Baixe o app para aproveitar ainda mais
Prévia do material em texto
REVISTA DE LA UNION MATEMATICA ARGENTINA Director: Dario J. Picco Redactores: A. Diego, E. Gentile, M. Herrera, D. Herrero, C. Trejo, O. E. VUlamayor Secretarios de Redacci6n: M. L. Gastaminza, A. G. de Pousa VOLUMEN 27, NUMERO 1 1974 BAHIA BLANCA 1975 La U;M.A. reconoce cuatrocategoriii.s protectores, tltulares y adheren~s. El ,mlle' ~ljrc:ir" p:r<it~jtQ~~~Pl~~ ta anual de $.200,00, por 10 menos;ei y eladherente (estudhmte solamertte>: pagos deberan efeetuarse por . cheque; .. ' orden de UNION MATEMATrCA ·· All~Gl~N':l'Jlli.t~, ... ,;,."Y"'"'''''' .~ .,,: Buenos Aires. Los autores de trabajQSi-eeib~n ; griatl~~~liro,~nte;, de . 50 ejemplares.Las eorreeeiones · eJl;";r;l",pl'.tl,lpap~'~ 'a,~ cuenta de los autores. · .. Presidente: Dr. A. GonzaJe.zDomihg-dez':; V~¢~l~t~t$I(i~ll~~i 16 e lng. E. GaspariSectet~i-io: t;i6; .< . Lic. N. D. patetta; TeSQiero:tie.b. ; . i),.I~~t~l~ij Castro; Director ' de Publieaciones: ~ D£:, Zona Buenos Aire~ ~ LaPl~ta.: " . A.Niell; Zona Cuyo: Dr.O.:!3:.o · rglll.; i~~jij,~~;:~I!~~ Noreste: lng. M. Ma.ran,gl1:Uj.~; na Sur: -Lie. J. A.D~iningliei; .f. ~e<!:l'e;~al'i'O)~ .J~o(~ M. L. Gurmendi; Cordoba: Jngf, .." •.. ·.H.'''H La Plata: Dra. S. salvioll; 'l\'~-~t ,('delt" Plat'~I"~l; d6za: Dr. E.Zarantonetiti; ' Rc;)s~;t:iI): Ovejero; San Luis: Li,c. ¥ .. R. :ae:tfli~l~~!:iJ~~~~;g~~s de Aranzadi; Tuc1lIllan:Dt; R>L ... ",J.v ....... . . fug. Rafael Laguardia ' (TT .. ·i1O',j.,.;,,\ ''1· .... ·,,'''''''' ..... Dr. Cesar Carranza (~ertl) Frueht (Chile), Di. ~M~HO , .. ' '~'!"~''''''~ ,-,ro"",,,':'! . .Io'L-; ru.±u< dara (MexiCO). . All administrative ·· COI·re!lponQ.!e~(:e a:nd"*l~cd;p~i6ril(:,,~ be addressed to: t;' :~-. REVISTA DE LA UNION MATEMATICA ARGENTINA Dkector: Dario 1. Picco Rec:tactores: A. Diego, E. Gentile, M. Herrera, D •. Herrero, C. Trejo, O. Eo Vlllamayor Secretarlos de Redaccl6n: M. L. GastAmtnza, A. O. de POUIa VOLUMEN 27, NUMERO 1 1974 BAHIA BLANCA 1976 Revista de la Union Matematica Argentina Volumen 21, 1974. AN ELEMENTARY PROOF OF THE JORDAN CANONICAL FORM Enzo R. Gent; i e Let V be a finite dimensional vector space over a field K. Let t be an endomorphism of V. Then, as it is well known, V can be written as a direct sum of cyclic subspaces. If mt(X) E Klxl de- notes the minimal - polynomial of t , in studying the structure of t one is reduced to consider the case where mt(X) = p(X)a, with p(X) an irreducible polynomial in Klxl and where a is a nat- ural number. A cyclic subspace of V admits the following matrix representation: P N P N (C) P N P where P is a block consisting of the companion matrix of p(X) and where N is the block of the same siz.e as P. o o o o o 0 o o o o o o o The rational canonical form of t consists of the matrix obtained by assembling blocks of the type (C). If p(X) = X - k, k E K (for instance if K is algebraically closed) then (C) becomes a Jordan block ,and the canonical form is called the Jordan canonical form. The proof of the Jord.an canonical form depends essentially on the canonical form of a nilpotent endomorphism, fact tediously proven in most books in linear algebra. , 2 In this Note we intend to give a direct proof of the Jordan canon- ical form. Nevertheless the ideas in the proof permit to prove the structure theorem of finitely generated torsion modules over a principal domain, which shall be done elsewhere. To start with, we set some terminology. Throughout this Note, sub- space means subspaae invariant (or stab~e) under t. Furthermore morphism means morphism aommuting with t. Recall that a subspace W of V is a direct summand of V if and only if there is amorphism f: V --+ W whose restriction fl W to W coincides with Idw' Let v E V. With (v > we denote the cyclic subspace of V generated by v. Any element of (v > can be written as a(t)v for some a (X) E K Ix I. From now on assume that t is a nilpotent morphism. For any v E V, v # 0 we define the order of v as the highest positive integer o(v) satisfying to(v)v = 0 Notice the following property of o(v): for any a(X) E Klxl, a(t)v = 0 implies that Xo(v) divides a(X). In fact, this follows from the property of being Klxl a principal domain and standard arguments on polynomials. Let v E V be an element of order s. We consider in (v the fo1- lowing sequences of subspaces: o c ( ts-1v > c ... c ( tiv > c ... c ( tv > c {v > We claim that if x E (v then In fact, part "if" is trivial. On the other hand, let tix = O. Write x = a(t)v. Then 0 = tix = tia(t)v and therefore XS divides Xia(X) which implies that a(X) is divisible by xS - i , a(X) = Xs-ir(X). Finally x = ts-ir(t)v E (ts-iv > as we wanted to prove. Next we start to prove that V is a direct sum of cyclic subspaces. Let v E V be an element of highest order in V. This implies that tSz = 0 for any z in V. Let (v > be the cyclic subspace of V generated by v. We shall de- fine a projection of V onto {v > • For this, let W be a subspace of V satisfying the following prope£ ties 3 i) {v} c W ii) There is a morphism f: W --+ (V) such that fl(v} iii) W is maximal with properties i) and ii). Obviously if W = V nothing has to be proved. Assu~e then V , W. Choose u E V - Wand consider the subspace W' = W + (u) • W' contains (v) and we shall extend f to W'. Let J be the ideal of all polynomials a(X) in Klxl satisfying a(t)u E W. J is generated by a monic polynomial g(XJ. Since tSu = 0 E W it follows that XS E J, therefore g(X) divides XS, so g(X) = Xd for some d ~ s. Notice that tdu E W therefore f(tdu) E (v ) . But since t s - d f(tds) = f(tSu) = 0, by an earlier remark we get that f (tdu) E ( tdv ) , that is for some x E (v ) . We set f': W+(u} ----..(v} f': w + a (t)u l---- f (w) + a (t)x and we claim that f' is a well d'efined morphism of W' onto (v ) • Let w,w' E W, a(X),a' (X) E Klxl satisfy w + a(t)u = w' + a' (t)u Hence (a' (t) - a(t))u w - w, E W implies that a' (X) - a(X) E J, that is a' (X) - a(X) = b(X)Xd Therefore few) - few') f(b(t)tc1U ) = b(t)tdx .. (a' (t) - a(t))x 4 f(w) + a(t)x = few') + a' (t)x which says that f' is well defined. Clearly f' is a morphism of W' onto (v > that extends f. Since W is properly contained in W', we have a contradiction. Therefore f is a projection of V onto (v and we can write V = (v > e V' But by an inductive argument, V' is a direct sum of cyclic sub- spaces, so V is a direct sum of cyclic subspaces and this was our claim. REMARK. Notice that the present proof holds for any endomorph i sm t whose minimal polynomial is mt(X) = p(X)a, with p(X) irreducible in Klxl. As we remarked at the beginning the general situation a . mt(X) = IT Pi(X) 1 of different irreducible factors reduces trivi- ally to the case above. Recibido en marzo de 1974 Universidad de Buenos Aires Argentina Revista de la Union Matematica Argentina Volumen 27, 1974. ACTIONS ON A GRAPH Antonio Diego ABSTRACT. Part of the theory of flows and tensions on a graph is extended to any kind of actions, i.e. to any subspace of the space of real functions defined on the arcs; in particular, the theorems on the existence of flows or tensions under bilateral restraints. 1. INTRODUCTION. Our basic setting is a graph G = (X,S); here it means that sex x X verifies: (i,i) ~ Sand (i,j) E S implies (j,i) E s, i,j E X. We assumeG connected. We consider functions f,g, ... defined on S and we denote by f.g L f ij gij' the usual scalar product. By Ewe denote the linear space of anti-symmetric functions: f.. + f.. = 0, (i,j) E S. ~J J ~ The subspaces of flows and tensions,<I>,e C E, are defined by: 'II E <1>, if L 'II .. = 0, i E X, and 0 E e, if o .. = t. - t i , t defined j ~J ~J J on X (0 = 6t) . <I> and e are orthogonal complements in E. If FeE is defined as the set of solutions f of the linear sys- tem: Aa.f = 0,~ EM, its orthogonal spa'ce G = pi in E, is gene- rated by the p'a, ~ E M, where J.I~j • Aj i - A~j When M '= X and the A~k vanish except for i, writing A~k = Aik , we have p.~j = -A ij , J.I~i = Aij and J.I~k = 0 for the remaining (h,k) E S: Then, the elements g E G are of the form gij = tjAji - tiAij (g = AA t ), where ~ is a function on X. The case A.: 5 1 corresponds to F = <1>, ~J G = e. Given an orientation to G - i.e. a subset U of S contain- ing for each (i,j) E S one and only one of the pairs (i,j), (j,i)- and a positive mi , i E X, defining Aik= mi , if (i,k) E U, and 6 x .. = 1, if (i,j) ~ U, we obtain the spaces of multiplicative ~J flows and tensions ([ 11 pp.22S). Actions of a certain kind can be thought as the elements f of a subspace F of E - J ij representing the · intensity of the action f . trasmited from i to j through the link (i,j) E S, Certain notions and results of the theory of flows and tensions on a graph, can be extended to any subspace F of E. Doing that, a unified linear treatment of the outstanding cases F = ~, F = e - that may be useful - is obtained. In 2 we give the notiQn of elementary action, corresponding to the notions of elementary cycles and cocycles, and a proposition on the decomposition of any action in elementary ones. It gives the known decomposition of a positive flow (tension) - on an ori- ented graph - as a positive linear combination of elementary cy- cles (cocycles) ((1] pp.143). In 3 we prove the analogue of Hoffman and Roy's theorems ([2], [3]) for actions of any kind, using the appropiate geometric ver- sion of the consistence theorem of a system of linear inequalities (Farkas-Minkowsky). 2. ELEMENTARY ACTIONS. I For fEE we denote s(f) = :{(i,j)/f .. > OJ, the (effective) sup- ~J port of f. It is seen that, for f,g E E: (A) ~~s(g) C s(f) implies s(f-Xg) C s(f), properly, for the posi- f · i X = max{-i.i / g > o} A function f E F, f # 0, is gij ij • tive number said to be an eLementary funation of F if for any g E F, s.(g) c c s (f) implies g = H. I This means that s(f) is) a minimal set of {s(g)/g E F}. In fact, s (f) minimal implies, flor each g E F with s (g) C s (f), that s(f-Xg) C s(f), proper~y, (A); since f - Xg E F it follows s(f -.Xg) • 0, f ·· Xg. , The converse is ~lear. I Of course, if f is an l~lementary function of F, so is xl, X ~ O. PROJ.>OSITION, Any . fEE. f ; , ot i.~ a 8Urn , of eLementary funation8 I 7 fn of F suah that s(fn) C s(f). Proof. Let gl be an elementary function of F such that s(gl) C C s(f). From (A), for some Xl > 0, it is s(f-Xlg l ) c s(f), proper- ly. If f - Xlg l "I 0, we apply to f - XlglE F the same argument and we get an elementary g2 E F, X2 > 0, such that s(g2) c c s(f-Xlg l ) and s(f-X l g l -X 2g2) C s(f-Xlg l ), properly. After a fi- nite number of steps we have elementary gl, ... ,gk E F, Xl ,X 2, ... , Xk > 0, such that f - Xlg l - lows with fn = Xng a , 1 ";a"; k. - Xkgk = O. The proposition fol~ For a set Z C S, such that (i,j) E Z implies (j,i) ~ Z, we define ~ = ~(Z) by ~ij = 1, -1, 0 according to (i,j) E z, (j,i) E Z or (i,j),(j,i) ~ Z, respectively. ~ E E and s(~) Z. If Z = {(i,j),(j,k), ... ,(h,l),(l,i)} is a cycle, ~ is a flow, if Z = {(i,j)/i E A, j ~ A} (A, X - A "I 0) is a cocycle, ~ is the tension ll( -1 A) • If the sequence i,j,k, ... ,h,l of the cycle Z has not repeated ele- ments Z is an elementary cycle. If A and X - A are connected, in the graph Gz obtained eliminating the (i,j),(j,i), with (i,j) E Z, the cocycle Z is said to be an elementary one. It is clear that ~ = ~ (Zl is an elementary flow, when Z is an ele- mentary cycle. For an elementary cocycle Z, if 0 llt is such that s(O) c Z = s(r), the connectedness of A and X - A in Gz im- plies tl A = a, t IX _ A = p, then llt = 0 = (p-a)r, with P ~ a. Hence ~ is an elementary tension. Conversely, if op ., 0 is a flow, sLOP) verifies that (i,j) E s(op) implies Cj,k) E sLOP) for some k., i (op .. + L op'k = 0, opJ"1' = J 1 k,li J = -op ij < 0). It follows that sLOP) contains a cycle, and then also an elementary cycle Z. Hence if op is an elementary flow, t = X.op, X > O. On the other ~and, if 0 = llt , 0 is a tension, taking a: min ti < a < < max t., the cocycle Z defined by means of A = {i/t. <: a} is con 1 1 - ,\:ained in s(8). Hence, if 8 is an elementary tension we have t = . • X8, X > O. Z has to be an elementary cocycle, otherwise we could 8 take a connected component A' of A (alternatively of X - A) in Gz and define Z' in terms of A'. But this wQuld imply set') c s(8), properly; which is a contradiction. Resuming, the multiples A.t, A > 0, t ~ t (Z), for Z elementary cy- cle (cocycle), are the elementary functions of 41(9). 3. EXISTENCE THEOREM. We consider a finite dimensional linear ~pace with the scalar pr~ duct x.y. For a cone Q - Q+Q c Q, AQ ~ Q for every A ~ 0 - the du al cone is defined by QO ~ {xix. y ;;. 0, for any y E Q}. We need the theorem of consistence of a system of linear inequali- ties under the following form: "Given the polyhedral set C and the polyhedral cone Q, (C, Q -1 0) it holds: Q n C -1 0 if and only if, for every x E QO there is a c E C such that x.c ~ 0". In fact, if Q n C 0, we can separate the closed convex (poly~ hedral) set Q - C from 0; i.e. there is x such that x.(c~q) < 0, for any c E C, q E Q. Taking Aq, A > 0, instead of q, we conclude that x.q ;;. 0, i.e. x E QO. For q ~ a we have x.c <0 for every c E C. The converse is clear. We will apPlY the theorem to a subspace Q. In this case QO ~ Ql. THEOREM. Let c ij +c j i ;;> 0, for any (i, j)E Stand F be a linear subspaae of E. In order that there e~ists f < c, f E F, it is nea essary and 8uffiaient that. for eaah e~ementary g E G ~ Fl, g+.c ;> o. + REMARK. As it is usual: g ~ max (g,O). The condition C .. +C .. ;;. 0 1J J 1 is obviously necessary for the existence of an anti-symmetric f "' c. Proof., Omitting the word "elementary", the equivalence follows from the theorem of consistence applie~ to Q = F, C = {xix", c} in the linear space E. In fact, Q n c ~ 0, i.e. there is f ~ F, f "' c, is equival~nt to 9 assert that for any g E G = QO there is x E E, x ~ c such that g.x ~ O. This implies g+.c ~ 0, since from (g+-g-)x ~ 0, C ~ x, it follows g+.c ~ g+.x ~ g- .x, and then 2g+.c ~ (g++g-)x = Iglx = 0 (I g I is symmetric). + And conversely,if g .c ~ 0, g E G, then defining, for a given g E G, x by x .. = c .. , - c .. , -1/2(c .. -c .. ), according to g . . > 0, l.J l.J l.J l.J J l. l.J < 0 or 0, we have g.x = 2g+.c ~ O. Finally, if g+.c ~ 0 for elementary functions g E G, the same holds for any g E G, since from the proposition we can write g+ L g:, for elementary functions ga of G. a REFERENCES [11 BERGE,C.and GHOUILA-HOURI,A., P~og~amm~ng, game¢ and ~~an¢ po~~a~~on ne~wo~~¢, John Wiley and Sons Inc. New York (1965). [2J HOFFMAN,A.J., Same ~ecen~ appl~ca~~on¢ 06 ~he ~heo~y 06 l~n ea~ ~nequal~~~e~ ~o ex~~emal comb~na~o~~al analy¢~~, Procee- dings of the Tenth Symposium in Applied Mathematics. A.M.S., New York, (1960). [31 ROY,B., Chem~nemen~ e~ connex~~~ dan¢ le¢ g~aphe¢: Appl~ca ~~on aux p~obleme~ d'o~donnancemen~,Metra, Serie speciale, N°1, (1962). Recibido en marzo de 1974 Universidad Nacional del Sur Bah1a Blanca, Argentina En el cu~~o de la ~p~e¢~6n del p~e~en~e ejempla~ de la Rev~~~a de la UMA hemo~ ~u6~~do la ~n60~~unada p~~d~da de nue~tM~ c.ompai!e~o¢ de t~abaj a y que~~do~ am~go¢ Evel~o T. Oklande~ y R~c.a~do Alemany, a cuya memo~~aded~camo~ e~~e ~~abaj o. Revista de 1a Union Matematica Argentina Vo1umen 27, 19~4. AN ELEMENTARY PROOF OF THE STRUCTURE THEOREM OF FINITELY GENERATED MODULES OVER A PRINCIPAL DOMAIN Enzo R. Gentile Let A be a principal domain. It is a well known result that if M is a finitely generated A-module then M can be written as a di- rect sum of a free submodule and thetorsion submodule. The free part is (up to isomorphism) completely determined by its rank. So, the main point is to characterize the torsion submodule. The- refore let M be a finitely generated torsion module over A. We intend to give an elementary proof of the classical result sta- ting that M can be written as a direct s~ of cyclic submodules. Let J be the ideal of A, annihilator ~f M. Since A is principal we have J = (a ). Let n a = IT i=1 r. E N 1 be a factorization of a in A. The elements Pi are irreducibles in A and furthermore Pi is not associated to P j if i ; j . . If, for any irreducible element p in A we denote with M the p-primary compo- p nent of M, that is M = {x I x E M and pix = 0 for some i E N} p we have that M splits in a direct sum of its Pi-primary components. We Can therefore ,assume that M = Mp for some irreducible element p in A. For any m E M, m , 0, we consider the integer oCm) as the maximal positive integer i satisfying: We call oCm) the order of m. It is ~lear that if m has order s then p. generates the ideal i~ A, .nnihilator of m, so (m) ... A/< p.) 11 We have the following filtration of {m } : c {pm} c {m Each {pim } is characterized by the property: xE{m} , iff s-i x E {p m In fact, "if" is trivial. On the other hand if pix 0, as x we have 0 = pix = pirm. This implies pS I pir, that is, and we get finally x E {ps-im }. s-i P rm r Let m E M be an element of maximal order in M. We shall prove the key result that {m} is a direct summand of M. To do this we shall define a projection of M onto {m } . The next Lemma is the main de- vice of our proof. LEMMA. Let N eM be a submodule of M and let f: N --+ {m,> be a morphism. Then f extends to a morphism of Minto {m } . Proof. If N = M we have nothing to do. Let N I M and take a E M-N. Let I be the ideal of A defined by I {k IkE A and ka E N} Then I = (d) for some d in A. Since pSa = 0 (by the maximality of 5) we have that pS I d , or pS = d.y. By the unique fa~torization of A is d pr.u with r EO; s . Without loss of generality we can. as- sume that u = 1 , that is d = r p We have p s-r f(da) = f (ps-rda) . f(psa) 0 but by an earlier remark we can write We intend to Let us. prove have f (da) = prx for some x E {m .} extend f to N + {a as follows: f' : N + {a ~ ( m. ) n + ta t----+ fen) + tx that f' is well defined. WIth n,n' E N, t,t' n + ta = n' + t'a * n - n' = (t' - t)a * t' - t e I .. t' - t • zd z e A E A we 12 Therefore f (n - n') f(zda) zf(ad) zdx (t' - t)x that is fen) + tx = fen') + t'x and this proves that ~' is well defined. We can repeat this pro- cess, but after a finite number of steps we have to arrive to M (M is a noetherian module !). This concludes the proof of the Lemma. Applying the Lemma to the situation: N = (m ) , f = 1d< m )' ,,!e get a morphism of Minto < m ) which is the ident:j. ty over < m ) . This is clearly a projection over (m ). So ~m) is a dire.ct summand of M: M = <m) ED M' .We can repeat the process with M'. But by the noe- therian property of M, the process must stop after a finite num- ber of steps. Then we ~et that M is a direct sum of cyclic sub- modules. REMARKS. Notice the revealing property of elements of maximal or- der. The present proof improves the one given in our Note appea- red in the American Mathematical Monthly, Vol.76, N°1. pp.60-61 (1 !;I69) • Recibido en julio de 1974 Universidad de Buenos Aires Argentina Revista de la Union Matematica Argentina Volumen 27, 1974. SEQUENTIAL ESTIMATION OF A TRUNCATION PARAMETER Eduardo Warner deWeerth 1. INTRODUCTION. Let Xl be a random variable defined on a probability space (ll,B,P) and whose distribution belongs to the family {Pa: a E 8} where e is a finite or infinite interval of the real line R. We want to estimate the true value of the parameter a on the basis of a series of independent observations XI ,X 2 , ... of the varia- ble Xl' After taking each observation we decide, on the basis of the observations that we have already taken, whether to take still another observation or to stop and estimate. We suppose that it costs us c units to take an observation and that if we estimate a by d when its true value is a we lose L(a,d). Our to- tal loss then, upon taking n observations and deciding d is L(a,d) + cn. A sequential decision procedure is a pair (8,t) where t is the stopping rule which tells us, for each possible sequence of obse~ vations x = (x I ,x2 , ... ), when to stop, and 8 = {8 n : n = 1,2, ..• } is a sequence of terminal decision functions. The meaning of 8 is as follows: for each n, if we stopped after taking n observa- tions (t(x) = n) and have observed x f ,x 2 , ••• ,xn ' we must decide a = 8 n (xl' ... ,x n ). If we use the procedure (8, t), our loss as "a function of x E ROO is given by: L[a,8 t (x)(x)] + ct(x) if a is the true value of the parameter. The average loss is called the risk and is given by: R(8,t,e) The most desirable sequential procedure would be one that mini- 14 mizes the risk uniformly in e. Such a procedure. however. is unat- tainable in all but trivial cases. In the Bayesian setup we assume that there is not one true value of the parameter but rather that e is a random variable with a distribution 'It. The variables X1 .X2 •..• are then assumed to be i!!, dependent and identically distributed given e. This determines the jo~nt distribution of e,x1,x2"~' . A procedure will be optimal in the Bayes sense (Bayes procedure) if it minimizes the Bayes risk: , fe R(6.t,e)'It(de). Assuming that Pe(dx1) = f(x1.e)p(dx1). where p is a a-additive measure. and 'It(de) = tjI(e)de, .the Bayes risk for a sample of fixed size n is: (1.1) J {J L[e,6 (x1.···.x )]f(x1.e) ... f(x .e)p(dx1)···Il(dx )}. e Rn n n n n • 1jJ(e)de + cn = J {f 1[e,6 (x1' .. ·.x )]tjI(slx1.···,x )de}Q (dx1.···.d:x: )+cn Rn 9 n n n n n is the condition al distribution of e given 11 •... rXn and Qn is the marginal distri bution of Xl'" .• Xn' Suppose that for each n there exists a measurable function of the observations, en = en(X'l •...• Xn).such that (1 .2) J L(e.e )tjI(elx:1.···.X )de 9 n n min f L(e .d)1jJ(e'\X 1 ..... X )de d . '9 · n From (1. 1) and (1. 2) it follows easily (cf.[ 1]) that the seq1,lence of terminal decision functions of a Bayes procedure must be: {e }. n The corresponding stopping rule is the one that minimizes: (1.3) 15 where Q is the marginal distribution of (X1'X Z ",,) and f 1(1),6 (X1'''''x )].(e\x1, .. ·,x )de 8 n n n is called the posterior (Bayes) risk after n observations. Bickel and Yahav introduced in [2] the notion of asymptotically pointwise optimal (A.P.O.) stopping rules. A stopping rule t Y +Ct will be called pointwise optimal if Q[ t .;;; 1] = 1 for any Y~+ct' other stopping rule t'. Pointwise optimal rules exist only in essentially deterministic situations. One such case obtains when there exists a random va- riable V, 0 < V < = such that (1. 5) Y n V n ' n=1,2, .... In this case it is easy to see that the stopping rule: "stop as soon as _V-,--_.;;; c" n (n+ 1) is pointwise optimal. In what follows we will call a function from the interval (0,=) to the set T of all possible stopping rules for the sequence {Yn } also a stopping rule. Then, a stopping rule t(c) is call~d A.P.O. if Yt(c)+ ct(c) lim ----~-~-~------ c+O inf{~ +cn: n=1,2, ..• } n . In their paper [3] Bickel and Yahav proved. THEOREM 1.1. If (1.6) Yn> 0 a.s. Q fo~ all. n a.s. Q (1.7). nBYn -- V a.B. Q. whe~e B > 0 and 0 < V <. a.s. Q then the stopping ~uZ.e t(c):stop fo~ the fi~stn ' suoh that (1.8) Y [1 - (~)B] ~ c n n+1 16 is A.P.O. REMARK. The theorem as stated in [3J is somewhat more general. It is not hard to see that any rule s(c) such that tltl_ iCc) a.s. Q is also A.P.O. In particular, the rule t'(c) defined by: "stop for thefirst n such that Y __ 13_" c" is A.P.O. The rule t' (c) n n+l is obtained when we replace in (l.a) the expression 1 - (-1!.....)t3 . by its firs t order approximation _13_ • n+l n+l The relation between the deterministic case (1.5) and (1.7) is clear. Bickel and Yahav proved furthermore (cf. also [4J) that the rela- tion (1.7) was fulfilled (in the case of quadratic loss" for 13=1) under very general conditions that may be described roughly as those insuring the existence and asymptotic normality of maximun likelihood estimators. It is the purpose of this work to extend the results of Bickel and Yahav to a case that clearly does not satisfy the above conditions, the case of the estimation of a truncation parameter. The model we consider is as follows. Let h be a strictly positive and continuous function on the open interval a = (a o '=)' Here ao can be finite or -=. For each a E a we assume that c(a) = f: h(x)dx < = o Then, for each a E e" the function ~, 0 " x " a, is the den· C (a) sity function of a probability distribution Pa concentrated on the interval (eo,a), and we seek to estimate the truncation point a sequentially. In section 2 we prove that (1.7) (with 13 = 2) holds in this model for loss functions of the type L(a,d) = B(a)(a-d)2 where B(e) is a positive, continuous function of e such that Je(1+e2)B(e)~(e)de < =. It then follows from the results uf Bickel 17 and Yahav that the stopping rules: are A.P.O. t(c): stop for the first n such that t' (c): stop for the first n such that Y _2_,.;; c n n+1 In section 3 we show, using a theorem of Bickel and Yahav [3], that the rules t(c) and t'(c) are also asymptotically optimal in the sense of Kiefer and Sacks [5] . It is interesting to remark that in our development the function ~ plays a role analogous to that played by the Fisher informa- C (e) tion (or information matrix) in the work of Bickel and Yahav. It is clear, furthermore, that if eo is finite we can always as- sume that eo = O. Finally, we wish to remark that all our results, obtained for a distribution truncated above, can easily be translated into the corresponding result for a distribution truncated below. 2. A LIMIT THEOREM. We turn now to the truncation parameter model described in sec- tion 1 and use the same notation as there". We assume that the loss function is (2.1) L(e ,d) = B(e) (e-d)2 where B(e) is a positive and continuous function of e. The prior density ~(e) is a positive continuous and bounded function of 9 and we assume that B(e) and ~(e) are such that (2.2) I~B(e)(1+e2)~(e)de < ~ and o [ "e2~(e)de < .. . e o 18 We 4enote by an the Bayes estimator based on n observations an~ we will now show that it exists and compute its value. The c;ondi- tional distribution of a given X1""'Xn has the density (2.3) (w(e)c-n(e)I[a~en]);f; ,(A)C-n(A)dA n where Ia denotes the indicator of the set A and ~n = max(x1,···,xn ) is as before the M.L.E. based on n observa- tions. It follows from C1.1) that the Bayes .estimator is the one that attains the minimun in the expression Y F,rom n inf d Yn i:f f; Bce)ca-d)2.ce)c-ncalde/l; .CA)C-n(A)dA - n n = i:f f; ca-d)2(B(a).ca)Cn ce) I r; BCA)wCA)C-nCA)dA)Ii~. n n f; BCA),CA)C-nCA)dA n J; .cA)CnCA)dA n it is clear that (2.4) a n and therefore (2. S) We now prove f;a CB(9).(e)C-nca)/J; BCA)~CA)C-nCA)dA)da n n Y n J '" Bce)(e-e )2.ca lx1 ... ·,x )de e n n o LEMMA 2.1. en -+ e a.B. Pe for every a e e. 19 But Since the term on the right goes to Oas N -+- co the lemma is pro- ved. In what follows we assume throughout that 9 is fixed and that we are dealing with a fixed sequence of observations for which 9 -+-9. n ,.. -1,.. n THEOREM 2.2. Let "n(s) = (C(9 n )/C(sn +9 n)) . T hen h (9) (2.6) J"'(1+s 2)IB(Sn- 1+e )~(sn-l+a )" (s)_B(9)~(9)e-C(9)slds + 0 o n n n a s n -+ 00. Proof. We can write the integral in (2.6) as the sum of the fol- lowing two integrals, where 8 is a positive number to be determi ned later. no _h(9)s J (1+s2) IB(sn-1+e )Hsn-1+e )" (s)-B(9)H9)e C(9) Ids o n n n (2.7) (2.8) We first consider (2.7) (2.9) where e < ~ < sn- 1+ 6 n n If 0 ~ s ~ n8 and n > N(8), we get 9 - 8 < ~n < ~ < ~n + 8 < 9 + 8 Therefore (2.10) " (s) .;;; e-W8 where w = inf {h....(& : Ix-al < 6} n C (x) It is clear, furthermore. that we can assume 6 sufficiently small and N sufficiently large so that, in addition 20 (2.11) -1 A B(sn +e n ) < 1 + B(e) for 0 ~ 5 < n6 $nd p > N . It then follows that if K is the upper bound of 1/1, ~(e) (1+5 2 ) I B(sn- 1+8 n H (sn- 1 +8 n )"n (5) -B(e)l/l(e)e- c(e)"1 I( O(:u~6) (2.12) _hills .;;; (1+5 2 )[ (l+B(e))Ke-ws + B(e)1/I(e)e c(e) ]1 [ ou] Since it is clear from (2.9) that "n(s) it follows from (2.12) and the dominated convergence theorem that the integral (2.7) goes to 0 as n --+ m . We now consider integr~l (2.8). From (2.9) it is clear that ~here w is as before. Therefore, putting s = nee-a ) n h(e) (2.1~) Jm (1 +5 2) I B(sn- 1+a n) 1jJ (sn- 1+e n)"n (5) -B (e)1/I (e)e -c(e) S Ids .;;; nil + B(e)1jJ(e) Jm n5 ne-nowfm A 5+9 n + Due to (2.2) this last expression goes to 0 as n --+ m and ~he proof of the theorem is thus concluded. We state for later use 21 COROI.LARY 2.3. (2.14) (2.15) Proof. Immediate. THEOREM .2. 4. Pe as n - a>. Proof. It follows from (2.4) that n(e -~ ) = Ja> nee-a )B(e)f(e)C-n(e)de/Ja> B(l)f(l)C-n(l)dl n n ~ n ~ n n and making the substitution s = nee-an) We get n(e -e ) n n The theorem now fOllows from a direct application of Corollary 2.3. Proof. Putting s n(e-a ) n f a> - 2 . A B(e)(e-e ) Helxl, ... ,x )de e n n n J a> A - A 2 .. B (e)[ n(e-e ) -nee -e )] He Ix l , •.• ,x )de e n n n n n fa>B(Sn-l+e )[s-n(e -e )]2f (sn- l +e )v (s)ds o n n n n n 11: . j J:f(Vn- l +9 n)Vn (V)dV By Theorem 2.4. for n > N sufficiently large and Therefore 22 nee -0 ) < £1iL + 1 n n h(8) [ (ill.l+ 1) 2 + 2]B( -1 A ) ( -1 A) () < h(8) s sn +8 n ~ sn +en vn S. and it follows from Theorem 2.2, the dominated convergence theo- rem (see Loeve [6] p.162) and (2.15) (with B = 1), that as was to be proved. It follows from Theorem 2.5 and Theorem 1.1 that the stopping ru- les t(c) and t'(c) defined in section 1 are A.P.O. in our trunca- tion parameter model with loss functions of the type L(e,d) = = B(e)(e-d)2 and subject to the stated conditions on • an~ B. 3. ASYMPTOTIC OPTIMALITY. Following Kiefer and Sacks [5] , we say that a stopping rule tec) is asymptotically optimal if (3.1) lim sup[E(Yt(c)+ct(c))/in£{E(Ys(c)+cs(c)):s(c) ET}] < 1 c .... O where T is the set of all stopping rules. Then we have THEOREM 3.1. (Bickel and Yahav (31) •. Under the oondiHons of Theorem 1.1 and if (3.2) sup nBE(Y ) < • n n 23 then the stopping rules t(c) and t'(c) are asymptotically optimal. The following corollary is an immediate consequence of Theorem 3.1, upon applying (1.1) and the definition of Bayes risk (1.4). COROLLARY 3.2. (Bickel anq Yahav [3]). If the conditions of Theo rem 1.1 hold and if there exists a sequence of estimates 8 such n that then the ru les t (c) and t I (c) are asymptotically optimal. We now apply this corollary to our truncation parameter model and prove THEOREM 3.3. In the truncation parameter model hlith loss func- tion L(e,d) = B(e)(e-d)2, if the conditions of section 2 are sa- tisfied and furthermore (3.3) hiil > a > 0 for every e c(e) then the rules t(c) and t'(c) are asymptotically optimal. Proof. Integrating by parts, (3.4) lim A+e o -1 Making the substitution: x = ~+sn , we get q.S) zJe (e-x)(~)ndx e Cf8T o because by (3.3) -1 (C (Han») n C(e) 24 -1 en[log c(e+sn )-log c(e)] for every n(eo-e) < s < Q (e+sn- 1 < ~ < e). From (3.4) and (3.5) it fo1~ows that where K is a constant. Then n2J: B(e)Ee[ (e-s n )2]1/J(6)d6..;; KJ: B(e)1jJ(e)de < CD o 0 by (2.2). The theorem now follows from Co~ollary 3.2. 4. REMARK. An important corollary can be obtained from Theorem 2.2. Puttin~ B_1 in (2.Q) and disregarding the s2 we obtain: (4.1) _hills J " IHsn- 1+& )v (s) - Ha)e C(e) Ids -+ 0 as n -+ .. o n n If we now divide by (4.2) we get (4.3) ds '" -+ 0 a,s n -+ • Furthermore (4.4) 2S _.hills - ~ e c(e) Ids -+ 0 c(e) because of (4.3) and (2.15). as n -+ co The limit theorem for the posterior distribution embodied in (4.4) is an analogue of the Bernstein-Von Mises theorem (cf. Bickel and Yahav [4] , Theorem 2.2). ACKNOWLEDGEMENT. This research is part of my Ph. D. Thesis com- pleted at the University of California, Berkeley, under the gui- dance of Prof. Peter J. Bickel. The problem was suggested as well by Prof. Bickel to whom I wish to express my deepest gratitude. 26 REFERENCES [1] ARROW,K., BLACKWELL,D. and GIRSHICK,M.A., Ba.ye.~ a.nd m.in..una.x ~otuz.ion~ 06 ~e.que.nz.ia.t de.e.i~.ion p40bte.m~, Econometrica 17, pp. 213-244 (1949). [2] BICKEL,P.J. and YAHAV, J.A., A~ympzoz.iea.tty po.inzw.i~e. opz..una.l p40ee.du4e.~ .in ~e.que.nz.ia.t a.na.ty~.i~, Proc. Fifth Berk. Symp. Math. Stat. Prob. (1965). [3] BICKEL,P.J. and YAHAV,J.A., A~ympzoz.iea.tty opz..una.t Ba.yu a.n,., m.in..una.x p40ee.du4e.~ .in ~e.que.nz.ia.t e.~z.£ma.z.ion, Ann. Math. Statist. 39, pp. 442-456(1968). [4] BICKEL,P. J. and YAHAV ,J .A., Some. eonz4.ibuz.ion~ ZO zhe. a.~ymp ... zoz.ie Zhe.04Y 06 Ba.ye.~ ~otuz.ion~, Zeit. Wahr. verw. Geb. 11, pp. 257-276 (1969). [5] KIEFER,J. and SACKS,J., A~ympzoz.iea.tty opt.£ma.t ~e.que.nz.ia.t .in6e.4e.nee. a.nd de.~.ign, Ann. Math. Statist. 34, pp. 705-750 (1963). [6] LO~VE,M., P40ba.b.it.izy Zhe.04Y, 3rd. edi, n. Van Nostrand Co. Inc. (1963). Recibido en marzo de 1974 Universidad Nacional del Sur Bahra Blanca, Argentin~ En e.t eu4~0 de. tao .£mp4e.~.i6n de.t p4e.4e.nze. e.je.mpta.4 de. ta.Re.V.i4Za. de. tao UMA he.m04 4u64.ido tao .in6Mzuna.da. pl4d.ida. de. nue.4Z40~ eompa.ne.404 de. z4a.ba.jo y que.4.ido~ a.m.ig04 Eve.t.io T. O~ta.nde4 y R.iea.4do Ate.ma.ny, a. euya. me.mo~.ia. de.d.iea.m04 e.4Ze. z4a.ba.jo. Revista de la Union Matematica Argentina Volumen 27, 1974. EL GRUPO DE WITT DE CIERTAS CLASES DE ANILLOS DE ENTEROS Ignacio Kaplan y Horacio H. O'Brien En este articulo calculamos el grupo de Witt del anillo de ente- ros de una extensi6n finita de Q, con la condici6n de que 2 no sea una unidad, y que tal anillo sea de Bezout. En este trabajo tratamos de generalizar ciertos resultados de ( 2) • Sea entonces K una extensi6n finita de Q, A el anillo de enteros de K, tal que 2 no sea inversible en A, y A es un anillo de idea les principales. 1. Denotamos con Q(A) el grupo de extensiones cuadraticas de A; dado que 2 no es inversible en A, se tiene por [3) un epimorfis- mo dis:Witt(A) ---+ Q(A) ---+ 0 dis(P,q) = Z(Co(P,q)) siendo Z(C (P,q)) el centro de la parte homogenea de grado 0 del o 4lgebra de Clifford asociada. Sea M(A) el nucleo de tal homomorfismo;rsigue que el diagrama: 0 - M(A) ---+ Witt (A) ---+ Q(A) - o · 1 1 JI 0- N(A) ---+ H(A) ---+ Q(A) ---+ 0 1 1 0 0 es exacto y conmutativo, siendo las aplicaciones verticales las can6nicas y H(A) el grupo de isomorfismos de algebras de Clifford. 2.1. PROPOSICION. Sea (P,q) un A m6du~0 de rango 4 y disorimi- nan .te 1. Entonoes q admite un aero no trivia~ en toda oomp~eta taoi6n no arquimediana de K. 28 DemoBtraaion. Sea L una tal extensi6n de K yAel anillode v v valuaci6n correspondiente. Como L es una extensi6n finita de Q , para cierto primo p, se v p tiene que Br (Av) = O. [1] En particular N(Av) = O. Entonces (P ~ Av ' q 8 Av) es un Av m6- dulo cuadratico de range 4 y discriminante trivial, cuya algebra de Clifford es nula. Por 10 tanto (P ® Av,q ® Av) y obviamente tenemos la tesis. 2.2. COROLARIO. Si A no eBta aontenido en R, y (P,q) es un A moduZo auadratiao de rango 4 y disariminante triviaZ entonaes (P,q) es un eBpaaio hiperb6Ziao. Demostraaion. Dado que q admite un cero no trivial en toda com- pletaci6n, sigue que q admite un cero no trivial en A [4]. Por ser A un anillo de Bezout, (P,q) admite un espacio hiperb6- lico como sumando directo. [Z] Entonces (P,q) = H(A) 1 (P' ,q'), donde (P' ,q') es un A m6dulo cuadratico de rango 2 y discriminante trivial. Dado que la aplicaci6n can6nica Witt(A) ---+ Witt(K) es un mQ- nomorfismo, se tiene que (P' ,q') = H(A), de donde obtenemos la tesis. 2.3. COROLARIO. Si A eBta aontenido en R, y (P,q) eB un A modu- Zo auadratiao de rango 4 y diBariminante 1, entonaes (P,q) es un espaaio hiperboZiao si y soZo si (P 8 R,q ® R) es una forma reaZ semidefinida. 3.1. PROPOSICION. Si A no esta aqntenido en R entonaes Witt(A)- = Q (A) • Demostraaion. Es claro que basta demostrar que M(A) = O. Dado que todo A m6dulo cuadratico de range mayor 0 igual que 4 y discriminante 1 admite un cero no trivial en A, basta consi- derar el caso de range 2 y discriminante trivial. Pero en tal caso sigue trivialmente que (P,q) = H(A), Y conse- cuentemente tenemos 1a tesis. 4.1. PROPOSICION. Si A seta aontsnido sn R sntonaee ta aptiaa~ 29 aion aanoniaa M(A) ---+ M(R) es inyeativa. Demostraaion. Sea (P,q) un A modulo cuadratico de range 2n y di! criminante trivial, tal que (P ® R,q ® R) es nulo en WitteR). Si la dimension real de P & R es 2, el problema es trivial. Si tal dimension es 4, sigue que q ® R es una forma cuadratica semidefinida y por (2.3) tenemos el resultado. Finalmente si la dimension es mayor 0 igual a 6, se tiene que q admite un cero no trivial en toda completaci6n de K, y por 10 tanto admite un espacio hiperb6lico sobre A como sumando direc· to. Iterande el procedimiento tenemos el resultado, pues resulta ser (P,q) un espacio trivial sobre A. 4.2. COROLARIO. Si A esta aontenido en R entonaes existe una suaesion exaata de La forma: o -+ nZ ---+ Witt(A) -+ Q(A) -+ 0 donde n es muLtipLo de 4. Demostraaion. Es inmediataa partir del heche que M(R) 4Z. 4.3. COROLARIO. Si A esta aontenido en R entonaes N(A) 0 6 N(A) = Z2' Demostraaion. En efecte, N(A) es un subgrupo de Br2 (A) por [3] y en este case es imagen de un grupo ciclico. 4.4. PROPOSICION. Si N(A) sion e::r:aata o entonaesM(A) at y z'a 8uae- o -+ az ---+ Witt(A) ---+ Q(A) ---+ 0 8e parte. Dem08t:roaoi6n. Sea (P ,q) un A m6dulo cuadrhico pe'rteneciente a MeA). Si el rango de P es 2, sea iv1 ,v2} una base de P y sea II C za zbOl1 la matrizde dq en tal base, donde dq es la for- ma bilineal asociada a q. 30 Dado que 4ab-c 2 = -1, sigue que x q y por 10 tanto (P,q) = H(A). - 2v1 + (c+1)v2 es un cero de Si rango de P es 4, dado que N(A) = 0, sigue que C(P,q) = 0 en Br2(A);entonces por [21 es un A modulo cuadratico trivial. Si rango de P=6, sigue que q admite un cero no trivial en toda completaci6n de K, y por 10 tanto .10 admi te en A. Sigue que (P,q) = (P',q') 1 H(A) , donde (P',q') es un A modulo cuadratico de rango 4 perteneciente a M(A}. Por 10 anterior se deduce que (P,q) esun espacio trivial . Si n=8 2 0 - 1 0 0 0 0 0 0 2 0 - 1 0 0 0 0 - 1 0 2 - 1 0 0 0 0 0 - 1 -1 2 -1 0 0 0 0 0 0 - 1 2 - 1 0 0 0 0 0 0 - 1 2 - 1 0 0 0 0 0 0 - 1 2 - 1 0 0 0 0 0 0 - 1 2 es la matriz correspondiente ·a un modulo cuadratico de rango 8 y discriminante 1. [ 21 Por ser (P ® R,q ® R) un R m6dulo cuadratico definido, se tiene que (P ,q) -I 0 en Witt(A). La primer par t e del teorema se deduce entonces del diagrama exacto y conmutativo: 0- M(A) - M(R) 2Z L L 0 N(A) - N(R) Z2 1 L 0 0 Veamos que la sucesion se parte. Sea B un elemento de Q(A), y sea (P,q) un A m6dulo cuadratico de rango 2 tal que Z(Co(P,q)) = B. Dado que N(A) = 0, se deduce que es el unico m6dulo de rango 2 (salvo isometrias) con esa propiedad. En efecto, sea (P',q') un modulo de rango 2 con discriminante B. Sigue entonces que C(P,q)C(P',q') en H(A). Dado entonces que rango P = rango P' = 2, sigue trivialmente que 31 C(P,q)(l) = C(P' ,q')(l)' Y obtenemos el resultado parcial. En las condiciones anteriores 2(P,q) = 0 en Witt(A). Sigue que la aplicaci6n Q(A) ---+ Witt(A) B ---+ (P,q) definida anteriormente es un homomorfismo. Para esto es suficiente ver que si Bl Y B2 son dos elementos de Q(A) y (P1,ql) ,(P2 ,q2) son las imagenes respectivas, entonces (P1,ql).1 (P 2 ,q2) = (P,q) donde este modulo cuadratico es ima- gen de Bl x B2 . Dado que (P1,ql) 1 (P 2 ,q2) 1 (P,q) es un elemento de. range 6 en M(A) se deduce que es ' nulo en Witt(A). Dado que 2(P,q) = 0 en Witt(A) , tenemos la tesis. 4.5. COROLARIO. Si A esta aontenido en R, N(A) = 0 y Q(A) = 0, entonaes Witt(A) = 8Z. En partiauZar,se tiene eZ resuZtado de [21 : Wi tt (Z) = 8Z. 4.6. PROPOSICION. Si N(A) = Z2 entonaes M(A) = Z2' . Demostraai6n. De todas las consideraciones anteriores, se deduce que M(A) = 4Z 6 M(A) = 8Z. Si M(A) = 8Z, entonces el m6dulo cuadratico de range 8, (P,q) asociado a la matriz de la proposici6n 4.4 , es un generador de M(A). Veamos que esto no puede suceder. En efecto, dado que la aplicaci6n M(A) ---+ N(A) es un epimor- fismo, sea (P,q) un A modulo cuadratico de dimensi6n minima, tal que C(P,q) # 0 en N(Al. Sigue que (P,q) = n(P,q), 10 cual es un absurdo pues obtenemos que C(P,q) = nC(P,q) = 0 en N(A). Se deduce que rango P • 4, Y obtenemos el resultado. 4.7. PROPOSICION. Sea A aontenido en R y (P,q) un A m6duZo ou£ drdtioo de rango 2. taZ que (P e R,q e R) tiene determinante n! gativ,O . Entonoes 2(P,q) • 0 en Witt(A) • . Demostraoidn. Sigue trivialmente que (P,q) ! (P,q) es unespacio hiperb6lico en toda completaci6n de K. 32 Por 10 tanto 2(P,q) admite un cero no trivial en A. Sigue entonces que 2(P,q) H(A) 1 (P',q'), donde este m6du10 es de rango 2. Es faci1 ver que (P' ,q') H(A), Y tenemos e1 resu1tado. 4.8. COROLARIO. Si A esta aontenido en R, N(A) suaesi6n exaata Z2' 'entonaes 1.a o ---+ 4Z ---+ Witt(A) ---+ Q(A) ---+ 0 se parte si y s61.0 si no existen m6du1.os de rango 2 auyo deter minante sea positivo. B IBLlOGRAF lA' [1] AUSLANDER,M., GOLDMAN,O., The B4aue4 g40Up 06 a commu~a~~ve 4btg, (T.A.M.S.), vol.97, 1960. [2] LAROTONDA,A., MlCALl,A., VILLAMAYOR,O., SU4 te g40upe de WU~ (1972). [3] MlCALl,A., VlLLAMAYOR,O., SU4 te~ atg~b4e~ de Ct~6604d, Ann. Scient. Ec. Norm. Sup., (1968). [4] O'MEARA, Quad4a~~c F04m~. Recibido en julio de 1974 Universidad Nacional de Rio Cuarto Argentina Revista de la Union Hatematica Argentina Volumen 27, 1974. ON THE PARTITIONS OF AN INTEGER Raul A. Chiappa As is well known (see for example [11, [21) the following formu- la (I) gives by recurrence the number of "partitions" - the "par- tages" of French authors - of a positive integer n; i.e., the nu!!!. ber of non decreasing sequences of p6sitive integers whose sum is n. Formula (I) is 7f n I (_l)i-l(7f i~1 3 ·2 . l. -l. n--- 2 - where we take 7f n = 0 for n < 0, and 7fo = 1. (I) We present here a direct calculation - direct, in the sense that no generating function or Euler identities are used - Further- more, for the proof we introduce a general lemma, which seems to merit some attention in itself. Let ~j,n denote the number of such sequences whose first element is j (j ~ 1). Then, 7f n _ 1 ~l for n ~ 2; and 7f = L~. for ,a n j~IJ,n n ~ 1. Clearly,~. = 1 whenever j = n or [n/31 < j ~ [n/21 and J ,n ~j ,n = 0 when n # j > [n/21 ([ xl denotes "integral part of x"). The following array gives the non zero values of~. and 7f for J ,n n 1 <;n<;14. 2 3 4 S 6 7 8 9 10 11 1 Z 13 14 2 3 4 5 6 7 2 3 5 7 11 2 2 1 15 22 4 4 1 2 1 30 7 2 1 42 8 3 1 S6 12 4 2 77 14 5 2 1 101 21 6 3 2 3 5 7 11 15 22 30 42 56 77 101 135 34 For convenience, we extend the function ~ to all nEZ, setting ~. = 0 whenever j ~ 1 and n ~ O. We then have the equalities: J,n A) lT n _ 1 = ~l n , ~. J,n for n , 0 We now proveC. It is clear, when j , p, ~. }: ~h .. J,P h~j ,P-J Then, for p , q we have: ~ = }: ~ ~ + }: ~ ~ + ~ q-l,p-l h~q-l h,p-q q-l,p-q h~q h,p-q q-l,p-q q-p and for p = q, ~q,p = ~q-l,p-l = and ~ = O. q-l,p-q In passing, we note that Property C and the values~. = 0 for J,n j ~ 1, n ~ 0, and ~l 1 , ~. 2' = 1 for all j ~ 1, determine uni-J, J quely the function ~. The following formula (II) is equivalent to (I), and will provi- de us with our basic approach. IT = 2 IT - IT + n n-l n-3 }: (_1)i-l (A i~2 + A 3i2 -i n--- 2 - where As ",- . lT s - lT s _1 ' lT n = 0 for n < 0 and lTD = lTi 1. (II) Clearly, evaluating lT n - lT n_I from (I) provides (II). On the oth- er hand, evaluating with the aide of (II) the sum of the values lT n , lT n _ I , lT n _ 2 , ... we have (I). Let us see now that the evaluation of lT n (n,O) - i.e., the sum of the elements of the n-th column of the array ~ - can be redu~ ced to the evaluation of a difference: subs tract th~ sum of the values of ~ on the set L = {(j,n-j-1) ; j ~ 1} from twice the sum of the elements of the (n-1)-th column of the array ~. Specifically, for n#O and from A), B) and C) we get: 'If n }: ~J',n j~l Therefore, using again A), B) and C) we have: 7r n = 2 11' n-l - 35 7r - n-3 for n ~ 2 The use of Property D, see below, reduces the evaluation of (1) L ~o 0 1 to the consideration of the elements of the array for j~3 J,n-J- another set T, which will be determined implicitly by the resulting equations. First, it will be convenient to visualize in the following array the sets Land T, whose elements are represented by means of "+" and "0,, respectively. n-13 n-12 n-11 n-10 n-9 n-8 n-7 n-6 n-5 n-4 n-3 n-2 n-1 n 0 0 0 0 0 0 0 0 + j = 1 0 0 0 0 + 2 ·0 0 0 + 3 0 0 + 4 0 + 5 + 6 The above mentioned Property D is the following: L l~h~q-l D) ~q,p = ~1,p-(q-l) - ~ h,p-q For the proof, it suffices to reiterate (q-1) times Property C. From D, we obtain: L ~Jo n- Jo-l = L (~l n-2 Jo - ~1,n-2Jo-l) - L L ~h,n-2Jo-l (2) . j~3' j23' h~2 j~h+l For convenience, we regroup the terms as follows: 'PI, m L (~l m-2s - ~l m-2s-1) S20' , (3) for h ;;;. 2 From C, we obtain 'P 2 ,m = 'P1,m-l ' and more generally, from D 'P r ,m = 'P 1 ,m-r+l - L 'Ph,m-r for r ~ 2 (4) 2~hsr-l Also, from A), C), and (3) we get: 'P 1 •m- i + 1 - 'P 2 ,m-i .. .7r m-i - 7r m-i-l D. m-i (5) 36 The last equality and the following proposition will permit us to express <p , and thus L 1/1. • I ' in terms of llk (k <; n-7). r,m j~3 J,n-J- PROPOSITION. Fo~ r ~ 3, we have: <Pr,m II + L (-1) k II : m-r we:W m-r-(wI+w2+w3+"'+w~) r where W r is the set of sequences of positive integers w. (1 <; i <; k) J. such that Proof. For r=3, the set Wr is empty, and thus, from (4),{5) 'P -<P =b. l,m-2 2,m-3 m-3 For r > 3, the use of (4),{5) and an inductive reasoning gives 'P1,m-r+1 - 'P 2 ,m-r - L 'Ph,m-r 3~hsr-1 b. m-r L (b. + L (-1) t b. ) m-r-h ue:U h m-r-h-(u l +u 2+ .•• +u t ) 3sh~r-l where, for each h, we have h-1 ~ u l > u 2 > u3 > ... > u t ~ 3. Setting wI = h and wi = u i _1 for 2 <; i <; k = t+1, we can write: 'P ' = II r,m m-r + L we:W r where the w-sequences satisfy the required conditions. Having obtained this, we return to (2). If in this formula we sub- stitute j by (3+s) in the indexes (1,n-2j), (1,n-2j-1) and j by (h+1+s) in the indexes (h,n-2j-1) for h ~ 2, we obtain: Finally, using (3),(5) and the proposition above: For h=l, the set Wh is empty, then: (6) where 37 with h-1 ~ wl > w2 > w3 > ... > wk ~ 3 for any h ~ 4. The only remaining task is now to evaluate S. This task is simpli- fied by using the following Lemma. Although we need it only for a=3. b:;l. we will present it for any a > O. b ~ O. LEMMA. For given arbitrary integers a > 0, b ~ 0, and every inte- ger s ~ a, tet Ws denote the set of subsetsof the set {a.a+l •...• s}. Then, if r is any funation for whiah there is some ~ suah that x > ~ impties rex) = 0, we witt have: R = s~a we:W s (-1)1 wi r(a(s+b) + r(a 2+ab) + L (-1)k{r(t(3(k+a)2 k~l + r(lC3(k+a)2 + (k+a) + a(2b-a-1)))} L W.e:w ~ (k+a) + a(2b-a-1))) + (8) Proof. The sums in (8) are finite. as follows easily from a > 0 and the above conditions on the support of r. For fixed s. we shall proceed to associate to each w E Ws - with one exception - an element w· E W .• where j satisfies either J s+l or j = s-l and in such a way that a(s+b) + L" Wi a(j+b) + W.e:w ~ L w!. with cardinals Iwl and Iw'l differing w~ EW' 1 ~ in one. Therefore. as r(a(s+b)+L Wi) is equal to r(a(j+b) , +L wi). these terms c~n be omitted in the evaluation of R. since they appear with opposite signs. Hence. for fixed s. the sum L (_1)1 wi r(a(s+b) + L Wi) will be reduced to a single term. we:W w.e:W S ~ To accomplish this. we start by identifying each w E Ws' w ; 0, with the decreasing sequence of its elements. That is to say, if k is the number of elements in w (w E We)' we will have w = C"':l.w2 ;"w3 ' ...• wk), with s ~ wl > w2 > w3 > ... > wk ~ a. Denoting by: Wl = {w / w s a} 38 w2 {w / w (WI' w2 ' ... , Wk) wk > a wI < s} s W3 {w / w (WI' w2 ' ... Wk) wk > a , wl = s} s we get: W = WI U W2 U W3 U {0 } , where 0 will denote the empty s s s s s s set as element of W s Clearly', W2 W2 2 are empty. a' a' and Wa+I We now define the rule of associat i on and, for the sake of clari- ty, we divide the problem in two parts. PART I. For fixed s ~ a, let as : W~ ping defined by: - 2 Ws+I U {0 s+I } be the map- Obviously, as is onto and one-to-one. We will then have : R = r (a(a+b) + I s<!a+I I 3(_1)lwl r(a(s+b) + I wi) WEW w. e:w s L because for 5 ~ a the terms in WI compensate with those in s 2 Ws+I u {0 s+l} , and for s ~ a+1 the same 6ccurs with W 2 H\} U s I Moreover, that in R appear, only the and Ws _I . we see now, terms corresponding to 0a and W 3 (s > a+1) . s PART II. First, let f, 1 ~ f ~ k, be the greatest integer such 3 that for WE Ws ' w = (w I ,w2 ' ... ,wk) we have wi = s - (i-1) whe- never 1 ~ i ~ f. For convenience, we set: wk = a+g and, since it wil1 be needed la ter, we single out the case f=k. Thus, the additional assumption g=f clearly implies s = 2k + a - 1 and reciprocally;. while g = f+1 implies s = 2k + a and reciprocally . That is to say, f=k=g if and only if (s-a) is odd, while f = k = g-l if and only if (s-a) is even. Thus, for both we have k = [s-2+ 11 , where [ 1 denotes the function "integral part of". Next, we consider in W3 two subsets:W4 , whose elements are the s~ s s quences such that wk = a + g ~ a + f but excluding that one with 39 f = k = g(if any); and w~, where wk = a + g > a + f, but e;x:clu- ding that one with f = k = g-l (if any). 455 Cl.early, Wa+I ' Wa+I ' · and Wa+2 are empty. The above considerations enable us to complete the rule of asso- ciation by defining the followirtg mapping: For fixed s ~ a+1, let 13 : W4 --+ WS+1 be given by: 888 13 8 (wI' w2 ' , wk )= (wi, wi, ... , wit_I) { w. + when 1 ,.; i ,.; g w! = 1. 1. when g+l ,.; i ,.; k-1 w. 1. where Clearly, I3 s is one to one. Moreover, it is onto, since for w' E WS 8+1 gives f3 8 (w) w. 1. w' = · (wi, wi, . ,. , wit), the sequence j w! - 1 if ,.; i ,.;f 1. = w! if f+1 ,.; i ,.; k 1. a + f if i k + w' . By means of f3 8' we see that for s ~ a+1 the terms corresponding to W: are equal to the terms corresponding to 5 W8+1 ' and for s ~ a+2 the same thing happens with the terms of WS and 4 8 W8_1 · Therefore, in the evaluation of R they compensate and hence can be omitted. Thus, for e:och W3 , only one element will remain for 8 evaluating R. Using this, we have the following reduced expres- sion: r(a2 (-1) k k R + ab) + L r(a(s+b) + L wi) 8~a+1 i-I where w. = s - (i-1) , 1 ,.; i ,.; k , and k = [ s-r1] 1. According to these restrictions, we have, for s = 2k + a - k a(s+b) + L w. - l (3(k+a)2 - (k+a) + a(2b-a-l)) i-I 1. and for s 2k + a (9) 40 k 1 2 a(s+b) + I w. - Z (3(k+a) + (k+a) + a(2b-a-l)) i=1 1 Replacing these expressions iri (9), we obt~in the asserte4 equali- ty (8). Returning to our problem, we substitute in the Lemma the values a=3, s=h-l, b=l, r(j) = t::. 3.' Then, the resulting value .o-f R is n- -J precisely the value of S given by (7). If moreover, we set k+3 = i, we obtain: S " I:' (_l)i-l {" = Un-IS + [. U h4 } 3i2+i n---- 2 (10) Finally, by replacing in (1) the results from (6) and (10), we obtain (II), as asserted. REFERENCES [1] M.HALL, Comb~na~o4~al Th~04Y, Ed. Blaisdell, 1967. [2] P.BERTIER, Pa4~ag~~, pa4~~~~, pa4~~~~on~, d~comp~~~ ~~ 4~p4~ ~~n~a~~on~. METRA, 1967, (VI-I) pg. 103-129. A\,;KNOWLEDGMENTS I wish to express my sincere thanks to Prof. A.Diego and M.L. Gastaminza for many stimulating discussions. Recibido en octubre de 1974 Universidad Nacional del Sur Bahia Blanca, Argentina En el CU4~O d~ la ~p4~~~6n d~l p4u~n~~ ~j~pla4 d~ la R~v~~~a d~ l.a UMA h~o~ ~U64~dola ~n604~unada pl4d'(;da d~ nuu~4o~ compaii~40~ d~ ~4abaj 0 y qU~4~do~ im~go~ Ev~l~o T. Okl.and~4 y R~ca4do Al~any, a cuya m~o4~a d~d~camo~ ~~~~ t4abajo. Revista de la Uni6n Matem~tica Argentina Volumen 27. 1974. CURVAS Y CUATERNIONES Luis A. Santa16 RESUMEN. Es bien sabido que las rotaciones alrededor de un punto, en E3 ' se representan biyectivamente por los puntos del es pacio eliptico S3. A una curva 'Y de S3 correspondera una familia de ro taciones dependientes de un parametro. El objeto de esta no ta es: 1° Calcular la curvatura y la torsion de la curva 'Y de S3 e nfu~ cion del vector V (velocidad instantanea) de la rotacion co rres- pondiente de E3 (formulas (4.15) y (4.18)). 2° Aplicar el resultado al caso de las "rotaciones de Fren et", 0 sea, a las rotaciones correspbndientes al triedro de Frene t (T.N,B) de una curva de E3 . Resultan as! los teoremas del final del trabajo. 1. INTRODUCCION: RESULTADOS CONOCIDOS. Sea E3 el espacio euclidiano de 3 dimensiones. Un cuaternion es un mlmero hipercomplejo (1 • 1 ) donde a i son num eros reales (componentes del cuaternion) y las unidades i. j, k cumplen, para el producto. las relaciones (1 .2) - 1 de las cuales ~e deducen (1 .3) i = jk = -kj j = ki = - ik k = ij = -ji de manera que los cuaterniones forman un algebra asociativa , no conmutativa. La primera componente ao se llama la compo nente es- calar y las aI' a 2 • a3 las comp onentes vectoriales. A todo cuaternion a asociamos el vector 42 (1.4) que tiene pOT componentes a 1 , a2 , a 3 respecto de una terna de vec tores ortonormales I, J, K. En todo 10 que sigue las letras mi- nusculas (sin indices) indicaran cuaterniones y las mismas letras, mayusculas, indicaran los vectores correspondientes. Reciproca- mente, a todo vector (1.4) haremos corresponderel cuaterni6n "puro" (con la parte escalar nula) que tiene por componentes veS toriales las componentes del vector. Observese que en esta corres pondencia, a los cuaterniones unidad i, j, k, corresponden los vectores de la base I, J, K. Representaremos por a* = aO-a1i-a2j-a3k al cuaterni6n aonjugado de a. La norma de un cuaterni6n es entonces (1 .5) N(a) * * a2 2 2 2 aa = a a = 0 + a 1 + a 2 + a3 y el inverso de a, es (1 .6) Se observa que es -1 a a* N (a) (1 .7) N(ab) = (ab)(ab)* Para un cuaterni6n puro es abb*a* (1 .8) u* -u N (u) = u (-u) Y si es de norma unidad, se cumple (1 .9) - 1 N (a)N (b) 2 -u Es bien sabido que una rotaci6n X -+ X' en E3 alrededor del ori- gen, se puede representar mediante cuaterniones por la expresi6n (1.10) -1 x, = q*xq = q xq donde x es el cuaterni6n puro correspondiente al vector X y q es un cuaterni6n de norma unidad. El cua.tern16n -q representa evidentemente la misma rotaci6n. Comodar q equivale a dar sus componentes qi ' que por ser q de norma unidad satisfacen la relaci6n q~ + q~ + q~ + q~ = 1,resul- ta que a todo punto de la esfera unidad de E4 corresponde una ro- 43 tacion de E3 y a puntos diametralmente opuestos corresponde la misma rotacion. 5e puede demostrar facilmente que esta correspon- dencia es biyectiva y como la esfera unidad de E4, con los puntos diametralmente opuestos identificados, es el espacio proyectivo P3, se puede enunciar: _ Las rotaciones de E3 alrededor de un punto, estan en corresponde~ cia biyectiva con los puntos del espacio proyectivo P3 . Mas exac- tamente, este espacio P3 es el espacio del grupo de las rotaciones de E3 alrededor de un punto. En el espacio proyectivo P3 se puede introducir la distancia ¢ entre dos puntos a, b por la formula (1.11) cos ¢ = t(a*b + b*a) ( a, b ) El espacio proyectivo P3 con esta metrica se llama el espaaio eLip- tiao 53' Tenemos as! una correspondencia biyectiva entre las rotaciones de E3 alrededor de un punto y los puntos de 53' A una familia de rotaciones dependientes de un parametro correspondera una curva de 53' Nuestro objeto es estudiar esta correspondencia. Para ello necesitaremos algunas formulas que vinculan los vecto res de E3 con los cuaterniones (que son los puntos de 53)' 5e introduce la notacion (1.12) y dos puntos se l1aman aonjugados (ortogonales) si su distancia es w/2, 0 sea (1.13) (a,b) = (b,a o 5i a,b son cuaterniones puros, es (1.14) (a,b) = A.B donde A.B indica el producto escalar de vectores de E3 . Entre 3 vectores A, B, C de E3 y sus correspondientes cuaternio- nes puros a~ b, c se puede comprobar la relacion (1.15) (A B C) = t (cba - abc) donde (A B C) es el producto mixto (A A B).C, 0 sea, el determinan 45 que son las f6rmulasde Frenet para las curvas del espacio elip- tico S3' Para futuros usos, conviene calcular ql, q2, q3 en funcion de qO q. Se tiene, por ser q de norma unidad (2.6) (2.7) De aqui (2.8) (q,q) = q*q = qq* = 1 1 q q' I( q , ,q ') ( q , ,q ') q It _ (q', q It) q , ( , ') 3/2 q ,q ( q ,q' ) o Para hallar q2 procederemos por coeficientes indeterminados, po- niendo (2.9) de donde, dadas las condiciones de ortonormalidad (2.1), resulta (2.10) ( , It) ( q It 1 ) = .:....:I.q --,-, .::I.q--,--"" ,q (") 1 / 2 q ,q ( qlt ,q ) y por tanto (2.11) 1 ( q , ,q It) /J (qlt - (qlt ,q) q - .:....:I._--'-_:l._--!- q') ( q' ,q ') El coeficiente /J se determina por ser q2 de norma unidad, resul- tando (2.12) ( q" ,q It) _ (q. q It) 2 _ (q'. q It) 2 (q' ,q ') La expresi6n de q3 es mas complicada y no la vamos a necesitar. 3. LA CURVATURA Y LA TORSION DE UNA CURVA DE 53' A partir de las f6rmulas de Frenet (2.5) se introducen las siguie~ tes definiciones (3.1) (3.2) (3.3) 46 ds = elementode arco d 2 wO 1 2 w 3 wO 1 k curvatura w= torsi6n Para calcular k a partir de la curva q que a partir de (2.5) se deduce (3.4) 1 1 2 1 2 w 2 = (dq ,q) = (q ',q) d t (q',q,)1/ 2 dt q (t) 0 · q (t), se observa [ ( 9 ' ,9 ') ( 9" ,q ") - (9', 9 ") 2 _ (9', 9 ')( g" ,9 ) 2] v (q',q,)3/2 [ ( 9 ' .9 ')( 9" , 9 ") - (9'.9')( 9 , 9 ") 2 - (g', 9 ") 2] 1 I 2 (q',q,)2 y por tanto (3.5) k 2 = (9',9')(9",9") - (9'.9')(9,9,,)2 - (9',9,,)2 (q',q,)3 Teniendo en cuenta que (q,q') = 0 ( q , q ") + (q' ,q , ) o ( q,q") -(q',q') se puede escribir tambien (3,.6) k 2 = (9', 9 ')( g" , 9 ") ( 9 , ! 9 ") 2 _ 1 (q',q,)3 Comparando (3. 5) con (2.12) resul ta (3.7) Para el c§lculo de w procederemos de manera indirecta. Sabemos que es q' = (q',q') 1/2 ql y pongamos q'" = ( .. ) gO + ( •• ) ql + ( •• ) q2 + 7jq3 , donde los va 47 10res de los parentesis no interesan. Por ser e1 determinante 1, sera (3.8) [q q' q" q'''1 = (q',q,)1/2" 11 E1 valor de " esta dado por (3.7). Para ha11ar 11 observemos que o y tambien (3.9) puesto que (ql, ,q3) = 0 por tener ql, solamente componentes se- 1 2 gun q, q , q . Por otra parte se tiene (3.10) w ,,( , ') 1/2 q ,q y comparando con (3.8) y ap1icando (3.7) resu1ta (3.11) [g g' g" g"'1 (q',q,)3 que, puesto que ya se conoce k, nos da e1 valor buscado de w en funcion de q y sus derivadas. 4. RELACIONES ENTRE LAS CURVAS OE S3 Y LAS FAMILIAS DE ROTACIONES EN E3 • Una rotacion en E3 a1rededor del origen, se puede considerar de- finida por 1a posicion de 1a terna ortonorma1 de vectores (El, E2, E3) transformada, por 1a rotacion dada, de 1a base ini- cia1 I, J, K. Por 1a rotacion definida por e1 cuaternion q = qO, los vectores Ei son los correspondientes a los cuaterniones (4.1) l" e = q* i q e 2 = q* j . q e 3 = q* k q o sea, 48 e l ( 2+ 2 2 2) . + 2 ( ) . qo ql-q2- q3 1 qlq2- qOq3 J + 2(qlq3+qOq 2)k (4.2) e 2 2(qOq3+q l q2)i + 2 2 2 2 . (qO-ql+q2- q3)J + 2(q2 q3-q Oql)k e3 2(Q1Q3- q Oq2)i + 2(QOQ1+Q2 Q3)j + 2 2 2 2 (QO-Ql-Q2+Q3)k Es decir, la terna ortonormal de vectores (E l , E2, E3) que defi- ne la rotacion correspondiente al cuaternion de norma unidad Q, es la de los vectores Ei que tienen las mismas componentes que los cuaterniones e i , dadas por las formulas anteriores. Consideremos l a familia de rotaciones correspondientes a los pun- tos de la curva Q(t) = QO(t). Por ser (E l , E2, E3) una terna or- tonormal, es Ei.dE i = 0, Ei.dEj + Ej.dE i = 0 (i # j) y por tanto los vectores dEi/dt se expresan en la base (E l , E2, E3) en la forma (4.3) Los coeficientes a, ~, ~ tienen claro significado geometrico. Ellos son, efectivamente, las componentes respectivas de la base movil (E l , E2, E3) del vector potaai6n instant&nea (4.4) H El nombre proviene de que el movimiento, en cada instante, es una rotacion alrededor de H, puesto que dEi/dt H 1\ Ei. Teniendo en cuenta que las componentes de Ei son las de e i en (4.2) , resulta a dE 2 E3 2(Qbql qiqo + QiQ3 - Qj q2) - dt (4.5) fJ dE 3 El 2(QOQ2 - Qi Q3 - QiQo + QjQl) dt ~ dEl E2 2(qOq3 + Qi Q2 QiQl - Q3 QO) - dt De aQui se deduce 49 2 qo a ql + (j q2 + 'Y q3 2 qi -a qo - (j q3 + 'Y q2 '(4.6) 2 qi a q3 -(i qo - 'Y ql Estas ecuaciones pueden condensarse en (4.7) -2 q' :0 v q v :0 -2 q' q-l donde (4.8) v :0 ai + (ij + 'Yk es el cuaterni6n correspondiente al vector (4.9) v :0 aI + (jJ + 'YK Nuestro objeto es c~lcular ahora la curvatura y _la torsi6n k, w de la curva q(t;) de S3' en funci6n del cuaterni6n v (4.8) cuyas componentes a, (i, 'Y tienen significado geometrico en el espacio de las rotaciones E3 . Para ello, de (4.7) se deduce (4.10) q':o_1. vq 2 q" -(1/2)(v'q + vq') - 1. (v'q - 1. v v q) 2 2 q'" 1 (v"q v'vq 1 v'q + 1. v v v q) - - -v 2 2 4 q'* 1 q* v* 2 q"* (q*v' * 1 - q* v* v*) 2 2 q"'* - 1.(q*v"* q*v*v'* 1. q*v'*v* + 1 q*v*v*) 2 2 4 Introduciendo el vector V definido por (4.9), tenemos (4.11) v2 :0 v*v :0 V v* VV' :0 (v*v' + v'*v) 2 50 que son, ambas, expresiones escalares, a sab~r: y2 = a2 + p2 + ~2. Y Y' = aa' + PP' + ~~'. Por otra parte, siendo v un cuaterni6n puro, es (4.12) v + v* = 0 v v = v* v* y por ser q de norma unidad, es (4.13) q*q = qq* = q*'q + q"'q' = 0 ,(q',q) Con estas relaciones, de (4.10) se deduce f~cilmente (q',q,)=ly2 4 con 10 cual queda (4.14) ( q" ,q") o sea, en notaci6n vectorial (4.15) (y2)2 + ly,2 ,(q',q") 16 4 Para hallar la torsi6n w, aplicamos la relaci6n (1.16), (q,q') =0 1 y Y' 4 (4.16)[q q'q"q"'] = l(q"q'*qq"'*+q"'q"'q'A"*-·qq'*q"q",*.q"'q"*q'q*) 4 Teniendo en cuenta los valores (4.10) y las relaciones (4.11), (4.12) y (4.13), despues de un c~lculo un poco largo, pero sin artificios, resulta (4.17) [q q'q"q"'] (Y Y'Y") + 1 ((Y y,)2 _ y2 y,2) 8 16 donde en el segundo miembro se trata de operaciones con el vec- tor Y. Por tanto, teniendo en cuenta (3.11) y (4.14), resulta (4.18) [g 9 '9"9"'] (q',q,)3 de donde, conociendo k por (4.15), se puede expresar w en funci6n de Y y sus derivadas. 51 5. ROTACIONESDE FRENET. L1amaremos rotaaiones de Frenet a las rotaciones a1rededor de -un punto de E3 que corresponden a1 movimiento del triedro de Frenet (T,N,B) de una curva del espacio E3 . Esto quiere decir, que los versores T,N,B que definen e1 movimiento (tangente, normal prin- cipal y binormal de una curva) satisfacen a las ecuaciones de Frenet dT KN ds (5.1) dN - KT + TB ds dB - TN ds Es decir, comparando con (4.3) resu1ta que e1 vector V tiene las componentes a = T , P = 0 , ~ = K, 0 sea, es e1 vector V = TI + KK, siendo ",T 1a curvatura y 1a torsion de 1a curva de E3 • Segun esto, 1a curva de S3 correspondiente a una rotacion de Frenet, tendra 1a curvatura y 1a torsion dada5 por (5.Z) k Z(KT I - KIT) (,,2 + T2)3/2 y, si k # 0, entonces (5.3) w - 1 be aqui 5e deduce: 1 0 A Las heLiaes de E3 (KIT (k=o) de S3. Z d arc tan ~ ds K constante) aorresponden Las reatas ZO A Las aurvas de E3 que no son heLiaes. aorresponden aurOas de to~ sion aonstante (w=-1) de S3. Para estas aurvas. La aurvatura to- taL Ik ds es un muLtipLo de 4w (inaLuido 0). Podemos preguntar, ~cua1 es 1a condicion para que 1a rotacion de- finida por las ecuaciones (4.3) sea un movimiento de Frenet? Para e110 sera necesario y suficiente que exista un cambio de ba- se ortonorma1, sea 52 (5.4) con la matriz (a i h) ortogonal, tal que el versor rotacion instan- tanea H (4.4) no tenga componente segun AZ, 0 sea, se verifique AZ.H = 0, 10 que equivale a (5.5) o de donde se deduce (puesto que las a i . son constantes) J (5.6) y para que este sistema (5.5), (5.6) tenga solucion debe ser (5.7) (V V'V") ex ' !3' 'Y ' o a: II (3 " 'Y" o Rec1procamente, si esta condici6n se cumple, el sistema (5.5), (5.6) tiene soluci6n y se tiene el versor AZ, que completado con AI, A3 de manera que formen una terna ortonormal, determinan un triedro de Frenet. Por tanto se puede enunciar: 3° La aondiai6n neaesaria y sufiaiente para que una rotaai6n de- finida por el sistema (4.3) sea una rotaai6n de Frenet, es que sea (V V'V") = 0, siendo V el veator (4.8), lo que equivale a la aondiai6n (5.7). Segun (4.18), la aondiai6n es equivalente a que sea w=-1, siendo w la torsi6n de la aurva de 83 asoaiada a la r~ taai6n, ~iempre que esta aurva sea de aurvatura k distinta de aero. BIBL IOGRAF IA [1) BLASCHKE,Wi1he1m, K~nema~~k und Qua~e~n~onen, Mathematische Monographien, Deutscher Verlag der Wissenschaften, Berlin, 1960. En este trabajo se citan otros de Clifford, Hje1ms1ev, Study re1a cionados con 1a representacion de rotaciones por cuaterniones y - por puntos de 53' Para problemas referentes a 1a integracion del sistema (4.3) ver tambien e1 cliisico tratado de DARBOUX, Th€.o~~e de~ S~~6ace~, Parte I, pggs.I-64, Gauthier-Villars, Paiis, 1914. Recibido en octubre de 1974 Universidad de Buenos Aires Argentina Revista de la Union Matematica Argentina Volumen 27, 1974. ACTIONS OF COMPACT, CONNECTED LIE GROUPS ON SPIN MANIFOLDS Cr;st;an U. Sanchez ABSTRACT. We study actions of compact connected Lie groups on Spin manifolds with only two fixed points obtaining results about the equivalence of the representations of the group on the tangent spaces at the fixed points. For semifree actions we get: Let X be a connected 2t-dimensional Spin manifold. Let G be a compact connected Lie group acting di,f fereptiably on X, with only two fixed points and semifreely. Then the two representations of G at the fixed points are equiv~ lent. 1. I NTRODUCT ION. As the title indicates our objective is to study actions of co~ pact, connected Lie groups on Spin Manifolds. We consider the specific case of an action with only two fixed points attemp- ting to obtain information on the relationship between the re- presentations of the group on the tangent spaces at the fixed points. The basic tools for this work are the results of Atiyah and Hirzebruch [3] (see (2.1) and (2.2)). In the case of a semi-free actions our results generalize a well known theorem · of Milnor ([ 1], p.478) for the particular case of connected groups. Our notation will be essentially from [1], [2] and [3]. We also indicate bY ,F(H,X) the fixed point set of H in X and by F(H,x,X) the connected component of F(H,X) containing the point x EX'. I wish to thank ·Professor Ted Petrie for his encouragement. 54 2. SPIN MANIFOLDS. Let X be a compact, oriented n-manifold with a Riemannian metric. Let Q be the oriented orthonormal frame bundle of the cotangent bundle to X. A Spin-structure of Xis a principal Spin(n) bundle P over X with a covering map w: P --4 Q of degree 2 such that the diagram P x Spin (n) -----+1 P 1 w x A w 1 Qx SO(n) ---_I Q co~utes, where A: Spin(n) ----+ SO(n) is the standard double covering and the horizontal arrows are the actions on the bun- dles. A compact, orientable manifold with a Spin-structure is called a Spin Manifold. If we have an action of a compact Lie group G on X (n=2t) by orientation preserving isometries and an action of G on P com- muting with the action of Spin(2t) from the right and compa- tible with w we can consider its Spinor-index Spin(G,X) which is by definition an element in R(G) ([1), p.479, [3], p.1S). The evaluation of Spin(G,X) in an eleme~t g of G is denoted by Spin(g,X). We shall indicate now two results from [3] and one from [11 which are necessary for our work. (2.1) THEOREM. (Atiyah-Hirzebruch) LetX be a aonneated Spin Manifold of dimension 2t and G a aompaat aonneated Li"e group aating non triviaHyorz X and aating on its Spin-struature P. Then Spin(G,X) = O. (2.2) THEOREM. (Atiyah-Hirzebruch) Let X be a aompaat, aonnea- ted, orientable differentiable manifold with Riemannian metria , on whiah a aompaat aonneated Lie group G operates effeatively by isometries. Suppose P is a Spin-struature for X. Then there exist aanoniaa Hy a Lie group Gland a homomorphism h: G 1 - G, whia~ is either identity 01' ~ double a6vering, suah that G1 aats on P induaing via h the given aation of G on X. (2.3) THEOREM. (Atiyah-Bott) Suppose that f:X - X is anisom£ 55 try of the 2t-dimensional compact oriented manifold X with only isolated fixed points x. Suppose further that X admits a Spin- structure P and that f has a lifting f 'to this Spin-structure. The Spin-number Spin(f,x) is then given by the expression Spin(f,X) = L vex) where x ranges over the fixed points of f and vex) ± (i/2)t 1r~=lt cosec(8 .12) J= J where 8 1 ".8 t is a coherent system of angles for dfx ([ 1] ,p.473). We can prove now (2.4) THEOREM. Let X be a connected 2t-dimensional Spin manifold and G be a compact, connected Lie group acting differentiably on X with only two fixed points x and y. Assume that each subgroup of prime order H of G satisfies the following condit i on: i) Either F(H,X) = {x,y} or F(H~x,X) = F(H,y,X) Then the representations of the group on the tangent spaces at the fixed points are equivalent. Proof. We can think that G acts effectively and by isometries. From (2.2) we obtain a Lie group G1 (which is connected by cons- truction [3], p.22) and a homomorphism h:G 1 --+ G, which is the identity or a double covering, such that G1 acts on P and indu- ces via h the action of G on X. The action of G1 on X is by orientation preserving isometries and its action on P commutes with the action of Spin(2t) from the right and is compatible with 1r. We can consid,er then the Spinor-index of the action of G1 on X and as a consequence of (2.1) we have Spin(G 1,X) = O. Since we want to prove the equivalence of the representations and the elements of prime order form a dense set in G, it is enough to prove the equivalence of the restrictions of the re- presentations to each cyclic subgroup of ,prime order p. Let H be a cyclic subgroup of prime order p;2 of G with generator f and assume that F(H,X) = {x,y}. Put Hl • h",l(H), then h-l(f) .. {f1,f2} (we could have one or two elements inthis set). Take f l , the~ Sign{fl,X) • 0 and we can apply (2.3) to obtain a condition on the coherent systems of angles of df and df . This condition will suffice to prove the x y equivalence of the representations. 56 Let us identify H with Z and f with the preferred generator. The p expression of vex) in (2.3) can be written as (2.5) v (x) = ± w J . J J . = t [ exp (i8 . /2) - exp ( - i8 . /2) ] j=l (1 - exp i8 j )(1 - expC -i8 j ))' Note that exp(i8 . ) ~ 1 since x is isoiated fixed point of f. J Let us put now hk exp(-i8 k/2). We have (2.6) vex) =± w~:~[h~h~ 1] Now the condition Sign(f1.X) = 0 gives vex) + v(y) turn implies 1 v (y) 12. 1 v (x) 1- 2 = 1. o which in In Iv(x)1 2 we have all the eigenvalues of dfx ' therefore (2.7) where aX i .s the number of eigenvalues of df which are equa;t to 6 X exp(iw/p) ) and Z* is the set of congruence classes mo- p dulo p which are prime to p. We have then (2.8) Put now a 6 (2.9) aX and wri te 6 sa -a w h S (h 26 _ 1) 6 6e:Z* where q ~ 6e:Z* p p sa S Clearly q E 0 (mod p). then (2.9) can be written as (2.10) (2.11) w 6e:Z* p = -a 6 -a (h 26 _ 1) 6 = 1 and wr.i te Since Ib s (2.12) o we have 1r se;Z* p 57 Now by a theorem of Kummer ([ 1], p.477) we have b s and then a = o. 0, s E Z* P s We have proved then the equivalence of the restrictions of the representations to H in the case F(H,X) = {x,y}. If this is not the case then F(H,x,X) = F(H,y,X) and in this situation the equivalence is obvious. The theorem is then proved. (2.13) REMARK. If the group G is a torus condition (i) of (2.4) can be replaced by i) Either dim F(H,X) = 0 or some of the subspaces F(H,x,X), F(H,y,X) has . positive dimension. (dim F(H,X) means the maximum of the dimensions of the components). If the action of G is semifree, that is free outside the fixed point set and F(G,X) = {x,y}, the theorem can be written as (2.14) THEOREM. Let X be a aonneated 2t~dimensional Spin mani- fold. Let G be a aompaat aonneated Lie group aating differen- tiabZy on X, with only two fixed points and semifreely. Then the two representations of G at the fixed points are equiva- lent. REFERENCES [1] M.F.ATIYAH and R.BOTT, The Le6~ehetz 6~xed-po~nt theo~em 6o~ eLL~pt~e eompLexe~ II, Ann. of Math. 88, 451-491. [2] M.F.ATIYAH and I.M.SINGER, The ~ndex 06 elL~pt~e ope~ato~4 111, Ann. of Math. 87, 546-604. [3] M.F.ATIYAH and F.HIRZEBRUCH, Sp~n-man~6oLd4 and g~oup ae- t~on4. Essays on Topology and Related Topics. Memoires de- dies a G. de Rham. Springer Verlag 1970, 18-28. Recibido en agosto de 1974 I.M.A.F. Universidad de Cordoba Cordoba, Argentina INDICE Volumen 27, Numero 1, 1974 An elementary proof of the Jordan canonical form Enzo R. Gentile ..... 0'0.......................... 1 Actions on a graph Antonio Diego An elementary proof of the structure theorem of finitely generated modules over a principal domain 5 Enzo R. Gentile ..................... '. . . . . . . . . . . 10 Sequential estimation of a truncation parameter Eduardo W. de Weerth ........................... 13 EI grupo de Witt de ciertas clases de anillos de enteros 1. Kaplan y H. H. O'Brien .................. . .. . .. 27 On the partitions of an integer Raul A. Chiappa ....................... '.......... 33 Curvas y cuaterniones Luis A. Santal6 ...... j........................... 41 Actions of compact, connected Lie groups on Spin manifolds Cristian U. Sanchez ............................ .. 53 01 Reg. Nac. de la Prop. Int. NQ 1.044.738 o.S f~ ~~ 0" «: C; ~o I'I~ Q)" 0;:1 olrll [(it-> 0 TARIFA REDUCIDA CONCESION N9 9120 FRANQUEO PAGADO CONCESION N9 3626 AUSTRAL IMPRESOS VILLARINO 739 - B. B. NORMAS PARA LA PRESENTACION DE ARTICULOS Los articulos que se presenten a esta revista' no deben haber sido publicados 0 estar slendo considerados para su publicacion en oka re~ vista. Cada trabajo debera ser enviado en su forma definitiva, con todas las indicaciones necesarias para su hnpresion. No se envian pruebas de imprenta a los autores. Cada articulo debe presentarse por duplicado, mecanografiado a doble espacio. Es deseable que comience con un resumen shnple de su contenido y resultados obtenidos. Debe ponerse especial cuidado en dis- tinguir indices y exponentes; distinguir entre la letra 0 y el nfunero cero, la letra 1 y el numero uno, la letra i y la t (iota), EyE, etc. Los diagramas deben dibujarse en tinta china. Los shnbolos manuscri- tos deben ser claramente legibles. Salvo en la primera pagina, deben evitarse en 10 posible notas al 'pie. El articulo debera acompafiarse de una lista completa de los shnbo- los utilizados en el texto. La recepcion de cada trabajo se comunicara a vuelta de correo y en su oportunidad, la aceptacion del mismo para su publicacion. Los trabajos deben enviarse a: DIRECTOR DE LA REVISTA DE LA U.M.A. INSTITUTO DE MATEMATICA. UNIVERSIDAD NACIONAL DEL SUR. BAHIA BLANCA. NOTES FOR 'THE AUTHORS Submission of a paper to this journal will be taken to hnply that it has not been previously published and that it is not being considered elsewhere for publication. Papers when submitted should ' tis, il), final form. Galley proofs are not sent to the authors. ., Papers should be submitted in duplicate, neatly typewritten, double spaced. It is desirable that every paper should begin with a shnple but explicit summary of its content and resUlts achieved. Speci~l care shoUld be taken with subscripts and superscripts; to show the difference bet- ween the letter 0 and the number zero, the letter 1 and the number one, the letter i and L (iota), E and E, etc. Diagrams should be drawn with black Indian ink. Symbols which have been inserted by hand shoUld be well spaced and cleary written. Footnotes not on the first page should be avoided as far as possible. A complete list of the symbols used in the paper should be attached to the manuscript. Reception of a paper will be acknowledged by return mail and its acceptance for publication will be communicated later on. Papers should be addressed to: DIRECTOR DE LA REVISTA DE LA U.M.A. INSTITUTO MATEMATICA. UNIVERSIDAD NACIONAL DEL SUR. BAHIA BLANCA. ARGENTINA.
Compartilhar