Differentiation in Rn
Let U be an open subset of Rn and let f : U → Rm be a map defined on U with values in Rm. 1. f is differentiable at p ∈ U if there exists a linear function L : Rn → Rm such that
x→plim
kf (x) − f (p) − L(x − p)k
kx − pk = 0.
2. If f is differentiable at p, then it is (locally) Lipschitz near p.
3. If the linear function exists at p, then it is unique.
4. If f is differentiable at p, then each of the partial derivatives ∂fi
∂xj(p), 1 ≤ i ≤ m and 1 ≤ j ≤ n, exists and
∂fi
∂xj(p) = the i-th compnent of L(ej) ∀ 1 ≤ i ≤ m, 1 ≤ j ≤ n.
5. If f is differentiable at p and if u is any element of Rn, then the partial derivative Duf (p) of f at p with respect to u exists. Moreover,
Duf (p) = Df (p)(u) = Df (p)(
n
X
i=1
uiei) =
n
X
i=1
uiDf (p)(ei)
and it is called the directional derivative of f at p in the direction of u provided that u is a unit vector of Rn.
Properties of the Derivative
Let U be an open subset of Rn, let f, g : U → Rm be maps defined on U with values in Rm, and let ϕ : U → R a function defined on U.
1. If f and g are differentiable at p ∈ U, and if α, β ∈ R, then h = αf + βg is differentiable at p and
Dh(p) = αDf (p) + βDg(p).
2. If ϕ and f are differentiable at p ∈ U, then k = ϕf : E → Rm is differentiable at p and Dk(p)(u) =Dϕ(p)(u)f (p) + ϕ(p)Df (p)(u) for u ∈ Rn.
3. If f and g are differentiable at p ∈ U, then h = hf, gi =
m
X
i=1
figi : E → R is differentiable at p and
Dh(p) = g(p) · Df (p) + f (p) · Dg(p) =
m
X
i=1
Dfi(p)gi(p) +
m
X
i=1
fi(p)Dgi(p).
Chain Rule Let f have domain E ⊆ Rn and range in Rm, and let g have domain B ⊆ Rm and range in R`. Suppose that f is differentiable at p and that g is differentiable at q = f (p). Then the composition h = g ◦ f is differentiable at p and
Dh(p) = Dg(q) ◦ Df (p).
Alternatively, we write
Mean Value Theorems
1. Let f be defined on an open subset U of Rn and have values in R. Suppose that the set U contains the points a, b and the line segment S joining them and that f is differentiable at every point of this segment. Then there exists a point c on S such that
f (b) − f (a) = Df (c)(b − a).
2. Let U ⊆ Rn be an open set and let f : U → Rm. Suppose that the set U contains the points a, b and the line segment S joining them and that f is differentiable at every point of this segment. Then there exists a point c on S such that
kf (b) − f (a)k ≤ kDf (c)(b − a)k.
Higher Derivatives
Let U be an open subset in Rn and let f : U → R be a differentiable function on U.
1. If Df : U → Rn is differentiable, we denote the derivative of Df by D2f : U → Rn×n, and we shall refer the n2 functions (entries) of D2f as the second partial derivatives of f denoted by
Dijf or ∂2f
∂xi∂xj
, i, j = 1, 2, . . . , n.
Similarly, if D2f : U → Rn2 is differentiable, we denote the derivative of D2f by D3f : U → Rn3, and we shall refer the n3 functions (entries) of D3f as the third partial derivatives of f denoted by
Dijkf or ∂3f
∂xi∂xj∂xk, i, j, k = 1, 2, . . . , n.
2. Note that if f is differentiable at a point p ∈ U, then Df (p) is a linear map on Rn to R defined by
Df (p)(z) =
n
X
i=1
Dif (p)zi =
n
X
i=1
∂f
∂xi(p)zi for z = (z1, . . . , zn) ∈ Rn.
If Df is differentiable at a point p ∈ U, then D2f (p) is a linear map on Rn2 to R defined by
D2f (p)(y, z) =
n
X
i, j=1
Dijf (p)yjzi =
n
X
i, j=1
∂2f
∂xi∂xj(p)yjzi for y, z ∈ Rn.
Similarly, if D2f is differentiable at a point p ∈ U, then D3f (p) is a linear map on Rn3 to R defined by
D3f (p)(y, z, w) =
n
X
i, j, k=1
Dijkf (p)ykzjwi =
n
X
i, j, k=1
∂3f
∂xi∂xj∂xk(p)ykzjwi for y, z, w ∈ Rn.
3. Taylor’s Theorem Suppose that f has continuous partial derivatives of order m in a neighborhood of every point of a line segment S joining two point a, b = a + u in U. Then there exists a point c on S such that
f (a + u) = f (a) +
m−1
X
k=1
1
k!Dkf (a)(u)k+ 1
m!Dmf (c)(u)m where Dkf (p)(u)k= Dkf (p)(u, u, . . . , u) for k = 1, . . . , m.
The Class C1(U )
Let U be an open subset in Rn, let f : U → Rm be differentiable on U and let Djfi = ∂fi
∂xj
. 1. A function L : Rn → Rm is said to be linear if
L(ax + by) = aL(x) + bL(y) for all a, b ∈ R and x, y ∈ Rn. Note that if L : Rn→ Rm is linear, then L(0) = 0 ∈ Rm.
2. Let L (Rn, Rm) = {L | L : Rn → Rm is linear}. Define (∗) kLknm= sup
z∈Rn, kzk≤1
kL(z)k for each L ∈L (Rn, Rm).
Since L ∈L (Rn, Rm) is linear,
kLknm = sup
z∈Rn, kzk≤1
kL(z)k = sup
z∈Rn, kzk=1
kL(z)k.
If L ∈L (Rn, Rm) is represented by an m × n matrix, then kLknm= sup
z∈Rn, kzk=1
kL(z)k = the maximum length of column vectors of L and the derivative of L equals itself, i.e. DL = L.
3. f ∈ C1(U ) if the derivative Df (x) exists and is continuous under the norm (∗) for all x ∈ U, i.e.
lim
y∈U, y→xkDf (y) − Df (x)knm= 0 for all x ∈ U.
4. For all x, y ∈ U,
kDf (x) − Df (y)knm ≤ ( m
X
i=1 n
X
j=1
|Djfi(x) − Djfi(y)|2 )1/2
≤√
mkDf (x) − Df (y)knm.
5. f ∈ C1(U ) if and only if Djfi is continuous on U for each 1 ≤ i ≤ m, 1 ≤ j ≤ n.
6. If f ∈ C1(U ), then det Df = det[Djfi] : U → R is continuous on U.
7. Suppose that U contains the points a, b and the line segment S joining them, and let p ∈ U.
Then we have
kf (b) − f (a) − Df (p)(b − a)k ≤ kb − ak sup{kDf (x) − Df (p)k }.
8. If f ∈ C1(U ), p ∈ U and ε > 0, then there exists δ(ε) > 0 such that if kxk− pk ≤ δ(ε), k = 1, 2, then xk ∈ U and
kf (x1) − f (x2) − Df (p)(x1− x2)k ≤ ε kx1− x2k.
9. If L = Df (p) is an injection for some p ∈ U, then there exists r > 0 such that (†) rkuk ≤ kDf (p)(u)k = kL(u)k for all u ∈ Rn.
10. If f ∈ C1(U ) and if L = Df (p) is an injection for some p ∈ U, then there exists a δ > 0 such that
(††)12r kx1− x2k ≤ kf (x1) − f (x2)k for all x1, x2 ∈ ¯Bδ(p) ⊂ U.
This prove that the restriction of f to ¯Bδ(p) is an injection; hence this restriction has an inverse function which we shall denote by g = (fB¯δ(p))−1. Note that g is uniformly continuous on f ( ¯Bδ(p)) by using (††).
11. If f ∈ C1(U ) and if L = Df (p) is a surjection for some p ∈ U, then there exists a bounded linear function M : Rm → Rn such that L ◦ M (y) = y for all y ∈ Rm.
12. If f ∈ C1(U ) and if L = Df (p) is a surjection for some p ∈ U, then there exists numbers d > 0 and α > 0 such that if y ∈ ¯Bα/2d(f (p)) ⊂ Rm, then there exists an x ∈ U ∩ ¯Bα(p) such that f (x) = y.
13. If f ∈ C1(U ) and if L = Df (p) is a surjection for all p ∈ U, then f (U ) is open in Rm. 14. If m = n, f ∈ C1(U ) and if L = Df (p) is a bijection for some p ∈ U, then there exist
open neighborhoods V and W of p and f (p), respectively, such that f : V → W has a differentiable inverse g = f−1 : W → V such that
Df−1(y) = Dg(y) = [Df (g(y))]−1 = [Df (f−1(y))]−1 for y ∈ W.
15. Let W be an open subset of Rn+m, f : W → Rm ∈ C1(W ) and let (a, b) ∈ W be a point at which f (a, b) = 0. Suppose that a ∈ Rn, b ∈ Rm and the m × m matrix
Dyf |(a,b)= ∂fi
∂yj(a, b)
1≤i,j≤m
is invertible.
Then there exist open neighborhoods A ⊂ Rn and B ⊂ Rm of a and b, respectively, and a unique g : A → Rm ∈ C1(A) such that
b = g(a) and F (x, g(x)) = 0 for all x ∈ A and
{(x, y) ∈ A × B ⊂ W | f (x, y) = 0}
= {(x, g(x)) | x ∈ A}
= the graph of g over A.
Extremum problems
1. Let U ⊆ Rn be open and let f : U → R have continuous second partial derivatives on U. If p ∈ U is a point of relative minimum [respectively, maximum] of f, then
D2f (p)(w)2 = wt ∂fi
∂xj
(p)
1≤i,j≤n
w =
n
X
i, j=1
Dijf (p)wiwj ≥ 0 for all w ∈ Rn
[respectively, D2f (p)(w)2 =
n
X
i, j=1
Dijf (p)wiwj ≤ 0 for all w ∈ Rn].
2. Lagrange’s Theorem Let U ⊆ Rn be open and suppose that f and g : U → R are real- valued functions in C1(U ). Suppose p ∈ U is such that g(p) = 0 and that there exists an r > 0 such that
f (x) ≤ f (p) or f (x) ≥ f (p) for all x ∈ Br(p) ⊂ U satisfying g(x) = 0.
Then there exist real numbers µ, λ, not both zero, such that µDf (p) = λDg(p).
Moreover, if Dg(p) 6= 0, we can take µ = 1.
Riemann Integrability
1. (Riemann Criterion for Integrability) f is Riemann integrable over K, i.e.
Z
K
f = Z
K
f, if and only if for each ε > 0, there exists a partition Pε of K such that
|U (Pε, f ) − L(Pε, f )| < ε.
2. (Cauchy Criterion for Integrability) f is Riemann integrable over K if and only if for each ε > 0, there exists a partition Pε of K such that if P and Q are refinements of Pε and SP(f, K) and SQ(f, K) are any corresponding Riemann sums, then
|SP(f, K) − SQ(f, K)| < ε.
Convergence and Differentiation
1. Let fn: [a, b] → C be a sequence of continuously differentiable functions defined on I = [a, b].
If fn(x0) converges for some x0 ∈ [a, b] and if fn0 : [a, b] → C converges uniformly on [a, b], then fn converges uniformly on [a, b] to a function f.
2. Term-by-Term Differentiation For each n ∈ N, let fn be a real-valued function on K = [a, b] which has a derivative fn0 on K. Suppose that
(i) P fn converges at x0 ∈ K, (ii) P fn0 converges uniformly on K,
there exists a real-valued function f on K such that (a) P fn converges uniformly on K to f,
(b) f is differentiable on K and f0 =X
fn0 on K.
Convergence and Integration
1. Let {fn} be a sequence of integrable functions that converges uniformly on a closed cell K ⊂ Rp to a function f. Then f is integrable and
Z
K
f = lim
n→∞
Z
K
fn.
2. Term-by-Term Integration For each n ∈ N, let fn be a real-valued integrable function on K = [a, b]. Suppose that the seriesP fn converges to f uniformly on K.
Then f is integrable on K and
Z
K
f =
∞
X
j=1
Z
K
fn.
3. Bounded Convergence Theorem Let {fn} be a sequence of integrable functions on a closed cell K ⊂ Rp. Suppose that there exists B > 0 such that kfn(x)k ≤ B for all n ∈ N, x ∈ K. If the function f (x) = lim
n→∞fn(x), x ∈ K, exists and is integrable, then Z
K
f = lim
n→∞
Z
K
fn.
4. Monotone Convergence Theorem Let {fn} be a monotone sequence of integrable func- tions on a closed cell K ⊂ Rp. If the function f (x) = lim fn(x), x ∈ K, exists and is integrable, then
Z
K
f = lim
n→∞
Z
K
fn.