Module stda_gpu 

GPU-accelerated sTDA Coulomb matrix construction.

Offloads the O(N²_singles × N²_atoms) J-integral computation to the GPU. The inner kernel evaluates: J_{ia,jb} = Σ_{A,B} q^A_{ia} · γ_{AB} · q^B_{jb}

This is a GEMM-like operation on the transition charge matrix: A_off_diag = 2 · Q^T · Γ · Q where Q is (n_atoms × n_singles) and Γ is (n_atoms × n_atoms).

Functions

compute_stda_j_matrix_gpu

GPU-accelerated sTDA J-integral matrix: A_{off} = 2 · Q^T · Γ · Q