Mathematics · Book 3 · Bachelor Year 1

University Mathematics — Year 1

University Mathematics — Year 1 · Bachelor Year 1

3Complex Numbers

Complex numbers were met in the High School volume as a computational device for quadratic equations. This chapter treats them as a central object: the exponential form and its consequences (de Moivre, nn-th roots, roots of unity), the systematic translation between C\C and plane geometry, and the use of eiθ\eu^{\iu\theta} as a machine for proving trigonometric identities.

3.1 The field C\C, modulus and conjugate

Definition 3.1 (The complex field)

C={a+ib:a,bR}\C = \{a + \iu b : a, b \in \R\} with the usual addition and the multiplication determined by i2=1\iu^2 = -1. Every nonzero z=a+ibz = a + \iu b has an inverse: z1=aiba2+b2z^{-1} = \frac{a - \iu b}{a^2 + b^2} — in the language of Chapter 7, C\C is a field. One writes a=(z)a = \Re(z), b=(z)b = \Im(z), z=aib\conj{z} = a - \iu b (the conjugate) and z=a2+b2\abs{z} = \sqrt{a^2 + b^2} (the modulus).

Proposition 3.2 (Rules for conjugate and modulus)

For z,wCz, w \in \C:

  1. z+w=z+w\conj{z + w} = \conj z + \conj w, zw=zw\conj{zw} = \conj z\, \conj w, z=z\conj{\conj z} = z;
  2. zz=z2z \conj z = \abs z^2;   (z)=z+z2\;\Re(z) = \frac{z + \conj z}{2}, (z)=zz2i\Im(z) = \frac{z - \conj z}{2\iu};
  3. zw=zw\abs{zw} = \abs z\, \abs w, and z1=z1\abs{z^{-1}} = \abs{z}^{-1} for z0z \neq 0;
  4. (triangle inequality) z+wz+w\abs{z + w} \leq \abs z + \abs w, with equality if and only if zz and ww lie on a common ray from 00 (w=λzw = \lambda z or z=λwz = \lambda w with λ0\lambda \geq 0);
  5. (reverse triangle inequality) zwzw\bigl|\abs z - \abs w\bigr| \leq \abs{z - w}.

Proof. (1) and (2) are direct computations on real and imaginary parts. (3): zw2=zwzw=zzww=z2w2\abs{zw}^2 = zw\,\conj{zw} = z\conj z\, w \conj w = \abs z^2 \abs w^2, then take square roots; apply to zz1=1z \cdot z^{-1} = 1 for the inverse.

(4) Both sides are nonnegative, so compare squares:

z+w2=(z+w)(z+w)=z2+w2+2(zw),\abs{z+w}^2 = (z+w)(\conj z + \conj w) = \abs z^2 + \abs w^2 + 2\,\Re(z \conj w),

and (zw)zw=zw\Re(z\conj w) \leq \abs{z \conj w} = \abs z \abs w gives z+w2(z+w)2\abs{z+w}^2 \leq (\abs z + \abs w)^2. Equality forces (zw)=zw\Re(z \conj w) = \abs{z \conj w}, i.e. zwR+z \conj w \in \R_+; if w0w \neq 0 this gives z=zww2w=λwz = \frac{z\conj w}{\abs w^2}\, w = \lambda w with λ0\lambda \geq 0 (and the case w=0w = 0 is trivial).

(5) z=(zw)+wzw+w\abs z = \abs{(z - w) + w} \leq \abs{z-w} + \abs w gives zwzw\abs z - \abs w \leq \abs{z - w}; swap zz and ww for the other sign.

Example 3.3 (A complete modulus-and-argument workout)

Put w=3+4i12iw = \dfrac{3 + 4\iu}{1 - 2\iu} in algebraic form and compute its modulus twice. Multiplying by the conjugate of the denominator:

w=(3+4i)(1+2i)(12i)(1+2i)=3+6i+4i81+4=5+10i5=1+2i.w = \frac{(3 + 4\iu)(1 + 2\iu)}{(1 - 2\iu)(1 + 2\iu)} = \frac{3 + 6\iu + 4\iu - 8}{1 + 4} = \frac{-5 + 10\iu}{5} = -1 + 2\iu .

Directly: w=1+4=5\abs w = \sqrt{1 + 4} = \sqrt5. Via the quotient rule (Proposition 3.2 (3)): w=3+4i12i=55=5\abs w = \frac{\abs{3 + 4\iu}}{\abs{1 - 2\iu}} = \frac5{\sqrt5} = \sqrt5 — same answer, no algebraic form needed. The lesson generalizes: moduli and arguments travel well through products and quotients, real and imaginary parts travel well through sums. Choose the representation that matches the operations at hand, and convert only when forced.

Example 3.4 (Equations involving the conjugate)

Solve in C\C:   z+2z=6+2i\;z + 2\conj z = 6 + 2\iu. An equation mixing zz and z\conj z is not polynomial in zz; the reliable move is to split into real coordinates. With z=x+iyz = x + \iu y:

z+2z=3xiy,z + 2\conj z = 3x - \iu y ,

so the equation reads 3x=63x = 6 and y=2-y = 2: the unique solution is z=22iz = 2 - 2\iu. (Check: (22i)+2(2+2i)=6+2i(2 - 2\iu) + 2(2 + 2\iu) = 6 + 2\iu.) Alternatively, conjugate the whole equation to get z+2z=62i\conj z + 2z = 6 - 2\iu and solve the linear system in the unknowns z,zz, \conj z — same answer, and a useful trick when the coefficients are complex. Equations in zz and z\conj z are really systems of two real equations; expecting “degree 11, one solution” is safe here, but zz=1z\conj z = -1 (no solution) shows the polynomial intuition failing as soon as products appear.

3.2 Exponential form

Definition 3.5 (Complex exponential of an imaginary argument)

For θR\theta \in \R one defines

eiθ=cosθ+isinθ.\eu^{\iu\theta} = \cos\theta + \iu \sin\theta .

Every z0z \neq 0 can be written z=reiθz = r\,\eu^{\iu\theta} with r=z>0r = \abs z > 0; θ\theta is an argument of zz, determined up to adding a multiple of 2π2\pi. The value in (π,π]\intoc{-\pi}{\pi} is the principal argument, written argz\arg z.

Example 3.6 (First values, and one famous identity)

Reading the definition at the cardinal angles:

eiπ/2=i,eiπ=1,e2iπ=1,eiπ/4=22(1+i).\eu^{\iu\pi/2} = \iu, \qquad \eu^{\iu\pi} = -1, \qquad \eu^{2\iu\pi} = 1, \qquad \eu^{\iu\pi/4} = \frac{\sqrt2}2\,(1 + \iu) .

The second one, rearranged as eiπ+1=0\eu^{\iu\pi} + 1 = 0, is Euler’s celebrated identity linking e\eu, i\iu, π\pi, 11 and 00; at this stage of the book it is a definition unwinding rather than a theorem, and its real content — why the analytic exponential function of Chapter 4, extended to imaginary arguments, deserves the same name — is settled by the power series of Chapter 17. Meanwhile the display above is worth memorizing as a conversion table: it is used silently every time an argument is read off a picture.

Theorem 3.7 (The functional equation)

For all θ,φR\theta, \varphi \in \R:

eiθeiφ=ei(θ+φ),eiθ=1,eiθ=eiθ=(eiθ)1.\eu^{\iu\theta}\, \eu^{\iu\varphi} = \eu^{\iu(\theta + \varphi)}, \qquad \abs{\eu^{\iu\theta}} = 1, \qquad \conj{\eu^{\iu\theta}} = \eu^{-\iu\theta} = (\eu^{\iu\theta})^{-1}.

Consequently zw=zw\abs{zw} = \abs z \abs w carries the arguments too: arg(zw)argz+argw(mod2π)\arg(zw) \equiv \arg z + \arg w \pmod{2\pi}.

Proof. Expand the product and use the addition formulas:

(cosθ+isinθ)(cosφ+isinφ)=(cosθcosφsinθsinφ)+i(sinθcosφ+cosθsinφ),(\cos\theta + \iu\sin\theta)(\cos\varphi + \iu\sin\varphi) = (\cos\theta\cos\varphi - \sin\theta\sin\varphi) + \iu\,(\sin\theta\cos\varphi + \cos\theta\sin\varphi),

which is cos(θ+φ)+isin(θ+φ)\cos(\theta+\varphi) + \iu\sin(\theta+\varphi). The modulus is cos2θ+sin2θ=1\sqrt{\cos^2\theta + \sin^2\theta} = 1, and the conjugate formula is the parity of cosine and sine; it inverts eiθ\eu^{\iu\theta} since eiθeiθ=e0=1\eu^{\iu\theta}\eu^{-\iu\theta} = \eu^0 = 1.

Corollary 3.8 (de Moivre’s formula)

For θR\theta \in \R and nZn \in \Z:   (cosθ+isinθ)n=cos(nθ)+isin(nθ)\;(\cos\theta + \iu\sin\theta)^n = \cos(n\theta) + \iu\sin(n\theta).

Proof. For n0n \geq 0, induct. The case n=0n = 0 reads 1=11 = 1. Assuming the formula for nn, the functional equation (Theorem 3.7) gives

(eiθ)n+1=(eiθ)neiθ=einθeiθ=ei(n+1)θ,\bigl(\eu^{\iu\theta}\bigr)^{n+1} = \bigl(\eu^{\iu\theta}\bigr)^{n}\,\eu^{\iu\theta} = \eu^{\iu n\theta}\,\eu^{\iu\theta} = \eu^{\iu(n+1)\theta} ,

which is the formula at rank n+1n + 1. For n<0n < 0, write n=mn = -m with m>0m > 0: since eimθ\eu^{\iu m\theta} has inverse eimθ\eu^{-\iu m\theta} (Theorem 3.7 again),

(eiθ)m=(eimθ)1=ei(m)θ.\bigl(\eu^{\iu\theta}\bigr)^{-m} = \bigl(\eu^{\iu m\theta}\bigr)^{-1} = \eu^{\iu(-m)\theta} . \qedhere

Example 3.9 (Expanding cos3θ\cos 3\theta by de Moivre)

Write c=cosθc = \cos\theta, s=sinθs = \sin\theta. De Moivre and the binomial theorem give

cos3θ+isin3θ=(c+is)3=c33cs2+i(3c2ss3),\cos 3\theta + \iu \sin 3\theta = (c + \iu s)^3 = c^3 - 3cs^2 + \iu\,(3c^2 s - s^3),

and identifying real parts, then substituting s2=1c2s^2 = 1 - c^2:

cos3θ=c33c(1c2)=4cos3θ3cosθ.\cos 3\theta = c^3 - 3c(1 - c^2) = 4\cos^3\theta - 3\cos\theta .

The imaginary part yields sin3θ=3sinθ4sin3θ\sin 3\theta = 3\sin\theta - 4\sin^3\theta for free: one complex identity always carries two real ones. Read backwards, the boxed identity is the key to the classical trisection equation: constructing cos(θ/3)\cos(\theta/3) from cosθ\cos\theta means solving the cubic 4x33x=cosθ4x^3 - 3x = \cos\theta, which is where the algebra of Chapter 8 takes over.

Proposition 3.10 (Euler’s formulas)

cosθ=eiθ+eiθ2,sinθ=eiθeiθ2i.\cos\theta = \frac{\eu^{\iu\theta} + \eu^{-\iu\theta}}{2}, \qquad \sin\theta = \frac{\eu^{\iu\theta} - \eu^{-\iu\theta}}{2\iu}.

Proof. Add, respectively subtract, eiθ=cosθ+isinθ\eu^{\iu\theta} = \cos\theta + \iu\sin\theta and eiθ=cosθisinθ\eu^{-\iu\theta} = \cos\theta - \iu\sin\theta.

Method 3.11 (Trigonometry via exponentials)

  1. Linearize cospθsinqθ\cos^p\theta \sin^q\theta (turn powers into a sum of coskθ\cos k\theta, sinkθ\sin k\theta): substitute Euler’s formulas, expand with the binomial theorem (Theorem 2.16), regroup conjugate terms.
  2. Expand cosnθ\cos n\theta as a polynomial in cosθ\cos\theta: write cosnθ=((cosθ+isinθ)n)\cos n\theta = \Re\bigl((\cos\theta + \iu\sin\theta)^n\bigr), expand, and convert even powers of sin\sin via sin2=1cos2\sin^2 = 1 - \cos^2.
  3. Sum trigonometric series like kcoskθ\sum_k \cos k\theta: recognize the real part of a geometric sum k(eiθ)k\sum_k (\eu^{\iu\theta})^k.
  4. The half-angle factorization: for all p,qp, q,

    eip+eiq=2cospq2  eip+q2,eipeiq=2isinpq2  eip+q2.\eu^{\iu p} + \eu^{\iu q} = 2 \cos\tfrac{p - q}{2}\; \eu^{\iu \frac{p+q}{2}}, \qquad \eu^{\iu p} - \eu^{\iu q} = 2\iu \sin\tfrac{p - q}{2}\; \eu^{\iu \frac{p+q}{2}} .

Example 3.12 (Linearization)

cos3θ=(eiθ+eiθ2) ⁣3=e3iθ+3eiθ+3eiθ+e3iθ8=cos3θ+3cosθ4.\cos^3\theta = \Bigl(\frac{\eu^{\iu\theta} + \eu^{-\iu\theta}}{2}\Bigr)^{\!3} = \frac{\eu^{3\iu\theta} + 3\eu^{\iu\theta} + 3\eu^{-\iu\theta} + \eu^{-3\iu\theta}}{8} = \frac{\cos 3\theta + 3\cos\theta}{4}.

This form integrates immediately — the very reason linearization matters in Chapter 15. A mixed product works the same way, only with both Euler formulas at once:

sin2θcos2θ=(sin2θ2) ⁣2=14(e2iθe2iθ2i) ⁣2=2e4iθe4iθ16=1cos4θ8,\sin^2\theta\cos^2\theta = \Bigl(\frac{\sin2\theta}2\Bigr)^{\!2} = \frac{1}{4}\cdot \Bigl(\frac{\eu^{2\iu\theta} - \eu^{-2\iu\theta}}{2\iu} \Bigr)^{\!2} = \frac{2 - \eu^{4\iu\theta} - \eu^{-4\iu\theta}}{16} = \frac{1 - \cos4\theta}{8} ,

where the double-angle shortcut in the first step saved one binomial expansion — always worth scanning for before mechanizing.

Example 3.13 (A binomial trigonometric sum)

For nNn \in \N and θR\theta \in \R, evaluate S=k=0n(nk)coskθS = \sum_{k=0}^{n}\binom nk \cos k\theta. Recognize the real part of a binomial expansion:

S=k=0n(nk)(eiθ)k=(1+eiθ)n,S = \Re\sum_{k=0}^n \binom nk \bigl(\eu^{\iu\theta}\bigr)^k = \Re\bigl(1 + \eu^{\iu\theta}\bigr)^n ,

then factor the half-angle (Method 3.11 (4)): 1+eiθ=2cosθ2eiθ/21 + \eu^{\iu\theta} = 2\cos\frac\theta2\,\eu^{\iu\theta/2}, so

S=(2ncosnθ2  einθ/2)=2ncosnθ2cosnθ2.S = \Re\Bigl(2^n\cos^n\frac\theta2\;\eu^{\iu n\theta/2}\Bigr) = 2^n \cos^n\frac\theta2\,\cos\frac{n\theta}2 .

The imaginary part yields k(nk)sinkθ=2ncosnθ2sinnθ2\sum_k\binom nk\sin k\theta = 2^n\cos^n\frac\theta2\sin\frac{n\theta}2 for free. Sanity checks: θ=0\theta = 0 recovers (nk)=2n\sum\binom nk = 2^n, and θ=π\theta = \pi gives S=0S = 0 for n1n \geq 1 (each factor cosπ2\cos\frac\pi2 vanishes), i.e. the alternating row sum of Example 2.17. The method — “see the cosine sum as the shadow of a complex power, then factor half-angles” — is exactly that of Exercise 3.6, with the binomial theorem replacing the geometric series.

3.3 Roots of complex numbers

Theorem 3.14 (nn-th roots)

Let a=reiα0a = r\,\eu^{\iu\alpha} \neq 0 and nNn \in \N^*. The equation zn=az^n = a has exactly nn solutions:

zk=r1/nei(αn+2kπn),k=0,1,,n1.z_k = r^{1/n}\, \eu^{\iu\left(\frac{\alpha}{n} + \frac{2k\pi}{n}\right)}, \qquad k = 0, 1, \dots, n - 1 .

Proof. Write z=ρeiθz = \rho\,\eu^{\iu\theta} (ρ>0\rho > 0). Then zn=ρneinθ=reiαz^n = \rho^n \eu^{\iu n\theta} = r \eu^{\iu\alpha} if and only if ρn=r\rho^n = r (moduli) and nθα(mod2π)n\theta \equiv \alpha \pmod{2\pi} (arguments), i.e. ρ=r1/n\rho = r^{1/n} and θ=αn+2kπn\theta = \frac{\alpha}{n} + \frac{2k\pi}{n} for some kZk \in \Z. It remains to see when two integers k,kk, k' yield the same number: this happens exactly when the angles differ by a multiple of 2π2\pi,

2kπn2kπn2πZ    kknZ    nkk.\frac{2k\pi}n - \frac{2k'\pi}n \in 2\pi\Z \iff \frac{k - k'}n \in \Z \iff n \mid k - k' .

By Euclidean division, every kZk \in \Z is congruent mod nn to exactly one element of {0,1,,n1}\{0, 1, \dots, n-1\}, so this range lists each solution once and the count is exactly nn. (The same argument, run inside Un\mathbb U_n, shows the roots form a regular nn-gon: consecutive values of kk rotate by the fixed angle 2πn\frac{2\pi}n.)

Example 3.15 (Cube roots of 27-27)

Solve z3=27z^3 = -27. Exponential form of the right side: 27=27eiπ-27 = 27\,\eu^{\iu\pi}, so the three roots are

zk=3ei(π3+2kπ3),k=0,1,2:z0=3eiπ/3=32+332i,z1=3,z2=z0.z_k = 3\,\eu^{\iu(\frac\pi3 + \frac{2k\pi}3)}, \quad k = 0, 1, 2 : \qquad z_0 = 3\eu^{\iu\pi/3} = \frac32 + \frac{3\sqrt3}2\,\iu, \quad z_1 = -3, \quad z_2 = \conj{z_0} .

Two checks. First, the real root 3-3 is the obvious one, and the other two are its rotations by ±2π3\pm\frac{2\pi}3 — equivalently 3j-3j and 3j2-3j^2. Second, algebra confirms: z3+27=(z+3)(z23z+9)z^3 + 27 = (z + 3)(z^2 - 3z + 9), and the quadratic has discriminant 936=27<09 - 36 = -27 < 0 with roots 3±3i32=z0,z0\frac{3 \pm 3\iu\sqrt3}2 = z_0, \conj{z_0}. The insight: for real right-hand sides, the non-real roots always come in conjugate pairs, so a picture of the solution set is symmetric about the real axis — a preview of the real factorization theorem of Chapter 8.

Example 3.16 (A non-real right-hand side)

Solve z4=8+8i3z^4 = -8 + 8\iu\sqrt3. Exponential form of the right side: modulus 64+192=16\sqrt{64 + 192} = 16, argument θ\theta with cosθ=12\cos\theta = -\frac12, sinθ=32\sin\theta = \frac{\sqrt3}2, i.e. θ=2π3\theta = \frac{2\pi}3. The four roots are

zk=2ei(π6+kπ2),k=0,1,2,3:z0=3+i,z1=iz0=1+i3,z_k = 2\,\eu^{\iu(\frac\pi6 + \frac{k\pi}2)}, \quad k = 0, 1, 2, 3 : \qquad z_0 = \sqrt3 + \iu,\quad z_1 = \iu z_0 = -1 + \iu\sqrt3,
z2=z0=3i,z3=iz0=1i3.z_2 = -z_0 = -\sqrt3 - \iu,\qquad z_3 = -\iu z_0 = 1 - \iu\sqrt3 .

(Check: z02=2+2i3z_0^2 = 2 + 2\iu\sqrt3, so z04=(2+2i3)2=412+8i3=8+8i3z_0^4 = (2 + 2\iu\sqrt3)^2 = 4 - 12 + 8\iu\sqrt3 = -8 + 8\iu\sqrt3.) Once one root is found, the other three come free: they are its successive rotations by π2\frac\pi2, i.e. its products with the fourth roots of unity — the general structure behind Theorem 3.14, worth exploiting before recomputing each root from scratch. No conjugate symmetry this time: the right-hand side is not real.

Definition 3.17 (Roots of unity)

The nn-th roots of unity are the solutions of zn=1z^n = 1:

Un={ωk:k=0,,n1},ω=e2iπ/n.\mathbb{U}_n = \bigl\{\, \omega^k : k = 0, \dots, n-1 \,\bigr\}, \qquad \omega = \eu^{2\iu\pi/n}.

They form a group under multiplication (Chapter 7) and sit at the vertices of a regular nn-gon inscribed in the unit circle.

The fifth roots of unity, = 2 π/5: a regular pentagon on the unit circle.
The fifth roots of unity, ω=e2iπ/5\omega = \eu^{2\iu\pi/5}: a regular pentagon on the unit circle.

Proposition 3.18 (Sum of the roots of unity)

For n2n \geq 2, the nn-th roots of unity sum to zero: k=0n1ωk=0\sum_{k=0}^{n-1} \omega^k = 0.

Proof. Geometric sum with ratio ω1\omega \neq 1: k=0n1ωk=ωn1ω1=0\sum_{k=0}^{n-1} \omega^k = \frac{\omega^n - 1}{\omega - 1} = 0 since ωn=1\omega^n = 1.

Example 3.19 (Reading real and imaginary parts)

Splitting k=0n1ωk=0\sum_{k=0}^{n-1}\omega^k = 0 into real and imaginary parts yields two trigonometric identities for free:

k=0n1cos2kπn=0,k=0n1sin2kπn=0(n2).\sum_{k=0}^{n-1}\cos\frac{2k\pi}n = 0, \qquad \sum_{k=0}^{n-1}\sin\frac{2k\pi}n = 0 \qquad (n \geq 2).

Geometrically: the centroid of a regular nn-gon inscribed in the unit circle is its center — the vertices balance out exactly. For n=5n = 5 the first identity gives 1+2cos2π5+2cos4π5=01 + 2\cos\frac{2\pi}5 + 2\cos\frac{4\pi}5 = 0 (pairing kk with nkn - k), the starting point of Exercise 3.8’s computation of cos2π5\cos\frac{2\pi}5.

Example 3.20 (Square roots in algebraic form)

To solve z2=3+4iz^2 = 3 + 4\iu without trigonometry, set z=x+iyz = x + \iu y:

x2y2=3,2xy=4,x2+y2=3+4i=5.x^2 - y^2 = 3, \qquad 2xy = 4, \qquad x^2 + y^2 = \abs{3 + 4\iu} = 5 .

Adding the first and last: x2=4x^2 = 4, so x=±2x = \pm 2, then y=2/x=±1y = 2/x = \pm 1 with the same sign pairing (xy=2>0xy = 2 > 0): z=±(2+i)z = \pm(2 + \iu). Combined with the usual formula, this solves every quadratic equation with complex coefficients (Exercise 3.7).

3.4 Complex numbers and plane geometry

Proposition 3.21 (Geometric dictionary)

Identify the point M(x,y)M(x, y) of the plane with its affix z=x+iyz = x + \iu y. For distinct points A,B,CA, B, C of affixes a,b,ca, b, c:

  1. ba\abs{b - a} is the distance ABAB;
  2. argcaba\arg\dfrac{c - a}{b - a} is the angle between the vectors AB\vect{AB} and AC\vect{AC} (mod 2π2\pi);
  3. A,B,CA, B, C are aligned if and only if cabaR\dfrac{c - a}{b - a} \in \R; the lines ABAB and ACAC are perpendicular if and only if cabaiR\dfrac{c - a}{b - a} \in \iu\R.

Proof. (1) is the definition of the modulus applied to bab - a, the affix of AB\vect{AB}. (2): write ba=reiθb - a = r\eu^{\iu\theta}, ca=seiφc - a = s\eu^{\iu\varphi}; then caba=srei(φθ)\frac{c-a}{b-a} = \frac sr \eu^{\iu(\varphi - \theta)} has argument φθ\varphi - \theta, the angle from AB\vect{AB} to AC\vect{AC}. (3): alignment means angle 00 or π\pi, i.e. argument in πZ\pi\Z, i.e. the quotient is real; perpendicularity means angle ±π2\pm\frac\pi2, i.e. the quotient is purely imaginary. (The quotient is nonzero since CAC \neq A.)

Remark 3.22 (Interlude: C\C is the plane with a multiplication)

It is worth pausing on what makes this chapter possible at all: the plane R2\R^2 carries additions in every direction, but no God-given multiplication — and C\C is the plane equipped with one, in which multiplying by a fixed number rotates and scales. This one structure will be milked three more times in this volume. In Chapter 21, multiplication by a+iba + \iu b reappears as the 2×22 \times 2 matrix with rows (a,b)(a, -b) and (b,a)(b, a): complex arithmetic is a first, entirely concrete family of matrix products. In Chapter 23, the formula (zw)\Re(\conj z\,w) turns out to be the dot product, and z\abs z the Euclidean norm: the triangle inequality proved here is the model for the Cauchy–Schwarz story there. And in Chapter 24, a moving point is best written tz(t)t \mapsto z(t), so that velocity and acceleration become complex-valued derivatives — circular motion, for one, is simply z(t)=Reiωtz(t) = R\,\eu^{\iu\omega t}. One good multiplication, four chapters of dividends.

Proposition 3.23 (Maps zaz+bz \mapsto az + b)

Let aCa \in \C^*, bCb \in \C. The transformation f(z)=az+bf(z) = az + b of the plane:

  • is a translation if a=1a = 1;
  • otherwise has a unique fixed point ζ=b1a\zeta = \frac{b}{1-a}, and f(z)ζ=a(zζ)f(z) - \zeta = a\,(z - \zeta): ff is the rotation of center ζ\zeta and angle arga\arg a, composed with the scaling of center ζ\zeta and ratio a\abs a.

In particular zeiθzz \mapsto \eu^{\iu\theta} z is the rotation of angle θ\theta about the origin, and z\conj z is the reflection in the real axis.

Proof. If a=1a = 1, f(z)=z+bf(z) = z + b translates by the vector of affix bb. If a1a \neq 1, the fixed-point equation z=az+bz = az + b has the unique solution ζ=b1a\zeta = \frac{b}{1-a}, and then f(z)ζ=az+b(aζ+b)=a(zζ)f(z) - \zeta = az + b - (a\zeta + b) = a(z - \zeta). Writing a=aeiargaa = \abs a\, \eu^{\iu\arg a}, multiplication by aa scales distances to ζ\zeta by a\abs a and adds arga\arg a to angles at ζ\zeta, which is the announced composition.

Example 3.24 (Classifying a map zaz+bz \mapsto az+b)

Take f(z)=iz+1f(z) = \iu z + 1. Here a=i1a = \iu \neq 1: the fixed point is

ζ=b1a=11i=1+i2,\zeta = \frac{b}{1 - a} = \frac1{1 - \iu} = \frac{1 + \iu}{2},

and since a=1\abs a = 1, there is no scaling at all: ff is the pure rotation of center 1+i2\frac{1+\iu}2 and angle argi=π2\arg \iu = \frac\pi2. Check: f(ζ)=i1+i2+1=i12+1=1+i2=ζf(\zeta) = \iu\,\frac{1+\iu}2 + 1 = \frac{\iu - 1}2 + 1 = \frac{1 + \iu}2 = \zeta, and f(0)=1f(0) = 1, f(1)=1+if(1) = 1 + \iu, f(1+i)=i(1+i)+1=if(1 + \iu) = \iu\,(1 + \iu) + 1 = \iu: the four points 0,1,1+i,i0, 1, 1+\iu, \iu of the unit square cycle around their center ζ\zeta, one quarter turn at a time — exactly what a rotation by π2\frac\pi2 about the square’s center must do.

The map f(z) = z + 1 of : a quarter-turn rotation about the fixed point = 1+ 2. The vertices of the unit square cycle 0 1 1+ 0; no point moves along a straight line, yet the whole square is rigidly turned.
The map f(z)=iz+1f(z) = \iu z + 1 of Example 3.24: a quarter-turn rotation about the fixed point ζ=1+i2\zeta = \frac{1+\iu}2. The vertices of the unit square cycle 011+ii00 \to 1 \to 1{+}\iu \to \iu \to 0; no point moves along a straight line, yet the whole square is rigidly turned.

Remark 3.25 (Common pitfalls with moduli and arguments)

  1. No inequalities in C\C. Writing zwz \leq w for non-real numbers is meaningless; only moduli, real parts and imaginary parts can be compared.
  2. z\sqrt{\phantom z} is reserved for R+\R_+. Every nonzero complex number has two square roots and neither is privileged: write “let δ\delta be a square root of Δ\Delta” (computed as in Example 3.20), never Δ\sqrt\Delta — the rule ab=ab\sqrt{ab} = \sqrt a\sqrt b already fails at a=b=1a = b = -1.
  3. Arguments live modulo 2π2\pi. From eiα=eiβ\eu^{\iu\alpha} = \eu^{\iu\beta} conclude αβ(mod2π)\alpha \equiv \beta \pmod{2\pi}, not α=β\alpha = \beta; forgetting the 2kπ2k\pi is how solution sets of zn=az^n = a lose n1n - 1 of their nn elements.
  4. z+w\abs{z + w} is not z+w\abs z + \abs w. Equality in the triangle inequality is the exceptional aligned case (Proposition 3.2 (4)); in general the modulus of a sum must be estimated, not computed.

Example 3.26

A,B,CA, B, C (affixes a,b,ca, b, c, pairwise distinct) form an equilateral triangle with the vertices in direct (counterclockwise) order if and only if caba=eiπ/3\frac{c - a}{b - a} = \eu^{\iu\pi/3}: the rotation of center AA and angle π3\frac\pi3 sends BB to CC. Both orientations together are captured by the symmetric equation a2+b2+c2=ab+bc+caa^2 + b^2 + c^2 = ab + bc + ca (Exercise 3.10).

Remark 3.27 (Where this chapter is used)

The exponential form is the single most reused computational device of the volume. Roots of unity become the standard example of a cyclic group in Chapter 7 and drive the factorization of Xn1X^n - 1 in Chapter 8; grouping conjugate roots there produces the real factorizations used by partial fractions in Chapter 9. Linearization (Method 3.11) is the standard preparation for integrating trigonometric powers in Chapter 15, and the characteristic equations of Chapter 5 have complex roots whose real and imaginary parts produce the oscillating solutions eλtcosωt\eu^{\lambda t}\cos\omega t. The geometric dictionary returns in matrix clothing: rotations and similarities become the orthogonal matrices of Chapters 21 and 23, and parametrized curves in Chapter 24 are often best written as maps tz(t)Ct \mapsto z(t) \in \C.

3.5 Exercises

Exercise 3.1

Put in exponential form: 1+i1 + \iu;   3i\;\sqrt 3 - \iu;   5\;-5;   1+i3i\;\dfrac{1 + \iu}{\sqrt 3 - \iu}. Deduce cos5π12\cos\frac{5\pi}{12} and sin5π12\sin\frac{5\pi}{12}.

Solution

Solution of Exercise 3.1.

1+i=2eiπ/41 + \iu = \sqrt 2\, \eu^{\iu\pi/4};   3i=2eiπ/6\;\sqrt 3 - \iu = 2\, \eu^{-\iu\pi/6};   5=5eiπ\;-5 = 5\, \eu^{\iu\pi};

1+i3i=22ei(π/4+π/6)=22e5iπ/12.\frac{1 + \iu}{\sqrt 3 - \iu} = \frac{\sqrt 2}{2}\, \eu^{\iu(\pi/4 + \pi/6)} = \frac{\sqrt 2}{2}\, \eu^{5\iu\pi/12}.

Computing the same quotient algebraically (multiply by the conjugate):

(1+i)(3+i)4=(31)+i(3+1)4.\frac{(1 + \iu)(\sqrt 3 + \iu)}{4} = \frac{(\sqrt 3 - 1) + \iu(\sqrt 3 + 1)}{4}.

Identifying with 22(cos5π12+isin5π12)\frac{\sqrt 2}{2}(\cos\frac{5\pi}{12} + \iu\sin\frac{5\pi}{12}):

cos5π12=624,sin5π12=6+24.\cos\frac{5\pi}{12} = \frac{\sqrt 6 - \sqrt 2}{4}, \qquad \sin\frac{5\pi}{12} = \frac{\sqrt 6 + \sqrt 2}{4}.

Exercise 3.2

Compute (1+i)20(1 + \iu)^{20}. For which nNn \in \N is (1+i)n(1 + \iu)^n a real number?

Solution

Solution of Exercise 3.2.

(1+i)2=2i(1+\iu)^2 = 2\iu, so (1+i)20=(2i)10=210i10=1024×(1)=1024(1+\iu)^{20} = (2\iu)^{10} = 2^{10}\,\iu^{10} = 1024 \times (-1) = -1024.

In exponential form (1+i)n=2n/2einπ/4(1+\iu)^n = 2^{n/2}\, \eu^{\iu n\pi/4}, real if and only if sinnπ4=0\sin\frac{n\pi}{4} = 0, i.e. 4n4 \mid n. So (1+i)nR(1+\iu)^n \in \R exactly for the multiples of 44 (value (4)n/4(-4)^{n/4}).

Exercise 3.3

Describe geometrically the set of zCz \in \C such that: z2=z+i\abs{z - 2} = \abs{z + \iu};   z1=2\;\abs{z - 1} = 2;   z1z+1iR\;\dfrac{z - 1}{z + 1} \in \iu\R (for z1z \neq -1).

Solution

Solution of Exercise 3.3.

z2=z+i\abs{z - 2} = \abs{z + \iu}: equidistance from the points 22 and i-\iu — the perpendicular bisector of the segment joining (2,0)(2, 0) and (0,1)(0, -1).

z1=2\abs{z - 1} = 2: the circle of center 11 and radius 22.

z1z+1iR\frac{z-1}{z+1} \in \iu\R: by Proposition 3.21 (3), the points M(z)M(z), A(1)A(1), B(1)B(-1) satisfy: lines MAMA and MBMB perpendicular (or z=1z = 1, where the quotient is 0iR0 \in \iu\R). The set is the circle of diameter [1,1][-1, 1] (the unit circle), minus the point 1-1 where the quotient is undefined. Check by computation: z=eiθz = \eu^{\iu\theta} gives z1z+1=eiθ/2(eiθ/2eiθ/2)eiθ/2(eiθ/2+eiθ/2)=itanθ2iR\frac{z-1}{z+1} = \frac{\eu^{\iu\theta/2}(\eu^{\iu\theta/2} - \eu^{-\iu\theta/2})}{\eu^{\iu\theta/2}(\eu^{\iu\theta/2} + \eu^{-\iu\theta/2})} = \iu\tan\frac\theta2 \in \iu\R.

Exercise 3.4

Linearize sin4θ\sin^4\theta, and expand cos4θ\cos 4\theta as a polynomial in cosθ\cos\theta.

Solution

Solution of Exercise 3.4.

Linearization:

sin4θ=(eiθeiθ2i) ⁣4=e4iθ4e2iθ+64e2iθ+e4iθ16=cos4θ4cos2θ+38.\sin^4\theta = \Bigl(\frac{\eu^{\iu\theta} - \eu^{-\iu\theta}}{2\iu}\Bigr)^{\!4} = \frac{\eu^{4\iu\theta} - 4\eu^{2\iu\theta} + 6 - 4\eu^{-2\iu\theta} + \eu^{-4\iu\theta}}{16} = \frac{\cos 4\theta - 4\cos 2\theta + 3}{8}.

Expansion: by de Moivre, cos4θ=((c+is)4)=c46c2s2+s4\cos 4\theta = \Re\bigl((c + \iu s)^4\bigr) = c^4 - 6c^2 s^2 + s^4 with c=cosθc = \cos\theta, s=sinθs = \sin\theta; substituting s2=1c2s^2 = 1 - c^2:

cos4θ=c46c2(1c2)+(1c2)2=8c48c2+1.\cos 4\theta = c^4 - 6c^2(1 - c^2) + (1 - c^2)^2 = 8c^4 - 8c^2 + 1 .

Exercise 3.5

Solve z3=8iz^3 = 8\iu and place the solutions on a picture. Solve z4=4z^4 = -4 and factor X4+4X^4 + 4 into two real quadratic polynomials.

Solution

Solution of Exercise 3.5.

8i=8eiπ/28\iu = 8\,\eu^{\iu\pi/2}, so the cube roots are 2ei(π/6+2kπ/3)2\,\eu^{\iu(\pi/6 + 2k\pi/3)}, k=0,1,2k = 0, 1, 2:

z0=2eiπ/6=3+i,z1=2e5iπ/6=3+i,z2=2e3iπ/2=2i:z_0 = 2\eu^{\iu\pi/6} = \sqrt 3 + \iu,\quad z_1 = 2\eu^{5\iu\pi/6} = -\sqrt 3 + \iu,\quad z_2 = 2\eu^{3\iu\pi/2} = -2\iu :

an equilateral triangle on the circle of radius 22.

4=4eiπ-4 = 4\eu^{\iu\pi}, so z4=4z^4 = -4 has solutions 2ei(π/4+kπ/2)\sqrt 2\, \eu^{\iu(\pi/4 + k\pi/2)}:   1+i\;1 + \iu, 1+i-1 + \iu, 1i-1 - \iu, 1i1 - \iu. Pairing conjugate roots:

X4+4=(X22X+2)(X2+2X+2),X^4 + 4 = \bigl(X^2 - 2X + 2\bigr)\bigl(X^2 + 2X + 2\bigr),

since (X(1+i))(X(1i))=X22X+2(X - (1+\iu))(X - (1-\iu)) = X^2 - 2X + 2 and similarly for the other pair. (Expand to check.)

Exercise 3.6 ★★

For θR\theta \in \R with eiθ1\eu^{\iu\theta} \neq 1 and nNn \in \N, compute

Cn=k=0ncoskθandSn=k=0nsinkθC_n = \sum_{k=0}^{n} \cos k\theta \qquad\text{and}\qquad S_n = \sum_{k=0}^{n} \sin k\theta

by summing the geometric series keikθ\sum_k \eu^{\iu k\theta} and using the half-angle factorization.

Solution

Solution of Exercise 3.6.

Cn+iSn=k=0neikθ=ei(n+1)θ1eiθ1C_n + \iu S_n = \sum_{k=0}^{n} \eu^{\iu k\theta} = \frac{\eu^{\iu(n+1)\theta} - 1}{\eu^{\iu\theta} - 1} (geometric sum, ratio eiθ1\eu^{\iu\theta} \neq 1). Half-angle factorization (Method 3.11 (4)) on numerator and denominator:

2isin(n+1)θ2ei(n+1)θ/22isinθ2eiθ/2=sin(n+1)θ2sinθ2einθ/2.\frac{2\iu\sin\frac{(n+1)\theta}{2}\, \eu^{\iu(n+1)\theta/2}} {2\iu\sin\frac{\theta}{2}\, \eu^{\iu\theta/2}} = \frac{\sin\frac{(n+1)\theta}{2}}{\sin\frac{\theta}{2}}\, \eu^{\iu n\theta/2}.

Taking real and imaginary parts:

Cn=sin(n+1)θ2sinθ2cosnθ2,Sn=sin(n+1)θ2sinθ2sinnθ2.C_n = \frac{\sin\frac{(n+1)\theta}{2}}{\sin\frac{\theta}{2}} \cos\frac{n\theta}{2}, \qquad S_n = \frac{\sin\frac{(n+1)\theta}{2}}{\sin\frac{\theta}{2}} \sin\frac{n\theta}{2}.

Exercise 3.7 ★★

Solve in C\C: z2(3+4i)z+(1+5i)=0z^2 - (3 + 4\iu) z + (-1 + 5\iu) = 0. (Compute the discriminant, extract its square roots as in Example 3.20.)

Solution

Solution of Exercise 3.7.

Discriminant: Δ=(3+4i)24(1+5i)=9+24i16+420i=3+4i\Delta = (3 + 4\iu)^2 - 4(-1 + 5\iu) = 9 + 24\iu - 16 + 4 - 20\iu = -3 + 4\iu. Square roots of 3+4i-3 + 4\iu: solve x2y2=3x^2 - y^2 = -3, 2xy=42xy = 4, x2+y2=5x^2 + y^2 = 5; then x2=1x^2 = 1, y=2/xy = 2/x, same signs: δ=±(1+2i)\delta = \pm(1 + 2\iu). Hence

z=(3+4i)±(1+2i)2{2+3i,  1+i}.z = \frac{(3 + 4\iu) \pm (1 + 2\iu)}{2} \in \{\, 2 + 3\iu,\; 1 + \iu \,\}.

Check: sum =3+4i= 3 + 4\iu and product (2+3i)(1+i)=1+5i(2+3\iu)(1+\iu) = -1 + 5\iu, as required by the coefficients.

Exercise 3.8 ★★

Let ω=e2iπ/5\omega = \eu^{2\iu\pi/5}.

  1. Justify 1+ω+ω2+ω3+ω4=01 + \omega + \omega^2 + \omega^3 + \omega^4 = 0.
  2. Set u=ω+ω4u = \omega + \omega^4 and v=ω2+ω3v = \omega^2 + \omega^3. Compute u+vu + v and uvuv, and deduce that uu and vv are the roots of X2+X1X^2 + X - 1.
  3. Conclude that cos2π5=514\cos\frac{2\pi}{5} = \frac{\sqrt 5 - 1}{4}.
Solution

Solution of Exercise 3.8.

  1. Proposition 3.18 with n=5n = 5.
  2. u+v=ω+ω2+ω3+ω4=1u + v = \omega + \omega^2 + \omega^3 + \omega^4 = -1 by (1). For the product, expand and reduce exponents modulo 55:

    uv=(ω+ω4)(ω2+ω3)=ω3+ω4+ω6+ω7=ω3+ω4+ω+ω2=1.uv = (\omega + \omega^4)(\omega^2 + \omega^3) = \omega^3 + \omega^4 + \omega^6 + \omega^7 = \omega^3 + \omega^4 + \omega + \omega^2 = -1 .

    So uu and vv have sum 1-1 and product 1-1: they are the two roots of X2+X1X^2 + X - 1.

  3. u=ω+ω=2cos2π5>0u = \omega + \conj\omega = 2\cos\frac{2\pi}{5} > 0 (the angle is acute), and the positive root of X2+X1X^2 + X - 1 is 1+52\frac{-1 + \sqrt 5}{2}. Hence cos2π5=514\cos\frac{2\pi}{5} = \frac{\sqrt 5 - 1}{4}.

Exercise 3.9 ★★

Prove that for all z,wCz, w \in \C (parallelogram identity):

z+w2+zw2=2z2+2w2,\abs{z + w}^2 + \abs{z - w}^2 = 2\abs z^2 + 2\abs w^2 ,

and interpret it geometrically in the parallelogram with vertices 0,z,w,z+w0, z, w, z + w.

Solution

Solution of Exercise 3.9.

Expand both squares as in the proof of Proposition 3.2 (4):

z+w2=z2+w2+2(zw),zw2=z2+w22(zw),\abs{z + w}^2 = \abs z^2 + \abs w^2 + 2\Re(z\conj w), \qquad \abs{z - w}^2 = \abs z^2 + \abs w^2 - 2\Re(z\conj w),

and add. Geometrically, z+w\abs{z+w} and zw\abs{z-w} are the lengths of the two diagonals of the parallelogram with vertices 0,z,z+w,w0, z, z+w, w, while z\abs z and w\abs w are the side lengths: the sum of the squares of the diagonals equals the sum of the squares of the four sides.

Exercise 3.10 ★★★

Prove that three pairwise distinct points of affixes a,b,ca, b, c form an equilateral triangle (either orientation) if and only if

a2+b2+c2=ab+bc+ca.a^2 + b^2 + c^2 = ab + bc + ca .

Hint: the direct case is caba=j2\frac{c-a}{b-a} = -j^2 and the indirect case caba=j\frac{c-a}{b-a} = -j, where j=e2iπ/3j = \eu^{2\iu\pi/3} satisfies j2+j+1=0j^2 + j + 1 = 0; or factor a+jb+j2ca + jb + j^2c and a+j2b+jca + j^2 b + jc.

Solution

Solution of Exercise 3.10.

Let j=e2iπ/3j = \eu^{2\iu\pi/3}, so j2+j+1=0j^2 + j + 1 = 0 and eiπ/3=j2\eu^{\iu\pi/3} = -j^2, eiπ/3=j\eu^{-\iu\pi/3} = -j. The triangle is equilateral direct iff ca=j2(ba)c - a = -j^2 (b - a), indirect iff ca=j(ba)c - a = -j(b - a) (Example 3.26).

Consider P=a+jb+j2cP = a + jb + j^2 c and Q=a+j2b+jcQ = a + j^2 b + jc. Using 1+j2=j1 + j^2 = -j and j3=1j^3 = 1:

ca+j2(ba)=c+j2b(1+j2)a=c+j2b+ja=j(a+jb+j2c)=jP,c - a + j^2(b - a) = c + j^2 b - (1 + j^2)\,a = c + j^2 b + ja = j\,(a + jb + j^2 c) = jP,

so the direct case reads jP=0    P=0jP = 0 \iff P = 0; the same computation with jj in place of j2j^2 gives ca+j(ba)=j2Qc - a + j(b - a) = j^2 Q, so the indirect case reads Q=0Q = 0. Hence: equilateral (either orientation)     PQ=0\iff PQ = 0. Expanding, with j+j2=1j + j^2 = -1:

PQ=a2+b2+c2+(j+j2)(ab+bc+ca)=a2+b2+c2(ab+bc+ca).PQ = a^2 + b^2 + c^2 + (j + j^2)(ab + bc + ca) = a^2 + b^2 + c^2 - (ab + bc + ca).

So the triangle is equilateral if and only if a2+b2+c2=ab+bc+caa^2 + b^2 + c^2 = ab + bc + ca.

Exercise 3.11 ★★★

For nNn \in \N^*, evaluate P=k=1n1(1ωk)P = \prod_{k=1}^{n-1} \bigl(1 - \omega^k\bigr) where ω=e2iπ/n\omega = \eu^{2\iu\pi/n}. Hint: Xn1=k=0n1(Xωk)X^n - 1 = \prod_{k=0}^{n-1} (X - \omega^k); divide by X1X - 1 and evaluate at X=1X = 1. Deduce k=1n1sinkπn=n2n1\prod_{k=1}^{n-1} \sin\frac{k\pi}{n} = \dfrac{n}{2^{n-1}}.

Solution

Solution of Exercise 3.11.

Since the ωk\omega^k, k=0,,n1k = 0, \dots, n-1, are exactly the nn roots of Xn1X^n - 1 (Theorem 3.14), and the polynomial is monic:

Xn1=k=0n1(Xωk)=(X1)k=1n1(Xωk).X^n - 1 = \prod_{k=0}^{n-1} (X - \omega^k) = (X - 1) \prod_{k=1}^{n-1} (X - \omega^k).

Dividing by X1X - 1:   1+X++Xn1=k=1n1(Xωk)\;1 + X + \dots + X^{n-1} = \prod_{k=1}^{n-1} (X - \omega^k). Evaluating at X=1X = 1 gives P=nP = n.

Now 1ωk=eikπ/n(eikπ/neikπ/n)=2ieikπ/nsinkπn1 - \omega^k = -\eu^{\iu k\pi/n}\bigl(\eu^{\iu k\pi/n} - \eu^{-\iu k\pi/n}\bigr) = -2\iu\,\eu^{\iu k\pi/n} \sin\frac{k\pi}{n}, so taking moduli in P=nP = n (each sinkπn>0\sin\frac{k\pi}n > 0 for 1kn11 \leq k \leq n-1):

n=P=k=1n12sinkπn=2n1k=1n1sinkπn,hencek=1n1sinkπn=n2n1.n = \abs P = \prod_{k=1}^{n-1} 2\sin\frac{k\pi}{n} = 2^{n-1} \prod_{k=1}^{n-1} \sin\frac{k\pi}{n}, \qquad\text{hence}\qquad \prod_{k=1}^{n-1} \sin\frac{k\pi}{n} = \frac{n}{2^{n-1}} .

Exercise 3.12 ★★

Let n2n \geq 2. Solve the equation (z+1)n=(z1)n(z + 1)^n = (z - 1)^n in C\C: show that it has exactly n1n - 1 solutions, all purely imaginary, namely

zk=icoskπnsinkπn,k=1,,n1.z_k = -\iu\,\frac{\cos\frac{k\pi}{n}}{\sin\frac{k\pi}{n}}, \qquad k = 1, \dots, n - 1 .

Hint: z=1z = 1 is not a solution, so divide and use the roots of unity; then apply the half-angle factorization of Method 3.11.

Solution

Solution of Exercise 3.12.

z=1z = 1 is not a solution (2n02^n \neq 0), so the equation is equivalent to (z+1z1)n=1\bigl(\frac{z+1}{z-1}\bigr)^n = 1, i.e. z+1z1=ωk\frac{z+1}{z-1} = \omega^k with ω=e2iπ/n\omega = \eu^{2\iu\pi/n} and k{0,,n1}k \in \{0, \dots, n-1\}. The value k=0k = 0 is excluded (z+1=z1z + 1 = z - 1 is impossible). For 1kn11 \leq k \leq n - 1, solving z+1=ωk(z1)z + 1 = \omega^k(z - 1) gives z(1ωk)=1ωkz(1 - \omega^k) = -1 - \omega^k, so with φ=2kπn\varphi = \frac{2k\pi}{n} and the half-angle factorizations 1+eiφ=2cosφ2eiφ/21 + \eu^{\iu\varphi} = 2\cos\frac\varphi2\, \eu^{\iu\varphi/2}, eiφ1=2isinφ2eiφ/2\eu^{\iu\varphi} - 1 = 2\iu\sin\frac\varphi2 \,\eu^{\iu\varphi/2}:

zk=1+ωkωk1=2coskπn2isinkπn=icoskπnsinkπn,z_k = \frac{1 + \omega^k}{\omega^k - 1} = \frac{2\cos\frac{k\pi}{n}}{2\iu\,\sin\frac{k\pi}{n}} = -\iu\,\frac{\cos\frac{k\pi}{n}}{\sin\frac{k\pi}{n}} ,

purely imaginary as claimed. The map tcost/sintt \mapsto \cos t/\sin t is injective on (0,π)\intoo0\pi (it is strictly decreasing), so the n1n - 1 values zkz_k are pairwise distinct: the equation, of degree n1n - 1 once expanded (the znz^n terms cancel), has exactly these n1n - 1 solutions.

3.6 Problem: Napoleon’s theorem

Problem 3.1

Erect an equilateral triangle outward on each side of an arbitrary triangle: the three centers of those triangles always form an equilateral triangle. This is Napoleon’s theorem — a statement with no visible reason to be true, which the algebra of j=e2iπ/3j = \eu^{2\iu\pi/3} proves in three lines of computation. This problem builds the full toolkit (rotations, direct similarities, the jj-criterion of Exercise 3.10), proves Napoleon’s theorem inner and outer, locates the degenerate case, and closes with a second jewel of the same school: van Schooten’s theorem on equilateral triangles and Ptolemy’s inequality. Throughout, points of the plane are identified with their affixes.

Part I — The number jj and rotations.

  1. Compute the algebraic form of jj, then j2j^2, j3j^3, 1+j+j21 + j + j^2, j\conj j and j1j^{-1}. Place 11, jj, j2j^2 on a sketch of the unit circle.
  2. Show that the rotation of center aa and angle θ\theta is r(z)=a+eiθ(za)r(z) = a + \eu^{\iu\theta}(z - a), and that the composition of two rotations, of angles θ\theta and θ\theta', is a rotation of angle θ+θ\theta + \theta' if θ+θ2πZ\theta + \theta' \notin 2\pi\Z, and a translation otherwise. (Use Proposition 3.23.)
  3. Let bcb \neq c. Show that there are exactly two points pp making (b,c,p)(b, c, p) equilateral, namely p=b+e±iπ/3(cb)p = b + \eu^{\pm\iu\pi/3}(c - b). For a direct triangle (a,b,c)(a, b, c), check on the example a=0a = 0, b=1b = 1, c=ic = \iu that the choice eiπ/3\eu^{-\iu\pi/3} is the one lying on the far side of the line BCBC from aa — the outward apex.
  4. From the solution of Exercise 3.10, (a,b,c)(a, b, c) is direct equilateral iff a+jb+j2c=0a + jb + j^2c = 0. Prove the two rotation identities

    b+jc+j2a=j2(a+jb+j2c),c+ja+j2b=j(a+jb+j2c),b + jc + j^2a = j^2\,(a + jb + j^2c), \qquad c + ja + j^2b = j\,(a + jb + j^2c),

    and deduce that the criterion is invariant under cyclic permutation of (a,b,c)(a, b, c).

  5. Show that if a+jb+j2c=0a + jb + j^2c = 0 and two of the three points coincide, then all three coincide. (So the criterion exactly characterizes: direct equilateral triangle, or a single point.)

Part II — Direct similarities.

  1. Show that the maps f(z)=az+bf(z) = az + b with aCa \in \C^* (direct similarities) are stable under composition and inversion: the composition of two of them, and the inverse of any of them, is again of this form. (In the language of Chapter 7: they form a group.)
  2. Given z1z2z_1 \neq z_2 and w1w2w_1 \neq w_2, show there is a unique direct similarity ff with f(z1)=w1f(z_1) = w_1 and f(z2)=w2f(z_2) = w_2, and give aa and bb explicitly.
  3. Show that a direct similarity f(z)=az+bf(z) = az + b multiplies all distances by a\abs a and preserves the shape caba\frac{c - a'}{b' - a'} of any triangle (a,b,c)(a', b', c') (numerator and denominator are both multiplied by aa). Deduce that two triangles are directly similar exactly when their shapes are equal.
  4. Determine completely the direct similarity with f(0)=1f(0) = 1 and f(1)=if(1) = \iu: give aa, bb, the fixed point, the ratio and the angle.
  5. Let α+β=1\alpha + \beta = 1. Show that the “weighted point” g(a,b)=αa+βbg(a', b') = \alpha a' + \beta b' commutes with every direct similarity: f(αa+βb)=αf(a)+βf(b)f(\alpha a' + \beta b') = \alpha f(a') + \beta f(b'). Deduce that centroids, midpoints, and the Napoleon centers below are all transported by similarities — the algebraic license behind every “without loss of generality, place the circumcircle at the unit circle” argument.

Part III — Napoleon’s theorem. Let (a,b,c)(a, b, c) be a direct triangle. On each side erect the outward equilateral triangle (question 3) and let nan_a, nbn_b, ncn_c be the centers (centroids) of the triangles erected on [b,c][b, c], [c,a][c, a], [a,b][a, b] respectively.

  1. Show that

    na=αb+βc,nb=αc+βa,nc=αa+βb,withα=3+i36, β=α.n_a = \alpha b + \beta c, \qquad n_b = \alpha c + \beta a, \qquad n_c = \alpha a + \beta b, \qquad\text{with}\quad \alpha = \frac{3 + \iu\sqrt3}{6},\ \beta = \conj\alpha .
  2. Check α+β=1\alpha + \beta = 1, and deduce that the triangle (na,nb,nc)(n_a, n_b, n_c) has the same centroid as (a,b,c)(a, b, c).
  3. Using the identities of question 4, show that

    na+jnb+j2nc=(αj2+βj)(a+jb+j2c).n_a + j\,n_b + j^2 n_c = (\alpha j^2 + \beta j)\,(a + jb + j^2c) .
  4. Compute αj+β\alpha j + \beta and conclude: αj2+βj=j(αj+β)=0\alpha j^2 + \beta j = j(\alpha j + \beta) = 0, so na+jnb+j2nc=0n_a + j n_b + j^2 n_c = 0 for every triangle: the outer Napoleon triangle is direct equilateral (or a point). Napoleon’s theorem is proved.
  5. Erect the equilateral triangles inward instead (choice e+iπ/3\eu^{+\iu\pi/3} in question 3) and let ma,mb,mcm_a, m_b, m_c be their centers. Show ma=βb+αcm_a = \beta b + \alpha c (and cyclically), then prove ma+j2mb+jmc=0m_a + j^2 m_b + j\,m_c = 0: the inner Napoleon triangle is equilateral too, with the opposite orientation.
  6. Locate the degeneracy: show that na=nb=ncn_a = n_b = n_c happens exactly when a+j2b+jc=0a + j^2 b + jc = 0, i.e. when (a,b,c)(a, b, c) is an indirect equilateral triangle. (Two points among na,nb,ncn_a, n_b, n_c coincide iff all three do, by question 5; then use na+j2nb+jnc=j(α+βj)(a+j2b+jc)n_a + j^2 n_b + j n_c = j(\alpha + \beta j)(a + j^2b + jc) and check α+βj0\alpha + \beta j \neq 0.)
  7. Carry out the whole computation on the triangle a=0a = 0, b=1b = 1, c=ic = \iu: give na,nb,ncn_a, n_b, n_c exactly, and verify by direct computation of the three squared side lengths that the triangle is equilateral, of squared side 2+33\frac{2 + \sqrt3}{3}.

Part IV — Ptolemy and van Schooten.

  1. Prove the identity, valid for all complex a,b,c,da, b, c, d:

    (ab)(cd)+(ad)(bc)=(ac)(bd).(a - b)(c - d) + (a - d)(b - c) = (a - c)(b - d) .
  2. Deduce Ptolemy’s inequality: for any four points A,B,C,DA, B, C, D,

    ACBD    ABCD+ADBC,AC \cdot BD \;\leq\; AB \cdot CD + AD \cdot BC ,

    with equality if and only if (ab)(cd)(a-b)(c-d) and (ad)(bc)(a-d)(b-c) lie on a common ray from 00.

  3. Show that for θ,φR\theta, \varphi \in \R, eiθeiφ=2sinθφ2\abs{\eu^{\iu\theta} - \eu^{\iu\varphi}} = 2\,\abs{\sin\frac{\theta - \varphi}2}.
  4. (van Schooten’s theorem) Let (a,b,c)=(1,j,j2)(a, b, c) = (1, j, j^2) — by Part II this loses no generality among direct equilateral triangles — and let p=eiθp = \eu^{\iu\theta} with θ(2π/3,4π/3)\theta \in \intoo{2\pi/3}{4\pi/3}, a point of the circumcircle on the arc BCBC not containing AA. Prove

    PA=PB+PC.PA = PB + PC .

    (Express the three distances with question 20 and use a sum-to-product formula.)

  5. Verify at p=1p = -1 that this is exactly the equality case of Ptolemy’s inequality for the cyclic order A,B,P,CA, B, P, C: compute (ab)(pc)(a - b)(p - c) and (ac)(bp)(a - c)(b - p) and check that their ratio is a positive real.

Part V — Synthesis.

  1. Compute the outer Napoleon triangle of the equilateral triangle (1,j,j2)(1, j, j^2) itself, and describe the result geometrically.
  2. Where exactly did the problem use: (i) multiplication as rotation; (ii) the group structure of similarities and the shape invariant; (iii) the half-angle factorization of Method 3.11? One sentence each.
  3. In a short paragraph, articulate the moral of the problem: what does the dictionary between plane geometry and the algebra of C\C buy, what does it cost, and which of the two proofsteps — the identity αj+β=0\alpha j + \beta = 0 or the classical picture — explains Napoleon’s theorem better in your view? Mention one place where the dictionary will reappear in matrix form later in this volume.
Solution

Solution of Problem 3.1.

1. j=cos2π3+isin2π3=12+i32j = \cos\frac{2\pi}3 + \iu\sin\frac{2\pi}3 = -\frac12 + \iu\frac{\sqrt3}2; j2=e4iπ/3=12i32=jj^2 = \eu^{4\iu\pi/3} = -\frac12 - \iu\frac{\sqrt3}2 = \conj j; j3=1j^3 = 1; 1+j+j2=01 + j + j^2 = 0 (sum of the cube roots of unity, Proposition 3.18); j1=j2j^{-1} = j^2 (since jj2=1j \cdot j^2 = 1). On the unit circle, 11, jj, j2j^2 are the vertices of a direct equilateral triangle.

2. The rotation of center aa and angle θ\theta fixes aa and turns every vector issued from aa by θ\theta: r(z)a=eiθ(za)r(z) - a = \eu^{\iu\theta}(z - a), i.e. r(z)=a+eiθ(za)r(z) = a + \eu^{\iu\theta}(z - a). Composing r(z)=a+eiθ(za)r(z) = a + \eu^{\iu\theta}(z-a) and r(z)=a+eiθ(za)r'(z) = a' + \eu^{\iu\theta'}(z-a'):

rr(z)=ei(θ+θ)z+constant,r' \circ r\,(z) = \eu^{\iu(\theta + \theta')} z + \text{constant},

a map of the form Az+BAz + B with A=ei(θ+θ)A = \eu^{\iu(\theta+\theta')} of modulus 11. By Proposition 3.23, it is a rotation of angle argA=θ+θ\arg A = \theta + \theta' if A1A \neq 1, and a translation if A=1A = 1, i.e. if θ+θ2πZ\theta + \theta' \in 2\pi\Z.

3. (b,c,p)(b, c, p) is equilateral iff pb=cb=pc\abs{p - b} = \abs{c - b} = \abs{p - c}. Writing q=pbcbq = \frac{p - b}{c - b}, the first equality says q=1\abs q = 1 and the second q1=1\abs{q - 1} = 1; together, q=e±iπ/3q = \eu^{\pm\iu\pi/3} (the two intersection points of the circles q=1\abs q = 1 and q1=1\abs{q - 1} = 1 are 12±i32\frac12 \pm \iu\frac{\sqrt3}2). Hence p=b+e±iπ/3(cb)p = b + \eu^{\pm\iu\pi/3}(c - b). For a=0a = 0, b=1b = 1, c=ic = \iu (a direct triangle): the choice eiπ/3\eu^{-\iu\pi/3} gives

p=1+(12i32)(i1)=1+32(1+i)1.37+1.37i,p = 1 + \Bigl(\tfrac12 - \iu\tfrac{\sqrt3}2\Bigr)(\iu - 1) = \tfrac{1 + \sqrt3}2\,(1 + \iu) \approx 1.37 + 1.37\,\iu ,

which lies on the far side of the line BCBC (x+y=1x + y = 1) from a=0a = 0: outward. The choice e+iπ/3\eu^{+\iu\pi/3} gives p0.370.37ip \approx -0.37 - 0.37\,\iu, on the same side as aa: inward.

4. Multiply P=a+jb+j2cP = a + jb + j^2c by j2j^2: j2P=j2a+j3b+j4c=b+jc+j2aj^2 P = j^2 a + j^3 b + j^4 c = b + jc + j^2 a (using j3=1j^3 = 1, j4=jj^4 = j); and by jj: jP=ja+j2b+cjP = ja + j^2 b + c. These are the two identities. If P=0P = 0 then j2P=jP=0j^2 P = jP = 0: the criterion holds for (b,c,a)(b, c, a) and (c,a,b)(c, a, b) as well — cyclic invariance (as it must be: an equilateral triangle does not care which vertex is listed first).

5. If a=ba = b: 0=a(1+j)+j2c=j2a+j2c0 = a(1 + j) + j^2 c = -j^2 a + j^2 c (using 1+j=j21 + j = -j^2), so c=ac = a. If b=cb = c: 0=a+b(j+j2)=ab0 = a + b(j + j^2) = a - b, so a=ba = b. If a=ca = c: 0=a(1+j2)+jb=ja+jb0 = a(1 + j^2) + jb = -ja + jb, so a=ba = b. In each case all three points coincide.

6. (az+b)(az+b)=aaz+(ab+b)(a'z + b') \circ (az + b) = a'a\,z + (a'b + b') with aa0a'a \neq 0: same form. The inverse of zaz+bz \mapsto az + b is z1azbaz \mapsto \frac1a z - \frac ba, again of the same form. With the identity map as neutral element, the direct similarities form a group under composition.

7. f(z)=az+bf(z) = az + b satisfies f(z1)=w1f(z_1) = w_1, f(z2)=w2f(z_2) = w_2 iff az1+b=w1a z_1 + b = w_1 and az2+b=w2a z_2 + b = w_2; subtracting, a(z2z1)=w2w1a(z_2 - z_1) = w_2 - w_1, so

a=w2w1z2z1  (0),b=w1az1a = \frac{w_2 - w_1}{z_2 - z_1} \;(\neq 0), \qquad b = w_1 - a z_1

are forced, and conversely this choice works: existence and uniqueness.

8. f(z)f(w)=a(zw)=azw\abs{f(z) - f(w)} = \abs{a(z - w)} = \abs a\,\abs{z - w}: all distances are multiplied by a\abs a. For the shape:

f(c)f(a)f(b)f(a)=a(ca)a(ba)=caba.\frac{f(c') - f(a')}{f(b') - f(a')} = \frac{a(c' - a')}{a(b' - a')} = \frac{c' - a'}{b' - a'} .

If two triangles (a,b,c)(a', b', c') and (a,b,c)(a'', b'', c'') have equal shapes, let ff be the unique direct similarity with f(a)=af(a') = a'' and f(b)=bf(b') = b'' (question 7); then the shape of (a,b,f(c))(a'', b'', f(c')) equals that of (a,b,c)(a', b', c'), hence that of (a,b,c)(a'', b'', c''), and the shape determines the third vertex from the first two: f(c)=cf(c') = c''. Conversely equal shapes follow from the displayed invariance.

9. b=f(0)=1b = f(0) = 1; a+b=f(1)=ia + b = f(1) = \iu gives a=i1a = \iu - 1. Ratio a=2\abs a = \sqrt2, angle arg(i1)=3π4\arg(\iu - 1) = \frac{3\pi}4. Fixed point:

ζ=b1a=12i=2+i5.\zeta = \frac{b}{1 - a} = \frac1{2 - \iu} = \frac{2 + \iu}5 .

So ff is the direct similarity of center 2+i5\frac{2+\iu}5, ratio 2\sqrt2 and angle 3π4\frac{3\pi}4.

10. With α+β=1\alpha + \beta = 1 and f(z)=az+bf(z) = az + b:

f(αa+βb)=aαa+aβb+b=α(aa+b)+β(ab+b)=αf(a)+βf(b),f(\alpha a' + \beta b') = a\alpha a' + a\beta b' + b = \alpha(a a' + b) + \beta(a b' + b) = \alpha f(a') + \beta f(b') ,

the key being b=(α+β)bb = (\alpha + \beta)b. Midpoints (α=β=12\alpha = \beta = \frac12), centroids (iterate), and the Napoleon centers αb+βc\alpha b + \beta c below are therefore equivariant: transforming the triangle transforms them accordingly. Hence proving a similarity-invariant statement for one well-placed triangle proves it for all — the license used in question 21.

11. The outward apex on [b,c][b, c] is pa=b+eiπ/3(cb)p_a = b + \eu^{-\iu\pi/3}(c - b) (question 3), so the center is

na=b+c+pa3=2b+c+1i32(cb)3=(3+i3)b+(3i3)c6=αb+βc,n_a = \frac{b + c + p_a}3 = \frac{2b + c + \frac{1 - \iu\sqrt3}2\,(c - b)}3 = \frac{(3 + \iu\sqrt3)\,b + (3 - \iu\sqrt3)\,c}6 = \alpha b + \beta c ,

with α=3+i36\alpha = \frac{3 + \iu\sqrt3}6 and β=3i36=α\beta = \frac{3 - \iu\sqrt3}6 = \conj\alpha. The same computation on the sides [c,a][c, a] and [a,b][a, b] gives nb=αc+βan_b = \alpha c + \beta a and nc=αa+βbn_c = \alpha a + \beta b (cyclic shift of the roles).

12. α+β=3+i3+3i36=1\alpha + \beta = \frac{3 + \iu\sqrt3 + 3 - \iu\sqrt3}6 = 1. Hence

na+nb+nc3=(α+β)(a+b+c)3=a+b+c3:\frac{n_a + n_b + n_c}3 = \frac{(\alpha + \beta)(a + b + c)}3 = \frac{a + b + c}3 :

the two triangles share their centroid.

13. Group by α\alpha and β\beta and apply question 4’s identities to P=a+jb+j2cP = a + jb + j^2c:

na+jnb+j2nc=α(b+jc+j2a)+β(c+ja+j2b)=αj2P+βjP=(αj2+βj)P.n_a + j n_b + j^2 n_c = \alpha\,(b + jc + j^2 a) + \beta\,(c + ja + j^2 b) = \alpha\,j^2 P + \beta\,j P = (\alpha j^2 + \beta j)\,P .

14. With j=1+i32j = \frac{-1 + \iu\sqrt3}2:

αj=(3+i3)(1+i3)12=3+3i3i3312=3+i36=β,\alpha j = \frac{(3 + \iu\sqrt3)(-1 + \iu\sqrt3)}{12} = \frac{-3 + 3\iu\sqrt3 - \iu\sqrt3 - 3}{12} = \frac{-3 + \iu\sqrt3}6 = -\beta ,

so αj+β=0\alpha j + \beta = 0, hence αj2+βj=j(αj+β)=0\alpha j^2 + \beta j = j(\alpha j + \beta) = 0, and question 13 gives na+jnb+j2nc=0n_a + jn_b + j^2n_c = 0 for every triangle (a,b,c)(a, b, c). By the criterion (questions 4–5), the centers form a direct equilateral triangle or a single point: Napoleon’s theorem.

15. The inward apex is b+e+iπ/3(cb)b + \eu^{+\iu\pi/3}(c - b), and the computation of question 11 with e+iπ/3=1+i32\eu^{+\iu\pi/3} = \frac{1 + \iu\sqrt3}2 swaps α\alpha and β\beta: ma=βb+αcm_a = \beta b + \alpha c, mb=βc+αam_b = \beta c + \alpha a, mc=βa+αbm_c = \beta a + \alpha b. Using the analogous identities b+j2c+ja=jQb + j^2c + ja = j\,Q and c+j2a+jb=j2Qc + j^2a + jb = j^2 Q for Q=a+j2b+jcQ = a + j^2b + jc:

ma+j2mb+jmc=β(b+j2c+ja)+α(c+j2a+jb)=(βj+αj2)Q=j(β+αj)Q=0,m_a + j^2 m_b + j m_c = \beta(b + j^2 c + ja) + \alpha(c + j^2 a + jb) = (\beta j + \alpha j^2)\, Q = j(\beta + \alpha j)\,Q = 0 ,

since αj=β\alpha j = -\beta (question 14). So the inner centers satisfy the indirect equilateral criterion: equilateral with the opposite orientation (or a point).

16. By question 5 applied to the triple (na,nb,nc)(n_a, n_b, n_c) (which satisfies the direct criterion), two centers coincide iff all three do; and all three coincide iff both criteria hold, i.e. iff additionally na+j2nb+jnc=0n_a + j^2 n_b + j n_c = 0. Compute as in question 13, with the identities b+j2c+ja=jQb + j^2c + ja = jQ, c+j2a+jb=j2Qc + j^2a + jb = j^2Q:

na+j2nb+jnc=αjQ+βj2Q=j(α+βj)Q.n_a + j^2 n_b + j n_c = \alpha\,jQ + \beta\,j^2 Q = j(\alpha + \beta j)\,Q .

Directly: βj=(3i3)(1+i3)12=3+3i3+i3+312=i33\beta j = \frac{(3 - \iu\sqrt3)(-1 + \iu\sqrt3)}{12} = \frac{-3 + 3\iu\sqrt3 + \iu\sqrt3 + 3}{12} = \frac{\iu\sqrt3}3, so α+βj=3+i3+2i36=1+i320\alpha + \beta j = \frac{3 + \iu\sqrt3 + 2\iu\sqrt3}6 = \frac{1 + \iu\sqrt3}2 \neq 0. Hence the outer Napoleon triangle degenerates iff Q=a+j2b+jc=0Q = a + j^2b + jc = 0, i.e. iff (a,b,c)(a, b, c) is an indirect equilateral triangle — in that case the “outward” constructions all point into the triangle’s circumscribed region and share one center.

17. With a=0a = 0, b=1b = 1, c=ic = \iu:

na=α+βi=(3+3)(1+i)6,nb=αi=3+3i6,nc=β=3i36.n_a = \alpha + \beta\iu = \frac{(3 + \sqrt3)(1 + \iu)}6, \qquad n_b = \alpha\iu = \frac{-\sqrt3 + 3\iu}6, \qquad n_c = \beta = \frac{3 - \iu\sqrt3}6 .

Squared sides: nanb=(3+23)+i36n_a - n_b = \frac{(3 + 2\sqrt3) + \iu\sqrt3}6 gives nanb2=(3+23)2+336=24+12336=2+33\abs{n_a - n_b}^2 = \frac{(3 + 2\sqrt3)^2 + 3}{36} = \frac{24 + 12\sqrt3}{36} = \frac{2 + \sqrt3}3; nbnc=(3+3)(1+i)6n_b - n_c = \frac{(3 + \sqrt3)(-1 + \iu)}6 gives nbnc2=2(3+3)236=24+12336\abs{n_b - n_c}^2 = \frac{2(3 + \sqrt3)^2}{36} = \frac{24 + 12\sqrt3}{36}; ncna=3i(3+23)6n_c - n_a = \frac{-\sqrt3 - \iu(3 + 2\sqrt3)}6 gives the same value again. All three squared sides equal 2+33\frac{2 + \sqrt3}3: equilateral, as promised.

18. Expand:

(ab)(cd)+(ad)(bc)=(acadbc+bd)+(abacbd+cd)=abadbc+cd,(a - b)(c - d) + (a - d)(b - c) = (ac - ad - bc + bd) + (ab - ac - bd + cd) = ab - ad - bc + cd ,

and (ac)(bd)=abadbc+cd(a - c)(b - d) = ab - ad - bc + cd: equal.

19. Take moduli in question 18 and apply the triangle inequality (Proposition 3.2 (4)):

ACBD=(ab)(cd)+(ad)(bc)abcd+adbc=ABCD+ADBC,AC \cdot BD = \abs{(a-b)(c-d) + (a-d)(b-c)} \leq \abs{a-b}\,\abs{c-d} + \abs{a-d}\,\abs{b-c} = AB \cdot CD + AD \cdot BC ,

with equality iff the two summands lie on a common ray from 00 (the equality case of the triangle inequality).

20. Half-angle factorization (Method 3.11 (4)): eiθeiφ=2isinθφ2  ei(θ+φ)/2\eu^{\iu\theta} - \eu^{\iu\varphi} = 2\iu\,\sin\frac{\theta - \varphi}2\;\eu^{\iu(\theta + \varphi)/2}, and taking moduli kills the unimodular factors: eiθeiφ=2sinθφ2\abs{\eu^{\iu\theta} - \eu^{\iu\varphi}} = 2\,\abs{\sin\frac{\theta - \varphi}2}.

21. With p=eiθp = \eu^{\iu\theta}, θ(2π/3,4π/3)\theta \in \intoo{2\pi/3}{4\pi/3}, question 20 gives

PA=2sinθ2,PB=2sin(θ2π3),PC=2sin(θ22π3).PA = 2\,\abs{\sin\tfrac\theta2}, \quad PB = 2\,\abs{\sin\bigl(\tfrac\theta2 - \tfrac\pi3\bigr)}, \quad PC = 2\,\abs{\sin\bigl(\tfrac\theta2 - \tfrac{2\pi}3\bigr)} .

On the arc, θ2(π/3,2π/3)\frac\theta2 \in \intoo{\pi/3}{2\pi/3}: then sinθ2>0\sin\frac\theta2 > 0; θ2π3(0,π/3)\frac\theta2 - \frac\pi3 \in \intoo0{\pi/3}, so the second sine is positive; θ22π3(π/3,0)\frac\theta2 - \frac{2\pi}3 \in \intoo{-\pi/3}0, so the third is negative and PC=2sin(2π3θ2)PC = 2\sin\bigl(\frac{2\pi}3 - \frac\theta2\bigr). Sum to product:

sin(θ2π3)+sin(2π3θ2)=2sinπ6cos(θ2π2)=sinθ2,\sin\Bigl(\frac\theta2 - \frac\pi3\Bigr) + \sin\Bigl(\frac{2\pi}3 - \frac\theta2\Bigr) = 2\,\sin\frac\pi6\,\cos\Bigl(\frac\theta2 - \frac\pi2\Bigr) = \sin\frac\theta2 ,

so PB+PC=2sinθ2=PAPB + PC = 2\sin\frac\theta2 = PA: van Schooten’s theorem.

22. At p=1p = -1: PA=2PA = 2, PB=PC=1PB = PC = 1 and all sides of the equilateral triangle have length 3\sqrt3, so both sides of Ptolemy read 232\sqrt3. Algebraically, using 1j2=j-1 - j^2 = j and 1+j=j21 + j = -j^2:

(ab)(pc)=(1j)(1j2)=(1j)j=jj2=i3,(a - b)(p - c) = (1 - j)\,(-1 - j^2) = (1 - j)j = j - j^2 = \iu\sqrt3 ,
(ac)(bp)=(1j2)(j+1)=1+jj2j3=jj2=i3.(a - c)(b - p) = (1 - j^2)(j + 1) = 1 + j - j^2 - j^3 = j - j^2 = \iu\sqrt3 .

The two terms are equal, so their ratio is 1R>01 \in \R_{>0}: the equality case of question 19, matching PABC=PBAC+PCABPA \cdot BC = PB \cdot AC + PC \cdot AB exactly.

23. For (a,b,c)=(1,j,j2)(a, b, c) = (1, j, j^2): na=αj+βj2=β+βj2n_a = \alpha j + \beta j^2 = -\beta + \beta j^2 (question 14) =β(j21)= \beta(j^2 - 1); numerically β(j21)=(3i3)6(32i32)=1\beta(j^2 - 1) = \frac{(3 - \iu\sqrt3)}6 \cdot \bigl(-\frac32 - \iu\frac{\sqrt3}2\bigr) = -1. Likewise nb=αj2+β=jn_b = \alpha j^2 + \beta = -j and nc=α+βj=j2n_c = \alpha + \beta j = -j^2. The outer Napoleon triangle of (1,j,j2)(1, j, j^2) is (1,j,j2)(-1, -j, -j^2): the original triangle reflected through its centroid 00 — same size, half-turn rotated. An equilateral triangle is a fixed shape of the Napoleon construction, not a shrinking limit.

24. (i) Multiplication by a unimodular number as rotation built the apexes and centers (questions 2–3, 11) and converted circle distances into sines (question 20). (ii) The group structure and the shape invariant justified normalizing the circumcircle to the unit circle and the triangle to (1,j,j2)(1, j, j^2) in question 21, via the equivariance of question 10. (iii) The half-angle factorization powered both question 20 and the sum-to-product step finishing van Schooten.

25. The dictionary converts geometric statements into polynomial identities in which every hypothesis is an equation: what it buys is mechanization — Napoleon’s theorem reduced to αj+β=0\alpha j + \beta = 0, one line of arithmetic in Q(i3)\Q(\iu\sqrt3); what it costs is geometric visibility — the computation certifies but does not show why the centers close up into an equilateral triangle. A fair answer is that the identity explains the theorem’s inevitability (it holds identically in a,b,ca, b, c, so no cleverness of configuration is involved), while the picture explains its content. The same dictionary returns in matrix form when rotations become the orthogonal 2×22 \times 2 matrices of Chapter 21 and Chapter 23, where “multiplication by eiθ\eu^{\iu\theta}” is the prototype of a linear isometry.