{"id":"7a3dbfd5-b869-4ffb-812a-212eedc91fad","arxiv_id":"2411.15427","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":12,"one_line_summary":"A day-to-day multimodal traffic model with optimal ridesharing matching and cancellations is applied to a benchmark network to assess vehicle restriction and pricing policies.","lead":"This paper builds a computer model of a city transport network with solo driving, ridesharing, buses, and metro, where drivers and passengers are matched by an optimization routine each day and may cancel. It then tests car restriction and pricing policies, showing trade-offs between congestion, emissions, travel cost, and fairness across traveler groups.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Fixed-point existence for the DTD model with discrete ridesharing matching is unproven; policy comparisons rest on a steady state that may not be well-defined.","rationale":"We read the paper in good faith. The methodological contribution is substantial: it is the first to embed centralized ridesharing matching with cancellations into a path-based DTD multimodal assignment model, and the Sioux-Falls experiments are clearly described. The sensitivity analysis is a reasonable demonstration of the model's capacity. However, the central claim that the model 'reaches stable states' is the foundation on which all policy implications are built. The reader's weakest-assumption analysis correctly identifies the unproven fixed point. We agree, and add that the mode-choice model's treatment of matching failure creates a further consistency issue: the logit choice of ridesharing is based on the cost of a successful match, while the actual probability of success and the cost of the fallback are not part of the choice utility. This means the steady state is not a standard behavioral equilibrium, so the comparative statics are hard to interpret. We therefore recommend keeping the CONDITIONAL verdict: the paper is publishable as a methodological proposal, but only if the authors either provide a convergence/stability analysis (at least for the tested cases) or clearly reframe the results as simulation outcomes without claiming a general fixed-point property. The proposed concrete test—running the most extreme policies under the least favorable learning parameters and multiple initial conditions—would directly determine whether the reported steady states are robust or artifacts.","tokens_in":17784,"tokens_out":9620,"duration_ms":85772,"concrete_test":"Run the DTD process for the three most extreme policy scenarios (ban=0.7, bus fare=10 yuan, ridesharing fare=15 yuan/km) with N=3 and θ1=θ2=0.01, the least favorable parameter settings for convergence, and also with 100 random perturbations of the initial day-0 mode flows. Record whether gap1 and gap2 from Eqs. (3.1)-(3.2) both fall below 10^-4 within 2000 days and whether the terminal mode splits are identical across perturbations for each scenario. If any scenario fails to converge or multiple attractors are found, the 'steady state' used in the sensitivity analysis is not a well-defined object and the policy rankings in Section 4 are unsupported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim that the integrated DTD model 'reaches stable states' is load-bearing because every policy conclusion in Sections 4–5 compares steady states under different fares and ownership bans. That claim rests on the fixed-point condition (2.29), C* = C(h_p*), h_p* = Λ(C*), where the loading map Λ embeds the BIPM ridesharing matching of Section 2.2.1. The matching is an integer program whose optimal solution can change discontinuously with the travel times that enter Eq. (2.2); no existence, uniqueness, or stability proof is given. Section 3.2 only reports numerical decay of gap1/gap2 for six parameter combinations and explicitly concedes that convergence 'may fail' on large networks or with highly sensitive travelers (small N, large θ). The policy rankings in Section 4 are therefore conditional on an unverified attractor. A second, related gap is that the mode-choice logit in Eq. (2.25) uses the perceived cost of a successful ridesharing match, while the matching failure rate and the resulting fallback costs are not embedded in the utility; unmatched travelers are reassigned after the mode choice, so the final mode shares need not be a consistent stochastic user equilibrium. This further undermines the interpretation of the 'stable state' as a behavioral equilibrium.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a path-based deterministic day-to-day (DTD) multimodal traffic assignment model that embeds a centralized ridesharing matching optimization (direct matching followed by a binary integer program, BIPM) with match cancellation and mode-adjustment rules. The model includes five travel modes (solo driving, ridesharing driver, ridesharing passenger, bus, metro) and two traveler groups (vehicle owners and non-owners). A weighted-average learning rule updates perceived costs, and a logit model governs mode and route choices. The steady state is characterized by the fixed-point condition C* = C(h_p*), h_p* = Λ(C*), and is investigated numerically on the Sioux-Falls network. Sensitivity analyses vary car-ownership bans, bus fares, and ridesharing fares, reporting effects on mode split, PCU, travel costs, and emissions, and drawing policy implications about trade-offs and social equity.","tokens_in":18231,"tokens_out":3332,"duration_ms":33558,"significance":"If the modeling framework is accepted, it provides a useful integration of an explicit ridesharing matching mechanism into a multimodal DTD assignment model, going beyond earlier studies that treated matching as a probability or capacity constraint. The paper is commendable for including cancellations and mode adjustment after failed matches, and for examining policy impacts separately for all OD pairs and mass-transit OD pairs, as well as by ownership group. It also reports numerical convergence over several parameter combinations and states its limitations candidly. However, the central policy conclusions all rely on comparing stable states of the DTD process, and the analysis is carried out on a single synthetic network with many hand-set parameters and no empirical validation. The main gaps are theoretical: the fixed point is assumed rather than proven under a discrete matching map, and the final mode shares are not consistent with the logit choice probabilities because unmatched travelers are reassigned after mode choice. These issues must be addressed before the policy implications can be regarded as robust.","major_comments":[{"comment":"The steady state is defined by the fixed-point condition C* = C(h_p*), h_p* = Λ(C*), but no existence, uniqueness, or stability proof is given for this DTD process. The loading map Λ embeds the BIPM ridesharing matching of Section 2.2.1.1, an integer program whose optimal solution can change discontinuously with travel times and costs. Section 3.2 only reports numerical convergence toward gaps below 10^-4 for six parameter settings, and the authors explicitly concede that convergence may fail on large networks or with highly sensitive travelers (small N, large θ). Because Sections 4 and 5 compare stable states under different policies, all policy conclusions are conditional on an unverified attractor. At minimum, the paper should prove existence (e.g., by showing the composite map C → Λ(C) → C is continuous or has a fixed point under the particular finite choice sets used) or restrict the policy claims to a stated convergence region and provide additional robustness checks over a wider parameter grid.","section":"§2.3.3, Eq. (2.29) and §3.2"},{"comment":"The logit mode-choice model in Eq. (2.25) is based on the perceived costs of the five modes, including ridesharing modes, as if all announced ridesharing trips are successful. However, after the BIPM matching solution and the cancellation threshold are applied, unmatched drivers and car-owning passengers are reassigned to solo driving and unmatched non-owners to bus or metro (Section 2.2.1.2). These fallback flows are added to the mode shares and path flows, so the final mode split is not the one predicted by the logit model. The steady state therefore cannot be interpreted as a stochastic user equilibrium over the five modes. This inconsistency should be addressed explicitly, for example by modeling the choice as a two-stage process (announcement followed by fallback) or by iterating between the matching and the mode-choice adjustment until the flows used in costs are the flows that generate those costs.","section":"§2.2.1.2 and §2.3.2, Eqs. (2.25)–(2.28)"},{"comment":"The matching cost rate β is set to 0 in the numerical study, and the authors state that this 'does not affect the analysis results.' This assertion is not supported by any sensitivity test. In Eq. (2.11), β scales the driver's net revenue from the passenger fare, so it directly changes the generalized cost of ridesharing driving and hence the supply of drivers and the matching outcomes. Since the paper later draws policy conclusions about ridesharing fare changes, the setting β=0 is not innocuous without demonstration. The claim should be removed or replaced by a brief sensitivity analysis over β.","section":"Table 2, note c"}],"minor_comments":[{"comment":"The phrase 'stranding for the mode choice' should read 'standing for the mode choice' or 'denoting the mode choice.'","section":"Section 2.3.2"},{"comment":"The solver name is misspelled: 'Gorubi' should be 'Gurobi.'","section":"Section 3.1"},{"comment":"There is a duplicated Chinese comma in the sentence defining the fixed point; the notation could be cleaned up for readability.","section":"Section 2.3.3, Eq. (2.29)"},{"comment":"Several minor typographical errors appear: 'moda split' should be 'modal split,' and 'mass transit ODs' appears inconsistently as 'mass-transit ODs.'","section":"Section 4"},{"comment":"The phrase 'with the highest SPRL is 1541424.8g' should read 'the highest SPRL is...' for grammatical consistency.","section":"Section 4.1"},{"comment":"In Eq. (2.30), the emission factors for modes 1, 2, and 4 are used, but the text does not explain whether bus emissions are assumed to be zero because buses are electrified; this should be stated explicitly near the emission model.","section":"Section 2.4"},{"comment":"The limitations paragraph is helpful, but it could also mention that the mode-choice inconsistency identified in Major Comment 2 is a modeling limitation worth acknowledging in future work.","section":"Section 6"}],"recommendation":"major_revision","confidential_remarks":"The paper is within the scope of the journal and addresses an interesting gap, but the theoretical foundations need strengthening before publication. The fixed-point issue is the most serious concern: since the matching step is an integer program, standard DTD convergence results do not apply, and the paper's own numerical section concedes potential failure. I would advise the editor to require either a formal existence proof for the specific finite network setting or a substantial robustness analysis, plus a clear revision of the mode-choice/matching consistency. The paper is not currently ready for publication, but the core idea is salvageable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First, the good: the paper actually does what it says. It embeds a centralized BIPM ridesharing matching with cancellations into a path-based deterministic DTD assignment with five modes and two ownership groups. That specific combination is not in Wei et al. (2020) or Yao & Bekhor (2023), and the paper is clear about where it sits relative to that work. The model is described in enough detail to be re-implemented, and the numerical experiments on Sioux-Falls show convergence for a range of memory and logit parameters. The sensitivity analyses are systematic, and the policy discussion is appropriately cautious about trade-offs and equity.\n\nThe soft spots are real but not fatal. The biggest one is the equilibrium claim. The paper states the fixed-point condition (2.29) but does not prove existence or stability of a fixed point for the full process including the integer matching step. The matching output can jump discontinuously as travel times change, so the fixed point is not guaranteed. The authors are honest about this: they say convergence is demonstrated numerically and \"may fail\" on large networks or with sensitive travelers. That is an acceptable limitation for a modeling paper, but it means the policy comparisons in Sections 4–5 are conditional on the numerical steady state, not on a proven equilibrium. The stress-test note also points out that unmatched travelers are reassigned after the logit mode choice, so the final mode shares are not exactly the logit-consistent shares. That is a fair point and worth an explicit discussion; the fixed point is defined for the whole process, so it is not a contradiction, but the paper should be clearer that the \"stable state\" is a process fixed point rather than a conventional SUE.\n\nThere are also a couple of internal inconsistencies in the emissions calculation. The paper says mass transit vehicles are fully electrified with zero on-road emissions, but then includes buses (mode 4) in the emission equations (2.30)–(2.31) and only provides emission factors for cars in Table 2. Either buses should be excluded or their factors should be stated. This is a minor fix, but it undermines the emission numbers as currently presented.\n\nOne more thing: no code or data are released. For a paper whose main contribution is a model, that is a reproducibility gap. It is not disqualifying, but it would be a better paper with the Sioux-Falls setup and parameters available.\n\nOverall, this is a serious piece of work. The novelty is real, the modeling is careful, and the limitations are acknowledged. The equilibrium concern and the emission inconsistency should be addressed in revision, but they do not sink the core idea. I would send it to peer review, and I would expect it to come back in much better shape after a round of revisions. For a reader working on multimodal assignment or ridesharing policy, this is worth a close look.\n\nRecommendation: accept for peer review, with the expectation of major revision.","headline":"A genuinely new integration of ridesharing matching into a DTD multimodal model, with honest numerical convergence but a few loose ends in the equilibrium interpretation and emissions accounting.","tokens_in":18633,"tokens_out":3053,"would_cite":true,"duration_ms":26543,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["90B20","90C10","90B06"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper's integrated model of optimal ridesharing matching within day-to-day multimodal traffic assignment shows that vehicle restrictions and pricing policies have uneven, sometimes non-monotonic effects on mode split, cost, and…","keywords":["Urban mobility","Multimodality","Ridesharing","Day-to-day traffic dynamics","Social equity","Vehicle emissions","Multimodal traffic assignment","Optimal matching"],"falsifier":"Run the day-to-day dynamics on the same Sioux Falls network from many random initial flow vectors with fixed parameters (N=30, θ1=θ2=0.004); if the path flows do not converge to the same fixed point (gap metrics below $10^{-4}$ within, say, 500 days), the uniqueness and stability needed to attribute the sensitivity results to the policies is falsified.","tokens_in":17627,"feed_emoji":"🚗","tokens_out":10855,"duration_ms":89976,"temperature":0.7,"pith_summary":"This paper sets out to show that ridesharing should be modeled as the outcome of a centralized matching optimization rather than as a generic mode choice, and that doing so changes how urban policies appear to work. It embeds an optimal matching process—direct matches plus a binary integer program maximizing total in-vehicle time savings, with cancellations when drivers face excessive empty mileage—into a path-based day-to-day multimodal traffic assignment model with five modes and two traveler groups. In numerical experiments on the Sioux Falls network, the model reaches a stable state for the tested memory lengths and traveler sensitivities, and the sensitivity analyses find that vehicle ownership bans, bus fare changes, and ridesharing fare changes affect mode split, travel cost, and emissions differently depending on traveler group and road link. The paper argues these results imply that vehicle restrictions and pricing strategies involve trade-offs and that their benefits and harms are not evenly distributed across society, which matters for designing equitable and sustainable transport policy.","feed_headline":"Car restrictions cut emissions but raise trip costs","feed_subtitle":"A day-to-day model with optimal ride-matching shows policy effects differ by group and by road link.","key_machinery":"The load-bearing mechanism is the coupling of the ridesharing matching optimization with the day-to-day updating loop. The matching step is a two-stage process: direct matching for identical origin-destination pairs, followed by a binary integer program (BIPM) that selects driver-passenger pairs to maximize total in-vehicle time savings S_kp, with each driver at most one passenger and vice versa; matched rides use the shortest path between origin and destination that day. Cancellations occur when the empty pick-up distance exceeds a threshold, sending unmatched drivers and car-owning passengers back to solo driving and non-owner passengers to bus or metro. The matched and adjusted flows enter the BPR congestion function, producing experienced costs; a weighted-average learning operator with memory N and weight λ forms perceived mode and path costs, and logit models convert these into next-day mode and route shares. The steady state is the fixed point C* = C(h_p*), h_p* = Λ(C*), where perceived cost equals experienced cost and path flows are constant. This integration is what allows the model to represent interactions between matching outcomes, congestion, mode substitution, and emissions.","core_discovery":"The paper's central claim is that a path-based deterministic day-to-day traffic assignment model can be extended to include a centralized optimal ridesharing matching step, with cancellation and mode adjustment, and still reach a fixed-point steady state in which perceived and experienced costs coincide and path flows are constant. The model covers five travel modes (solo driving, ridesharing as driver, ridesharing as passenger, bus, and metro) and two traveler groups (vehicle owners and non-owners). On each day, ridesharing requests are matched first directly when origins and destinations coincide, then by a binary integer program that maximizes total in-vehicle time savings subject to one-to-one matching constraints; matched pairs whose empty pick-up distance exceeds a threshold cancel and switch modes. The resulting flows feed a BPR link cost function, and perceived costs are updated by a weighted-average learning operator, with next-day mode and route choices given by logit models. The numerical experiments on the Sioux Falls network with 24 nodes and 76 links show convergence to a stable state under different memory lengths (N=3,6,30) and logit sensitivities (θ=0.001,0.004,0.01). Sensitivity analyses then show that ownership bans lower road load and emissions but raise trip time and, at moderate restriction levels, monetary cost; bus-fare increases shift travelers toward metro, ridesharing, and solo driving; and ridesharing-fare increases produce non-monotonic shifts in modal split and PCU that differ between owners and non-owners and between mass-transit and all OD pairs.","pith_inferences":["The paper leaves implicit that the cancellation threshold d_cancel is a policy lever: lowering it would reduce successful matches and push more car owners back to solo driving, which the model would predict to raise PCU and emissions; this is a testable extension of the sensitivity analysis.","Because the matching step is a discrete optimization embedded in a continuous day-to-day process, the fixed-point condition (2.29) may admit multiple equilibria; the paper's convergence results do not rule this out, so policy comparisons should be re-examined for uniqueness.","The link-level emission map the paper reports could be combined with population exposure data to quantify health impacts rather than just total emissions, turning the stated equity concern into a measurable quantity."],"forward_implications":["Vehicle ownership restrictions on mass-transit OD pairs reduce total road load and network emissions, but increase average travel time and, at 30–50% restriction levels, raise average monetary cost before it falls at 70%.","Changing the bus fare has the smallest effect on emissions and travel time among the tested policies, while strongly affecting the bus-versus-metro split.","Ridesharing fare increases first attract non-owners to ridesharing and then, above 10 yuan/km, push them back to public transit, producing inflection points in PCU and modal split that differ between mass-transit and all OD pairs.","Because the same policy affects owners and non-owners, and mass-transit and non-mass-transit OD pairs, differently, policy evaluation that only looks at aggregate mode split can miss distributional or equity consequences."],"supporting_citations":[{"why":"Supplies the basis for the ridesharing matching formulation used in the paper, including the binary integer program that maximizes total time savings.","marker":"Agatz et al. (2011)"},{"why":"Provides the dynamic ride-sharing matching approach and evidence on congestion that the BIPM step extends.","marker":"Alisoltani et al. (2021)"},{"why":"Frames ridesharing state of the art and motivates the cancellation mechanism based on excessive pick-up distances.","marker":"Furuhata et al. (2013)"},{"why":"Supplies the day-to-day learning operator (weighted average of past experienced costs) and the convergence metrics used to define the steady state.","marker":"Yu et al. (2020)"},{"why":"Provides the fixed-point steady-state condition (2.29) and the gap metrics used to certify numerical convergence.","marker":"Liu and Geroliminis (2017)"},{"why":"The doubly dynamic multimodal network model with ridesharing that this paper extends by adding explicit optimal matching and path-based choices.","marker":"Wei et al. (2020)"},{"why":"Underwrites the fixed-point and stochastic-process framework for multi-vehicle day-to-day assignment models.","marker":"Cantarella and Fiori (2022)"},{"why":"Argues for explicitly integrating ridesharing matching into multimodal traffic models, which is the gap this paper addresses.","marker":"Yao and Bekhor (2023)"}],"fun_headline_variants":["Ride-matching model shows car bans cut emissions but raise costs","Multimodal model with ride-matching exposes policy trade-offs","Ownership cap lowers road load, but trips get pricier and slower","Fare changes ripple unevenly across traveler groups in new model","Day-to-day rideshare model reveals social equity impacts of pricing"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the day-to-day process converges to a fixed-point steady state; the matching optimization can jump discontinuously with costs, and convergence is only demonstrated numerically for a few parameter settings on the Sioux Falls network, with the authors conceding it may fail on larger networks or with highly cost-sensitive travelers.","fun_headline_variants_meta":{"raw":{"variants":["Ride-matching model shows car bans cut emissions but raise costs","Multimodal model with ride-matching exposes policy trade-offs","Ownership cap lowers road load, but trips get pricier and slower","Fare changes ripple unevenly across traveler groups in new model","Day-to-day rideshare model reveals social equity impacts of pricing"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000287,"raw_usage":{"total_tokens":1749,"prompt_tokens":1075,"completion_tokens":674,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":691,"completion_tokens_details":{"reasoning_tokens":584}},"tokens_in":691,"tokens_out":674,"duration_ms":6442,"temperature":1.0,"reasoning_tokens":584,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T14:18:36.205695+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the day-to-day dynamics on the same Sioux Falls network from many random initial flow vectors with fixed parameters (N=30, θ1=θ2=0.004); if the path flows do not converge to the same fixed point (gap metrics below $10^{-4}$ within, say, 500 days), the uniqueness and stability needed to attribute the sensitivity results to the policies is falsified.","supporting_citations":[],"review_version":1}