arXiv:2611.03851
High-fidelity quantum gate learning via physics-informed neural networks
Luis Fabra, Cristian E. Susa, Alcides Montoya, Ariel Norambuena
Abstract
We propose a physics-informed neural network (PINN) approach for high-fidelity quantum gate learning in both one- and two-qubit systems. We train a PINN using the time-dependent Schrödinger equation, where smooth control fields with fixed boundary conditions are obtained by minimizing a loss function that considers model dynamics, gate error, physical constraints, regularization terms, and multiple initial conditions. Training considers pairs of mutually unbiased bases, while independent validation combines Haar-random input states with the average gate fidelity. Unitary simulations yield an average gate infidelity of $8.49\times10^{-8}$ for Hadamard and values between $3.39\times10^{-7}$ and $4.96\times10^{-6}$ for controlled-NOT under $XX$, $YY$, and $ZZ$ interactions. We examine how the learned control fields depend on the coupling, training configuration, and available local fields, including five exchange-type dipolar scenarios. We test the learned control fields under the Lindblad master equation with dephasing and amplitude damping, theoretically reproducing the expected first-order fidelity loss in the weak-dissipation regime. Finally, boundary-adapted Chebyshev expansions represent 139 learned controls with a minimum common count of $K_\ast=24$ coefficients per control for all eleven family-averaged mean squared errors to fall below $10^{-6}$. Separately, square-pulse and Strang propagation at $K=30$ preserve all 31 continuous polynomial references within infidelity $10^{-5}$ at $N=128$ and all larger tested resolutions, under both sampling rules. These results connect multi-state gate synthesis, control-resource characterization, and compact square-pulse playback.
AI Summary
The paper studies physics-informed neural networks (PINNs) for synthesizing smooth Hadamard and CNOT control pulses in one- and two-qubit Hamiltonians. A network jointly represents density-matrix trajectories for multiple input states and shared control fields. Training uses pairs of mutually unbiased bases, dynamical residuals, target-state errors, trace constraints, and weight regularization. An analytic envelope enforces zero pulse amplitude and first derivative at both endpoints. Independent propagation and average gate fidelity, cross-checked against Haar-random inputs, give unitary infidelities of $8.49\times10^{-8}$ for Hadamard and $3.39\times10^{-7}$, $4.96\times10^{-6}$, and $1.17\times10^{-6}$ for CNOT with $XX$, $YY$, and $ZZ$ interactions. Five exchange-type $XY$ configurations explore different local controls and detunings. A symmetry argument establishes $F_{\mathrm{avg}}\leq2/5$ for specified restricted two-control configurations, independently of pulse shape or duration. Lindblad simulations reuse unitary-trained pulses and reproduce established first-order dissipative fidelity losses; they do not demonstrate decoherence suppression. Boundary-adapted Chebyshev fits compress 139 scalar controls across 31 protocols: 24 coefficients per control suffice for all eleven family-averaged waveform MSEs to fall below $10^{-6}$. Using 30 coefficients, square-pulse and Strang playback preserve every continuous fitted evolution within infidelity $10^{-5}$ at 128 intervals and all larger tested resolutions. The formulation imposes no hard amplitude or bandwidth bounds, and observed control patterns do not establish minimum resources. Waveform MSE does not guarantee gate accuracy; one protocol retains baseline infidelity $1.56\times10^{-3}$. Physical delivery requires device-specific calibration, timing, bandwidth validation, and, for split playback, generator isolation or refocusing.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 22 pages, 18 figures