13.1. SARD’S LEMMA AND APPROXIMATION 303

be in ∂K. Therefore, K = K1 but this also is a contradiction because if x ∈ ∂K then x /∈ Kthanks to the first part that ∂U =U \U . ■

Note that for an open set U ⊆ Rp, and h : U → Rp, dist(h(∂U) ,y)≥ dist(h(U),y)

because U ⊇ ∂U .The following lemma will be nice to keep in mind.

Lemma 13.0.4 f ∈C(Ω× [a,b] ;Rp

)if and only if

t→ f (·, t) ∈C([a,b] ;C

(Ω;Rp))

Also∥f∥

∞,Ω×[a,b] = maxt∈[a,b]

(∥f (·,t)∥

∞,Ω

)Proof:⇒By uniform continuity, if ε > 0 there is δ > 0 such that if |t− s|< δ , then for

all x ∈Ω, ∥f (x,t)−f (x,s)∥< ε

2 . It follows that

∥f (·, t)−f (·,s)∥∞≤ ε

2< ε

⇐Say (xn, tn)→ (x,t) . Does it follow that f (xn, tn)→ f (x,t)?

∥f (xn, tn)−f (x,t)∥ ≤ ∥f (xn, tn)−f (xn, t)∥+∥f (xn, t)−f (x, t)∥≤ ∥f (·, tn)−f (·, t)∥

∞+∥f (xn, t)−f (x, t)∥

both terms converge to 0, the first because f is continuous into C(Ω;Rp

)and the second

because x→ f (x, t) is continuous.The claim about the norms is next. Let (x, t) be such that ∥f∥

∞,Ω×[a,b] < ∥f (x, t)∥+ε .Then

∥f∥∞,Ω×[a,b] < ∥f (x, t)∥+ ε ≤ max

t∈[a,b]

(∥f (·, t)∥

∞,Ω

)+ ε

and so ∥f∥∞,Ω×[a,b] ≤ maxt∈[a,b] max

(∥f (·,t)∥

∞,Ω

)because ε is arbitrary. However, the

same argument works in the other direction. There exists t such that

∥f (·, t)∥∞,Ω = max

t∈[a,b]

(∥f (·, t)∥

∞,Ω

)by compactness of the interval. Then by compactness of Ω, there is x such that

∥f (·,t)∥∞,Ω = ∥f (x, t)∥ ≤ ∥f∥

∞,Ω×[a,b]

and so the two norms are the same. ■

13.1 Sard’s Lemma and ApproximationFirst are easy assertions about approximation of continuous functions with smooth ones.

The following is the Weierstrass approximation theorem. It is Corollary 5.5.3 presentedearlier.

13.1. SARD’S LEMMA AND APPROXIMATION 303be in OK. Therefore, K = K but this also is a contradiction because if « € 0K then a ¢ Kthanks to the first part that0U =U \U.Note that for an open set U C R?, and hh: U > R?, dist(h (dU) ,y) > dist (h (U) ,y)because U D OU.The following lemma will be nice to keep in mind.Lemma 13.0.4 f €C(Q x [a,b];R”) if and only ift— f(-,t) €C([a,b];C (Q;R?’))Also|Fllaaian) = max (IF Olea)Proof: =By uniform continuity, if € > 0 there is 6 > 0 such that if |t — s| < 6, then forall x € Q, || f (x,t) — f (a,)|| < 4. It follows thatEIFC -F( slo S53 <€<Say (&p,t») —> (x,t) . Does it follow that f (an,t) > f (at)?IF (@nstn) — f (@n,t) || + IF (@n.t) — f (@,0)||lf (@nstn) — Ff (ast)|| <SWF om) -F COM. +NF (@n.t) — F (@,0)|both terms converge to 0, the first because f is continuous into C (Q;R? ) and the secondbecause x — f (x,f) is continuous.The claim about the norms is next. Let (a,t) be such that I| Flo. 2{a,) <||f (v,1)|| +e.ThenIF,2x(a,b] < IF (w,t)|| +e < mar (lif (lle) Leand so ||f Iloo. D(a.) < Max; efa,p] Max (I f (-s1)llna) because € is arbitrary. However, thesame argument works in the other direction. There exists ¢ such thatIF (Dla = max ([IF Dla)by compactness of the interval. Then by compactness of Q, there is x such thatIF Ct) lon = INF (@ 0 SF lle (a.6)and so the two norms are the same.13.1 Sard’s Lemma and ApproximationFirst are easy assertions about approximation of continuous functions with smooth ones.The following is the Weierstrass approximation theorem. It is Corollary 5.5.3 presentedearlier.