math.wiki
Abstract algebra › Fields › Compass-and-straightedge constructions, and why an angle cannot be trisected
Sign in

Compass-and-straightedge constructions, and why an angle cannot be trisected

With a compass and a straightedge one can bisect any angle and build a square of twice the area of a given one, yet nobody ever trisected a 60^\circ angle or doubled a cube that way. Is that a failure of ingenuity, or are those lengths out of reach?

Where you are 5 in Fields

Why this exists

Compass and straightedge copy, bisect and build equilateral triangles (Why a construction works), but three problems resisted every attempt: trisect any angle, double a cube, square a circle. To prove that no construction exists one needs a property every constructed length has and the wanted length lacks. Fields supply it: each step solves at most a quadratic equation, so by the tower law every constructed number has degree a power of 2 over \mathbb{Q}, while \displaystyle \sqrt[3]{2} and \cos 20^\circ have degree 3. Pierre Wantzel published this argument in 1837.

60° 20° 0 1
\operatorname{cis} 60^\circ
\operatorname{cis} 20^\circ
Figure 1The angle 60^\circ is constructible from 0 and 1. The dashed ray at 20^\circ is not (Theorem 6), so no construction trisects every angle.

Definition

In plain words: start with two points, call their distance 1, and keep drawing lines through points you have and circles centered at points you have; a number is constructible when it shows up as a coordinate of some point you reach.

Identify the plane with \mathbb{C}, the point (a, b) being a + bi. Given a set S of points, a straightedge line is the line through two distinct points of S, and a compass circle is a circle whose center is in S and whose radius is the distance between two points of S (the compass carries a length, as in Copying a segment and an angle).

  • A point P is constructible if there is a finite sequence P_0 = 0, P_1 = 1, P_2, \dots, P_n = P in which each P_j with j \ge 2 is a point of intersection of two distinct straightedge lines or compass circles drawn from \{P_0, \dots, P_{j-1}\}. A complex number is constructible when its point is.
  • A real number a is constructible if the point a on the real axis is constructible. The set of constructible real numbers is \mathcal{K}.
  • An angle \theta is constructible if the point \operatorname{cis}\theta = \cos\theta + i\sin\theta is constructible (Polar form of a complex number). Equivalently, the ray from 0 at angle \theta to the positive real axis can be drawn: it meets the circle about 0 through 1 at \operatorname{cis}\theta.
  • To trisect the angle \theta is to construct \operatorname{cis}(\theta/3) from 0, 1 and \operatorname{cis}\theta.

Intuition

Sums and differences lay one length after another with the compass; products and quotients come from parallel lines and similar triangles. Square roots come from a semicircle: the height of the semicircle on a segment of length 1 + a, above the point that splits it into 1 and a, is \displaystyle \sqrt{a}.

-1
0 a
\displaystyle \sqrt{a}
Figure 2Theorem 1(b): the semicircle on the segment from -1 to a meets the imaginary axis at height \displaystyle \sqrt{a}, since y^2 = r^2 - c^2 = a (drawn for a = 3).
0 1 b
a\,i
ab\,i
Figure 3Theorem 1(a): the parallel through b to the line through 1 and ai meets the imaginary axis at abi.

Conversely, intersecting lines and circles with equations over a field F solves at most one quadratic, so each step adds at most one square root. The constructible numbers are those built from 1 by +, -, \times, \div and \displaystyle \sqrt{\ }.

Results

Theorem 1 (the constructible field). (a) \mathcal{K} is a subfield of \mathbb{R}. (b) If a \in \mathcal{K} and a > 0, then \displaystyle \sqrt{a} \in \mathcal{K}. (c) For real a, b, the point a + bi is constructible if and only if a, b \in \mathcal{K}. A real d \ge 0 lies in \mathcal{K} if and only if d is the distance between two constructible points. (d) The constructible complex numbers form the subfield \mathcal{K}(i) = \{a + bi : a, b \in \mathcal{K}\} of \mathbb{C}.

Corollary 2 (angles). (a) An angle \theta is constructible if and only if \cos\theta \in \mathcal{K}. (b) If \theta and \varphi are constructible angles, so are \theta + \varphi, \theta - \varphi and \theta/2.

Theorem 3 (one step, one square root). Let F be a subfield of \mathbb{R}. Call a line an F-line if its equation is ax + by + c = 0 with a, b, c \in F, and a circle an F-circle if its center has coordinates in F and the square of its radius is in F. Every point where two distinct F-lines or F-circles meet has both coordinates in \displaystyle F(\sqrt{d}) for some d \in F with d \ge 0.

Theorem 4 (towers of square roots). A real number \alpha is constructible if and only if there are subfields \mathbb{Q} = F_0 \subseteq F_1 \subseteq \dots \subseteq F_k of \mathbb{R} with \alpha \in F_k and \displaystyle F_i = F_{i-1}(\sqrt{d_i}) for some d_i \in F_{i-1}, d_i > 0 (1 \le i \le k).

Corollary 5 (the degree test). Every \alpha \in \mathcal{K} is algebraic over \mathbb{Q}, and [\mathbb{Q}(\alpha) : \mathbb{Q}] = 2^m for some integer m \ge 0. In particular, if f \in \mathbb{Q}[x] has degree 3 and no rational root, then no real root of f is constructible.

Theorem 6 (the classical problems). (a) \displaystyle \sqrt[3]{2} \notin \mathcal{K}: from a cube of edge 1 one cannot construct the edge of a cube of volume 2. (b) The angle 60^\circ is constructible and 20^\circ is not: 60^\circ cannot be trisected, so no construction trisects every angle. (c) Granting that \pi is transcendental, \displaystyle \sqrt{\pi} \notin \mathcal{K}: from a circle of radius 1 one cannot construct the side of a square of the same area.

The transcendence of \pi (Lindemann) is stated without proof in Algebraic elements and minimal polynomials.

Why it is true

Midpoints (Bisecting a segment and an angle), perpendiculars at a point of a line (Perpendicular and parallel lines through a point) or from a point off it (bisect the chord a circle about the point cuts), and parallels (a perpendicular to a perpendicular) use only straightedge lines and compass circles, so they produce constructible points from constructible points.

Proof of Theorem 1. The real axis is the line through 0 and 1. The circle about 0 through 1 meets it again at -1, and the perpendicular bisector of -1 and 1 is the imaginary axis.

(c) Let P = a + bi be constructible. The perpendicular from P to the real axis meets it at a, and the perpendicular from P to the imaginary axis meets it at bi; the circle about 0 through bi meets the real axis at \pm b. So a, b \in \mathcal{K}. Conversely, for a, b \in \mathcal{K} the circle about 0 through b gives bi, and the perpendiculars to the axes at a and at bi meet at a + bi. If d = |PQ| for constructible P, Q, the circle about 0 of radius |PQ| meets the real axis at d; and d \in \mathcal{K} is the distance from 0 to d.

(a) 0, 1 \in \mathcal{K}. Let a, b \in \mathcal{K}. The circle about a of radius |b| meets the real axis at a + b and a - b. For products and quotients take a, b \ne 0; by (c) the points 1, ai and b are constructible. The line through 1 and ai is x + y/a = 1; its parallel through b is x + y/a = b, which meets the imaginary axis at abi. The line through b and ai is x/b + y/a = 1; its parallel through 1 is x/b + y/a = 1/b, which meets the imaginary axis at (a/b)i. By (c), ab, a/b \in \mathcal{K}, so \mathcal{K} is closed under subtraction, multiplication and division by nonzero elements: a subfield.

(b) The midpoint of -1 and a is \displaystyle c = \frac{a - 1}{2}, and the circle about c through a has radius \displaystyle r = \frac{a + 1}{2}. On the imaginary axis its points satisfy y^2 = r^2 - c^2 = a, so it meets that axis at \displaystyle \sqrt{a}\,i, and \displaystyle \sqrt{a} \in \mathcal{K} by (c).

(d) By (c) the constructible complex numbers are the set \mathcal{K}(i) described. It is a subfield because (a + bi)(c + di) = (ac - bd) + (ad + bc)i and \displaystyle (a + bi)^{-1} = \frac{a - bi}{a^2 + b^2} have parts in \mathcal{K} by (a). \square

Proof of Corollary 2. (a) By Theorem 1(c), \operatorname{cis}\theta is constructible if and only if \cos\theta, \sin\theta \in \mathcal{K}. If \cos\theta \in \mathcal{K}, then \displaystyle \sin\theta = \pm\sqrt{1 - \cos^2\theta} lies in \mathcal{K} by Theorem 1(a) and (b) (it is 0 when 1 - \cos^2\theta = 0). (b) \operatorname{cis}(\theta \pm \varphi) = \operatorname{cis}\theta \cdot (\operatorname{cis}\varphi)^{\pm 1} lies in the field \mathcal{K}(i). From \cos\theta = 2\cos^2(\theta/2) - 1 (Trigonometric identities), \displaystyle \cos(\theta/2) = \pm\sqrt{(1 + \cos\theta)/2}, which is in \mathcal{K}; apply (a). \square

Proof of Theorem 3. A line through two distinct points (x_1, y_1), (x_2, y_2) with coordinates in F has the equation (y_2 - y_1)(x - x_1) - (x_2 - x_1)(y - y_1) = 0, an F-line; a circle with center (p, q) and radius r, r^2 \in F, is

\begin{aligned} &x^2 + y^2 + dx + ey + f = 0, \\ &d = -2p,\ e = -2q,\ f = p^2 + q^2 - r^2 \in F. \end{aligned}

Two lines a_1x + b_1y + c_1 = 0 and a_2x + b_2y + c_2 = 0 that meet in one point have \Delta = a_1b_2 - a_2b_1 \ne 0, and Cramer’s rule gives x = (b_1c_2 - b_2c_1)/\Delta and y = (a_2c_1 - a_1c_2)/\Delta, in \displaystyle F = F(\sqrt{0}).

A line and a circle. If b \ne 0, substitute y = -(ax + c)/b into the circle: Ax^2 + Bx + C = 0 with A = 1 + a^2/b^2 \ne 0 and B, C \in F. A real common point has x a real root, so D = B^2 - 4AC \ge 0 and \displaystyle x = (-B \pm \sqrt{D})/(2A) \in F(\sqrt{D}) (The quadratic formula); then \displaystyle y \in F(\sqrt{D}). If b = 0, then x = -c/a \in F and y solves a quadratic over F, with the same conclusion.

Two circles. Subtracting their equations gives (d_1 - d_2)x + (e_1 - e_2)y + (f_1 - f_2) = 0. If d_1 = d_2 and e_1 = e_2, the circles are distinct with one center, so they do not meet. Otherwise this is an F-line, and the common points of the circles are the common points of this line and the first circle: the previous case. \square

Proof of Theorem 4. (\Leftarrow) Every subfield of \mathbb{R} contains 1, hence every n \cdot 1 and every quotient of these, so F_0 = \mathbb{Q} \subseteq \mathcal{K}. If F_{i-1} \subseteq \mathcal{K}, then \displaystyle \sqrt{d_i} \in \mathcal{K} by Theorem 1(b), and every element of F_i is \displaystyle u + v\sqrt{d_i} with u, v \in F_{i-1} (Theorem 1(e) of Algebraic elements and minimal polynomials, as \displaystyle \sqrt{d_i} has degree at most 2), so F_i \subseteq \mathcal{K} by Theorem 1(a). By Proof by induction, \alpha \in F_k \subseteq \mathcal{K}.

(\Rightarrow) Let P_0, \dots, P_n = \alpha be a construction, and L_1 = \mathbb{Q}. Suppose the coordinates of P_0, \dots, P_j lie in L_j, the top of a chain of real fields, each the one below with some \displaystyle \sqrt{d}, d \ge 0, adjoined. Lines and circles drawn from these points are L_j-lines and L_j-circles (a squared radius is a sum of squares of differences of coordinates), so by Theorem 3 the coordinates of P_{j+1} lie in \displaystyle L_{j+1} = L_j(\sqrt{d}) for some d \in L_j, d \ge 0. So \alpha \in L_n; deleting the steps with d = 0 leaves a chain as in the theorem. \square

Proof of Corollary 5. Each \displaystyle \sqrt{d_i} is a root of x^2 - d_i \in F_{i-1}[x], so [F_i : F_{i-1}] is 1 or 2, and by the tower law (Field extensions and degree) [F_k : \mathbb{Q}] = 2^s for some s. Since \alpha \in F_k, Theorem 4 of Algebraic elements and minimal polynomials makes \alpha algebraic of degree dividing 2^s, hence a power of 2 (Every number has exactly one prime factorization). For the cubic: a factorization of f into nonconstant factors in \mathbb{Q}[x] would include a linear factor, hence a rational root. So f is irreducible, and for a real root \alpha the minimal polynomial m_\alpha is f divided by its leading coefficient (Theorem 1(d) there), of degree 3, which is not a power of 2. \square

Proof of Theorem 6. (a) By the The rational root theorem the only candidates for a rational root of x^3 - 2 are \pm 1, \pm 2, where it takes -1, -3, 6, -10. Corollary 5 applies.

(b) \displaystyle \cos 60^\circ = \frac{1}{2} \in \mathcal{K}, so 60^\circ is constructible (Corollary 2; the point \operatorname{cis} 60^\circ is the apex of the equilateral triangle on 0 and 1, Equilateral triangle, square and hexagon in a circle). For any t, put z = \operatorname{cis} t, so z + z^{-1} = 2\cos t and z^3 + z^{-3} = 2\cos 3t (Roots of a complex number and roots of unity). Expanding (z + z^{-1})^3 = z^3 + z^{-3} + 3(z + z^{-1}) gives, for \beta = 2\cos t,

\beta^3 - 3\beta = 2\cos 3t. \qquad (\ast)

At t = 20^\circ, \beta is a root of x^3 - 3x - 1, which takes -3 and 1 at \pm 1, so Corollary 5 gives \beta \notin \mathcal{K}, hence \cos 20^\circ = \beta/2 \notin \mathcal{K} and 20^\circ is not constructible. A trisection of 60^\circ would construct \operatorname{cis} 20^\circ from 0, 1 and \operatorname{cis} 60^\circ; putting a construction of \operatorname{cis} 60^\circ in front of it would construct \operatorname{cis} 20^\circ from 0 and 1 alone.

(c) If \displaystyle \sqrt{\pi} \in \mathcal{K}, then \displaystyle \pi = (\sqrt{\pi})^2 \in \mathcal{K} would be algebraic by Corollary 5. \square

Worked examples

Example 1

A nested square root

Show that \displaystyle \alpha = \sqrt{1 + \sqrt{3}} is constructible, describe a construction, and find [\mathbb{Q}(\alpha) : \mathbb{Q}].

Solution

Take \displaystyle F_1 = \mathbb{Q}(\sqrt{3}) (d_1 = 3) and F_2 = F_1(\alpha) (\displaystyle d_2 = 1 + \sqrt{3} > 0); Theorem 4 gives \alpha \in \mathcal{K}. Concretely: the semicircle on [-1, 3] gives \displaystyle \sqrt{3} (the figure under Intuition); the circle about 1 of radius \displaystyle \sqrt{3} meets the real axis at \displaystyle 1 + \sqrt{3}; the semicircle on \displaystyle [-1, 1 + \sqrt{3}] gives \alpha.

For the degree, (\alpha^2 - 1)^2 = 3, so \alpha is a root of x^4 - 2x^2 - 2. This is Eisenstein at 2 (Irreducible polynomials and Eisenstein's criterion): 2 divides every coefficient below the leading one (0, -2, 0, -2), 2 \nmid 1 and 4 \nmid -2. So [\mathbb{Q}(\alpha) : \mathbb{Q}] = 4 = 2^2, and both steps of the tower have degree exactly 2, as Corollary 5 allows.

Example 2

The regular pentagon

Show that \displaystyle \cos 72^\circ = \frac{\sqrt{5} - 1}{4} and that the regular pentagon inscribed in the circle about 0 through 1 is constructible.

Solution

The vertices are \zeta^k for \zeta = \operatorname{cis} 72^\circ, 0 \le k \le 4 (Regular polygons and their circles). Since \zeta^5 = 1 \ne \zeta and x^5 - 1 = (x - 1)(x^4 + x^3 + x^2 + x + 1), dividing \zeta^4 + \zeta^3 + \zeta^2 + \zeta + 1 = 0 by \zeta^2 gives

(\zeta^2 + \zeta^{-2}) + (\zeta + \zeta^{-1}) + 1 = 0.

With \beta = \zeta + \zeta^{-1} = 2\cos 72^\circ and \zeta^2 + \zeta^{-2} = \beta^2 - 2, this is \beta^2 + \beta - 1 = 0. As \cos 72^\circ > 0, \displaystyle \beta = \frac{-1 + \sqrt{5}}{2} and \displaystyle \cos 72^\circ = \frac{\sqrt{5} - 1}{4} \in \mathbb{Q}(\sqrt{5}) \subseteq \mathcal{K}. By Corollary 2, \zeta is constructible, and its powers lie in the field \mathcal{K}(i): all five vertices are constructible, with the one-step tower \displaystyle \mathbb{Q} \subseteq \mathbb{Q}(\sqrt{5}).

Example 3

The regular heptagon

Show that no regular heptagon is constructible.

Solution

A constructed regular heptagon gives its center (Constructions with circles) and its central angle 360^\circ/7, which can be copied onto the ray from 0 through 1 (Copying a segment and an angle). So it suffices that 360^\circ/7 is not constructible.

Let \zeta = \operatorname{cis}(360^\circ/7) and \beta = \zeta + \zeta^{-1}. As in Example 2, 1 + \zeta + \dots + \zeta^6 = 0; divided by \zeta^3,

(\zeta^3 + \zeta^{-3}) + (\zeta^2 + \zeta^{-2}) + (\zeta + \zeta^{-1}) + 1 = 0.

With \zeta^2 + \zeta^{-2} = \beta^2 - 2 and \zeta^3 + \zeta^{-3} = \beta^3 - 3\beta (from (\ast)), this is \beta^3 + \beta^2 - 2\beta - 1 = 0. The cubic takes -1 and 1 at \pm 1, its only rational candidates, so by Corollary 5, \beta = 2\cos(360^\circ/7) \notin \mathcal{K}, and by Corollary 2 the angle is not constructible.

Example 4

An angle that can be trisected

Let \theta be the angle in [0^\circ, 180^\circ] with \displaystyle \cos\theta = -\frac{11}{16}. Show that \theta can be trisected and find \cos(\theta/3).

Solution

\theta is constructible, as \cos\theta \in \mathbb{Q}. By (\ast), \beta = 2\cos(\theta/3) satisfies \displaystyle \beta^3 - 3\beta = -\frac{11}{8}, that is 8\beta^3 - 24\beta + 11 = 0. Testing candidates finds the root \displaystyle \frac{1}{2}, and division gives

8x^3 - 24x + 11 = (2x - 1)(4x^2 + 2x - 11),

with roots \displaystyle \frac{1}{2} and \displaystyle \frac{-1 \pm 3\sqrt{5}}{4} (about 1.427 and -1.927). Since \cos\theta < 0, 90^\circ < \theta < 180^\circ, so 30^\circ < \theta/3 < 60^\circ and \displaystyle 1 < \beta < \sqrt{3}: only \displaystyle \frac{3\sqrt{5} - 1}{4} qualifies. So \displaystyle \cos(\theta/3) = \frac{3\sqrt{5} - 1}{8} \in \mathbb{Q}(\sqrt{5}) \subseteq \mathcal{K}, and one square root trisects \theta.

Example 5

Angles of a whole number of degrees

Prove that for a positive integer n, the angle n^\circ is constructible if and only if 3 \mid n.

Solution

Proof. The angles 72^\circ (Example 2) and 60^\circ are constructible, so by Corollary 2(b) so are 12^\circ = 72^\circ - 60^\circ, its half 6^\circ and its quarter 3^\circ, and every multiple of 3^\circ by repeated addition.

Conversely, let n^\circ be constructible with 3 \nmid n, and let d = \gcd(n, 360). There are integers u, v with d = un + 360v (Bezout's identity and the extended Euclidean algorithm), so \operatorname{cis}(d^\circ) = \operatorname{cis}(n^\circ)^u, as \operatorname{cis} 360^\circ = 1. A power of a nonzero element of the field \mathcal{K}(i) (negative powers included) lies in it, so d^\circ is constructible. Now d \mid 360 = 2^3 \cdot 3^2 \cdot 5 and 3 \nmid d, so d \mid 40; then 40^\circ, a multiple of d^\circ, and its half 20^\circ are constructible, against Theorem 6(b). \square

Example 6

Infinitely many square roots

Prove that \mathcal{K} is an algebraic extension of \mathbb{Q} of infinite degree.

Solution

Proof. Every element of \mathcal{K} is algebraic over \mathbb{Q} by Corollary 5. For m \ge 1, \gamma_m = 2^{1/2^m} is obtained from 2 by m square roots, so \gamma_m \in \mathcal{K} by Theorem 1(b) and induction on m. It is a root of x^{2^m} - 2, Eisenstein at 2, so [\mathbb{Q}(\gamma_m) : \mathbb{Q}] = 2^m. If [\mathcal{K} : \mathbb{Q}] = N were finite, Theorem 4 of Algebraic elements and minimal polynomials would give 2^m \mid N for every m, which is impossible. \square

Boundaries

  • A power of 2 is necessary, not sufficient. The real roots of x^4 - 2x - 2 (Eisenstein at 2) have degree 4 but are not constructible, as its splitting field has degree 24. The exact criterion, stated here without proof and following from the The fundamental theorem of Galois theory: \alpha is constructible if and only if a splitting field of m_\alpha over \mathbb{Q} has degree a power of 2.
  • The tools are part of the theorem. A straightedge with two marks (a construction attributed to Archimedes) or paper folding trisects every angle.
  • “Cannot be trisected” means not every angle. 90^\circ and the angle of Example 4 can be. As \mathbb{Q} \subseteq \mathcal{K} is dense in \mathbb{R}, every length can be approximated; only exactness fails.

Where this goes next

Gauss and Wantzel showed (stated here without proof) that the regular n-gon is constructible exactly when n is a power of 2 times distinct primes of the form 2^{2^k} + 1: the 17-gon is constructible, the 7-gon is not. The proof reads the degree of \cos(360^\circ/n) off the Field automorphisms and the Galois group of \mathbb{Q}(\operatorname{cis}(360^\circ/n)). With towers of radicals in place of square roots, the same pattern shows in Why the quintic has no formula that no formula in radicals solves every quintic.

Exercises

Constructible or not

  1. Give a tower of square roots showing that \displaystyle \alpha = \sqrt{1 + \sqrt{1 + \sqrt{2}}} is constructible, and find m_\alpha and [\mathbb{Q}(\alpha) : \mathbb{Q}].
  2. Decide which real numbers are constructible: (a) \displaystyle \sqrt[3]{5}; (b) \displaystyle \sqrt[4]{5}; (c) \displaystyle \sqrt[3]{2\sqrt{2}}; (d) \displaystyle \sqrt[3]{7 + 5\sqrt{2}}.
  3. Show that 2\cos 40^\circ is a root of x^3 - 3x + 1, and conclude that the regular 9-gon is not constructible.
  4. Let \theta be the angle in [0^\circ, 180^\circ] with \displaystyle \cos\theta = \frac{1}{4}. Show that \theta is constructible but cannot be trisected.
  5. Let \theta be the angle in [0^\circ, 180^\circ] with \displaystyle \cos\theta = -\frac{9}{16}. Show that \theta can be trisected and find \cos(\theta/3).

Polygons and angles

  1. Using Example 5, decide which of the regular 12-, 18-, 20-, 24- and 40-gons are constructible.
  2. Let \zeta = \operatorname{cis}(360^\circ/13) and \beta = \zeta + \zeta^{-1}. Given that 1 + x + \dots + x^{12} is irreducible over \mathbb{Q}, show that [\mathbb{Q}(\beta) : \mathbb{Q}] = 6, and conclude that the regular 13-gon is not constructible.
  3. Find integers u, v with 24 = 72u + 120v, and use them to show that the regular 15-gon is constructible.

At the engraver’s

Mei and Hiro run an engraving shop and lay out designs with compass and straightedge from a unit segment.

  1. Mei wants a square plate with the same area as a 2 by 7 rectangle. Show that its side is constructible, using a semicircle as in Theorem 1(b).
  2. Hiro wants a cube-shaped box with three times the volume of a cube of edge 1. Can he construct its edge?
  3. A customer’s design uses the angle \theta in [0^\circ, 90^\circ] with \displaystyle \cos\theta = \frac{3}{5}, and asks for \theta/2 on one plate and \theta/3 on another. Which can Mei construct?

Proofs

  1. Prove that \mathcal{K} is the smallest subfield of \mathbb{R} containing the square root of each of its positive elements: \mathcal{K} has this property, and every subfield of \mathbb{R} with it contains \mathcal{K}.
  2. Prove that every square root of a constructible complex number is constructible.
  3. Prove that if the regular m-gon and the regular n-gon are constructible and \gcd(m, n) = 1, then the regular mn-gon is constructible.
  4. Let \theta be an angle with c = \cos\theta \in \mathbb{Q}. Prove that \theta can be trisected if and only if x^3 - 3x - 2c has a rational root.
  5. Prove that for an integer n that is not the cube of an integer, \displaystyle \sqrt[3]{n} is not constructible.
Answers
  1. \displaystyle \mathbb{Q} \subseteq \mathbb{Q}(\sqrt{2}) \subseteq \mathbb{Q}(\sqrt{1 + \sqrt{2}}) \subseteq \mathbb{Q}(\alpha), adjoining square roots of 2, \displaystyle 1 + \sqrt{2} and \displaystyle 1 + \sqrt{1 + \sqrt{2}}, all positive. From ((\alpha^2 - 1)^2 - 1)^2 = 2, \alpha is a root of x^8 - 4x^6 + 4x^4 - 2, Eisenstein at 2, so m_\alpha = x^8 - 4x^6 + 4x^4 - 2 and the degree is 8.
  2. (a) Not constructible: x^3 - 5 takes -4, -6, 120, -130 at \pm 1, \pm 5, so it has no rational root (Corollary 5). (b) Constructible: \displaystyle \sqrt[4]{5} = \sqrt{\sqrt{5}}. (c) Constructible: \displaystyle (\sqrt{2})^3 = 2\sqrt{2}, so the real cube root is \displaystyle \sqrt{2}. (d) Constructible: \displaystyle (1 + \sqrt{2})^3 = 1 + 3\sqrt{2} + 6 + 2\sqrt{2} = 7 + 5\sqrt{2}, so the number is \displaystyle 1 + \sqrt{2}.
  3. By (\ast) with t = 40^\circ, \beta^3 - 3\beta = 2\cos 120^\circ = -1. The cubic takes -1 and 3 at \pm 1, so 2\cos 40^\circ \notin \mathcal{K} (Corollary 5), the central angle 40^\circ is not constructible, and neither is the 9-gon (as in Example 3).
  4. \cos\theta \in \mathbb{Q}, so \theta is constructible. By (\ast), \beta = 2\cos(\theta/3) satisfies \displaystyle \beta^3 - 3\beta = \frac{1}{2}, a root of 2x^3 - 6x - 1. The rational candidates \displaystyle \pm 1, \pm \frac{1}{2} give \displaystyle -5, 3, -\frac{15}{4}, \frac{7}{4}, so \beta \notin \mathcal{K} and \theta/3 is not constructible.
  5. \beta = 2\cos(\theta/3) is a root of 8x^3 - 24x + 9 = (2x - 3)(4x^2 + 6x - 3), with roots \displaystyle \frac{3}{2} and \displaystyle \frac{-3 \pm \sqrt{21}}{4} (about 0.396 and -1.896). As 90^\circ < \theta < 180^\circ, \displaystyle 1 < \beta < \sqrt{3}, so \displaystyle \beta = \frac{3}{2} and \displaystyle \cos(\theta/3) = \frac{3}{4}. Check: \displaystyle 4(\frac{3}{4})^3 - 3 \cdot \frac{3}{4} = \frac{27 - 36}{16} = -\frac{9}{16}.
  6. The central angles are 30^\circ, 20^\circ, 18^\circ, 15^\circ, 9^\circ. All but 20^\circ are multiples of 3: the 12-, 20-, 24- and 40-gons are constructible and the 18-gon is not.
  7. \beta = 2\cos(360^\circ/13) is real, and \zeta is a root of x^2 - \beta x + 1 \in \mathbb{Q}(\beta)[x], not real, so [\mathbb{Q}(\zeta) : \mathbb{Q}(\beta)] = 2. As \zeta is a root of the irreducible 1 + x + \dots + x^{12}, [\mathbb{Q}(\zeta) : \mathbb{Q}] = 12, and the tower law gives [\mathbb{Q}(\beta) : \mathbb{Q}] = 6, not a power of 2. So \cos(360^\circ/13) \notin \mathcal{K}.
  8. u = 2, v = -1: 24 = 144 - 120. So \operatorname{cis} 24^\circ = (\operatorname{cis} 72^\circ)^2 (\operatorname{cis} 120^\circ)^{-1}, a product in \mathcal{K}(i) (72^\circ by Example 2, 120^\circ = 60^\circ + 60^\circ), and 24^\circ = 360^\circ/15.
  9. The side is \displaystyle \sqrt{14}. The semicircle on the segment from -2 to 7 has center \displaystyle \frac{5}{2} and radius \displaystyle \frac{9}{2}, and meets the imaginary axis at height y with \displaystyle y^2 = \frac{81}{4} - \frac{25}{4} = 14.
  10. No. The edge is \displaystyle \sqrt[3]{3}, and x^3 - 3 takes -2, -4, 24, -30 at \pm 1, \pm 3: no rational root, so Corollary 5 rules it out.
  11. \theta/2 only. \displaystyle \cos(\theta/2) = \sqrt{(1 + \frac{3}{5})/2} = \frac{2}{\sqrt{5}} = \frac{2\sqrt{5}}{5} \in \mathcal{K}. For \theta/3, (\ast) makes 2\cos(\theta/3) a root of \displaystyle x^3 - 3x - \frac{6}{5}, that is of 5x^3 - 15x - 6, which is Eisenstein at 3 (3 \mid -15, -6, 3 \nmid 5, 9 \nmid -6), so irreducible of degree 3 and \theta/3 is not constructible.
  12. Proof. \mathcal{K} has the property by Theorem 1(a) and (b). Let L \subseteq \mathbb{R} be a subfield with it, and \alpha \in \mathcal{K} with a tower F_0 \subseteq \dots \subseteq F_k as in Theorem 4. Then F_0 = \mathbb{Q} \subseteq L, and if F_{i-1} \subseteq L, then d_i \in L is positive, so \displaystyle \sqrt{d_i} \in L and \displaystyle F_i = \{u + v\sqrt{d_i} : u, v \in F_{i-1}\} \subseteq L. By induction \alpha \in F_k \subseteq L. \square
  13. Proof. Let z = a + bi \ne 0 be constructible (0 is trivial). Then \displaystyle r = |z| = \sqrt{a^2 + b^2} \in \mathcal{K} and z = r\operatorname{cis}\theta with \cos\theta = a/r \in \mathcal{K}, so \theta/2 is constructible (Corollary 2). Then \displaystyle w = \sqrt{r}\operatorname{cis}(\theta/2) \in \mathcal{K}(i) (Theorem 1(b), (d)), w^2 = z, and the square roots of z are \pm w. \square
  14. Proof. As in Examples 2 and 3, a regular k-gon is constructible if and only if the angle 360^\circ/k is. With um + vn = 1 (Bezout's identity and the extended Euclidean algorithm), \displaystyle \frac{360^\circ}{mn} = u \cdot \frac{360^\circ}{n} + v \cdot \frac{360^\circ}{m}, so \displaystyle \operatorname{cis}\frac{360^\circ}{mn} = (\operatorname{cis}\frac{360^\circ}{n})^u (\operatorname{cis}\frac{360^\circ}{m})^v \in \mathcal{K}(i). \square
  15. Proof. As \theta is constructible, trisecting it means \beta = 2\cos(\theta/3) \in \mathcal{K} (Corollary 2), and \beta is a root of f = x^3 - 3x - 2c by (\ast). If f has a rational root r, then f = (x - r)(x^2 + Bx + C) over \mathbb{Q}, and either \beta = r or \displaystyle \beta = \frac{-B \pm \sqrt{D}}{2} with D = B^2 - 4C \ge 0 as \beta is real; either way \beta \in \mathcal{K}. Conversely, if \beta \in \mathcal{K}, its degree is a power of 2 (Corollary 5) and at most 3. Degree 1 makes \beta a rational root; degree 2 makes m_\beta a quadratic factor of f in \mathbb{Q}[x] (Theorem 1(c) of Algebraic elements and minimal polynomials), and the other factor is linear over \mathbb{Q}. \square
  16. Proof. Suppose x^3 - n had a rational root p/q in lowest terms. By the The rational root theorem, q \mid 1, so the root is an integer k with k^3 = n, against the hypothesis. So x^3 - n has no rational root, and Corollary 5 shows its real root \displaystyle \sqrt[3]{n} is not constructible. \square
math.wiki · updated 2026-11-18 · CC-BY-SA-4.0