1386 CHAPTER 42. MAXIMAL MONOTONE OPERATORS, HILBERT SPACE

(∫ 0

−h

∣∣∣∣∣∣∣∣1h∫ t+h

tg̃(u)du

∣∣∣∣∣∣∣∣pX

dt)1/p

+

(∫ T

T−h

∣∣∣∣∣∣∣∣1h∫ t+h

tg̃(u)du

∣∣∣∣∣∣∣∣pX

dt)1/p

≤ 1h

(∫ 0

−h

(∫ h

−h||g̃(u)||X du

)p

dt)1/p

+1h

(∫ T

T−h

(∫ T+h

T−h||g̃(u)||X du

)p

dt)1/p

which by Minkowski’s inequality for integrals is no larger than

≤ 1h

∫ h

−h

(∫ 0

−h||g̃(u)||pX dt

)1/p

du+1h

∫ T+h

T−h

(∫ T

T−h||g̃(u)||pX dt

)1/p

du

≤ 1h

∫ h

−hεdu+

1h

∫ T+h

T−hεdu = 4ε

whenever h is small enough because of the fact that ||g̃||pX ∈ L1 (R;X). Since ε is arbitrary,this shows ∥∥∥∥g̃∗φ h (·)−

1h

∫ (·)+h

(·)g(s)dsX[0,T−h] (·)

∥∥∥∥Lp(R;X)

→ 0

and also, it was shown above that

∥g̃∗φ h (·)− g̃∥Lp(R;X)→ 0

It follows that1h

∫ (·)+h

(·)g(s)dsX[0,T−h] (·)→ g̃

in Lp (R;X) and consequently in Lp ([0,T ] ;X) as well. But g̃ = g on [0,T ] .The following theorem is a form of the chain rule in which the derivative is replaced by

the subgradient.

Theorem 42.3.13 Suppose u ∈W 1,p ([0,T ] ;X) ,z ∈ Lp′ ([0,T ] ;X ′), and z(t) ∈ δφ (u(t))a.e t ∈ [0,T ] . Then the function, t→ φ (u(t)) is in L1 (0,T ) and its weak derivative equals⟨z,u′⟩ . In particular,

φ (u(t))−φ (u(0)) =∫ t

0

⟨z(s) ,u′ (s)

⟩ds

Proof: Modify u on a set of measure zero such that δφ (u(t)) ̸= /0 for all t. Next modifyz on a set of measure zero such that for ũ and z̃ the modified functions, z̃(t) ∈ δφ (ũ(t)) forall t. First I claim t→ φ (ũ(t)) is in L1 (0,T ). Pick t0 ∈ [0,T ] and let

z̃(t0) ∈ δφ (ũ(t0)) .

Then for t ∈ [0,T ],⟨z̃(t0) , ũ(t)− ũ(t0)⟩+φ (ũ(t0))≤

1386 CHAPTER 42. MAXIMAL MONOTONE OPERATORS, HILBERT SPACE‘Olly ptth P 1/p T ||] ptth Dp 1/p(/ 7 | (u)du ar) “(| 7 | F(u)du ar)—h h t xX T—h h t xX1 0 h Pp \1/p<= e@(u du) dt<i (L, (fi) le@ellxan) ar)1 T T+h Pp \1/pEL, Cf letolleaw) ar)T-h T—hwhich by Minkowski’s inequality for integrals is no larger than1 fh 0] > \/p 1 opT+h ro , \/p<7 _EE (fi ieeoiear) Sauer EL" (LT iecoiear) a1 h 1 T+h<- é€du+ — Edu = 4€~h l, hJT—nwhenever h is small enough because of the fact that ||g||% € L' (IR;X). Since € is arbitrary,this showszo,)-} [8 * h . __—A J(.)+0LP(R;X)8(s)ds2ior—nj()and also, it was shown above thatIIe* On (-) — 8llio~e:x) +0It follows that1 pote _if, 8O)4s%orn 48in L? (R;X) and consequently in L? ({0,7];X) as well. But g= gon [0,7]. BfThe following theorem is a form of the chain rule in which the derivative is replaced bythe subgradient.Theorem 42.3.13 Suppose u € W! ({0,7];X),z € L? ([0,7];X’), and z(t) € 56 (u(t))a.et €[0,T]. Then the function, t + @ (u(t)) is in L! (0,T) and its weak derivative equals(z,u’) . In particular,6 (u(0)) 9 (w(0)) = [ (266) ul (9))asProof: Modify u on a set of measure zero such that 6@ (u(t)) 40 for all t. Next modifyzon a set of measure zero such that for w and Z the modified functions, z(t) € 6 (w(t)) forall t. First I claim t + @ (a(t) is in L' (0,7). Pick to € [0,7] and letZ(to) € 5 (u(to)).Then for ¢ € [0,7],(Z(to) w(t) — u(to)) +9 (u(to)) <