For x > 0, we set
Γ(x) = Z ∞
0
tx−1e−tdt.
We can show that Γ(x) is convergent. Hence Γ defines a function from (0, ∞) to R. It is not difficult for us to show that
(1) Γ(x + 1) = xΓ(x) for any x > 0;
(2) Γ(1) = 1.
By induction, Γ(n + 1) = n! for any n ∈ N. In other words, (1.1)
Z ∞ 0
tne−tdt = n!, n ∈ N.
We can compute the integral (1.1) via other method. For each s > 0, we define F (s) =
Z ∞ 0
e−stdt = 1 s. Let us assuming that the following formula holds
F0(s) = Z ∞
0
∂
∂s e−st dt.
By F0(s) = −1/s2, we find
Z ∞ 0
te−stdt = 1 s2. Inductively, we obtain that
Z ∞ 0
tne−stdt = n!
sn+1
for any natural number n ∈ N. Plugging s = 1, we obtain (1.1). In this note, we are going to learn techniques of evaluating definite integrals or improper integrals via the method of differentiation. Let us take a look at other examples.
Example 1.1. Evaluate the integral Z ∞
0
e−txsin x
x dx for t > 0.
We know that | sin x/x| ≤ 1 for any x > 0. Then 0 ≤ e−tx
sin x x
≤ e−tx
for any x > 0 if t > 0. Since Z ∞
0
e−txdx is convergent, by comparison test, the above integral is absolutely convergent if t > 0. Therefore the integral defines a function from (0, ∞) to R.
We define this function by F (t). Let us assume that F is differentiable. This fact can be verified and is left to the reader. Let us assume further that the following formula holds:
F0(t) = Z ∞
0
∂
∂t
e−txsin x x
dx = − Z ∞
0
e−txsin xdx.
By elementary calculus, Z
eaxsin bxdx = a sin bx − b cos bx a2+ b2 + C.
1
Hence we obtain that
Z ∞ 0
e−txsin xdx = 1 1 + t2.
This implies that F0(t) = −1/(1 + t2) and therefore F (t) = C − tan−1t for t > 0. On the other hand,
|F (t)| ≤ Z ∞
0
e−tx
sin x x
dx ≤ Z ∞
0
e−txdx = 1 t. We find that lim
t→∞F (t) = 0. Therefore
t→∞lim F (t) = lim
t→∞ C − tan−1t = C −π 2. Thus C = π/2. Furthermore, F (0) = π/2 which gives
Z ∞ 0
sin x
x dx = π 2.
Example 1.2. (The Gaussian integral and the its moments) For any nonnegative integer n, evaluate the integral cn=
Z ∞
−∞
xne−12x2dx.
We can prove that the integrals converges absolutely via the limit comparison test. The convergence of the integrals are left to the readers. Let us define
F (t) = Z ∞
−∞
etxe−12x2dx for any t. By taking the derivatives of F, we find
F(k)(t) = Z ∞
−∞
xketxe−12x2dx.
Then cn= F(n)(0) for any n ≥ 0. When n = 2k + 1 is odd, the function x 7→ x2k+1e−12x2 is an odd function on R. The integral
Z ∞
−∞
x2k+1e−12x2dx is zero, i.e. c2k+1= 0. Now we only need to compute c2k for any k ≥ 0. By completing the square we know
tx −x2 2 = −1
2(x + t) + t2 2. Therefore
F (t) = Z ∞
−∞
e−12(x+t)+t22 dx = et22 Z ∞
−∞
e−12(x+t)2dx.
Since the integral is invariant under the change of variable change of variable u = x + t, F (t) = e12t2c0.
Using the Taylor expansion of F (t), we know that F (t) =
∞
X
n=0
c0
2nn!t2n, t ∈ R.
By uniqueness of the Taylor expansion, F2n(0)
(2n)! = c0
2nn!, n ≥ 0.
We find that
c2n= Z ∞
−∞
x2ne−12x2dx = (2n)!
2nn!c0. Now we only need to compute c0. To do this, we set
G(t) = Z ∞
0
e−12t2(1+x2) 1 + x2 dx.
Differentiating G with respect to t, one has G0(t) =
Z ∞
−∞
e−12t2(1+x2)· (−t)dx.
Making a change of variable u = xt, we find Z ∞
0
e−12t2(1+x2)· (−t)dx = − Z ∞
0
e−12t2−12u2du = −e−12t2 Z ∞
−∞
e−12u2du.
Hence G0(t) = −e−12t2c0/2. By fundamental Theorem of calculus, G(b) − G(0) = −c0
2 Z b
0
e−12t2dt.
On the other hand, lim
b→∞G(b) = 0 and G(0) = π/2. We find that
−π
2 = −c0
2 Z ∞
0
e−12t2dt = −c20 4. This implies that c0 =√
2π.
Let us study when these calculation make senses.
Lemma 1.1. Let R = [a, b] × [c, d] be a compact interval in R2 and h : R → R be a continuous function. Define H : [a, b] → R by
H(x) = Z d
c
h(x, y)dy.
Then H : [a, b] → R is uniformly continuous; hence
x→tlimH(x) = H(t) ⇐⇒ lim
x→t
Z d c
h(x, y)dy = Z d
c
x→tlimh(x, y) dy.
Proof. Since R is compact and h : R → R is continuous, h is uniformly continuous. For any
> 0, we can find δ > 0 so that
|h(x, y) − h(x0, y0)| < d − c
whenever k(x, y) − (x0, y0)k < δ. If |x − x0| < δ, then for any y ∈ [c, d], k(x, y) − (x0, y)k =
|x − x0| < δ. In this case, |h(x, y) − h(x0, y)| < /(d − c). If |x − x0| < δ,
|H(x) − H(x0)| ≤ Z d
c
|h(x, y) − h(x0, y)|dy <
Z d
c
d − cdy = .
We complete the proof of our assertion.
Theorem 1.1. Let U be an open subset of R2 and f : U → R be a function such that fx exist and continuous on U. For any compact subinterval [a, b] × [c, d] of U, we define F : [a, b] → R by
F (x) = Z d
c
f (x, y)dy.
Then F ∈ C1[a, b] such that F0(x) = Z d
c
fx(x, y)dy.
Proof. Let x be an interior point of [a, b]. Choose δ > 0 so that (x − δ, x + δ) is contained in [a, b]. For 0 < |h| < δ, we compute the difference quotient
F (x + h) − F (x)
h =
Z d c
f (x + h, y) − f (x, y)
h dy.
By the mean value theorem, there exists θ ∈ [0, 1] so that f (x + h, y) − f (x, y) = fx(x + θh, y)h.
When 0 < |h| < δ,
F (x + h) − F (x)
h =
Z d c
fx(x + θh, y)dy.
Since fx is continuous on U, by Lemma 1.1, we know
h→0lim
F (x + h) − F (x)
h = lim
h→0
Z d c
fx(x + θh, y)dy = Z d
c
h→0limfx(x + θh, y)dy = Z d
c
fx(x, y)dy.
Therefore F0(x) exists and
F0(x) = Z d
c
fx(x, y)dy.
We leave to the reader to verify that the left derivative of F at b and right derivative of F at a both exist. By Lemma 1.1 again, F0(x) is continuous on [a, b]. We complete the proof
of our result.
Corollary 1.1. Let g ∈ C2(U ) where U is an open subset of R2. Then gxy = gyx on U.
Proof. We verify that gxy(x0, y0) = gyx(x0, y0) for any (x0, y0) ∈ U. We choose a compact interval R = [a, b] × [c, d] containing (x0, y0). Let f : U → R be the function f (x, y) = gy(x, y). By the fundamentla Theorem of calculus,
g(x, y) = g(x, d) + Z y
d
f (x, t)dt for any (x, y) ∈ R. By Theorem 1.1, we find that
gx(x, y) = gx(x, d) + Z y
d
fx(x, t)dt, (x, y) ∈ R.
It follows from the fundamental Theorem of calculus again, (differentiating the above equa- tion with respect to y)
gxy(x, y) = fx(x, y) = gyx(x, y)
for any (x, y) ∈ R. We complete the proof of our result.
In calculus, we have learned the following formula d
dx
Z ψ2(x) ψ1(x)
f (t)dt = f (ψ2(x))ψ02(x) − f (ψ1(x))ψ01(x), where f, ψ1, ψ2 are some C1-functions. Now we would like to investigate
d dx
Z ψ2(x) ψ1(x)
f (x, y)dy.
More precisely, let f ∈ C2(U ) and ψ1, ψ2 : [a, b] → R be two C1 functions on [a, b] such that a < ψ1(x) < ψ2(x) < b
for any x ∈ [a, b]. Define F : [a, b] → R by F (x) =
Z ψ2(x) ψ1(x)
f (x, y)dy.
Then we would like to compute F0(x). To do this, we set G(u, v, x) =
Z v u
f (x, y)dy,
where (u, v) ∈ [c, d]2 and x ∈ [a, b]. The function F (x) can be rewritten as F (x) = G(ψ1(x), ψ2(x), x), x ∈ [a, b].
By chain rule,
F0(x) = Gu(ψ1(x), ψ2(x), x)ψ01(x) + Gv(ψ1(x), ψ2(x), x)ψ20(x) + Gx(ψ1(x), ψ2(x), x).
It is not difficult to verify that
Gu(u, v, x) = −f (x, u), Gv(u, v, x) = f (x, v), Gx(u, v, x) = Z v
u
fx(x, y)dy.
We conclude that
F0(x) = f (x, ψ2(x))ψ02(x) − f (x, ψ1(x))ψ01(x) +
Z ψ2(x) ψ1(x)
fx(x, y)dy, i.e.
d dx
Z ψ2(x) ψ1(x)
f (x, y)dy = f (x, ψ2(x))ψ20(x) − f (x, ψ1(x))ψ10(x) +
Z ψ2(x) ψ1(x)
fx(x, y)dy, for any x ∈ [a, b].
Example 1.3. Evaluate Z 1
0
(x − 1)xk
log x dx for k > −1.
We leave it to the reader to verify that (1) lim
k→∞kF (k) = 1 (2) F0(k) = 1
k + 2 − 1 k + 1.
From here, you will be able to determine F (k).
Corollary 1.2. Let f ∈ C([a, b] × [c, d]) and I =
Z b a
Z d c
f (x, y)dy
dx and J = Z d
c
Z b a
f (x, y)dx
dy.
Then I = J.
Proof. Let us define two functions u, v : [a, b] × [c, d] → R by v(x, y) =
Z y c
f (x, s)ds u(x, y) =
Z x a
v(t, y)dt.
Then u(b, d) = I and vy = f (x, y). Differentiating u with respect to y, we find uy(x, y) =
Z x a
vy(t, y)dt = Z x
a
f (t, y)dt.
On the other hand, by the fundamental theorem of calculus, u(x, y) = u(x, c) +
Z y c
uy(x, s)ds = Z y
c
Z x a
f (t, s)dt
ds
This implies that u(b, d) = J. We conclude that I = J.
Now let us assume that I × J is an unbounded closed interval in R2 where I and J are intervals of R such that I or J could be unbounded; in other words, I and J could be one of the following integrals (−∞, ∞) or [a, ∞) or (−∞, b] for some a, b ∈ R. Assume that f is a function defined on an open subset U of R2 containing I × J so that f and fx are continuous on U. We define
F : I → R, F (x) = Z
J
f (x, y)dy.
We would like to investigate the differentiability of F.
Theorem 1.2. Suppose that there exist nonnegative A, B : J → R such that (1) |f (x, y)| ≤ A(y) and |fx(x, y)| ≤ B(y) for any (x, y) ∈ I × J and (2)
Z
J
A(y)dy and Z
J
B(y)dy are both convergent.
Then F is differentiable and
F0(x) = Z
J
fx(x, y)dy.