The kissing number
is the largest number of congruent nonoverlapping hyperspheres
in
-dimensional Euclidean
space that can touch a congruent central hypersphere.
It is also called the Newton number, contact number, coordination number, or ligancy.
Equivalently, the points of tangency form a spherical
code on
with pairwise angles of at least
. In the standard notation for spherical
codes,
The three-dimensional problem was discussed by Newton and David Gregory in 1694. Newton asserted that 12 spheres could touch the central
sphere, while Gregory considered whether 13 might be possible.
A configuration of 12 is obtained by placing the points of tangency at the polyhedron
vertices of a regular icosahedron, but
enough space remains to make the impossibility of a 13th sphere
far from visually obvious. The first generally accepted proof that was published by Schütte and van der Waerden (1953),
followed by a shorter proof by Leech (1956).
The asymmetric kissing configuration shown above creates a larger hole containing a smaller sphere of radius . This smaller sphere
touches the central sphere and four surrounding spheres.
Since
, it does not provide a 13th unit
sphere.
The exact values currently known are ,
,
,
,
, and
(OEIS A257479).
Musin (2008) proved
using a modification of Delsarte's method. The values for
and 24 are realized by the
root lattice and the Leech lattice, respectively (Odlyzko and Sloane 1979).
For every other
, the exact value remains
unknown.
The following table gives the best general lower bounds and upper
bounds
known for
as of August 2026; equal entries are exact (Cohn). Cohn's continuously maintained
table at https://cohn.mit.edu/kissing-numbers/
lists current lower bounds and upper
bounds in these and higher dimensions, with references for each bound.
| 1 | 2 | 2 | 13 | ||
| 2 | 6 | 6 | 14 | ||
| 3 | 12 | 12 | 15 | ||
| 4 | 24 | 24 | 16 | ||
| 5 | 40 | 44 | 17 | ||
| 6 | 72 | 77 | 18 | ||
| 7 | 126 | 134 | 19 | ||
| 8 | 240 | 240 | 20 | ||
| 9 | 306 | 363 | 21 | ||
| 10 | 510 | 553 | 22 | ||
| 11 | 593 | 868 | 23 | ||
| 12 | 840 | 24 |
As , an AI-generated proof given by OpenAI (2026)
established
where
is little-o notation. This improves the
optimized classical exponent
of Kabatyanskii and Levenshtein (1978).
Ho (2026) constructed 11948 kissing spheres in dimension 19 using a 1280-word subcode of a punctured binary Golay code. On the ambient 4096-word binary code, the graph joining codewords at Hamming distance below 5 consists of four identical parts with no edges between them. In each part, grouping the codewords into cosets of size 64 gives the Clebsch graph, where two cosets are joined when an edge joins their members. A five-element independent vertex set in that graph therefore gives 320 codewords in each part, for 1280 in total.
Kolosov (2026) proved that the 1280-word added-vector code in Ho's 19-dimensional construction is optimal among subsets of the specified 4096-word binary code with minimum Hamming distance at least 5. The upper bound partitions the code into 256 classes whose forbidden-distance graphs are Clebsch graphs, each with independence number 5. This establishes optimality within that code and does not give an upper bound of 11948 for the unrestricted 19-dimensional kissing number. The proof was co-developed with AI and its exact certificate was replayed by VibeMathed (2026), but independent specialist review had not been reported as of Sep. 7, 2026.
Allowing different ball radii leads to the average kissing number, which takes the supremum of the average contact degree over finite ball packings. Bockhaus and Glazyrin (2026) proved that this quantity equals the kissing number in dimensions 8 and 24.
Lin (2026, p. 1) constructed kissing configurations in dimensions 32-35 and 37-39. The construction in dimension 38 adds antipodal pairs of vectors from the Leech lattice. The other constructions use binary constant-weight codes. Lin reports that Claude developed and ran the searches and wrote the exact verification programs.
Kamenetsky (2026) constructed the following lower bounds in dimensions 49 through 61. They are bounds for the unrestricted kissing number, not asserted exact values.
| lower bound for | |
| 49 | 52,430,388 |
| 50 | 52,459,164 |
| 51 | 52,502,324 |
| 52 | 52,588,576 |
| 53 | 52,703,384 |
| 54 | 52,932,310 |
| 55 | 53,316,308 |
| 56 | 54,117,390 |
| 57 | 54,574,792 |
| 58 | 55,960,474 |
| 59 | 56,583,474 |
| 60 | 57,573,374 |
| 61 | 60,057,914 |
Kamenetsky (2026) reports that an AI agent wrote and executed the mathematical search and exact verification code, and that the author checked the code and distributed certificates.