Changelog

Release notes for quimb.

v1.16.0 (unreleased)

Breaking changes

Decomposition and compression defaults

• array_split and low-level rank-revealing decompositions now default to cutoff_mode="rel" instead of "rsum2". 1D zipup compression also defaults to "rel".

• Randomized SVD and batched decompositions reject active cumulative cutoff modes. Use "abs" or "rel" instead.

• tensor_network_1d_compress with an arbitrary-geometry method always returns a 1D network, inserting identities along long-range bonds. To keep the geometry, call tensor_network_ag_compress directly, as 2D boundary contraction now does for periodic boundaries, which therefore need an ag method.

Hilbert spaces and Hubbard models

• HilbertSpace: site ordering is now immutable. set_ordering raises TypeError. Use with_ordering to create a space with a different ordering.

• fermi_hubbard_from_edges: default to order="interleaved", alternating spins at each coordinate to avoid MPO bond dimensions growing with system size, with slightly slower matrix-vector products. Rebuild anything keyed by rank or flat configuration. Use order="blocked" for the old layout.

Partial traces and local expectations

• 2D compute_local_expectation: compute tr(rho G) from each term’s reduced density matrix and normalize locally by default (normalized=True). With return_all=True, return plain values, or (expec, trace) pairs with normalized="return".

• 2D compute_local_expectation and the new 2D partial trace, expectation, and environment methods require max_bond. Supply None explicitly for no limit, which is not recommended in 2D.

• 2D partial_trace and local_expectation use the boundary or envs routes. For the previous arbitrary-geometry compressed contraction with optimize, call TensorNetworkGenVector.partial_trace or TensorNetworkGenVector.local_expectation directly.

• Partial traces derive bra index names by replacing a leading k with b, falling back to random uuids on a name collision. This affects partial_trace_exact, make_reduced_density_matrix, and 3D partial traces. See get_bra_inds.

• Cluster expectations default to smudge=1e-12 and optimize="auto-hq", matching cluster partial traces.

Enhancements

Tensor decompositions

• Split functions accept cutoff="auto": 1e-10 for exact decompositions, disabled for randomized SVD.

• svd_rand_truncated: support "abs" and "rel" cutoffs on the sketched spectrum, and noise_dist="normal" or "rademacher" for the random sketch.

• Batched "abs" and "rel" cutoffs treat blocks as one spectrum and retain a common bond dimension. Both modes support renormalization, including for batched decompositions.

1D compression and MPS

• Add a 1D compression guide with method comparisons, benchmarks, and usage examples.

• Add successive deterministic compression (method="sdc"), sdc-oversample, and randomized variants sdcr and sdcr-oversample, supporting fermionic networks, based on https://arxiv.org/abs/2601.19650.

• direct, dm, zipup, sdc, sdcr, src, srcmps, fit, and their oversampling variants handle long-range bonds directly, supporting symmray abelian and fermionic tensors (for srcmps and fit, with a supplied tn_fit).

• dm uses QR-SVD when a site’s density matrix would exceed its rank bound, reducing layered PEPS norm boundary contraction costs (https://arxiv.org/abs/2406.09769), including for fermionic networks.

• src, srcmps, fit, and their oversampling variants accept a seed or random generator and match the backend’s device and dtype (autoray v0.10.0 or newer). By default, use the backend’s global random state.

• tensor_network_1d_compress defaults to cutoff="auto", choosing the cutoff for each method: 0.0 for src and srcmps, disabled for sdcr with its default randomized SVD, and 1e-10 otherwise.

• zipup, sdc, and sdcr oversampling accept cutoff_mode_oversample. Zipup also accepts cutoff_oversample="auto". All 1D oversampling methods accept compress_opts_final to configure the final direct sweep separately.

• fit with bsz=1 now supports fermionic tensor networks, including odd-parity tensors.

• Add MatrixProductState.from_product, also used by MPS_product_state, with support for block-sparse single-site vectors.

• Compression functions accept tag groups in site_tags; see parse_site_tag_groups. The 1D, 2D, and arbitrary-geometry wrappers also accept custom method callables.

Partial traces, environments, and 2D contraction

• Add quimb.tensor.environments for computing many block or plane environments in one sweep on open or periodic geometries. EnvironmentPlan supports "tree", "cut" and "cutpair" schedules. gen_exact_environments and gen_compressed_environments yield environments as they become ready. Choose a 1D, 2D, or arbitrary-geometry compressor with compress_fn.

• 1D partial traces and local expectations support periodic MPS with route=None, "canonical", or "envs". MatrixProductState.partial_trace returns a dense reduced density matrix; compute_partial_traces computes many at once. Add local_expectation alongside compute_local_expectation, with explicit methods for each route.

• Add TensorNetwork1D.gen_block_environments and compute_block_environments for exact block environments, and is_cyclic to check for periodic chains.

• 2D partial traces and local expectations support periodic PEPS with route=None, "boundary", or "envs" and options max_bond, cutoff, and method. Add TensorNetwork2DVector.partial_trace, compute_partial_traces, and local_expectation, with explicit methods for each route.

• Add TensorNetwork2D.gen_block_environments and compute_block_environments for compressed row or column block environments, and compute_plaquette_environments_via_envs for plaquettes, including those crossing periodic boundaries.

• The 2D "envs" route supports second_dense to control exact contraction of the remaining strip, by default only for strips one plane wide, alongside autogroup, first_contract, custom layer_tags, fixed plaquette_sizes, and callback_rhos to save reduced density matrices as they finish. max_separation=0.5 enables half-layer compression, including with a single periodic environment.

• Add the 2D "cluster_boundary" route: partial_trace_cluster_boundary, compute_partial_traces_cluster_boundary, and compute_local_expectation_cluster_boundary. Control cluster size with max_distance, including across periodic boundaries or over a whole periodic direction, optionally insert simple-update gauges, and use 1D boundary compression methods. max_separation controls uncompressed lines around kept sites, with 0.5 enabling half-layer compression even with only one boundary.

• contract_boundary with around now respects max_separation on each side of the target region.

• Add compute_partial_traces_exact and compute_partial_traces_cluster for computing many reduced density matrices at once.

• Exact and cluster partial traces now accept get="tn" for an uncontracted network. D2BP.partial_trace now accepts normalized="return" for (rho, trace) without normalization.

• 2D contract_boundary with method="full-bond" accepts compress_opts to configure the similarity decomposition, or similarity_method as a shortcut.

• Norm, overlap, and reduced density matrix constructors, and 2D norm, normalization, partial trace, and expectation methods accept layer_tags=True or 1 for the default ket and bra tags and False or 0 to disable them. Methods operating on an existing network reject these boolean and integer options with a clear error. 2D normalize now honors custom layer tags.

Gates and circuits

• Add dagger and transpose options to tensor-network gating methods, including tensor_network_gate_inds, tensor_network_gate_sandwich_inds, tensor_network_ag_gate, tensor_network_ag_gate_simple, and gate_sandwich_with_op_lazy. These apply \(G^\dagger\) or \(G^T\) instead of \(G\). MatrixProductState.gate_nonlocal also accepts transpose.

• Add MatrixProductOperator.gate_sandwich_with_auto_swap for two-site gate sandwiches that preserve canonical form, including long-range gates, with optional exponent stripping.

• CircuitPEPOSimpleUpdate: add dtype, to_backend, and convert_eager options, matching CircuitPEPSSimpleUpdate.

Time evolution and run management

• Add LocalHamGen.get_trotter_gates for first-, second-, and fourth-order Trotter approximations over multiple steps, inherited by 1D, 2D, and 3D Hamiltonians. TrotterGate objects unpack as U, where. Consecutive uses of the same commuting layer are fused by default. trotter_schedule returns the product formula alone.

• build_mpo_propagator_trotterized and build_pepo_propagator_trotterized accept order (default 1). The 1D method also gains ordering, matching the 2D method.

• TEBD and simple-update classes accept logdir and log_every to write progress.json. Use plot_progress_log with watch=True to monitor a running job, or load_progress_log for raw data.

• These classes accept checkpoint_every to save checkpoint.pkl in logdir, including a final checkpoint on normal or graceful completion. Continue with resume=True or load with from_checkpoint; resumed runs restore state and evolution options, so omit psi0 and ham.

• log_every and checkpoint_every accept sweep counts or durations such as "10mins"; see parse_time_spec.

• A STOP file in logdir or the first Ctrl-C during evolve stops the run after the current sweep. The file is removed once seen; a second interrupt stops immediately. Use graceful_interrupt=False for the previous interrupt behavior.

Belief propagation and loop expansions

• Support fermionic BP projector compression with method="projector", canonize="bp" and fermionic gates with D2BP.gate_.

• All BP classes accept diis and damping in both __init__ and run. Values passed to run become the new defaults.

• D1BP, D2BP, L1BP, L2BP, and TensorNetwork.gauge_all_simple accept sweep_order, with an efficient alternating schedule for trees. See SweepScheduler for other options.

• Add gen_gloops_edge_induced (tn.gen_gloops_edge_induced), yielding NetworkPatch objects that distinguish loops using different bonds on the same tensors.

• Infinite 2D generalized-loop expectations accept max_size and num_joins. Reuse their info cache across loop settings while the state, gauges, and operators remain unchanged.

Network construction

• TN_from_strings: add join_prefer="short" or "long" to favor smaller or larger loops, and join_avoid_loop_length (default 2) to avoid loops up to a given length. Set it to 0 to disable loop avoidance. join_prefer does not support join="all".

• Add hidden-cactus networks TN2D_rand_hidden_cactus, TN3D_rand_hidden_cactus, and TN_rand_hidden_cactus to extend hidden correlations while preserving lattice bond dimensions and low exact contraction cost. Also available through TN_from_strings(..., join_trees=True).

• Add TN_rand_hidden_loop, the arbitrary-graph counterpart of TN2D_rand_hidden_loop.

• Add TensorNetwork.insert_projectors_between_regions to insert precomputed projector arrays between regions.

• TN_matching: add edges to choose which sites to join, for example a plain chain ignoring long-range bonds.

Hilbert spaces and utilities

• HilbertSpace: support U1U1 sectors with any site ordering and a species argument selecting each site’s conserved charge. Accept sector forms {species: filling}, (ka, kb), and ((na, ka), (nb, kb)). order="blocked" groups species; "interleaved" alternates them at each position.

• Add distance_from_overlaps for normalized distances and infidelities from three precomputed overlaps.

• save_to_disk: add atomic to prevent partially written targets.

Deprecations

• Tensor.gate: rename transposed to transpose. The old name raises FutureWarning.

• 2D boundary contraction, environment, and norm methods: rename mode to method, with FutureWarning for the old name. With mode="full-bond", a method supplied as well is taken as similarity_method.

• MatrixProductState.compute_local_expectation: rename method to route, with FutureWarning for the old name.

• Rename MatrixProductState.partial_trace_to_dense_canonical to partial_trace_canonical. The old name still works but raises a warning.

• TN_from_strings: deprecate join_avoid_self_loops in favor of join_avoid_loop_length.

Bug fixes

Sparse matrices

• csr_matrix @ vector: fix wrong results for tall matrices and segfaults for wide matrices on the parallel path, used for nnz > 50000 once quimb is imported (GH #432).

Operators and time evolution

• SpinHam1D.build_sparse: include the last term with cyclic=True (GH #419).

• ikron: raise ValueError for out-of-range or repeated indices, which previously placed fewer operators than supplied without an error.

• build_matrix_ikron: fix incorrect matrices with non-identity ordering or non-integer site labels.

• HilbertSpace: prevent incorrect rank-to-configuration mappings after reordering, including invalid configurations for mixed dimensions. Ordering is now immutable, as described above.

• TEBD: fix order=4 to achieve fourth-order accuracy. Correct Suzuki weights include backward evolution during part of each step. Also accept order=1.

• build_mpo_propagator_trotterized: fix cyclic wrapping terms that are not symmetric under a site swap.

• FullUpdate: reuse the cached plaquette map in compute_energy.

Decompositions and compression

• compute_oblique_projectors: avoid inf or nan projectors for rank-deficient environments with cutoff=0.0. The same fix applies to shared diagonal division helpers and D2BP.gauge_insert with return_gauges="inverse".

• safe_inverse: fix TypeError in projector boundary contractions with block-sparse backends such as symmray.

• tensor_split: fix method="svd:rand" for complex and single-precision arrays, preserving dtype and device, and method="svd:eig" with nonzero cutoff for single precision and Numba.

• Oblique projector compression, including L2BP.compress, CTMRG, and HOTRG: fix shape-mismatch errors with symmray block-sparse arrays.

• 3D boundary contraction: remove temporary site tags from replaced tensors, including those shared with cached environments.

• 1D fit with max_iterations=1: build any initial guess at max_bond rather than padding a smaller one, and fix a TypeError when only cutoff is given.

• 1D srcmps: conjugate the sampling MPS, so a supplied guess helps for complex arrays, and odd-parity fermionic networks work.

Fermionic tensors

• 1D compression: fix fit for fermionic networks, and dm for mixed bond orientations and truncated odd-parity norm networks.

• Projector compression (method="projector"): fix zero projectors for odd-parity fermionic norm networks with every canonize option.

• D2BP.compress and D2BP.gauge_symmetric: preserve positive messages and full-rank identity matrices for fermionic networks.

• Tensor.conj: add output_inds to select output legs for fermionic conjugation phases. Fix conjugation phases for subnetworks and paired and unpaired odd-parity dummy modes in TensorNetwork.conj and Tensor.conj.

Belief propagation and overlaps

• D1BP.contract_loop_series_expansion and D2BP.contract_loop_series_expansion: include all distinct loops in each region and correct suppression factors. Multi-excitation corrections raise RuntimeError if they fail to converge.

• D2BP.normalize_tensors and single-site D2BP.gate_: keep contractions consistent after updates. Reject distinct supplied networks, and fix message ordering for complex two-site gates.

• D2BP.gauge_temp: restore gauges when its body raises. Fix relative smudge scaling for unsorted symmetry-block spectra.

• tensor_network_distance: remove incorrect dependence of infidelity on relative phase.

Partial traces and general tensor operations

• partial_trace_exact: fix AttributeError with get="tensor" and normalized=True.

• Exact, cluster, and arbitrary-geometry partial traces and local expectations accept a single site as where instead of raising TypeError.

• local_expectation_cluster: support max_bond on MPS.

• 2D compute_local_expectation: accept reversed two-site terms without KeyError in the boundary route.

• compute_local_expectation_exact and compute_local_expectation_cluster: correctly sum normalized expectations with normalized="return" and return_all=False. Methods that compute terms individually also complete their progress bars.

• MPS_product_state: fix TypeError for backends requiring a single shape tuple when reshaping single-site vectors.

• TensorNetwork.tids_are_connected: make results independent of tensor order.

v1.15.0 (2026-08-10)

Breaking changes (1.15.0)

Circuits

• Circuit.uni: return the non-transposed unitary, so circ.uni.to_dense() gives U acting as U @ psi. Use circ.get_uni(transposed=True) for the old convention.

• Non-exact circuit simulators now inherit CircuitBase rather than Circuit. Use isinstance(circ, CircuitBase) for any circuit type. Exact-only methods such as get_uni, get_psi_simplified, and sample_gate_by_gate are no longer available on these simulators.

• MPS circuit methods operate directly on the current, possibly compressed state and no longer accept simplify_* options. Unsupported representation-specific methods raise NotImplementedError consistently.

Gauging and loop expansions

• TensorNetwork.gauge_all_simple: arguments after max_iterations and tol, including smudge, power, and damping, are keyword-only.

• Simple-update gauging: smudge and gauge_smudge are relative to the largest gauge value. Results change for unnormalized gauges, but remain unchanged when max(g) == 1.

• Automatic generalized-loop sizing (max_size=None in tn.gen_gloops, or gloops=None in loop expansions) covers every target tensor or site that belongs to a loop. Targets outside loops are ignored, with UserWarning for explicit targets or trees with no loops. Loops larger than the resolved size are excluded. Use max_size="min" or gloops="min" for the old behavior.

Enhancements (1.15.0)

Circuit simulators and OpenQASM

• Add CircuitPEPSSimpleUpdate for arbitrary-geometry PEPS circuits with nearest-neighbor simple-update gates and cluster expectations. Define geometry with edges or infer it from gates or psi0; control accuracy with max_bond and gauge equilibration with equilibrate.

• Add CircuitPEPOSimpleUpdate, a Heisenberg-picture simulator using backward PEPO evolution with simple-update compression and reverse-light-cone filtering. Use get_evolved_operator for the evolved observable or local_expectation for its |00...0> expectation.

• Add CircuitBase, the shared base for all circuit simulators. Its from_* constructors accept representation-specific options.

• Add CircuitMPSLazy for deferred gate evaluation and periodic compression, which can be more efficient for long-range gates with src compression. MPS simulators also gain compute_marginal and sample_chaotic.

• Add OpenQASM 3 parsing through from_openqasm3_str, from_openqasm3_file, and from_openqasm3_url, supporting custom gates, register broadcasting, and symbolic input tracking.

• CircuitDense: support controlled gates through controls= without building the full dense operator.

Infinite 2D tensor networks

• Add quimb.tensor.tn2dinf for infinite-PEPS ground states: GeometryInfinite2D defines the unit cell, PEPSInfinite2D stores the state, LocalHamInfinite2D defines the Hamiltonian, and SimpleUpdateInfinite2D performs imaginary-time simple update. Support dense, Abelian-symmetric, and fermionic symmray backends, with cluster (max_distance) or generalized-loop (gloops) expectations. Hamiltonians can include longer-range terms than the state geometry.

• Add GeometryInfinite2D.square for square-lattice unit cells. Set neighbor range with couplings (neighbor shells or displacement vectors) or radius.

Gates, gauges, and lattice bonds

• tensor_network_ag_gate_simple: support long-range gates between sites without a direct bond through tensor_network_ag_gate_simple_long_range.

• TensorNetwork.gauge_all_simple: add fuse_multibonds to preserve multi-index bonds during gauging. tensor_compress_bond supports explicit bond-index selection; SimpleUpdateGen preserves multi-index bonds by default.

• Add tensor_gauge_simple_bond to gauge a single bond using a shared gauges dictionary.

• tensor_canonize_bond: add swap_inds to move indices between tensors during canonization and bond_ind for explicit bond selection.

• Add LatticeBondMap for consistent lattice bond indices across open and periodic boundaries.

• LocalHamGen.get_auto_ordering: return commuting layers with group=True for sort, random, and None. order="random-ungrouped" groups shuffled terms while preserving their order.

Belief propagation: gauging and loop expansions

• Add gauge_d2bp, TensorNetwork.gauge_all_belief_propagation, and D2BP.gauge_symmetric for symmetric gauging with dense 2-norm belief propagation.

• D2BP: add power, relative smudge conditioning, and converge_d2bp. Support BP messages as projector environments, including canonize="bp" in tensor_network_ag_compress_projector.

• TensorNetworkGenVector.norm_gloop_expand: add an info cache, shared with compute_local_expectation_gloop_expand for normalized="global". Supplied gauges are copied rather than modified. combine_local_contractions gains power.

• D1BP.contract_gloop_expand: include singleton regions and optionally remove dangling tensors without modifying the target network.

Norms, decompositions, and backend conversion

• TensorNetwork.norm, norm_gloop_expand, and normalize_simple accept strip_exponent to return the norm or normalization factor as separate mantissa and base-10 exponent values.

• eigh_truncated: add shift for optional diagonal regularization.

• Add Tensor.to and TensorNetwork.to to change backend, dtype, or device (autoray v0.9.0 or newer).

• trace_distance: speed up Hermitian trace distances using eigenvalues. Use isherm=False for the previous singular-value calculation.

Bug fixes (1.15.0)

Circuits

• CircuitDense: fix ValueError in psi, partial_trace, and local_expectation. get_uni raises the same clear ValueError as uni because the dense state cannot provide the unitary.

• CircuitPermMPS: return amplitude, to_dense, and local_expectation results in logical qubit order after lazy permutations.

• CircuitPermMPS and CircuitMPSLazy: preserve subclass attributes in copy() (GH #387).

• CircuitMPSLazy: apply pending gates and configured compression before state accessors run. Results no longer depend on accessor order (GH #387).

• CircuitMPS and subclasses: raise NotImplementedError for unsupported schrodinger_contract instead of an internal IndexError (GH #387).

• Circuit.get_rdm_lightcone_simplified: prevent partial_trace and local_expectation from returning results cached before later gates were applied (GH #398).

Belief propagation: messages and loop corrections

• TensorNetwork.gen_gloops: fix automatic sizing and joins for dangling target tensors. Add allow_dangling, on by default for local expectations and partial traces. "alldangle" and "anydangle" remain aliases.

• D1BP.contract_loop_series_expansion and HD1BP.contract_gloop_expand: generate loops with gloops=None instead of raising TypeError.

• D2BP.partial_trace_loop_series_expansion: fix incorrect reduced density matrices for complex states (GH #380).

• D2BP.normalize_tensors: prevent repeated reduced-density-matrix calculations from drifting after rescaling (GH #381).

• D2BP.compress: make later message updates use the compressed tensor data after in-place compression.

Compression, lattice boundaries, and norms

• tensor_network_1d_compress_src and tensor_network_1d_compress_srcmps: fix compression of networks with long-range bonds that skip sites.

• enforce_1d_like: connect long-range bonds correctly when site_tags reverse their site order, including sweep_reverse=True.

• PEPS, PEPO, and PEPS3D: keep open and periodic bonds distinct for cyclic dimensions of length 1 or 2, including bond dimension 1.

• TensorNetwork3D.contract_peps_sweep: preserve norm exponents and support separate mantissa/exponent returns with strip_exponent=True.

• TN_from_strings and random hidden-loop networks: fix normalization when the tensor-network exponent is nonzero.

• TensorNetwork2DVector.compute_norm: always return a scalar.

• TensorNetworkGenVector.norm_gloop_expand: include the tensor-network exponent.

• TensorNetwork.isel: support slices.

Internal (1.15.0)

• Convert quimb.tensor.circuit from a module to a package, preserving public imports and existing pickles.

Documentation (1.15.0)

• Sphinx AutoAPI documents each object only where it is defined. Support short suffix links such as Tensor.

v1.14.0 (2026-05-10)

Breaking Changes

• tensor_compress_bond: rename input tensor args ta and tb

Enhancements:

• D2BP: support fermionic tensor networks (only computing norm^2 so far, gating/compression need work).

• tensor_compress_bond: add reduce_opts for controlling the decomposition options used when reducing each tensor before the main truncating decomposition. For example reduce_opts={"method": "qr:cholesky"}.

• add TensorNetworkGen.select_sites as a convenience method for selecting a sub network given a list of sites.

• add PEPO_product_operator for bond-dimension-1 PEPOs given by a product of on-site operators, including cyclic boundary conditions via cyclic=True or cyclic=(cyclic_x, cyclic_y).

• PEPO: accept explicit cyclic kwarg in the constructor, to override shape-based boundary-condition inference (required for bond dimension 1 cyclic PEPOs).

• TensorNetworkGenOperator: add generic apply (dispatching on operator/vector tensor networks), trace and partial_transpose methods. These now work for arbitrary geometry operator tensor networks (including MPO and PEPO); partial_transpose supports arbitrary hashable site labels. apply also gains an inplace option that propagates to the acting operator rather than the one being acted on.

• add TensorNetworkGen.has_site to test whether an object is a valid site label of a tensor network. The generic implementation checks membership in the site set; 1D, 2D and 3D tensor networks override it with a fast bounds check.

• add LocalHam2D.build_pepo_propagator_trotterized for a first-order Trotter decomposition of expm(x H) as a PEPO. Accepts an ordering argument to control the order in which terms are applied.

• TensorNetwork.split_simplify: consider all candidate bipartitions for each tensor and accept the one that minimizes the resulting maximum tensor size, rather than the first reduction found.

• contract_hotrg, coarse_grain_hotrg, their 3D counterparts, and tensor_network_ag_compress_projector: add gauge_power parameter to control the power applied to the bond gauge weights when canonize=True before computing the compressed projectors.

• RegionGraph: add get_maximal_regions, get_minimal_regions, and get_maximal_ancestors helpers for querying the region hierarchy.

Drawing and schematic updates:

• Drawing: add orthographic projection mode alongside the existing axonometric projection via the new projection parameter (replaces a/b). Named presets include "orthographic", "axonometric", and "isometric".

• Drawing.translate: new context manager to temporarily offset all draw operations in coordinate space (before projection).

• Drawing.translate_screen: new context manager to temporarily offset all draw operations in screen space (after projection).

• Drawing.grid3d: automatically select back-facing planes based on projection so grids always appear behind the scene, use readable tick label orientations for all projections, and place axis labels correctly.

Bug fixes:

• CircuitPermMPS.sample: fix output bitstring ordering when the internal MPS qubit order is permuted (GH #327).

• TensorNetwork.split_tensor: fix handling of absorb=None, adding all tensors returned by the split (GH #260).

• D2BP.gate_: correctly mark touched tensors and rebuild local contraction expressions after applying gates.

• contract_hotrg and 3D counterpart: fix bug when specifying strip_exponent in final_contract_opts.

v1.13.0 (2026-03-19)

Breaking Changes

• ham_hubbard_hardcore fix description and sign convention of hopping strength t.

• heisenberg_from_edges fix sign convention of magnetic field terms.

• the quimb.tensor submodule structure has been refactored with tn1d, tn2d, tn3d, and tnag submodules for better organization. Imports from old locations will still work, but are deprecated. Public classes and functions such as MatrixProductState are directly accessible from the top level quimb.tensor module as before.

Enhancements:

Major updates to splitting/decomposing individual tensors/arrays:

• add array_split and array_svals as the primary array-level entry points for matrix decomposition, consolidating dispatch logic that was previously internal to tensor_core.

• add register_split_driver and register_svals_driver decorators for registering custom matrix decomposition methods with array_split and array_svals.

• allow array_split to handle batches of matrices (for most methods).

• array_split: automatically detect and forward valid kwargs to underlying decomposition methods.

• tensor_split and array_split: expand absorb options significantly beyond "left", "both", "right", None to include "lorthog", "rorthog", "lfactor", "rfactor", "lsqrt", "rsqrt and "s" for returning partial results (single factors or singular values only). Default changed from "both" to "auto", which uses each method’s natural default.

• add method "svd:eig" with main implementation svd_via_eig for efficient SVD via hermitian eigen-decomposition, with shortcuts for all absorb modes. This can be faster (especially e.g. on GPU) than the standard SVD, but entails some loss of precision.

• tensor_split: rename method option "eig" to "svd:eig" to make it clearer that this is an SVD split via eigen-decomposition. "eig" remains as a deprecated alias for "svd:eig".

• add method "svd:rand" with main implementation svd_rand_truncated for randomized SVD with truncation, with shortcuts for all absorb modes. (This is a new and backend agnostic implementation as opposed to the existing 'rsvd' method).

• add method "qr:cholesky" qr_via_cholesky for efficient QR or LQ like decompositions via cholesky decomposition, with shortcuts for all absorb modes. This can be faster than the standard QR (especially on GPU) but entails some loss of precision.

• tensor_split and array_split: add "lsqrt" and "rsqrt" absorb options, update cholesky decomposition to cholesky_regularized with shift as exposed parameter.

• compute_oblique_projectors: allow method kwarg.

• QR decomposition: add stabilize kwarg for controlling QR stabilization behavior.

• decomposition methods: various compatibility improvements for JAX backend.

Other enhancements:

• add shift and clock operators.

• add Tensor.isfermionic and TensorNetwork.isfermionic methods.

• add Tensor.isblocksparse and TensorNetwork.isblocksparse methods.

• add phase_dual option to TensorNetwork.conj.

• rename tensor_network_1d_compress_zipup_first to tensor_network_1d_compress_zipup_oversample and standardise oversample arguments.

• add tensor_network_1d_compress_srcmps_oversample and tensor_network_1d_compress_fit_oversample methods.

• add connected_bipartitions for finding all connected bipartitions of a tensor network

• tn.distribute_exponent: add new_exponent option for specifying the new exponent value (default 0.0).

• tensor_network_1d_compress: correctly handle input networks with non-zero exponents and equalize_norms.

• add tensor_network_gate_sandwich_inds for applying a gate and its conjugate like \(G A G^\dagger\) to a tensor network.

• tensor_network_ag_gate: add which="sandwich" option for applying a gate and its conjugate like \(G A G^\dagger\) to a tensor network, default to this if the supplied tensor network is a TensorNetworkGenOperator.

• add function tensor_network_ag_gate_simple for applying a gate to an arbitrary geometry tensor network vector or operator, using simple update style gauges to perform any compression.

• insert_compressor_between_regions and upstream CTMRG/HOTRG methods: add explicit contract_opts, reduce_opts, and compress_opts keyword arguments for fine-grained control.

• TensorNetwork2D.contract_boundary, contract_ctmrg, contract_hotrg, coarse_grain_hotrg and their 3D counterparts: add strip_exponent parameter and equalize_norms="auto" default.

• TensorNetwork3D.contract_hotrg: use updated projecting/gauging scheme.

• all compression methods: accept an explicit compress_opts kwarg.

• tensor_network_ag_compress: allow fine-grained control over split options via compress_opts.

• TensorNetworkGen.flatten: add arbitrary geometry flatten method, used in 1D/2D/3D.

• RegionGraph: various improvements.

• add hash_kwargs_to_int utility for hashing keyword arguments to a deterministic integer.

Bug fixes:

• fix isometrize_qr for complex torch arrays (GH #346).

• fix right_canonicalize to return the right canonicalized tensor network (GH #347)

• ensure all belief propagation contraction methods correctly propagate the target tensor network’s .exponent.

• fix cutoff mode bug in array_split decomposition truncation.

• fix tensor_network_1d_compress_zipup equalize_norms exponent accumulation.

• fix final_contract_opts inplace handling in boundary contraction methods.

• fix squared_op_to_reduced_factor argument handling.

• fix cholesky decomposition shift kwarg forwarding and absorb="right" direction.

• fix sample_hd1bp sub-progress bar display.

• fix gate tag propagation in tensor_network_gate_inds.

• handle equalize_norms correctly in TensorNetwork2D.compute_environments (GH #352).

v1.12.1 (2026-01-12)

Breaking Changes

• bump minimum required python version to 3.11

Bug fixes:

• fix SimpleUpdateGen mixin inheritance order.

• fix insert_compressor_between_regions for fermionic tensor networks with bond signature +-.

v1.12.0 (2026-01-09)

Enhancements:

• move the experimental operatorbuilder module to the main quimb.operator module.

• add basic introduction to the operator module - Basics

• add new example on tracing tensor network functions Tracing tensor network functions and reusing intermediates

• tensor_split: add an info kwarg, supplying this with an empty dict or with the entry 'error' will store the truncation error when using method in {"svd", "svd:eig"}.

• update infrastructure for TEBD and SimpleUpdate based algorithms.

• schematic.Drawing: add grid, grid3d, bezier, star, cross and zigzag methods.

• schematic.Drawing: add relative option to arrowhead, shorten option to text_between and text_left and text_right options to line.

• add Drawing.scale_figsize for automatically setting the absolute figsize based on placed elements.

• refactor TEBDGen and SimpleUpdateGen

• update the 2d specific SimpleUpdate to use the new infrastructure.

• tn.draw(): show abelian signature if using symmray arrays.

• tn.draw(): add adjust_lims option

• TNOptimizer: allow autodiff_backend="torch" with jit_fn=True to work with array backends with general pytree parameters, e.g. symmray arrays.

• tn.gen_gloops and tn.gen_gloops_sites: add join_overlap option. When building cluster by joining smaller generalized loops, this option controls how many nodes they need to overlap by to be joined together.

• all message passing routines: add callback option

• GBP: allow a message initilization function.

• D1BP: allow messages to be a callable initialization function.

• MatrixProductState.gate_nonlocal: add method="lazy" option for lazily applying a non-local gate as a sub-MPO without contraction or compression.

• LocalHamGen.apply_to_arrays: support pytree parameter arrays such as symmray.

• add Tensor.get_namespace and TensorNetwork.get_namespace for getting a reusable data array namespace

• TensorNetwork.isel: use take where possible to better support e.g. torch.vmap across amplitudes.

• MatrixProductState.measure, and MatrixProductState.sample: add backend_random option for specifying which backend to use for random number generation when sampling, this can be set for example to jax to make the whole process jittable, but by default is numpy, regardless of the actual array backend.

Bug fixes:

• fix insert_compressor_between_regions when insert_into is None.

• tensor network drawing, ensure hyper indices can be specified as output_inds.

• fix MatrixProductState.measure when using jax arrays (GH #340).

• fix MatrixProductState.measure when projecting and keeping a site site (GH #344).

v1.11.2 (2025-07-30)

Enhancements:

• Update the introduction to tensor contraction docs

• Improve efficiency of 1D structured contractions when default optimize is used, especially for large bond dimension overlaps.

Bug fixes:

• fixes for MPS and MPO constructors when L=1, (GH #314)

• tensor splitting with absorb=”left” now correctly marks left indices.

• tn.isel: fix bug when value could not be compared to string "r"

• truncated svd, make n_chi comparison more robust to different backends

v1.11.1 (2025-06-20)

Enhancements:

• add create_bond to tensor_canonize_bond and tensor_compress_bond for optionally creating a new bond between two tensors if they don’t already share one. Add as a flag to TensorNetwork1DFlat.compress and related functions (GH #294).

• add ensure_bonds_exist for ensuring that all bonds in a 1D flat tensor network exist. Use this in the permute_arrays methods and optionally in the expand_bond_dimension method.

• tn.draw(): permit empty network, and allow color=True to automatically color all tags.

• tn.add_tag: add a record: Optional[dict] kwarg, to allow for easy rewinding of temporary tags without tracking the actual networks.

• add qu.plot as a quick wrapper for calling matplotlib.pyplot.plot with the quimb style.

• quimb.schematic: add zorder_delta kwarg for fine adjustments to layering of objects in approximately the same position.

• operatorbuilder: big performance improvements and fixes for building matrix representations including Z2 symmetry. Add default symmetry and sector options that can be overridden at build time. Add lazy (slow, matrix free) ‘apply’ method. Add pauli_decompose transformation. Add experimental PEPO builder for nearest neighbor operators. Add unit tests.

Bug fixes:

• Fix TensorNetwork2D.compute_plaquette_environments for mode="zipup" and other boundary contraction methods that use the generic 1D compression algorithms.

• parse_openqasm2_str allow custom gate names to start with the word gate (GH #312).

• MatrixProductState.gate_with_mpo: fix bug to do with inplace argument (GH #313).

v1.11.0 (2025-05-14)

Breaking Changes

• move belief propagation to quimb.tensor.belief_propagation

• calling tn.contract() when an non-zero value has been accrued into tn.exponent now automatically re-absorbs that exponent.

• binary tensor operations that would previously have errored now will align and broadcast

Enhancements:

• Tensor: make binary operations (+, -, *, /, **) automatically align and broadcast indices. This would previously error.

• MatrixProductState.measure: add a seed kwarg

• belief propagation, implement DIIS (direct inversion in the iterative subspace)

• belief propagation, unify various aspects such as message normalization and distance.

• belief propagation, add a plot method.

• belief propagation, add a contract_every option.

• HV1BP: vectorize both contraction and message initialization

• add qu.plot_multi_series_zoom for plotting multiple series with a zoomed inset, useful for various convergence plots such as BP

• add info option to tn.gauge_all_simple for tracking extra information such as number of iterations and max gauge diffs

• Tensor.gate: add transposed option

• TensorNetwork.contract: add strip_exponent option for return the mantissa and exponent (log10) separately. Compatible with contract_tags, contract_cumulative, contract_compressed sub modes.

• tensor_split: add matrix_svals option, if True any returned singular values are put into the diagonal of a matrix (by default, False, they are returned as a vector).

• add Tensor.new_ind_pair_diag for expanding an existing index into a pair of new indices, such that the diagonal of the new tensor on those indices is the old tensor.

• TNOptimizer: add ‘cautious’ ADAM

• TensorNetwork.pop_tensor: allow tid or tags to be specified.

• add an example notebook for converting hyper tensor networks to normal tensor networks, for approximate contraction - Converting a hyper TN for approximate contraction

• add “SX” and “SXDG” gates to Circuit (PR #277)

• add “XXPLUSYY” and “XXPLUSYY” gates to Circuit (PR #279)

• add progress bar to various Circuit methods (PR #288)

• quimb.operator: fix MPO building for congested operators (GH #296 and GH #301), allow arbitrary dtype (GH #289). Fix building of sparse and matrix representations for non-translationally symmetric operators and operators with trivial (all identity) terms.

Bug fixes:

• fix MatrixProductState.measure for cupy backend arrays (GH #276).

• fix linalg.expm dispatch (GH #275)

• fix ‘dm’ 1d compress method for disconnected subgraphs

• fix docs source lookup in quimb.tensor module

• fix raw gate copying in Circuit (GH #285)

v1.10.0 (2024-12-18)

Enhancements:

• tensor network fitting: add method="tree" for when ansatz is a tree - tensor_network_fit_tree

• tensor network fitting: fix method="als" for complex networks

• tensor network fitting: allow method="als" to use a iterative solver suited to much larger tensors, by default a custom conjugate gradient implementation.

• tensor_network_distance and fitting: support hyper indices explicitly via output_inds kwarg

• add tn.make_overlap and tn.overlap for computing the overlap between two tensor networks, \(\langle O |T \rangle\), with explicit handling of outer indices to address hyper networks. Add output_inds to tn.norm and tn.make_norm also, as well as the squared kwarg.

• replace all numba based paralellism (prange and parallel vectorize) with explicit thread pool based parallelism. Should be more reliable and no need to set NUMBA_NUM_THREADS anymore. Remove env var QUIMB_NUMBA_PAR.

• Circuit: add dtype and convert_eager options. dtype specifies what the computation should be performed in. convert_eager specifies whether to apply this (and any to_backend calls) as soon as gates are applied (the default for MPS circuit simulation) or just prior to contraction (the default for exact contraction simulation).

• tn.full_simplify: add check_zero (by default set of "auto") option which explicitly checks for zero tensor norms when equalizing norms to avoid log10(norm) resulting in -inf or nan. Since it creates a data dependency that breaks e.g. jax tracing, it is optional.

• schematic.Drawing: add shorten kwarg to line drawing and curve drawing and examples to schematic - manual drawing.

• TensorNetwork: add .backend and .dtype_name properties.

v1.9.0 (2024-11-19)

Breaking Changes

• renamed MatrixProductState.partial_trace and MatrixProductState.ptr to MatrixProductState.partial_trace_to_mpo to avoid confusion with other partial_trace methods that usually produce a dense matrix.

Enhancements:

• add Circuit.sample_gate_by_gate and related methods Circuit.reordered_gates_dfs_clustered and Circuit.get_qubit_distances for sampling a circuit using the ‘gate by gate’ method introduced in https://arxiv.org/abs/2112.08499.

• add CircuitBase.draw for drawing a very simple circuit schematic.

• Circuit: by default turn on simplify_equalize_norms and use a group_size=10 for sampling. This should result in faster and more stable sampling.

• Circuit: use numpy.random.default_rng for random number generation.

• add qtn.circ_a2a_rand for generating random all-to-all circuits.

• expose qtn.edge_coloring as top level function and allow layers to be returned grouped.

• add docstring for tn.contract_compressed and by default pick up important settings from the supplied contraction path optimizer (max_bond and compress_late)

• add Tensor.rand_reduce for randomly removing a tensor index by contracting a random vector into it. One can also supply the value "r" to isel selectors to use this.

• add fit-zipup and fit-projector shorthand methods to the general 1d tensor network compression function

• add MatrixProductState.compute_local_expectation for computing many local expectations for a MPS at once, to match the interface for this method elsewhere. These can either be computed via canonicalization (method="canonical"), or via explicit left and right environment contraction (method="envs")

• specialize CircuitMPS.local_expectation to make use of the MPS form.

• add PEPS.product_state for constructing a PEPS representing a product state.

• add PEPS.vacuum for constructing a PEPS representing the vacuum state \(|000\ldots0\rangle\).

• add PEPS.zeros for constructing a PEPS whose entries are all zero.

• tn.gauge_all_simple: improve scheduling and add damping and touched_tids options.

• qtn.SimpleUpdateGen: add gauge difference update checking and tol and equilibrate settings. Update .plot() method. Default to a small cutoff.

• add psi.sample_configuration_cluster for sampling a tensor network using the simple update or cluster style environment approximation.

• add the new doc Exploring circuit sampling with various methods

---

v1.8.4 (2024-07-20)

Bug fixes:

• fix for MPS sampling with fixed seed (GH #247 and PR #248)

• fix for mps_gate_with_mpo_lazy (GH #246).

---

v1.8.3 (2024-07-10)

Enhancements:

• support for numpy v2.0 and scipy v1.14

• add MPS sampling: MatrixProductState.sample_configuration and MatrixProductState.sample (generating multiple samples) and use these for CircuitMPS.sample and CircuitPermMPS.sample.

• add basic .plot() method for SimpleUpdate classes

• add edges_1d_chain for generating 1D chain edges

• operatorbuilder: better coefficient placement for long range MPO building

---

v1.8.2 (2024-06-12)

Enhancements:

• TNOptimizer can now accept an arbitrary pytree (nested combination of dicts, lists, tuples, etc. with TensorNetwork, Tensor or raw array_like objects as the leaves) as the target object to optimize.

• TNOptimizer can now directly optimize Circuit objects, returning a new optimized circuit with updated parameters.

• Circuit: add .copy(), .get_params() and .set_params() interface methods.

• Update generic TN optimizer docs.

• add tn.gen_paths_loops for generating all loops of indices in a TN.

• add tn.gen_inds_connected for generating all connected sets of indices in a TN.

• make SVD fallback error catching more generic (PR #238)

• fix some windows + numba CI issues.

• approx_spectral_function add plotting and tracking

• add dispatching to various tensor primitives to allow overriding

---

v1.8.1 (2024-05-06)

Enhancements:

• CircuitMPS now supports multi qubit gates, including arbitrary multi-controls (which are treated in a low-rank manner), and faster simulation via better orthogonality center tracking.

• add CircuitPermMPS

• add MatrixProductState.gate_nonlocal for applying a gate, supplied as a raw matrix, to a non-local and arbitrary number of sites. The kwarg contract="nonlocal" can be used to force this method, or the new option "auto-mps" will select this method if the gate is non-local (GH #230)

• add MatrixProductState.gate_with_mpo for applying an MPO to an MPS, and immediately compressing back to MPS form using tensor_network_1d_compress

• add MatrixProductState.gate_with_submpo for applying an MPO acting only of a subset of sites to an MPS

• add MatrixProductOperator.from_dense for constructing MPOs from dense matrices, including an only subset of sites

• add MatrixProductOperator.fill_empty_sites for ‘completing’ an MPO which only has tensors on a subset of sites with (by default) identities

• MatrixProductState and MatrixProductOperator, now support the sites kwarg in common constructors, enabling the TN to act on a subset of the full L sites.

• add TensorNetwork.drape_bond_between for ‘draping’ an existing bond between two tensors through a third

• add Tensor.new_ind_pair_with_identity

• TN2D, TN3D and arbitrary geom classical partition function builders (TN_classical_partition_function_from_edges) now all support outputs= kwarg specifying non-marginalized variables

• add simple dense 1-norm belief propagation algorithm D1BP

• add qtn.enforce_1d_like for checking whether a tensor network is 1D-like, including automatically adding strings of identities between non-local bonds, expanding applicability of tensor_network_1d_compress

• add MatrixProductState.canonicalize as (by default non-inplace) version of canonize, to follow the pattern of other tensor network methods. canonize is now an alias for canonicalize_ [note trailing underscore].

• add MatrixProductState.left_canonicalize as (by default non-inplace) version of left_canonize, to follow the pattern of other tensor network methods. left_canonize is now an alias for left_canonicalize_ [note trailing underscore].

• add MatrixProductState.right_canonicalize as (by default non-inplace) version of right_canonize, to follow the pattern of other tensor network methods. right_canonize is now an alias for right_canonicalize_ [note trailing underscore].

Bug fixes:

• CircuitBase.apply_gate_raw: fix kwarg bug (PR #226)

• fix for retrieving opt_einsum.PathInfo for single scalar contraction (GH #231)

---

v1.8.0 (2024-04-10)

Breaking Changes

• all singular value renormalization is turned off by default

• TensorNetwork.compress_all now defaults to using some local gauging

Enhancements:

• add quimb.tensor.tn1d.compress.py with functions for compressing generic 1D tensor networks (with arbitrary local structure) using various methods. The methods are:

  • The ‘direct’ method: tensor_network_1d_compress_direct

  • The ‘dm’ (density matrix) method: tensor_network_1d_compress_dm

  • The ‘zipup’ method: tensor_network_1d_compress_zipup

  • The ‘zipup-oversample’ method: tensor_network_1d_compress_zipup_oversample

  • The 1 and 2 site ‘fit’ or sweeping method: tensor_network_1d_compress_fit

  • … and some more niche methods for debugging and testing.

  And can be accessed via the unified function tensor_network_1d_compress. Boundary contraction in 2D can now utilize any of these methods.

• add quimb.tensor.tnag.compress.py with functions for compressing arbitrary geometry tensor networks using various methods. The methods are:

  • The ‘local-early’ method: tensor_network_ag_compress_local_early

  • The ‘local-late’ method: tensor_network_ag_compress_local_late

  • The ‘projector’ method: tensor_network_ag_compress_projector

  • The ‘superorthogonal’ method: tensor_network_ag_compress_superorthogonal

  • The ‘l2bp’ method: tensor_network_ag_compress_l2bp

  And can be accessed via the unified function tensor_network_ag_compress. 1D compression can also fall back to these methods.

• support PBC in tn2d.contract_hotrg, tn2d.contract_ctmrg, tn3d.contract_hotrg and the new function tn3d.contract_ctmrg.

• support PBC in gen_2d_bonds and gen_3d_bonds, with cyclic kwarg.

• support PBC in TN2D_rand_hidden_loop and TN3D_rand_hidden_loop, with cyclic kwarg.

• support PBC in the various base PEPS and PEPO construction methods.

• add tensor_network_apply_op_op for applying ‘operator’ TNs to ‘operator’ TNs.

• tweak tensor_network_apply_op_vec for applying ‘operator’ TNs to ‘vector’ or ‘state’ TNs.

• add tnvec.gate_with_op_lazy method for applying ‘operator’ TNs to ‘vector’ or ‘state’ TNs like \(x \rightarrow A x\).

• add tnop.gate_upper_with_op_lazy method for applying ‘operator’ TNs to the upper indices of ‘operator’ TNs like \(B \rightarrow A B\).

• add tnop.gate_lower_with_op_lazy method for applying ‘operator’ TNs to the lower indices of ‘operator’ TNs like \(B \rightarrow B A\).

• add tnop.gate_sandwich_with_op_lazy method for applying ‘operator’ TNs to the upper and lower indices of ‘operator’ TNs like \(B \rightarrow A B A^\dagger\).

• unify all TN summing routines into `tensor_network_ag_sum, which allows summing any two tensor networks with matching site tags and outer indices, replacing specific MPS, MPO, PEPS, PEPO, etc. summing routines.

• add rand_symmetric_array, rand_tensor_symmetric TN2D_rand_symmetric for generating random symmetric arrays, tensors and 2D tensor networks.

Bug fixes:

• fix scipy sparse monkey patch for scipy>=1.13 (GH #222)

• fix autoblock bug where connected sectors were not being merged (GH #223)

---

v1.7.3 (2024-02-08)

Enhancements:

• qu.randn: support dist="rademacher".

• support dist and other randn options in various TN builders.

Bug fixes:

• restore fallback (to scipy.linalg.svd with driver=’gesvd’) behavior for truncated SVD with numpy backend.

---

v1.7.2 (2024-01-30)

Enhancements:

• add normalized=True option to tensor_network_distance for computing the normalized distance between tensor networks: \(2 |A - B| / (|A| + |B|)\), which is useful for convergence checks. Tensor.distance_normalized and TensorNetwork.distance_normalized added as aliases.

• add TensorNetwork.cut_bond for cutting a bond index

Bug fixes:

• removed import of deprecated numba.generated_jit decorator.

---

v1.7.1 (2024-01-30)

Enhancements:

• add TensorNetwork.visualize_tensors for visualizing the actual data entries of an entire tensor network.

• add ham.build_mpo_propagator_trotterized for building a trotterized propagator from a local 1D hamiltonian. This also includes updates for creating ‘empty’ tensor networks using TensorNetwork.new, and building up gates from empty tensor networks using TensorNetwork.gate_inds_with_tn.

• add more options to Tensor.expand_ind and Tensor.new_ind: repeat tiling mode and random padding mode.

• tensor decomposition: make eigh_truncated backend agnostic.

• tensor_compress_bond: add reduced="left" and reduced="right" modes for when the pair of tensors is already in a canonical form.

• add qtn.TN2D_embedded_classical_ising_partition_function for constructing 2D (triangular) tensor networks representing all-to-all classical ising partition functions.

Bug fixes:

• fix bug in kruas_op when operator spanned multiple subsystems (GH #214)

• fix bug in qr_stabilized when the diagonal of R has significant imaginary parts.

• fix bug in quantum discord computation when the state was diagonal (GH #217)

---

v1.7.0 (2023-12-08)

Breaking Changes

• Circuit : remove target_size in preparation for all contraction specifications to be encapsulated at the contract level (e.g. with cotengra)

• some TN drawing options (mainly arrow options) have changed due to the backend change detailed below.

Enhancements:

• TensorNetwork.draw: use quimb.schematic for main backend="matplotlib" drawing. Enabling:

  1. multi tag coloring for single tensors

  2. arrows and labels on multi-edges

  3. better sizing of tensors using absolute units

  4. neater single tensor drawing, in 2D and 3D

• add quimb.schematic.Drawing from experimental submodule, add example docs at schematic - manual drawing. Add methods text_between, wedge, line_offset and other tweaks for future use by main TN drawing.

• upgrade all contraction to use cotengra as the backend

• Circuit : allow any gate to be controlled by any number of qubits.

• Circuit : support for parsing openqasm2 specifications now with custom and nested gate definitions etc.

• add is_cyclic_x, is_cyclic_y and is_cyclic_z to TensorNetwork2D and TensorNetwork3D.

• add TensorNetwork.compress_all_1d for compressing generic tensor networks that you promise have a 1D topology, without casting as a TensorNetwork1D.

• add MatrixProductState.from_fill_fn for constructing MPS from a function that fills the tensors.

• add Tensor.idxmin and Tensor.idxmax for finding the index of the minimum/maximum element.

• 2D and 3D classical partition function TN builders: allow output indices.

• quimb.tensor.belief_propagation: add various 1-norm/2-norm dense/lazy BP algorithms.

Bug fixes:

• fixed bug where an output index could be removed by squeezing when performing tensor network simplifications.

---

v1.6.0 (2023-09-10)

Breaking Changes

• Quantum circuit RZZ definition corrected (angle changed by -1/2 to match qiskit).

Enhancements:

• add OpenQASM 2.0 parsing support: CircuitBase.from_openqasm2_file,

• Circuit: add RXX, RYY, CRX, CRY, CRZ, toffoli, fredkin, givens gates

• truncate TN pretty html reprentation to 100 tensors for performance

• add Tensor.sum_reduce and Tensor.vector_reduce

• contract_compressed, default to ‘virtual-tree’ gauge

• add TN_rand_tree

• experimental.operatorbuilder: fix parallel and heisenberg builder

• make parametrized gate generation even more robost (ensure matching types so e.g. tensorflow can be used)

Bug fixes:

• fix gauge size check for some backends

---

v1.5.1 (2023-07-28)

Enhancements:

• add MPS_COPY().

• add ‘density matrix’ and ‘zip-up’ MPO-MPS algorithms.

• add drop_tags option to tensor_core.tensor_contract()

• compress_all_simple(), allow cutoff.

• add structure checking debug methods: Tensor.check() and TensorNetwork.check().

• add several direction contraction utility functions: get_symbol, inds_to_eq() and array_contract().

Bug fixes:

• Circuit: use stack for more robust parametrized gate generation

• fix for gate_with_auto_swap() for i > j.

• fix bug where calling tn.norm() would mangle indices.

---

v1.5.0 (2023-05-03)

Enhancements

• refactor ‘isometrize’ methods including new “cayley”, “householder” and “torch_householder” methods. See decomp.isometrize().

• add TensorNetwork.compute_reduced_factor() and TensorNetwork.insert_compressor_between_regions() methos, for some RG style algorithms.

• add the mode="projector" option for 2D tensor network contractions

• add HOTRG style coarse graining and contraction in 2D and 3D. See TensorNetwork2D.coarse_grain_hotrg(), TensorNetwork2D.contract_hotrg(), TensorNetwork3D.coarse_grain_hotrg(), and TensorNetwork3D.contract_hotrg(),

• add CTMRG style contraction for 2D tensor networks: TensorNetwork2D.contract_ctmrg()

• add 2D tensor network ‘corner double line’ (CDL) builders: TN2D_corner_double_line()

• update the docs to use the furo theme, myst_nb for notebooks, and several other sphinx extensions.

• add the 'adabelief' optimizer to TNOptimizer as well as a quick plotter: TNOptimizer.plot()

• add initial 3D plotting methods for tensors networks ( TensorNetwork.draw(dim=3, backend='matplotlib3d') or TensorNetwork.draw(dim=3, backend='plotly') ). The new backend='plotly' can also be used for 2D interactive plots.

• Update HTN_from_cnf() to handle more weighted model counting formats.

• Add cnf_file_parse()

• Add random_ksat_instance()

• Add TN_from_strings()

• Add convert_to_2d()

• Add TN2D_rand_hidden_loop()

• Add convert_to_3d()

• Add TN3D_corner_double_line()

• Add TN3D_rand_hidden_loop()

• various optimizations for minimizing computational graph size and construction time.

• add 'lu', 'polar_left' and 'polar_right' methods to tensor_split().

• add experimental arbitrary hamilotonian MPO building

• TensorNetwork: allow empty constructor (i.e. no tensors representing simply the scalar 1)

• TensorNetwork.drop_tags(): allow all tags to be dropped

• tweaks to compressed contraction and gauging

• add jax, flax and optax example

• add 3D and interactive plotting of tensors networks with via plotly.

• add pygraphiviz layout options

• add TensorNetwork.combine() for unified handling of combining tensor networks potentially with structure

• add HTML colored pretty printing of tensor networks for notebooks

• add quimb.experimental.cluster_update.py

Bug fixes:

• fix qr_stabilized() bug for strictly upper triangular R factors.

---

v1.4.2 (2022-11-28)

Enhancements

• move from versioneer to to setuptools_scm for versioning

---

v1.4.1 (2022-11-28)

Enhancements

• unify much functionality from 1D, 2D and 3D into general arbitrary geometry class TensorNetworkGen

• refactor contraction, allowing using cotengra directly

• add Tensor.visualize() for visualizing the actual data entries of an arbitrarily high dimensional tensor

• add Gate class for more robust tracking and manipulation of gates in quantum Circuit simulation

• tweak TN drawing style and layout

• tweak default gauging options of compressed contraction

• add TensorNetwork.compute_hierarchical_grouping()

• add Tensor.as_network()

• add TensorNetwork.inds_size()

• add TensorNetwork.get_hyperinds()

• add TensorNetwork.outer_size()

• improve tensor_core.group_inds()

• refactor tensor decompositiona and ‘isometrization’ methods

• begin supporting pytree specifications in TNOptimizer, e.g. for constants

• add experimental submodule for new sharing features

• register tensor and tensor network objects with jax pytree interface (PR #150)

• update CI infrastructure

Bug fixes:

• fix force atlas 2 and weight_attr bug (GH #126)

• allow unpickling of PTensor objects (GH #128, PR #131)

---

v1.4.0 (2022-06-14)

Enhancements

• Add 2D tensor network support and algorithms

• Add 3D tensor network infrastructure

• Add arbitrary geometry quantum state infrastructure

• Many changes to TNOptimizer

• Many changes to TN drawing

• Many changes to Circuit simulation

• Many improvements to TN simplification

• Make all tag and index operations deterministic

• Add tensor_network_sum(), tensor_network_distance() and TensorNetwork.fit()

• Various memory and performance improvements

• Various graph generators and TN builders

---

v1.3.0 (2020-02-18)

Enhancements

• Added time dependent evolutions to Evolution when integrating a pure state - see Time-Dependent Evolutions - as well as supporting LinearOperator defined hamiltonians (PR #40).

• Allow the Evolution callback compute= to optionally access the Hamiltonian (PR #49).

• Added Tensor.randomize() and TensorNetwork.randomize() to randomize tensor and tensor network entries.

• Automatically squeeze tensor networks when rank-simplifying.

• Add TensorNetwork1DFlat.compress_site() for compressing around single sites of MPS etc.

• Add MPS_ghz_state() and MPS_w_state() for building bond dimension 2 open boundary MPS reprentations of those states.

• Various changes in conjunction with autoray to improve the agnostic-ness of tensor network operations with respect to the backend array type.

• Add tensor_core.new_bond() on top of Tensor.new_ind() and Tensor.expand_ind() for more graph orientated construction of tensor networks, see Graph Orientated Tensor Network Creation.

• Add the operators.fsim() gate.

• Make the parallel number generation functions use new numpy 1.17+ functionality rather than randomgen (which can still be used as the underlying bit generator) (PR #50)

• TN: rename contraction_complexity to TensorNetwork.contraction_width().

• TN: update TensorNetwork.rank_simplify(), to handle hyper-edges.

• TN: add TensorNetwork.diagonal_reduce(), to automatically collapse all diagonal tensor axes in a tensor network, introducing hyper edges.

• TN: add TensorNetwork.antidiag_gauge(), to automatically flip all anti-diagonal tensor axes in a tensor network allowing subsequent diagonal reduction.

• TN: add TensorNetwork.column_reduce(), to automatically identify tensor axes with a single non-zero column, allowing the corresponding index to be cut.

• TN: add TensorNetwork.full_simplify(), to iteratively perform all the above simplifications in a specfied order until nothing is left to be done.

• TN: add num_tensors and num_indices attributes, show num_indices in __repr__.

• TN: various improvements to the pytorch optimizer (PR #34)

• TN: add some built-in 1D quantum circuit ansatzes: circ_ansatz_1D_zigzag(), circ_ansatz_1D_brickwork(), and circ_ansatz_1D_rand().

• TN: add parametrized tensors PTensor and so trainable, TN based quantum circuits – see Tensor Network Training of Quantum Circuits.

Bug fixes:

• Fix consistency of fidelity() by making the unsquared version the default for the case when either state is pure, and always return a real number.

• Fix a bug in the 2D system example for when j != 1.0

• Add environment variable QUIMB_NUMBA_PAR to set whether numba should use automatic parallelization - mainly to fix travis segfaults.

• Make cache import and initilization of petsc4py and slepc4py more robust.

---

v1.2.0 (2019-06-06)

Enhancements

• Added kraus_op() for general, noisy quantum operations

• Added projector() for constructing projectors from observables

• Added calc.measure() for measuring and collapsing quantum states

• Added cprint() pretty printing states in computational basis

• Added calc.simulate_counts() for simulating computational basis counts

• TN: Add TensorNetwork.rank_simplify()

• TN: Add TensorNetwork.isel()

• TN: Add TensorNetwork.cut_iter()

• TN: Add 'split-gate' gate mode

• TN: Add TNOptimizer for tensorflow based optimization of arbitrary, contstrained tensor networks.

• TN: Add Dense1D.rand()

• TN: Add connect() to conveniently set a shared index for tensors

• TN: make many more tensor operations agnostic of the array backend (e.g. numpy, cupy, tensorflow, …)

• TN: allow align_TN_1D() to take an MPO as the first argument

• TN: add SpinHam1D.build_sparse()

• TN: add Tensor.unitize() and TensorNetwork.unitize() to impose unitary/isometric constraints on tensors specfied using the left_inds kwarg

• Many updates to tensor network quantum circuit (Circuit) simulation including:

  • CircuitMPS

  • CircuitDense

  • 49-qubit depth 30 circuit simulation example Quantum Circuits

• Add from quimb.gates import * as shortcut to import X, Z, CNOT, ....

• Add U_gate() for parametrized arbitrary single qubit unitary

Bug fixes:

• Fix pkron for case len(dims) == len(inds) (GH #17, PR #18).

• Fix qarray printing for older numpy versions

• Fix TN quantum circuit bug where Z and X rotations were swapped

• Fix variable bond MPO building (GH #22) and L=2 DMRG

• Fix norm(X, 'trace') for non-hermitian matrices

• Add autoray as dependency (GH #21)