Module contract 

Multi-tensor contraction with optimal contraction order.

This module provides functions to contract multiple tensors efficiently using hyperedge-aware einsum optimization via the tensorbackend (tenferro-backed implementation).

This module works with concrete types (DynIndex, TensorDynLen) only.

Main Functions

• contract_multi: Contracts tensors, handling disconnected components via outer product
• contract_connected: Contracts tensors that must form a connected graph

Diag Tensor Handling

When Diag tensors share indices, their diagonal axes are unified to create hyperedges in the einsum optimizer.

Example: Diag(i,j) * Diag(j,k):

• Diag(i,j) has diagonal axes i and j (same index)
• Diag(j,k) has diagonal axes j and k (same index)
• After union-find: i, j, k all map to the same representative ID
• This creates a hyperedge that the einsum optimizer handles correctly

Structs

AxisUnionFind

Union-Find data structure for grouping axis IDs.

Functions

build_diag_union

Build a union-find structure from a collection of tensors.

collect_sizes

Collect dimension sizes for remapped IDs.

contract_connected

Contract multiple tensors that form a connected graph.

contract_multi

Contract multiple tensors into a single tensor, handling disconnected components.

print_and_reset_contract_profile

Print and clear the aggregated multi-tensor contraction profile.

remap_output_ids

Remap output IDs using the union-find structure.

remap_tensor_ids

Remap tensor indices using the union-find structure.

reset_contract_profile

Reset the aggregated multi-tensor contraction profile.