[v0.11.0] - 2024-02-20

Release Notes

Several fixes and additions have accumulated since the last release.

We now officially support python 3.12.

If you have ever defined a custom model and used add_multi_coupling_term() with plus_hc=True, please note issue #218!

Additionally, there are several bugfixes listed in the changelog below that may have caused silent problems (wrong results without raising errors) in some cases.

Changelog

Backwards incompatible changes

• nothing yet

Added

• tJModel and SpinHalfHoleSite

• QRBasedTEBDEngine

• ClockModel, ClockChain and ClockSite

• Simulation parameters Simulation.measure_at_algorithm_checkpoints and Simulation.canonicalize_before_measurement

• BaseMPSExpectationValues parent class to unify the framework of computing expectation values and correlation functions in MPS and MPSEnvironment

• Abstract BaseEnvironment parent class for MPSEnvironment and MPOEnvironment

• Add phi_ext parameter to FermionModel, BoseHubbardModel and FermiHubbardModel.

• Option allow_incommensurate for from_lat_product_state().

• Most “important” objects (this obvious involves some judgement calls, regarding what counts as important) are now directly exposed in the top-level namespace of the tenpy package, i.e. you can now from tenpy import MPS, tensordot, TwoSiteDMRGEngine or import tenpy as tp and then use tp.tensordot etc. All objects which are not “private” and/or “implementation details” are exposed in the subpackage namespace, e.g. you can from tenpy.networks import MPOGraph.

• Overlaps of finite, shifted/translated MPS in overlap_translate_finite().

• New MPS construction method from_product_mps_covering() as a generalization of from_singlets.

Changed

• Change the build system and metadata declaration to pyproject.toml format. This makes installation more future-proof and stable, but should not affect how tenpy is used, once installed.

• Allow couplings parameters in the MixedXKModel methods add_inter_ring_hopping(), add_intra_ring_hopping(), add_inter_ring_interaction(), and add_inter_ring_interaction() to vary with x.

• Renamed the module lanczos to tenpy.linalg.krylov_based.

• The move_right attribute of Sweep now supports a third value, None, in addition to True, False. None means that the sweep will not move at all, i.e. the next update will be at the same position than the current one. This happens e.g. in TDVP.

• Mixers have been generalized and are no longer specialized for use in DMRG. Method names and signatures have been changed. The mixer classes are now implemented in tenpy.linalg.algorithms.mps_common. Backwards-compatible wrappers with the old method names and signatures will be kept in tenpy.linalg.algorithms.dmrg until v1.0.

• Introduce IterativeSweeps to generalize algorithms that repeat sweeps until convergence. In particular this makes the convergence check a modular method which can be easily adapted via subclass.

• New estimate_simulation_RAM() to estimate the RAM requiremensts of a simulation before actually running it (issue #305 and issue #342).

Fixed

• Potentially serious bug issue #260 that the sorted flag of LegCharge was not set correctly in qr().

• from_infiniteT_canonical() should now work with arbitrary charges of the original model, and has the option to double the number of charges to separately conserve the charges on each the physical and ancilla legs.

• Fix a wrong total charge in apply_zipup().

• Fix issue #218 that add_multi_coupling_term() with plus_hc=True didn’t correctly add the hermitian conjugate.

• Fix issue #210 that psi_AKLT() had typos and wrong canonical form for finite systems.

• Fix that the MPS apply_local_op() with local multi-site operators didn’t correctly track the norm with renormalize=False.

• We now update the norm of the MPS in update_bond(). If the parameter preserve_norm is True (which is the default for real time evolution) this has no effect when using run() or similar, since the MPS norm is reset after the timestep anyway. It does, however, change the behavior if preserve_norm is False.

• issue #265 that MPO methods make_U_I(), make_U_II, apply_naively and apply_zipup just ignored the explicit_plus_hc flag of the MPO, possibly giving completely wrong results without raising errors.

• Make sure that eigh() doesn’t have a LegPipe on the second (=new) leg.

• issue #289 that correlation_length() raised errors for charge_sector np ndarrays. Further, allow to pass multiple charge sectors to loop over at once, add argument return_charges. Also, provide a correlation_length_charge_sectors() convenience function to return valid charge sectors.

• issue #153 that DMRG energy convergence criterion was verified after an arbitrarily large energy increase.