Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Existence and uniqueness results for variational inequalities Lecture 2 Basis results (end) and existence results Didier Aussel Univ. de Perpignan – p. 1 Outline of lecture 2 I- Set-valued or not? (end) a- The pointwise case b- The local case II- First Existence results a- The linear case b- The finite dimensional case III- Other classes of V.I. – p. 2 Notations X a topological vector space X ∗ its topological dual (w∗ -top.) h·, ·i the duality product – p. 3 Notations X a topological vector space X ∗ its topological dual (w∗ -top.) h·, ·i the duality product Stampacchia variational inequality : X∗ Let T : X → 2 be a map and C be a nonempty subset of X. Find x̄ ∈ C such that there exists x̄∗ ∈ T (x̄) for which hx̄∗ , y − x̄i ≥ 0, ∀ y ∈ C. Notation : S(T, C) set of solutions (⊂ C). – p. 3 I - Set-valued or not? (cont.) The pointwise case – p. 4 Pointwise Single-valuedness Theorem 1 (Kenderov 1975) X∗ Let X be a Banach space and T : X → 2 be a monotone set-valued map. If T is lower semicontinuous at a point x0 , then T is single-valued at x0 . T is lsc at x0 if, for every open V such that V ∩ T (x0 ) 6= ∅, there exists a neigh. U of x0 such that V ∩ T (x) 6= ∅, for any x ∈ U . – p. 5 X∗ Definition 2 A set-valued map T : K → 2 is said to be single-directional at x ∈ dom T if, T (x) ⊆ R+ {x∗ } for some x∗ ∈ T (x). X∗ Proposition 3 Let T : X → 2 be a quasimonotone set-valued map. If T is lower semicontinuous at x ∈ X, then T is single-directional at x; – p. 6 Recovering Kenderov’73 (monotone lsc) Proposition 4 (D.A., J.-N. Corvellec & M. Lassonde ’94) X∗ A map T : X → 2 is monotone if and only if, for any α∗ ∈ X ∗ , the map T + {α∗ } is quasimonotone – p. 7 Recovering Kenderov’73 (monotone lsc) Proposition 6 (D.A., J.-N. Corvellec & M. Lassonde ’94) X∗ A map T : X → 2 is monotone if and only if, for any α∗ ∈ X ∗ , the map T + {α∗ } is quasimonotone X∗ Proposition 7 Let T : X → 2 be a set-valued map and x be a point of its domain dom T be such that, for any α∗ ∈ X ∗ , the map T + α∗ is single-directional at x, then T is single-valued at x. – p. 7 Recovering Kenderov’73 (monotone lsc) Hyp : T is monotone and lsc at x. ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is quasimonotone lsc at x ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is single-directional at x ⇒ T is single-valued at x. – p. 8 Recovering Kenderov’73 (monotone lsc) Hyp : T is monotone and lsc at x. ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is quasimonotone lsc at x ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is single-directional at x ⇒ T is single-valued at x. Theorem 9 (Kenderov 1975) X∗ Let X be a Banach space and T : X → 2 be a monotone set-valued map. If T is lower semicontinuous at a point x0 , then T is single-valued at x0 . – p. 8 I - Set-valued or not? (cont.) The local case – p. 9 Local Single-valuedness Theorem 10 (Dontchev and Hager 1994) X∗ Let T : X → 2 be a monotone set-valued map. If T is Lipschitz-like around (x0 , y) ∈ gr T , then T is single-valued in a neigh. of x0 . T : X → 2Y is Lipschitz-like around (x, y) ∈ gr T if it exist a neighb. U of x, a neighb. V of y and l > 0 such that T (u) ∩ V ⊂ T (u′ ) + lku′ − ukB Y (0, 1), ∀ u, u′ ∈ U If T is single-valued T is Lipschitz-like around (x, T (x)) ⇔ T is loc. Lipschitz at x – p. 10 X∗ Definition 11 A set-valued map T : K → 2 is said to be - single-directional at x ∈ dom T if, T (x) ⊆ R+ {x∗ } for some x∗ ∈ T (x). - locally single-directional at x ∈ dom T if it exists a neigh. U of x such that, for any x′ ∈ U , T (x′ ) is single-directional at x′ . Theorem 12 ∗ Let T : X → 2X be a quasimonotone set-valued map. If T satisfies the Lipschitz-like property around a point (x, x∗ ) of its graph and x∗ 6= 0, then T is alocally single-directional at x. – p. 11 Proposition 13 X∗ Let T : X → 2 be a quasimonotone set-valued map with convex values and satisfying the Lipschitz-like property around a point (x, x∗ ) of its graph. Then T is locally single-directional at x. – p. 12 Proofs (quasimonotone Lipschitz-like case) direct proof thanks to the following results ∗ Proposition 14 Let T : X → 2X be a set-valued map satisfying the Lipschitz-like property around a point (x, x∗ ) of its graph (with U and V the corresponding neighborhoods). Then the mapping TU ×V , whose graph is given by gr TU ×V = gr T ∩ (U × V ) is lower semicontinuous at x. ∗ Y 2 be a set-valued map with convex Proposition 15 Let T : Y → values. If T admits a single-directional localization around a point (y, y ∗ ) of its graph, then T is locally single-valued at y. – p. 13 Recovering Dontchev-Hager’94 (monotone Lipschitz-like) Hyp : T is monotone and Lipschitz-like around (x, x∗ ). ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is quasimonotone and Lipschitz-like around (x, x∗ ) ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is locally single-directional at x ⇒ T is single-valued on a neigh. of x. – p. 14 Recovering Dontchev-Hager’94 (monotone Lipschitz-like) Hyp : T is monotone and Lipschitz-like around (x, x∗ ). ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is quasimonotone and Lipschitz-like around (x, x∗ ) ⇒ ∀ α∗ ∈ X ∗ , T + {α∗ } is locally single-directional at x ⇒ T is single-valued on a neigh. of x. In the case of a convex valued operator... Theorem 17 (Dontchev and Hager 1994) X∗ Let T : X → 2 be a monotone set-valued map. If T is Lipschitz-like around (x0 , y) ∈ gr T , then T is single-valued in a neigh. of x0 . – p. 14 References 1- Facchinei, F. and Pang, J.S. Finite Dimensional Variational Inequalities, Springer, (2003). 2- Kenderov, P., Semi-continuity of set-valued mappings, Fund. Math., 88 (1975), 61-69. 3- Dontchev, A.L. and Hager, W., Implicit functions, Lipschitz maps, and stability in optimization, Math. Oper. Reasearch, 19 (1994), 753-768. 5- Aussel, D., Garcia Y. and Hadjisavvas, N., Single-directional property of multivalued maps and variational systems, SIAM J. Optim (2009). 6- Aussel, D. and Garcia Y., Single-valuedness of quasimonotone and monotone set-valued maps, preprint, (2009). – p. 15 II First Existence results a- The linear case b- The finite dimension case c- V.I. under monotonicity d- Other assumptions – p. 16 II - a - The linear case Let H be an Hilbert space, f an element of H ∗ = H and a : H × H → H a bilinear application. Theorem 18 (Stampacchia 64) Let C be a nonempty closed convex subset of H. If a is continuous and satisfies the following property ∃ α > 0 such that a(v, v) ≥ αkvk2 , ∀ v ∈ H, then there exists x̄ ∈ C such that a(x̄, y − x̄) ≥ hf, y − x̄i, ∀ y ∈ C. D. Kinderlehrer and G. Stampacchia, An introduction to Variational Inequalities and Their Applications, 1980, Academic Press – p. 17 Theorem 19 (Lax-Milgram) If a is continuous and satisfies the following property ∃ α > 0 such that a(v, v) ≥ αkvk2 , ∀ v ∈ H, then there exists a unique x̄ ∈ H such that a(x̄, u) = hf, ui, ∀ u ∈ H. Proof. From the Theorem of Stampacchia, there exists u ∈ H such that a(u, v − u) ≥ hf, v − ui, or in other words ϕu (v) ≥ ϕu (u), ∀v ∈ H ∀ v ∈ H where ϕu (v) = a(u, v) − hf, vi. But ϕu is linear and therefore ϕu ≡ 0. – p. 18 II - b - The finite dimension case Let f : X → X ∗ be a function and C be a nonempty subset of X. Find x̄ ∈ C such that hf (x̄), y − x̄i ≥ 0, ∀ y ∈ C. First property : S(f, C) ∩ int(C) = {x ∈ int(C) : f (x) = 0} – p. 19 II - b - The finite dimension case Proposition 20 (Stampacchia 1966) Let C be a nonempty convex compact subset of IRn and f : C → IRn a continuous function. Then S(f, C) is nonempty, that is, there exists x̄ ∈ C such that hf (x̄), y − x̄i ≥ 0, ∀ y ∈ C. – p. 20 II - b - The finite dimension case Proposition 21 (Stampacchia 1966) Let C be a nonempty convex compact subset of IRn and f : C → IRn a continuous function. Then S(f, C) is nonempty, that is, there exists x̄ ∈ C such that hf (x̄), y − x̄i ≥ 0, ∀ y ∈ C. Proof: Let us consider the application ψ: C → C x 7→ ψ(x) = PC ◦ (Id − f )(x) = PC (x − f (x)) ψ is continuous on the convex compact set C and therefore, according to the Brouwer fixed point theorem, there exists a point x̄ ∈ C such that x̄ = ψ(x̄) ⇔ x̄ = PC (x̄ − f (x̄)) ⇔ hx̄ − f (x̄) − x̄, y − x̄i ≤ 0, ∀ y ∈ C ⇔ x̄ ∈ S(f, C). – p. 20 Without compacity The compacity hypothesis can be replaced by a "coercivity condition": Proposition 22 Let C be a nonempty closed convex subset of IRn and f : C → IRn a continuous function. The following conditions are equivalent: - S(f, C) is nonempty. - ∃ r > 0 such that S(f, C ∩ B(0, r)) ∩ B(0, r) 6= ∅, – p. 21 Without compacity 2 The compacity hypothesis can be replaced by a "coercivity condition" in term of the function: Proposition 23 Let C be a nonempty closed convex subset of IRn and f : C → IRn a continuous function. If f is coercive, that is, hf (x) − f (x0 ), x − x0 i ∃ x0 ∈ C such that lim = +∞ kx − x0 k then S(f, C) is nonempty. – p. 22 Multivalued form Theorem 24 Let C be a nonempty convex compact subset of IRn IRn and F : C → 2 a upper semicontinuous map with convex compact values. Then S(F, C) is nonempty, that is, there exist x̄ ∈ C and x̄∗ ∈ F (x̄) such that hx̄∗ , y − x̄i ≥ 0, ∀ y ∈ C. – p. 23 Multivalued form Theorem 24 Let C be a nonempty convex compact subset of IRn IRn and F : C → 2 a upper semicontinuous map with convex compact values. Then S(F, C) is nonempty, that is, there exist x̄ ∈ C and x̄∗ ∈ F (x̄) such that hx̄∗ , y − x̄i ≥ 0, ∀ y ∈ C. can be found, e.g., Borwein, J. and Lewis, A., Convex Analysis and Nonlinear Optimization, 2000 – p. 24 Multivalued form Theorem 24 Let C be a nonempty convex compact subset of IRn IRn and F : C → 2 a upper semicontinuous map with convex compact values. Then S(F, C) is nonempty, that is, there exist x̄ ∈ C and x̄∗ ∈ F (x̄) such that hx̄∗ , y − x̄i ≥ 0, ∀ y ∈ C. F is upper semicontinuous at x if ∀ V neighb. of F (x), ∃ U neighb. of x such that F (U ) ⊂ V . – p. 24 Multivalued form: proof Proof. Since F is cusco, its range F (C) is compact and the map T = (Id − F ) ◦ PC is also cusco since PC is continuous (Exercise). Now one can observe that T (co(C − F (C))) = co(C − F (C)). Therefore T has a fixed point (say x0 ) on C − F (C), that is x0 ∈ T (x0 ) ⇔ x0 ⇔ ⇔ ∈ (Id − F ) ◦ PC (x0 ) x̄ = PC (x0 ) x̄∗ = x̄ − x0 ∈ F (x̄) hx0 − x̄, y − x̄i ≤ 0, ∀y ∈ C x̄∗ = x̄ − x0 ∈ F (x̄) and thus x̄ ∈ S(F, C). – p. 25 Equivalence Actually the above existence theorem 24 is equivalent to the Kakutani Fixed point theorem. Indeed, let C be a nonempty convex compact subset of IRn and T : C → 2C be a cusco map. Observing that the set-valued map F = Id − T is also cusco, Theorem 24 implies the solvability of the set-valued Stampacchia variational inequality defined by F and C, that is there exists x̄ ∈ C such that x̄ ∈ S(Id − T, C) ⇔ ⇔ 8 < ∃ u∗ ∈ (Id − T )(x̄) such that : hu∗ , y − x̄i ≥ 0, ∀ y ∈ C 8 < ∃ x̄∗ ∈ T (x̄) such that : hx̄ − x̄∗ , y − x̄i ≥ 0, ∀ y ∈ C Taking y = x̄∗ , one has hx̄ − x̄∗ , x̄∗ − x̄i ≥ 0 showing that x̄ is a fixed point of T . – p. 26 II Existence of solution c - V.I. under monotonicity – p. 27 Monotone case Theorem 27 (1966) Let X be a reflexive Banach space, C be a weakly compact convex nonempty subset of X. If f : C → X ∗ is a monotone function, continuous on finite dimension subsets, then S(f, C) is nonempty. Hartmann, G. and Stampacchia, G., Acta Mathematica, 115 (1966), 271-310 – p. 28 Monotone case Theorem 28 (1966) Let X be a reflexive Banach space, C be a weakly compact convex nonempty subset of X. If f : C → X ∗ is a monotone function, continuous on finite dimension subsets, then S(f, C) is nonempty. Theorem 30 Let X be a reflexive Banach space, C be a nonempty X∗ convex and weakly compact subset of X. If F : C → 2 is a monotone map with convex weakly compact values and continuous on finite dimension subsets, then S(F, C) is nonempty. can be found, e.g., in Shih, M.H. and Tan, K.K., JMAA, 134 (1988), 431-440 – p. 29 Monotonicities T is monotone iff ∀ x, y ∈ X, ∀ x∗ ∈ T (x) and ∀ y ∗ ∈ T (x) hy ∗ − x∗ , y − xi ≥ 0. T is pseudomonotone iff ∃ x∗ ∈ T (x) : hx∗ , y − xi ≥ 0 ⇒ hy ∗ , y − xi ≥ 0, ∀ y ∗ ∈ T (y). T is quasimonotone iff ∃ x∗ ∈ T (x) : hx∗ , y − xi > 0 ⇒ hy ∗ , y − xi ≥ 0, ∀ y ∗ ∈ T (y). monotone ⇓ pseudomonotone ⇓ quasimonotone – p. 30 Weakening monotonicty: why? Let f : X → IR be differentiable f is convex iff df is monotone iff df (x)(y − x) > 0 ⇒ df (y)(y − x) ≥ 0 f is pseudoconvex iff df is pseudomonotone iff df (x)(y − x) > 0 ⇒ df (y)(y − x) ≥ 0 f is quasiconvex iff df is quasimonotone iff df (x)(y − x) > 0 ⇒ df (y)(y − x) ≥ 0 Let f : X → IR ∪ {+∞} be lsc f is convex iff ∂f is monotone iff ∀ x∗ ∈ ∂f (x), y ∗ ∈ ∂f (y), hy ∗ − x∗ , y − xi ≥ 0 f is pseudoconvex iff ∂f is pseudomonotone iff ∃ x∗ ∈ ∂f (x) : hx∗ , y − xi ≥ 0 ⇒ ∀ y ∗ ∈ ∂f (y), hy ∗ , y − xi ≥ 0 f is quasiconvex iff ∂f is quasimonotone iff ∃ x∗ ∈ ∂f (x) : hx∗ , y − xi > 0 ⇒ ∀ y ∗ ∈ ∂f (y), hy ∗ , y − xi ≥ 0 – p. 31 monotone ⇓ pseudomonotone ⇓ quasimonotone Some previous attempts: J.C. Yao (1994) with pseudomonotone map on reflexive space N. Hadjisavvas and S. Schaible (1996) with quasimonotone + existence of inner points Dinh The Luc (2001) with singlevalued densely pseudomonotone maps – p. 32 Existence: simplified form Theorem 32 C nonempty convex compact subset of X. X∗ T : C → 2 quasimonotone + upper hemi-continuous + T (x) 6= ∅ convex w∗ -compact. Then S(T, C) 6= ∅. T is upper hemicontinuous on X if the restriction of T to any line segment is usc with respect to the w∗ -topology – p. 33 Knaster-Kuratowski-Mazurkiewicz (1929) : C nonempty subset of X. ϕ : C × C → IR is said to be a KKM-application if ∀ x1 , . . . , xn ∈ C, ∀ x ∈ co{x1 , . . . , xn } ∃ i ∈ 1, . . . , n such that ϕ(xi , x) ≥ 0. Theorem 34 C nonempty closed convex subset of X. ϕ : C × C → IR KKM-application + ϕ(x, .) usc quasiconcave, ∀ x + ∃ x̃ ∈ C with {y ∈ C : ϕ(x̃, y) ≥ 0} compact. Then ∃ ȳ ∈ C such that ϕ(x, ȳ) ≥ 0, ∀ x ∈ C. – p. 34 A very simple proof: Let us define ϕ : C × C → IR (x, y) → inf x∗ ∈T (x) hx∗ , x − yi The function ϕ is continuous concave relatively to y and {y ∈ C : ϕ(x, y) ≥ 0} is compact, ∀ x ∈ C. C AS 1 : ϕ is KKM T h.KKM =⇒ ∃ ȳ ∈ C such that ϕ(x, ȳ) ≥ 0, ∀ x ∈ C. C AS 2 : ∃ x1 , . . . , xn ∈ C, ∃ ȳ ∈ co{x1 , . . . , xn } such that ϕ(xi , ȳ) < 0, ∀ i = 1, . . . , n that is ∀ i = 1, . . . , n, ∃ x∗i ∈ T (xi ) : hx∗i , ȳ − xi i > 0 and, for some ρ > 0 and all z ∈ B(ȳ, ρ) ∩ C ∀ i, ∃ x∗i ∈ T (xi ) : hx∗i , z − xi i > 0. – p. 35 By quasimonotonicity, ∀ z ∈ B(ȳ, ρ) ∩ C, ∀ z ∗ ∈ T (z), ∀ i = 1, . . . , n, hz ∗ , z − xi i ≥ 0. In both cases, ∀ z ∈ B(ȳ, ρ) ∩ C, ∀ z ∗ ∈ T (z), hz ∗ , z − ȳi ≥ 0. (1) Let us prove that ȳ ∈ S(T, C): Let x ∈ C (x 6= ȳ), t0 > 0 be such that zt = ȳ + t(x − ȳ) ∈ B(ȳ, ρ) ∩ C for all t ∈]0, t0 [. From (1), one have, hzt∗ , x − ȳi ≥ 0, ∀ t ∈]0, t0 [, ∀ zt∗ ∈ T (zt ). (2) Let us define the open subset W = {y ∗ ∈ X ∗ : hy ∗ , x − ȳi < 0}. – p. 36 If T (ȳ) ⊂ W , by upper hemicontinuity, T (zt ) ⊂ W , for any t sufficiently closed to 0, which is a contradiction with (2). Finally ∃ ȳ ∗ ∈ T (ȳ) : hȳ ∗ , x − ȳi ≥ 0, that is inf sup hy ∗ , x − ȳi ≥ 0 x∈C y ∗ ∈T (ȳ) thus ȳ ∈ S(T, C) by a classical minimax theorem (Sion). – p. 37 Upper sign-continuity X∗ • T : X → 2 is said to be upper sign-continuous on C iff for any x, y ∈ C, one have : ∀ t ∈ ]0, 1[, ∗ inf hx , y − xi ≥ 0 ∗ x ∈T (xt ) =⇒ sup hx∗ , y − xi ≥ 0 x∗ ∈T (x) where xt = (1 − t)x + ty. – p. 38 Upper sign-continuity X∗ • T : X → 2 is said to be upper sign-continuous on C iff for any x, y ∈ C, one have : ∀ t ∈ ]0, 1[, ∗ inf hx , y − xi ≥ 0 ∗ x ∈T (xt ) =⇒ sup hx∗ , y − xi ≥ 0 x∗ ∈T (x) where xt = (1 − t)x + ty. upper semi-continuous ⇓ upper hemicontinuous ⇓ upper sign-continuous – p. 38 locally upper sign continuity X∗ Definition 35 Let T : C → 2 be a set-valued map. T est called locally upper sign-continuous on C if, for any x ∈ C there exist a neigh. Vx of x and a upper sign-continuous set-valued X∗ map Φx : Vx → 2 with nonempty convex w ∗ -compact values such that Φx (y) ⊆ T (y) \ {0}, ∀ y ∈ Vx – p. 39 Existence: complete form C nonempty convex compact subset of X. X∗ T : C → 2 quasimonotone + T (x) 6= ∅ convex w∗ -compact + upper hemicontinuous Then S(T, C) 6= ∅. – p. 40 Existence: complete form C nonempty convex compact subset of X. X∗ T : C → 2 quasimonotone + T (x) 6= ∅ convex w∗ -compact + upper sign-continuous Then S(T, C) 6= ∅. – p. 40 Existence: complete form C nonempty convex compact subset of X. X∗ T : C → 2 quasimonotone + T (x) 6= ∅ convex w∗ -compact + locally upper sign continuous on C Then S(T, C) 6= ∅. – p. 40 Existence: complete form C nonempty convex subset of X. X∗ T : C → 2 quasimonotone + locally upper sign continuous on C + coercivity condition: ∃ ρ > 0, ∀ x ∈ C \ B(0, ρ), ∃ y ∈ C with kyk < kxk such that ∀ x∗ ∈ T (x), hx∗ , x − yi ≥ 0 and there exists ρ′ > ρ such that C ∩ B(0, ρ′ ) is weakly compact (6= ∅). Then S(T, C) 6= ∅. – p. 40 II - d - Other assumptions Hartman and Stampacchia : sum of monotone operator and continuous function J.C. Yao, N.D. Yen, B.T. Kien and many others: continuous maps with degre technics – p. 41 III Other classes of V.I. – p. 42 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. – p. 43 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. Example : Mixed var. ineq. Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that hx̄∗ , y − x̄i + h(y) − h(x̄) ≥ 0, Also called hemi-variational inequality ∀ y ∈ C. – p. 43 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. Example : General var. ineq. Find x0 ∈ X such that hf (x0 ), g(x) − g(x0 )i ≥ 0, ∀ x ∈ X avec g(x) ∈ C. – p. 43 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. Example : Equilibrium problem Find x0 ∈ C such that f (x0 , x) + h(x) − h(x0 ) ≥ 0, ∀ x ∈ C. – p. 43 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. • Quasimonotonicity –> Φ-quasimonotonicity • Upper-sign continuity –> Φ-upper sign-continuity – p. 43 a - A general framework X, Y l.c.t.v.s., C subset of X T : C → Y set-valued map Coupling function : Φ : Y × C → IR • Variational inequality: Find x̄ ∈ C and x̄∗ ∈ T (x̄) such that Φ(x̄∗ , y) − Φ(x̄∗ , x̄) ≥ 0, ∀ y ∈ C. • Quasimonotonicity –> Φ-quasimonotonicity • Upper-sign continuity –> Φ-upper sign-continuity Hyp.: Φ(x∗ , ·) convex continue and Φ(·, x) concave. See Aussel-Luc, Bull. Austral. Math. Soc. (2005) – p. 43 b - Minty and local Minty V. I. Let C ⊆ X and T : C → ∗ X 2 be a map. We denote by • Sw (T, C) set of weak solutions of Stampacchia V.I. Find x̄ ∈ C such that ∀ y ∈ C, ∃ x̄∗ ∈ T (x̄) : hx̄∗ , y − x̄i ≥ 0 – p. 44 b - Minty and local Minty V. I. Let C ⊆ X and T : C → ∗ X 2 be a map. We denote by • Sw (T, C) set of weak solutions of Stampacchia V.I. Find x̄ ∈ C such that ∀ y ∈ C, ∃ x̄∗ ∈ T (x̄) : hx̄∗ , y − x̄i ≥ 0 • M (T, C) set of solutions of the Minty V.I. Find x̄ ∈ C such that ∀ y ∈ C and ∀ y ∗ ∈ T (y) : hy ∗ , y − x̄i ≥ 0. – p. 44 b - Minty and local Minty V. I. Let C ⊆ X and T : C → ∗ X 2 be a map. We denote by • Sw (T, C) set of weak solutions of Stampacchia V.I. Find x̄ ∈ C such that ∀ y ∈ C, ∃ x̄∗ ∈ T (x̄) : hx̄∗ , y − x̄i ≥ 0 • M (T, C) set of solutions of the Minty V.I. Find x̄ ∈ C such that ∀ y ∈ C and ∀ y ∗ ∈ T (y) : hy ∗ , y − x̄i ≥ 0. • LM (T, C) set of local solutions of Minty V.I., i.e. x̄ ∈ LM (T, C) if it exists a neighb. V of x̄ such that x̄ ∈ M (T, V ∩ C). – p. 44 Links ? X∗ Proposition 37 C nonempty convex subset of X and T : C → 2 a map. i) If T is pseudomonotone, then LM (T, C) = M (T, C). ii) If T is locally upper sign-continuous at x then LM (T, C) ⊆ Sw (T, C). iii) If, additionally to ii), the maps Sx are convex valued, then LM (T, C) ⊆ Sw (T, C) = S(T, C). – p. 45 Proper Quasimonotonicity: X∗ Let C ⊆ X and T : C → 2 . • T is properly quasimonotone if, for any x1 , . . . , xn ∈ dom T , and any x ∈ co {x1 , x2 , . . . , xn }, there exists i ∈ {1, 2, . . . , n} such that ∀x∗ ∈ T (xi ) : hx∗ , xi − xi ≥ 0. monotone ⇓ pseudomonotone ⇓ properly quasimonotone ⇓ quasimonotone – p. 46 Existence of Minty V.I. Proposition 39 X∗ Let T : X → 2 be a set-valued map. Then the following are equivalent: i) T is properly quasimonotone ii) ∀ C nonempty convex compact, M (T, C) 6= ∅. R. John, Proc. of the GCM7, Ed. Kluwer (1999) – p. 47 Existence of Minty V.I. Proposition 41 X∗ Let T : X → 2 be a set-valued map. Then the following are equivalent: i) T is properly quasimonotone ii) ∀ C nonempty convex compact, M (T, C) 6= ∅. The proof is based on the Ky Fan theorem: Let C be a closed convex set of X and D ⊂ C . If {Cx : x ∈ D} is a KKM family of closed convex subsets of C and one of them is compact, then ∩{Cx : x ∈ D} is nonempty. A family {Cx : x ∈ D} is a KKM if ∀ x1 , . . . , xn ∈ D, conv{x1 , . . . , xn } ⊂ ∪x∈D Cx – p. 47 Existence of Minty V.I. Proof of i) ⇒ ii). Let us define the set-valued map G : C → C by G(y) = {x ∈ C : ∀ y ∗ ∈ T (y), hy ∗ , y − xi ≥ 0}. By hypothesis, G(y) is nonempty, convex and compact. Now reformulating the proper quasimonotonicity of T , we have that x1 , . . . , xn ∈ C and ∀ x ∈ conv{x1 , . . . , xn }, x ∈ ∪i=1,n G(xi ) thus showing that the family {G(x) : x ∈ C} is a KKM family. By applying Ky Fan’s theorem, we have ∩y∈C G(y) 6= ∅. But any element of this intersection (say x̄) satisfies ∀ y ∈ C, ∀ y ∗ ∈ T (y), hy ∗ , y − x̄i ≥ 0 and is therefore a solution of the Minty variational inequality. – p. 48 Existence of Minty V.I. Proof of ii) ⇒ i). Let x1 , . . . , xn be elements of X. Then conv{x1 , . . . , xn } is a nonempty convex compact subset of X and consequently, M (T, conv{x1 , . . . , xn }) is non empty. This means that ∃ x̄ ∈ conv{x1 , . . . , xn } such that ∀ x ∈ conv{x1 , . . . , xn }, ∀ x∗ ∈ T (x), hx∗ , x − x̄i ≥ 0 and thus ∃ x̄ ∈ conv{x1 , . . . , xn } such that ∀ i = 1, . . . , n, x̄ ∈ G(xi ). As a conclusion ∀ x1 , . . . , xn ∈ X, ∃ x̄ ∈ conv{x1 , . . . , xn } such that x̄ ∈ ∩i=1,n G(xi ) or in other word, the family of closed convex {G(x) : x ∈ X} satisfies the Finite Intersection Property. It can be shown by induction on n (a good exercise) that the family {G(x) : x ∈ X} is a KKM family. – p. 49 Existence of Minty V.I. Proposition 43 X∗ Let T : X → 2 be a set-valued map. Then the following are equivalent: i) T is quasimonotone ii) ∀ x, y ∈ X, M (T, [x, y]) 6= ∅. Indeed T is quasimonotone ⇔ its restriction to any line segment is quasimonotone ⇔ its restriction to any line segment is properly quasimonotone – p. 50 Existence of local solutions of Minty V.I. Proposition 45 Let C be a nonempty convex subset of X and X∗ T : C → 2 be quasimonotone. Then one of the following assertions holds: i) T is properly quasimonotone ii) LM (T, C) 6= ∅. If, C is weakly compact, then LM (T, C) 6= ∅ in both cases. – p. 51 Existence of local solutions of Minty V.I. Proof. Suppose that i) does not hold. Thus there exists x1 , . . . , xn ∈ C, x∗i ∈ T (xi ) and x ∈ conv{x1 , . . . , xn } such that hx∗i , x − xi i > 0, ∀ i. By continuity of the x∗i ’s, there exists a neighb. U of x for which hx∗i , x − xi i > 0, ∀y ∈ C ∩ U and thus, according to the quasimonotonicity of T , for any y ∗ ∈ T (y), hy ∗ , y − xi i ≥ 0. Now since x ∈ conv{x1 , . . . , xn }, for any y ∈ C ∩ U and any y ∗ ∈ T (y), one has hy ∗ , y − xi ≥ 0 and x ∈ LM (T, C). – p. 52 c - Quasivariational inequalities X Banach space X ∗ its topological dual (w∗ -top.) h·, ·i the duality product Stampacchia Quasivariational inequality: X∗ Let C be a nonempty subset of X and T : X → 2 and ϕ : C → 2C be set-valued maps. Find x̄ ∈ C such that x̄ ∈ ϕ(x̄) and there exists x̄∗ ∈ T (x̄) for which hx̄∗ , y − x̄i ≥ 0, ∀ y ∈ ϕ(x̄). – p. 53 Theorem 47 Let C a convex compact and nonempty subset of X Assume that i) ϕ : C → 2C is usc with convex compact nonempty values ii) f : C × C → IR is a function such that ∀ y ∈ C, f (·, y)is lsc ∀ x ∈ C, f (x, ·)is quasiconcave ∀ x ∈ C, f (x, x) ≤ 0 iii) the subset {x ∈ C : supy∈ϕx f (x, y) > 0} is open Then there exists x̄ ∈ C such that x̄ ∈ ϕ(x̄) and supy∈ϕx̄ f (x̄, y) ≤ 0. can be found in Lassonde, Master course, UAG – p. 54 References 1- D. A. & N. Hadjisavvas, On Quasimonotone Variational Inequalities, JOTA 121 (2004), 445-450. 2- D. A. & D.T. Luc, Existence Conditions in General Quasimonotone Variational Inequalities, Bull. Austral. Math. Soc. 71 (2005), 285-303. 3- J.-C. Yao, Multivalued variational inequalities with K-pseudomonotone operators, JOTA (1994). 4- J.-P. Crouzeix, Pseudomonotone Variational Inequalities Problems: Existence of Solutions, Math. Programming (1997). 5- D. T. Luc, Existence results for densely pseudomonotone variational inequality, J. Math. Anal. Appl. (2001). 6- R. John, A note on Minty Variational Inequality and Generalized Monotonicity, Proc. GCM7 (2001). – p. 55 Brouwer fixed point theorem Theorem 48 Let C be a nonempty compact convex subset of IRn and f : C → IRn a continuous function. Then f admits on C at least a fixed point. – p. 56 Kakutani fixed point theorem Theorem 49 Let C be a nonempty compact convex subset of IRn and IRn F : C → 2 a cusco map. Then F admits on C at least a fixed point. – p. 57