{"id":"603525c0-1c0f-41a0-ae8b-063c166dfb06","arxiv_id":"2608.12591","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"A theta-symmetric scaled relative graph is introduced, its phase is computed by semidefinite programming, and necessary and sufficient cyclic robust stability conditions are extended to cactus networks.","lead":"This paper introduces a rotated version of the scaled relative graph, a geometric tool for feedback analysis, and uses it to give stability tests for multi-loop networks with cactus structure. It also connects a phase measure to a matrix norm minimization problem, making the phase computable by semidefinite programming.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 7's sufficiency proof is invalid as written: it invokes Theorem 6, but the -1 avoidance condition does not imply Theorem 6's ray-avoidance hypothesis; the homotopy condition requires [-∞,-1]∩S=∅.","rationale":"The reader's weakest_assumption points to Proposition 5 (interpolation) as the load-bearing concern. I find that less convincing: Proposition 5 is exactly the 0-symmetric case needed for Theorem 7, and its proof via small-gain interpolation is essentially correct, provided one chooses the parameter μ corresponding to the disk R[β,γ] (μ=γ cosβ/(1+sinβ)); the omitted M^c necessity in Theorem 5 is a real proof gap but it is a proof-completeness issue rather than a demonstrated false step. The more direct and more load-bearing defect is in the sufficiency proof of Theorem 7: the text claims it follows from Theorem 6, but Theorem 6 requires ray avoidance, while Theorem 7 only checks pointwise -1 avoidance. The homotopy argument used in Theorem 6's proof needs nonsingularity of I+τ∏P_i for all τ∈[0,1], which is equivalent to avoiding the entire ray [-∞,-1]. The scalar example with P_k=-4, R=[1/3,1] shows -1 may be avoided while the ray is intersected, so Theorem 6 does not apply. This is not a mere cosmetic issue: it breaks the only stated sufficiency argument for the central theorem. The concrete test would settle whether the theorem's statement is actually false or just under-proved; either way the manuscript requires substantial revision. Therefore the verdict remains CONDITIONAL, and the reader's verdict is unchanged, though the primary concern differs from the reader's stated weakest assumption.","tokens_in":25438,"tokens_out":35699,"duration_ms":301363,"concrete_test":"Test the sufficiency implication on a minimal scalar case. Let N=2, P_k(s)=-4 (constant), β=π/6, γ=1 so that R[β,γ] is the real interval [1/3,1]. Choose θ(ω)=π/2; then SRG_θ(P_k)={-4,4}, and -1∉SRG_θ(P_k)R[β,γ] because -4R=[-4,-4/3] and 4R=[4/3,4]. However, [−∞,−1]∩(-4R)≠∅, so Theorem 6's ray condition (20) fails. Now take P_2(s)=0.5, which lies in R[β,γ]. The homotopy condition I+τP_kP_2 = 1-2τ is singular at τ=0.5, precisely the scenario Theorem 6's proof must rule out; nonetheless, (1+P_kP_2)^{-1} is stable. This example demonstrates that the claimed derivation from Theorem 6 is invalid. Then perform a small numerical search over first-order real-rational P_k and P_2 in P[β,γ] where -1∉S holds pointwise, checking whether the Nyquist curve of P_kP_2 encircles -1. Any such case would refute Theorem 7's sufficiency; if none exists, the theorem may survive but needs a new proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 7's necessary-and-sufficient robust stability condition. The sufficiency direction is dismissed with 'The sufficiency follows from Theorem 6', but Theorem 6 requires [−∞,−1]∩∏SRG_{θ_i}(P_i(jω)) = ∅ for all ω, i.e., the entire negative ray from -1 to -∞ is avoided. Theorem 7 only assumes -1∉SRG_{θ(ω)}(P_k(jω))∏_{i∈K'}R[β_i,γ_i]. Even after setting θ_i=0 for the uncertain systems (allowed because P_i∈P[β_i,γ_i] implies SRG(P_i(jω))⊂R[β_i,γ_i]), the best one obtains from Theorem 7 is -1∉S, not [−∞,−1]∩S=∅. These are inequivalent: a scalar example with P_k=-4 and R[β,γ]=[0.333,1] gives S=[-4,-1.333], which avoids -1 but intersects the ray at -2. The proof of Theorem 6 itself uses the homotopy condition (21) with τ∈[0,1], requiring nonsingularity of I+τ∏P_i for all τ, which is exactly the ray condition. No lemma in the paper bridges the point condition to the ray condition, so the sufficiency proof of Theorem 7 is incomplete. This is a load-bearing gap: if -1∉S does not actually preclude encirclement of -1 by eigenloci, the 'if' direction of the central theorem fails; if it does, the paper needs a new argument, since Theorem 6 cannot be invoked as written.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript introduces the θ-symmetric scaled relative graph (SRG), a phase-parameterized variant of the SRG designed to retain the sign of phase information for complex matrices and systems. It develops the basic properties of this object, including a spectrum-containment lemma and inverse/rotation rules, and gives an LMI/norm-minimization characterization of the associated θ-segmental phase. The paper then proves subadditive and submultiplicative properties for sums and products of matrix sets, derives nonsingularity criteria for sum-type and product-type return-difference matrices, and states a necessary and sufficient robust-stability condition for cyclic interconnections under mixed gain-phase uncertainty. The final section extends the framework to cactus networks, with a detailed sunflower-graph theorem and an algorithmic hierarchical procedure for general cactus graphs.","tokens_in":25791,"tokens_out":18735,"duration_ms":178963,"significance":"If the results hold, the paper provides a genuinely compositional graphical method for MIMO feedback stability that generalizes the classical Nyquist plot and handles mixed gain-phase uncertainty in a less conservative way than separate gain or phase conditions. The SDP characterization of segmental phase in Theorem 1 and the interpolation construction in Proposition 5 are nontrivial and likely to be of independent use. The submultiplicative/subadditive calculus of θ-symmetric SRGs is clean and well suited to network analysis. The paper does not provide code or machine-checked proofs, but the main derivations are self-contained except for the specific gaps identified below.","major_comments":[{"comment":"The statement 'The sufficiency follows from Theorem 6' is not justified as written. Theorem 6 requires the ray avoidance condition [−∞,−1] ∩ ∏ SRG_{θ_i(ω)}(P_i(jω)) = ∅ for every ω, whereas Theorem 7 assumes only −1 ∉ SRG_{θ(ω)}(P_k(jω)) ∏ R[β_i, γ_i]. The gap is repairable: each R[β_i, γ_i] is star-shaped about 0 and contains 0; finite products of star-shaped sets are star-shaped; and multiplication of a star-shaped set by any fixed complex set preserves star-shapedness, so the set S(ω) = SRG_{θ(ω)}(P_k(jω)) ∏ R[β_i, γ_i] is star-shaped. If S(ω) met (−∞,−1], it would contain −1. The authors should add this argument explicitly, and should also state that the uncertain subsystems are assigned θ_i(ω) = 0 with over-approximations R[β_i, γ_i], before invoking Theorem 6.","section":"Section VII-B, Theorem 7 (sufficiency)"},{"comment":"The necessity direction invokes Theorem 5 but does not specify which variant of Theorem 5 is being used. The uncertainty set P[β, γ] in Theorem 7 is defined by the 0-symmetric SRG, i.e., by SRG_0(P(jω)) ⊂ R[β, γ], which corresponds to the M_c[−β, β, γ] variant of Theorem 5, not to the M[−β, β, γ] variant whose proof is supplied. The appendix only proves the M-case, saying that the M_c case follows similarly. Since the entire necessity argument depends on the M_c case, the paper should either provide the M_c proof or give a precise reduction to the M-case; as written, this is an omitted proof in a load-bearing position.","section":"Section VII-B, Theorem 7 (necessity) and Appendix I, proof of Theorem 5"},{"comment":"No correctness theorem is stated or proved for the general cactus-network algorithm. Theorem 8 covers sunflower graphs only; the text asserts that the recursive decomposition extends naturally and then presents Algorithm 1, but there is no proof that the local checks plus loop absorption certify stability of every cactus network, nor a statement of conservatism. Since the abstract and introduction claim a unified framework for general cactus networks, this is a load-bearing omission: either add a formal theorem with an induction proof over the loop tree, or restrict the claim to the sunflower case.","section":"Section VIII-B, Algorithm 1"}],"minor_comments":[{"comment":"The notation [1, e^{2jθ_i(ω)}] is used without definition; presumably it denotes the line segment between 1 and e^{2jθ_i(ω)}, but the text should define it and explain how it arises from SRG_{θ_i(ω)}(I + Σ_q L_q(jω)).","section":"Section VIII-A, Theorem 8"},{"comment":"The region Ω is described as {r e^{jθ} : |r| ≤ 0.6 or −15° ≤ θ ≤ 10°}; the use of 'or' and the relationship between r and θ are ambiguous, and the polar-coordinate specification should be rewritten for clarity.","section":"Example 7"},{"comment":"In the equation for the equivalent central loop, the product over i ∈ V_1 is described as being taken 'in the order consistent with the connection along the central loop'; this ordering should be made explicit, since cyclic permutations of the factors and the accompanying phase parameters affect the formal statement.","section":"Section VIII-A, Theorem 8 proof"},{"comment":"The proof states that condition (20) implies [−∞,−1] ∩ Λ(P_1(jω)⋯P_N(jω)) = ∅; this is correct because the product of the SRGs contains the spectrum of the product, but the step should be stated explicitly so that the role of spectrum containment is transparent.","section":"Section VII-A, Theorem 6 proof"},{"comment":"The output line 'Stable if all verification steps pass' should name the exact inequalities verified at each layer (e.g., the analogues of (26) and (27)) and the admissible over-approximation choices, so that the algorithm is reproducible without reconstructing the derivation from Theorem 8.","section":"Section VIII-B, Algorithm 1"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a serious systems-theory contribution, and the central cyclic robust-stability result appears salvageable: the point-versus-ray gap in the proof of Theorem 7 can be closed by a star-shapedness argument that is not currently in the paper. The more substantial omissions are the unproved correctness of Algorithm 1 for general cactus graphs and the omitted M_c-case in the proof of Theorem 5; both should be addressed in revision. The authors should also clarify the incremental contribution over their closely related conference papers [34] and [35]."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"What you should know: this is a real contribution to MIMO gain-phase stability analysis, but the main necessary-and-sufficient robust stability theorem has a proof gap in the sufficiency direction. The gap is load-bearing, not a technicality.\n\nThe new θ-symmetric SRG is a good idea. It fixes the sign ambiguity of the 0-symmetric SRG and gives a more faithful MIMO analogue of the Nyquist plot. Theorem 1's connection between θ-segmental phase and a norm minimization problem is genuinely useful—it provides an SDP for a quantity that was previously hard to compute. The subadditivity and submultiplicativity results in Theorem 3 are carefully derived and form the right algebraic foundation for the later tests. I also credit the necessity direction of Theorem 7: the construction using Proposition 5 to interpolate single-frequency data into stable transfer matrices is a serious argument, and Proposition 5 itself seems plausible, building on classical small-gain interpolation.\n\nThe problem is Theorem 7's sufficiency. The proof says 'follows from Theorem 6.' Theorem 6 requires avoiding the entire ray (−∞,−1] in the SRG product; Theorem 7 only assumes −1 is avoided. The scalar example in the stress-test note is the right counterexample: if the product set is the interval [−4,−1.33], then −1 is outside but points like −2 are inside. Pointwise avoidance of −1 does not prevent eigenloci from encircling −1, and the homotopy argument (I+τP nonsingular for τ∈[0,1]) needs the ray condition because −1/τ≤−1. So the sufficiency proof does not go through. The theorem might be salvageable with an extra argument—for instance, showing the product set is star-shaped around 0, or using a different stability criterion—but that argument is not in the paper.\n\nTwo smaller issues: the general cactus test in Section VIII-B is an algorithm without a formal correctness theorem, and the proof of Theorem 5 for the M^c case is skipped with 'similar lines.' These are minor relative to the main gap, but they matter in a paper that aspires to be a unified framework.\n\nWho gets value from this: control theorists working with phase-based MIMO stability and anyone interested in compositional network tests. The algebraic core is worth knowing. But I would not cite Theorem 7's iff claim as it stands. Send it to peer review, but the referee should push hard on the sufficiency argument. If the authors close that gap, this is a good IEEE TAC paper; if not, they need to weaken the claim to a sufficient condition.","headline":"A genuinely new SRG variant with a solid algebraic core, but Theorem 7's sufficiency proof is missing the key encirclement argument.","tokens_in":26332,"tokens_out":10076,"would_cite":false,"duration_ms":84407,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93D09","93D05","93C80","15A60"],"pacs":[],"model":"deepseek-v4-flash","headline":"The θ-symmetric scaled relative graph records signed phase, not just magnitude, and its submultiplicative and subadditive structure yields necessary and sufficient robust stability tests for cyclic and cactus feedback networks.","keywords":["θ-symmetric scaled relative graph","phase lead and lag","Nyquist plot","cactus network","cyclic interconnection","robust stability","semidefinite programming","return difference matrix"],"falsifier":"Check Proposition 5 numerically for the symmetric case: pick $\\beta=30^\\circ$, $\\gamma=1$ and a random matrix $\\Delta$ with $\\mathrm{SRG}(\\Delta)\\subset R[\\beta,\\gamma]$, then attempt to construct a real stable rational $P$ with $P(j\\omega_0)=\\Delta$ as in the proof. If for some such $\\Delta$ no real-rational stable $P$ exists, the necessity direction of Theorem 7 cannot be carried through and the iff claim fails.","tokens_in":25203,"feed_emoji":"📈","tokens_out":13075,"duration_ms":101352,"temperature":0.7,"pith_summary":"The paper tries to establish that a phase-aware redefinition of the scaled relative graph (SRG)—the device that records how a matrix rotates and scales each vector direction—captures both gain and the sign of phase for multivariable systems, and that this one object cleanly certifies feedback stability of cyclic cascades and, recursively, of entire cactus networks. The authors first prove that the graph's phase aspect equals a scalar norm-minimization problem, so the phase interval is computable by semidefinite programming. They then prove submultiplicative and subadditive inequalities for the graph, which convert nonsingularity of product- and sum-type return difference matrices into the geometric condition that $-1$ avoid a certain region. For cyclic interconnections with symmetric gain-phase uncertainty they prove a necessary and sufficient robust stability condition, and they extend it to cactus networks by absorbing petal loops into their hubs layer by layer. A sympathetic reader would take this as a compositional MIMO generalization of the Nyquist plot: subsystem-level regions, not eigenloci, decide stability.","feed_headline":"Phase-aware graph test certifies exact stability for MIMO loops","feed_subtitle":"A phase-aware redefinition of the scaled relative graph turns Nyquist-style checks into an iff condition for cactus networks.","key_machinery":"The load-bearing object is the θ-symmetric scaled relative graph, $\\mathrm{SRG}_\\theta(C)=\\{\\frac{\\|Cx\\|}{\\|x\\|}e^{j(\\theta\\pm\\angle_\\theta(x,Cx))}:x\\neq0\\}$, with $\\angle_\\theta(x,y)=\\arccos\\frac{\\operatorname{Re}\\langle x,e^{-j\\theta}y\\rangle}{\\|x\\|\\|y\\|}$. The rotation $e^{-j\\theta}$ before the inner product is what restores phase sign: a scalar matrix $zI$ now has $\\mathrm{SRG}_{\\angle z}(zI)=\\{z\\}$ instead of the conjugate pair $\\{z,\\bar z\\}$. The argument then runs on three rails: Theorem 1 ties the phase interval to $\\arcsin\\min_{\\gamma>0}\\|\\gamma e^{-j\\theta}C-I\\|$ (hence SDP-computable), Theorem 3 gives subadditivity and submultiplicativity under θ-chord and θ-arc hull over-approximations, and Proposition 5—the single-frequency interpolation of a matrix with prescribed SRG by a stable real-rational transfer function—converts the region-avoidance certificate into a necessary and sufficient robust stability test.","core_discovery":"On the paper's own terms, the central discovery is Theorem 7: for the feedback interconnection of a cascade of stable systems $P_1,\\dots,P_N$, with one nominal system $P_k$ and the rest uncertain in $P[\\beta_i,\\gamma_i]$ (sets whose 0-symmetric SRG lies in the gain-phase region $R[\\beta_i,\\gamma_i]$ at every frequency), the closed loop is robustly stable for all admissible plants if and only if there exists a frequency-dependent angle $\\theta(\\omega)$ such that $-1\\notin \\mathrm{SRG}_{\\theta(\\omega)}(P_k(j\\omega))\\prod_{i\\in K'}R[\\beta_i,\\gamma_i]$ for all $\\omega$. The 'only if' is the sharp part: it uses a constructive interpolation (Proposition 5) to turn a single-frequency violation into an actual destabilizing plant. Section VIII then lifts this cyclic test to cactus networks through a hierarchical 'absorb the leaves' procedure, replacing the global return-difference inverse by a sequence of local loop checks.","pith_inferences":["Editorial inference: the same structure suggests a phase-domain analogue of structured singular value analysis: choosing the $\\theta_i$ that maximize the distance from $-1$ to the product of SRG regions (Remark 5) would automate robustness certificates, much as $D$-scaling tightens $\\mu$ bounds.","Editorial inference: if the open asymmetric real-coefficient interpolation problem (Remark 6) is solved, Theorem 7 should extend to uncertainties with different upper and lower phase bounds, which matters because physical plant uncertainty is rarely symmetric in phase.","Editorial inference: the leaf-absorption procedure applies to any network whose simple-loop intersection graph is a tree, so quantifying the approximation error for networks slightly denser than cactus is a natural next test.","Editorial inference: because the original SRG was defined for nonlinear operators, the θ-symmetric version may carry the same compositional stability checks to nonlinear multivariable systems; the paper leaves this, and controller design, to future work."],"forward_implications":["$-1$ avoidance conditions like (20) and Theorem 7 can be checked frequency by frequency from each subsystem's SRG region, so stability verification does not require computing or plotting MIMO eigenloci.","For SISO systems the frequency-wise canonical SRG coincides exactly with the Nyquist plot, so all the new tests specialize to the classical picture.","The θ-segmental phase is computed by a semidefinite program, giving a concrete computational route for the previously open problem of computing segmental phase.","Cactus network stability can be certified by a recursive leaf-absorption algorithm, so the cost and the insight scale with local loops rather than with the global return difference matrix.","In the symmetric uncertainty description the cyclic robust stability test is exact (if and only if), eliminating the conservatism of sufficient-only gain-phase separation methods."],"supporting_citations":[{"why":"supplies the generalized Nyquist criterion and the small-gain interpolation construction used in the sufficiency proof and in Proposition 5.","marker":"[2]"},{"why":"introduces the cyclic small phase theorem and segmental phase that the θ-symmetric SRG extends and compares against.","marker":"[12]"},{"why":"provides the unitary-orbit construction that the necessity proof of Theorem 5 follows, cited as similar to [17, Theorem IV.1].","marker":"[17]"},{"why":"defines the original scaled relative graph and the chord/arc property ingredients used by Lemma 1 and Theorem 3.","marker":"[26]"},{"why":"brings SRG into feedback-system analysis and establishes the 0-symmetric SRG results that Corollary 2 and the comparisons generalize.","marker":"[28]"},{"why":"gives the rational matrix interpolation theory used to realize single-frequency matrices as stable transfer functions in Proposition 5.","marker":"[45]"},{"why":"defines cactus graphs and their stability background, the network class the paper's hierarchical framework targets.","marker":"[46]"}],"fun_headline_variants":["θ-SRG gives iff condition for robust MIMO stability","Exact robust stability test for cactus networks via θ-SRG","Phase-aware SRG yields iff stability for cyclic interconnections","θ-symmetric SRG: necessary and sufficient robust stability test"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The 'only if' direction of the main robust-stability theorem rests on the interpolation claim that every matrix whose SRG lies in the allowed symmetric gain-phase region at one frequency can be realized by a stable real-rational transfer matrix, a claim the paper proves only for symmetric phase bounds and explicitly leaves open for asymmetric θ-symmetric bounds.","fun_headline_variants_meta":{"raw":{"variants":["θ-SRG gives iff condition for robust MIMO stability","Exact robust stability test for cactus networks via θ-SRG","Phase-aware SRG yields iff stability for cyclic interconnections","θ-symmetric SRG: necessary and sufficient robust stability test"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000978,"raw_usage":{"total_tokens":4189,"prompt_tokens":1015,"completion_tokens":3174,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":631,"completion_tokens_details":{"reasoning_tokens":3105}},"tokens_in":631,"tokens_out":3174,"duration_ms":18915,"temperature":1.0,"reasoning_tokens":3105,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T00:07:31.146385+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Check Proposition 5 numerically for the symmetric case: pick $\\beta=30^\\circ$, $\\gamma=1$ and a random matrix $\\Delta$ with $\\mathrm{SRG}(\\Delta)\\subset R[\\beta,\\gamma]$, then attempt to construct a real stable rational $P$ with $P(j\\omega_0)=\\Delta$ as in the proof. If for some such $\\Delta$ no real-rational stable $P$ exists, the necessity direction of Theorem 7 cannot be carried through and the iff claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the generalized Nyquist criterion and the small-gain interpolation construction used in the sufficiency proof and in Proposition 5."},{"cited_title":"A cyclic small phase theorem,","cited_arxiv_id":null,"evidence_quote":"introduces the cyclic small phase theorem and segmental phase that the θ-symmetric SRG extends and compares against."},{"cited_title":"Scaled relative graphs: nonexpan- sive operators via 2D Euclidean geometry,","cited_arxiv_id":null,"evidence_quote":"defines the original scaled relative graph and the chord/arc property ingredients used by Lemma 1 and Theorem 3."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"gives the rational matrix interpolation theory used to realize single-frequency matrices as stable transfer functions in Proposition 5."},{"cited_title":"Diagonal stability on cactus graphs and application to network stability analysis,","cited_arxiv_id":null,"evidence_quote":"defines cactus graphs and their stability background, the network class the paper's hierarchical framework targets."}],"review_version":1}