Chapter 5

 Combining results with Latin square literature  and general conclusion

5.1. Tabulating order- dependent numbers and structural                     parameters

We have collected significant parameters for LS’s sets of orders up to 7 in Table 4.2.1.

Order 2 3 4 5 6 7
1. Total LS-number in general format 2 12 576 161280 812851200 61479419904000
2. Total number in standard format 1 2 24 1344 1128960 12198297600
3. Default multiplicity per structure type 32 72 128 200 288 392
4. Average multiplicity per structure type 1 2 5 75 200 300
5. Number of Latin patterns 1 1 2 4 10 27
6. Number of composition families 1 1 3 14 ? ?
7. Number of structure types 1 1 5 18 100 2000
8. Number of structures within a family 1 1 1 or 2 1 or 2 ? ?
9. PG ( par. 3.5 ) 12 192 * 145164** ?
10. TG ( par. 3.5) 7 60 ? ?
11. Number of isotopy classes 1 1 2 2 22 564
12. Number of orthogonal pairs (st. format) - 1 6 216 0 123292800

*first determined by Barink
**determined by Joriki
numbers in bold determined by the author
author’s purely order-of-size guesses in italics

Table 5.1. Numbers of Latin squares and related quantities with increasing order

Ad 3

Default multiplicity would be the multiplicity in the absence of symmetry like we have in order 5 structure types 4 and 11. The maximum (default) multiplicity in standard format of a Latin square structure within an order n set is 8. n². (8. n² . n! in general format). Multiplicities in the lower order domain are far lower  because of symmetry. Symmetry is expected to be (far) less dominant in higher order LS’s so the relative (actual versus default) multiplicity) might rapidly rise towards 1 for higher n.

Ad 4

Set multiplicity divided by the number of types. Very low at orders up to 5 but likely to rise towards default multiplicity at high orders due to low number of symmetrical LS’s

Ad 5

Remarkably slow rise. Increasing by a factor 5 from n 4 to 6 where as the standard format set multiplicity rises by a factor of 50000.

Ad 7

The most intriguing number missing is the number of structure types for n 6. Averaging various extrapolations the author settles for something of the order of 100…..

5.2 General conclusion

We have studied the geometry of Latin squares in order to reach a classification of LS’s in structure types; rather similar to identifying building trends within a class of related crystal structures. Our approach included:

● Generalizing the Latin square construct as such into a cyclic or on-going structure by either putting it on the torus or repeating it wallpaper-wise in a Latin structure; at the same time extending the “same structure” criterion of two LS’s to isomorphism of the Latin periodic structures

● subjecting same-order sets of Latin squares to a rigorous symmetry analysis, including colour symmetry,  in order to establish the (number of) independent structure types:

● proposing the Latin pattern concept: the geometric pattern defined by the positions of sites with identical symbol, adopting Latin patterns as the major constituents of a LS, and setting up a Latin-patterns composition formula for describing a LS, similar to the chemical formula of a compound. Latin patterns: the DNA of Latin squares!

● showing that the LS sets of order up to 5 can be partitioned in a highly limited number of families with members sharing the same Latin pattern composition (“formula”)

● establishing a powerful vehicle in this lower order domain: LS’s with the same composition family share the same- or no more than two different structures-types.

● showing that our division of LS’s in (sub)families(= in structure types) is a subdivsion of the isotopy classification.

The decomposition of a LS in Latin patterns, the very slow rise of Latin pattern numbers with n, the low numbers of independent structure types within same-order sets: 5 and 18 for order 4 and 5, and the subdivision of the isotopy classes in structure type families would seem novel contributions to the knowledge of Latin squares. Our geometric analyses correlate well with often quoted examples of orthogonality. The same holds for obtaining the numbers of LS’s in reduced form (par. 3.3) and for the correlation of short range order and structure type.

The approach for obtaining most results could be termed semi empirical; many of them could well have been obtained faster in a more analytic manner. But then the article would have been fairly incomprehensible for non-combinatorial readers

The approaches offered in this paper are extensible to Latin squares of larger magnitude, given appropriate computational means. Identifying LS’s as combinations of Latin patterns, and placing them in composition families, is a straightforward means to ”format” the immense populations of higher order Latin squares, at the same time eliminating what is not really relevant: multiplication by permutation, different orientation and – boundaries, thus establishing their true geometric diversity.

Latin square presentations in which the site symbols are (coloured) objects (boards, circles, connected in nets and threads or highlighted as “singles”, were shown to be useful in assessing structure types, symmetries, multiplicities. They might also be inspiring for designs of tile fillings (tilings), mosaics, fabrics, quilts (ref. 6) and perhaps for recreational toys and educational games.

An example of a plane filling (tesselation) based on the couple-couple Latin pattern in 4×4 Latin squares is presented underneath. The order 5 LS-structures offer abundant inspiration for further aesthetic endeavours.


Figure 5.2.1. Tile filling, known as snub square tiling (ref. 13), based on the couple-couple (CC) Latin pattern (in black) in 4×4 Latin squares (par. 3.4, fig. 3.4.1) as an example of designs which can be taken from Latin squares and Latin patterns.
Creation of sets of LS’s up to order 6, and all calculations, permutations, symmetry – , orthogonality – and other numeric or symbolic operations on LS structures required throughout this article were programmed and performed on a Hewlett Packard scientific(graphic) calculator (HP48 and later versions). Handy programs – for putting LS’s on standard numeric format, checking for symmetry relationship or orthogonality, can be made available on request.                                                                                                                                                  Some figures were prepared using a version of Processing,  some others with a homemade graphic program based on Forth.

The author is indebted to Marc van der Zalm for creating the font cover header.


1. Cited for instance in Wolfram MathWorld

2. Douglas Dunham, University of Minnesota, Creating Repeating Patterns with Color Symmetry                Web Site: ddunham/

3. Henry D. Shapiro, Generalized Latin squares on the torus, Discrete Mathematics, volume 24, issue 1,  1978, p. 63 -77

4. John J.  Watkins, Across the Board, 2007, Princeton University Press

5. Wikipedia, Wallpapergroup

6. Latin Hexagon,, page presently(2017) not accessible   but try Latin Hexagon, origame live

7. Bill Lombard (2009), try Google Analytic: 4x4.html


9. H. W. Barink, webside: How many structurally different Latin squares order 5 do exist,                  https://latinsquares5x5,

10. Joriki, Mathematics,1381193  page presently (2017) not          accessible.

11. Terry Ritter

12. G.P. Graham & C.E. Roberts, 2006. Enumeration and isomorphic classification of self-orthogonal Latin squares, Journal of Combinatorial Mathematics and Combinatorial Computing, 59, pp. 101–1114.

13. Wikipedia, Snub square Tiling,