{"id":"d40d1129-581e-4937-a1c9-ca9c614ef346","arxiv_id":"2412.01135","paper_version":1,"verdict":"ACCEPT","confidence":"HIGH","novelty_score":4.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"Known indistinguishability results for skeletal path compartmental models are reproved using graph-theoretic forest sums.","lead":"This paper gives new, purely graph-based proofs of two known results about when different linear compartmental models produce the same input-output dynamics. The proofs replace linear algebra with combinatorial counting of forests, offering a possible template for proving indistinguishability in other model families.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified.","rationale":"The paper is a careful reproof of two known indistinguishability results. Its contribution is a graph-theoretic proof strategy, not new theorems, and the authors are explicit about relying on Theorem 2.3 from prior work. I re-examined the two propositions that convert the forest-sum formula into elementary-symmetric-polynomial expressions. In Proposition 3.1, the only possible violation of the incoming-forest condition is the leak vertex having two outgoing edges, because ~P_n is otherwise a directed path with a pendant leak edge and no cycles. In Proposition 3.3, the directed path plus return edge gives exactly one cycle, the two-cycle at the last two vertices, and no vertex has outdegree greater than one; hence the only forbidden subsets are those containing both cycle edges. These structural claims are correct for the stated family. The indistinguishability maps are valid bijections: Theorem 3.4 maps the leak parameter to the return edge and leaves the path edge out of the penultimate vertex fixed, while Theorem 3.5 swaps the two path edges leaving the two leak positions and maps leak to leak. Under these maps, the subtracted products in the coefficient formulas correspond, the remaining parameter sets correspond, and the d_0 coefficient products correspond since multiplication is commutative. I also checked small cases in the text and found the examples consistent with the general formulas. No hidden assumption, indexing error, or circularity surfaced. The only non-reproved ingredient is the cited forest-sum formula, but reliance on a published theorem is standard and does not by itself create a correctness risk in the absence of evidence against it. The reader's identification of Theorem 2.3 as the weakest assumption matches my own reading, but I do not view this as a flaw requiring any change to the verdict.","tokens_in":14268,"tokens_out":15053,"duration_ms":133884,"concrete_test":"As a sanity check, symbolically compute the input-output equation for P_5 with a single leak at each vertex i = 1, 2, 3, 4 and for P_5 ∪ {5→4} directly from the compartmental matrix using differential elimination, then compare each result with the formulas in Propositions 3.1 and 3.3. If all computed coefficients match the elementary-symmetric-polynomial formulas for every leak position, the application of the forest-sum theorem and the enumeration of forbidden edge subsets are confirmed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I find no load-bearing flaw in the paper's central argument. The reproofs rest on Theorem 2.3, the forest-sum formula from prior work, and the structural enumeration of forest-violating edge subsets in Propositions 3.1 and 3.3. I checked these enumerations: in ~P_n with a leak at i, the only way an edge subset violates the incoming-forest condition is by containing both a_{0i} and a_{(i+1)i}, since every other vertex has at most one outgoing edge and the underlying graph is acyclic. In ~H = P_n ∪ {n→n−1}, every vertex has exactly one outgoing edge and the only cycle is the two-cycle between n−1 and n, so the only forbidden subsets are those containing both a_{n(n−1)} and a_{(n−1)n}. The permutation maps in Theorems 3.4 and 3.5 are bijections that send the excluded pairs to the corresponding excluded pairs, so the elementary-symmetric-polynomial expressions transform as claimed, and the d_0 coefficients match. The only external load-bearing ingredient is Theorem 2.3, which is cited from published work and is standard; I found no indication that it fails for these leak and cycle models. Thus the central claim appears sound.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper gives a graph-theoretic reproof of two permutation indistinguishability results for skeletal path linear compartmental models that were originally proved by Bortner and Meshkat using linear algebra. The key tool is Theorem 2.3, a forest-sum formula for the coefficients of input-output equations. The authors derive explicit coefficient formulas for a path model with a single leak (Proposition 3.1) and for a path model with a return edge from the last vertex to the penultimate vertex (Proposition 3.3). Using these formulas, they construct explicit parameter bijections that preserve the input-output equation, yielding the two indistinguishability theorems (Theorems 3.4 and 3.5). The proofs are elementary and rely only on elementary symmetric polynomial identities and a correct enumeration of the edge subsets that violate the incoming-forest condition in the associated graphs.","tokens_in":14487,"tokens_out":5507,"duration_ms":45892,"significance":"The central contribution is a clean proof-of-concept showing that the two known indistinguishability theorems follow directly from a graphical coefficient formula, avoiding the comparatively heavy linear algebra of Cramer's rule or differential elimination. The coefficient formulas in Propositions 3.1 and 3.3 are explicit and checkable, and I verified that the forest-violating edge subsets identified in the proofs are exactly the pairs of edges leaving the leak vertex (for the path-with-leak models) and the two-cycle edges (for the return-edge model). The permutation maps in Theorems 3.4 and 3.5 are simple bijections that send each excluded pair to the corresponding excluded pair, so the symmetric-polynomial expressions transform as claimed. The paper does not introduce new indistinguishability theorems, but it establishes a transparent combinatorial framework that is likely to be useful for future sufficient conditions based on graph structure.","major_comments":[],"minor_comments":[{"comment":"The input-output equation (8) omits the c_0 y term, and the statement defines c_j only for j=1,...,n-1, even though Example 3.2 correctly computes c_0=0. Please state explicitly that c_0=0 (or include j=0 in the formula range) and similarly note that d_i=0 for i>0 in both Proposition 3.1 and Proposition 3.3.","section":"Section 3, Proposition 3.1 and Eq. (8)"},{"comment":"The sentence in the proof of Proposition 3.1 that the only non-incoming forests are those containing both a_{0i} and a_{(i+1)i} is correct, but the justification should explicitly mention that the underlying graph is acyclic and that every vertex other than the leak vertex has at most one outgoing edge; a one-sentence enumeration would make the argument fully rigorous. The analogous claim in Proposition 3.3, namely that the only forbidden subsets are those containing both edges of the two-cycle, is also correct and could be stated with the same level of explicitness.","section":"Section 3, proofs of Propositions 3.1 and 3.3"},{"comment":"There is a small notational slip in the set complments: Example 3.2 writes Q_3 \\ {a_{03}a_{(3+1)3}} instead of Q_3 \\ {a_{03}, a_{43}}, and Proposition 3.3 writes Q_n \\ {a_{n(n-1)}a_{(n-1)n}} instead of a set with a comma between the two excluded parameters. Please fix these to avoid ambiguity.","section":"Section 3, Example 3.2 and Proposition 3.3"},{"comment":"In the sentence referring to \"the left-hand coefficient\" of the input-output equations, the plural \"coefficients\" would be more accurate, since the proof compares the full sets of left-hand side coefficients. This is a wording issue only.","section":"Section 3, Theorem 3.4 proof"}],"recommendation":"minor_revision","confidential_remarks":"The paper is a well-focused proof-of-concept for a combinatorial approach to indistinguishability, and the central arguments are sound. The only reservations are notational and expository. The manuscript cites the relevant prior work, including the source of the forest-sum formula and the original indistinguishability theorems, so there are no attribution concerns. The paper should be acceptable after the minor revisions listed above."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThis is a small but genuinely clean paper. It reproves two indistinguishability results for skeletal path models—results that Bortner and Meshkat already proved by linear algebra—using the forest-sum coefficient formula from the group's earlier work. The authors are upfront about the reproof; the stated goal is a proof-of-concept for graph-theoretic arguments.\n\nThe new content is Propositions 3.1 and 3.3, which express the input-output coefficients as an elementary symmetric polynomial minus a correction term for the single family of edge subsets that violate the incoming-forest condition. In the leak model that's the two edges leaving the leak vertex; in the return-edge model it's the two-cycles. I checked the enumerations and the bijections in Theorems 3.4 and 3.5; the maps swap exactly the forbidden pairs and preserve the path-edge product, so the proofs work. The writing is transparent about assumptions: the derivation leans on Theorem 2.3 from prior work, which is published and standard. No circularity—the target theorems are not assumed.\n\nThe soft spots are minor but real. The main theorems are not new, so the contribution is a proof technique rather than a result. The algebra is simple symmetric-polynomial manipulation; the graph theory only identifies the forbidden subsets. There's no attempt to extend the framework to other model families, so it remains a proof-of-concept. But it is exactly what it claims to be.\n\nThe stress-test note found no load-bearing flaw, and I agree. This is a correct, honest, well-scoped paper. It should go to peer review; a referee can verify it quickly and the likely outcome is acceptance with minor revisions. It would be a useful reference for anyone exploring graph-based approaches to indistinguishability, though I wouldn't cite it in my own work unless I were working directly on that technique.","headline":"A clean, honest reproof of two known indistinguishability results via the forest-sum formula; the explicit coefficient formulas are the real new content, and they check out.","tokens_in":15010,"tokens_out":5014,"would_cite":false,"duration_ms":41685,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C20","05C05","92B05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves that two families of skeletal path models—paths with a single leak and the path with a terminal reverse edge—have identical input–output equations up to parameter relabeling, established through a graph-theoretic…","keywords":["linear compartmental models","skeletal path models","permutation indistinguishability","incoming forests","elementary symmetric polynomials","input-output equations","graph-theoretic proofs","forest-sum formula"],"falsifier":"Compute the input–output equation of $M_2=(P_4,\\{1\\},\\{4\\},\\{2\\})$ by direct substitution and compare the coefficient of $y'$ with the value $\\sigma_3(Q)-a_{02}a_{32}\\sigma_1(Q\\setminus\\{a_{02},a_{32}\\})$; any mismatch would falsify Proposition 3.1. Alternatively, for $n=5$ and generic numerical parameters, apply the relabeling $\\Phi$ of Theorem 3.5 to the coefficients of $M_2$, and check whether they equal the coefficients of $M_4$; a single generic failure would falsify the theorem.","tokens_in":14094,"feed_emoji":"🔀","tokens_out":9006,"duration_ms":74808,"temperature":0.7,"pith_summary":"Linear compartmental models are used to describe how quantities move through compartments in pharmacokinetics, cell biology, and ecology. Two models are permutation indistinguishable when their input–output equations coincide after renaming parameters, meaning that experiments cannot tell them apart. This paper proves two such indistinguishability results for skeletal path models—directed paths with one leak, or with a terminal reverse edge—using a graph-based formula for the coefficients of the input–output equation. The proof shows that the coefficients are elementary symmetric polynomials minus a single correction term for the edge pairs that violate the incoming-forest condition, and that this correction term is preserved by an explicit parameter relabeling. The earlier linear-algebra proofs are thereby replaced by a combinatorial argument.","feed_headline":"Graph identities prove leaky path models indistinguishable","feed_subtitle":"Two skeletal path model families share input–output equations; graph structure alone decides.","key_machinery":"The load-bearing mechanism is the forest-sum formula of Theorem 2.3: it expresses each coefficient $c_j$ as the sum of productivities of all incoming forests on the augmented graph with $n-j$ edges, and each $d_i$ as the sum over incoming forests with a path from the input to the output in the edge-reduced graph. In these path models, an incoming forest is any acyclic subgraph in which no vertex has two outgoing edges. The key observation is that the only subsets of parameters that fail this condition are precisely the two-edge sets $\\{a_{0i},a_{i+1,i}\\}$ for a leak at $i$, or $\\{a_{n,n-1},a_{n-1,n}\\}$ for the terminal reverse edge. Therefore the full elementary symmetric polynomial $\\sigma_k(Q)$ decomposes into the coefficient plus a single product times $\\sigma_{k-2}$ of the remaining parameters, and this decomposition is what the indistinguishability maps preserve.","core_discovery":"The discovery is that, for skeletal path models, the input–output coefficients have a uniform graph-theoretic shape. For a path model $M_i=(P_n,\\{1\\},\\{n\\},\\{i\\})$ with parameter set $Q_i$, Proposition 3.1 gives $$c_j=\\sigma_{n-j}(Q_i)-a_{0i}a_{i+1,i}\\,\\sigma_{n-j-2}(Q_i\\setminus\\{a_{0i},a_{i+1,i}\\})$$ for the coefficient of $y^{(n-j)}$, while for the no-leak model $M_n$ with edge $n\\to n-1$ and parameter set $Q_n$, Proposition 3.3 gives the same expression with the terminal pair $a_{n,n-1}a_{n-1,n}$ playing the role of the forbidden pair. The reason is that in the augmented graph $\\tilde G$, the only edge subsets that violate the incoming-forest condition are exactly the subsets containing both edges that leave the leak vertex, or both edges of the two-cycle at the end of the path. The authors define bijections between parameter sets that swap the forbidden pairs, so the elementary symmetric polynomials and the correction terms match term by term. This proves Theorems 3.4 and 3.5 as purely combinatorial identities.","pith_inferences":["The “one forbidden pair” mechanism suggests a testable sufficient condition: any two graphs whose non-incoming-forest edge subsets can be paired by an edge bijection that preserves productivity sums will be permutation indistinguishable, and this could be checked on other structured families.","For models with several leaks, the correction term would become a sum over all pairs of edges leaving the same vertex; if that sum is symmetric under some parameter bijection, the same proof pattern would apply.","One could test the boundary of the claim by adding an extra edge to a path model; the enumeration of forbidden edge subsets would have to remain a matching set of pairs for the coefficient identity to survive.","The identity may also be read as an identifiability statement: the correction term is determined by the rest of the coefficients, so the graph structure fixes how much the leak can be moved without changing the model's behavior."],"forward_implications":["The penultimate-leak model $M_{n-1}$ and the no-leak model $M_n$ with edge $n\\to n-1$ have identical input–output equations under the parameter bijection that sends the leak parameter $a_{0,n-1}$ to the reverse-edge parameter $b_{n-1,n}$.","Any two single-leak path models $M_i$ and $M_k$ with $1\\le i