{"id":"2b0e5968-4168-433e-989b-8c73782d4266","arxiv_id":"2411.17575","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":3.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"QAOA on a 10-qubit trapped-ion processor finds near-simulated energy values for a 3-product, 2-shelf warehouse allocation QUBO instance.","lead":"This paper runs a quantum approximate optimization algorithm (QAOA) on a trapped-ion quantum processor for a small warehouse storage allocation problem. The hardware results roughly match simulations, but the problem instance is tiny and classically trivial.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The QUBO objective in Eq. (1b) is inherited from ref [18] without validation, and its binary variables encode only shelf assignment, not within-shelf order; without a derivation linking fB to reinsertion frequency, the claim of identifying the optimal warehouse configuration is ungrounded.","rationale":"The reader's weakest assumption correctly flags that the QUBO formulation is inherited from ref [18] without derivation or validation. I agree this is the least secure pillar of the central claim: even a perfect QAOA execution would not validate the warehouse conclusion if the objective function is wrong. My stress-test sharpens the concern by pointing out a specific structural mismatch: the QUBO variables encode only shelf assignment, whereas the paper's own description of gravity-shelf reinsertion (Fig. 1a) depends on the order of products within a shelf. Without an argument that lambda aggregates these positional effects, fB may be an arbitrary quadratic form rather than a faithful cost. The proposed test directly checks this for the demonstrated instance, which is small enough to enumerate exhaustively. I do not see a need to change the reader's CONDITIONAL verdict: the hardware demonstration of QAOA on a 10-qubit QUBO appears technically sound, and the required condition is to supply the missing validation (or to moderate the conclusion). I therefore keep the verdict unchanged. The agreement is partial because the reader focused on the general inherited equivalence and the capacity-encoding assumption, while I emphasize the within-shelf ordering gap as the concrete mechanism by which the equivalence could fail.","tokens_in":11020,"tokens_out":15066,"duration_ms":133190,"concrete_test":"For the paper's instance (P=3 products, M=2 shelves, L=2), build a discrete-event simulation of the JIS gravity-shelf system with FIFO retrieval: assign each product a demand frequency consistent with the off-diagonal lambda values (e.g., products 1 and 3 co-requested more often), generate many demand sequences, and compute the expected number of reinsertions for each of the six ordered allocations (two shelves, with the shelf of size 2 having two possible internal orders). Compare this expected-cost ranking with the QUBO objective f = 10fA + 0.5fB + 0.25fC evaluated on the corresponding assignments. If the assignment minimizing f does not also minimize the simulated expected reinsertions, then Eq. (1b) does not represent the stated warehouse cost and the paper's central claim is unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim in Sec. V is that the implemented QAOA algorithm identified the optimal warehouse configuration. This requires that the QUBO minimized by QAOA faithfully represents the actual warehouse cost. In Sec. II A, the paper defines the goal as minimizing the frequency of item reinsertions in a JIS gravity-shelf system, where reinsertion depends on the retrieval order and the position of products on a shelf (as illustrated in Fig. 1a). However, the QUBO objective fB in Eq. (1b) is only a sum of pairwise interproduct costs for products sharing a shelf; the binary variables x_m^alpha encode shelf assignment but not the within-shelf position or order. The paper provides no derivation connecting fB to reinsertion frequency, explicitly deferring the formulation to ref [18], and the interproduct cost matrix lambda is introduced without a demonstrated mapping to operational costs. Consequently, the optimal configuration found by QAOA may be optimal only for the artificial QUBO objective, not for the warehouse problem the paper claims to solve.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents a QAOA implementation for a warehouse allocation problem formulated as a QUBO. The authors consider a JIS gravity-shelf warehouse instance with P=3 products, M=2 shelves of capacity L=2, map the QUBO to a 10-qubit Hamiltonian using a Pauli-X encoding that is compatible with trapped-ion MS gates, and run the algorithm both in noiseless simulation and on AQT's IBEX trapped-ion processor. Simulation results include an energy landscape, a random multi-start strategy, and a recursive parameter-fixing strategy; hardware results report a mean expected energy of 2.23 compared with 2.40 in simulation. The paper concludes that the QAOA identified the optimal warehouse configuration in both simulation and hardware execution.","tokens_in":11293,"tokens_out":7566,"duration_ms":65953,"significance":"If the central claim were properly supported, the paper would be a useful proof-of-principle demonstration of QAOA for a small industrial logistics problem on trapped-ion hardware, with a native MS-gate-friendly Hamiltonian mapping and explicit resource estimates. The strengths include the clear presentation of the Hamiltonian construction, the resource scaling table, and the direct simulation-to-hardware comparison. However, the significance is currently limited by the very small instance size, the absence of any classical baseline or success-probability metric, and the reliance on an unvalidated QUBO model inherited from prior work.","major_comments":[{"comment":"The concluding claim that 'the implemented QAOA algorithm identified the optimal warehouse configuration in both simulation and execution on real quantum computing hardware' is not supported by the reported data. The hardware run (Sec. IV C) reports only the mean expected energy (E_hardware = 2.23) and does not report the probability of measuring a ground state or the frequency of the optimal feasible configuration among the 200 shots. The simulation results in Sec. IV B show at most 5.60% probability of the ground states at the global minimum of the energy landscape, and the recursive strategy reports only mean energies without reporting the ground-state energy or the success probability. To support the optimality claim, the paper must report the ground-state fidelity or the sampling probability of the optimal solution, and ideally the constraint-violation rate.","section":"Sec. V and Sec. IV C"},{"comment":"The QUBO objective f_B in Eq. (1b) is presented as the cost to be minimized for the warehouse problem, but the paper defers the derivation of this objective to ref. [18] and does not provide any validation linking the pairwise interproduct cost matrix λ to the reinsertion frequency in a JIS gravity-shelf system. The binary variables x_m^α encode only shelf assignment, not the within-shelf position or retrieval order, so the objective may not capture the cost mechanism illustrated in Fig. 1(a), where reinsertion depends on the position of items on the shelf. Because the central claim is about the optimal warehouse configuration, the manuscript must either derive the mapping from the operational cost to Eq. (1b) or limit the claim to optimality with respect to the adopted QUBO model.","section":"Sec. II A, Eq. (1b)"},{"comment":"The virtual-product encoding in Eq. (1c) is only exercised for the case L = 2, yet the paper states that the formulation 'is very general and can be applied to arbitrary number of objects to be allocated and arbitrary number of positions on the shelves.' For general L (or non-power-of-two capacities), the encoding requires proof that the penalty term exactly enforces the capacity constraint and does not introduce spurious low-energy states. This is particularly relevant because the number of virtual-product qubits is log2 L, which presumes binary encoding of the empty-space count; the manuscript does not discuss how the encoding behaves for general L. The generalization claim should be either proven or explicitly scoped to the power-of-two case demonstrated.","section":"Sec. II A, Eq. (1c) and Sec. II B"},{"comment":"The values of the penalty weights A = 10, B = A/20, and C = B/2 are chosen heuristically, and the paper does not report whether the final sampled states satisfy the constraints in Eqs. (1a) and (1c). If the reported mean energies include states that violate the 'one shelf per product' or capacity constraints, then the comparison to the ground state of the QUBO is not meaningful. The authors should report the fraction of samples that are feasible (e.g., no product assigned to two shelves, capacity respected) and ideally show that the infeasible states have energies sufficiently penalized.","section":"Sec. IV B"}],"minor_comments":[{"comment":"The gate error rates are given, but the paper does not state the number of shots used for the final energy estimation separately from the 200 shots per optimization step; no error bars or standard deviation for the hardware mean energy are reported.","section":"Sec. III"},{"comment":"The description of the recursive strategy is terse; it is not clear how the parameters of the first p-1 layers are fixed from a previous execution and whether the optimization is re-run for the last layer only. A precise algorithm pseudocode would improve reproducibility.","section":"Sec. IV B"},{"comment":"The resource table reports only asymptotic scalings; it would be useful to state the constant factors and the number of classical optimization parameters (2p) that dominate the classical optimization cost.","section":"Sec. IV A"},{"comment":"The sentence 'The algorithm yielded a 1.40% probability of identifying the possible ground states' for a single random start is useful, but the paper should also state how the ground states are defined (the two global minima) and how the probability is computed (sum over both states).","section":"Sec. IV B"},{"comment":"The phrase 'global maximum probability of 5.60%' is ambiguous; it should be 'probability of measuring one of the two ground states.'","section":"Sec. V"},{"comment":"The validation of the claim that lower expected energy corresponds to a more organized warehouse is deferred to ref. [50], a patent; this source is not accessible for verification, so the paper should either include a direct argument or cite a peer-reviewed reference.","section":"Sec. IV C"},{"comment":"The comparison of simulation and hardware would be more informative if the panels used the same vertical scale and if a quantitative distance (e.g., total variation distance) were reported.","section":"Fig. 3"},{"comment":"The notation a_m is not defined explicitly; clarify that a_m is the m-th shelf's virtual-product binary string and specify the range of the summation index i in the definition of ⟨2|a_m⟩.","section":"Eq. (2b)"}],"recommendation":"major_revision","confidential_remarks":"The paper relies heavily on ref. [18], a contemporaneous arXiv preprint by the same group, for the QUBO formulation; the referee should note that this model is not independently validated in the present manuscript. The demonstration is very small (10 qubits), does not benchmark against classical solvers, and the optimality claim is overstated relative to the reported data. These issues are fixable with additional analysis and revised claims, but they currently affect the central message."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a legitimate small proof-of-principle that QAOA with a σx mapping runs on a trapped-ion device, but the conclusion that the optimal warehouse configuration was identified is not supported by the paper's own numbers.\n\nThe new thing here is practical: the authors rewrite the QUBO so the problem Hamiltonian becomes ∑ J_ij σxσx + ∑ h_i σx, which compiles directly to native MS and single-qubit gates on AQT's IBEX. That is a sensible, useful trick for ion-trap QAOA. They also run a real 10-qubit experiment, report gate errors, give a resource-scaling table, and test a recursive parameter-initialization strategy that clearly improves the average expected energy. The writing is clear and the limitations of the simulator study (low ground-state probability, local minima) are acknowledged in the results section.\n\nThe soft spots are mostly in the interpretation. The central claim in Section V is that the algorithm 'identified the optimal warehouse configuration' in simulation and on hardware. The data do not support that. For the chosen parameters (A=10, B=0.5, C=0.25), the ground-state energy is 0.1 (shelf 1 holds products 1 and 3, cost 0.2·0.5). The best average expected energy they report is 2.03 (p=5 recursive), and the hardware run gives 2.23. The single-layer landscape has only 5.6% ground-state probability. So the algorithm never approaches the true optimum; it merely lowers a mean energy that is still far above the ground state. That is a proof-of-concept of the hardware and mapping, not a demonstration of optimization success.\n\nSecond, the QUBO itself is inherited from the companion paper [18] and the patent [50] with no derivation. The stress-test point holds: fB is a sum of pairwise costs for products that share a shelf, but the binary variables only encode shelf assignment, not position within the shelf. In a gravity shelf, reinsertion frequency depends on retrieval order and within-shelf position, so the link between fB and the operational cost is assumed, not shown. The authors should either derive it or explicitly label the objective as a proxy.\n\nMinor but worth fixing: no classical baseline (the instance has 32 feasible configurations, so it is trivially enumerable), no error bars on the hardware energies, and penalty weights A, B, C are hand-picked without sensitivity analysis.\n\nThe math and citation pattern are fine; the self-citation is justified. Overall, this is a honest, workmanlike hardware demonstration that overstates its conclusion. A serious referee should engage with it, but the paper needs revision before acceptance.","headline":"A legitimate small trapped-ion QAOA demo with a conclusion that overclaims: the reported energies are far above the true ground state.","tokens_in":11837,"tokens_out":4024,"would_cite":false,"duration_ms":93020,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A trapped-ion quantum processor identifies an optimal warehouse layout in a small test.","keywords":["warehouse optimization","QAOA","trapped-ion quantum processor","QUBO","combinatorial optimization","inventory management","variational quantum algorithms"],"falsifier":"Run a brute-force enumeration of all 1024 assignments for the three-product, two-shelf, capacity-two instance and compare the true ground state of Eq. (3) against the state most frequently sampled from the hardware; a mismatch would disprove the claim that the optimal warehouse configuration was identified.","tokens_in":10853,"feed_emoji":"📦","tokens_out":7113,"duration_ms":62081,"temperature":0.7,"pith_summary":"Warehouse managers want to place products on gravity shelves so that items requested together are stored together, reducing costly reinsertions. This paper adapts that allocation problem into a quadratic unconstrained binary optimization (QUBO) instance and solves it with the Quantum Approximate Optimization Algorithm (QAOA) on a trapped-ion quantum processor. The central claim is that the QAOA implementation, using ten qubits for a three-product, two-shelf, capacity-two warehouse, found the optimal configuration both in noiseless simulation and on real hardware. The paper also shows that a recursive parameter-initialization scheme lowers the average expected energy much faster than random multi-start as the circuit depth grows. If correct, this is a concrete demonstration that near-term trapped-ion devices can tackle practical logistics problems encoded as QUBO.","feed_headline":"Trapped-ion quantum computer finds best shelf layout","feed_subtitle":"Three products, two shelves: QAOA on real hardware matched simulation and picked the optimal assignment.","key_machinery":"The central object is the mapping of the warehouse allocation variables $x^m_\\alpha$ (product $\\alpha$ placed on shelf $m$) into spin operators $x^m_\\alpha \\to \\frac{1}{2}(I - \\hat\\sigma_x)$, which converts the QUBO cost $f = A f_A + B f_B + C f_C$ into an Ising Hamiltonian $\\hat H = \\sum_{i\\neq j} J_{ij} \\hat\\sigma_x^i \\hat\\sigma_x^j + \\sum_i h_i \\hat\\sigma_x^i$. Because only $\\hat\\sigma_x$ and $hat\\sigma_x\\hat\\sigma_x$ terms appear, the time evolution factorizes into rotations native to trapped-ion hardware via the Mølmer–Sørensen gate. The QAOA circuit alternates evolution under this problem Hamiltonian with a mixer $\\hat H_M = -\\sum_i \\hat\\sigma_z^i$, and the parameters are classically optimized. Capacity limits are enforced by adding 'virtual products' terms $\\langle 2|a^m \\rangle$, so empty shelf slots are treated as fillers; in this work the capacity is fixed at $L=2$.","core_discovery":"The paper claims that the implemented QAOA algorithm identified the optimal warehouse configuration in both simulation and execution on real quantum hardware. For the instance studied—three products distributed over two shelves of capacity two, with an interproduct cost matrix chosen so that products 1 and 3 are cheaper to store together—the most populated output states in both simulation and hardware correspond to the two global minima of the problem Hamiltonian. The hardware run reached an expected energy of $\\bar E_{\\mathrm{hardware}} = 2.23$, slightly lower than the simulated $\\bar E_{\\mathrm{sim}} = 2.40$, a difference the authors attribute to shot noise and device imperfections. Beyond the single run, the paper establishes that a recursive strategy, which fixes earlier-layer parameters based on previous runs, drives the average expected energy from $\\bar E = 15.24$ at one layer down to $\\bar E = 2.03$ at five layers, approaching the global minimum, while a random multi-start strategy improves only linearly with depth.","pith_inferences":["If the QUBO formulation faithfully represents reinsertion cost, the approach could be extended to larger JIS gravity-shelf warehouses, but the exponential state space and hardware noise will require error mitigation or error correction before practical advantage appears.","The recursive parameter initialization could be tested on other QUBO problems, such as portfolio optimization or scheduling, to see whether the convergence speedup is general or specific to this cost landscape.","A direct benchmark against classical exact solvers for the same instance would quantify the quantum overhead; the paper does not provide such a comparison.","The capacity constraint's virtual-product encoding is only exercised for $L=2$; checking $L=4$ or $L=8$ against brute-force enumeration would validate or invalidate the generalization to arbitrary shelf capacities."],"forward_implications":["For the ten-qubit instance tested, the QAOA output on real trapped-ion hardware is consistent with the optimal assignment found in simulation, indicating that current NISQ devices can run this QUBO encoding.","The recursive parameter-fixing strategy reduces the average expected energy from about 15.2 to 2.0 with five layers, offering a practical recipe for improving convergence in low-depth QAOA.","The resource scaling of about $M(P+1+\\log_2 L)$ qubits and $O(p M P (M+P+\\log_2 L))$ two-qubit gates sets a concrete target for scaling to realistic warehouses, such as the 1900 qubits estimated for 15 products on 100 shelves.","The same QUBO-to-Ising mapping with $\\hat\\sigma_x$ interactions could be applied to other allocation problems on trapped-ion processors."],"supporting_citations":[{"why":"Supplies the QUBO formulation of the warehouse allocation problem that this paper adapts and implements.","marker":"[18]"},{"why":"Introduces the Quantum Approximate Optimization Algorithm, the core hybrid quantum-classical solver used.","marker":"[10]"},{"why":"Provides the Mølmer–Sørensen gate, the native two-qubit entangling operation enabling the problem Hamiltonian's time evolution on trapped-ion hardware.","marker":"[33]"},{"why":"The derivative-free classical optimizer used to find the variational parameters.","marker":"[38]"},{"why":"Supports the use of multistart and recursive parameter-initialization strategies to handle local minima.","marker":"[39]"},{"why":"Describes the trapped-ion quantum computer system used for the real-device execution, including its native gate set and all-to-all connectivity.","marker":"[40]"}],"fun_headline_variants":["Trapped-ion QAOA picks optimal shelf assignment","Quantum algorithm finds optimal warehouse storage on real hardware","Trapped-ion quantum solves small warehouse optimization problem","Quantum computer matches simulation in warehouse layout test","QAOA on trapped-ion hardware finds optimal warehouse configuration"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The cost function's fidelity to the real warehouse operation is inherited from the authors' earlier work (ref. [18]) and is not rederived or validated here, so the hardware result optimizes that QUBO objective rather than a directly measured reinsertion cost.","fun_headline_variants_meta":{"raw":{"variants":["Trapped-ion QAOA picks optimal shelf assignment","Quantum algorithm finds optimal warehouse storage on real hardware","Trapped-ion quantum solves small warehouse optimization problem","Quantum computer matches simulation in warehouse layout test","QAOA on trapped-ion hardware finds optimal warehouse configuration"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000697,"raw_usage":{"total_tokens":3087,"prompt_tokens":822,"completion_tokens":2265,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":438,"completion_tokens_details":{"reasoning_tokens":2194}},"tokens_in":438,"tokens_out":2265,"duration_ms":17129,"temperature":1.0,"reasoning_tokens":2194,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T11:58:04.229804+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run a brute-force enumeration of all 1024 assignments for the three-product, two-shelf, capacity-two instance and compare the true ground state of Eq. (3) against the state most frequently sampled from the hardware; a mismatch would disprove the claim that the optimal warehouse configuration was identified.","supporting_citations":[{"cited_title":"Potvin, Genetic algorithms for the traveling salesman problem, Ann","cited_arxiv_id":null,"evidence_quote":"Introduces the Quantum Approximate Optimization Algorithm, the core hybrid quantum-classical solver used."},{"cited_title":"Bittel, S","cited_arxiv_id":null,"evidence_quote":"Provides the Mølmer–Sørensen gate, the native two-qubit entangling operation enabling the problem Hamiltonian's time evolution on trapped-ion hardware."},{"cited_title":"Shaydulin, I","cited_arxiv_id":null,"evidence_quote":"Describes the trapped-ion quantum computer system used for the real-device execution, including its native gate set and all-to-all connectivity."}],"review_version":1}