Speaker
Description
Telum.jl is a non-Abelian tensor network library written entirely in Julia. Its internal Clebsch–Gordan engine, LurCGT.jl, extends the approaches of QSpace and TensorKit.jl to compute symmetry-related quantities for SU(N), Sp(2N), SO(N), and G2 exactly using integer arithmetic and in a deterministic manner. Telum.jl is designed to handle tensors with arbitrary combination of these symmetries, with ready-to-use support for spinful fermionic spaces involving U(1), SU(2), and SU(N) charge, spin, and channel symmetries.
Telum.jl is designed for users to implement high-performance tensor network algorithms with simple and readable syntax. It provides an ITensor-style interface, together with utility methods for explicitly controlling contracted legs. On the performance side, Telum.jl implements an efficient SVD algorithm that avoids explicit leg fusion, as well as lazy evaluation for simple operations such as conjugation. In benchmark calculations performed so far, Telum.jl achieves CPU runtimes comparable to QSpace while reducing memory usage. Further optimizations, including lazy evaluation for tensor contractions and GPU support, are planned for future releases.