Eulerian Digraph & Definition with Implications

• Eulerian digraphs are directed graphs where every vertex has equal in-degree and out-degree, and whose relevant non-isolated part is connected.
• These digraphs enable the existence of directed Euler circuits, which are closed directed trails traversing every arc exactly once.
• Applications of Eulerian digraphs range from graph theory and network design to toric Calabi–Yau manifolds, significantly impacting theoretical and practical fields.

An Eulerian digraph is a directed graph in which every vertex has equal in-degree and out-degree and whose relevant non-isolated part is connected in the appropriate sense. For a vertex $v$, write $d^+(v)$ and $d^-(v)$ for its out-degree and in-degree. The balance condition is

$d^+(v)=d^-(v)\qquad\text{for every }v.$

For finite digraphs, Euler’s directed theorem identifies this condition together with weak connectivity of the underlying undirected graph with the existence of a directed Euler circuit: a closed directed trail traversing every arc exactly once. Equivalently, an Eulerian digraph is strongly connected and its arc set can be partitioned into arc-disjoint directed cycles. Conventions vary: some works use “Eulerian” for balance alone, allowing disconnected unions of Eulerian components, whereas others include connectivity explicitly (Medeiros, 2010, Sorge, 2011, Janzer et al., 2021).

1. Definitions and fundamental structure

A digraph $\vec G$ is balanced when

$\deg^+(v)=\deg^-(v)$

at every vertex. A balanced digraph is $k$-regular if every vertex has out-degree $k$; if it has $n$ vertices, then it has $e=kn$ arcs and its underlying undirected graph is $d^+(v)$0-regular. Loops contribute one to both in-degree and out-degree. In every digraph,

$d^+(v)$1

The directed Euler theorem gives the equivalence

$d^+(v)$2

Weak connectivity refers to connectivity after orientations are ignored. Strong connectivity means that every ordered pair of vertices is joined by a directed path. Thus strong connectivity alone is insufficient: a strongly connected digraph with unequal in- and out-degrees need not be Eulerian. Conversely, in a balanced digraph, weak connectivity forces strong connectivity on the non-isolated part.

The balance condition implies the cut identity

$d^+(v)$3

for every vertex set $d^+(v)$4. The number of arcs leaving $d^+(v)$5 equals the number entering $d^+(v)$6, because internal arcs contribute equally to the in-degree and out-degree sums over $d^+(v)$7. This identity is central in structural, extremal, embedding, and algorithmic arguments.

Every Eulerian digraph admits an arc-disjoint directed-cycle decomposition. Directed variants of Fleury’s algorithm construct Euler circuits, while the BEST theorem counts them. In a $d^+(v)$8-regular Eulerian digraph, an Euler circuit can be encoded by a chord diagram: each vertex occurs twice on the circuit, and two vertices are interlaced when their occurrences alternate. The associated interlace graph changes by graph pivoting under transpositions of Euler circuits; its interlace polynomial determines circuit-partition enumerators (Medeiros, 2010).

A useful parameter is

$d^+(v)$9

In the toric construction associated with an Eulerian digraph, $d^-(v)$0 is the complex dimension of the resulting affine toric Calabi–Yau variety, while its defining lattice polytope lies in $d^-(v)$1 (Medeiros, 2010).

2. Elementary operations and decompositions

Finite Eulerian digraphs admit a structure theory based on four elementary operations:

1. loop addition;
2. subdivision of an arc;
3. contraction of an undirected simple arc;
4. simple immersion, the reverse of an appropriate vertex-splitting operation.

Loop addition adds a loop at a vertex. Since a loop contributes one to both degrees, balance and Eulerianity are preserved: $d^-(v)$2

Subdivision replaces an arc $d^-(v)$3 by

$d^-(v)$4

The new vertex has in-degree and out-degree one, and

$d^-(v)$5

so $d^-(v)$6 is unchanged. Subdivision of a loop produces a directed $d^-(v)$7-cycle.

Contraction of an undirected simple arc identifies the endpoints of an arc whose underlying undirected edge is the only edge between them. It preserves balance and changes

$d^-(v)$8

The simplicity requirement prevents the contraction from creating a loop in the loopless class.

Splitting a vertex of out-degree two separates its incoming and outgoing arc ends into two balanced groups. It changes

$d^-(v)$9

The reverse operation is a simple immersion; it introduces a new $d^+(v)=d^-(v)\qquad\text{for every }v.$0-regular vertex and replaces two arcs by four arcs.

Every finite Eulerian digraph can be generated from the one-vertex graph with no arcs by iterating these four operations. Loopless smooth representatives are obtained by removing loops and smoothing degree-two vertices. For fixed $d^+(v)=d^-(v)\qquad\text{for every }v.$1, every loopless smooth Eulerian digraph is obtainable from a loopless smooth $d^+(v)=d^-(v)\qquad\text{for every }v.$2-regular Eulerian digraph by contracting undirected simple arcs. The $d^+(v)=d^-(v)\qquad\text{for every }v.$3-regular class is generated from a family of obstruction digraphs $d^+(v)=d^-(v)\qquad\text{for every }v.$4 by simple immersion.

The distinction between immersion and other containment relations is important. An immersion maps vertices injectively and replaces each arc by a directed path, with the paths pairwise edge-disjoint; internal vertices may be shared. A strong immersion additionally forbids branch vertices from occurring internally on immersion paths. Immersion is weaker than subdivision, which generally requires internal vertex-disjointness, and differs from a minor relation based on vertex and edge contraction (Girão et al., 2021, Cavallaro et al., 30 Sep 2025).

3. Quivers, toric Calabi–Yau varieties, and chip-firing

An Eulerian digraph naturally defines an abelian quiver gauge theory. Associate a gauge factor $d^+(v)=d^-(v)\qquad\text{for every }v.$5 to each vertex and a bifundamental field to each arc. The incidence matrix $d^+(v)=d^-(v)\qquad\text{for every }v.$6 assigns $d^+(v)=d^-(v)\qquad\text{for every }v.$7 to the head of arc $d^+(v)=d^-(v)\qquad\text{for every }v.$8, $d^+(v)=d^-(v)\qquad\text{for every }v.$9 to its tail, and $\vec G$0 elsewhere. Its column sums vanish: $\vec G$1 For an Eulerian digraph, balance is exactly the row condition

$\vec G$2

This is the quiver anomaly-cancellation condition.

For a weakly connected digraph, the diagonal $\vec G$3 lies in the kernel, leaving an effective gauge group

$\vec G$4

Choose lattice vectors $\vec G$5 satisfying

$\vec G$6

The cone generated by these vectors defines an affine toric variety of complex dimension

$\vec G$7

Because the row sums of $\vec G$8 vanish, there is an integral vector $\vec G$9 such that

$\deg^+(v)=\deg^-(v)$0

for every arc. Hence all generators lie in a common characteristic hyperplane, the canonical bundle is trivial, and

$\deg^+(v)=\deg^-(v)$1

The associated lattice polytope $\deg^+(v)=\deg^-(v)$2 is the convex hull of the projected generators.

The four graph operations have corresponding polytope transformations. Loop addition produces a pyramid and increases the toric dimension by one. Subdivision leaves the generating set, polytope, and toric variety unchanged. Contraction deletes the lattice point corresponding to the contracted arc and corresponds physically to Higgsing a relative $\deg^+(v)=\deg^-(v)$3. Simple immersion increases the dimension by one and produces a Cayley polytope. In special cases, the resulting polytopes include Lawrence prisms, joins of simplices, and products of independent unit squares. Distinct Eulerian quivers can define the same toric Calabi–Yau geometry, reflecting toric duality and, physically, Seiberg duality (Medeiros, 2010).

Eulerianity also governs chip-firing. After choosing a sink $\deg^+(v)=\deg^-(v)$4, deleting arcs leaving $\deg^+(v)=\deg^-(v)$5 produces a digraph with a global sink. Recurrent configurations form the sandpile group, whose order is

$\deg^+(v)=\deg^-(v)$6

the number of spanning arborescences oriented toward $\deg^+(v)=\deg^-(v)$7. For Eulerian digraphs, the appropriately shifted generating function of recurrent configurations is independent of the sink: $\deg^+(v)=\deg^-(v)$8 It specializes to the undirected partial Tutte polynomial $\deg^+(v)=\deg^-(v)$9, and

$k$0

The sink-independence theorem depends on the Eulerian burning criterion and on the maximum-chip property of recurrent configurations (Perrot et al., 2013).

4. Extremal properties and long directed structures

Eulerian balance imposes strong restrictions on feedback arc sets and directed cycles. If a simple Eulerian digraph has $k$1 vertices and $k$2 arcs, its minimum feedback arc set has size

$k$3

This is optimal for infinitely many pairs $k$4, including suitable Cayley digraphs. The feedback bound implies the short-cycle estimate

$k$5

where $k$6 is the directed girth. It also yields an Eulerian subgraph $k$7 with minimum degree

$k$8

These estimates are tight up to constant factors (Huang et al., 2012).

The same framework gives long-cycle lower bounds. Every Eulerian digraph with $k$9 vertices and $k$0 arcs contains a directed cycle of length at least

$k$1

A later result improved the general square-root estimate: if the average out-degree is $k$2, then the circumference satisfies

$k$3

The proof uses a final spanning out-branching. Its levels are independent, arcs from upper levels to lower levels are forced to be back arcs, and the Eulerian cut identity bounds forward arcs by

$k$4

when $k$5 is the maximum cycle length. Combining forward and back arcs yields the stated inequality (Ai et al., 30 Oct 2025).

For directed paths, a stronger exponent is known. Every connected Eulerian digraph of average degree at least $k$6 contains, from every prescribed starting vertex, a simple directed path of length

$k$7

The proof combines $k$8-full Eulerian subgraphs, robust strong connectivity after deleting small vertex sets, random path selection, shortcutting through high-degree vertices, and Eulerian-preserving replacement of two-edge paths by fake edges (Janzer et al., 2021).

A further density theorem establishes an oriented analogue of the Erdős–Sós phenomenon. If an Eulerian digraph with $k$9 vertices and $n$0 arcs satisfies

$n$1

then it contains every oriented tree with $n$2 edges. The threshold is sharp: a disjoint union of complete bidirected graphs on $n$3 vertices has equality and contains no oriented tree with $n$4 edges. Quantitatively, the number of embeddings of any such tree is at least

$n$5

The proof uses permutation prefixes, balanced-prefix inequalities, prefix gluing, and induction on the oriented tree (Mubayi et al., 10 Sep 2026).

Minimum directed degree also forces dense immersions. There is an absolute constant $n$6 such that

$n$7

in an Eulerian digraph implies an immersion of the complete digraph $n$8. The proof extracts robust bidirectional expansion, routes many edge-disjoint paths, and converts a dense Eulerian multidigraph into a complete-digraph immersion. Eulerian cut balance is essential: without it, large in- and out-degree does not force even a $n$9-immersion (Girão et al., 2021).

5. Algorithms and parameterized problems

The Eulerian Extension problem asks whether arcs of total weight at most a threshold can be added to a directed multigraph so that the result is Eulerian. Added arcs must simultaneously repair degree imbalances and connect components. The problem is NP-hard. If $e=kn$0 is the number of added arcs, it can be solved in

$e=kn$1

time. Parameterizing by the number $e=kn$2 of connected components and the total positive imbalance

$e=kn$3

the problem admits an algorithm with running time

$e=kn$4

A matching reformulation, Conjoining Bipartite Matching, represents imbalance correction by perfect matching and connectivity by conjoining constraints. Fixed-parameter tractability for the component parameter alone remains open, and no polynomial-size kernel is expected for parameters $e=kn$5, $e=kn$6, $e=kn$7, or $e=kn$8 unless $e=kn$9 (Sorge, 2011).

A related problem, Eulerian Strong Component Arc Deletion, deletes a minimum number of arcs so that every remaining strongly connected component is internally Eulerian. This is weaker than making the entire digraph Eulerian: arcs between different strongly connected components are ignored in the internal balance equations. The problem is W[1]-hard parameterized by the deletion budget, para-NP-hard parameterized by maximum degree, and W[1]-hard parameterized by the vertex-cover number. It is in XP parameterized by treewidth and is FPT parameterized by treewidth plus maximum degree or by treewidth plus solution size. The problem remains NP-hard even when degree pairs belong to

$d^+(v)$00

(Crowston et al., 2011, Blažej et al., 2024).

For semicomplete digraphs, spanning Eulerian subdigraphs behave more regularly. Every $d^+(v)$01-arc-strong semicomplete digraph has a spanning Eulerian subdigraph containing any prescribed arc, one avoiding any prescribed arc, and a spanning trail between every ordered pair of distinct vertices. For $d^+(v)$02 prescribed arcs, $d^+(v)$03-arc-strongness is conjectured to suffice for avoidance; the proved general upper bound is

$d^+(v)$04

while $d^+(v)$05, and equality is proved for $d^+(v)$06 and for prescribed arc sets forming forests of stars (Bang-Jensen et al., 2019).

Spectral methods provide separation bounds for connected Eulerian digraphs whose out-degree Laplacian

$d^+(v)$07

is normal. If $d^+(v)$08 are disjoint and no arc goes from $d^+(v)$09 to $d^+(v)$10, then the sizes of $d^+(v)$11 and $d^+(v)$12 are bounded in terms of the complex Laplacian eigenvalues. The proof uses a symmetric block dilation of a shifted Laplacian and eigenvalue interlacing. The normality hypothesis is restrictive but includes Cayley digraphs of abelian groups and regular tournaments (Guo, 2015).

6. Embeddings, topology, and enumerative theory

A directed embedding is one in which incoming and outgoing half-arcs alternate around every vertex; equivalently, every facial boundary follows arc directions. A connected Eulerian digraph always admits an orientable directed embedding. A bi-Eulerian embedding has exactly two faces, each bounded by a directed Euler circuit.

For quartic Eulerian digraphs, diplanarity means an embedding in the plane with alternating local orientations. Cycle removal followed by suppression defines a partial order under which diplanarity is hereditary. Minimal non-diplanar obstructions have no loops, at most two undirected edges between any pair of vertices, are $d^+(v)$13-edge-connected, and are strongly $d^+(v)$14-edge-connected. Obstructions with digons are generated from digon-free obstructions by admissible vertex splitting and digon insertion. Known families include doubled directed cycles, directed Möbius ladders, anti-ladders, cyclic block families, $d^+(v)$15, and $d^+(v)$16. A complete obstruction classification remains open (Archdeacon et al., 2017).

Degree modulo four controls orientable bi-Eulerian embeddings. If every vertex of an Eulerian digraph has total degree $d^+(v)$17, then any prescribed directed Euler circuit can bound one face of an orientable bi-Eulerian embedding. If exactly two vertices have degree $d^+(v)$18, the same conclusion holds when those vertices are interlaced along the prescribed circuit. More generally, if $d^+(v)$19 vertices have degree $d^+(v)$20, there is an orientable directed embedding with one prescribed face and at most $d^+(v)$21 additional faces, yielding orientable genus at least

$d^+(v)$22

Every nontrivial Eulerian digraph also has a one-face nonorientable directed embedding, giving maximum Euler genus $d^+(v)$23 (Ellingham et al., 2024).

For sufficiently dense underlying simple graphs, the orientable embedding problem becomes sharper. If

$d^+(v)$24

and $d^+(v)$25 is any prescribed directed circuit decomposition, then $d^+(v)$26 has an oriented directed embedding in which the circuits of $d^+(v)$27 are profaces and the number of additional antifaces is exactly one or two, according to the parity of

$d^+(v)$28

This embedding has maximum orientable genus relative to the prescribed circuits. For a specified Euler circuit $d^+(v)$29, the parity condition $d^+(v)$30 yields a bi-Eulerian embedding with $d^+(v)$31 as one face (Ellingham et al., 2024).

Eulerian digraphs also support several enumerative correspondences. Dyck words are in bijection with a restricted family of ordered, cycle-labelled Eulerian multidigraphs through Dyck matrices. Matrix rows represent directed cycles, columns represent vertices, and the row-transition rules encode controlled intersections of consecutive cycles. The bijection transfers Catalan enumeration to these cycle-matrix structures (Codara et al., 2014).

In chip-firing, the sink-independent polynomial $d^+(v)$32 extends the partial Tutte specialization. In circuit enumeration, if $d^+(v)$33 denotes partitions of Eulerian circuits into $d^+(v)$34 circuits, then every connected Eulerian digraph other than a single directed cycle satisfies

$d^+(v)$35

The proof uses the refinement semilattice of connected Eulerian edge partitions, heaps of directed cycles, unique-sink orientations of cycle-intersection graphs, bond-lattice Möbius inversion, and NBC theory. The cancellation yields an application to the classical Harary–Sachs theorem for graphs (Cooper et al., 2 Feb 2025).

Finally, Eulerian digraphs provide overlap graphs for universal cycles. Allowed words, lattice paths, random walks, monotone words, Lipschitz words, augmented-onto functions, and cyclic-category words become arcs of a shift-overlap digraph. If the overlap digraph is weakly connected and balanced, its Euler circuit is a universal cycle containing every allowed object exactly once as a cyclic block (Gardner et al., 2017). This illustrates the central role of Eulerian balance: it converts local extension compatibility into a global cyclic enumeration.