best
>>> W.
>>>
>>> --
>>> ------------------------------------------------------------------------
>>> Wolfgang Bangerth email: bang...@colostate.edu
>>> www:
>>> http://www.math.colostate.edu/~bangerth/
>>>
>>>
>
--
The deal.II project is located at http://www.dealii.org/
For mailing list/forum options, see
https://groups.google.com/d/forum/dealii?hl=en
---
You received this message because you are subscribed to the Google Groups
"deal.II User Group" group.
To unsubscribe from this group and stop receiving emails from it, send an email
to dealii+unsubscr...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.
