The periodic-Dirichlet problem for some semilinear wave equations

JOURNAL OF DIFFERENTIAL EQUATIONS 34&354 (1992) %, The Periodic- Dirichlet Problem for Some Semilinear Wave Equations K. BEN-NAOUM DPpartement de Mathematique, Universitk d’Oran, Oran, Alg&ie AND J. MAWHIN Institut MathPmatique, Universitc!de Louvain, B-1348, Louvain-la-Neuve, Belgique Received November DEDICATED TO THE 1. 6, 1990; revised February MEMORY OF LAMBERTO 12, 1991 CESARI INTRODUCTION The aim of this paper is to prove the existence and uniqueness of the solution for equations of the form Lu+Nu=f, (1) in a Hilbert space H, with L: dom L c H -+ H linear and self-adjoint, N: H + H a possibly nonlinear operator. First, by a direct use of the Banach contraction theorem, we are able to obtain simpler proofs and improvements of recent results of Smiley [13]. Applications are then given to the periodic-Dirichlet problem for multi-dimensional semilinear wave equations of the form u,, - Au +g(u) =A?, x), on rectangles of R” with sides commensurable with the time period. In the one-dimensional space case with space length incommensurable to the time period, we then show the equivalence between some number theoretical assumptions introduced by McKenna [9] with other ones used earlier in [6] and we improve some existence results of [9]. In this case, we use some results of the theory of numbers obtained by Naparstek in 340 0022-0396192 $3.00 Copyright 0 1992 by Academic Press, Inc. All rights of reproduction in any form reserved. THE PERIODIC-DIRICHLET PROBLEM 341 [lo] which are proved in a much simpler way in the Appendix. All the above mentioned authors are Ph.D. students of Lamberto Cesari whose pioneering work in the functional analytic treatment of the periodic solutions of semilinear hyperbolic equations and systems is well described in the survey papers [4, 51. 2. EXISTENCE AND UNIQUENESS RESULTS FOR LIPSCHITZIAN PERTURBATIONS OF SELF-ADJOINT OPERATORS IN A HILBERT SPACE Let H be a real Hilbert space with inner product (x, y) and corresponding norm 1XI = (x, x) ‘j2, L: dom L c H + H a linear self-adjoint operator, N: H + H a (possibly) nonlinear operator and fe H. We denote by o(L) the spectrum of L. For I$ o(L), we denote by dA the distance of 1 to o(L). As o(L) is closed, d, > 0. The following simple existence result, modelled on [9, Theorem 43 will be useful in the sequel. LEMMA 1. Assume that there exists 2 # o(L), u E [0, dA[ and v 2 0 such that the conditions (i) (ii) ]Nu+,Iu-NV-AU] <d, Iu-II], ]Nu+luI 6~ 1241 +v, hold for ail u, v E H. Then equation (1) Lu+Nu=f; admits at least one solution for each f E H. Zf (i) and (ii) are replaced by (iii) I NU + 1~ - (NV + no)1< p I u - u 1, then (1) has a unique solution which can be obtained, from any u0 E dom L, by the iteration process defined by Lu k+l-h+l= Proof f-(NuI,+hh Equation (1) is clearly equivalent to the fixed point problem in H u=(L-AZ)-‘[f-(Nu+Au)]=T>,u and, L being self-adjoint, l(L-2Z))‘I 505/96/Z-IO kEiV. =d,‘. 342 BEN-NAOUM AND MAWHIN Consequently, if condition (i) holds, T,: Ei --+H is a nonexpansive operator. On the other hand, for 1u / < R, we have, if condition (ii) holds, provided in which case T,: B[R] + B{R], if B[ R] denotes the closed ball in N of center 0 and radius R. It then follows from Browder-GGhde-Kirk fixed point theorem [15] that Tl has a fixed point in B[R], i.e., that (1) has a solution in dom L n B[R]. Jf condition (iii) holds instead, the same reasoning shows that TA is a strict contraction on H and the result follows from the Banach fixed point theorem. l Remark 1. The first part of Lemma 1 is a slight generalization of [9, Theorem 11, which corresponds to 1= 0 and hence covers situations where 0 E#e(t), i.e., L is invertible. A result similar to the case of assumption (iii) was considered in [8]. As an application of this Lemma 1 with i = 0, we can consider, like in [9] the existence of weak solutions of the following ~~odi~-Diri~hlet problem for a one-dimensional semilinear wave equation on IO,f&/B x IO,74 on CO,dJ;il (2) u,, - ux.r+ g(u) =ff4 xl u(t, 0) = u(2,71)= 0 u(0, x) - u(?&/!i, x) = 24,(0, x) - u, (7&b, x) = 0 on EO,~1, where g: R + Iw is continuous and f~ L2(]0, ~/fi[ x 10, rt[ ). A weak ~olu~~o~of (2) is some u~L’(]0, n/+%[ x IO, x[) such that n/ar - 4,)+(g(u) -f) #Ikcdt=0 s s Cu(h, 0 0 for all #E C’(CO, ~/fiJ x [0, z]) such that 4(t, 0) = &t, n) = 0 MA x) - cswfi, xl = 4,(0, xl - $w/Jz, on K4 dfil, xl = 0 on CO,n]. Denoting by L the abstract realization in H= L*(]O, n/fi[ x IO, n[) of conditions on the wave operator with the periodic-Di~~hlet 30, rc,/&[ x 10, R[, it is standard to show that L is self-adjoint and c(L)= (j2-2k2:jdY,,kdU). THE PERIODIC-~IRI~HLET PROBLEM 343 Thus 0 ff atI,) and, by the theory of Pell’s equation in number theory, each eigenvalue of L has an infinite multiplicity [ 111. As do = 1, we shall have, by Lemma 1, existence of a weak solution of (2} for eachSo L* if for some p f CO,l[, v 3 0 and all u E R. For example, the assumptions are satisfied by g(u) = SinMu)), where, for some R > 0 and 0 dp < 1, h is defined by h(u) = u if Iuj~R,~“u+(l-~)flifu>Rand~~-fl-C1)Rifuc-R. ff we now write cr(L f = { 1, : n E Z) with I.,, c A, + $, a direct application of Lemma I with I = (A,, 1+ n,)/2, so that d, = (I.,, 1- &,)/2, implies the existence of a weak solution of (2) for eachf E L2 whenever and for some O<fl< (;tn+i -&},C?, v>O and u, veU% Similar results hold for the case where ]O, =:,:“i[ is replaced by 1% kJ&E for some square free positive integers m and n. RESUKBFOR STRONGLY 3. EXSTIBCEAND ~~~~~E~ MONOTONE ~TURB~~ONS OF SELF-ADJO~T OPERATORS m A HILBERT SPACE The following consequence of Lemma 1 will cover caseswhere 0 E V(L), If rW; (resp. rW,t) denotes the set of negative (resp. positive) real numbers, we shall set d; = distf0, o(L) n iR1,f, with the convention d; = + co if g(L)\(O) c: 0;s:. The following result generalizes in several ways L:13, Theorem 3.1J. 344 BEN-NAoU~ AND MAWHIN THEOREM 1. Assume that 0 < d; < co and that there exist positive constants PO,p,, y,,, yl, 6,, S, such that the assumptions (1) (Nu-N~,u-~)),~o~u-uJ~, (2) /Ah--NV/ (3) (4) (JQ4 U)ZY, b12-&, INUldYl IUI +d,, 6p1/u-vl, are satisfied for all u, v, E H. If the following conditions hoid: (if (ii) 8:6d;13y y:<d;y,, then Eq. (1) has at least one solution for eachf E H. Zf conditions (1) and (2) hold together with the inequality (iii) j3: cd; #lo, then Eq. (1) has, for eachfe H, a unique solution which can be obtained by the iterative process defined by u0E dom L and kEN. Lu k+,+(d~/2)uk+,=f-(Nuk-(d~/2)Uk), ProojI For each 1~ 0, we have, using conditions (1) and (2), /Nu+~u-(Nu+h)~* = INu-Nv[2+2R(Nu--NO, u-u)+12~#-v~2 <((8:+up,+n2)lu-v12. Now, taking R= -d;/2, (3) we have d, = d;/2 and, by (i), /3:-d,-j?,+(d,-/2)*%(d&‘2f2=d;, and condition (i) of Lemma 1 holds. Similarly, l~u-(d~/2)u12=l~ul~-d~(~~,u)+(d~/2)2~u~2 < Cy:-d;yo+(d,-,‘2)*f /~1~+21!,6~ /uJ +S:+d;&, and hence, by (ii), there exists 0 < p < d; /2 and v > 0 such that INu-(d,-/2)ul~~~u~+t for all u&II, so that condition (ii) of Lemma 1 holds and the first conclusion follows. In the second case, it fohows from (iii) with the same choice of A and from conditions (1) and (2) that INu+;lu--(Nv+h)l <p Ju-UJ, THE PERIODIC-DIRICHLET PROBLEM 34.5 for some 0 < p -Cd&/2 and all U, U,E H, so that condition (iii) of Lemma 1 holds and the proof is complete for d; finite. 1 Remark 2. It follows immediately from assumptions (1) to (4) of Theorem 1 and Schwarz inequality that necessarily Remark 3. The proof of Theorem 1 is motivated by that of Zarantone110[ 141 in his pioneering work on monotone Lipschitzian operators. Remark 4. If d; = + co, then (Lu, u) 30 for all UE dom L and L is maximal monotone (see, e.g., [2]). Then it follows from a result of Browder [3] that (1) has a solution for each f~ H if N: H+ H is monotone, hemi-continuous, takes bounded sets into bounded sets and is such that When N: H--t H is a continuous gradient operator, Theorem 1 can be replaced by the following sharper result. THEOREM 2. Assume that N: H + H is a contjn~o~s gradient operator, that 0 < d; < co and that there exist positive constants PO,j?, , y, ,6, such that the assumptions (i) (ii) ~o~u-u~2~(Nu-Nv,u-o)~~pl INu-(d;/2)uI<y, lul+6, l~-uI* are satisfied for all u, v E EI. Zf d; is finite and the foZIowing conditions hold (iii) fll <d; (iv) y1 < 412 then Eq. (1) has at least one solution for eachf E H. Zf condition (i) holds together with the ineq~aZity then Eq. (1) has, for eachf E H, a unique solution which can be obtained by the iterative process defined in Theorem 1. ProojI AE 52, It follows from assumption (i) and [7, Lemma 1J that, for each INu+h--NV-Au1 <max(l/Z+p,l, lA+~ll)lu-vl (4) 346 BEN-NAOUM AND MAWHIN for all u, v, E H. Taking A = -d;/2, (iii), we have we have d, = d; 12 and, by condition so that the result follows from Lemma 1. 1 We apply this result to a periodic-Dirichlet problem in an interval for a semilinear wave equation in R”. Let cli (1 < i < n) be positive rational numbers, Sz= nr= r] 0, CliX[, A the Laplacian in R”, g: R + R, J= ] 0,2z[, f~ L2(J x Q). We consider the existence of weak solutions for the problem inJxQ u,, - Au +g(u) =f(t, xl inJxdQ u(t, x)=0 u(2n, x) - u(0, x) = z&(271,x) - u,(O, x) = 0 (5) in .G, i.e., the existence of u E L2( J x Q) such that IJxR Cu(~,,-A~)+(g(u)-f)~l dtdx=O for all 4 E C2(J x a) such that inJxfX2 #(&x)=0 &% THEOREM 3. x) - i(O, x) = 4, (271,x) - $, (0, x) = 0 Let C.Q =pi/qi in a. with pi and qi positive relatively prime integers (1 < i < n), and let 11p 11= nr= 1pi. Assume that there exist such that the assumption (9 B. G CM4 -&Mu - 41 G PI is satisfied for all u # v in R. if, moreover, one has (ii) lim suPlul+ m CsWul < IIP II-2T then problem (5) has at least one weak solution. If condition (i) holds with (iii) B,<IIPII~*~ then problem (5) has a unique weak solution. ProoJ Let H = L2(Jx a) with the usual inner product and norm, and THE PERioDlC-DIRICHLETPROBLEM 347 let L be the abstract realization in H of the wave operator with the periodic-Dirichlet boundary conditions in (5). Then L is self-adjoint and Consequently, OE(T(L) and d; 3 11p[j -‘. If we define N by (A%)(?,X) =g(u(t, x)), then N: H -+ H is a continuous gradient operator and p,~u-v~z~(Nu-Nv,u-~)~‘p, /u--u/2. (6) Without loss of generality (modifying J) we can assume that g(0) =O. Hence, by condition (ii), there exists y0 < I(p II-* and R > 0 such that for 1u 12 R, and, using also condition (ii) -(d,/2)<Bo-(d,/2)~Cg(u)-(d,/2)ullu ~Y0-~~,-/~~~II~/1-*-~~~/~~~:do/~, so that for some 0 < y, < d; /2 and all ( u 12 R, which easily implies assumptions (ii) and (iv) of Theorem 2. The case of a unique solution follows directly from conditions (i), (iii), and Theorem 2. 1 Repark 5. Theorem 3 improves Theorem 5.1 of Smiley in [ 133 which requires condition (i) with condition (iii) replaced by the stronger assumption 4. THE PERIODIC-DIRICHLET PROBLEM FOR SEMILINEAR WAVE EQUATIONS FOR SOMEIRRATIONAL RATIOS BETWEENTHE PERIODAND INTERVAL LENGTH Let us consider now the existence of weak solutions for the following periodic-Dirichlet problem for a one-dimensional semilinear wave equation u,r - u.xx-g(u) =f(h xf u(t,O)=u(t, n)=O u(0, x) - u(27c/a,x) = ~~(0, x) - u,(2n/a, x) = 0 on IO, &W x IO, XT, on EO,Wal, on LO,~1, (7) 348 BEN-NAOU~ AND MAWHIN where a is a positive irrational number which is not the square root of an integer, g: R --) Iw is continuous and f~ L2(]0, 27c/u[ x 10, R[ ). A weak solution of (7) is defined as for Eq. (2) with 2$ replaced by ~1,and we shall denote by L the abstract realization in H= L’(]O, 27$x(: x ]O, n[) of the wave operator with the periodic-Di~chlet conditions on ]O, 2n/n[ x JO, n[. Thus, L is self-adjoint and its spectrum o(L) is the closure of the set of the eigenvalues (j’- a*k’:j~ N,, k E N >. We refer to the Appendix for the concepts and results of number theory used in this section. We first recall a special case of [6, Theorem 11, already observed in [lo], which insures that 0 does not belong to the spectrum of L. LEMMA 2. The linear periodic-Dirichlet problem u,, - 4, =f(t, xl u(t, 0) = u(t, n) =o u(0, x) - u(Zn/ff, x) = U$(O,x) - u,(27$%,x) = 0 on IO, 27daCx IO, 71[, on CO,27+1, on CO,xl, has a weak solution for eachf~ L2( ]0,2n/a [ x 10,~ [) if and only if c, = inf ](~rm)~--n~~>O, (m,nfcZx& in which case one has dist(O, o(L)) = c, and IL-’ I= c;‘. THEOREM4. Assume that c1has a bounded sequenceof partial quotients. Then there exists E> 0 such that the problem (7) has a unique weak solution for eachf E H when the condition holds for all u, v E 88,u # u. Proof. It follows from our assumptions, Corollary of the Appendix and Lemma 2 that there exists s1= c, > 0 such that o(L) n ] -el, .sl [ = $3, and if we choose any 0 < E< ei, then E< d, = c,, so that the result follows from Lemma 1. g This result was already given in [9] under the slightly more restrictive condition that g is of class C’ and 1g’(u)! < E for all UE R. We can now use a result of Amann [ 11, for which a simpler proof based upon Cesari’s alternative method is given in [7, Corollary 11, to improve [9, Theorem 41. We shall denote by c,~,(Z,) the essential spectrum of L. 349 THE PERIODIC-DIRICHLET PROBLEM THEOREM 5. Assume again that 01 has a bounded sequence of partial quotients. Assume moreover that there exist real numbers a and b with a < b such that the following conditions hold. 6) Ca,bl~fl,,,(~)=k5; (ii) a<(g(u)-g(v)&-udbforallu, (iii) tlim influ, j a: Mu)/u), ~~82, u#v; lim wlul -t m Mu)lu)l i-74L) = 53. Then problem (7) has at least one weak solution for each f E H. Proof: We shall show that the conditions of [7, Corollary l] are satisfied. Assumption B in this corollary follows from conditions (i) and (ii). Letting g_. = lim inf gO l=l-m u and g, = lim sup gfi, /~/--lix. u it follows from condition (iii) that we can find 1, PE a(L) such that 11, p[ c=p(L), with p(L) the resolvant of L, and A<g- Gg, <p. Let /I > 0 be such that fi-=zmin(p--g+,g- --A). Then there exists R>O such that for all I u 12 R, and hence IF--yi<min(g+ +/3-+,T-g-+8) =y<q=dist(y,o(L)). The conclusion follows then from [7, Corollary 1] as clearly one has (2 + PIP E Ca,bl\G). I 5. APPENDIX:A PROBLEM IN NUMBER THEORY The existence theorems of Section 4 require some results of number theory. Those results can essentially be found in [lo] but we reproduce 350 BEN-NAOUM AND MAWHIN them here for the reader’s convenience, because of the lack of availability of [lo] and because our presentation is simpler. Let c(E [w\CI and let Q, be the quadratic form defined on Z x 22, by Q, (m, n) = (am)’ - n*. Following the discussion of [6] (which is easily adapted from the periodic-periodic case to the periodic-Dirichlet one), we want to determine a class of a such that IQ,(m,n)l2 c,> 0, for some c, > 0 and all such that QM(nr,n) # 0. Now, 1QE(O,n)j = n2 2 1 for all n E Z,, and hence we can restrict ourself to the (m, n) E Z, x Z, such that Q, (m, n) # 0, i.e., to all (m, n)~i&xZ,, because, a being irrational, Q,(m, n) # 0 for (m,n)~Z,xZ,. As we can further assume, without loss of generality, that w.> 0 and (m, n) l No x N,. Define r, and r;, respectively, by Clearly, rl 6 r; and r; > 0 if and only if r, > 0. Indeed, if r; > 0, there exists R > 0 such that and, CIbeing irrational, IQll(m,n)l=Iam+nIIam-nI#O, for all (m, n) # (0, 0), and hence has a positive lower bound on the finite set {(m,n)#(O,O):Iml+lnl<R}. Let a = [a,, al, ....I THE PERIODIC-DIRICH~ET PROBLEM 351 be the continuous fraction decomposition of cc Recall that it is obtained as follows; put a,= [a], where [. J denotes the integer part. Then a=a,+ l/a, with a,> 1, and we set a,= [a,]. If ao,u,, .... a,_1 and al, a2, ..-, ~1,~1 are known, then a,_, = a,t. , + i/a,, with a, > 1 and we set a, = [a,]. It can be shown [ 111 that this process does not terminate if and only if a is irrational. The integers ao, a,, ,.., are the partial quotients of a; the numbers aI, a2, .... are the complete quotients of a and the rationals +,,,a, n ,...)uJ=uo+-&afi...~, 1 2 with p,,, q,, relatively prime integers, are the convergents of a and are such that pnlqn -+aasn+co. It is well known that the pn, qn are recursively defined by the relations The following lemma is useful for finding r;. To each irrational number a corresponds a unique (extended) LEMMA. number M(a) E [$, CC] hauing the ~ollowjng properties (i) For each positive number jt< M(E) there exist infinitely many pairs (p,, qi) with qi # 0, such that I a-&<1 9; ! ‘ii7 (ii) Zf M(a) is finite, then, for each ,u > M(a), there are on/y finitely many pairs (p,, qi) satisfying the inequality Proof Let It then follows from the elementary properties of the upper limit that M(a) satisfies the conditions of the lemma, with the exception of the estimate 352 BUN-NAOUM AND klAWHIN M(E) 3 $. But a weIl-known theorem of Hurwitz [ 111 asserts that for infinitely many pairs (pi, qi) one has so that the proof is complete. # If we set infinifel~~lnany (pi, qi) satisfy then the above lemma clearly states that M(a) = sup A’(a). PROPOSITION 1. M(CY) is finite if and only if the sequence (ai)ifN of partial quut~entsuf a is ~aunded. Proof: We have -I =4zr2 = lC4+1, ai+ I(-l)igi(cxi+lq;+qi-l)l -.I + CO,ai,ajwl, .... allI = lCai+,l+~i+?il, with 0 < oi, qi < 1 for all positive integers i. Thus, if (LZ,)~ EN is unbounded, one has limsup~i~limsup([ai+,]-2)=+c0, *-cc i-00 and M(cw)= co. If (ai)ieN is bounded, say, by A#, then M(a) = lim sup pi < lim sup ( [ai+, ] + 2) < co. a i-a, PROPOSITION i*m 2. Zf a E R + \Q, then r; = Za/M( a). Proof We have THE PERIODIC-DIRIC~L~T PROBLEM 353 and hence lim inf 1Q,(pi, qi)l =r 2a/M(s). i-4) Now let Jlr(a)={hMR,+: infinitely many pairs of integers (p, q) withq#OsatisfyIcr-(p/q)l<l/Mq*)=A’(a). It is known [l l] (see also the interesting paper [12]) that if M> 2 and A&EM(~), then ME&(~), and that, for each aE R\Q &~Jt(a). Thus, M(a) = sup M(a) = sup N(a), and hence, for p > M(a), only finitely many pairs of integers (p, q) with q # 0 satisfy the inequalities lQ.(~~4)/d1r-‘(a+(~/q))~~~‘(2a+(~/~lq~))~ which implies that Fe= lim inf I Q, (p, q)l - --&I > 2dl.l. lPl+lYl-+C= i Consequently, I‘: > 2a/M(rx), so that the equality holds. 4 Now, as I”: > 0 if and only if rz > 0, we also have the following COROLLARY. r, > 0 IY and only if M(x) < co, i.e., if and only if the sequence (txi),, N is hounded above. This corollary shows the equivalence between the number theoretical conditions upon a introduced in [6] and in [9]. The authors thank the referee for his careful reading of the manuscript. REFERENCES 1. Ii. AMANN, Saddle points and multiple solutions of differential equations, Math. Z. 109 (i979), 127-166. 2. H. Bdz~s, “Analyse fonctionnelle. Thkorie et applications,” Masson, Paris, 1983. 3. F. BROWDER,Problemes non liniaires, in “Sbminaire de Mathematiques Supirrieures,” Vol. 1.5,Pressesde 1’UniversitC de Montrkal, Montreal, 1966. 354 BEN-NAOUM AND MAWHIN 4. L. CESARI, Functional analysis, nonlinear differential equations, and the alternative method, in “Nonlinear Functional Analysis and Differential Equations” (L. Cesari, R. Kannan, and J. D. Schuur, Eds.), pp. 1-197, Lecture Notes in Pure and Applied Math., Vol. 19, Dekker, New York, 1976. 5. L. CESARI,Nonlinear analysis, Boll. Un. h4al. ital. A (6) 4 (1985), 157-216. 6. J. MAWHIN, Solutions p&riodiques d’tquations aux derivees partielles hy~rboliques non lineaires, in “Melanges Th. Vogel” (B. Rybak, P. Janssens,et M. Jesse&Eds.), pp. 301-319, Presses Univ. Bruxelles, Bruxelles, 1978. 7. J. MAWHIN, Semilinear equations of gradient type in Hilbert spaces and applications to differential equations, in “Nonlinear Differential Equations: Invariance, Stability, and Bifurcation” (E. de Mottoni and L. Salvadori, Eds), pp. 269-282, Academic Press, New York, 1981. 8. J. MAWHIN, Periodic oscillations of nonlinear wave systems, in “Ninth Intern. Conf. on Nonlinear Oscillations, Kiev, 1981” (Yu. Mitropolsky, Ed.), Vol. 1, pp. 47-53. Kiev Naukova Dumka, 1984. 9. P. J. MCKENNA, On solutions of a nonlinear wave equation when the ratio of the period to the length of the interval is irrational, Proc. Amer. Math. Sot. 93 (1985), 59-64. 10. A. NAPARSTEK.“Periodic Solutions of Certain Weakly Nonlinear Hyperbolic Partial Differential Equations,” Ph.D. Thesis, University of Michigan, University Mi~ro~lms, Ann Arbor, Michigan, 1968. Il. I. NIVE~ AND H. S. ZUCKERMAN, “The Theory of Numbers,” 4th ed., Wiley, New York, 1980. 12. G. R. SELL, The prodigal integral, Amer. Math. Monfhly 84 (1977), 162-167. 13. M. W. SMILEY, Eigenfunction methods and nonlinear hyperbolic boundary value problems at resonance, J. Math. Anal. Appi. 122 (1987), 129151. 14. E. H. ~RA~~NELLO, Solving functional equations by contractive averaging, in “Mathematics Research Center,” Rep. 160, Madison, Wisconsin, 1960. 15. E. ZEIDLER, “Nonlinear Functional Analysis and its Applications. I. Fixed-Point Theorems,” Springer-Verlag, New York, 1986.