**Introduction to Latin squares**

*1.1. Latin square description and presentation*

Latin squares are normally presented as arrays with *indexed sites (cells), (i, j),* occupied by numbers. These may be replaced by symbols or other entities (values). Throughout this article both numbers 1, 2, 3, 4, (5)… and coloured boards will be used to fill and highlight the sites of LS’s. Symmetry and structure of a LS show up particularly vivid in *colour presentation*. Sites are grounded upon the sites-lattice: a sub-lattice of the Bravais-lattice which is defined by the periodicities of the entire LS structure* (par.1. 2)

Site positions will be expressed as (i, j) starting top-left (0,0) with i , j equal to 0, 1, 2 ,3, (4); intersite distances as 1 (adjacent), V2 (diagonal) , 2, V5 (knight’s move in chess), etc. Edge length (LS-order) is n. Areas are in corresponding units.

*Figure 1.1.1. Various representations of the same Latin square*

* (3) is the corresponding Latin-pa**tterns composition diagram (par. 1.2)*

Fig. 1.1.1 illustrates various ways in which a Latin square with a specific geometric structure can be presented. LS(1) is in* standard numeric format*: first row in sequential order: 1 2 3 4; LS (2) in our *standard colour format* with first row colours red yellow green blue (and purple in order 5 LS’s). The nomenclature-correspondence 1 -> red, 2 -> yellow, etc. (first rows of LS’s (1) and (2)), is maintained throughout both squares. We have to establish a correspondence of numbers and colours to conclude that LS’s (1) and (2) *are equal*.

The equality holds if we produce “another” Latin square, (1gf), in *general format*, by permutation of the site symbols of (1); here according to 1234 → 2314. The correspondence here is a permutation. LS’s are *different* * if such a correspondence does not exist. LS’s (1), (2) and (1gf), although symbolically and numerically different looking , are* equal**.

We will present LS’s in the standard formats* (sf)* wherever possible. Converting to* sf* is a fast way to see whether two LS’s are equal, or not.

A LS will represent n! equal LS’s obtainable by permutation. This will bring the numbers of different LS’s of order 4 and 5: 576 and 161280, down already to 24 and 1344(ref 1).

*Latin patterns*. A meta-description of the structure of a Latin square is shown in Fig. 1.1.1 (3). It is obtained by connecting sites with equal symbol in LS(1), (1gf) or (2). The specificity (number, colour) of site entities can be entirely disregarded in (3); it is replaced by the coherence of the equally occupied sites. Each network defined in this manner – already termed a* Latin pattern (Lp)* in the summary – is an intermediate constituent of a LS-square. A *Latin patterns diagram*: Fig. 1.1.1 (3), the superposition of its constituting Lp’s, is now representing the LS. It is independent of (= the same for) any specific representation or nomenclature of the LS such as ( LS(1), (1gf) and (2).

A *Lp composition formula* can also be formed for each LS, by naming and counting the constituting Lp’s per unit cell (Chapter III, formula 3.1.1), somewhat similar to assigning a chemical formula to a compound. Lp-diagram and Lp-formula will prove highly useful when classifying LS’s in terms of symmetry and structure (par. 3.1).

Latin patterns often have a marked appearance (see Figs. 3.1.1 and 3.6.1 , reminiscent of structural components in crystals: layers, macromolecules, fibres, networks etc.

* Terminology which may be specific for this article is explained in the glossary (Appendix I)

*1.2. Shape and crystal symmetries of Latin squares*

You may wonder what can be said about the shape of a (Latin) square. In fact, *all Latin squares*, although nearly always drawn as squares, *lack the (conventional) point group symmetry properties for square symmetry:* a fourfold rotation axis, or a diagonal a horizontal and a vertical mirror line. The first and both last requirements contradict the Latin square condition for even and odd sized LS’s. Instead, site/value arrangements within LS’s, including (2) of Fig. 1.1.1, conform with a rectangular or oblique axis system on the basis (or absence) of internal symmetry.

The oblique axes system of many LS’s would readily show up if we would materialize them as a close-packed structure of discs and replace different olours by unequal disc-diameters.

Of course one is free to write and draw LS’s as is suitable for well-ordered presentation and easy establishing (in)equalities between them, as illustrated in par. 1.4. However, when attributing planar symmetry *systems,* *classes* and *planar* *groups* to Latin squares (Chapter 2 and 3) we will respect their true crystallographic symmetry originating from their spacial arrangements of values.

###### Wherever necessary we may adopt a more general description of a LS: a Latin “square” taken as a geometric object, can be a parallelopid, a rhombus, a hexagon (par 1.7), a rectangle or a square) depending on the crystallographic symmetry of its internal value/site arrangement

*1.3. Conventional and colour symmetry*

Let us consider the symmetry of the coloured LS (2) once more. The only conventional symmetry element directly visible is an *inversion centre* (a twofold axis perpendicular to the paper) moving all sites towards sites with identical colour. *Mirror* *lines* through the centre are absent: horizontal and vertical reflections switch the colour of all sites; reflection across the diagonal lines turn red into red and blue into blue, but yellow and green switch colour.

We note though that the diagonal reflection induces a switch of* all* yellow and green sites. Similarly, the central horizontal/vertical lines, switch all red- blue and yellow-green pairs.

In other words: these reflections perform *permutations* of colours of the LS. So, in addition to its conventional symmetry elements, a LS may carry *colour mirror lines*; more generally: elements of colour symmetry. We accept them as symmetry elements because we consider a LS invariant under switches of colours.

In different words: An object is said to display colour symmetry if application of one or more of the isometric movements of the object followed by a permutation of its values – not being the unity permutation- can fully restore the original status. Or: if a full colour switch or a step in the colour sequence occurs in conjunction with the isometric movements of a square.

If an observer will test a possible element of crystal symmetry (mirror line, rotation axis etc.) in or on an object he must confirm first that the original and the image obtained by an *isometric* transformation (mapping) coincide fully as drawings* (ref.2)* , that is that their Latin patterns diagrams coincide.

If he next does note distinct values, say colours, he will meet one of three possibilities:

– each colour site will be mapped in the same colour of the original on the corresponding site: we have *conventional symmetry*,

– all sites with common colour have adopted a single colour not being the original. We have a permutation of colours; we have *colour symmetry*, in this case a colour shift. If we proceed with the image in its new colour(s) we may return to the original colour or find still another colour etc. : we have a colour sequence. The sequence number will be the LS order n or a factor of n.

– the colours of the original and the mapping are not permutation related. The operation under consideration is not a symmetry operation.

Accepting colour symmetry elements in LS- structures means that we may raise the crystal system to that of its Latin patterns diagram The actual conventional and colour symmetries are given in par 2.4 and 2.5

*1.4. Point symmetry relationships and structure types*

We will use structure distinction or -correspondence of LS’s as a vehicle to identify *structure types*, highly similar to what is practice in crystallographic classification. As an example we refer to two *different* LS’s, (4) and (5), Fig. 1.4.1. left, and show that (5) is obtained by rotating (4) clockwise over 90 degrees (4rot) and next converting to standard format (sf).

LS(4) and (5), although *different* as LS’s (par.1.1), are *(point) symmetry related*. They* share the same structure type; *they are* dependent*. All point symmetry relationships in any listing of same-order LS’s can be established – and more structure types obtained – by applying the symmetry qA movements of the *dihedral group for a square*.

*Figure. 1.4.1. (5) is obtained by rotating(4) over 90 degrees and bringing it back to sf; **(6) is obtained by cyclic shifting of (1) and converting to sf. Latin squares**(4) (5) (1) (6) are LS02, 03, 09, 16 respectively of the full order 4 set (Fig. 2.2.1)*

*1.5. Cyclic symmetry operations*

In LS literature, operations on LS’s include permutations of rows and columns. Such operations split a same-order set in so-called *isotopy classes* depending on whether LS’s are mutually row-row (etc.) permutation-related or not. We will show (Chapters 2, 3) that our distinction in terms of structure types (above) is a subdivision of the isotopy classification. It follows that one may well “hop” from one structure type to the other with a row/row (etc.) switch; in other words: a LS is generally *not* geometrically invariant under row/row etc. permutations.

A special case however – paramount for our entire approach – is *cyclic permutation* or – *shifting,* csh) of rows or columns This can be visualized by bending a LS, for instance (1) of Figs. 1.1.1 and 1.1.2, into vertical cylindrical shape by connecting the end columns, reopen the cylinder elsewhere, obtain (1csh) and bring the result back in sf (6) . The resulting LS (6) is *(translational) symmetry related* to the original (1).. So, again, we have *dependent, symmetry related*, structure types. Translations can be repeated. The unit translation equals the period of the sites lattice.

*1.6. Torus and Latin structure*

Cyclic shifting can be generalized further by bending the cylinder into a *torus*, (Fig. 1.6.1). A torus allows for shifting a LS of order n in two directions: along and around the torus, thus exposing n²-1 translation-related representations of the original. Shapiro (*1978) * seems to have been the first to propose the torus as a natural object to *extend the notion of a Latin square* *(ref. 3).* A Latin square in toroidal representation has origin nor edges.

If we map the sites-lattice on the torus we find that the rows and columns are circles now, around and along the torus; the diagonals follow ellipses. There are two finite groups to consider: the translations group with n²-1 elements, and the point symmetry group which is the order 8 version of the dihedral group. Translation movements following the sites lattice can be numerous (n²-1) whereas the number of point symmetry operations is 8

We note (Fig. 1.6.1) that the point group – and translations group – movements have similarity on the torus; both groups are finite; the translations move sites (“cells”) and values (colours) towards other sites; the point symmetry movements do the same but may leave certain site-values, or all values along certain lines, unchanged. This is why we consider the movements connecting translational-symmetry related LS’s: (1) and (6) of Fig. 1.4.1), similar to those connecting point symmetry related LS’s (4) and (5)..

*Figure 1.6.1. Latin square LS01 (Table 2.1.1), in toroidal representation.*

Since a Latin square on a torus is somewhat difficult to draw and overlook we may as well extend the structure of a Latin square *wallpaper-wise*, terming the resulting 2D-periodic array a *Latin-periodic structure,* or briefly:* Latin structure* (Fig. 1.6.2). An order 5 example is Fig. 3.1.2 . Latin patterns and Lp diagrams (par.1.1) can be likewise extended.

Note that extension of a LS as such is necessary to fully describe the symmetry (conventional and/or colour -) of the structure. Looking once more at (2) we might conclude that conventional symmetry is limited to the inversion centre at the centre. However by adding a row and a column in all four directions we note a system of conventional mirror lines, diagonally following yellow- green-green-yellow sites.

*1.7. Latin squares in an omniverse of related patterns*

Once we generalize a Latin square* as such, nxn*, versus an *ongoing unbounded Latin structure* we can compare the Latin square world with other similarly related families of patterns, 3D or 2D, such as crystal structures and wallpaper designs in particular, quilts , tessellations, tilings, settings on (chess) boards *(Ref.4)* and various architectic, decorative and artistic *(Escher!, Ref.2)* creations. *Ref (5 )* on wallpaper designs is an excellent introduction to planar symmetry groups in general

Subsequently, we can follow the symmetry-hierarchy of crystalline -and wallpaper structures (2D) with 7 (4) crystal systems (monoclinic, triclinic etc) and 230 space groups (17 planar groups), and apply it to any Latin square/structure..

The overall variety of symmetry systems and groups of Latin type squares is hampered by the Latin square conditions: lattice arrangement of sites; all values once and only once per period along the two main axes. We saw that square symmetry is impossible unless we accept colour symmetry and so is trigonal unless we accept a *Latin hexagon *for order n=3 . *Ref. (6)* invites readers to design a Latin hexagon for n 6. This illustrates once more the order-dependency of the symmetry-variety . The possible crystal systems are reduced to 2 or 3 with subsequent reduction of the spatial groups to 17 (planar groups). However, colour symmetry may raise symmetry and diversity. Also, keep in mind that many different structures may carry the same planar group symmetry (Chapter 3).

*Figure 1.7. 1. Order 4 Latin square (LS01, Table 2.1.1), extended towards an ongoing Latin periodic structure, or briefly Latin structure, Latin structures may (or may not!) be attractive suggestions for “wallpaper” design, garage door embellishment, or other mosaic structure.*

*1.8. Translation movements and space (planar) group symmetry*

The overall translational periodicity in a (Latin) structures is pinpointed by its Bravais-lattice. Some patterns repeat themselves completely *within* their Latin square boundaries as can best be seen in an ongoing structure, Fig. 1.6.2 a centred LS repetition of the structure halfway the body diagonals. This is why the number of Bravais lattices exceeds the crystal systems number : 14 (5)versus 7 (3) ;2D in parenthesis.

Glide lines, combining reflection and translation halfway a full translation period, and all other symmetry operations of a LS under consideration, constitute its *planar group* (Chapter 2 and 3).

We summarize the various symmetry- and structure aspects (to be) raised in the present and following chapters:

1 : the varous descriptions and representations of LS’s including the Latin patterns diagram;

2 ; the *point group symmetry movements* of any square (the dihedral group of order 8, 4mm);

3 ; the *conventional* *group* versus the *colour group* of a LS;

4a; extending the *LS notion* by putting a LS *unbounded* on a *torus* and collect the translation movements in a finite group;

4b; alternatively: having the LS 2D-repeated in a wallpaper-wise *Latin- periodic structure*;

5a ; finding the point group- and/or translation *symmetry relationships* between LS’s, and:

5b; partitioning LS’s of same order in *structure types* according to common symmetry;

6 ; determining the IUC designation (point and planar group notation) of Latin squares and – structures

.