[Paper Review] Notes on Some Questions in Mathematical Physics and Quantum Information

M. B. Hastings|arXiv (Cornell University)|Apr 16, 2014

Random Matrices and Applications4 citations

TL;DR

This paper presents a collection of independent notes on diverse topics in mathematical physics and quantum information, including tightened bounds on correlation decay in matrix product states with manifestly Hermitian transfer matrices, challenges in constructing PEPS for 2D fermionic systems with quantized Hall conductance, connections between almost commuting matrices and vector bundles, and an open question on quantum channel simulation with non-deterministic decoders. A key contribution is a proof of a 'Matthew principle' for quantum error correction, showing that if error correction improves performance at a given noise level, it also does so at higher noise levels.

ABSTRACT

This is a set of notes on some unrelated topics in mathematical physics, at varying levels of detail. First, I consider certain questions relating to the decay of correlation functions in matrix product states, in particular those generated by quantum expanders. This is discussed in relation to recent results of Brandao and Horodecki on area laws on systems with exponentially decaying correlation function\cite{areaexp}. Second, I consider some difficulties in trying to construct a tensor product state (or PEPS) describing a two-dimensional fermionic system with non-vanishing Hall conductance. Third, I present some relations between the theory of almost commuting matrices and that of vector bundles, making the connection between the classifications more explicit. Fourth, I present an open question about quantum channels, and some partial results.

Motivation & Objective

• To analyze and tighten bounds on exponential correlation decay in matrix product states, particularly when the transfer matrix is manifestly Hermitian.
• To investigate the fundamental challenges in constructing projected entangled pair states (PEPS) for two-dimensional fermionic systems exhibiting non-zero Hall conductance.
• To clarify and make explicit the connections between the classification of almost commuting matrices and topological K-theory via vector bundles.
• To explore an open problem in quantum channel theory concerning the simulation of noisy channels using non-deterministic decoders.

Proposed method

• Derive improved correlation decay bounds in matrix product states by exploiting the manifestly Hermitian property of the transfer matrix, leading to tighter prefactor dependence on the bond dimension.
• Use gauge freedom to transform matrix product state representations, enabling the construction of a Hermitian transfer operator that simplifies spectral analysis.
• Apply induction and differential inequalities to prove that the success probability of a quantum error-correcting code satisfies a monotonicity condition under the Matthew principle.
• Analyze the operator norm of the transfer matrix and its spectral gap to establish exponential decay of correlations with improved prefactor control.
• Use the monotonicity of the success-to-failure ratio $ g(p) = \tilde{f}(p)/(1 - \tilde{f}(p)) $ to derive a lower bound on its logarithmic derivative, linking it to the classical $ 1/[p(1-p)] $ form.
• Examine the structure of quantum channel equivalence classes and non-additivity to identify non-simulable channel pairs, particularly in the context of non-deterministic decoding.

Experimental results

Research questions

• RQ1 Can tighter bounds on the prefactor in correlation decay be derived for matrix product states when the transfer matrix is manifestly Hermitian?
• RQ2 Is it possible to construct a PEPS for a two-dimensional fermionic system with non-zero Hall conductance, and what are the fundamental obstructions?
• RQ3 How do classifications of almost commuting matrices relate to topological invariants in vector bundle theory, and can this connection be made more explicit?
• RQ4 Under what conditions can a quantum channel $ \mathcal{C}_p $ simulate another channel $ \mathcal{C}_q $ with $ q > p $ using non-deterministic decoders?
• RQ5 Does the Matthew principle for quantum error correction hold: if error correction helps at a low noise level $ p $, does it also help at higher noise levels $ p' > p $?

Key findings

• For matrix product states with a manifestly Hermitian transfer matrix, the prefactor in the exponential correlation decay bound is significantly improved, reducing dependence on the bond dimension $ k $.
• The paper proves a 'Matthew principle' for quantum error correction: if $ \tilde{f}(p) > p $ for some $ p < 1/2 $, then $ \tilde{f}(p) + \tilde{f}(1-p) > 1 $, implying improved performance at higher noise levels.
• The logarithmic derivative of the success-to-failure ratio $ g(p) $ satisfies $ \partial_p \ln g(p) \geq 1/[p(1-p)] $, which is a key inequality in proving the Matthew principle.
• The proof relies on induction and algebraic manipulation showing that $ \tilde{f}_1 - \tilde{f}_0 \geq (\tilde{f}_1 - \tilde{f}_0)^2 $, which holds due to the boundedness of probabilities in $[0,1]$.
• The analysis reveals that quantum channel simulation is not symmetric: there exist pairs of channels that cannot simulate each other, indicating a complex hierarchy of equivalence classes.
• The paper identifies a potential open problem: whether $ \mathcal{C}_p $ can simulate $ \mathcal{C}_q $ for $ p \leq 1/2 $ and $ q > p $ using non-deterministic decoders, highlighting non-trivial structure in channel simulation.