42.3. SUBGRADIENTS 1389
whenever (y,a) ∈C1 and (y1,a1) ∈C2.Claim: β > 0.Proof of claim: If β < 0 let
a = φ 1 (x)− x∗ (x− x)−φ 1 (x)+1,
a1 = φ 2 (x)−φ 2 (x) , and y = y1 = x.
Then from 42.3.25
β (φ 1 (x)− x∗ (x− x)−φ 1 (x)+1)≥ β (φ 2 (x)−φ 2 (x)) .
Dividing by β yields
φ 1 (x)− x∗ (x− x)−φ 1 (x)+1≤ φ 2 (x)−φ 2 (x)
and soφ 1 (x)+φ 2 (x)− (φ 1 (x)+φ 2 (x))+1≤ x∗ (x− x)
≤ φ 1 (x)+φ 2 (x)− (φ 1 (x)+φ 2 (x)),
a contradiction. Therefore, β ≥ 0.Now suppose β = 0. Letting
a = φ 1 (x)− x∗ (x− x)−φ 1 (x)+1,
(x,a) ∈ int(C1) ,
and so there exists an open set U containing 0 and η > 0 such that
x+U× (a−η ,a+η)⊆C1.
Therefore, 42.3.25 applied to (x+ z,a) ∈C1 and (x,φ 2 (x)−φ 2 (x)) ∈C2 for z ∈U yields
w∗ (x+ z)≥ w∗ (x)
for all z ∈U . Hence w∗ (z) = 0 on U which implies w∗ = 0, contradicting 42.3.24. Thisproves the claim.
Now with the claim, it follows β > 0 and so, letting z∗ = w∗/β , 42.3.25 and Lemma18.2.15 implies
z∗ (y)+a≥ z∗ (y1)+a1 (42.3.26)
whenever (y,a) ∈C1 and (y1,a1) ∈C2. In particular,
(y,φ 1 (y)−φ 1 (x)− x∗ (y− x)) ∈C1 (42.3.27)
becauseφ 1 (y)−φ 1 (x)− x∗ (y− x)≤ φ 1 (y)− x∗ (y− x)−φ 1 (x)
and(y1,φ 2 (x)−φ 2 (y1)) ∈C2. (42.3.28)