Tree Augmentation Problem (TAP)

Updated 21 January 2026

TAP is a combinatorial optimization problem that seeks to make a tree 2-edge-connected by adding the fewest supplemental links or minimizing cost.
It leverages techniques like LP relaxations, CG cuts, and SDP methods to reduce the integrality gap and address the problem’s APX-hardness.
Recent advances have achieved approximation ratios as low as 1.393 for general instances and 4/3 for special cases, improving network design resilience.

The Tree Augmentation Problem (TAP) is a central problem in combinatorial optimization and network design, with applications spanning survivable network design, communication infrastructures, and distributed computing. The core challenge in TAP is to transform a given tree into a 2-edge-connected graph by adding the minimum number (or cost) of supplementary edges, known as links. Despite its seemingly simple structure, TAP is APX-hard, and achieving approximation ratios close to 1 for general instances remains a major open challenge.

1. Formal Definition and Problem Models

Given an undirected tree $T=(V, E_T)$ and a set of potential augmentation edges ("links") $L \subseteq \binom{V}{2}$ (with $E_T \cap L = \emptyset$ ), TAP seeks a minimum-cardinality subset $F \subseteq L$ such that the augmented graph $(V, E_T \cup F)$ is 2-edge-connected. In the weighted variant (WTAP), each link $\ell \in L$ is assigned a nonnegative cost $c_\ell$ , and the objective is to minimize $\sum_{\ell \in F} c_\ell$ . The principal requirement is that every tree-edge $e \in E_T$ must be covered; that is, there must be at least one link in $F$ whose unique tree path in $T$ contains $e$ .

Formally, the standard integer programming (cut cover) formulation is: $\min \sum_{\ell \in L} x_\ell \quad \text{subject to} \quad \forall e \in E_T:\, \sum_{\ell\colon e \in P_T(\ell)} x_\ell \geq 1, \quad x_\ell \in \{0,1\}.$ where $P_T(\ell)$ denotes the unique path in $T$ between the endpoints of $\ell$ .

2. Polyhedral Relaxations, LP and SDP Techniques

The baseline linear relaxation is the "cut-LP": $\min \sum_{\ell \in L} c_\ell x_\ell \quad\text{s.t.}~\forall e \in E_T:\, \sum_{\ell: e \in P_T(\ell)} x_\ell \geq 1,~ x_\ell \geq 0.$ The cut-LP has integrality gap exactly 2 for arbitrary costs. Substantial progress has occurred by introducing combinatorial and lift-and-project polyhedral strengthening, such as:

Bundle (or $k$ -bundle) constraints: For any collection of up to $k$ subtrees (bundles), enforce $\sum_{\ell: P_\ell \cap B \neq \emptyset} c_\ell x_\ell \geq \mathrm{OPT}(B)$ , where $\mathrm{OPT}(B)$ is the integrally minimal cover cost for bundle $B$ .
Chvátal–Gomory (CG) cuts, especially {0,½}-odd-cut constraints: For any odd-degree cut $S \subset V$ , enforce $\sum_\ell \lceil |P_\ell \cap \delta_G(S)|/2 \rceil x_\ell \geq (|\delta_G(S)|+1)/2$ .
Lasserre/SoS hierarchy SDPs: Tighten the relaxation by positive semidefinite moment and localizing constraints, making fractional solutions closer to integral (Cheriyan et al., 2015).

Recent work exploits the intersection of these strengthenings. The odd-cut-bundle LP achieves a $(3/2+\epsilon)$ -approximation—matching the best known SDP-based integrality ratio for unweighted TAP (Fiorini et al., 2017). These relaxations support both rounding and credit-based combinatorial methods.

3. Approximation Algorithms and Complexity Landscape

3.1 Historical Progression

A series of breakthroughs have steadily reduced the approximation guarantee for unweighted TAP:

2-approximation: Frederickson & JáJá (1981), primal-dual and iterative LP methods.
1.8: Even, Feldman, Kortsarz, Nutov (2009) via ear-decomposition.
1.75 and 1.5: Kortsarz & Nutov (2016), using primal-dual and credit invariant frameworks (Kortsarz et al., 2015).
$(1.5+\epsilon)$ : Fiorini et al. (tightening with CG cuts and LP structure).
1.458: Grandoni, Kalaitzis, Zenklusen (rewiring and matching-in-expectation arguments, stack rounding) (Grandoni et al., 2018).
1.393: Traub, Zenklusen, Cecchetto (use of stack analysis and advanced convex combination techniques; unified with CAP) (Cecchetto et al., 2020).
$4/3$: Deferred local ratio, leaf-to-leaf and matching-based plus stack methods for special subcases (Cecchetto et al., 2022, Kortsarz, 14 Jan 2026).

3.2 Sample of Algorithmic Milestones

Reference	Guarantee	Model(s)/Techniques	Notes
(Kortsarz et al., 2015, Cheriyan et al., 2015)	1.5 - 1.5+ε	Primal-dual, credit invariants, SDP	Fundamental for lower bounds
(Fiorini et al., 2017, Cecchetto et al., 2020)	1.5+ε, 1.393	{0,½}-CG cuts, stack analysis, LP	Unified with CAP
(Grandoni et al., 2018)	1.458	Rewiring, k-wide LPs, sampling	k-wide decomposition
(Kortsarz, 14 Jan 2026, Cecchetto et al., 2022)	4/3, 1.29	Matching-based, deferred LR, stack	Leaf-to-leaf, O(1)-wide

For TAP with unit costs, the best known integrality gap for the cut-LP is at most $28/15 \approx 1.8667$ ; for the 3-Branch-LP it is $7/4=1.75$ (Nutov, 2017).

4. Representative Techniques: Local Ratio, Matching, Stack, and Deferred LR

Local Ratio & Credit Invariants: Augmentation is performed by greedily matching leaves and tracking "coupons" (tokens) to manage legal contractions and exploit the combinatorial structure of TAP subtrees (“semi-closed subtrees”), ensuring each operation can be “paid for” under tight budget (Kortsarz et al., 2015).
Rewiring and Stack Rounding: After splitting the tree into principal O(1)-wide subtrees, combine LP-optimal fractional solutions by randomized sampling and strategically recombine cross-links via matching and rewiring, saving over the naive $1.5$ barrier (Grandoni et al., 2018, Cecchetto et al., 2020).
Deferred Local Ratio: Departing from recursive disjointness in local-ratio, allows temporary overlap via “shadow-sets” and uses “golden-ticket” nodes to globally control charge. This yields the first purely combinatorial $4/3$-approximation for unweighted TAP on general trees (Kortsarz, 14 Jan 2026).
Bundle/CG Cut Lifting: Carefully chosen bundle constraints and odd-cut CG inequalities narrow the LP gap and can be solved efficiently for bounded link costs, leading to $3/2$-type bounds for WTAP (Fiorini et al., 2017, Adjiashvili, 2016).

5. Special Cases, Hardness, and Structural Results

Leaf-to-leaf TAP: Restricting all links to have both endpoints as leaves captures the hardest known TAP instances and enables sharper analysis. Here, a $(4/3+\epsilon)$ -approximation is available, further improved to $1.29$ via stack-based analysis (Cecchetto et al., 2022).
Stemless TAP: Excluding so-called stems (degree-2 nodes whose two incident tree-edges are always covered together by the same augmentation edge) permits structural decomposition leading to tight $3/2+\epsilon$ -approximation via SDP/[SoS] relaxations (Cheriyan et al., 2015).
APX-hardness: TAP is APX-hard even for the leaf-to-leaf case and when each link is only between leaves (Kortsarz, 14 Jan 2026).

6. Distributed and Online Variants

Recently, distributed algorithms for TAP were developed in the CONGEST model, delivering:

Weighted TAP: 2-approximation in $O(h)$ rounds, where $h$ is the tree height.
Unweighted TAP: 4-approximation in $O(D + \sqrt{n}\log^* n)$ rounds, where $D$ is diameter, $n$ the node count (Censor-Hillel et al., 2017).
Online WTAP: Deterministic $O(\log n)$ -competitive, tight up to constants, via path decomposition and online set cover (Joseph et al., 2019).

7. Open Problems and Recent Advances

Breaking $1.393$ barrier for general TAP: Theoretical evidence suggests the true integrality gap of natural LP relaxations may be as low as $4/3$; no counterexamples known, but best presently-achievable guarantee for all general TAP is $1.393$ (Cecchetto et al., 2020).
Combinatorial vs LP/SDP Approaches: New combinatorial techniques such as deferred local-ratio may supersede dependence on heavy LP or SDP machinery in practice (Kortsarz, 14 Jan 2026).
Extensions: General $k$ -edge-connectivity augmentation, higher survivability, and the weighted case with arbitrary costs remain the subject of ongoing research (Cecchetto et al., 2020, Adjiashvili, 2016).
Dynamic and distributed TAP: Designing fully dynamic (edge insert/delete) data structures with near-optimal approximations and extending sub-2 approximations to distributed settings are major open problems (Censor-Hillel et al., 2017).

8. Summary Table: Best Known Approximations and Key References

Setting/Restriction	Approximation	Main Technique(s)	Reference
General TAP (unit costs)	1.393	Stack analysis, CG cuts	(Cecchetto et al., 2020)
Leaf-to-leaf TAP	$4/3+\epsilon$	Matching + O(1)-width decomp	(Cecchetto et al., 2022)
Weighted TAP, costs in $[1, M]$	$3/2+\epsilon$	Odd-cut bundle LP	(Fiorini et al., 2017)
Unweighted TAP, combinatorial	$4/3$	Deferred local ratio	(Kortsarz, 14 Jan 2026)
Weighted TAP, online	$O(\log n)$	Path-decomp, online covering	(Joseph et al., 2019)
Distributed TAP, CONGEST	2, 4 (weighted, unweighted)	Virtual tree, scans	(Censor-Hillel et al., 2017)

References:

This landscape demonstrates that substantial progress has been made in both the theoretical guarantee and algorithmic structure for TAP, though tightening the approximation ratio for both weighted and unweighted cases, and understanding the precise combinatorial barriers, remains an area of active research.