{"id":"14a90ab8-39bc-4d2f-835e-6547859b4943","arxiv_id":"2411.10444","paper_version":2,"verdict":"REJECT","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":3,"one_line_summary":"The paper proposes a joint MPC dispatch policy for shared autonomous electric vehicles that balances passenger service with grid power restoration, and an ADMM-based distributed solver, claiming passenger queue stability whenever any policy can achieve it.","lead":"Researchers designed a dispatch system for shared autonomous electric vehicles that, after a disaster, both carries passengers and uses the car batteries to help restore power to the grid. The system coordinates a fleet operator and a grid operator, and a distributed algorithm makes the calculations fast enough for realistic networks.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The maximum-throughput guarantee rests on an unproven transfer of the Kang–Levin stability theorem; Section II-E asserts the transfer and defers the proof to thesis [34], leaving the paper's central claim unsupported.","rationale":"The reader's weakest assumption identifies the same load-bearing concern: the maximum-throughput claim is transferred from Kang–Levin to the SAEV/grid setting by assertion, not by proof. My reading of Section II-E and objective (35) confirms that the demand-responsive price term is intended to force passenger service when queues grow, but the actual Lyapunov/drift conditions are not established. Grid revenue, battery degradation, charging-station capacity, state-of-charge dynamics, and the receding-horizon MPC with forecast demand all enter the model in ways that the original theorem does not directly cover. This is an internal-support problem, not a disagreement with consensus: the claim could be true, but the paper does not supply the argument needed to support it. The ADMM contribution is framed honestly as a heuristic, so the core problem is isolated to the maximum-throughput guarantee. The numerical results are suggestive but not sufficient, and one passage even appears to state that service rate is less than arrival rate, which would contradict bounded queues if taken literally. Since the reader already rejected on this basis, my stress-test does not move the verdict.","tokens_in":17145,"tokens_out":6168,"duration_ms":73203,"concrete_test":"Independently re-derive the stability transfer: restate the Kang–Levin drift condition for the full SAEV model and check the key step on the 5-node toy network. Compute whether there exists a threshold W such that whenever max_qr w_qr(t) > W, the optimum of (35) under constraints (1)-(34) with a short horizon serves at least one waiting passenger in the first implemented step and decreases the Lyapunov function, regardless of µ2, µ3, and µ4. If a single reachable state with w_qr(t) > W yields an MPC optimum that only charges/discharges and serves no passenger, the transfer fails as stated. If no such state is found on the toy network, repeat on the Sioux Falls/IEEE-85 case; absence of a counterexample is still not a proof, so a full drift inequality or a softening of the abstract's claim would be required.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central guarantee—maximum passenger throughput if any policy can—depends on Section II-E's claim that the Kang–Levin [30] maximum-stability theorem transfers to this SAEV/grid model, but the claim is asserted, with the proof deferred to thesis [34]. The transfer is not automatic. The Kang–Levin Lyapunov/drift argument requires that, whenever queues are long enough, a feasible action exists that reduces a Lyapunov function and that the queue-length-dependent term in the objective dominates all other terms. In objective (35) the demand-responsive term uses the current queue length w_rs(t) as a coefficient on pickups at every horizon time, while grid revenue µ2, generation cost µ3, and battery degradation µ4 are fixed per unit energy; even if these are finite, the proof must show that the queue term dominates uniformly over feasible actions and that battery bounds (13)-(14), charger limits (12), and power-flow/topology constraints do not prevent the policy from mimicking a stabilizing dispatch at the required times. It must also handle the receding-horizon MPC with predicted demand in Eq. (4), since finite-horizon forecasts can differ from realized queues, and the maximum-stability theorem is ordinarily stated for a policy that reacts to current queues. The numerical experiments (Figs. 4-7, 10-11) check only two scenarios and do not verify a maximum-throughput bound; the text around Fig. 4 even says the service rate is less than the incoming demand rate, which, if literal, contradicts the bounded-queue claim. The load-bearing proof is missing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a model predictive control (MPC) dispatch policy for shared autonomous electric vehicles (SAEVs) that coordinates passenger pickup with grid services (charging/discharging) in a distribution network. The policy is claimed to provide maximum passenger throughput if any policy can, based on an asserted transfer of a maximum-stability theorem for shared autonomous vehicles. The paper also develops a hierarchical distributed solution method based on ADMM, with local vehicle subproblems and a central coordinator, and reports case studies on a small 5-node system and a combined Sioux Falls / IEEE 85-node network comparing SAEVs, SA Vs, and TESSs.","tokens_in":17525,"tokens_out":6225,"duration_ms":62035,"significance":"If the maximum-throughput guarantee were rigorously established, the work would be a valuable extension of max-stability dispatch to coupled transportation-power systems, with clear practical relevance for disaster recovery. The model is comprehensive and the case studies illustrate that joint modeling of passenger and grid constraints matters. The ADMM decomposition with privacy preservation is a reasonable practical contribution. However, the central theoretical claim is not proven in the manuscript, the ADMM heuristic is acknowledged to lack convergence guarantees, and the numerical evaluation does not quantify near-optimality. As it stands, the paper's advertised guarantee is unsupported.","major_comments":[{"comment":"The maximum-throughput guarantee is asserted, not derived. The paper states that the Kang-Levin stability proof [30] applies 'within this electric vehicle framework the same is true' and defers the proof to thesis [34]. Since this property is the paper's central claimed contribution, the manuscript must include a self-contained proof or at least a precise statement of conditions under which the Lyapunov drift argument carries over. The transfer is nontrivial because objective (35) includes finite grid revenue, generation cost, and battery degradation terms that must be dominated by the queue-dependent term, and because battery limits (14), charger limits (12), and radiality constraints (15)-(24) can restrict the set of stabilizing actions.","section":"Section II-E and Abstract"},{"comment":"The text says 'the service rate is less than the incoming demand rate (meaning in the long run all passengers will be served).' This is internally inconsistent: if the service rate is below the arrival rate, queues grow without bound. If the intended statement is that the service rate is greater than the incoming demand rate, the reported queue growth should be reconciled with the bounded-queue claim. In the larger network, the statement that neither fleet stabilizes passenger demands within the first 3 hours also needs to be reconciled with the maximum-throughput guarantee; a transient explanation should be made explicit and supported.","section":"Section IV-A (paragraph after Fig. 4)"},{"comment":"The ADMM heuristic is acknowledged to have no convergence guarantees for the MILP, and the near-optimality claim is supported only by two scenarios. For the toy network, the comparison with the centralized solution is limited to 3 hours and no optimality gap is reported. For the Sioux Falls network, the text states that centralized solutions can take hours or may not be found, so no exact benchmark is available. The claim 'near-optimal solutions quickly' should be either quantified with objective-value gaps on instances where an exact solution is available, or softened to 'empirically good performance on the tested instances.'","section":"Section III-B and Section IV"},{"comment":"The passenger queue evolution inside the MPC horizon uses predicted demand \\tilde d_{qr}(t\\tau). The maximum-stability theorem in [30] is stated for a policy that reacts to realized queue lengths. The paper does not show that the receding-horizon implementation with forecast errors and finite horizon preserves the bounded-queue guarantee. This is a distinct issue from the grid constraints and should be addressed as part of the missing proof.","section":"Section II-A and Eq. (4)"}],"minor_comments":[{"comment":"The sentence 'we also need to ensure that vehicle are only allowed to charge and discharge' should be 'vehicles are only allowed.'","section":"Section II-C (near Eq. (12))"},{"comment":"The phrase 'retaining a the vehicle dispatcher' should be 'retaining the vehicle dispatcher.'","section":"Section III-A"},{"comment":"The symbol \\bar{e}^{sv}_q is used but not defined; please define it in the text.","section":"Eq. (9)"},{"comment":"The caption 'points are lower-level iterations and lines are upper-level iterations' is unclear; please clarify what the lines connect.","section":"Figure 9 caption"},{"comment":"The central theoretical claim is supported by citation to thesis [34]; if this remains the basis, please include the relevant argument or a rigorous sketch in the paper itself, since readers of a journal article should not need to consult a thesis to verify the main theorem.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The main concern is that the paper's headline guarantee is not proven in the manuscript. The simulation section also contains a contradictory sentence about service rate versus arrival rate. These are load-bearing issues, but they may be fixable if the authors provide a rigorous proof (or clearly stated conditions) and correct the simulation interpretation. The ADMM heuristic is presented honestly as a tuned heuristic, but the near-optimality language should be moderated unless gaps are reported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a solid engineering paper on integrating maximum-stability SAEV dispatch with distribution grid restoration, but the central theoretical claim—maximum passenger throughput if any policy can—is not actually established in the manuscript. The proof is deferred to the first author's thesis [34], and the transfer is not automatic. I'd send it to review, but with the clear expectation that the authors either prove the transfer or explicitly soften the abstract and Section II-E.\n\nWhat's genuinely new: the combination of Kang–Levin maximum-stability dispatch with LinDistFlow-based grid restoration in one MPC framework, and the hierarchical ADMM decomposition that keeps vehicle-level routing private and parallelizes well. The larger Sioux Falls/IEEE-85 case study is a reasonable demonstration that the heuristic converges in minutes and produces sensible trade-offs. The comparisons of SAVs, TESSs, SAEVs, and a 50/50 mix make the point that ignoring either side's constraints misleads.\n\nWhere it's soft: the maximum-throughput claim. Section II-E says 'within this electric vehicle framework the same is true' and points to [34]. But the Kang–Levin theorem is for a policy reacting to current queues, while this is a receding-horizon MPC with predicted demand (Eq. 4) and extra battery/grid constraints. The µ0 term uses the queue length at horizon start, not a running cost that provably dominates every feasible action; the dominance argument needs real work. The text around Fig. 4 also says 'the service rate is less than the incoming demand rate' while claiming long-run stability—that's either a typo or a sign the bounded-queue claim isn't being checked as stated. The ADMM heuristic is honestly labeled as lacking convergence guarantees, and the free parameters (µ's, ρ's, α, ε) are tuned, so the near-optimality claim rests on the two scenarios shown.\n\nIs it a serious thinker? Yes—the authors are clear about what's heuristic and what's inherited, and the integration is careful. The problem is a missing proof, not confusion.\n\nWho it's for: researchers working on mobility-on-demand and grid resilience who want a concrete coupled formulation and a scalable heuristic. The max-throughput guarantee should not be cited without checking the thesis.\n\nRecommendation: send to peer review. The central flaw is fixable—supply the proof or soften the claim to 'stable in simulation'—and the engineering contribution is worth refereeing. I would not accept it in present form.","headline":"A useful integration of maximum-stability dispatch with grid restoration, but the headline guarantee is asserted, not proven—worth refereeing with the expectation of fixing or softening that claim.","tokens_in":18005,"tokens_out":3791,"would_cite":true,"duration_ms":33598,"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 single dispatch policy can keep passenger queues bounded while using fleet batteries to restore critical electric loads after a disaster.","keywords":["shared autonomous electric vehicles","dispatch policy","maximum throughput","grid resilience","service restoration","ADMM","model predictive control","vehicle-to-grid"],"falsifier":"Solve the centralized model on a small network with grid-service revenue $\\mu_2$ set well above the demand-responsive price $\\mu_0$ and queues that grow long; if the resulting dispatch lets the expected waiting queue diverge while a stabilizing schedule exists, the maximum-throughput claim fails. The paper's simulations use one fixed set of prices and do not probe this boundary.","tokens_in":16967,"feed_emoji":"⚡","tokens_out":9288,"duration_ms":82940,"temperature":0.7,"pith_summary":"This paper tries to establish that a fleet of shared autonomous electric vehicles can do two jobs at once after a disaster: keep carrying passengers (especially essential workers) and use its batteries to restore electric loads that the damaged grid cannot serve. The proposed model predictive control policy maximizes combined revenue from passenger fares and energy delivery, and it is claimed to provide maximum passenger throughput—meaning it keeps the expected number of waiting passengers bounded whenever any dispatch policy could do so. The paper also proposes a two-level distributed solution method based on the alternating direction method of multipliers (ADMM), so that the fleet operator and grid operator solve separate problems and only exchange predicted power flows at charging stations. If these claims are right, the practical consequence is that disaster recovery does not have to choose between mobility and power: one coordinated policy can serve both, and ignoring either side leads to misleading performance estimates.","feed_headline":"One dispatch policy stabilizes passenger queues and restores power","feed_subtitle":"After a disaster, the same electric vehicles can carry critical workers and deliver power to loads the grid can't reach.","key_machinery":"The carrying mechanism is the objective function of the centralized MILP: a demand-responsive pricing term $\\mu_0(r,s,t)w_{rs}(t)$ multiplying passenger pickups, which grows without bound as queue lengths grow. That term makes passenger service eventually dominate all other revenues and costs, which is what forces vehicles to serve the longest queues first and keeps expected waiting passengers bounded; the paper asserts the same argument holds within the electric vehicle framework as in the maximum-stability dispatch theorem. Around this objective, the paper builds a two-level distributed algorithm based on the alternating direction method of multipliers (ADMM): the grid operator and vehicle dispatcher solve separate subproblems and exchange only predicted power flows at charging stations until consensus, and the dispatcher further splits per-vehicle routing subproblems so they run in parallel. The LinDistFlow model, a linearized radial distribution power-flow model, together with radial-topology constraints, describes how vehicle power transfers move through the distribution network.","core_discovery":"The central claim is that the dispatch policy obtained by solving the model predictive control problem—a mixed-integer linear program over the joint transportation and distribution network—stabilizes passenger demand whenever stabilization is possible, even while vehicles charge and discharge to serve electric loads. The objective includes a demand-responsive price term that grows without bound as passenger queues grow, so that eventually serving the longest queues dominates any revenue from grid service, battery degradation, or electricity generation. The paper presents this as an extension of the maximum-stability dispatch theorem for shared autonomous vehicles to the SAEV/grid setting, with the detailed proof deferred to the first author's thesis. Simulations on small and large networks show queues staying bounded while nearly all unserved electric demand is restored, and the distributed ADMM heuristic reproduces the centralized behavior in a fraction of the computation time.","pith_inferences":["Editorial inference: because the demand-responsive price term is what makes queues stable, capping fares or letting grid-service revenue dominate could break the guarantee; this is a regime the paper's fixed-price simulations do not test.","Editorial inference: the same two-level ADMM decomposition could serve day-to-day vehicle-to-grid functions such as peak shaving and frequency regulation, not only disaster restoration, since the model already captures charging and discharging dynamics.","Editorial inference: because vehicles only report aggregate pickups and charging amounts, competing fleet operators could contribute to grid restoration without revealing routing or payment data, which the paper motivates but does not formalize."],"forward_implications":["If the maximum-throughput claim holds, the same fleet can serve every passenger demand that any feasible dispatch policy could serve while also restoring critical loads, so disaster response planners do not have to sacrifice mobility for power.","Ignoring grid constraints when planning SAEV dispatch understates passenger queues; ignoring passenger service overstates how much energy the fleet can deliver, so joint models are needed for accurate restoration estimates.","The ADMM heuristic finds near-optimal solutions in 1–5 minutes on the 150-vehicle Sioux Falls / IEEE 85-node test case, where centralized solvers take tens of minutes to hours, making real-time implementation plausible.","Queues grow more stable as congestion increases because the distributed subproblems have fewer conflicts, so the algorithm is fastest exactly when optimal dispatch matters most.","Because only aggregate pickup and charging decisions are exchanged, the hierarchical approach preserves privacy for individual vehicle routes and for each operator's internal costs."],"supporting_citations":[{"why":"Supplies the maximum-stability dispatch theorem that this paper extends to the SAEV/grid setting.","marker":"[30]"},{"why":"First author's thesis, cited as the source of the proof that the stability property carries over to the electric-vehicle framework.","marker":"[34]"},{"why":"Prior real-time dispatching strategy for shared automated electric vehicles with performance guarantees that motivates the maximum-throughput objective.","marker":"[4]"},{"why":"Maximum-throughput dispatch for shared autonomous vehicles including rebalancing, used as the aggregate-flow baseline this approach builds on.","marker":"[25]"},{"why":"The alternating-direction method of multipliers formulation that provides both levels of the distributed solution approach.","marker":"[35]"},{"why":"LinDistFlow radial power-flow model used to represent how vehicle power transfers propagate through the distribution network.","marker":"[39]"},{"why":"Source of the proximal-term adjustment used to break vehicle homogeneity and stabilize ADMM convergence.","marker":"[43]"}],"fun_headline_variants":["One dispatch policy stabilizes riders and restores grid post-disaster","Shared EVs carry workers and deliver power after disaster","Balanced dispatch keeps passengers moving and grid alive","Optimal policy balances passenger flow and electric load"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The bounded-queue guarantee rests on the demand-responsive price term growing without limit as queues grow, and the paper assumes this dominance still holds when vehicles must also charge and discharge for the grid, deferring the proof to the first author's thesis.","fun_headline_variants_meta":{"raw":{"variants":["One dispatch policy stabilizes riders and restores grid post-disaster","Shared EVs carry workers and deliver power after disaster","Balanced dispatch keeps passengers moving and grid alive","Optimal policy balances passenger flow and electric load"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.00061,"raw_usage":{"total_tokens":2811,"prompt_tokens":888,"completion_tokens":1923,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":504,"completion_tokens_details":{"reasoning_tokens":1860}},"tokens_in":504,"tokens_out":1923,"duration_ms":19145,"temperature":1.0,"reasoning_tokens":1860,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T19:36:48.603024+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Solve the centralized model on a small network with grid-service revenue $\\mu_2$ set well above the demand-responsive price $\\mu_0$ and queues that grow long; if the resulting dispatch lets the expected waiting queue diverge while a stabilizing schedule exists, the maximum-throughput claim fails. The paper's simulations use one fixed set of prices and do not probe this boundary.","supporting_citations":[{"cited_title":"Maximum-Stability Dispatch Policy for Shared Autonomous Vehicles,","cited_arxiv_id":null,"evidence_quote":"Supplies the maximum-stability dispatch theorem that this paper extends to the SAEV/grid setting."},{"cited_title":"Resilience and Operational Benefits of Electric Vehicle and Grid Integration,","cited_arxiv_id":null,"evidence_quote":"First author's thesis, cited as the source of the proof that the stability property carries over to the electric-vehicle framework."},{"cited_title":"A Real-Time Dispatching Strategy for Shared Automated Electric Vehicles with Performance Guarantees,","cited_arxiv_id":null,"evidence_quote":"Prior real-time dispatching strategy for shared automated electric vehicles with performance guarantees that motivates the maximum-throughput objective."},{"cited_title":"Maximum Throughput Dis- patch for Shared Autonomous Vehicles Including Vehicle Re- balancing,","cited_arxiv_id":null,"evidence_quote":"Maximum-throughput dispatch for shared autonomous vehicles including rebalancing, used as the aggregate-flow baseline this approach builds on."},{"cited_title":"Distributed Optimization and Statistical Learning via the Al- ternating Direction Method of Multipliers,","cited_arxiv_id":null,"evidence_quote":"The alternating-direction method of multipliers formulation that provides both levels of the distributed solution approach."},{"cited_title":"Distributed Control of Reactive Power Flow in a Radial Distribution Circuit with High Photovoltaic Penetration,","cited_arxiv_id":null,"evidence_quote":"LinDistFlow radial power-flow model used to represent how vehicle power transfers propagate through the distribution network."},{"cited_title":"Two-Stage Fully Distributed Approach for Unit Commitment with Consensus Admm,","cited_arxiv_id":null,"evidence_quote":"Source of the proximal-term adjustment used to break vehicle homogeneity and stabilize ADMM convergence."}],"review_version":1}