---
title: Linear Arboricity Conjecture Insights
url: https://www.emergentmind.com/topics/linear-arboricity-conjecture
type: topic
---

# Linear Arboricity Conjecture Insights

Searching arXiv for recent and foundational papers on the Linear Arboricity Conjecture.
arxiv_search(query="Linear Arboricity Conjecture", max_results=10, sort_by="relevance")
arxiv_search({"query":"Linear Arboricity Conjecture","max_results":10,"sort_by":"relevance"})
I’ll look up relevant arXiv papers on the Linear Arboricity Conjecture and related list/directed variants.
<tool name="arxiv_search">
{"query":"Linear Arboricity Conjecture","max_results":10,"sort_by":"relevance"}
</tool>
The Linear Arboricity Conjecture concerns decompositions of graphs into path-forests. A linear forest is a forest whose connected components are paths, and the linear arboricity of a graph $G$, denoted $\operatorname{la}(G)$, is the minimum number of linear forests whose edge sets partition $E(G)$. Introduced by Akiyama, Exoo, and Harary, the conjecture asserts that every graph of maximum degree $\Delta$ satisfies
\[
\left\lceil \frac{\Delta}{2}\right\rceil \le \operatorname{la}(G) \le \left\lceil \frac{\Delta+1}{2}\right\rceil.
\]
The problem sits at the intersection of edge-coloring, sparse decomposition, and Hamiltonian methods, and has developed into a broader theory including list versions, directed analogues, dense-graph confirmations, random-graph confirmations, and recent logarithmic-error bounds for general graphs [1712.05006][2507.20500][2512.20195].

## 1. Definitions, lower bounds, and equivalent forms

A linear forest is equivalently a graph of maximum degree at most $2$ with no cycles. In some parts of the literature the linear arboricity is denoted $\chi'_l(G)$, but the notation $\operatorname{la}(G)$ is now standard. A linear coloring of $G$ is an edge-coloring in which each color class induces a linear forest; then $\operatorname{la}(G)$ is the minimum number of colors in such a coloring [2007.06066].

The basic lower bound is immediate. If $v$ is a vertex of degree $\Delta(G)$, then each linear forest contributes at most two edges incident to $v$, so
\[
\operatorname{la}(G)\ge \left\lceil \frac{\Delta(G)}{2}\right\rceil.
\]
For regular graphs there is a stronger obstruction. Harary observed that for $\Delta$-regular graphs one has
\[
\operatorname{la}(G)\ge \left\lceil \frac{\Delta(G)+1}{2}\right\rceil,
\]
and because every graph is a subgraph of a $\Delta(G)$-regular graph, settling the conjecture for regular graphs suffices. In particular, the usual “regular form” of the conjecture is that every $d$-regular graph satisfies
\[
\operatorname{la}(G)=\left\lceil \frac{d+1}{2}\right\rceil.
\]
This is equivalent to the maximum-degree formulation [2007.06066][1712.05006].

A common misunderstanding is to treat the conjecture as a universal $\lceil \Delta/2\rceil$ statement. The standard formulation is $\lceil (\Delta+1)/2\rceil$, and the distinction matters when $\Delta$ is even. The parity obstruction is explicit in regular graphs: if $d$ is even and one tried to decompose a $d$-regular graph into $d/2$ linear forests, then every vertex would have degree exactly $2$ in every color class, forcing each class to be $2$-regular and hence a union of cycles rather than a forest [1712.05006].

The conjecture is tight in natural classes. Complete graphs satisfy $\operatorname{la}(K_n)=\lceil n/2\rceil$, which equals $\lceil (\Delta(K_n)+1)/2\rceil$ since $\Delta(K_n)=n-1$ [2405.18494]. More generally, when $\Delta(G)$ is odd, the lower bound $\lceil \Delta(G)/2\rceil$ and the conjectured upper bound $\lceil (\Delta(G)+1)/2\rceil$ coincide; when $\Delta(G)$ is even, the exact value may be either $\Delta(G)/2$ or $\Delta(G)/2+1$, and deciding which occurs is NP-complete in general [2405.18494].

## 2. Classical statement and historical context

The conjecture emerged as a degree-$2$ analogue of edge-coloring and arboricity problems. Whereas matchings correspond to color classes of maximum degree $1$, linear forests allow degree $2$ but forbid cycles, so the conjecture interpolates between edge-coloring and acyclic decompositions. This perspective is explicit in later surveys and extensions, including the list version and degree-$f$ generalizations [1712.05006][2301.09933].

The early literature established the conjecture in several exact cases. It is known for graphs with $\Delta(G)\le 6$ and for $\Delta(G)\in\{8,10\}$, and exact results were also obtained for complete graphs, complete bipartite graphs, series-parallel graphs, and planar graphs [1712.05006][2207.07169]. These exact confirmations showed that the conjectured bound is not merely asymptotically plausible but structurally sharp in a wide range of families.

The conjecture has also generated stronger local versions. One such strengthening is the List Linear Arboricity Conjecture, due to An and Wu, which asserts that the list analogue satisfies
\[
\left\lceil \frac{\Delta}{2}\right\rceil \le \operatorname{la}(G)=\operatorname{lla}(G)\le \left\lceil \frac{\Delta+1}{2}\right\rceil.
\]
Another line extends the problem to digraphs, where the components are directed paths rather than undirected paths [1712.05006][2512.20195].

## 3. Quantitative progress for general graphs

For unrestricted graphs, progress was long measured by how close one could get to $\Delta/2$. The asymptotic program began with Alon’s proof that
\[
\operatorname{la}(G)\le \frac{\Delta}{2}+O\!\left(\Delta\cdot \frac{\log\log\Delta}{\log\Delta}\right),
\]
followed by Alon–Spencer’s improvement to
\[
\operatorname{la}(G)\le \frac{\Delta}{2}+O\!\big(\Delta^{2/3}\log^{1/3}\Delta\big),
\]
and Ferber–Fox–Jain’s further reduction to $O(\Delta^{2/3-\alpha})$ for some $\alpha>0$ [2008.04251][1809.04716].

| Regime | Bound for $\operatorname{la}(G)$ | Source |
|---|---:|---|
| Asymptotic general bound | $\frac{\Delta}{2}+O\!\left(\Delta\frac{\log\log\Delta}{\log\Delta}\right)$ | [2008.04251] |
| Improved asymptotic bound | $\frac{\Delta}{2}+O\!\big(\Delta^{2/3}\log^{1/3}\Delta\big)$ | [2008.04251] |
| Polynomial improvement | $\frac{\Delta}{2}+O(\Delta^{2/3-\alpha})$ | [1809.04716] |
| Nibble/list-era bound | $\frac{\Delta}{2}+3\sqrt{\Delta}(\log\Delta)^4$ | [2008.04251] |
| Logarithmic-error bound | $\frac{\Delta}{2}+O(\log n)$ | [2507.20500] |

A major step came from the modified nibble framework of Lang and Postle, which gave
\[
\operatorname{la}(G)\le \frac{\Delta}{2}+3\sqrt{\Delta}(\log\Delta)^4
\]
for sufficiently large $\Delta$, and did so in the stronger list setting. The proof duplicated each color into “twin colors,” reduced the problem to a proper edge-coloring problem in product lists, and then prevented twin-alternating cycles so that each original color class became an acyclic degree-$2$ subgraph, hence a linear forest [2008.04251].

A further development in 2025 broke the square-root barrier. The bound
\[
\operatorname{la}(G)\le \frac{\Delta}{2}+(\log n)
\]
was proved for every $n$-vertex graph of maximum degree $\Delta$, where “$(\log n)$” denotes $O(\log n)$ with an absolute constant. The same work proved
\[
\operatorname{fla}(G)\le \frac{\Delta}{2}+(\log\Delta)
\]
for fractional linear arboricity. The method generalized Pósa rotations from single endpoints of a path to simultaneous rotations of multiple endpoints of a linear forest, and when $\Delta=\Omega(n^\varepsilon)$ it gives an exponential improvement over the previous best error term on the log-scale [2507.20500].

These quantitative results collectively reposition the conjecture. The obstruction is no longer merely asymptotic validity: for general graphs, the gap is now logarithmic in $n$ rather than polynomial in $\Delta$.

## 4. Exact and near-exact results in structured graph classes

A large body of exact work concerns sparse classes described by degeneracy. A graph is $k$-degenerate if every subgraph has a vertex of degree at most $k$. Chen, Hao, and Yu proved that every $k$-degenerate graph $G$ satisfies
\[
\operatorname{la}(G)=\left\lceil \frac{\Delta(G)}{2}\right\rceil
\quad\text{whenever}\quad
\Delta(G)\ge 2k^2-k,
\]
and hence satisfies the conjectured upper bound already when $\Delta(G)\ge 2k^2-2k$ [2207.07169]. This result is constructive and shows that on sufficiently high-degree degenerate graphs the trivial lower bound is attained exactly.

Subsequent work sharpened the low-degeneracy thresholds. For $3$-degenerate graphs,
\[
\chi'_l(G)=\left\lceil \frac{\Delta(G)}{2}\right\rceil
\quad\text{for}\quad
\Delta(G)\ge 9,
\]
improving the earlier threshold $\Delta(G)\ge 15$. The same paper gave a different proof of the Linear Arboricity Conjecture for all $3$-degenerate graphs and proved that every $2$-degenerate graph satisfies
\[
\chi'_l(G)=\left\lceil \frac{\Delta(G)}{2}\right\rceil
\quad\text{if}\quad
\Delta(G)\ge 5.
\]
These results came with linear-time algorithms and detailed control over monochromatic vertices [2007.06066].

Dense graphs admit a different exact regime. For every fixed $0<\varepsilon<1$, there exists $n_0$ such that if $G$ has $n\ge n_0$ vertices and minimum degree
\[
\delta(G)\ge \frac{(1+\varepsilon)n}{2},
\]
then
\[
\operatorname{la}(G)\le \left\lceil \frac{\Delta(G)+1}{2}\right\rceil.
\]
The proof regularizes the graph by path removals, uses robust expansion and Hamilton-decomposition theorems, and then converts Hamilton cycles into path decompositions by breaking them at auxiliary vertices [2405.18494].

Random graphs satisfy an even stronger statement in a sparse regime. If $C>0$ is sufficiently large and
\[
C\frac{\log n}{n}\le p(n)\le n^{-2/3},
\]
then with high probability
\[
\operatorname{la}(G(n,p))\le \left\lceil \frac{\Delta(G(n,p))}{2}\right\rceil.
\]
Thus, in this range the random graph attains the optimum lower bound with high probability, which is “even slightly less than in the linear arboricity conjecture” when $\Delta$ is even [2310.11580].

Large girth provides another structured route. For a $2k$-regular graph $G$ with girth $g(G)\ge 2k$, one has
\[
\operatorname{la}(G)\le k+1,
\]
which is exactly the conjectured value for $\Delta=2k$. More generally, the same work proved girth-indexed bounds such as $\operatorname{la}(G)\le k+2$ when $g(G)\ge k$ and $\operatorname{la}(G)\le k+3$ when $g(G)\ge k/2$, using a 2-factor decomposition and an auxiliary flow network with lower bound constraints to find a sparse cycle-transversal subgraph [2512.11240].

## 5. List linear arboricity

The list version replaces a common palette by edge-specific permissible sets. A list assignment $L$ gives each edge $e$ a set $L(e)$ of colors, and a linear $L$-coloring is a map $\phi:E(G)\to \bigcup_e L(e)$ with $\phi(e)\in L(e)$ such that each color class induces a linear forest. The list linear arboricity $\operatorname{lla}(G)$ is the minimum $k$ such that every list assignment with $|L(e)|\ge k$ for all edges admits a linear $L$-coloring. Clearly,
\[
\operatorname{la}(G)\le \operatorname{lla}(G)
\]
[1712.05006][2512.20195].

An and Wu’s conjectural strengthening states that the list parameter should equal the ordinary one:
\[
\left\lceil \frac{\Delta}{2}\right\rceil
\le
\operatorname{la}(G)=\operatorname{lla}(G)
\le
\left\lceil \frac{\Delta+1}{2}\right\rceil.
\]
Kim and Postle proved this asymptotically: for every $\varepsilon>0$ and sufficiently large $\Delta(G)$,
\[
\operatorname{lla}(G)\le \left(\frac{\Delta(G)}{2}\right)(1+\varepsilon),
\]
and more strongly, in a color-degree formulation, if $|L|\ge (d/2)(1+\varepsilon)$ and $\Delta_G^L\le d$, then $G$ admits a linear $L$-coloring [1712.05006].

The strongest quantitative list bound in the data is
\[
\operatorname{lla}(G)\le \frac{\Delta}{2}+3\sqrt{\Delta}(\log\Delta)^4
\]
for sufficiently large $\Delta$. This bound was obtained by tying linear arboricity to the List Colouring Conjecture through color duplication: each original color $c$ is replaced by twins $(c,1)$ and $(c,2)$, one constructs a proper edge-coloring from the product lists, and then forbids twin-alternating cycles. Properness at the twin level gives maximum per-color degree at most $2$, while the cycle condition forces acyclicity of each original color class [2008.04251].

This list framework is not merely technical strengthening. It is the setting in which the best undirected asymptotic bound was first proved, and it became the template for later directed extensions [2512.20195].

## 6. Directed analogues and related generalizations

For a digraph $D$, a directed linear forest is a disjoint union of directed paths, and $\operatorname{la}(D)$ is the minimum number of directed linear forests needed to partition $A(D)$. The degree parameter is
\[
\Delta(D):=\max\{\Delta^+(D),\Delta^-(D)\},
\]
where $\Delta^+(D)$ and $\Delta^-(D)$ are the maximum out-degree and in-degree. Nakayama and Péroche conjectured that every digraph satisfies
\[
\operatorname{la}(D)\le \Delta(D)+1
\]
[2512.20195].

The directed problem has a sharper small-exception structure than the undirected one. For a $d$-regular digraph, every directed forest has at most $|V(D)|-1$ arcs, so
\[
\operatorname{la}(D)\ge
\left\lceil \frac{d\,|V(D)|}{|V(D)|-1}\right\rceil
\ge d+1.
\]
Nakayama–Péroche’s conjecture is therefore equivalent to $\operatorname{la}(D)=d+1$ for all $d$-regular digraphs. However, He et al. found counterexamples: the symmetric complete digraphs $K_3^*$ and $K_5^*$ satisfy $\operatorname{la}(K_3^*)=4=d+2$ and $\operatorname{la}(K_5^*)=6=d+2$ [2512.20195].

The strongest directed/list asymptotic result in the data is the theorem
\[
\operatorname{lla}(D)\le \Delta(D)+6\sqrt{\Delta(D)}(\log\Delta(D))^4
\]
for sufficiently large $\Delta(D)$. It extends the Lang–Postle undirected bound to digraphs with a matching asymptotic error term, and it works in the stronger list setting. The proof adapts the semi-random Rödl nibble to enforce a partial $(1,1)$-edge-coloring, introduces suspicious directed paths to prevent monochromatic directed cycles, and combines the Lovász Local Lemma, Talagrand’s inequality, and reserve-color arguments in the style of Molloy–Reed. As an immediate consequence, any Eulerian orientation of a $2d$-regular graph admits a decomposition into at most
\[
d+6\sqrt d\,(\log d)^4
\]
directed linear forests [2512.20195].

The conjecture has also been placed inside broader degree-constrained coloring theories. One direction studies degree-$f$ arboricity and its directed branching analogue. In that setting, Truszczyński’s conjecture was disproved for general multigraphs, but positive large-girth and asymptotic results were proved for simple graphs and digraphs. For bounded $f$, this framework still yields asymptotic confirmations of the linear arboricity conjecture in the case $f\equiv 2$ [2301.09933]. Another orientation-based approach proved that the Linear Arboricity Conjecture holds for all $k$-degenerate loopless multigraphs when
\[
\Delta(G)\ge 4k-2,
\]
improving a previously announced quadratic threshold for simple graphs and extending the conclusion to multigraphs [2108.11816].

## 7. Methods, significance, and open problems

Three methodological strands dominate the modern theory. The first is probabilistic and local: nibble arguments, concentration inequalities, and the Lovász Local Lemma. This line underlies the best list and directed asymptotic bounds, where the main difficulty is to control both degree constraints and monochromatic cycles while lists evolve under random deletions [2008.04251][2512.20195]. The second is Hamiltonian and global: robust expansion, Hamilton decompositions, and path-removal regularization, which have proved decisive in dense graphs and random graphs [2405.18494][2310.11580]. The third is structural and combinatorial: generalized Pósa rotations, flow transversals, and pivot-based recoloring schemes, which drive the 2025 logarithmic-error bound, the girth-based results, and the low-degeneracy exact theorems [2507.20500][2512.11240][2007.06066].

Algorithmic status is mixed. Some sparse-class results are fully constructive and even linear-time, notably for $2$- and $3$-degenerate graphs [2007.06066]. The dense minimum-degree theorem yields a polynomial-time algorithm, though with very large hidden constants from the regularity and expansion machinery [2405.18494]. By contrast, the $O(\log n)$ general bound is explicitly non-constructive: the rotation-component analysis and the distribution on minimum linear forests are not currently accompanied by a polynomial-time implementation [2507.20500].

Several open problems remain central. The primary one is the conjecture itself for arbitrary graphs. On the quantitative side, the directed/list work highlights the challenge of removing polylogarithmic factors and pushing the error term from $\Delta/2+O(\sqrt{\Delta}(\log\Delta)^4)$ or $\Delta+O(\sqrt{\Delta}(\log\Delta)^4)$ toward $O(\sqrt{\Delta})$, and ultimately toward the exact conjectured constants [2512.20195]. In sparse classes, the exact threshold for $k$-degenerate graphs is unresolved: the quadratic threshold $2k^2-k$ is unlikely to be optimal, and reducing it to a linear function of $k$, ideally $2k$, is explicitly posed as a natural direction [2207.07169][2007.06066]. For dense graphs, it remains open whether the exact bound can be proved under the bare Dirac threshold $\delta(G)\ge n/2$ rather than $\delta(G)\ge (1+\varepsilon)n/2$ [2405.18494]. In the directed setting, the gap between the asymptotic list bound and the conjectural $\Delta(D)+1$ bound remains substantial, especially outside the known small symmetric complete exceptions [2512.20195].

Taken together, these developments show that the Linear Arboricity Conjecture is no longer only a classical extremal statement. It has become a focal point for interactions among edge-coloring, list-coloring, degeneracy theory, Hamilton decomposition, probabilistic methods, and directed graph decomposition. The exact conjecture remains open, but the current landscape is markedly sharper: logarithmic-error bounds are known in full generality, exact confirmations are known in dense, random, and several sparse regimes, and the list and directed theories now mirror much of the undirected asymptotic structure [2507.20500][2310.11580][2512.20195].

Source: https://www.emergentmind.com/topics/linear-arboricity-conjecture