arXiv:2609.20739
Efficient Non-Uniform Quantum Hermite Transform through Adaptive Sampling
Nitay Mayo, Aryeh Lev Zabokritskiy
Abstract
On the span of the first $N$ oscillator modes, Gauss--Hermite quadrature gives an exact change of basis between mode coefficients and $N$ weighted position space samples. We implement this transform with $O(N\operatorname{polylog}(N,1/\varepsilon))$ logical gates and polylogarithmic quantum width. The operator-error bound $\varepsilon$ holds on arbitrary superpositions and includes all auxiliary registers. The construction uses signed averages on adaptive windows to convert uniform-grid samples into weighted Hermite-root samples. Their varying widths control the amplification cost, giving the near-linear bound.
AI Summary
The paper asks how efficiently a quantum circuit can implement the finite Hermite transform: the exact change of basis between the first $N$ harmonic-oscillator modes and $N$ weighted samples at Gauss–Hermite roots. Quadrature makes this sampling map unitary on the retained mode space; the encoded amplitudes are weighted samples, not raw position values. For $N=2^n$ and $0<\varepsilon\leq1/2$, it constructs a Clifford+$T$ circuit using $O(N\operatorname{polylog}(N,1/\varepsilon))$ logical gates and $\operatorname{polylog}(N,1/\varepsilon)$ qubits. The operator-norm error bound includes all auxiliary registers, fixes phases, and holds for arbitrary superpositions and external reference systems. Reversing the circuit implements the inverse transform with the same resources. The construction converts an existing coherent uniform-grid Hermite sampler into weighted root samples using signed Legendre-kernel averages over disjoint windows. Window radii follow local root spacing, including near the spectral edges. Christoffel-weight estimates and complex Mehler bounds control normalization and approximation error. Kernel degree $d=O(\log(N/\varepsilon))$ and polynomial grid size suffice; block encoding and coherent amplification produce the clean transform. Harmonic-oscillator evolution and fractional Fourier rotation follow by conjugating diagonal mode phases. The resource bounds exclude input loading, full classical readout, and polynomial-time classical certification and compilation. Numerical constants, optimized gate counts, and hardware performance are not supplied. Inputs outside the retained space incur separate truncation and sampled-tail aliasing errors; small $L^2$ truncation error alone does not control aliasing. The oscillator examples and proposed noise diagnostic use ideal dense transforms, not compiled circuits or hardware experiments.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 30 pages, including supplementary material. Minor corrections to notation and exposition; improved figure legibility and ancillary documentation. Results unchanged