---
title: Integral Spanning in Graph Theory
url: https://www.emergentmind.com/topics/integral-spanning
type: topic
---

# Integral Spanning in Graph Theory

Searching arXiv for recent and foundational papers related to “integral spanning” and adjacent graph-theoretic uses of “spanning” to ground the article.
First, I’ll look for papers specifically using “integral spanning” or closely related spanning structures in graph theory, combinatorics, and optimization.
“Integral spanning” is not presented in the literature as a single standardized formalism. As an *Editor’s term*, it usefully groups work on spanning trees, forests, subgraphs, walks, configurations, and polytopes in which integrality is the governing feature: the spanning object itself is required to be integral, a fractional relaxation is rounded to an integral spanning solution, or the relevant invariants appear as integral cohomology, integral roots, or integral formulas. Under that umbrella, the subject includes expected minimal spanning tree length via Tutte-polynomial integrals, half-integral LP rounding for edge- and arc-connectivity problems, degree- and leaf-constrained spanning trees, moduli of spanning subspace configurations, increasing spanning-forest enumerators, and spanning lattice polytopes [1501.03758] [2008.03327] [2011.06108] [1903.07579] [1610.05093] [1608.03166].

## 1. Conceptual scope of integrality in spanning problems

The literature uses integrality in several mathematically distinct senses. In combinatorial optimization, integrality usually means that the output is a genuine spanning tree or spanning subgraph rather than a fractional LP point. This is the setting of the Minimum $2$-Edge Connected Spanning Multisubgraph problem, the weighted minimum strongly connected spanning subgraph problem, and the $k$-edge-connected spanning subgraph problem, where half-integral or approximately optimal fractional solutions are converted into integral spanning objects with explicit approximation guarantees [2008.03327] [2011.06108] [2205.14978].

A second sense concerns integer-valued local constraints on spanning structures. Here the spanning object is still a tree or walk, but its existence is controlled by prescribed degree bounds, visit bounds, leaf counts, or designated internal vertices. Theorems in this direction give sufficient component-count conditions for bounded-degree spanning trees and bounded-visit spanning closed walks, prove contiguity of the attainable numbers of leaves in spanning trees, and analyze minimum-weight spanning trees in which a prescribed set of vertices must be internal [1702.06203] [2312.13674] [2310.05494].

A third sense is algebraic or topological. The moduli space of spanning configurations in $\mathbb{C}^k$ has an integral cohomology ring presented explicitly as a quotient of $\mathbb{Z}[\mathbf{x}_n]$, increasing spanning forests are counted by a polynomial with nonpositive integral roots, and spanning lattice polytopes are characterized by a gap-free $h^*$-vector [1903.07579] [1610.05093] [1608.03166].

A fourth sense is analytic or enumerative. The expected length of a minimal spanning tree can be written as an integral whose integrand is a polynomial determined by spanning subgraphs, and spanning tree generating functions for regular lattices are expressed through integral representations, Mahler measures, and hypergeometric functions [1501.03758] [1207.2815].

## 2. Integral formulas and enumeration for spanning trees

A foundational probabilistic result in this area is Steele’s formula for the expected length of the minimal spanning tree of a finite, connected, simple graph whose edge weights are independent $\mathrm{Uniform}[0,1]$ variables. If $T(G;x,y)$ is the Tutte polynomial, then
\[
\mathbb{E}[L(G)] = \int_0^1 \frac{1-t}{t} \frac{T_x\left(G; \frac{1}{t}, \frac{1}{1-t}\right)}{T\left(G; \frac{1}{t}, \frac{1}{1-t}\right)} \, dt .
\]
The paper "Polynomial representation for the expected length of minimal spanning trees" shows that the integrand is in fact a polynomial
\[
p_m(t) = -1 + \sum_{A \in S(G)} k(A)\, t^{|A|}(1-t)^{m-|A|}
      = -1 + \sum_{i=0}^m a_i t^i,
\]
of degree at most the number of edges $m$, with explicit coefficient formula
\[
a_i = \sum_{\ell=0}^i (-1)^{i-\ell} \binom{m-\ell}{m-i}\sum_{A\in S_\ell} k(A).
\]
The first coefficients are structural: $a_0=n$, $a_1=-m$, $a_2=0$, $a_3=c_3$, $a_4=c_4$, $a_5=c_5-k_3^5$, and $a_6=c_6+2k_4-c_{5,1}-k_{3,2}$ [1501.03758].

For complete graphs $K_n$, the first coefficients become
\[
a_0=n,\quad a_1=-\binom{n}{2},\quad a_2=0,\quad a_3=\binom{n}{3},\quad a_4=3\binom{n}{4},
\]
\[
a_5=12\binom{n}{5}-6\binom{n}{4},\qquad
a_6=60\binom{n}{6}-60\binom{n}{5}-2(n-5)\binom{n}{4},
\]
and the polynomial factors as $p_m(t)=(1-t)^{n-1}q(t)$ with $\deg q\leq \binom{n-1}{2}$. The paper explicitly notes that these patterns may be leveraged for asymptotic analysis and for deeper understanding of random spanning tree properties [1501.03758].

Enumeration with a prescribed integral substructure leads to a different but related theory. For a complete multipartite graph $K_{n_1,\dots,n_s}$ and a fixed spanning forest $F$ with components $T_1,\dots,T_c$, the count $\tau_F(G)$ of spanning trees containing $F$ reduces by contraction to
\[
\tau_F(G)=\tau(G/F).
\]
The paper "On enumeration of spanning trees of complete multipartite graphs containing a fixed spanning forest" derives a determinantal formula for $\tau_F(K_{n_1,\dots,n_s})$ using the Generalized Matrix Determinant Lemma and Jacobi’s formula for the derivative of a determinant. If $\alpha_p=\sum_{i=1}^s n_i(m_p-n_{ip})$, $\mathbf b=(n_1,\dots,n_s)^\top$, and $C_{ij}$ is the $(i,j)$-cofactor of a specified matrix $M$, then
\[
\tau_F(K_{n_1,\ldots,n_s})=
\frac{C_{ij}}{n_i(n-n_j)}\prod_{p=1}^c \alpha_p
\]
for any $i,j\in\{1,\dots,s\}$; the case $s=2$ reduces to the previously known bipartite formula [2602.03602].

A further analytic development is the spanning tree generating function (STGF) for a regular lattice $\mathcal L$,
\[
T_{\mathcal L}(z)=\log q+\frac{1}{(2\pi)^d}\int_{-\pi}^{\pi}\cdots\int_{-\pi}^{\pi}
\log\!\left[\frac{1}{z}-\Lambda(\vec k)\right]\, d^d k,
\]
whose value at $z=1$ is the spanning tree constant $\lambda_{\mathcal L}$. The derivative satisfies
\[
-\frac{dT_{\mathcal L}(z)}{dz}=P(\vec 0;z),
\]
where $P(\vec 0;z)$ is the lattice Green function, so that
\[
T_{\mathcal L}(z)=\log q-\int_0^z P(\vec 0;t)\,dt,\qquad
\lambda_{\mathcal L}=\log q-\int_0^1 \frac{P(\vec 0;z)-1}{z}\,dz .
\]
For the square lattice, the paper gives $\lambda_{\mathrm{sq}}=\frac{4G}{\pi}$, and for the standard two- and three-dimensional lattices it expresses STGFs as Mahler measures and hypergeometric functions [1207.2815].

## 3. Half-integrality, LP rounding, and integral spanning subgraphs

In survivable network design, integral spanning emerges through rounding the natural LP relaxations. For the Minimum $2$-Edge Connected Spanning Multisubgraph problem, the LP is
\[
\min\left\{\sum_{e\in E} c_e x_e \mid x(\delta(S))\ge 2,\ x_e\ge 0\right\},
\]
and the half-integral case is characterized by $x_e\in\{0,\tfrac12,1\}$. Carr and Ravi had shown that the integrality gap is at most $\frac43$ in this case. The paper "A $4/3$-Approximation Algorithm for the Minimum $2$-Edge Connected Multisubgraph Problem in the Half-Integral Case" gives a simpler proof via an extension of Lovász’s splitting-off theorem and, crucially, an explicit deterministic algorithm: given a half-integral LP solution $x$, one can in $O(n^2)$ time construct a $2$-edge connected spanning multisubgraph of cost at most $\frac43 c^Tx$ [2008.03327].

The mechanism is recursive. From a half-integral feasible $x$, one forms the multigraph $G'$ with $2x_e$ parallel copies of each edge; $G'$ is then $4$-regular and $4$-edge connected. Splitting-off admissible pairs of incident edges preserves the required connectivity while reducing the instance by one vertex, and careful edge-cost assignment maintains the inductive cost bound. The key induction theorem states that for a $4$-regular, $4$-edge connected multigraph $G$ and any edge $e$, one can in $O(n^2)$ time find a $2$-edge connected spanning subgraph $H$ of $G-e$ such that $c(H)\le \frac23 c(G-e)$ and $H$ uses at most one copy of each multiedge; this yields the $\frac43$ guarantee for the original LP solution [2008.03327].

The directed analogue is the weighted minimum strongly connected spanning subgraph problem. Its LP relaxation requires $x(\delta^+(S))\ge 1$ for all nonempty proper $S\subset V$. The paper "An Optimal Rounding for Half-Integral Weighted Minimum Strongly Connected Spanning Subgraph" proves that half-integral LP solutions can be rounded in polynomial time at multiplicative cost $1.5$, and more generally that if every positive variable is at least $f>0$, the multiplicative bound is $2-f$. The construction uses distributions over in-arborescences and out-arborescences rooted at a chosen vertex $r$, with
\[
\Pr(a\in I)=\Pr(a\in O)=x_a,
\]
and therefore
\[
\Pr(a\in I\cup O)=2x_a-x_a^2.
\]
For the half-integral case $f=\frac12$, this gives the optimal factor $1.5$, matching a known $1.5$ integrality-gap lower bound for a half-integral instance [2011.06108].

For undirected $k$-edge connectivity, the paper "Approximating $k$-Edge-Connected Spanning Subgraphs via a Near-Linear Time LP Solver" separates the fractional and integral aspects. It computes a $(1+\epsilon)$-approximate fractional solution in time $\tilde O(m/\epsilon^2)$ and then rounds it to an integral $k$ECSS of cost at most $(2+\epsilon)$ times optimum in time
\[
\tilde O\!\left(\frac{m}{\epsilon^2}+\frac{k^2 n^{1.5}}{\epsilon^2}\right).
\]
The rounding proceeds through cut sparsification and a Khuller–Vishkin reduction to $k$ edge-disjoint arborescences in a directed support of size $\tilde O(nk)$ [2205.14978].

These results make clear that integrality is not merely a modeling convention. In this part of the literature it is a quantitative objective: the aim is to certify how close an integral spanning subgraph can be to the optimal fractional benchmark.

## 4. Degree bounds, walks, leaves, and designated internal vertices

The language of integral spanning structures also appears in existence theorems with local constraints. The paper "Spanning trees and spanning closed walks with small degrees" proves that if for all $S\subseteq V(G)$,
\[
\omega(G\setminus S)<\sum_{v\in S}(f(v)-2)+2+\omega(G[S]),
\]
then $G$ has a spanning tree $T$ containing an arbitrary given matching such that $d_T(v)\le f(v)$ for all vertices. It also proves that if for all $S\subseteq V(G)$,
\[
\omega(G\setminus S)\le \sum_{v\in S}(f(v)-1)+1,
\]
then $G$ admits a spanning closed walk passing through the edges of an arbitrary given matching and meeting each vertex $v$ at most $f(v)$ times; this solves a long-standing conjecture of Jackson and Wormald [1702.06203].

The attainable number of leaves in spanning trees exhibits a different integrality phenomenon. If
\[
L(G)=\{k\in\mathbb Z: G \text{ has a spanning tree with exactly } k \text{ leaves}\},
\]
then the paper "Spanning trees for many different numbers of leaves" proves that $L(G)$ is contiguous for every connected graph. For connected and locally connected graphs of order at least $4$, including triangulations, this yields spanning trees with exactly $k$ leaves for every
\[
2\le k\le \left\lfloor \frac n2+1\right\rfloor,
\]
and for plane $4$-connected $n$-vertex triangulations there are polynomial-time constructions for every integer $k$ in that range. The same paper also constructs infinitely many plane $4$-connected $n$-vertex triangulations with $\max L(G)=\frac{2n}{3}$ [2312.13674].

A complementary optimization problem is the Minimum Weight Non-Terminal Spanning Tree problem, where a specified set $S$ must consist of internal vertices in the spanning tree, equivalently $\deg_T(v)\ge 2$ for all $v\in S$. This generalizes the standard spanning tree problem and contains $s$-$t$ Hamiltonian Path as the special case $S=V\setminus\{s,t\}$. The problem is NP-hard, W[1]-hard when parameterized by clique-width, admits a $3k$-vertex kernel for $k=|S|$, and has an $O^*(2^k)$-time algorithm for polynomially bounded integral edge weights; parameterization by $\ell=|E(G[S])|$ gives an $O^*(2^\ell)$-time algorithm for arbitrary weights [2310.05494].

Exact integral modeling also appears in the constrained optimum communication spanning tree problem. The paper "Mixed-Integer Approaches to Constrained Optimum Communication Spanning Tree Problem" models pairwise distances directly by integral variables $d_{ij}$ together with binary edge variables $x_{ij}$, using Bellman-type recursions and the algebraic relation
\[
\mathcal L D + 2I = 2(\mathbbm 1 - 2d)\mathbbm 1^T.
\]
In the reported experiments, the novel linear distance-based formulations are superior to the traditional multicommodity flow model, and the distance-based MILP formulations are stated to be 35 times faster for $n=15$ and $n=20$ [2101.03872].

## 5. Integral cohomology, integral roots, and spanning geometry

In algebraic combinatorics and geometry, spanning objects are often studied through integral invariants rather than through optimization. A spanning configuration in $\mathbb C^k$ is a sequence $(W_1,\dots,W_r)$ of subspaces with prescribed dimensions $\dim(W_i)=\alpha_i$ such that
\[
W_1+\cdots+W_r=\mathbb C^k.
\]
Its moduli space
\[
X_{\alpha,k}=\{(W_1,\dots,W_r): \dim(W_i)=\alpha_i,\ W_1+\cdots+W_r=\mathbb C^k\}
\]
is a Zariski open subset of a product of Grassmannians. The paper "Spanning subspace configurations" gives the integral cohomology presentation
\[
H^*(X_{\alpha,k};\mathbb Z)=\left(\mathbb Z[\mathbf x_n]/I_{\alpha,k}\right)^{S_\alpha},
\]
where $I_{\alpha,k}$ is generated by the top $k$ elementary symmetric polynomials in all variables and by the complete homogeneous symmetric polynomials $h_d(\mathbf x_n^{(i)})$ for $d>k-\alpha_i$. This simultaneously generalizes Borel’s presentation for partial flag varieties and the cohomology of spanning line configuration spaces associated with the Haglund–Remmel–Wilson Delta Conjecture [1903.07579].

The polynomial theory of increasing spanning forests exhibits a different form of integrality. If $G$ has vertices $\{1,\dots,n\}$ and $E_k=\{jk\in E(G): j<k\}$, then the generating polynomial for increasing spanning forests is
\[
ISF(G,t)=\prod_{k=1}^n (t+|E_k|).
\]
Hence all roots are nonpositive integers. Hallam and Sagan had shown this factorization, and the later paper provides new combinatorial proofs, bounds the coefficients using broken circuits, and extends the theory to cage-free simplicial complexes and labeled multigraphs. Moreover,
\[
ISF(G,t)=(-1)^n P(G,-t)
\]
if and only if the labeling is a perfect elimination order [1610.05093].

The phrase “spanning” also enters Ehrhart theory. A lattice polytope $P\subseteq \mathbb Z^d$ is spanning if every lattice point in $\mathbb Z^d$ is an integral affine combination of lattice points of $P$. The paper "Ehrhart Theory of Spanning Lattice Polytopes" proves that the $h^*$-vector of a spanning lattice polytope has no gaps: if $P$ is spanning, then
\[
h_i^*(P)\ge 1 \qquad \text{for all } i=0,\dots,\deg(P).
\]
As a consequence, the paper derives the inequality
\[
h_1^*+\deg(P)\le \operatorname{Vol}_{\mathbb Z^d}(P),
\]
described there as a polyhedral consequence of the Eisenbud–Goto conjecture [1608.03166].

## 6. Structural synthesis, terminology, and open directions

Taken together, these works suggest that “integral spanning” is best understood as a structural motif rather than a single theorem or model. The common thread is that a spanning requirement is coupled to an integrality phenomenon: exact selection of edges or arcs, integral multiplicities, integral root or cohomology behavior, or integral analytic representations. The principal methods are correspondingly diverse: Tutte-polynomial differentiation and spanning-subgraph expansions for random MSTs, Laplacian determinants and contraction for enumeration, splitting-off and arborescence decompositions for approximation, Bellman-type recursions for mixed-integer formulations, mixed reduction and orbit harmonics for cohomology, and half-open triangulations for $h^*$-theory [1501.03758] [2602.03602] [2008.03327] [2101.03872] [1903.07579] [1608.03166].

A persistent misconception is that “integral” in spanning problems refers only to integer edge weights. The cited literature shows otherwise. In 2ECM and WMSCSS, the salient issue is rounding half-integral LP solutions to integral spanning subgraphs [2008.03327] [2011.06108]. In spanning configurations, the decisive property is integral cohomology over $\mathbb Z$ [1903.07579]. In increasing spanning forests, it is complete factorization with nonpositive integral roots [1610.05093]. In spanning lattice polytopes, it is the arithmetic spanning condition and the induced no-gap behavior of the $h^*$-vector [1608.03166].

Several open-ended directions remain visible within the surveyed material. For complete graphs, the coefficient patterns in the MST integrand are explicitly said to hint at asymptotic analysis [1501.03758]. For $2$ECM, the Four-Thirds Conjecture remains the long-standing benchmark beyond the half-integral case [2008.03327]. In Ehrhart theory, the no-gap theorem is presented as a modest analog of unimodality, while unimodality questions themselves remain open for important classes such as IDP polytopes [1608.03166]. These directions indicate that integrality in spanning problems is not a peripheral regularity condition: it is a source of structure, complexity, and algebraic rigidity across combinatorics, optimization, and geometry.

Source: https://www.emergentmind.com/topics/integral-spanning