Data: Operator Learning for efficient Quantum Computation

Data of the operator learning experiments by P. Over, S. Bengoechea, L. Borello Busilacchi, M. Kiffner, T. Rung, and A. A. Michailidis, DOI: 10.25592/uhhfdm.18962.

The repository contains one archive per experiment:

Archive Application
exp1.zip Transverse-field Ising model (quantum simulation)
exp2.zip Pariser–Parr–Pople model of butadiene (quantum chemistry)
exp3.zip Finite-difference second derivative (Laplacian)
exp4.zip Hess–Smith panel method, NACA 0012 airfoil

Folder structure

exp1.zip – TFIM

N<n>/G<k>/dt<δt>/
  • N<n>: number of qubits n (4, 6, 8, ...).
  • G<k>: trainable ansatz G1 (two-qubit gates, one layer), G2 (two-qubit gates, two layers), G3 (three-qubit gates, one layer).
  • dt<δt>: time step, written without decimal point (dt001 = 0.01, dt002 = 0.02, dt005 = 0.05, dt01 = 0.1, dt02 = 0.2, dt05 = 0.5, dt1 = 1.0).

exp2.zip – PPP butadiene

0_3qubit/<δt>/     stage 0: six three-qubit gates
1_2qubit/<δt>/     stage 1: fifteen two-qubit gates
  • <δt>: time step (0.01, 0.02, 0.05, 0.1, 0.2, 0.5, 1.0).
  • The two folders correspond to the two gate groups of the ansatz G = G₂G₁: the three-qubit hopping gates (0_3qubit) and the long-range two-qubit gates (1_2qubit).

exp3.zip – Laplacian

a/n<ns>/LNN/       linear nearest-neighbor ansatz
a/n<ns>/2D/        two-dimensional lattice ansatz
b/A/               no hierarchical optimization
b/B_0/             hierarchical, without stabilizing term (µ = 0)
b/B/               hierarchical, µ = 10⁻⁶
b/C/               hierarchical + regularization
  • n<ns>: number of system qubits n_s (4–9); one ancilla qubit is added.
  • The runs in b/ use n_s = 6 and the LNN ansatz.

exp4.zip – Panel method

noreg/             without regularization (ρ = 0)
reg/               with regularization (ρ = 10⁻⁶)
  • P = 31 panels, n_s = 5 system qubits plus one ancilla qubit.