Area laws and tensor networks for two-dimensional gapped systems

The explanation

Picture a 1,000 × 1,000 grid of tiny magnets, quantum spins, one per square. They interact only with nearby spins. Cut out a 10 × 10 block in the middle. It holds 100 spins but only 40 edges cross its border. For a 100 × 100 block the counts are 10,000 spins and 400 border edges. The question is how much quantum information the block shares with the outside.

That shared information is the entanglement entropy. A simple count bounds it by the number of spins in the block. One shared pair of entangled spins across the border adds log 2, about 0.693. An area law says the entropy is at most a constant times the number of border edges instead. Here area means boundary size, a name from three dimensions.

The idea came from black-hole physics. Bombelli, Koul, Lee and Sorkin related entropy from hidden quantum fields to black-hole areaSource 1. For spin systems, the gap-to-area conjecture asks for an area law whenever the energy gap stays above a fixed sizeSource 2. The gap is the energy jump from the lowest state, the ground state, to the next. In one dimension, Hastings proved it in 2007Source 3. In higher dimensions, strict area laws were known only for restricted model classesSource 4. The obstacle is that a gap gives decay of correlations, but entropy needs more. It requires control of the reduced state's full spectrumSource 5.

The first paper claims the answer in two dimensionsSource 6. For any set of sites on a finite square-grid domain, the entropy is at most a constant times the number of crossing edgesSource 7. The constant depends only on the spin size, the interaction range and strength, and the gap. It does not depend on the domain's size or shape. No gap is assumed for smaller pieces of the system, only for the wholeSource 8. So a 100 × 100 block in the middle would hold at most 400 times that constant, not something growing like 10,000.

The second paper uses this to describe the state compactly. A projected entangled-pair state, or PEPS, stores the state as small tables of numbers, one per site, linked along edges. The bond dimension is the size of each link. The paper claims a PEPS approximation with bond dimension polynomial in L and error at most 1/LSource 9. For L = 100 the error is at most 0.01. It makes no claim that the tables can be found or used efficientlySource 10.

How sure is this? There is no Lean formalization for these papers. The claims rest on two long papers. No specialist review of them has been published.

What it doesn't show. This does not give an efficient algorithm to find the ground state or its tensors, and it does not cover three-dimensional lattices.