Are there any plans to expose block-sparse tensor support in cuTensorNet, now that cuTENSOR provides cutensorBlockSparseContract (Beta)?
The block-sparse layout (per-mode sections + non-zero block coordinates) maps naturally onto symmetric tensors with Abelian (e.g. U(1)) symmetry, the standard data structure in tensor-network libraries for quantum many-body physics.
Path optimization and decomposition primitives (cutensornetTensorSVD, QR) aware of symmetry sectors would save TN users from rolling their own dispatcher on top of cuTENSOR.
Are there any plans to expose block-sparse tensor support in cuTensorNet, now that cuTENSOR provides
cutensorBlockSparseContract(Beta)?The block-sparse layout (per-mode sections + non-zero block coordinates) maps naturally onto symmetric tensors with Abelian (e.g. U(1)) symmetry, the standard data structure in tensor-network libraries for quantum many-body physics.
Path optimization and decomposition primitives (
cutensornetTensorSVD, QR) aware of symmetry sectors would save TN users from rolling their own dispatcher on top of cuTENSOR.