 1. Introduction
 2. An overview of crystallographic studies of variants
 3. Crystals oriented in space, internal symmetries
 4. Orientation relationships and external symmetries
 5. Algebraic definition of the orientational variants
 6. Definition of the operators
 7. Groupoid of orientational variants
 8. Calculation of the possible parent crystals
 9. Computation and application to reconstructive transitions
 10. Applications and perspectives
 11. Conclusions
 References
 1. Introduction
 2. An overview of crystallographic studies of variants
 3. Crystals oriented in space, internal symmetries
 4. Orientation relationships and external symmetries
 5. Algebraic definition of the orientational variants
 6. Definition of the operators
 7. Groupoid of orientational variants
 8. Calculation of the possible parent crystals
 9. Computation and application to reconstructive transitions
 10. Applications and perspectives
 11. Conclusions
 References
research papers
of orientational variants
^{a}CEAGrenoble DRT/LITEN, 17 rue des Martyrs, 38054 Grenoble, France
^{*}Correspondence email: cyril.cayron@cea.fr
Daughter crystals in orientation relationship with a parent crystal are called variants. They can be created by a structural
(Landau or reconstructive), by or by precipitation. Internal and external classes of transformations defined from the point groups of the parent and daughter phases and from a transformation matrix allow the orientations of the distinct variants to be determined. These are algebraically identified with left cosets and their number is given by the Lagrange formula. A simple equation links the numbers of variants of the direct and inverse transitions. The equivalence classes on the transformations between variants are isomorphic to the double cosets (operators) and their number is given by the Burnside formula. The orientational variants and the operators constitute a whose composition table acts as a crystallographic signature of the transition. A general method that determines if two daughter variants can be inherited from more than one parent crystal is also described. A computer program has been written to calculate all these properties for any some results are given for Burgers transitions and for martensitic transitions in steels. The complexity, irreversibility and of fractal systems constituted by orientational variants generated by thermal cycling are briefly discussed.Keywords: coset; groupoid; orientation; phase transition; symmetry; variant.
1. Introduction
Phasetransition materials are constituted by grains of the same phase (for example β) that, when some thermodynamical conditions are changed, are transformed into many crystals of the same phase (for example α) with an orientation relationship (OR). These daughter crystals are called variants. The physical and mechanical properties of phasetransition materials highly depend on the microstructure formed by these variants. For example: (i) the martensite laths obtained by quenching or rapid cooling increase significantly the yield strength of steels; (ii) the twins that appear during deformation or recrystallization allow a stress relaxation in copper; or (iii) the nanoprecipitates that nucleate during ageing treatments give a structural hardening to aluminium alloys. These cases and many more (ferroelectric, ferroelastic and magnetic domains) have been treated in different fields of materials science with different approaches.
The aim of the present study is to show that a unique crystallographic description of these different situations can be given by using a unifying algebraic structure called a ). As for groups, they have a direct analogy in geometry. They can be used in many scientific domains because, as clearly pointed out by Stewart (2004), they are `more flexible and often more appropriate than the betterknown groups'. They are actually the ideal tool for describing symmetries that apply only to parts of systems containing repeated units, such as, in our case, systems constituted by variants.
Groupoids were first introduced to determine the quadratic forms of a given norm (Brandt, 1926Since this study has been motivated by the understanding and the exploitation of some experimental diffraction results by electron backscattered diffraction (EBSD) and orientations of the crystals are treated (and not their positions nor the atomic structures of the interfaces). From now on, the word `symmetry' will refer only to orientational symmetries (and the associated point group) and the word `variants' will refer to orientational variants. Firstly, we give a brief overview of the different crystallographic studies about variants. These studies use similar algebraic tools but most of them do not introduce explicitly the orientation relationship between the parent and the daughter crystals, and the variants are not mathematically identified. Secondly, we define the orientational variants by introducing the notions of internal and external equivalence classes on transformation matrices and by using some general algebraic tools of group theory. Direct and inverse transitions will be treated. Operators will be introduced as classes of transformations between variants. Finally, we show that the set of variants and the set of transformations that link them do not form a group but a A method that gives the possible parent crystals from the knowledge of some daughter crystals is also described. A simple geometrical case (Fig. 1) will be used as an example during the study. However, for complex reconstructive transitions, as the geometrical considerations are limited (they require a great capacity for visualizing complicated threedimensional objects), a computer program has been written to calculate all the variants, the transformations between them, the different possibilities of parent crystals and the composition table of the Its application to Burgers transitions and martensitic transitions in steels will be presented. We will briefly explain on these examples how to use the composition table of groupoids to reconstruct parent crystals from EBSD data [for more details on this application, see Cayron et al. (2006)]. The problem of the generation of highorder variants during thermal cycling and the related fundamental problems will be discussed in the last section.
(TEM), only theWe have decided to sum up the algebraic notions used in this paper in Appendices A and B. The reader is invited to go over them to at least be familiar with the mathematical notations. Some are described by Janovec et al. (2003). The reader can also refer to Milne (2003) or to Kargapolov & Merzliakov (1985) for an introduction to group theory.
2. An overview of crystallographic studies of variants
2.1. Landau transitions
Landau has studied the thermodynamic of phase transitions when a group–subgroup relation exists between the parent and daughter symmetries (Landau, 1937a,b; Tolédano & Tolédano, 1987). The of the transition depends on the nature of the transition (it is an atomic position or a degree of occupancy etc.) and the theory relies on the assumption that the form of the free energy should continue to respect the symmetries of the higher symmetric phase (the phase stable at high temperature) even at temperatures lower than the transition temperature. Although restrictive, this approach was fruitful: it has permitted us to show that crystallography has a direct effect on the thermodynamics of a In this special case of group–subgroup relationships between the parent and daughter crystals, geometrical objects called orientational domains or variants have been introduced (but not explicitly mathematically defined) from the algebraic decomposition of the symmetry group of the parent phase G into left cosets of the of one of the daughter crystals (denoted H or F, depending on the authors). The algebraic developments were done for ferroelectric domains (Janovec, 1972, 1976) and for variants generated by Landau order–disorder transitions (Van Tendeloo & Amelinckx, 1974).
Janovec introduced the term `domain states' to refer to any physical property attached to the daughter phase crystals such as the polarization of ferroelectric domains. The first domain state is denoted S_{1}. All the symmetry elements of G that leave S_{1} invariant form a group called the of S_{1}, denoted F_{1} (with F for ferroic). It was shown that the number of distinct domain states is given by the decomposition of G into left cosets of F_{1}:
A deep algebraic study of orientational and translational variants is also given in Dirl et al. (1997) by considering the space groups of the two phases.
The algebraic base used by Van Tendeloo is similar to that of Janovec. A variant is an `object' V_{i} invariant by a of G denoted H_{i}, i.e. H_{i}V_{i} = V_{i}. The other variants V_{j} are defined from V_{i} by an element g of G such that gV_{i} = V_{j} and g H_{i}. The point or of the parent phase is then decomposed as left (or right) cosets of H_{i} such as in equation (1). Van Tendeloo uses the term `variant' to refer to a crystal oriented with reference to the parent crystal but no mathematical definition was given. Since the orientation can be considered as a crystal property, Janovec's description includes Van Tendeloo's. In both cases, the number of variants results from the Lagrange theorem: N = G/H, where G and H are the order of the groups.
2.2. Reconstructive transitions
In most minerals and alloys (Fe, Ti, Zr etc.), phase transitions do not respect a group–subgroup relationship. A new thermodynamic theory of developed by Tolédano & Dmitriev (1996) resolves this problem by generalizing the Landau approach. However, the general crystallographic and algebraic developments of these transitions are not as advanced as those presented before for the Landau transitions. The study of `nondisruptive' transitions^{1} introduced by Guymont (1981) tries to reduce the gap: the special case of phase transitions that produce two variants is extensively described but the algebraic developments for the general case are more limited.
Most of the studies on the reconstructive transitions have actually been performed in metallurgy for many different aims: (i) to determine by EBSD the exact orientation relationship in steels (Morito et al., 2003; Kitahara et al., 2005); (ii) to understand the texture evolution during different heat treatments in Ti alloys (Gey & Humbert, 2003) or in brass (Standford & Bate, 2005); or (iii) to correlate the fracture mechanisms of steels to their microstructures (Gourgues et al., 2000). Indeed, after the transition, the parent grains are generally hardly visible. Therefore, a key point is the reconstruction of the parent grains from the data obtained by EBSD on the daughter grains. The most advanced theoretical studies on this subject have been proposed by Humbert et al: a method to calculate the number of orientational variants for Burgers transitions is given by Humbert et al. (1992) and a method to determine the orientation of the βparent grains is proposed by Humbert et al. (1995) and Humbert & Gey (2002). However, these studies lack generality. The algebraic tools used for the Landau transitions (for example the decomposition and the Lagrange theorem) are not explicitly recognized. Moreover, some results are incomplete (see §9.1).
2.3. Precipitation
Precipitates in an orientation relationship inside a matrix have been mainly studied to simulate the complex diffraction patterns obtained by TEM (Akbay et al., 1994; Cayron, 2000). General theoretical concepts have been given by Cahn & Kalonji (1981): the orientation relationship between two crystals of phase I (parent) and phase II (variant) is explicitly introduced by the linear transformation (R/t), where R is the rotation that orientates crystal II with reference to crystal I and t is the translation that gives the position of the lattice of crystal II with reference to the lattice of crystal I. The set of symmetry operations of the of crystal II, denoted , is written in a reference coordinate system associated with crystal I:
This equation leads directly to the set of common symmetry operations: H = ∩ G^{II}, where H is a of called the `intersection group'. It also determines the precipitate morphology. With arguments similar to those given for the Landau transitions, the variants of crystals II in matrix I appear from the left decomposition of H into , and their number is given, here again, by the Lagrange formula . The concepts introduced by Cahn & Kalonji (1981) go far beyond the morphologies. They have led to a general approach of grain boundaries (Gratias & Portier, 1982; Kalonji, 1985) that unifies and clarifies the notions of coincidence site lattice (CSL), Olattice and displacement shift lattice, previously introduced by Bollmann (1970, 1982). However, the orientation relationships between the variants are not studied and the algebraic tools (orbits, stabilizers) are not used as is done for the Landau transitions.
2.4. Twinning
) and nowadays is often treated as a (by deformation or recrystallization). Thus, one can be tempted to use the algebraic notions reported in §2.1, but one must be very cautious because all the concepts developed for the Landau transitions cannot be directly applied for In Hahn & Klapper (2003), K, the socalled `composite symmetry group' is introduced from H, the of an untwinned crystal, and k, a by a decomposition similar to equation (1). K is believed to be a of H (i.e. H is a of K, as is the case for the Landau transitions); however, in general, the composite `crystal', strictly speaking, is not a crystal (no can be attributed) and the operations between the `partners' (i.e. the variants that constitute this composite structure) are only partial operations. Thus, K is in general not a group. A rigorous approach to by applying some concepts of reconstructive transitions (in terms of the nondisruption condition) and by using the intersection group presented in equation (2) is given by Wadhawan (1997).
has been studied for more than a hundred years (Friedel, 1904We present now our own approach on the subject. It includes and unifies all the notions presented before (but of course not all the particular developments of each case). Comparisons with past studies are discussed in the Remarks at the end of each section. The important aim of the present paper is to show that the algebraic structure constituted by the variants (created by precipitation or twinning) is not a group but a groupoid.
3. Crystals oriented in space, internal symmetries
The term `phase' usually designates a class of materials with particular physical or chemical properties. In crystallography, for solid materials, it is restricted to particular types of atomic bonding and, if the materials are crystalline, the term also refers to the atomic arrangements, i.e. the atomic positional symmetry (spacegroup type) and the metric (metric tensor). The term `crystal' designates a solid constituted by a crystalline phase and orientated in space. In this paper, phases will be denoted by a Greek character (β) and crystals will also be denoted by a Greek character (β) with an index when required (β_{i}). From now on, we consider a crystal only by its orientation (and not by its position) in a fixed reference coordinate system. A crystal is then defined by (i) its internal orientational symmetries (one of the 32 point groups) and (ii) its orientation in space. We point out that a crystal is more than a simple oriented lattice (the would be reduced to the group). The orientation of a crystal is usually given by three Euler angles; see for example (Kocks et al., 1998) for the numerous conventions of Euler angles. However, many different triplets of Euler angles may express the same orientation because of the internal symmetries. In the present study, we choose another way to define the orientation of a crystal. A crystal β will be defined by the set of its equivalent bases. Two crystals are `distinct' if their sets of equivalent bases are disjoint and `in coincidence' (or `equal') if their sets are equal. With this definition, a crystal appears as an equivalence class on the set of bases (the equivalence classes are defined in Appendix A). If a unique reference coordinate system is used, the geometrical notion of `basis' can be transformed into a useful algebraic tool, the transformation matrix.
3.1. Transformation matrices
Let us consider two bases B_{1} and B_{2}. The basis B_{2} is formed by three noncollinear vectors trivially expressed by
The lower index (with /) represents the basis in which the vector is written and the upper index represents the basis constituted by the vector. These vectors can also be expressed in B_{1} and they become:
We denote [B_{1}▹B_{2}] as the transformation matrix between the two bases (B_{1},B_{2}). It is given by
Since [B_{1}▹B_{2}][B_{2}▹B_{1}] = E (the 3 × 3 identity matrix), it follows that [B_{1}▹B_{2}]^{−1} = [B_{2}▹B_{1}]. Moreover, for any intermediate basis B_{0}: [B_{1}▹B_{2}] = [B_{1}▹B_{0}][B_{0}▹B_{2}]. If f is a linear function expressed in a basis B_{1} by the matrix F, f is expressed in a basis B_{2} by the matrix [B_{1}▹B_{2}]^{−1}F[B_{1}▹B_{2}]. One must be cautious when interpreting transformation matrices in terms of linear functions.^{2}
3.2. Orientational symmetries expressed as transformation matrices, composition rule
We consider a crystal β with the set of its equivalent bases denoted B^{β} = . Let us arbitrarily choose one of the possible bases = as a fixed reference coordinate system. The bases of B^{β} are given by the transformation matrices ▹ = , where is an orientational symmetry of the crystal β. We denote as G^{β} = the set of all the symmetry operations^{3}, i.e. the of the crystal β. If another basis of the crystal β is chosen as a new reference coordinate system, for example ∈ B^{β} instead of , then becomes in this system = [▹]^{−1}[▹] = , the conjugate of by . Since a group acts on itself by conjugation (G^{β})′ = G^{β}, we conclude that G^{β} does not depend on the choice of the reference basis if this one belongs to B^{β}. In this way, the crystal is `selfreferred' and G^{β} constitutes its internal symmetries. There is no distinction between the of a phase β and the of a β crystal if this one is selfreferred.
In order to reduce the length of equations that will be developed in the following, we note:
The fact that the matrices are implicitly expressed in the basis appears in = E and = . One may check that the matrices can be composed according to the following rule:
The composition (6) has a mathematical meaning: the indexes i, k and l follow the composition rule of the trivial pair two pairs (i, k) and (k′, l) can be composed if and only if k = k′ and then (i, k)(k′, l) = (i, l) as developed in §7.1. We compare our notation with the one chosen by Janovec.^{4}
3.3. Internal classes of transformation matrices
Now, let us consider a crystal α oriented in space, and B_{0} an arbitrary fixed reference coordinate system. One can orientate this crystal with the help of any transformation matrix from B_{0} to one of the crystal bases. In other words, the crystal α is oriented by any matrix of the set {[B_{0}▹], ∈ }. Each orientation matrix of this set can be decomposed according to: [B_{0}▹] = [B_{0}▹][▹] = T, with T = [B_{0}▹]. Hence, the set of all the equivalent transformation (or orientation) matrices of a crystal α becomes
We can then introduce an `internal' equivalence relation, denoted (with I for internal), on the matrices that define (the OR of) the same crystal α: two transformation matrices T_{k} and T_{l} define the same crystal α if and only if ∃(, ) ∈ (G^{α})^{2} and ∃ T an invertible matrix such that T_{k} = T and T_{l} = T. This condition can be written without explicit reference to T (or to the fixed reference coordinate system B_{0}) by noticing that T_{k}^{1} T_{l} = = . It becomes
TG^{α}, the set of matrices of the orientation of the crystal α, does not form a group. However, since the `left class of equivalence' found in group theory (Appendix A) is similar to condition (9), we will also use the same term for (9). One may also notice that G^{α} acts trivially transitively at the right of the set TG^{α} (Appendix B).
4. Orientation relationships and external symmetries
4.1. Matrix expression of an orientation relationship
We denote by β a crystal with a fixed orientation and by α_{i} the crystals of phase α in equivalent orientation relationships (OR) with β. The crystals α_{i} are generated by a β → α structural by or by precipitation. We denote as G^{α} = the of the phase α. At each crystal α_{i} can be associated the set of equivalent bases = such that ▹ = G^{α}.
In order to compare a crystal α_{i} to another crystal α_{j}, a unique reference coordinate system must be chosen, for example . The OR between the variant α_{1} and the parent crystal β is given by the transformation matrix ▹ which will be denoted T_{>}. The subscript > represents the direction of the transition (see §5.2).
T_{>} can be explicitly calculated from the usual form of an OR given by two couples of parallel and similarly oriented vectors expressed in their own  and  with and not parallel (see example in Table 1). A third couple of vectors  can then be created with = and = (the calculations of the cross product are not reported here, they imply expression of the vectors in an orthogonal coordinate system with the help of the structure tensors). Now, a new basis B_{c} common to both crystals α_{1} and β is introduced and defined by
and
Then, T_{>} is explicitly deduced from the OR by
We point out that the directions of the `parallel and similarly oriented' vectors are important if the crystals are not centrosymmetric.

Since equation (10) gives the OR of the first variant, we write
4.2. External classes of transformation matrices
The parent crystal β has its own internal symmetries, which, in general, do not correspond to the internal symmetries of the crystals α_{i}. Using the formalism (Weinstein, 1996), we will say that the symmetries of the crystal β act as `external' symmetries for the crystals α_{i}. We now choose the following indexing rule for the crystals α_{i}:
Then, the matrix that defines the orientation of a crystal α_{i} is given by ▹ = ▹▹ = T_{>}. One may then write
The set of the transformation matrices of all the crystals α_{i} inherited from the same parent crystal β is
For example, it means that for an m3m parent crystal the number of daughter crystals is at maximum equal to G^{β} = 48. However, some of the daughter crystals are in coincidence and we will see in the next section that their number is actually a divisor of G^{β}.
We can then introduce an `external' equivalence relation, denoted (with E for external), on the matrices that define (the OR of) the crystals inherited from the same parent crystal: two transformation matrices T_{k} and T_{l} define two daughter crystals α_{k} and α_{l} if and only if ∃ ∈ (G^{β})^{2} and ∃ T_{>} an invertible matrix such that T_{k} = T_{>} and T_{l} = T_{>}. This condition can be written without explicit reference to T_{>} by noticing that T_{k}T_{l}^{1} = . The condition becomes
G^{β}T_{>}, the set of orientations of the crystals α_{i}, does not form a group. However, since the `right class of equivalence' (Appendix A) is similar to condition (15), we will also use the same term for (15). One may also notice that G^{β} acts trivially transitively at the left of the set G^{β}T_{>} (Appendix B).
Remark 1
If the two phases α and β are the same, the crystals β and α are called twinned `partners' and, if the condition of twin orientation relationship is obeyed (the two crystals have a common lattice row and a parallel lattice plane), the transformation matrix of equation (11) corresponds to the `twin law' (Hahn & Klapper, 2003). With cubic crystals, a is given by a rotation matrix with rational elements , such that a_{ij} are integers with no integral factor (Warrington & Bufalini, 1971) and Σ is the ratio of the unit volume of the CSL lattice referred to the unit volume of the (Grimmer et al., 1974). The present approach generalizes the or the CSL lattice because no restriction exists on the vectors u, v, w introduced to define T_{>}; they can have real coordinates.
5. Algebraic definition of the orientational variants
5.1. The set of variants
We consider two crystals α_{i} and α_{j} that are variants inherited from the same parent crystal β, with the associated externally equivalent transformation matrices = T_{>} and = T_{>}. These two matrices define the orientation of the same crystal (i.e. the crystals are in coincidence, `α_{i} = α_{j}') if and only if they are internally equivalent [see equation (9)]:
= . Since is also an element of G^{β}, the condition becomes
It must be remembered that T_{>} = ▹; therefore, in equation (16), constitutes the set of symmetry matrices of the crystal α_{1} expressed in the basis of the crystal β. As the mapping
is an isomorphism, = ψ(G^{α}) is also a group. Since the intersection of two groups is not always a group, the reader is invited to verify that the elements of G^{β} ∩ effectively form a group that is a of G^{β}. We denote this H^{β} and call it the intersection group as done by Cahn & Kalonji (1981). More precisely, H^{β} can also be written because H^{β} depends on T_{>}, which is a function of the two bases and (the sign / just means `with reference to'). H^{β} is the set of symmetries common to crystal β and crystal α_{1}. We recall here that the external morphology of precipitates embedded in a matrix has the symmetries of the H^{β} group (Cahn & Kalonji, 1981).
To sum up, two external transformations and (elements of G^{β}T_{>}) point to the same daughter crystal (`α_{i} = α_{j}') if and only if they are also elements of T_{>}G^{α}, i.e. if and only if is an element of
The reader may now check that the transformation matrices giving equivalent ORs of the crystal α_{i} can be deduced from the matrix presented in equation (13); they form the set
Reciprocally, two crystals α_{i} and α_{j} are distinct if and only if H^{β} and H^{β} are distinct sets. We denote by H^{β} each left H^{β} in G^{β}. The canonical decomposition of G^{β} into left cosets leads to
where = E. Now, each is treated as a unique mathematical entity. Their set (i.e. a set of sets) is called the quotient set; it is denoted :
We have proved that each orientation of a crystal α_{i} (each orientational crystal) is defined by one left H^{β} i.e. an element of G^{β}/ H^{β}. In all the following, we will associate with the crystal α_{i} (geometric notion) a left (algebraic notion) by directly writing
The distinct α_{i} will be called orientational variants. Their number N^{α} in the decomposition (20) is given by the Lagrange formula
In the following, we will simply denote V^{α} as the set of the α_{i} variants:
In this paper, we will consider that G^{β} acts on V^{α} by classical left product: for any g ∈ G^{β} and α_{i} ∈ V^{α}, gα_{i} = g(H^{β}) = H^{β}. For this action,^{5} the of a variant α_{i} is defined by
It forms a G^{β} also called the isotropy group of α_{i}. The of α_{1} is = = . The of each variant α_{i} is the intersection group of α_{i}; it is given by
ofis the conjugate of H^{β} by . One may verify that it does not depend on the choice of the representative in the α_{i} and that equation (26) can be directly written
then appears as the product of two sets of matrices.^{6}
The present approach can be considered as a generalization of the previous studies of ):
and (see §2It is of primary importance to note that V^{α} = G^{β}/H^{β} is a quotient set and, in consequence, does not have a group structure in general. V^{α} can be considered as a group if and only if H^{β} is a of G^{β} (denoted H^{β} G^{β}, see Appendix A). Only in this case can V^{α} be called a `supergroup' or `composite group'. Moreover,
which means that all variants have the same intersection group H^{β}. There is a special case that implies H^{β}G^{β}: when there are only two left cosets in the decomposition (20), i.e. when there are only two variants; see Janovec et al. (2003, p. 384) and Guymont (1981).
The twodimensional example proposed in Fig. 1 is very illustrative. The of the parent crystal is G^{β} = , where, for instance, is the matrix of the mirror symmetry through the x axis and the matrix of the plane rotation of angle . All matrices of G^{β} are expressed in the same basis . The of the daughter phase is G^{α} = . For nearly all the orientation relationships (Fig. 1a), H^{β} is reduced to {E} and the number of variants is N^{α} = G^{β} = 8. In these cases, H^{β} is trivially normal in G^{β}; thus, each variant is equal to a singleton of G^{β}, for example α_{1} = {E}, α_{2} = , α_{3} = etc. We note the trivial isomorphism p(x) = {x}; the set of variants V^{α} is isomorphic to G^{β} (V^{α} ≅_{p} G^{β}) and, in consequence, has a group structure.
However, there are two special cases that differ from this `quasigeneral' case. In Fig. 1(b), H^{β} increases to because of the crystal β coincides with of the crystal α_{1}. In Fig. 1(c), H^{β} increases to because coincides with . In these two cases, the number of variants is reduced to N^{α} = 4. Each variant is described by a couple of symmetry operations:
and = = , = = . The very interesting fact is that V^{α} is not a group any more because now H^{β} is not a of G^{β}. For example, the reader can check in the case of Fig. 1(b) that and do not commute.^{7}
Remark 2
The orientational variants are algebraically similar to the domain states S_{i} = g_{i}F_{1} introduced in Janovec (1972) for the Landau transitions. Therefore, many of the results found by Janovec are also valid for reconstructive transitions if one takes the precaution to substitute G^{α} by H^{β} in some parts of his equations. For example, if we call n_{F} the number of distinct intersection groups (i.e. the number of distinct conjugate subgroups) and d_{F} the number of variants with the same intersection group: since ≤ G^{β} (see Appendix B), it follows that n_{F} = and d_{F} = and then that N^{α} = n_{F}d_{F}.
Remark 3
The equality (23) has already been proved by Humbert et al. (1992) in the specific case of Burgers transitions. Moreover, the demonstration is based on the decomposition of the rotational group of G^{β} into generator elements, but restrict G^{β} to the R^{β} = is in general incomplete. Indeed, a parent phase that has no rotational symmetries can generate variants; for example, G^{β} = {E, m} can generate two variants.
Remark 4
We have shown that V^{α} can be considered as a group if and only if H^{β}G^{β}. Another special condition has been described by Van Tendeloo & Amelinckx (1974): when ∃ K^{β} ≤ G^{β} such that G^{β} = H^{β}K^{β}, H^{β} ∩ K^{β} = {E} and H^{β}K^{β} = G^{β}. In this particular case, it is said that it is always possible to find in each left α_{i} one (a representative) such that the set of these representatives has a group structure. This set is called the variant generating group VGG. However, this group is not representative of the complete and complex structure of V^{α}. Indeed, the two examples in Figs. 1(b) and (c) have the same VGG = , whereas these two cases are distinct; this means that the information about the OR of the variants with their parent crystal is lost if only the VGG is considered.
5.2. Duality between the direct and inverse transitions
This part is only relevant for structural phase transitions. In the previous section, we have introduced for the β → α the intersection group (18): H^{β} = G^{β} ∩ T_{>}G^{α} with T_{>} = ▹ . Let us now consider the inverse transition α → β that can be obtained for example by increasing the temperature. The intersection group for this transition is H^{α} = G^{α} ∩ T_{<}G^{β} ≤ G^{α} with T_{<} = ▹ . One can choose α = α_{1} and β_{1} = β; it follows that
and, therefore,
Thus, the two groups H^{α} and H^{β} are linked by the isomorphism ψ introduced in equation (17): H^{β} ≅_{ψ} H^{α}. Both groups may be understood as the largest subgroups of G^{α} and G^{β} linked by the isomorphism ψ. One may also notice that
Thus, (H^{β}, H^{α}) appears as the solutions of the equation with unknown ∈ (G^{β},G^{α}): T_{>} = T_{>}, i.e. as the elements of G^{β}T_{>} ∩ T_{>}G^{α}, i.e. as the intersection of the trivial external G^{β} orbit with the trivial internal G^{α} orbit on the set G^{β}T_{>}G^{α} (see §§3.3 and 4.2).
Since H^{β} ≅_{ψ} H^{α}, and, since and , we deduce a simple equation that links the numbers of variants of the direct and inverse transitions:
To illustrate this equation, we consider the twodimensional case of Fig. 1: since G^{β} = 8 and G^{α} = 6, the number of variants of the inverse transition is N^{β} = 6 in the case of Fig. 1(a) (because N^{α} = 8), and N^{β} = 3 in the cases Figs. 1(b) and (c) (because N^{α} = 4).
From equation (33), one may also notice that, if G^{β} = G^{α}, then N^{β} = N^{α}. This means that the number of variants generated by a transition is equal to the number of variants generated by the inverse transition, whatever the orientation relationship. This is the case for the in steels γ → α (see §9.2).
The β → α Landau transition corresponds to H^{β} = ψ(G^{α}), which implies that H^{β} = G^{α}. Equation (23) applied to the inverse α → β transition leads to N^{β} = G^{α}/H^{α} = G^{α}/H^{β} = 1. This means that all the crystals α_{i} (inherited from the same parent crystal β_{1}) generate only one variant of phase β by the inverse transition (which is the initial parent crystal β_{1}). Thus, Landau transitions appear as reversible ones. As proved by (33), this is not the case for reconstructive transitions.
6. Definition of the operators
6.1. Transformations from one variant to another
Equation (19), = H^{β}T_{>}, can also be written with the help of equations (32):
This set of matrices corresponds to the transformations from a basis to bases such that: = ▹ ∈ G^{β}, T_{>} = ▹ and ▹ = ∈ H^{α}. Hence, the set of transformation matrices from a variant α_{i} to a variant α_{j}, denoted , is given by the matrices of the form [▹] = [▹]^{−1}[▹], with [▹] ∈ and [▹] ∈ . Therefore,
In the following we note
Since α_{ij} = with ψ the isomorphism defined in equation (17), there is a onetoone correspondence between the set and the set H^{β}H^{β}. Moreover, a trivial property of α_{ij} is that = α_{ji}, which means that the set of transformation matrices from α_{j} to α_{i} is the inverse of the set of transformation matrices from α_{i} to α_{j}.
6.2. The set of operators
It can be easily checked from equation (36) that all the sets α_{ij} with (i,j) ∈ [1,N^{α}]^{2} are either disjoint or equal. Each distinct α_{ij} is called an operator and denoted (the term `operator' will be justified later). Each operator is a class of equality on the set of the sets of transformation matrices {α_{ij}}. In other words, α_{ij} and α_{kl} are elements of the same class if and only if α_{ij} = α_{kl} (= ). An operator can be geometrically imagined as a color attributed to the arrows (transformations) that link the variants. The index n = 0 is applied for all α_{ii} with i ∈ [1,N^{α}]. One may verify that:
Each operator that transforms α_{i} into α_{j} is equal to
This shows that each operator takes the algebraic form of a H^{β}H^{β}. In consequence, the set of operators, which will be denoted O^{α}, can be deduced from the canonical decomposition of the group G^{β} into double cosets:
Each H^{β}\G^{β}/H^{β}:
(operator) can be treated as a unique mathematical entity. The set of operators is the double quotient set, algebraically denotedis the number of operators, it is equal to the order of this set:
Remark 5
The ) with details in Janovec & Dvorakova (1989).^{8} To keep Janovec's vocabulary for domain pairs, we say that an operator is ambivalent if = α_{ij} = α_{ji} = α_{ij} H^{β}H^{β} = H^{β}()^{−1}H^{β} = . For ambivalent operators, the set of transformations from the variant α_{i} to the variant α_{j} is equal to the set of transformations between the variant α_{i} and the variant α_{j}. One can then write instead of . If the operator is not ambivalent, i.e. if ≠ ≠ , the operator is called polar. However, an ambivalent equivalence class can be created from polar operators by writing = ∪ = with = ∪ .
decomposition was introduced for Landau transitions by Janovec (1972To illustrate our reasoning, let us consider the example of Fig. 1(a). One may verify with equation (38) that the set of variants V^{α} and the set of transformations that link the variants, O^{α}, are isomorphic by the trivial application p(x) = {x}: O^{α} ≅_{p} V^{α}. Thus the number of operators is equal to the number of variants and O^{α} has the same group structure as V^{α}. In the case of Fig. 1(b), it can be checked algebraically that = , = α_{12} = α_{21} = α_{23} = α_{32} = α_{34} = α_{43} = α_{41} = α_{14} = and = α_{13} = α_{31} = α_{24} = α_{42} = . We may see that all the operators are ambivalent. We also point out that neither V^{α} nor O^{α} has a group structure.
Although the decomposition (39) is canonical (see footnote 8), Janovec & Dvorakova (1989) did not give a general formula to calculate the number of double cosets. However, we will show now that such a generalization is possible. The reader is invited to look over Appendix B before reading the next section.
6.3. Numbering the operators
We consider the operator = H^{β}H^{β}. In this set, each right H^{β} can be written as a set of matrices {, ∈ H^{β}}. Hence, any operator (double coset) can be written as a set of left cosets:
Thus, according to the vocabulary of Janovec & Dvorakova (1989), if each operator comprises one it is called simple or, if it comprises many cosets, it is called multiple. The isotropic operator = {α_{1}} is always a simple operator. In the example of Fig. 1(b), the three operators can be written = {α_{1}}, = {α_{2}, α_{4}} and = {α_{3}}; only is multiple.
If H^{β} is normal in G^{β}, H^{β} = H^{β} (any right is equal to a left coset); consequently, any operator = {H^{β}} = {α_{i}}. The trivial isomorphism p(x) = {x} links each to a left and, in consequence, H^{β}\G^{β}/H^{β} ≅_{p} G^{β}/ H^{β}:
Then, if , one can associate one variant to each operator (the operator is simple): the set of variants and the set of operators have the same group structure G^{β}/H^{β}. A consequence is = . This is the case in Fig. 1(a).
More generally (whatever H^{β} is in G^{β}), from equation (39) one can always notice that H^{β} acts at the left of the set of variants V^{α}. Then, each operator can be identified to a left H^{β} orbit on the set V^{α}. Since H^{β} acts closely on V^{α}, these orbits can be called internal orbits. To sum up, an operator is expressed by a and represented by an internal H^{β} orbit on V^{α}. A direct consequence is that the number of operators is the number of H^{β} orbits on V^{α}; this number appears in the class equation (Appendix B):
Since ∣V^{α}∣ = N^{α} and = {α_{1}}, it follows that
where is the number of variants contained in each orbit and = {h^{β} ∈ H^{β}, h^{β} = } is the of in H^{β}.
Since = H^{β} ∩ H^{β} with ∈ (see the end of Appendix B) and since H^{β}( = [equation (26)], it follows that
From this formula, appears as the set of elements that leaves invariant both α_{1} and α_{i}, i.e. that leaves invariant the ordered pair (α_{1},α_{i}).
Equation (44) proves that the number of operators is always lower than the number of variants, and that the number of variants is the addition of the unity to some terms that divide H^{β}. To illustrate this property, we consider the case of Fig. 1(b): there are N^{α} = 4 variants that are partitioned into three operators {α_{1}}, {α_{2},α_{4}} and {α_{3}} whose number of elements divides H^{β} = 2 such that finally 4 = 1+2+1. Since there are three terms in this decomposition, = 3. The reader can also check in this example that equation (45) gives the of the three operators {α_{1}}, {α_{2},α_{4}} and {α_{3}}.
The number of operators given by equation (44) can be also directly obtained from the Burnside formula (see Appendix B):
where = {α_{i} ∈ V^{α}, h^{β}α_{i} = α_{i}} is the fixer of h^{β} in V^{α}.
In equation (42), {α_{i}} = {H^{β} with ∈ H^{β}} is also equal to {H^{β}α_{i}, ∈ H^{β}} = {α_{i} with ∈ H^{β}}; consequently,
This means that the variants α_{i} that constitute the operator in equation (42) are all the `images' of the variant α_{1} by the operator . Equation (47) can also be written
which should be read from left to right: `first take α_{1}, then apply and you obtain different variants α_{i}'. With this equation, appears as an operator. Moreover, acts as a multivalued operation that transforms α_{1} into many α_{i}. The pairs (α_{i},α_{j}) verifying α_{ij } = are the pairs (source, target) for the operator (see §7.1).
Remark 6
In this work, the information required to build the different variants inherited from the same parent crystal is obtained from the set of equivalent transformations defined in (19): = T_{>}H^{α}. However, the whole set of transformations from the basis to all the bases of any crystal α_{i} (not necessarily considered as a variant) is given by = T_{>}G^{α}, with the index t for total. The difference between the two sets and forms excessive information that is not useful to determine the algebraic structure constituted by the variants. The total set of transformation matrices from a crystal α_{i} to a crystal α_{j} is = G^{α}T_{>}G^{α}, which includes the set presented in equation (35). This is actually the set used in all the metallurgical studies devoted to the reconstruction of parent grains from EBSD maps, for example on martensitic steels (Gourgues et al., 2000) or on titanium alloys (Gey & Humbert, 2003). Indeed, if G^{α} is not reduced to {E} or {E,I}, each set is a set of matrices that contains at least one matrix with positive determinant, i.e. a rotational matrix. For each rotational matrix of the set , the rotation angle and axis (δ,U) can be easily deduced from its eigenvalues and eigenvectors: cos(δ) is known from the trace of the matrix and the sign of δ can be attributed if the reference coordinate system is oriented as a direct basis. Then, at each set , one may associate a representative pair (δ_{min},U) with the minimum angle and the corresponding rotation axis in order to create a class of transformations: two transformations and are declared `equal' or `distinct' by comparing their corresponding pairs (δ_{min},U)_{ij} and (δ_{min},U)_{kl}. Actually, in the previous metallurgical studies, only ∣δ∣ has been taken into consideration [probably because only cos(δ) was considered in the rotational matrices]. The operators presented in these studies (they are called `special misorientations') are therefore less precise since they are all forced to be ambivalent: they correspond to the operators of type . We have determined by computation of the operators and the corresponding pairs (∣δ_{min}∣,U) from the operators defined by equation (47) for any the results for Burgers transitions are presented in §9.1
7. of orientational variants
We have seen that in general V^{α}, the set of variants, does not have a group structure. What could be the algebraic structure of such a set? Moreover, what is the algebraic structure of O^{α}, the set of operators that link these variants? Since all the terms have been mathematically defined in the previous sections, it will be quite direct to show that the combination of two sets V^{α} and O^{α} forms a structure. These considerations will allow us to show that the composition of operators is a multivalued operation and how it is possible to establish a composition table for the operators. This section must be considered as a theoretical approach; an example of the application of the groupoids to materialsscience problems will be given in §10.1 with more details in Cayron et al. (2006).
7.1. Introduction to groupoids
A group defines a structure of actions without explicitly presenting the objects on which these actions are applied. Indeed, the actions of the group G applied to the identity element e implicitly define the objects of the set G by ge = g; in other terms, in a group, actions and objects are two isomorphic entities. A enlarges the notion of group by explicitly introducing, in addition to the actions, the objects on which the actions are applied. By this approach, many identities may exist (they correspond to the actions that leave an object invariant). We present here a definition adapted from the Weinstein (1996) paper about the transformation groupoids: let x be a solid and Γ a finite subset of the group of isometries O(3), and let us call X the set of similar solids formed by the actions of Γ on x. The transformation formed by X and Γ is the set:
with the partially defined operations: (x,γ,y)(y,ν,z) = (x,γν,z). One may imagine γ of Γ as an arrow (without reference to any object of X) and g of G(X,Γ) as the arrow γ from the object x to the object y. Contrary to the convention used by Weinstein, the composition of the binary operations γν must be understood as `first apply γ and then apply ν'. This convention is the direct consequence of the definition of transformation matrices presented in §3.1 and of their composition rule (6).
We denote by s and t two maps (called source and target) from G(X,Γ) onto X such that:
The operations on G(X,Γ) have the following properties.
1. The product is defined only for certain pairs of elements of G: gh is defined only when t(g) = s(h).
2. It is associative: if either the product (gh)k or g(hk) is defined, then so is the other and they are equal.
3. For each g ∈ G, there are left and right identity elements λ_{g} and ρ_{g} such that λ_{g}g = g = gρ_{g}.
4. Each g of G has an inverse g^{−1} for which gg^{−1} = λ_{g} and g^{−1}g = ρ_{g}.
More generally, a X is a set G(X,Γ) with mappings s and t from G(X,Γ) onto X and a partially defined (g,h) → gh satisfying the four previous conditions. One may consider each g of G as the arrow s(g) → t(g). This direction is more natural than the one resulting from the Weinstein convention.^{9} Some concepts are also important in theory:
with a base(a) The isotropy group of x ∈ X consists of those g in G(X,Γ) with s(g) = x = t(g).
(b) An orbit of the G(X,Γ) containing x is O_{x} = {g = (x,γ,y), xγ = y and γ ∈ Γ}. The set of orbits over X defines an equivalence class with x ∼_{G} y if and only if there is a element g such that s(g) = x and t(g) = y; i.e. xγ = y.
(c) We need to introduce a new notion, the `operator', to explain the results obtained in the present study. Let us consider (x_{1}, y_{1}) and (x_{2}, y_{2}) ∈ X^{2} such that x_{1}γ = y_{1} and x_{2}ν = y_{2}. We introduce the equivalence relation on the pairs by: (x_{1},y_{1}) ∼ (x_{2},y_{2}) γ = ν. It is the relation of pairs of elements (x,y) similarly placed in the The equivalence classes partition the G according to the sets of elements g = (x,γ,y) such that x and y are linked by the same The associated with each set of this partition is what we call an operator.
In order to illustrate the definition of G(S,+), defined by the set of elements g = (z_{1},z_{2}−z_{1},z_{2}) with (z_{1},z_{2}) ∈ S^{2} and the composition rule (z_{1},z_{2}−z_{1},z_{2})(z_{2},z_{3}−z_{2},z_{3}) = (z_{1},z_{3}−z_{1},z_{3}). One may check that G(S,+) is a This kind of is called a pair or banal The identities are g_{0} = (z_{1},0,z_{1}) and, if g = (z_{1},z_{2}−z_{1},z_{2}), g^{−1} = (z_{2},z_{1}−z_{2},z_{1}). For any fixed number Z, the equivalence class of elements (z_{1},z_{2}−z_{1},z_{2}) linked by the same operation z_{2}−z_{1} = +Z forms the operator +Z. This must be understood by: z_{1}+Z = z_{2} (`first consider z_{1}, then apply +Z, and you obtain z_{2}').
let us consider a simple onedimensional example with the set of the relative integers associated with the operation + (addition), but omitting, for instance, the numbers 5 and 11: . Of course, (,+) is a group, but not (S,+). However, one can associate with S the structure7.2. Definition and properties of the of orientational variants
In the previous section, we have defined the set of variants V^{α} = {α_{i}} = G^{β}/H^{β} for the β → α transition. The transformation from the variant α_{i} to the variant α_{j} was defined by = α_{ij}T_{>}, with α_{ij} = being an element of the set O^{α} = H^{β}\G^{β}/H^{β}. Now, it may be checked that V^{α}, the set of variants, associated with O^{α}, the set of operators, forms a that we call a of orientational variants:
with the partially α_{i},α_{ij},α_{j})(α_{j},α_{jk},α_{k}) = (α_{i},α_{ij} α_{jk},α_{k}) = (α_{i},α_{ik},α_{k}), as illustrated in Fig. 2. In all the following, we will directly associate α_{ij} with (α_{i},α_{ij},α_{j}). The variants α_{i} and α_{j} appear as the source and target of α_{ij}: α_{i} = s(α_{ij}) and α_{j} = t(α_{ij}). One may imagine α_{ij} as an arrow pointing from α_{i} to α_{j}.
(The G^{β→α} is geometrically rather simple, i.e. its lowlevel properties can be visualized or verified from the algebraic definition (50).
1. The α_{ij}α_{jk})α_{kl} and α_{ij}(α_{jk}α_{kl}) are equal (to α_{il}).
is associative: the products (2. For each α_{ij} in G^{β→α}, there are left and right identity elements that are both equal to α_{ii} = α_{jj} = H^{α} = , which also constitutes the isotropy group of each variant α_{i}.
3. Each element α_{ij} in G^{β→α} has an inverse equal to = α_{ji}.
In the G^{β→α}, all the binary operations between the variants are defined; consequently, there is only one trivial orbit on the base V^{α}. However, there are many operators in G^{β→α} and their set is O^{α}. For each operator of O^{α}, one can associate a set of pairs {(α_{i},α_{j}), (α_{k},α_{l}), …} of variants similarly oriented in the parent crystal, i.e. such that α_{ij} = α_{kl} = … = .
This definition of N crystals in this polycrystal, there will be, in general, 2C(N,2) operators that link these crystals. What makes the of orientational variants interesting relies on the fact that the variants are not randomly oriented and, consequently, the number of operators is far lower than 2C(N,2). Moreover, these operators can be composed in a special way and this composition may be seen as the signature, as shown in the following.
is very general. Any polycrystal (a set of randomly oriented crystals) can actually be imagined as a If there are7.3. Composition of operators
We can perform two interesting compositions between the operators of the
of orientational variants.The first one is: ∈ (O^{α})^{2} → = ∈ P(O^{α}), where P(O^{α}) is the set of partitions of the set O^{α}. This composition can be easily computed from the associativity condition of the Indeed, we write as a set of α_{ij} and as a set of α_{kl}. Then, we find in each set the elements that verify t(α_{ij}) = s(α_{kl}), i.e. j = k, and write α_{il} = α_{ij}α_{kl}. Since each α_{il} belongs to a class , the product of operators takes the form of a set .
The second one is: ∈ (O^{α})^{2} → = ∈ P(O^{α}). This composition can also be easily computed by taking α_{1} as a reference. Indeed, we write as a set of α_{i} (= α_{1i}) and as a set of α_{j} (= α_{1j}). Then, we directly write α_{ij} = . In general, the composition of two operators is neither reduced to ∅ nor to a singleton. It means that the composition is not a classical application but a multivalued product. The multivalued composition of operators is geometrically visible. In the example of Fig. 1(b), since = , = α_{12} = α_{21} = α_{23} = α_{32} = α_{34} = α_{43} = α_{41} = α_{14} and = α_{13} = α_{31} = α_{24} = α_{42}, it may be checked that = , = and = . In this simple case, the multivalued result obtained with comes from the ambivalence of ; but this is not always the reason (as will be shown for a more complex case given in §9.1). The composition has the advantage of allowing the representation of the entire structure of the (variants and operators) in the same table that we will call table of composition of the (see Fig. 6 for example, with details in §9.1). This table plays the same role as a group table; it is the signature of the of orientational variants and is characteristic of the crystallographic aspects of the phase transition.
7.4. Comparison to space groupoids
Groupoids were first introduced in crystallography for order–disorder (OD) structures by DornbergerSchiff & GrellNiemann (1961); after that the same authors noticed the partial character of some symmetry operations in twin structures (DornbergerSchiff, 1959). OD structures are crystallographic structures constituted by twodimensional periodic layers linked by partial (or local) operations (PO). The name `order–disorder' is probably too broad since only the order in the stacking of the layers is not complete (the positions of the atoms inside each layer are entirely determined). The OD theory is actually a crystallographic (and algebraic) description of but order–disorder transitions are far more universal and complex than order–disorder in DornbergerSchiff points out that the set of POs does not form a group but a since, for example, the operation that transforms the layer L_{p} in L_{q} and the operation that transforms the layer L_{r} in L_{s} cannot be combined, unless q = r. Such considerations explain some diffraction enhancements of symmetry not due to Friedel's law^{10} in (Sadanaga, 1978; Sadanaga & Ohsumi, 1979) and in quasicrystals (Yamamoto & Ishihara, 1988). Those works introduce the concept of space in order to define the operations on a assuming that this one is constituted by a repeated Thus, the space and the of orientational variants could appear very different since the related problems are different. However, both can be considered as the algebraic description of a composite structure built from specially placed subunits and therefore are similar from a geometrical point of view. Their algebraic definitions are actually very close.^{11}
8. Calculation of the possible parent crystals
8.1. Motivations for the calculation
We denote by θ_{c} the transition temperature for Landau transitions, and by θ_{s} and θ_{f} the start and finish temperatures for reconstructive transitions. We call `half a cycle' the operation that consists in cooling the material from a temperature θ_{1} > θ_{s} (or θ_{c}) to a temperature θ_{2} < θ_{f} (or θ_{c}) or in heating it from θ_{2} to θ_{1}. By `transition cycle' we mean the succession of the two operations: cooling from θ_{1} to θ_{2} and heating from θ_{2} to θ_{1}. Now, let us consider a transition cycle: we start at θ_{1} with a material constituted only by one crystal β_{1}; this material is then cooled to θ_{2} and completely transformed by the transition β_{1} → {α_{i}}. The material is now constituted by N^{α} orientational crystals {α_{i}}. It is then reheated at θ_{1} such that each crystal α_{i} becomes a parent crystal for the transformation α_{i} → {β_{k}} (end of a cycle). We call these new crystals of phase β, which are variants of variants, β crystals of second generation. Their number is denoted as . Since many of them are identical (in the same orientation), one may verify that
For Landau transitions, since N^{β} = 1, all the variants α_{i} recreate the same orientational crystal, which is β_{1}; and, in consequence, during thermal cycles, = = = = N^{β} = 1. For reconstructive transitions, to our knowledge, there is no general formula that gives .
This theoretical `thermal cycling problem' is close to the one encountered in parent grain reconstruction from EBSD data. Only the orientations of a limited number of daughter grains is known in the data but one would like to reconstruct the parent grains and know their location and orientation. This is useful to determine if cracks occur and propagate at parent grain boundaries (Gourgues et al., 2000), or to better understand why and how variant selection occurs during successive heating and cooling treatments (Gey & Humbert, 2003; Standford & Bate, 2005). In order to determine the possible parent crystals of a finite set of variants {α_{i}}, one may imagine the inverse transition such that each α_{i} becomes a parent crystal of a set of crystals {β_{k}}. The number of possible parent crystals of the crystals α_{i} is then given by the set of all the daughter crystals β_{k} that have α_{i} as parent crystals. This problem has been treated by Humbert et al. (1995) for Burgers transitions. This section generalizes this approach with the help of the algebraic notions presented in the previous sections. Both problems (calculation of and determination of parent grains) require the resolution of the same intermediate problem: assuming that the orientations of two variants are known, how many distinct crystals (i.e. with distinct orientations) can be their parent?
8.2. Parent crystals common to two daughter crystals
Let us consider two α variants inherited from a crystal β_{1}, for example α_{i} = H^{β} and α_{j} = H^{β} ∈ (V^{α})^{2}. These two variants are determined from their respective orientations with the crystal β_{1} given by equation (34):
We want to know if these two variants can also be inherited from another parent crystal, which we will call, for example, β_{k}. It is essential to notice that the index k of this crystal is given with reference to the other crystals α_{i} or α_{j}. Hence, if β_{k} is inherited from the parent crystal α_{i} by the inverse transition, it must be denoted and, if it is inherited from the parent crystal α_{j}, it must be denoted (see Fig. 3). Of course, and are both elements of the same set V^{β} but in general k_{i} ≠ k_{j}. What is the algebraic relation between and ?
By inverting each member of the indexing convention chosen (12), we obtain
Then, by applying again the indexing rule (12), we can generalize:
It follows that ▹ = ▹▹ = T_{<}. It must be noticed that is an element of G^{α}. By indicating its index in this list as k_{i}, we may write = . The crystal β_{k} is then determined from two transformation matrices:
These two transformation matrices are respectively elements of
with and elements of V^{β}. One may notice that the matrices of and are elements of the external orbit G^{α}T_{<} around α_{i} and α_{j}, respectively (see Fig. 3). From equations (56), β_{k} can be calculated in the same reference coordinate system in two ways:
These two transformation matrices define the same crystal β_{k} if and only if, from equation (9),
By using equation (32), one may verify that equation (58) does not depend on the choice of representative and in their respective cosets H^{α} and H^{α} (both elements of V^{β}). Consequently, only one representative can be chosen in each of V^{β} when using equation (58). This simplification is not possible for the choice of and , however, since these two matrices are elements of α_{i} and α_{j}, respectively, the matrix = ∈ , where is the operator that links α_{i} to α_{j}. Testing all the ∈ generates all the possible solutions of equation (58). Therefore, we conclude that two crystals α_{i} and α_{j} inherited from the same parent crystal β_{1} and linked by can also be inherited from another crystal β_{k} if and only if
with ∈ . One may check, with the help of equation (32), that β_{1} = H^{α} always obeys equation (59). We recall that k_{i} and k_{j} represent the indices of the crystal β_{k} with reference to the crystals α_{i} and α_{j}, respectively. One can also verify that the operator that links β_{1} to β_{k} is given by = = = H^{α} = H^{β}.
8.3. Number of variants of second generation
The variants of second generation corresponding to the cases of Figs. 1(b) and (c) are presented in Figs. 4(a) and (b), respectively; their number can be geometrically determined; it is = 3. In the present study, we do not propose a general algebraic formula that gives this number. We propose, however, a method that allows its calculation and that can be easily computed. We first consider the crystal α_{1} (inherited from β_{1}) and all its variants . Then, we consider the second crystal α_{2} and all its variants . By applying equation (59), we deduce all the variants of identical (internally equivalent) to one of the variants of , and then the variants of that are `really new' (those different from any . Now, we consider the third variant α_{3} and all its variants , and keep in this set only the new crystals (i.e. different from and ) etc. This algorithm has been applied to calculate the number of variants of second generation for complex cases such as those presented in §9.1. Of course, the mathematical study of the series of variant numbers , , …, of the nth generation would require a deeper study.
9. Computation and application to reconstructive transitions
The twodimensional cases like those presented in Fig. 1 are in general geometrically simple. However, threedimensional crystallographic cases encountered in materials science (phase transition, precipitation, twinning) are more complex. Therefore, we have written a computer program in Python language (Marteli, 2004), according to the equations presented in the previous sections, which calculates the set of variants, the set of operators, the composition table of the and the possible parent crystals inherited from two daughter variants. It takes a few seconds to compute all the variants and operators with a 1 GHz PC for any crystalline Such speed could not have been obtained without the theoretical considerations described in this paper. We give here the details of the results obtained for Burgers transitions; some results obtained for martensitic transitions in steels are also briefly reported.
9.1. Burgers transitions
The β → α Burgers transitions (Burgers, 1934) are transitions from a hightemperature bodycentered cubic phase (β) to a lowtemperature hexagonal closepacked phase (α). This transition occurs for example in Ti or Zr alloys. One of the possible ORs that links an α crystal to a β parent crystal is given in Table 1. The intersection group H^{β} is given in Table 2 and the computed variants α_{i} are presented in Fig. 5. We found that H^{β} = 4 and, since G^{β} = 48, the number of α variants is N^{α} = 12. The operators associated with the pairs of variants (α_{i}, α_{j}) are determined by calculating all the distinct sets of matrices α_{ij} = , as presented in Table 3. We find that = {α_{1}}, = {α_{2},α_{3}}, = {α_{4}}, = {α_{7},α_{9}}, = {α_{6},α_{12}}, = {α_{5},α_{10}} and = {α_{8},α_{11}}. The number of operators is = 7, in agreement with the class equation (44): 12 = 1+2+1+2+2+2+2 = addition of seven terms that each divides H^{β} = 4. For each operator, the list of source and target pairs (α_{i},α_{j}) is directly deduced from Table 3 and presented in Table 4. We have also calculated the set of transformation matrices (see Remark 6) and we have reported some rotations with the minimum angle and with special angles (60, 90, 120 or 180°) in the last column of Table 4. One may notice that the operators and are complementary polar operators and that the other ones are ambivalent. If the operators are forced to be ambivalent (i.e. and considered as the same operator), we find that = 6, as published by Gey & Humbert (2003). The composition of the operators is presented in Fig. 6. All the parent crystals of two variants α_{1} and α_{i}, {β_{1},…,β_{k}} are calculated according to equation (59) and are given in Table 5. From this table, we recursively calculate if a set {α_{1},…,α_{N}} of N daughter variants can have only one parent crystal (β_{1}). We found that four variants determine unambiguously the parent crystal. Indeed, in contrast to the results given by Humbert et al. (1995), in some special cases, even three variants {α_{1},α_{6},α_{12}} can be inherited from two different parent crystals β_{1} and β_{6}, where β_{6} is a twin of β_{1} through the mirror plane. The situation can be understood by observing on Fig. 5(a) that α_{1}, α_{6}, α_{12} have their basic vector = [100] normal to the plane of β_{1} and, consequently, are invariant by the rotation of 180° around [100]_{α1} = [11]_{β1}, which transforms β_{1} into β_{6}. This solution has been forgotten by Humbert et al. (1995) probably because in that study only the rotational symmetries were taken into consideration. Moreover, from Table 5 and Table 4, we have determined that the number of variants of second generation is = 40.



9.2. Martensitic transitions in steels
In the γ → α martensitic transitions in steels, the parent phase is the facecentered cubic austenite (γ) and the daughter phase, at low carbon levels, is the bodycentered cubic martensite (α′). There are two possible orientation relationships: the Nishiyama–Wasserman (NW) OR (Nishiyama, 1934; Wassermann, 1933) and the Kurdjumov–Sachs (KS) OR (Kurdjumov & Sachs, 1930). Some results on the of variants are briefly reported here; more details will be given by Cayron et al. (2006). Since the parent and daughter phases have the same pointgroup order, the number of variants is independent of the direction of the transition arrow (heating or cooling). We find that H^{γ} = H^{α} = 4 and N^{α} = N^{γ} = 12 variants for a NW OR, and that H^{γ} = H^{α} = 2 and N^{α} = N^{γ} = 24 for a KS OR, which is well known in metallurgy. The operators are determined by computation, we find = 7 with a NW OR and = 24 with a KS OR. If the operators are forced to be ambivalent, we find that = 6 with a NW OR and 17 with a KS OR, as published by Gourgues et al. (2000). All the parent crystals of two variants α_{1} and α_{i}, {β_{1},…,β_{k}}, are determined by computation; we have deduced that four and seven distinct variants determine unambiguously the parent crystal for a NW OR and a KS OR, respectively.
10. Applications and perspectives
This study generalizes the previous crystallographic studies on the Landau transitions to any type of crystalline transitions (Landau and reconstructive). We have introduced two equivalence classes of transformation matrices (internal and external) on the orientations of crystals and then we have associated the geometric object `daughter crystal' and the algebraic object `coset' into a unique object `orientational variant'. We have defined the operators as classes of transformations between variants and have identified them as double cosets. We have shown that the set of variants and the set of operators constitute a
which we have called a of orientational variants. But what could be the applications and implications of such concepts?10.1. Reconstruction of parent grains from EBSD maps
This theoretical study takes its origin from an applied materials science problem. We initially wanted to determine, from EBSD maps acquired on different α titanium alloys used for space applications and α steels used for nuclear applications) the size and the orientation of the parent grains that have produced these variants, i.e. the βTi and γFe grains, respectively. Some studies of this problem already exist for Burgers transitions; the method consists in calculating the disorientation between two α grains and checking if it corresponds to a `special misorientation' (i.e. an operator of 〈O^{α}〉). As we will explain and quantify in Cayron et al. (2006), the probability of finding an operator between two randomly oriented α crystals is low for Burgers transitions but is far from being negligible for martensitic transitions in steels. The only way to avoid abusive grain regrouping is to check the coherency of all the α crystals considered as variants at each step of the algorithm. This can be performed with the help of the composition table of the (§7.3), without solving any equation as in Gey & Humbert (2003). For example, to know if three adjacent grains are variants inherited from the same parent crystals, one must check (i) if the misorientations between them correspond to ones of the theoretical operators and (ii) if the composition of the operators is coherent. In this way, if the grains G_{1}, G_{2} and G_{3} are found to be linked by the operators G_{12 } = , G_{13} = and G_{23 } = , then since G_{23} = (G_{12})^{−1}G_{13} ∈ (see Fig. 6), the triplet of operators is coherent. If grains G_{1}, G_{2} and G_{3} were found to be linked by the operators G_{12 } = , G_{13} = and G_{23 } = , the triplet is not coherent (see Fig. 6), so we would be able to deduce that these grains are in fact accidentally linked by misorientations close to operators. More details will be given in Cayron et al. (2006).
materials (10.2. Groupoids and complexity
It may be noticed that the algebraic structure describing the `general' case of Fig. 1(a) is a group and the one describing the `particular' cases of Figs. 1(b) and (c) is a an algebraic structure that is more general than groups. This is a surprising point; indeed, one could expect the opposite and think that `particular' cases are described by one subfamily of the `general' case. Two elements of response can be given.
1. If we link an orthogonal basis to the parent cubic crystal and to the triangular daughter crystals (with the help of the structure tensors, for example), T_{>} can be substituted by a rotation matrix R_{>}. Then, the number of variants N^{α} appears as a function of ϕ, the rotation angle of R_{>}. In the case of Fig. 1, this function is always equal to 8, except for the rotation angles δ = Nπ/4 (with N integer), where N^{α} is then reduced to 4. The function N^{α}(ϕ) appears as a Dirac comb with a period T = π/4. Each Dirac of this comb represents an `algebraic structure breaking' (with reference to the usual term `symmetry breaking' used for phase transitions). Therefore, the words `particular' and `general' probably do not have any mathematical meaning in this case: Figs. 1(a) and (b) are actually two distinct cases.
2. The fact that Fig. 1(a) seems geometrically `more complex' than Fig. 1(b) comes from its higher number of triangles. However, this is only an apparent complexity. Indeed, the number of components in a system is only one part of its complexity; the interconnections between the components are also very important (see for example Allegrini et al., 2004). One may actually conceive that the structures of Figs. 1(b) and (c) are more complex than that of Fig. 1(a) because these structures are `constrained'; any modification of ϕ degenerates the algebraic structure of Figs. 1(b) or (c) (a groupoid) into a simpler one (a group). More studies would be required to clearly define an `algebraic complexity'.
10.3. Transition cycles: irreversibility and entropy
For Landau transitions, as presented in §8.1, the number of variants of the nth generation is always equal to 1: = = = = N^{β} = 1. Consequently, if only the orientations are considered, the system remains algebraically stable (the algebraic structure of the initial crystal, G^{β}, and that of orientational variants, V^{α}, are isomorphic), and the physical structure is recovered after one cycle. We conclude therefore that Landau transitions are reversible during thermal cycling. This implies that the irreversible part of the variation at each cycle is zero: ΔS_{i} (one cycle) = 0. We recall that the present system is open (and not isolated) and that ΔS = ΔS_{e}+ΔS_{i} where ΔS_{e} is the flow of due to the heat exchanges and ΔS_{i} is the production owing to irreversible processes inside the system (Prigogine et al., 1972).
However, the situation is completely different for reconstructive transitions. For example, for Burgers transitions, one monocrystal β_{1} can be transformed by cooling into 12 variants by β_{1} → α_{i} and, if we reheat the material, each variant becomes the parent crystal for the transition α_{i} → such that the number of variants of second generation is = 40 (see §8.3). At each cycle of reconstructive transitions, more and more variants are introduced, the complexity of the system increases and the structure will never recover. Consequently, we find algebraically that the reconstructive transitions are intrinsically irreversible. The irreversible part of the variation on the first cycle is strictly positive: ΔS_{i}(1st cycle) > 0. For the moment, no general formula that gives could have been proposed. But is it possible to find such a formula? More generally, how many variants do we obtain after N cycles? What is their algebraic structure? The number of distinct crystals increases after each cycle , but does it stabilizes for an nth cycle? This would imply that ΔS_{i}(nth cycle) = 0. This problem constitutes a fascinating algebraic subject. Experimentally speaking, if we heat and cool a phasetransition material around its transition temperatures (in order to decrease its grain size and increase its mechanical properties) and perform Xray diffraction, will we obtain an isotropic (random) or textured pattern after a very high number of cycles? Geometrically speaking, if we assume that the crystals keep their size, we must study how to make a threedimensional pavement with two kinds of solids like in a Penrose problem. If we assume that the variants are smaller and smaller at each cycle (such as presented in Fig. 4), the different generated variants will form a threedimensional fractal. For example, we will show in another study (Cayron, 2006) that the Σ3^{n} multiple in cubic materials can be algebraically represented by a and geometrically by a threedimensional fractal.
We recall that the complexity of the structure produced by reconstructive transitions comes from the fact that the H^{β} is not in general a normal group in G^{β}, which is a consequence of the noncommutativity between the transformation matrices. Therefore, it could be interesting to see if some mathematical concepts developed in noncommutative geometry (Connes, 1990) could be applied to describe mathematically the transition cycles.
11. Conclusions
In this study, α daughter crystals in an orientation relationship with a β parent crystal have been defined from their orientations in a fixed reference basis with the help of transformation matrices. They can be created by β → α phase transitions (Landau or reconstructive), by or by precipitation. We called T_{>} one of the possible transformation matrices from a basis of a parent crystal β to a basis of a daughter crystal α_{1}. Two equivalence classes on the transformation matrices have been introduced: an internal class for the matrices that point to the same crystal α_{i} and an external class for the matrices that point from the same crystal β. The set of internally and externally equivalent transformation matrices are T_{>}G^{α} and G^{β}T_{>}, respectively. These two classes are equal for matrices of the intersection group H^{β} = G^{β} ∩ T_{>}G^{α}. The set of distinct daughter crystals α_{i}, denoted V^{α}, has been identified as the quotient set G^{β}/H^{β}. In this way, we have associated the geometric notion of orientational variant α_{i} with the algebraic notion of left H^{β} and created a unique concept: the orientational variants. Their number directly results from the Lagrange formula N^{α} = G^{β}/H^{β}. By considering the β → α and α → β transitions, we proved the isomorphism H^{α} ≅ H^{β}, and hence, that N^{α}G^{α} = N^{β}G^{β}. The transformations from one variant to another were partitioned into distinct equivalence classes. The set formed by these classes is isomorphic to the set of the left H^{β} orbits on the quotient set G^{β}/H^{β}, which is the double quotient set H^{β}\G^{β}/H^{β}. These orbits were called `operators'. Their number is given by the Burnside formula. The set of orientational variants associated with the set of operators constitutes a We have proposed a composition table for this that acts as a crystallographic signature of the A method to determine if two daughter variants α_{i} and α_{j} can be inherited from more than one parent crystal was given by calculating the intersection of external orbits. A computer program has been written to determine the variants, the operators, the composition table of the and the possible parent crystals for any Some results obtained with the Burgers transition and with the in steels are given and compared to published results. This work opens the way for deeper algebraic researches on fractal structures formed by reconstructive phasetransition materials after thermal cycling. The complexity, irreversibility and of such systems were briefly discussed.
APPENDIX A
Elements of group theory
An equivalence relation ∼ is a relation among elements a of a set X satisfying: (a) reflexivity: a ∼ a; (b) symmetricity: a ∼ b → b ∼ a; (c) transitivity: a ∼ b, b ∼ c → a ∼ c. An important property of an equivalence relation is that it partitions the set X into distinct equivalence classes C_{a} = {b ∈ X, b ∼ a} such that for any (a, b) ∈ X^{2} either C_{a} = C_{b} or C_{a} ∩ C_{b} = ∅. It may be noted that this property is not linked to the algebraic structure of the set X.
A group is a nonempty set G together with a law of composition (a,b) → (ab): G × G → G satisfying the following axioms: (i) associativity: for all a, b, c ∈ G, (ab)c = a(bc); (ii) existence of an identity element: ∃ e ∈ G such that for all a ∈ G: ae = ea = a; and (iii) existence of inverses: for all a ∈ G, ∃ a^{−1} ∈ G such that aa^{−1} = a^{−1}a = e. We will use G to denote, for a finite group G, the number of elements of G, also called order of G. A H on a group G, denoted H ≤ G, is a nonempty subset of G such that (a) for (a, b) ∈ H^{2}, ab ∈ H and (b) for a ∈ H, a^{−1} ∈ H. When specifying a group–subgroup relation, the sign ≤ is preferred to because a set X can be included in a group G without being itself a of G.
A homomorphism from a group (G, .) to a second group (G′,*) is a map α: G → G′ such that, for all a, b ∈ G, α(ab) = α(a)*α(b). The image of α is defined as Im(α) = {g′ ∈ G′, ∃ g ∈ G with g′ = α(g)}. It is a of G′. The kernel of α is defined as Ker(α) = {g ∈ G, α(g) = e′} with e′ being the identity element of G′. It is a of G. An isomorphism between G and G′ is simply a bijective homomorphism (a onetoone map between G and G′). α is an isomorphism if and only if Ker(α) = {e} and Im(α) = G′.
Let H be a of G. For any fixed a ∈ G, a left of H in G is defined by the set of the form aH = {ah for h ∈ H}. Similarly, a right is defined by Ha = {ha for h ∈ H}. It can be checked that (i) two left cosets are either disjoint or equal, that (ii) aH = bH if and only if a^{−1}b ∈ H, and that (iii) any two left cosets have the same number of elements equal to H. Similarly, two right cosets Ha = Hb if and only if ab^{−1} ∈ H.
The left and right cosets of H in G partition G. The condition `a and b lie in the same left is a left class of equivalence. We denote G/H the set of all the left cosets and G\H the set of all the right cosets. Card(G/H), denoted G/H or [G:H], is the number of left cosets of H in G, it is also equal to the number of right cosets. A direct consequence of the partitioning of G into left equivalence classes is the Lagrange formula: for any H of a finite group G, the order of H divides the order of G, i.e. G = G/HH.
Two subgroups H_{1} and H_{2} are conjugate if ∃g ∈ G such that H_{1} = gH_{2}g^{−1}. One may verify that the property `being conjugate' constitutes an equivalent class. A N of G is normal (or invariant, or selfconjugated), denoted N G, if and only if, for all g ∈ G, gNg^{−1} = N. This condition is also equivalent to: for all g ∈ G, gN = Ng or to: for all gN in G/N, (gN)^{−1} = Ng^{−1}. In other words, for a N, each left of N in G is equal to a right of N in G. The kernel of a homomorphism α: G → G′ is a of G. If NG, a natural composition law `.' can be defined on G/N the set of cosets of N in G by
Indeed, the law `.' is well defined (i.e. it does not depend on the choice of the representative a in the equivalence class aN) and it confers on G/N a natural group structure. Reciprocally, a natural group structure exists on G/N only if N G. In this case, G/N is called the quotient group.
There are many important isomorphism theorems in group theory that will not be described here. One of their corollaries is the decomposition of a group into a ): Consider two subgroups H and K of a group G such that G = HK = {hk, h ∈ H, k ∈ K} and H ∩ K = {e}. Then:
of two subgroups – see for example Milne (2003(a) if H G, then G/H is isomorphic to K; there is a onetoone correspondence between the left cosets and the elements of K;
(b) if H G and if K G, then G is isomorphic to H × K = {(h, k), h ∈ H, k ∈ K}; the group G can be canonically decomposed as the product of the two subgroups H and K.
APPENDIX B
Groups and actions on sets
Let us now introduce some notions on groups acting on sets. Let X be a set and G be a group. A left action of G on X, denoted `.', is a map (g, x) → g.x: (G, X) → X such that: e.x = x for all x ∈ X and (gh).x = g(h.x) for all gh ∈ G and x ∈ X.
For x and y ∈ X, we write x ∼_{G} y if and only if ∃ g ∈ G such that y = g.x. It can be checked that this relation is an equivalence relation on X. The left equivalence classes are called the left G orbits. For a fixed x ∈ X, we denote the left G orbit containing x as
Note that all the distinct G orbits partition X. The group G is said to act transitively on X if there is only one orbit, i.e. if, ∀ x,y ∈ X, ∃ g ∈ G/y = g.x.
Let us call for any x ∈ X, the at the left of x:
It is a G, also called the isotropy group. One can then prove that, if G acts at the left on X, then the application defined by g ∈ G and x ∈ X,
ofis a bijection. Therefore, = G/ Stab_{G}(x).
We call O_{i} all the G orbits in X and N_{o} their number: .
If we choose for each O_{i} a point x_{i} ∈ X such that O_{i } = O_{xi }^{}, we obtain the class equation:
Now let us introduce, for g ∈ G, Fix_{X}(g) = {x ∈ X/g.x = x}. Then
We obtain the Burnside formula giving the number of distinct orbits by
The number of distinct orbits N_{o} is the average of the elements fixed by the orbits.
For a H, the set of all g ∈ G such that gHg^{−1} = H is called the of H in G and is denoted N_{G}(H). For a of G, N_{G}(N) = G. Clearly, N_{G}(H) is the largest of G containing H as a We also call = . It has been proved, see for example Milne (2003), that N_{0}(H) is the largest contained in H.
Let us now consider two interesting examples of the class equation (which will be used in the present study).
1. With X = {H, H ≤ G} the set of all the subgroups of G, then G acts on X by conjugation: for any g ∈ G, g.H = gHg^{−1}. One may check that for this action the is = N_{G}(H). The superscript c means `by conjugation'. For a fixed H of G, the number of distinct subgroups that are conjugate is given by the class equation applied to X = {gHg^{−1}, g ∈ G}, it is G/N_{G}(H).
2. With X = G/H, the partition formed by the left H cosets on G, then:
(a) G acts on G/H by the classical left product: for any g′ ∈ G, g′·gH = g′gH. Note that, for this action, the is = gHg^{−1}, the conjugate of H by g and that in general ≠ unless H G. The class equation applied to G/H is equivalent to Lagrange's formula.
(b) H acts at the left of G/H: for any h ∈ H, h·gH = hgH. Note that, for this operation, the is = H ∩ gHg^{−1}.
Thus, for any H ≤ G and g ∈ G,
and, if H = N G,
Footnotes
^{1}The term `nondisruptive' relies on the assumption that some symmetry elements of the new structure (daughter phase) can be described in the frame of reference of the old one (parent phase). Guymont does not consider these transitions as reconstructive ones but we think they are, according to the definition given by Tolédano & Dmitriev (1996).
^{2}Clearly, if we denote by f and g the linear functions associated respectively with the matrices F and G expressed in the same coordinate system B_{0}, their composition (`first g then f') is expressed by the product FG in B_{0}. However, if we denote by f and g the linear functions associated with the matrices F = [B_{0}▹B_{1}] and G = [B_{1}▹B_{2}], it must be noted that these matrices are expressed in different coordinate systems. We can write them in the same system B_{0}; they become F and FGF^{−1}, respectively, and now it is the composition (`first f then g') that is expressed by FG in B_{0}. We conclude that the notation fg (= ) is more appropriate to write the composition of functions associated with transformation matrices.
^{3}Sometimes the term `symmetry operators' is used, but we will keep this term for another mathematical concept, i.e. a class of equivalence of symmetry operations (see §6.2).
^{4}The notation introduced in equation (5) slightly differs from the one used by Janovec & Přívratská (2003): we use g_{ki} instead of Janovec's g_{ik}. Actually, Janovec notes the `switching operation' that transforms the domain state S_{i} into S_{k}: . With our notation (5), we write and the composition rule (6) becomes: if and , then with g_{li} = g_{lk}g_{ki}. The reader may see that this composition is more intuitive than with the Janovec notation.
^{5}One must be very careful not to be confused by the type of action used by the author. For example, Janovec usually uses the conjugate action of on its subgroups H defined by . He uses for and, as detailed in Appendix B, for this action, the of .
^{6}We recall that the product of two sets X and Y is XY = {x_{i} y_{j}, x_{i} ∈ X and y_{j} ∈ Y} and that X^{−1} = {x_{i}^{1}, x_{i} ∈ X}.
^{7}We recall that the product of matrices m_{xy}m_{x} corresponds to the composition of linear functions (first apply m_{x} and then m_{xy}); see footnote 2. It follows that and .
^{8}Janovec & Dvorakova (1989) use the decomposition to define classes of ordered pair of states . Two domain pairs and ( are called `crystallographically equivalent' if ∃ such that and . There is a onetoone correspondence between the class of pair domains and the double coset.
^{9}Weinstein's convention leads to the opposite: t(g) → s(g) which appears confusing. This linguistic ambiguity comes from the fact that Weinstein has chosen the convention of the functional composition (`first apply g and then f'). In this paper, we have preferred the notation fg (`first apply f and then g') the composition of the linear functions f and g associated with transformation matrices F and G (footnote 2).
^{10}This approach generalizes Friedel's law, but still respects Curie's principle (Curie, 1894): the effects (diffraction) are at least as symmetric as the causes (the crystallographic structure).
^{11}In his approach, Sadanaga uses the of the form , where H is the of the (what they called the kernel of the i.e. what we call the isotropy group of the substructure) and T is the hull of the (the set of actions of the i.e. the operators). In Sadanaga's approach, the composite structure must be built from only one as is the case in the of orientational variants. We think that the difference between his M and the of orientational variants with G the of the parent crystal and H a of G is due to the difference in the composition rules.
Acknowledgements
We are very grateful to Professor Michéa for having pointed our attention to the special mathematical concept of
We also thank him for checking some algebraic developments of the paper. We are grateful to Dr Jouneau for his precious advice on crystallography and his help in the Python language. Acknowledgments are also due to Dr Briottet for supporting this work. The notations have been simplified and the bibliography has been improved thanks to the referees' remarks.References
Akbay, B., Aydinol, K. & Bor, A. S. (1994). J. Appl. Cryst. 27, 379–384. CrossRef CAS Web of Science IUCr Journals Google Scholar
Allegrini, P., Giuntoli, M., Grigolini, P. & West, B. J. (2004). Chaos, Solitons Fractals, 20, 11–32. Google Scholar
Bollmann, W. (1970). Crystal Defects and Crystalline Interfaces. Berlin: SpringerVerlag. Google Scholar
Bollmann, W. (1982). Crystal Lattices, Interfaces, Matrices. Published by the author. Google Scholar
Bradley, C. J. & Cracknell, A. P. (1972). The Mathematical Theory of Symmetry in Solids. Oxford: Clarendon Press. Google Scholar
Brandt, W. (1926). Math. Ann. 96, 360–366. CrossRef Google Scholar
Burgers, W.G. (1934). Physica (Utrecht), 1, 561–569. CrossRef CAS Google Scholar
Cahn, J. W. & Kalonji, G. (1981). SolidSolid Phase Transformations, edited by H. I. Aaronson, D. E. Laughlin, R. F. Skerka & C. M. Marvin Wayman, pp. 3–14. Washington: The Metallurgical Society of AIME. Google Scholar
Cayron, C. (2000). PhD thesis No. 2246, EPFL, Lausanne, Switzerland, https://cimewww.epfl.ch/people/cayron/these.html. Google Scholar
Cayron, C. (2006). In preparation for Acta Cryst. A. Google Scholar
Cayron, C., Artaud, B. & Briottet, L. (2006). Submitted to Mater. Charact. Google Scholar
Connes, A. (1990). Géométrie Non Commutative. Paris: Dunod (InterEditions U). Google Scholar
Curie, P. (1894). J. Phys. 3, 393–415. Google Scholar
Dirl, R., Zeiner, P., Davies, B. L. & Janovec, V. (1997). Acta Cryst. A53, 456–466. CrossRef CAS Web of Science IUCr Journals Google Scholar
DornbergerSchiff, K. (1959). Acta Cryst. 12, 246. CrossRef IUCr Journals Web of Science Google Scholar
DornbergerSchiff, K. & GrellNiemann, H. (1961). Acta Cryst. 14, 167–177. CrossRef IUCr Journals Web of Science Google Scholar
Friedel, G. (1904). Étude sur les Groupements Cristallins. Extrait du Bulletin de la Société de l'Industrie Minérale, Quatrième Série, Tomes III et IV. SaintÉtienne: Société de l'Imprimerie Thèolier J. Thomas et C. Google Scholar
Gey, N. & Humbert, M. (2003). J. Mater. Sci. 38, 1289–1294. Web of Science CrossRef CAS Google Scholar
Gottstein, G. (1984). Acta Metall. 32, 1117–1138. CrossRef CAS Web of Science Google Scholar
Gourgues, A. F., Flower, H. M. & Lindley, T. C. (2000). Mater. Sci. Tech. 16, 26–40. Web of Science CrossRef CAS Google Scholar
Gratias, D. & Portier, R. (1982). J. Phys. 43, C6, 15–24. Google Scholar
Grimmer, H., Bollmann, W. & Warrington, D. H. (1974). Acta Cryst. A30, 197–207. CrossRef IUCr Journals Web of Science Google Scholar
Guymont, M. (1981). Phys. Rev. B, 24, 2647–2655. CrossRef CAS Web of Science Google Scholar
Hahn, Th. & Klapper, H. (2003). International Tables for Crystallography, Vol. D, edited by A. Authier, Section 3.3, pp. 393–448. Dordrecht: Kluwer Academic Publishers. Google Scholar
Humbert, M. & Gey, N. (2002). J. Appl. Cryst. 35, 401–405. Web of Science CrossRef CAS IUCr Journals Google Scholar
Humbert, M., Wagner, F. & Esling, C. (1992). J. Appl. Cryst. 25, 724–730. CrossRef Web of Science IUCr Journals Google Scholar
Humbert, M., Wagner, F., Moustahfid, H. & Esling, C. (1995). J. Appl. Cryst. 28, 571–576. CrossRef CAS Web of Science IUCr Journals Google Scholar
Janovec, V. (1972). Csech. J. Phys. B22, 974–994. Google Scholar
Janovec, V. (1976). Ferroelectrics, 12, 43–53. CrossRef Google Scholar
Janovec, V. & Dvorakova, E. (1989). Acta Cryst. A45, 801–802. CrossRef Web of Science IUCr Journals Google Scholar
Janovec, V., Hahn, Th. & Klapper, H. (2003). International Tables for Crystallography, Vol. D, edited by A. Authier, Section 3.2, pp. 377–391. Dordrecht: Kluwer Academic Publishers. Google Scholar
Janovec, V. & Přívratská, J. (2003). International Tables for Crystallography, Vol. D, edited by A. Authier, Section 3.4, pp. 449–505. Dordrecht: Kluwer Academic Publishers. Google Scholar
Kalonji, G. (1985). J. Phys. 46, C4, 249–256. Google Scholar
Kargapolov, M. & Merzliakov, I. (1985). Eléments de la Théorie des Groupes. Moscow: Mir. Google Scholar
Kitahara, H., Ueji, R., Ueda, M., Tsuji, N. & Minamino, Y. (2005). Mater. Charact. 54, 378–386. Web of Science CrossRef CAS Google Scholar
Kocks, U. F., Tomé, C. N. & Wenk, H. R. (1998). Texture and Anisotropy. edited by U. F. Kocks, pp. 45–102. Cambridge University Press. Google Scholar
Kurdjumov, G. & Sachs, G. (1930). Z. Phys. 64, 325–343. CrossRef Google Scholar
Landau, L. D. (1937a). Phys. Z. Sowjun. 11, 26–47. CAS Google Scholar
Landau, L. D. (1937b). Phys. Z. Sowjun. 11, 545–555. CAS Google Scholar
Marteli, A. (2004). Python en Concentré. Paris: O'Reilly. Google Scholar
Milne, J. S. (2003). Group Theory, https://www.jmilne.org/math/. Google Scholar
Morito, S., Tanaka, H., Konishi, R., Furuhara, T. & Maki, T. (2003). Acta Mater. 51, 1789–1799. Web of Science CrossRef CAS Google Scholar
Nishiyama, Z. (1934). Sci. Rep. Tohoku Imp. Univ. 23, 637–644. CAS Google Scholar
Prigogine, I., Nicolis, G. & Babloyantz, A. (1972). Phys. Today, 25, 23–28. CrossRef CAS Google Scholar
Sadanaga, R. (1978). Rec. Prog. Nat. Sci. Jpn, 3, 143–151. Google Scholar
Sadanaga, R. & Ohsumi, K. (1979). Acta Cryst. A35, 115–122. CrossRef CAS IUCr Journals Web of Science Google Scholar
Standford, N. & Bate, P. S. (2005). Acta Mater. 53, 859–867. Web of Science CrossRef Google Scholar
Stewart, I. (2004). Nature (London), 427, 601–604. CrossRef PubMed CAS Google Scholar
Tolédano, J. C. & Tolédano, P. (1987). The Landau Theory of Phase Transition. Singapore: World Scientific. Google Scholar
Tolédano, P. & Dmitriev, V. (1996). Reconstructive Phase Transitions. Singapore: World Scientific. Google Scholar
Van Tendeloo, G. & Amelinckx, S. (1974). Acta Cryst. A30, 431–440. CrossRef CAS IUCr Journals Web of Science Google Scholar
Wadhawan, V. K. (1997). Acta Cryst. A53, 546–555. CrossRef CAS Web of Science IUCr Journals Google Scholar
Warrington, D. H. & Bufalini, P. (1971). Scr. Metall. 5, 771–776. CrossRef CAS Web of Science Google Scholar
Wassermann, G. (1933). Arch Eisenhüttenwes. 16, 149–155. Google Scholar
Weinstein, A. (1996). Am. Math. Soc. 43, 744–752. Google Scholar
Yamamoto, A. & Ishihara, K. N. (1988). Acta Cryst. A44, 707–714. CrossRef CAS Web of Science IUCr Journals Google Scholar
This is an openaccess article distributed under the terms of the Creative Commons Attribution (CCBY) Licence, which permits unrestricted use, distribution, and reproduction in any medium, provided the original authors and source are cited.