Existence and Uniqueness Theorems John Stalker∗ March 3, 2015 In these notes, Br (ξ) denotes the open ball of ra- By the Fundamental Theorem of the Calculus, dius r about ξ and B r (ξ) is the closed ball of radius Z 1 ∂f r about ξ. This notation will be used in Euclidean f (1, x, y , y ) − f (0, x, y , y ) = (t, x, ya , yb ) dt. a b a b spaces of any dimension, including dimension 1. 0 ∂t Theorem 1 Suppose that ξ ∈ R, η ∈ Rn , r > 0 and By the Chain Rule, s > 0. Suppose ∂f ∂F ∂u u(t, x, ya , yb ) = (x, u(t, ya , yb )) (t, ya , yb ). n ∂t ∂y ∂t F : B r (ξ) × B s (η) → R From the definition of u we see that satisfies ∂F ∂y (x, y) ≤ M ∂u (t, ya , yb ) = (yb − ya ). ∂t for each x ∈ B r (ξ) and y ∈ B s (η), where the ∂F/∂y Since the matrix norm is submultiplicative is a matrix of partial derivatives and the norm is the ∂f usual matrix norm. Then u(t, x, ya , yb ) ≤ M kya − yb k. ∂t kF (x, ya ) − F (x, yb )k ≤ M kya − yb k Integrating this inequality gives for all x ∈ B r (ξ) and ya , yb ∈ B s (η). kF (x, ya ) − F (x, yb )k ≤ M kya − yb k Before beginning the proof, note that the norm is continuous and B r (ξ) × B s (η) is closed and bounded, as required. so if ∂F/∂y is continuous then there must be such From now on we assume that an M . F : B r (ξ) × B s (η) → Rn Proof: Let satisfies u(t, ya , yb ) = tya + (1 − t)yb . kF (x, ya ) − F (x, yb )k ≤ M kya − yb k and f (t, x, ya , yb ) = F (x, u(t, ya , yb )). for all x ∈ B r (ξ) and ya , yb ∈ B s (η) and that Then kF (x, η)k ≤ N F (x, ya ) − F (x, yb ) = f (1, x, ya , yb ) − f (0, x, ya , yb ). ∗ School for all x ∈ B r (ξ). If F is continuous then there is such an N . of Mathematics, TCD, Copyright 2015 1 Theorem 2 Suppose 0 < σ < s. There is a ρ > 0 Now such that if y0 ∈ Bσ (η) and y: Bρ (ξ) → Rn satisfies kz(ξ)k = ky(ξ) − ηk = ky0 − ηk < σ y ′ (x) = F (x, y(x)) and for x ∈ Bρ (ξ) and |x − ξ| < ρ′ , y(ξ) = y0 so then kz(x)k ≤ y(x) ∈ Bs (η) for all x ∈ Bρ (ξ). N N σ+ < s. exp(M ρ′ ) − M M The Bootstrap Lemma from the preceding chapter then allows us to conclude that Proof: It will be shown that this holds with 1 Ms + N ρ = min r, . log M Mσ + N kz(x)k < s for all x ∈ Bρ (ξ). By continuity there is a ρ′ > 0 such that Theorem 3 If ya and yb satisfy the differential equation y ′ (x) = F (x, y(x)) y(x) ∈ Bs (η) for all x ∈ Bρ′ (ξ). Suppose that x ∈ Bρ′ (ξ). The derivative of the the norm of a matrix valued function is bounded by the norm of the derivative. We apply for all x ∈ Bρ (ξ) with initial conditions this to ya (ξ) = za , yb (ξ) = zb z(x) = y(x) − η, obtaining d kz(x)k ≤ kz ′ (x)k = ky ′ (x)k = kF (x, y(x))k. dx then kya (x) − yb (x)k ≤ kza − zb k exp(M |x − ξ|). Proof: If ya and yb are solutions then d kya (x) − yb (x)k ≤ ky ′ (x) − y ′ (x)k a b dx = kF (x, ya (x)) − F (x, yb (x))k ≤ M kya (x) − yb (x)k. By the triangle inequality kF (x, y(x))k ≤ kF (x, y(x)) − F (x, η)k + kF (x, η)k. By hypothesis kF (x, y(x)) − F (x, η)k ≤ M ky(x) − ηk = M kz(x)k The last step is justified because, by the preceding theorem, ya (x), yb (x) ∈ Bs (η) and kF (x, η)k ≤ N. Thus d kz(x)k ≤ M kz(x)k + N. dx for all x ∈ Bρ (ξ). Applying Gronwall again, kya (x) − yb (x)k ≤ kya (ξ) − yb (ξ)k exp(M |x − ξ|). By Gronwall’s inequality N N exp(M |x − ξ|) − . kz(x)k ≤ kz(ξ)k + M M There are two important corollaries. 2 Now Theorem 4 The initial value problem y ′ (x) = F (x, y(x)), kz − ηk < σ y(ξ) = y0 and has at most one solution in the interval Bρ (ξ). kF (t, η)k ≤ N Assume there are two solutions, ya and yb . Then the preceding theorem, with za = zb = y0 gives Proof: so ky1 (x) − y0 (x)k ≤ σ + N |x − x0 |. By generalised induction we show that for all n ≥ 1 and all x ∈ Bρ (ξ) • kya (x) − yb (x)k = 0 and hence ya = yb . Theorem 5 Suppose that for each z ∈ Bσ (x) there is a solution to the initial value problem y ′ (x) = F (x, y(x)), kyn (x) − yn−1 (x)k ≤ gn (x) where y(ξ) = y0 in Bρ (ξ). This solution depends continuously on z, uniformly in x. gn (x) = σM n−1 |x − x0 |n−1 M n−1 N |x − x0 |n + , (n − 1)! n! • Proof: Suppose ǫ > 0. Let δ = ǫ exp(−M ρ). Then if kza − zb k < δ kyn (x) − ηk ≤ n X gm (x). m=1 and ya and yb are the solutions with those initial conditions then, by the theorem, In case n = 1 both of these are just the statement kya (x) − yb (x)k ≤ δ exp(M |x − ξ|) < ǫ ky1 (x) − y0 (x)k ≤ σ + N |x − x0 |. for all x ∈ Bρ (ξ). This is the definition of continuity. It is uniform because the δ we found depends only on ǫ, not on x. To make the preceding theorem non-trivial we still need to show that solutions exist. proved above. For the induction step we assume the two inequalities above and the corresponding inequalities with n replaced by any smaller integer. Using these, we prove the corresponding inequalities with n replaced by n + 1. First of all, gm (x) is non-negative, so Theorem 6 For any z ∈ Bσ (η) there is a contin ∞ nously differentiable y: Bρ (ξ) such that X X g (x) ≤ gm (x) m y ′ (x) = F (x, y(x)), y(ξ) = z. m=1 m=1 N N exp(M |x − x0 |) − = σ + Proof: We define, inductively, M M Z x < s. F (t, yn (t)) dt. y0 (x) = η, yn+1 (x) = z + x0 Thus Z x yn (x) ∈ Bs (η) (y1′ (t) − y0′ (t)) dt y1 (x) − y0 (x) = y1 (ξ) − y0 (ξ) + for all x ∈ Bρ (ξ). Since we also have the correspondx0 Z x ing inequality with n replaced by n − 1. From this it F (t, y0 (t)) dt. =z−η+ follows that Zxx0 F (t, η) dt. =z−η+ kF (x, yn (x)) − F (x, yn−1 (x)k ≤ M kyn (x) − yn−1 (x) x0 3 for all x ∈ Bρ (ξ). By the other part of the induction hypothesis the right hand side is bounded by M gn (x). Since Z x (F (x, yn (x)) − F (x, yn−1 (x)) dx yn+1 (x)−yn (x) = x0 we see that kyn+1 (x) − yn (x)k ≤ Z x M gn (t) dt = gn+1 (x). x0 That’s half of what we needed to prove for the induction step. The other half follows because, by the triangle inequality, kyn+1 (x) − ηk = kyn+1 (x) − yn (x)k + kyn (x) − ηk . For the first norm on the right we use the inequality just obtained and for the second we use the induction hypothesis. The result is yn+1 (x)−yn (x) ≤ gn+1 (x)+ n X m=1 gm (x) = n+1 X gm (x). m=1 This is the other half of what we needed to prove, thus completing the induction. 4