Neighborhood-Balanced k-Coloring

• Neighborhood-balanced k-coloring is a graph coloring paradigm that partitions vertices into k classes so that every vertex sees an equal number of neighbors in each color class.
• It applies to diverse graph families such as regular, circulant, and Hamming graphs, with necessary conditions based on degree divisibility and uniform edge-distribution identities.
• The topic also encompasses algorithmic challenges, structural generalizations, and NP-completeness issues in determining both open and closed neighborhood-balanced colorings.

Neighborhood-balanced $k$-coloring is a local color-distribution constraint on a graph $G=(V,E)$ in which the open neighborhood of every vertex is required to be perfectly equidistributed among $k$ colors. In the standard formulation, a coloring $c:V\to\{1,\dots,k\}$ is neighborhood-balanced if for every vertex $v$ and all colors $i,j$, $|N(v)\cap V_i|=|N(v)\cap V_j|$, where $V_i=c^{-1}(i)$ (Almeida et al., 7 Sep 2025). The notion generalizes the 2-color setting of Freyberg and Marr and the 3-color case studied by Minyard et al., and it has since expanded into a broader family of open-neighborhood, closed-neighborhood, quasi-balanced, and $\lambda$-balanced theories, with distinct structural and algorithmic behavior (Almeida et al., 7 Sep 2025).

1. Formal definition and algebraic encodings

For a simple graph $G=(V,E)$ and an integer $G=(V,E)$0, a neighborhood-balanced $G=(V,E)$1-coloring is a partition

$G=(V,E)$2

such that every vertex sees the same number of neighbors in each color class: $G=(V,E)$3 Equivalently, if $G=(V,E)$4, then each vertex satisfies

$G=(V,E)$5

Thus the definition is intrinsically local: it constrains neighborhood color histograms rather than forbidding monochromatic edges (Almeida et al., 7 Sep 2025).

Several papers use signed encodings of the colors. For odd $G=(V,E)$6, colors may be encoded by $G=(V,E)$7; for even $G=(V,E)$8, by $G=(V,E)$9. If

$k$0

then neighborhood-balance is equivalent to $k$1 for all vertices $k$2. In the odd-prime formulation, this symmetric encoding is used systematically to express $k$3-neighborhood balance as a zero-sum condition on every neighborhood (Almeida, 12 May 2025).

The underlying colorings are not proper colorings. In the generalized $k$4-colored framework for neighborhood-balance and $k$5-balance, a coloring is a partition into color classes, and adjacent vertices are permitted to receive the same color (Collins et al., 5 Mar 2026).

2. Necessary conditions and counting identities

The most immediate necessary condition is divisibility of degrees. If $k$6 admits a neighborhood-balanced $k$7-coloring, then every degree is a multiple of $k$8: $k$9 If $c:V\to\{1,\dots,k\}$0 has no isolated vertices, then $c:V\to\{1,\dots,k\}$1. There is also a coarsening principle: if $c:V\to\{1,\dots,k\}$2 admits a neighborhood-balanced $c:V\to\{1,\dots,k\}$3-coloring and $c:V\to\{1,\dots,k\}$4, then $c:V\to\{1,\dots,k\}$5 also admits a neighborhood-balanced $c:V\to\{1,\dots,k\}$6-coloring (Almeida et al., 7 Sep 2025).

Neighborhood-balance imposes rigid edge-distribution identities. If $c:V\to\{1,\dots,k\}$7 is a neighborhood-balanced $c:V\to\{1,\dots,k\}$8-coloring with color classes $c:V\to\{1,\dots,k\}$9, then for all $v$0,

$v$1

and for each $v$2,

$v$3

Hence every unordered pair of distinct colors supports the same number of edges, and each monochromatic color class induces the same number of edges (Almeida et al., 7 Sep 2025).

For $v$4-regular graphs, the consequences are stronger. Every color class has the same size: $v$5 so $v$6. In addition,

$v$7

These are necessary conditions only; they are not sufficient in general (Almeida et al., 7 Sep 2025).

The 3-color case sharpens these constraints. If a graph is 3-balanced, then $v$8 for every vertex, hence $v$9. More precisely,

$i,j$0

so $i,j$1. For an $i,j$2-regular 3-balanced graph, $i,j$3, and in the cubic case one gets

$i,j$4

hence $i,j$5 (Minyard et al., 2024).

3. Canonical graph classes and constructive families

Complete graphs are excluded from the open-neighborhood theory except in the trivial one-vertex case. More precisely, no $i,j$6 with $i,j$7 admits a neighborhood-balanced $i,j$8-coloring, because the simultaneous conditions $i,j$9 and $|N(v)\cap V_i|=|N(v)\cap V_j|$0 are impossible for $|N(v)\cap V_i|=|N(v)\cap V_j|$1 (Almeida et al., 7 Sep 2025).

Complete multipartite graphs admit an exact characterization. A complete multipartite graph $|N(v)\cap V_i|=|N(v)\cap V_j|$2 is neighborhood-balanced $|N(v)\cap V_i|=|N(v)\cap V_j|$3-colorable if and only if every part size is divisible by $|N(v)\cap V_i|=|N(v)\cap V_j|$4: $|N(v)\cap V_i|=|N(v)\cap V_j|$5 The sufficiency construction colors each part evenly with all $|N(v)\cap V_i|=|N(v)\cap V_j|$6 colors, and the converse follows from the neighborhood equations for vertices inside each part (Almeida et al., 7 Sep 2025).

Hamming graphs provide a high-dimensional regular family with a clean criterion. The Hamming graph $|N(v)\cap V_i|=|N(v)\cap V_j|$7 is neighborhood-balanced $|N(v)\cap V_i|=|N(v)\cap V_j|$8-colorable if and only if

$|N(v)\cap V_i|=|N(v)\cap V_j|$9

The proof is constructive: for $V_i=c^{-1}(i)$0, a coloring is built inductively using cyclic shifts across nested decompositions into smaller Hamming graphs. The hypercube $V_i=c^{-1}(i)$1 is the special case $V_i=c^{-1}(i)$2, so it is neighborhood-balanced 2-colorable exactly when $V_i=c^{-1}(i)$3 is even (Almeida et al., 7 Sep 2025).

Circulant graphs furnish another systematic source of examples. If

$V_i=c^{-1}(i)$4

then the circulant graph $V_i=c^{-1}(i)$5 is neighborhood-balanced $V_i=c^{-1}(i)$6-colored. A second sufficient condition requires that $V_i=c^{-1}(i)$7 and the number of generators be multiples of $V_i=c^{-1}(i)$8, and that the step set be equidistributed modulo $V_i=c^{-1}(i)$9; then the cyclic coloring $\lambda$0 yields balanced neighborhoods (Almeida et al., 7 Sep 2025).

The property is also stable under several graph operations. If $\lambda$1 and $\lambda$2 admit neighborhood-balanced $\lambda$3-colorings, then so do the Cartesian product $\lambda$4, the strong product $\lambda$5, and the lexicographic product $\lambda$6. The direct product $\lambda$7 is neighborhood-balanced if at least one factor is. Joins $\lambda$8 are neighborhood-balanced provided both factors admit neighborhood-balanced $\lambda$9-colorings with equal color-class sizes; in particular this holds for regular balanced factors (Almeida et al., 7 Sep 2025).

4. The 3-color regime and cubic structure

The 3-color theory is unusually rigid. For a 3-balanced coloring $G=(V,E)$0, the generalized Petersen graph $G=(V,E)$1 is 3-balanced if and only if

$G=(V,E)$2

The sufficiency coloring is simply $G=(V,E)$3, while the nonexistence direction for $G=(V,E)$4 is obtained from a linear system built from circulant blocks and an argument using roots of unity (Minyard et al., 2024).

Generalized Pappus graphs satisfy an analogous criterion. The graph $G=(V,E)$5 is 3-balanced if and only if

$G=(V,E)$6

Again, the positive direction uses the residue coloring $G=(V,E)$7, and the negative direction uses an aggregated linear system over residue classes modulo powers of $G=(V,E)$8 (Minyard et al., 2024).

Among other cubic families, the Möbius ladder $G=(V,E)$9 is 3-balanced exactly when $G=(V,E)$00. More broadly, cubic 3-balanced graphs admit strong internal characterizations. If a cubic graph is 3-balanced, then the induced edge-labeling

$G=(V,E)$01

is a Tait coloring: a proper 3-edge-coloring in which each edge color class is a perfect matching, and the union of any two edge colors is a vertex-disjoint union of alternating cycles. Thus no snark is 3-balanced (Minyard et al., 2024).

Two structural characterizations are especially notable. First, a cubic graph with a Tait coloring is 3-balanced if and only if a cycle-based alternating-sum invariant depends only on the base vertex; if any even cycle passes through that vertex, the alternating sum is forced to be $G=(V,E)$02. Second, every cubic 3-balanced graph can be encoded by a partition $G=(V,E)$03 together with bijections

$G=(V,E)$04

satisfying $G=(V,E)$05 and with $G=(V,E)$06 fixed-point-free; conversely, such a dataset reconstructs a cubic 3-balanced graph (Minyard et al., 2024).

At small order the classification is explicit. Both connected cubic graphs on 6 vertices, namely $G=(V,E)$07 and the triangular prism, are 3-balanced. Among the 85 connected cubic graphs on 12 vertices, exactly 17 are 3-balanced (Minyard et al., 2024).

5. Variants, relaxations, and structural generalizations

The standard notion concerns open neighborhoods, but a parallel theory studies closed neighborhood-balanced $G=(V,E)$08-coloring, where every closed neighborhood $G=(V,E)$09 is equally colored. If $G=(V,E)$10 admits such a coloring, then

$G=(V,E)$11

Complete graphs behave differently in this setting: $G=(V,E)$12 is closed neighborhood-balanced $G=(V,E)$13-colorable if and only if $G=(V,E)$14. The Hamming graph $G=(V,E)$15 is closed neighborhood-balanced $G=(V,E)$16-colorable if and only if

$G=(V,E)$17

Closed balance is also preserved by several graph operations, including strong products, certain Cartesian products, lexicographic products under equal-class hypotheses, and joins under equal-class hypotheses; by contrast, if both $G=(V,E)$18 and $G=(V,E)$19 are closed neighborhood-balanced $G=(V,E)$20-colored, then $G=(V,E)$21 is not closed neighborhood-balanced $G=(V,E)$22-colored (Almeida et al., 18 Oct 2025).

A broader relaxation is given by neighborhood $G=(V,E)$23-balance. In a $G=(V,E)$24-colored graph, a coloring is $G=(V,E)$25-balanced if for every vertex $G=(V,E)$26,

$G=(V,E)$27

for all colors $G=(V,E)$28; it is $G=(V,E)$29-balanced if the same inequality holds in $G=(V,E)$30; and it is $G=(V,E)$31-balanced if each vertex may satisfy either the open or the closed version. The associated invariants are the open, closed, and local $G=(V,E)$32-balance numbers,

$G=(V,E)$33

These satisfy

$G=(V,E)$34

and all three bounds are tight. The same work introduces color degree matrices and color 2-switches, proving that two $G=(V,E)$35-colored graphs have the same color degree matrix if and only if one can be obtained from the other by a sequence of color 2-switches. Because color 2-switches preserve the number of neighbors of each color at every vertex, they preserve $G=(V,E)$36-, $G=(V,E)$37-, and $G=(V,E)$38-balanced colorings (Collins et al., 5 Mar 2026).

For $G=(V,E)$39 and $G=(V,E)$40, the theory refines further into OSB, CSB, SBV, and parity-balanced classes. In this regime, every tree is OSB, and complete multipartite graphs satisfy

$G=(V,E)$41

with each bound sharp. The same paper gives complete characterizations of parity-balanced and CSB caterpillars and a recurrence-based counting formula for CSB-colored caterpillars of fixed spine length (Collins et al., 5 Mar 2026).

Another relaxation is quasi neighborhood balanced coloring, a 2-color notion in which each vertex has red and blue neighbor counts differing by at most one, with at least one vertex attaining difference exactly one. Uniform, positive, and negative variants specify the sign of the imbalance at odd-degree vertices. A negative quasi neighborhood balanced coloring on a graph whose vertices all have odd degree is also a closed neighborhood balanced coloring. Representative class results include: $G=(V,E)$42 is negative quasi neighborhood balanced if and only if $G=(V,E)$43 is even; $G=(V,E)$44 is uniformly quasi neighborhood balanced if and only if at least one of $G=(V,E)$45 is odd; every path admits a quasi neighborhood balanced coloring (Almeida, 30 Dec 2025).

6. Complexity, non-heredity, and open directions

The exact open-neighborhood problem is computationally hard. For every fixed $G=(V,E)$46, the decision problem asking whether a graph admits a neighborhood-balanced $G=(V,E)$47-coloring is NP-complete. The reduction is from $G=(V,E)$48-Equal Sum Subsets and uses the $G=(V,E)$49-house gadget, whose key property is that in any neighborhood-balanced $G=(V,E)$50-coloring all index vertices in the gadget receive the same color. This forces a coloring of the reduction graph to encode an equal-sum partition (Almeida et al., 7 Sep 2025).

The class is also non-hereditary. Every graph occurs as an induced subgraph of a neighborhood-balanced $G=(V,E)$51-colored graph, via a cloning construction that replaces each original vertex $G=(V,E)$52 by $G=(V,E)$53 copies $G=(V,E)$54 and connects all copies of adjacent originals in all $G=(V,E)$55 color combinations. Consequently, the class of neighborhood-balanced $G=(V,E)$56-colorable graphs has no forbidden induced subgraph characterization (Almeida et al., 7 Sep 2025).

The same two phenomena recur in related variants. Closed neighborhood-balanced $G=(V,E)$57-coloring is NP-complete for each fixed $G=(V,E)$58, via a reduction from proper $G=(V,E)$59-coloring using edge cliques and padding gadgets; moreover, every graph is an induced subgraph of some closed neighborhood-balanced $G=(V,E)$60-colored graph, so that class is likewise non-hereditary and admits no forbidden induced subgraph characterization. The status of the closed $G=(V,E)$61 problem is left open (Almeida et al., 18 Oct 2025). Quasi neighborhood balanced coloring is NP-complete as well, and its class is also non-hereditary, with no forbidden induced subgraph characterization (Almeida, 30 Dec 2025).

Several open directions are explicit. For the exact open-neighborhood theory, a central unresolved problem is to characterize regular graphs that admit neighborhood-balanced $G=(V,E)$62-colorings (Almeida et al., 7 Sep 2025). For the closed-neighborhood theory, the corresponding characterization problem for regular graphs, and the complexity of the 2-color case, remain open (Almeida et al., 18 Oct 2025). In the $G=(V,E)$63-balanced framework, proposed extensions include distance-two neighborhoods, alternative imbalance objectives, and edge-color analogues (Collins et al., 5 Mar 2026). Taken together, these questions indicate that neighborhood-balanced $G=(V,E)$64-coloring sits at the intersection of local equitable structure, graph products, modular counting, and computational intractability, with a theory that is already rich in special families but still incomplete in general.