arXiv:2610.03823
Predicting Noise in a Trapped-ion Quantum Simulation of a Lattice Gauge Theory
Hassan Ugail
Abstract
A recent trapped-ion experiment realised a lattice gauge theory in which the internal states of the ions carry the gauge field and their vibrations carry bosonic matter. In that experiment, a charge moving around a loop was frozen in place by Aharonov--Bohm interference. The noise of such a device is usually simulated after the measurements have been made. Here we show that it can be predicted before the machine is run. For each quantity that the experiment measures, we derive, from the reported noise rates alone, a formula for how it decays together with a bound it can never exceed. The results reveal a simple pattern. The local conservation law of the theory is a parity of the vibrational mode, so heating violates it at a rate that grows as the mode warms, and the violation accelerates rather than accumulating steadily. At zero electric field, each of the three noise processes of the device leaves its own fingerprint. Heating affects only the conservation law, motional dephasing only the interference, and qubit errors only the stored flux and the frozen state. Once the electric field is switched on, the frozen state begins to leak, and we write an exact formula and a guaranteed bound for that leak.
AI Summary
The paper asks whether observable-specific noise effects in a trapped-ion $\mathbb Z_2$ lattice-gauge simulator can be predicted from a Lindblad model without full noisy evolution. Bosonic matter occupies vibrational modes and gauge fields occupy link qubits. It combines adjoint-Lindbladian calculations, approximate decay laws, trace-norm bounds and small-system numerical tests. For a Gauss generator, it derives the exact heating equation $\partial_t\langle G_m\rangle=-2\dot n_m\langle G_m(2n_m+1)\rangle$, plus a qubit-dephasing contribution. The occupation-dependent rate motivates a factorized decay law and a bound using total phonon-number growth. At zero electric field, the loop flux obeys $\langle W_C(t)\rangle=\langle W_C(0)\rangle e^{-2\gamma_x|C|t}$ under independent bit flips, while motional noise and qubit dephasing preserve it. On a two-link loop initially at $\pi$ flux, the zero-flux population is exactly $(1-e^{-4\gamma_xt})/2$. Heating adds a state-independent population contribution at zero field, whereas motional dephasing damps interference contrast; the reported contrast rate is $369\,\mathrm{s}^{-1}$. At finite field, single-phonon coherent leakage reduces to a three-level system, permitting complete transfer out of the frozen sector at $h=J$. Small fixed-number-sector reference evolutions and Duhamel bounds provide leakage certificates. Heating-induced population shifts become non-additive numerically. Stated limitations include assumed Markovian dephasing, unreported experimental qubit-error rates, modest Fock truncations, loose leakage bounds, and no analytic law for finite-field heating non-additivity. Larger geometries and direct comparison with experimental curves remain unaddressed; a calibration test accepts only 35 held-out instances, giving a coverage interval of $[0.90,1.00]$.
Central claim may not hold: The abstract overstates coverage because the finite-field heating-induced population law remains open and predictions involving qubit errors require rates that the experiment did not report. The claimed accelerating Gauss-law violation is false in general: an initially vacuum, uncoupled mode under the stated heating channel has violation 2rt/(1+2rt), which is strictly concave. The stated derivation of the Gauss certificate yields twice its printed quadratic contribution, leaving the advertised tighter bound unestablished by that argument. The claimed disjoint zero-field noise fingerprints fail for the full noise model because qubit dephasing also violates Gauss's law, as the paper's exact equation explicitly shows. The Bell-state mixture uses the reported fidelity as a depolarizing mixture weight, thereby modeling fidelity 0.95725 instead of 0.943.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.