Derandomised tensor product gap amplification for quantum Hamiltonians
Thiago Bergamaschi, Tony Metger, Thomas Vidick, Tina Zhang
TL;DR
This work advances the quantum PCP program by introducing a derandomised gap amplification for local Hamiltonians that mimics Dinur’s classical approach but operates within the quantum setting. The core method combines derandomised tensor-product amplification with expander-walk sampling, implemented on layered, commuting components of the Hamiltonian, and analyzed via a novel miser’s de Finetti technique to manage entanglement. The authors establish completeness and soundness guarantees for the amplified Hamiltonian H^{(2t)}, derive an auxiliary energy measurement as a diagnostic tool, and show that the procedure is iteratable to yield streaming quantum PCP and locality-gap tradeoffs. Overall, the paper provides a concrete, quasi-derandomised framework to amplify energy gaps in quantum Hamiltonians while keeping the expanded Hamiltonian tractable, with potential implications for MIP*-style protocols and sparse quantum PCP conjectures.
Abstract
The quantum PCP conjecture asks whether it is QMA-hard to distinguish between high- and low-energy Hamiltonians even when the gap between "high" and "low" energy is large (constant). A natural proof strategy is gap amplification: start from the fact that high- and low-energy Hamiltonians are hard to distinguish if the gap is small (inverse polynomial) and amplify the Hamiltonians to increase the energy gap while preserving hardness. Such a gap amplification procedure is at the heart of Dinur's proof of the classical PCP theorem. In this work, following Dinur's model, we introduce a new quantum gap amplification procedure for Hamiltonians which uses random walks on expander graphs to derandomise (subsample the terms of) the tensor product amplification of a Hamiltonian. Curiously, our analysis relies on a new technique inspired by quantum de Finetti theorems, which have previously been used to rule out certain approaches to the quantum PCP conjecture.
