Imagine you're assembling a massive jigsaw puzzle, but every piece subtly changes shape depending on which pieces are already placed nearby — and not just the adjacent ones, but pieces three or four rows away. That's the computational challenge of simulating Rydberg atom arrays in two dimensions: the long-range van der Waals interactions mean you can't just worry about nearest neighbors. Most tensor-network methods handle short-range entanglement beautifully but choke when distant atoms talk to each other. This paper says: projected entangled-pair states (PEPS), optimized with a sampling-based variational approach, handle those long-range terms on equal footing with short-range ones — no special tricks, no truncation guilt. The committed claim: finite PEPS can resolve debated ground-state phase diagrams on frustrated 2D Rydberg geometries — specifically, spin-glass behavior on the Kagomé lattice and topological order on the Ruby lattice beyond the PXP approximation — where other numerical methods either disagree or simply haven't been applied at sufficient fidelity. This is not a new algorithm paper. It's a deployment paper: take an existing tool (sampling-based PEPS optimization) and point it at physically contested questions that the experimental Rydberg community actually cares about. The architectural choice matters. PEPS is the 2D generalization of matrix product states (MPS), the workhorse of 1D quantum simulation. In 2D, PEPS contraction is #P-hard in general, so you need approximate contraction schemes — here, sampling-based optimization sidesteps the exact contraction bottleneck. The key structural advantage over, say, quantum Monte Carlo (QMC) is that PEPS doesn't suffer from the sign problem on frustrated lattices. And compared to the PXP model often used for Ruby-lattice Rydberg simulations, this work keeps the full interaction Hamiltonian, which is the honest thing to do when you want to predict what an actual experiment will see. The ladder question is where things get interesting and also where the paper is most honest by implication. The competitors here are exact diagonalization (limited to tiny systems), DMRG/MPS (great in 1D, struggles in 2D), QMC (sign problem on frustrated geometries), and the PXP approximation (which throws away interaction details). PEPS doesn't claim to beat all of these on speed — it claims to be the only method that handles 2D frustrated geometries with long-range interactions at reasonable system sizes without an uncontrolled approximation. The paper's contribution is less about a single headline number and more about demonstrating that PEPS is the right tool class for this specific experimental platform. Integrity-wise, this is a computational study validated against itself and against known phase-diagram expectations. There's no independent experimental comparison in the paper — the framing is that PEPS can serve as a 'digital twin' for Rydberg experiments, but the twin hasn't been checked against the physical system yet. The choice of Kagomé and Ruby lattices is well-motivated by the active experimental and theoretical debates, so the benchmarks aren't cherry-picked in topic, though the specific parameter regimes shown could be selectively favorable. No code release is mentioned. The milestone to watch: the gap between 'PEPS predicts phase X on geometry Y' and 'Rydberg experiment confirms PEPS prediction on geometry Y.' That's the real validation event. Rydberg array experiments on Kagomé and Ruby geometries are actively being pursued by groups at Harvard, MIT, and elsewhere. If a PEPS simulation correctly predicts a phase boundary that an experiment subsequently confirms — especially one where other methods disagreed — that would be the moment PEPS earns its 'digital twin' label for real. The obvious experiment not run: direct benchmarking against a Rydberg experiment's measured observables. The honest read is (a) — they don't have access to the experimental data at the precision needed, or the experiments on these specific geometries haven't been completed yet. This is a theory paper laying groundwork, not a theory-experiment comparison paper. That's fine, but it means the strongest possible validation is still pending.
