{"id":"0cd06791-015c-49fa-b246-f3950d896df7","arxiv_id":"2505.11522","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":9,"one_line_summary":"Closed-form delay and optimal cycle length formulas are proposed for mixed automated and human traffic, but the optimal cycle length is invalid because the model's own proof shows delay grows monotonically with cycle length.","lead":"The paper derives formulas to estimate traffic delay and choose signal cycle length at an intersection where automated and human-driven vehicles mix. The intended use is to help traffic engineers adjust signals as the share of automated vehicles rises.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The optimal-cycle-length argument is internally contradictory: Proposition 1 proves total delay is increasing in C for all C, so no interior optimum exists, while Eq. (30) is presented as a stationary-point formula and Fig. 9 reports positive optima; moreover, under the λ definition in Eq.","rationale":"The reader's verdict of REJECT is confirmed. My main concern, however, is not the reader's named weakest assumption about CAV/HDV departure behavior; it is a more internal and more decisive problem: the paper's own Proposition 1 rules out the interior optimum that Eq. (30) purports to compute, and Eq. (30) itself is negative under the paper's stated definition of λ as the green split. This is a mathematical contradiction that holds regardless of whether real CAVs depart with zero start-up lost time or whether the quadratic HDV acceleration profile is accurate. I agree partially with the reader because their rationale mentions the monotonicity/negative-value inconsistency, but their formal weakest_assumption field identifies the behavioral departure assumptions instead. A derived optimal cycle formula that is negative, and a proof that no interior optimum exists, cannot be rescued by empirical calibration or more realistic CAV behavior. The capacity derivation in Eq. (5) also appears to use steady-state probabilities that do not sum to one (the state-n self-loop is not handled correctly), which would corrupt Eq. (9) and hence Eq. (27) and Eq. (30); I flag this as a secondary issue rather than the primary one. Since the existing REJECT verdict already captures the failure of the central claim, my independent stress test does not change the recommended verdict, only strengthens the basis for it.","tokens_in":12678,"tokens_out":8324,"duration_ms":77877,"concrete_test":"Evaluate Eq. (30) using Table 1 defaults for a single approach, e.g., q=0.25 veh/s, c from Eq. (9), p=0.5, λ=0.55 as the green split, Tr=2 s, Ta=3 s. If the numerator is negative and C*<0, Eq. (30) contradicts Fig. 9. Then independently re-derive dD_total/dC from Eq. (31) with R=(1−λ)C and λ fixed as the green split, and check whether the derivative is nonnegative over C in (20, 240) s and whether it ever vanishes. If the derivative never vanishes, any formula obtained by setting it to zero cannot be the optimal cycle length; also recompute with λ reinterpreted as q/c to confirm that the two interpretations give different derivatives, exposing the equivocation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central deliverable, the closed-form optimal cycle length in Eq. (30), fails on internal grounds. First, Section 3 defines λi as the green split G/C (Eq. 29), with 0<λi<1. In Eq. (30), the numerator contains cqi/(c−qi) times [Ta(λi−λip+p−1)+2Tr(λi−λip+p−1)]. Since λi−λip+p−1=(1−p)(λi−1)<0 and cqi/(c−qi)>0 under the stated under-saturation condition, the numerator is negative while the denominator is positive; Eq. (30) therefore returns a negative C* for every parameter setting. Figure 9, however, reports positive optimal cycle lengths ranging from about 220 s down to 15 s, so the figure cannot be generated by the displayed formula. Second, Proposition 1 states that D_total is monotonically increasing in C and proves D'_total(C) ≥ 0 for all C>0. If that proof is correct, setting the derivative to zero has no solution and no interior optimum exists; the only minimum would be the smallest feasible cycle length Cmin. The paper nevertheless introduces Eq. (30) by differentiating total delay and setting the result to zero, which is logically incompatible with Proposition 1. Third, the derivative in Eq. (32) equivocates on λ: Eq. (29) uses λi=G/C, whereas the proof of Proposition 1 redefines λi=qi/c as the demand-to-capacity ratio. These cannot both be right, because red time is not (1−q/c)C unless the green split happens to equal q/c. This equivocation invalidates the derivation of both Eq. (30) and Eq. (32). None of these problems depends on the realism of the CAV/HDV departure assumptions; they are internal to the presented mathematics. The delay expression Eq. (27) may also inherit errors from the unnormalized steady-state probabilities in Eq. (5), but the optimization contradiction is already sufficient to invalidate the paper's main contribution.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a stochastic analytical framework for estimating delay and optimizing cycle length at an isolated signalized intersection with mixed connected automated vehicle (CAV) and human-driven vehicle (HDV) traffic. A discrete-time Markov chain models the distribution of CAV platoon lengths, from which an expected capacity is derived using car-following headway relations. Queuing-theoretic delay expressions are developed separately for CAV-led and HDV-led platoons, combined through the CAV penetration rate, and then differentiated with respect to the cycle length to obtain a closed-form optimal cycle length. Numerical experiments illustrate the sensitivity of average delay to arrival rate, green ratio, CAV penetration rate, and cycle length, and report an optimal cycle length that decreases with CAV penetration.","tokens_in":13211,"tokens_out":7336,"duration_ms":65655,"significance":"If correct, the framework would give traffic engineers a simple closed-form tool for signal timing in mixed traffic, which is currently lacking. The paper has the virtue of attempting to connect stochastic platoon composition, car-following headways, and queuing delay in one analytical model, and it provides explicit formulas that could be tested against simulation. However, the central deliverables—the capacity formula and especially the optimal cycle length—are undermined by internal mathematical inconsistencies that make the main claims untenable as stated.","major_comments":[{"comment":"The steady-state probabilities in Eq. (5) do not sum to 1. Summing π_i = p^i / (p^n/(1-p) + Σ_{m=0}^{n-1} p^m) over i=0..n gives (1-p^{n+1})/(1-p) / (1/(1-p)) = 1 - p^{n+1}, not 1. For the DTMC defined by Eq. (1) with the self-loop at state n, the correct steady state is π_i=(1-p)p^i for i=0,...,n-1 and π_n=p^n. This normalization error propagates directly into the capacity formula Eq. (9) and hence into every delay expression derived from it, so it must be corrected before the model can be evaluated.","section":"Eq. (5)"},{"comment":"The optimal-cycle-length formula Eq. (30) is internally contradictory with Proposition 1. Proposition 1 proves that D_total(C) is monotonically increasing for all C>0, so no interior stationary point exists and the minimum occurs at the boundary C_min. Yet Eq. (30) is presented as the result of setting the derivative to zero. Furthermore, for under-saturated approaches (0<λ_i<1, 0<p<1), the numerator of Eq. (30) contains (T_a+2T_r)(λ_i−λ_i p+p−1) = (T_a+2T_r)(1-p)(λ_i−1) < 0, while the denominator (λ_i−1)^2 > 0, so Eq. (30) returns a negative C* for every parameter combination. Figure 9 reports positive optimal cycle lengths between about 220 s and 15 s, which cannot be produced by Eq. (30). This is a load-bearing flaw in the paper's central claim.","section":"Eq. (30) and Proposition 1"},{"comment":"The derivation equivocates on the meaning of λ_i. In Eq. (29), λ_i is defined as the green split G/C, so that red time is R_i=(1−λ_i)C. In the proof of Proposition 1, the text states that λ_i = q_i/c, the demand-to-capacity ratio. The derivative in Eq. (32) uses the latter definition, but the expression R_i=(1−λ_i)C only holds when λ_i is the green split. These two definitions are not generally equal, so the derivative computation leading to both Eq. (32) and Eq. (30) is not valid. The contradiction between Proposition 1 and Eq. (30) is a direct consequence of this equivocation, and it cannot be resolved without re-deriving the optimality condition.","section":"Eq. (32) and the definition of λ_i"}],"minor_comments":[{"comment":"The manuscript contains several typos that should be fixed: 'DMTC' should be 'DTMC', 'Illustative' should be 'Illustrative', and the summation index in Eq. (36) is written as 'n' rather than a phase index.","section":"Throughout"},{"comment":"The presentation of Eq. (9) is difficult to parse because the denominator appears to combine terms in a way that does not match the standard formula c = 1/(E[τ]+L/v_free). After correcting the steady-state probabilities, the capacity expression should be re-derived and written as a single unambiguous fraction.","section":"Eq. (9)"},{"comment":"Figure 7 contains a dark blue region described as over-saturation, but the model explicitly excludes over-saturated conditions. The paper should clarify why this region is shown and whether the plotted delay values in that region are produced by the model or by an extrapolation.","section":"Figure 7"},{"comment":"The text claims that the optimal cycle length decreases with CAV penetration, but it does not state whether Figure 9 was generated using Eq. (30), the boundary minimum C_min from Eq. (34), or a numerical search. Since Eq. (30) yields negative values, the figure must come from some other procedure, and the authors should specify it explicitly.","section":"Section 4.3 and Figure 9"},{"comment":"The car-following headway relation in Eq. (7) is adopted from Chen et al. [27] without derivation. Given that this relation is a key input to the capacity and delay formulas, the paper should at least state its origin prominently and discuss conditions under which it applies to mixed CAV/HDV platoons.","section":"References"}],"recommendation":"reject","confidential_remarks":"The paper's central contribution—a closed-form optimal cycle length—is invalidated by the internal inconsistency between Proposition 1 and Eq. (30), the negative values produced by Eq. (30), and the equivocation on λ_i. The normalization error in Eq. (5) further undermines the capacity model. These issues are not local presentation problems; they affect the core derivation and cannot be fixed within the scope of a minor revision. The reader's assessment aligns with mine."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This paper is aiming at a real need: an analytical, closed-form signal timing model for mixed CAV/HDV traffic that could complement microsimulation. That goal is worthwhile, and the authors put together a plausible narrative combining a Markov-chain capacity model with Webster-style delay analysis. The capacity component extends Ghiasi et al. (2017) to multiple CAV platoon lengths in a straightforward way, and the separate treatment of CAV-led and HDV-led departure behaviors is a reasonable modeling choice. I want to give credit where it is earned: the problem is well motivated, and the structure of the derivation is clear enough to follow.\n\nUnfortunately, the central deliverable does not survive contact with the math. The stress-test note is accurate on every point. First, the steady-state probabilities in Eq. (5) do not normalize to 1, which undermines the capacity expression. Second, Proposition 1 proves that total delay is monotonically increasing in cycle length for all C, so no interior optimum exists. Yet Eq. (30) is presented as the result of differentiating and setting to zero. Third, λ is used inconsistently: in Eq. (29) it is the green split G/C, but in the proof of Proposition 1 it is redefined as the demand-to-capacity ratio q/c. These cannot both be right, and the equivalence invalidates both Eq. (30) and the derivative in Eq. (32). Even taken at face value, Eq. (30) gives a negative optimal cycle length whenever 0<λ<1 and p<1, because the numerator carries a factor (1−p)(λ−1) which is negative. Figure 9 reports positive values between 15 and 220 s, so that figure cannot have been generated by the displayed formula.\n\nThese are not minor typos; they are load-bearing contradictions in the paper's main contribution. The delay expressions in Eqs. (18)–(27) may or may not be salvageable, but as written they inherit the normalization problem from Eq. (5), and there is no simulation or field validation to fall back on. The conclusion that higher CAV penetration reduces delay is intuitive and consistent with the literature, but that is not the paper's contribution.\n\nFor a reader: this paper is a cautionary example, not a usable tool. The authors would need to re-derive the optimization, fix the normalization, and validate against microsimulation before this becomes a real contribution. My recommendation: desk reject. It does not deserve referee time in its current state.","headline":"The paper's central claim—a closed-form optimal cycle length for mixed CAV/HDV traffic—is invalid: the derivation contradicts its own monotonicity result and produces negative values, so the main contribution collapses.","tokens_in":13709,"tokens_out":3264,"would_cite":false,"duration_ms":32146,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper derives closed-form expressions for expected average delay and optimal cycle length at an isolated signalized intersection with mixed connected automated and human-driven traffic, using a Markov-chain platoon model and…","keywords":["signal timing","mixed traffic","connected automated vehicles","delay approximation","Markov chain","queuing theory","optimal cycle length","intersection capacity"],"falsifier":"Measure cumulative departures from an isolated approach over many signal cycles and compare CAV-led and HDV-led platoons. The model predicts zero start-up lost time for CAV-led platoons and an optimal cycle near 15 seconds as the CAV penetration approaches 1; if real CAV-led queues show measurable reaction delay, or if the measured optimum is far above 15 seconds, the central claim fails.","tokens_in":12515,"feed_emoji":"🚦","tokens_out":7114,"duration_ms":66896,"temperature":0.7,"pith_summary":"The paper sets out to give traffic engineers a simple analytical replacement for simulation when estimating delay at a signalized intersection where some vehicles are connected automated vehicles (CAVs) and the rest are human-driven. It models the alternating sequence of CAV and HDV platoons as a discrete Markov chain, converts the resulting steady-state platoon probabilities into an expected capacity, and then integrates queuing-theory delay formulas for the two departure behaviors. The output is a closed-form expected average delay and a closed-form optimal cycle length. This matters because signal timing is currently tuned with simulation-heavy or HDV-only formulas, and a closed form makes mixed-traffic signal planning fast and transparent.","feed_headline":"Optimal signal cycle shrinks as CAV share rises","feed_subtitle":"A closed-form Markov delay model: at full automation the best cycle is about 15 s, and green ratio beats penetration.","key_machinery":"The machinery is a discrete-time Markov chain whose state is the number of consecutive CAVs in a platoon: each following vehicle is a CAV with probability $p$ and an HDV with probability $1-p$, returning the state to 0. Its steady-state distribution, Eq. (5), feeds an expected time-gap $\\mathbb{E}[\\tau]$, Eq. (6), in which successive CAV time gaps follow the string-stability condition of Eq. (7), giving the mixed-capacity expression Eq. (9). On top of that, delay is computed as the area between the cumulative arrival curve and a piecewise departure curve, Eq. (12), whose HDV branch is a reaction-time delay $T_r$ plus a quadratic acceleration phase of length $T_a$; the closed-form integrals (18) and (24), mixed by penetration probability, yield the final delay and cycle-length formulas.","core_discovery":"The central claim is that for an isolated approach with arrival rate $\\bar{q}$, capacity $c$ derived from the Markov chain, red time $R$, green time $G$, CAV penetration $p$, HDV reaction time $T_r$, and HDV acceleration duration $T_a$, the expected average delay is given by Eq. (27), formed as the penetration-weighted combination of a CAV-led delay $D_{\\mathrm{CAV}}$ and an HDV-led delay $D_{\\mathrm{HDV}}$, and the optimal cycle length is Eq. (30), obtained by differentiating the summed approach delay with respect to cycle length. The paper further proves Proposition 1: total delay is monotonically increasing in cycle length, so the shortest safe cycle is the best cycle. Under the model's assumptions, increasing CAV penetration lowers both delay and optimal cycle length, while green ratio changes affect delay more than penetration changes.","pith_inferences":["If the constant-arrival assumption is relaxed to stochastic arrivals, extra queueing variance terms should appear in the delay expression, and the monotone-in-cycle result in Proposition 1 would likely need an oversaturation caveat.","Extending the Markov state space to include a 'CAV leader with lost communication' state would yield a testable prediction: the optimal cycle length as a function of communication reliability, interpolating between the penetration-driven cycle and the HDV-only cycle.","The model suggests a feedback loop not analyzed in the paper: if signal controllers adopt shorter cycles as CAV share grows, platoon formation itself changes, which would alter the very headway distribution the Markov chain computes.","A direct empirical test of the mechanism would compare cumulative departure curves of CAV-led versus HDV-led platoons; the predicted difference is a start-up lost time measurable from trajectory data."],"forward_implications":["At zero CAV penetration the model reproduces HDV-only delay behavior; at full penetration the optimal cycle falls to about 15 seconds and average delay to roughly 4.5 seconds per cycle in the numerical settings.","Green ratio has a larger effect on expected delay than penetration rate, so signal timing optimization remains a high-leverage policy even before CAVs are widespread.","Because total delay is monotonically increasing in cycle length, the minimum permissible cycle set by clearance and lost-time constraints is also the delay-minimizing cycle under the model.","Higher CAV penetration is more beneficial at longer cycle lengths; shorter cycles compress the delay gap between penetration levels.","Closed-form expressions allow signal retiming to be evaluated instantaneously without simulation, making the model suitable for real-time or planning applications."],"supporting_citations":[{"why":"It supplies the string-stability time-gap condition for successive CAVs and the base stochastic capacity derivation that the paper extends to mixed traffic.","marker":"(27)"},{"why":"It supplies the optimal-cycle-length methodology: differentiate total delay with respect to cycle length and set the derivative to zero.","marker":"(14)"},{"why":"It supplies the quadratic acceleration profile used to model the HDV-led departure phase.","marker":"(28)"},{"why":"It supports the assumption that CAV-led platoons depart without start-up lost time.","marker":"(29)"},{"why":"It supplies the Highway Capacity Manual delay framework and the minimum-cycle-length and lost-time constraints used in the model.","marker":"(15)"},{"why":"It supplies the Markov-chain mixed-capacity analysis that motivates modeling platoon composition by penetration rate.","marker":"(11)"}],"fun_headline_variants":["Markov delay model: more CAVs mean shorter optimum signal cycle","CAV share drops optimal cycle to 15 s, green ratio matters more","Markov proof: shortest safe cycle is best for mixed signals","Closed-form signal timing: CAV penetration trims optimal cycle"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The formulas stand or fall on the departure-behavior assumption: CAV-led platoons discharge at full saturation flow the instant green begins, with zero reaction or start-up lost time, while HDV-led platoons lose a fixed reaction time followed by a quadratic acceleration phase; if either behavior differs in the field, the closed-form delay and the optimal cycle length both change.","fun_headline_variants_meta":{"raw":{"variants":["Markov delay model: more CAVs mean shorter optimum signal cycle","CAV share drops optimal cycle to 15 s, green ratio matters more","Markov proof: shortest safe cycle is best for mixed signals","Closed-form signal timing: CAV penetration trims optimal cycle"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000568,"raw_usage":{"total_tokens":2667,"prompt_tokens":904,"completion_tokens":1763,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":520,"completion_tokens_details":{"reasoning_tokens":1688}},"tokens_in":520,"tokens_out":1763,"duration_ms":13508,"temperature":1.0,"reasoning_tokens":1688,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T22:42:38.311356+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure cumulative departures from an isolated approach over many signal cycles and compare CAV-led and HDV-led platoons. The model predicts zero start-up lost time for CAV-led platoons and an optimal cycle near 15 seconds as the CAV penetration approaches 1; if real CAV-led queues show measurable reaction delay, or if the measured optimum is far above 15 seconds, the central claim fails.","supporting_citations":[],"review_version":1}