Typical $T$-free graphs

Bruce Reed; Yelena Yuditsky

REVIEW 3 major objections 4 minor 24 references

Typical $T$-free graphs

T0 review · 3 major / 4 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read For every tree T with alpha(T)>2, almost all T-free graphs admit a certifying partition into alpha(T)-1 parts.

desk verdict Strong structural theorem for typical T-free graphs, with the perfect-matching regime genuinely new but resting on an unpublished companion paper that needs to be made available. read the letter →

arxiv 2506.01067 v1 pith:YTCRCK7Y submitted 2025-06-01 math.CO

Bruce Reed , Yelena Yuditsky This is my paper

classification math.CO MSC 05C0505C3005C80

keywords T-freegraphsinducedsubgraphscertifyingpartitionstypicalstructureasymptoticenumerationP4-freetreeshereditaryproperties

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper proves that for every tree $T$ with $\alpha(T)>2$, almost every graph on $n$ vertices that avoids $T$ as an induced subgraph carries visible evidence of why it avoids it: its vertices can be partitioned into $\alpha(T)-1$ parts such that each part is $P_4$-free, the first part is the one with largest independence number, and every other part contains a clique as large as $T$. Such a partition certifies $T$-freeness, because $T$ itself cannot be partitioned into $\alpha(T)-1$ pieces of the required simple kinds. The theorem implies that the family of $T$-free graphs is dominated by graphs assembled from a small number of simple pieces plus arbitrary edges between pieces, enough to count them nearly exactly and to determine their typical chromatic number.

What carries the argument

The mechanism is an iterative pattern-counting argument. Theorem 3.1, quoted from the manuscript [R], supplies a first approximation: for almost every $T$-free graph one can delete $o(n)$ vertices and obtain a $T$-freeness witnessing partition whose parts have nearly equal size, each vertex has nearly the same neighbours inside its part as some vertex of a bounded core $B$, and each part is almost $P_4$-free. Over such a partition one studies 'patterns'—the graphs induced on the parts—and asks how many $T$-free graphs extend a pattern. A subset of a part is called dangerous when it could be completed to a copy of $T$ using pervasive copies in the other parts; Lemma 3.5 and Lemma 3.6 show that dangerous sets are 'choice-destroying' and reduce the number of extensions by an exponential factor. The proof repeatedly prunes patterns with too many dangerous sets, until only patterns whose parts literally satisfy the certifying conditions survive. Counting then uses Seinsche's lemma (every $P_4$-free graph is disconnected or co-disconnected) and Bell-number bounds to estimate the number of allowed patterns inside each part.

What would settle it

For a fixed tree $T$ with $\alpha(T)>2$, count the $T$-free graphs on $n$ vertices that admit no $T$-freeness certifying partition. Theorem 1.2 predicts this number is $o(2^{(1-1/(\alpha(T)-1))n^2/2})$; finding a single $T$ for which it is $\Omega(2^{(1-1/(\alpha(T)-1)+\epsilon)n^2})$ for some $\epsilon>0$ would refute the theorem. A more direct check is whether Theorem 3.1 of [R] holds for every graph $H$ with witnessing partition number at least two, since the approximation step depends on it; the trees $P_6$ and $M_6$ are natural first test cases.

Watch

Extended reading notes

Core claim

The central claim is Theorem 1.2: for any tree $T$ with $\alpha(T)>2$, almost all $T$-free graphs $G$ have a $T$-freeness certifying partition of $V(G)$ into $\alpha(T)-1$ parts $X_1,\dots,X_{\alpha(T)-1}$ satisfying (i) $\alpha(G[X_1])\ge \alpha(G[X_i])$ for $i\ge 2$, (ii) each $G[X_i]$, $i\ge 2$, contains a clique of size $|V(T)|$ and $G[X_1]$ contains a clique or a stable set of size $|V(T)|$, and (iii) every $G[X_i]$ is $P_4$-free. The partition certifies $T$-freeness because no such piece can host the corresponding part of $T$ in any partition of $T$ into $\alpha(T)-1$ parts. The paper sharpens this by classifying the possible certifying partitions according to a taxonomy of $T$: for example, when $T$ has no perfect matching and is not a subdivided star the parts are cliques; when $T$ is a spiked star the parts are complements of matchings; when $T$ is a double star other than $P_6$ one part is a join of stable sets and clique-plus-vertex graphs while the rest are cliques.

Load-bearing premise

The argument takes as a black box the unpublished results listed as Theorem 24 and Claim 34 of the manuscript [R], which say that after deleting $o(n)$ vertices almost every $T$-free graph already has an approximate certifying partition with a bounded core; if those results fail in the stated generality, the proof of the cases where $T$ has a perfect matching collapses.

Editorial extensions

If this is right

The number of $T$-free graphs on $n$ vertices is $2^{(1-1/(\alpha(T)-1))n^2/2+O(n\log n)}$, strengthening the earlier upper bound of [PS92] and matching the clique-partition lower bound up to the polynomial factor.
For trees without a perfect matching that are not subdivided stars, almost every $T$-free graph splits into $\alpha(T)-1$ cliques.
For spiked stars, almost every $T$-free graph splits into $\alpha(T)-1$ parts each inducing the complement of a matching; for double stars other than $P_6$, into cliques plus one part built from stable sets and clique-plus-vertex pieces.
The same structure is used in a follow-up paper to show that almost every $T$-free graph has chromatic number equal to its clique number.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The pattern-pruning machinery is set up for arbitrary graphs $H$, so the same route may extend Theorem 1.2 beyond trees to every $H$ whose witnessing partition number is at least two; the paper itself completes only the tree case.
Because almost all $T$-free graphs are extensions of a small collection of $P_4$-free patterns on a balanced partition, one could sample typical $T$-free graphs by choosing a balanced partition, a random allowed pattern, and independent cross edges; this also yields a computational test of the theorem on small $n$.
The remaining gap between upper and lower bounds in the counting corollaries is almost entirely a Bell-number factor, so improved Bell-number estimates would immediately tighten the enumeration of $T$-free graphs in the perfect-matching classes.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Dear [Colleague],

The short version: this is a serious paper with a real theorem—for every tree T (except an edge), almost every induced-T-free graph has a certifying partition into alpha(T)-1 parts that proves it is T-free. That unifies a string of earlier results (triangles, cycles, cliques) and gives near-tight enumeration bounds. The genuinely new territory is the perfect-matching regime: M6, even paths, spiked stars, double stars, and the full taxonomy of certifying partitions. The no-perfect-matching cases are credited to Balogh–Butterfield, which is honest and correct.

The paper is well organized: a taxonomy of trees, a reduction to patterns, and a long but structured elimination argument in Section 4. Claims 2.1–2.10 are careful and checkable. The authors also state plainly which results are new and which are imported. That counts for something.

The soft spot is not in the internal logic; it is the foundation. Theorem 3.1 (from [R]) is stated for every graph H with wpn(H) >= 2, and Lemma 2.21 (Claim 34 of [R]) is used to control double counting in several places. But [R] is an unpublished manuscript whose title says 'when H is a cycle.' If Theorem 24 or Claim 34 do not actually have the claimed generality, the perfect-matching case—the core new contribution—is not yet proved. The paper gives no proof sketch of these inputs, so a referee cannot verify the preprint on its own. This is a real verification burden, not a manufactured flaw. I agree with the stress-test note here. There are also some LaTeX garbles in the counting formulas, but the intended arguments are clear enough.

That said, the paper is not circular: the cited results give an approximate partition after deleting o(n) vertices, and the paper's work is to kill the exceptional vertices and pin down the exact shapes. No sign of data-fitting or invented entities.

Who is this for? Researchers in extremal and random graph theory who want the typical structure of T-free graphs for all trees. It deserves a serious referee. My recommendation: send it to review, but require the authors to either make [R] available or prove the needed statements in an appendix. The theorem is important enough that the referee time is justified.

Referee Report

3 major / 4 minor

Summary. This paper studies the typical structure of graphs on n vertices that contain no fixed tree T as an induced subgraph. The main result (Theorem 1.2) states that for every tree T with alpha(T)>2, almost every T-free graph has a so-called T-freeness certifying partition into alpha(T)-1 parts, each part being P4-free and with additional part shapes depending on T. The authors classify the certifying partitions for each tree class (Lemmas 2.14-2.20), derive matching asymptotic bounds for the number of T-free graphs (Corollaries 2.24, 2.30, 2.37, 2.39, 2.41), and reduce the problem, for trees with perfect matchings, to an approximate partition theorem (Theorem 3.1) plus a sequence of pattern-refinement arguments in Section 4.

Significance. If the stated results are correct, the paper would resolve the typical structure and counting problem for induced subtrees in a strong form: it extends earlier work on cliques and cycles to all trees, gives bounds that are tight up to polynomial factors, and provides a structural classification that is used in the companion paper for the chi=omega result. The proof strategy is coherent and detailed: it combines a useful taxonomy of trees with a modular pattern-counting framework, and the lower-bound constructions via Bell numbers and matching counts are natural. The principal weakness is that the manuscript is not self-contained in its two central external inputs, both attributed to an unpublished manuscript by the first author; this makes the verification of the main theorem and of several counting corollaries conditional on unpublished work.

major comments (3)

[§3.1, Theorem 3.1] Theorem 3.1 is stated for every graph H with wpn(H)>=2, but the cited reference [R] is titled 'when H is a cycle' and no proof or proof sketch is given in this paper. Section 4 applies Theorem 3.1 to every tree T with alpha(T)>2, including the perfect-matching trees that constitute the main new regime of Theorem 1.2. If Theorem 24 of [R] is only proved for cycles, or if its full generality is not verified here, then the proof of Theorem 1.2 for perfect-matching trees is not established. The authors should either include a proof of Theorem 3.1 or make the relevant part of [R] available and verify that its hypotheses cover all trees used in Section 4.
[§2.3, Lemma 2.21] Lemma 2.21, cited to [R] Claim 34, is used without proof to control double counting in Corollaries 2.24, 2.30, 2.37, 2.39, 2.41 and throughout Section 4. The printed statement is unclear: the condition 'The graphs in F_i are P4-free or have girth at least five' appears to apply to all i, while the minimum-degree condition applies only to i>1; it is not verified for the specific families used here (P4-free graphs, complements of matchings, and the join-type part shapes of Lemmas 2.16-2.20). This lemma is load-bearing for the counting upper bounds; a proof or precise reference with a statement matching the usage is required.
[§2.3, Corollaries 2.30-2.41 and §1 overview] Several displayed asymptotic formulas are corrupted by LaTeX errors and cannot be checked as printed. Examples include the '2 n2n2/4' bound in the overview (Section 1), the expression in Corollary 2.30 ('n e(alpha(T)-1) lnn) n alpha(T)-1'), the formulas in Corollary 2.32 ('n ew n 2 e sqrt(nw)'), and the formula in Corollary 2.33. Since these formulas are the paper's headline counting bounds, they must be corrected before the claims can be verified.

minor comments (4)

[Throughout] The LaTeX spacing of graph names is inconsistent: 'P 4-free', 'P 4', 'P 3', 'M 6' appear with spurious spaces in many places, which makes the text harder to read; please use consistent math-mode formatting.
[§2.1.2, Observation 2.6] The partition list for P6 contains the entry '(P3,P3)' three times in a row; please check whether the intended list has a typo or whether the repetition is deliberate.
[§2.3, Lemma 2.28] In the proof for i=4, the sentence 'if we add v_{k+1} to a component which is not a stable set in G' should likely read 'component in the complement'; the surrounding case analysis is hard to follow as printed.
[References] The entries [R] and [RY] are listed as 'To be submitted'; given the central role of [R], the authors should update these entries or provide a version of record, ideally with a URL or repository link.

Circularity Check

2 steps flagged · score 4.0 of 10

The perfect-matching case of Theorem 1.2 is handed to an unpublished, self-authored theorem ([R] Theorem 24) whose stated scope is cycles, so the central derivation chain for that case terminates in a load-bearing self-citation rather than an internally proved result.

self citation load bearing [Section 3.1, Theorem 3.1; also References entry [R]]
"Theorem 3.1 ( [R] Theorem 24). For every graph H with wpn(H) ≥ 2 and constant ϵ > 0, there are ρ = ρ(H, ϵ) > 0 and b = b(H, ϵ) ∈ N, such that the following holds. For almost all H-free graphs G with V(G) = [n], there exists a partition (π1, π2, ..., πwpn(H)) of V(G) and a set B ⊂ V(G) of at most b vertices for which we can partition πi into Zi and π′i such that setting Z = ∪ wpn(H) i=1 Zi, (I) the partition (π′1, π′2..., π′wpn(H)) is an H-freeness witnessing partition of G−Z"

The perfect-matching case of Theorem 1.2 is built on this theorem as the 'starting point', yet it is attributed to [R], the first author's own unpublished manuscript, whose listed title restricts H to cycles ('when H is a cycle. To be submitted'). The paper uses Theorem 3.1 for every H with wpn(H)≥2, including every perfect-matching tree, and Lemma 2.21 ([R] Claim 34) repeatedly controls double counting. No proof or proof sketch of either is provided here. The central claim is therefore not established by the paper's own derivation chain; it reduces to a load-bearing self-citation with a stated scope narrower than the use made of it.
self citation load bearing [Section 2.3, Lemma 2.21; applied in Corollaries 2.30, 2.37, 2.39 and Section 4]
"We can apply the following result due to Reed([R] Claim 34) to show that the effect of such double counting is negligible."

The counting upper bounds for perfect-matching trees call Lemma 2.21 repeatedly (e.g., Corollaries 2.30, 2.37, 2.39 and the Section 4 arguments), and the only proof offered is a citation to the same unpublished [R]. Thus the claimed growth rates inherit their double-counting control from the same self-citation; if Claim 34 does not cover the partition families used here (complements of matchings, P4-free parts, etc.), the bounds are unsupported. This is not a minor reference: the double-counting control is essential when summing over all certifying partitions, and the paper supplies no independent derivation.

full rationale

The paper's taxonomy and partition characterizations in Section 2 are internal and do not assume Theorem 1.2: they are proven from the structure of T and from Seinsche's P4-free characterization. The large-stable-set, two-far, one-far, and somewhat-far lemmas in Section 4 do substantial independent work: starting from [R]'s approximate partition with o(n) exceptional vertices, they eliminate the exception set and pin the exact partition shapes. No equation in the paper fits a parameter to the target theorem, and no quantity called a prediction is defined in terms of the output. The non-perfect-matching regime is explicitly derived from [BB11], an independent published source. The only circularity-type concern is the load-bearing use of the first author's unpublished [R], both through Theorem 3.1 (the approximate witnessing partition) and Lemma 2.21 (double-counting control), despite the reference being titled for cycles and 'to be submitted.' If [R] in fact contains the stated 'every H with wpn(H)≥2' result and Claim 34 in the needed generality, the paper's derivation is coherent conditional on that black box; if not, the perfect-matching case of Theorem 1.2 is unproved. That is a genuine load-bearing self-citation rather than a minor reference, but it is not a by-construction equivalence of Theorem 1.2 with its own inputs, so the score is 4 rather than higher.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

No empirical fits and no invented entities. The only 'parameters' in the proof are asymptotic constants such as epsilon, rho, b, and mu, chosen sufficiently small or large by standard arguments; they are not fitted to data. The ledger instead records the external theorems on which the perfect-matching case rests, notably two unpublished results by Reed.

assumptions (5)

standard math Ramsey's Theorem guarantees that each large part contains a clique or a stable set of size |V(T)|
Invoked in the remark after the definition in Section 1 to justify focusing on interesting partitions.
standard math Seinsche's characterization: P4-free graphs are exactly graphs whose complement is disconnected or that are disconnected
Lemma 2.11, used throughout the structural lemmas on parts.
domain assumption Theorem 3.1 (Reed, [R], unpublished): for every H with wpn(H)>=2, almost all H-free graphs have a near-certifying partition after deleting o(n) vertices
Load-bearing input for Sections 3 and 4; stated but not proved in this preprint.
domain assumption Lemma 2.21 (Reed, [R] Claim 34) bounding double counting of graph-partition pairs
Used in Section 2.3 and in Corollaries 2.24, 2.30, 2.32, 2.33, and 2.37; proof only appears in the unpublished [R].
standard math Bell number bounds of Berend and Tassa (Theorem 2.25)
Used to estimate numbers of partition-patterns in Section 2.3.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Typical $T$-free graphs." pith.science (2026). https://pith.science/paper/YTCRCK7Y

@misc{pith2026250601067,
  author       = {Pith},
  title        = {Pith review of: Typical $T$-free graphs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/YTCRCK7Y}},
  note         = {Machine review of arXiv:2506.01067}
}

abstract

We prove that for every tree $T$ which is not an edge, for almost every graph $G$ which does not contain $T$ as an induced subgraph, $V(G)$ has a partition into $\alpha(T)-1$ parts certifying this fact. Each part induces a graph which is $P_4$-free and has further properties which depend on $T$. As a consequence we obtain good bounds (often tight up to a constant factor) on the number of $T$-free graphs and show in a follow-up paper~\cite{RY} that almost every $T$-free graph $G$ has chromatic number equal to the size of its largest clique.

Figures

Figures reproduced from arXiv: 2506.01067 by the authors.

**Figure 1.** The graph M6. If T has a near perfect matching then almost every T-free graph has such a partition unless T is a subdivided star. If T is a subdivided star, then almost every T4 graph has either such a partition or a partition into a stable set and α − 2 cliques certifying it is T-free. Again we give a very short and very straightforward proof that these result follows from the main result in [BB11]. If T has a perf… view at source ↗

**Figure 2.** Partitions of M6. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_2.png] view at source ↗

**Figure 3.** Partitions of P6. Observation 2.6. P6 can be partitioned into the following graphs. • (P3, P3), (P3, P3), (P3, P3), (S3, S3),(S3, P3), and • (K2, P4), (K2, 2K2), (K2, P3 + S1), (S2, P4), (S2, 2K2), (S2, P3 + S1). Observation 2.7. Every tree T with at least 6 vertices and a perfect matching can be partitioned into α(T) − 3 edges and any of the following pairs of graphs. • (P3, P3), (P3, P3), (S3, S3),(S3, P3), and •… view at source ↗

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

24 extracted references · 22 canonical work pages

[1]

N. Alon, J. Balogh, B. Bollob\' a s and R. Morris. The structure of almost all graphs in a hereditary property. J. Combin. Theory B 101: 85-110 (2011)

work page 2011
[2]

Balogh, B

J. Balogh, B. Bollob\'as, and M. Simonovits. The fine structure of octahedron-free graphs, J. Combin. Theory Ser. B 101(2) (2011), 67--84

work page 2011
[3]

Balogh and J

J. Balogh and J. Butterfield. Excluding induced subgraphs: Critical graphs. Random Structures and Algorithms, 38(1-2): 100-120 (2011)

2011
[4]

Berend and T

D. Berend and T. Tassa. Improved bounds on Bell numbers and on moments of sums of random variables. Probability and Mathematical Statistics 30(2): 185-205 (2010)

work page 2010
[5]

Bollob\' a s and A

B. Bollob\' a s and A. Thomason, Hereditary and monotone properties of graphs. The mathematics of Paul Erd o s II 14: 70-78 (1997)

work page 1997
[6]

Bondy and U.S.R Murty

J.-A. Bondy and U.S.R Murty. Graph theory. Graduate texts in mathematics, Springer, (2007)

work page 2007
[7]

N. G. de Bruijn. Asymptotic Methods in Analysis. Dover, New York, NY: 103-108 (1958)

work page 1958
[8]

A. Cayley. A theorem on trees. Quart. J. Pure Appl. Math. 23: 376-378 (1889); Collected Mathematical Papers Vol. 13, Cambridge University Press: 26-28, (1897)

work page

Show all 24 references

[9]

Chernoff

H. Chernoff. A note on an inequality involving the normal distribution. Ann. Probab., 9: 533-535 (1981)

1981
[10]

Erd os, D

P. Erd os, D. J. Kleitman, B. L. Rothschild. Asymptotic enumeration of K_n -free graphs, International Colloquium on Combinatorial Theory, Atti dei Convegni Lincei 17 (1976), 19--27

1976
[11]

Erd o s and G

P. Erd o s and G. Szekeres. A combinatorial problem in geometry. Compositio Math. 2: 463-470 (1935)

1935
[12]

J. Kim, D. K\" u hn, D. Osthus, T. Townsend. Forbidding induced even cycles in a graph: typical structure and counting. J. Comb. Theory, Ser. B 131 170-219 (2018)

2018
[13]

D.E. Knuth. The Art of Computer Programming, Volume 3, 2nd ed. Addison-Wesley: 73-75 (1997)

1997
[14]

Kolaitis, H.J

Ph.G. Kolaitis, H.J. Pr \"o mel and B.L. Rothschild. K_ +1 -free graphs: asymptotic structure and a 0 - 1 law. Trans. Amer. Math. Soc. 303: 637-671 (1987)

1987
[15]

J. Pach, B. Reed and Y. Yuditsky. Almost all string graphs are intersection graphs of plane convex sets. Proceedings of the 34th International Symposium on Computational Geometry (SoCG), (2018)

2018
[16]

H. J. Pr \"o mel and A. Steger. Excluding induced subgraphs: Quadrilaterals. Random Structures and Algorithms 2: 55-71 (1991)

1991
[17]

H. J. Pr \"o mel and A. Steger. Excluding induced subgraphs III: A general asymptotic. Random Structures and Algorithms 3(1): 19-31 (1992)

1992
[18]

H. J. Pr \"o mel and A. Steger. Almost all Berge graphs are perfect. Combinatorics, Probability and Computing 1: 53-79 (1992)

1992
[19]

H. J. Pr \"o mel and A. Steger. Excluding induced subgraphs II: Extremal graphs. Discrete Applied Mathematics 44(1-3): 283-294 (1993)

1993
[20]

F.P. Ramsey. On a Problem of Formal Logic. Proceedings of the London Mathematical Society, s2-30: 264-286 (1930)

1930
[21]

B. Reed. The Global Structure of a Typical Graph without H as an induced subgraph when H is a cycle. To be submitted
[22]

Reed and Y

B. Reed and Y. Yuditsky. The Asymptotic -Boundedness of Hereditary Families To be submitted
[23]

Scheinerman and J

E. Scheinerman and J. Zito. On the size of hereditary classes of graphs. J Combin Theory Ser B 61: 16-39 (1994)

1994
[24]

Seinsche On a property of the class of n -colorable graphs

D. Seinsche On a property of the class of n -colorable graphs. J. Combin. Theory Ser. B 16: 191-193 (1974)

1974

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

N. Alon, J. Balogh, B. Bollob\' a s and R. Morris. The structure of almost all graphs in a hereditary property. J. Combin. Theory B 101: 85-110 (2011)

work page 2011

[2] [2]

Balogh, B

J. Balogh, B. Bollob\'as, and M. Simonovits. The fine structure of octahedron-free graphs, J. Combin. Theory Ser. B 101(2) (2011), 67--84

work page 2011

[3] [3]

Balogh and J

J. Balogh and J. Butterfield. Excluding induced subgraphs: Critical graphs. Random Structures and Algorithms, 38(1-2): 100-120 (2011)

2011

[4] [4]

Berend and T

D. Berend and T. Tassa. Improved bounds on Bell numbers and on moments of sums of random variables. Probability and Mathematical Statistics 30(2): 185-205 (2010)

work page 2010

[5] [5]

Bollob\' a s and A

B. Bollob\' a s and A. Thomason, Hereditary and monotone properties of graphs. The mathematics of Paul Erd o s II 14: 70-78 (1997)

work page 1997

[6] [6]

Bondy and U.S.R Murty

J.-A. Bondy and U.S.R Murty. Graph theory. Graduate texts in mathematics, Springer, (2007)

work page 2007

[7] [7]

N. G. de Bruijn. Asymptotic Methods in Analysis. Dover, New York, NY: 103-108 (1958)

work page 1958

[8] [8]

A. Cayley. A theorem on trees. Quart. J. Pure Appl. Math. 23: 376-378 (1889); Collected Mathematical Papers Vol. 13, Cambridge University Press: 26-28, (1897)

work page