Skip to main contentIBM Quantum Documentation
You are viewing the API reference for an old version of Qiskit SDK. Switch to latest version


class AerSimulator(configuration=None, properties=None, provider=None, **backend_options)

GitHub(opens in a new tab)

Bases: qiskit_aer.backends.aerbackend.AerBackend

Noisy quantum circuit simulator backend.

Configurable Options

The AerSimulator supports multiple simulation methods and configurable options for each simulation method. These may be set using the appropriate kwargs during initialization. They can also be set of updated using the set_options() method.

Run-time options may also be specified as kwargs using the run() method. These will not be stored in the backend and will only apply to that execution. They will also override any previously set options.

For example, to configure a density matrix simulator with a custom noise model to use for every execution

noise_model = NoiseModel.from_backend(backend)
backend = AerSimulator(method='density_matrix',

Simulating an IBMQ Backend

The simulator can be automatically configured to mimic an IBMQ backend using the from_backend() method. This will configure the simulator to use the basic device NoiseModel for that backend, and the same basis gates and coupling map.

backend = AerSimulator.from_backend(backend)

Returning the Final State

The final state of the simulator can be saved to the returned Result object by appending the save_state() instruction to a quantum circuit. The format of the final state will depend on the simulation method used. Additional simulation data may also be saved using the other save instructions in qiskit.provider.aer.library.

Simulation Method Option

The simulation method is set using the method kwarg. A list supported simulation methods can be returned using available_methods(), these are

  • "automatic": Default simulation method. Select the simulation method automatically based on the circuit and noise model.
  • "statevector": A dense statevector simulation that can sample measurement outcomes from ideal circuits with all measurements at end of the circuit. For noisy simulations each shot samples a randomly sampled noisy circuit from the noise model.
  • "density_matrix": A dense density matrix simulation that may sample measurement outcomes from noisy circuits with all measurements at end of the circuit.
  • "stabilizer": An efficient Clifford stabilizer state simulator that can simulate noisy Clifford circuits if all errors in the noise model are also Clifford errors.
  • "extended_stabilizer": An approximate simulated for Clifford + T circuits based on a state decomposition into ranked-stabilizer state. The number of terms grows with the number of non-Clifford (T) gates.
  • "matrix_product_state": A tensor-network statevector simulator that uses a Matrix Product State (MPS) representation for the state. This can be done either with or without truncation of the MPS bond dimensions depending on the simulator options. The default behaviour is no truncation.
  • "unitary": A dense unitary matrix simulation of an ideal circuit. This simulates the unitary matrix of the circuit itself rather than the evolution of an initial quantum state. This method can only simulate gates, it does not support measurement, reset, or noise.
  • "superop": A dense superoperator matrix simulation of an ideal or noisy circuit. This simulates the superoperator matrix of the circuit itself rather than the evolution of an initial quantum state. This method can simulate ideal and noisy gates, and reset, but does not support measurement.

GPU Simulation

By default all simulation methods run on the CPU, however select methods also support running on a GPU if qiskit-aer was installed with GPU support on a compatible NVidia GPU and CUDA version.

MethodGPU Supported

Running a GPU simulation is done using device="GPU" kwarg during initialization or with set_options(). The list of supported devices for the current system can be returned using available_devices().

For multiple shots simulation, OpenMP threads should be exploited for multi-GPUs. Number of GPUs used for multi-shots is reported in metadata gpu_parallel_shots_ or is batched execution is done reported in metadata batched_shots_optimization_parallel_gpus. For large qubits circuits with multiple GPUs, number of GPUs is reported in metadata chunk_parallel_gpus in cacheblocking.

If AerSimulator is built with cuStateVec support, cuStateVec APIs are enabled by setting cuStateVec_enable=True.

Additional Backend Options

The following simulator specific backend options are supported

  • method (str): Set the simulation method (Default: "automatic"). Use available_methods() to return a list of all availabe methods.

  • device (str): Set the simulation device (Default: "CPU"). Use available_devices() to return a list of devices supported on the current system.

  • precision (str): Set the floating point precision for certain simulation methods to either "single" or "double" precision (default: "double").

  • executor (futures.Executor or None): Set a custom executor for asynchronous running of simulation jobs (Default: None).

  • max_job_size (int or None): If the number of run circuits exceeds this value simulation will be run as a set of of sub-jobs on the executor. If None simulation of all circuits are submitted to the executor as a single job (Default: None).

  • max_shot_size (int or None): If the number of shots of a noisy circuit exceeds this value simulation will be split into multi circuits for execution and the results accumulated. If None circuits will not be split based on shots. When splitting circuits use the max_job_size option to control how these split circuits should be submitted to the executor (Default: None).

    a noise model exceeds this value simulation will be splitted into sub-circuits. If None simulator does noting (Default: None).

  • enable_truncation (bool): If set to True this removes unnecessary qubits which do not affect the simulation outcome from the simulated circuits (Default: True).

  • zero_threshold (double): Sets the threshold for truncating small values to zero in the result data (Default: 1e-10).

  • validation_threshold (double): Sets the threshold for checking if initial states are valid (Default: 1e-8).

  • max_parallel_threads (int): Sets the maximum number of CPU cores used by OpenMP for parallelization. If set to 0 the maximum will be set to the number of CPU cores (Default: 0).

  • max_parallel_experiments (int): Sets the maximum number of qobj experiments that may be executed in parallel up to the max_parallel_threads value. If set to 1 parallel circuit execution will be disabled. If set to 0 the maximum will be automatically set to max_parallel_threads (Default: 1).

  • max_parallel_shots (int): Sets the maximum number of shots that may be executed in parallel during each experiment execution, up to the max_parallel_threads value. If set to 1 parallel shot execution will be disabled. If set to 0 the maximum will be automatically set to max_parallel_threads. Note that this cannot be enabled at the same time as parallel experiment execution (Default: 0).

  • max_memory_mb (int): Sets the maximum size of memory to store a state vector. If a state vector needs more, an error is thrown. In general, a state vector of n-qubits uses 2^n complex values (16 Bytes). If set to 0, the maximum will be automatically set to the system memory size (Default: 0).

  • cuStateVec_enable (bool): This option enables accelerating by cuStateVec library of cuQuantum from NVIDIA, that has highly optimized kernels for GPUs (Default: False). This option will be ignored if AerSimulator is not built with cuStateVec support.

  • blocking_enable (bool): This option enables parallelization with multiple GPUs or multiple processes with MPI (CPU/GPU). This option is only available for "statevector", "density_matrix" and "unitary" (Default: False).

  • blocking_qubits (int): Sets the number of qubits of chunk size used for parallelizing with multiple GPUs or multiple processes with MPI (CPU/GPU). 16*2^blocking_qubits should be less than 1/4 of the GPU memory in double precision. This option is only available for "statevector", "density_matrix" and "unitary". This option should be set when using option blocking_enable=True (Default: 0). If multiple GPUs are used for parallelization number of GPUs is reported to chunk_parallel_gpus in cacheblocking metadata.

  • chunk_swap_buffer_qubits (int): Sets the number of qubits of maximum buffer size (=2^chunk_swap_buffer_qubits) used for multiple chunk-swaps over MPI processes. This parameter should be smaller than blocking_qubits otherwise multiple chunk-swaps is disabled. blocking_qubits - chunk_swap_buffer_qubits swaps are applied at single all-to-all communication. (Default: 15).

  • batched_shots_gpu (bool): This option enables batched execution of multiple shot simulations on GPU devices for GPU enabled simulation methods. This optimization is intended for statevector simulations with noise models, or statevecor and density matrix simulations with intermediate measurements and can greatly accelerate simulation time on GPUs. If there are multiple GPUs on the system, shots are distributed automatically across available GPUs. Also this option distributes multiple shots to parallel processes of MPI (Default: True). If multiple GPUs are used for batched exectuion number of GPUs is reported to batched_shots_optimization_parallel_gpus metadata.

  • batched_shots_gpu_max_qubits (int): This option sets the maximum number of qubits for enabling the batched_shots_gpu option. If the number of active circuit qubits is greater than this value batching of simulation shots will not be used. (Default: 16).

  • num_threads_per_device (int): This option sets the number of threads per device. For GPU simulation, this value sets number of threads per GPU. This parameter is used to optimize Pauli noise simulation with multiple-GPUs (Default: 1).

These backend options only apply when using the "statevector" simulation method:

  • statevector_parallel_threshold (int): Sets the threshold that the number of qubits must be greater than to enable OpenMP parallelization for matrix multiplication during execution of an experiment. If parallel circuit or shot execution is enabled this will only use unallocated CPU cores up to max_parallel_threads. Note that setting this too low can reduce performance (Default: 14).
  • statevector_sample_measure_opt (int): Sets the threshold that the number of qubits must be greater than to enable a large qubit optimized implementation of measurement sampling. Note that setting this two low can reduce performance (Default: 10)

These backend options only apply when using the "stabilizer" simulation method:

  • stabilizer_max_snapshot_probabilities (int): set the maximum qubit number for the ~qiskit_aer.extensions.SnapshotProbabilities instruction (Default: 32).

These backend options only apply when using the "extended_stabilizer" simulation method:

  • extended_stabilizer_sampling_method (string): Choose how to simulate measurements on qubits. The performance of the simulator depends significantly on this choice. In the following, let n be the number of qubits in the circuit, m the number of qubits measured, and S be the number of shots (Default: resampled_metropolis).

    • "metropolis": Use a Monte-Carlo method to sample many output strings from the simulator at once. To be accurate, this method requires that all the possible output strings have a non-zero probability. It will give inaccurate results on cases where the circuit has many zero-probability outcomes. This method has an overall runtime that scales as n^{2} + (S-1)n.
    • "resampled_metropolis": A variant of the metropolis method, where the Monte-Carlo method is reinitialised for every shot. This gives better results for circuits where some outcomes have zero probability, but will still fail if the output distribution is sparse. The overall runtime scales as Sn^{2}.
    • "norm_estimation": An alternative sampling method using random state inner products to estimate outcome probabilites. This method requires twice as much memory, and significantly longer runtimes, but gives accurate results on circuits with sparse output distributions. The overall runtime scales as Sn^{3}m^{3}.
  • extended_stabilizer_metropolis_mixing_time (int): Set how long the monte-carlo method runs before performing measurements. If the output distribution is strongly peaked, this can be decreased alongside setting extended_stabilizer_disable_measurement_opt to True (Default: 5000).

  • extended_stabilizer_approximation_error (double): Set the error in the approximation for the extended_stabilizer method. A smaller error needs more memory and computational time (Default: 0.05).

  • extended_stabilizer_norm_estimation_samples (int): The default number of samples for the norm estimation sampler. The method will use the default, or 4m^{2} samples where m is the number of qubits to be measured, whichever is larger (Default: 100).

  • extended_stabilizer_norm_estimation_repetitions (int): The number of times to repeat the norm estimation. The median of these reptitions is used to estimate and sample output strings (Default: 3).

  • extended_stabilizer_parallel_threshold (int): Set the minimum size of the extended stabilizer decomposition before we enable OpenMP parallelization. If parallel circuit or shot execution is enabled this will only use unallocated CPU cores up to max_parallel_threads (Default: 100).

  • extended_stabilizer_probabilities_snapshot_samples (int): If using the metropolis or resampled_metropolis sampling method, set the number of samples used to estimate probabilities in a probabilities snapshot (Default: 3000).

These backend options only apply when using the matrix_product_state simulation method:

  • matrix_product_state_max_bond_dimension (int): Sets a limit on the number of Schmidt coefficients retained at the end of the svd algorithm. Coefficients beyond this limit will be discarded. (Default: None, i.e., no limit on the bond dimension).

  • matrix_product_state_truncation_threshold (double): Discard the smallest coefficients for which the sum of their squares is smaller than this threshold. (Default: 1e-16).

  • mps_sample_measure_algorithm (str): Choose which algorithm to use for "sample_measure" (Default: “mps_apply_measure”).

    • mps_probabilities: This method first constructs the probability vector and then generates a sample per shot. It is more efficient for a large number of shots and a small number of qubits, with complexity O(2^n * n * D^2) to create the vector and O(1) per shot, where n is the number of qubits and D is the bond dimension.
    • mps_apply_measure: This method creates a copy of the mps structure and measures directly on it. It is more efficient for a small number of shots, and a large number of qubits, with complexity around O(n * D^2) per shot.
  • mps_log_data (str): if True, output logging data of the MPS structure: bond dimensions and values discarded during approximation. (Default: False)

  • mps_swap_direction (str): Determine the direction of swapping the qubits when internal swaps are inserted for a 2-qubit gate. Possible values are “mps_swap_right” and “mps_swap_left”. (Default: “mps_swap_left”)

These backend options apply in circuit optimization passes:

  • fusion_enable (bool): Enable fusion optimization in circuit optimization passes [Default: True]
  • fusion_verbose (bool): Output gates generated in fusion optimization into metadata [Default: False]
  • fusion_max_qubit (int): Maximum number of qubits for a operation generated in a fusion optimization. A default value (None) automatically sets a value depending on the simulation method: [Default: None]
  • fusion_threshold (int): Threshold that number of qubits must be greater than or equal to enable fusion optimization. A default value automatically sets a value depending on the simulation method [Default: None]

fusion_enable and fusion_threshold are set as follows if their default values (None) are configured:

other methods514

Aer class for backends.

This method should initialize the module and its configuration, and raise an exception if a component of the module is not available.


  • configuration (BackendConfiguration) – backend configuration.
  • properties (BackendProperties or None) – Optional, backend properties.
  • defaults (PulseDefaults or None) – Optional, backend pulse defaults.
  • provider (Provider) – Optional, provider responsible for this backend.
  • backend_options (dict or None) – Optional set custom backend options.


AerError – if there is no name in the configuration




Return the available simulation methods.



Return the available simulation methods.



Reset the simulator options to default values.



Return the simulator backend configuration.


the configuration for the backend.

Return type




Return the simulator backend pulse defaults.


The backend pulse defaults or None if the

backend does not support pulse.

Return type



classmethod AerSimulator.from_backend(backend, **options)

Initialize simulator from backend.


Format backend name string for simulator


Return the simulator backend properties if set.


The backend properties or None if the

backend does not have properties set.

Return type




Return the backend Provider.


the Provider responsible for the backend.

Return type


run, validate=False, parameter_binds=None, **run_options)

Run a qobj on the backend.


  • circuits (QuantumCircuit or list) – The QuantumCircuit (or list of QuantumCircuit objects) to run
  • validate (bool) – validate the Qobj before running (default: False).
  • parameter_binds (list) – A list of parameter binding dictionaries. See additional information (default: None).
  • run_options (kwargs) – additional run time backend options.


The simulation job.

Return type



AerError – If parameter_binds is specified with a qobj input or has a length mismatch with the number of circuits.

Additional Information:

  • Each parameter binding dictionary is of the form:

        param_a: [val_1, val_2],
        param_b: [val_3, val_1],

    for all parameters in that circuit. The length of the value list must be the same for all parameters, and the number of parameter dictionaries in the list must match the length of circuits (if circuits is a single QuantumCircuit object it should a list of length 1).

  • kwarg options specified in run_options will temporarily override any set options of the same name for the current run.


ValueError – if run is not implemented


AerSimulator.set_option(key, value)

Special handling for setting backend options.

This method should be extended by sub classes to update special option values.


  • key (str) – key to update
  • value (any) – value to update.


AerError – if key is ‘method’ and val isn’t in available methods.



Set the simulator options



Return backend status.


the status of the backend.

Return type




Return the options for the backend

The options of a backend are the dynamic parameters defining how the backend is used. These are used to control the run() method.


Default value: 1

Was this page helpful?
Report a bug or request content on GitHub.