{"id":"559ba990-0d49-4b5a-880e-aa92bc543f70","arxiv_id":"2505.14457","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"New sum-of-squares conditions allow globally stabilizing controllers for structured polynomial systems, including a data-driven version that is robust to bounded noise and prior parameter knowledge.","lead":"The authors give new ways to design stabilizing feedback controllers for a class of polynomial systems, both when the equations are known and when only noisy data are available. The method lets Lyapunov functions that stay bounded at infinity be used for global stability, and it can use prior knowledge about system parameters to make data-driven design less conservative.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 2 is sound under the stated noise model; the load-bearing condition is that Φ11 is known a priori and the true noise energy obeys it, which the examples only verify post hoc.","rationale":"A stress-test pass should distinguish between a theorem being internally incorrect and a theorem being conditional on an assumption that is hard to validate. Theorems 1 and 2 are internally coherent: I re-derived the key Schur-complement and S-lemma steps in the proofs and found no algebraic gap. Lemma 1 is a standard sufficient condition for global asymptotic stability without radial unboundedness, and condition (b) of Theorem 1 supplies exactly the uniform decay outside a compact set that Lemma 1 needs. In Theorem 2, the matrix (30) is affine in the decision variables P and L for fixed ε2, ε3, so the SOS formulation is a valid LMI relaxation; the proof's passage from (30) to (29) via Schur complements is consistent with the stated specialized S-lemma. The genuine load-bearing assumption is that Φ11 is known and that the true noise satisfies the energy bound (20). If this fails, the true system lies outside Σ and no guarantee is claimed; the paper is explicit about this, but the numerical section does not demonstrate how Φ11 would be chosen in practice. The reader's conditional verdict is appropriate, and the concrete validation test above would either confirm the examples respect the assumption or reveal a gap in the reported numerical verification.","tokens_in":16961,"tokens_out":43464,"duration_ms":400444,"concrete_test":"For the reported Example 2 data, compute the residual W = dotX - [A1^s;A2^s]F - [0;B2^s]GU using the stated true parameter matrices, and verify that ||W||_F^2 ≤ Φ11 = ω²T. Then repeat the synthesis with the same data but with Φ11 reduced by 20%: if the SOS program remains feasible even though the true system is excluded from Σ, simulate the closed-loop system to confirm that the formal guarantee disappears exactly at the boundary where the true noise energy exceeds the assumed bound. This check isolates whether the numerical verification is consistent with the assumption the theorem requires.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central data-driven guarantee is conditional on the true noise W satisfying the quadratic bound (19)/(20) with a known Φ11. If the actual disturbance energy exceeds Φ11, the true parameter vector v_s is not in the compatibility set Σ, the S-lemma reasoning in Theorem 2 never applies, and the synthesized controller K(x)=L(x)P^{-1}(x1)Z(x) carries no formal stability guarantee. This is not an internal inconsistency, but it is the point where the claim that the data are informative for stabilization is least secure. In the numerical examples, Φ11 is set to ω²T after the noise samples are generated, so the experiments show feasibility of the SOS conditions for the realized noise, not robustness to a misspecified prior bound. A user must know Φ11 before seeing data; the paper does not provide any data-driven way to validate or adapt this bound, and no margin is built in. The model-based Theorem 1 itself appears mathematically sound: the Schur-complement step in its proof is valid, and Lemma 1 correctly replaces radial unboundedness by the uniform decay condition (12). The proof of Theorem 2 also checks out: the specialized S-lemma is applicable because N22<0 and N|N22≥0, and the algebra from (30) to (29) is consistent. Thus the weakest link is not the theorem structure but the a priori noise energy assumption, exactly as the reader identified.","agreement_with_reader":"agree"},"referee_report":null,"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a solid paper that deserves a serious referee. The model-based part is a modest but clean extension of Prajna et al. [6], replacing radial unboundedness with the uniform decay condition from Lemma 1. The data-driven part is the real novelty: a way to synthesize globally stabilizing controllers for all systems compatible with noisy data, with a handle for prior knowledge. The proofs are internally consistent and the examples illustrate the method.\n\nThe reader raised two technical concerns that don't survive close reading. First, the alleged missing square on epsilon_2 in the Schur complement of Theorem 1: the derivation gives M >= epsilon_2 * epsilon_3^{-1} P P, because the square on epsilon_2 cancels against the (2,2) block. No typo. Second, condition (30) is not bilinear in P and L. R(x,y) is affine in those unknowns, so the whole SOS constraint is affine in them; this is a standard SDP, not a bilinear problem.\n\nThe real soft spot is the a priori noise energy bound in (19)/(20). The guarantee covers every system in Sigma, but Sigma itself is defined by a bound Phi_11 that the user must know before seeing data. In the examples, Phi_11 = omega^2 T is set after generating the noise, so the numerics demonstrate feasibility for the realized noise, not robustness to an underestimated bound. This is a standard assumption in the informativity literature, but the paper could be clearer that the user must commit to Phi_11 in advance.\n\nTwo minor caveats. Lemma 2 is from the authors' own companion paper [23], not yet published; the referee should verify that lemma independently, since Theorem 2 leans on it. And the numerical section gives P and L but no code or certificates, so the claims are reproducible only in principle. For a theory paper that's a minor complaint.\n\nOverall this is honest, incremental work. The contribution is clearly scoped, the math checks out as far as I can see, and the data-driven setup is genuinely useful. I'd send it to peer review with a request to check [23] and to state the noise-bound assumption more prominently.","headline":"A competent, incremental paper: the model-based relaxation is clean, the data-driven extension is real, and the reader's two technical objections don't hold up on close reading—the actual caveat is the a priori noise bound.","tokens_in":17787,"tokens_out":4148,"would_cite":true,"duration_ms":42306,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-07T15:35:05.342035+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}