Technical reports and preprints from the Quantuva group. Every result is described well enough to reproduce; where a claim is still under review, we say so.
We present a qubit-mapping pass that folds SWAP insertion into the search itself rather than patching it on after placement. Routing is scored against a backend's measured gate-error profile, not a uniform cost model, and we show where greedy depth-minimization stops being a reliable proxy for output fidelity.
Trapped-ion devices expose a different native two-qubit gate than superconducting ones. We describe a decomposition that targets the ion-trap entangling interaction directly, using fewer entangling gates than fixed template-based approaches, and report the conditions under which the advantage appears.
Standard ZNE fits an extrapolation curve through noise-scaled estimates, which bakes in an assumption about how the noise grows. We describe a fit-free estimator and give the precise conditions under which it degrades gracefully — and where it does not.
Pauli twirling converts coherent errors into stochastic ones, making them tractable for modelling. We catalogue the situations where the twirled model diverges from the noise a running application actually sees, and argue for reporting both twirled and untwirled estimates.
Readout calibration matrices drift between measurement rounds, which quietly degrades correction over the course of a run. We study a partial-calibration scheme that keeps the correction effective on a budget, and characterize when a fully re-measured matrix is worth its cost.
A statevector simulator stores 2ⁿ amplitudes and hits a hard memory wall; matrix-product-state methods trade exactness for width. We compare the two across the shallow, high-qubit regime and give a practical crossover heuristic for choosing between them.
A simulator that models decoherence must choose which physical processes to include. We work through amplitude damping, dephasing, and leakage, and identify which must be present to reproduce an observed decay curve and which can be safely omitted without distorting mitigation results.
The surface code is usually introduced through its stabilizers and syndromes. We start from the geometry instead — a grid of data and ancilla qubits — and build up the X and Z checks, the syndrome, and the decoder for readers who think in software rather than physics.