1.1 The Complex Plane and Polynomials
The complex numbers as a field and as a plane: modulus and argument, De Moivre's theorem, the roots of unity, circles and lines, the Riemann sphere, and what the coefficients of a polynomial say about its roots.
The complex numbers were invented to solve equations. Renaissance algebraists found that Cardano's formula for the roots of a real cubic sometimes passes through square roots of negative numbers even when all three roots are real, and the only way to get the right answer was to calculate with $\sqrt{-1}$ as if it were a number. Three centuries later the reward was clear: in $\mathbb{C}$ every nonconstant polynomial has a root (the fundamental theorem of algebra), so $\mathbb{C}$ is the natural home of algebra. At the same time $\mathbb{C}$ is a plane, and multiplication by a complex number is a rotation followed by a scaling. That double nature — a field and a plane at once — is what makes complex analysis so powerful, and it is the subject of this section.
Everything here is used constantly in the rest of the course: polar form and roots of unity reappear in residue computations (Chapter 3), in Möbius maps (Chapter 4), in cyclic groups (Chapter 6) and in cyclotomic fields (Chapter 8). CSIR NET tests the material directly too. Part B likes the number of solutions of an equation such as $z^5=\bar z$, the value of a sum or product over roots of unity, or the centre and radius of a locus; Part C likes statements about where the roots of a polynomial can lie and which "obvious" identities for $\operatorname{Arg}$ fail. The ⚠️ box collects the traps.
The Field of Complex Numbers
A complex number is an ordered pair $z=(x,y)$ of real numbers. The set of all of them is $\mathbb{C}=\mathbb{R}^2$ with the operations
$$(x,y)+(u,v)=(x+u,\ y+v),$$ $$(x,y)\cdot(u,v)=(xu-yv,\ xv+yu).$$We identify the real number $x$ with $(x,0)$ and write $i=(0,1)$. Then $i^2=(0,1)(0,1)=(-1,0)=-1$ and every complex number is $z=x+iy$. We call $x=\operatorname{Re}z$ the real part and $y=\operatorname{Im}z$ the imaginary part (a real number!). Numbers with $\operatorname{Re}z=0$ are purely imaginary.
Proof. (a) Addition is that of $\mathbb{R}^2$, so it is associative and commutative with identity $0=(0,0)$ and inverses $-(x,y)=(-x,-y)$. Multiplication is commutative because the formula is symmetric in the two factors; associativity and distributivity are routine expansions (each side of $(z_1z_2)z_3=z_1(z_2z_3)$ has real part $x_1x_2x_3-x_1y_2y_3-y_1x_2y_3-y_1y_2x_3$, and similarly for the imaginary part). The identity is $1=(1,0)$. If $z\ne0$ then $x^2+y^2\gt0$, and $$(x+iy)\cdot\frac{x-iy}{x^2+y^2}=\frac{x^2+y^2+i(-xy+yx)}{x^2+y^2}=1.$$ Finally $(x,0)+(u,0)=(x+u,0)$ and $(x,0)(u,0)=(xu,0)$, so the embedding respects both operations.
(b) In an ordered field every square is $\ge0$: if $a\ge0$ then $a^2\ge0$, and if $a\lt0$ then $-a\gt0$ and $a^2=(-a)^2\gt0$. Also $1=1^2\gt0$. In $\mathbb{C}$ we would get $-1=i^2\ge0$, and adding $1\gt0$ gives $0\gt0$ — a contradiction. ∎
Two further descriptions of $\mathbb{C}$ are worth knowing. The matrices $\begin{pmatrix}a&-b\\ b&a\end{pmatrix}$ form a field isomorphic to $\mathbb{C}$ (the matrix of the $\mathbb{R}$-linear map $w\mapsto(a+ib)w$ in the basis $1,i$), whose determinant $a^2+b^2$ is the square of the modulus defined next. And $\mathbb{C}\cong\mathbb{R}[x]/(x^2+1)$, the quotient of a polynomial ring by an irreducible polynomial — the model for every field extension in Chapter 8.
- $\overline{z+w}=\bar z+\bar w$, $\ \overline{zw}=\bar z\,\bar w$, $\ \overline{\bar z}=z$; $\ \operatorname{Re}z=\tfrac{z+\bar z}{2}$, $\ \operatorname{Im}z=\tfrac{z-\bar z}{2i}$; $z$ is real iff $z=\bar z$.
- $|z|^2=z\bar z$, $\ |\bar z|=|z|$, $\ |zw|=|z|\,|w|$, and $|z/w|=|z|/|w|$ for $w\ne0$.
- $|\operatorname{Re}z|\le|z|$ and $|\operatorname{Im}z|\le|z|$.
- Triangle inequality: $|z+w|\le|z|+|w|$, with equality iff one of $z,w$ is a non-negative real multiple of the other.
- Reverse triangle inequality: $\big||z|-|w|\big|\le|z-w|$.
- Parallelogram law: $|z+w|^2+|z-w|^2=2|z|^2+2|w|^2$.
Proof. (1) and the first part of (2) are direct computations; for instance $\overline{(x+iy)(u+iv)}=(xu-yv)-i(xv+yu)$, which is $(x-iy)(u-iv)$; and $z\bar z=x^2+y^2$. Then $|zw|^2=zw\,\overline{zw}=z\bar z\,w\bar w=|z|^2|w|^2$; take square roots. (3) $|x|=\sqrt{x^2}\le\sqrt{x^2+y^2}$.
(4) Expand: $|z+w|^2=(z+w)(\bar z+\bar w)$, which is $|z|^2+|w|^2+z\bar w+\bar z w$, that is, $|z|^2+|w|^2+2\operatorname{Re}(z\bar w)$. By (3) and (2), $\operatorname{Re}(z\bar w)\le|z\bar w|=|z||w|$, so $$|z+w|^2\le(|z|+|w|)^2.$$ Equality holds iff $\operatorname{Re}(z\bar w)=|z\bar w|$, i.e. iff $z\bar w$ is a non-negative real number. If $w\ne0$ this says $z=(z\bar w)/|w|^2\cdot w$ is a non-negative multiple of $w$; if $w=0$ the statement holds trivially.
(5) $|z|=|(z-w)+w|\le|z-w|+|w|$, so $|z|-|w|\le|z-w|$; exchange $z$ and $w$. (6) Add the expansion of $|z+w|^2$ in (4) to the same expansion with $w$ replaced by $-w$; the cross terms cancel. ∎
In words, (6) says that the sum of the squares of the diagonals of a parallelogram equals the sum of the squares of its four sides. The next inequality is the algebraic heart of the Schwarz lemma and of the automorphisms of the unit disc (§2.3 and §4.1).
Proof. Using $|u-v|^2=|u|^2+|v|^2-2\operatorname{Re}(u\bar v)$: $|1-\bar az|^2=1+|a|^2|z|^2-2\operatorname{Re}(\bar az)$ and $|z-a|^2=|z|^2+|a|^2-2\operatorname{Re}(z\bar a)$. The real parts are equal, so the difference is $1+|a|^2|z|^2-|z|^2-|a|^2$, which factors as $(1-|a|^2)(1-|z|^2)$. If $|a|\lt1$ and $|z|\le1$ then $|\bar az|\lt1$, so $1-\bar az\ne0$ and $\varphi_a(z)$ is defined. The right-hand side is $\gt0$ when $|z|\lt1$ and $=0$ when $|z|=1$, which gives $|z-a|\lt|1-\bar az|$, respectively $|z-a|=|1-\bar az|$. ∎
Polar Form, De Moivre's Theorem and Roots of Unity
A point $z\ne0$ of the plane is also described by its distance $r=|z|$ from the origin and the angle $\theta$ that the ray from $0$ to $z$ makes with the positive real axis. The angle is determined only up to multiples of $2\pi$, and keeping track of this ambiguity is the source of most errors with arguments and, in §1.2, with logarithms.
For $z\ne0$, a real number $\theta$ is an argument of $z$ if $z=|z|(\cos\theta+i\sin\theta)$. The set of all arguments is $\arg z=\{\theta_0+2\pi k:k\in\mathbb{Z}\}$ for any one argument $\theta_0$. The principal argument $\operatorname{Arg}z$ is the unique argument in $(-\pi,\pi]$. We write
$$e^{i\theta}=\cos\theta+i\sin\theta\qquad(\theta\in\mathbb{R}),$$so that $z=re^{i\theta}$ with $r=|z|$ (the polar form). The notation is justified in §1.2, where $e^{w}$ is defined for all $w\in\mathbb{C}$ by its power series. The number $0$ has no argument.
For $z=x+iy$: $\operatorname{Arg}z=\arctan(y/x)$ if $x\gt0$; $\arctan(y/x)+\pi$ if $x\lt0\le y$; $\arctan(y/x)-\pi$ if $x\lt0$, $y\lt0$; and $\pm\pi/2$ if $x=0$, $y\gtrless0$. So $\operatorname{Arg}(-1)=\pi$, $\operatorname{Arg}(-1-i)=-3\pi/4$ and $\operatorname{Arg}(-\sqrt3+i)=5\pi/6$. Always locate the quadrant before using a calculator's $\arctan$.
Proof. Multiplying out, $(\cos\theta+i\sin\theta)(\cos\phi+i\sin\phi)$ has real part $\cos\theta\cos\phi-\sin\theta\sin\phi=\cos(\theta+\phi)$ and imaginary part $\sin\theta\cos\phi+\cos\theta\sin\phi=\sin(\theta+\phi)$, by the addition formulas of trigonometry. The formulas for $zw$ follow, and $z/w$ is the number whose product with $w$ is $z$, namely $\tfrac rs e^{i(\theta-\phi)}$. Every argument of $zw$ has the form $\theta+\phi+2\pi k$, and every such number is an argument of $zw$, which is the set statement. De Moivre: for $n\ge0$ induct, $e^{i(n+1)\theta}=e^{in\theta}e^{i\theta}$; for $n=-m\lt0$, $\left(e^{i\theta}\right)^{-m}=1/e^{im\theta}=e^{-im\theta}$ because $e^{im\theta}e^{-im\theta}=e^{0}=1$. ∎
Geometrically: multiplication by $w=se^{i\phi}$ rotates the plane through the angle $\phi$ and scales it by the factor $s$. Multiplication by $i$ is a quarter-turn. The set identity $\arg(zw)=\arg z+\arg w$ does not give $\operatorname{Arg}(zw)=\operatorname{Arg}z+\operatorname{Arg}w$: for $z=w=-1+i$, $\operatorname{Arg}z+\operatorname{Arg}w=3\pi/2$, whereas $zw=-2i$ has $\operatorname{Arg}(zw)=-\pi/2$. The two differ by $2\pi$, which is all that the set identity promises.
Proof. Write $z=re^{i\theta}$ with $r\gt0$ (note $z=0$ is not a solution). By De Moivre, $z^n=r^ne^{in\theta}$, and two nonzero numbers are equal iff their moduli are equal and their arguments differ by a multiple of $2\pi$. So $z^n=w$ iff $r^n=|w|$ and $n\theta=\operatorname{Arg}w+2\pi k$ for some $k\in\mathbb{Z}$, i.e. iff $r=|w|^{1/n}$ and $\theta=(\operatorname{Arg}w+2\pi k)/n$. Two values of $k$ give the same $z$ iff the angles differ by a multiple of $2\pi$, i.e. iff the values of $k$ differ by a multiple of $n$. Hence $k=0,\dots,n-1$ give all solutions, each once. Consecutive solutions differ by the angle $2\pi/n$ and have the same modulus. ∎
- $U_n$ is a group under multiplication, generated by $\omega$. The root $\omega^k$ has order $n/\gcd(k,n)$; it is primitive iff $\gcd(k,n)=1$, so there are exactly $\varphi(n)$ primitive $n$-th roots of unity ($\varphi$ is Euler's function, §5.3).
- $z^n-1=\displaystyle\prod_{k=0}^{n-1}(z-\omega^k)$. Consequently $\displaystyle\sum_{k=0}^{n-1}\omega^k=0$ and $\displaystyle\prod_{k=0}^{n-1}\omega^k=(-1)^{n-1}$.
- For every integer $j$: $\displaystyle\sum_{k=0}^{n-1}\omega^{jk}=\begin{cases}n,&n\mid j,\\ 0,&n\nmid j.\end{cases}$
- $\displaystyle\prod_{k=1}^{n-1}\left(1-\omega^k\right)=n$, and therefore $\displaystyle\prod_{k=1}^{n-1}\sin\frac{k\pi}{n}=\frac{n}{2^{n-1}}$.
Proof. (1) Products and inverses of $n$-th roots of unity are $n$-th roots of unity ($(\zeta\eta)^n=\zeta^n\eta^n=1$, $(\zeta^{-1})^n=1$), and every element is a power of $\omega$. Let $g=\gcd(k,n)$. Then $(\omega^k)^m=1\iff n\mid km$, and $n\mid km\iff\tfrac ng\mid\tfrac kg m\iff\tfrac ng\mid m$, the last step because $\tfrac ng$ and $\tfrac kg$ are coprime. So the order of $\omega^k$ is $n/g$, which equals $n$ iff $g=1$; the number of $k\in\{0,\dots,n-1\}$ coprime to $n$ is $\varphi(n)$ by definition.
(2) The $n$ distinct numbers $\omega^k$ are roots of the monic polynomial $z^n-1$ of degree $n$; by the factor theorem (Theorem 1.11 below) $z^n-1$ is divisible by $\prod(z-\omega^k)$, a monic polynomial of the same degree, so the two are equal. Compare coefficients (Vieta, Theorem 1.13): the coefficient of $z^{n-1}$ is $-\sum\omega^k$ on the right and $0$ on the left ($n\ge2$); the constant term is $(-1)^n\prod\omega^k$ on the right and $-1$ on the left.
(3) If $n\mid j$, every term $\omega^{jk}=(\omega^n)^{(j/n)k}=1$ and the sum is $n$. Otherwise $q=\omega^j\ne1$ but $q^n=(\omega^n)^j=1$, and the geometric sum is $\sum_{k=0}^{n-1}q^k=\dfrac{q^n-1}{q-1}=0$.
(4) Divide (2) by $z-1$: for $z\ne1$, $$1+z+\cdots+z^{n-1}=\prod_{k=1}^{n-1}(z-\omega^k);$$ both sides are polynomials, so the identity holds at $z=1$ too, giving $n=\prod_{k=1}^{n-1}(1-\omega^k)$. Now $1-e^{i\alpha}=e^{i\alpha/2}\left(e^{-i\alpha/2}-e^{i\alpha/2}\right)$, which equals $-2i\,e^{i\alpha/2}\sin\tfrac\alpha2$, so $|1-\omega^k|=2\sin\tfrac{k\pi}{n}$ for $1\le k\le n-1$ (the sine is positive there). Taking moduli, $n=2^{n-1}\prod_{k=1}^{n-1}\sin\tfrac{k\pi}{n}$. ∎
Part (3) is the discrete form of the orthogonality of $e^{ij\theta}$ on $[0,2\pi]$; it is the engine of the discrete Fourier transform and of the "roots-of-unity filter" in Exercise II.6. Taking real parts in (2) gives, e.g., $\cos\tfrac{2\pi}{7}+\cos\tfrac{4\pi}{7}+\cos\tfrac{6\pi}{7}=-\tfrac12$ (Example 4).
- Sums of powers $\sum\omega^{jk}$: geometric series — the answer is $n$ or $0$ according as $n\mid j$ or not.
- Products $\prod(a-\omega^k)$: evaluate the factorisation $z^n-1=\prod(z-\omega^k)$ at $z=a$, or $1+z+\dots+z^{n-1}=\prod_{k\ge1}(z-\omega^k)$ when the factor $k=0$ is absent.
- Sums of $f(\omega^k)$ with $f$ a polynomial: expand and apply step 1 term by term (this picks out the coefficients of $f$ whose index is divisible by $n$).
- Conjugate pairs: $\overline{\omega^k}=\omega^{n-k}=\omega^{-k}$; pairing $k$ with $n-k$ makes sums real.
- Counting solutions of equations like $z^m=\bar z^{\,p}$: take moduli first (this fixes $|z|$), then use $\bar z=1/z$ on the unit circle.
Geometry in the Complex Plane
Because $|z-w|$ is the Euclidean distance, plane geometry can be written in complex numbers, and the algebra is often shorter than with coordinates. Two facts do most of the work. First, $\dfrac{z_3-z_1}{z_2-z_1}$ has modulus $\dfrac{|z_3-z_1|}{|z_2-z_1|}$ and argument equal to the angle from the vector $z_2-z_1$ to the vector $z_3-z_1$; so $z_1,z_2,z_3$ are collinear iff this ratio is real, and the angle at $z_1$ is a right angle iff it is purely imaginary. Second, every line and every circle has an equation in $z$ and $\bar z$.
Proof. Note $\bar\beta z+\beta\bar z=2\operatorname{Re}(\bar\beta z)$ is real, so the equation is a real equation. (a) With $\beta=p+iq$ and $z=x+iy$, $2\operatorname{Re}(\bar\beta z)=2(px+qy)$, and $\Gamma$ is the line $2px+2qy+C=0$, whose normal vector is $(p,q)$, i.e. $\beta$. (b) Divide by $A$ and complete the square: with $c=-\beta/A$, $$z\bar z-\bar cz-c\bar z=(z-c)\overline{(z-c)}-|c|^2,$$ so the equation becomes $|z-c|^2=|c|^2-\tfrac CA=\tfrac{|\beta|^2-AC}{A^2}$. This is a circle of the stated radius when the right side is positive, the single point $c$ when it is zero, and empty when it is negative. Conversely a line $px+qy=d$ is case (a) with $\beta=\tfrac{p+iq}2$, $C=-d$, and a circle $|z-a|=r$ expands as stated. ∎
Lines and circles are therefore one family — "generalised circles" — distinguished only by whether the coefficient of $z\bar z$ vanishes. On the Riemann sphere below they become literally the same kind of object, and Möbius maps (§4.1) permute them.
- the perpendicular bisector of the segment $[a,b]$ if $k=1$;
- if $k\ne1$, the circle with $$\text{centre } c=\frac{a-k^2b}{1-k^2},\qquad \text{radius } r=\frac{k\,|a-b|}{|1-k^2|}.$$ The points $a$ and $b$ lie on one ray from $c$ with $|a-c|\cdot|b-c|=r^2$: they are inverse (symmetric) points with respect to the circle.
Proof. Square: $|z-a|^2=k^2|z-b|^2$, i.e. $$\begin{aligned}&(1-k^2)|z|^2-2\operatorname{Re}\big(z\,\overline{(a-k^2b)}\big)\\ &\qquad+|a|^2-k^2|b|^2=0.\end{aligned}$$ If $k=1$ the $|z|^2$ terms cancel and we get a line (Theorem 1.7(a)); it is the set of points equidistant from $a$ and $b$, the perpendicular bisector. If $k\ne1$ divide by $1-k^2$: by Theorem 1.7(b) this is a circle with centre $c=\frac{a-k^2b}{1-k^2}$ and $$r^2=|c|^2-\frac{|a|^2-k^2|b|^2}{1-k^2}=\frac{|a-k^2b|^2-(1-k^2)(|a|^2-k^2|b|^2)}{(1-k^2)^2}.$$ Expanding the numerator, the terms $|a|^2$ and $k^4|b|^2$ cancel and what remains is $k^2\left(|a|^2-2\operatorname{Re}(a\bar b)+|b|^2\right)=k^2|a-b|^2$, which gives $r$. Finally $a-c=\frac{k^2(b-a)}{1-k^2}$ and $b-c=\frac{b-a}{1-k^2}$; their ratio is the positive number $k^2$, so $a,b$ lie on the same ray from $c$, and $|a-c||b-c|=\frac{k^2|a-b|^2}{(1-k^2)^2}=r^2$. ∎
As $k$ runs through $(0,\infty)$ these circles fill the plane minus $\{a,b\}$: small $k$ gives small circles around $a$, large $k$ small circles around $b$, and $k=1$ the separating line. This family reappears as the level curves of $\left|\frac{z-a}{z-b}\right|$ — the image of concentric circles under a Möbius map.
The Extended Plane and Stereographic Projection
Many statements of complex analysis become cleaner if we add one point at infinity: $1/z$ is then defined at $0$, a polynomial of degree $n$ has a "pole" at $\infty$, and lines are circles through $\infty$. A sphere makes this concrete.
Proof. The points of the line through $N$ and $(x,y,0)$ are $(tx,\,ty,\,1-t)$, $t\in\mathbb{R}$. Such a point lies on $S$ iff $t^2|z|^2+(1-t)^2=1$, i.e. $t\left(t(|z|^2+1)-2\right)=0$. The root $t=0$ is $N$; the other, $t=2/(|z|^2+1)$, gives $P(z)$ as stated. Conversely, if $P=(x_1,x_2,x_3)\ne N$ then $x_3\lt1$ and the line from $N$ through $P$ meets $x_3=0$ where $1-t\,(1-x_3)=0$, at $z=\frac{x_1+ix_2}{1-x_3}$. The two constructions are inverse to each other, which gives the bijection. From the formula, $x_3=\frac{|z|^2-1}{|z|^2+1}$ is negative, zero or positive according as $|z|\lt1$, $=1$ or $\gt1$.
For the chordal distance, $|P(z)|=|P(w)|=1$, so $|P(z)-P(w)|^2=2-2P(z)\cdot P(w)$. With $D=(1+|z|^2)(1+|w|^2)$ and $E=4\operatorname{Re}(z\bar w)+(|z|^2-1)(|w|^2-1)$ we have $P(z)\cdot P(w)=E/D$, and expanding, $$2D-2E=4|z|^2+4|w|^2-8\operatorname{Re}(z\bar w)=4|z-w|^2.$$ Hence $\chi(z,w)^2=4|z-w|^2/D$. Letting $|w|\to\infty$ gives $\chi(z,\infty)$, the distance from $P(z)$ to $N$. Antipodal means $P(w)=-P(z)$, i.e. $\chi(z,w)=2$; with $w=-1/\bar z$ one checks $P(w)=-P(z)$ directly from the formula (numerator $\left(-2x,-2y,1-|z|^2\right)/|z|^2$, denominator $(1+|z|^2)/|z|^2$), and the antipode is unique. ∎
Proof. A circle on $S$ is the intersection of $S$ with a plane $a_1x_1+a_2x_2+a_3x_3=a_0$ (with $a_1^2+a_2^2+a_3^2\gt a_0^2$, so that the plane cuts the sphere in more than a point). Substituting the formula for $P(z)$ and multiplying by $|z|^2+1$: $$(a_3-a_0)|z|^2+2a_1x+2a_2y-(a_3+a_0)=0.$$ This has the form of Theorem 1.7 with $A=a_3-a_0$, and it is a line exactly when $a_3=a_0$, i.e. when $N=(0,0,1)$ lies on the plane. Conversely, given a circle or line $A|z|^2+Bx+Cy+D=0$ (real coefficients), the plane with $a_1=B/2$, $a_2=C/2$, $a_3=(A-D)/2$, $a_0=-(A+D)/2$ meets $S$ exactly in the points corresponding to the curve; since the curve has more than one point, so does the intersection, which is therefore a circle. ∎
Stereographic projection also preserves angles between curves (it is conformal); together with Theorem 1.10 this is why the sphere is the natural stage for Möbius transformations in Chapter 4.
Polynomials: Roots, Coefficients and Where the Roots Lie
Proof. (a) Existence by induction on $\deg f$: if $\deg f\ge\deg g=d$, subtract $\frac{a}{b}z^{\deg f-d}g$ ($a,b$ the leading coefficients) to lower the degree. Uniqueness: if $qg+r=q'g+r'$ then $(q-q')g=r'-r$, and the left side has degree $\ge d$ unless $q=q'$. With $g=z-\alpha$ the remainder is a constant $r$, and evaluating at $\alpha$ gives $r=p(\alpha)$. (b) If $\alpha_1,\dots,\alpha_k$ are distinct roots with multiplicities $m_j$, then $\prod(z-\alpha_j)^{m_j}$ divides $p$ (the factors for different $j$ have no common root, and we divide them out one at a time using (a)), so $\sum m_j\le n$. (c) If $p=(z-\alpha)^mq$ then by Leibniz' rule $p^{(k)}(\alpha)=0$ for $k\lt m$ and $p^{(m)}(\alpha)=m!\,q(\alpha)$. Conversely, Taylor's formula for polynomials, $p(z)=\sum_k\frac{p^{(k)}(\alpha)}{k!}(z-\alpha)^k$ (expand $z=(z-\alpha)+\alpha$), shows that the stated vanishing makes $(z-\alpha)^m$ a factor with $q(\alpha)=p^{(m)}(\alpha)/m!\ne0$. ∎
About the proof. The complete proof is given in §2.3 as a two-line consequence of Liouville's theorem: if $p$ had no root, $1/p$ would be a bounded entire function, hence constant. The "consequently" part follows from Theorem 1.11 by induction: a root $z_1$ gives $p=(z-z_1)q$ with $\deg q=n-1$, and we repeat. Here is the geometric reason why a root must exist, which the argument principle (§3.2) makes rigorous. Let $p$ be monic with $p(0)\ne0$. On a huge circle $|z|=R$, $p(z)\approx z^n$, so as $z$ goes once around the circle the point $p(z)$ winds $n$ times around $0$. On a tiny circle $|z|=\varepsilon$, $p(z)\approx p(0)$ stays near a nonzero point and winds $0$ times. As the radius grows from $\varepsilon$ to $R$ the closed curve $p(\{|z|=\rho\})$ deforms continuously, and its winding number can change only when the curve passes through $0$ — that is, at a root.
The theorem fails over $\mathbb{R}$ ($x^2+1$) and over $\mathbb{Q}$ ($x^2-2$): $\mathbb{C}$ is algebraically closed, while $\mathbb{R}$ and $\mathbb{Q}$ are not.
Proof. Expanding $\prod_{j}(z-z_j)$, a term $z^{n-k}$ arises by choosing $-z_j$ from exactly $k$ of the factors and $z$ from the rest; summing over the choices gives the coefficient $(-1)^ke_k$. Multiply by $a_n$ and compare with $p$. For Newton's identities with $n=3$ (the general case is the same computation): $s_1=e_1$; $e_1^2=s_2+2e_2$ gives $s_2$; and each root satisfies $z_j^3=e_1z_j^2-e_2z_j+e_3$ (it is a root of $z^3-e_1z^2+e_2z-e_3$), so summing over $j$ gives $s_3=e_1s_2-e_2s_1+3e_3$. For general $n$, write $T=\sum_{j\ne k}z_jz_k^2$ (ordered pairs of distinct indices). Then $e_1s_2=s_3+T$, while in $e_2s_1=\sum_{j\lt l}z_jz_l\sum_kz_k$ the terms with $k\in\{j,l\}$ give $T$ and those with $k\notin\{j,l\}$ give every product of three distinct roots three times; so $e_2s_1=T+3e_3$, and $e_1s_2-e_2s_1=s_3-3e_3$. ∎
Proof. Since $\bar a_k=a_k$, Theorem 1.2(1) gives $\overline{p(z)}=\sum\bar a_k\bar z^k=p(\bar z)$; so $p(\alpha)=0\iff p(\bar\alpha)=0$. If $p=(z-\alpha)^mq$ with $q(\alpha)\ne0$, conjugating all coefficients gives $p=\bar p=(z-\bar\alpha)^m\bar q$ with $\bar q(\bar\alpha)=\overline{q(\alpha)}\ne0$, so $\bar\alpha$ has the same multiplicity $m$. For non-real $\alpha$, $(z-\alpha)(z-\bar\alpha)=z^2-2(\operatorname{Re}\alpha)z+|\alpha|^2$ has real coefficients and negative discriminant $4(\operatorname{Re}\alpha)^2-4|\alpha|^2=-4(\operatorname{Im}\alpha)^2$. Grouping the factorisation of Theorem 1.12 into such pairs gives the real factorisation. The non-real roots contribute an even number to the degree, so odd degree forces a real root. ∎
The same argument works for any field automorphism: if $p$ has rational coefficients and $a+b\sqrt d$ is a root ($a,b\in\mathbb{Q}$, $\sqrt d\notin\mathbb{Q}$), then so is $a-b\sqrt d$ — a first glimpse of Galois theory (Chapter 8).
Where can the roots of a polynomial lie? Locating them without solving is a constant theme — Rouché's theorem (§3.2) counts roots in a disc — and the next three results are the elementary tools.
Proof. If $M=0$ then $p=z^n$ and the only root is $0$. Otherwise let $|z|\ge1+M$, so $|z|\gt1$. By the triangle inequality and the geometric sum, $$\left|a_{n-1}z^{n-1}+\cdots+a_0\right|\le M\left(|z|^{n-1}+\cdots+1\right)=M\,\frac{|z|^n-1}{|z|-1}\lt\frac{M}{|z|-1}\,|z|^n\le|z|^n,$$ using $|z|-1\ge M$ in the last step. Hence $|p(z)|\ge|z|^n-|a_{n-1}z^{n-1}+\cdots+a_0|\gt0$, and $z$ is not a root. ∎
Proof. Multiply by $1-z$: $$\begin{aligned}(1-z)p(z)&=a_0+(a_1-a_0)z+\cdots\\&\quad+(a_n-a_{n-1})z^n-a_nz^{n+1}.\end{aligned}$$ Let $|z|\gt1$. The coefficients $a_0$ and $a_k-a_{k-1}$ are $\ge0$, $|z|^k\le|z|^n$ for $k\le n$, and $a_0|z|^0\lt a_0|z|^n$ because $a_0\gt0$. Hence $$\left|a_0+\sum_{k=1}^n(a_k-a_{k-1})z^k\right|\lt|z|^n\Big(a_0+\sum_{k=1}^n(a_k-a_{k-1})\Big)=a_n|z|^n,$$ and so $|(1-z)p(z)|\gt a_n|z|^{n+1}-a_n|z|^n$, that is, $|(1-z)p(z)|\gt a_n|z|^n(|z|-1)\gt0$. Thus $p(z)\ne0$. For the second statement apply the first to $z^np(1/z)=a_n+a_{n-1}z+\cdots+a_0z^n$, whose roots are the reciprocals of those of $p$ (note $p(0)=a_0\ne0$). ∎
Proof. Write $p=a_n\prod_{j}(z-z_j)$. Logarithmic differentiation (the product rule for each factor) gives, wherever $p(z)\ne0$, $$\frac{p'(z)}{p(z)}=\sum_{j=1}^n\frac{1}{z-z_j}=\sum_{j=1}^n\frac{\bar z-\bar z_j}{|z-z_j|^2}.$$ Let $p'(w)=0$. If $p(w)=0$ then $w$ is one of the $z_j$ and lies in the hull. Otherwise the sum vanishes at $w$; conjugating, $\sum_j\lambda_j(w-z_j)=0$ with $\lambda_j=|w-z_j|^{-2}\gt0$. Hence $$w=\sum_j\frac{\lambda_j}{\Lambda}\,z_j,\qquad\Lambda=\sum_j\lambda_j,$$ a convex combination of the roots (positive weights summing to $1$). ∎
For a real polynomial all of whose roots are real, the convex hull is an interval of the real axis — so all roots of $p'$ are real too. This is a complex-variable proof of a fact usually obtained from Rolle's theorem. But Rolle's theorem itself has no complex analogue: $p(z)=z^3-1$ vanishes at $1$ and at $\omega=e^{2\pi i/3}$, yet $p'(z)=3z^2$ vanishes only at $0$, which is not on the segment $[1,\omega]$.
- $\mathbb{C}$ is not ordered. "$z\lt w$" is meaningless for non-real numbers; compare moduli, real parts or imaginary parts instead (Theorem 1.1(b)).
- $\operatorname{Arg}$ is not additive. $\operatorname{Arg}\big((-1+i)^2\big)=-\pi/2\ne3\pi/2$; only the set identity $\arg(zw)=\arg z+\arg w$ holds. Likewise $\operatorname{Arg}\bar z=-\operatorname{Arg}z$ fails on the negative real axis: $\operatorname{Arg}\overline{(-1)}=\pi$.
- $\sqrt{zw}\ne\sqrt z\sqrt w$. With principal roots, $\sqrt{-1}\sqrt{-1}=i\cdot i=-1$ but $\sqrt{(-1)(-1)}=1$. An equation $z^n=w$ has $n$ solutions; "the" $n$-th root needs a chosen branch (§1.2).
- Equality in the triangle inequality needs the same direction: $|z+w|=|z|+|w|$ iff one is a non-negative multiple of the other. $z=1$, $w=-1$ are parallel but $|z+w|=0$.
- $\sum_{k=0}^{n-1}\omega^{jk}=0$ needs $n\nmid j$. For $n\mid j$ the sum is $n$; e.g. $\sum_{k=0}^{2}\omega_3^{3k}=3$.
- Not every root of unity is primitive. $\omega^k$ has order $n/\gcd(k,n)$; $-1$ is a $6$-th root of unity of order $2$.
- Conjugate roots need real coefficients. $z^2-iz=z(z-i)$ has the root $i$ but not $-i$. And complex polynomials of odd degree need not have real roots: $z-i$.
- A real polynomial may have no real root ($z^2+1$); only odd degree guarantees one.
- $|z-a|=k|z-b|$ is a circle only for $k\ne1$; $k=1$ gives a line. The centre of the circle is not the midpoint of $a$ and $b$ — it lies outside the segment $[a,b]$.
- $\operatorname{Arg}\frac{z-a}{z-b}=\theta$ is an arc, not a full circle; the complementary arc has $\operatorname{Arg}=\theta-\pi$ (Example 5(c)).
- No Rolle theorem in $\mathbb{C}$. $z^3-1$ vanishes at $1$ and $e^{2\pi i/3}$, but $3z^2$ does not vanish on the segment between them. Gauss–Lucas is the correct replacement.
- Eneström–Kakeya needs positive, monotone coefficients. $2+z$ has decreasing coefficients and the root $-2$ outside the unit disc; $1-2z+3z^2$ has a negative coefficient, and its roots $\tfrac{1\pm i\sqrt2}{3}$ happen to lie inside, so no conclusion follows either way.
- There is one point at infinity. In $\mathbb{C}_\infty$, $-\infty=\infty$; $\infty-\infty$ and $0\cdot\infty$ stay undefined. Stereographic antipodes are $z$ and $-1/\bar z$, not $z$ and $-z$ (these agree only on $|z|=1$).
Exercises
- 1 Write in the form $a+ib$: (a) $(2-i)^3$; (b) $\dfrac{1+2i}{3-4i}$; (c) $i^{2026}$.
- 2 Find both square roots of $8-6i$.
- 3 Solve $z^2-(3+i)z+(4+3i)=0$.
- 4 Prove: if $|z|=1$ and $z\ne1$, then $\operatorname{Re}\dfrac{1}{1-z}=\dfrac12$.
- 5 For $z=-\sqrt3+i$ and $w=i$ find $\operatorname{Arg}z$, $\operatorname{Arg}w$ and $\operatorname{Arg}(zw)$. Is $\operatorname{Arg}(zw)=\operatorname{Arg}z+\operatorname{Arg}w$?
- 6 Use De Moivre's theorem to express $\cos5\theta$ as a polynomial in $\cos\theta$.
- 7 Compute $(\sqrt3-i)^{12}$, and find the least $n\in\mathbb{N}$ for which $(1+i)^n$ is a positive real number.
- 8 Prove that if $|z|=1$ and $az+b\ne0$, then $\left|\dfrac{az+b}{\bar bz+\bar a}\right|=1$.
- 9 Prove Lagrange's identity $\left|\sum_{k=1}^nz_kw_k\right|^2=\sum_k|z_k|^2\sum_k|w_k|^2-\sum_{j\lt k}|z_j\bar w_k-z_k\bar w_j|^2$ for $n=2$, and deduce the Cauchy–Schwarz inequality $\left|z_1w_1+z_2w_2\right|^2\le(|z_1|^2+|z_2|^2)(|w_1|^2+|w_2|^2)$.
- 1 Find all solutions of $z^6=-64$ in the form $a+ib$.
- 2 Solve $z^3=1+i$ in polar form.
- 3 List the primitive $12$-th roots of unity and compute their sum.
- 4 Let $\rho\ne1$ be a cube root of unity. Evaluate $(1-\rho+\rho^2)(1+\rho-\rho^2)$.
- 5 Evaluate $\prod_{k=1}^{5}\sin\dfrac{k\pi}{6}$ using Theorem 1.6(4), and check it directly.
- 6 Let $\omega=e^{2\pi i/n}$. Prove that $\sum_{k=0}^{n-1}(1+\omega^k)^n=2n$.
- 7 Let $p$ be prime. Show that the sum of the primitive $p$-th roots of unity is $-1$ and the sum of the primitive $p^2$-th roots of unity is $0$.
- 8 Let $\zeta\ne1$ satisfy $\zeta^n=1$. Prove that $\sum_{k=1}^{n}k\,\zeta^{k-1}=\dfrac{n}{\zeta-1}$.
- 1 Describe the locus $|z-1|=|z+i|$.
- 2 Find the centre and radius of the circle $|z+2|=3|z-2|$, and verify with two points.
- 3 Identify the curve $|z-2|+|z+2|=6$ and write its Cartesian equation.
- 4 Prove that $z_1,z_2,z_3$ are collinear iff $\operatorname{Im}\left(z_1\bar z_2+z_2\bar z_3+z_3\bar z_1\right)=0$.
- 5 Show that the set $\{z: |z|^2+(1+i)z+(1-i)\bar z+1=0\}$ is a circle and find its centre and radius.
- 6 Find the point of the Riemann sphere corresponding to $z=2-i$, and the complex number corresponding to $\left(0,-\tfrac35,\tfrac45\right)$.
- 7 Prove that $z,w\in\mathbb{C}$ correspond to antipodal points of the sphere iff $z\bar w=-1$.
- 8 Compute the chordal distances $\chi(1,i)$, $\chi(0,\infty)$ and $\chi(1,-1)$. What is the largest possible chordal distance?
- 9 Prove that the map $z\mapsto1/z$ corresponds on the sphere to the rotation by $\pi$ about the $x_1$-axis, $(x_1,x_2,x_3)\mapsto(x_1,-x_2,-x_3)$.
- 1 Find the monic real polynomial of least degree having $1+2i$ and $3$ as roots.
- 2 Let $\alpha,\beta,\gamma$ be the roots of $z^3+2z+5$. Compute $\alpha^2+\beta^2+\gamma^2$ and $\alpha^3+\beta^3+\gamma^3$.
- 3 Given that $i$ is a root of $z^4-2z^3+3z^2-2z+2$, find all its roots.
- 4 Show that every root of $z^5-3z^2+2z-7$ satisfies $|z|\lt8$, and that at least one root satisfies $|z|\ge7^{1/5}$.
- 5 Show that all roots of $1+2z+3z^2+6z^3$ lie in $|z|\le1$ and all roots of $z^3+2z^2+3z+6$ in $|z|\ge1$.
- 6 Find the roots of $p'$ for $p(z)=z(z-2)(z-2i)$ and verify the Gauss–Lucas theorem.
- 7 Prove: if all roots of $p$ lie in the open half-plane $\operatorname{Re}z\gt0$, so do all roots of $p'$.
- 8 Let $p$ be a real polynomial with $p(x)\gt0$ for every real $x$. Prove that $\deg p$ is even and that $p=c\prod_j(z^2+b_jz+c_j)$ with $c\gt0$ and $b_j^2\lt4c_j$.
E1. CSIR Part B GATE MA The number of complex numbers $z$ satisfying $z^5=\bar z$ is (A) $1$; (B) $5$; (C) $6$; (D) $7$.
E2. CSIR Part B The set $\{z\in\mathbb{C}:|z-3|=2|z|\}$ is (A) the circle with centre $-1$ and radius $2$; (B) the circle with centre $1$ and radius $2$; (C) the circle with centre $4$ and radius $2$; (D) a straight line.
E3. CSIR Part B Let $p(z)=z^4+az^3+bz^2+cz+d$ have real coefficients, and suppose $1+i$ and $2-i$ are roots of $p$. Then $p(1)$ equals (A) $0$; (B) $1$; (C) $2$; (D) $10$.
E4. CSIR Part C Let $n\ge3$ and $\omega=e^{2\pi i/n}$. Which of the following hold for every such $n$? (One or more options may be correct.) (A) $\sum_{k=0}^{n-1}\omega^{3k}=0$; (B) $\prod_{k=1}^{n-1}(1-\omega^k)=n$; (C) $\sum_{k=0}^{n-1}|1-\omega^k|^2=2n$; (D) exactly $n-1$ of the numbers $\omega,\omega^2,\dots,\omega^n$ are primitive $n$-th roots of unity.
E5. CSIR Part C GATE MA Let $p$ be a polynomial of degree $5$ with real coefficients. Which of the following are necessarily true? (A) $p$ has at least one real root. (B) If $2i$ is a root of $p$, then $z^2+4$ divides $p$. (C) $p'$ has at least one real root. (D) If all roots of $p$ are real, then all roots of $p'$ are real.
E6. CSIR Part C Identify $\mathbb{C}_\infty$ with the unit sphere $S\subset\mathbb{R}^3$ by stereographic projection from $N=(0,0,1)$. Which of the following are true? (A) The unit circle corresponds to the equator $x_3=0$. (B) For every $z\ne0$, $z$ and $-z$ correspond to antipodal points. (C) Every straight line in $\mathbb{C}$, together with $\infty$, corresponds to a circle on $S$ through $N$. (D) The disc $|z|\lt2$ corresponds to the part of $S$ below the plane $x_3=\tfrac35$.
Suggested plan — 3 lectures × 60 min.
- The field and the plane. Definition 1.1, Theorem 1.1 with the one-line proof that $\mathbb{C}$ cannot be ordered; conjugate and modulus (Theorem 1.2) — prove the triangle inequality with its equality case on the board. Theorem 1.3 and Example 2 (these maps return in Chapters 2 and 4). Example 1(c): square roots without polar form.
- Polar form and roots of unity. Theorem 1.4 with the picture "multiply = rotate and scale"; the $\operatorname{Arg}$ trap. Theorems 1.5–1.6; project Interactive 1 with $n=6$, first $j=1$ (hexagon), then $j=2$ (a triangle traced twice), then $j=6$ (the chain runs off to $6$). Examples 3–4; Exam-corner E1 and E4 live.
- Geometry, the sphere, polynomials. Theorems 1.7–1.8 and Example 5; a sketch of the sphere with the formula of Theorem 1.9 and the proof of Theorem 1.10. Theorems 1.11–1.14 and Example 7; Cauchy's bound and Gauss–Lucas with Interactive 2 (mode 1, sweep $t$ through $\pm2$). Finish with E5.
Discussion questions. (1) We showed $\mathbb{C}$ cannot be ordered as a field. Can $\mathbb{C}$ be given some total order (e.g. lexicographic)? Which field axiom of order does it break? (2) Why does the triangle-inequality proof need $\operatorname{Re}u\le|u|$ — where exactly is geometry hidden in it? (3) The winding-number picture of the fundamental theorem of algebra: what goes wrong with the same picture for $\bar z^{\,n}$ or for $|z|^2+1$?
Misconceptions to watch for. Treating $\operatorname{Arg}$ as additive; forgetting $z=0$ when counting solutions; "every root of unity of order dividing $n$ is primitive"; using the conjugate-root theorem for complex coefficients; reading $|z-a|=k|z-b|$ as a circle centred at a point between $a$ and $b$; believing that Rolle's theorem holds for complex polynomials.
Suggested homework. I(3, 4, 5, 8), II(1, 3, 6, 8), III(2, 4, 7, 9), IV(2, 4, 6, 8); Exam corner E2, E3, E6.
Research question. Where do the roots of a random polynomial live? If the coefficients are chosen at random, how many roots are real, and how close are the others to the unit circle?
Why it matters. Random polynomials model the characteristic equations of large random systems — vibrations of a structure with uncertain parameters, stability of a power grid, digital filters with noisy coefficients in signal processing. Engineers care whether roots stay inside the unit circle (a stable discrete-time filter); physicists study the striking fractal pictures formed by the roots of all polynomials with coefficients $\pm1$.
What to do.
- With Python (
numpy.roots), generate $10\,000$ polynomials of degree $20$ whose $21$ coefficients are independent standard normal numbers. Record the number of real roots of each and the modulus of every root. - Plot a histogram of the root moduli and a scatter plot of all roots. Where do they concentrate?
- Repeat with coefficients chosen uniformly from $\{-1,1\}$. Prove, using Cauchy's bound (Theorem 1.15) applied to $p$ and to $z^{20}p(1/z)$, that every root then satisfies $\tfrac12\lt|z|\lt2$, and compare with your data.
- Compare the average number of real roots with Kac's asymptotic formula $\tfrac{2}{\pi}\log n$ for Gaussian coefficients. Explain why the real roots must come with the conjugate pairs around them (Theorem 1.14).
Tools: Python with numpy and matplotlib (or any language with a polynomial root-finder).
Hand in: Two plots, a table of averages, the proof of the annulus bound, and a paragraph on how the roots approach the unit circle as the degree grows.
Going further. Read about the Erdős–Turán theorem on the angular equidistribution of roots, and plot all roots of all $\pm1$ polynomials of degree $16$ to see the "Littlewood fractal".
🧪 Self-check quiz
Q1. The principal argument of $-1-i$ is
Q2. With $\omega=e^{2\pi i/5}$, the sum $\sum_{k=0}^{4}\omega^{10k}$ equals
Q3. The set $\{z:|z-i|=|z+i|\}$ is
Q4. The least degree of a nonzero polynomial with real coefficients having both $1$ and $i$ as roots is
Q5. Under stereographic projection from $N=(0,0,1)$ onto the unit sphere, $z=i$ corresponds to