Source-linked AI summary
Adaptive Pauli Shadows for Energy Estimation
Charles Hadfield
TL;DR
Energy estimation with classical shadows concerns estimating a Hamiltonian's energy from a single linear combination of Pauli observables, while derandomized shadows add classical computation. APS adapts LBCS with additional basis-selection computation, and the paper reports similar performance to derandomized shadows with apparent accuracy gains in many cases, especially on larger instances.
Problem
Locally-biased classical shadows estimate Hamiltonian energies, while derandomized shadows claim greater accuracy at the cost of additional classical computation.
Source: §1 · p. 1Method
APS adapts LBCS by choosing Pauli measurement bases through an optimization that uses bases assigned to preceding qubits.
Source: §2 · pp. 3–4Results
APS performs similarly to derandomized shadows and appears more accurate in many cases, particularly on larger instances.
Source: §1 · p. 1Takeaways & Limitations
APS's basis-selection runtime is O(n_H · n), better than derandomized shadows' preparation-count scaling, but higher than LBCS's O(1).
Source: §2 · p. 4
Abstract
from arXiv · showhide
Locally-biased classical shadows allow rapid estimation of energies of quantum Hamiltonians. Recently, derandomised classical shadows have emerged claiming to be even more accurate. This accuracy comes at a cost of introducing classical computing resources into the energy estimation procedure. This present note shows, by adding a fraction of this classical computing resource to the locally-biased classical shadows setting, that the modified algorithm, termed Adaptive Pauli Shadows is state-of-the-art for energy estimation.
1. Introduction
Adaptive Pauli Shadows (APS) modifies locally-biased classical shadows (LBCS) by adding classical computation to basis selection. The introduction reports performance similar to derandomized shadows and, particularly on larger instances, apparent accuracy gains.
1. Introduction
- LBCS adapts classical shadows to estimate a single linear combination of Pauli observables, the energy-estimation task used for quantum Hamiltonians.The paper notes this task's ubiquity in quantum computing through variational quantum eigensolver algorithms.Source: §1 · p. 1
- APS performs similarly to derandomized shadows and appears more accurate in many cases, particularly on larger instances.The comparison is framed as allowing LBCS similar, but asymptotically better, classical computation.Source: §1 · p. 1
- The benchmark setup gives each algorithm 1,000 preparations of ground states for five molecules mapped to 8–16 qubits across three fermionic encodings.The molecules are H2, LiH, BeH2, H2O, and NH3; the reported error is the average absolute ground-energy estimation error.Source: §1 · p. 1
- The introduction compares random-Pauli classical shadows, LBCS, derandomized shadows, and APS, using different procedures to estimate their average errors.CS and LBCS use a variance-based estimate; derandomized shadows and APS use ten simulations and a root-square-mean deviation.Source: §1 · p. 2
2. Algorithm of Adaptive Pauli Shadows
APS estimates Hamiltonian energy by sampling Pauli measurement bases and adapting each qubit's basis choice to those already assigned to earlier qubits. Its basis-selection optimization adds classical computation, with runtime scaling as O(n_H · n).
2. Algorithm of Adaptive Pauli Shadows
- APS estimates energy from random Pauli-basis measurements, updating each Pauli term's current estimate and coverage count when the chosen basis covers it.A Pauli term is covered when each non-identity factor matches the corresponding measurement basis.Source: §2 · p. 2
- The measurement routine initializes an estimate and coverage count for each Hamiltonian Pauli term, selects a basis, then measures the prepared state qubit by qubit.The outcome for a covered term is used to update its estimate and count.Source: §2 · p. 3
- APS chooses each qubit's measurement basis using an optimized distribution that can depend on bases assigned to preceding qubits.The algorithm randomly permutes the qubit order, then samples each basis from its distribution.Source: §2 · pp. 3–4
- At each stage, APS optimizes a probability distribution over X, Y, and Z using the Pauli terms in Ω and the measurement bases assigned earlier.The sets Ω_B group terms in Ω by their factor at the current qubit; the optimization is convex.Source: §2 · p. 3
- The basis-selection optimization has runtime O(n_H · n), better than derandomized shadows, whose basis choice scales linearly with the number of state preparations S.The comparison is against LBCS runtime O(1); derandomized shadows also require S and certain hyperparameters to be fixed beforehand.Source: §2 · p. 4