0.1. Solution of Problem 1. Step 1 Let x ∈ U be a point where the maximum of |Du| is achieved. If |Du|(x) = 0 there is nothing to prove. Otherwise set ν = Du(x)/|Du|(x) and consider the function v = ν · Du. v is harmonic in U and C 1 on U . Therefore, by the maximum principle, there is a y ∈ ∂U such that |Du|(x) = v(x) ≤ v(y) = ν · Du(y) ≤ |Du|(y) . We thus conclude that max |Du| ≤ max |Du| . ∂U U (0.1) Step 2 Without loss of generality we assume that |Du| achieves its maximum in 0 ∈ ∂U and that U ⊂ {x1 ≥ 0}. Let ν := Du/|Du|(0) and e1 := (1, 0, . . . , 0). If ν · e1 = 0, then |Du|(0) = |Dτ u|(0) = |Dτ g|(0) ≤ |Dg(0)|, where Dτ f denotes the tangential derivative of f on ∂U . We therefore assume ν · e1 6= 0. Moreover, since −u and −g satisfy the assumptions of the exercise, we can also assume ν · e1 > 0. Consider next the points xε := εν. For ε small enough we have xε ∈ U and hence |Du(0)| = lim ε↓0 u(xε ) − u(0) . ε (0.2) Step 3 Let M := maxU |∆g| and d := diam U and consider the function w(x) := g(x) + M dM x1 − x21 2 2 Observe that ∆w = ∆g − M ≤ 0. Thus ∆w ≤ 0. Moreover, if x ∈ U , then 0 ≤ x1 ≤ |x| ≤ diam (U ) = d. We therefore conclude that M x1 (d − x1 ) ≥ g(x) 2 Therefore, w ≥ g = u on ∂U and by the maximum principle w(x) = g(x) + u(x) ≤ w(x) for all x ∈ U . (0.3) We can therefore use (0.2) and (0.3) to estimate w(xε ) − u(0) g(εν)) + dM εν · e1 /2 − M ε2 (ν · e1 )2 /2 − g(0) = lim ε↓0 ε↓0 ε ε g(εν) − g(0) dM dM = lim + ν · e1 = Dg(0) · ν + ν · e1 ε↓0 ε 2 2 dM ≤ |Dg|(0) + . (0.4) 2 Recalling that |Du|(0) = max∂U |Du|, d = diam (U ) and M = maxU |∆g|, (0.1) and (0.4) give the desired inequality. |Du|(0) ≤ lim 1 2 0.2. Solution of Problem 2. First of all, by the density of Cc∞ (Rn ) in W N,p (Rn ), it suffices to show the inequality under the additional assumption u ∈ Cc∞ (Rn ). Step 1 We prove the inequality in the special case n = 1, N = 2 and j = 1. Indeed we will show that ku0 kLp ≤ 2kukLp + ku00 kLp for every u ∈ Cc∞ (R). Having fixed u ∈ Cc∞ (R) and x ∈ R we have ˆ ˆ x+1 ˆ x+1 0 0 0 0 u (x) − |u (s) − u (x)| ds ≤ u (s) ds ≤ x x ˆ x+1 x+1 s |u00 (τ )| dτ ds x ˆ x+1 |u00 (τ )| dτ . |u00 (τ )| dτ ds = x (0.6) x x ˆ x+1 1 |u00 (x+τ )| dτ . (0.7) 00 |u (τ )| dτ ≤ |u(x+1)|+|u(x)|+ 0 x x ˆ x ˆ ≤ We therefore conclude ˆ x1 ˆ 0 0 |u (x)| ≤ u (s)ds + x+1 (0.5) Integrating this last inequality we obtain p 1/p ˆ ˆ 1 00 0 dx |u (x + τ )| dτ ku kLp ≤ 2kukLp + R ˆ ˆ 0 1/p 1 00 p |u (x + τ )| dτ dx ≤ 2kukLp + R ≤ 2kukLp + ku00 kLp . (0.8) 0 Step 2 Assume n ∈ N and u ∈ Cc∞ (Rn ). From the previous step we conclude ˆ ˆ p k∂x1 ukLp dx = |∂x1 u(x1 , x0 )|p dx1 dx0 n−1 R R ˆ ˆ ˆ (0.5) 2 p 0 p 0 p p 2 |u(x1 , x )| dx1 + |∂x1 u(x1 , x )| dx1 dx0 ≤ Rn−1 ≤ C̄kukpLp R R + C̄k∂x21 ukpLp . (0.9) Summing (0.9) over i and taking the p-th rooth we achieve kDukLp ≤ CkukLp + CkD2 ukLp . (0.10) Next, fix any ε ∈]0, +∞[ and define v(x) := ε−n+1 u(x/ε). Note that ˆ x p dx 1/p kDukLp = Du n ε ε (0.10) = kDvkLp ≤ C̄kvkLp + C̄kD2 vkLp = C̄ε1/p kukLp + C̄ kD2 ukLp . ε1/p (0.11) 3 Moreover, from (0.11), for any fixed N , n and p we conclude the existence of a constant C̃ = C̃(N, n, p) such that kDN ukLp ≤ C̃δkDN −1 ukLp + C̃ kDN +1 ukLp δ ∀u ∈ Cc∞ (Rn ), ∀δ ∈]0, +∞[ . (0.12) Step 3 Having fixed n and p we now prove the desired inequality by induction on N . First of all observe that (0.10) provides the starting point of the induction, i.e. the case N = 2 of the desired inequality. Next, we fix N > 1 and assume the existence of a constant CN such that kDj ukLp ≤ CN kukLp + CN kDN ukLp ∀j ∈ {1, . . . , N − 1}, ∀u ∈ Cc∞ (Rn ) . (0.13) We want to show the existence of a constant CN +1 such that kDj ukLp ≤ CN +1 kukLp +CN +1 kDN +1 ukLp ∀j ∈ {1, . . . , N }, ∀u ∈ Cc∞ (Rn ) . (0.14) Observe that (0.14) follows from (0.13) and kDN ukLp ≤ CkukLp + CkDN +1 ukLp ∀u ∈ Cc∞ (Rn ) . (0.15) To show (0.15) we fix δ > 0 (whose choice will be specified later) and observe kDN ukLp (0.12) ≤ (0.13) ≤ C̃δkDN −1 ukLp + C̃ kDN +1 ukLp δ C̃CN δkukLp + C̃CN δkDN ukLp + C̃ kDN +1 ukLp . δ At this point choose δ = (2C̃CN )−1 to infer 1 1 kDN ukLp ≤ kukLp + kDN ukLp + CkDN +1 ukLp , 2 2 from which (0.15) follows easily. (0.16)