Symmetry And Block-Sparse Documentation

This directory owns quantum-number semantics and the design of the symmetry-aware block-sparse Tensor path.

Implemented Foundation

• Quantum Numbers and Symmetry describes Symmetry, QNum, QNumList, and BlockSpace.
• Bosonic Abelian BlockTensor Prototype describes the implemented key-coordinate and dense-axis model, sparse and complete block patterns, host and packed CUDA storage, mapped permutation and repartition views, linear operations, generalized adjacent contraction, diagonal storage, and staged per-charge SVD and truncation.

Active Design

• Spaces, Duals, and Tensor Morphisms is the canonical boundary, space-identity, wire-bending, and contraction model.
• BlockTensor is the central symmetry-typed container design.
• Block-Sparse Tensor and Layout develops the storage and layout model refined by BlockTensor.
• Block Coalescing covers GEMM grouping and placement interactions.
• Axis Labels, Contraction, and Braiding defines the intended user-facing axis and permutation policy.
• Raw Primitives and Symmetric Lowering identifies the primitive operations needed below symmetry-aware algorithms.

The implemented bosonic Abelian path now supports host and single-device CUDA execution, including matrix-free Krylov operations and finite U(1) two-site DMRG. Broader symmetry categories, multi-device and MPI-distributed block placement, and wider specialized operation coverage remain active work.

Source Navigation

• Implemented symmetry foundations
• Low-level tensor kernels
• Dense tensor operations
• CPU reference linalg kernels