Toward Generalized Tensor Algebra for ab initio Quantum Chemistry Methods
summary
Proposes a concrete block-sparse tensor algebra and C++ API that encodes index spaces, dependency maps and tiled index sequences so sparse/compressed formats drive code generation and loop nests. Shows how dependent index sequences let the compiler reorder or intersect iteration spaces for correct sparse contraction and explains storage-agnostic addressing for distributed data. Demonstrates the framework on real ab initio kernels (Hartree–Fock/exchange-style contractions), reports working prototypes and ongoing performance tuning for multiple architectures.
AI-written summary · about