Appendix A Constructibility
This book is dedicated to the synthetic (or axiomatic) approach to geometry, di- rectly inspired and motivated by the work of Greek geometers of antiquity. The Greek geometers achieved great work in geometry; but as we have seen, a few prob- lems resisted all their attempts (and all later attempts) at solution: circle squaring, trisecting the angle, duplicating the cube, constructing all regular polygons, to cite only the most popular ones. We conclude this book by proving why these various problems are insolvable via ruler and compass constructions. However, these proofs concerning ruler and compass constructions completely escape the scope of syn- thetic geometry which motivated them: the methods involved are highly algebraic and use theories such as that of fields and polynomials. This is why we these results are presented as appendices. Moreover, we shall freely use (with precise references) various algebraic results developed in [5], Trilogy II. We first review the theory of the minimal polynomial of an algebraic element in a field extension. Furthermore, since it will be essential for the applications, we prove the Eisenstein criterion for the irreducibility of (such) a (minimal) polynomial over the field of rational numbers. We then show how to associate a field with a given geometric configuration and we prove a formal criterion telling us—in terms of the associated fields—when one can pass from a given geometrical configuration to a more involved one, using only ruler and compass constructions.
A.1 The Minimal Polynomial
This section requires some familiarity with the theory of polynomials in one variable over a field K. The reader not familiar with this material will find some essential aspects of the theory in Appendix A of [5], Trilogy II.
Proposition A.1.1 If K ⊆ L is a field extension, then is a K-vector space.
Proof The scalar multiplication of k ∈ K and l ∈ L is their product in L.
F. Borceux, An Axiomatic Approach to Geometry, DOI 10.1007/978-3-319-01730-3, 355 © Springer International Publishing Switzerland 2014 356 A Constructibility
Under the conditions of Proposition A.1.1, it thus makes sense to speak of the (vectorial) dimension of L over K, which we shall write as dim[L : K].
Proposition A.1.2 Given finite dimensional field extensions K ⊆ L ⊆ M, one has
dim[M : K]=dim[M : L]×dim[L : K].
Proof Let l1,...,lr be a basis of L as a K-vector space and m1,...,ms a basis of M as an L-vector space. It suffices to observe that the rs elements limj constitute a ∈ basis of M over K. Indeed an element μ M can be uniquely written as j λj mj , ∈ = with λj L, while each λj can itself be uniquely written as λj i ki,j li with ki,j ∈ K. The conclusion follows at once.
Proposition A.1.3 Given a finite dimensional field extension K ⊆ L, every element l ∈ L is a root of a polynomial with coefficients in K.
Proof Suppose that dim[L : K]=n. Then the n + 1 elements of L
1,l,l2, ..., ln cannot be linearly independent over K, proving the existence of scalars ki ∈ K such that 2 n k01 + k1l + k2l +···+knl = 0. This proves that l is a root of the polynomial
2 n p(X) = k0 + k1X + k2X +···+knX ∈ K[X].
The situation described in Proposition A.1.3 has been given a name:
Definition A.1.4 An element l ∈ L in a field extension K ⊆ L is algebraic over K when it is a root of a polynomial with coefficients in K. When all elements of L are algebraic over K, the extension K ⊆ L itself is called algebraic.
The main result of this section is then:
Theorem A.1.5 Consider a field extension K ⊆ L and an element l ∈ L which is algebraic over K. There exists a unique polynomial p(X) ∈ K[X] such that 1. p(l) = 0; 2. the leading coefficient of p(X) is 1; 3. p(X) is irreducible in K[X]. Moreover, under these conditions: 1. p(X) divides in K[X] all other polynomials admitting l as a root; 2. the dimension of the subfield K(l) ⊆ L generated by K and l is equal to the degree of the polynomial p(X). A.1 The Minimal Polynomial 357
The polynomial p(X) is called the minimal polynomial of l over K.
Proof Among all the polynomials in K[X] admitting l as a root, choose a non-zero one whose degree is minimal. Possibly dividing by the leading coefficient (which does not change the roots), we can assume that this leading coefficient is 1. This is our polynomial p(X). If q(X) ∈ K[X] is such that q(l) = 0, the Euclidean division of q(X) by p(X) (see Theorem A.2.1 in [5], Trilogy II) yields
q(X) = p(X)d(X) + r(X), degree r(X)< degree p(X).
Since q(l) = 0 = p(l), this implies r(l) = 0. By minimality of the degree of p(X), we conclude that r(X)= 0. Thus p(X) divides q(X). This also proves that p(X) is irreducible in K[X]. Indeed otherwise l would be a root of a proper factor q(X) of p(X), which is a contradiction since by what we have just proved, p(X) would then have to divide q(X), which is of strictly lower degree. The uniqueness of p(X) also follows immediately. Indeed given another irre- ducible polynomial q(X) ∈ K[X] such that q(l) = 0, p(X) divides q(X) which is irreducible, thus q(X) = kp(X) with k ∈ K. Since both have the same leading coefficient 1, necessarily k = 1. It remains to prove the result concerning dimensions. If p(X) has degree n,the sequence of elements − 1,l,l2, ..., ln 1 is linearly independent over K. Indeed otherwise (as in the proof of Proposi- tion A.1.3) there would be a polynomial of degree n − 1 with coefficients in K[X] admitting l as a root; this would contradict the choice of p(X) with minimal degree. It remains to observe that − 1,l,l2, ..., ln 1 generate K(l) as a vector space over K. Writing
n n−1 p(X) = X + kn−1X +···+k1X + k0,ki ∈ K we get at once n n−1 l =−kn−1l −···−k1l − k0. This already proves that the subset n−1 M = α0 + α1l +···+αn−1l |αi ∈ K ⊆ L is not only stable under addition and subtraction, but also under multiplication, since each power ln after multiplication can be replaced by a combination of strictly lower powers; repeating this operation as many times as needed, one ends up with an expression where all powers of l are at most n − 1, thus with an element of M. 358 A Constructibility
Moreover, the inverse in L of every element m ∈ M is still in M, proving that M is in fact a subfield of L; and then trivially, M = K(l). When m ∈ K, this is trivial. Otherwise, m = q(l) for a non-constant polynomial
n−1 q(X) = β0 + β1(X) +···+βn−1X ,βi ∈ K.
But q(X) and p(X) do not have a common non-constant factor (they are “relatively prime”), because p(X) is irreducible and has a larger degree than q(X).BytheBe- zout theorem (see Theorem A.3.2 in [5], Trilogy II), there exist polynomials r(X), s(X) in K[X] such that
r(X)p(X) + s(X)q(X) = 1 with moreover
degree r(X)< degree q(X), degree s(X) < degree p(X) = n.
Taking the values at l we find s(l)q(l) = 1, proving that s(l) is the inverse of m = q(l). Indeed, s(l) ∈ M since degree s(X) < n.
Let us conclude with the classical terminology for “non-algebraic” elements:
Definition A.1.6 An element l ∈ L in a field extension K ⊆ L which is not alge- braic over K is said to be transcendental over K.
As an immediate consequence of Proposition A.1.3 we have
Proposition A.1.7 If a field extension K ⊆ L contains a transcendental element over K, its dimension is infinite.
A.2 The Eisenstein Criterion
Most impossibility proofs that we shall consider in Appendices B and C will be in terms of algebraic elements over of the field of rational numbers. Therefore, in order to easily identify the corresponding minimal polynomials, it is important to have an efficient criterion telling us when a polynomial in Q[X] is irreducible (see Theorem A.1.5). Of course, multiplying all the coefficients of a polynomial in Q[X] by a common multiple of their denominators, we obtain a polynomial with coefficients in Z, which still has the same roots. But more amazingly:
Proposition A.2.1 Let p(X) ∈ Z[X] be a polynomial which is irreducible in Z[X]. Then p(X) is also irreducible in Q[X]. A.2 The Eisenstein Criterion 359
Proof It is equivalent to prove that when p[X]∈Z[X] is reducible in Q[X],itis reducible in Z[X]. Suppose therefore that p(X) = r(X)s(X) with r(X) and s(X) two non-constant polynomials in Q[X]. Multiplying by a common multiple n of the denominators of the coefficients of r(X) and s(X), we obtain np(X) = r(X)s(X), where r(X) and s(X) now have their coefficients in Z. Let us write
k l r(X)= rkX +···+r0, s(X) = slX +···+s0,ri,sj ∈ Z.
Since p(X) has coefficients in Z, n divides all the coefficients of np(X). There- fore every prime factor d of n also divides all the coefficients of np(X) = r(X)s(X). Let us prove that d necessarily divides all the coefficients of r(X) or all the coeffi- cients of s(X). We prove this by a reductio ad absurdum. Assuming this is not the case, write respectively i and j for the smallest indices such that d does not divide i+j ri and sj . The coefficient of X in np(X) = r(X)s(X) is
risj + ri−1sj+1 + ri−2sj+2 +··· .
By choice of i and j, the prime number d does not divide the first term of this sum (see Proposition 3.7.5), but divides all the other terms. Thus d does not divide the sum itself. Since this sum is a coefficient of np(X), we have a contradiction. So d divides all the coefficients of one of the two polynomials r(X) or s(X).It is thus possible to simplify d on both sides maintain equality np(X) = r(X)s(X) and still maintain equality between polynomials with coefficients in Z. Repeating the operation for all the prime factors of n, we end up with p(X) presented as the product of two polynomials with coefficients in Z[X].
The following criterion (a sufficient condition for being irreducible) will prove to be very useful:
Theorem A.2.2 (The Eisenstein criterion) A polynomial
n n−1 p(X) = anX + an−1X +···+a1X + a0,ai ∈ Z is irreducible in Q[X] as soon as there exists a prime number d such that:
1. d does not divide an; 2. d divides an−1,...,a0; 2 3. d does not divide a0.
Proof Once more we use a reductio ad absurdum. Suppose that p(X) is reducible in Q[X]; then by Proposition A.2.1, p(X) is reducible in Z[X]. Write p(X) = r(X)s(X) with
k l r(X)= rkX +···+r0,s(X)= slX +···+s0,ri,sj ∈ Z,k,l>0.
By assumption, d divides a0 = r0s0, thus divides r0 or s0 (see Proposition 3.7.5). Let us say that d divides r0; the other case is analogous. On the other hand d does not 360 A Constructibility divide an = rksl , thus certainly d does not divide rk. Let us write i for the smallest index such that d does not divide ri ; we therefore have 0
ai = ris0 + ri−1s1 + ri−2s2 +··· .
2 By choice of i, d does not divide ri . And since d divides r0 but d does not divide a0 = r0s0, d does not divide s0. Thus d being prime, it does not divide the first term ris0 of the sum (see Proposition 3.7.5). Since by choice of i, d divides all the other terms of the sum, d does not divide the sum. Since k,l > 0, we also have k,l < n and therefore by assumption, d divides ai . This is the expected contradiction.
Counterexample A.2.3 The Eisenstein criterion is not a necessary condition for irreducibility in Q[X].
Proof Consider the polynomial
p(X) = X2 + X + 2.
The only prime number dividing a0 = 2 is 2 and it does not divide a1 = 1. Thus the Eisenstein criterion does not apply. Nevertheless, the polynomial p(X) is irreducible over Q[X] because it is irre- ducible over Z[X] (see Proposition 3.11.11). Indeed if the polynomial p(X) was reducible in Z[X], it would be the product of two polynomials of degree 1
X2 + X + 2 = (aX + b)(cX + d) = acX2 + (ad + bc)X + bd, a,b,c,d ∈ Z.
From ac = 1 we get at once a =±1 = c and from bd = 2 we obtain b =±2, d =±1 (or the converse). So in any case we end up with ad + bc =±3, which is a contradiction since a1 = 1.
A.3 Ruler and Compass Constructibility
This section intends only to make precise some common notions of constructibility in the Euclidean plane R2 (see Example 8.5.3).
Definition A.3.1 A geometric configuration is a finite set of points, straight lines and circles in the Euclidean plane R2. A given point of the configuration, or an intersection point of two distinct lines of the configuration—straight or circular—is called a node of the configuration.
Let us thus make very clear that given a line or a circle in a configuration, the points of this line or this circle are generally not nodes. For example, the config- uration comprising two parallel straight lines has no node at all, while the config- uration comprising two intersecting lines has the corresponding intersection point A.3 Ruler and Compass Constructibility 361 as unique node. If you want additional points on these lines to be nodes of the configuration, you have to include these points at once as given elements of the configuration. Let us also recall that Proposition 5.10.5 indicates that we could as well reduce our attention to configurations comprising only of points and circles; but we shall not make this quite unnatural choice.
Lemma A.3.2 A geometric configuration has finitely many nodes.
Proof Two lines intersect in at most one point, two circles in at most two points, a line and a circle in at most two points. This implies the result, because a configu- ration contains only finitely many points, straight lines and circles.
The problem of ruler and compass constructions is thus to determine, starting from a given geometric configuration, if it is possible to construct with ruler and compass some specified more involved configuration.
Definition A.3.3 An elementary extension of a given geometric configuration is one which is obtained by adding one line to this configuration, namely: the straight line passing through two of the nodes, or the circle with center some node and pass- ing through some other node. A finite extension of a given geometric configuration is one which can be obtained via a finite sequence of successive elementary exten- sions.
And thus as certainly expected:
Definition A.3.4 A point of the Euclidean plane R2 is constructible from a given geometric configuration when it can be presented as a node in some finite extension of that configuration.
Notice that it follows trivially from Definitions A.3.3 and A.3.4 that: • no point at all is constructible from a geometric configuration without any nodes; • if a configuration has a unique node P , only that point P is constructible from the configuration. So the problem of constructible points with respect to a configuration becomes sen- sible only when there are at least two nodes. Of course, there is no loss of generality in choosing a system of coordinates based on these two nodes. It is therefore conve- nient to use the following terminology:
Definition A.3.5 A geometric configuration in standard position is a geometric configuration, together with the choice of an orthonormal system of Cartesian co- ordinates, so that the two points with coordinates (0, 0) and (1, 0) are nodes of the configuration. 362 A Constructibility
Fig. A.1
We shall often refer to the following:
Definition A.3.6 By the basic geometric configuration we mean the configura- tion, in some orthonormal system of Cartesian coordinates, comprising only the two points (0, 0) and (1, 0).
Notice at once that
Lemma A.3.7 Given a geometric configuration in standard position, the point with coordinates (0, 1) is constructible and the two coordinate axes thus belong to some finite extension of the configuration.
Proof The results of Sect. 3.1—which we freely use without further reference— justify at once the following argument; we refer to Fig. A.1. Write A = (0, 0) and B = (1, 0); the line through A and B is the first coordinate axis. The circle with center A and radius AB cuts this first axis at a second point C. The circle with center C and radius CB cuts the circle with center B and radius BC at two points D and E. The line through D and E is perpendicular to the first axis at A and is therefore the second axis. The circle with center A and radius AB cuts this second axis at two points F and G which have the coordinates (0, ±1).
It will also be useful to use the following terminology:
Definition A.3.8 A real number r is constructible from a given geometric config- uration in standard position when the point with coordinates (r, 0) is constructible from that configuration.
As expected:
Proposition A.3.9 A point P with coordinates (a, b) is constructible from a given geometric configuration in standard position if and only if both real numbers a and b are constructible from that configuration. A.4 Constructibility Versus Field Theory 363
Fig. A.2
Proof We freely use the results of Sect. 3.1, without explicit mention; we refer to Fig. A.2.GivenP = (a, b), one gets the two points (a, 0) and (0,b) when drawing from P perpendiculars to the axis of coordinates. Drawing a circle with center the origin and passing through (0,b) then yields the intersection (b, 0) with the first axis. Conversely, given the points (a, 0) and (b, 0), a circle with center the origin and passing through (b, 0) yields (0,b) as intersection with the second axis. Next drawing through (a, 0) and (0,b)the perpendiculars to the axis yields as intersection the point P = (a, b).
A.4 Constructibility Versus Field Theory
The key point in handling problems of constructibility is to associate a field to each geometric configuration.
Definition A.4.1 The field of coordinates of a geometric configuration in standard position is the subfield K of R generated by the Cartesian coordinates of all the nodes of the configuration.
Thus if Ai = (ai,bi), i ∈ I , are the various nodes of the configuration, K is the subfield of the reals generated by all the elements ai and bi . So trivially:
Proposition A.4.2 The field of coordinates of the basic configuration is the field Q of rational numbers.
Observe at once that:
Lemma A.4.3 If K ⊆ R is the field of coordinates of a geometric configuration in standard position, all the numbers k ∈ K are constructible from that configuration (see Definition A.3.8).
Proof By Definition A.3.5,thesetS of real numbers appearing as one of the coor- dinates of a node of the configuration contains already the numbers 0 and 1, whose presence in a subfield is compulsory. The field K is then trivially the set of all real numbers which can be obtained via a finite sequence of arithmetical opera- tions (addition, subtraction, multiplication, division) from the numbers in S. It thus suffices to observe that all four arithmetical operations on real numbers r can be 364 A Constructibility performed via ruler and compass constructions on the corresponding points of coor- dinates (r, 0). This is trivial for addition and subtraction. The cases of multiplication and division follow at once from Proposition 3.6.8: Given three quantities α, β, α, one can construct from these, with ruler and compass, the α = α quantity β such that β β . = = × = = β Choosing α 1 we find β β α and choosing α 1, we get β α .
On the other hand:
Lemma A.4.4 Consider a geometric configuration in standard position and its field K of coordinates. Every line joining two nodes, and every circle centered at some node and passing through another node, admit equations whose coefficients are in K.
Proof The line passing through the points (a, b) and (c, d) admits the equation
(d − b)(X − a) = (c − a)(Y − b) while the circle centered at (a, b) and passing through (c, d) has equation
(X − a)2 + (Y − b)2 = (c − a)2 + (d − b)2.
Since by assumption a, b, c, d are in K, so are all the coefficients of these equa- tions.
We now arrive at the key result for proving the impossibility of some ruler and compass constructions:
Theorem A.4.5 Let K be the field of coordinates of a geometric configuration in standard position. A real number r is constructible from that configuration if and only if there is a finite sequence of field extensions
K = K0 ⊂ K1 ⊂ K2 ···⊂Kn such that r ∈ Kn and each field Ki+1 has dimension 2 over Ki .
Proof Assume first the existence of a finite sequence of field extensions as in the statement. By Proposition A.1.3, every element r ∈ Ki+1 is algebraic over Ki . Since the field Ki(r) is contained in Ki+1, the dimension of Ki(r) over Ki is at most 2. By Theorem A.1.5, the corresponding minimal polynomial p(X) of r over Ki then has degree at most 2. Trivially, r ∈ Ki if and only if the minimal polynomial has degree 1, that is, if p(X) = X − r. So when r ∈ Ki , the minimal polynomial p(X) has degree 2 and
2 p(X) = aX + bX + c, a,b,c ∈ Ki,a= 0. A.4 Constructibility Versus Field Theory 365
Since r is a root of p(X) √ −b ± b2 − 4ac r = . 2a Since the four arithmetical operations can be performed via ruler and com- pass constructions (see the proof of Lemma A.4.3), as well as square roots (see Lemma 3.2.4), the number r is constructible with ruler and compass from the three numbers a, b, c in Ki . Thus given a number r ∈ Kn, it is constructible with ruler and compass from finitely many numbers in Kn−1; each of these numbers is itself constructible from finitely many numbers in Kn−2 and so on. So eventually, r is constructible with ruler and compass from finitely many numbers in K,thatis,viaLemmaA.4.3, from the nodes of the given configuration. Conversely, consider a finite extension of the given configuration and its field L of coordinates. We must prove the existence of a sequence of field extensions as in the statement, ending with Kn = L. Of course it suffices to prove the same result for every elementary extension (see Definition A.3.3). Furthermore, since in an elementary extension one adds finitely many nodes, it suffices to prove the result for the addition of one single node. We shall consider separately the cases of adding a node which is the intersection of two straight lines, of a straight line and a circle, or of two circles. So let us consider a geometric configuration in standard position with field of co- ordinates M. Two intersecting lines joining nodes of the configuration admit equa- tions of degree 1 whose coefficients are elements of M (see Lemma A.4.4). The solution of the corresponding system of two equations aX + bY = c rX + sY = t is simply det cb det ac = ts = rt X ab ,Y ab . det rs det rs These two quantities lie still in M. Next consider the case of a line joining two nodes and intersecting a circle cen- tered at a node and passing through another node. We have to solve a system of equations aX + bY = c (X − r)2 + (Y − s)2 = t2 = = c ∈ with all coefficients in M (see Lemma A.4.4 again). If a 0, we obtain Y b M. Introducing this value into the second equation, it remains an equation of degree 2 in X with coefficients in M. Solving that equation yields the value of X.Ifa = 0, then = 1 − X a (c bY). Introducing this value into the second equation yields an equation of 366 A Constructibility degree 2 in Y , with all coefficients still in M. Solving this equation yields the value = 1 − of Y , from which by further arithmetical operations we get the value X a (c bY). Thus in order to add to M the coordinates of the intersection points it suffices, if needed, to add the solutions of one equation of degree 2 with coefficients in M. The solutions of an equation
αZ2 + βZ + γ = 0 of degree 2 with coefficients in M are given by the well-known formula −β ± β2 − 4αγ . 2α If l = β2 − 4αγ is in M, the solutions of the equation lie in M as well. Otherwise l is a root of the polynomial Z2 −l2. This polynomial has coefficients in M and is ir- reducible in M[X], because l ∈ M. By Theorem A.1.5, it is the minimal polynomial of l over M. Therefore, still by Theorem A.1.5, M(l) is an extension of degree 2 of M and this extension contains the roots of the given second degree equation. This remains the case for two intersecting circles, both centred at some node and passing through a node. We have this time to solve a system of equations (X − a)2 + (Y − b)2 = c2 (X − r)2 + (Y − s)2 = t2 with coefficients in M (see Lemma 3.11.11). Subtracting these two equations yields
−2aX + 2rX + a2 − r2 − 2bY + 2sY + b2 − s2 = c2 − t2.
This is an equation of degree 1 in X and Y with coefficients in M.Itnowremains to solve the system comprising this equation of degree 1 and one of the two orig- inal equations of degree 2; but this is precisely the second case treated above: the intersection of a line and a circle.
Corollary A.4.6 Consider a geometric configuration in standard position and its field K of coordinates. If a real number r is constructible from that configuration, then it is algebraic over K with a minimal polynomial whose degree is a power of 2.
Proof By Theorem A.4.5 and adopting the notation used there, r belongs to some n extension Kn of K. By Proposition A.1.2,dim[Kn : K]=2 . By Proposition A.1.3, r is algebraic over K. By Theorem A.1.5, the degree of the minimal polynomial of r over K is equal to dim[K(r) : K]. But since K ⊆ K(r) ⊆ Kn, Proposition A.1.2 implies n 2 = dim[Kn : K]=dim Kn : K(r) × dim K(r) : K . So dim[K(r) : K] is a power of 2 as a divisor of a power of 2.
Let us highlight the fact that Corollary A.4.6 is only a necessary condition. A.4 Constructibility Versus Field Theory 367
Corollary A.4.7 Consider a geometric configuration in standard position and its field K of coordinates. A number r which is transcendental over K is not con- structible from the given configuration.
Proof This follows by Corollary A.2.2, since a transcendental number over K is non-algebraic. Appendix B The Classical Problems
This appendix is devoted to proving the impossibility—via ruler and compass constructions—of the three so-called “classical problems” of Greek geometry: du- plicating the cube, trisecting the angle and squaring the circle.
B.1 Duplicating the Cube
See Sect. 2.5 for an historical approach to this problem.
Proposition B.1.1 Duplicating the cube is impossible via ruler and compass con- structions.
Proof Taking as unit length the√ edge of the given cube, we must construct a cube√ of volume 2, that is, of edge 3 2. The question is therefore to determine if 3 2 is constructible from the basic configuration (see Definition A.3.6). The field of coordinates of the basic configuration is Q (see Proposition A.4.2). Choosing d = 2 in the Eisenstein criterion (Theorem A.2.2), the polynomial X3 − 2 is irreducible√ over Q. Therefore, by Theorem A.2.2, X3 − 2 is the mini- 3 mal polynomial of 2 over√ Q. Since the degree 3 of that polynomial is not a power of 2, by Corollary A.4.6, 3 2 is not constructible.
B.2 Trisecting the Angle
See Sect. 2.3 for an historical approach to this problem.
Proposition B.2.1 Trisecting the angle is impossible via ruler and compass con- structions.
F. Borceux, An Axiomatic Approach to Geometry, DOI 10.1007/978-3-319-01730-3, 369 © Springer International Publishing Switzerland 2014 370 B The Classical Problems
Fig. B.1
Proof Start with the basic configuration (see Definition A.3.6) whose field of coor- dinates is the field Q of rational numbers (see Proposition A.4.2). Draw an equilat- eral triangle ABC on the segment joining A = (0, 0) and B = (0, 1) (see Proposi- tion 3.1.5 and Fig. B.1). The measure of the angle at A is 60 degrees. If this angle can be trisected, the trisector will meet the circle of centre A and radius 1 (that is, the circle passing through B) at the point P with coordinates P = (cos 20◦, ± sin 20◦). To prove the impossibility of this trisection, it thus suffices to prove that cos 20◦ is not constructible over the rationals (see Proposition A.3.9). We start from the well-known trigonometric formula cos 3θ = 4 cos3 θ − 3 cos θ. Choosing θ = 20◦, we conclude that cos 20◦ is a root of the polynomial 1 q(X) = 4X3 − 3X − , 2 ◦ = 1 ◦ just because cos 60 2 . Equivalently, cos 20 is a root of the polynomial 2q(X) = 8X3 − 6X − 1. Perform now a change of variables Y + 1 X = ,Y= 2X + 1. 2 The polynomial 2q(X) transforms as r(Y) = Y 3 + 3Y 2 − 3. Putting d = 3 in the Eisenstein criterion (Theorem A.2.2), we get that r(Y) is irreducible in Q[X]. Therefore 2q(X) is already irreducible in Q[X]. By Theo- rem A.1.5 1 3 1 p(X) = 2q(X) = X3 − X − 8 4 8 is then the minimal polynomial of cos 20◦ over Q. By Corollary A.4.6, since the degree 3 of that polynomial is not a power of 2, cos 20◦ is not constructible. B.3 Squaring the Circle 371
B.3 Squaring the Circle
See Sect. 2.4 for an historical approach to this problem.
Lemma B.3.1 To prove the impossibility of squaring the circle with ruler and com- pass, it suffices to prove that the number π is transcendental over the field Q of rational numbers.
Proof Squaring√ a circle of radius 1, thus of area π, is the construction√ of a square with side π. The problem thus reduces to proving that π is not constructible from the basic geometric configuration (see Definition A.3.6). But since multipli- cations (see Proposition 3.6.8) and square roots√ (see Proposition 3.2.4) can be per- formed by ruler and compass constructions, π is constructible from the basic con- figuration if and only if π is constructible from that configuration. Now the field of coordinates of the basic configuration is the field Q of rational numbers (see Proposition A.4.2). Thus if π is transcendental over Q, by Corollary A.4.7, it is not constructible with ruler and compass from the basic configuration.
The first proof of the transcendence of π over the rationals is due to Lindemann, in 1882. Many other proofs are known: some of them are rather elegant, using so- phisticated theories which are definitely beyond the scope of this book; others rely on more elementary but highly technical results. We propose such a technical ele- mentary proof, based only on results explicitly proved in this “trilogy”; in fact our proof is fairly close to that of Lindemann. A key ingredient of the proof will be the theory of symmetric polynomials de- veloped in Appendix E of [5], Trilogy II. A polynomial p(X1,...,Xn) with coeffi- cients in Z is symmetric when it remains unchanged under every permutation of the variables. The “elementary” symmetric polynomials with two variables are
X1 + X2,X1X2, with three variables
X1 + X2 + X3,X1X2 + X1X3 + X2X3,X1X2X3, and so on. The so-called “structural theorem” for symmetric polynomials over the integers attests that
Given a symmetric polynomial p(X1,...,Xn,) with coefficients in Z, there exists a poly- nomial q(X1,...,Xn) with coefficients in Z, with degree less than or equal to the degree of p(X1,...,Xn), such that p(X1,...,Xn) = q σ1(X1,...,Xn),...,σn(X1,...,Xn)
where the σi (X1,...,Xn) are the elementary symmetric polynomials with n variables. (See Theorem E.2.1 and Remark E.2.2 in [5], Trilogy II.)
Theorem B.3.2 (Lindemann) The number π is transcendental over Q. 372 B The Classical Problems
Proof We develop the proof by a reductio ad absurdum: that is, we suppose that π is algebraic over Q and we infer a contradiction. √ We show first that the algebraicity of π forces that of iπ, with i = −1 ∈ C.It suffices to consider the field extensions
Q ⊆ Q(π) ⊆ Q(π)(i).
By assumption, π is algebraic over Q and certainly, i is algebraic over Q(π), since it is root of the polynomial X2 + 1 (see Definition A.1.4). By Theorem A.1.5, both extensions are finite dimensional, thus so too is the composite of these, by Propo- sition A.1.2. Therefore the element iπ ∈ Q(π)(i) is algebraic over Q by Proposi- tion A.1.3. Multiplying the corresponding minimal polynomial by a common mul- tiple of the denominators of its coefficients, we obtain an irreducible polynomial q(X) ∈ Z[X]
n n−1 1 q(X) = anX + an−1X +···+a1X + a0,ai ∈ Z such that q(iπ) = 0 (see Theorem A.1.5 again). Viewed as a polynomial q(X) ∈ C[X], we know that q(X) is a product of factors of degree 1 (see Theorem F.3.3 and Corollary A.6.6 in [5], Trilogy II). Let us write
q(X) = an(X − r1)(X − r2) ···(X − rn), ri ∈ C with r1 = iπ and an ∈ Z. We next use the complex exponential function
+ ea bi = ea(cos b + i sin b), a,b ∈ R.
Let us recall that exey = ex+y for x,y ∈ C (see Proposition F.2.12 in [5], Trilogy II). We have in particular
eiπ = cos π + i sin π =−1 that is er1 + 1 = 0.
This implies of course er1 + 1 · er2 + 1 ··· ern + 1 = 0.
Expanding the product, this can be re-written as erj = 0 J ⊆{1,...,n} j∈J with the usual convention that the empty product equals 1. But the property ex+y = exey of the exponential allows us to further rewrite the equality as e( j∈J rj ) = 0. J ⊆{1,...,n} B.3 Squaring the Circle 373
For every integer m ∈{1,...,n}, we shall now prove the existence of a polyno- mial fm(X) ∈ Z[X] whose roots in C are precisely all the quantities rj ,J⊆{1,...,n}, card(J ) = m, j∈J where card(J ) indicates the number of elements of J . Trivially, when m = 1, it suffices to choose f1(X) = q(X) ∈ Z[X]. We shall write further 1 q(X) = (X − r1)(X − r2) ···(X − rn) = q(X). an The coefficients of q(X) are thus rational numbers; but they are also the elemen- tary symmetric expressions in r1,...rn. This proves that all these expressions are rational numbers. 1 2 − The polynomial f2(X) must admit the 2 (n n) roots = + rjj rj rj ,j We consider first the polynomial in C[X] = − ∈ C[ ] f 2(X) (X rjj ) X j rjjj = rj + rj + rj ,j g(X) = f1(X) · f2(X) ···fn(X) with coefficients in Z, whose roots in C are precisely the quantities rj , ∅=J ⊆{1,...,n}. j∈J 374 B The Classical Problems Of course it can be the case that some of the roots are equal to 0, that is, that g(X) contains some factors X. Canceling all these factors X, we end up with a polynomial h(X), still with coefficients in Z, and whose roots are the non-zero quantities rj = 0, ∅=J ⊆{1,...,n}. j∈J Notice that r1 = iπ remains one of the roots of h(X). Thus h(X) has degree k ≥ 1; we shall write it as k k−1 h(X) = bkX + bk−1X +···+b1X + b0,bi ∈ Z. Notice that b0 = 0 because 0 is not a root of h(X). Of course bk = 0 since we have called k the degree of h(X). Let us now go back to the equality er1 + 1 · er2 + 1 ··· ern + 1 = e( j∈J rj ) = 0. J ⊆{1,...,n} The quantity j∈J rj is equal to zero for all the zero-roots of g(X), but also when J is the empty set. Since e0 = 1, the equality can therefore be re-written as