Chapter III · Complex numbers

Roots of complex numbers and the roots of unity

§§47–49 35 min

The equation zn=az^n = a has exactly nn roots, at the corners of a regular polygon centred at 00; the nn-th roots of unity are the powers of one of them; and De Moivre's theorem extends to rational powers.

Why this matters

Among the real numbers, x2=−1x^2 = -1 has no solution, x3=8x^3 = 8 has one and x4=16x^4 = 16 has two. The rule is erratic. Among the complex numbers it is perfectly regular: zn=az^n = a has exactly nn solutions for every a≠0a \ne 0, and they sit at the corners of a regular nn-gon. This lesson proves that, using nothing but De Moivre’s theorem and the fact that amplitudes are only fixed up to 2π2\pi. It finishes the chapter by giving a meaning to am/na^{m/n} for complex aa, and it is the first case of the fundamental theorem of algebra that we can prove completely.

The idea

Write a=ρ(cos⁡ϕ+isin⁡ϕ)a = \rho(\cos\phi + i\sin\phi) and look for z=r(cos⁡θ+isin⁡θ)z = r(\cos\theta + i\sin\theta). By De Moivre, zn=rn(cos⁡nθ+isin⁡nθ)z^n = r^n(\cos n\theta + i\sin n\theta). Two polar forms are equal when the moduli are equal and the angles differ by a multiple of 2π2\pi (lesson on the Argand diagram). So we need

rn=ρ,nθ=ϕ+2kπ for some integer k.r^n = \rho, \qquad n\theta = \phi + 2k\pi \text{ for some integer } k.

The first equation has one positive solution, r=ρnr = \sqrt[n]\rho. The second gives θ=(ϕ+2kπ)/n\theta = (\phi + 2k\pi)/n, and here is the whole secret: dividing by nn spreads the extra turns 2kπ2k\pi into nn different directions, spaced 2π/n2\pi/n apart. After k=n−1k = n - 1 the directions repeat.

Explore The nn roots of ww

Try this. Drag ww and watch the nn roots (the larger dot is the principal root). Change nn. Then drag ww once round the origin and follow the principal root.

ReIm
  • w (drag)
  • the n roots
  • unit circle
w=1.2+1.6i=2 (cos⁡0.927+isin⁡0.927)w = 1.2 + 1.6i = 2\,(\cos 0.927 + i\sin 0.927)
The 5 roots of z5=w=1.2+1.6iz^{5} = w = 1.2 + 1.6i have modulus ∣w∣1/5=1.149|w|^{1/5} = 1.149 and arguments arg⁡w+2kπ5\frac{\arg w + 2k\pi}{5}, k=0,…,4k = 0,\dots,4: 1.13+0.212i,  0.147+1.14i,  −1.04+0.492i,  −0.789−0.835i,  …1.13 + 0.212i,\; 0.147 + 1.14i,\; -1.04 + 0.492i,\; -0.789 - 0.835i,\;\dots
What to notice

The roots always form a regular nn-gon on the circle of radius ∣w∣1/n|w|^{1/n}. When ww turns through an angle, each root turns through 1/n1/n of it. When ww makes a full turn, every root moves on to the next corner: the polygon is the same, but the roots have swapped places. That is why no single root can be singled out continuously, and why a “principal value” is a convention.

Precise statements

Definition Roots and rational powers Hardy: §47

For a positive integer nn, an\sqrt[n]{a} or a1/na^{1/n} means any complex number zz with zn=az^n = a. For an integer mm, am/na^{m/n} means (a1/n)m(a^{1/n})^m. These are definitions of what the symbols stand for; whether such numbers exist is a separate question.

Theorem Solution of zn=az^n = a Hardy: §48

Let a=ρ(cos⁡ϕ+isin⁡ϕ)a = \rho(\cos\phi + i\sin\phi) with ρ>0\rho > 0 and −π<ϕ≤π-\pi < \phi \le \pi. The equation zn=az^n = a has exactly nn roots, namely

zk=ρn (cos⁡ϕ+2kπn+isin⁡ϕ+2kπn),k=0,1,…,n−1,z_k = \sqrt[n]{\rho}\,\left(\cos\frac{\phi + 2k\pi}{n} + i\sin\frac{\phi + 2k\pi}{n}\right), \qquad k = 0, 1, \dots, n - 1,

where ρn\sqrt[n]\rho is the positive real nn-th root. The root z0z_0 is the principal value of an\sqrt[n]a.

Corollary The roots of unity Hardy: §48; Examples XXII, 2, 4

The nn-th roots of 11 are 1,ωn,ωn2,…,ωnn−11, \omega_n, \omega_n^2, \dots, \omega_n^{n-1}, where ωn=cos⁡2πn+isin⁡2πn\omega_n = \cos\dfrac{2\pi}n + i\sin\dfrac{2\pi}n. Their sum is 00 (for n≥2n \ge 2). The nn-th roots of any a≠0a \ne 0 are the principal root multiplied by the nn-th roots of unity.

Theorem De Moivre's theorem for rational powers Hardy: §49

If α=p/q\alpha = p/q is rational, then one of the values of (cos⁡θ+isin⁡θ)α(\cos\theta + i\sin\theta)^\alpha is cos⁡αθ+isin⁡αθ\cos\alpha\theta + i\sin\alpha\theta.

Proofs

Proof Solution of zn=az^n = a
Strategy

Put the unknown in polar form, apply De Moivre, and compare moduli as numbers and amplitudes up to multiples of 2π2\pi. Then count how many genuinely different angles arise.

Reduce to two real equations

z=0z = 0 is not a root since a≠0a \ne 0, so write z=r(cos⁡θ+isin⁡θ)z = r(\cos\theta + i\sin\theta) with r>0r > 0. By De Moivre, zn=az^n = a reads rn(cos⁡nθ+isin⁡nθ)=ρ(cos⁡ϕ+isin⁡ϕ)r^n(\cos n\theta + i\sin n\theta) = \rho(\cos\phi + i\sin\phi).

Compare

By the theorem on equal polar forms, this holds iff rn=ρr^n = \rho and nθ−ϕ=2kπn\theta - \phi = 2k\pi for some integer kk. So r=ρnr = \sqrt[n]\rho (the only positive solution) and θ=(ϕ+2kπ)/n\theta = (\phi + 2k\pi)/n. Every such θ\theta gives a root, so the roots are exactly the numbers zkz_k for all integers kk.

Only nn different values of kk matter

Divide kk by nn: k=pn+qk = pn + q with 0≤q<n0 \le q < n. Then θ=2pπ+(ϕ+2qπ)/n\theta = 2p\pi + (\phi + 2q\pi)/n, and adding 2pπ2p\pi to an angle changes nothing. So zk=zqz_k = z_q, and z0,…,zn−1z_0, \dots, z_{n-1} already include every root.

These nn are distinct

For 0≤q<q′<n0 \le q < q' < n the angles of zqz_q and zq′z_{q'} differ by 2π(q′−q)/n2\pi(q' - q)/n, which lies strictly between 00 and 2π2\pi, so it is not a multiple of 2π2\pi. Hence zq≠zq′z_q \ne z_{q'}: there are exactly nn roots. Their angles are spaced 2π/n2\pi/n apart on the circle of radius ρn\sqrt[n]\rho, the corners of a regular nn-gon.

Proof The roots of unity, and rational powers
Strategy

Specialise to a=1a = 1; use De Moivre for powers of ωn\omega_n; sum a geometric progression. For §49, take a qq-th root and raise it to the power pp.

Powers of ωn\omega_n

With ρ=1\rho = 1, ϕ=0\phi = 0, the roots are cos⁡2kπn+isin⁡2kπn\cos\frac{2k\pi}n + i\sin\frac{2k\pi}n, which by De Moivre is ωnk\omega_n^k.

Their sum is 00

ωn≠1\omega_n \ne 1 for n≥2n \ge 2, and ωnn=1\omega_n^n = 1, so the geometric sum gives 1+ωn+⋯+ωnn−1=ωnn−1ωn−1=01 + \omega_n + \dots + \omega_n^{n-1} = \dfrac{\omega_n^n - 1}{\omega_n - 1} = 0.

Roots of aa

If z0n=az_0^n = a then (z0ωnk)n=a⋅1=a(z_0\omega_n^k)^n = a\cdot 1 = a, so the nn distinct numbers z0ωnkz_0\omega_n^k are all the roots.

Rational powers (§49)

Let α=p/q\alpha = p/q with q>0q > 0. By De Moivre, c=cos⁡(θ/q)+isin⁡(θ/q)c = \cos(\theta/q) + i\sin(\theta/q) satisfies cq=cos⁡θ+isin⁡θc^q = \cos\theta + i\sin\theta, so cc is one value of (cos⁡θ+isin⁡θ)1/q(\cos\theta + i\sin\theta)^{1/q}. Then cp=cos⁡(pθ/q)+isin⁡(pθ/q)c^p = \cos(p\theta/q) + i\sin(p\theta/q) (De Moivre for the integer pp) is one value of (cos⁡θ+isin⁡θ)p/q(\cos\theta + i\sin\theta)^{p/q}.

All the roots of unity are powers of ωn\omega_n. But they need not all be powers of some other root ωnk\omega_n^k; the walk below shows when they are.

Explore Walking by ωk\omega^k

Try this. With n=8n = 8, try the steps k=1,2,3,4k = 1, 2, 3, 4. Which walks visit every root? Then try n=7n = 7 with several kk, and n=12n = 12. Use “steps drawn” to follow a walk one step at a time.

ReIm
  • the n-th roots of unity
  • the walk 1 → ωᵏ → ω²ᵏ → …
  • average of 1, ωᵏ, …, ω⁽ⁿ⁻¹⁾ᵏ
ω=cos⁡2π8+isin⁡2π8\omega = \cos\tfrac{2\pi}{8} + i\sin\tfrac{2\pi}{8}; the walk by ω3\omega^{3} returns to 11 after 8/gcd⁡(8,3)=88/\gcd(8, 3) = 8 steps, visiting every root: ω3\omega^{3} is a primitive 8th root of unity.
1+ω3+ω6+⋯+ω21=01 + \omega^{3} + \omega^{6} + \dots + \omega^{21} = 0 (the points balance about the origin).
What to notice

The walk by ωk\omega^k closes after n/gcd⁡(n,k)n/\gcd(n, k) steps. It visits every root exactly when gcd⁡(n,k)=1\gcd(n, k) = 1; for prime nn this happens for every kk from 11 to n−1n - 1. Whatever kk is, the sum 1+ωk+ω2k+⋯+ω(n−1)k1 + \omega^k + \omega^{2k} + \dots + \omega^{(n-1)k} is 00 unless nn divides kk: the points it sums are a regular polygon (traversed gcd⁡(n,k)\gcd(n, k) times), and they balance about the origin.

Worked examples and exercises

Exercise Small roots of unity ●○○ Hardy: Examples XXII, 1

Find the cube roots and the fourth roots of 11 in the form x+yix + yi, and show that the fifth roots other than 11 are 14[±5−1±i10±25]\frac14\bigl[\pm\sqrt5 - 1 \pm i\sqrt{10 \pm 2\sqrt5}\bigr] with suitable choices of sign.

Hint
For the fifth roots you need cos⁡72∘\cos 72^\circ. Since ω=ω5\omega = \omega_5 satisfies 1+ω+ω2+ω3+ω4=01 + \omega + \omega^2 + \omega^3 + \omega^4 = 0, the number c=ω+ω4=2cos⁡72∘c = \omega + \omega^4 = 2\cos 72^\circ satisfies a quadratic: compute c2c^2.

Solution

Cube roots. cos⁡2kπ3+isin⁡2kπ3\cos\frac{2k\pi}{3} + i\sin\frac{2k\pi}3: 11 and 12(−1±i3)\frac12(-1 \pm i\sqrt3). Fourth roots: 1,i,−1,−i1, i, -1, -i.

Fifth roots. Let ω=ω5\omega = \omega_5, c=ω+ω4=2cos⁡72∘c = \omega + \omega^4 = 2\cos 72^\circ and d=ω2+ω3=2cos⁡144∘d = \omega^2 + \omega^3 = 2\cos 144^\circ. Then c+d=−1c + d = -1 (sum of the roots minus 11) and cd=ω3+ω4+ω6+ω7=ω3+ω4+ω+ω2=−1cd = \omega^3 + \omega^4 + \omega^6 + \omega^7 = \omega^3 + \omega^4 + \omega + \omega^2 = -1. So c,dc, d are the roots of t2+t−1=0t^2 + t - 1 = 0, and as c>0>dc > 0 > d: c=5−12c = \frac{\sqrt5 - 1}2, d=−5−12d = \frac{-\sqrt5 - 1}2. Hence cos⁡72∘=5−14\cos 72^\circ = \frac{\sqrt5 - 1}4 and sin⁡72∘=1−cos⁡272∘=1410+25\sin 72^\circ = \sqrt{1 - \cos^2 72^\circ} = \frac14\sqrt{10 + 2\sqrt5}; also cos⁡144∘=−5−14\cos 144^\circ = \frac{-\sqrt5 - 1}4 and sin⁡144∘=1410−25\sin 144^\circ = \frac14\sqrt{10 - 2\sqrt5}. The four roots are 14[5−1±i10+25]\frac14[\sqrt5 - 1 \pm i\sqrt{10 + 2\sqrt5}] and 14[−5−1±i10−25]\frac14[-\sqrt5 - 1 \pm i\sqrt{10 - 2\sqrt5}].

Exercise ●○○ Hardy: Examples XXII, 3

With ω=ω3\omega = \omega_3, prove that (x+yω+zω2)(x+yω2+zω)=x2+y2+z2−yz−zx−xy(x + y\omega + z\omega^2)(x + y\omega^2 + z\omega) = x^2 + y^2 + z^2 - yz - zx - xy.

Hint
Use ω3=1\omega^3 = 1 and ω+ω2=−1\omega + \omega^2 = -1.

Solution

Multiplying out: x2+y2ω3+z2ω3+xy(ω2+ω)+xz(ω+ω2)+yz(ω2+ω4)x^2 + y^2\omega^3 + z^2\omega^3 + xy(\omega^2 + \omega) + xz(\omega + \omega^2) + yz(\omega^2 + \omega^4). Since ω3=1\omega^3 = 1, ω4=ω\omega^4 = \omega, and ω+ω2=−1\omega + \omega^2 = -1 (sum of the cube roots of unity is 00), this is x2+y2+z2−xy−xz−yzx^2 + y^2 + z^2 - xy - xz - yz.

Exercise Square roots by formula ●●○ Hardy: Examples XXII, 5

In Examples XXI, 14 Hardy found the square roots of α+βi\alpha + \beta i (β≠0\beta \ne 0) to be

±[12(α2+β2+α)+i σ12(α2+β2−α)],\pm\left[\sqrt{\tfrac12\bigl(\sqrt{\alpha^2 + \beta^2} + \alpha\bigr)} + i\,\sigma\sqrt{\tfrac12\bigl(\sqrt{\alpha^2 + \beta^2} - \alpha\bigr)}\right],

where σ\sigma is the sign of β\beta. Show that this agrees with §48.

Hint
With ρ=α2+β2\rho = \sqrt{\alpha^2 + \beta^2}, §48 gives the roots ±ρ(cos⁡ϕ2+isin⁡ϕ2)\pm\sqrt\rho(\cos\frac\phi2 + i\sin\frac\phi2). Use the half-angle formulas cos⁡2ϕ2=12(1+cos⁡ϕ)\cos^2\frac\phi2 = \frac12(1 + \cos\phi) and sin⁡2ϕ2=12(1−cos⁡ϕ)\sin^2\frac\phi2 = \frac12(1 - \cos\phi).

Solution

Write α+βi=ρ(cos⁡ϕ+isin⁡ϕ)\alpha + \beta i = \rho(\cos\phi + i\sin\phi), −π<ϕ≤π-\pi < \phi \le \pi; since β≠0\beta \ne 0, ϕ≠0,π\phi \ne 0, \pi. By §48 the roots are ±ρ(cos⁡ϕ2+isin⁡ϕ2)\pm\sqrt\rho(\cos\frac\phi2 + i\sin\frac\phi2), with ϕ2∈(−π2,π2)\frac\phi2 \in (-\frac\pi2, \frac\pi2), so cos⁡ϕ2>0\cos\frac\phi2 > 0 and sin⁡ϕ2\sin\frac\phi2 has the sign of ϕ\phi, which is the sign σ\sigma of β\beta. Now ρcos⁡ϕ2=ρ⋅12(1+α/ρ)=12(ρ+α)\sqrt\rho\cos\frac\phi2 = \sqrt{\rho\cdot\frac12(1 + \alpha/\rho)} = \sqrt{\frac12(\rho + \alpha)} and ρ ∣sin⁡ϕ2∣=12(ρ−α)\sqrt\rho\,|\sin\frac\phi2| = \sqrt{\frac12(\rho - \alpha)}. That is the formula. (The rule “like signs if β>0\beta > 0, unlike if β<0\beta < 0” is the factor σ\sigma.)

Exercise Real factors of x2m−a2mx^{2m} - a^{2m} ●●○ Hardy: Examples XXII, 6

Let a>0a > 0. Show that x2m−a2m=(x2−a2)∏s=1m−1(x2−2axcos⁡sπm+a2)x^{2m} - a^{2m} = (x^2 - a^2)\displaystyle\prod_{s=1}^{m-1}\left(x^2 - 2ax\cos\frac{s\pi}{m} + a^2\right).

Hint
The roots of x2m=a2mx^{2m} = a^{2m} are a ω2msa\,\omega_{2m}^s, s=0,…,2m−1s = 0, \dots, 2m - 1. Pair ss with 2m−s2m - s.

Solution

The 2m2m roots are a(cos⁡sπm+isin⁡sπm)a(\cos\frac{s\pi}m + i\sin\frac{s\pi}m), s=0,1,…,2m−1s = 0, 1, \dots, 2m - 1, and by the factorisation theorem of §43 (leading coefficient 11), x2m−a2mx^{2m} - a^{2m} is the product of xx minus each root. The roots s=0s = 0 and s=ms = m are aa and −a-a, giving x2−a2x^2 - a^2. For 1≤s≤m−11 \le s \le m - 1, the roots for ss and 2m−s2m - s are conjugate, and (x−a(cos⁡sπm+isin⁡sπm))(x−a(cos⁡sπm−isin⁡sπm))=x2−2axcos⁡sπm+a2.\bigl(x - a(\cos\tfrac{s\pi}m + i\sin\tfrac{s\pi}m)\bigr)\bigl(x - a(\cos\tfrac{s\pi}m - i\sin\tfrac{s\pi}m)\bigr) = x^2 - 2ax\cos\tfrac{s\pi}m + a^2. These m−1m - 1 pairs use up the remaining 2m−22m - 2 roots. Dividing by x2−a2x^2 - a^2 gives Hardy’s form of the result.

Exercise ●●○ Hardy: Examples XXII, 9 (Math. Trip. 1910)

Find all the roots of x6−2x3+2=0x^6 - 2x^3 + 2 = 0.

Hint
It is a quadratic in x3x^3.

Solution

(x3)2−2x3+2=0(x^3)^2 - 2x^3 + 2 = 0 gives x3=1±ix^3 = 1 \pm i. Now 1+i=2(cos⁡π4+isin⁡π4)1 + i = \sqrt2(\cos\frac\pi4 + i\sin\frac\pi4), whose cube roots are 21/6(cos⁡θ+isin⁡θ)2^{1/6}(\cos\theta + i\sin\theta) with θ=π12,3π4,−7π12\theta = \frac{\pi}{12}, \frac{3\pi}4, -\frac{7\pi}{12} (that is, π12+2kπ3\frac\pi{12} + \frac{2k\pi}3 reduced to the principal range). The cube roots of 1−i1 - i are their conjugates, θ=−π12,−3π4,7π12\theta = -\frac\pi{12}, -\frac{3\pi}4, \frac{7\pi}{12}. So the six roots are 21/6(cos⁡θ+isin⁡θ)2^{1/6}(\cos\theta + i\sin\theta) with θ=±π12,±7π12,±3π4\theta = \pm\frac{\pi}{12}, \pm\frac{7\pi}{12}, \pm\frac{3\pi}{4}, as expected for a real equation, in conjugate pairs.

Check yourself

Check yourself Enter a number

What is the real part of the principal value of −83\sqrt[3]{-8}?

Why

−8=8(cos⁡π+isin⁡π)-8 = 8(\cos\pi + i\sin\pi) with principal amplitude π\pi, so the principal cube root is 2(cos⁡π3+isin⁡π3)=1+3 i2(\cos\frac\pi3 + i\sin\frac\pi3) = 1 + \sqrt3\,i, with real part 11. The real cube root −2-2 is one of the three values, but not the principal one (k=1k = 1). This is a frequent surprise.

Check yourself Enter a number

With Hardy’s definition am/n=(a1/n)ma^{m/n} = (a^{1/n})^m, how many distinct values does 12/41^{2/4} have?

Why

The fourth roots of 11 are 1,i,−1,−i1, i, -1, -i; their squares are 1,−1,1,−11, -1, 1, -1. So there are 22 values, the same as for 11/21^{1/2}. In general (a1/n)m(a^{1/n})^m has n/gcd⁡(m,n)n/\gcd(m, n) values, which is why one usually reduces m/nm/n to lowest terms first.

Check yourself True or false?

Every sixth root of unity is a power of ω62\omega_6^2.

Why

False. ω62=ω3\omega_6^2 = \omega_3, and its powers are only the three cube roots of unity 1,ω62,ω641, \omega_6^2, \omega_6^4. The walk by ω2\omega^2 with n=6n = 6 closes after 6/gcd⁡(6,2)=36/\gcd(6, 2) = 3 steps. Only ω6\omega_6 and ω65\omega_6^5 are primitive sixth roots.

Check yourself Choose one

The roots of z4=−16z^4 = -16 lie

Why

−16=16(cos⁡π+isin⁡π)-16 = 16(\cos\pi + i\sin\pi), so the roots have modulus 161/4=216^{1/4} = 2 and angles (π+2kπ)/4(\pi + 2k\pi)/4, i.e. ±2±2 i\pm\sqrt2 \pm \sqrt2\,i. The points ±2,±2i\pm2, \pm2i are the roots of z4=+16z^4 = +16. The last option is true only for real numbers.

Hardy’s language

That the n roots are distinct is easily seen by plotting them on Argand’s diagram.

— G. H. Hardy, §48

Summary

Summary
  • zn=az^n = a (a≠0a \ne 0) has exactly nn roots: modulus ∣a∣n\sqrt[n]{|a|}, amplitudes (ϕ+2kπ)/n(\phi + 2k\pi)/n for k=0,…,n−1k = 0, \dots, n - 1, the corners of a regular nn-gon.
  • The extra turns 2kπ2k\pi are what produce the different roots; the principal value takes k=0k = 0 with −π<ϕ≤π-\pi < \phi \le \pi.
  • The nn-th roots of unity are 1,ωn,…,ωnn−11, \omega_n, \dots, \omega_n^{n-1}; they sum to 00; the roots of aa are the principal root times them.
  • One value of (cos⁡θ+isin⁡θ)p/q(\cos\theta + i\sin\theta)^{p/q} is cos⁡pθq+isin⁡pθq\cos\frac{p\theta}q + i\sin\frac{p\theta}q.