{"id":"6398caa8-ea95-4abb-8351-000fc62bb374","arxiv_id":"1908.00934","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For affine single-input nonlinear systems, the paper extends prior sampled-data stabilization results by weakening the required Lie algebraic conditions and adding bounded feedback guarantees.","lead":"This paper proves new sufficient Lie algebraic conditions under which a nonlinear control system can be stabilized by sampled-data feedback, generalizing the Artstein-Sontag theorem. The generalization covers systems with nonzero drift and gives conditions for bounded feedback, shown on a worked example.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of Prop. 2 rests on the unproved CBH polynomial properties (3.9c) and I–III; for n=3 the asserted nonzero/independent behavior of Π_{3,1} is not forced by the hypotheses, so the induction lacks a key premise.","rationale":"The paper's intended contribution is a Lie-algebraic sufficient condition for sampled-data stabilization. The main theorems are clearly stated, and the overall strategy—two-stage pulse control plus CBH expansion—is coherent. I spot-checked the n=1 and n=2 formulas and the general structure of (3.9c); these are consistent with a direct expansion. The weakest point is exactly the unproved polynomial machinery I–III in Section III, which the Reader also identified. My independent computation sharpens this from 'unproved' to 'apparently false as stated': for n=3 the polynomial Π_{3,1} is proportional to a scalar that the hypotheses do not prevent from vanishing, so the stated nonzero/independence property cannot hold for every fixed x. This does not by itself disprove Proposition 2, because a repaired induction might still go through, but it means the submitted proof is incomplete at a load-bearing step. The illustrative example also contains bracket-sign computations that are left largely unchecked, reinforcing the need for a written or machine-verified derivation of (3.9c). I therefore keep the Reader's conditional verdict: acceptance should require a complete, independent derivation of the CBH expansion and the three polynomial properties, with special attention to the cases where Π_{n,i} vanishes.","tokens_in":16017,"tokens_out":41978,"duration_ms":392309,"concrete_test":"Independently re-derive (3.9c) for n=3 and n=4 from the Campbell–Baker–Hausdorff expansion of e^{ρtX}e^{tY} with X=f+u1g and Y=f−ρu1g. For n=3, i=1, verify whether Π_{3,1}(ρ;x)=−ρ(ρ+1)^2(3f[f,g]V+2[[f,g],f]V)(x) can vanish at a point satisfying (2.4) with N=2; construct such an example with a nilpotent pair f,g and a positive definite V whose relevant mixed derivatives vanish. If Π_{3,1} is zero there while λ_{3,1} is not forced zero, the induction's use of (3.10b) fails as written. If instead an independent symbolic derivation proves the degree-n and linear-independence claims, the concern is dismissed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central induction in Section III needs, for every fixed x, the polynomial Π_{n,i}(ρ;x) to be nonzero of degree n and linearly independent from q(ρ)=ρ^{n-1}(ρ+1). These properties are asserted in (3.10a)–(3.10b) without derivation, and they are not consequences of the stated hypotheses. A direct CBH computation for n=3, i=1 gives the u1 coefficient of m^(3)(0) as −3ρ(ρ+1)^2 f[f,g]V − ρ(ρ+1)(ρ+2)[[f,g],f]V. Splitting off the q·λ_{3,1} term as in (3.9c) yields Π_{3,1}(ρ;x)=−ρ(ρ+1)^2(3f[f,g]V + 2[[f,g],f]V)(x). The hypotheses (2.4) with N=2 require fV=0, f^2V=0 and [f,g]V=0, but they do not require this scalar to be nonvanishing; indeed f[f,g]V and [[f,g],f]V can both vanish at a point satisfying those equalities. At such a point Π_{3,1} is the zero polynomial, so (3.10a) is false and {Π_{3,1}, q} is linearly dependent, directly contradicting (3.10b). The proof then invokes (3.10b) to conclude, when the combined coefficient vanishes, that both Π=0 and λ=0; without genuine independence this dichotomy is unjustified. Since the whole induction for Properties P1/P2(i)–P2(iii) runs through these polynomial steps, this is a load-bearing gap rather than a merely cosmetic omission.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper extends sampled-data feedback stabilization results for affine single-input nonlinear systems x' = f(x)+u g(x) with nonzero drift term. Under the existence of a smooth, proper, positive definite V and a hierarchy of Lie-algebraic conditions (2.3)-(2.9), Proposition 2 asserts SDF-SGAS, and Proposition 3 adds conditions (2.10)-(2.11) to obtain BSDF-SGAS. The proof follows [24] by writing the composition of the flows X_{\\rho t} \\circ Y_t and expanding the derivatives of V along this composite via the Campbell-Baker-Hausdorff formula, reducing the problem to finding constants u1,u2 and rho that make the first N derivatives of m(t)=V(R(t)) vanish and the (N+1)-st negative. An illustrative planar example with V(x,y)=W(x)+y^2 is analyzed.","tokens_in":16286,"tokens_out":10411,"duration_ms":107784,"significance":"If correct, the result is a genuine generalization of [24, Proposition 3] and of the Artstein-Sontag theorem in the sampled-data setting: it replaces conditions involving the full Lie algebra Lie{f,g} with conditions on a smaller subalgebra L{f,g} and relaxes (2.7) by allowing odd j rather than j=N. The proposed control construction is explicit and the bounded-feedback variant in Proposition 3 is a useful addition. However, the central proof is not fully supported: the polynomial properties I-III, which drive the inductive case analysis, are asserted without derivation and are not consequences of the stated hypotheses. The paper has no machine-checked proofs or code; the main value is the statement and control construction, whose correctness remains conditional on filling this gap.","major_comments":[{"comment":"Properties I-III of the polynomials Pi_{n,i} are asserted without proof and are in fact false as stated. For n=3, i=1, a direct CBH computation gives Pi_{3,1}(\\rho;x)=-\\rho(\\rho+1)^2(3f[f,g]V+2[[f,g],f]V)(x). Under (2.4) with N=2, the terms fV(x), f^2V(x), and [f,g]V(x) vanish, but 3f[f,g]V(x)+2[[f,g],f]V(x) is not required to be nonzero; at a point where it vanishes, Pi_{3,1} is the zero polynomial, contradicting (3.10a). Likewise, (3.10b) cannot hold for a zero polynomial. Since the inductive proof in Cases 2 and 3 repeatedly invokes (3.10a,b), the proof of Proposition 2 lacks a key premise.","section":"Section III, Eqs. (3.9c)-(3.11)"},{"comment":"The case analysis selects u1 based on the coefficient Pi_{N+1,i}(\\rho;x)+\\rho^N(\\rho+1)(\\lambda_{N+1,i}V)(x) and uses (3.10b) to justify the dichotomy 'either this coefficient is nonzero for some rho, or Pi=0 and lambda=0'. Because (3.10b) is not established, the exclusion of the intermediate possibility 'Pi is a nonzero multiple of q' is not justified. At the very least, the proof must either prove the polynomial properties for the actual Pi_{n,i} or replace the induction with a direct computation of the relevant coefficients.","section":"Section III, after Eq. (3.15), Cases 2 and 3"},{"comment":"The CBH expansion (3.9c) is not derived in the paper; the text refers to the proof of [24, Proposition 3], but that reference does not contain the family Pi_{n,i} or properties I-III. Property III, namely membership of Pi_{n,i} in the span of iterated derivatives of V with total order n and g-order i, is essential for the derivations of (3.12), (3.16), (3.22a), and (3.27a). Without a proof of (3.9c) and (3.11), the bridge between the Lie-algebra assumptions and the polynomial inequality m^{(N+1)}(0)<0 is unsupported. The authors should supply the expansion and a proof of the polynomial properties, or state them as explicit additional assumptions.","section":"Section III, Eq. (3.9c) and Eq. (3.11)"}],"minor_comments":[{"comment":"The verification of the hypotheses for the set E5 is ended with 'Details are left to the reader'; since this is the most delicate case of the illustrative example, the proof should be written out or moved to an appendix.","section":"Section IV, Case 5"},{"comment":"The definition of the control u(.,x) is given on [0,t] and (t,t+rho t] and then extended to [0,epsilon]; the sentence 'for every sufficiently small sigma=sigma(x)>0 and epsilon in (0,sigma]' is grammatically ambiguous about whether sigma is chosen after x and whether epsilon depends on sigma.","section":"Section III, Eqs. (3.29)-(3.30)"},{"comment":"The phrase 'Property P2(ii) is strengthened by assuming that is fulfilled with j = N' should read 'assuming that it is fulfilled' for clarity.","section":"Proposition 3, bullet 2"},{"comment":"The notation L{f,g} and Lie{f,g} is used with very similar typography; in printed form this distinction may be hard to see, and a display of the definition (2.1)-(2.2) with explicit examples would be helpful.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The paper's central theorem may well be true, and the example suggests that the intended construction works, but the proof as written is not reliable: the asserted polynomial properties are false in the form stated for n=3. I recommend major revision, not rejection, because the gap is localized to the unproved CBH/polynomial machinery and might be repaired either by proving a corrected version of properties I-III under additional hypotheses or by replacing the inductive argument with a direct CBH coefficient computation. The authors should be shown the explicit n=3 expression for Pi_{3,1} so they can address it concretely."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nTwo things you should know about arXiv:1908.00934. First, the paper is a real technical extension of Tsinias's sampled-data stabilization work. It introduces a Lie subalgebra L{f,g} generated by f and bracketed monomials, relaxes the strong condition (λ_{N+1,N}V)≠0 of [24] to a weaker odd-index condition (2.7), and adds boundedness conditions in Prop. 3 for BSDF-SGAS. That is a legitimate advance.\n\nSecond, the proof of Prop. 2 is not rigorous as written. The authors state in (3.10a,b)–(3.11) three properties of polynomials Π_{n,i}: nonzero, degree n, linear independence from q(ρ)=ρ^{n-1}(ρ+1), and membership in a span of bracket derivatives. No derivation is given. A direct CBH computation for n=3, i=1 yields Π_{3,1}(ρ;x)=−ρ(ρ+1)^2(3 f[f,g]V + 2[[f,g],f]V)(x). Under hypotheses (2.4) with N=2, fV, f^2V and [f,g]V vanish, but the scalar factor need not be nonzero; at points where it vanishes, Π_{3,1} is the zero polynomial. This contradicts the 'nonzero' claim, and the asserted linear independence from q is false for a zero polynomial. The induction in Section III repeatedly uses that independence to split into subcases and to ensure that a nonzero λ term yields a usable coefficient. Without a correct property, the central argument has no foundation. This is a load-bearing gap.\n\nWhat is good: the subalgebra is a genuine innovation, the conditions are plainly weaker than [24] (Remark 1 is accurate), and the example is worked through in detail for several cases. The delegation of CBH to [24] is reasonable if the formula is correct, but the new Π terms need a complete proof. Case 5 of the example leaves 'details to the reader' at the delicate point, which is unsatisfying.\n\nThe framework is plausible and likely salvageable, but as it stands the main theorem is not proven. I would send this to a serious referee because the problem is meaningful and the gap is concrete, but the referee should require a corrected proof of the polynomial properties or a strengthening of hypotheses. I would not cite it in its current form.\n\nBest,","headline":"The main generalization is real, but the proof of Prop. 2 rests on polynomial properties that are asserted without proof and are false as stated; the central theorem is not established.","tokens_in":16922,"tokens_out":17635,"would_cite":false,"duration_ms":164149,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93D15","93D20","93C57"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that affine single-input nonlinear systems with nonzero drift are sampled-data semiglobally asymptotically stabilizable when a generalized control Lyapunov function satisfies certain Lie-bracket conditions, and that…","keywords":["sampled-data feedback stabilization","semiglobal asymptotic stabilization","affine nonlinear systems","control Lyapunov functions","Lie algebraic conditions","Artstein-Sontag theorem","Campbell-Baker-Hausdorff formula","bounded feedback"],"falsifier":"Compute the explicit expansion (3.9c) for a concrete low-dimensional system satisfying (2.4) and one of the properties P1/P2, for instance the example system (4.1) with $N=3$, and symbolically evaluate $m^{(N+1)}(0)$ as a polynomial in $\\rho$ and $u_1$; if the claimed identities for $\\Pi_{N+1,i}$ fail, or if every choice of small $u_1$ and $\\rho$ leaves $m^{(N+1)}(0)\\ge 0$, the claim is refuted. A simpler numerical test is to simulate the two-stage control (3.29)-(3.30) on that example and check whether $V$ strictly decreases over every sampling interval.","tokens_in":15704,"feed_emoji":"🎛️","tokens_out":16456,"duration_ms":148331,"temperature":0.7,"pith_summary":"This paper asks when a nonlinear system of the form $\\dot x=f(x)+u g(x)$ can be steered to zero by feedback that is held constant between sampling instants, even when the uncontrolled part $f$ is nonzero. It proves that a generalized control Lyapunov function $V$—a smooth positive definite proper function whose decrease certifies stabilization—suffices, provided that at every nonzero state either $(gV)(x)\\neq 0$, or $(gV)(x)=0$ forces $(fV)(x)<0$, or a finite hierarchy of Lie-bracket derivatives of $V$ vanishes and one of three listed algebraic properties holds. Proposition 2 establishes sampled-data feedback semiglobal asymptotic stabilization under these conditions, and Proposition 3 adds two hypotheses under which the feedback can be chosen bounded. The conditions are weaker than those of the paper's preceding sampled-data Lie-bracket result, so the method covers a broader class of affine systems, and a concrete example shows how the abstract conditions can be checked.","feed_headline":"Broader Lyapunov conditions secure sampled-data stabilization","feed_subtitle":"A generalized control Lyapunov function plus Lie-bracket hypotheses stabilizes affine systems with drift under weaker assumptions.","key_machinery":"The central object is the Lie subalgebra $L\\{f,g\\}$ spanned by iterated brackets $\\lambda_{\\kappa,j}$, each a sum of Lie monomials built from $f$ and $g$ with total order $\\kappa$ and exactly $j$ occurrences of $g$. The argument runs through the Campbell-Baker-Hausdorff expansion of the two-flow composition, which yields the derivative formula $m^{(n)}(0)=(\\rho+1)^n(f^nV)(x)+\\sum_i u_1^i(\\Pi_{n,i}(\\rho;x)+\\rho^{n-1}(\\rho+1)(\\lambda_{n,i}V)(x))$ plus boundary terms. The decisive assumptions are that each polynomial $\\Pi_{n,i}$ is independent of $u$, has degree $n$, is linearly independent of $\\rho^{n-1}(\\rho+1)$, and lies in the span of $V$-derivatives of the bracketed order. Those properties allow an induction in $i$ to choose $u_1$ so that the first $N$ derivatives of $m$ vanish and the $(N+1)$-st is negative.","core_discovery":"The paper's central claim is that sampled-data feedback can overcome the obstruction at states where the control vector field does not change the Lyapunov function, provided the drift and bracket vector fields generate enough higher-order information. At such a point the controller applies two constant inputs in sequence over each sampling interval, $u_2=-\\rho u_1$ for a short piece and $u_1$ for the rest. The Campbell-Baker-Hausdorff formula expresses the derivatives of $m(t)=V(X_{\\rho t}\\circ Y_t(x))$ as a combination of iterated Lie-bracket derivatives of $V$; condition (2.4) makes the derivatives up to order $N$ vanish, and the properties P1 or P2(i)-(iii) make the $(N+1)$-st derivative strictly negative for a suitable choice of $u_1$ and $\\rho$. This produces a decrease of $V$ on every sampling interval, which by the paper's Proposition 1 implies semiglobal asymptotic stabilization. The paper thereby extends the classical Artstein-Sontag theorem to the sampled-data setting and, with the additional conditions (2.10) and (2.11), gives a bounded-feedback version.","pith_inferences":["A natural next step, not taken in the paper, is to derive the three polynomial properties of $\\Pi_{n,i}$ by an independent induction on $n$; doing so would turn the proof's combinatorial core into explicit formulas and make the hypotheses symbolically checkable.","The odd/even conditions in (2.8) suggest an undeveloped link with homogeneous or nilpotent approximations: the parity of the number of $g$-occurrences may act like a higher-order controllability index, so the same $V$ could yield practical stabilization for an approximating system.","A testable extension is to look for a state-independent bound $N\\le N_0$ in (2.4); if such a uniform bound exists, the two-stage controller becomes structurally constant and stabilization over bounded sets should be uniform.","The example's region-by-region verification indicates a general template: check the bracket conditions separately on regions and use boundary relations like (4.3) to glue local decreases into a global Lyapunov decrease."],"forward_implications":["Under Proposition 2, every affine single-input system satisfying the hypotheses is SDF-SGAS: for any prescribed bounded set of initial states and any bounded sequence of sampling intervals, a piecewise-constant feedback makes the origin stable and attracts every trajectory from that set.","With the extra conditions (2.10) and (2.11), the same conclusion holds with a feedback whose magnitude is bounded on each bounded neighborhood of the origin (BSDF-SGAS), which matters for systems with actuation limits.","Because the assumptions use the subalgebra $L\\{f,g\\}$ instead of the full Lie algebra and only require some odd $j$ in (2.7), the new proposition covers systems not covered by the earlier sampled-data Lie-bracket result.","The example system (4.1), with state $(x,y)$ and control direction $\\partial_y$, satisfies both propositions; the partition of $\\mathbb R^n\\setminus\\{0\\}$ into the five regions $E_1,\\ldots,E_5$ shows that the abstract conditions are checkable in practice.","The paper states that the single-input restriction is made for clarity and that the same technique extends to multi-input affine systems, so the result is not limited to one scalar control."],"supporting_citations":[{"why":"Supplies the baseline sampled-data Lie-bracket proposition whose assumptions the present Proposition 2 weakens, and the proof strategy that is extended.","marker":"[24]"},{"why":"The relaxed-control stabilization theorem that the Artstein-Sontag generalization starts from.","marker":"[2]"},{"why":"The universal construction of an almost smooth stabilizing feedback that the sampled-data construction parallels.","marker":"[19]"},{"why":"Earlier sufficient Lyapunov-like stabilization conditions that the Lie-bracket hypotheses refine.","marker":"[20]"},{"why":"The Hermes controlled-stability condition is the model for assumption (2.8) comparing spans of Lie monomials.","marker":"[8]"},{"why":"Introduced sampled-data feedback stabilization and its first Lie-algebraic sufficient conditions.","marker":"[21]"},{"why":"Provided the preceding Lie-algebraic sampled-data sufficient conditions that the present work extends.","marker":"[23]"}],"fun_headline_variants":["Sampled-data feedback overcomes drift via Lie-bracket conditions","Generalized Lyapunov functions stabilize sampled-data affine systems","Artstein-Sontag theorem extended to sampled-data control","Two-step sampled inputs force Lyapunov decrease despite drift","Lie brackets relax Lyapunov conditions for sampled-data stabilization"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the Campbell-Baker-Hausdorff expansion has exactly the claimed algebraic form: the polynomials $\\Pi_{n,i}$ are independent of $u$, have degree $n$, are linearly independent of $\\rho^{n-1}(\\rho+1)$, and lie in the asserted span of bracket derivatives; if any of these three properties fails for some $n,i$, the induction that forces the $(N+1)$-st derivative of $V$ to be negative no longer goes through.","fun_headline_variants_meta":{"raw":{"variants":["Sampled-data feedback overcomes drift via Lie-bracket conditions","Generalized Lyapunov functions stabilize sampled-data affine systems","Artstein-Sontag theorem extended to sampled-data control","Two-step sampled inputs force Lyapunov decrease despite drift","Lie brackets relax Lyapunov conditions for sampled-data stabilization"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000577,"raw_usage":{"total_tokens":2692,"prompt_tokens":889,"completion_tokens":1803,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":505,"completion_tokens_details":{"reasoning_tokens":1721}},"tokens_in":505,"tokens_out":1803,"duration_ms":13319,"temperature":1.0,"reasoning_tokens":1721,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:27:47.418409+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the explicit expansion (3.9c) for a concrete low-dimensional system satisfying (2.4) and one of the properties P1/P2, for instance the example system (4.1) with $N=3$, and symbolically evaluate $m^{(N+1)}(0)$ as a polynomial in $\\rho$ and $u_1$; if the claimed identities for $\\Pi_{N+1,i}$ fail, or if every choice of small $u_1$ and $\\rho$ leaves $m^{(N+1)}(0)\\ge 0$, the claim is refuted. A simpler numerical test is to simulate the two-stage control (3.29)-(3.30) on that example and check whether $V$ strictly decreases over every sampling interval.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the baseline sampled-data Lie-bracket proposition whose assumptions the present Proposition 2 weakens, and the proof strategy that is extended."},{"cited_title":"Artstein, ”Stabilization with relaxed controls,” Nonlinear Analysis TMA, vol.7, pp","cited_arxiv_id":null,"evidence_quote":"The relaxed-control stabilization theorem that the Artstein-Sontag generalization starts from."},{"cited_title":"Sontag, ”A ”universal” construction of Artstein’s theorem on nonlinear stabilization,” Systems and Control Lett","cited_arxiv_id":null,"evidence_quote":"The universal construction of an almost smooth stabilizing feedback that the sampled-data construction parallels."},{"cited_title":"Tsinias, Sufficient Lyapunov-like conditions for stabilization, Math","cited_arxiv_id":null,"evidence_quote":"Earlier sufficient Lyapunov-like stabilization conditions that the Lie-bracket hypotheses refine."},{"cited_title":"Hermes , ”Controlled Stability,” Ann","cited_arxiv_id":null,"evidence_quote":"The Hermes controlled-stability condition is the model for assumption (2.8) comparing spans of Lie monomials."},{"cited_title":"Tsinias, Remarks on asymptotic controllability and sampled-data feedback stabilization for autonomous systems, IEEE Trans","cited_arxiv_id":null,"evidence_quote":"Introduced sampled-data feedback stabilization and its first Lie-algebraic sufficient conditions."},{"cited_title":"Tsinias, New results on sampled-data feedback stabilization for autonomous nonlinear systems, Systems and Control Lett","cited_arxiv_id":null,"evidence_quote":"Provided the preceding Lie-algebraic sampled-data sufficient conditions that the present work extends."}],"review_version":1}