The -hypercube
graph, also called the
-cube graph and commonly denoted
or
, is the graph, for a nonnegative integer
, whose vertices are
the
binary strings
where
or 1 and two vertices
are adjacent iff the
symbols differ in exactly one coordinate.
For , each vertex
of
has vertex degree
. The handshaking lemma
therefore gives
edges. Including the singleton
graph
, the edge counts are
0, 1, 4, 12, 32, 80, ... (OEIS A001787).
The graph of the -hypercube is given by the graph
Cartesian product of path graphs
. The hypercube graph
is isomorphic
to the Hasse diagram for the Boolean
algebra of subsets of an
-element set.
is also isomorphic
to the simplex graph of the complete
graph
(Alikhani and Ghanbari 2024).
The locally twisted cube graph (
) modifies the recursive
perfect matching between two copies of the preceding
graph. It retains the
vertices and
edges of
, while reducing the graph
diameter from
to approximately
(Yang et al. 2005). Unlike hypercube graphs, these graphs
are not bipartite for
.
The above figures show orthographic projections of some small -hypercube
graphs using the first two of each vertex's
set of
coordinates. Note that
above is a projection of the usual cube
looking along a space diagonal so that the top
and bottom vertices coincide, and hence only seven
of the cube's eight vertices are visible. In addition,
three of the central edges connect to the upper vertex, while the other three connect to the lower vertex.
Hypercube graphs may be computed in the Wolfram Language using the command HypercubeGraph[n], and precomputed properties of hypercube graphs are implemented in the Wolfram Language as GraphData[{"Hypercube", n}].
Special cases are summarized in the following table.
For , hypercube graphs are
Hamiltonian, and any Hamiltonian
cycle of a labeled hypercube graph defines a cyclic
binary Gray code (Skiena 1990, p. 149; Mütze
2024). The numbers of Hamiltonian cycles in
hypercube graphs were considered by Dixon and Goodman (1975).
Hypercube graphs are also graceful (Maheo 1980, Kotzig 1981, Gallian 2026), antipodal, and (most likely) dominating unique.
The numbers of (directed) Hamiltonian paths whose endpoints are nonadjacent on an -hypercube graph for
, 2, ... are 0, 0, 48, 48384, 129480729600, ... (OEIS
A006070; extending the result of Gardner 1986,
pp. 23-24). The numbers of (directed) Hamiltonian
cycles for
, 3, ... are 2, 12, 2688, 1813091520,
... (Harary et al. 1988; OEIS A003042).
For
, the graph
has no Hamiltonian
cycle under the convention that a graph cycle
has at least three vertices.
Closed formulas for the numbers of cycles of length
in
are given by
for
odd and
|
(1)
| |||
|
(2)
| |||
|
(3)
| |||
|
(4)
|
(E. Weisstein, Nov. 16, 2014 and Apr. 19, 2023).
Hypercube graphs are distance-transitive, and therefore also distance-regular.
Alspach (2010) reports that Ringel showed in 1954 that the hypercube graphs admit Hamilton
decompositions whenever
is a power of 2. Alspach et al.
(1990) showed that every
for
admits a Hamilton
decomposition.
For , the hypercube graphs
are also unit-distance (Gerbracht 2008), as
illustrated above for the first few hypercube graphs. This
follows by induction starting with a single unit-length edge.
Given a unit-distance embedding of
, choose a unit translation vector that is not the
difference of any two existing vertices. There are
only finitely many forbidden vectors, so such a choice exists. The original embedding
and its translate have disjoint vertex sets. Joining
corresponding vertices by unit-length edges
constructs
.
Determining the domination number is intrinsically difficult (Azarija et al. 2017)
and exact values are known for
(Östergård and Blass 2001, Bertolo et
al. 2004) and the infinite families
and
for
(Azarija et al. 2017, Theorem 3). Azarija et
al. (2017) showed that domination and total
domination numbers of the hypercube graph are related
by
.
is planar
for
, so has graph
crossing number
for
. Eggleton and Guy (1970) claimed to have discovered
an upper bound for the graph
crossing number of
for
, where
|
(5)
| |||
|
(6)
|
The first few values for , 4, ... are 0, 8, 56, 352,
1760, 8192, 35712, ... (OEIS A307813).
An error was subsequently found, but Erdős and Guy (1973) then conjectured that not only was the original bound correct (though not yet proved), but that (Clancy et al. 2020, pp. 258-259).
While it is known that
, exact values for larger
are not known (Clancy et al. 2020,
p. 259). However, upper bounds are directly computable
using QuickCross (Haythorpe) which correspond to the Eggleton and Guy values for
(E. Weisstein, Apr. 30,
2019). In addition, the Erdős and Guy (1973) conjecture has now been refuted
since it is known that
(Clancy et al. 2020, p. 259).
The coil-in-the-box problem asks for the maximum length of an induced cycle in . The induced condition excludes edges
between nonconsecutive vertices of the cycle.
The geodesic transversal number is the minimum cardinality of a vertex set meeting every inclusion-maximal graph geodesic. Chen et al. (2026) proved
|
(7)
|
for , where
and
denote the ceiling function
and floor function.