ResearchInteger sequencesGSI-R-0008 · Note · Count

We computed that ℤ27, ℤ29 and ℤ31 have 164,135,223,171,169,785; 18,674,231,142,723,249,003; and 2,427,915,096,775,283,929,115 complete mappings that fix 0, values that would extend OEIS A003111 to a(15).

A complete mapping of a group G is a permutation f with x ↦ x·f(x) also a permutation, normalised if f fixes the identity. A003111 lists normalised counts a(k) of ℤn, n = 2k + 1; A006717, full counts n·a(k). We give both for ℤ27, ℤ29, ℤ31 and full counts for 23 groups of orders 24 to 28 and 32.

Before this work
OEIS A003111 [1] and A006717 [2], read on 15 November 2026, list a(0) to a(12); [1] credits the last (ℤ25) to Yuh-Pyng Shieh, 10 January 2006. Wanless’s survey [3] gives the counts of all groups of order 3 to 23 and of ℤ25.
Evidence
One engine. All counts come from programs written by our own system: 2 agree on a(15), at least 3 on a(13) and a(14), one written without reading the others’ code, after their values were known. All share counting identities [4], [5], properties of the problem and our cloud machines. No review by a person is on record.
What you can check
Counts only. The 3 programs below, each under a second in CPython 3.15, test the definition for n ≤ 11, necessary conditions on the values, and that the 23 files are group tables. They cannot recompute or confirm a count.
Not established
That a(15) is correct: 2 programs of ours agree, and no third has run at n = 31.
Method
Not published.

What was known

Shieh and co-authors computed the counts of the groups of order 3 to 23 [3], [6]; [3] reports that McKay, McLeod and Wanless [7] confirmed all but the cyclic groups of order 21 and above, and gives the cyclic counts to n = 25 as known from Shieh’s thesis and a private communication of 2006. Priority for these values, ℤ25 included, is theirs. Our counts for orders 3 to 23 equal Table 4 of [3] as multisets for each of the 16 orders.

To our knowledge no source lists the counts below (OEIS and web searches of 15 November 2026).

The result

nTermNormalised count a(k) (A003111)All complete mappings n·a(k) (A006717)
27a(13)1641352231711697854431651025621584195
29a(14)18674231142723249003541552703138974221087
31a(15)242791509677528392911575265368000033801802565
The full count equals the number of transversals of the Cayley (addition) table of ℤn [3]: sets of n cells, one in each row and column, with distinct entries.

The 23 groups are the 21 of order 24 to 28 with complete mappings, other than ℤ25, and 2 of the 51 of order 32. By Hall and Paige [8] the other 6 of order 24 to 28, with cyclic nontrivial Sylow 2-subgroups, have none: ℤ3⋊ℤ8, ℤ24, D26, ℤ26, ℤ7⋊ℤ4 and ℤ28.

SmallGroupGroupAll complete mappings
(24,3)SL(2,3)532708548673536
(24,4)ℤ3⋊Q8711496367603712
(24,5)ℤ4×S31424254213029888
(24,6)D24715293909319680
(24,7)ℤ2×(ℤ3⋊ℤ4)1421775848079360
(24,8)(ℤ6×ℤ2)⋊ℤ2714149386518528
(24,9)ℤ12×ℤ24263330069872640
(24,10)ℤ3×D82137306357039104
(24,11)ℤ3×Q82133607509393408
(24,12)S4356469972664320
(24,13)ℤ2×A41067993165463552
(24,14)ℤ22×S31429082153680896
(24,15)ℤ6×ℤ224270246393282560
(25,2)ℤ5241611960482858625
(27,1)ℤ274431651025621584195
(27,2)ℤ9×ℤ34431730326458552091
(27,3)ℤ32⋊ℤ31477670782269469248
(27,4)ℤ9⋊ℤ31477532807935799904
(27,5)ℤ334431953778920809155
(28,3)D286916031483664400384
(28,4)ℤ14×ℤ248224355681195524096
(32,45)ℤ4×ℤ23932009054420727968563200
(32,51)ℤ25935373426152330923868160
Not normalised. SmallGroup is the GAP identifier (order, index); Dn has order n. The ℤ27 row is 27 × a(13).

How to check it

A first program reproduces A003111 and A006717 [1], [2] for n = 3 to 11. The map g(x) = f(−x) turns the condition on x + f(x) into A003111’s condition on g(x) − x and keeps f(0) = 0.

def a(n):  # complete mappings f of Z_n with f(0) = 0: f and x -> x + f(x) are both permutations
    def go(x, used, sums):
        if x == n:
            return 1
        return sum(go(x + 1, used | 1 << y, sums | 1 << (x + y) % n) for y in range(n)
                   if not used >> y & 1 and not sums >> (x + y) % n & 1)
    return go(1, 1, 1)  # f(0) = 0 uses the value 0 and the sum 0
for n in (3, 5, 7, 9, 11):
    print(n, a(n), n * a(n))
3 1 3
5 3 15
7 19 133
9 225 2025
11 3441 37851
Output, 0.1 s: n, the normalised count, the full count.

A second program tests, on the 3 new values and the last 3 listed terms, necessary conditions from Wanless’s line in A003111 [1]: for odd n the count a is odd, divisible by 3 if n > 3 and n ≢ 1 (mod 3), ≡ −2 (mod n) if n is prime, and divisible by n if n is composite. It prints the ratio of the full count to e−1/2 n!2/nn−1 (Eberhard, Manners and Mrazović [9], as listed in A006717 [2]).

from math import exp, lgamma, log
OLD = {21: 275148653115, 23: 19686730313955, 25: 1664382756757625}  # a(10), a(11), a(12) of A003111
NEW = {27: 164135223171169785, 29: 18674231142723249003, 31: 2427915096775283929115}
for n, a in {**OLD, **NEW}.items():
    prime = all(n % p for p in range(2, n))
    tests = [a % 2 == 1,                                 # a is odd
             n % 3 == 1 or a % 3 == 0,                   # 3 | a unless n = 1 (mod 3), for n > 3
             (a + 2) % n == 0 if prime else a % n == 0]  # a = -2 (mod n) for prime n, n | a otherwise
    ratio = n * a / exp(-0.5 + 2 * lgamma(n + 1) - (n - 1) * log(n))  # to e^{-1/2} n!^2 / n^{n-1}
    print(n, n * a, all(tests), a % 8, round(ratio, 4))
21 5778121715415 True 3 1.0154
23 452794797220965 True 3 1.0141
25 41609568918940625 True 1 1.013
27 4431651025621584195 True 1 1.0121
29 541552703138974221087 True 3 1.0112
31 75265368000033801802565 True 3 1.0105
Output, under 0.1 s: n, the full count, whether all 3 tests pass, a mod 8, the ratio to the asymptotic.

A third program checks that the 23 files are group tables: 9 abelian, 14 non-abelian.

import sys
for f in sys.argv[1:]:
    rows = [list(map(int, l.split())) for l in open(f)]
    n, t = rows[0][0], rows[1:]
    assert len(t) == n and all(sorted(r) == list(range(n)) for r in t + [list(c) for c in zip(*t)])  # Latin square
    assert all(t[0][j] == j and t[j][0] == j for j in range(n))  # element 0 is the identity
    assert all(t[t[x][y]][z] == t[x][t[y][z]] for x in range(n) for y in range(n) for z in range(n))  # associative
    print(f, n, "abelian" if all(t[x][y] == t[y][x] for x in range(n) for y in range(n)) else "non-abelian")
Save as tables.py and run python3 tables.py G*.tab in the folder of the files: 23 lines, 0.03 s.

A wrong integer passes the congruences with probability about 1/54, 1/174 and 1/62 for n = 27, 29, 31; the ratio tests leading digits only. The residues a ≡ 1, 3, 3 (mod 8) continue the pattern 1, 1, 3, 3 that [3] observed to n = 25 and calls an open question. A different value from another program would refute it.

What it changes

Our counts extend A003111 and A006717 from n = 25 to 31, and Table 4 of [3] from order 23 to orders 24 to 28 and 32.

Limits

  • a(15) is not confirmed: 2 programs of ours agree, and no third has run at n = 31. A program written without reading their code, or a recount by others, would close this.
  • No value has a proof object or a recorded review by a person; no counter is published, so no count can be recomputed here.
  • The 23 group counts come from one family of programs; for 15 of them (the 14 non-abelian groups and ℤ4×ℤ23) the 2 programs implement the same method. Only ℤ27 has a check by a program written without reading the others’ code.
  • Evans’s books, Cavenagh and Wanless (2010), Shieh’s thesis, Shieh, Hsiang and Hsu (2000), Novaković (2000) and [7] beyond its abstract were not read, and the page of Shieh linked from A003111 did not resolve. Any may list some of these values; reading them would close this.

Files

References

  1. [1]OEIS A003111, Number of complete mappings of the cyclic group Z2n+1, read 15 November 2026: author N. J. A. Sloane; extensions by J. Hsiang, D. F. Hsu and Y. P. Shieh (3 June 2002) and a(12) from Yuh-Pyng Shieh (10 January 2006); formula line of Ian Wanless (30 July 2010), citing Cavenagh and Wanless, McKay, McLeod and Wanless, and Stones and Wanless. Link
  2. [2]OEIS A006717, Number of ways of arranging 2n+1 nonattacking semi-queens on a (2n+1) X (2n+1) toroidal board, read 15 November 2026: formula lines of Sam Spiro (16 April 2019, corrected by Sean Eberhard, 21 July 2023) and Andrew Howroyd (28 September 2020), who gives a(n) = (2n+1)·A003111(n). Link
  3. [3]I. M. Wanless, Transversals in Latin squares, Quasigroups and Related Systems 15 (2007) 169–190, arXiv:0903.5142v1 (30 March 2009): Theorems 6 and 8, Table 4 and the values of the normalised cyclic counts to n = 25. Link
  4. [4]D. Best, I. M. Wanless, Parity of transversals of Latin squares, Algebraic Combinatorics 3 (2020) 539–557, arXiv:1912.11230 (24 December 2019), Section 2. Link
  5. [5]D. G. Glynn, The permanent of a square matrix, European Journal of Combinatorics 31 (2010) 1887–1891. Link
  6. [6]J. Hsiang, D. F. Hsu, Y.-P. Shieh, On the hardness of counting problems of complete mappings, Discrete Mathematics 277 (2004) 87–100 (received 2 May 2002): states that Shieh computed the counts of all groups of order up to 23 and lists the cyclic counts to ℤ23. Link
  7. [7]B. D. McKay, J. C. McLeod, I. M. Wanless, The number of transversals in a Latin square, Designs, Codes and Cryptography 40 (2006) 269–284. Link
  8. [8]M. Hall, L. J. Paige, Complete mappings of finite groups, Pacific Journal of Mathematics 5 (1955) 541–549; the necessary condition on the Sylow 2-subgroup, as stated in Theorem 8 of Wanless’s survey [3]. Link
  9. [9]S. Eberhard, F. Manners, R. Mrazović, Additive triples of bijections, or the toroidal semiqueens problem, arXiv:1510.05987 (v1 20 October 2015, v3 22 March 2016); the asymptotic is as listed in A006717 [2], and we did not read its statement in the paper. Link

Cite asKishan Kumar, Complete mappings of the cyclic groups of order 27, 29 and 31 and of 23 groups of orders 24 to 28 and 32, GSI-R-0008, https://generalised.si/research/complete-mappings-cyclic-groups, 2026.

Published16 November 2026

History
  • Published.

All results