Release Notes¶

0.4.0¶

Prelude¶

Following up on the deprecation of BlueprintsCircuit in the 2.1 version of Qiskit, most tests have been updated to use the replacements functions instead. These circuits are still supported by qiskit-algorithms, though their use is now deprecated, and their support will be removed when the oldest supported Qiskit version is 3.0. The AmplificationProblem and Grover classes now support being passed Gate objects to support the use of PhaseOracleGate in addition to PhaseOracle. The AdaptVQE class now supports a new way to specify its ansatz, following up on the deprecation of EvolvedOperatorAnsatz.

New Features¶

Added Python 3.13 support.

The AmplificationProblem and Grover classes now support being passed Gate objects to support the use of PhaseOracleGate in addition to PhaseOracle. For instance, the following is a valid way to instantiate the AmplificationProblem class:
```
from qiskit.circuit.library.phase_oracle import PhaseOracleGate
from qiskit_algorithms import AmplificationProblem

problem = AmplificationProblem(PhaseOracleGate("(a & b)"))
```

The AdaptVQE class now supports a new way to specify its ansatz, following up on the deprecation of EvolvedOperatorAnsatz. For instance, the following code to create a AdaptVQE instance in the 0.3 version:
```
AdaptVQE(
    VQE(
        StatevectorEstimator(),
        EvolvedOperatorAnsatz(operators, initial_state=initial_state),
        SLSQP(),
    )
)
```
is now deprecated, and should instead be written as:
```
AdaptVQE(
    VQE(
        StatevectorEstimator(),
        QuantumCircuit(1),
        SLSQP(),
    ),
    operators=operators,
    initial_state=initial_state,
)
```
Note that the ansatz passed to the class:.VQE: solver is ignored when using this method. More generally, AdaptVQE now accepts (as keyword arguments only) every possible argument that can be passed to evolved_operator_ansatz(), with the addition of an initial_state to prepend to the ansatz.

Support for SBPLX optimizer from NLopt library has been added. SBPLX is a local gradient-free optimizer based on Nelder-Mead and is expected to show better convergence behavior. Further information about this optimizer and the others can be found in the API ref for the optimizers.

Upgrade Notes¶

Previously, the compute_minimum_eigenvalue method set the operators attribute of its solver’s ansatz, which was assumed to be of type EvolvedOperatorAnsatz. Since this behavior was undocumented, and since the ansatz’s type can’t be assumed to still be EvolvedOperatorAnsatz at the end because of transpilation, this feature has been removed.

Deprecation Notes¶

Previously, it was possible to pass to VQE, SamplingVQE and VQD a BlueprintCircuit as an ansatz without its number of qubits being set, the algorithm taking care of setting it. Since BlueprintCircuits are now deprecated, and those being the only ones that can have their number of qubits set after their initialization, this behavior is now also deprecated, and won’t be supported once the oldest supported Qiskit version is 3.0. As such, users that made use of this feature would now need to ensure that the ansatz they pass to these algorithms have their number of qubits set and matching with that of the operator they wish to run the algorithm on.

Other Notes¶

Aspects of the gradients internal implementation, which manipulate circuits more directly, have been updated now that circuit data is being handled by Rust so it’s compatible with the former Python way as well as the new Qiskit Rust implementation.

0.3.0¶

Prelude¶

This release now supports Qiskit 1.0 while continuing to work with the Qiskit 0.46 release for those still navigating the changes brought about in the Qiskit 1.0 release.

New Features¶

Added input argument insert_barriers to TrotterQRTE to add barriers between Trotter layers.

Added support for using Qiskit Algorithms with Python 3.12.

Bug Fixes¶

Fixes state fidelity ComputeUncompute to correct an issue arising from the threading used for the job result. This issue could be seen sometimes if more than one job was created and their results fetched back-to-back, such that more than one internal thread was active processing these results.

Fixes internal cache used by state fidelities so that circuits are cached using the same generated key method as that used by the reference primitives. This avoids a potential incorrect result that could have occurred with the key as it was before when id() was used.

Fixes GSLS optimizer minimize() so that if the bounds parameter is passed with tuples that have entries of None then the entry is treated as equivalent to infinity.

Fixed the AQGD optimizer grouping objective function calls by default so that a single point is now passed to the objective function. For algorithms that can handle more than one gradient evaluations in their objective function, such as a VQE in the algorithms here, the number of grouped evaluations can be controlled via the max_grouped_evals parameter. Grouped evaluations allows a list of points to be handed over so that they can potentially be assessed more efficiently in a single job.

0.2.0¶

Upgrade Notes¶

The qiskit-algorithms code uses a common random number generator, which can be seeded for reproducibility. The algorithms code here, having originated in Qiskit and having been moved here, used random function from qiskit.utils which was seeded as follows:
```
from qiskit.utils import algorithm_globals
algorithm_globals.random_seed = 101
```
Now this will continue to work in qiskit-algorithms, until such time as the algorithm_globals are removed from Qiskit, however you are highly recommended to already change to import/seed the algorithm_globals that is now supplied by qiskit-algorithms thus:
```
from qiskit_algorithms.utils import algorithm_globals
algorithm_globals.random_seed = 101
```
As can be seen it’s simply a change to the import statement, so as to import the qiskit_algorithms instance rather than the one from qiskit.

This has been done to afford a transition and not break end-users code by supporting seeding the random generator as it was done before. How does it work - well, while the qiskit.utils version exists, the qiskit-algorithms version simply delegates its function to that instance. However the main codebase of qiskit-algorithms has already been altered to use the new instance and the delegation function, that accesses the random generator, will warn with a message to switch to seeding the qiskit-algorithms version if it detects a difference in the seed. A difference could exist if the random_seed was set direct to the qiskit.utils instance.

At such a time when the qiskit.utils algorithm_globals version no longer exists, rather than delegating the functionality to that, it will use identical logic which it already has, so no further user change will be required if you already transitioned to seeding the qiskit_algorithms.utils algorithms_globals.

A couple of algorithms here, PVQD, AdaptVQE and optimizer SNOBFIT, directly raised a QiskitError. These have been changed to raise an AlgorithmError instead. Algorithms have now been moved out of Qiskit and this better distinguishes the exception to the algorithms when raised. Now AlgorithmError was already raised elsewhere by the algorithms here so this makes things more consistent too. Note, that as AlgorithmError internally extends QiskitError, any code that might have caught that specifically will continue to work. However we do recommend you update your code accordingly for AlgorithmError.

The deprecated threshold input argument of AdaptVQE has been removed, and replaced by gradient_threshold. This change was made to avoid confusion with the later introduced eigenvalue_threshold argument. The updated AdaptVQE use would look like this:

from qiskit_algorithms import VQE, AdaptVQE

adapt_vqe = AdaptVQE(
              VQE(Estimator(), ansatz, optimizer),
              gradient_threshold=1e-3,
              eigenvalue_threshold=1e-3
            )

Other Notes¶

Removed the custom __str__ method from SamplingMinimumEigensolverResult so that string conversion is based on the method of its parent AlgorithmResult which prints all the result fields in a dictionary like format. The overridden method had only printed a select couple of fields, unlike when normally printing a result all fields are shown, and the lack of fields expected to be shown caused confusion when printing results derived from that, such as returned by SamplingVQE and QAOA.

0.1.0¶

Prelude¶

Qiskit’s qiskit.algorithms module has been superseded by this new standalone library, qiskit_algorithms.

As of Qiskit’s 0.25 release, active development of new algorithm features has moved to this new package.

If you’re relying on qiskit.algorithms you should update your requirements to also include qiskit-algorithms and update the imports from qiskit.algorithms to qiskit_algorithms.

Note

If you have not yet migrated from QuantumInstance-based to primitives-based algorithms, you should first follow the migration guidelines in https://qisk.it/algo_migration, to complete the migration of your code, as this package does not include any deprecated algorithm function.

The decision to migrate the qiskit.algorithms module to a separate package was made to clarify the purpose of Qiskit and make a distinction between the tools and libraries built on top of it.

New Features¶

The primitive-based algorithms in qiskit_algorithms.eigensolvers and qiskit_algorithms.minimum_eigensolvers are now directly importable from qiskit_algorithms. For example, the primitive-based VQE can now be imported using from qiskit_algorithms import VQE without having to specify from qiskit_algorithms.minimum_eigensolvers import VQE. This short import path used to be reserved for QuantumInstance-based algorithms (now deprecated and removed from the codebase). If you have not yet migrated from QuantumInstance-based to primitives-based algorithms, you should follow the migration guidelines in https://qisk.it/algo_migration.