REVIEW 2 major objections 3 minor 18 references
On Spectra of $\mathbb{T}$-Gain Digraphs
T0 review · 2 major / 3 minor · reviewed 2026-08-28 · deepseek-v4-flash
Pith's one-line read For directed graphs with unit complex gains, the paper proves that cycle balance is equivalent to cospectrality with the underlying digraph, and that the spectral radius never exceeds the digraph's except under μ-balanced assignments.
desk verdict A genuinely useful unifying framework that is one convention away from being correct; the switching rule as written is inconsistent with the matrix similarity and leaves three central theorems unproved until fixed. read the letter →
The pith
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
The reading
What carries the argument
The central object is the $\mathbb{T}$-gain digraph $\Phi=(D,\varphi)$ with gain adjacency matrix $A(\Phi)_{ij}=\varphi(v_i,v_j)$ when $(v_i,v_j)$ is an arc and $0$ otherwise; the gains lie in the circle group $\mathbb{T}=\{z:|z|=1\}$ and opposite arcs are independent. The argument-carrying mechanism is switching: conjugation by a diagonal unitary $S=\operatorname{diag}(\sigma(v_1),\dots,\sigma(v_n))$ with $\sigma:V\to\mathbb{T}$, giving $A(\Phi_\sigma)=S A(\Phi) S^{-1}$. Cycle gains $\varphi(C)$ and elementary subdigraphs (vertex-disjoint unions of directed cycles) convert the determinant into a sum over cycle covers, which is what makes balance visible in the characteristic polynomial. A potential function $\psi:V\to\mathbb{T}$ is the switching datum intended to send every arc gain to $1$ under the reciprocal convention, and $\mu$-balance generalizes balance to a constant cycle gain $\mu^{|C|}$.
What would settle it
On the strongly connected digraph with vertices $v_1,v_2,v_3$ and arcs $(v_1,v_2),(v_2,v_3),(v_3,v_1),(v_2,v_1)$, set $\varphi(v_1,v_2)=i$, $\varphi(v_2,v_3)=-i$, $\varphi(v_3,v_1)=1$, $\varphi(v_2,v_1)=-i$. Both directed cycles (the triangle and the digon) then have gain $1$, so Theorem 2.6 predicts $\det(\lambda I-A(\Phi))=\det(\lambda I-A(D))$; direct computation settles this claim. Applying the printed product-form switching with the potential function constructed for this example will not produce the all-ones matrix, isolating the reciprocal-form assumption on which the proof depends.
Extended reading notes
Core claim
The paper's central discovery is that the non-Hermitian gain adjacency matrix $A(\Phi)$ obeys a determinant expansion over spanning elementary subdigraphs, $\det A(\Phi)=\sum_{E\in\mathcal{E}_n(D)}(-1)^{n-c(E)}\varphi(E)$, and that the coefficients of its characteristic polynomial enumerate elementary subdigraphs with their gains. From this it proves that $\Phi$ is cycle balanced if and only if $A(\Phi)$ and $A(D)$ are cospectral: balance forces every gain product in the expansion to be $1$, and cospectrality forces the trace of every power to match, so every closed walk—and therefore every directed cycle—must have gain $1$. For strongly connected digraphs this is equivalent, under the reciprocal switching convention used in the proofs, to switching equivalence with the all-ones gain and to the existence of a potential function. The spectral-radius theorem states $\rho(\Phi)\le\rho(D)$, with equality if and only if $\Phi$ is $\mu$-balanced (every cycle of length $k$ has gain $\mu^k$); a second bound gives $\rho(\Phi)\le\Delta^+$, with equality iff the underlying digraph is $\Delta^+$-out-regular and $\Phi$ is $\mu$-balanced. Unicyclic gain digraphs then have spectrum $\{0^{n-\ell}\}\cup\{\lambda:\lambda^\ell=g\}$, and for strongly connected digraphs the Laplacian has eigenvalue $0$ exactly in cycle balance while the signless Laplacian has eigenvalue $0$ exactly in cycle antibalance.
Load-bearing premise
The load-bearing premise is that switching multiplies each arc gain by $\sigma(v_i)\varphi(v_i,v_j)\sigma(v_j)^{-1}$ rather than by the product form printed in Definition 2.3, since only the reciprocal form lets a potential function turn every gain into $1$; if the printed form is taken literally, the proofs of the balance, switching, and cospectrality equivalence no longer go through.
Editorial extensions
If this is right
- For any $\mathbb{T}$-gain digraph, cycle balance is spectrally testable: comparing the characteristic polynomial of $A(\Phi)$ with that of $A(D)$ decides whether every directed cycle has gain $1$, no matter how the gains are distributed.
- Gains can only lower or preserve the spectral radius of a strongly connected digraph; the only gain assignments that preserve it are the $\mu$-balanced ones, so $\rho(D)$ is a sharp upper bound over all gain choices.
- In a unicyclic gain digraph, the tree arcs are spectrally invisible: the whole spectrum is the $(n-\ell)$-fold eigenvalue $0$ together with the $\ell$-th roots of the single cycle gain.
- For strongly connected digraphs, zero is a Laplacian eigenvalue exactly when the digraph is cycle balanced, and a signless Laplacian eigenvalue exactly when it is cycle antibalanced, so these two matrices detect balance by their stability.
- Because signed digraphs, mixed graphs, and complex unit gain graphs are special cases, the balance and spectral-radius results transfer automatically to those classes, recovering and unifying earlier criteria.
Reading between the lines
- The trace argument in Theorem 2.6 suggests a quantitative refinement: if only the first $s$ spectral moments of $A(\Phi)$ and $A(D)$ are known to agree, unbalanced cycles of length greater than $s$ could in principle remain undetected; full cospectrality, not finitely many moments, is what forces balance.
- The $\mu$-balance equality condition implies that extremal gain assignments on a fixed strongly connected digraph are organized by cycle-length parity: over $\mathbb{T}$, only the case $\mu=\pm1$ can occur when every digon carries conjugate gains, so balance/antibalance is the generic extremal regime.
- A natural next test is whether the strong-connectivity hypothesis in Theorems 2.7 and 3.7 can be relaxed to weaker connectivity hypotheses; the diamond example shows the hypothesis is not merely technical, and the paper leaves weakly connected digraphs open.
Formalized claims in Lean
-
Claim #1: The paper's central discovery is that the non-Hermitian gain adjacency matrix $A(\Phi)$ obeys a determinant expansion over spanning elementary subdigraphs, $\det A(\Phi)=\sum_{E\in\mathcal{E}_n(D)}(-1)^{n-c(E)}\varphi(E)$, and that the coefficients of its characteristic polynomial enumerate elementary subdigraphs with their gains. From this it proves that $\Phi$ is cycle balanced if and only if $A
/-- @claim 1 The paper's central discovery is that the non-Hermitian gain adjacency matrix $A(\Phi)$ obeys a determinant expansion over spanning elementary subdigraphs, $\det A(\Phi)=\sum_{E\in\mathcal{E}_n(D)}(-1)^{n-c(E)}\varphi(E)$, and that the coefficients of its characteristic polynomial enumerate elementary subdigraphs with their gains. From this it proves that $\Phi$ is cycle balanced if and only if $A -/ def central_claim : Prop :=
Signed reviews
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper develops the spectral theory of T-gain digraphs, in which every arc carries a complex unit gain and opposite arcs are unrestricted. It introduces gain adjacency, Laplacian, and signless Laplacian matrices, derives determinant and characteristic-polynomial expansions in terms of elementary subdigraphs, and uses them to characterize cycle balance through cospectrality with the underlying digraph. For strongly connected underlying digraphs it characterizes balance via switching equivalence, bounds the spectral radius by that of the underlying digraph and by the maximum out-degree, and identifies the equality cases with mu-balance. It also determines the spectra of T-gain directed cycles and unicyclic digraphs, and gives zero-eigenvalue characterizations for the Laplacian and signless Laplacian, with the spectra shown to lie in the closed right half-plane.
Significance. If the switching convention is corrected, the paper gives a genuinely unifying framework: signed digraphs, mixed graphs, complex unit gain graphs, and ordinary digraphs all appear as special cases. The determinant expansion and the trace-based cospectrality argument for Theorem 2.6 are self-contained and transparent, and the mu-balanced equality case is derived from the Horn-Johnson theorem rather than assumed. The paper also makes crisp falsifiable spectral predictions for unicyclic digraphs. The central flaw is a systematic reciprocal-versus-product inconsistency in the switching definition that propagates into several named theorems; the repair is routine and local, so the manuscript is suitable for major revision rather than rejection.
major comments (2)
- [Definition 2.3; Theorems 2.7, 3.7, 4.1, 5.3] The stated switching rule phi_sigma(v_i,v_j)=sigma(v_i)phi(v_i,v_j)sigma(v_j) is incompatible with the claimed similarity A(Phi_sigma)=S A(Phi) S^{-1}. For unit sigma, (S A(Phi) S^{-1})_{ij}=sigma(v_i)phi(v_i,v_j)sigma(v_j)^{-1}, and the two expressions agree only when sigma(v_j)=+/-1. The proof of Theorem 2.7 constructs psi with phi(v_i,v_j)=psi(v_i)^{-1}psi(v_j) and then identifies this with psi(v_i)psi(v_j); for unit psi this substitution is valid only when psi(v_i)^2=1. Under the stated product rule, switching by psi gives phi_sigma(v_i,v_j)=psi(v_j)^2 rather than 1, so the implication (i)=>(ii) is not proved as written. The same reciprocal convention is used in Lemma 3.4, in the equality case of Theorem 3.7 (where the correction should read phi(v_i,v_j)=mu sigma(v_i)sigma(v_j)^{-1}), in the switching similarity of Theorem 4.1, and in the proof of Theorem 5.3. The repair is to define phi_sigma(v_i,v_j)=sigma(v_i)phi(v_i,v_j)sigma(v_j)^{-1} and a potential by phi(v_i,v_j)=sigma(v_i)^{-1}sigma(v_j); with this convention the named results go through.
- [Example 2.8] The alleged contradiction in Example 2.8 is arithmetically wrong. As written, the gains on u->x, x->w, u->y, y->w are all zeta, so both paths u->x->w and u->y->w have product zeta^2, not zeta^2 and zeta^{-2}. In fact the displayed gains admit a potential function (for example, under the product convention, take psi(u)=1 and psi(x)=psi(y)=psi(w)=zeta? this gives psi(u)psi(x)=zeta, psi(x)psi(w)=zeta^2? no; a correct choice under the product convention is psi(u)=1, psi(x)=psi(y)=zeta, psi(w)=1, which yields phi(u,x)=phi(u,y)=zeta and phi(x,w)=phi(y,w)=zeta). Thus the example does not establish that strong connectivity cannot be dropped in Theorem 2.7, and the figure caption's assertion that the digraph has no potential function is false for the written gains. The example should be corrected or replaced.
minor comments (3)
- [Theorem 5.3] The sentence that the equivalence of (i), (ii), and (iii) 'follows directly from Theorem 2.7' is inaccurate, because Theorem 2.7 concerns A(Phi), not L(Phi). The implication (ii)=>(iii) does follow from the diagonal similarity S, and the proof's actual chain (iii)=>(iv)=>(ii) makes the theorem valid, but the sentence should be rephrased to avoid claiming an unproved converse for Laplacian cospectrality.
- [Introduction, paragraph on the circle group] The sentence stating that in T 'z^{-1}=z' should read 'z^{-1}=overline{z}' (or 'z^{-1}=z^{-1}' trivially); as written it is false for generic unit complex numbers and likely contributes to the switching-convention confusion.
- [Corollary 2.5 and surrounding text] There are several typographical slips, including 'F or' for 'For' and the phrase 'the function yields phi_sigma equiv 1' in Definition 2.3, which is false under the stated product rule; these should be fixed in the same pass as the switching convention.
Circularity Check
No circularity: the derivation chain is self-contained and uses only external matrix theorems; the only notable flaw is an inconsistent switching convention, which is a correctness issue, not a circular reduction.
full rationale
All central results are proved from within the paper: Theorem 2.4's determinant expansion is a direct permutation argument; Theorem 2.6 uses that expansion plus trace and triangle-inequality arguments; Theorem 2.7 uses strong connectivity and path gains; Theorem 3.7 uses the externally cited Perron-Frobenius theorem and the Horn-Johnson equality theorem, neither of which is authored by the present authors or encodes the target conclusion; Lemma 3.4 and Theorem 5.3 reduce to earlier proved equivalences. There are no fitted parameters, no quantities are renamed as predictions, and the reference list contains no self-citations. The skeptical note about Definition 2.3 is a genuine mathematical inconsistency: as written, phi_sigma(v_i,v_j)=sigma(v_i)phi(v_i,v_j)sigma(v_j) is not the same as (S A(Phi) S^{-1})_{ij}=sigma(v_i)phi(v_i,v_j)sigma(v_j)^{-1}, and the potential-function algebra in Theorems 2.7, 3.7, and 5.3 uses the reciprocal form. This is a correctness or typo issue that would be repaired by changing the switching convention; it is not circularity because no theorem is assumed as its own input.
Assumptions & free parameters
assumptions (5)
- standard math Perron-Frobenius theorem for nonnegative irreducible matrices (Theorem 3.1, from Horn and Johnson [9])
- standard math Horn-Johnson modulus theorem and its equality case (Theorems 3.5 and 3.6, from [9])
- standard math Gershgorin circle theorem (Theorem 5.6, from [9])
- standard math Newton identities tying trace powers to the characteristic polynomial (used in Theorem 2.6)
- domain assumption Finite loopless digraphs with at most one arc per ordered pair and gains in the circle group T
Cite this review
Pith. "Pith review of On Spectra of $\mathbb{T}$-Gain Digraphs." pith.science (2026). https://pith.science/paper/6APMXIXK
@misc{pith2026260823655,
author = {Pith},
title = {Pith review of: On Spectra of $\mathbbT$-Gain Digraphs},
year = {2026},
howpublished = {\url{https://pith.science/paper/6APMXIXK}},
note = {Machine review of arXiv:2608.23655}
}
abstract
A \(\mathbb{T}\)-gain digraph is a directed graph with complex unit gains on its arcs, allowing for no restrictions on oppositely directed arcs. This framework unifies various graph types, such as signed graphs, mixed graphs, complex unit gain graphs, digraphs, and signed digraphs. The gain adjacency, Laplacian, and signless Laplacian matrices are generally non-Hermitian with complex spectra. We develop the spectral theory of these matrices, extending classical results from signed digraphs and complex unit-gain graphs. For the gain adjacency matrix, we establish determinant and characteristic polynomial formulas, characterize cycle balance through switching equivalence and cospectrality with the underlying digraph. We further bound the spectral radius in terms of the underlying digraph and its maximum out-degree, with equality characterized by $\mu$-balance. As a consequence, we determine the spectra of $\mathbb{T}$-gain unicyclic digraphs. Moreover, we characterize cycle balance and antibalance for the Laplacian and signless Laplacian matrices, respectively, through the presence of a zero eigenvalue.
Figures
Reference graph
Works this paper leans on
-
[1]
B. D. Acharya. Spectral criterion for cycle balance in networks.Journal of Graph Theory, 4(1):1–11, 1980. 15
work page 1980
-
[2]
R. B. Bapat.Graphs and Matrices. Universitext. Springer, London, 2nd edition, 2014
work page 2014
-
[3]
M. A. Bhat and S. Pirzada. On equienergetic signed graphs.Discrete Applied Mathematics, 189:1–7, 2015
work page 2015
-
[4]
M. A. Bhat and S. Pirzada. Spectra and energy of bipartite signed digraphs.Linear and Multilinear Algebra, 64(9):1863–1877, 2016
work page 2016
-
[5]
R. Brualdi and H. J. Ryser.Combinatorial Matrix Theory, volume 39 ofEncyclopedia of Mathematics and Its Applications. Cambridge University Press, 1991
work page 1991
- [6]
- [7]
-
[8]
F. Harary. On the notion of balance of a signed graph.Michigan Mathematical Journal, 2(2):143–146, 1953
work page 1953
Show all 18 references
-
[9]
R. A. Horn and C. R. Johnson.Matrix Analysis. Cambridge University Press, Cambridge, 2nd edition, 2013
2013
-
[10]
M. Khan, R. Farooq, and J. Rada. Complex adjacency matrix and energy of digraphs.Linear and Multilinear Algebra, 65(11):2170–2186, 2017
2017
-
[11]
Liu and X
J. Liu and X. Li. Hermitian-adjacency matrices and hermitian energies of mixed graphs. Linear Algebra and its Applications, 466:182–207, 2015
2015
-
[12]
Mehatari, M
R. Mehatari, M. R. Kannan, and A. Samanta. On the adjacency matrix of a complex unit gain graph.Linear and Multilinear Algebra, 70(9):1798–1813, 2022
2022
-
[13]
Pirzada and M
S. Pirzada and M. A. Bhat. Energy of signed digraphs.Discrete Applied Mathematics, 169:195–205, 2014
2014
-
[14]
H. S. Ramane, H. B. Walikar, S. B. Rao, B. D. Acharya, P. R. Hampiholi, S. R. Jog, and I. Gutman. Equienergetic graphs.Kragujevac Journal of Mathematics, 26:5–13, 2004
2004
-
[15]
N. Reff. Spectral properties of complex unit gain graphs.Linear Algebra and its Applications, 436(9):3165–3176, 2012
2012
-
[16]
Wissing and E
P. Wissing and E. R. van Dam. Spectral fundamentals and characterizations of signed directed cycles.Journal of Combinatorial Theory, Series A, 187:Paper No. 105573, 2022
2022
-
[17]
Zaslavsky
T. Zaslavsky. Biased graphs. I. Bias, balance, and gains.Journal of Combinatorial Theory, Series B, 47(1):32–52, 1989
1989
-
[18]
A mathematical bibliography of signed and gain graphs and allied areas
Thomas Zaslavsky. A mathematical bibliography of signed and gain graphs and allied areas. The Electronic Journal of Combinatorics, # DS 8:1–524, 2018. 16
2018
Reviewed August 28, 2026 · model on record in the stance chip above.
Discussion (0). Continue with ORCID to comment.