Differentiability and Power Series
Definition Let S ⊆ C be a subset of C, and let f : S → C be a complex-valued function defined on S. We say that f is complex-differentiable (or differentiable) at an interior point z of S if
h→0lim
f (z + h) − f (z) h
exists in C. In that case, the limit is denoted f0(z).
Remark If f is differentiable at z = x + yi then, by identifying z = x + iy with (x, y), we have f0(z) = lim
h∈C, h→0
f (z + h) − f (z) z + h − z
= lim
h∈R, h→0
f ((x + h) + yi) − f (x + yi) (x + h) + yi − (x + yi)
= lim
h∈R, h→0
f (x + h, y) − f (x, y) h
= fx(z)
= lim
h∈R, h→0
f (x + (y + h)i) − f (x + yi) x + (y + h)i − (x + yi)
= lim
h∈R, h→0
f (x, y + h) − f (x, y) ih
= −ify(z).
Proposition Let U ⊂ C and let f, g : U → C be differentiable at z ∈ U. Then (a) f + g is differentiable at z with (f + g)0(z) = f0(z) + g0(z).
(b) f g is differentiable at z with (f g)0(z) = f0(z)g(z) + f (z)g0(z).
(c) f
g is differentiable at z with f g
0
(z) = f0(z)g(z) − f (z)g0(z)
g2(z) provided that g(z) 6= 0.
Definition Let S ⊆ C be a subset of C, and let f : S → C be a complex-valued function defined on S. We say thatf is analytic (or holomorphic) at zif f is differentiable in an open neighborhood of z. Hence, we say that f is analytic on a set S ⊆ U if f is differentiable at all points of some open set containing S. If additionally S = C, the analytic function f : C → C is called an entire function.
Examples
(a) Let P (z) = a0+a1z+· · ·+anznbe a polynomial of degree n ≥ 0. Show that P is differentiable at all points z and P0(z) = a1+ 2a2z + · · · + nanzn−1.
(b) Show that the function f (z) = ¯z is not complex-differentiable.
Proposition Let u and v be real-valued functions and let f (x, y) = u(x, y) + iv(x, y) :=
(u(x, y), v(x, y)).
(a) If f = u + vi is differentiable at z = x + yi, then the Cauchy-Riemann equations hold at z fy(z) = ifx(z)
⇐⇒ uy + ivy = i(ux+ ivx) = −vx+ iux
⇐⇒ ux = vy; uy = −vx. (b) If f is differentiable at z = x + iy = (x, y), then
|f0(z)| = |fx(z)| = |fy(z)|,
=⇒ |f0(z)| =q
u2x(x, y) + u2y(x, y) =q
vx2(x, y) + vy2(x, y), and det Df (x, y)=
ux uy vx vy
=
ux uy
−uy ux
= u2x(x, y) + u2y(x, y)= |f0(z)|2, by the Cauchy-Riemann equations.
(c) Consider the Jacobian of f : R2 → R2
Jf :=ux uy vx vy
and the rotation matrices
cos θ − sin θ sin θ cos θ
.
Since multiplication by i is equivalent to multiplication by the rotation matrix (by θ = π/2) I =0 −1
1 0
, and since
JfI =uy −ux vy −vx
and IJf =−vx −vy ux uy
,
theJacobian matrix commutes with the rotation by π/2 is equivalent to the Cauchy Riemann equations, i.e.
JfI = IJf ⇐⇒ ux = vy; uy = −vx.
Proposition A polynomial P (x, y) is analytic on C if and only if Py = iPx on C.
Proof The necessity of the condition follows from the preceding Proposition part (a).
To show that it is also sufficient, note that if
Py = iPx,
the condition must be met separately by the terms of any fixed degree. Suppose that P has n-th degree terms of the form
Q(x, y) = C0xn+ C1xn−1y + C2xn−2y2+ · · · + Cnyn. Since Qy = iQx,
C1xn−1+ 2C2xn−2y + · · · + nCnyn−1
= i[nC0xn−1+ (n − 1)C1xn−2y + · · · + Cn−1yn−1].
Comparing coefficients,
C1 = i n C0 = in 1
C0 C2 = i(n − 1)
2 C1 = i2 n(n − 1)
2 C0 = i2n 2
C0, and in general
Ck = ikn k
C0 so that
Q(x, y) =
n
X
k=0
Ckxn−kyk= C0
n
X
k=0
n k
xn−k(iy)k= C0(x + yi)n= C0zn which is analytic on C. Thus P is analytic on C.
Examples
(a) A non-constant analytic polynomial cannot be real-valued, for then both Px and Py would be real and the Cauchy-Riemann equations would not be satisfied.
(b) Using the Cauchy-Riemann equations, one can verify that x2− y2+ 2ixy is analytic every- where while x2+ y2− 2ixy is not.
(c) Let
f (x, y) =
xy(x + iy)
x2+ y2 if z 6= 0
0 if z = 0
Show that ∂f
∂x(0, 0) = 0 = ∂f
∂y(0, 0) and thus the Cauchy-Riemann equations are satisfied at z = 0, but f is not complex-differentiable. Note that ∂f
∂x and ∂f
∂y are not continuous at (0, 0).
Theorem Let fn : [a, b] → C be a sequence of continuously differentiable functions defined on [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.
Proof For any x ∈ [a, b] and for any m, n ∈ N, there exists a y depending on m, n such that [fm(x) − fn(x)] − [fm(x0) − fn(x0)] = [fm0 (y) − fn0(y)](x − x0)
=⇒ sup
x∈[a,b]
|fm(x) − fn(x)| ≤ |fm(x0) − fn(x0)| + (b − a) sup
y∈[a,b]
|fm0 (y) − fn0(y)|.
Hence fn converges uniformly on [a, b] to a function f and f is continuous on [a, b].
Remark In the proof, we have applied the following Mean Value Theorem to the real and imaginary parts of fm(x) − fn(x).
Mean Value Theorem If f, g are continuous on [a, b] and differentiable on (a, b), then there exists c ∈ (a, b) such that:
[f (b) − f (a)]g0(c) = [g(b) − g(a)]f0(c).
Note if g(x) = x, this gives the Mean Value Theorem.
Proof For x ∈ [a, b], let
h(x) = [f (b) − f (a)]g(x) − [g(b) − g(a)]f (x).
Since h(a) = h(b), there is a c ∈ (a, b) such that h0(c) = 0 and the Mean Value Theorem follows from Rolle’s Theorem.
Theorem (Term-by-Term Differentiation) For each k ∈ N, let fk be a real-valued function on I = [a, b] which has a derivative fk0 on I. Suppose that
(i)
∞
X
k=0
fk converges at x0 ∈ I,
(ii)
∞
X
k=0
fk0 converges uniformly on I,
there exists a real-valued function f on I such that (a)
∞
X
k=0
fk converges uniformly on I to f,
(b) f is differentiable on I and f0 =
∞
X
k=0
fk0 on I.
Proof
(a) Note that for each x ∈ I and for any m, n ∈ N, by the Mean Value Theorem, there exists y lying between x and x0 such that
sm(x) − sn(x) = sm(x0) − sn(x0) + (x − x0)(s0m(y) − s0n(y)).
Since
∞
X
k=0
fk converges at x0 ∈ I and
∞
X
k=0
fk0 converges uniformly on I, i.e.
sn(x0) =
n
X
k=0
fk(x0) converges and s0n=
n
X
k=0
fk0 converges uniformly on I,
sn =
n
X
k=0
fk converges uniformly on I to f =
∞
X
k=0
fk.
(b) Suppose that
∞
X
k=0
fk0 converges uniformly on I to g.For each x 6= c ∈ I and for any m, n ∈ N, by the Mean Value Theorem, there exists y lying between x and c such that
sm(x) − sn(x) = sm(c) − sn(c) + (x − c)(s0m(y) − s0n(y))
=⇒
sm(x) − sm(c)
x − c −sn(x) − sn(c) x − c
≤ ks0m− s0nkI = sup
y∈I
|s0m(y) − s0n(y)|.
For each ε > 0, since s0n converges uniformly on I, there exists a M (ε) such that if m, n ≥ M (ε) then ks0m− s0nkI < ε. Taking the limit with respect to m, we get
f (x) − f (c)
x − c − sn(x) − sn(c) x − c
≤ ε when n ≥ M (ε).
Next, since g(c) = lim
n→∞s0n(c), there exists an N (ε) such that if n ≥ N (ε), then
|s0n(c) − g(c)| < ε when n ≥ N (ε).
Let L = max{M (ε), N (ε)}. Since sL is differentiable at c ∈ I, there exists a δL(ε) > 0 such that if 0 < |x − c| < δL(ε), then
sL(x) − sL(c)
x − c − s0L(c)
< ε.
Therefore, it follows that if 0 < |x − c| < δL(ε), then
f (x) − f (c)
x − c − g(c)
< 3ε.
This shows that f0(c) exists and equals g(c).
Definition A power series in z is an infinite series of the form
∞
X
n=0
ckzk.
To study the convergence of a power series, we recall the notion of thelim sup = limof a positive real-valued sequence. That is,
n→∞liman= lim
n→∞
sup
k≥n
ak
. Since sup
k≥n
ak= sup{ak | k ≥ n} is a non-increasing function of n, the limit always exists or equals
∞.
Also note that if lim
n→∞an = L, then
(i) for each N and for each ε > 0, there exists some k > N such that ak > L − ε.
(ii) for each ε > 0, there is some N such that ak≤ L + ε for all k > N.
(iii) lim
n→∞c an= c L for any nonnegative constant c.
Theorem Suppose lim
k→∞|ck|1/k = L.
1. If L = 0, then
∞
X
k=0
ckzk converges (absolutely) for all z ∈ C.
2. If L = ∞, then
∞
X
k=0
ckzk converges for z = 0 only.
3. If 0 < L < ∞, set R = 1/L. Then
∞
X
k=0
ckzk converges for |z| < R and diverges for |z| > R.
(R is calledthe radius of convergence of the power series.) Proof
1. L = 0 Since lim
k→∞|ck|1/k = 0, lim
k→∞|ck|1/k|z| = 0 for all z ∈ C. Thus for each z ∈ C, there exists N ∈ N such that if k > N, then
|ck|1/k|z| ≤ 1
2 =⇒ |ckzk| ≤ 1 2k =⇒
∞
X
k=0
ckzk converges (absolutely) for all z ∈ C.
2. L = ∞
For any z 6= 0, there exist infinitely many values of k such that
|ck|1/k ≥ 1
|z| =⇒ |ckzk| ≥ 1 =⇒ lim sup
k→∞
|ckzk| 6= 0 =⇒
∞
X
k=0
ckzk diverges.
3. 0 < L < ∞, R = 1/L.
Assume first that |z| < R and set |z| = R(1 − 2δ). Then since lim
k→∞|ck|1/k|z| = |z|
R = 1 − 2δ, there exists N ∈ N such that if k > N, then
|ck|1/k|z| < 1 − δ < 1 =⇒
∞
X
k=0
ckzk is absolutely convergent.
On the order hand, if |z| > R, since lim
k→∞|ck|1/k|z| = |z|
R > 1, so that for infinitely many values of k, ckzk has absolute value greater than 1 and
∞
X
k=0
ckzk diverges.
Examples 1. Since lim
k→∞k1/k = 1,
∞
X
k=1
k zk−1 the power series converges for |z| < 1 and diverges for |z| > 1.
Also the series diverges for |z| = 1 since lim
k→∞|k zk−1| = ∞.
2. The power series
∞
X
k=1
(zk/k2) has radius of convergence equal to 1. In this case, however, the series converges for all points z on the unit circle since
zk k2
= 1
k2 for |z| = 1.
3.
∞
X
k=1
(zk/k) has radius of convergence equal to 1. In this case, the series converges for all points on the unit circle except z = 1. [Hint: Let z = cis θ and analyze the real and imaginary parts of the series separately.]
4.
∞
X
k=0
(zk/k!) converges for all z ∈ C since
k→∞lim 1
(k!)1/k = 0.
5.
∞
X
k=0
[1 + (−1)k]kzk has radius of convergence 12 since lim sup
k→∞
[1 + (−1)k] = 2.
Example Show that
∞
X
k=0
zk= 1
1 − z for |z| < 1.
Theorem Suppose f (z) =
∞
X
k=0
ckzk converges for |z| < R. Then f0(z) exists and equals to
∞
X
k=0
kckzk−1 throughout |z| < R.
Proof For each 0 < δ < R, since
∞
X
k=0
|ck||R − δ|k converges, f (z) =
∞
X
k=0
ckzk converges uniformly on |z| ≤ R − δ.
Since
∞
X
k=1
kckzk−1 has the same radius of convergence as that of
∞
X
k=0
ckzk,
∞
X
k=1
kckzk−1 converges uniformly on |z| ≤ R − δ.
By the Term-by-Term Differentiation Theorem, f0(z) exists and equals to
∞
X
k=0
kckzk−1throughout
|z| ≤ R − δ. Since 0 < δ < R is arbitrary, f0(z) exists and equals to
∞
X
k=0
kckzk−1 throughout
|z| < R.
Example Show that
∞
X
k=0
(k + 1)zk = 1
(1 − z)2 for |z| < 1.
Corollary Inside its radius of convergence, a power series is infinitely differentiable, with the expected derivatives as convergent power series.
Corollary If R > 0, then ck = f(k)(0) k! .
Lemma If f (z) =
∞
X
k=0
ckzk is a convergent power series and f (zn) = 0 for a sequence {zn}∞n=1 with lim
n→∞zn = 0 and zn 6= 0 for all n ∈ N, then ck= 0 for all k.
Proposition If f (z) =
∞
X
k=0
akzk, g(z) =
∞
X
k=0
bkzk and these agree on some set accumulating to 0, then ak= bk for all k and f (z) = g(z).
Proof Consider
∞
X
k=0
(ak − bk)zk and apply the Lemma. Note that if lim
k→∞|ak|1/k ≤ L and
k→∞lim |bk|1/k ≤ L, then lim
k→∞|ak− bk|1/k ≤ lim
k→∞21/kmax{|ak|, |bk|}1/k ≤ L.