arXiv:1005.2247v2 [math.FA] 14 Aug 2010

The sum of a maximal monotone operator of type (FPV) and a maximal monotone operator with full domain is maximal monotone Liangjin Yao∗ August 13, 2010

Abstract The most important open problem in Monotone Operator Theory concerns the maximal monotonicity of the sum of two maximal monotone operators provided that Rockafellar’s constraint qualification holds. In this paper, we prove the maximal monotonicity of A + B provided that A and B are maximal monotone operators such that dom A ∩ int dom B 6= ∅, A + Ndom B is of type (FPV), and dom A ∩ dom B ⊆ dom B. The proof utilizes the Fitzpatrick function in an essential way.

2010 Mathematics Subject Classification: Primary 47H05; Secondary 49N15, 52A41, 90C25 Keywords: Constraint qualification, convex function, convex set, duality mapping, Fitzpatrick function, linear relation, maximal monotone operator, monotone operator, monotone operator of type (FPV), subdifferential operator.

1

Introduction

Throughout this paper, we assume that X is a real Banach space with norm k · k, that X ∗ is the continuous dual of X, and that X and X ∗ are paired by h·, ·i. Let A : X ⇉ X ∗ be Mathematics, Irving K. Barber School, UBC Okanagan, Kelowna, British Columbia V1V 1V7, Canada. E-mail: [email protected]

1

∗ a set-valued operator (also known as multifunction) from X to X , i.e., for every x ∈ X, ∗ ∗ ∗ ∗ Ax ⊆ X , and let gra A = (x, x ) ∈ X × X | x ∈ Ax be the graph of A. Recall that A is monotone if

(1)

hx − y, x∗ − y ∗i ≥ 0,

∀(x, x∗ ) ∈ gra A ∀(y, y ∗) ∈ gra A,

and maximal monotone if A is monotone and A has no proper monotone extension (in the sense of graph inclusion). Let A : X ⇉ X ∗ be monotone and (x, x∗ ) ∈ X × X ∗ . We say (x, x∗ ) is monotonically related to gra A if hx − y, y − y ∗i ≥ 0,

∀(y, y ∗) ∈ gra A.

Let A : X ⇉ X ∗ be maximal monotone. We say A is of type (FPV) if for every open convex set U ⊆ X such that U ∩ dom A 6= ∅, the implication x ∈ Uand (x, x∗ ) is monotonically related to gra A ∩ U × X ∗ ⇒ (x, x∗ ) ∈ gra A holds. We say A is a linear relation if gra A is a linear subspace. Monotone operators have proven to be a key class of objects in modern Optimization and Analysis; see, e.g., the books [6, 7, 8, 10, 17, 18, 15, 24] and the references therein. We adopt standard notation used in these books: dom A = x ∈ X | Ax 6= ∅ is the domain of A. Given a subset C of X, int C is the interior of C, and C is the norm closure of C. The indicator function of C, written as ιC , is defined at x ∈ X by ( 0, if x ∈ C; (2) ιC (x) = ∞, otherwise. We set dist(x, C) = inf c∈C kx − ck, for x ∈ X. If D ⊆ X, we set C − D = {x − y | x ∈ C, y ∈ D}. For every x ∈ X, the normal cone operator of C at x is defined by NC (x) = x∗ ∈ X ∗ | supc∈C hc − x, x∗ i ≤ 0 , if x ∈ C; and NC (x) = ∅, if x ∈ / C. For x, y ∈ X, we set [x, y] = {tx+(1−t)y | 0 ≤ t ≤ 1}. Given f : X → ]−∞, +∞], we set dom f = f −1 (R) and f ∗ : X ∗ → [−∞, +∞] : x∗ 7→ supx∈X (hx, x∗ i − f  (x)) is the Fenchel conjugate of f . If f is convex and ∗ ∗ ∗ ∗ dom f 6= ∅, then ∂f : X ⇉ X : x 7→ x ∈ X | (∀y ∈ X) hy − x, x i + f (x) ≤ f (y) is the ∗ ∗ subdifferential operator of f . We also  set PX : X × X → X : (x, x ) 7→ x. Finally, the open unit ball in X is denoted by BX = x ∈ X | kxk < 1 , and N = {1, 2, 3, . . .}.

Let A and B be maximal monotone operators from X to X ∗ . Clearly, the sum operator A+  ∗ ∗ ∗ ∗ ∗ B : X ⇉ X : x 7→ Ax + Bx = a + b | a ∈ Ax and b ∈ Bx is monotone. Rockafellar’s [14, Theorem 1] guarantees maximal monotonicity of A + B under Rockafellar’s constraint qualification dom A ∩ int dom B 6= ∅ when X is reflexive — this result is often referred to as “the sum theorem”. The most famous open problem concerns the maximal monotonicity of A + B in nonreflexive Banach spaces when Rockafellar’s constraint qualification holds. 2

See Simons’ monograph [18] and [4, 5, 23] for a comprehensive account of some recent developments. Now we focus on the case when A and B satisfy the following three conditions: dom A ∩ int dom B 6= ∅, A + Ndom B is of type (FPV), and dom A ∩ dom B ⊆ dom B. We show that the sum A + B is maximal monotone in this setting. We note in passing that in [20, Corollary 2.9(a)], Verona and Verona derived the same conclusion when A is the subdifferential operator of a proper lower semicontinuous convex function, and B is maximal monotone with full domain. In [2, Theorem 3.1], it was recently shown that the sum theorem is true when A is a linear relation and B is the normal cone operator of a closed convex set. In [22], Voisei confirmed [17, Theorem 41.5] that the sum theorem is also true when A is type of (FPV) with convex domain, and B is the normal cone operator of a closed convex set. Our main result, Theorem 3.4, generalizes all the above results and it also contains a result due to Heisler [11, Remark, page 17] on the sum theorem for two operators with full domain. The remainder of this paper is organized as follows. In Section 2, we collect auxiliary results for future reference and for the reader’s convenience. The main result (Theorem 3.4) is proved in Section 3.

2

Auxiliary Results

Fact 2.1 (Rockafellar) (See [13, Theorem 3(b)], [18, Theorem 18.1], or [24, Theorem 2.8.7(iii)].) Let f, g : X → ]−∞, +∞] be proper convex functions. Assume that there exists a point x0 ∈ dom f ∩ dom g such that g is continuous at x0 . Then ∂(f + g) = ∂f + ∂g. Fact 2.2 (See [10, Theorem 2.28].) Let A : X ⇉ X ∗ be monotone with int dom A 6= ∅. Then A is locally bounded at x ∈ int dom A, i.e., there exist δ > 0 and K > 0 such that sup ky ∗k ≤ K,

∀y ∈ (x + δBX ) ∩ dom A.

y ∗ ∈Ay

Fact 2.3 (Fitzpatrick) (See [9, Corollary 3.9].) Let A : X ⇉ X ∗ be maximal monotone, and set  (3) FA : X × X ∗ → ]−∞, +∞] : (x, x∗ ) 7→ sup hx, a∗ i + ha, x∗ i − ha, a∗ i , (a,a∗ )∈gra A

the Fitzpatrick function associated with A. Then for every (x, x∗ ) ∈ X × X ∗ , the inequality hx, x∗ i ≤ FA (x, x∗ ) is true, and the equality holds if and only if (x, x∗ ) ∈ gra A. 3

Fact 2.4 (See [21, Theorem 3.4 and Corollary 5.6], S or [18, Theorem 24.1(b)].) Let A, B : X ⇉ X ∗ be maximal monotone operators. Assume λ>0 λ [PX (dom FA ) − PX (dom FB )] is a closed subspace. If (4)

FA+B ≥ h·, ·i on X × X ∗ ,

then A + B is maximal monotone. Fact 2.5 (Simons) (See [18, Thereom 27.1 and Thereom 27.3].) Let A : X ⇉ X ∗ be maximal monotone with int dom A 6= ∅. Then int dom A = int [PX dom FA ], dom A = PX [dom FA ] and dom A is convex. Now we cite some results on maximal monotone operators of type (FPV). Fact 2.6 (Simons) (See [18, Theorem 48.4(d)].) Let f : X → ]−∞, +∞] be proper, lower semicontinuous, and convex. Then ∂f is of type (FPV). Fact 2.7 (Simons) (See [18, Theorem 46.1].) Let A : X ⇉ X ∗ be a maximal monotone linear relation. Then A is of type (FPV). Fact 2.8 (Simons and Verona-Verona) (See [18, Thereom 44.1] or [19].) Let A : X ⇉ X ∗ be a maximal monotone. Suppose that for every closed convex subset C of X with dom A ∩ int C 6= ∅, the operator A + NC is maximal monotone. Then A is of type (FPV). The following statement first appeared in [17, Theorem 41.5]. However, on [18, page 199], concerns were raised about the validity of the proof of [17, Theorem 41.5]. In [22], Voisei recently provided a result that generalizes and confirms [17, Theorem 41.5] and hence the following fact. Fact 2.9 (Voisei) Let A : X ⇉ X ∗ be maximal monotone of type (FPV) with convex domain, let C be a nonempty closed convex subset of X, and suppose that dom A∩int C 6= ∅. Then A + NC is maximal monotone. Corollary 2.10 Let A : X ⇉ X ∗ be maximal monotone of type (FPV) with convex domain, let C be a nonempty closed convex subset of X, and suppose that dom A ∩ int C 6= ∅. Then A + NC is of type (F P V ). Proof. By Fact 2.9, A + NC is maximal monotone. Let D be a nonempty closed convex subset of X, and suppose that dom(A + NC ) ∩ int D 6= ∅. Let x1 ∈ dom A ∩ int C and x2 ∈ dom(A + NC ) ∩ int D. Thus, there exists δ > 0 such that x1 + δBX ⊆ C and x2 + δBX ⊆ D. Then for small enough λ ∈ ]0, 1[, we have x2 + λ(x1 − x2 ) + 21 δBX ⊆ D. Clearly, B ⊆ C ∩ D. Since dom A x2 + λ(x1 − x2 ) + λδBX ⊆ C. Thus x2 + λ(x1 − x2 ) + λδ 2 X 4

is convex, x2 + λ(x1 − x2 ) ∈ dom A and x2 + λ(x1 − x2 ) ∈ dom A ∩ int(C ∩ D). By Fact 2.1 , A + NC + ND = A + NC∩D . Then, by Fact 2.9 (applied to A and C ∩ D), A + NC + ND = A + NC∩D is maximal monotone. By Fact 2.8, A + NC is of type (F P V ).  Corollary 2.11 Let A : X ⇉ X ∗ be a maximal monotone linear relation, let C be a nonempty closed convex subset of X, and suppose that dom A ∩ int C 6= ∅. Then A + NC is of type (F P V ). Proof. Apply Fact 2.7 and Corollary 2.10.

3



Main Result

The following result plays a key role in the proof of Theorem 3.4. The first half of its proof follows along the lines of the proof of [18, Theorem 44.2]. Proposition 3.1 Let A, B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that A + Ndom B is maximal monotone of type (FPV), and dom A ∩ dom B ⊆ dom B. Then PX [dom FA+B ] = dom A ∩ dom B. Proof. By [9, Theorem 3.4], dom A ∩ dom B = dom(A + B) ⊆ PX [dom FA+B ]. It suffices to show that (5)

PX [dom FA+B ] ⊆ dom A ∩ dom B.

After translating the graphs if necessary, we can and do assume that 0 ∈ dom A ∩ int dom B and that (0, 0) ∈ gra B. To show (5), we take z ∈ PX [dom FA+B ] and we assume to the contrary that (6)

z∈ / dom A ∩ dom B.

Thus α = dist(z, dom A ∩ dom B) > 0. Now take y0∗ ∈ X ∗ such that (7)

ky0∗k = 1 and hz, y0∗ i ≥ 23 kzk.

Set (8)

Un = [0, z] +

α B , 4n X

5

∀n ∈ N.

Since 0 ∈ Ndom B (x), ∀x ∈ dom B, gra B ⊆ gra(B + Ndom B ). Since B is maximal monotone and B + Ndom B is a monotone extension of B, we must have B = B + Ndom B . Thus A + B = A + Ndom B + B.

(9)

Since dom A ∩ dom B ⊆ dom B by assumption, we obtain dom A ∩ dom B ⊆ dom(A + Ndom B ) = dom A ∩ dom B ⊆ dom A ∩ dom B. Hence dom A ∩ dom B = dom(A + Ndom B ).

(10)

By (6) and (10), z ∈ / dom(A + Ndom B ) and thus (z, ny0∗ ) ∈ / gra(A + Ndom B ), ∀n ∈ N. For every n ∈ N, since z ∈ Un and since A + Ndom B is of type (FPV) by assumption, we deduce the existence of (zn , zn∗ ) ∈ gra(A + Ndom B ) such that zn ∈ Un and hz − zn , zn∗ i > nhz − zn , y0∗i,

(11)

∀n ∈ N.

Hence, using (8), there exists λn ∈ [0, 1] such that (12)

kz − zn − λn zk = kzn − (1 − λn )zk < 14 α,

∀n ∈ N.

By the triangle inequality, we have kz − zn k < λn kzk + 14 α for every n ∈ N. From the definition of α and (10), it follows that α ≤ kz − zn k and hence that α < λn kzk + 14 α. Thus, 3 α 4

(13)

< λn kzk,

∀n ∈ N.

By (12) and (7), (14)

hz − zn − λn z, y0∗ i ≥ −kzn − (1 − λn )zk > − 41 α,

∀n ∈ N.

By (14), (7) and (13), (15)

hz − zn , y0∗ i > λn hz, y0∗i − 14 α >

23 α 34

− 41 α = 14 α,

∀n ∈ N.

Then, by (11) and (15), (16)

hz − zn , zn∗ i > 41 nα,

∀n ∈ N.

α By (8), there exist tn ∈ [0, 1] and bn ∈ 4n BX such that zn = tn z + bn . Since tn ∈ [0, 1], there exists a convergent subsequence of (tn )n∈N , which, for convenience, we still denote by (tn )n∈N . Then tn → β, where β ∈ [0, 1]. Since bn → 0, we have

(17)

zn → βz. 6

By (10), zn ∈ dom A ∩ dom B; thus, kzn − zk ≥ α and β ∈ [0, 1[. In view of (9) and (16), we have, for every z ∗ ∈ X ∗ , FA+B (z, z ∗ ) = FA+Ndom B +B (z, z ∗ ) ≥

sup {n∈N,y ∗ ∈X ∗ }



(18)

sup {n∈N,y ∗ ∈X ∗ }

We now claim that

[hzn , z ∗ i + hz − zn , zn∗ i + hz − zn , y ∗i − ιgra B (zn , y ∗)]   hzn , z ∗ i + 41 nα + hz − zn , y ∗i − ιgra B (zn , y ∗) . FA+B (z, z ∗ ) = ∞.

(19) We consider two cases. Case 1 : β = 0.

By (17) and Fact 2.2 (applied to 0 ∈ int dom B), there exist N ∈ N and K > 0 such that (20)

Bzn 6= ∅ and

sup ky ∗ k ≤ K, y ∗ ∈Bz

∀n ≥ N.

n

Then, by (18),   hzn , z ∗ i + 41 nα + hz − zn , y ∗i − ιgra B (zn , y ∗) {n≥N,y ∗ ∈X ∗ }   ≥ sup −kzn k · kz ∗ k + 41 nα − kz − zn k · ky ∗k {n≥N,y ∗ ∈Bzn }   ≥ sup −kzn k · kz ∗ k + 41 nα − Kkz − zn k (by (20))

FA+B (z, z ∗ ) ≥

sup

{n≥N }

= ∞ (by (17)).

Thus (19) holds. Case 2 : β 6= 0. Take vn∗ ∈ Bzn . We consider two subcases. Subcase 2.1 : (vn∗ )n∈N is bounded. By (18),   FA+B (z, z ∗ ) ≥ sup hzn , z ∗ i + 41 nα + hz − zn , vn∗ i {n∈N}

  ≥ sup −kzn k · kz ∗ k + 14 nα − kz − zn k · kvn∗ k {n∈N}

= ∞ (by (17) and the boundedness of (vn∗ )n∈N ). 7

Hence (19) holds. Subcase 2.2 : (vn∗ )n∈N is unbounded. We first show lim sup hz − zn , vn∗ i ≥ 0.

(21)

n→∞

Since (vn∗ )n∈N is unbounded and after passing to a subsequence if necessary, we assume that kvn∗ k = 6 0, ∀n ∈ N and that kvn∗ k → +∞. By 0 ∈ int dom B and Fact 2.2, there exist δ > 0 and M > 0 such that (22)

sup ky ∗k ≤ M,

By 6= ∅ and

∀y ∈ δBX .

y ∗ ∈By

Then we have hzn − y, vn∗ − y ∗ i ≥ 0, ∀y ∈ δBX , y ∗ ∈ By, n ∈ N ⇒ hzn , vn∗ i − hy, vn∗ i + hzn − y, −y ∗i ≥ 0, ∀y ∈ δBX , y ∗ ∈ By, n ∈ N ⇒ hzn , vn∗ i − hy, vn∗ i ≥ hzn − y, y ∗i, ∀y ∈ δBX , y ∗ ∈ By, n ∈ N ⇒ hzn , vn∗ i − hy, vn∗ i ≥ −(kzn k + δ)M, ∀y ∈ δBX , n ∈ N (by (22)) ⇒ hzn , vn∗ i ≥ hy, vn∗ i − (kzn k + δ)M, ∀y ∈ δBX , n ∈ N ⇒ hzn , vn∗ i ≥ δkvn∗ k − (kzn k + δ)M, ∀n ∈ N (23)

⇒ hzn , kvvn∗ k i ≥ δ − ∗

n

(kzn k+δ)M , ∗k kvn

∀n ∈ N.

By the Banach-Alaoglu Theorem (see [16, Theorem 3.15]), there exist a weak* convergent subnet (vγ∗ )γ∈Γ of (vn∗ )n∈N , say vγ∗ w* ∗ ⇁w kvγ∗ k

(24)

∈ X ∗.

Using (17) and taking the limit in (23) along the subnet, we obtain hβz, w ∗i ≥ δ.

(25) Since β > 0, we have

hz, w ∗ i ≥

(26)

δ β

> 0.

Now we assume to the contrary that lim suphz − zn , vn∗ i < −ε, n→∞

for some ε > 0. 8

Then, for all n sufficiently large, hz − zn , vn∗ i < − 2ε , and so (27)

hz − zn , kvvn∗ k i < − 2kvε∗ k . ∗

n

n

Then by (17) and (24), taking the limit in (27) along the subnet again, we see that hz − βz, w ∗ i ≤ 0. Since β < 1, we deduce hz, w ∗ i ≤ 0 which contradicts (26). Hence (21) holds. By (18),   FA+B (z, z ∗ ) ≥ sup hzn , z ∗ i + 14 nα + hz − zn , vn∗ i {n∈N}   ≥ sup −kzn k · kz ∗ k + 14 nα + hz − zn , vn∗ i {n∈N}   ≥ lim sup −kzn k · kz ∗ k + 14 nα + hz − zn , vn∗ i n→∞

= ∞ (by (17) and (21)).

Hence (28)

FA+B (z, z ∗ ) = ∞.

Therefore, we have verified (19) in all cases. However, (19) contradicts our original choice that z ∈ PX [dom FA+B ]. Hence PX [dom FA+B ] ⊆ dom A ∩ dom B and thus (5) holds. Thus PX [dom FA+B ] = dom A ∩ dom B.  Corollary 3.2 Let A : X ⇉ X ∗ be maximal monotone of type (FPV) with convex domain, and B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that dom A ∩ dom B ⊆ dom B. Then PX [dom FA+B ] = dom A ∩ dom B. Proof. Combine Fact 2.5, Corollary 2.10 and Proposition 3.1.



Corollary 3.3 Let A : X ⇉ X ∗ be a maximal monotone linear relation, and let B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that dom A ∩ dom B ⊆ dom B. Then PX [dom FA+B ] = dom A ∩ dom B. Proof. Combine Fact 2.5, Corollary 2.11 and Proposition 3.1. Alternatively, combine Fact 2.7 and Corollary 3.2.  We are now ready for our main result. 9

Theorem 3.4 (Main Result) Let A, B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that A + Ndom B is maximal monotone of type (FPV), and that dom A ∩ dom B ⊆ dom B. Then A + B is maximal monotone. Proof. After translating the graphs if necessary, we can and do assume that 0 ∈ dom A ∩ int dom B and that (0, 0) ∈ gra A ∩ gra B. By Fact 2.3, dom A ⊆ PX (dom FA ) and dom B ⊆ PX (dom FB ). Hence, [  (29) λ PX (dom FA ) − PX (dom FB ) = X. λ>0

Thus, by Fact 2.4, it suffices to show that

FA+B (z, z ∗ ) ≥ hz, z ∗ i,

(30)

∀(z, z ∗ ) ∈ X × X ∗ .

Take (z, z ∗ ) ∈ X × X ∗ . Then (31)

FA+B (z, z ∗ ) = sup [hx, z ∗ i + hz, x∗ i − hx, x∗ i + hz − x, y ∗ i − ιgra A (x, x∗ ) − ιgra B (x, y ∗ )] . {x,x∗ ,y ∗ }

Assume to the contrary that FA+B (z, z ∗ ) < hz, z ∗ i.

(32)

Then (z, z ∗ ) ∈ dom FA+B and, by Proposition 3.1, z ∈ dom A ∩ dom B = PX [dom FA+B ].

(33) Next, we show that

FA+B (λz, λz ∗ ) ≥ λ2 hz, z ∗ i,

(34)

∀λ ∈ ]0, 1[ .

Let λ ∈ ]0, 1[. By (33) and Fact 2.5, z ∈ PX dom FB . By Fact 2.5 again and 0 ∈ int dom B, 0 ∈ int PX dom FB . Then, by [24, Theorem 1.1.2(ii)], we have λz ∈ int PX dom FB = int [PX dom FB ] .

(35)

Combining (35) and Fact 2.5, we see that λz ∈ int dom B. We consider two cases. Case 1 : λz ∈ dom A. By (31), FA+B (λz, λz ∗ ) ≥ sup [hλz, λz ∗ i + hλz, x∗ i − hλz, x∗ i + hλz − λz, y ∗ i − ιgra A (λz, x∗ ) − ιgra B (λz, y ∗ )] {x∗ ,y ∗ }

= hλz, λz ∗ i. 10

Hence (34) holds. Case 2 : λz ∈ / dom A. Using 0 ∈ dom A ∩ dom B and the convexity of dom A ∩ dom B (which follows from (33)), we obtain λz ∈ dom A ∩ dom B ⊆ dom A ∩ dom B. Set Un = λz + n1 BX ,

(36)

∀n ∈ N.

/ gra(A+Ndom B ), λz ∈ Un , and A+Ndom B Then Un ∩dom(A+Ndom B ) 6= ∅. Since (λz, λz ∗ ) ∈ is of type (FPV), there exists (bn , b∗n ) ∈ gra(A + Ndom B ) such that bn ∈ Un and (37)

hλz, b∗n i + hbn , λz ∗ i − hbn , b∗n i > λ2 hz, z ∗ i,

∀n ∈ N.

Since λz ∈ int dom B and bn → λz, by Fact 2.2, there exist N ∈ N and M > 0 such that (38)

bn ∈ int dom B

and

sup kv ∗ k ≤ M,

∀n ≥ N.

v∗ ∈Bbn

Hence Ndom B (bn ) = {0} and thus (bn , b∗n ) ∈ gra A for every n ≥ N. Thus by (31), (37) and (38), FA+B (λz, λz ∗ ) ≥ sup [hbn , λz ∗ i + hλz, b∗n i − hbn , b∗n i + hλz − bn , v ∗ i] , ∀n ≥ N {v∗ ∈Bbn }   ≥ sup λ2 hz, z ∗ i + hλz − bn , v ∗ i , ∀n ≥ N (by (37)) {v∗ ∈Bbn }

(39)

  ≥ sup λ2 hz, z ∗ i − Mkλz − bn k , ≥ λ2 hz, z ∗ i (by bn → λz).

∀n ≥ N

(by (38))

Hence FA+B (λz, λz ∗ ) ≥ λ2 hz, z ∗ i. We have verified that (34) holds in both cases. Since (0, 0) ∈ gra A ∩ gra B, we obtain (∀(x, x∗ ) ∈ gra(A + B)) hx, x∗ i ≥ 0. Thus, FA+B (0, 0) = 0. Now define f : [0, 1] → R : t → FA+B (tz, tz ∗ ). Then f is continuous on [0, 1] by [24, Proposition 2.1.6]. From (34), we obtain (40)

FA+B (z, z ∗ ) = lim− FA+B (λz, λz ∗ ) ≥ lim− hλz, λz ∗ i = hz, z ∗ i, λ→1

λ→1

which contradicts (32). Hence (41)

FA+B (z, z ∗ ) ≥ hz, z ∗ i.

Therefore, (30) holds, and A + B is maximal monotone. Theorem 3.4 allows us to deduce both new and previously known sum theorems. 11



Corollary 3.5 Let f : X → ]−∞, +∞] be proper, lower semicontinuous, convex, and let B : X ⇉ X ∗ be maximal monotone with dom f ∩ int dom B 6= ∅. Assume that dom ∂f ∩ dom B ⊆ dom B. Then ∂f + B is maximal monotone. Proof. By Fact 2.5 and Fact 2.1, ∂f + Ndom B = ∂(f + ιdom B ). Then by Fact 2.6, ∂f + Ndom B is type of (FPV). Now apply Theorem 3.4.  Corollary 3.6 Let A : X ⇉ X ∗ be maximal monotone of type (FPV), and let B : X ⇉ X ∗ be maximal monotone with full domain. Then A + B is maximal monotone. Proof. Since A + Ndom B = A + NX = A and thus A + Ndom B is maximal monotone of type (FPV), the conclusion follows from Theorem 3.4.  Corollary 3.7 (Verona-Verona) (See [20, Corollary 2.9(a)] or [18, Theorem 53.1].) Let f : X → ]−∞, +∞] be proper, lower semicontinuous, and convex, and let B : X ⇉ X ∗ be maximal monotone with full domain. Then ∂f + B is maximal monotone. Proof. Clear from Corollary 3.5. Alternatively, combine Fact 2.6 and Corollary 3.6.



Corollary 3.8 (Heisler) (See [11, Remark, page 17].) Let A, B : X ⇉ X ∗ be maximal monotone with full domain. Then A + B is maximal monotone. Proof. Let C be a nonempty closed convex subset of X. By Corollary 3.7, NC + A is maximal monotone. Thus, A is of type (FPV) by Fact 2.8. The conclusion now follows from Corollary 3.6.  Corollary 3.9 Let A : X ⇉ X ∗ be maximal monotone of type (FPV) with convex domain, and let B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that dom A ∩ dom B ⊆ dom B. Then A + B is maximal monotone. Proof. Combine Fact 2.5, Corollary 2.10 and Theorem 3.4.



Corollary 3.10 (Voisei) (See [22].) Let A : X ⇉ X ∗ be maximal monotone of type (FPV) with convex domain, let C be a nonempty closed convex subset of X, and suppose that dom A ∩ int C 6= ∅. Then A + NC is maximal monotone. Proof. Apply Corollary 3.9.



Corollary 3.11 Let A : X ⇉ X ∗ be a maximal monotone linear relation, and let B : X ⇉ X ∗ be maximal monotone with dom A ∩ int dom B 6= ∅. Assume that dom A ∩ dom B ⊆ dom B. Then A + B is maximal monotone. 12

Proof. Combine Fact 2.7 and Corollary 3.9.



Corollary 3.12 (See [2, Theorem 3.1].) Let A : X ⇉ X ∗ be a maximal monotone linear relation, let C be a nonempty closed convex subset of X, and suppose that dom A∩int C 6= ∅. Then A + NC is maximal monotone. Proof. Apply Corollary 3.11.



Corollary 3.13 Let A : X ⇉ X ∗ be a maximal monotone linear relation, and let B : X ⇉ X ∗ be maximal monotone with full domain. Then A + B is maximal monotone. Proof. Apply Corollary 3.11.



Example 3.14 Suppose that X = L1 [0, 1], let  D = x ∈ X | x is absolutely continuous, x(0) = 0, x′ ∈ X ∗ , and set

A : X ⇉ X ∗ : x 7→

(

{x′ }, if x ∈ D; ∅, otherwise.

By Phelps and Simons’ [12, Example 4.3], A is an at most single-valued maximal monotone linear relation with proper dense domain, and A is neither symmetric nor skew. Now let J be the duality mapping, i.e., J = ∂ 12 k · k2. Then Corollary 3.13 implies that A + J is maximal monotone. To the best of our knowledge, the maximal monotonicity of A + J cannot be deduced from any previously known result. Remark 3.15 In [3], it was shown that the sum theorem is true when A is a linear relation, B is the subdifferential operator of a proper lower semicontinuous sublinear function, and Rockafellar’s constraint qualification holds. When the domain of the subdifferential operator is closed, then that result can be deduced from Theorem 3.4. However, it is possible that the domain of the subdifferential operator of a proper lower semicontinuous sublinear function does not have to be closed. For an example, see [1, Example 5.4]: Set C = {(x, y) ∈ R2 | 0 < 1/x ≤ y} and f = ι∗C . Then f is not subdifferentiable at any point in the boundary of its domain, except at the origin. Thus, in the general case, we do not know whether or not it is possible to deduce the result in [3] from Theorem 3.4.

Acknowledgment The author thanks Dr. Heinz Bauschke and Dr. Xianfu Wang for their valuable discussions and comments. The author also thanks Dr. Robert Csetnek for his pertinent comments. 13

References [1] S. Bartz, H.H. Bauschke, J.M. Borwein, S. Reich, and X. Wang, “Fitzpatrick functions, cyclic monotonicity and Rockafellar’s antiderivative”, Nonlinear Analysis, vol. 66, pp. 1198–1223, 2007. [2] H.H. Bauschke, X. Wang, and L. Yao, “An answer to S. Simons’ question on the maximal monotonicity of the sum of a maximal monotone linear operator and a normal cone operator”, Set-Valued and Variational Analysis, vol. 17, pp. 195–201, 2009. [3] H.H. Bauschke, X. Wang, and L. Yao, “On the maximal monotonicity of the sum of a maximal monotone linear relation and the subdifferential operator of a sublinear function”, to appear Proceedings of the Haifa Workshop on Optimization Theory and Related Topics. Contemp. Math., Amer. Math. Soc., Providence, RI ; http://arxiv.org/abs/1001.0257v1, January 2010. [4] J.M. Borwein, “Maximal monotonicity via convex analysis”, Journal of Convex Analysis, vol. 13, pp. 561–586, 2006. [5] J.M. Borwein, “Maximality of sums of two maximal monotone operators in general Banach space”, Proceedings of the American Mathematical Society, vol. 135, pp. 3917– 3924, 2007. [6] J.M. Borwein and J.D. Vanderwerff, Convex Functions, Cambridge University Press, 2010. [7] R.S. Burachik and A.N. Iusem, Set-Valued Mappings and Enlargements of Monotone Operators, Springer-Verlag, 2008. [8] D. Butnariu and A.N. Iusem, Totally Convex Functions for Fixed Points Computation and Infinite Dimensional Optimization, Kluwer Academic Publishers, 2000. [9] S. Fitzpatrick, “Representing monotone operators by convex functions”, in Workshop/Miniconference on Functional Analysis and Optimization (Canberra 1988), Proceedings of the Centre for Mathematical Analysis, Australian National University, vol. 20, Canberra, Australia, pp. 59–65, 1988. [10] R.R. Phelps, Convex Functions, Monotone Operators and Differentiability, 2nd Edition, Springer-Verlag, 1993. [11] R.R. Phelps, “Lectures on maximal monotone operators”, Extracta Mathematicae, vol. 12, pp. 193–230, 1997; http://arxiv.org/abs/math/9302209v1, February 1993.

14

[12] R.R. Phelps and S. Simons, “Unbounded linear monotone operators on nonreflexive Banach spaces”, Journal of Convex Analysis, vol. 5, pp. 303–328, 1998. [13] R.T. Rockafellar, “Extension of Fenchel’s duality theorem for convex functions”, Duke Mathematical Journal, vol. 33, pp. 81–89, 1966. [14] R.T. Rockafellar, “On the maximality of sums of nonlinear monotone operators”, Transactions of the American Mathematical Society, vol. 149, pp. 75–88, 1970. [15] R.T. Rockafellar and R.J-B Wets, Variational Analysis, 2nd Printing, Springer-Verlag, 2004. [16] R. Rudin, Functinal Analysis, Second Edition, McGraw-Hill, 1991. [17] S. Simons, Minimax and Monotonicity, Springer-Verlag, 1998. [18] S. Simons, From Hahn-Banach to Monotonicity, Springer-Verlag, 2008. [19] A. Verona and M.E. Verona, “Regular maximal monotone operators”, Set-Valued Analysis, vol. 6, pp. 303–312, 1998. [20] A. Verona and M.E. Verona, “Regular maximal monotone operators and the sum theorem”, Journal of Convex Analysis, vol. 7, pp. 115–128, 2000. [21] M.D. Voisei, “The sum and chain rules for maximal monotone operators”, Set-Valued and Variational Analysis, vol. 16, pp. 461–476, 2008. [22] M.D. Voisei, “A Sum Theorem for (FPV) operators and normal cones ”, Journal of Mathematical Analysis and Applications, vol. 371, pp. 661–664, 2010. [23] M.D. Voisei and C. Z˘alinescu, “Maximal monotonicity criteria for the composition and the sum under weak interiority conditions”, Mathematical Programming (Series B), vol. 123, pp. 265–283, 2010. [24] C. Z˘alinescu, Convex Analysis in General Vector Spaces, World Scientific Publishing, 2002.

15

The sum of a maximal monotone operator of type (FPV) and a maximal ...

Aug 14, 2010 - Keywords: Constraint qualification, convex function, convex set, duality mapping, Fitz- patrick function, linear relation, maximal monotone ...

184KB Sizes 0 Downloads 165 Views

Recommend Documents

On the maximal monotonicity of the sum of a maximal monotone linear ...
Jan 1, 2010 - Throughout, we shall identify X with its canonical image in the bidual ..... Council of Canada and by the Canada Research Chair Program.

Examples of discontinuous maximal monotone linear operators and ...
Sep 14, 2009 - http://arxiv.org/abs/0802.1375v1, February 2008. [5] H.H. ... [9] R. Cross, Multivalued Linear Operators, Marcel Dekker, Inc, New York, 1998.

Equivalence of Utilitarian Maximal and Weakly Maximal Programs"
Feb 5, 2009 - of utilitarian maximal programs coincides with the set of weakly .... A program 0kt1 from k % 0 is good if there exists some G ) R such that,. N.

Maximal charge injection of a uniform separated electron ... - inspire-hep
Dec 21, 2015 - A charge sheet model is proposed to study the space charge effect and ... bution of this work must maintain attribution to the author(s) and.

A Maximal Property of Cyclic Quadrilaterals
May 10, 2005 - This completes the proof that among quadrilaterals of given side lengths, the cyclic one has greatest area. Reference. [1] N. D. Kazarinoff, Geometric Inequalities, Yale University, 1961. Antreas Varverakis: Department of Mathematics,

Efficient Mining of Large Maximal Bicliques - CiteSeerX
Graphs can be used to model a wide range of real world applications. In this ... increasingly large multigene data sets from sequence databases [18]. Deter-.

On the Computation of Maximal-Correlated Cuboids Cells
since the introduction of data warehousing, OLAP, and the data cube ... In this paper, we propose a new iceberg cube mining method for reducing cuboids.

ON THE VERTEX SEPARATION OF MAXIMAL ...
sense, binary 2-trees (we explain later precisely what we mean). A fast and ... cycle. The length of a path p is the number of vertices in it and is denoted by. |p|.

Maximal charge injection of consecutive electron pulses with uniform ...
Aug 13, 2015 - charge effect will limit the maximum injected current den- sity. ... In the past 50 years .... easy illustration, a diagram of the injection of two charge.

Operator Type Operator Java Flow of Control - GitHub
Operator Type. Operator. Java Flow of Control. Unary ! ++ -- + -. Java Evaluates First. Mul plica ve. * / %. Addi ve. + -. Rela onal. < > = Equality. == !=

Continuous Maximal Flows and Wulff Shapes
combinatorial) domain often results in intrinsically data- parallel algorithms, that can be significantly accelerated. e.g. by modern graphics processing units.

Mining maximal quasi-bicliques: Novel algorithm and ...
Oct 15, 2009 - Market and Protein Networks ... 2 School of Computer Engineering, Nanyang Technological University, Singapore .... tions do not have a good constraint on the vertices to have .... Application 2: Mining protein networks.