Math. Proc. Camb. Phil. Soc. (1996), 120, 547 5 4 7 Printed in Great Britain Singularity theory and equivariant bifurcation problems with parameter symmetry BY JACQUES-ELIE FURTER, Department of Mathematics and Statistics, Brunei University, Uxbridge UBS 3PM ANGELA MARIA SITTA, Departamento de Matemdtica, IBILCE-UNESP, Brazil AND IAN STEWART* Mathematics Institute, University of Warwick, Coventry CV4 7AL. (Received 26 January 1995; revised 30 August 1995) 1. Introduction The study of equivariant bifurcation problems via singularity theory (Golubitsky and Schaeffer[8], Golubitsky, Stewart and Schaeffer[9]) has been mainly concerned with models exhibiting spontaneous symmetry-breaking. The solutions of such bifurcation problems lose symmetry as the parameters vary, but the equations that they satisfy retain the same symmetry throughout. There is another category of equivariant problems where the bifurcation equations possess less symmetry when some parameters are non-zero; this is called forced or induced symmetry breaking. In [16], Vanderbauwhede, using classical techniques in bifurcation theory, has described some of what might happen in that case. Although many of the results are fairly general they have so far mainly been used for forced symmetry-breaking of a full orbit of a continuous Lie group; for instance in the case of periodic forcing of autonomous systems or, in mechanics, in the case of rigid bodies. See, for example, Chillingworth[2]. In this work we apply techniques from singularity theory to multiparameter bifurcation problems f(x, A) = 0 with symmetry on both the state variables and on the bifurcation parameters. That is, we consider a map-germ/: 0?" x Uk->IRn that yf{x,X) VyeF, where F is a compact Lie group acting linearly on Un and on K*. (We will often omit the word 'germ' and refer to maps, but we work throughout with germs, so all results are valid locally near the origin.) This means that we study forced symmetry-breaking when some constraints are imposed via an action of the symmetry group F on the parameters. Let S c F b e the isotropy subgroup of A. In terms of spontaneous symmetry-breaking, the full equation is S-equivariant, since when A 6 Fix (F) the map f(x, A) is F-equivariant in x. In the language of singularity theory, without additional conditions, we would consider the recognition problem for F-equivariant problems and unfold them in the S-theory. In our case we do not have a 'full' unfolding in the S-theory because F remains as a residue of symmetry when we enlarge the space to encompass the * Research supported in part by a grant from the SERC (now EPSRC). 548 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART parameters. In that sense the problem is more constrained than the problems considered in Vanderbauwhede[16] and Chillingworth[2]. In [8] and [9] Golubitsky, Stewart and Schaeffer studied, via singularity theory, one-parameter bifurcation problems with symmetry only on the state variables. Peters [12] classified bifurcation problems with a one-dimensional state variable and two bifurcation parameters, and carried out an extension to multiparameter bifurcations with a symmetry group acting independently on both the state variables and the bifurcation parameters - that is, by a 'diagonal' action (x, A)h+(yx,yA). (It is convenient to use the term ' diagonal' even though the two actions are in general different, to indicate that the state space and the parameter space are invariant under the group action.) Simultaneously, in his Ph.D. thesis, Lari-Lavassani[10] wrote about multiparameter bifurcation problems with symmetry on the state variables, following [9] and some algebraic results introduced by Damon in [3]. In Section 2 we set up the general machinery needed to study F-equivariant multiparameter bifurcation problems via singularity theory, for a diagonal linear action of a compact Lie group F on the state variables a; and on the multiparameter A. We define an equivalence relation for such bifurcation problems using changes of coordinates (contact equivalence) preserving the bifurcation structure (A-slices) and the symmetry (F-action) of the problem. Indeed, two maps f,g are said to be equivalent if there exist T,X,A such that g(x,A) = T(x,A)f(X(x,A),A(A)), where T(x, A) is an invertible matrix and (x, A)^*(X(x, A), A(A)) is a diffeomorphism germ. Both T and (X, A) must be symmetry- and orientation-preserving, that is, T(yx, yX) y = yT(x, A),X(yx, yA) = yX(x, A), A(yA) = yA(A) and T(0,OJ.X^O,0), AA(0) are in the connected components of their respective identity operators. We show that this context fits into the general framework introduced by Damon [3]. We can derive the main results, the Finite Determinacy and Unfolding theorems, directly or from their abstract equivalent in [3]. The finite co-dimension of the 'extended tangent space' of such/implies both that/is contact equivalent to a finite segment of its Taylor series expansion (finite determinacy) and that any perturbation of/ can be represented in a special perturbation F with k parameters, where k = cod/is the co-dimension of/ (universal unfolding). One can also improve some aspects of the results in a similar spirit to Lari-Lavassani [10]. Observe that the theory introduced here extends to cases when the groups acting on the state and the parameter spaces are different (because this reduces to a suitable action of the direct product of the two groups). In Section 3 we describe an extended version of the theory. We classify, up to topological co-dimension 1, bifurcation problems with two state variables and two bifurcation parameters that are equivariant for a particular action of the dihedral group D4. In complex coordinates z, A, this action is defined by f(z,A)=f(z,A) and f(iz,A) = if(z,A). This action is motivated by a mathematical model describing the buckling of a square plate when a pair of forces acts on the four edges of the plate, see Section 5-1. We use the classical framework of singularity theory to find the appropriate tangent spaces and higher order terms relative to our chosen equivalence relation, which Singularity theory 549 leads to a list of normal forms. We give a corrected version of the generic normal form derived in Peters [12] and extend the classification to all map-germs of topological co-dimension one. The main challenge in carrying out the classification for multiparameter problems comes from the calculation of the tangent spaces. They are not in general finitely generated modules over a unique ring, but over a system of rings in the sense of Damon [3]. For AeIR this is not important because we can first disregard some contributions to the tangent space and consider the remaining 'restricted tangent space' (which is now a finitely generated module). This has finite co-dimension if and only if the total module does. Then we can reintroduce the missing part of the tangent space to compute the final quotient. However, for a multi-dimensional parameter A this approach is no longer possible, because the restricted tangent space is in general of infinite co-dimension. We must therefore make more careful use of the algebraic structure. In doing so, a different but very natural approach appears: the path formulation. This views the bifurcation equation f(x,A) = 0 as a multiparameter 'path', parametrized by A, through a universal unfolding off(x, 0). In a sister paper, [4], we analyse that path formulation in an algebraic context, with applications to a more extended classification of the D4-equivariant bifurcation problems from Section 3. The path formulation was introduced in a geometrical context by Golubitsky and Schaeffer[7, 8]. They relate bifurcation problems in one state variable without symmetry to A-paths in the parameter space of the miniversal unfolding, in the sense of catastrophe theory, of the function germ xk, k ^ 2. However, at that time the techniques of singularity theory were not powerful enough to handle the full power of the path formulation efficiently - not even in theory. This is why contact equivalence with distinguished parameters was developed in [7]. However, current techniques, strengthened by the use of symbolic algebra software, have made the path formulation a much more feasible method, for example see Mond and Montaldi[ll]. This section ends with remarks on the relationship between the classification by Golubitsky and Roberts [6] of D4-equivariant problems with one parameter (upon which the action is trivial) and our classification. We also comment on the use of our classification to tackle gradient D4-equivariant bifurcation problems. Some bifurcation problems, like the buckling of elastic shells in Section 5, have a natural gradient structure, which acts as an additional constraint. Even though contact equivalence does not preserve the set of gradients S^ A, it still induces an equivalence relation on S^ K. Moreover, the perturbation (unfolding) theory extends to the gradient case: see Bridges and Furter [1] for general theoretical results on such questions. Section 4 deals with the geometrical description of the normal forms of D4- equivariant bifurcation problems in terms of their bifurcation diagrams. First, we describe for a general function the type of solutions we can expect (via their isotropy subgroup) with their (linearized) stability. In the multiparameter situation it is important to understand what structure is preserved by contact equivalence. In general, only the relative position of open regions in parameter space, where the structure of the zero-set does not change in any important qualitative manner, is preserved. Usually, specific families of one-dimensional slices have no invariant meaning. However, in our situation the symmetry on parameters implies that the 550 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART Fig . 1. Bifurcat ion d i ag ram of the generic normal form when e5 = {,m> 1. Lines d rawn in the form - x - x - x - indicate branches t h a t are always uns table . F ig . 2. Bifurcat ion d i ag ram of the generic normal form when e5 = — l,m > 1. axis A2 = 0 is invariant under contact equivalence, so the structure in each half-plane is preserved. Moreover, we can show that the generic normal form (Io in Theorem 3-2-2) is also generic for the stronger contact equivalence respecting Aj-slices at A2 = constant, that is, with A(A1;A2) = (A^Aj, A2), A(A2)). The proof of that fact is easier in the path formulation, so we refer to the proof in [4]. Hence, in that case, the A2- succession of Aj-slices has a perfectly good invariant meaning. We analyse the generic normal form in detail. We also draw some schematic diagrams to show the behaviour of the solutions of these bifurcation problems, which show that the trivial solution loses stability to one of the competing single modes. Then we distinguish cases where the solutions without any symmetry (mixed-mode) are stable (Figs 2, 4) or unstable (Figs 1, 3). There is an exchange of stability between the pure and mixed modes: they are never both stable for the same parameter values. Singularity theory 551 Fig. 3. Solution sets for fixed A for the generic normal form when e5 = \,m > 1. In Section 5 we briefly discuss two applications of the theory for D4-equivariant problems. We consider the buckling of thin elastic shells with a square base under lateral compression, and the bifurcation of steady states of symmetric rings of interacting cells arranged in a square. We concentrate on showing that those two problems can lead to situations where our theory applies. We show that the linearization at the bifurcation has two competing interchangeable modes and the full equations possess the appropriate D4-symmetry. We refer to [4] for an analysis of the nonlinear equations. 2. General theory In this section we present a general theory of unfoldings, finite determinacy and the recognition problem for multiparameter equivariant bifurcation problems when a compact Lie group acts diagonally on both the state variables and the bifurcation parameters. Our discussion of this question is based on results of Damon [3] and Lari- Lavassani[10]. 2-1. Notation and definitions The state variable is x = (xl...xn)eWl and the bifurcation parameter A = (A,... A,)e W. The derivatives are denoted by subscripts, fx for df/dx,..., and the superscript ° denotes the value of any function at the origin, 552 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART Fig. 4. Solut ion sets for fixed A for the generic normal form when e5 = — l , m > 1. Let (?x denote the ring of smooth germs/: (Rn,0)->R and Jl' x its maximal ideal. For ?/elRm let gxy denote the $\x A,fl> u(a:,A,/!)'UM,/J)- Perturbations of any fe S\x A) are described by ̂ -parameter unfoldings of/, which are map-germsFeS\x x ^fi = (/?x .../?fc), such that ^(a:, A,0) = ^ , A). We denote by 554 JACQTTES-ELIE F U R T E R , ANGELA MARIA SITTA AND I A N STEWART J f £ un(k) the group of F-equivalences for unfoldings with k parameters. It is a natural extension of Cff\ in the following sense: •*T«»(*) = HT,X, A, O ) e M [ x ^ x 0 ^ * ©&,% x •*,,,» I (^.X, A) is a & parameter unfolding of an element of JTA and is a diffeomorphism germ}. The action of ^\un(k) on Fei\xX^ is denned by (T,X, A, O) .F(x, A, p) = T(x, A, fi)F(X{x, A, /?), A(A, /?), *(/?)). We say that F,Gt$\x x ^ are JfA un(k)-equivalent if they belong to the same For later purposes, we distinguish two subgroups which together generate „„(&). The first, denoted by X~? un(k), is given by = 1^; for the second, CfT eq{k), the function (T,X, A) is an unfolding of the identity of C/f\. 2-2-3. Tangent spaces. Associated with J>TA we can define several different tangent spaces to / . The fundamental ones we need in the rest of this paper are the following. Let / e $\x A). Then the extended tangent space of f is Observe that this has only the structure of a ^[-module. The extended normal space to / is defined by and the F-co-dimension oifeS\xX) is To simplify some algebraic manipulations it is also useful to introduce the following more restrictive equivalence which brings in tangent spaces that are $\X,X)- modules. We say that two germs / , g € Jiv (x A) are strongly 3f\-equivalent if they are J f ^-equivalent with A(A) = A. For our purposes the appropriate tangent space to / associated with strong equivalence is Observe that 3%STv{j) is a finitely generated S\x A)-module. When AeU, its use is fundamental because the finite co-dimension of / is equivalent to the finite co- dimension of &^~r(f). This is not the case with a multidimensional A; nevertheless, the S\x A)-module structure makes 0tSTv{f) of some use in explicit computations. For the recognition problem (Section 2-5-2) we will encounter yet another type of tangent space associated to the subgroup of unipotent equivalence. 2-3. Unfolding theory Let FeS\xX^ and GeS\x x a)) be two unfoldings of fsd>^x X) with k and r parameters respectively. We say that G maps into or factors through F if there exist TeM^a),XeQlA X(x, A, 0) = x and A(A, 0) = A such that G{X,A,OL) = T(x,A,a)F(X{x,A,OL),A(A,a.),A{*)). Singularity theory 555 The unfolding F is versal if any unfolding G of/ maps into F. IfF is versal and has minimal number of parameters, it is called a miniversal unfolding of/. The miniversal unfoldings of /are unique up to JTA un(£)-equivalence (cf. Theorem 2-3-3). From Damon's general theory ([3], cf. Section 2-6 here), or directly (Sitta[14]), we have the usual results from singularity theory, as follows. THEOREM 2-3-1. (Unfolding Theorem). LetfeS\x X) andFeS\x K a) be an unfolding of f with k parameters, a = (ax... ak). Then 1. The unfolding F is versal if and only if 2. Two versal unfoldings of a germ in S\x A) are equivalent as unfoldings if and only if they have the same number of unfolding parameters. COROLLARY 2-3-2. Let geS\xX) be of finite co-dimension and let W <= $\xj<) be a complement of Jf\(g) as a vector space, S\xX) = ^r e(g) © W. Let {pJjLj be a basis for W. Then k G(x, A, a) = g(x, A) + £ a.jP}(x, A) j-i is a miniversal unfolding of g. From this corollary we get: THEOREM 2-3-3. (Uniqueness of Miniversal Unfolding). If f and ge$\x A) are two JfA- equivalent germs of finite co-dimension and F and GeS\x A a), with a = (ax... ak), are two miniversal unfoldings of f and g respectively, thenF and G are $f\ un(k)-equivalent. In this case we say that F,G are universal unfoldings. 2-4. Determinacy For any mapping/we let jk(f) denote the Taylor polynomial of order k, or &-jet, of/. A germ fe $ \x A) is k-$C\-determined if every germ gs$\x ^ with jk(g) =jk(f) is JfJ-equivalent to/ . A germ is finitely ^{-determined if it is k-Z£ ̂ -determined for some integer k. As usual, there is a close relationship between being finitely determined and being of finite co-dimension. The first theorem follows from the general theory. THEOREM 2-4-1. {Finite Determinacy). LetfeS\x A). Then f is finitely tf\-determined if and only if codr (/) is finite. The following theorem gives a necessary condition for a germ p e S\x A) to be in the set of higher order terms of/; that is, those terms that can be removed (by an equivalence) from the Taylor expansion of/. See theorem 2-2 of Gaffney[5]. THEOREM 2-4-2. Let feS\x X) and let p be any germ in S\x A). Then 1. lfm$~T(f+tp) = @3Tr(f'),Vte[0,1], thenf+p is strongly ^{-equivalent to f. 2. / / cod r ( / ) is finite and 3Tr e(f+tp) = ^( / ) ,VO(/ ) = T.fo (X, A). Say that a submodule M cz S\xX) is intrinsic if O(/)eMfor all feM and for all d>e JT£. Let V c S\X{f) is an intrinsic submodule of'S\x A). 2-5-2. Unipotent Jf {-equivalences. Another useful subgroup of Jf J is the group of (restricted) unipotent equivalences. The kernel of the projection map n sending o n t o (T°,Xx,Kl) is m\ = {(T,X,A)e*r\T° = Im,Xx = In, AS = I,}. It is a normal subgroup of tf\ consisting of unipotent diffeomorphisms and it is called the subgroup of unipotent Y-equivalences. Its associated tangent space to fe 3TW(f) = {Tf+fxX+f,A\Tem\x,X),Xe®\x%,Ae®l-°, T° = 0,X°x = 0, A° = 0}. For computational reasons we also define a restricted unipotent tangent space at/ , given by = {Tf+fxX\ (T,X,lf)eTeMr (x»,XeQrx°A), T° = 0,X°x = 0}. By theorem 1-17 of Gaffney [5] we have: PROPOSITION 2-5-3. LetfsS\xK) be of finite V-co-dimension. Then0>(f) => In COROLLARY 2-5-4. Let pelntr&~tftr{f), then f+p is ^[-equivalent to f. 2-6. Geometric subgroups and proofs 2-6-1. Definitions. We recall some basic definitions and some results due to Damon [3] which are necessary to establish the Unfolding and Finite Determinacy Theorems. Let yeW and y'eUm. A DA-algebra A [differentiable algebra) consists of an U- Singularity theory 557 algebra A and a surjective algebra homomorphism i/r: 8y^>-A. These algebras are local rings with maximal ideals JtA. If ^ : $y.^-B defines another DA-algebra, then a homomorphism of DA-algebras Sy, with a = g* for some g: (Rm,0)->(IRn,0) and aof = (j)od. For example, Sv (x A) and S\ are DA-algebras. Moreover, M\x A) and Jt\ are Jacobson ideals (that is, l + / i s always invertible for f&Jl). Let (3>,<) be a finite partially ordered set of indices. A system of DA-algebras (associated to) Si consists of a set of DA-algebras {RJaeS together with homo- morphisms of DA-algebras ay, for a ^ /? ̂ y, and (j)^ = id. A system of ideals {^fx} of {R^ consists of ideals ^ c Ra so that 4>ap{ IR9+1. Thus we can compute the extended spaces ^e^un{q) and ^ ^ „(' differ from the tangent spaces 2T^ and 2T0F by 'constant vector fields'. Exponential maps. We first define a group ^eq(q) of equivalences of unfoldings with q parameters consisting of those elements oi defined in a neighbourhood of the origin of W. See §2-2-2 of Damon[3]. Again, we can consider 'unfoldings' of elements of Geg{q) to lie in Geq{q+l) via a one-parameter unfolding of the identity of Geq(q). Computing the extended tangent space at the identity, we have: The restriction of the exponential map for J#~ induces a map where d/dt(j>t = E,.(pt and 0 = id. Filtration preservation. The group %n{q) preserves the filtration {{Jtx}. ^un{q)} on ^un{q) and it induces an action on the quotient ®" (a) Singularity theory 559 We now show that ^ = $C\ is a geometric subgroup, that is, it satisfies the four conditions listed above. To do so, take SF = S\x X), ^un(q) = &\X,KFI and %n(q) = {(T,X,A)eJf[un(q)l = y Naturality is clear. The tangent space condition follows from Section 2-2-3 and preliminary results of previous sections. To verify the exponential map condition we need to define %g(q) = i(T,X, A)eJfr Aun(q)\ (T,X, A) is an unfolding of the identity in JiT{}. Again, we omit the details because everything extends readily to our context. Finally, the filtration preservation condition holds because it already holds for the non-equivariant situation, and we consider a linear F-action. The Unfolding and Finite Determinacy theorems are then a rewording in our situation of the corresponding results in the general theory of [3]. 3. Example: T)4-equivariant problems A I'-bifurcation problem is a germ Fe$[xjy) such that / " = 0 and f° = 0. In this section our aim is to solve the recognition problem for D4-bifurcation problems. We identify IR2 with C, and in complex notation the action of D4 on C x C is generated by where K denotes the flipK(Z) = z and fi denotes the action fi(z) = (e*%).z = iz. 3-1. Preliminaries Define the D4-invariant functions 8 = - u4 = A2. Let u = (u1,u2,u3,ui,ub) with wx = N, u2 = A, u3 = Alt u4 = A\, us = SA2 and v— (AJ,M4). Note that there is a unique relation U^ — 1^2 1t>4. Since it is much more convenient to work with free modules we take u = (u1, u2, u3, M4). From Peters [12] we have the following results. PROPOSITION 3-1-1. 1. S^A) is freely generated by z, Sz, A2z and SA2z as an Su- module. 2. M°'A) is freely generated as a Su-module by the following linear maps (in complex notation): A A A A w, S7(z, A) w = —iSanv, A /\ A A S2(z, A)w = A2 id, Sr(z, A) iv = 8A2 w, S6(z, A) w = — z'A2 cow, S8(z, A)w — iSA2 a>w. 560 JACQUES-ELIE FTJRTER, ANGELA MARIA SITTA AND IAN STEWART Table 1. Normal forms. Here top-cod denotes the topological D4-co-dimension and cod the smooth D^-co-dimension Case Normal form top-cod cod Il I2 [mN+nNu,, + eoA1,eb,Al + eaui + a.,O] II III IV 0 1 1 1 1 1 1 1 2 3 1 1 2 2 where u> = i/^z^ — z2). 3. c?°4 = v*Sv and $\^* is freely generated as an S^-module by (1,0) and (0, A2). Thus, fe$fz lx) may be written as f{z,X) = p(u) z + q(u) Sz + r(u) A2z + s(u) SA2z and we identify / with [p,q,r, s]. Observe also that since n = m we have Oĵ vft = •M%\\y 3-2. Classification theorems THEOREM 3-2-1. (Generic normal form). Let f= [mN+egA^e^, 1,0], where e0 = + 1 , e5 = + 1 and ra 4= ± 1,0. Then f has D^co-dimension 1 (topological co- dimension 0,misa modal parameter) and all D4-bifurcation problems h = [p, q, r, s] that satisfy p° = 0 and p°N.p°x -r°.(p°N + q°) 4= 0 are JT°4-equivalent to f with the coefficients e0, e5 and m given by eQ = sgnp\ , e5 = sgnq" and m = p°N/\q°\. We now solve the classification problem for the first degeneracies of the generic normal form given in Theorem 3-2-1. Let Ax,y(p,q)=po xq o y-p o yq°x. THEOREM 3-2-2. (Classification Theorem). A germf = [p, q, r, s] e $®£ A) is of topological co-dimension 0 or 1 if and only if it belongs to the list shown in Table 1. Moreover, f is J#~\°4-equivalent to the given normal form in each case below if and only if it satisfies the correspondent sets of defining and non-degeneracy conditions listed below in Section 3-2-1. In all cases p° = 0. The parameter a is the unfolding parameter and m,m1,m2,n are moduli. 3-2-1. Additional Information. Case I1. Defining condition: p°x = 0. Non-degeneracy conditions: p%. p0^ .q°.r°. (p% ± q°) (p°NXi q° - q«yN) 4= 0. Case I2. Defining condition: r° = 0. Non-degeneracy conditions: q°-P°l-{p%±q°)-^vUt(p,r).^3.^4 4= 0. Case II. Defining condition: p% = 0. Non-degeneracy conditions: p°NN.p° .q°.r° 4= 0. Singularity theory 561 Case III. Denning condition: p°N = —q". Non-degeneracy conditions: p\ .q°.r°.^ 4= 0. Case IV. Denning condition: p°N = q. Non-degeneracy conditions: p°N-P° .r° .{p°NN + 2pl — 2q°N).E,i 4= 0. Case V. Denning condition: q° = 0. Non-degeneracy conditions: p°N-plrr°.AN A(p,q).AN Ai(p,q) 4= 0. Coefficients e0 = , e1 = sgnp°N, e2 = sgnp°NN, e3 = ebsgn(p°NAiq° -q°Aip°N), e 4 = S e5 = sgnq0, e. = sgn£1; e8 = eosgn- n2 «2i -P%r°N) + r°t. {pi - q°*), m = \q°\' e, ra, = \p°).\AN ^p,q)\i' 2\q°\ n = • i. (S°q°-P% r%) + f°Xi. (p% - q°') 33. Proofs 3-3-1. Tangent spaces. The D4-extended tangent space at feSf/A), denned in Section 2, is given by 562 JACQTJES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART where g2 = [u4r,u4s,p,q], 9a = [Aq,p,As,r], g4 = [Np — Aq,p—Nq, —Nr + As, —r+Ns], gh = [Ap—NAq,Np — Aq,NAs — Ar,As—Nr], g6 = [—Nu4r + Au4s, — u4r+Nu4s,Np — Aq,p—Nq], g7 = [Au4s,u4r,Aq,p], g8 = [ — Au4 r +NAu4 s, —Nu4 r + Au4 s, Ap —NAq, Np — Aq], g9 = [2NpN + , jy99'^99^i9^'U't99>9io>i We define the T>4-topological codimension of a D4-bifurcation problem as its D4-co- dimension minus the number of modal parameters. A modal parameter changes the differentiable type of the singularity but not its topological type, see Golubitsky and Schaeffer[8]. 3'3-2. Change of coordinates. Let e=(T,X, A) be a J^°4-equivalence where X — [a,b.c,d] with a,b,c,deSu and a" > 0, A = (A1; A2 A2) with A1eJ?°t, (AJ" > 0 and A2e<^P4 and T = 2f_,-4*5, with AteSu and A\ > 0. To simplify the notation, we define the following expressions: Nt = a2N- 2abA + b2AN+ c2Nu4 - 2cdAu4 + d2NAu4, N2 = -2ac + 2adN+2bcN-2bdA, (3-1) D2 = - 2acN + 2adA + 2bcA- 2bdNA, Al=D\A+D\u4 and ^ = 20^ Singularity theory 563 Using (3-3) below we find that = AjO A = = uiK\, (3-2) For heSu, we define h = h(N,A,A1,u4). Let/e + + + ^ N iA). (3-3) (ii) / " = (T,I8>I2)./ ' = [p",q",r",s"], where (A2+A6N+A8A)u4s', + (A3+A4+A5N)As', s" = (A1+A4N+AiA)s'. (3-4) 3-3'3. Intrinsic ideals and submodules. Recall that an intrinsic submodule of <̂ °*A) is a submodule that is invariant under the action of JTf4 and an intrinsic ideal in ^U!A) 'S invariant under the group of coordinate changes (x, A) i-> (X(x, A), A(A)) where X and A satisfy the conditions of Section 3-3-2. Clearly, (^/?°4 A))* is intrinsic for all k. Moreover, it follows from (3-2) that sums and products of the intrinsic ideals (A^w,,) and }. LEMMA 3-3-1. [£,f2>&> A] ^s ̂ '-intrinsic if and only if / 4 c / 4 c 564 J A C Q U E S - E L I E F U R T E R , ANGELA MARIA SITTA AND I A N STEWART Proof. This follows by straightforward calculations using (3-4). I Now, define the following ideals , A, We verify by induction on k ^ 1 that 3-3-4. Proof of Theorem 3-2-1. The proof of Theorem 3-2-1 is a consequence of the following lemma. LEMMA 3-3-2. Let f= [mN+e0A1,ei, 1,0] as defined above. Then the D^-unipotent tangent space is intrinsic and is given by Proof. A calculation shows that, since m =f= +1,0, I t follows that is generated by [N, 0,0,0], [1,0,0,0], [0,1,0,0] and [0,0,1,0] as a real vector space. Let R = {Jt\ + , Jtu, Jfu, Su\ We will show that R = (i) &~<%D*(f) c R From the Preparation Theorem, is generated by [N, 0,0,0], [1,0,0,0], [0,1,0,0] and [0,0,1,0] as an £{Xj ^-module. Therefore, (j>eR can be represented as 4> = fJN, 0,0,0] + 02[1,0,0,0] + &[0,1,0,0] + &[0,0,1,0] + $, where {dUi <=• ^^uj a n d ^e^^r%D'(f). Since M2T°ll°^f)eR, this implies that [^1N+4>2,^3,(pi,0]eR and so 0i>03=04e<^o]>, (3-6) showing that 3~%u'{f) c R. (ii) « Stejo 1. Consider the following set of generators obtained from ^^"^D*(/) modulo (̂A,,«4> • <[<*?> 0.0,0], |>4, 0,0,0], [0,0, A1; 0], [0,0, uA, 0]>, provided that TO #= + 1,0. ^ = [mATA+ 6 0 ^ , 6 5 A, A,0], h'e = [eB A, TOZV+ e0 Ax, 0 , 1 ] , h'7 = [0,(m-eb)N+e0Av -N,0], K = [(m-e5)iVA + eoAA1,TOiV2 + eoiVA1-e5A, - A , -N], h'lo = [0, « 4 , e5 A, ?7iiV+ e0 A J , A'n = [ —AM 4 J —iVit4,(m —e5)iVA + e0AA1,?wiV2 + e0iVA1 —e5A], h'12 = [e0NA1,0,N,0], Ai, = [e o AA 1 ,0 ,A ,0 ] , A'15 = [0,0,e6JV, TO], Step 2. Le t / be the S(X u ,-module generated by {h^... h'16, Ah'12, fch'^Nh^Nh'^ and let S be the ,,f) = S(K «4) • o, o, o], [0, \lt o, o], [0, «4I o, o]>. Therefore, that is R c ST%^{j) and, from (i) and (ii), , tt4>, Jtu, Jlu, Su\ 566 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART Proof of Theorem 3-2-1. From Lemma 3-3-2, y%u*{J) = \_J and &?*(/) = 3~WD'(f) + M.{g1,g9,gn,A1g13,gu}- Hence 3T?l{f) = £r®v,(f) + U.{[mN,e5,0,0], [1,0,0,0], [A^ 0,0,0], [0,0,1,0]}, so that Sfyx) = 3r?'(f) + M.{[N,0,0,0]}. Therefore,/ has D4-co-dimension 1 and topological codimension 0 since TO is a modal parameter. From Lemma 3-3-1 and Proposition 2-5-2, 5"<^D<(/) = In t r^^ D <( / ) c P(/). Following the classical framework, we make explicit changes of coordinates modulo ^~^ D l ( / ) , which is contained in the set of higher order terms of/. For h = \p,q,r,s]eSfz*A), let h = h modulo terms in F%D*{f). Then, and it is a routine calculation to show that h can be reduced into the normal form / ^es, 1,0] if p° = 0 and p°N.p0 Ai.r° .(p%±q°) # 0 , with a n d e5 = 3-3-5. Pre-normal form. Before proving the Classification Theorem, we present some preliminary results to obtain a pre-normal form for D4-bifurcation problems / = [p, q, r, s] when q° 4= 0. By this we mean a simplified form for the germ, which goes part way towards the final normal form that will be derived and paves the way for obtaining it. These results are very useful for the calculation of higher co- dimensions cases and can be readily generalized to other groups. The ^ '-restricted tangent space and the ^--unipotent restricted tangent space at /denoted, respectively by &&~%(J) and ^ST%'€{f) are A)} = , where the {g^*^ are previously defined. The following proposition is the "^^-equivalence version of theorem 1-3 ([9], p. 168). PROPOSITION 3-3-3. Letfed>fz*A) andp any germ in S^X). Suppose that for all te[0,1]. Thenf+tp is Jff'-equivalent to f for all te[0,1]. Proof. The proof is similar to the proof of theorem 1-3 mentioned above and it is done in detail in ([9], §2*, p. 172). The basic idea is to prove that the argument in Proposition 3-3-3 is true in a neighbourhood of t = t. for all te [0,1]. The result follows from compactness and connectedness of the interval. I Singularity theory 567 PROPOSITION 3-3-4. Let fbe a D ̂ bifurcation problem andpe£fz*A). Proof. Hpe0i3r^(f) then P(z, A) = £ « » uj+ S fi}(u) TJ t - l 3-2 with a4,/?;e- equivalent to a pre-normal form h = \j>, e, f, 0] with e = sgnq0 and p, f depending on N, Ax and «4 only. Proof. See [14]. I This theorem applies to cases Io to IV in Theorem 3-2-2. 3-3-6. Proof of Theorem 3-2-2. We follow the classical framework as described in the proof of Theorem 3-2-1. We omit the details of the calculations. For each given normal form g, we find an intrinsic submodule R' contained in the intrinsic part of the D4-unipotent tangent space of g and from Proposition 2-5-2, R' is contained in the set of higher order terms of g. Then we make the change of coordinates modulo R' for a D4-bifurcation problem/ = [p, q, r, s] establishing tha t / is J^°4-equivalent to g if and only if / satisfies the defining and non-degeneracy conditions stated in each case. Then we apply Corollary 2-3-2 to get the required information about miniversal unfoldings. Case Io was proved in Theorem 3-2-1. For Case Il5 take g = [mN+e3NA1 + e4A 2 ve5,1,0] where el = e\ = e\ = 1 and m + ± l , 0 . Then 2, A, ut>, Jt\ + e5 A1( 0, and + IR.{[m.V,e5,0,0]![A2,0!0,0],[e3iV 7+2e4A1,0,0,0],[iVA1)0,0,0],[0,0,l,0]}. 568 J A C Q U E S - E L I E FTJRTER, ANGELA MARIA SITTA AND IAN STEWART As a consequence, g has D4-co-dimension 2 and topological co-dimension 1 since m is a modal parameter and [1,0,0,0] is also an unfolding term. Now, let From Lemma 3-3-1, R' is intrinsic. F o r C a s e I 2 , t a k e g = [mN+7iNu4 + e0A1,e5,A1 + esui,0] where ra=t= + l , 0 , el = e\ = el = 1 and n =t= 0. Then , e5 u4,0,0], [0,0,N, eb m], [0,0, e0 Alt e5(l - m 2 ) ] , [e0u4,0, ^4,0]>, ,ei,0,0],[e0,0,1,0], The unfolding terms are [N, 0,0,0], [ ^ , 0 , 0 , 0 ] , and [0,0,1,0]. I t follows that g has D4-co-dimension 3 and topological co-dimension 1 since m and n are modal parameters. In this caseiT = {Ji\ + Jlu.i^^Ji\KUi),Jl\,Jll,Mv\. For Case II , take g = [e2iV 2 + eoAj,e5,1,0] where ejj = eij = ejj = 1. Then As 3T»'{g) = 3rqi»<(g) + U.{[N\ 0,0,0], [1,0,0,0], [A^0,0,0], [0,1,0,0], [0,0,1,0]}, the unfolding term is [N, 0,0,0] and by definition g has D4 and topological co- dimension equal to 1. In this case R' = [J^3 u + (NA1,A 2 l,A,ui),^u,J/u,S'u]. For Case III, take g = [ — esN+e0Ave5, l + e7N, 0] where e2, = e| = e2 = 1. Then + {[N2, -N,0,0], [e0NAlt0,N,0l [A, -iV,0,0], [-iVA^A^0,0], [eoJVA^e^.O.O.OHO.O.-iV.l]). As "̂f«(flr) = ST^l^ig) + 01 .{[-iV, 1,0,0], [1,0,0,0], [A15 0,0,0], [0,0,N, 0], [0,0,1,0]}, we take [N, 0,0,0] as the unfolding term. So, g has both smooth and topological D4- co-dimensions equal to 1. Here i ? '= [ ^ + ^4. + ,^2+,^/2 t-l-,^4]. For Case IV, take g = [etN+e9N 2 + e0Avel + ra2A1; 1,0] where e% = e\ = el = i and ni2 =f= 0. Then, + (Ui,A\,A\ AA1: <[^A1,A1,0,0],[iV2,iV,0,0], [N2-A,0,0,0],[0,e0Av0,1], [0,0,A7, Singularity theory 569 We choose [iV, 0,0,0] and [0. Aj, 0,0] as unfolding terms and so g has D4-co-dimension 2 and topological co-dimension 1 since m2 is a modal parameter. For Case V, take g = [e^+CoAuCg A + rax A1; 1,0] where e% = e\ = e\= i and x 4= 0. Then, foA^m^, 0,0], [0,0,1,0]}. Let [0, Aj, 0,0] and [0,1,0,0] be the unfolding terms. So, g has D4-co-dimension 2 and topological co-dimension 1 since m1 is a modal parameter. HereiT = [Jtl + 0. In some cases, the diagram changes when R varies. The asymptotic orbital stability of the solutions, determined by the eigenvalues of the Jacobian, is also shown. Although stability is not in general invariant under contact equivalence there are many cases when it is - a feature that we discuss below when it arises. 4-1. Isotropy types for D^-problems When F acts trivially on the parameter space W', the isotropy subgroup S(a. A) of (x,A)eMn+t', given by 2(a. A) = {yeF\y.x = x}, is independent of A. The fixed-point subspace for a subgroup 2 of F is the following subspace of Un+f: Fix(S) = {xeUn\(T.x = Z, Vcr e £} x IR'. In our situation, however, we must take into account the action of F on W. In that case it is easy to show the following, remembering that the actions yn, y( of ye F on !R™ and IR'' are in general not equal: LEMMA 4-1. Under the previous hypotheses, and FixR«+^(£) = FixR«(S) x Singularity theory 571 Table 2. Fixed-point data Fix (2) Xj = x2 = 0 Z2(K) Z\(K) Z2(xK) ZJ(lVf) 1 x2 X\ x l x l u* = 0 = 0 = x2, A2 = 0 ^ = —*^2' 2 ^ For our situation, the trivial solution (0,0) eIR4 is of maximal isotropy D4. On the state space R2 there are 4 fixed-point subspaces of dimension 1 with isotropy subgroups isomorphic to Z2:a;2 = 0 (Z2(K)), x1 = 0 (Z2( —/«•)), â = x2 (Z2(wc)) and ^i = — x2 (Z2( —i/c)), where the pairs Z2(K),ZC 2(K) and Z2(i/e),Z!;(tK) are conjugate. The corresponding action on the A-space is obtained by applying the group homomorphism i^K.K^l to D4, hence it is equal to Z2(K) for Z2 c)(«c) and 1 for Z2 C)(AC). This means that the isotropy subgroups and their fixed-point subspaces for the chosen D4-action are as listed in Table 2 above. Note that (y,0)eFix(Z2(i<)) at {AVA2) if and only if (0,?/)eFix (ZC 2(K)) at (Av -A2). The space Fix (1) is globally invariant under the symmetry A2t-̂ —A2, so it is enough to consider A2 ^ 0. 4-2. Linearized stability Our linear stability information is given for the system that is, asymptotic stability is given by positive eigenvalues. Modulo the usual restrictions (Vanderbauwhede[16]) they are also valid when g = 0 is a bifurcation equation obtained via Liapunov-Schmidt reduction. The matrix gx of the linearization of g is diagonal on the fixed-point subspaces with non-trivial isotropy, as is the restriction of any jT°4-equivalence, so stability is preserved for solutions with non-trivial isotropy. Moreover, the determinant of the linearization on Fix(l) is also preserved (as for any contact equivalence). The general form for a D4-problem is given by g(z, A) = p(u) z + q(u) Sz + r(u) A2 z + s(u) A2 Sz, where u = (N, A, Ax, tt4). The full solution of g — 0 and the eigenvalue data are listed in Table 3. General formulas for the determinant and trace of the linearization of g are given by Ij/d = Ij/zlHs/zl a n d trgr,. = 2re<72, where g is considered as a function of two independent variables z,z. The trace and determinant of the linearization on Fix (1) are: tr gx = NpN + 2ApA -Nq - AqN - 2A2% + SA2( -rN- 2iWA + a +NsN + 2A« J , 8(p1rN-pwrJ - qA sN)] + u4[4:SrN + 8A(sA rN - sN r j ] . (4- 572 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART Table 3. Solutions and stability for T>4-problems Fixed-point subspaces Fix (D4) Fix (Z2(K)) Fix(Z2(-*)) Fix (Zf(iK)) Fix(l) X2r + Sq = 0 Solutions a;, = x2 = 0 x2 = 0 p = qx\ — X2r + sX2 x\ a;, = 0 p = qx\ + X2r — sX2 x\ x1 = +x2 and A2 = 0 p + sSX2 = 0 Eigenvalues p + X2r and p — X2r (pN-q+(2pi-qN)xl-2qAx{) and (qx\ — X2 r) (PN ~ ? + (2Pi — ?w) x\ ~~ 2?A ̂ 2) + A2( - r w + s — (2rA —«w) a:̂ + 2s^x\) and (ga;| + A2 r) pN and — g Trace and determinant given below in (4-1) Note that solutions in Fix (Z(c)(i/c)) are always contained in Fix (1). Because of the non-trivial action of T on IR'', the linearization satisfies When y( is trivial this means that the linearizations are conjugate along the group orbits (for A fixed), that is, the stability is constant along them. Hence (4-2) also means that stability is constant along the group orbits, but this time this also involves the A-component. Note also that because the actions of D4 on the fixed-points subspaces of the non- trivial isotropy subgroups are absolutely irreducible there cannot be a Hopf bifurcation from points in those subspaces. So periodic orbits can be born only through tertiary bifurcation inside Fix(l). 4-3. Diagrams for the generic normal form We now study the bifurcation diagrams for the generic normal form, namely gf(2,A1,A2) = (??iiV+e0A1)2:-l-e552 + A22, (4-3) with m =t= ± 1,0 and e0 = ± 1, e5 = + 1. It is enough to consider e0 = — 1. Moreover, it is possible to show (see [4]) that (4-3) is also the generic normal form for the more restrictive contact equivalence where Aj and A2 are not mixed but At is of higher status. This is represented in the change of coordinates by A(A1,A2) = (A1(A1,A2),A2(A2)). I t means that the A2-sequence of A^slices is an invariant of contact equivalence. In cartesian coordinates (xltx2, A1; A2) the miniversal unfolding G of g takes the form (— Aj + A2) xx + (m — e5) x\ + (m + e5) xx x\ K — (Aj + A2) x2 + (m — e5) x\ + (m + e5) x\ x2 where m = m + 0. Recall that vi is a modal parameter, that is, small changes in /? have no topological influence on the bifurcation diagrams (except perhaps for isolated 'exceptional' values of vi). We analyse each case according to the variation of m. Singularity theory 573 Table 4. Solutions and their stability for the generic normal form Isotropy subgroup Fix (Z2(*)) Fix (ZJ(#r)) Fix (Xf(iK)) Fix(l) Regions of the (A,, A2)-plane ( T O - £ 5 ) ( A 1 - A 2 ) > 0 (m-e^fA^A,,) > 0 TOA, > 0 A2 = 0 TOAJ > 0 1 m x2 o-2 Xl xl xl A Solutions = 0 (A,-A2) (TO-65) (TO-65) Eigenalues TO-e5 e5A!-TOA2 TO-e5 e 5 A 1 + TOA2 TO-e5 TO and — e5 «*(,., = *(*-*, sg det (ffj) = sg (— e5wi) The trivial solution exists for A e R2 and its stability is given by the signs of the eigenvalues, which are — A1 — A2, — Aj + A2. We have chosen e0 such that for Aj large negative (A2 small) then the trivial solution is stable. It looses stability when we cross the lines Aj + A2 = 0, with a bifurcation to solutions in Fix(Z2 c) K). The rest of the solutions are given in Table 4. One way of representing the bifurcation diagrams is to consider first the A-plane and to describe the boundaries of the regions where the solution set does not change qualitatively. Note that there is an exchange principle for the stability of non-trivial solutions, that is, the solutions in Fix (1) and in Fix (Z2 c)(/c)) cannot be stable at the same time. So we consider two situations, depending on whether the solutions in Fix(l) are stable or not. We consider the representation of the diagrams by a parametrization in polar coordinates: (Aj, A2) = (Rcoss,Rsins) with R > 0. The solution branches are functions of the bifurcation parameters. They are drawn in the space (s,x1,x2) with axes oriented as shown in the figure. In the generic case the qualitative behaviour of the diagrams does not depend on the values of the radius R, so we have ' factored out' the variable R. First, let e5 = 1 with m > 1. Then the solutions in Fix(l) are unstable. What actually happens is that as Ax grows from — oo the trivial solution goes unstable, bifurcating to stable Z2 c)(/e)-solutions (which group occurs depends on the sign of A2) which then go unstable when solutions with trivial symmetry bifurcate. Secondly, if e5 = — 1 and m > 1, the solutions with trivial symmetry are now stable. They come about from secondary bifurcation destabilizing solutions in Fix(Z2 c)(/e)). Note that when m > 1 and e5 = 1 we have the intersection of two ' pitchforks' in the (a^sj-plane. The same occurs in the (s,x2)-plane. In the regions of the A-plane defined by Al > 0 and A2 + Aj/ra > 0, or Aj > 0 and A2 — A1/m < 0, the branches in the space (s,xvx2) have the property that for each fixed s there are four points as solutions. The contact point of these branches with the 'pitchforks' are those 574 JACQUES-ELIE FTJRTER, ANGELA MARIA SITTA AND IAN STEWART corresponding with the coincidence of the solutions which arise from the fixed-point subspaces, as described previously. To draw the bifurcation diagrams we analyse the number of solutions for the bifurcation problem defined by the generic normal form for each fixed (A^Ag). Schematic representations of these solutions are shown in Figs. 3 and 4. The corresponding bifurcation diagrams are drawn in Figs. 1 and 2 respectively. 5. Sample applications In this section we describe two problems where the theory that we have developed can be applied. We concentrate on showing that the mathematical models lead to bifurcation equations which are D4-equivariant of the type studied in Sections 3 and 4. The main requirement is a linearization with two independent but interchangeable modes. Nonlinear effects can then be taken care of via the classification of contact equivalent bifurcation problems in &Yzl» (provided the model leads to a problem of finite co-dimension). The detailed study of the full nonlinear symmetry is postponed to a future paper. In broad terms, provided no additional symmetry has been forgotten or introduced in the model, and if there are k perturbation parameters (in addition to the two bifurcation parameters), then we expect the bifurcation problem to be of co-dimension k and we expect to encounter singularities that arise in the universal deformations of co-dimension k problems. 5*1. Buckling of a thin elastic shell In [8] there is a study of the buckling of a thin isotropic rectangular elastic plate subject to a compressive load A, uniformly distributed over the end faces as shown in Figure 5. A mathematical formulation of the plate problem is given by the non- dimensionalized von Karman equations for w, the vertical displacement, and/, the stress function: A«/~M, J A*w=[w,f] + A[w,f0]. j Here iv,feHl(Q.),[u,v] = uxxvxx — 2uxyvxy + uyyvyy, and the appropriate remaining boundary conditions depend on how the plate is attached along its edges. The excess stress due to extra compression beyond the critical buckling load is given by A/o. Generically, there is a one-dimensional kernel for the linearization of (5-l) at the critical load. However, depending on the aspect ratio of the plate, there are situations where two modes compete. Those modes represent two buckled configurations with consecutive wave numbers. When this happens (5-1) reduces to a Z2 © Z2-equivariant bifurcation problem with the bifurcation parameter being the load A (without symmetry acting on it). Note that in the case of a square plate the classical von Karman equations (5-1) have only a one-dimensional kernel at the principal bifurcation; that is, there are no competing modes. A general situation leading to our framework needs a two-dimensional kernel {(x, y)} with two parameters (Ax, A2), with x and y being interchangeable when A2 = 0. This suggest studying the buckling of a thin anisotropic elastic square plate when a Singularity theory 575 Fig. 5. Rectangular plate subject to lateral compressive load. Fig. 6. Square plate subject to uniform compression. pair of forces F1 and F2 act on the four edges as shown in Fig. 6. Our general theory can be applied to the bifurcation equations obtained from the von Karman approximation to this problem. Those equations represent a D4-equivariant problem with symmetry on the bifurcation parameters, given by the forces Ax = F1 and A2 = F2—F1. But we still have a one-dimensional kernel at the bifurcation and so only a Z2-action in space. We can take care of the situation in two ways. We can consider the buckling of a thin elastic shell, or consider some anisotropic material (or non-uniform compressive load). In the first case we consider the shell as some buckled state of the original isotropic elastic square plate and look for secondary bifurcation with two competing modes. A fairly general formulation of the shallow shell equations is A2/ =-\[w,w\- ky wxx - kx wyy, Ahv = [ where kx and ky represent the sectional curvatures of the shell and A is the Laplacian. As a simple example of the buckling of a shell, assume the following for a shallow shell: the (non-negative) radii of curvature Rx,Ry are roughly constant in the mid- region of the surface, the projection of the shell is the unit square [0,1] x [0,1], and the shell is hinged at the boundary of the square (Dirichlet boundary conditions). We moreover assume that the equilibria correspond to states where the normal compressive stresses N^ =fxx,*Sy =fyy are constant with fxy = 0. Then the model reduces to the following 8th order differential equation in w: wxxxx + 2kx ky %oxxyy wy wx w y) = 0yyyy) + AA(N^ wxx + Sy wyy) with constants kx = l/Rx and ky = 1/Ry, and zero boundary conditions. The linearized eigenvectors are (f>nm = sin nnx sinmny, where n,m~^ 1. The corresponding eigenvalues are = nm~ = (n2 + m2)4 + (ky n 2 + kx m 2)2)2 576 JACQUES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART © © © 0 Fig. 7. Four-cell network. If we want to be able to exchange x and y we must assume that Nx = Ny = N and kx = ky = k. In that case nm N(n2 Note that when Anm, considered as a function on N2, achieves its minimum over the integers for a unique pair n #= m, we have a two-dimensional kernel with the required D4-symmetry. We therefore expect the generic bifurcations in such circumstances to be described by the appropriate normal form derived above. A full nonlinear analysis, confirming this expectation, is in preparation. 5-2. Four-cell rings Another different type of model with the same abstract structure is given by four interconnected cells as in Fig. 7. Here the model equations take the form x = F(x,A,a), (5-2) where a;e[R4n,A = (Aj,A2)elR2 are the main bifurcation parameters and ae[Rfc represents unfolding parameters. We consider the question of forced symmetry breaking as a parameter varies. We suppose first that A2 = 0; F is a D4-equivariant problem which represents identical cells coupled such that all the near-neighbour interactions and all the diagonal interactions are the same. The generators of D4 are a, the mid-edge symmetry, represented by the per- mutation (1, 2,3,4) H> (2,1,4, 3), and /?, the diagonal symmetry (1,2, 3,4) H* (3,2,1,4). When A2 becomes non-zero we lose the D4-symmetry. Generically, we get one of two representations of Z2 © Z2 or Z4: 1. (a) Z 2 ©Z 2 , generated by (the symmetry of the rectangle), (b) Z 2 ©Z 2 , generated by . 2. Z4, generated by . We assume here that symmetry-breaking occurs with the nearest-neighbour interactions equal only on opposite pairs of edges; that is, when A2 =j= 0 we have the first situation 1 a. It is easy to see that the components of F satisfy T = \ 1 1 1 1 1 - 1 1 - 1 1 1 - 1 - 1 - 1 1 1 - 1 Singularity theory 577 for some function / with the following (x, A)-combined symmetry: f(x1,x2,x3,x4,A1.A2) = f(x1,xl,x3,x2,A1, -A2). (5-3) We introduce the change of variables on IR4" given by the matrix >!„• The new coordinates represent X1 ~ (x,x,x,x), X2 ~ (x, —x,x, —x), X3 ~ (x,x, —x, —x) and X4 ~ (x, —x, —x,x). In those coordinates the linearization of X = G(X, A, a) = TTF(TX, A, a) at the origin X = x = 0 is diagonal with elements A+C+B+D, A+C—B—D, A-C+B-D and A-C-B+D, where A=f°i, B = f0^, C = f°x3 and D =f°Xi. Moreover the symmetry (5-3) means that A and C are even in A2 and D(A2) = B{ — A2). Thus if B and D are also even in A2 there is no symmetry-breaking, so (5-1) is still D4- equivariant when A2 =)= 0. On the other hand, when B or D have an odd component in A2, we do see symmetry-breaking. When A2 = 0 we find that B° = D°, so that a-(DG°) = (T(A0 + C0 + 2B0)\Ja-(A0 + C0-2B0)\Ja(A0-C0) (twice), where er denotes the spectrum. Generically, we see that to get a two-dimensional kernel with the required D4-symmetry we must ensure that the trivial solution loses stability through a (double) eigenvalue of A0 — C° crossing the imaginary axis to the right. In the exceptional case when B° = D° = 0, that is, when the linear near- neighbour interactions vanish, it is also possible to work with the eigenvalues of A° + C°. The quantity A° — C° represents the difference of the self-attraction and the diagonal interaction. In the original coordinates, the kernel ofDF" represents steady states of the type (u, v, —u, —v); that is, of diagonally opposite states. The exceptional case when cr(A° + C°) changes sign corresponds to (u,v,u,v), diagonally identical states. REFERENCES [1] T. J. BRIDGES and J. E. FURTEE. Singularity theory and equivariant symplectic maps. Lecture Notes in Math. 1558 (Springer-Verlag, 1993). [2] D. R. J. CHILLINGWORTH. Bifurcation from an orbit of symmetry. In Singularities and dynamical systems (ed. S. N. Pnevmatikos), 258-294 (North-Holland, 1986). [3] J. DAMON. The Unfolding and Determinacy Theorems for subgroups of si and X'. Memoirs of the American Mathematical Society 50 no. 306, 1984. [4] J. E. FURTER, A. M. SITTA and I. STEVVAKT. Algebraic path formulation for equivariant bifurcation problems, preprint, 1995. [5] T. GAFFNEY. Some new results in the classification theory of bifurcation problems. In Multiparameter bifurcation theory. Contemporary Mathematics 56 (Amer. Math. Soc, 1986). [6] M. GOLUBITSKY and M. ROBERTS. A classification of degenerate Hopf bifurcations \vithO(2) symmetry. J. Biff. Eq. 69 (1987), 216-264. [7] M. GOLUBITSKY and D. G. SCHAEFFER. A theory for imperfect bifurcation theory via singularity theory. Commun. Pure Appl. Math. 32 (1979), 21-98. 578 JACQTJES-ELIE FURTER, ANGELA MARIA SITTA AND IAN STEWART [8] M. GOLUBITSKY and D. G. SCHAEFFER. Singularities and groups in bifurcation theory, vol. i, Applied Math. Sci. 51 (Springer-Verlag, 1985). [9] M. GOLUBITSKY, I. STEWART and D. G. SCHAEFFER. Singularities and groups in bifurcation theory, vol. n, Applied Math. Sci. 69 (Springer-Verlag, 1988). [10] A. LARI-LAVASSANI. Multiparameter bifurcation with symmetry via singularity theory, Ph.D. Thesis (Ohio State U., 1990). [11] D. MOND and J. MONTALDI. Deformations of maps on complete intersections, Damon's 3CV- equivalence, and bifurcations, preprint, U. of Warwick, 1991. [12] M. PETERS. Classification of two-parameter bifurcations, Ph.D. Thesis, U. of Warwick, 1991. [13] G. SCHWARTZ. Smooth functions invariant under the action of a compact Lie group. Topology 14 (1975), 63-68. [14] A. M. SITTA. Singularity theory and equivariant bifurcation problems with parameter symmetry, PhD Thesis, U. of Warwick and USP-Sao Carlos, 1993. [15] A. VANDERBAUWHEDE. Stability of bifurcating equilibria and the principle of reduced stability; in Bifurcation theory and applications, Montecatini 1983 (ed. L. Salvadori), Lectures Notes in Math. 1057 (Springer-Verlag, 1984), 209-233. [16] A. VANDERBAUWHEDE. Local bifurcation and symmetry. Research Notes in Math. 75 (Pitman, 1982).