On Directed Graphs with the Same Sum over Arborescence Weights

We show that certain digraphs with the same vertex set but different arc sets have the same sum over the weights of all arborescences with a given root vertex. We relate our results to the Matrix-Tree Theorem and show how they provide a graphical approach for factoring matrix determinants.

💡 Research Summary

The paper investigates a surprising invariance property of weighted directed graphs (digraphs) that share the same vertex set but differ in their arc sets. Specifically, it proves that for a given root vertex, the total weight summed over all arborescences (directed spanning trees rooted at that vertex) remains unchanged under two elementary graph transformations: moving an arc to a different source while keeping the same target, and merging parallel arcs into a single arc whose weight is the sum of the original weights. These results are formalized as the “Moving‑Arc Theorem” (Theorem 2.1) and the “Combining‑Arcs Theorem” (Theorem 2.2).

The Moving‑Arc Theorem states that if an arc e = (a→b) is replaced by an arc e′ = (c→b) of identical weight, and neither pair (a,b) nor (c,b) is strongly connected in the respective graphs, then there exists a bijection between the sets of arborescences of the original and the modified graph that preserves the product of arc weights. The proof constructs this bijection by separating arborescences that contain the moved arc from those that do not, and showing that the presence of a strong connection would create a directed cycle, violating the arborescence property.

The Combining‑Arcs Theorem deals with parallel arcs e₁ and e₂ that share the same source a and target b. Replacing them by a single arc e_c with weight w(e_c)=w(e₁)+w(e₂) leaves the total arborescence weight unchanged. The argument pairs each arborescence containing e₁ with the unique arborescence containing e₂, noting that both contribute the same factor W(H)·(w(e₁)+w(e₂)) where H is the set of remaining arcs.

Both theorems are linked to the classical Matrix‑Tree Theorem, which relates the determinant of a Laplacian‑type matrix A to the sum of arborescence weights in a corresponding digraph Γ₀ that includes an extra root vertex 0. The paper shows that the graph operations above correspond precisely to elementary row/column operations on A (adding a multiple of one row/column to another, or merging columns), which are known to leave the determinant unchanged. Consequently, the invariance of the arborescence sum under the two graph transformations provides a purely combinatorial explanation for the determinant’s invariance under those linear algebraic operations.

Beyond the theoretical statements, the authors develop a constructive “vertex‑isolation” procedure that systematically reduces any weighted digraph associated with a nonsingular matrix to a collection of fully isolated graphs. The procedure works as follows:

1. Choose a vertex i (starting with the smallest index) and “root” the graph at i, thereby partitioning the set of arborescences into those that contain the arc (0→i) and those that do not.
2. For the subgraph where i is not the root, move every outgoing arc (i→k) to an arc (0→k). Because i is not strongly connected to any other vertex in that subgraph, the Moving‑Arc Theorem guarantees that the total arborescence weight is preserved.
3. If multiple arcs now point from 0 to the same vertex k, apply the Combining‑Arcs Theorem to merge them.

Repeating steps 1–3 for all vertices yields n! fully isolated digraphs, each consisting solely of arcs from the root 0 to the other vertices. The weight of each isolated digraph is simply the product of the corresponding arc weights, and the sum of these n! products equals the determinant of the original matrix.

The paper illustrates the method with two concrete examples. First, an upper‑triangular matrix A is transformed into a diagonal matrix A′ by moving arcs to the root and merging parallel arcs; the resulting product of diagonal entries matches det(A). Second, a more intricate 3×3 matrix M is processed through the vertex‑isolation algorithm, producing six isolated graphs whose weight products sum to the familiar expansion of det(M) obtained by the Leibniz formula.

The significance of these results lies in providing a graphical, combinatorial perspective on matrix determinants. The Moving‑Arc and Combining‑Arcs theorems give a clear visual interpretation of row/column operations, while the isolation procedure offers a systematic way to decompose a determinant into a sum over permutation‑like graph configurations. Potential applications include reliability analysis of networks (where arborescences represent functional configurations), electrical circuit analysis (where Laplacian determinants relate to effective resistance), and the study of Markov chains (where spanning arborescences appear in stationary distribution formulas). The work bridges algebraic graph theory and linear algebra, suggesting new tools for both theoretical investigations and practical computations of determinants.