Topological classification of oriented cycles of 4 linear mappings˚ 1 0 2 n Tetiana Rybalkina Vladimir V. Sergeichuk a J rybalkina [email protected] [email protected] 1 Institute of Mathematics, Tereshchenkivska 3, Kiev, Ukraine 1 ] T R . Abstract h t a We consider oriented cycles m [ 1 Fm1 hh A1 // Fm2 A2 // ... At´2// Fmt´1 At´1 //Fmt v At 0 5 of linear mappings over F “ C or R, and reduce the problem of their 5 2 classification up to homeomorphisms in the spaces Fm1,...,Fmt to . the case t “ 1, which was studied by N.H. Kuiper and J.W. Robbin 1 0 [Invent. Math. 19 (no.2) (1973) 83–106] and by other authors. 4 1 Keywords: Oriented cycles of linear mappings; Topological equiva- : lence v i 2000 MSC: 15A21; 37C15 X r a 1 Introduction We consider the problem of topological classification of oriented cycles of linear mappings. ˚PrintedinUkrainian Math. J.,2014. Thesecondauthorwassupportedinpartbythe FoundationforResearchSupportofthe StateofSa˜oPaulo(FAPESP),grant2012/18139- 2. 1 Let A : V1 hh A1 // V2 A2 // ... At´2// Vt´1 At´1 // Vt (1) At and B : W1 hh B1 //W2 B2 // ... Bt´2// Wt´1 Bt´1 // Wt (2) Bt be two oriented cycles of linear mappings of the same length t over a field F. We say that a system ϕ “ tϕ : V Ñ W ut of bijections transforms A to B i i i i“1 if all squares in the diagram V1 jj A1 // V2 A2 // ... At´2 //Vt´1 At´1 //Vt At ϕ1 ϕ2 ϕt´1 ϕt (3) (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) W1 jj B1 //W2 B2 // ... Bt´2// Wt´1 Bt´1 //Wt Bt are commutative; that is, ϕ A “ B ϕ , ..., ϕ A “ B ϕ , ϕ A “ B ϕ . (4) 2 1 1 1 t t´1 t´1 t´1 1 t t t Definition 1. Let A and B be cycles of linear mappings of the form (1) and (2) over a field F. (i) A and B are isomorphic if there exists a system of linear bijections that transforms A to B. (ii) A and B are topologically equivalent if F “ C or R, V “ Fmi, W “ Fni for all i “ 1,...,t, i i and there exists a system of homeomorphisms1 that transforms A to B. 1By [2, Corollary 19.10] or [17, Section 11], m1 “n1,...,mt “nt. 2 The direct sum of cycles (1) and (2) is the cycle A‘B : V ‘W A1‘B1 // V ‘W A2‘B2 // ... At´1‘Bt´1// V ‘W 1 1jj 2 2 t t At‘Bt The vector dimA :“ pdimV ,...,dimV q is the dimension of A. A cycle 1 t A is indecomposable if its dimension is nonzero and A cannot be decomposed into a direct sum of cycles of smaller dimensions. AcycleAisregular ifallA ,...,A arebijections, andsingular otherwise. 1 t Each cycle A possesses a regularizing decomposition A “ A ‘A ‘¨¨¨‘A , (5) reg 1 r in which A is regular and all A ,...,A are indecomposable singular. An reg 1 r algorithmthatconstructs aregularizingdecomposition ofanonorientedcycle of linear mappings over C and uses only unitary transformations was given in [21]. The following theorem reduces the problem of topological classification of oriented cycles of linear mappings to the problem of topological classification of linear operators. Theorem 1. (a) Let F “ C or R, and let A : Fm1 ii A1 //Fm2 A2 // ... At´2// Fmt´1 At´1 //Fmt (6) At and B : Fn1 hh B1 // Fn2 B2 // ... Bt´2// Fnt´1 Bt´1 //Fnt (7) Bt be topologically equivalent. Let A “ A ‘A ‘¨¨¨‘A , B “ B ‘B ‘¨¨¨‘B (8) reg 1 r reg 1 s be their regularizing decompositions. Then their regular parts A and B reg reg are topologically equivalent, r “ s, and after a suitable renumbering their indecomposable singular summands A and B are isomorphic for all i “ i i 1,...,r. 3 (b) Each regular cycle A of the form (6) is isomorphic to the cycle A1 : Fm1 ii 1 //Fm2 1 // ... 1 // Fmt´1 1 //Fmt (9) At¨¨¨A2A1 If cycles (6) and (7) are regular, then they are topologically equivalent if and only if the linear operators A ¨¨¨A A and B ¨¨¨B B are topologically t 2 1 t 2 1 equivalent pas the cycles Fm1ýA ¨¨¨A A and Fn1ýB ¨¨¨B B of length t 2 1 t 2 1 1q. Kuiper and Robbin [18, 16] gave a criterion for topological equivalence of linear operators over R without eigenvalues that are roots of 1. Budnitska [4, Theorem 2.2] found a canonical form with respect to topological equivalence of linear operators over R and C without eigenvalues that are roots of 1. The problem of topological classification of linear operators with an eigenvalue that is a root of 1 was studied by Kuiper and Robbin [18, 16], Cappell and Shaneson [6, 7, 8, 9, 10], and Hsiang and Pardon [15]. The problem of topologicalclassification ofaffineoperatorswasstudied in[4, 12, 1, 3, 5]. The topological classifications of pairs of counter mappins V ÝÐÑÝV (i.e., oriented 1 2 cycles of length 2) and of chains of linear mappings were given in [19] and [20]. 2 Oriented cycles of linear mappings up to isomorphism This section is not topological; we construct a regularizing decomposition of an oriented cycle of linear mappings over an arbitrary field F. A classification of cycles of length 1 (i.e., linear operators V ý) over any field is given by the Frobenius canonical form of a square matrix under similarity. The oriented cycles of length 2 (i.e., pairs of counter mappins V ÝÐÑÝV ) are classified in [11, 14]. The classification of cycles of arbitrary 1 2 lengthandwitharbitraryorientationofitsarrows iswell known inthetheory of representations of quivers; see [13, Section 11.1]. For each c P Z, we denote by rcs the natural number such that 1 ď rcs ď t, rcs ” c mod t. 4 By the Jordan theorem, for each indecomposable singular cycle V ýA there exists a basis e ,...,e of V in which the matrix of A is a singular 1 n Jordan block. This means that the basis vectors form a Jordan chain A A A A A e ÝÑ e ÝÑ e ÝÑ ¨¨¨ ÝÑ e ÝÑ 0. 1 2 3 n Inthesamemanner, eachindecomposable singularcycleAofanarbitrary length t also can be given by a chain e ÝAÑp e ÝAÝrÝp`ÝÑ1s e ÝAÝrÝp`ÝÑ2s ¨¨¨ ÝAÝrÝq´ÝÑ1s e ÝAÝrÑqs 0 p p`1 p`2 q in which 1 ď p ď q ď t and for each l “ 1,2,...,t the set te |i ” l mod tu is i a basis of V ; see [13, Section 11.1]. We say that this chain ends in V since l rqs e P V . The number q ´p is called the length of the chain. q rqs For example, the chain ee6 rr❞rr❡❞❡❞❡//❞//e❡e❞❡7❞❡❞❡❞❡❞❡❞//❡//e❞0❡❞8❡❞❡❞❡❞❡//❞❡ee❞❡49❞❡❞❡❞❡❞////ee150 11 12 of length 8 gives an indecomposable singular cycle on the spaces V “ Fe ‘ 1 6 Fe , V “ Fe ‘Fe , V “ Fe , V “ Fe ‘Fe , V “ Fe ‘Fe . 11 2 7 12 3 8 4 4 9 5 5 10 Lemma 1. Let A : V1 hh A1 //V2 A2 // ... At´2 //Vt´1 At´1 // Vt At be an oriented cycle of linear mappings, and let (5) be its regularizing decom- position. (a) Write Aˆ :“ A ¨¨¨A A : V Ñ V (10) i ri`t´1s ri`1s i i i and fix a natural number z such that V˜ :“ AˆzV “ Aˆz`1V for all i “ 1,...,t. i i i i i Let A˜: V˜1 hh A˜1 // V˜2 A˜2 // ... A˜t´2 //V˜t´1 A˜t´1 // V˜t A˜t 5 be the cycle formed by the restrictions A˜ : V˜ Ñ V˜ of A : V Ñ i i ri`1s i i V . Then A “ A˜ pand so the regular part is uniquely determined ri`1s reg by Aq. (b) The numbers k :“ dimKerpA ¨¨¨A A q, i “ 1,...,t and j ě 0, ij ri`js ri`1s i determine the singular summands A ,...,A of regularizing decompo- 1 r sition (5) up to isomorphism since the number n pl “ 1,...,t and lj j ě 0q of singular summands given by chains of length j that end in V l can be calculated by the formula n “ k ´k ´k `k (11) lj rl´js,j rl´js,j´1 rl´j´1s,j`1 rl´j´1s,j in which k :“ 0. i,´1 Proof. (a) Let (5) be a regularizing decomposition of A. Let V “ V ‘V ‘¨¨¨‘V , i “ 1,...,t, i i,reg i1 ir be the corresponding decompositions of its vector spaces. Then AˆzV “ i i,reg V (since all linear mappings in A are bijections) and AˆzV “ ¨¨¨ “ i,reg reg i i1 AˆzV “ 0. Hence V “ V˜, and so A “ A˜. i ir i,reg i reg (b) Denote by σ :“ n `n `n `¨¨¨ ij ij i,j`1 i,j`2 the number of chains of length ě j that end in V . Clearly, k “ σ , i i0 i0 k “ σ `σ , ..., and i1 i0 ri`1s,1 k “ σ `σ `¨¨¨`σ ij i0 ri`1s,1 ri`js,j for each 1 ď i ď t and j ě 0. Therefore, σ “ k ´k , l :“ ri`js lj ij i,j´1 (recall that k “ 0). This means that l ” i ` j mod t, i ” l ´ j mod t, i,´1 i “ rl´js, and so σ “ k ´k . lj rl´js,j rl´js,j´1 We get n “ σ ´σ “ k ´k ´k `k . lj lj l,j`1 rl´js,j rl´js,j´1 rl´j´1s,j`1 rl´j´1s,j 6 3 Proof of Theorem 1 In this section, F “ C or R. (a) Let A and B be cycles (6) and (7). Let them be topologically equiv- alent; that is, A is transformed to B by a system tϕ : Fmi Ñ Fniut of i i“1 homeomorphisms. Let (8) be regularizing decompositions of A and B. First we prove that their regular parts A and B are topologically reg reg equivalent. In notation (10), Aˆ “ A ¨¨¨A A , Bˆ “ B ¨¨¨B B . i ri`t´1s ri`1s i i ri`t´1s ri`1s i Let z be a natural number that satisfies both AˆzFmi “ Aˆz`1Fmi and BˆzFni “ i i i Bˆz`1Fni for all i “ 1,...,t. By (3), the diagram i Fmi Aˆz //Fmi ϕi ϕi (cid:15)(cid:15) (cid:15)(cid:15) Fni Bˆz // Fni is commutative. Then ϕ ImAˆz “ ImBˆz for all i. Therefore, the restriction i i i ϕˆ : ImAˆz Ñ ImBˆz is a homeomorphism. The system of homeomorphisms i i i ϕˆ ,...,ϕˆ transforms A˜ to B˜, which are the regular parts of A and B by 1 t Lemma 1(a). Let us prove that r “ s, and, after a suitable renumbering, A and B are i i isomorphic for all i “ 1,...,r. Since all summands A and B with i ě 1 can i i be given by chains of basic vectors, it suffices to prove that n “ n1 for all i ij ij and j, where n1 is the number of singular summands B ,...,B in (8) given ij 1 s by chains of length j that end in the ith space Fni. Due to (11), it suffices to prove that the numbers k are invariant with ij respect to topological equivalence. In the same manner as k is constructed by A, we construct k1 by B. ij ij Let us fix i and j and prove that k “ k1 . Write ij ij A :“ A ¨¨¨A A , B :“ B ¨¨¨B B , q :“ ri`j `1s ri`js ri`1s i ri`js ri`1s i and consider the commutative diagram Fmi A //Fmq (12) ϕi ϕq (cid:15)(cid:15) (cid:15)(cid:15) Fni B // Fnq 7 which is a fragment of (3). We have k “ dimKerA “ m ´dimImA, k1 “ n ´dimImB. ij i ij i Because ϕ : Fmi Ñ Fni is a homeomorphism, m “ n (see [2, Corol- i i i lary 19.10] or [17, Section 11]). Since the diagram (12) is commutative, ϕ pImAq “ ImB. Hence, the vector spaces ImA and ImB are homeomor- q phic, and so dimImA “ dimImB, which proves k “ k1 . ij ij (b) Each regular cycle A of the form (6) is isomorphic to the cycle A1 of the form (9) since the diagram Fm1 ii 1 //Fm2 1 //Fm3 1 //Fm4 1 // ... 1 //Fmt At¨¨¨A2A1 1 A1 A2A1 A3A2A1 At´1¨¨¨A2A1 (13) (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) (cid:15)(cid:15) Fm1 ii A1 //Fm2 A2 // Fm3 A3 // Fm4 A4 // ... At´1 // Fmt At is commutative. Let A and B be regular cycles of the form (6) and (7). Let them be topologically equivalent; that is, A is transformed to B by a system ϕ “ pϕ ,...,ϕ q of homeomorphisms; see (3). By (4), 1 t ϕ A A ¨¨¨A “ B ϕ A ¨¨¨A 1 t t´1 1 t t t´1 1 “ B B ϕ A ¨¨¨A “ ¨¨¨ “ B B ¨¨¨B ϕ , t t´1 t´1 t´2 1 t t´1 1 1 and so the cycles Fm1ýA ¨¨¨A A and Fm1ýB ¨¨¨B B are topologically t 2 1 t 2 1 equivalent via ϕ . 1 Conversely, let Fm1ýA ¨¨¨A A and Fm1ýB ¨¨¨B B be topologically t 2 1 t 2 1 equivalent via some homeomorphism ϕ , and let A1 and B1 be constructed 1 by A and B as in (9). Then A1 and B1 are topologically equivalent via the system of homeomorphisms ϕ “ pϕ ,ϕ ,...,ϕ q. Let ε and δ be systems of 1 1 1 linear bijections that transform A1 to A and B1 to B; see (13). Then A and B are topologically equivalent via the system of homeomorphisms δϕε´1. 8 References [1] J. Blanc, Conjugacy classes of affine automorphisms of Kn and linear automorphisms of Pn in the Cremona groups, Manuscripta Math. 119 (2) (2006) 225–241. [2] G.E. Bredon, Topology and Geometry, New York: Springer-Verlag, 1993. [3] T.V. Budnitska, Classification of topological conjugate affine mappings, Ukrainian Math. J. 61 (2009) 164–170. [4] T. Budnitska, Topological classification of affine operators on unitary and Euclidean spaces, Linear Algebra Appl. 434 (2011) 582–592. [5] T. Budnitska, N. Budnitska, Classification of affine operators up to biregular conjugacy, Linear Algebra Appl. 434 (2011) 1195–1199. [6] S.E. Cappell, J.L. Shaneson, Linear algebra and topology, Bull. Amer. Math. Soc., New Series, 1 (1979) 685–687. [7] S.E. Cappell, J.L. Shaneson, Nonlinear similarity of matrices, Bull. Amer. Math. Soc., New Series, 1 (1979) 899–902. [8] S.E. Cappell, J.L. Shaneson, Non-linear similarity, Ann. of Math. 113 (2) (1981) 315–355. [9] S.E. Cappell, J.L. Shaneson, Non-linear similarity and linear similarity are equivariant below dimension 6, Contemp. Math. 231 (1999) 59–66. [10] S.E. Cappell, J.L. Shaneson, M. Steinberger, J.E. West, Nonlinear sim- ilarity begins in dimension six, Amer. J. Math. 111 (1989) 717–752. [11] N.M. Dobrovol1skaya, V.A. Ponomarev, A pair of counter-operators, Us- pehi Mat. Nauk. 20 (6) 80–86 (in Russian). [12] W. Ephra¨mowitsch, Topologische Klassifikation affiner Abbildungen der Ebene, Mat. Sb. 42 (1) (1935) 23–36. [13] P. Gabriel, A.V. Roiter Representations of Finite-Dimensional Algebras, Encyclopaedia of Math. Sci., V. 73 (Algebra VIII), Berlin: Springer- Verlag, 1992. 9 [14] R.A. Horn, D.I. Merino, Contragredient equivalence: a canonical form and some applications, Linear Algebra Appl. 214 (1995) 43–92. [15] W.C. Hsiang, W. Pardon, When are topologically equivalent orthogonal transformations linearly equivalent, Invent. Math. 68 (2) (1982) 275– 316. [16] N.H. Kuiper, J.W. Robbin, Topological classification of linear endomor- phisms, Invent. Math. 19 (2) (1973) 83–106. [17] J. McCleary, A First Course in Topology: Continuity and Dimension, Providence: Amer. Math. Soc., 2006. [18] J. W.Robbin, Topologicalconjugacyandstructural stability fordiscrete dynamical systems, Bull. Amer. Math. Soc. 78 (1972) 923–952. [19] T. Rybalkina, Topological classification of pairs of counter linear maps, Mat. Stud. 39 (1) (2013) 21–28 (in Ukrainian). [20] T. Rybalkina, V.V. Sergeichuk, Topological classification of chains of linear mappings, Linear Algebra Appl. 437 (2012) 860–869. [21] V.V. Sergeichuk, Computation of canonical matrices for chains and cy- cles of linear mappings, Linear Algebra Appl. 376 (2004) 235–263. 10