qoro-maestro-pyscf

PySCF integration plugin for the Maestro quantum simulator by Qoro Quantum.

Run quantum chemistry VQE calculations — works on CPU out of the box, upgrade to GPU for speed.

Installation

pip install qoro-maestro-pyscf

Dependencies: PySCF, OpenFermion, SciPy, NumPy, and the Maestro runtime. No compiler needed — pre-built wheels for Linux and macOS (x86_64, arm64). Fully functional on CPU-only machines; no GPU required to install or run.

Quick Start — CASCI with VQE

No GPU needed. Install and run:

from pyscf import gto, scf, mcscf
from qoro_maestro_pyscf import MaestroSolver

mol = gto.M(atom="Li 0 0 0; H 0 0 1.6", basis="sto-3g", verbose=0)
hf = scf.RHF(mol).run()

# Exact reference (PySCF's built-in FCI)
cas_fci = mcscf.CASCI(hf, 2, 2)
cas_fci.verbose = 0
fci_e = cas_fci.kernel()[0]

# VQE with Maestro
cas = mcscf.CASCI(hf, 2, 2)
cas.fcisolver = MaestroSolver(ansatz="uccsd", maxiter=500)
vqe_e = cas.kernel()[0]

print(f"FCI energy:  {fci_e:.8f} Ha")
print(f"VQE energy:  {vqe_e:.8f} Ha")
print(f"Error:       {abs(vqe_e - fci_e) * 1000:.4f} mHa")

Runs on CPU by default — no license key required. You should see ~0.01 mHa error, well within chemical accuracy (1.6 mHa).

Running Your Own Molecule

Swap in your molecule, basis set, and active space:

Choosing Your Setup

Automatic Active Space Selection

Not sure which orbitals to include? Let PySCF + MP2 natural orbitals decide:

from pyscf import gto, scf, mcscf
from qoro_maestro_pyscf import MaestroSolver, suggest_active_space_from_mp2

mol = gto.M(atom="Li 0 0 0; H 0 0 1.6", basis="sto-3g", verbose=0)
hf = scf.RHF(mol).run()

norb, nelec, mo_coeff = suggest_active_space_from_mp2(hf)
cas = mcscf.CASCI(hf, norb, nelec)
cas.mo_coeff = mo_coeff
cas.fcisolver = MaestroSolver(ansatz="uccsd", maxiter=500)
cas.run()

Scaling Up

Hitting limits on larger active spaces? Two options:

MPS Mode (still CPU, no license needed)

Switch to Matrix Product State simulation for larger systems without needing a GPU:

cas.fcisolver = MaestroSolver(
    ansatz="hardware_efficient",
    ansatz_layers=3,
    simulation="mps",
    mps_bond_dim=128,
)

GPU Acceleration (fastest)

For maximum performance, add GPU support. Get your key instantly at maestro.qoroquantum.net, then:

cas.fcisolver = MaestroSolver(
    ansatz="uccsd",
    backend="gpu",
    license_key="XXXX-XXXX-XXXX-XXXX",
)

GPU + MPS for the largest active spaces:

cas.fcisolver = MaestroSolver(
    ansatz="hardware_efficient",
    ansatz_layers=3,
    backend="gpu",
    simulation="mps",
    mps_bond_dim=128,
)

GPU Setup & Licensing

GPU simulation requires an NVIDIA GPU and a license key. Get your key instantly at maestro.qoroquantum.net.

Three ways to provide your key:

Option 1 — Pass directly to the solver:

cas.fcisolver = MaestroSolver(
    ansatz="uccsd",
    backend="gpu",
    license_key="XXXX-XXXX-XXXX-XXXX",
)

Option 2 — Set it once in your script:

from qoro_maestro_pyscf import set_license_key
set_license_key("XXXX-XXXX-XXXX-XXXX")

Option 3 — Environment variable (recommended for production):

export MAESTRO_LICENSE_KEY="XXXX-XXXX-XXXX-XXXX"

Note: First activation requires an internet connection (one-time). After that, the license is cached locally for offline use.

Migrating from Qiskit

Features

• Works on CPU out of the box — no GPU or license needed to get started
• GPU-accelerated statevector & MPS simulation via Maestro's CUDA backend (with license)
• Drop-in PySCF solver — implements the full fcisolver protocol (kernel, make_rdm1, make_rdm1s, make_rdm12, make_rdm12s)
• CASCI and CASSCF support (CASCI recommended; CASSCF works but VQE convergence can be tricky in the macro-iteration loop)
• Multiple ansatze — hardware-efficient, UCCSD, UpCCD, ADAPT-VQE, and custom (inject any QuantumCircuit callable, e.g. QCC)
• Custom Pauli evaluation — evaluate_custom_paulis() bypasses Hamiltonian generation for measurement-reduction research
• Raw state extraction — get_final_statevector() for exact fidelity benchmarking against exact diagonalisation
• UHF support — handles spin-unrestricted integrals
• Pre-computed amplitudes — skip VQE with maxiter=0 + initial_point
• State fidelity — compare circuit states via compute_state_fidelity()

Architecture

qoro_maestro_pyscf/
├── maestro_solver.py   # MaestroSolver — PySCF fcisolver drop-in
├── hamiltonian.py      # PySCF integrals → QubitOperator (Jordan-Wigner)
├── ansatze.py          # HF initial state, hardware-efficient, UCCSD, UpCCD
├── adapt.py            # ADAPT-VQE adaptive circuit growing
├── expectation.py      # Maestro circuit evaluation wrapper
├── rdm.py              # RDM reconstruction from VQE circuit
├── properties.py       # Dipole moments, natural orbitals
└── backends.py         # GPU/CPU/MPS backend configuration

Dependencies

Examples

See the examples/ directory for 13 worked examples and a full workflow notebook covering H₂ dissociation, LiH UCCSD, GPU benchmarking, MPS bond dimensions, CASSCF, NEVPT2, dipole moments, geometry optimisation, UpCCD paired doubles, GPU benchmarks, ADAPT-VQE, custom QCC ansatze, and iterative QCC with fidelity benchmarking.

License

Apache 2.0 — see LICENSE for details.