Exercise Set 1 (Updated)

(1) For a, b, c, d fixed but arbitrary elements of R, define f : C → C by

∀ z ∈ C, f (z) ..= (aRe(z) + bIm(z)) + i(cRe(z) + dIm(z)).
Show that f is C-differentiable at 0 iff it is C-differentiable on C iff a = d and b = −c. Can you
find a more natural way to write f ?
(2) Show that, for a ∈ C \ {0} arbitrary, the limit of
h
a
h
as h → 0 does not exist.
(3) Define f : C → C as follows:
∀ z ∈ C, f (z) ..= |z|2 .
Show that f is C-differentiable only at 0.
(4) Define f : C → C as follows:
∀ z ∈ C, f (z) ..= Re(z)Im(z) + iIm(z).
Show that f is C-differentiable at exactly one point of C, and find this point.
(5) (A definition of the exponential function) We want to extend the exponential function
exp, defined on R, to C. Note that the most important feature of exp is its behaviour under
addition of its inputs:
∀ x, y ∈ R, exp(x + y) = exp(x) exp(y);
note also that every complex number z can be written uniquely as x+iy with x, y ∈ R. Therefore,
if we want the additive property above to be satisfied, we see that, in order to extend exp to
C, it is sufficient to define exp(iy) for y ∈ R. Also check that, if we somehow manage to define
exp on C so that it satisfies the additive property above and agrees with exp on R, then we
must have exp(0) = 1. Let us assume that exp(iy) = u(y) + iv(y) with u, v real-valued functions
and try to determine what u, v must be. First conclude, using the previous remark, that one
must have u(0) = 1 and v(0) = 0. Let us also assume, for the sake of simplicity, that u, v are
differentiable (this is a very mild assumption).

Then our tentative definition of exp becomes
exp(x + iy) = exp(x) u(y) + iv(y) . Show that, with this definition, exp is C-differentiable if
and only if u(y) = v ′ (y) and u′ (y) = −v(y). Surely we want exp to be C-differentiable (it will,
in fact, turn out to be infinitely differentiable, but we’ll see that later.) Therefore, the equations
above should suggest something! Show that, if we define exp(iy) ..= cos(y) + i sin(y) for y ∈ R
arbitrary, then exp, defined on C as follows:


∀ z ∈ C, f (z) ..= exp(Re(z)) cos(Im(z)) + i sin(Im(z))
is C-differentiable everywhere. (It is a theorem that if (u, v) is a pair of differentiable functions
on R such that u(0) = 1, v(0) = 0, u′ = −v and v ′ = u then u = cos and v = sin. This shows
that the definition
∀ y ∈ R, exp(iy) = cos(y) + i sin(y)
is not arbitrary but is the only reasonable definition.)
(6) With the complex exponential defined as above, define f : C → C as follows:
∀ z ∈ C, f (z) ..= exp(Im(z)) exp(iRe(z)).
Show that f is C-differentiable nowhere.
1
2

(7) (Challenging — not important for exam purposes) Recall that given an open subset G ⊂ C,
given a map f : G → C, and given a point a ∈ G, f is said to be differentiable at a if there
exists an R-linear map Ta : C → C such that
f (a + h) − f (a) − Ta (h)
lim = 0.
h→0 ∥h∥
(In the above, Ta (h) denotes the linear map Ta acting on h, which is regarded as a vector in R2 ;
we have to put ∥ · ∥ in the denominator because it is meaningless to divide by a vector. Note
how this definition is different from, yet similar to that of C-differentiability: there we were
demanding the existence of a complex number c such that
f (a + h) − f (a) − ch
lim = 0,
h→0 h
where h is regarded as a complex number, ch simply denotes the product of c and h as complex
numbers, and we can divide by h because it is a complex number.) Therefore the derivative
at a point is an R-linear map from C to C; this will help motivate what follows.
(a) Prove that the set of all R-linear maps from C to itself forms a vector space over R. Denote
it by LR (C). What is its dimension?
We will use the maps P and Q introduced in the write-up on R-linear and C-linear maps
uploaded to Moodle. Let e1 , e2 denote the standard basis of R2 , where e1 ..= (1, 0)T and
e2 ..= (0, 1)T . Note that P(e1 ) = 1 and P(e2 ) = i. Let e1,1 , e1,2 , e2,1 , e2,2 denote the R-linear
maps from C to C given, respectively, by:
e1,1 (z) ..= Re(z);
e1,2 (z) ..= iRe(z);
e2,1 (z) ..= Im(z);
e2,2 (z) ..= iIm(z).
(b) Show that {e1,1 , e1,2 , e2,1 , e2,2 } forms an R-basis of LR (C) (note that this answers the ques-
tion above).
(c) Show that LR (C) is also a vector space over C.
(d) Show that {e1,1 , e2,1 } forms a C-basis of LR (C). (Hence the dimension of LR (C) as a vector
space over C is 2.)
(e) Write f1 ..= e1,1 + ie2,1 , f2 ..= e1,1 − ie2,1 (note that these linear combinations make sense
because LR (C) is a C-vector space) and show that {f1 , f2 } also forms a C-basis of LR (C).
In the write-up mentioned previously, we also learned the definition of a C-linear map from
C to itself. Denote the set of all C-linear maps from C to itself by LC (C); obviously, LC (C) ⊂
LR (C). Now define a new subset of LR (C) as follows:
LC (C) ..= {T ∈ LR (C) | ∀ z, w ∈ C, T (wz) = wT (z)};
this is called the set of all C-antilinear maps from C to itself (“antilinear” simply means that,
instead of the complex number itself, its conjugate pops out when you apply the map).
(f) Show that LC (C) and LC (C) are C-vector subspaces of LR (C).
(g) Show that {f1 } is a C-basis of LC (C) and {f2 } is a C-basis of LC (C).
Since {f1 , f2 } forms a C-basis of LR (C), it follows that LR (C) = LC (C) ⊕ LC (C) (this simply
means that every element g ∈ LR (C) can be written uniquely in the form g = λ1 f1 + λ2 f2 with
λ1 , λ2 ∈ C).
Now suppose that X is any set and that H : X → LR (C) is any function. Then, for every
x ∈ X, since H(x) ∈ LR (C) and {e1,1 , e2,1 } forms a C-basis of LR (C), there exists exactly one
 3

element (dx , dˇx ) ∈ C2 such that H(x) = dx e1,1 + dˇx e2,1 . But since {f1 , f2 } also forms a C-basis
of LR (C), there also exists exactly one element (cx , čx ) ∈ C2 such that H(x) = cx f1 + čx f2 .
(h) Show that cx = 2−1 (dx − idˇx ) and čx = 2−1 (dx + idˇx ).
Since x was arbitrary, this means that there exist functions dX,H , ďX,H , cX,H , čX,H : X → C
such that, for every x, H(x) = dX,H (x)e1,1 + ďX,H (x)e2,1 = cX,H (x)f1 + čX,H (x)f2 (one simply
has dX,H (x) = dx , ďX,H (x) = dˇx , cX,H (x) = cx and čX,H (x) = čx for every x ∈ X, where
cx , čx , dx , dˇx are as above). By the exercise above, one has cX,H = 2−1 (dX,H − iďX,H ) and
čX,H = 2−1 (dX,H + iďX,H ).
Now note that, if G ⊂ C is some open set and f : G → C is some continuously differentiable
function (“continuously differentiable” simply means that all the first partial derivatives of f
exist and are continuous; continuously differentiable functions are also called C 1 functions (the
superscript 1 indicates that only the first partial derivatives are required to be continuous),
and we will also use this terminology), then the total derivative Df of f is a map from G to
LR (C) (there is no mystery here! Note that the total derivative at a point z, Df (z), is an
element of LR (C); that is, for every z ∈ G, we have an element Df (z) ∈ LR (C) — this means
exactly that Df is a function (or, synonymously, a map) from G to LR (C)). Therefore, by the
above, there exist functions dG,Df , ďG,Df , cG,Df , čG,Df : G → C such that, for every z ∈ G,
Df (z) = dG,Df (z)e1,1 + ďG,Df (z)e2,1 = cG,Df (z)f1 + čG,Df (z)f2 . Instead of the cumbersome
dG,Df , ďG,Df , cG,Df , čG,Df , we write simply df , ďf , cf , čf (they depend on the given function
f ). An equivalent way of writing the equation above is Df = df e1,1 + ďf e2,1 = cf f1 + čf f2
(both sides of the equation are now functions from G to LR (C)). One has, by our remarks above,
cf = 2−1 (df − iďf ) and čf = 2−1 (df + iďf ).
(i) Show that, for every z ∈ C, f1 (z) = z and f2 (z) = z.
Now e1,1 and e2,1 are sometimes denoted — think of their definitions! — dx and dy respectively.
Also (bear in mind the above exercise!) f1 and f2 are sometimes denoted dz and dz, respectively.
Therefore one has dz = dx + idy and dz = dx − idy. As natural as you like! Note that, with
these new notations, we can write, for G ⊂ C an arbitrary open set and f : G → C an arbitrary
C 1 map, Df = df dx + ďf dy = cf dz + čf dz.
(j) (Challenging, but involving only clever use of language that you already know) Show that,
given G ⊂ C and f : G → C as above, df and ďf are nothing but ∂f ∂f
∂x and ∂y , respectively.
∂f ∂f
Therefore it is reasonable to define ∂z and ∂z to be cf and čf above, respectively.
∂ ∂
So we have obtained the so-called partial differential operators ∂z and ∂z (the terminology
∂ ∂
shouldn’t scare you! Remember that ∂x and ∂y are also nothing but partial differential opera-
tors). But how do we compute with them? The answer is contained in the following exercises.
Let G ⊂ C be an open set, let f, g : G → C be C 1 functions, and let λ ∈ C.
(k) Show that ∂z ∂z ∂z ∂z
∂z = 1, ∂z = 0, ∂z = 0 and ∂z = 1. (z and z in the numerators are simply a
lazy way of writing the functions z 7→ z : C → C and z 7→ z : C → C.)
∂(f +g) ∂f ∂g ∂(f +g) ∂f ∂g
(l) Show that ∂z = ∂z + ∂z and ∂z = ∂z + ∂z .
∂ ∂f ∂ ∂f
(m) Show that ∂z (λf ) = λ ∂z and ∂z (λf ) = λ ∂z .
∂(f g) ∂f ∂(f g)
(n) Show that ∂z = ∂z g + f ∂g
∂z and ∂z = ∂f
∂z g + f ∂g
∂z .
∂ ∂
Note that the above exercises show that ∂z and ∂z satisfy the sum and product rules for
differentiation, as expected. The chain rule is somewhat more subtle, although it shouldn’t
∂
surprise you if your bear in mind the chain rule for the familiar differential operator ∂x .
4

(o) Suppose that G ⊂ C is an open set, that f : G → C is a C 1 map, and that g is a C 1 map
defined on an open set that contains f (G). Then g ◦ f is defined on G. Moreover,
∂(g ◦ f ) ∂g ∂f ∂g ∂f
= (f (z)) + (f (z)) , and
∂z ∂z ∂z ∂z ∂z
∂(g ◦ f ) ∂g ∂f ∂g ∂f
= (f (z)) + (f (z)) .
∂z ∂z ∂z ∂z ∂z
Now note that, given f as above, f : G → C is also a function, given by: ∀ z ∈ G, f (z) ..= f (z).
If f is C 1 , then f is also clearly C 1 .
(p) Show that
∂f ∂f
= .
∂z ∂z
Note that from this it also follows that
∂f ∂f
=
∂z ∂z
(just replace f by f in the preceding formula). Now we are ready to present an elegant charac-
terization of C-differentiability.
(q) Show that if G ⊂ C is an open set and f : G → C is a C 1 map then f is C-differentiable if
and only if
∂f
≡ 0.
∂z
Try to figure out what the equation above is really saying. It should remind you of equations
that you have encountered already several times! Now we are ready to present some immediate
corollaries of (7o).
(r) In the situation of (7o),
(i) Suppose that both f and g are C-differentiable. Then
∂(g ◦ f ) ∂g ∂f ∂(g ◦ f )
= (f (z)) and ≡ 0.
∂z ∂z ∂z ∂z
(ii) Suppose that f is C-differentiable (but g is not known to be C-differentiable). Then
∂(g ◦ f ) ∂g ∂f ∂(g ◦ f ) ∂g ∂f
= (f (z)) and = (f (z)) .
∂z ∂z ∂z ∂z ∂z ∂z
(iii) Suppose that g is C-differentiable (but f is not known to be C-differentiable). Then
∂(g ◦ f ) ∂g ∂f ∂(g ◦ f ) ∂g ∂f
= (f (z)) and = (f (z)) .
∂z ∂z ∂z ∂z ∂z ∂z
Let inv denote the function z 7→ 1/z : C \ {0} → C (inv for “inversion”). We have already seen
in class that it is C-differentiable, and its C-derivative is given by z 7→ −1/z 2 : C \ {0} → C.
(s) Show that, if G ⊂ C is an open set and f : G → C is a C-differentiable function, then ∂f ∂z ,
as defined above, is nothing but the ordinary C-derivative of f as defined in class, i.e.,
∂f
= f ′ (z).
∂z
Applying this to the function inv gives us:
∂ 1
(inv) = − 2 .
∂z z
∂ ∂
Using the above, (7r) and (7n) finally gives us the quotient rule for the operators ∂z and ∂z :
 5

(t) In the situation of (7n), if g is nowhere-vanishing (i.e., g never takes the value 0),
g ∂f ∂g
∂z − f ∂z g ∂f ∂g
∂z − f ∂z
   
∂ f ∂ f
= and = .
∂z g g2 ∂z g g2
Conclusion: Given any C 1 function f : G → C, where G ⊂ C is an open set, in order to
calculate ∂f ∂f
∂z and ∂z , we simply express f as a function of z and z (always possible, since f
can always be expressed as a function of x and y, and x and y, in turn, can be expressed as
functions of z and z, namely x = (z + z)/2 and y = (z − z)/2i) and differentiate mechanically
partially with respect to z and z, respectively, treating them as independent variables. In
this situation, f is C-differentiable at an arbitrary point z0 if and only if ∂f ∂z (z0 ) = 0 and is
C-differentiable on G if and only if ∂f ∂z ≡ 0 on G.
This gives us an enormously powerful method of checking whether a given function is C-differentiable
or not. For example, if we define f (z) ..= z 2 for all z ∈ C, we see instantly, by partial differentiation,
that ∂f ∂f
∂z = 0 and that ∂z = 2z. Since the latter vanishes only at 0, it follows immediately that f is
C-differentiable at 0 and not C-differentiable anywhere else.
To take another example, let us consider the exponential function exp defined for complex inputs as
above. Note that exp, as defined, is C-differentiable and (easy to check) exp′ (z) = exp(z). Now define
f : C → C as follows: f (z) ..= exp(z) ∀ z ∈ C. By the foregoing (use (7(r)iii)), ∂f
∂z = exp)(z). Since this
never vanishes on C, it follows immediately that f is C-differentiable nowhere.