T(h+1)-Free Edge Deletion Problem

• T(h+1)-Free Edge Deletion is defined as deleting at most k edges from a graph so that no component or induced subgraph forms a tree on h+1 vertices.
• The problem is NP-complete for h ≥ 2 in the induced variant and for h ≥ 3 in the component-size variant, highlighting a sharp complexity dichotomy.
• Positive results include FPT algorithms and kernelization for parameters like cluster-vertex deletion and neighborhood diversity when paired with h.

The T(h+1)-Free Edge Deletion problem is a fundamental question in graph modification theory. It asks: given a graph $G$ and integers $k$ and $h$, can one delete at most $k$ edges from $G$ so that the resulting graph has no (induced or non-induced, depending on variant) subgraph isomorphic to any tree on $h+1$ vertices? Analogously, in the case of the "every component is small" variant, the goal is to delete at most $k$ edges so that every connected component of the resulting graph contains at most $h$ vertices, equivalently ensuring the graph is $T_{h+1}$-free for the family $T_{h+1}$ of all trees on $k$0 vertices. This problem has received extensive attention due to both its intrinsic combinatorial interest and its relevance in applications such as epidemic containment, network resilience, and structural graph theory.

1. Formal Problem Definition

Let $k$1 be a finite simple graph and $k$2 an integer. Define $k$3 as a fixed tree on $k$4 vertices, or alternatively as the family of all (labeled) trees on $k$5 vertices.

Induced variant:

Given a fixed tree $k$6, the task is to decide, for input $k$7, whether there exists a set $k$8 with $k$9 such that $h$0 contains no induced copy of $h$1.

Component-size variant:

Given input $h$2, does there exist $h$3, $h$4, such that every connected component of $h$5 has at most $h$6 vertices—i.e., $h$7 is $h$8-free in the sense that no component has order greater than $h$9? This is equivalent to forbidding all (not necessarily induced) copies of any $k$0 as a subgraph (Gaikwad et al., 31 Jan 2026, Gaikwad et al., 2021).

2. Classical Complexity Landscape

For the induced variant with fixed tree $k$1 (on $k$2 vertices or $k$3 edges), T(h+1)-Free Edge Deletion is NP-complete (Aravind et al., 2015, Aravind et al., 2015). The dichotomy result is sharp: for $k$4, the problem is trivial or polynomial-time solvable; for $k$5, it is NP-complete.

For the component-size variant (forbidding all trees of $k$6 vertices), the problem remains NP-complete for $k$7 and is polynomial-time solvable for $k$8:

• For $k$9, one deletes all edges;
• For $G$0, the solution corresponds to finding a maximum matching (Gaikwad et al., 31 Jan 2026, Aravind et al., 2015).

Moreover, it is shown that unless the Exponential Time Hypothesis (ETH) fails, there is no algorithm solving T(h+1)-Free Edge Deletion in time $G$1 (Aravind et al., 2015, Aravind et al., 2015).

The following table summarizes classical complexity:

$G$2	Complexity	Reference
$G$3	Polynomial time	(Gaikwad et al., 31 Jan 2026)
$G$4	NP-complete, no $G$5	(Aravind et al., 2015)

3. Parameterized Complexity and Hardness

The parameterized complexity landscape is notably intricate, especially when $G$6 is allowed to grow. The following are the central findings for the component-size variant:

W[1]/W[2]-hardness:

• T(h+1)-Free Edge Deletion is W[1]-hard parameterized by treewidth (Gaikwad et al., 2021), feedback-vertex-set number, pathwidth, feedback-edge-set number (Gaikwad et al., 31 Jan 2026), and even cluster-vertex-deletion number, treedepth, modular width, and twin cover. This implies that, unless FPT=W[1], no FPT algorithms exist for these structural parameters alone when $G$7 is unbounded (Gaikwad et al., 31 Jan 2026, Gaikwad et al., 2021).
• The problem is W[2]-hard parameterized by solution size $G$8, $G$9, or $h+1$0, even on restricted graph classes such as planar or bipartite graphs (Gaikwad et al., 2021).

This series of negative results establishes that most "classical" parameterizations are insufficient to yield tractability for large-$h+1$1 instances, even on graphs close to cluster graphs (cluster-vertex deletion), those with bounded treedepth, or low modular width (Gaikwad et al., 31 Jan 2026).

4. Positive FPT and Kernelization Results

Despite broad intractability for many parameters, several parameterizations restore tractability when combined with $h+1$2 or under certain graph class restrictions.

• FPT in cluster-vertex deletion number plus $h+1$3: T(h+1)-Free Edge Deletion can be solved in $h+1$4 time, where $h+1$5 is the size of a cluster-vertex deletion set. The method first reduces the input graph by handling large twin classes and then applies dynamic programming on a path decomposition of width $h+1$6 (Gaikwad et al., 31 Jan 2026).
• FPT in neighborhood diversity plus $h+1$7: For graphs with neighborhood diversity $h+1$8, the problem is solvable by ILP in time $h+1$9, using integer variables to model part sizes in the solution (Gaikwad et al., 31 Jan 2026).
• FPT in vertex cover number (fixed $k$0): In $k$1 time, where $k$2 is the vertex cover number and $k$3 the Bell number. The algorithm partitions the independent set into twin classes and models the assignment of vertices and component splits as an ILP (Gaikwad et al., 2021).
• Kernelization: For parameter $k$4, T(h+1)-Free Edge Deletion admits a kernel of $k$5 vertices and $k$6 edges (Gaikwad et al., 2021). For bounded-degree graphs ($k$7 fixed), T(h+1)-Free Edge Deletion has a kernel with $k$8 vertices (Aravind et al., 2014).

For the induced variant, no polynomial kernel exists on general graphs for non-star trees with at least seven vertices, unless NP $k$9 coNP/poly (Aravind et al., 2015).

5. Algorithmic Methods and Reductions

Multiple reduction and algorithmic constructions underlie the core results:

• Base Case Reductions: For stars and twin-stars (trees of diameter 2 and 3), reductions respectively from P₃-free and P₄-free edge deletion apply by attaching cliques to each vertex, ensuring that solution transfer entails solution transfer for the original instance (Aravind et al., 2015).
• General Tree Reductions: An inductive "leaf-pruning" reduction grows larger trees from smaller ones, ensuring that solution sets correspond via gadgets (Construction 1) that branch off $h$0 "completions" for each instance of the smaller tree (Aravind et al., 2015, Aravind et al., 2015).
• Parameterized Reductions: For cluster-vertex deletion and neighborhood diversity, the main algorithmic approach involves partitioning residual graphs into twin classes and solving (typically via ILP) the assignment of components or part sizes, under constraints linking deletions to forbidden structures (Gaikwad et al., 31 Jan 2026, Gaikwad et al., 2021).

These methods yield both hardness constructions (demonstrating parameterized intractability via gadgets simulating Hitting Set or Unary Bin Packing) and FPT algorithms where additional structure is available.

6. Special Cases, Approximability, and Restricted Graph Classes

• Bicriteria FPT Approximation in $h$1: Although W[1]-hard for $h$2, there is an FPT bicriteria approximation that in $h$3 time produces an edge deletion set of size at most $h$4 if a solution of size $h$5 exists (Gaikwad et al., 31 Jan 2026). The algorithm uses recursive application of FPT Minimum Bisection to disassemble the graph into small components.
• Split Graphs: The problem is NP-complete on split graphs, but FPT in $h$6 via reduction to vertex cover or greedy branching, leveraging the graph’s structural dichotomy between clique and independent set (Gaikwad et al., 31 Jan 2026).
• Interval Graphs: No NP-hardness is known for interval graphs (with $h$7 possibly unbounded), but T(h+1)-Free Edge Deletion is FPT in $h$8 in $h$9 time through dynamic programming over a clique-path decomposition, using strong restrictions on component intersections in bags (Gaikwad et al., 31 Jan 2026).
• Directed Variant: For the problem of deleting arcs in digraphs such that no node can reach more than $T_{h+1}$0 others, the problem is W[2]-hard for $T_{h+1}$1, even on DAGs, via reductions from Hitting Set (Gaikwad et al., 31 Jan 2026).

7. Summary Table: Parameterized Complexity Landscape

Parameter(s)	Complexity	Reference
$T_{h+1}$2	W[2]-hard	(Gaikwad et al., 2021)
$T_{h+1}$3 feedback-edge-set	W[1]-hard	(Gaikwad et al., 31 Jan 2026)
treewidth, pathwidth, feedback vertex	W[1]-hard	(Gaikwad et al., 2021, Gaikwad et al., 31 Jan 2026)
cluster-vertex-deletion $T_{h+1}$4 + $T_{h+1}$5	FPT	(Gaikwad et al., 31 Jan 2026)
neighborhood diversity $T_{h+1}$6 + $T_{h+1}$7	FPT	(Gaikwad et al., 31 Jan 2026)
vertex cover number $T_{h+1}$8 (fixed $T_{h+1}$9)	FPT	(Gaikwad et al., 2021)
bounded degree ($T_{h+1}$0), fixed $T_{h+1}$1	poly kernel	(Aravind et al., 2014)

The T(h+1)-Free Edge Deletion problem thus defines a sharply divided parameterized landscape. Most classic structural parameters do not yield FPT algorithms unless combined with $T_{h+1}$2, but FPT is available for cluster-vertex deletion plus $T_{h+1}$3 or neighborhood diversity plus $T_{h+1}$4. A plausible implication is that the interplay of "component-size control" and "global connectivity structure" is the core driver of the parameterized complexity for this class of edge deletion problems. For the induced variant, incompressibility results rule out polynomial kernels beyond certain bounded parameters, especially for trees that are not stars (Aravind et al., 2015).

• (Aravind et al., 2015) Parameterized lower bound and NP-completeness of some $T_{h+1}$5-free Edge Deletion problems
• (Aravind et al., 2015) Parameterized Lower Bounds and Dichotomy Results for the NP-completeness of $T_{h+1}$6-free Edge Modification Problems
• (Aravind et al., 2014) On Polynomial Kernelization of $T_{h+1}$7-free Edge Deletion
• (Gaikwad et al., 2021) Edge Deletion to Restrict the Size of an Epidemic
• (Gaikwad et al., 31 Jan 2026) Hardness and Tractability of T_{h+1}-Free Edge Deletion