Cochran's multiplier conjecture for generalized action graphs

From papers

Let (sn)(s_n) be an appropriate sequence of positive integers, and let GnG_n be the action graph associated with sns_n. For n1n\geq 1, write znz_n for the number of new vertices labeled nn that must be added to the vertex labeled 00. Cochran's conjecture. The required number is

zn=sni=1n1zisni.z_n=s_n-\sum_{i=1}^{n-1}z_i\cdot s_{n-i}.

This formula is intended to determine the multiplier on the edge from the vertex labeled 00 to the vertex labeled nn in condensed notation, thereby giving a construction of generalized action graphs for any sequence satisfying the stated property. The supplied text gives no resolution status.

Progress summary

Open

The proposed counting rule has been proved to build these graphs when it works, but no source found proves that every valid graph must obey it.

Cochran’s conjecture asserts that the number of new vertices attached to the root at level nn is determined by the recurrence zn=sni=1n1zisniz_n=s_n-\sum_{i=1}^{n-1}z_i s_{n-i}. The available literature attributes this conjecture to Cochran but gives no posing date.

Known results

  • Generalized action graphs require s0=1s_0=1 and satisfy further structural constraints, including s2s12s_2\geq s_1^2 (Klanderman, McDicken, and Tebbe, 2025).

July 2025 sufficient-condition result

Klanderman, McDicken, and Tebbe proved that if s0=1s_0=1 and positive integers znz_n satisfy the recurrence, then generalized action graphs exist. Their paper explicitly leaves open whether this condition is necessary; the scan found no later proof, counterexample, or verification.

Current status (as of August 2026): The recurrence is a sufficient construction condition, but its necessity for all generalized action graphs remains open.

Sources
Sources & referencesView supporting material

Primary source

Sarah Klanderman, Katy McDicken and Amelia Tebbe, “Conditions for building generalized action graphs from sequences”, arXiv:2507.22861 (2025).

Solutions 1

Proof

Cochran's action-graph multiplier conjecture and a complete characterization

Source and version. Sarah Klanderman, Katy McDicken, and Amelia Tebbe, Conditions for Building Generalized Action Graphs From Sequences, The PUMP Journal of Undergraduate Research 9 (2026), 158–179, formulate Cochran's multiplier conjecture as Conjecture 3.3 and explicitly leave its necessity direction open. The earlier arXiv version labels the same assertion Conjecture 5.1. Crucially, the final published Definition 1.1 specifies directed, labeled, rooted trees; the older preprint only says “graphs.” The rooted-tree hypothesis is essential below.

We prove the conjecture and obtain the exact necessary-and-sufficient condition for a positive integer sequence to admit generalized action graphs. In particular, the auxiliary multipliers need only be nonnegative, not strictly positive as assumed in the sufficient direction of the source's Theorem 1.1.

1. The published axioms

Let

s0,s1,s2,Z>0s_0,s_1,s_2,\ldots\in\mathbb Z_{>0}

and let G0,G1,G2,G_0,G_1,G_2,\ldots be a sequence of generalized action graphs in the sense of the published Definition 1.1. Thus:

  1. Each GnG_n is a directed, labeled, rooted tree. The tree G0G_0 has s0s_0 vertices labeled 00 and no edges, and GnG_n extends Gn1G_{n-1} by adding exactly sns_n vertices labeled nn.
  2. For every vertex vv of GnG_n, its full descendant subtree is isomorphic, after the appropriate uniform shift of vertex labels, to one of the earlier action graphs GjG_j. The source explains this label-shift convention in Section 3.1 and makes the shift explicit in Theorem 1.1.
  3. Every leaf of GnG_n has label nn.

Since G0G_0 is a rooted tree with no edges, necessarily

s0=1.(1)s_0=1. \tag{1}

Write ρ\rho for the unique vertex labeled 00, and define

zi=#{v: v is a child of ρ and has label i}(i1).(2)z_i = \#\{v:\ v\text{ is a child of }\rho \text{ and has label }i\} \qquad(i\geq1). \tag{2}

The quantity ziz_i is independent of the choice of GnG_n with nin\geq i: vertices labeled ii first occur in GiG_i, and subsequent graphs extend the existing rooted tree without changing its previous edges.

2. Every root branch has a forced type

Fix n1n\geq1, and let vv be any child of ρ\rho whose label is ii. The descendant subtree rooted at vv is, by the second axiom, a copy of some GjG_j with all labels shifted upward by ii. The leaves of GjG_j have label jj by the third axiom, so the leaves of this shifted copy have label i+ji+j.

Every leaf of a full descendant subtree in a rooted tree is also a leaf of the whole tree. Therefore those same leaves have label nn in GnG_n. It follows that

i+j=n,and henceDescGn(v)Gniwith all labels shifted by i.(3)i+j=n, \qquad\text{and hence}\qquad \operatorname{Desc}_{G_n}(v) \cong G_{n-i} \quad\text{with all labels shifted by }i. \tag{3}

The number of vertices labeled nin-i in GniG_{n-i} is exactly snis_{n-i} by the first axiom. Consequently, the branch below each root child labeled ii contains precisely

sni(4)s_{n-i} \tag{4}

vertices labeled nn.

Because GnG_n is a rooted tree, the descendant subtrees of its distinct root children are disjoint, and together they contain every nonroot vertex. In particular, they partition all sns_n vertices labeled nn. Grouping these branches according to the label ii of their first vertex gives

sn=i=1nzisni(n1).(5)\boxed{ s_n=\sum_{i=1}^{n}z_i s_{n-i} \qquad(n\geq1). } \tag{5}

Since s0=1s_0=1, the summand with i=ni=n is simply znz_n. Solving (5) for that summand proves Cochran's exact conjectured formula:

zn=sni=1n1zisni(n1).(6)\boxed{ z_n=s_n-\sum_{i=1}^{n-1}z_i s_{n-i} \qquad(n\geq1). } \tag{6}

The argument uses the final published rooted-tree condition essentially: for a general directed graph, distinct root branches can share descendants, and the counting step (5) would not follow.

3. The complete generating-function classification

Define the formal power series

S(x)=n0snxn,Z(x)=n1znxn.(7)S(x)=\sum_{n\geq0}s_nx^n, \qquad Z(x)=\sum_{n\geq1}z_nx^n. \tag{7}

Multiplication of (5) by xnx^n and summation over n1n\geq1 yields

S(x)1=Z(x)S(x).(8)S(x)-1=Z(x)S(x). \tag{8}

Since S(0)=1S(0)=1, formal inversion is valid, and therefore

Z(x)=11S(x),S(x)=11Z(x).(9)\boxed{ Z(x)=1-\frac{1}{S(x)}, \qquad S(x)=\frac{1}{1-Z(x)}. } \tag{9}

Every znz_n counts actual root children, so

z1=s1>0,zn0(n2).(10)z_1=s_1>0, \qquad z_n\geq0\quad(n\geq2). \tag{10}

We next prove that these necessary conditions are also sufficient.

Suppose s0=1s_0=1 and that the uniquely determined coefficients of

Z(x)=11S(x)=i1zixi(11)Z(x)=1-\frac{1}{S(x)}=\sum_{i\geq1}z_ix^i \tag{11}

satisfy (10). For each i1i\geq1, introduce an alphabet AiA_i consisting of ziz_i distinguishable letters, each assigned weight ii. Alphabets with zi=0z_i=0 are simply empty. A word

w=a1a2ar,ajAij,w=a_1a_2\cdots a_r, \qquad a_j\in A_{i_j},

has weight

w=i1+i2++ir.(12)|w|=i_1+i_2+\cdots+i_r. \tag{12}

The empty word \varnothing has weight 00.

Construct GnG_n as follows: its vertices are all words of weight at most nn; the label of a vertex is its weight; its root is the empty word; and its directed edges are

wwa(aAi, w+in).(13)w\longrightarrow wa \qquad \bigl(a\in A_i,\ |w|+i\leq n\bigr). \tag{13}

Each nonempty word has exactly one parent, obtained by deleting its final letter. Thus GnG_n is a finite directed rooted tree. The ordinary generating function for all words by weight is

wxw=r0Z(x)r=11Z(x)=S(x).(14)\sum_w x^{|w|} = \sum_{r\geq0}Z(x)^r = \frac{1}{1-Z(x)} =S(x). \tag{14}

Hence precisely sis_i vertices have label ii, and passing from Gn1G_{n-1} to GnG_n adds exactly the sns_n words of weight nn. This is the first axiom.

For a word ww of weight ii, every descendant has the unique form wuwu, where uu is a word of weight at most nin-i. Deleting the prefix ww gives a directed rooted-tree isomorphism

DescGn(w)Gni,wuu,(15)\operatorname{Desc}_{G_n}(w) \longrightarrow G_{n-i}, \qquad wu\longmapsto u, \tag{15}

and the labels differ by exactly ii. This is the second axiom.

Finally, z1>0z_1>0 guarantees the existence of a letter of weight 11. Every word of weight strictly less than nn can therefore be extended by one such letter, while no word of weight nn can be extended within GnG_n. Consequently,

w is a leaf of Gnw=n.(16)w\text{ is a leaf of }G_n \quad\Longleftrightarrow\quad |w|=n. \tag{16}

This is the third axiom. We have therefore proved the sharp equivalence

(sn)n0 admits generalized action graphss0=1,[x](11S(x))>0,[xn](11S(x))0(n2).(17)\boxed{ \begin{aligned} &(s_n)_{n\geq0}\text{ admits generalized action graphs} \\ &\quad\Longleftrightarrow\quad s_0=1,\quad [x]\left(1-\frac1{S(x)}\right)>0, \quad [x^n]\left(1-\frac1{S(x)}\right)\geq0 \quad(n\geq2). \end{aligned} } \tag{17}

For the positive sequences considered in the source, the strict inequality for the coefficient of xx is automatic because it equals s1s_1.

4. Zero multipliers and structural uniqueness

The source's Theorem 1.1 assumes zi>0z_i>0 for every i1i\geq1. Condition (17) shows that this is stronger than necessary: a root can have no child with some particular label.

For example, take

sn=1(n0).(18)s_n=1\quad(n\geq0). \tag{18}

Then

S(x)=11x,Z(x)=11S(x)=x,(19)S(x)=\frac1{1-x}, \qquad Z(x)=1-\frac1{S(x)}=x, \tag{19}

so

z1=1,zi=0(i2).(20)z_1=1, \qquad z_i=0\quad(i\geq2). \tag{20}

The corresponding action graph GnG_n is simply the directed path

01n,(21)0\longrightarrow1\longrightarrow\cdots\longrightarrow n, \tag{21}

which satisfies all three published axioms. Thus zero higher multipliers must be permitted in the complete characterization.

More generally, (3) applies to every vertex, not only root children. Inside the subtree below a vertex labeled jj, the number of children whose labels are j+ij+i is exactly ziz_i. Therefore every admissible action graph is, up to a rooted label-preserving tree isomorphism, the weighted-word tree (13). The multiplier sequence is uniquely determined by (6), and the entire graph structure is determined by those multipliers.

This proves Cochran's multiplier conjecture, completes the missing necessity direction of the published paper, weakens its sufficiency hypothesis from strict positivity to the optimal nonnegativity condition, and gives the requested complete classification of admissible sequences.

0 endorsements
Shivam Patel ·