Release Notes#

0.5.0#

Prelude#

Qiskit Dynamics 0.5.0 introduces a major change to the internal array dispatching mechanism. Users can now work directly with their supported array library of choice, without the need to personally manage the dispatching behaviour using the Array class as in previous versions. Internally, Qiskit Dynamics now uses Arraylias for dispatching. See the user guide entry on using different array libraries with Qiskit Dynamics for more details on the updated approach.
To accompany this change, arguments that explicitly control the array library used to store matrices (e.g. evaluation_mode) in models and related classes have also been changed. See the upgrade notes for a list of such changes, as well as the updated user guide on configuring simulations for performance for how to use the new versions of these arguments.
Due to the scale of these changes, this release will generally break any pre-existing Dynamics code utilizing Array or any of the aformentioned model and solver arguments. The Array class and array and dispatch submodules have been deprecated, and will be removed in 0.6.0. The documentation for the previous release 0.4.5 is still available here.

New Features#

  • Adds the DynamicsBackend.solve() method for running simulations of circuits and schedules for arbitrary input types, and returning the ODE simulation results.

Upgrade Notes#

  • Support for Python 3.12 has been added, and Python 3.8 has been dropped.

  • The upper bound on JAX and Diffrax in the last version of Qiskit Dynamics has been removed. Users should try to use the latest version of JAX.

  • The interface for GeneratorModel, HamiltonianModel, and LindbladModel have been modified. The copy method has been removed, and all setter methods other than in_frame_basis and signals have been removed. The evaluation_mode construction argument has been replaced by array_library, which controls which array library is used internally to store and evaluate operations, and the additional vectorized boolean argument has been added to LindbladModel to control whether the equation is evaluated in vectorized mode. Note that, regardless of array library used, dense arrays must be supplied to the constructors of these classes, due to peculiarities of the internal setup for sparse libraries.

  • In conjunction with the change to the evaluation_mode argument in the model classes, the Solver class has been updated to take the array_library constructor argument, as well as the vectorized constructor argument (for use when Lindblad terms are present).

  • Similar to the Solver initialization method, the DynamicsBackend.from_backend() constructor method argument evaluation_mode has been removed and replaced by the arguments array_library and vectorized.

  • The logic in Solver.solve() for automatic jit compiling when using JAX and simulating a list of schedules has been updated to no longer be based on when Array.default_backend() == "jax". The attempted automatic jit now occurs based purely on whether method="jax_odeint", or method is a Diffrax integration method. A warning will be raised if the array_library is not known to be compatible with the compilation routine. (For now, "scipy_sparse" is the only array_library not compatible with this routine, however a warning will still be raised if no explicit array_library is provided, as in this case the JAX-compatibility is unknown.)

Deprecation Notes#

  • The array and dispatch submodules of Qiskit Dynamics have been deprecated as of version 0.5.0, and will be removed in version 0.6.0.