Bounded Components of Positive Solutions of Nonlinear Abstract Equations

In this work a general class of nonlinear abstract equations satisfying a generalized strong maximum principle is considered in order to show that any bounded component of positive solutions bifurcating from the curve of trivial states (λ, u) = (λ, 0) at a nonlinear eigenvalue λ = λ₀ must meet the c...

Full description

Saved in:
Bibliographic Details
Date:2005
Main Authors: Cano-Casanova, S., Lopez-Gomez, J., Molina-Meyer, M.
Format: Article
Language:English
Published: Інститут прикладної математики і механіки НАН України 2005
Series:Український математичний вісник
Online Access:https://nasplib.isofts.kiev.ua/handle/123456789/124580
Tags: Add Tag
No Tags, Be the first to tag this record!
Journal Title:Digital Library of Periodicals of National Academy of Sciences of Ukraine
Cite this:Bounded Components of Positive Solutions of Nonlinear Abstract Equations / S. Cano-Casanova, J. Lopez-Gomez, M. Molina-Meyer // Український математичний вісник. — 2005. — Т. 2, № 1. — С. 38-51. — Бібліогр.: 13 назв. — англ.

Institution

Digital Library of Periodicals of National Academy of Sciences of Ukraine
id nasplib_isofts_kiev_ua-123456789-124580
record_format dspace
spelling nasplib_isofts_kiev_ua-123456789-1245802025-02-10T00:29:05Z Bounded Components of Positive Solutions of Nonlinear Abstract Equations Cano-Casanova, S. Lopez-Gomez, J. Molina-Meyer, M. In this work a general class of nonlinear abstract equations satisfying a generalized strong maximum principle is considered in order to show that any bounded component of positive solutions bifurcating from the curve of trivial states (λ, u) = (λ, 0) at a nonlinear eigenvalue λ = λ₀ must meet the curve of trivial states (λ, 0) at another singular value λ₁ ≠ λ₀. Since the unilateral theorems of P. H. Rabinowitz [13, Theorems 1.27 and 1.40] are not true as originally stated (c.f. the counterexample of E. N. Dancer [6]), in order to get our main result the unilateral theorem of J. Lopez-Gomez [11, Theorem 6.4.3] is required. 2005 Article Bounded Components of Positive Solutions of Nonlinear Abstract Equations / S. Cano-Casanova, J. Lopez-Gomez, M. Molina-Meyer // Український математичний вісник. — 2005. — Т. 2, № 1. — С. 38-51. — Бібліогр.: 13 назв. — англ. 1810-3200 2000 MSC. 34A34, 34C23, 35B32, 35B50, 35J25. https://nasplib.isofts.kiev.ua/handle/123456789/124580 en Український математичний вісник application/pdf Інститут прикладної математики і механіки НАН України
institution Digital Library of Periodicals of National Academy of Sciences of Ukraine
collection DSpace DC
language English
description In this work a general class of nonlinear abstract equations satisfying a generalized strong maximum principle is considered in order to show that any bounded component of positive solutions bifurcating from the curve of trivial states (λ, u) = (λ, 0) at a nonlinear eigenvalue λ = λ₀ must meet the curve of trivial states (λ, 0) at another singular value λ₁ ≠ λ₀. Since the unilateral theorems of P. H. Rabinowitz [13, Theorems 1.27 and 1.40] are not true as originally stated (c.f. the counterexample of E. N. Dancer [6]), in order to get our main result the unilateral theorem of J. Lopez-Gomez [11, Theorem 6.4.3] is required.
format Article
author Cano-Casanova, S.
Lopez-Gomez, J.
Molina-Meyer, M.
spellingShingle Cano-Casanova, S.
Lopez-Gomez, J.
Molina-Meyer, M.
Bounded Components of Positive Solutions of Nonlinear Abstract Equations
Український математичний вісник
author_facet Cano-Casanova, S.
Lopez-Gomez, J.
Molina-Meyer, M.
author_sort Cano-Casanova, S.
title Bounded Components of Positive Solutions of Nonlinear Abstract Equations
title_short Bounded Components of Positive Solutions of Nonlinear Abstract Equations
title_full Bounded Components of Positive Solutions of Nonlinear Abstract Equations
title_fullStr Bounded Components of Positive Solutions of Nonlinear Abstract Equations
title_full_unstemmed Bounded Components of Positive Solutions of Nonlinear Abstract Equations
title_sort bounded components of positive solutions of nonlinear abstract equations
publisher Інститут прикладної математики і механіки НАН України
publishDate 2005
url https://nasplib.isofts.kiev.ua/handle/123456789/124580
citation_txt Bounded Components of Positive Solutions of Nonlinear Abstract Equations / S. Cano-Casanova, J. Lopez-Gomez, M. Molina-Meyer // Український математичний вісник. — 2005. — Т. 2, № 1. — С. 38-51. — Бібліогр.: 13 назв. — англ.
series Український математичний вісник
work_keys_str_mv AT canocasanovas boundedcomponentsofpositivesolutionsofnonlinearabstractequations
AT lopezgomezj boundedcomponentsofpositivesolutionsofnonlinearabstractequations
AT molinameyerm boundedcomponentsofpositivesolutionsofnonlinearabstractequations
first_indexed 2025-12-02T04:17:41Z
last_indexed 2025-12-02T04:17:41Z
_version_ 1850368654387970048
fulltext Український математичний вiсник Том 2 (2005), № 1, 38 – 51 Bounded Components of Positive Solutions of Nonlinear Abstract Equations Santiago Cano-Casanova, Julián López-Gómez and Marcela Molina-Meyer (Presented by I. V. Skrypnik) Abstract. In this work a general class of nonlinear abstract equations satisfying a generalized strong maximum principle is considered in order to show that any bounded component of positive solutions bifurcating from the curve of trivial states (λ, u) = (λ, 0) at a nonlinear eigenvalue λ = λ0 must meet the curve of trivial states (λ, 0) at another singu- lar value λ1 6= λ0. Since the unilateral theorems of P. H. Rabinowitz [13, Theorems 1.27 and 1.40] are not true as originally stated (c.f. the counterexample of E. N. Dancer [6]), in order to get our main result the unilateral theorem of J. López-Gómez [11, Theorem 6.4.3] is required. 2000 MSC. 34A34, 34C23, 35B32, 35B50, 35J25. Key words and phrases. Positive Solutions, Compact solution com- ponents, Nonlinear abstract equations, Bifurcation theory. 1. Introduction Throughout this work U stands for an ordered real Banach space whose positive cone, P , is normal and it has nonempty interior, and we consider the nonlinear abstract equation F(λ, u) := L(λ)u+ R(λ, u) = 0 , (λ, u) ∈ X := R × U , (1.1) where (HL) The family K(λ) := IU − L(λ) ∈ L(U), λ ∈ R, is compact and real analytic, and L(λ̂) is a linear topological isomorphism for some λ̂ ∈ R, where IU stands for the identity map of U and L(U) is the space of linear continuous endomorphisms of U . Received 13.01.2004 The work has been partially supported by grants BFM2000-0797 and REN2003-00707 of the Spanish Ministry of Science and Technology. ISSN 1810 – 3200. c© Iнститут прикладної математики i механiки НАН України S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 39 (HR) R ∈ C(R×U ;U) is compact on bounded sets and limu→0 R(λ,u) ‖u‖ = 0 uniformly on compact intervals of R . (HP) The solutions of (1.1) satisfy the strong maximum principle in the sense that (λ, u) ∈ R × (P \ {0}) and F(λ, u) = 0 imply u ∈ IntP , where IntP stands for the interior of the cone P . Subsequently, given u1, u2 ∈ U , we write u1 > u2 if u1−u2 ∈ P \{0}, and u1 ≫ u2 if u1 − u2 ∈ IntP . Also, it will be said that (λ, u) is a positive solution of (1.1), if (λ, u) is a solution of (1.1) with u > 0. Thanks to Assumption (HP), any positive solution (λ, u) of (1.1) must be strongly positive, in the sense that u≫ 0. Under Assumptions (HL) and (HR), F(λ, 0) = 0 for each λ ∈ R . The main result of this paper concerns the bounded components of positive solutions of (1.1) emanating from (λ, u) = (λ, 0) at a nonlinear eigenvalue λ0 ∈ R with geometric multiplicity one. By a component of positive solutions of (1.1) it is meant a maximal (for the inclusion) relatively closed and connected subset of the set of positive solutions of (1.1) (in R × IntP ). A value σ ∈ R is said to be an eigenvalue of the family L(λ) if dimN [L(σ)] ≥ 1. The set of eigenvalues of L(λ) will be denoted by S. Thanks to (HL), L(λ) is Fredholm of index zero for any λ ∈ R and, hence, S provides us with the set of singular values of the family L(λ). In other words, L(λ) is a linear topological isomorphism if λ ∈ R \ S. Moreover, it follows from [11, Theorem 4.4.4] that S is discrete and that it consists of algebraic eigenvalues of L, i.e., for any σ ∈ S there exist C > 0, ε > 0 and ν ≥ 1 such that for any λ ∈ (σ − ε, σ + ε) \ {σ} the operator L−1(λ) is well defined and ‖L−1(λ)‖L(U) ≤ C |λ−σ|ν if 0 < |λ − σ| < ε. Thus, thanks to the abstract spectral theory developed in [11, Chapter 4], the algebraic multiplicity χ[L; ·] : S → N introduced by J. Esquinas and J. López-Gómez in [8] and [7] is well defined. Actually, χ[L;λ0] = 1 if λ0 ∈ S is a simple eigenvalue of L(λ) as discussed by M. G. Crandall and P. H. Rabinowitz [4], i.e., if dimN [L0] = 1 and L1(N [L0]) ⊕R[L0] = U , (1.2) where L0 := L(λ0), L1 := dL dλ (λ0) = −dK dλ (λ0), and, for any T ∈ L(U), N [T ] and R[T ] stand for the null space and the range of T . A value σ ∈ S is said to be a nonlinear eigenvalue of L(λ) if (σ, 0) is a bifurcation point of (1.1) from (λ, 0), λ ∈ R, for any R(λ, u) satisfying (HR). According to [11, Theorem 4.3.4], χ[L;σ] ∈ 2N+1 if σ is a nonlinear eigenvalue of L(λ), 40 Bounded Components of Positive Solutions... and, thanks to [11, Theorem 6.6.2], for any σ ∈ S there is η ∈ {−1, 1} such that Ind (0,K(λ)) = η sign (λ− σ)χ[L;σ] , λ ∼ σ , λ 6= σ , where Ind (0,K(λ)) is the local index of K(λ) at zero (the topological de- gree of L(λ) in any open bounded set containing zero). Thus, Ind (0,K(λ)) changes as λ crosses σ if, and only if, χ[L;σ] ∈ 2N + 1. Consequently, thanks to [11, Theorem 6.2.1], σ ∈ S is a nonlinear eigenvalue of L(λ) if χ[L;σ] ∈ 2N + 1, and, therefore, σ ∈ S is a nonlinear eigenvalue of L(λ) if, and only if, χ[L;σ] ∈ 2N + 1. The main result of this paper can be stated as follows. Theorem 1.1. Suppose λ0 ∈ S is a nonlinear eigenvalue of L(λ) such that N [L(λ0)] = span [ϕ0] , ϕ0 ∈ P \ {0} , (1.3) and K(λ0) is strongly positive, i.e., K(λ0)(P \ {0}) ⊂ IntP . (1.4) Then, there exists a component CP λ0 of the set of positive solutions of F(λ, u) = 0 emanating from (λ, u) = (λ, 0) at λ = λ0. Moreover, there exists λ1 ∈ S \ {λ0} such that (λ1, 0) ∈ C̄P λ0 if CP λ0 is bounded in X := R × U . Note that, thanks to (HL) and (1.4), K(λ0)ϕ0 = ϕ0 ≫ 0 . (1.5) Thus, the existence of CP λ0 follows by adapting some of the unilateral results of P. H. Rabinowitz [13]; in Section 2 complete details will be provided. The distribution of this paper is as follows. In Section 2 we show the existence of CP λ0 , in Section 3 we complete the proof of Theorem 1.1, and in Section 4 we derive from Theorem 1.1 a celebrated unilateral the- orem attributable to E. N. Dancer [5]. Finally, in Section 5 we give an application of Theorem 1.1, and in Section 6 we construct an example showing the necessity of condition χ[L;λ0] ∈ 2N + 1 for the validity of Theorem 1.1. Throughout the remaining of this paper it will be assumed that λ0 ∈ S satisfies all the requirements of Theorem 1.1. S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 41 2. Unilateral Bifurcation. The Existence of CP λ0 The set of non-trivial solutions of (1.1) is defined through S := F−1(0) \ [(R \ S) × {0}] . Note that (λ, u) ∈ S if either u 6= 0, or else u = 0 and λ ∈ S. Since χ[L;λ0] ∈ 2N+1, thanks to [11, Corollary 6.3.2] there exists a component of S, subsequently denoted by Cλ0 , such that (λ0, 0) ∈ Cλ0 . Subsequently, we suppose that ϕ0 has been normalized so that N [L(λ0)] = span [ϕ0] , ‖ϕ0‖ = 1 , ϕ0 > 0 . (2.1) Thanks to (1.5), ϕ0 ≫ 0. Now, let Y be a closed subspace of U such that U = N [L(λ0)]⊕Y . Thanks to Hahn-Banach’s theorem, there exists ϕ∗ 0 ∈ U ′ such that Y = {u ∈ U : 〈ϕ∗ 0, u〉 = 0 } , 〈ϕ∗ 0, ϕ0〉 = 1 , where 〈·, ·〉 stands for the duality between U and U ′. Now, for each η ∈ (0, 1) and sufficiently small ε > 0 we set Qε,η := { (λ, u) ∈ X : |λ− λ0| < ε , |〈ϕ∗ 0, u〉| > η ‖u‖ } . Since the mapping u 7→ |〈ϕ∗ 0, u〉| − η ‖u‖ is continuous, Qε,η is an open subset of X consisting of the two disjoint components Q+ ε,η and Q− ε,η defined though Q+ ε,η := { (λ, u) ∈ X : |λ− λ0| < ε , 〈ϕ∗ 0, u〉 > η ‖u‖ } , Q− ε,η := { (λ, u) ∈ X : |λ− λ0| < ε , 〈ϕ∗ 0, u〉 < −η ‖u‖ } . (2.2) The following result collects the main consequences from [11, Theorem 6.2.1, Proposition 6.4.2], which are consequences from the reflection ar- gument of P. H. Rabinowitz [13]. Subsequently, we denote by BR(x) the open ball of radius R > 0 centered at x ∈ X. Theorem 2.1. For each sufficiently small δ > 0, Cλ0 ∩Bδ(λ0, 0) ⊂ Qε,η ∪ {(λ0, 0)} and each of the sets S \ [ Q− ε,η ∩Bδ(λ0, 0) ] and S \ [ Q+ ε,η ∩Bδ(λ0, 0) ] contains a component, denoted by C+ λ0 and C− λ0 , respectively, such that (λ0, 0) ∈ C+ λ0 ∩ C− λ0 and Cλ0 ∩Bδ(λ0, 0) = ( C+ λ0 ∪ C− λ0 ) ∩Bδ(λ0, 0) . (2.3) Moreover, for each (λ, u) ∈ (Cλ0 \ {(λ0, 0)}) ∩ Bδ(λ0, 0), there exists a unique pair (s, y) ∈ R × Y such that u = sϕ0 + y and |s| > η ‖u‖. Furthermore, λ = λ0 + o(1) and y = o(s) as s→ 0. 42 Bounded Components of Positive Solutions... It should be noted that if (λ, u) ∈ C+ λ0 ∩ Bδ(λ0, 0), u 6= 0, then u = sϕ0 + y with s > η ‖u‖ > 0, and, hence, u s = ϕ0 + y s . Thus, since lims→0 y s = 0, for sufficiently small s > 0, u s ∈ IntP and, consequently, u ∈ IntP . Therefore, for any sufficiently small δ > 0, we have that [ C+ λ0 \ {(λ0, 0)} ] ∩Bδ(λ0, 0) ⊂ R × IntP . (2.4) This shows the existence of the component CP λ0 of R × IntP containing (λ0, 0) (cf. the statement of Theorem 1.1). Actually, CP λ0 is the maximal sub-continuum of C+ λ0 in R × IntP . The following result, which is [11, Theorem 6.4.3], provides us with an updated version of the unilateral theorem of P.H. Rabinowitz [13, Theorem 1.27], which is not true as originally stated (cf. E. N. Dancer [6]). Theorem 2.2. For each ∗ ∈ {−,+}, the component C∗ λ0 satisfies some of the following alternatives: 1. C∗ λ0 is unbounded in X. 2. There exists λ1 ∈ S \ {λ0} such that (λ1, 0) ∈ C∗ λ0 . 3. C∗ λ0 contains a point (λ, y) ∈ R × (Y \ {0}). Thanks to (HL), (1.4) and (1.5), the theorem of M. G. Krein and M. A. Rutman [10] (cf. H. Amann [1, Theorem 3.2]) as well), shows the validity of the following result, which is needed to conclude the proof of Theorem 1.1. Theorem 2.3. Let Spr (K(λ0)) denote the spectral radius of K(λ0). Then, (a) Spr (K(λ0)) = 1 is an algebraically simple eigenvalue of K(λ0) and, hence, N [L0] = N [L2 0] = span [ϕ0] . (2.5) Thus, 0 is an algebraically simple eigenvalue of L0, i.e., U = N [L0] ⊕R[L0] . (2.6) Moreover, no other eigenvalue of K(λ0) admits a positive eigenvec- tor. (b) For every y ∈ IntP , the equation u − K(λ0)u = y cannot admit a positive solution. S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 43 Proof. As Spr (K(λ0)) is the unique eigenvalue of K(λ0) associated with it there is a positive eigenvector, and 1 is an eigenvalue to the eigenfunction ϕ0, we have that Spr (K(λ0)) = 1. Moreover, 1 is algebraically simple, i.e., N [IU − K(λ0)] = N [(IU − K(λ0)) 2] , and, hence, due to (HL), (2.5) holds. Now, since L0 is Fredholm of index zero, to prove (2.6) it suffices to show that ϕ0 6∈ R[L0]. On the contrary, suppose that ϕ0 ∈ R[L0]. Then, there exists u ∈ U \ N [L0] such that L0u = ϕ0, and, hence, L2 0u = L0ϕ0 = 0. Thus, u ∈ N [L2 0] \N [L0], which contradicts (2.5) and concludes the proof of (2.6). This completes the proof of Part (a). Part (b) is an straightforward consequence from H. Amann [1, Theorem 3.2]. As a consequence from (2.6) we can make the choice Y = R[L0] , (2.7) which will be maintained throughout the remaining of the proof of The- orem 1.1. 3. Completion of the Proof of Theorem 1.1 Assume CP λ0 is bounded. Since CP λ0 ⊂ C+ λ0 , some of the following alternatives occurs. Either CP λ0 = C+ λ0 \ {(λ0, 0)} , (3.1) or else CP λ0 is a proper subset of C+ λ0 \ {(λ0, 0)} . (3.2) Suppose (3.1). Then, C̄P λ0 = C+ λ0 satisfies some of the alternatives of Theorem 2.2. Alternative 1 cannot be satisfied, since C̄P λ0 is compact. Suppose Alternative 3 occurs. Then, thanks to the choice (2.7), there exists (λ, y) ∈ R × R[L0], y 6= 0, such that (λ, y) ∈ C̄P λ0 ⊂ R × P . Since y 6= 0, necessarily y ∈ IntP , by (HP). Thus, there exists u ∈ U such that L0u = u− K(λ0)u = y . Since ϕ0 ∈ IntP , for each sufficiently large α > 0 we have that uα := u+ αϕ0 ≫ 0. Moreover, uα − K(λ0)uα = y , 44 Bounded Components of Positive Solutions... because K(λ0)ϕ0 = ϕ0, which contradicts Theorem 2.3(b). Therefore, Alternative 2 of Theorem 2.2 must be satisfied. This concludes the proof of Theorem 1.1 in case (3.1). Now, suppose (3.2). Then, since C+ λ0 ∩Bδ(λ0, 0) = [ CP λ0 ∩Bδ(λ0, 0) ] ∪ {(λ0, 0)} for each sufficiently small δ > 0, fixing one of these δ’s, there exists (λ1, u) 6∈ Bδ(λ0, 0) such that (λ1, u) ∈ C+ λ0 ∩ (R × ∂P ) ∩ ∂CP λ0 . Let {(µn, un)}n≥1 be any subsequence of CP λ0 such that limn→∞(µn, un) = (λ1, u). Then, F(λ1, u) = 0 and u ∈ P . If u > 0, then, thanks to (HP), u ∈ IntP , which contradicts u ∈ ∂P . Thus, u = 0, and, hence, (λ1, 0) ∈ C̄P λ0 . Moreover, λ1 6= λ0, since (λ1, u) = (λ1, 0) 6∈ Bδ(λ0, 0), which concludes the proof of Theorem 1.1. Note that, thanks to [11, Lemma 6.1.2], λ1 ∈ S, since these are the unique possible bifurcation values from (λ, 0) for (1.1). 4. Improving Dancer’s Unilateral Theorem As an immediate consequence from Theorem 1.1, the following uni- lateral result holds. Theorem 4.1. Suppose λ0 ∈ S is a nonlinear eigenvalue of L(λ) satis- fying (1.3) and (1.4), and, in addition, no other σ ∈ S \ {λ0} admits an eigenfunction in P \ {0}. Then, the component CP λ0 is unbounded in X. Proof. On the contrary, suppose that CP λ0 is bounded. Then, thanks to Theorem 1.1, there exists λ1 ∈ S \ {λ0} such that (λ1, 0) ∈ C̄P λ0 . Let {(µn, un)}n≥1 be any sequence of CP λ0 such that limn→∞(µn, un) = (λ1, 0) and set vn := un ‖un‖ , n ≥ 1. Then, vn = K(µn)vn − R(µn, un) ‖un‖ , n ≥ 1 , and, hence, by (HL) and (HR), there exists a subsequence of {vn}n≥1, again labeled by n, such that limn→∞ vn = ψ. Necessarily ψ > 0. More- over, passing to the limit as n → ∞ gives ψ = K(λ1)ψ, or, equivalently, L(λ1)ψ = 0, which is impossible, since we are assuming that λ0 is the unique element of S to a positive eigenvector. This contradiction shows that CP λ0 is unbounded and concludes the proof. S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 45 In the special case when K(λ) = λK, λ ∈ R, for some linear strongly positive compact operator K ∈ L(U), necessarily λ0 := 1 Spr K is the unique element of S associated with it there is a positive eigenvector, and, therefore, thanks to Theorem 4.1, CP λ0 must be unbounded. Consequently, Theorem 4.1 is a substantial improvement of [11, Theorem 6.5.5], which is a well known result attributable to E. N. Dancer [5]. 5. An Application of Theorem 1.1 In this section we consider the semi-linear weighted boundary value problem { Eu = λW (x)u− a(x)ur in Ω , u = 0 on ∂Ω , (5.1) where Ω is a bounded domain of RN with boundary ∂Ω of class C2+ν for some ν ∈ (0, 1), r ∈ (1,∞), λ ∈ R is regarded as a bifurcation parameter, E is a second order uniformly elliptic operator of the form E := − N∑ i,j=1 αij ∂2 ∂xi∂xj + N∑ i=1 αi ∂ ∂xi + α0 with αij = αji ∈ Cν(Ω̄), αi, α0 ∈ Cν(Ω̄), 1 ≤ i, j ≤ N , and (Ha) a ∈ Cν(Ω̄) and, setting a+ := max{a, 0}, a− := a+ − a, Ω0 a+ := Ω \ supp a+, Ω0 a− := Ω \ supp a−, Ω0 a+ and Ω0 a− are two proper open subsets of Ω of class C2+ν with a finite number of well separated components. Moreover, either N ∈ {1, 2}, or else N ≥ 3 and for some constant γ > 0 the following is satisfied [dist (·, ∂Ω0 a−)]−γa− ∈ C ( supp a−, (0,∞) ) , r < max { N + 2 N − 2 , N + 1 + γ N − 1 } . (Hw) W ∈ Cν(Ω̄) changes of sign in Ω0 a+ and max λ∈R σ[E − λW ; Ω] > 0 , (5.2) where, for any elliptic operator L in a bounded domain D, σ[L;D] stands for the principal eigenvalue of L in D under homogeneous Dirichlet boundary conditions. 46 Bounded Components of Positive Solutions... Thanks to (5.2), for each D ∈ { Ω0 a+ ,Ω } the weighted boundary value problem { Eϕ = λWϕ in D , ϕ = 0 on ∂D , (5.3) possesses two principal eigenvalues, λD 1 < λD 2 . By a principal eigenvalue of (5.3) it is meant a value of λ for which (5.3) possesses an eigenfunction ϕ > 0. It should be noted that, necessarily, σ[E − λD j W ;D] = 0 for each (D, j) ∈ { Ω0 a+ ,Ω } × {1, 2}. Moreover, by the monotonicity of σ[·;D] with respect to D, λ Ω0 a+ 1 < λΩ 1 < λΩ 2 < λ Ω0 a+ 2 , and, thanks to a celebrated result by P. Hess and T. Kato [9], setting Σ(λ) := σ[E − λW ; Ω], λ ∈ R, one has that Σ′(λΩ 1 ) > 0 and Σ′(λΩ 2 ) < 0, where ′ = d dλ . Set λ0 := λΩ 1 and denote by ϕ0 the principal eigenfunction associated to Σ(λ0) = 0, normalized so that ‖ϕ0‖ = 1. Then, since Σ(λ) is a simple eigenvalue, there is an analytic mapping λ 7→ ϕ(λ) ∈ C2+ν 0 (Ω̄) such that ϕ(λ0) = ϕ0 and (E − λW )ϕ(λ) = Σ(λ)ϕ(λ) . Now, differentiating with respect to λ and particularizing at λ = λ0 gives (E − λ0W )ϕ′(λ0) = Wϕ0 + Σ′(λ0)ϕ0 and, hence, Σ′(λ0) = −〈ϕ∗ 0,Wϕ0〉, where N [E∗ − λ0W ] = span[ϕ∗ 0] with 〈ϕ∗ 0, ϕ0〉 = 1. Therefore, since Σ′(λ0) > 0, we find that Wϕ0 6∈ R[E − λ0W ] . (5.4) Now, suppose u is a positive solution of (5.1). Then, u|∂Ω0 a+ > 0 and (E − λW )u = −aur = a−ur ≥ 0 in Ω0 a+ . Thus, thanks to characterization of the maximum principle of J. López- Gómez and M. Molina-Meyer [12], it is apparent that σ[E−λW ; Ω0 a+ ] > 0, and, therefore, λ ∈ (λ Ω0 a+ 1 , λ Ω0 a+ 2 ), by the strict concavity of λ 7→ Σ(λ). Thus, by the a priori bounds found by S. Cano-Casanova [2], there exists a constant C > 0 such that for any positive solution (λ, uλ) of (5.1), ‖uλ‖Cν(Ω̄) ≤ C. Now, let M > 0 be sufficiently large so that σ[E + M ; Ω] > 0 and λΩ 1W (x) +M > 0 for each x ∈ Ω̄. By elliptic regularity, the positive solutions of (5.1) are given by the zeroes in U := Cν 0 (Ω̄) of the equation L(λ)u+ R(λ, u) = 0 , (5.5) where, for each (λ, u) ∈ R × U , we have denoted L(λ)u := u− (E +M)−1[(λW +M)u] , S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 47 and R(λ, u) := (E +M)−1(aur) . It should be noted that the inverse operator (E +M)−1 ∈ L(U) is com- pact, by elliptic regularity and Ascoli-Arzela’s theorem, and strongly or- der preserving, by the strong maximum principle; the space U being ordered by the cone of point-wise non-negative functions. Thus, (5.5) fits into the abstract setting of Section 1 with K(λ)u := (E +M)−1[(λW +M)u] , u ∈ U . By the choice of M , K(λ0) is strongly positive. Moreover, setting L0 := L(λ0) , L1 := dL dλ (λ0) = −(E +M)−1(W ·) , one has that N [L0] = span[ϕ0] and L1ϕ0 6∈ R[L0]. Indeed, if L0u = −(E +M)−1(Wϕ0) for some u ∈ U , then u− (E +M)−1[(λ0W +M)u] = −(E +M)−1(Wϕ0) and, by elliptic regularity, u ∈ C2+ν 0 (Ω̄). Thus, (E − λ0W )u = −Wϕ0, which contradicts (5.4). Hence, the transversality condition of M. G. Crandall and P. H. Rabinowitz [4] is satisfied and, consequently, χ[L;λ0] = 1. Therefore, since λΩ 1 and λΩ 2 are the unique values of λ where positive solutions of (5.1) can bifurcate from (λ, 0), as an immediate con- sequence from Theorem 1.1 the following result is obtained. Theorem 5.1. There is a bounded component of the set of positive solutions of (5.1), say CP , such that (λΩ 1 , 0), (λΩ 2 , 0) ∈ C̄P . Moreover, PλCP ⊂ (λ Ω0 a+ 1 , λ Ω0 a+ 2 ), where Pλ stands for the λ-projection operator. It should be noted that, thanks to Theorem 4.1 and Theorem 5.1, (5.1) cannot admit an abstract representation as a fixed point equation of the form (1.1) with K(λ) = λK for some fixed compact strongly positive operator K, and, consequently, even if the unilateral results of P. H. Rabinowitz [13] would be correct as originally stated, Theorem 5.1 could not be a consequence from them. 48 Bounded Components of Positive Solutions... 6. Three Different Types of Bounded Components Under the assumptions of Theorem 1.1, L(λ) must possess two dif- ferent eigenvalues λ ∈ S, at least λ0 and λ1, associated with each the operator has a positive eigenfunction. This is far from being true if, in- stead of assuming that λ0 is a nonlinear eigenvalue, one assumes that χ[L;λ0] ∈ 2N, since, in this case, the component CP λ0 might emanate from the curve (λ, 0) exclusively at λ0. Therefore, the oddity of the mul- tiplicity is crucial for the validity of Theorem 1.1. Actually, there are boundary value problems of the form (5.1) that possess bounded com- ponents exhibiting each of these behaviors. For example, consider the one-dimensional prototype model in Ω = (0, 1) { −u′′ + µu = λ sin(2πx)u− a(x)u2 , u(0) = u(1) = 0 , (6.1) where a(x) =    −0.2 sin ( π 0.2(0.2 − x) ) if 0 ≤ x ≤ 0.2 , sin ( π 0.6(x− 0.2) ) if 0.2 < x ≤ 0.8 , −0.2 sin ( π 0.2(x− 0.8) ) if 0.8 < x ≤ 1 , (6.2) and (λ, µ) ∈ R2 are regarded as two real parameters. Note that a > 0 in (0.2, 0.8), a < 0 in (0, 0.2)∪ (0.8, 1), and a(0) = a(0.2) = a(0.8) = a(1) = 0. For an adequate choice of the parameter µ, this problem fits into the abstract setting of Section 5 by choosing Eµ := − d2 dx2 +µ, W := sin(2π · ) and r = 2. Indeed, since N = 1, Ω0 a+ = (0, 0.2) ∪ (0.8, 1) and max λ∈R σ[Eµ − λW ; Ω] = σ[− d2 dx2 + µ; Ω] = π2 + µ , because of the symmetry of the problem around 0 (cf. [3] for further details), it turns out that condition (5.2) holds as soon as µ > −π2. Actually, for each µ > −π2, there exist λΩ 1 (µ) < 0 < λΩ 2 (µ) = −λΩ 1 (µ) such that σ[Eµ − λΩ 1 (µ)W ; Ω] = σ[Eµ − λΩ 2 (µ)W ; Ω] = 0 . Moreover, As µ decreases approaching −π2, λΩ 1 (µ) increases, and, hence, λΩ 2 (µ) decreases, approaching 0, i.e., limµ↓−π2 λΩ 1 (µ)=0=limµ↓−π2 λΩ 2 (µ). As a result, Theorem 5.1 applies when µ > −π2 while it cannot be applied if µ ≤ −π2. Actually, the mapping λ 7→ Σµ(λ) := σ[Eµ−λW ; Ω] satisfies Σ−π2(0) = 0, Σ′ −π2(0) = 0, and Σ−π2(λ) < 0 for each λ ∈ R \ {0}. Therefore, χ[E−π2 −λW ; 0] = 2 and Theorem 5.1 cannot applied to cover S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 49 this transition situation, though the problem still possesses a bounded component of positive solutions emanating from (λ, 0) at λ = 0 if µ = −π2 (cf. the second plot of Figure 6.1). When µ < −π2 there are no bifurcation points to positive solutions from (λ, 0) and, actually, if µ is sufficiently close to −π2, then (3.1) exhibits an isola of positive solutions (cf. the third plot of Figure 6.1). The bifurcation diagrams of Figure 6.1 were computed by coupling a pseudo-spectral method with collocation and a path-continuation solver (cf. [3]). The left plot of Figure 6.1 shows the component for µ = 0. In this case, λΩ 1 ∼ −28.0233 and λΩ 2 ∼ 28.0233. The central plot of Figure 6.1 show the perturbations of the positive solutions of the left plot as µ decreases from zero up to reach the value µ = −9.8693 > π2 = −9.86960... . Now, λΩ 1 ∼ −0.13861 and λΩ 2 ∼ 0.13861; as these values are very close, the central plot of Figure 6.1 shows them super-imposed. As the computational model is discrete and π2 is irrational there is no way to get the bifurcation diagram for µ = −π2, though it must be very similar to the central diagram. The right plot shows the isola of solutions obtained for µ = −40, for which (λ, 0) always is linearly unstable. −50 0 50 −200 0 200 400 600 800 1000 1200 1400 −50 0 50 −200 0 200 400 600 800 1000 1200 1400 −50 0 50 −200 0 200 400 600 800 1000 1200 1400 Figure 6.1 Three components of positive solutions for µ = 0,−9.8693,−40, respectively. 50 Bounded Components of Positive Solutions... In Figure 6.1 we are plotting the value of λ against the L∞-norm of the corresponding positive solution. Stable solutions are indicated by solid lines, unstable by dotted lines. As there are some ranges of values of λ where the model possesses at least two solutions with very similar L∞- norms, the plot did not allow us distinguishing them, but rather plotted twice these pieces. This is why the bifurcation diagrams exhibit a darker arc of curve. It should be clear that in case µ = −π2 the component of positive solutions of (3.1) bifurcating from (λ, 0) at λ = 0 must be bounded and that it emanates from the curve (λ, 0) exclusively at λ = 0. References [1] H. Amann, Fixed point equations and nonlinear eigenvalue problems in ordered Banach spaces, SIAM Rev. 18 (1976), 620–709. [2] S. Cano-Casanova, Compact Components of positive solutions for Superlinear In- definite Elliptic Problems of Mixed Type // Top. Meth. Non. Anal. 23 (2004), 45–72. [3] S. Cano-Casanova, J. López-Gómez, and M. Molina-Meyer, Isolas: compact solu- tion components separated away from a given equilibrium curve, Hiroshima Math. J. 34 (2004), 177-199. [4] M. G. Crandall, and P. H. Rabinowitz, Bifurcation from simple eigenvalues // J. Funct. Anal. 8 (1971), 321–340. [5] E. N. Dancer, Global solution branches for positive mappings // Arch. Rat. Mech. Anal. 52 (1973), 181–192. [6] E. N. Dancer, Bifurcation from simple eigenvalues and eigenvalues of geometric multiplicity one // Bull. London Math. Soc. 34 (2002), 533–538. [7] J. Esquinas, Optimal multiplicity in local bifurcation theory, II: General case // J. Diff. Eqns. 75 (1988), 206–215. [8] J. Esquinas, and J. López-Gómez, Optimal multiplicity in local bifurcation theory, I: Generalized generic eigenvalues // J. Diff. Eqns. 71 (1988), 72–92. [9] P. Hess, and T. Kato, On some linear and nonlinear eigenvalue problems with an indefinite weight function // Comm. Part. Diff. Eqns. 5 (1980), 99–1030. [10] M. G. Krein, and M. A. Rutman, Linear operators leaving invariant a cone in a Banach space // Amer. Math. Soc. Transl. 10 (1962), 199–325. [11] J. López-Gómez, [2001] Spectral Theory and Nonlinear Functional Analysis, Re- search Notes in Mathematics 426, CRC Press, Boca Raton 2001. [12] J. López-Gómez, and M. Molina-Meyer, The maximum principle for cooperative weakly elliptic systems and some applications // Diff. Int. Eqns. 7 (1994), 383– 398. [13] P. H. Rabinowitz, Some global results for nonlinear eigenvalue problems // 7 (1971), 487–513. S. Cano-Casanova, J. López-Gómez, M. Molina-Meyer 51 Contact information Santiago Cano-Casanova Departamento de Matemática Aplicada y Computación Universidad Pontificia Comillas de Madrid 28015-Madrid, Spain E-Mail: scano@dmc.icai.upco.es Julián López-Gómez Departamento de Matemática Aplicada Universidad Complutense de Madrid 28040-Madrid, Spain E-Mail: Lopez_Gomez@mat.ucm.es Marcela Molina-Meyer Departamento de Matemáticas Universidad Carlos III de Madrid 28911-Leganés, Madrid, Spain E-Mail: mmolinam@math.uc3m.es