1. Derivatives Let (X, k · kX) and (Y, k · kY) be normed vector spaces.
Definition 1.1. A linear map T : X → Y is said to be bounded if there exists M > 0 such that
kT (x)kY ≤ M kxkX for any x ∈ X.
Proposition 1.1. Let T : X → Y be a linear map. The following three statements are equivalent.
(1) T is bounded.
(2) T is continuous.
(3) T is continuous at 0.
Proof. The proof is left to the reader as an exercise.
Definition 1.2. Let f : U → Y be a function defined on a nonempty open subset U of X.
We say that f is differentiable at x0 ∈ U if there exists a bounded linear map T : X → Y such that
khklimX→0
kf (x0+ h) − f (x0) − T (h)kY
khkX = 0.
In this case, T is called a derivative of f at x0.
Proposition 1.2. If f is differentiable at x0, its derivative is unique.
The proof follows from the following fact:
Lemma 1.1. Let S : X → Y be a linear map such that S(x) = 0 for any x ∈ X with kxkX = 1. Then S = 0 is the zero linear map.
Proof. Since S is linear, S(0X) = 0Y. Here 0X and 0Y are the zero vectors of X and of Y respectively. For x 6= 0X, set w = x/kxkX. Then w is a unit vector and x = kxkXw. Since w is a unit vector, S(w) = 0Y. We see that
S(x) = S(kxkXw) = kxkXS(w) = kxkX · 0Y = 0Y.
This shows that S = 0 is the zero linear map.
Let us go back to prove Proposition 1.2. Since f is differentiable at x0, there exist bounded linear maps T1, T2: X → Y such that for any > 0, there exists δ > 0 with
kf (x0+ h) − f (x0) − T1(h)kY ≤ 2khkX, kf (x0+ h) − f (x0) − T2(h)kY ≤
2khkX,
whenever 0 < khkX < δ. By triangle inequality for norms, we obtain that whenever 0 <
khkX < δ,
kT1(h) − T2(h)k ≤ kf (x0+ h) − f (x0) − T1(h)kY + kf (x0+ h) − f (x0) − T2(h)kY
≤
2khkX+ ≤
2khkX = khkX.
Let S = T1 − T2 and xh = h/khkX for 0 < khkX < δ. Then xh is a unit vector and kS(xh)kY < for any > 0. This implies that kS(x)k = 0 for any x ∈ X with kxkX = 1.
Lemma 1.1 implies that S = 0 and hence T1 = T2.
Remark. Since the derivative of f at x0 is unique if it is differentiable at x0, its derivative is denoted by Df (x0).
1
2
Example 1.1. Let f : R2 → R2, f (x, y) = (x2− y2, 2xy). Show that f is differentiable at p(2, 1) and find Df (2, 1).
Proof. Let h = (h1, h2) ∈ R2. Then
f (p + h) − f (p) = (4h1− 2h2+ h21− h22, 2h1+ 4h2+ 2h1h2).
Take T (h) = (4h1− 2h2, 2h1+ 4h2). Then T is a linear map and f (p + h) − f (p) − T (h) = (h21− h22, 2h1h2).
We find that
kf (p + h) − f (p) − T (h)k = q
(h21− h22)2+ (2h1h2)2 = khk2 and that
khk→0lim
kf (p + h) − f (p) − T (h)k
khk = lim
khk→0khk = 0.
Since any linear map T on R2 is bounded, the above equation implies that f is differentiable at p and Df (2, 1) = T.
Proposition 1.3. If f is differentiable at x0, then it is continuous at x0.
Proof. Since f is differentiable at x0, choose δ1> 0 such that kf (x0+ h) − f (x0) − T (h)kY < khkX
whenever 0 < khkX < δ1. Here T = Df (x0). Since T is bounded, there exists M > 0 such that kT (x)kY ≤ M kxkX. For any x ∈ B(x0, δ1), we have kx − x0k < δ1. Thus
kf (x) − f (x0)kY ≤ kf (x) − f (x0) − T (x − x0)kY + kT (x − x0)kY
≤ kx − x0kX + M kx − x0kx
= (M + 1)kx − x0kX. Let C = M + 1. We have
kf (x) − f (x0)kY ≤ Ckx − x0kX
whenever x ∈ B(x0, δ). This implies that f is continuous at x0. Example 1.2. Let X = C([0, 1]) be the space of all real valued continuous functions on [0, 1] equipped with the sup norm
kf k∞= sup
x∈[0,1]
|f (x)|
and Y = R be the space of all real numbers with absolute value. Define F : X → Y by F (u) =
Z 1 0
u(x)2dx.
Show that F is differentiable at all u0∈ X.
Proof. We observe that
F (u0+ h) − F (u0) = Z 1
0
(2u0(x)h(x) + h(x)2)dx for any h ∈ X. Define T : X → R by
T (h) = 2 Z 1
0
u0(x)h(x)dx, h ∈ X.
3
For any h ∈ X,
|F (u0+ h) − F (u0) − T (h)| =
Z 1 0
h(x)2dx
≤ khk2∞ and hence
0 ≤ |F (u0+ h) − F (u0) − T (h)|
khk∞ ≤ khk∞.
By Sandwich principle,
khklim∞→0
|F (u0+ h) − F (u0) − T (h)|
khk∞ = 0.
Hence if we can show that T is a bounded linear map, then F is differentiable at u0 with DF (u0) = T.
We leave it to the reader to verify that T is a linear map. Now let us prove that T is bounded. If u0 = 0, T is the zero linear map and hence T is bounded. Assume that u0 6= 0.
For any h ∈ X, we have
|T (h)| =
Z 1 0
2u0(x)h(x)dx
≤ 2ku0k∞khk∞.
This implies that T is bounded. We complete the proof of our assertion.