Complex Analysis Notes

Math 444 - Fall 2024

Jump to: Syllabus, Week 1 , Week 2, Week 3, Week 4, Week 5, Week 6, Week 7, Week 8, Week 9, Week 10, Week 11, Week 12, Week 13, Week 14, Week 15

Week 1 Notes

Tentative Schedule

Day Section Topic
Mon, Aug 26 1.1 Definitions and algebraic properties
Wed, Aug 28 1.2 From algebra to geometry and back
Fri, Aug 30 1.3 Geometry of complex numbers

Mon, Aug 26

Today we introduced the complex numbers \mathbb{C} which are a field. We defined the real and imaginary parts of a complex number and also the absolute value and argument of a complex number. We did the following examples.

  1. Find the real and imaginary parts of 13+4i\dfrac{1}{3+4i} by rationalizing the denominator.

  2. Factor the polynomial z2+3iz2z^2 + 3iz - 2.

  3. The polynomial z2+1z^2 + 1 is irreducible over \mathbb{R}, but not over \mathbb{C}. Show this by factoring it over \mathbb{C}.

Later in the course we will use complex analysis to prove one of most important theorems in algebra:

The Fundamental Theorem of Algebra. Every polynomial p(z)p(z) with coefficients in \mathbb{C} factors into a product of linear expressions of the form (az+b)(az + b) where a,ba, b \in \mathbb{C}.

We finished with a discussion of the polar form z=r(cosθ+isinθ)z = r (\cos \theta + i \sin \theta) of a complex number.

Wed, Aug 28

Following Beck et. al., we will define eiθ:=cosθ+isinθ.e^{i \theta} := \cos \theta + i \sin \theta. Later, when we define the complex exponential function, we will revisit this definition. For now, we will do some calculations that suggest the definition is a good one.

  1. Use the angle addition formulas to show that (eiα)(eiβ)=ei(α+β).(e^{i \alpha}) (e^{i \beta}) = e^{i (\alpha + \beta)}.

  2. Simplify 1eiθ\frac{1}{e^{i \theta}}.

  3. Calculate ddθeiθ\dfrac{d}{d \theta} e^{i \theta}. Hint: Use the definition of eiθe^{i \theta} and treat ii like any other constant.

  4. What is (eiθ)n(e^{i \theta})^n?

  5. What is i\sqrt{i}?

  6. Explain why ddθeiθ=ieiθ\dfrac{d}{d \theta} e^{i \theta} = i e^{i \theta} makes sense in the context of the velocity of a point moving counterclockwise along the unit circle.

After introducing eiθe^{i\theta}, we discussed the n-th roots of unity which are the complex numbers zz such that zn=1z^n = 1. They are given by the formula e2πik/n where k{0,1,,n1}.e^{2 \pi i k/ n} \text{ where } k \in \{0, 1, \ldots, n-1\}.

  1. Find the 3rd roots of 8.

  2. Find the 3rd roots of (22+i22)\left(\frac{\sqrt{2}}{2} + i \frac{\sqrt{2}}{2} \right).

We finished with a discussion of square and cube roots of complex numbers and how they relate to roots of unity. We also defined the complex conjugate of z=a+ibz = a+ib to be z=aib\bar{z} = a - ib.

Fri, Aug 30

We reviewed some of the useful formulas involving complex conjugates including the following:

We also introduced and proved the triangle inequality for complex numbers |z+w||z|+|w|.| z + w | \le |z| + |w|.

  1. Use the triangle inequality to prove the reverse triangle inequality |zw||z||w| | z - w | \ge |z| - |w| by introducing a substitution u=zwu = z - w.

  2. If |zw|ϵ|z - w| \le \epsilon, prove that |1z1w|ϵ|z||w|.\left| \frac{1}{z} - \frac{1}{w} \right| \le \frac{\epsilon}{|z| \, |w|}.

A convex combination of z,wz, w \in \mathbb{C} is any point tz+(1t)w,0t1. t z + (1-t) w, ~ 0 \le t \le 1. The set of all convex combinations of zz and ww is a line segment connecting zz to ww, and the formula for the convex combinations is a parametric formula for the line segment. If you allow any tt \in \mathbb{R}, then you get an affine combination of zz and ww. The set of all affine combinations of zz and ww is line passing through zz and ww.


Week 2 Notes

Tentative Schedule

Day Section Topic
Mon, Sep 2 Labor Day, no class
Wed, Sep 4 1.4 Elementary topology of the plane
Fri, Sep 6 1.4 Elementary topology of the plane - con’d

Wed, Sep 4

Let BR(z)={w:|zw|<R}B_R(z) = \{w \in \mathbb{C}: |z-w| < R\} denote the open disk of radius RR around zz. We used BR(z)B_R(z) to define interior and boundary points for a subset of \mathbb{C}. We proved the following trichotomy result

Theorem. If AA \subseteq \mathbb{C} and xx \in \mathbb{C}, then exactly one of the following holds.

  1. xx is an interior point of AA.
  2. xx is a boundary point of AA.
  3. xx is an interior point of AcA^c (the complement of AA).

We also defined open and closed sets, and proved the following theorem.

Theorem. A set AA \subseteq \mathbb{C} is closed if and only if its complement is open.

  1. If a set is both open and closed, then it can’t have any boundary points, since it includes its boundary (because it is closed) but every point is an interior point (because it is open). Are there any sets with no boundary points?

We also defined bounded sets. We finished by talking about paths which are continuous functions γ:[a,b]\gamma: [a,b] \rightarrow \mathbb{C}. The range of a path is a subset of \mathbb{C} called a curve (note that not all books follow this terminology). We defined the derivative of a path, and discussed how to define a smooth path so that it matches our intuition for a curve with no sharp turns.

  1. Why isn’t the path γ(t)=t2+it3\gamma(t) = t^2 + i t^3 for t[1,1]t \in [-1,1] smooth?

A path γ:[a,b]\gamma: [a,b] \rightarrow \mathbb{C} is simple if γ(t1)γ(t2)\gamma(t_1) \ne \gamma(t_2) for all t1t2t_1 \ne t_2, except possibly at the endpoints t1=at_1 = a and t2=bt_2 = b. Intuitively, a path is simple if it cannot cross itself, except possibly at the endpoints. A path is closed if γ(a)=γ(b)\gamma(a) = \gamma(b).

Fri, Sep 6

A set AA \subseteq \mathbb{C} is path connected if for any w,zAw, z \in A, there is a (continuous) path γ:[a,b]A\gamma: [a,b] \rightarrow A such that γ(a)=w\gamma(a) = w and γ(b)=z\gamma(b) = z. The following theorem seems obvious, but it is actually famously tricky to prove (see https://en.wikipedia.org/wiki/Jordan_curve_theorem).

The Jordan Curve Theorem. The complement of the range of a simple closed curve in \mathbb{C} consists of two disjoint open path connected sets, one of which (the inside) is bounded and the other (the outside) is not bounded.

A sequence is a function s:s: \mathbb{N}\rightarrow \mathbb{C}. We use the notation sns_n to mean the same thing as s(n)s(n) for sequences. A sequence sns_n converges to LL \in \mathbb{C} if for every ϵ>0\epsilon > 0, there is an N>0N > 0 such that |snL|<ϵ|s_n - L | < \epsilon for every nNn \ge N. Intuitively, this means that for every open disk around LL, there can only be a finite number of nn such that sns_n is not in the disk. When a sequence converges, the number it converges to is called its limit.

If we don’t know the limit of a sequence, we can still use Cauchy’s criterion to show that it must converge. A Cauchy sequence is a sequence sns_n such that for every ϵ>0\epsilon > 0, there exists N>0N > 0 such that |snsm|<ϵ|s_n - s_m | < \epsilon whenever m,nNm, n \ge N.

Theorem (Cauchy’s Criterion). If sns_n is a Cauchy sequence, then it converges to some limit in \mathbb{C}.

We talked about how the Cauchy criterion applies to sequences of real numbers and the complex numbers, in fact the property that Cauchy sequences converge in a set is known as completeness and it is one of the defining properties of the real numbers.

We used completeness to prove the following theorem about real number sequence:

Theorem (Monontone convergence). If sns_n is a sequence of real numbers that is monotone (non-decreasing) and bounded, then it converges.

We ran out of time, but a related application of completeness is the following result.

Theorem. If AA \subset \mathbb{C} and AcA^c are both nonempty sets, then AA has at least one boundary point.


Week 3 Notes

Tentative Schedule

Day Section Topic
Mon, Sep 9 7.1 Sequences and completeness
Wed, Sep 11 7.2 Series
Fri, Sep 13

Mon, Sep 9

The complex numbers are a metric space because they have a distance function d(z,w)=|zw|d(z,w) = |z-w| that satisfies the following properties for all x,y,zCx, y, z \in C:

  1. Definiteness d(x,x)=0d(x,x) = 0
  2. Positivity d(x,y)>0d(x,y) > 0 if xyx \ne y
  3. Symmetry d(x,y)=d(y,x)d(x,y) = d(y,x)
  4. Triangle Inequality d(x,z)d(x,y)+d(y,z)d(x,z) \le d(x,y) + d(y,z)

It helps when working with absolute values in \mathbb{C} to remember that |zw||z-w| is the distance between zz and ww. We didn’t have time for a proof of the last theorem from Friday, so we worked through a proof using the fact that \mathbb{C} is a complete metric space today. Recall that complete means that all Cauchy sequences converge. We proved the theorem by proving the following claims:

  1. Any line segment that connects a point in AA to a point in AcA^c contains a line segment that is half as long and also connects a point in AA to a point in AcA^c.

  2. The midpoints of the line segments above form a Cauchy sequence.

  3. The limit of the midpoints is a boundary point of AA.

This proof definitely doesn’t work without assuming completeness. For example, if A={x:x<π}A = \{x \in \mathbb{Q} : x < \pi \}, then AA has no boundary points in \mathbb{Q}. After discussing that example, we talked about infinite series.

A series of complex numbers k=0ak\sum_{k = 0}^\infty a_k converges if its sequence of partial sums Sn=k=0nakS_n = \sum_{k = 0}^n a_k converges. It helps to know some example infinite series, so we talked about these three:

Then we reviewed geometric series which are series of the form k=0ark\sum_{k = 0}^\infty a r^k where aa is the first term and rr is the common ratio. We proved:

Theorem. A geometric series k=0ark\sum_{k = 0}^\infty a r^k converges if and only if |r|<1|r| < 1, and in that case k=0ark=a1r.\sum_{k = 0}^\infty a r^k = \frac{a}{1-r}.

Wed, Sep 11

Class was canceled since I was out with COVID.

Fri, Sep 13


Week 4 Notes

Tentative Schedule

Day Section Topic
Mon, Sep 16 2.1 Limits and continuity
Wed, Sep 18 2.2 Differentiability and holomorphicity
Fri, Sep 20 2.2 The Cauchy-Riemann equations

Mon, Sep 16

Today we reviewed the questions from the infinite series workshop. Then we talked about functions f:f: \mathbb{C}\rightarrow \mathbb{C}. We defined limits of functions several ways, and we did the following examples.

  1. limz0zz\lim_{z \rightarrow 0} \frac{\bar{z}}{z}.

  2. Exercise 2.6

  3. Use the ϵδ\epsilon-\delta definition of limits to prove that if limzz0f(z)=L0\lim_{z \rightarrow z_0} f(z) = L \ne 0 then limzz01f(z)=1L\lim_{z \rightarrow z_0} \frac{1}{f(z)} = \frac{1}{L}.

    Consider |1f(z)1L|=|Lf(z)||Lf(z)|\left| \frac{1}{f(z)} - \frac{1}{L} \right| = \frac{|L - f(z)|} {|L f(z)|} We can use the fact that f(z)Lf(z) \rightarrow L to make the top as small as we want. We just need to make sure that the bottom doesn’t get close to zero at the same time. The trick is to use the triangle inequality to show that |f(z)|>|L|ϵ|f(z)| > |L| - \epsilon when |Lf(z)|<ϵ|L - f(z)| < \epsilon,

  4. Use the sequential definition of limits to prove the product rule for limits, i.e., limzz0f(z)g(z)=(limzz0f(z))(limzz0g(z))\lim_{z \rightarrow z_0} f(z) g(z) = (\lim_{z \rightarrow z_0} f(z)) (\lim_{z \rightarrow z_0} g(z) ).

Wed, Sep 18

We introduced the complex derivative for functions f:Df: D \rightarrow \mathbb{C}:

f(z)=limh0f(z+h)f(z)h.f'(z) = \lim_{h \rightarrow 0} \frac{f(z+h) - f(z)}{h}.

  1. Find the derivative of f(z)=z2f(z) = z^2.

What does it mean for a function to be differentiable? For functions f:nmf: \mathbb{R}^n \rightarrow \mathbb{R}^m, we say that ff is differentiable at a vector xnx \in \mathbb{R}^n if there is a matrix JJ (called the Jacobian) such that f(x+Δx)f(x)+JΔxf(x + \Delta x) \approx f(x) + J \Delta x more specifically, limΔx0f(x+Δx)(f(x)+JΔx)Δ=0.\lim_{\Delta x \rightarrow 0} \frac{f(x + \Delta x) - (f(x) + J \Delta x)}{\|\Delta \|} = 0. The expression f(x)+JΔxf(x) + J \Delta x is an affine linear approximation of ff near xx.

  1. Show that multiplication by a+iba + ib is a linear transformation on \mathbb{C} that corresponds to multiplying vectors in 2\mathbb{R}^2 by the matrix (abba).\begin{pmatrix} a & -b \\ b & a \end{pmatrix}. Note that the multiplication by a complex number can rotate and/or scale, but it cannot reflect or skew.

A complex function f:f: \mathbb{C}\rightarrow \mathbb{C} can be thought of as a real function f:22f: \mathbb{R}^2 \rightarrow \mathbb{R}^2, and f(z)f'(z) corresponds to the Jacobian matrix J=(Ref/xRef/yImf/xImf/y)J = \begin{pmatrix} \partial \operatorname{Re}f / \partial x & \partial \operatorname{Re}f/ \partial y \\ \partial \operatorname{Im}f / \partial x & \partial \operatorname{Im}f/ \partial y \end{pmatrix}

  1. Observe that if z=x+iyz = x+iy, then z2=x2y2+2xyiz^2 = x^2 - y^2 + 2 x y i. Find the Jacobian of the function f(x,y)=(x2y22xy)f(x,y) = \begin{pmatrix} x^2 - y^2 \\ 2xy \end{pmatrix} and show that it corresponds to the complex number 2z2z.

  2. Show that the function f(z)=zf(z) = \bar{z} is not complex differentiable. Explain why that makes sense, even though the corresponding map on 2\mathbb{R}^2, f(x,y)=(xy)f(x,y) = \begin{pmatrix} x \\ -y \end{pmatrix}, does have a Jacobian matrix.

Fri, Sep 20

We introduced the Cauchy-Riemann equations.

Theorem. Suppose f:Df: D \rightarrow \mathbb{C} is complex differentiable at z0z_0, and f(x+iy)=u(x,y)+iv(x,y)f(x+iy) = u(x,y) + i v(x,y). Then ff satisfies the Cauchy-Riemann equations ux=vy and vx=uy.\frac{\partial u}{\partial x} = \frac{\partial v}{\partial y} \text{ and } \frac{\partial v}{\partial x} = -\frac{\partial u}{\partial y}. Conversely, if f(x+iy)=u(x,y)+iv(x,y)f(x+iy) = u(x,y) + i v(x,y) and the partial derivatives of uu and vv are continuous in a disk around z0=x0+iy0z_0 = x_0 + i y_0 and satisfy the Cauchy-Riemann equations, then ff is complex differentiable at z0z_0.

  1. f(z)=ezf(z) = e^z

  2. This is Exercise 2.19 in the book: Define f(z)=0f(z) = 0 if zz is on either purely real or purely imaginary, and f(z)=1f(z) = 1 otherwise. Show that ff satisfies the Cauchy–Riemann equations at z=0z = 0, yet ff is not differentiable at z=0z = 0. Why doesn’t this contradict the theorem about a complex function being complex differentiable if and only if it satisfies the Cauchy-Riemann equations?

We say that a function ff that is complex differentiable in an open disk around z0z_0 \in \mathbb{C} is holomorphic at z0z_0.

Then we talked about smooth paths which are paths that are differentiable. If γ\gamma is a smooth path in \mathbb{C} and ff is holomorphic at every point on the curve, then the chain rule holds and says that ddtf(γ(t))=f(γ(t))γ(t).\frac{d}{dt} f(\gamma(t)) = f'(\gamma(t)) \cdot \gamma'(t).

Idea. Notice that the angle of the tangent vector gets rotated by the argument of f(γ(t))f'(\gamma(t)). If two different smooth curves both intersect at a point z0z_0, and ff is holomorphic at z0z_0, then since their tangent vectors are both rotated by the same angle, the angle between the two tangent vectors stays the same. This implies that holomorphic maps are conformal, that is they preserve angles between curves.

  1. f(z)=z2/5f(z) = z^2 / 5 (see Complex Grapher)

We also looked at the function eze^z on the complex grapher and saw that it was also conformal.


Week 5 Notes

Tentative Schedule

Day Section Topic
Mon, Sep 23 3.4 Exponential and trigonometric functions
Wed, Sep 25 3.5 Logarithms and complex exponentials
Fri, Sep 27 3.1 Möbius transforms

Mon, Sep 23

We started by reviewing the properties of the exponential function including its algebraic properties, domain, range and that it is periodic with period 2πi2\pi i. Then we introduced the complex natural logarithm which is the inverse of eze^z. There is one problem with defining the inverse: eze^z is an \infty-to-1 function, every w{0}w \in \mathbb{C}\backslash \{0\} has infinitely many pre-images. So we have two options:

  1. logz\log z denotes the multivalued inverse of eze^z. It has the form logz=ln|z|+iargz\log z = \ln |z| + i \operatorname{arg}z where argz\operatorname{arg}z is the multivalued argument function.

  2. Logz\operatorname{Log}z is the principal branch of logz\log z. It is a single valued function of the form Logz=ln|z|+iArgz\operatorname{Log}z = \ln |z| + i \operatorname{Arg}z where Argz\operatorname{Arg}z is the single valued principal branch of the argument function that takes values in (π,π](-\pi, \pi].

We did the following exercises.

  1. Calculate log(1+i)\log(1+i).

  2. If z=x+iyz = x+iy where x>0x > 0, then Argz=arctan(y/x)\operatorname{Arg}z = \arctan(y/x). Calculate the Cauchy-Riemann equations for Logz\operatorname{Log}z to verify that Logz\operatorname{Log}z is complex differentiable when x>0x > 0 (in fact it is complex differentiable everywhere except at its branch cut (,0](-\infty, 0]). What is the derivative of Logz\operatorname{Log}z?

Wed, Sep 25

We looked at the Cauchy-Riemann equations for Logz=12ln(x2+y2)+iarctan(yx)\operatorname{Log}z = \tfrac{1}{2} \ln(x^2 + y^2) + i \arctan \left( \tfrac{y}{x} \right) again. We also did the following.

  1. Give an example where Log(z+w)Logz+Logw\operatorname{Log}(z+w) \ne \operatorname{Log}z + \operatorname{Log}w.

We introduced principle values of a complex power zα=eαLogzz^\alpha = e^{\alpha \operatorname{Log}z} for any α,z\alpha, z \in \mathbb{C}, with z0z \ne 0. We calculated the following examples.

  1. i1/3i^{1/3}

  2. (1+i)2i(1+i)^{2-i}

  3. Find all other values of (1+i)2i(1+i)^{2-i} in addition to the principal value.

  4. Find all solutions of the equation e1/z=ie^{1/z} = i.

We finished by talking about the reciprocal map f(z)=1zf(z) = \tfrac{1}{z}. We looked at how it appears to transform lines & circles into lines and circles. We’ll look at why next time.

Fri, Sep 27

We started by proving this theorem about the reciprocal map.

Theorem. Let f(z)=1zf(z) = \tfrac{1}{z}. Then ff transforms lines and circles to lines in circles.

The key to the proof is the fact that the solution of the algebraic equation (with real coefficients) a(x2+y2)+b1x+b2y+c=0a(x^2 + y^2) + b_1 x + b_2 y + c = 0 is a circle (possibly degenerate to a point or \varnothing) if a0a \ne 0, and it is a line if a=0a = 0.

It helps to think of the reciprocal map f(z)=1zf(z) = \tfrac{1}{z} as a bijection (1-to-1 and onto map) from {}{}\mathbb{C}\cup \{ \infty\} \rightarrow \mathbb{C}\cup \{ \infty \}. We define 1=0 and 10=.\frac{1}{\infty} = 0 \text{ and } \frac{1}{0} = \infty. We call {}\mathbb{C}\cup \{ \infty \} the extended complex plane.

The reciprocal map is a special case of an important family of bijections on {}\mathbb{C}\cup \{\infty\} called Möbius transforms. A Möbius transform (also known as a Linear Fractional Transform) is a map f(z)=az+bcz+df(z) = \frac{a z + b}{c z + d} where a,b,c,da, b, c, d \in \mathbb{C} satisfy adbc0ad - bc \ne 0. We proved the following facts.

  1. A Möbius transform always has two fixed points in {}\mathbb{C}\cup \{ \infty \}.

  2. For any invertible matrix A=(a11a12a21a22)A = \begin{pmatrix} a_{11} & a_{12} \\ a_{21} & a_{22} \end{pmatrix}, we can define an associated Möbius transform TA(z)=a11z+a12a21z+a22.T_A(z) = \frac{a_{11} z + a_{12}}{a_{21} z + a_{22}}. Then if A,B2×2A, B \in \mathbb{C}^{2 \times 2} are any two intertible matrices, TATB=TAB.T_A \circ T_B = T_{AB}.

In particular the inverse of a Möbius transform can be found by inverting its matrix: TA1=TA1.T_A^{-1} = T_{A^{-1}}. Notice also that if you multiply a matrix by a constant, the Möbius transform doesn’t change, so TcA=TA.T_{cA} = T_A. That is convenient because the inverse of a 2-by-2 matrix is (abcd)1=1adbc(dbca).\begin{pmatrix} a & b \\ c & d \end{pmatrix}^{-1} = \frac{1}{ad - bc} \begin{pmatrix} d & -b \\ -c & a \end{pmatrix}. Therefore you can invert a Möbius transform f(z)=az+bcz+df(z) = \frac{a z + b}{c z + d} by swapping aa and dd and making bb and cc negative: f1(z)=dzbcz+a.f^{-1}(z) = \frac{d z - b}{-cz + a}.


Week 6 Notes

Tentative Schedule

Day Section Topic
Mon, Sep 30 3.2 Möbius transforms - con’d
Wed, Oct 2 Review
Fri, Oct 4 Midterm 1

We started by proving that

Theorem. Any Möbius transformation maps lines & circles to lines & circles.

To prove this, we showed that if c0c \ne 0, then T(z)=az+bcz+dT(z) = \frac{az + b}{cz + d} can be expressed as a composition of three maps T3(z)=(bcdac)z+ac,T2(z)=1z, and T1(z)=cz+d.T_3(z) = \left(\frac{bc-da}{c} \right) z + \frac{a}{c}, ~~~~~ T_2(z) = \frac{1}{z}, \text{ and } T_1(z) = cz + d.

Theorem 2. A Möbius transform is uniquely determined by where it maps any three points in {}.\mathbb{C}\cup \{ \infty \}.

Proof. If z1,z2,z3z_1, z_2, z_3 are any three distinct points in {}\mathbb{C}\cup \{\infty\}, suppose that there are two Möbius transforms TT and SS such that S(zi)=T(zi)S(z_i) = T(z_i) for each ii. Then S1TS^{-1} T has three distinct fixed points which is impossible unless S1=T1S^{-1} = T^{-1} which means that SS and TT are the same. \square

  1. Find a Möbius transform that sends \infty to 2, ii to \infty, and 11 to 11.

  2. Find a general formula for a Möbius transform that sends z1,z2,z3z_1, z_2, z_3 to w1w_1, w2w_2, w3w_3. Hint: It might help to start with a formula to send z1z_1 to 00, z2z_2 to 11 and z3z_3 to \infty.

  3. Find a Möbius transform that leaves +1+1 and 1-1 fixed but maps 0 to \infty. Actually, this wasn’t a great example because it turned out to be the reciprocal map!

  4. How does the Möbius transform from the previous example transform the circles in this figure where the blue circle is centered at ii?

Video: Möbius Transformations Revealed

Wed, Oct 2

We reviewed for the exam by talking about these two problems.

  1. How does the Möbius transform T(z)=ziz1T(z) = \dfrac{z-i}{z-1} transform the three shapes shown below?
1 i
  1. Suppose sns_n is a sequence in \mathbb{C} such that |sn+1sm+1|0.9|snsm||s_{n+1} - s_{m + 1}| \le 0.9 |s_n - s_m| for all m,nm, n \in \mathbb{N}.

    1. Show that sns_n is bounded by finding an upper bound for |sn||s_n|.

    2. Show that sns_n is a Cauchy sequence by finding an upper bound for |snsm||s_n - s_m| when n>mn > m.


Week 7 Notes

Tentative Schedule

Day Section Topic
Mon, Oct 7 4.1 Complex integrals
Wed, Oct 9 4.3 Cauchy’s theorem
Fri, Oct 11 7.3 & 7.4 Power series

Mon, Oct 7

We introduced complex integrals, which are defined for any piecewise smooth path γ:[a,b]\gamma: [a, b] \rightarrow \mathbb{C} by γf(z)dz=abf(γ(t))γ(t)dt.\int_\gamma f(z) \, dz = \int_a^b f(\gamma(t)) \, \gamma'(t) \, dt. We talked about why this definition makes sense and we did these examples.

  1. γz2dz\int_\gamma z^2 \, dz on the upper half of the unit circle from 11 to 1-1. We started by using Python to numerically approximate the integral with a Riemann sum:
from cmath import *

n = 1000
total = 0
for k in range(1000):
    delta_z = exp(1j * pi * (k + 1) / n) - exp(1j * pi * k / n)
    z = exp(1j * pi * k / n)
    total += z ** 2 * delta_z
    
print(total)
  1. γ(z¯)2dz\int_\gamma (\overline{z})^2 \, dz on the path γ(t)=t(1+i)\gamma(t) = t(1+i) with t[0,1]t \in [0,1].

  2. δ(z¯)2dz\int_\delta (\overline{z})^2 \, dz on the path δ(t)=t+it2\delta(t) = t+it^2 with t[0,1]t \in [0,1]. (Link: Sympy code)

from sympy import *
x, y, z = symbols('x y z')
t = symbols('t',real=True)

f = conjugate(z)**2
g = t+t**2*I

print(integrate(f.subs(z,g)*diff(g,t),(t,0,1)))
  1. Use the formula length(γ)=ab|γ(t)|dt\operatorname{length}(\gamma) = \int_a^b |\gamma'(t)| \, dt to find the length of the unit circle.

Wed, Oct 7

Today we introduced

Cauchy’s Theorem. If ff is holomorphic on a simply connected open set DD and CC is a simple, closed, piecewise, smooth curve in DD, then Cf(z)dz=0.\oint_C f(z) \, dz = 0.

We defined what simply connected means (intuitively it means that DD has no holes), and we reviewed simple closed curves. We also talked about why the specific parametrization of a curve does not matter for integrals (see Proposition 4.2), but the orientation does, and the symbol \oint indicates that the orientation of a simple closed curve is positive (counterclockwise). We calculated

  1. |z|=11zdz\oint_{|z| = 1} \frac{1}{z} \, dz

and saw that Cauchy’s theorem does not apply because the reciprocal function is not holomorphic at 0.

Then we talked about how to prove Cauchy’s theorem using

Green’s Theorem. If P(x,y)P(x,y) and Q(x,y)Q(x,y) are real-valued functions with continuous partial derivatives in a simply connected domain DD with a piecewise smooth boundary curve CC, then CPdx+Qdy=D(QxPy)dA.\oint_C P \, dx + Q \, dy = \iint_D \left( \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y} \right) \, dA.

Here is a nice video explanation of the intuition behind Green’s theorem. One problem with using Green’s theorem to prove Cauchy’s theorem is that it requires us to assume that f(z)f'(z) is continuous (so that the partial derivatives are all continuous. It turns out that is true for all holomorphic functions, as we will see later. The Complex Variables book has a different proof of Cauchy’s theorem that doesn’t require this assumption.

We finished by starting to talk about how useful Cauchy’s theorem is. One application of Cauchy’s theorem is that it implies that every holomorphic function has a holomorphic antiderivative. You can define the antidervative F(z)F(z) of a holomorphic function ff on a simply connected open domain this way: F(z)=γf(z)dzF(z) = \int_\gamma f(z) \, dz where γ\gamma is any piecewise smooth path from a fixed z0Dz_0 \in D to zz. We talked about why F(z)F(z) is well-defined.

Friday, Oct 11

Last time we showed that when a function f(z)f(z) is holomorphic on a simply connected open set DD, then the integral on any two paths γ1\gamma_1 and γ2\gamma_2 that both start at a point z0z_0 and end at z1z_1 and stay inside DD must be the same:

γ1f(z)dz=γ2f(z)dz.\int_{\gamma_1} f(z) \, dz = \int_{\gamma_2} f(z) \, dz.

In other words, the value of the integral is independent of the path. Today we proved some important theorems to make working with complex integeral easier.

Evaluation Theorem. If ff has a holomorphic antiderivative FF in a path connected open domain DD, then for any piecewise smooth path γ:[a,b]D\gamma: [a,b] \rightarrow D, γf(z)dz=F(γ(b))F(γ(a)).\int_\gamma f(z) \, dz = F(\gamma(b)) - F(\gamma(a)).

This theorem follows immediately from the definition of a complex path integral and the chain rule, combined with the evaluation theorem from real variable calculus. An application of this theorem is the following.

  1. Calculate γzndz\int_\gamma z^n \, dz for any piecewise smooth path γ\gamma that begins at zz \in \mathbb{C} and ends at ww \in \mathbb{C}. Note: this solution works as long as n1n \ne -1 with the caveat that the path shouldn’t pass through 0 if nn is negative. Why won’t this work if n=1n = -1?

ML-Inequality. If ff is a continuous complex-valued function on a path connected open domain DD and γ\gamma is a piecewise smooth path in DD, then |γf(z)dz|maxzrange(γ)|f(z)|length(γ).\left| \int_\gamma f(z) \, dz \right| \le \max_{z \in \operatorname{range}(\gamma)} |f(z)| \cdot \operatorname{length}(\gamma).

This is like a triangle inequality for complex integrals. In fact, you can use the triangle inequality to prove it.

Antiderivative Theorem. If ff is holomorphic in a simply connected open domain DD, then ff has a holomorphic antiderivative FF in DD.

Last time we used independence of path to define F(z)=γf(z)dzF(z) = \int_\gamma f(z) \, dz where γ\gamma is any path in DD from z0z_0 to zz. We just have to show that F(w)=f(w)F'(w) = f(w) for every wDw \in D to prove that this function FF is the antiderivative needed for the antiderivative theorem. Two key ideas in the proof are the following:

  1. Let δ\delta be a parametrization of the line segment from ww to w+hw+h in DD. Show that δf(w)dz=hf(w)\int_\delta f(w) \, dz = h f(w).

  2. Use the ML-inequality to estimate |δf(z)f(w)dzh|.\left| \frac{\int_\delta f(z) - f(w) \, dz}{h} \right|.


Week 8 Notes

Tentative Schedule

Day Section Topic
Mon, Oct 14 Fall break, no class
Wed, Oct 16 4.4 Cauchy’s integral formula
Fri, Oct 18 5.1 Holomorphic implies infinitely differentiable

Wed, Oct 16

Cauchy Integral Formula. Let ff be holomorphic in a simply connected open domain DD \subseteq \mathbb{C} that contains a simple, closed, piecewise smooth curve CC around a point ww. Then f(w)=12πiCf(z)zwdz.f(w) = \frac{1}{2 \pi i} \, \oint_C \frac{f(z)}{z-w} \, dz.

That’s kind of weird if you think about it. This theorem says that the value of ff at a point inside a curve is completely determined by the values of ff on the curve. The proof required two key insights:

  1. If ff is complex differentiable at ww, then f(z)=f(w)+f(w)(zw)+ϵ(z),f(z) = f(w) + f'(w) (z-w) + \epsilon(z), where the error term ϵ(z)\epsilon(z) satisfies limzwϵ(z)zw=0\lim_{z \rightarrow w} \frac{\epsilon(z)}{z-w} = 0.

  2. If C1C_1 and C2C_2 are two different simple, closed, piecewise smooth curves in a region where ff is holomorphic, then C1f(z)zwdz=C2f(z)zwdz.\oint_{C_1} \frac{f(z)}{z-w} \, dz = \oint_{C_2} \frac{f(z)}{z-w} \, dz.

Use the Cauchy integral formula to evaluate the following integrals.

  1. Cezz1dz\oint_C \frac{e^z}{z-1} \, dz where CC is the square with vertices at 10,10i,10,10i10, 10i, -10, -10i. (https://youtu.be/NJap6Vm5mEk)

  2. |zi|=11z2+1dz\oint_{|z-i|=1} \frac{1}{z^2+1} \, dz

  3. |z|=3ezz22zdz\oint_{|z|=3} \frac{e^z}{z^2 - 2z} \, dz

  4. |z|=1z2+1z(2z+1)dz\oint_{|z| = 1} \frac{z^2+1}{z(2z+1)} \, dz. (https://youtu.be/APoh2B5S2ok)

We finished by looking at the following application of the Cauchy integral formula:

  1. 1x4+1dx\int_{-\infty}^\infty \frac{1}{x^4 + 1} \, dx.

The key is to integrate over the boundary of the upper half of the disk of radius RR centered at the origin (where RR is very large). Then use the ML-inequality to show that the integral over the circular part of the boundary goes to zero, leaving just the real integral we want.

Friday, Oct 18

Last time we ran into the problem of figuring out how to decompose functions like

g(z)=1z4+1 into the form f(z)zw.g(z) = \frac{1}{z^4 + 1} \text{ into the form } \frac{f(z)}{z - w}.

That isn’t always easy to do. Here is a shortcut that you can use. If g(z)=f(z)zwg(z) = \frac{f(z)}{z-w} where ff is holomorphic in a neighborhood around ww, then f(z)=g(z)(zw)f(z) = g(z) \cdot (z-w) for every zwz \ne w, and since holomorphic functions are continuous: f(w)=limzwg(z)(zw).f(w) = \lim_{z \rightarrow w} g(z) \cdot (z - w). The value of this limit is called the residue of g(z)g(z) at ww. You can often use L’Hospital’s rule to find the residue.

  1. Find the residues of 1z4+1\dfrac{1}{z^4+1} at w1=eπi/4w_1 = e^{\pi i/ 4} and w2=e3πi/4w_2 = e^{3 \pi i /4}.

A function f:Df:D \rightarrow \mathbb{C} is analytic on an open domain DD \subseteq \mathbb{C} if it is infinitely (complex) differentiable and for every wDw \in D, there is a power series k=0ak(zw)k\sum_{k = 0}^\infty a_k (z-w)^k that absolutely converges to f(z)f(z) for every zz in an open disk around ww.

Theorem (Holomorphic implies Analytic). If ff is holomorphic in an open set containing a closed disk BR(w)B_R(w), then f(z)f(z) has a power series f(z)=k=0ak(zw)kf(z) = \sum_{k = 0}^\infty a_k (z-w)^k which converges absolutely inside the disk BR(w)B_R(w). The coefficients of the power series are ak=12πi|zw|=Rf(z)(zw)k+1dz.a_k = \frac{1}{2\pi i} \oint_{|z-w|=R} \frac{f(z)}{(z-w)^{k+1}} \, dz.

And immediate corollary is:

Cauchy Integral Formula for Derivatives. Let ff be holomorphic in a simply connected open domain DD \subseteq \mathbb{C} that contains a simple, closed, piecewise smooth curve CC around a point ww. Then for any nn \in \mathbb{N}, f(n)(w)=n!2πiCf(z)(zw)n+1dz.f^{(n)}(w) = \frac{n!}{2 \pi i} \, \oint_C \frac{f(z)}{(z-w)^{n+1}} \, dz.

The key to proving the first theorem is to confirm that the following expression converges when |uw|<R|u - w| < R. 2πif(u)=f(z)zudz=f(z)(zw)(uw)dz=f(z)(zw)1(uwzw)dz2\pi i f(u) = \oint \frac{f(z)}{z - u} \, dz = \oint \frac{f(z)}{(z - w) - (u - w)} \, dz = \oint \frac{\frac{f(z)}{(z-w)}}{1 - \left( \frac{u - w}{z-w}\right)} \, dz =k=0f(z)(zw)k+1(uw)kdz=k=0f(z)(zw)k+1(uw)kdz.= \oint \sum_{k = 0}^\infty \frac{f(z)}{(z-w)^{k+1}} (u-w)^k \, dz =\sum_{k = 0}^\infty \oint \frac{f(z)}{(z-w)^{k+1}} (u-w)^k \, dz.

We ran out of time before we could do this exercise, but we’ll start with it on Monday.

  1. |z|=3eiz(z+i)2dz\oint_{|z| = 3} \frac{e^{iz}}{(z+i)^2} \, dz. (https://youtu.be/WJOf4PfoHow)

Week 9 Notes

Tentative Schedule

Day Section Topic
Mon, Oct 21 8.1 Power series and holomorphic functions
Wed, Oct 23 5.3 Winding number and the Fund. Thm. of Algebra
Fri, Oct 25 5.3 Entire functions & Liouville’s theorem

Mon, Oct 21

We ran out of time last class before we could do this example.

  1. |z|=3eiz(z+i)2dz\oint_{|z| = 3} \frac{e^{iz}}{(z+i)^2} \, dz. (https://youtu.be/WJOf4PfoHow)

Kade asked what happens if the function f(z)f(z) in Cauchy’s integral formula is not holomorphic everywhere in the domain, which was a great questions, so we looked at a variant of the previous problem where we replaced the function eize^{iz} with the function 1/z1/z.

  1. |z|=31/z(z+i)2dz\oint_{|z| = 3} \frac{1/z}{(z+i)^2} \, dz.

Then we talked about the following partial converse to Cauchy’s theorem.

Morera’s Theorem. Let f:Df : D \rightarrow \mathbb{C} be continuous on a simply connected open domain DD and suppose that Cf(z)dz=0\oint_C f(z) \, dz = 0 for every piecewise smooth simple closed curve CC in DD. Then ff is holomorphic in DD.

We outlined the proof by reviewing the proof of the antiderivative theorem. Since integrals of ff are independent of path in DD, we can define an antiderivative F(w)=z0wf(z)dz.F(w) = \int_{z_0}^w f(z) \, dz. The same proof as before shows that the antiderivative is holomorphic, but Cauchy’s integral formula for derivatives implies that ff is also complex differentiable in DD, so it is also holomorphic.

After that, we reviewed several important facts about power series.

Taylor Series Radius of Convergence. If ff is holomorphic in an open domain DD, and wDw \in D, then the Taylor series for ff centered at ww f(z)=k=0f(k)(w)k!(zw)kf(z) = \sum_{k = 0}^\infty \frac{f^{(k)}(w)}{k!} (z-w)^k converges absolutely for all zz inside an open disk around ww with radius of convergence equal to the distance from ww to the nearest singularity of ff.

Another fact which we stated but did not prove is

Theorem (Analytic implies Holomorphic). If ff is analytic at ww, then ff is holomorphic in an open disk around ww.

You can use Morera’s theorem to prove this. But the proof also requires knowing that analytic functions are continuous in an open disk too. The easiest way to prove that is to argue that the partial sums of the Taylor series converge uniformly to f(z)f(z) in the disk, but we haven’t talked about how continuity is preserved under uniform convergence.

We left the following exercise to try on your own.

  1. Suppose that the power series k=0akzk\sum_{k = 0}^\infty a_k z^k has radius of convergence RR. Prove that its derivative k=0kakzk1\sum_{k = 0}^\infty k a_k z^{k-1} also has radius of convergence RR.

Wed, Oct 23

The winding number of a (piecewise smooth) closed path γ\gamma around a point ww \in \mathbb{C} is 12πiγ1zwdz\frac{1}{2\pi i} \int_\gamma \frac{1}{z-w} \, dz (assuming that γ\gamma does not intersect ww).

Theorem (Polynomial Transformations of the Unit Circle). If p(z)p(z) is a polynomial with no roots on the unit circle, then the winding number of p(eit),0t2πp(e^{it}), 0 \le t \le 2 \pi, around the origin is equal to the number of roots of pp inside the unit circle (counting multiplicity).

We used the (now updated) complex grapher to look at some examples. Then we proved this theorem by doing the following exercises in class:

  1. If γ(t)=p(eit)\gamma(t) = p(e^{it}), show that γ1zz=|z|=1p(z)p(z)dz.\int_\gamma \frac{1}{z} \, z = \oint_{|z|=1} \frac{p'(z)}{p(z)} \, dz.
  1. For a polynomial p(z)=(zr1)m1(zr2)m2(zrk)mk,p(z) = (z-r_1)^{m_1} (z-r_2)^{m_2} \cdots (z-r_k)^{m_k}, show that p(z)p(z)=m1zr1+m2zr2++mkzrk.\frac{p'(z)}{p(z)} = \frac{m_1}{z-r_1} + \frac{m_2}{z-r_2} + \ldots + \frac{m_k}{z-r_k}. Hint: Use mathematical induction on kk if you want to make this rigorous.

We can use this theorem to derive one proof of the Fundamental Theorem of Algebra.

Fundamental Theorem of Algebra. Any non-constant polynomial with complex coefficients has a root in \mathbb{C}.

Assume without loss of generality that pp is a monic (leading coefficient is one) degree nn polynomial. Note that pp has a root if and only if the polynomial p(Rz)/Rnp(Rz)/R^n has a root for any R>0R > 0.

  1. Show that there exists RR large enough so that |znp(Rz)Rn|<1|z^n - \frac{p(Rz)}{R^n}| < 1 for all zz on the unit circle.

You can get a proof of the Fundamental Theorem of Algebra by combining this inequality with the following intuitive result from topology.

Dog on a Leash Theorem. Suppose that γ\gamma and δ\delta are closed, piecewise smooth paths from [a,b][a,b] to \mathbb{C} such that |γ(t)δ(t)|<|γ(t)||\gamma(t) - \delta(t)| < |\gamma(t)| for all t[a,b]t \in [a,b], then the winding numbers of γ\gamma and δ\delta around 0 are the same.

Fri, Oct 25

We started by talking about Homework 9. We did the following example, which is similar to problem #4:

  1. Find the Maclaurin series for the function 1z38.\frac{1}{z^3 - 8}.

A function f:f: \mathbb{C}\rightarrow \mathbb{C} is entire if it is holomorphic everywhere in \mathbb{C}.

Liouville’s Theorem. Any bounded entire function is constant.

To prove Liouville’s theorem, we need a quick observation and an idea. The observation is that if f(z)=0f'(z) = 0 everywhere, then ff must be constant. The idea is that we can use Cauchy’s integral formula for derivatives to calculate f(z)f'(z) if ff is bounded and entire: f(z)=12πi|ξz|=Rf(ξ)(ξz)2dξ.f'(z) = \frac{1}{2\pi i} \oint_{|\xi - z| = R} \frac{f(\xi)}{(\xi - z)^2} \, d\xi. What happens if the radius RR gets really really big? Use the max-times-length inequality to estimate |f(z)||f'(z)|.

We finished by applying Liouville’s theorem to re-prove the Fundamental Theorem of Algebra. Unlike the proof we gave last time, using Liouville’s theorem to prove the FToA only guarantees one root. But that is enough if you use this result from algebra.

Division Algorithm for Polynomials. If p(z)p(z) and d(z)d(z) are polynomials with complex coefficients and degree(d)>0\operatorname{degree}(d) > 0, then there are unique polynomials q(z)q(z) and r(z)r(z) with degree(r)<degree(d)\operatorname{degree}(r) < \operatorname{degree}(d) such that p(z)=q(z)d(z)+r(z).p(z) = q(z) d(z) + r(z).

A corollary of the division algorithm and the Fundamental Theorem of Algebra is:

Theorem. An n-th degree polynomial p(z)=anzn++a0p(z) = a_n z^n + \ldots + a_0 with complex coefficients always factors into the form an(zr1)(zr2)(zrn)a_n(z-r_1) (z-r_2) \cdots (z-r_n) where reach rkr_k is a root of p(z)p(z) (repeats are allowed).

We used these ideas to completely factor the polynomial:

  1. p(z)=z33z2+z3p(z) = z^3 - 3z^2 + z - 3.

Another good example to factor completely would be:

  1. p(z)=z38p(z) = z^3 - 8.

Week 10 Notes

Tentative Schedule

Day Section Topic
Mon, Oct 28 7.4 Radius of convergence
Wed, Oct 30 7.3 Uniform convergence
Fri, Nov 1 8.2 Zeroes of analytic functions

Mon, Oct 28

  1. Consider the following power series for Log(z)=(z1)(z1)22+(z1)33(z1)44+.\operatorname{Log}(z) = (z-1) - \frac{(z-1)^2}{2} + \frac{(z-1)^3}{3} - \frac{(z-1)^4}{4} + \ldots.
    1. Re-write this series in Σ\Sigma-notation.
    2. What are the coefficients ana_n of this power series?
    3. What is the center and radius of convergence?
    4. How would you find a Taylor series for Log(z)\operatorname{Log}(z) centered at w=1+iw = -1 + i? What is the radius of convergence for that Taylor series?
  2. Find power series for the following functions by using the Maclaurin series for eze^z, sinz\sin z, and cosz\cos z:
    1. ez2/2e^{-z^2/2}
    2. sinzz\dfrac{\sin z}{z}

Why do power series have a radius of convergence?

Lemma. Let n=0an(zw)n\sum_{n = 0}^\infty a_n (z-w)^n be a power series that converges for one z0z_0 \in \mathbb{C}. Then it converges absolutely for all zz \in \mathbb{C} such that |zw|<|z0w||z - w| < |z_0 - w|.

We proved this in class, and then we talked about why the least upper bound of the set

{r0:n=0anrn converges}\left\{ r \ge 0 ~:~ \sum_{n = 0}^\infty a_n r^n \text{ converges} \right\}

must be the radius of convergence.

Theorem. Every power series k=0ak(zw)k\sum_{k = 0}^\infty a_k (z- w)^k has a radius convergence R[0,]R \in [0, \infty]. If |zw|<R|z-w| < R, then the power series converges absolutely and if |zw|>R|z-w| > R, then the series diverges.

Another result that you can derive from the above is:

Ratio Test. If limn|an||an+1|=R\lim_{n \rightarrow \infty} \frac{|a_n|}{|a_{n+1}|} = R exists, then RR is the radius of convergence of the power series n=0an(zw)n\sum_{n = 0}^\infty a_n (z - w)^n.

We talked about the intuition of why this is true: because there is an NN after which, our power series will be sandwiched between two geometric series, one with radius of convergence R+ϵR+\epsilon and the other with radius of convergence RϵR-\epsilon.

Here are some exercises we didn’t have time for in class.

  1. Use the ratio test to find the radius of convergence of the power series 1+z+34z2+48z3+516z4+.1 + z + \frac{3}{4} z^2 + \frac{4}{8} z^3 + \frac{5}{16} z^4 + \ldots.
  1. Why can’t you use the ratio test as written to find the radius of convergence for the Maclaurin series for sinz\sin z?

Wed, Oct 30

A sequence of functions fn:Df_n: D \rightarrow \mathbb{C} converges uniformly to a function f:Df: D \rightarrow \mathbb{C} if for every ϵ>0\epsilon > 0, there exists N>0N >0 such that |fn(z)f(z)|<ϵ|f_n(z) - f(z)| < \epsilon for all nNn \ge N and zDz \in D.

  1. Show that if ff is a holomorphic function with a power series n=0an(zw)n\sum_{n = 0}^\infty a_n (z-w)^n that has radius of convergence R>0R > 0, then the partial sums converge uniformly to ff inside any closed disk around ww with a radius r<Rr < R.

  2. Show that if fnf_n is a sequence of continuous functions that converge uniformly to ff on an open set DD, then ff is continuous on DD. You have to show that for any wDw \in D and any ϵ>0\epsilon > 0, there is a δ>0\delta > 0 such that |zw|<δ implies |f(z)f(w)|<ϵ.|z - w | < \delta \text{ implies } |f(z) - f(w)| < \epsilon. Hint: Fix ϵ\epsilon. Choose NN big enough so that |fnf|<ϵ3|f_n - f| < \tfrac{\epsilon}{3} for all nNn \ge N. Then use the continuity of fnf_n and the triangle inequality (see the image below).

After that we talked about the idea that if a function is defined by a power series in a open disk, then you can often analytically continue the function by constructing a new power centered at a different point inside the open disk. In this way you can extend the function past its original disk of convergence. This has one immediate consequence. If a holomorphic function f:Df:D \rightarrow \mathbb{C} on an open connected domain DD has fn(w)=0f^{n}(w) = 0 for all n0n \ge 0 at one point wDw \in D, then f=0f = 0 everywhere in DD.

Theorem (Classification of Zeros) If ff is a holomorphic function in an open connected domain DD \subseteq \mathbb{C} and f(w)=0f(w) = 0 for some wDw \in D, then either

  1. f=0f = 0 everywhere on DD or

  2. The Taylor series for ff centered at ww has a first nonzero term am(zw)ma_m (z-w)^m. In that case f(z)=(zw)mg(z)f(z) = (z-w)^m g(z) where gg is holomorphic in an open disk around ww and g(w)0g(w) \ne 0. Then we say that ww is a zero of order mm.

Find the orders for the following zeros:

  1. w=0w = 0 for f(z)=zsinzf(z) = z-\sin z.

  2. w=πiw = \pi i for g(z)=ez+1g(z) = e^z + 1.

Fri, Nov 1

Theorem (Zeros are Isolated) If ff is a non-constant analytic function in an open connected domain DD \subseteq \mathbb{C}, then the zeros of ff are isolated which means that for every zero wDw \in D, there is an open disk around ww that does not contains any other zeros of ff.

Proof. By the classification of zeros theorem, f(z)=(zw)mg(z)f(z) = (z-w)^m g(z) for some m>0m > 0 where gg is holomorphic and g(w)0g(w) \ne 0. Then by continuity f0f \ne 0 in an open disk around ww, except at ww. \square

Identity Principle. If f:Df: D \rightarrow \mathbb{C} is holomorphic on an open connected domain DD and ff has a sequence of distinct zeros wnw_n that converge to a point wDw \in D, then f=0f = 0 everywhere on DD.

  1. Why doesn’t the identity principle apply to the function sin(1/z)\sin(1/z) which has infinitely many zeros in the open disk of radius one around 1?

  2. This theorem is called the identity principle because it implies that if two holomorphic functions are the same on a compact (closed and bounded) infinite set inside DD, then they must the same on all of DD.

If DD is an open set, wDw \in D, and f:D{w}f: D \backslash \{ w \} \rightarrow \mathbb{C} is holomorphic, then we say that ww is an isolated singularity of ff. There are three types of isolated singularities.

Here are examples of the three types of singularity.

  1. sinzzπ\frac{\sin z}{z- \pi} at w=πw = \pi.

  2. z(z5)2\frac{z}{(z - 5)^2} at w=5w = 5.

  3. e1/ze^{1/z} at w=0w = 0.


Week 11 Notes

Tentative Schedule

Day Section Topic
Mon, Nov 4 9.1 Classification of singularities
Wed, Nov 6 9.3 The argument principle and Rouche’s theorem
Fri, Nov 8 8.3 Laurent series

Mon, Nov 4

Classification of Singularities. If ff has an isolated singularity at ww, then

  1. ww is a removable singularity if and only if there is a holomorphic function gg on an open disk around ww such that f(z)=g(z)f(z) = g(z) everywhere, except (technically) ww on the disk.

  2. ww is a pole if and only if there is a holomorphic gg and positive integer mm such that f(z)=g(z)(zw)mf(z) = \frac{ g(z) }{(z-w)^m}
    for all zwz \ne w in an open disk around ww and g(w)0g(w) \ne 0. We call mm the order of the pole at w.w.

  1. Prove that if limzw(zw)f(z)=0\lim_{z \rightarrow w} (z-w) f(z) = 0, then g(z)={(zw)2f(z) if zw,0 if z=wg(z) = \begin{cases} (z-w)^2 f(z) & \text{ if } z \ne w, \\ 0 & \text{ if } z = w \end{cases} is holomorphic at z=wz = w.

  2. What can you say about the power series for gg centered at ww?

  3. If limzw|f(z)|=\lim_{z \rightarrow w} |f(z)| = \infty, then the function g(z)=1/f(z)g(z) = 1/f(z) has a removable singularity at ww. What can you say about the power series for gg centered at ww?

With this classification, we can prove a really cool theorem:

Argument Principle. Suppose that ff is holomorphic on an open simply connected domain DD, except for isolated poles. If γ\gamma is a piecewise smooth, positively oriented, simple closed path in DD and γ\gamma does not pass through any zeros or poles of ff, then the winding number of the path fγf \circ \gamma around the origin is 12πiγf(z)f(z)dz={number of zerosof f inside γ}{number of polesof f inside γ}.\frac{1}{2 \pi i} \int_\gamma \frac{f'(z)}{f(z)} \, dz = \left\{ \begin{array}{l} \text{number of zeros} \\ \text{of }f\text{ inside } \gamma \end{array} \right\} - \left\{ \begin{array}{l} \text{number of poles} \\ \text{of }f\text{ inside } \gamma \end{array} \right\}.

  1. If ff has a zero at ww of order mm, then f(z)=(zw)mg(z)f(z) = (z-w)^m g(z) where gg is holomorphic, g(w)0,g(w) \ne 0, and f(z)f(z)=g(z)g(z)+m(zw).\frac{f'(z)}{f(z)} = \frac{g'(z)}{g(z)} + \frac{m}{(z-w)}. Similarly, if ff has a pole at ww of order mm, then f(z)=(zz0)mg(z)f(z) = (z-z_0)^{-m} g(z) where gg is holomorphic, g(w)0g(w) \ne 0, and f(z)f(z)=g(z)g(z)m(zw).\frac{f'(z)}{f(z)} = \frac{g'(z)}{g(z)} - \frac{m}{(z-w)}.

  2. If ww is the only zero or pole of ff inside γ\gamma, explain why this proves that the winding number of f(γ)f(\gamma) around ww is either mm or m-m (depending on whether ww is a zero or a pole).

  3. What if ff has several zeros and/or poles inside γ\gamma?

Wed, Nov 6

The main application of the argument principle is not to find the winding number. It is actually to use the winding number to confirm the existence of zeros or poles.

Rouche’s Theorem. Suppose that f,gf, g are holomorphic on an open, simply connected domain DD, and γ\gamma is a positively oriented, simple, closed, piecewise smooth path in DD. If |f(z)g(z)|<|f(z)||f(z) - g(z)| < |f(z)| for all zz in the range of γ\gamma, then ff and gg have the same number of zeros inside γ\gamma (counting multiplicity).

Proof. The inequality condition is the same as the one for the dog on a leash theorem. Therefore fγf \circ \gamma and gγg \circ \gamma have the same winding numbers around the origin and then by the argument principle they must have the same number of zeros (they can’t have poles because they are holomorphic on DD). \square

  1. Use Rouche’s theorem to do Exercise 9.21 from the book.

Open Mapping Principle. If ff is a non-constant holomorphic function on an open set UU, then f(U)f(U) is an open set.

Proof. We have to show that for every w0f(U)w_0 \in f(U) is an interior point of f(U)f(U). In other words, there is an open disk around w0w_0 that is completely contained inside f(U)f(U). By definition, there is a z0z_0 such that f(z0)=w0f(z_0) = w_0, and w0w_0 is a zero of the function fw0f-w_0. We just have to show that g=fwg = f-w has a zero for all ww sufficiently close to w0w_0.

Since zeros are isolated, we can choose a circle around z0z_0 in UU that doesn’t have any zeros on its boundary. Let γ\gamma be a positively oriented parametrization of that circle. Let δ\delta be the distance from the closest point on fγf \circ \gamma to w0w_0. If |ww0|<δ|w - w_0| < \delta, then Rouche’s theorem guarantees that fwf- w has a zero inside γ\gamma and therefore wf(U)w \in f(U). \square

Fri, Nov 8

We ran out of time last class before we could write down this next corollary of the open mapping principle.

Maximum Modulus Principle. If ff is a non-constant holomorphic function on an open set UU, then |f||f| cannot have a local maximum in UU. If in addition, UU is bounded and ff is continuous on the closure of UU, then the maximum of |f||f| occurs on the boundary of UU.

Proof. Suppose that z0z_0 is a local maximum, that is, |f(z0)||f(z)||f(z_0)| \ge |f(z)| for all zz in a small disk around z0z_0. This is a contradiction because f(Br(z0))f(B_r(z_0)) is an open set, so it contains an open disk around f(z0)f(z_0). \square

A Laurent series centered at ww \in \mathbb{C} is a series of the form k=ak(zw)k\sum_{k = -\infty}^\infty a_k (z-w)^k where the coefficients aka_k are complex numbers and zz is a complex variable. We say that a function f(z)f(z) has a Laurent series in a domain DD if there is a Laurent series that converges absolutely to ff for all zDz \in D. The sum of the terms of a Laurent series with negative powers is called the principal part and the sum of terms with nonnegative powers is the analytic part of a Laurent series.

We calculated Laurent series for the following functions in class.

  1. coszz5\dfrac{\cos z}{z^5}.

  2. e1/ze^{1/z}.

We talked about how you can tell from the Laurent series above that 0 is an order 5 pole for the first example and an essential singularity in the second.

  1. Find the Laurent series for f(z)=z3(z1)2f(z) = \dfrac{z^3}{(z-1)^2} centered at w=1w=1. Hint: Use the substitution u=z1u = z - 1 to simplify ff.

A function might have different Laurent series with the same center (unlike power series which are unique). Each Laurent series will converge in a different region. It is not hard to show that the region of convergence for a Laurent series is always an annulus, but we didn’t prove that in class.

  1. Find the Laurent series for g(z)=2z41z1g(z) = \dfrac{2}{z-4} - \dfrac{1}{z-1} on the annulus with 1<|z|<41 < |z| < 4. Then find a Laurent series on the annulus 4<|z|<4 < |z| < \infty.

Theorem. If f(z)f(z) has a Laurent series k=ak(zw)k\sum_{k = -\infty}^{\infty} a_k (z-w)^k in an open annulus, then Cf(z)dz=2πia1\oint_C f(z) \, dz = 2 \pi i a_{-1} for any simple, closed, piecewise sooth curve CC in the annulus that winds around ww.

We proved this by using Cauchy’s theorem and Cauchy’s integral formulas to evaluate Cak(zw)kdz\oint_C a_k (z-w)^k \, dz for every kk, and we saw that the only integral that wasn’t zero was when k=1k = -1. The coefficient a1a_{-1} is called the residue of ff at ww.


Week 12 Notes

Tentative Schedule

Day Section Topic
Mon, Nov 11 9.2 Residues
Wed, Nov 13 Review
Fri, Nov 15 Midterm 2

Mon, Nov 11

Residue Theorem. If ff is holomorphic in an open simply connected domain DD, except at some isolated singularities, and γ\gamma is a positively oriented, simple, closed, piecewise smooth path in DD that avoids the singularities of ff, then there are only finitely many singularities z1,,znz_1, \ldots, z_n inside γ\gamma and γf(z)dz=2πik=1nRes(f,zk)\int_\gamma f(z) \, dz = 2 \pi i \sum_{k = 1}^n \operatorname{Res}(f, z_k) where Res(f,zk)\operatorname{Res}(f,z_k) is the residue of ff at zkz_k.

We talked about why this theorem is true. Then we looked at some examples.

  1. |z|=1coszz5dz\oint_{|z| = 1} \dfrac{\cos z}{z^5} \, dz.

  2. Evaluate |z2|=3ezsinzdz\oint_{|z - 2| = 3} \dfrac{e^z}{\sin z} \, dz.

We got bogged down with the second example because we were trying to derive the formula for a residue every time from scratch.

Wed, Nov 13

Since we got held up last time with some tedious residue computations, I introduced the following formula today.

Theorem (Residue Formula). If ff has a pole at ww with order no more than mm, then Res(f,w)=1(m1)!limzw[(ddz)(m1)(zw)mf(z)].\operatorname{Res}(f, w) = \frac{1}{(m-1)!} \lim_{z \rightarrow w} \left[ \left(\frac{d}{dz}\right)^{(m-1)} (z-w)^m f(z)\right].

Proof. This follows from the fact that (zw)mf(z)(z-w)^m f(z) has a removable singularity at ww. So we could (in theory) find a power series for (zw)mf(z)(z-w)^m f(z). Then the residue corresponds the (m1)(m-1)-th order coefficient of the power series. \square

We applied the formula to the following examples:

  1. Find the orders and residues of each pole of f(z)=1(z1)2(z3)f(z) = \dfrac{1}{(z-1)^2 (z-3)}.

  2. Evaluate |z2|=3zsinzdz\displaystyle \oint_{|z-2|=3} \dfrac{z}{\sin z} \, dz.

We also did some review problems to prepare for the midterm including:

  1. Use Rouche’s theorem to determine how many zeros f(z)=z7+3z2+1f(z) = z^7 + 3z^2 + 1 has in the annulus with 1|z|21 \le |z| \le 2.

  2. Exercise 9.2


Week 13 Notes

Tentative Schedule

Day Section Topic
Mon, Nov 18 Contour integrals
Wed, Nov 20 6.1 Harmonic functions
Fri, Nov 22 Vector fields & holomorphic functions

Mon, Nov 18

Today we talked about applications of residues to real integrals. We started with the following result.

Theorem (Integrals of Rational Functions on the Real Line). If pp and qq are polynomials such that qq has no real zeros and the degree of qq is at least two greater than pp, then p(x)q(x)dx\int_{-\infty}^{\infty} \frac{p(x)}{q(x)} \, dx is 2πi2 \pi i times the sum of the residues of p/qp/q at all zeros of qq in the upper half plane.

After we sketched a proof, we applied this theorem to solve:

  1. x2(x2+1)(x2+4)dx\int_{-\infty}^\infty \dfrac{x^2}{(x^2+1)(x^2 + 4)} \, dx.

Another common type of problem that we can often solve using residues is R(x)sinxdx\int_{-\infty}^\infty R(x) \sin x \,dx or R(x)cosxdx\int_{-\infty}^\infty R(x) \cos x \, dx where RR is rational function that is continuous on the real axis. To evaluate these integrals, we can apply the Residue Theorem to either the real or imaginary part of f(z)=R(z)eizf(z) = R(z) e^{iz}. It helps to know the following inequality.

Jordan’s Lemma. Let γ(t)=Reit\gamma(t) = Re^{it}, with 0tπ0 \le t \le \pi. So γ\gamma parametrizes the upper half of a circle of radius RR around the origin. Then
|γeizdz|<π.\left| \int_\gamma e^{iz} \, dz \right| < \pi.

Unfortunately the ML-inequality is not strong enough to prove this theorem. So we need to dig a little deeper into the integral.

Step 1. Show that if z=γ(t)z = \gamma(t), then |eiz|=eRsint|e^{iz}| = e^{-R\sin t}.

Step 2. Show that |γeizdz|=|0πeRsintReiteReitdt|0πReRsintdt.|\int_\gamma e^{iz} \, dz| = \left|\int_0^\pi e^{-R\sin t} R e^{it} e^{Re^{it}} \, dt \right| \le \int_0^\pi R e^{-R \sin t} \, dt.

Step 3. Show that 2πtsint\tfrac{2}{\pi} t \le \sin t for all 0tπ20 \le t \le \tfrac{\pi}{2}.

Therefore 0πReRsintdt0π/22Re2Rπtdt=π(1eR)<π.\int_0^\pi R e^{-R \sin t} \, dt \le \int_0^{\pi/2} 2R e^{-\frac{2R}{\pi} t} \, dt = \pi(1-e^{-R}) < \pi.

  1. Compute xsinxx2+9dx\int_{-\infty}^\infty \dfrac{x \sin x}{x^2 + 9} \, dx.

Wed, Nov 20

We started with this real integral

  1. 02π1106cosθdθ\int_0^{2 \pi} \dfrac{1}{10 - 6 \cos \theta} \, d\theta.

To convert integrals of functions involving sine & cosine from 0 to 2π, you can use the following substitutions, all based on letting z=eiθz = e^{i \theta} and integrating over the unit circle:

02π1106cosθdθ=|z|=1(1103z3z1)(dziz)\int_0^{2\pi} \dfrac{1}{10 - 6 \cos \theta} \, d\theta = \oint_{|z|=1} \left(\dfrac{1}{10 - 3z - 3z^{-1}}\right) \left( \dfrac{dz}{iz} \right) =1i|z|=1110z3z23dz=\frac{1}{i} \oint_{|z|=1} \dfrac{1}{10z - 3z^2 - 3} \, dz The roots of 10z3z2+310z - 3z^2 + 3 are 3 and 1/31/3, and only the later is inside the unit disk, so we just need to find the residue at 1/31/3. 2πlimz13z1310z3z23=2πlimz131106z=π4.2\pi \lim_{z \rightarrow \tfrac{1}{3}} \frac{z - \tfrac{1}{3}}{10 z - 3z^2 - 3} = 2 \pi \lim_{z \rightarrow \tfrac{1}{3}} \frac{1}{10 - 6z} = \frac{\pi}{4}.

After that, we introduced harmonic functions which are functions u:2u:\mathbb{R}^2 \rightarrow \mathbb{R} with continuous partial derivatives such that 2ux2+2uy2=0.\frac{\partial^2 u}{\partial x^2} + \frac{\partial^2 u}{\partial y^2} = 0.

  1. Show that u(x,y)=x33xy2u(x,y) = x^3 - 3xy^2 is harmonic.

  2. Construct a real-valued function v(x,y)v(x,y) such that u+ivu+iv satisfies the Cauchy-Riemann equations.

Theorem (Harmonic Conjugates). If u:Du: D \rightarrow \mathbb{R} is harmonic on a simply connected open domain D2D \subseteq \mathbb{R}^2, then there exists a harmonic function v:Dv:D \rightarrow \mathbb{R} such that u+ivu+iv is holomorphic. The function vv is called a harmonic conjugate of uu. The converse is true as well, if ff is holomorphic, then both the real and imaginary parts of ff are harmonic functions.

  1. Show that u(x,y)=ln(x2+y2)u(x,y) = \ln(x^2 + y^2) is harmonic on its domain. Why can’t you find a harmonic conjugate of uu on all of \mathbb{C}?

A corollary of the previous theorem and the open mapping principle is this:

Maximum Modulus Principle for Harmonic Functions. A harmonic function on a open domain D2D \subseteq \mathbb{R}^2 cannot have a local max or local min in DD.

Fri, Nov 22

For a domain D2D \subseteq \mathbb{R}^2, we call a function F:D2F:D \rightarrow \mathbb{R}^2 a vector field on DD. If F(x,y)=(P(x,y),Q(x,y)),F(x,y) = (P(x,y), Q(x,y)), then the divergence and curl of FF are defined by divF=F=Px+Qy,\operatorname{div} F = \nabla \cdot F = \frac{\partial P}{\partial x} + \frac{\partial Q}{\partial y}, curlF=×F=QxPy.\operatorname{curl} F = \nabla \times F = \frac{\partial Q}{\partial x} - \frac{\partial P}{\partial y}.

Green’s Theorem (both versions). If F=(P,Q)F = (P, Q) is a vector field in an open simply connected domain in 2\mathbb{R}^2 that includes a simple, closed, piecewise smooth curve CC and let DD be the inside of CC. Then

CFdγ=DcurlFdA\oint_C F \cdot d \gamma = \iint_D \operatorname{curl} F \, dA CFndt=DdivFdA\oint_C F \cdot n \, dt = \iint_D \operatorname{div} F \, dA

where n=n(t)n = n(t) is the normal vector to γ(t)\gamma(t) at each point.

The div and curl versions of Green’s theorem explain the intuition that curl measures the net rotation of a flow (positive is counterclockwise, negative is clockwise), while divergence measures the net flux for the flow (inward is negative, outward is positive). When a vector field FF has curlF=0\operatorname{curl} F = 0 everywhere, we say it is irrotational. When divF=0\operatorname{div} F = 0 everywhere, we say that it is incompressible.

curlF=0\operatorname{curl} F = 0 curlF<0\operatorname{curl} F < 0
divF>0\operatorname{div} F > 0 divF=0\operatorname{div} F = 0

If DD is an open simply connected domain in \mathbb{C}, then any function f:Df: D \rightarrow \mathbb{C} can be thought of as a vector field. Furthermore, if ff is holomorphic on DD, then we know by Cauchy’s theorem that Cfdz=0.\oint_C f \, dz = 0. But keep in mind that CfdzCfdz.\oint_C f \, dz \ne \oint_C f \cdot dz. That’s because complex multiplication is not the same as the dot-product on 2\mathbb{R}^2.

Recall that if z=x+iyz = x + iy and w=u+ivw = u + iv, then Re(zw)=(x,y)(u,v).\operatorname{Re}(z \bar{w}) = (x,y) \cdot (u, v). Therefore, if we let F=fF = \bar{f}, then CFdz=Re(Cfdz)=0.\oint_C F \cdot dz = \operatorname{Re}\left( \oint_C f \, dz \right) = 0. Since this is true on every small simple closed curve CC in DD, it follows that the vector field FF has curlF=0\operatorname{curl} F = 0 everywhere.

The vector field F=fF = \bar{f} is known as the Polya field of a holomorphic function ff. It turns out that the Polya field also has divF=0\operatorname{div} F = 0 everywhere. To prove that, follow these exercises.

  1. If f=u+ivf = u + i v, show that Im(fdz)=vdx+udy\operatorname{Im}(f \, dz) = v\, dx + u \, dy.

  2. If n(t)n(t) is the normal vector for a path z(t)z(t), then ndt=idzn \, dt = -i \, dz.

  3. Show that Fndt=vdx+udvF \cdot n \, dt = v \, dx + u \, dv.

From these you can conclude that: CFndt=Im(Cfdz)=0.\oint_C F \cdot n \, dt = \operatorname{Im}\left( \oint_C f \, dz \right) = 0.

So the Polya field of a holomorphic function is always both incompressible and irrotational. We ended with one more exercise that we didn’t have time to finish.

  1. Show that if f=u+ivf = u + i v is holomorphic, then the Polya field for f(z)f'(z) is the gradient u\nabla u.

Week 14 Notes

Tentative Schedule

Day Section Topic
Mon, Nov 25 The heat equation
Wed, Nov 27 Thanksgiving break, no class
Fri, Nov 29 Thanksgiving break, no class

Mon, Nov 25

We began with this exercise that we didn’t have time to finish last time.

  1. Show that if f=u+ivf = u + i v is holomorphic, then the Polya field for f(z)f'(z) is the gradient u\nabla u. Hint, recall that f(z)=ux+ivx.f'(z) = \frac{\partial u}{\partial x} + i \frac{\partial v}{\partial x}.

Since f(z)f'(z) is holomorphic, its Polya field is both irrotational and incompressible. We talked about why it makes sense that the gradient of a harmonic function u(x,y)u(x,y) would have those two properties, in light of the maximum modulus principle for harmonic functions.

  1. Graph the harmonic function u=x33xy2u = x^3 - 3x y^2 on Desmos 3D. Based on the graph, what should the vector field u\nabla u look like?

Recall that a vector field FF on 2\mathbb{R}^2 is conservative if there is a potential function uu such that FF is the gradient of uu. The Polya fields of a holomorphic function are all conservative. There are lots of applications of conservative vector fields. There are lots of examples of conservative vector fields in nature.

Application Potential Function Tangent Curves Normal Curves
Fluid flow Pressure or elevation Streamlines Equipotential curves
Heat flow Temperature Lines of heat flux Isotherms
Electrostatic force Electric potential Field lines Equipotential curves

Then we looked at the 2-dimensional heat equation:

Tt=2Tx2+2Ty2.\frac{\partial T}{\partial t} = \frac{\partial^2 T}{\partial x^2} + \frac{\partial^2 T}{\partial y^2}.

This is a partial differential equation for modeling how the temperature TT at a point (x,y)(x,y) on a 2-dimensional object (like a thin metal plate) changes with respect to time tt. The solution would be a function T(t,x,y)T(t, x, y) that would tell us the temperature at any location and time. To understand where this equation comes from, consider the following:

The net flow of heat into a disk DD with boundary CC is equal to CTn(t)dt\oint_C \nabla T \cdot n(t) \, dt where n(t)=(y(t),x(t))n(t) = (y'(t), - x'(t)) is the (outward) normal vector and T\nabla T is the gradient of the temperature function (at a fixed time).

Note that the gradient points in the direction of greatest increase in temperature. If that is outwards (like in the image above), then it is hotter outside the disk DD, so heat tends to flow inwards (opposite the gradient vectors).

  1. Use the divergence form of Green’s theorem to show that the net flow of heat into DD is D2Tx2+2Ty2dA,\iint_D \dfrac{\partial^2 T}{\partial x^2} + \frac{\partial^2 T}{\partial y^2} \, dA, which explains why the heat equation makes sense as a model of heat flow.

Steady State Solutions

When the temperature reaches equilibrium, it stops changing so the left side of the heat equation (T/t\partial T/ \partial t) becomes zero. When that happens, T=T(x,y)T = T(x,y) is a harmonic function. We’ve talked about functions that are harmonic on the entire complex plane. But an important problem is to find functions that are harmonic on a domain DD \subset \mathbb{C} that satisfy a boundary condition. This is called the Dirichlet problem.

The Dirichlet Problem. Let D2D \subset \mathbb{R}^2 be a simply connected open domain with boundary C.C. Given a continuous real valued function gg defined on the boundary CC, find a continuous function uu on DCD \cup C such that uu is harmonic in DD and u=gu = g on C.C.


Week 15 Notes

Tentative Schedule

Day Section Topic
Mon, Dec 2 The Dirichlet problem
Wed, Dec 4 Fourier transform
Fri, Dec 6 The heat equation
Mon, Dec 9 Recap & review

Mon, Dec 2

We looked at one solution to the Dirichlet problem on the upper-half plane: T(z)=11πArgz. T(z) = 1 - \tfrac{1}{\pi} \operatorname{Arg} z.

Level curves for T
  1. Show that the function T(z)=11πArgzT(z) = 1-\tfrac{1}{\pi} \operatorname{Arg} z is harmonic. Hint: Is there a holomorphic function with T(z)T(z) as its real or imaginary part?

An important fact about harmonic functions and their conjugates is the following:

Theorem. If TT is a harmonic function and VV is its harmonic conjugate, then the level curves of TT are orthogonal to the level curves of VV.

  1. Prove this! Hint: it might help to assume that T=u(z)T = u(z) and V=v(z)V = v(z) are the real and imaginary parts of a holomorphic function. Then you can prove that their level curves are orthogonal by showing that their gradients are orthogonal.

Theorem. A harmonic function T:D1T: D_1 \rightarrow \mathbb{R} that is defined on one domain D1D_1 \subset \mathbb{C} can be turned into a harmonic function on another domain D2D_2 \subset \mathbb{C} if you can find an invertible holomorphic function f:D1D2f: D_1 \rightarrow D_2. In that case, Tf1T \circ f^{-1} is a harmonic function on D2D_2.

  1. Prove this. Hint: there is a holomorphic function gg such that T=RegT = \operatorname{Re}g.

  2. The Cayley transform is the Möbius transform f(z)=ziz+if(z) = \frac{z - i}{z+i} which maps the upper half-plane into the unit disk. If T(z)=Arg(z)T(z) = \operatorname{Arg}(z), then what do the level curves of Tf1T \circ f^{-1} look like? Hint: recall that Möbius transforms send lines and circles to lines and circles.

  1. Let f(z)=sinzf(z) = \sin z. This function maps the open vertical strip D1={w:π<Rew<π}D_1 = \{ w \in \mathbb{C}: -\pi < \operatorname{Re}w < \pi \} onto the set D2=([1,)(,1])D_2 = \mathbb{C}\backslash ([1,\infty) \cup (-\infty, -1]). If we have flow lines in D1D_1 parametrized by c+itc + it where π2<c<π2-\tfrac{\pi}{2} < c < \tfrac{\pi}{2}, then by the angle addition formula: w=sin(c+it)=sin(c)cos(it)+cos(c)sin(it).w = \sin(c+it) = \sin(c) \cos(it) + \cos(c) \sin(it). If we let w=u+ivw = u+iv, then it is not hard to see that u=sin(c)cos(it)u = \sin(c) \cos(it) and iv=cos(c)sin(it)iv = \cos(c) \sin(it). Then u2(sinc)2v2(cosc)2=cos2(it)+sin2(it)=1.\frac{u^2}{(\sin c)^2} - \frac{v^2}{(\cos c)^2} = \cos^2(it) + \sin^2(it) = 1. In particular the graph w(t)=u(t)+iv(t)=sin(c)cos(it)+cos(c)sin(it)w(t) = u(t) + iv(t) = \sin(c) \cos(it) + \cos(c) \sin(it) is a hyperbola in \mathbb{C}.
    These hyperbolas are the flow lines corresponding to a fluid flowing through a narrow opening in the real line.

Wed, Dec 4

Today we talked about the Fourier transform of a real function f:f:\mathbb{R}\rightarrow \mathbb{R} which is defined to be:

(f)=f(x)eiαxdx=F(α).\mathcal{F}(f) = \int_{-\infty}^\infty f(x) e^{-i\alpha x} \, dx = F(\alpha).

In class we used a contour integral to find the Fourier transform of

  1. f(x)=11+x2f(x) = \dfrac{1}{1+x^2}. Hint: There are two cases: if α>0\alpha > 0 and if α<0\alpha < 0.

We also calculated the Fourier transform of the Gaussian f(x)=ex2f(x) = e^{-x^2}. Before calculating that, we needed to know that ex2dx=π\int_{-\infty}^\infty e^{-x^2} \, dx = \sqrt{\pi}. Here is a video that explains this.

Then we used a rectangular contour to show that (ex2)=πeα2/4\mathcal{F}(e^{-x^2}) = \sqrt{\pi} e^{-\alpha^2/4}.

  1. Show that ex2eiαx=eα2/4exp([x+iα2]2)e^{-x^2} e^{-i\alpha x} = e^{-\alpha^2/ 4} \exp(-[x+\tfrac{i \alpha}{2}]^2).

  2. Use the contour below to show that in the limit as RR \rightarrow \infty, Cez2dz=πexp([x+iα2]2)dx=0,\oint_C e^{-z^2} \, dz = \sqrt{\pi} - \int_{-\infty}^\infty \exp(-[x+\tfrac{i \alpha}{2}]^2) \, dx = 0, so exp([x+iα2]2)dx=π.\int_{-\infty}^\infty \exp(-[x+\tfrac{i \alpha}{2}]^2) \, dx = \sqrt{\pi}.

R -R iα/2
  1. Conclude that (ex2)=πeα2/4\mathcal{F}(e^{-x^2}) = \sqrt{\pi} e^{-\alpha^2 / 4}.

We finished by talking about some of the properties of the Fourier transform.

  1. Show that \mathcal{F} is linear, i.e.,

    1. (f+g)=(f)+(g)\mathcal{F}(f+g) = \mathcal{F}(f) + \mathcal{F}(g), and
    2. (cf)=c(f)\mathcal{F}(cf) = c \mathcal{F}(f).
  2. Prove the translation property: If g(x)=f(x+c)g(x) = f(x+c), then (g)=eiαcF(α).\mathcal{F}(g) = e^{i \alpha c} F(\alpha).

  3. Prove the derivative property: (f)=iαF(α).\mathcal{F}(f') = i \alpha F(\alpha). Hint: Use integration by parts and use the fact that limx±f(x)=0\lim_{x \rightarrow \pm \infty} f(x) = 0 since |f||f| must have a finite integral.

Fri, Dec 6

Today we used the Fourier transform to solve the 1-dimensional heat equation: Tt=2T2x\frac{\partial T}{\partial t} = \frac{\partial^2 T}{\partial^2 x} with initial condition T(x,0)=f(x)T(x,0) = f(x). The goal is to find an equation T=T(x,t)T = T(x,t) that can predict the temperature at any location xx and time tt on a thin rod.

We started by defining the convolution of two functions f,g:f, g : \mathbb{R}\rightarrow \mathbb{R} as (f*g)(x)=f(y)g(xy)dy.(f \ast g) (x) = \int_{-\infty}^\infty f(y) g(x-y) \, dy.

Convolutions play very nicely with the Fourier transform.

  1. Prove the convolution property of the Fourier transform. If (f)=F(α)\mathcal{F}(f) = F(\alpha) and (g)=G(α)\mathcal{F}(g) = G(\alpha), then (f*g)=F(α)G(α).\mathcal{F}(f \ast g) = F(\alpha) G(\alpha). Hint: You can prove this by swapping the order of integration and applying a u-substitution.

We need one more property of the Fourier transform before solving the heat equation.

  1. Prove the scaling poperty: For any c>0c>0 (f(xc))=cF(cα).\mathcal{F}\left(f\left(\frac{x}{c}\right)\right) = c F(c \alpha).

Then we used these steps to solve the heat equation. Before we did that, we briefly talked about why the 1-dimensional heat equation makes sense.

In some cases you can calculate the value of this convolution by hand, but usually it is easier to use a computer to calculate it numerically. Here is a graph of an example solution of the heat equation on Desmos (click on the image below):

  1. See if you can use Desmos to graph the solution to the heat equation if the initial condition is f(x)=max(2|x|,0).f(x) = \max(2 - |x|, 0).

Mon, Dec 9

We went over some of the questions from the final exam. In particular, we talked about how to find the Fourier transform of a function that is only non-zero in a finite interval like the one in the exam. We did this example:

  1. Find the Fourier transform of the function f(x)={0 if |x|>1x+1 if 1<x<01x if 0<x<1f(x) = \begin{cases} 0 & \text{ if } |x| > 1 \\ x+1 & \text{ if } -1 < x < 0 \\ 1-x & \text{ if } 0 < x < 1 \end{cases} One issue with this example is that it requires integration by parts, so it is a little tedious to calculate without tabular integration. We didn’t finish the computation, but we got the idea across. Another issue is that this function is continuous, unlike the example on the final exam, but the continuity isn’t that important. As long as the discontinuities only occur at a finite number of points, you can ignore those points when calculating the integrals.

We also looked at what the solution of the heat equation would look like with this function as the initial condition. Surprisingly we don’t really need to know the Fourier transform of the initial condition since it goes away in the solution when we apply the inverse Fourier transform to get our final convolution equation for the solution.

We also talked about how to find a Laurent series.

  1. Let f(z)=sinzz5f(z) = \dfrac{\sin z}{z - 5}. Find a Laurent series for ff centered at 0 which converges when |z|>5|z| > 5. The solution is to find infinite series for sinz\sin z and 1z5\dfrac{1}{z-5} separately and then multiply the Laurent series for 1z5\dfrac{1}{z-5} with the MacLaurin series for sinz\sin z.