[deal.II] Re: multiply constrained dofs (hanging nodes+periodic) fails a simple test case
I understand my mistake now: there are extra hanging nodes constraints in
the constraint matrix with (hanging nod constraints + PBC) compared to the
constraint matrix with only hanging node constraints.
That is why the minimal example fails.
Sambit
On Wednesday, April 11, 2018 at 1:00:25 PM UTC-5, Sambit Das wrote:
>
>
>>> true, but I don't see why you would have the same norms if you
>> distribute with constraints from hanging nodes only or constraints from
>> hanging nodes+ PBC.
>> I think we can agree that the two ConstraintMatrix objects should be
>> different as in the case of PBC you additionally need to make sure that FE
>> space on the refined boundary matches that on the opposite, non-refined
>> side.
>>
>
> Yes I agree the ConstraintMatrix objects are different but the
> coefficients (a_{ij}) of the hanging node constraint equations
>
> x_{i} = a_{ij} x_{j}
>
> would be the same in both cases, only x_{j}'s would be different in both
> cases. Now x_{j}'s are nodes without any constraints which are set to the
> correct values explicitly in both cases:
>
> if(!constraints.is_constrained(globalDofIndex))
>vec1[globalDofIndex]=nodalCoor.norm();
>
> if(!onlyHangingNodeConstraints.is_constrained(globalDofIndex))
>vec2[globalDofIndex]=nodalCoor.norm();
>
> So the hanging nodes in both cases should have the same value after
> calling distribute.
>
>
>
>> If you suspect that there is a bug in constraints, you could check this
>> by simply choosing some more-or-less random vector, distribute and
>> plot-over-line in Paraview / Visit.
>> More cumbersome comparison would be to evaluate random field at the
>> opposite points.
>> You can use FEField function and then choose L/2-\delta and
>> -L/2+\delta with \delta = 1e-8 or so for X coordinate and then
>> whatever you want to Y/Z. This should give you the same value anywhere on
>> the two periodically matching points for a random input vector after
>> constraints are distributed.
>>
>> I will try doing this.
>
> Best,
> Sambit
>
--
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.
[deal.II] Re: multiply constrained dofs (hanging nodes+periodic) fails a simple test case
On Wednesday, April 11, 2018 at 7:00:18 PM UTC+2, Sambit Das wrote: > > Hi Denis, > > >> I don't think that's the case. The domain is indeed periodic, but this is >> completely detached from location of support/nodal points. >> Same applies to geometry, you will have different coordinates of vertices >> across the PBC so >> >> > I agree, the location of nodal points is detached from the periodicity of > the domain, but in this case the origin is at the center of the hypercube. > This artificially enforces that the nodal_coordinate.norm() is periodic. > true, but I don't see why you would have the same norms if you distribute with constraints from hanging nodes only or constraints from hanging nodes+ PBC. I think we can agree that the two ConstraintMatrix objects should be different as in the case of PBC you additionally need to make sure that FE space on the refined boundary matches that on the opposite, non-refined side. If you suspect that there is a bug in constraints, you could check this by simply choosing some more-or-less random vector, distribute and plot-over-line in Paraview / Visit. More cumbersome comparison would be to evaluate random field at the opposite points. You can use FEField function and then choose L/2-\delta and -L/2+\delta with \delta = 1e-8 or so for X coordinate and then whatever you want to Y/Z. This should give you the same value anywhere on the two periodically matching points for a random input vector after constraints are distributed. Denis. -- 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.
[deal.II] Re: multiply constrained dofs (hanging nodes+periodic) fails a simple test case
Hi Denis, > I don't think that's the case. The domain is indeed periodic, but this is > completely detached from location of support/nodal points. > Same applies to geometry, you will have different coordinates of vertices > across the PBC so > > I agree, the location of nodal points is detached from the periodicity of the domain, but in this case the origin is at the center of the hypercube. This artificially enforces that the nodal_coordinate.norm() is periodic. Best, Sambit -- 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.
