Path-count recurrence for generalized action graphs of the super Catalan numbers

Let GnG_n be the graph associated with S(0,n)S(0,n), and let Kℓ,v,nK_{\ell,v,n} denote the number of paths of length ℓ\ell in GnG_n that start at a vertex labeled vv and end at a vertex labeled nn. Path-count recurrence conjecture. The number of paths of length ℓ\ell in Gn+1G_{n+1} that start at a vertex labeled vv and end at a vertex labeled n+1n+1 is given by

Kℓ,v,n+1=∑i=0n+1−ℓ−v22iKℓ−1+i,v,n.K_{\ell,v,n+1} = \displaystyle \sum_{i=0}^{n+1-\ell-v} \frac{2}{2^i} K_{\ell-1+i,v,n}.

The recurrence is intended to compute the entries of the nn-tables without explicitly counting every path. It has been checked through the 77-tables, but the general case remains unproved.

References

Primary source

Drew Caldwell, Ali Cochran, Nathan Glisson, Bryce Jennings, Katy McDicken, Luke Proctor, Sarah Klanderman and Amelia Tebbe, “Catalan number sequences and generalized action graphs”, arXiv:2507.22719 (2025).

Progress summary

Refreshed
Claimed solved

The conjecture has a complete proof posted in discussion, but it has not been independently verified, so the general question is not settled.

Caldwell, Cochran, Glisson, Jennings, McDicken, Proctor, Klanderman, and Tebbe proposed the recurrence in 2025 for path counts Kℓ,v,nK_{\ell,v,n} in generalized action graphs of the super Catalan numbers. It gives the number of paths in Gn+1G_{n+1} from label vv to label n+1n+1 using counts from GnG_n.

Known results

  • The recurrence was checked computationally through the 77-tables (Caldwell et al., 2025).
  • The published preprint states it as a conjecture and supplies no general proof, counterexample, or correction.

Posted attempt

A posted argument claims a complete proof: it establishes integrality of the construction, derives vertex-level path-count identities, and obtains the recurrence by deleting the final edge of each path. The argument has not been independently verified.

Current status (as of August 2026): The recurrence remains unproved in the published record; a complete discussion proof is posted but unverified.

Sources

Solutions 1

ProofThis solution needs a summarySee full solutionHide full solution

Path-count recurrence for the super-Catalan action graphs

We prove Conjecture 5.5 of Caldwell, Cochran, Glisson, Jennings, McDicken, Proctor, Klanderman and Tebbe, arXiv:2507.22719v1.

For an actual vertex xx of GnG_n, let pn(x,r)p_n(x,r) be the number of directed paths of length rr from xx to a vertex labeled nn. For integers vv and rr, let

Kr,v,n=∑x∈V(Gn)label⁡(x)=vpn(x,r).K_{r,v,n}=\sum_{\substack{ x\in V(G_n)\\ \operatorname{label}(x)=v }}p_n(x,r).

Thus Kr,v,nK_{r,v,n} counts all directed paths of length rr in GnG_n whose initial vertex has label vv and whose terminal vertex has label nn, exactly as in the source.

Integrality of the construction

Definition 5.2 asks us, for each x∈V(Gn)x\in V(G_n) and each 0≤i≤n0\le i\le n, to add

22ipn(x,i)\frac{2}{2^i}p_n(x,i)

new vertices. We first prove that these numbers are integers. In fact, simultaneously with the construction of the graphs, we prove the stronger assertion

2r∣pn(x,r)for every x∈V(Gn) and every r≥0.(1)\begin{gathered} 2^r\mid p_n(x,r)\\ \text{for every }x\in V(G_n)\\ \text{ and every }r\ge 0. \end{gathered} \tag{1}

For G0G_0, the only nonzero path count is p0(x,0)=1p_0(x,0)=1, so (1) holds. Suppose that GnG_n has been constructed and satisfies (1). For i=0i=0, the required number of new vertices is 2pn(x,0)2p_n(x,0); for i≥1i\ge1, divisibility (1) shows that 21−ipn(x,i)2^{1-i}p_n(x,i) is an integer. Hence Gn+1G_{n+1} is well-defined.

It remains to verify (1) for Gn+1G_{n+1}. A newly added vertex has one path of length zero to label n+1n+1 and no positive-length path to that label, so the assertion is immediate for new vertices. Let xx be an old vertex and let r≥1r\ge1. Delete the final edge of a path of length rr from xx to a new vertex. If its old endpoint is zz, then the construction supplies, for every ii, exactly 21−ipn(z,i)2^{1-i}p_n(z,i) choices for the final edge. Concatenating the initial path of length r−1r-1 with the path of length ii counted at zz gives

pn+1(x,r)=∑i=0n22ipn(x,r−1+i).(2)p_{n+1}(x,r) =\sum_{i=0}^{n}\frac{2}{2^i}p_n(x,r-1+i). \tag{2}

Every summand in (2) is divisible by 2r2^r: when r−1+i>0r-1+i>0, this follows from

2r−1+i∣pn(x,r−1+i),2^{r-1+i}\mid p_n(x,r-1+i),

and the factor 21−i2^{1-i}; the sole remaining case is r=1,i=0r=1,i=0, where the factor 22 gives the same conclusion. Thus (1) holds for Gn+1G_{n+1}, completing the simultaneous induction.

The aggregate recurrence

Now fix n≥0n\ge0, 0≤v≤n+10\le v\le n+1, and ℓ≥1\ell\ge1. Every path of length ℓ\ell in Gn+1G_{n+1} from a vertex labeled vv to a vertex labeled n+1n+1 has a unique last edge x→yx\to y, where x∈V(Gn)x\in V(G_n) and yy is one of the new vertices. Removing that last edge leaves a path of length ℓ−1\ell-1 from a vertex labeled vv to xx. Conversely, any such old path, followed by any new edge out of xx, gives exactly one path counted by Kℓ,v,n+1K_{\ell,v,n+1}. Expanding the construction count at xx therefore gives

Kℓ,v,n+1=∑i=0n22i∑P: ∣P∣=ℓ−1label⁡(start⁡P)=vpn(end⁡P,i).(3)\begin{aligned} &K_{\ell,v,n+1}\\ &=\sum_{i=0}^{n}\frac{2}{2^i} \sum_{\substack{P:\ |P|=\ell-1\\ \operatorname{label}(\operatorname{start}P)=v}} p_n(\operatorname{end}P,i). \end{aligned} \tag{3}

For fixed ii, the pairs counted by the inner sum consist of a path PP of length ℓ−1\ell-1 followed by a path of length ii from its endpoint to a vertex labeled nn. Concatenation is a bijection from these pairs to the paths counted by Kℓ−1+i,v,nK_{\ell-1+i,v,n}. Hence (3) becomes

Kℓ,v,n+1=∑i=0n22iKℓ−1+i,v,n.(4)K_{\ell,v,n+1} =\sum_{i=0}^{n}\frac{2}{2^i}K_{\ell-1+i,v,n}. \tag{4}

Every edge in these inductively constructed graphs goes from a smaller label to a larger label. Consequently, a path from label vv to label nn has length at most n−vn-v. Thus

Kℓ−1+i,v,n=0wheneveri>n+1−ℓ−v.\begin{gathered} K_{\ell-1+i,v,n}=0\\ \text{whenever}\qquad i>n+1-\ell-v. \end{gathered}

Deleting the zero terms from (4) gives precisely the conjectured recurrence

Kℓ,v,n+1=∑i=0n+1−ℓ−v22iKℓ−1+i,v,n.(5)\boxed{ K_{\ell,v,n+1} =\sum_{i=0}^{n+1-\ell-v}\frac{2}{2^i}K_{\ell-1+i,v,n}. } \tag{5}

When n+1−ℓ−v<0n+1-\ell-v<0, the sum in (5) is empty and equals zero; the left side is also zero because no path with those parameters can exist. The case ℓ=0\ell=0 is the separately stated table boundary: it counts the newly added vertices themselves and is not part of this recurrence. This proves Conjecture 5.5.