Spanning Tree Isomorphism Problem: Analysis & Algorithms

• Spanning Tree Isomorphism Problem is the decision problem of determining if a graph contains a spanning tree isomorphic to a given target tree.
• The literature covers diverse approaches including parameterized algorithms by redundant set size, counting isomorphism classes in special graphs, and decompositions into isomorphic spanning trees.
• Insights from this topic inform efficient algorithm design and complexity analyses, with implications for related NP-complete spanning tree decomposition problems.

The Spanning Tree Isomorphism Problem (STIP) is the decision problem of determining whether a graph contains a spanning tree that is isomorphic to a prescribed target tree on the same number of vertices. In the undirected formulation, given an undirected graph $G=(V_G,E_G)$ and a tree $T=(V_T,E_T)$ with $|V_G|=|V_T|=n$, one asks whether there exists a bijection $\phi:V_G\rightarrow V_T$ and an edge subset $E_H\subseteq E_G$ with $|E_H|=n-1$ such that, for any pair of vertices $u,v\in V_G$, $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$. A directed analogue asks whether a digraph contains a spanning directed tree isomorphic to a given directed tree. The research landscape around this problem is heterogeneous: it includes direct parameterized algorithms for STIP, counting isomorphism classes of spanning trees in special host graphs, decomposition results for multiple pairwise isomorphic spanning trees, and precursor work on rooted tree isomorphism that is relevant but not itself STIP (Shen et al., 7 Aug 2025).

1. Problem formulations and scope

A central formulation in the recent literature parameterizes STIP by the size of a redundant set, namely a set of edges or arcs whose deletion transforms the host graph into a spanning tree. If a graph on $n$ vertices has redundant set size $k$, then it has exactly $T=(V_T,E_T)$0 edges or arcs; the parameter $T=(V_T,E_T)$1 therefore measures how far the input is from a tree. In the directed setting, the definition is mediated by the underlying graph of a digraph, obtained by replacing each arc $T=(V_T,E_T)$2 with an undirected edge $T=(V_T,E_T)$3 (Shen et al., 7 Aug 2025).

The same phraseology does not cover all nearby spanning-tree questions. Some papers study the number of isomorphism classes of spanning trees occurring in a fixed host graph; others study whether a host graph can be decomposed into several pairwise isomorphic spanning trees; still others study constrained decompositions into spanning trees with orientation requirements. These are related to STIP, but they are not equivalent to the decision problem above.

Variant	Formulation in the cited literature	Representative result
Undirected STIP	Does $T=(V_T,E_T)$4 contain a spanning tree isomorphic to $T=(V_T,E_T)$5?	FPT by redundant set size $T=(V_T,E_T)$6 (Shen et al., 7 Aug 2025)
Directed STIP	Does $T=(V_T,E_T)$7 contain a spanning directed tree isomorphic to $T=(V_T,E_T)$8?	FPT by redundant set size $T=(V_T,E_T)$9 (Shen et al., 7 Aug 2025)
Counting classes	How many isomorphism classes of spanning trees occur in $|V_G|=|V_T|=n$0?	Lower bounds via partition counts (Johnson et al., 6 Feb 2026)
Isomorphic decomposition	Can a host graph be partitioned into pairwise isomorphic spanning trees?	Results for properly colored $|V_G|=|V_T|=n$1 (Fu et al., 2014)
Rooted tree isomorphism	Are two trees themselves isomorphic?	Quantum and algebraic recognition results (Rosenbaum, 2010, Li et al., 2019)

This separation of formulations is important. A tree-isomorphism algorithm solves comparison once a spanning tree has already been specified, whereas STIP asks for the existence of an appropriate spanning tree inside a larger graph.

2. Rooted-tree isomorphism as a precursor

Two cited works address tree isomorphism rather than spanning tree isomorphism. One gives a quantum algorithm for deciding whether two rooted trees are isomorphic by preparing a state $|V_G|=|V_T|=n$2 for each rooted tree such that, when the trees have the same dimension,

$|V_G|=|V_T|=n$3

The construction is recursive, labels isomorphic child subtrees by multiplicity tags $|V_G|=|V_T|=n$4, forms a uniform superposition over permutations in $|V_G|=|V_T|=n$5, applies conditional permutations, and uncomputes auxiliary data to obtain a permutation-invariant encoding of the rooted tree. The same paper states that by considering all possible roots in one of the trees, one can decide isomorphism of unrooted trees efficiently as well; it explicitly does not study the spanning tree isomorphism problem as a graph problem (Rosenbaum, 2010).

A second precursor uses the Terwilliger algebra of a rooted tree $|V_G|=|V_T|=n$6. With

$|V_G|=|V_T|=n$7

the paper shows that the principal module $|V_G|=|V_T|=n$8 recognizes the rooted tree. Its main structural theorem identifies

$|V_G|=|V_T|=n$9

where $\phi:V_G\rightarrow V_T$0, and proves that the Terwilliger algebra determines the rooted isomorphism class. It also establishes that, for trees,

$\phi:V_G\rightarrow V_T$1

This is an algebraic recognition theorem for rooted trees, not a method for selecting or matching spanning trees inside general graphs (Li et al., 2019).

These precursor results clarify a recurrent boundary in the subject. Once a spanning tree is given, rooted or unrooted tree isomorphism can be handled by specialized techniques; STIP is harder because the tree must first be found inside the ambient graph. This suggests that the combinatorial difficulty of STIP is concentrated in subgraph selection rather than in tree comparison alone.

3. Parameterized algorithms by redundant set size

The most direct algorithmic treatment of STIP in the cited corpus studies both undirected and directed versions parameterized by the redundant-set size $\phi:V_G\rightarrow V_T$2. For the undirected problem, the algorithm runs in

$\phi:V_G\rightarrow V_T$3

and for the directed problem in

$\phi:V_G\rightarrow V_T$4

where $\phi:V_G\rightarrow V_T$5 is the number of vertices (Shen et al., 7 Aug 2025).

In the undirected case, the method begins with an easy $\phi:V_G\rightarrow V_T$6 case: if the graph has exactly one cycle, one can enumerate cycle edges, delete each in turn to obtain a spanning tree, and test isomorphism to the target tree using the AHU tree isomorphism algorithm. The general case applies a kernelization procedure, Make-Contractible, consisting of Leaf Trimming and Path Contraction, until no degree-1 or degree-2 vertices remain. The resulting graph kernel satisfies the bound

$\phi:V_G\rightarrow V_T$7

The proof uses invariance of $\phi:V_G\rightarrow V_T$8, the minimum degree condition in the kernel, and the handshaking lemma. The subsequent search is organized around anchor points, tree-like neighbors, neighbor anchor sets, and a DFS order on the target tree. Tree-like components are matched deterministically, while non-tree-like components are handled recursively through permutations of anchor points in the kernel. A correctness theorem states that reordering the matching of tree-like and non-tree-like neighbors does not lose valid mappings.

The directed case reuses the kernel of the underlying undirected graph but introduces anchor chains

$\phi:V_G\rightarrow V_T$9

where each adjacent pair supports at least one arc direction. Two structural lemmas are central: every redundant arc lies on an anchor chain corresponding to a kernel edge, and if there are $E_H\subseteq E_G$0 redundant arcs, then there exists a subset of $E_H\subseteq E_G$1 kernel edges such that each corresponding anchor chain contains exactly one redundant arc. A further theorem states that the direction pattern along an anchor chain determines the redundant arc uniquely or leaves only a constant-size ambiguity. The runtime bound follows from the facts that the kernel has at most $E_H\subseteq E_G$2 vertices and at most $E_H\subseteq E_G$3 edges, giving

$E_H\subseteq E_G$4

and hence the overall bound $E_H\subseteq E_G$5.

Within this line of work, STIP is treated as fixed-parameter tractable when the host graph is close to a tree. The parameter $E_H\subseteq E_G$6 functions as the size of the combinatorial core that remains after contracting away tree-like structure.

4. Counting isomorphism classes in complete bipartite graphs

A distinct but closely related problem asks for the number of isomorphism classes of spanning trees in a fixed host graph. For the complete bipartite graph $E_H\subseteq E_G$7, $E_H\subseteq E_G$8, the paper defines $E_H\subseteq E_G$9 as the number of isomorphism classes of spanning trees of $|E_H|=n-1$0. Every such spanning tree has

$|E_H|=n-1$1

The paper’s central observation is that a spanning tree of $|E_H|=n-1$2 can be described by the degree sequences on the two bipartition classes, and these degree sequences are integer partitions of $|E_H|=n-1$3 (Johnson et al., 6 Feb 2026).

A structural lemma shows that spanning trees of $|E_H|=n-1$4 are exactly the connected bipartite graphs whose degree sequences on the two sides are integer partitions

$|E_H|=n-1$5

of $|E_H|=n-1$6. Equivalently, every spanning tree gives rise to two integer partitions of $|E_H|=n-1$7, and every such pair is realized by at least one spanning tree. This creates a direct bridge between spanning-tree isomorphism classes and partition theory.

The main lower bound for $|E_H|=n-1$8 is

$|E_H|=n-1$9

where $u,v\in V_G$0 denotes the number of integer partitions of $u,v\in V_G$1 of length $u,v\in V_G$2. In the balanced case $u,v\in V_G$3, with $u,v\in V_G$4, the bound becomes

$u,v\in V_G$5

Using Scoins’ formula for the total number of labeled spanning trees,

$u,v\in V_G$6

the paper derives

$u,v\in V_G$7

and, in the balanced case,

$u,v\in V_G$8

The results are explicitly lower bounds, not a full classification. The paper states that it does not prove tightness in general and does not classify all isomorphism classes. Its contribution is to show that the number of abstract tree types realized as spanning trees of $u,v\in V_G$9 is already forced to be large by partition-count considerations.

5. Isomorphic spanning-tree decompositions in colored complete graphs

Another branch of the literature studies whether the edges of a host graph can be partitioned into several pairwise isomorphic spanning trees. In a proper $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$0-edge-coloring of the complete graph $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$1, a multicolored spanning tree is a spanning tree all of whose edges have distinct colors. The paper investigates whether $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$2 can be decomposed into spanning trees that are both multicolored and pairwise isomorphic, situating the question near the Brualdi–Hollingsworth and Constantine conjectures (Fu et al., 2014).

Its main decomposition theorem states that if $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$3 is a proper $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$4-edge-coloring of $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$5, $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$6, such that any two colors form a $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$7-factor, then the edges of $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$8 can be partitioned into $(u,v)\in E_H\iff (\phi(u),\phi(v))\in E_T$9 isomorphic multicolored spanning trees. A key lemma shows that this hypothesis is highly rigid: if any two colors induce a 2-factor of 4-cycles, then $n$0 for some $n$1, and for each $n$2, the graph contains a clique of order $n$3 using exactly $n$4 colors. The proof uses Latin squares, specifically the direct product $n$5 of the $n$6-Latin square, together with the proposition that $n$7 has $n$8 disjoint transversals for each $n$9.

The same paper weakens the hypotheses in two steps. First, it proves that in any proper $k$0-edge-coloring of $k$1, there exist two disjoint isomorphic multicolored spanning trees for $k$2. Second, it proves that there exist three disjoint isomorphic multicolored spanning trees in any proper $k$3-edge-coloring of $k$4 for $k$5. The constructive mechanism is a sequence of local edge exchanges that preserve the tree property, multicoloredness, disjointness, and isomorphism.

These results do not solve STIP as a subgraph-existence problem against an arbitrary target tree. Instead, they address the existence of multiple pairwise isomorphic spanning trees inside highly structured hosts, especially properly edge-colored complete graphs.

6. Complexity boundaries and neighboring problems

The boundaries of STIP are sharpened by work on nearby spanning-tree decomposition problems. One such problem asks whether the edge set $k$6 of a graph $k$7 can be partitioned into two spanning trees, one blue and one red, such that the blue tree admits a prescribed $k$8-orientation and the red tree admits a prescribed $k$9-orientation, where $T=(V_T,E_T)$00. The paper calls this a $T=(V_T,E_T)$01-partition of $T=(V_T,E_T)$02, and proves that deciding whether such a partition exists is NP-complete (Gevigney, 2013).

The NP-completeness proof is by reduction from NotAllEqual 3-SAT. The construction uses clause gadgets, copied clause gadgets, variable cycles $T=(V_T,E_T)$03, and a special vertex $T=(V_T,E_T)$04 with

$T=(V_T,E_T)$05

A key forced-orientation condition is

$T=(V_T,E_T)$06

Two propositions drive the reduction: one forces each clause gadget to contain both a blue and a red edge, and the other forces all original occurrences of a variable to receive the same color while copied occurrences receive the opposite color. The result refutes Recski’s conjecture unless $T=(V_T,E_T)$07.

This problem is not STIP: it is a constrained decomposition problem rather than an isomorphism problem. Its significance for the present topic is indirect. It shows that modest-looking global conditions on spanning-tree structure, once coupled to orientation or outdegree prescriptions, are already computationally intractable. In combination with the direct parameterized results for STIP and the specialized counting and decomposition results above, the literature indicates that “spanning tree isomorphism” is best understood as a family of related but non-equivalent problems. The exact STIP concerns finding a spanning tree of prescribed isomorphism type; neighboring questions concern how many such types occur, whether several isomorphic spanning trees can coexist or partition the host, and how additional combinatorial constraints alter the complexity landscape.