"Negative Jacobian" when using GapHeatTransfer

[WayneX-Wong]

Hi there,

I was using GapHeatTransfer, but encountered a "negative jacobian" error.

Model:
A cylinder and a hollow cylinder with a 1cm gap between the bottom surface of the former and the top surface of the latter, GapHeatTransfer is set to this gap.

Input:
Shown as main.txt

Error:

Elem Information
   id()=invalid, unique_id()=invalid, subdomain_id()=1, processor_id()=0
   type()=QUAD4
   dim()=2
   n_nodes()=4
   mapping=LAGRANGE_MAP
    0  Node id()=84581, processor_id()=0, Point=(x,y,z)=(-0.0962917, 0.0269797,      0.5)
    DoFs=(0/0/84581) (1/0/169162) (1/1/169163) 
    1  Node id()=84056, processor_id()=0, Point=(x,y,z)=(-0.112731,   0.0372,      0.5)
    DoFs=(0/0/84056) (1/0/168112) (1/1/168113) 
    2  Node id()=84451, processor_id()=0, Point=(x,y,z)=(-0.113464, 0.0171475,      0.5)
    DoFs=(0/0/84451) (1/0/168902) (1/1/168903) 
    3  Node id()=84580, processor_id()=0, Point=(x,y,z)=(-0.0990686, 0.0136167,      0.5)
    DoFs=(0/0/84580) (1/0/169160) (1/1/169161) 
   n_sides()=4
    neighbor(0)=nullptr
    neighbor(1)=nullptr
    neighbor(2)=nullptr
    neighbor(3)=nullptr
   hmin()=0.0136485, hmax()=0.027255
   volume()=0.000269658
   active()=1, ancestor()=0, subactive()=0, has_children()=0
   parent()=nullptr
   level()=0, p_level()=0
   refinement_flag()=DO_NOTHING
   p_refinement_flag()=DO_NOTHING
   DoFs=
libMesh terminating:
ERROR: negative Jacobian -4.1359e-25 at point (x,y,z)=(-0.0533052, 0.0001775,      0.5) in element 18446744073709551615

Negative Jacobian error occurs at the inner hollow on the top surface of the hollow cylinder, with no node exists there

Troubleshooting attempt:

1. I first checked the mesh volume and conformality using the MeshDiagnosticsGenerator and found no negative volume or non-confromal issues

*** Info ***
Number of elements below prescribed minimum volume : 0


*** Info ***
Number of elements with negative volume : 0


*** Info ***
Number of elements above prescribed maximum volume : 0


*** Info ***
Number of non-conformal nodes: 0

2. I then tried to change the hollow cylinder to a solid cylinder in the model, which was able to run normally without any error, and got a result which seems reasonable.

Is the problem due to the fact that GapHeatTransfer cannot applied to situation when two surfaces have different shapes?

Thanks!

[GiudGiud]

Hello

When using the MeshDiagnosticsGenerator in 1), did you use the element jacobian check?
If this reports an error, then your input mesh has a problem.

A close up to the elements around (-0.0962917, 0.0269797, 0.5) (from the log) could show the issue

Guillaume

[maxnezdyur]

Can you post the input file with the mesh. The link above is not accessible.

[GiudGiud]

the link isnt set up well
copy paste it in the browser

[maxnezdyur]

It fails in fe_side->reinit(side, &points); below (FindContactPoint.C). I don't think this has anything to do with going out of bounds. For this example we are 4 iterations in and then it fails. I think we just need to have a try and catch here or something like that. If it can't converge then maybe it wasn't a good contact point? But I am very confused about how findContactPoint works.

  for (; nit < 12 && update_size > TOLERANCE * TOLERANCE; nit++)
  {
    d = secondary_point - phys_point[0];

    DenseMatrix<Real> jac(dim - 1, dim - 1);

    jac(0, 0) = (d2xyz_dxi2[0] * d) - (dxyz_dxi[0] * dxyz_dxi[0]);

    if (dim - 1 == 2)
    {
      jac(1, 0) = (d2xyz_dxieta[0] * d) - (dxyz_dxi[0] * dxyz_deta[0]);

      jac(0, 1) = (d2xyz_detaxi[0] * d) - (dxyz_deta[0] * dxyz_dxi[0]);
      jac(1, 1) = (d2xyz_deta2[0] * d) - (dxyz_deta[0] * dxyz_deta[0]);
    }

    DenseVector<Real> rhs(dim - 1);

    rhs(0) = -dxyz_dxi[0] * d;

    if (dim - 1 == 2)
      rhs(1) = -dxyz_deta[0] * d;

    DenseVector<Real> update(dim - 1);

    jac.lu_solve(rhs, update);

    ref_point(0) += update(0);

    if (dim - 1 == 2)
      ref_point(1) += update(1);

    points[0] = ref_point;
    fe_side->reinit(side, &points);
    d = secondary_point - phys_point[0];

    update_size = update.l2_norm();
  }

[GiudGiud]

@maxnezdyur you were on the right track

With a try/catch we can get further
@WayneX-Wong please check out this branch and see if the results are good for you
#28149

It still has to skip some contact nodes because it cant find a solution. But at least the solve starts

[maxnezdyur]

Oh nice, happy to help. @BoZeng1997 I believe you had some of these problem many months ago. I think the solution is finally found! Look at this thread and check #28149.

[WayneX-Wong]

The case runs successfully, and the results seem reasonable now!

Thanks again for your help @GiudGiud @maxnezdyur !