{"id":"af171d96-8af8-4f19-82e2-202f0f5219b7","arxiv_id":"2507.14620","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For fragile power domination, two graphs with equal-size sensor sets have identical expected observed-vertex polynomials exactly when their k-subset observation sums agree, and gadget constructions can freely control all but the lowest two coefficients.","lead":"This paper studies fragile power domination, a graph model of power grid monitoring where sensors independently fail, and characterizes when two networks have the same expected number of observed nodes. It also builds graph gadgets that let mathematicians control the coefficients of the expected value polynomial, and shows that a common placement heuristic can fail.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the asserted identity (19) in Theorem 3.7 verifies on direct computation, and the other flagged issues are typographical.","rationale":"I read the paper as establishing three things: (i) Theorem 3.1 characterizes equality of fragile power domination expectations via equality of k-subset sums, using the Bernstein basis; (ii) Theorem 2.2/2.3 shows the coefficients q^k for k≥2 can be arbitrarily prescribed by affixing Ψ and Φ gadgets; (iii) Theorem 3.3 characterizes linear polynomials and Theorem 3.7 provides a partial converse to Conjecture 3.4 under a largeness condition on the linear coefficient. I checked the most delicate steps. Lemma 2.1's probabilities are correct: the two-leaf condition prevents A from forcing, and the Ψ connection vertex retains multiple unobserved subdivision vertices unless all A-sensors succeed. Theorem 2.2's descending induction correctly isolates each target coefficient because higher-index hi(q) affect lower coefficients in a controlled way. The proof of Theorem 3.7's identity (19) was verified by direct counting of |Obs| on the gadget graph, as detailed in load_bearing_attack. The residual structural identity (18) for the replacement graph H follows from the component decomposition independently of (19); the polynomial match only serves to transfer the k-subset sums via Theorem 3.1. The sign error in Proposition 4.2 (λk(S)-λk(A) ≥ ℓ - ... instead of ... - ℓ) is a typo in the displayed inequality; the surrounding text and choice ℓ > s^2 2^{s+1} show the intended bound is satisfied. The reader's call for a derivation of (19) is reasonable as an editorial request but does not indicate a correctness risk. I therefore find no load-bearing objection and would not adjust the verdict on mathematical grounds.","tokens_in":20198,"tokens_out":27128,"duration_ms":293912,"concrete_test":"Run an exhaustive brute-force enumeration of all 2^s sensor failure subsets on the small construction H(t,w,d) for, say, (t,w,d)=(1,2,1) and s=4, using an independent implementation of the power domination process, and confirm that the computed expected observed count equals the right-hand side of (19); repeat for at least three parameter triples.","verdict_should_be":"UNCHANGED","load_bearing_attack":"After direct verification, the reader's primary concern does not land. Identity (19) in Theorem 3.7 is asserted without derivation, but it is correct: for the gadget graph H'=(2P3 ⊟_A Ψ^2_t) ⊟_A Φ^2_w, a single surviving sensor at v0 yields |Obs(H';{v0})|=w+9 (the 2P3 contributes 4 observed vertices, the Ψ gadget contributes 2, and the Φ gadget contributes 3+w), while two surviving sensors observe all 13+t+w vertices. These counts give E(H';A,q)=2(w+9)q(1-q)+(t+w+13)(1-q)^2 = (t-w-5)q^2-(2t+8)q+(t+w+13); adding the K_{d+1} component (d+1)(1-q) and the s-3 isolated vertices (s-3)(1-q) reproduces exactly the claimed (t-w-5)q^2-(s+2t+d+6)q+(s+t+w+d+11). The remaining flagged items in the paper—the sign of the lower bound in Proposition 4.2's proof and the K_d/K_{d+1} labeling in Theorem 3.7's proof—are typographical and do not affect the argument.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies fragile power domination, where each PMU fails independently with probability q, and analyzes the expected number of observed vertices E(G;S,q). The main results are: a Bernstein-basis characterization of when two graphs with sensor sets of equal size have the same expected value polynomial (Theorem 3.1); a necessary and sufficient condition for E(G;S,q) to be linear (Theorem 3.3); partial progress on a conjecture characterizing polynomials of degree at most ℓ (Theorem 3.6 and Theorem 3.7); and a construction showing that minimum power dominating sets using only degree-3 vertices can be strictly better, in the fragile-power-domination sense, than sets containing a high-degree vertex (Section 4). Section 2 develops Ψ and Φ gadgets intended to control individual coefficients of the expected value polynomial.","tokens_in":20422,"tokens_out":53840,"duration_ms":408263,"significance":"The characterization in Theorem 3.1 is elegant and useful: it reduces polynomial equality to finitely many integer conditions, and it underpins the comparison machinery in Section 4. Theorem 3.3 gives a clean linearity criterion. The Section 4 counterexample to the high-degree heuristic is explicit and appears convincing. If the coefficient-control machinery of Section 2 were fully rigorous, it would be a valuable engineering tool for fragile power domination. However, the proof of Theorem 2.2 currently has a serious gap in the treatment of multiple overlapping gadgets, which affects one of the paper's advertised main results.","major_comments":[{"comment":"The proof of Theorem 2.2 assumes that each affixed gadget behaves independently, with the path-head observation probabilities given in Lemma 2.1, but Lemma 2.1 is only proved for a single affixed gadget. When several gadgets are affixed to overlapping sets A_i, a failed affix vertex can be observed by another gadget, which changes the observation probability of larger Φ gadgets. Concretely, take S={a,b,c,d}, f≡1, and affix Φ^2 at {a,c}, Φ^2 at {b,d}, and Φ^4 at {a,b,c,d}. If sensors a and b succeed and c and d fail, the two Φ^2 gadgets force c and d to be observed (in a Φ^2, a successful affix forces the cycle vertex y, which forces the failed affix). In the Φ^4 gadget, every B-vertex, i.e. every 3-subset of {a,b,c,d}, contains at least one of a or b, so all B-vertices are observed by domination; the B-vertex {a,c,d} then has all three of its A-neighbors observed, so it forces the connection vertex x, and x forces the path head. Thus the Φ^4 path head is observed despite only two of the four sensors in A succeeding, contradicting the formula in Lemma 2.1, which for a=4 requires at least three successes. The expressions for h_i(q) in equations (5)-(7) therefore do not account for cross-gadget contributions, and the coefficient-control argument in Theorem 2.2 is unsupported as written.","section":"Section 2.2, Lemma 2.1 and Theorem 2.2"}],"minor_comments":[{"comment":"The graph G' := K_{ℓ-s+1} ⊔ K_{s-1} should use the empty graph on s-1 vertices (overline{K}_{s-1}), not the complete graph; otherwise the graph has only two components and cannot contain one sensor in each of s components. The notation in the Appendix appears to have lost the overline.","section":"Theorem 3.3 proof"},{"comment":"The component H2 is defined as K_d but the earlier construction uses K_{d+1}; the proof should read H2 = K_{d+1}, and K_{s-3} denotes the empty graph on s-3 vertices.","section":"Theorem 3.7 proof"},{"comment":"The identity E(H;S',q) = (t-w-5)q^2 - (s+2t+d+6)q + (s+t+w+d+11) is asserted as 'straightforward to verify'. The identity is correct, but the verification should be sketched, since it is load-bearing for the parameter choices that follow.","section":"Theorem 3.7, equation (19)"},{"comment":"The displayed inequality has a sign error: the correct bound is λ_k(S)-λ_k(A) ≥ ℓ - binom(s,k)2s^2, not binom(s,k)2s^2 - ℓ. With ℓ > s 2^{2s+1}, the corrected expression is positive, so the argument can be repaired, but the current text is inconsistent.","section":"Proposition 4.2 proof"},{"comment":"The statement 'the only vertices v ∈ V(G) which have the property that L_i ⊆ Obs(G;v) are vertices in S' is too strong; a sensor placed inside a pendant path would observe that path. What is needed and true is that the specific vertex v_{1,2} ∈ K has no L_i contained in its observation set, which is what the case relies on.","section":"Proposition 4.2, Case 1"},{"comment":"The polynomial is written as E(G;S,q) = ax^2 - bx + c but the variable should be q throughout.","section":"Theorem 3.7 statement"}],"recommendation":"major_revision","confidential_remarks":"The main characterization (Theorem 3.1), the linearity criterion (Theorem 3.3), and the Section 4 construction appear sound and are the strongest parts of the paper. The serious issue is in Section 2: the proof of Theorem 2.2 does not justify that multiple affixed gadgets behave independently, and there is a concrete configuration where a Φ^4 gadget is observed with only two successes because smaller Φ^2 gadgets observe the failed affix vertices. This affects one of the paper's advertised main results. I would recommend major revision rather than rejection, because the rest of the paper is independent and valuable, and it may be possible to repair the gadget construction or the analysis. The authors should also correct the notation and sign errors listed in the minor comments."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"You should read this one. It is a genuine advance in the fragile power domination line, not just a re-coordinatization of earlier work. The new things: the Ψ/Φ gadget constructions that give coefficient-level control of the expected value polynomial (Theorem 2.3), the Bernstein-basis characterization of when two networks have identical expectation (Theorem 3.1), the concrete counterexample to the converse of the earlier linearity condition from [4], and the necessary-and-sufficient linearity characterization (Theorem 3.3). Section 4's comparison of sensor placements and the high-degree heuristic counterexample are also worth a look.\n\nWhat the paper does well: the proofs are mostly complete and the gadget probability computations in Lemma 2.1 are careful and correct. Theorem 3.1 is a clean reduction of polynomial equality to finitely many integer conditions, and it genuinely makes the later arguments work. The example in Figure 3 is well chosen.\n\nThe one spot that made me nervous is identity (19) in the proof of Theorem 3.7, where the expected value of the replacement graph H is asserted as \"straightforward to verify\" without derivation. The reader flagged this as load-bearing. I sat down and computed it: the count is correct. A single surviving sensor at v0 observes w+9 vertices, two surviving sensors observe all 13+t+w vertices, and adding K_{d+1} and the s−3 isolated vertices reproduces exactly the claimed quadratic. So the concern does not land. That said, the authors should still add a short derivation; a referee cannot be expected to take an unproven count on faith for a theorem that uses it essentially.\n\nThe other issues are genuinely minor. In the proof of Theorem 3.3, the notation K_{s−1} is ambiguous: the argument only works if it means the empty graph, which the appendix does define but the body does not say. In the proof of Proposition 4.2, the displayed inequality has a sign error: the correct bound is λk(S) − λk(A) ≥ ℓ − (s choose k)2s^2, not the reverse. The conclusion still follows once you fix the sign, and it does not affect the theorem statement. Similarly, in Theorem 3.7's proof, the component label K_d should be K_{d+1}.\n\nCentral argument holds up. The partial forward direction of Conjecture 3.4 is honestly labeled, and the restriction to large negative linear coefficient is reasonable given Proposition 2.4. No circularity, no fitted parameters, no invented entities.\n\nWho is this for? Anyone working on power domination, zero forcing, or probabilistic graph searching. It deserves a serious referee; the authors are pointed at the right open questions and the machinery is reusable. I would engage with it, and with a request for the short derivation of (19) plus the typo fixes, I would be comfortable accepting it.","headline":"Solid gadget-based toolkit and a correct characterization theorem for fragile power domination; the one scary asserted identity checks out, and the remaining issues are expositional typos.","tokens_in":21003,"tokens_out":1716,"would_cite":true,"duration_ms":20683,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C69"],"pacs":[],"model":"deepseek-v4-flash","headline":"The expected number of observed vertices under fragile power domination is equal for two graphs exactly when, for every subset size k, the total observations over all k-subsets of the sensor set are equal; add-on gadgets can force most…","keywords":["fragile power domination","expected value polynomial","phasor measurement units","sensor failure","power domination","zero forcing","graph gadgets","coefficient control"],"falsifier":"Recompute identity (19) directly for small parameters, for example s=3, t=w=d=0, by enumerating every subset W of the sensor set S' in the gadget graph H and summing |Obs(H;W)|; the resulting polynomial should equal -$5q^{2}$-9q+14, and any mismatch would refute the gadget count on which Theorem 3.7 rests.","tokens_in":19975,"feed_emoji":"⚡","tokens_out":8121,"duration_ms":87351,"temperature":0.7,"pith_summary":"This paper studies the expected number of observed vertices in a power grid when each PMU fails independently with probability q. Its central theorem is a characterization: two graphs with sensor sets of the same size produce the same expected-value polynomial exactly when, for every subset size k, the total number of vertices observed by all k-subsets of sensors is the same in both graphs. The paper also constructs Ψ and Φ gadgets that, when attached to the sensor set, can force the coefficients of q^k for k at least 2 to any prescribed integer, while proving that the constant and linear coefficients are not freely controllable. It gives a necessary and sufficient condition for the expected polynomial to be linear, and it builds a graph family in which a minimum sensor set made of degree-3 vertices strictly outperforms any minimum set containing a high-degree vertex.","feed_headline":"Power-grid coverage under failures is captured by subset sums","feed_subtitle":"A finite subset-sum test decides when two sensor layouts give identical expected coverage; gadget tools and a high-degree counterexample…","key_machinery":"The load-bearing object is the expected value polynomial E(G;S,q), expressed in the Bernstein-type basis {$q^{{s-k}}$(1-q)^k}; the coefficient of each basis vector is the sum of |Obs(G;W)| over all k-subsets of the sensor set, which makes polynomial equality equivalent to finitely many integer equalities. The coefficient-control machinery consists of two gadget families, the Ψ and Φ gadgets, affixed to sensor vertices by identifying a set of affix vertices; Lemma 2.1 gives the probability that a gadget's path head is observed, and choosing appended path lengths lets the proof set the coefficient of q^k to any integer for k at least 2. Forts play a supporting role in lower bounds, because a sensor must lie in a fort or its entrance for the fort to be observed.","core_discovery":"The paper's central claim is that the fragile power domination expectation E(G;S,q), as a polynomial in q, is completely encoded in the quantities a_k = sum over k-subsets W of S of |Obs(G;W)|. Theorem 3.1 shows that E(G;S,q)=E(G';S',q) iff a_k=a'_k for every 1≤k≤s, because the polynomial is written in the basis {$q^{{s-k}}$(1-q)^k} and these sums are exactly the basis coefficients. From this, the paper derives a sharp linearity criterion (Theorem 3.3), proves the reverse direction of a conjectured degree bound for general ℓ (Theorem 3.6), and proves the forward direction in the quadratic case when the linear coefficient is sufficiently negative (Theorem 3.7). It also proves a monotonicity result: if every k-subset total for one placement is at least as large as the corresponding total for another, then the first placement is at least as good at every failure probability, and a constructed graph family shows that degree is not a reliable guide to placement quality.","pith_inferences":["The subset-sum characterization suggests a canonical numerical signature for expected coverage: the vector of the s quantities Σ_{W∈C(S,k)}|Obs(G;W)|. Comparing signatures is finite and could be automated, though computing all |Obs(G;W)| for large s is computationally expensive and may require new algorithms.","The gadget construction shows that expected-value polynomials are extremely flexible; one could design graphs whose expected coverage matches a target polynomial on [0,1], which in principle allows constructing test instances that expose weaknesses in PMU placement heuristics.","If Conjecture 3.4 is true, degree-ℓ expected polynomials would be determined by only ℓ integer parameters, so equality and dominance questions would collapse to small linear-algebra comparisons; the partial Theorem 3.7 is evidence that the conjecture holds at least for quadratics with a large negative linear coefficient.","The counterexample to the high-degree heuristic suggests a testable refinement: on random or structured grids, compare placements chosen by maximum degree versus placements chosen by maximum single-sensor observed set size, and check which achieves better expected coverage at intermediate failure probabilities."],"forward_implications":["Two sensor placements of the same size give identical expected coverage at every failure probability if and only if their k-subset observation totals agree for all k, so equality of networks can be decided by a finite integer comparison.","Attaching Ψ and Φ gadgets to a network can force all coefficients of q^k with k≥2 to arbitrary integer values, while the constant term stays at |Obs(G;S)| and the linear coefficient stays nonpositive.","A placement A is at least as good as a placement B at every q whenever each k-subset total of A is at least the corresponding total of B, and strictly better when some total is strictly larger.","The necessary-and-sufficient linearity condition (11) shows that linear expected coverage is equivalent to every k-subset total being exactly the binomial multiple of the sum over single sensors, generalizing the earlier sufficient condition of [4].","In the constructed graph family, minimum power dominating sets of degree-3 vertices are strictly better under fragile power domination than minimum sets containing high-degree clique vertices, so the low-degree avoidance heuristic is not valid in this random-failure setting."],"supporting_citations":[{"why":"introduces the fragile power domination model, the expected value polynomial, and the earlier sufficient linearity condition that Theorem 3.3 refines","marker":"[4]"},{"why":"defines the power domination process as a domination step followed by zero forcing and supplies the observation about avoiding low-degree vertices that the Section 4 heuristic challenges","marker":"[11]"},{"why":"establishes the fort-entrance condition used to derive lower bounds on minimum power dominating sets in Proposition 4.2","marker":"[6]"},{"why":"introduces forts in the zero forcing setting, which the paper uses as the obstruction structure in its arguments","marker":"[10]"},{"why":"provides the Bernstein-type basis used in Theorem 3.1 to identify polynomial coefficients with k-subset observation sums","marker":"[3]"},{"why":"supplies the binomial theorem used in the derivations of Theorems 3.3 and 3.6","marker":"[7]"}],"fun_headline_variants":["Fragile PMU failures: expectation equals subset sums","When do power grids fail alike? Check subset sums","Sensor failure math: subset sums determine coverage","Expected grid coverage from subset sums of observations","Fragile power domination: subset sums decide equal behavior"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof that the quadratic case of the conjecture works for polynomials with a sufficiently negative linear term relies on the asserted gadget count in identity (19), which is stated as 'straightforward to verify' without displaying the derivation; if that count is wrong, the replacement-graph argument for that direction collapses.","fun_headline_variants_meta":{"raw":{"variants":["Fragile PMU failures: expectation equals subset sums","When do power grids fail alike? Check subset sums","Sensor failure math: subset sums determine coverage","Expected grid coverage from subset sums of observations","Fragile power domination: subset sums decide equal behavior"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000277,"raw_usage":{"total_tokens":1613,"prompt_tokens":873,"completion_tokens":740,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":489,"completion_tokens_details":{"reasoning_tokens":666}},"tokens_in":489,"tokens_out":740,"duration_ms":8551,"temperature":1.0,"reasoning_tokens":666,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T15:53:48.292852+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Recompute identity (19) directly for small parameters, for example s=3, t=w=d=0, by enumerating every subset W of the sensor set S' in the gadget graph H and summing |Obs(H;W)|; the resulting polynomial should equal -$5q^{2}$-9q+14, and any mismatch would refute the gadget count on which Theorem 3.7 rests.","supporting_citations":[{"cited_title":"Power domination with random sensor failure","cited_arxiv_id":"2312.12259","evidence_quote":"introduces the fragile power domination model, the expected value polynomial, and the earlier sufficient linearity condition that Theorem 3.3 refines"},{"cited_title":"Haynes, S","cited_arxiv_id":null,"evidence_quote":"defines the power domination process as a domination step followed by zero forcing and supplies the observation about avoiding low-degree vertices that the Section 4 heuristic challenges"},{"cited_title":"Bozeman, B","cited_arxiv_id":null,"evidence_quote":"establishes the fort-entrance condition used to derive lower bounds on minimum power dominating sets in Proposition 4.2"},{"cited_title":"Fast and I","cited_arxiv_id":null,"evidence_quote":"introduces forts in the zero forcing setting, which the paper uses as the obstruction structure in its arguments"},{"cited_title":"D´ emonstration du th´ eor` eme de weierstrass fonde´ e sur le calcul des probabilit´ es","cited_arxiv_id":null,"evidence_quote":"provides the Bernstein-type basis used in Theorem 3.1 to identify polynomial coefficients with k-subset observation sums"},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"supplies the binomial theorem used in the derivations of Theorems 3.3 and 3.6"}],"review_version":1}