gydZddlmZddlZddlmZddlmZddlmZddl Z ddl m Z dd l mZmZejrddlZGd d eZGd d ZdS)z9A generic implementation of Approximate Quantum Compiler.) annotationsN)partial)Callable)Protocol)Operator)ApproximateCircuitApproximatingObjectivec eZdZdZ ddd ZdS) MinimizeraCallable Protocol for minimizer. This interface is based on `SciPy's optimize module `__. This protocol defines a callable taking the following parameters: fun The objective function to minimize. x0 The initial point for the optimization. jac The gradient of the objective function. bounds Parameters bounds for the optimization. Note that these might not be supported by all optimizers. and which returns a SciPy minimization result object. NfunCallable[[np.ndarray], float]x0 np.ndarrayjac)Callable[[np.ndarray], np.ndarray] | Nonebounds list[tuple[float, float]] | Nonereturnscipy.optimize.OptimizeResultcdS)aGMinimize the objective function. This interface is based on `SciPy's optimize module `__. Args: fun: The objective function to minimize. x0: The initial point for the optimization. jac: The gradient of the objective function. bounds: Parameters bounds for the optimization. Note that these might not be supported by all optimizers. Returns: The SciPy minimization result object. N)selfr rrrs V/granary/edev/quantum/lib/python3.11/site-packages/qiskit/synthesis/unitary/aqc/aqc.py__call__zMinimizer.__call__4s , NN) r rrrrrrrrr)__name__ __module__ __qualname____doc__rrrrr r s?0:>37        rr c6eZdZdZ ddfd Z dddZxZS)AQCa A generic implementation of the Approximate Quantum Compiler. This implementation is agnostic of the underlying implementation of the approximate circuit, objective, and optimizer. Users may pass corresponding implementations of the abstract classes: * The *optimizer* is an implementation of the :class:`~.Minimizer` protocol, a callable used to run the optimization process. The choice of optimizer may affect overall convergence, required time for the optimization process and achieved objective value. * The *approximate circuit* represents a template which parameters we want to optimize. Currently, there's only one implementation based on 4-rotations CNOT unit blocks: :class:`.CNOTUnitCircuit`. See the paper for more details. * The *approximate objective* is tightly coupled with the approximate circuit implementation and provides two methods for computing objective function and gradient with respect to approximate circuit parameters. This objective is passed to the optimizer. Currently, there are two implementations based on 4-rotations CNOT unit blocks: :class:`.DefaultCNOTUnitObjective` and its accelerated version :class:`.FastCNOTUnitObjective`. Both implementations share the same idea of maximization the Hilbert-Schmidt product between the target matrix and its approximation. The former implementation approach should be considered as a baseline one. It may suffer from performance issues, and is mostly suitable for a small number of qubits (up to 5 or 6), whereas the latter, accelerated one, can be applied to larger problems. * One should take into consideration the exponential growth of matrix size with the number of qubits because the implementation not only creates a potentially large target matrix, but also allocates a number of temporary memory buffers comparable in size to the target matrix. N optimizerMinimizer | Noneseed int | Nonecddl}t|pt|jjddddi|_||_dS)ab Args: optimizer: an optimizer to be used in the optimization procedure of the search for the best approximate circuit. By default, the scipy minimizer with the ``L-BFGS-B`` method is used with max iterations set to 1000. seed: a seed value to be used by a random number generator. rNrzL-BFGS-Bmaxiteri)argsmethodoptions)scipy.optimizesuper__init__roptimizeminimize _optimizer_seed)rr$r&scipy __class__s rr/z AQC.__init__jse  # w N #"Z)UYIZ( ( (  r target_matrixrapproximate_circuitr approximating_objectiver initial_pointnp.ndarray | NonerNonec|jd}tj|}tj|ds%|tj|d|z t z }d}n|}d}||_|Wtj |j tj ddtj z|j }||j||j} || jt'|j} |rYtjtjtj| j|} | |_dSdS) a Approximately compiles a circuit represented as a unitary matrix by solving an optimization problem defined by ``approximating_objective`` and using ``approximate_circuit`` as a template for the approximate circuit. Args: target_matrix: a unitary matrix to approximate. approximate_circuit: a template circuit that will be filled with the parameter values obtained in the optimization procedure. approximating_objective: a definition of the optimization problem. initial_point: initial values of angles/parameters to start optimization from. rr)dtypeTFN)r rr)shapenplinalgdetisclosepowercomplexr6randomr&r3uniformpi num_thetasr2 objectivegradientbuildxrdataangletracedotconjT global_phase) rr6r7r8r9 matrix_dim target_det su_matrixglobal_phase_required opt_result approx_matrixalphas rcompile_unitaryzAQC.compile_unitarys[&#(+ Y]]=11 z*a(( *%a*nU\(](](]]I$( ! !%I$) !1:-  INN4: & & &I--aRUrfs@?"""""" $$$$$$((((((CCCCCCCC + + + + + + + + \b5b5b5b5b5b5b5b5b5b5r