Hello,
is it possible to solve high dimensional PDEs with Moduls, for example 6 dimensions, defining hypercubes geometry?
Thanks.
I’m not with nvidia, so take what I say with a grain of salt.
Technologically speaking, I don’t see a reason why it fundamentally wouldn’t work so long as your PDE’s support the added dimensionality. However, the included code and PDE’s (e.g. Navier Stokes) appear to be capped at 3 dimensions, so you would likely need to implement the functionality yourself. Once that’s done though, I would imagine it works.
Thank you for your reply.
The reason why I am asking is that it seems from the documentation that in the geometry classes only 1D, 2D, and 3D geometries can be defined (Modulus Geometry - NVIDIA Docs), with output being only [‘x’], [‘x’,’y’], or [‘x’,’y’,’z’].
For example, how would I define a custom 4D hypercube geometry class to be used in PointwiseInteriorConstraint and PointwiseBoundaryConstraint?
Thanks.
4D geometry is not possible in the geometry module. You can always generate a input variable dataset ahead of time that reflects your higher-dimensional system you are interested in. This should be pretty easy to achieve if you can get your inputs to be numpy dictionary:
Alternatively, if you want to a custom geometry object that is possible like you mention. It can be used in the current constraints given it has the proper functionality (typically this is boundary shapes, sdf function, etc).