{"id":"73dccb65-615b-423d-b8df-c3642fdeda00","arxiv_id":"2606.22223","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"An SDP-based algorithm estimates both control gains and minimum dwell times online for switched LQR systems with unknown dynamics, achieving O(|M|^{1/4} n_s^{3/4} + n_m) expected regret while keeping state norms bounded.","lead":"The paper designs an online controller for switched linear-quadratic systems whose mode dynamics are unknown and whose mode sequence is revealed one step at a time, guaranteeing bounded state growth and bounded regret versus a full-information benchmark. It is worth reading because it is among the first to convert data-driven dwell-time estimation into a safety-plus-regret guarantee for switched systems.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Covariance excitation lower bound (Lemma 17) is the linchpin of the regret theorem and is only sketched for ζ=1/4; its proof must be verified.","rationale":"The reader's weakest assumption identifies the same linchpin. I agree. The paper's other listed issues are real but less decisive: the benchmark optimality claim is not proved, but the formal regret theorem is stated relative to the DARE-plus-dwell-time benchmark, so a suboptimal benchmark would weaken the interpretation, not the theorem; unquantified constants do not affect the rate; the D1 inequality appears to contain a typo but is likely repairable. Lemma 17 is the only place where the proof is explicitly deferred, and the sketch leaves a technical ambiguity about the time-varying excitation coefficient. Because Theorem 13, Lemma 20, and Theorem 15 all inherit from it, the CONDITIONAL verdict is appropriate. The concern is not a demonstrated falsification; it is an unverified step that must be checked before the central claim can be accepted.","tokens_in":33446,"tokens_out":15286,"duration_ms":127933,"concrete_test":"Independently re-derive Lemma 17 for ζ=1/4 from Lemmas 9–12 of Chekan & Langbort (2024a), keeping \\bar{p}_t time-varying, and check whether the lower bound (68) follows with \\bar{p}_t or only with a smaller constant. If the proof needs \\bar{p}_t to be nondecreasing and the cumulative-sum lower bound uses \\bar{p}_1, verify that the resulting bound still satisfies the stability condition (75) with the chosen \\bar{c}_i. A numerical Monte Carlo simulation of the closed loop with the proposed Γ_t for ζ=1/4, comparing empirical λ_min(Σz_kz_k^T) to (68), would provide a useful corroboration.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central regret bound (Theorem 15) depends on the covariance excitation bound of Lemma 17: Theorem 10 uses it to guarantee (κ,γ)-strong stability, Lemma 20 uses it to bound µ∥V^{-1}∥, and Lemmas 23–26 propagate this into the regret rate. The paper states the proof for general ζ∈(0,1/2) is 'given in Chekan & Langbort (2024a) for the special case ζ=1/2. We sketch the modifications...' The sketch is not a proof: it asserts E[z_t z_t^T|F_{t-1}] ⪰ (σ_ω² \\bar{p}_t/(2(t+\\bar{c})^ζ))I and then invokes Lemmas 10–12 of the prior work, but does not show that the martingale/covering argument remains valid when \\bar{p}_t is time-varying. Moreover, (68) writes \\bar{p} without subscript while the intermediate steps use \\bar{p}_t; if \\bar{p}_t grows with t, the factor cannot be moved out of the sum in the lower bound, and if \\bar{p} is meant as a constant, its value and the sufficiency of condition (75) are not established. This is the single point of failure for the dwell-time error bound and the regret theorem.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies switched LQR systems with unknown mode dynamics, where the next mode is revealed online only at switching times. It introduces an (α,β)-controllability condition that bounds expected state-norm growth across switches, proposes a known-model benchmark based on DARE feedback plus minimum dwell times, and then presents an SDP-based online algorithm (SAS) that jointly learns feedback gains and dwell times from data. The main claim is an expected regret bound of O(|M|^{1/4} n_s^{3/4} + n_m) relative to the DARE-plus-dwell-time benchmark, where n_s is the number of switches and n_m is the number of malignant switches. The technical machinery follows the authors' earlier ARSLO work, adapted to switching and to exploration noise exponents ζ < 1/2.","tokens_in":33760,"tokens_out":15160,"duration_ms":142333,"significance":"If the result is correct, it is a meaningful step: it extends anytime regret guarantees from fixed LQR to switched systems with externally revealed modes, while enforcing a state-norm safety constraint through learned dwell times. The regret decomposition, the benign/malignant switch classification, and the Lyapunov-based dwell-time sufficient condition are attractive ideas. The paper also gives explicit tuning choices for the exploration noise and regularization. However, the central regret theorem currently rests on an unproved covariance excitation lemma and on a dwell-time formula that appears to contain a sign error. These are load-bearing, so the significance is conditional on substantial revision.","major_comments":[{"comment":"The dwell-time formula contains a sign error. From the paper's own inequality (65), the condition \\bar X \\bar ρ (1 - \\bar η)^{τ} ≤ \\bar α is equivalent to τ ≥ [ln \\bar α - ln \\bar ρ - ln \\bar X] / ln(1 - \\bar η), since ln(1 - \\bar η) < 0. The manuscript instead states τ ≥ [ - ln \\bar ρ + ln \\bar X - ln \\bar α ] / ln(1 - \\bar η). Because \\bar α < 1, these are not equivalent; for \\bar ρ = \\bar X = 1, \\bar α = 0.9, \\bar η = 0.5, the published formula gives a negative lower bound while the correct bound is positive. Thus in typical cases the max{1, ·} in (25) and (42) collapses the dwell time to one step, destroying the safety guarantee in Lemma 12. The same sign problem appears in Lemma 18: the right-hand side of (94) is negative for \\bar κ_* > 1 because the denominator is ln(1 - 1/\\bar κ_*^2) rather than its negative; hence the bound τ_{k,k+1} ≤ τ* and the conversion t ≤ n_s τ* are not est","section":"Section 3, Theorem 6 / eqs. (13), (25), (42), Lemma 18"},{"comment":"The covariance excitation lower bound is the linchpin of the stability and regret analyses: Theorem 10 uses it to guarantee strong stability, Lemma 20 uses it to bound μ ||V^{-1}||, and Lemmas 23-26 propagate this into Theorem 15. For general ζ ∈ (0,1/2), the proof is not supplied; the text says the detailed proof is in Chekan & Langbort (2024a) for ζ = 1/2 and only sketches modifications. The sketch does not justify the martingale/covering argument when \\bar p_t is time-varying: (83) defines \\bar p_t as a log-growing function of n_i(t), while (68) writes \\bar p without a subscript and the intermediate steps use \\bar p_t. If \\bar p_t varies, the factor cannot be moved out of the time sum in (70)-(71) without additional argument; if \\bar p is intended as a constant, its value and the sufficiency of condition (75) are not established. Because every downstream dwell-time error and regret bo","section":"Appendix B.2, Lemma 17 / eqs. (67)-(71)"},{"comment":"The abstract asserts that, in the known-model setting, the strategy that minimizes the average expected cost consists of applying the DARE feedback gain with a suitably chosen dwell time. Section 3, however, only introduces the DARE policy as 'a natural and appealing candidate' and does not solve Problem 1 or prove optimality. The regret bound is therefore relative to a heuristic benchmark, not to an optimal known-model policy. The theorem statements themselves are stated relative to the DARE-plus-dwell-time benchmark, so this may be fixable by restating the claims, but the current framing overstates the result and should be corrected in the abstract and introduction.","section":"Abstract and Section 3"}],"minor_comments":[{"comment":"Typo: 'Discrete Algebric Riccati Equation' should be 'Algebraic'.","section":"Section 3"},{"comment":"The appendix heading 'Expected sate norm bound' should read 'state'.","section":"Appendix B.4"},{"comment":"Typo: 'in teh later one' should be 'in the latter one'.","section":"Proof of Theorem 13"},{"comment":"The proof lists two items labeled '(2)' after '(1)'; the second should be '(3)'.","section":"Lemma 22 proof"},{"comment":"Equation (68) uses \\bar p without a subscript while the proof and (83) use \\bar p_t. Even apart from the substantive gap, the notation should be made consistent.","section":"Lemma 17 notation"}],"recommendation":"major_revision","confidential_remarks":"The manuscript relies heavily on the authors' own previous arXiv work (Chekan & Langbort 2024a), especially for the covariance excitation lemma. This is not by itself disqualifying, but the missing ζ < 1/2 proof is the main technical risk. The sign errors in the dwell-time formulas appear to be genuine and must be corrected before the safety and regret claims can be accepted. I see no evidence of deliberate concealment; the sources are cited transparently. The paper may be publishable after a careful revision that supplies the missing proof and restates the benchmark claim accurately."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"One or two things to know up front: this is, to my knowledge, the first regret analysis for switched LQR in the externally forced, online-revealed-mode setting with unknown dynamics and learned minimum dwell times. The headline rate O(|M|^{1/4} n_s^{3/4} + n_m) is genuinely interesting if the proof holds. But the proof chain rests on Lemma 17, a covariance excitation lower bound that the paper only sketches for the ζ=1/4 regime it actually needs. The sketch is not a complete proof.\n\nWhat is genuinely new and good: the problem setup itself is a real step beyond the bandit-based switching work of Li et al. (2023) — only the next mode is revealed, the mode dynamics are unknown, and safety is encoded as (α,β)-controllability of the state norm. The benign/malignant switch dichotomy in Definition 14 is a natural way to isolate the costly cases in the regret decomposition. The known-model dwell-time sufficient condition in Theorem 6 is cleanly derived from a Lyapunov recursion, and the regret decomposition in Appendix B.6 is careful. The paper is also honest about what it defers: it states plainly that Theorem 10 follows prior work and that Lemma 17 is “given in Chekan & Langbort (2024a)” for ζ=1/2.\n\nThe soft spots are real and I agree with the reader’s conditional verdict. The most serious is Lemma 17. The stress-test note is correct: equation (68) writes ̅p without a subscript while the derivation uses ̅p_t; if ̅p_t grows with t, the factor cannot be moved out of the sum, and the covering argument from the prior paper does not automatically transfer. This is not a cosmetic gap — Lemma 20 and the regret bounds in Lemmas 23–26 all inherit it. Second, the abstract overstates the benchmark: it says the DARE-plus-dwell-time policy “minimizes” the average expected cost, but Section 3 only calls it “a natural and appealing candidate.” The regret is still well-defined relative to that policy, but the optimality claim should be removed or proved. Third, the constants in Theorem 13 are never quantified, and the D1 bound in Appendix B.5 has a sloppy inequality that mixes λ and log λ and does not obviously go in the claimed direction. Fourth, there are no numerical experiments, so the behavior of the constants and the ζ trade-off is unchecked.\n\nWho is this for: learning-based control people working on switched systems. It deserves a serious referee — I would send it out rather than desk-reject — but with the clear request that the authors supply a complete proof of Lemma 17 for ζ<1/2 and correct the benchmark claim. If the excitation lemma can be proven, this is a solid contribution; as it stands, the main theorem is conditional on that proof.","headline":"A serious switched-LQR learning paper whose core regret bound hinges on a covariance excitation lemma that is only sketched for the regime used; worth refereeing seriously, but the main theorem is conditional until that proof is supplied.","tokens_in":34351,"tokens_out":3747,"would_cite":true,"duration_ms":40759,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93C30","90C22","93B52","93E35"],"pacs":[],"model":"deepseek-v4-flash","headline":"Online learning can match the known-dynamics benchmark in switched LQR, with expected regret O(|M|^{1/4} n_s^{3/4} + n_m), while keeping expected state-norm growth bounded at every switch.","keywords":["switched linear systems","linear quadratic regulator","regret bounds","online learning","dwell time","(α,β)-controllability","semidefinite programming","confidence ellipsoids"],"falsifier":"Run the proposed switching algorithm on switched LQR instances with ζ = 1/4 and monitor the minimum eigenvalue of the accumulated state-action covariance in each mode. Look for any epoch of length at least 400(n+m+log(1/δ)) where λ_min falls below (σ_ω^2 p̄/80)(t+c̄)^(1−ζ) − σ_ω^2 c̄/80; a single violation, or a counterexample to the claimed excitation bound for any ζ < 1/2, would invalidate the dwell-time error theorem and the regret bound.","tokens_in":33207,"feed_emoji":"🎛️","tokens_out":5155,"duration_ms":50357,"temperature":0.7,"pith_summary":"This paper asks whether a switched linear-quadratic regulator can be run safely when each mode's dynamics are unknown and the next mode is revealed only at the moment of switching. The authors first characterize a known-dynamics benchmark: apply the Riccati-optimal gain in each mode and stay just long enough to keep the expected state norm from growing past a user-set contraction ratio. They then give an online algorithm that learns both the feedback gain and the minimum dwell time from noisy state measurements, using semidefinite relaxations around confidence ellipsoids. The main claim is that this learner's expected extra cost over the benchmark is O(|M|^{1/4} n_s^{3/4} + n_m), where |M| is the number of modes, n_s is the number of switches, and n_m counts 'malignant' switches; the same algorithm keeps expected state-norm growth bounded at every switch. If true, online learning can replace full parameter knowledge in switched LQR without sacrificing either safety or near-optimal cost.","feed_headline":"Safe switching learned online: O(|M|^1/4 n_s^3/4) regret","feed_subtitle":"Algorithm learns feedback gains and minimum dwell times from state data while keeping state growth bounded.","key_machinery":"The central object is the minimum mode-dependent dwell-time identity τ = max{1, (−ln ρ + ln X − ln α)/ln(1 − η)}, evaluated on eigenvalue ratios of the Lyapunov/Riccati matrices of successive modes. In the learned setting, those matrices are replaced by solutions of relaxed primal–dual SDPs formed from confidence ellipsoids around the unknown dynamics. This single formula carries both safety (the (α,β)-controllability state-norm bound at every switch) and regret (errors in τ translate directly into excess cost).","core_discovery":"In the known-dynamics case, the paper identifies a sufficient condition for safe switching: dwell at least τ = max{1, (−ln ρ + ln X − ln α)/ln(1 − η)}, where ρ, X, and η are eigenvalue ratios of the Lyapunov/Riccati solutions of the two consecutive modes. In the unknown-dynamics case, the same formula is evaluated using optimal solutions of a relaxed pair of primal–dual semidefinite programs built from confidence ellipsoids around the unknown parameters, plus injected exploration noise whose power decays as (t+c)^(−ζ). The paper proves a high-probability upper bound on the error between the estimated and true dwell times, and from that bound derives the regret guarantee of O(|M|^{1/4} n_s^{3","pith_inferences":["The safety guarantee is stated for expected state-norm growth at switching instants; a natural testable extension is to tighten it to high-probability state bounds, which would matter in safety-critical deployments.","Because the regret bound is tuned by a single exploration-noise exponent ζ (optimal value 1/4), the framework suggests a directly testable empirical question: does the observed regret-minimizing noise schedule match ζ = 1/4 on synthetic switched LQR instances?","The same confidence-set machinery would plausibly extend to unknown cost parameters or partially observed modes, since the identification step targets the dynamics parameters, not the costs.","The healthcare-motivated setting suggests a concrete application: using observed patient responses to choose both the duration and the dosage of the next therapy, with the (α,β) condition acting as a safety constraint on physiological state excursions."],"forward_implications":["Learned dwell times can replace fixed or conservative ones: the algorithm computes the minimum safe stay in each mode from data, so the system can switch faster as confidence grows.","The regret scales sublinearly in the number of switches and only as the fourth root of the number of modes, so the method remains viable as the mode set grows.","Malignant switches, where the Riccati cost matrix increases, appear only additively in the regret, while benign switches contribute only through the sublinear term.","Under the paper's assumptions, the DARE-plus-dwell-time benchmark can be replaced by a fully online procedure with no asymptotic loss in cost."],"fun_headline_variants":["Learn safe switching times in LQR with regret guarantee","Unknown-mode LQR: safe switching with provable regret bound","Bounded state, low cost: LQR switching without known dynamics","Switching LQR online: O(|M|^{1/4} n_s^{3/4}) regret, safe state"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The load-bearing premise is the covariance excitation lower bound of Lemma 17: the injected exploration noise must make the minimum eigenvalue of the accumulated state-action covariance grow at least like (σ_ω^2 p̄/80)(t+c̄)^(1−ζ) for ζ = 1/4 at every epoch, and the authors prove the general-ζ case only by a sketch, with the detailed proof given for a special case—so if that excitation fails, the dwell-time error bound and the O(|M|^{1/4} n_s^{3/4}) regret bound collapse.","fun_headline_variants_meta":{"raw":{"variants":["Learn safe switching times in LQR with regret guarantee","Unknown-mode LQR: safe switching with provable regret bound","Bounded state, low cost: LQR switching without known dynamics","Switching LQR online: O(|M|^{1/4} n_s^{3/4}) regret, safe state"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000331,"raw_usage":{"total_tokens":1725,"prompt_tokens":837,"completion_tokens":888,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":581,"completion_tokens_details":{"reasoning_tokens":805}},"tokens_in":581,"tokens_out":888,"duration_ms":8181,"temperature":1.0,"reasoning_tokens":805,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T10:34:04.451258+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the proposed switching algorithm on switched LQR instances with ζ = 1/4 and monitor the minimum eigenvalue of the accumulated state-action covariance in each mode. Look for any epoch of length at least 400(n+m+log(1/δ)) where λ_min falls below (σ_ω^2 p̄/80)(t+c̄)^(1−ζ) − σ_ω^2 c̄/80; a single violation, or a counterexample to the claimed excitation bound for any ζ < 1/2, would invalidate the dwell-time error theorem and the regret bound.","supporting_citations":[],"review_version":2}