Geometry-Aware Decoherence Maps for Tunable Quantum-Dot Qubits: A Physics-Informed Surrogate Modeling Study

Abstract

Quantum-dot qubits offer strong tunability, but their sensitivity to geometry, external fields, and model parameters makes decoherence prediction computationally demanding. Here, we present a physics-informed surrogate modeling framework to construct geometry-aware decoherence maps for tunable quantum-dot qubits. Using an effective two-level model with parabolic and triangular confinement and a physically motivated phenomenological Lindblad-rate model, we compute the transition frequency, an entropy-based information metric, and a coherence-time proxy over a controlled multidimensional parameter sweep.

The resulting 250-sample dataset is used to train two compact, toolbox-free surrogate models: a quadratic polynomial model and a Gaussian radial basis function surrogate. Both reproduce the physics model with $R^2>0.97$ under 5-fold cross-validation and provide $\sim$260–950$\times$ faster predictions than direct diagonalization. A composite operating-point score is evaluated across 66 weighting choices to assess robustness.

Within this simplified model, near-parabolic confinement yields a more stable spectral response and broader robust operating region, whereas triangular distortion enhances electric-field tunability but narrows the coherence-preserving window. The electric field is the dominant control parameter, while the magnetic field has a weaker secondary role. These results illustrate a transferable surrogate-assisted strategy for rapid qubit design-space exploration.