Adaptive approximate quantum compiling

An open-source implementation of ADAPT-AQC [1], an approximate quantum compiling (AQC) and matrix product state (MPS) preparation algorithm. As opposed to assuming any particular ansatz structure, ADAPT-AQC adaptively builds an ansatz, adding a new two-qubit unitary every iteration.

ADAPT-AQC is the successor to ISL [2], using much of the same core code, routine and optimiser. The most significant difference however is its use of MPS simulators. This allows it to compile circuits at 50+ qubits, as well as directly prepare MPSs.

[1] https://arxiv.org/abs/2503.09683
[2] https://github.com/abhishekagarwal2301/isl

Installation

This repository can be easily installed using pip. You have two options:

Use a stable version based on the last commit to master

pip install git+ssh://[email protected]/qiskit-community/adapt-aqc.git

Use an editable local version (after already cloning this repository)

pip install -e PATH_TO_LOCAL_CLONE

Contributing

To make changes to ADAPT-AQC, first clone the repository. Then navigate to your local copy, create a Python environment and install the required dependencies

pip install .

You can check your development environment is ready by successfully running the scripts in /examples/.

Minimal examples

Compiling with statevector simulator

A circuit can be compiled and the result accessed with only 3 lines if using the default settings.

from adaptaqc.compilers import AdaptCompiler
from qiskit import QuantumCircuit

# Setup the circuit
qc = QuantumCircuit(3)
qc.rx(1.23, 0)
qc.cx(0, 1)
qc.ry(2.5, 1)
qc.rx(-1.6, 2)
qc.ccx(2, 1, 0)

# Compile
compiler = AdaptCompiler(qc)
result = compiler.compile()
compiled_circuit = result.circuit

# See the compiled output
print(compiled_circuit)

Compiling matrix product states

Circuits beyond the size accessible to statevector simulators can be compiled via their representation as matrix product states. To give a very simple example where most the qubits are left in the $|0\rangle$ state.

from qiskit import QuantumCircuit

from adaptaqc.backends.aer_mps_backend import AerMPSBackend
from adaptaqc.compilers import AdaptCompiler

n = 50
qc = QuantumCircuit(n)
qc.h(0)
qc.cx(0, 1)
qc.h(2)
qc.cx(2, 3)
qc.h(range(4, n))

# Create compiler with the default MPS simulator, which has very minimal truncation.
adapt_compiler = AdaptCompiler(qc, backend=AerMPSBackend())

result = adapt_compiler.compile()
print(f"Overlap between circuits is {result.overlap}")

Specifying additional configuration

For more advanced examples, please see examples/advanced_mps_example.py and advanced_sv_example.py. For a full overview of the different configuration options, in addition to the documentation, see docs/running_options_explained.md.

Citing usage

We respectfully ask any publication, project or whitepaper using ADAPT-AQC to cite the following work:

Jaderberg, B., Pennington, G., Marshall, K.V., Anderson, L.W., Agarwal, A., Lindoy, L.P., Rungger, I., Mensa, S. and Crain, J., 2025. Variational preparation of normal matrix product states on quantum computers. arXiv preprint arXiv:2503.09683