arXiv:2611.03207
Spin SYK-dot chains: entanglement bound and resource estimates towards a quantum advantage
Sayantan Dey, Abel Carreras, Siddhartha Patra
Abstract
Ground states of Sachdev-Ye-Kitaev (SYK)-type models are volume-law entangled, which makes them costly both for tensor networks and for quantum hardware simulations. Thus, it remains unclear where quantum hardware could outperform classical simulation of such models. In this work, we study a spin SYK-dot chain, in which q dense SYK dots of p qubits are joined by single ZZ bonds of strength K. The entanglement across any cut is then bounded by floor(p/2) ln 2 plus the inter-dot entanglement, and the two-qubit depth of one Trotter step is set by p alone. A resource estimate shows that a quantum advantage for the ground state could appear only for dots of about 15 to 21 qubits or more. There, a single Trotter step already requires two-qubit error rates of order 10^{-6}, beyond the reach of noisy intermediate-scale quantum (NISQ) hardware. We test the assumptions of this estimate with an imaginary-time matrix product state (MPS) that applies each dot as one block. It agrees with exact diagonalisation at (p, q) = (6, 2). At (5, 2), a variational circuit with the same dot-plus-link layout comes within 3.6% of the exact energy, and noisy simulations show that its fidelity decays as assumed in the resource estimate. At (8, 5), i.e., N = 40 qubits, the energy converges at bond dimension chi = 768, far below the exact value 2^20, and the inter-dot entanglement stays small and is set mainly by K. Out-of-time-order correlators across the links decay on a time scale that shrinks rapidly with |K|. Thus, the dot size p and the inter-dot coupling K together set the entanglement of the chain, and with it where a quantum advantage could appear.
AI Summary
Based on v1The paper asks where ground-state simulation of SYK-type spin models might offer quantum advantage when compared against tensor-network methods. It studies chains of $q$ dense, randomly coupled four-body spin dots, each containing $p$ qubits, connected by single $ZZ$ bonds of strength $K$. Subadditivity and the Araki–Lieb inequality give $S_{\mathrm{cut}}\le\lfloor p/2\rfloor\ln2+\max_x S_x$, where $S_x$ is the entropy across an inter-dot boundary. This is a state-independent relation, not a bound on the inter-dot entropy itself. Parallel dot evolution makes the estimated two-qubit depth per Trotter step independent of $q$, scaling as $O(p^3)$ with all-to-all intra-dot connectivity. Imaginary-time block-MPS simulations are checked against exact diagonalisation at small sizes. For $(p,q)=(8,5)$, a single-realisation energy converges to about $10^{-4}$ per site at bond dimension $768$, versus exact rank $2^{20}$. The disorder-averaged first-link entropy is $0.200\pm0.014$ at $K=2J$. A ten-qubit variational circuit reaches energy within $3.6\%$ of exact, with noise-induced fidelity loss approximately exponential in gate count. Using an operation-count estimate for an unimplemented one-site-per-dot MPS, a $10^{18}$-operation/1-TB budget, and modest link ranks, the estimated classical crossover is $p\approx15$–$21$ at $q=5$. One unoptimised Trotter step there requires gate errors around $10^{-6}$ or below. These estimates do not establish advantage: preparation gaps and step counts are unstudied, link-rank scaling is uncertain, large-system convergence sampling is limited, and OTOC Trotter errors are unquantified.
Central claim may not hold: The abstract's 'only' overstates a 15–21-qubit crossover derived from a chosen classical budget and reference algorithm rather than a necessary threshold for quantum advantage. The crossover extrapolates assumed link bond dimensions to p=15–21 despite numerical tests only at p≤8 and no established scaling with dot size or chain length. The stated approximately 10^-6 error requirement is specific to an unoptimized Trotter circuit and a global-fidelity noise criterion, not a demonstrated necessity for ground-state simulation generally.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- Comments
- 17 pages, 10 figures