Directed Expander Decomposition

• Directed expander decomposition is a graph partitioning primitive that divides directed graphs into robust expander subgraphs while ensuring a globally acyclic inter-cluster structure.
• It employs techniques such as witness graphs, cut–matching games, and dynamic push–pull–relabel frameworks to certify expansion and manage sparse cuts efficiently.
• Recent advances achieve near-optimal static, dynamic, and capacitated algorithms that enhance performance in flow routing, SCC maintenance, and combinatorial optimization.

Searching arXiv for papers on directed expander decomposition and closely related variants. {"query":"directed expander decomposition arXiv directed expander decomposition near-optimal algorithm capacitated length-constrained", "max_results": 10} I found several relevant arXiv papers:

• (Sulser et al., 2024) — “Near-Optimal Algorithm for Directed Expander Decompositions”
• (Fleischmann et al., 13 Jul 2025) — “Improved Directed Expander Decompositions”
• (Haeupler et al., 29 Mar 2025) — “Length-Constrained Directed Expander Decomposition and Length-Constrained Vertex-Capacitated Flow Shortcuts”
• (Hua et al., 2022) — “Maintaining Expander Decompositions via Sparse Cuts”
• (Goranci et al., 2020) — “The Expander Hierarchy and its Applications to Dynamic Graph Algorithms”

I’ll use these to ground the article. Directed expander decomposition is a family of structural and algorithmic frameworks for partitioning a directed graph into induced subgraphs that are themselves directed expanders, while isolating only a controlled set of inter-component edges that violate a global acyclic orientation. In the conductance-based formulation used by recent work, the central difficulty is that directed acyclic graphs cannot contain nontrivial directed expanders, so a decomposition cannot hope to bound all inter-cluster edges; instead, it bounds only those edges that go “backwards” relative to a topological order of the quotient graph. Since 2024, the area has moved from almost-optimal unweighted decremental frameworks to near-optimal static and dynamic algorithms, near-linear capacitated algorithms, and length-constrained variants with explicit multicommodity routing characterizations (Sulser et al., 2024, Fleischmann et al., 13 Jul 2025, Haeupler et al., 29 Mar 2025).

1. Directed expansion and the decomposition primitive

For a directed graph $G=(V,E)$, recent work uses the undirected degree

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$

and the corresponding volume

$\vol_G(S)=\sum_{v\in S}\deg_G(v)$

for $S\subseteq V$. If $E_G(A,B)$ denotes directed edges from $A$ to $B$, and $e_G(A,B)=|E_G(A,B)|$, then a cut $(S,\bar S)$ is $\phi$-out-sparse when $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$0 and $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$1. A graph is a $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$2-out-expander if it has no $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$3-out-sparse cut. A directed $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$4-expander is defined symmetrically: both $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$5 and its reversal $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$6 must be $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$7-out-expanders, equivalently every nontrivial $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$8 with $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$9 satisfies

$\vol_G(S)=\sum_{v\in S}\deg_G(v)$0

This is a purely combinatorial notion; no spectral definition is involved (Sulser et al., 2024).

The basic directed decomposition problem asks for a partition $\vol_G(S)=\sum_{v\in S}\deg_G(v)$1 of $\vol_G(S)=\sum_{v\in S}\deg_G(v)$2 and a set of removed or special edges $\vol_G(S)=\sum_{v\in S}\deg_G(v)$3 such that every induced subgraph $\vol_G(S)=\sum_{v\in S}\deg_G(v)$4, $\vol_G(S)=\sum_{v\in S}\deg_G(v)$5, is a directed $\vol_G(S)=\sum_{v\in S}\deg_G(v)$6-expander, the number of special edges is bounded by $\vol_G(S)=\sum_{v\in S}\deg_G(v)$7, and the quotient graph $\vol_G(S)=\sum_{v\in S}\deg_G(v)$8 is acyclic. The DAG condition is the directed analogue of the small-boundary condition in undirected expander decomposition. It records that inter-component edges are globally well-oriented except for the few edges in $\vol_G(S)=\sum_{v\in S}\deg_G(v)$9. A common misconception is that directed decomposition should simply minimize all inter-cluster edges; the DAG obstruction shows why this is impossible in general (Sulser et al., 2024).

A related decremental formulation defines a directed expander decomposition of a directed graph $S\subseteq V$0 as a partition $S\subseteq V$1 and a set $S\subseteq V$2 such that each $S\subseteq V$3 is a directed $S\subseteq V$4-expander, $S\subseteq V$5, and the contracted graph $S\subseteq V$6 is a DAG. In that setting, the decomposition is maintained under edge deletions, vertex splits, and self-loop insertions, and the algorithmic objective is often stronger than a one-shot decomposition: the partition should refine over time, while $S\subseteq V$7 should only grow (Hua et al., 2022).

Capacitated variants replace edge counts by total capacities. For a directed capacitated graph $S\subseteq V$8, one writes $S\subseteq V$9 for the total capacity from $E_G(A,B)$0 to $E_G(A,B)$1, and some formulations measure conductance relative to a vertex weighting $E_G(A,B)$2: $E_G(A,B)$3 A graph is then a $E_G(A,B)$4-expander if this quantity is at least $E_G(A,B)$5 for all $E_G(A,B)$6. This weighted formulation is the basis for near-linear capacitated decomposition algorithms (Fleischmann et al., 13 Jul 2025).

2. Witnesses, cut–matching, and flow certificates

A defining feature of the directed literature is the use of witness graphs. In the augmented framework, a pair $E_G(A,B)$7 is a $E_G(A,B)$8-witness for $E_G(A,B)$9 if $A$0 is itself a $A$1-expander on the same vertex set, $A$2 embeds every edge of $A$3 into a path of $A$4 with bounded congestion, and degree bounds satisfy

$A$5

A standard fact quoted in the 2024 near-optimal algorithm is that if $A$6 is a $A$7-witness for $A$8, then $A$9 is a $B$0-expander. Intuitively, a sparse cut in $B$1 would induce a sparse cut in $B$2, contradicting that $B$3 is an expander. This witness formalism is what the dynamic algorithms actually maintain (Sulser et al., 2024).

The witness viewpoint also appears in decremental sparse-cut frameworks. There, an $B$4-witness of $B$5 is a directed graph $B$6 on the same vertex set together with an embedding into $B$7, a slack vector $B$8, and a balance vector $B$9, such that $e_G(A,B)=|E_G(A,B)|$0 certifies out-expansion in both $e_G(A,B)=|E_G(A,B)|$1 and $e_G(A,B)=|E_G(A,B)|$2. The key implication is that if such a witness exists for $e_G(A,B)=|E_G(A,B)|$3, then $e_G(A,B)=|E_G(A,B)|$4 is a $e_G(A,B)=|E_G(A,B)|$5-expander. Witness degradation under deletions is handled by repairing the embedding or extracting a genuine sparse cut (Hua et al., 2022).

Static decomposition algorithms typically begin with a cut–matching game. In the 2024 near-optimal algorithm, the directed cut–matching game either returns a balanced sparse cut or returns a witness $e_G(A,B)=|E_G(A,B)|$6 certifying that $e_G(A,B)=|E_G(A,B)|$7 is a $e_G(A,B)=|E_G(A,B)|$8-expander for a small set of fake edges $e_G(A,B)=|E_G(A,B)|$9. The fake edges are then removed by a pruning subroutine, which extracts a large expander core and recursively decomposes the remaining pieces. This “witness plus pruning” paradigm is central to the 2024 framework (Sulser et al., 2024).

The 2025 improved framework replaces the fake-edge cleanup paradigm by the first implementation and analysis of the non-stop cut-matching game for directed, capacitated graphs. The algorithm outputs disjoint sparse cuts $(S,\bar S)$0, and either the total removed weight is already substantial or the remaining vertex weighting mixes in the graph with congestion $(S,\bar S)$1, where $(S,\bar S)$2. This is the directed analogue of the Saranurak–Wang undirected template, but its analysis requires active/deleted vertex clones, directed bisections, and separate row- and column-mixing arguments (Fleischmann et al., 13 Jul 2025).

3. Algorithmic development and quantitative guarantees

The modern literature can be organized around three milestones: a decremental sparse-cut framework, a near-optimal witness-and-pruning framework, and a capacitated non-stop cut-matching framework.

Paper	Model	Stated guarantee
“Maintaining Expander Decompositions via Sparse Cuts” (Hua et al., 2022)	Directed dynamic graph under edge deletions, vertex splits, and self-loop insertions	Maintain a $(S,\bar S)$3-expander decomposition in total time $(S,\bar S)$4, with refinement and monotone $(S,\bar S)$5
“Near-Optimal Algorithm for Directed Expander Decompositions” (Sulser et al., 2024)	Static and decremental directed graphs	Static time $(S,\bar S)$6; dynamic initialization $(S,\bar S)$7; amortized deletion time $(S,\bar S)$8
“Improved Directed Expander Decompositions” (Fleischmann et al., 13 Jul 2025)	Directed capacitated graphs	Uncapacitated time $(S,\bar S)$9; capacitated time $\phi$0

In the decremental sparse-cut framework, the guiding meta-algorithm is direct: repeatedly find a $\phi$1-sparse cut inside some cluster, split the cluster, and add the smaller directed side of the cut to the global removed-edge set. The significance of the 2022 result is not only the $\phi$2-type update bound, but also the structural guarantees: the maintained partition is always a refinement of the previous one, and the cut set is monotonically nondecreasing. The paper emphasizes that this is the first approach guaranteeing a sublinear $\phi$3 bound on the total number of edges crossing between clusters across the entire update sequence (Hua et al., 2022).

The 2024 near-optimal algorithm improves over previous directed algorithms that were optimal only up to subpolynomial factors. For $\phi$4, it maintains a $\phi$5-directed expander decomposition under edge deletions with

$\phi$6

initialization time

$\phi$7

and amortized deletion time

$\phi$8

Its static algorithm computes a $\phi$9-expander decomposition in time

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$00

For constant $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$01 up to polylogarithmic factors, this is near-linear and near-optimal (Sulser et al., 2024).

The 2025 improved results match the guarantees of Saranurak and Wang for undirected expander decomposition, now in directed graphs. For uncapacitated graphs, the algorithm computes a partition $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$02 and an acyclic subgraph $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$03 such that each induced $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$04 is a $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$05-expander, the number of inter-component edges in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$06 is $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$07, and the running time is $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$08. For capacitated graphs with capacities in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$09, it computes a partition and acyclic subgraph with each part a $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$10-expander, total capacity of inter-component edges in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$11 equal to $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$12, and runtime

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$13

This is stated as the first directed expander decomposition algorithm for capacitated graphs in near-linear time with optimal dependence on $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$14 (Fleischmann et al., 13 Jul 2025).

4. Dynamic maintenance and the push–pull–relabel framework

The decisive technical innovation of the 2024 near-optimal decremental algorithm is the push–pull–relabel framework, a generalization of classical push–relabel from pre-flows to arbitrary pseudo-flows. Classic push–relabel warm-starts successfully in undirected expander pruning because restricting a solved flow to an induced subgraph preserves the pre-flow property. In the directed setting, after cutting off a subset, the restricted object may have negative excess; the resulting object is only a pseudo-flow. This breaks the standard invariants and explains the extra complexity of earlier directed frameworks (Sulser et al., 2024).

The new framework maintains a state $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$15, where $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$16 is a pseudo-flow and $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$17 is a label function satisfying a saturation condition on edges that are “too far downhill.” A valid state additionally requires vertices with negative excess to lie at level $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$18 and vertices with positive excess to lie at level $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$19. The data structure supports two updates: IncreaseSource(δ), which increases the source vector by a nonnegative integer vector, and RemoveVertices(S), which restricts the flow problem to the induced subgraph on the remaining vertices. These are exactly the operations needed when deleted or fake edges create new source mass and when recursive pruning cuts off a set of vertices (Sulser et al., 2024).

Algorithmically, the framework combines two symmetric routines. PushRelabel resolves positive excess using push operations from higher to lower labels and relabels by increasing $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$20. PullRelabel resolves negative excess using a new pull operation that pulls flow back from a higher-level neighbor $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$21 to $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$22 when $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$23, with relabels now decreasing $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$24. Because labels can move both upward and downward, monotonicity is lost; the analysis instead uses the potential

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$25

The resulting complexity bound states that over a sequence of source increases and vertex removals, the total time to maintain valid states is

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$26

where $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$27. This is the mechanism that makes full dynamization possible with only polylogarithmic overhead (Sulser et al., 2024).

The 2022 decremental sparse-cut framework solves the same high-level maintenance problem differently. It maintains multiple levels of witnesses $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$28, slack vectors $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$29, and balance vectors $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$30 for each cluster $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$31, and triggers repair whenever $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$32 exceeds a scale-dependent threshold. Repair invokes a local directed flow computation PruneOrRepair, which either returns a directed sparse out-cut or produces a new witness with reduced $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$33. The resulting decomposition always refines previous decompositions, and the removed-edge set only grows. This monotonicity is central in applications such as decremental SCC maintenance (Hua et al., 2022).

The 2025 capacitated framework returns to a cut-matching-plus-trimming organization, but now with a native directed analysis. Partially matched vertices are handled by splitting each vertex into active and deleted clones; mixing is certified through two potentials, one for rows and one for columns of the flow matrix; and a grafting step either reincorporates difficult vertices into the near-expander or identifies additional sparse cuts. This shows that the natural undirected non-stop cut-matching approach applies even to directed graphs, though only after substantial modifications (Fleischmann et al., 13 Jul 2025).

5. Capacitated and length-constrained variants

Capacitated directed expander decomposition is not a straightforward corollary of unweighted results. The 2025 improved algorithms regularize the vertex weighting by

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$34

where $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$35 is the unweighted degree. This weighting satisfies $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$36 and $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$37, and it allows the dynamic push-relabel analysis inside trimming to bound the total work by $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$38 times the height factor. Because $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$39 dominates the usual degree measure, a $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$40-expander is automatically a $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$41-expander. This regularization is what permits near-linear capacitated runtimes with optimal $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$42 dependence (Fleischmann et al., 13 Jul 2025).

A distinct generalization is length-constrained directed expander decomposition. Here the graph has edge lengths $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$43 and capacities $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$44, demands are constrained to pairs at directed distance at most $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$45, and “cuts” become moving cuts $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$46 that increase edge lengths rather than delete edges. The size of a moving cut is

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$47

and its effect is to replace $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$48 by $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$49. Separation is measured by how much demand becomes farther than a threshold after the length increase (Haeupler et al., 29 Mar 2025).

Given a node-weighting $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$50, the $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$51-length conductance $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$52 is the minimum, over $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$53-length moving cuts, of the worst-case sparsity against symmetric $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$54-respecting $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$55-length demands. One says that $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$56 is $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$57-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$58-expanding if $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$59. An $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$60-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$61-expander decomposition for $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$62 with cut slack $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$63 is then an $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$64-length moving cut $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$65 such that $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$66 is $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$67-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$68-expanding in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$69 and

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$70

This is a genuine extension of the decomposition paradigm: the inter-cluster defect is encoded as added length, not deleted capacity (Haeupler et al., 29 Mar 2025).

The corresponding routing theorem states that if $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$71 is $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$72-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$73-expanding in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$74, then every $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$75-respecting symmetric $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$76-length demand can be routed with congestion $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$77 and length at most $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$78. Conversely, if $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$79 is not $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$80-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$81-expanding, then there exists an $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$82-respecting symmetric $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$83-length demand that cannot be routed with congestion at most $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$84 and length at most $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$85. The decomposition theorem then states that for any directed graph $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$86, node-weighting $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$87, $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$88, $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$89, $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$90, and $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$91, there exists a directed $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$92-length $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$93-expander decomposition for $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$94 with cut slack

$\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$95

This extends length-constrained expander decomposition beyond the undirected edge-capacitated setting (Haeupler et al., 29 Mar 2025).

6. Structural significance, applications, and open directions

Directed expander decomposition has become a standard primitive in modern graph algorithms for flows, shortest paths, and dynamic reachability. The 2024 near-optimal algorithm is already used as a building block in the currently fastest min-cost flow algorithm, and the authors further state that it can speed up and simplify recent breakthroughs in combinatorial graph algorithms toward fast maximum flow algorithms. The same paper explicitly connects directed expander decompositions to min-cost flow and to recent combinatorial maximum-flow and bipartite-matching algorithms, where decomposition localizes augmenting paths, reduces global congestion, and supports dynamic maintenance of residual structures (Sulser et al., 2024).

In decremental directed graphs, the monotone decomposition of the 2022 sparse-cut framework yields especially clean applications. The paper derives a simple decremental SCC algorithm by combining the monotone set $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$96 with a theorem of Łącki for maintaining SCCs under monotone edge deletions outside the current SCC partition. The conceptual point is that refinement of clusters and monotonicity of removed edges are not merely cosmetic; they are algorithmically load-bearing properties (Hua et al., 2022).

The relation to expander hierarchies is more delicate. The 2020 “expander hierarchy” framework is formulated for undirected graphs and explicitly states that there is no sensible generalization of expander hierarchies to directed graphs in that development. Its boundary-linked decomposition and hierarchy tree are nevertheless relevant context because they explain why dynamic graph algorithms seek stronger cluster interfaces than plain conductance guarantees. A plausible implication is that directed decomposition still lacks a comparably canonical hierarchy formalism, even though directed dynamic maintenance now exists (Goranci et al., 2020).

Several misconceptions can be stated precisely. Directed expander decomposition is not an undirected decomposition of the underlying graph followed by arbitrary orientation; the decomposition primitive itself is directional, and the DAG condition is essential. Nor is the area confined to ordinary edge deletion: length-constrained decomposition replaces deletion by moving cuts that increase edge lengths. Finally, fake edges are not intrinsic to the directed setting. The 2024 near-optimal algorithm still uses them as an intermediate certificate that must be pruned away, but the 2025 improved algorithms show that a direct non-stop cut-matching approach can avoid them entirely, even in capacitated graphs (Sulser et al., 2024, Fleischmann et al., 13 Jul 2025).

The open directions stated across the recent literature are concrete. The 2024 near-optimal paper points to reducing the $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$97 loss in $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$98, the $\deg_G(v)=|\{e\in E:e\text{ incident to }v\}|$99 factor in $\vol_G(S)=\sum_{v\in S}\deg_G(v)$00, and the $\vol_G(S)=\sum_{v\in S}\deg_G(v)$01 factor in the update time, extending beyond the decremental model to fully dynamic directed graphs, and generalizing more sharply to weighted or capacitated settings. The length-constrained work identifies almost-linear-time algorithms as a target for future work and suggests that directed flow shortcuts remain out of reach despite the new decomposition theorem. The 2025 improved results indicate no fundamental barrier to matching the exact Saranurak–Wang polylogarithmic factors for capacitated directed graphs. Taken together, these results suggest that directed expander decomposition has progressed from a difficult specialized tool to a central, modular primitive, but that the fully dynamic, hierarchy-level, and length-aware theories are still evolving (Sulser et al., 2024, Haeupler et al., 29 Mar 2025, Fleischmann et al., 13 Jul 2025).