Hi Guido

First, sorry about sending to the old mailing list; I had forgotten 
about the new one. (I actually sent a copy to the new one, so sorry if 
you got this twice.)

Second - maybe I am misunderstanding terminology? Regularization, I 
thought, is making sure that a cell is no more than once more refined 
than any of its neighbors? I thought that, in 2D and 3D, 
prepare_coarsening_and_refinement () does this even with none of the 
flags set that you mentioned below..? I am talking about the case when 
all flags in the member smooth_grid are false.

Regardless, in 1D, there is not the option to set any of those flags, 
and there is no regularization. The only thing that 
prepare_coarsening_and_refinement () does (according to the subversion) 
is make a call to fix_coarsen_flags. So I guess what I'm talking about 
is implementing the regularization step in 
prepare_coarsening_and_refinement() (STEP 6 according to the comments in 
the source) and then calling fix_coarsen_flags. (I'll go by the version 
6.0 source, there's no need to add all the anisotropic refinement stuff, 
of course.) I assume there's no reason I can't do this?

thanks
dan

Guido Kanschat wrote:
>
> Daniel,
>
> first, please replace the email address for the mailing list in your 
> browser: it is [EMAIL PROTECTED] The old address is obsolete.
>
>
>
> it seems to me, that the following smoothing algorithms make sense in 1D:
>
> /    limit_level_difference_at_vertices       
> eliminate_unrefined_islands       patch_level_1       
> coarsest_level_1       eliminate_refined_inner_islands       
> eliminate_refined_boundary_islands       
> do_not_produce_unrefined_islands   
> Additionally, these should do the proper things in 1D as well.
>
>    smoothing_on_refinement       smoothing_on_coarsening       
> maximum_smoothing
>
> I think, what you describe is the first method only. If you want to 
> implement any of those, I suggest to read the source code as well as 
> the docs.
>
> On the other hand, the documentation is a bit bold there: if your 
> error estimator does not indicate refinement, yhere should be no 
> degradation of accuracy. That is, if it IS a reliable and efficient 
> estimator. Therefore, I'd suggest you try without smoothing.
>
> Best,
> Guido
>
> /Daniel Goldberg wrote:
>> Hi all
>>
>> I'm using a 1D model to try to test out an a posteriori estimator (I 
>> figure 1D will make it easier to identify whether the estimator is 
>> more effective than a generic one, and also I have some nice 
>> quasi-analytic 1D solutions). I realized that there is no mesh 
>> regularization implemented for the call to the 1-D version of 
>> prepare_coarsening_and_refinement(), which could degrade accuracy, 
>> right?
>>
>> I figure it should not be so difficult to implement the 
>> regularization step in 1D on my own, but I was wondering if I could 
>> get some assistance from someone who has tried it or thought about 
>> it? Would I do a similar reverse sweep, from highest down to (2nd) 
>> lowest level, to the one described in the docs, where at each level I 
>> check that if a cell's level after refinement would be 2 higher than 
>> that of its neighboring cell, then the neighboring cell is marked for 
>> refinement? And I noticed the fix_coarsen_flags() function is called 
>> even in 1D.. would it just need to be called once after the 
>> aforementioned sweep?
>>
>> Thanks very much,
>> Dan
>>
>


_______________________________________________

Reply via email to