Factorising tensor

TTNOpt also provides a tool for decomposing a high-rank tensor into a tensor network representation.

We assume that the input tensor is given as a .npy file. Then, a input file for the factrizing tensor is created as follows:

factorising.yml

target:
    tensor: tensor.npy
numerics:
    initial_bond_dimension: 4
    fidelity:
        opt_structure:
            type: 1 # 0: no optimization, 1: structural optimization
        init_random: 0
        max_bond_dimensions: [4, 8, 16] # Maximum bond dimension for each repetition
        max_num_sweeps: [100, 50, 25]
        truncation_error: 0.0
        fidelity_convergence_threshold: 1e-10
        entanglement_convergence_threshold: 1e-14

output:
    dir: TTN_
    tensors: 1

Then, we run the decomposition by the following command:

ft factorising.yml

Then, the results including the decomposed tensors and the connectivity information of the tensor network are output to the specified folder. Examples of the input files are provided in the sample directory, which reproduces the results of the paper.