On Triangles as Real Analytic Varieties of an Extended Fermat Equation
Giri Prabhakar∗
Abstract On the one hand, triangles play an important role in the solution of elliptic curves, as exemplified in studies by Tunnell on the congruent number problem, or Ono relating arbitrary triangles to elliptic curves. On the other hand, Frey-Hellegouarch curves are deeply connected to Fermat’s equation. Thus, we are motivated to explore the relationship between triangles and Fermat’s equation. We show that triangles in Euclidean space can be exactly described by real analytic varieties of an extension of the Fermat equation. To the 3 best of our knowledge, this formulation has not been explored previously. Let S = {(a, b, c) ∈ R≥0 | (c > φ φ φ 0 ∧ ∃ φ ∈ R≥1)[a + b = c ]} be the set of real non-negative solutions to the Fermat equation extended to admit real valued exponents φ ≥ 1, which we term the Fermat index. Let E be the set of triplets of side lengths of all non-degenerate and degenerate 2-simplices. We show that S = E, and that each triangle has a unique Fermat index. We show that there exists a deformation retraction F based on the extended Fermat equation, that maps all possible sides a ≤ c and b ≤ c of 2-simplices (a, b, c) ∈ S, to the retract c of fixed length at a fixed angle to a. However, (a, b, c) is also governed by the law of cosines; thus an alternate map P exists based on the law of cosines, which must be geometrically identical to F . Setting F = P generates an algebraic equation that yields a class of polynomials closely related to the 2mth cyclotomic polynomial. We propose criteria for the irreducibility of these polynomials in Q[x], and present an extension of the Schoenemann-Eisenstein theorem, which we employ along with additive shifts and Gauss’s Lemma to prove the propositions. The approach yields new geometric insights about triangles: in particular, a simple proof that no rational triangle exists with a Fermat index of 4, and in general, the result that no non-degenerate oblique rational triangle can have an integer Fermat index. Keywords: Euclidean geometry, Pythagorean theorem, Diophantine equations 2010 MSC: 51N20, 51M04, 51M05
1. Introduction
Trigonometric methods are indispensable in a wide spectrum of modern scientific and technological fields, but the fundamental representations and analyses of triangles is one of the oldest and most established subjects [1]. Therefore, the expectation that studies in these areas have little that is new to offer should come as no surprise. Nevertheless, there is still a possibility that one comes across undiscovered insight, such as the work by Kendig extending the interpretation of Heron’s formula to the complex plane[2]. Considerations in plane trigonometry result from the root presumption of the Pythagorean theorem. The cosine law is a prime example that relates all three sides of a triangle, and we will call this the Pythagorean representation of a triangle. Note that the term “triangle” in this paper refers to 2-simplices in Euclidean space. On the one hand, many studies arising from the congruent number problem and generalizations of this problem have shown that rational triangles are closely related to solutions of elliptic curves[1, 3–6]. Tunnell related the congruent number problem to the Birch and Swinnerton-Dyer conjecture [6]. Ono describes
∗Corresponding author Email address: [email protected] (Giri Prabhakar) On Triangles and an Extended Fermat Equation Giri Prabhakar a method by which it is possible to generate an infinite number of elliptic curves over an algebraic field by replacing right triangles in the congruent number problem with arbitrary triangles [7]. On the other hand, elliptic curves are also strongly connected to Fermat’s equation; Frey-Hellegouarch curves of the form 2 m m m m m 3 y = x(x − a )(x + b ) are connected to Fermat’s equation a + b = c where (a, b, c) ∈ Z>0 and m is an odd prime [8]. The contributions of Frey, Serre, Shimura, Taniyama and others on the modularity of these elliptic curves over the field of rational numbers played a critical role in the breakthrough and celebrated proof of Fermat’s theorem by Wiles[8–15]. The strong relationship between triangles and elliptic curves on the one hand, and elliptic curves and Fermat’s equation on the other, motivated us to study the connection between triangles and an extension of Fermat’s equation. We have discovered an alternate representation of the relationship between the sides of a triangle, that is not a general consequence of the Pythagorean theorem. We call this the Fermat representation. In addition to making connections to Kendig’s observations and cyclotomic polynomials, we also present an extension of the Schoenemann-Eisenstein theorem and fresh insight into the properties of triangles that, to the best of our knowledge, have been previously unexplored. We begin in this section by showing that the extended Fermat equation offers an alternate, exact, real analytic description of any triangle. It is naturally interesting to then consider the relationship between the Fermat and Pythagorean representations, and its consequent implications. Therefore, in section 2, we estab- lish the geometric framework which is the basis upon which the Fermat and Pythagorean representations are related in section 3. The relationship suggests the formulation of an analytic function, which takes the form of univariate polynomials for positive integer values of the exponent. We then propose the irreducibility of the resultant polynomials in section 4, and prove these propositions by means of an extension of the Schoenemann-Eisenstein theorem [16] and Gauss’s Lemma [17]; these are our main results.
3 φ φ φ Theorem 1. Let S = {(a, b, c) ∈ R≥0 | (c > 0 ∧ ∃ φ ∈ R≥1)[a + b = c ]} be triplets of non-negative real numbers that satisfy aφ + bφ = cφ. (1)
The number φ will be called Fermat index. Let E be the set of triplets of side lengths of all non-degenerate and degenerate 2-simplices. Then, S = E, and each 2-simplex has a unique Fermat index. Proof. First let b > a and φ ∈ (1, ∞); and consider the expression
(a + b)φ = bφ[1 + (a/b)]φ,
φ which, on expanding [1 + (a/b)] as a convergent binomial series with k ∈ Z≥0, becomes
∞ X φ φ bφ (a/b)k, where := φ(φ − 1) ... (φ − k + 1)/k!, k k k=0
∞ X φ = bφ + φabφ−1 + bφ (a/b)k, k k=2 which holds for both integer and non-integer real values of φ. Since aφ−1 < φbφ−1,
∞ X φ (a + b)φ > bφ + aφ + bφ (a/b)k. (2) k k=2
th P∞ φ k The k term of k=2 k (a/b) is
k tk = [φ(φ − 1) ... (φ − k + 1)/k!](a/b) , so that tk + tk+1 = tk{1 + [(φ − k)/(k + 1)](a/b)}. 2 On Triangles and an Extended Fermat Equation Giri Prabhakar
n Let n be the first integer that is greater than φ; then tn = [φ(φ − 1) ... (φ − n + 1)/n!](a/b) ≥ 0, and in fact φ ∀ i ∈ Z≥0 3 r = n + 2i, tr ≥ 0, because r is either 0 (when φ is an integer) or contains the product of an even number of negative terms in the numerator (when φ is a non-integer real number). Since 1 < φ < r, it follows that r − φ < r − 1 < r + 1. Therefore, φ − r < 0 and |(φ − r)/(r + 1)| < 1 =⇒ 0 < tr + tr+1 ≤ tr, with the equality applying when φ = n − 1. Hence
∞ ∞ n−1 X φ X X (a/b)k = t = t + (t + t ) + (t + t ) + ..., k k k n n+1 n+2 n+3 k=2 k=2 k=2 which implies that n−1 ∞ X X φ 0 < t < (a/b)k < (1 + a/b)n, k k k=2 k=2 and from (2) leading to the inequality
(a + b)φ > aφ + bφ = cφ, from which we conclude that a + b > c. Since c > a and c > b, we get b + c > a and c + a > b. Thus, for a 6= b and φ > 1, (1) implies the triangle inequalities, which are necessary and sufficient for the triplet (a, b, c) to form a 2-simplex [18]. When φ = 1, (1) implies that (a, b, c) is a degenerate 2-simplex. For a = b, the triangle inequality is trivially satisfied in S, since in (1), 21/φa = c (taking only the positive real root), thus 2a > c, but also c > a. Clearly, a triplet (a, b, c) satisfying (1) is equivalent to (λa, λb, λc) ∀ λ ∈ R, and hence S ⊂ P2(R). Conversely, consider a triangle (a, b, c) ∈ E which is either degenerate or non-degenerate (in the former case with c and at least one of a, b non-zero), with the length of all non-zero sides greater than 1. The assumption about the length does not lose generality, since E ⊂ P2(R), but is simpler for analysis. Let c be (one of) the longest side(s). We have, in the degenerate case (without loss of generality) a+b = c, a < c, b ≤ c, and in the non-degenerate case, a + b > c, a ≤ c, b ≤ c. First we assume strict inequality, and defer the equality cases to Lemmas 2 and 3 (where they will be shown to correspond to the extreme cases φ → 1 and φ → ∞). Thus, a + b > c, a < c, b < c. Note also that ln(c) > ln(a) and ln(c) > ln(b). For every real number x > 1, let y(x) = cx/(ax + bx). Then
y0(x) = cx{ax[ln(c) − ln(a)] + bx[ln(c) − ln(b)]}/(ax + bx)2
= [cx/(ax + bx)]{ax[ln(c) − ln(a)] + bx[ln(c) − ln(b)]}/(ax + bx), thus y0(x) = y(x){ax[ln(c) − ln(a)] + bx[ln(c) − ln(b)]}/(ax + bx). (3) Let ε = min([ln(c) − ln(a)], [ln(c) − ln(b)]). Moreover, when we take away the terms ln(a) and ln(b) in (3) we see that 0 < ε < {ax[ln(c) − ln(a)] + bx[ln(c) − ln(b)]}/(ax + bx) < ln(c), hence we have 0 < εy(x) < y0(x) < ln(c)y(x), and y is smooth and real analytic, so that for some value of z > 1 Z z Z z Z z εdx < dy/y < ln(c)dx, 1 1 1 hence ε(z − 1) < ln[y(z)/y(1)] < ln(c)(z − 1) =⇒ y(1)eε(z−1) < y(z) < y(1)e[ln(c)](z−1).
Therefore, it is possible to choose z = z1 and z = z2 such that
ε(z1−1) y(1) < y(1)e = 1 =⇒ z1 = 1 − (1/ε)ln[y(1)] > 1, 3 On Triangles and an Extended Fermat Equation Giri Prabhakar
ln(c)(z2−1) y(1)e = 1 =⇒ z2 = 1 − [1/ln(c)]ln[y(1)] > 1.
Since ε < ln(c), 1 − (1/ε)ln[y(1)] > 1 − [1/ln(c)]ln[y(1)]. Thus, there exists 0 < 1, 2 < 1 such that 1 − 1 ≤ y(z) ≤ 1 + 2 whenever 1 − [1/ln(c)]ln[y(1)] ≤ z ≤ 1 − (1/ε)ln[y(1)]. Since y is continuous in this interval, from Cauchy’s Intermediate Value Theorem [19], there exists φ ∈ [1 − [1/ln(c)]ln[y(1)], 1 − (1/ε)ln[y(1)]] such that y(φ) = 1, or cφ = aφ + bφ. Moreover, y0(x) > εy(x) > 0 ∀ x. Then, from Cauchy’s Mean Value Theorem [20], there exists some ξ such that ∀ δ > 0, [y(φ + δ) − y(φ)]/δ = y0(ξ) > 0, hence y(φ + δ) > y(φ), and therefore φ is the unique value at which y(x) = 1. For completeness, the degenerate case a + b = c, a < c, b ≤ c, and the equality case a + b > c, a ≤ c, b ≤ c must also be considered. Lemma 2. The Fermat index of a degenerate 2-simplex is 1. Proof. Let γ be the angle opposite the longest side c of triangle (a, b, c) with Fermat index φ. As φ → 1, a2 + b2 − c2 −2ab (a + b)2 → c2, therefore, cos γ = → = −1. Hence, 2ab 2ab lim (γ | aφ + bφ = cφ)γ = π. φ→1 Clearly the converse is true since a + b = c =⇒ φ = 1, and both a, b cannot be 0 together. Lemma 3. The Fermat index of isoceles triangles, in which equal sides are the longest sides, is unbounded,
∀ (a, b, c) ∈ S, lim φ → ∞. c→max(a,b) Proof. We first assume b > a. As c → b, for some small positive real , let c = b + . Dividing the equation aφ + bφ = (b + )φ throughout by b and expanding the equation as a (convergent) binomial series, we get a φ ( )φ + 1 = (1 + )φ = 1 + + δ(), b b b ∞ X φ φ where δ() = (/b)k, with := φ(φ − 1) ... (φ − k + 1)/k!, k k k=2 (a/b)φ δ() thus, = + , φ b φ φ and since lim→0 δ() → 0, lim→0(a/b) /φ → 0; the limit on the left hand side of the equation can be evaluated with the application of the L’Hˆopitalrule [21], by differentiating the numerator and denominator by φ in the interval (1, ∞) to get lim(a/b)φln(a/b) → 0 =⇒ (a/b)φ → 0 =⇒ lim(φ)ln(a/b) → −∞, →0 →0 φ φ φ and since a < b, ln(a/b) < 1, and lim→0 φ → ∞. Conversely, given φ → ∞ in a + b = c =⇒ (a/c)φ + (b/c)φ = 1, the left hand side goes to 0 since a < c and b < c, unless, as φ → ∞, either a → c or b → c, so that the triangle must have two equal longest sides if φ becomes unbounded. As a corollary, we also see that the Fermat index of an equilateral triangle is unbounded. An alternate way to prove this is: let 1 and 2 both be small positive real numbers, and let a = c − 1 and b = c − 2. φ φ Now letting 1 → → 0 and 2 → → 0, we get from the last equation 2[c − ] = c , which, on taking th positive real φ roots (since c and c − ∈ R>0), leads to the equation 1/φ c ln(2) 2 = =⇒ φ = c c − ln( ) c − from which we conclude that lim→0 φ → ∞; conversely limφ→∞ → 0. From Theorem 1, and Lemmas 2 and 3, we see that 3 ∀ (a, b, c) ∈ R≥0 [(a, b, c) ∈ S ⇐⇒ (a, b, c) ∈ E], therefore, S = E. 4 On Triangles and an Extended Fermat Equation Giri Prabhakar
Corollary 1. If (a, b, c) ∈ S and φ is the Fermat index of (a, b, c), then
x x x δ(a, b, c) = {(a , b , c ) | x ∈ [1, φ]} ⊂ S.
Proof. Since aφ + bφ = cφ, ∀ x ∈ [1, φ], (ax)φ/x + (bx)φ/x = (cx)φ/x ∧ φ/x ≥ 1 =⇒ (ax, bx, cx) ∈ S from Theorem 1 and Lemma 2. The condition is trivially satisfied for a degenerate triangle since φ = 1. For an isoceles triangle in which one or both of a and b equal c, the triangle inequality for (ax, bx, cx) is clearly satisfied regardless of the value of x ≥ 1. Since (ax, bx, cx) is a triangle for 1 ≤ x ≤ φ, with the triangle being degenerate at x = φ, it follows that ax + bx > cx for 1 ≤ x < φ. From Heron’s formula, A(x) = ps(s − ax)(s − bx)(s − cx), is the area of the triangle, where s = (ax + bx + cx)/2, so that s − cx = (ax + bx − cx)/2. Intuitively, one may visualize the triangle (ax, bx, cx) as “stretching”, with a continuously decreasing area; thus A(φ) = 0, and A(x) is imaginary for x > φ, questions around which have been explored by Kendig [2]). Triangles with imaginary areas therefore have Fermat indices less than 1. Corollary 1 can be used to determine the Fermat index φ of a non-isoceles and non-degenerate triangle (a, b, c). If one took the logarithms of the sides of any triangles in δ(a, b, c) with Fermat index x, the triplet thus obtained: [xln(a), xln(b), xln(c)] is the same point in P2(R). Let (ar, br, cr) be any triangle in δ(a, b, c). We therefore have ln(ar)/ln(br) = ln(a)/ln(b). Without losing generality, we will assume that c = 1 (any triangle may be brought to this condition by normalizing the sides by c). Let us assume that (a0, b0, 1) is the right triangle in the set δ(a, b, 1). Let a0 := t0 < 1 be the initial assumption. Then
ln(b)/ln(a) ln(b0) := ln(t0)ln(b)/ln(a) =⇒ b0 := t0 . q p 2 [2ln(b)/ln(a)] Compute the corrected value a0 := t1 = 1 − b0 = 1 − t0 , and iterate until convergence. Thus, we summarize the algorithm as Algorithm 1.
Initialize 0 < t0 < 1, r = ln(b)/ln(a) For i = 1, 2, 3, ··· q 2r ti := 1 − ti−1 (4)
Terminate at i = k when |tk − tk−1| < 2 a0 = lim tn ≈ tk, φ = 2ln(a0)/ln(a) = ln(1 − a0)/ln(b). n→∞
The convergence of the iterations in Algorithm 1 can be seen as follows: the sequence of values tk are 3 defined in R≥0 and hence in a metric space; the relevant subspace to which all tk belong, we will call tδ, and ∀ k, tk ∈ [0, 1], hence tδ is compact. Assume without loss of generality that b > a (the case b = a is easily 2 solvable: φ = ln(2)/ln(c/a)). Let us rewrite (4) with xi = 1 − ti to obtain
r xi = (1 − xi−1) . (5)
The solution to (5) is the intersection of the curve f(x) = (1 − x)r and g(x) = x; hence this point is unique for a given value of r, is the only solution, and always exists for x ∈ [0, 1]. Moreover, since f(x) is continuous and defined on the same domain and range as g(x), f(x) = x is a Brouwer fixed point. Let this fixed point be (x∗, f(x∗)). The slope of the curve is f 0(x) = −r(1 − x)r−1, and is strictly decreasing with the limits f 0(0) = −r and f 0(1) = −∞. Therefore, for x < x∗, f(x) > f(x∗), and for x > x∗, f(x) < f(x∗). Since ∗ r < 1, (5) shows that xi > 1−xi−1, and xi−1 > 1−xi−2, ∴ xi > xi−2. Now let x0 < x , =⇒ f(x0) = x1 > ∗ ∗ ∗ ∗ x and f(x1) = x2 < x , but x0 > x2 =⇒ f(x0) = x1 < f(x2) = x3, thus x0 < x2 < x , and x > x3 > x1. In this manner one of the subsequences generated by has the ordering sh = {x0 < ··· < xi−2 < xi < ∗ ∗ xi+2 < ··· < x }, and the other, sl = {x > ··· > xi+1 > xi−1 > ··· > x3 > x1}. Let us define the distance ∗ ∗ ∗ ∗ d(x, y) = |x−y|. Then ∀ i ∈ Q≥−1 , d(xi+1, x ) < d(xi−1, x ) =⇒ ∃ q ∈ (0, 1] 3 d(xi+1, x ) ≤ qd(xi−1, x ), 5 On Triangles and an Extended Fermat Equation Giri Prabhakar
∗ ∗ ∗ ∗ with such a q being chosen from the interval (d(x2, x )/d(x0, x ), 1) for sh and (d(x3, x )/d(x1, x ), 1) for ∗ φ/2 sl. Thus, from the Banach Fixed Point Theorem [22], the recursion (5) converges to x = 1 − a as two convergent sequences sl and sh under the metric d(x, y) = |x − y|. We now explore the relationship between the Fermat index and the geometry of the triangle. Lemma 4. For obtuse triangles, 1 ≤ φ < 2, and for acute triangles 2 < φ < ∞. Proof. Note from Lemma 2 that the line segment is an obtuse degenerate 2-simplex, so this takes care of the case φ = 1, and we will assume φ ∈ (1, ∞). Let us assume a triangle (a, b, c) with c the longest side, and x γ the (largest) angle opposite c. Without loss of generality, let a, b, c > 1. For any x ∈ R>0, let h(x) = c and l(x) = ax + bx. From Theorem 1, there is a unique number x = φ such that (1) holds. Set x = φ + y, so that h(φ + y) = c(φ+y) = cφeyln(c) = aφeyln(c) + bφeyln(c), and similarly l(φ + y) = aφeyln(a) + bφeyln(b). Since ln(c) > ln(a) and ln(c) > ln(b), for 1 < φ + y < φ =⇒ 1 − φ < y < 0, l(φ + y) > h(φ + y) and for φ < φ + y < ∞ =⇒ 0 < y < ∞, l(φ + y) < h(φ + y). Therefore,
aφ+y + bφ+y > cφ+y, 1 − φ < y < 0; aφ+y + bφ+y = cφ+y, y = 0; (6) aφ+y + bφ+y < cφ+y, y > 0.
When the Fermat index of (a, b, c) is 1 < φ < 2, (6) implies a2 + b2 < c2, and cos γ = (a2 + b2 − c2)/2ab < 0, thus π/2 < γ < π and the triangle is obtuse. When 2 < φ < ∞, then from (6), a2 + b2 > c2, and cos γ = (a2 + b2 − c2)/2ab > 0, therefore, γ < π/2 and the triangle is acute. We further note from the preceding definitions of h(φ + y) and l(φ + y) that,
If y2 > y1 > 0, then
l(φ + y2) − h(φ + y2) < l(φ + y1) − h(φ + y1) =⇒ a(φ+y2) + b(φ+y2) − c(φ+y2) < a(φ+y1) + b(φ+y1) − c(φ+y1) (7)
and if 1 − φ < y2 < y1 < 0, then
l(φ + y2) − h(φ + y2) > l(φ + y1) − h(φ + y1) =⇒ a(φ+y2) + b(φ+y2) − c(φ+y2) > a(φ+y1) + b(φ+y1) − c(φ+y1)
We will now establish a geometric framework that enables systematic analysis of all possible triangles. The geometric framework comprises of definitions and terminology, along with the construction of a geometric mapping that will enable the creation of an effective procedure, in other words, an algorithm, for choosing triangles in an ordered sequence.
2. Geometric Framework
2.1. Definitions and Terminology The following definitions and terminology will hold unless explicitly mentioned otherwise. The triplet (a, b, c) will always refer to an element of E (equivalently, of S), with non-zero b and c (one of) the longest side(s), and we will assume the triplet (a, b, c) scaled such that all non-zero values of a, b, c > 1 without loss of generality. We will also call c the hypotenuse, regardless of whether (a, b, c) is right, oblique or degenerate. This is a convenient choice for us as we need to frequently refer to the longest side. A triangle is called integer if the sides all have positive integer lengths, primitive if the sides of an integer triangle do not have a greatest common divisor except 1, and rational if the sides are rational numbers. The Fermat index φ of a triangle (a, b, c) will be alternatively be indicated as φ(a, b, c), or other parametric forms introduced when needed. The angle between a and c is denoted by θ, while that between b and c is denoted by λ. The angle opposite c is denoted by γ. We will denote altitudes perpendicular to sides a and b as ha (= c sin θ) and hb (= c sin λ) respectively, and the projection of c on sides a and b, as α (= c cos θ) and β (= c cos λ) respectively. Therefore 6 On Triangles and an Extended Fermat Equation Giri Prabhakar
(α, ha, c) and (β, hb, c) are right triangles. As a convention, we will adopt the symbol p (respectively, r) whenever α (respectively, β) is an integer, and p/q (respectively, r/s) whenever α (respectively, β) is a proper fraction, with p, q, r, s ∈ Z>0. Further, when α = p/q, we scale the triangle (a, b, c) to (qa, qb, qc) = (A, B, C). Thus, in triangle (A, B, C), α = p ∈ Z>0. Note that (A, B, C) = (a, b, c) ∈ P2(R), but the triangle (A, B, C) cannot be primitive. We will denote similarity of the triangles by (A, B, C) ∼ (a, b, c) (which is equality in P2(R)). The notation is depicted in Fig. 1.
Figure 1: Notation for a triangle (a, b, c).
2.2. Triangle Deformation Retraction Map Having established the basic definitions and terminology, we will now construct a geometric map between orthogonal and oblique triangles, with a goal that all triangles can be selected in a systematic manner allowing us to define an alogrithm for the selection process. First, we describe an example of how an oblique triangle maps to a corresponding orthogonal triangle, and then we will investigate the nature of this mapping.
2.2.1. Construction relating oblique and orthogonal triangles
Figure 2: Construction Choose the point of intersection of two lines in general linear position in R2 as the origin of a two dimensional Cartesian coordinate system. Draw a circle with radius c, and consider only the first quadrant (hence all coordinates are non-negative). Select a right triangle OJL, with hypotenuse c, at an angle θ to the X−axis, and restricted to the interval [0, π/2). Let the base of the right triangle be α = c cos θ, and height 7 On Triangles and an Extended Fermat Equation Giri Prabhakar ha = c sin θ. Any triangle (a, b, c) sharing the hypotenuse c at angle θ to a, can be constructed by choosing a side of length a = 2ηα, with η restricted to values within [0, ηc] where ηc = min[1, c/(2α)]. The reason for the restriction of η is that the length of this side does not exceed the length of the hypotenuse. We denote such a triangle by Tη(c, α). Hence, triangle OJL is denoted by T1/2(c, α). At the limit η → 0, the triangle degenerates to a line segment (OL), which is the hypotenuse of T1/2(c, α), and is denoted by T0(c, α). The triangle Tηc (c, α) is always isoceles. Following the notation described in Sect. 2.1, γ is the angle opposite c (between a and b), and λ the angle between b and c. There can only be one obtuse angle in a triangle, and as γ is defined as the largest angle in (a, b, c), we see that 0 ≤ θ < π/2. Since, by definition, a and b cannot exceed c, we will denote the value of η at which max(a, b) = c, as ηc. When θ ≥ π/3, ηc = 1 and for θ < π/3, ηc < 1: specifically, when θ < π/3, a = c, b < c, when θ > π/3, a < c, b = c, and when θ = π/3, a = b = c. Hence we have the restriction ηc = min[1, c/(2α)], and 0 ≤ η ≤ ηc. All triangles Tη(c, α) will be called as identifiable triangles. Now we need to examine if the set of all identifiable triangles is equal to the set of all triangles in E. 3 Lemma 5. Let F = {(a, b, c) = Tη(c, α) | (a, b, c) ∈ R≥0 ∧ η, α ∈ R≥0)[max(a, b, c) = c > 0, η ∈ [0, ηc], ηc = min[1, c/(2α)], α/c ∈ (0, 1]]} be the set of identifiable triangles. Then F = E. 3 Proof. A triplet (a, b, c) ∈ R≥0 with the constraint c > 0 can be formed by choosing 3 numbers independently of each other. On further imposing the triangle inequality [18] as a condition to admit the triplet into the set E, the triplet is a triangle if the condition holds. Note that the triangle inequality is suitably modified to include degenerate cases (for example, a + b ≥ c, b + c > a, c + a ≥ b will account for the degenerate case a = 0, b 6= 0). Every triangle that can be formed from the symmetric group of (a, b, c) is the same triangle, and will be a member of E. Let us now generate the minimal set of triangles, by choosing one member from each of the sets Sym[(a, b, c)], such that c ≥ a, b. Then the angle between c and either a or b is acute. Therefore, it is possible find this triangle in F. Let Pij be a permutation operation that exchanges the element in the ith position with that the one in the jth position. For example P12(x, y, z) = (y, x, z). Also, let o(x) represent the numerical position of the element x in the triplet (a, b, c). 3 Then ∀ (a, b, c) ∈ R≥0 [(a, b, c) ∈ E ⇐⇒ Po[max(a,b,c)]3(a, b, c) ∈ F], hence F = E.
Let us denote all values of η > ηc by the symbol η+, thus ηc < η+ < ∞. Whenever a > c or b > c, then (a, b, c) 6∈ F. However, the permutation Po[max(a,b,c)]3(a, b, c) ∈ F. We therefore see that, for a right triangle with a given hypotenuse c and a fixed angle θ to a, it is possible to generate a continuous sequence of triangles parametrized by η = [0, ηc]. Each triangle in this sequence has a Fermat index φ which varies with its largest angle γ. We will now explore the relationship between φ and γ as η varies.
2.2.2. Mapping φ to γ Lemma 6. Let ω = max[θ, (π −θ)/2]. For constant c and θ, γ in the interval (ω, π) is a continuous, strictly decreasing and (hence) bijective function of φ. Proof. As implied in the proof to Theorem 1, given a triangle (a, b, c), (ax + bx)/cx, and hence ax + bx − cx, is strictly decreasing for x ≥ 1. Refer to Fig. 2; let us hold c and α constant (hence angle θ constant) while η increases in the interval [0, ηc). From the construction, we see that γ is strictly decreasing with respect to η; since θ is constant, as η increases, a = 2ηα increases, hence λ increases, and γ decreases. This implies that, since 0 ≤ γ ≤ π, cos γ is strictly increasing with respect to η. For each triangle Tη, (1) can also be written in the following alternate (dual) form:
sinφθ + sinφλ = sinφγ = sinφ(θ + λ) (8)
Equation (8) is derived by invoking the sine rule [21]: sin λ/a = sin θ/b = sin γ/c = 1/(2R) where R is the circumradius of the triangle with sides a, b and c. Consider two triangles Tη1 and Tη2 at values η1 < η2 respectively with angles γ1 > γ2 and λ1 < λ2, sides (a1, b1, c) = (α + δ1, b1, c) and (a2, b2, c) = (α + δ2, b2, c), circumradii R1, R2 and Fermat indices φ1, φ2. We will consider the following two cases: 1) γ1, γ2 ∈ (π/2, π), and 2) γ1, γ2 ∈ (0, π/2), with the condition that γ1 > γ2 in both cases.
8 On Triangles and an Extended Fermat Equation Giri Prabhakar
Case 1 (π > γ1 > γ2 > π/2): Since we have 2R1sin γ1 = 2R2sin γ2 = c, and a2 > a1 =⇒ R2sin λ2 > th R1sin λ1, the relations together yield: sin λ2/sin γ2 > sin λ1/sin γ1. Taking φ1 powers and subtracting 1 φ1 φ1 φ1 φ1 φ1 φ1 from both sides then gives [sin λ2 − sin γ2]/sin γ2 > [sin λ1 − sin γ1]/sin γ1 (note that φ1 > 1). φ1 φ1 φ1 φ1 φ1 φ1 Using sin γ2 > sin γ1 > 0 we further obtain [sin λ2 − sin γ2]/sin γ2 > [sin λ1 − sin γ1]/sin γ2, which leads to φ1 φ1 φ1 φ1 sin λ2 − sin γ2 > sin λ1 − sin γ1, (9)
φ1 φ1 φ1 in which the right hand side, from (8), gives sin λ1 − sin γ1 = −sin θ, which upon substitution in (9) yields φ1 φ1 φ1 sin θ + sin λ2 > sin γ2. (10)
φ1 Now we multiply (10) throughout with the factor (2R2) and get
φ1 φ1 φ1 a2 + b2 > c , (11) which from (7) in Lemma 4 leads to φ2 > φ1. 2 2 2 Case 2 (π/2 > γ1 > γ2 > 0): Recalling the law of cosines [21] as cos γ = (a + b − c )/2ab, η1 < 2 2 2 2 2 2 η2 ⇐⇒ γ1 > γ2 ⇐⇒ cos γ1 < cos γ2, =⇒ (a1 + b1 − c )/2a1b1 < (a2 + b2 − c )/2a2b2, ∵ a1 < 2 2 2 2 2 2 a2 ∧ b1 < b2, =⇒ a1 + b1 − c < a2 + b2 − c , and since 2 < φ < ∞, this implies from (7) in Lemma 4, that 2 2 2 2+ 2+ 2+ 2+ 2+ 2+ ∃ > 0 3 a1+b1−c1 = (a2) +(b2) −(c2) , at which value of we also have (a1) +(b1) −(c1) < 2+ 2+ 2+ (a2) + (b2) − (c2) , repeating which leads to the relationships holding ∀ > 0. Since φ1, φ2 > 2, we φ1 φ1 φ1 φ1 φ1 φ1 can choose = φ1 − 2 to get a1 + b1 − c1 = 0 < a2 + b2 − c2 , which leads to φ2 > φ1. From Cases 1 and 2, η1 < η2 ⇐⇒ γ1 > γ2 ⇐⇒ φ1 < φ2. (12) Assuming θ ≤ π/3, an isoceles triangle occurs when a = c (or b = c, if θ ≥ π/3 to begin with). For the condition a = c, γ = (π − θ)/2, and if b = c, γ = θ. When γ = ω, η = ηc, and Tηc (c, α) is isoceles, and from Lemma 3, φ is unbounded. Thus, the interval (ω, π) is a bijection of φ ∈ [1, ∞), with the special points φ(π) = φ(0) = 1, φ(π/2) = 2 and φ(γ → ω) → ∞ defining degenerate, right and isoceles triangles respectively.
Moreover, there exists a one-to-one mapping between every point on ha and a corresponding point on b or c. Thus, we have
Theorem 7. For a constant c and α, let Tη(c, α) represent any identifiable triangle in F, with η ∈ [0, ηc) ⊂ φ φ 1/φ Iηc and Fermat index φ = φη(c, α). Then, Tη(c, α) = [2ηα, fη(c, α), c], where fη(c, α) = [c − (2ηα) ] , 2 and the mapping F : T × Iηc → T with F (c, α, η) = Tη(c, α) is a deformation retraction of T ∈ ∆ onto 1 T0 ∈ ∆ .
2.3. Algorithm for Selection of Rational Triangles For a given c and θ, the mapping F orders the sequence of triangles in strictly increasing order of φ relative to η, as shown in (12). While for fixed values of c and α,(c, α, η) is strictly ordered in product order [23], the fact that (a, b, c) = P12(a, b, c) = (b, a, c) shows that for two triplets (c1, α1, η1) (c2, α2, η2) =6 ⇒
Tη1 (c1, α1) Tη2 (c2, α2). Thus, F induces only a partial ordering on S, which is a strict ordering over η (equivalently, γ) for each fixed value of c and α (equivalently, θ). Nevertheless, the ordering induced by F on S makes it possible to uniquely identify any triangle by the triplet [c, θ, γ] or, equivalently, [c, θ, 2ηα]. Hence we determine the algorithm to traverse all rational triangles as follows: Algorithm 2. Initialize c = 1.
Step 1: Fix the value of c ∈ Z>0 (without loss of generality).
Step 2: For the fixed c, fix a rational value of cos θ ∈ (0, 1] =⇒ α ∈ Q>0 ≤ c.
Step 3: With c and α thus fixed, select η ∈ Q ∈ [0, ηc], so that a = 2ηα is rational.
9 On Triangles and an Extended Fermat Equation Giri Prabhakar
Step 4: Check for the rationality of side b, discard η if b is irrational.
Step 5: Repeat steps (iii) and (iv) for all rational values of η ∈ [0, ηc]. Step 6: Set c := c + 1 and go to step (i)
Since for a fixed c and α all triangles Tη(c, α) share a common hypotenuse and θ, all maps F (c, α, η) with the same ratio α/c differ in R2 only by a scaling factor, and are (respectively) mapped to the same sequence of points (triangles) in P2(R). The ratio cos θ = α/c therefore defines an equivalence class of maps F (c, α, η), and we will denote the sequence of all triangles Tη(c, α) in this class by the symbol τ(c, α). This equivalence property is later used in simplifying our analysis over rational triangles by restriction to integer triangles. In Algorithm 2, α ∈ Q>0, therefore, α can be either an integer or a proper fraction. In further analysis, we will be considering both cases of α.
2.4. Pythagorean Representation Consider a map P : T × I → T with P (c, α, η) = Tη(c, α) = [2ηα, p(c, α, η), c], where p(c, α, η) = [c2 + 4ηα2(η − 1)]1/2 derives from the law of cosines [21, 24]. This can be easily seen by applying the law of cosines to the side b = LQ in Fig. 2, so that b = [c2 +a2 −2(a)(c)cos θ]1/2 = [c2 +(2ηα)2 −2(2ηα)(c)(α/c)]1/2. For a given c, θ and η, each such triangle generated by P will also have a Fermat index, φη(c, α), which depends only on c, α and η, and the latter is also defined for the triangle at η generated by the map F . Having established the necessary geometric framework, in which we defined the Fermat representation (F ), and the Pythagorean representation (P ), we now examine the relationship between them.
3. The Congruence Between F and P Since F = P , Tη(c, α) = [2ηα, p(c, α, η), c] = [2ηα, f(c, α, η), c], (13) which motivates the definition of a function for any t ∈ R≥1: 2 2 t t t 2 ψt(c, α, x) = [c + x − 2xα] − [c − x ] , (14) where 0 < α < c ∈ R>0. When c and α are fixed, and t = n ∈ Z>0, then ψn(c, α, x) is a univariate polynomial in x that we denote by the symbol un(c, α, x), in order to emphasize the polynomial nature of this function as distinguished from ψt(c, α, x) with any real value of t. As can be verified by a binomial series expansion of terms in (14), only triangles with integer values of the Fermat index can generate such polynomials, while triangles with non-integer values of the Fermat index will generate infinite series, so ψt(c, α, x) acts as a filter, by which we can separate triangles with integer Fermat indices from those with non-integer Fermat indices, and “enforce” the integer Fermat index as a constraint in our analysis. The congruence F = P is satisfied for integer values of the Fermat index whenever un(c, α, x) attains a zero.
3.1. Roots of the polynomial un(c, α, x) We will denote the positive integer values of t by the symbol n, and expand un(c, α, x) to get,
2 2 n n n 2 2 n n n 2 un(c, α, x) = [c + x − 2xα] − [c − x ] = [c + x(x − 2α)] − [c − x ] (15)
n−1 X n = nc2(n−1)x(x − 2α) + xn(x − 2α)n + c2(n−k)xk(x − 2α)k + 2cnxn − x2n (15a) k k=2 Thus 2(n−1) n−1 n un(c, α, x) = xsn(c, α, x) = x[nc (x − 2α) + x (x − 2α) + n−1 X n (15b) c2(n−k)xk−1(x − 2α)k + 2cnx(n−1) − x(2n−1)] k k=2 10 On Triangles and an Extended Fermat Equation Giri Prabhakar
n n Let us now examine the zeros of un(c, α, x). Firstly, setting un(c, α, x) = 0 in (15) leads to [c − x ] = p n 2 2 2 2 2 ± [c2 + x2 − 2xα]n = ±b , which is always real since ∀ x ∈ R, c +x −2xα = (x−α) +c −α > 0 ∵ c > α. Here b can be regarded as the third side of a triangle with sides c, x and b, and Fermat index n; the side b can also be seen to result from the cosine law applied to this triangle. Therefore, un(c, α, x) = 0 represents two forms of the Fermat equation: cn = xn + bn (a triangle with c as hypotenuse), and cn + bn = xn (a triangle with x as hypotenuse). This means that there are exactly two real solutions for x. Note the term −2xα in un(c, α, x): this term in the cosine law is sign dependent, and our assumption that 0 ≤ α < 1, along with (x, b, c) being acute geometrically implies that x > 0 is the only possibility in the non-trivial solution. Secondly, since un(c, α, x) = xsn(c, α, x) in (15b), the term x is either a trivial zero, or factors out of un(c, α, x). The value x = 0 corresponds to one of a = 0 or b = 0 in (1). Also note that for n = 1 in (15b), s1[c, α, x] = 2x(c − α), and hence there is just the trivial solution x = 0 (corresponding to the degenerate triangle, or 1-simplex, ∆1). Hence we will assume n ≥ 2, and there are two non-trivial real solutions to un(c, α, x) = 0, both of which are necessarily triangles with Fermat index n. Refer to Fig. 3; let x = α + h. We divide the real line x into three domains or cases: (a) h < 0, (b) 0 ≤ h ≤ min(c − α, α), and (c) min(c − α, α) < h < ∞. Consider case (a): for all values of h < 0, the triangle formed by line segments x and c is obtuse, for example Q2OL in the figure. From Lemma 4, the Fermat index of such triangles is not a whole number: 1 < φ < 2, and for n ≥ 2 ∈ Z, sn(c, α, x) 6= 0 in this case, because the Fermat index of any possible obtuse triangle can never equal n (except n = 1, which is the degenerate case). Therefore, in case (a), sn(c, α, x) cannot have a real root. In case (b) φ is a strictly increasing function of h as shown in (12), and 2 ≤ φ < ∞, with φ unbounded as h → min(c − α, α). ∗ Therefore, there is exactly one value of h (corresponding to α + h = 2η α) at which φη∗ (c, α) = n, and ∗ ∗ ∗ sn(c, α, x ) = sn(c, α, α+h ) = sn(c, α, 2αη ) = 0. This is the first non-trivial real solution for un(c, α, x) = 0 with c as hypotenuse. In this case, the corresponding Fermat equation is (x∗)n + bn = cn. Multiplying this ∗ n n ∗ n 2 ∗ n equation throughout by the factor (c/x ) , we get c +(bc/x ) = (c /x ) . Since the hypotenuse of Tη(c, α) is already fixed as c, for increasing values of x, we expect a second solution, at x = c2/x∗. Given there are ∗ only two possible real solutions, and given that the first zero of sn(c, α, x) = 0 at x = x , the second must occur at x = c2/x∗. Since x∗ < c, c2/x∗ > c which means (x∗, b, c) ∈ F, but (c2/x∗, cb/x∗, c) 6∈ F. However, as shown earlier, (c, cb/x∗, c2/x∗) ∈ τ(c, α) ∈ F.
Figure 3: Geometric interpretation of (15): x is the variable side of the triangle on the x−axis, c is fixed, and b is determined simultaneously by (13). ∗ Summarizing, un(c, α, x) has exactly one trivial zero, and two nontrivial real zeros of the form x and 2 ∗ c /x . Note from (15) that sn(c, α, x) is a polynomial of degree 2(n − 1). From the fundamental theorem of algebra [25], sn(c, α, x) must have 2(n − 2) complex roots, which comprise of n − 2 complex conjugate pairs.
3.1.1. Computed Example: Roots of u7(15, 7, x) 2 n n n 2 A sample plot of un(c, α, x) = [c + x(x − 2α)] − [c − x ] for c = 15, α = 7 and n = 7 is shown in Fig. (4a) [26]. u7(15, 7, x) may be expanded as [26]:
12 11 10 9 8 u7(15, 7, x) = xs7(15, 7, x) = −x(98x − 5691x + 228340x − 7038185x + 172149054x
−3442681627x7 + 56343906554x6 − 774603366075x5 + 8715045858750x4 − 80169326015625x3
11 On Triangles and an Extended Fermat Equation Giri Prabhakar
+585210445312500x2 − 3281717373046875x + 12715141113281250) As expected from (15), there are a total of 2(7 − 1) = 12 non-trivial roots corresponding to the polynomial ∗ s7(15, 7, x) of degree 12, of which 2 are real: one occurs before 2α = 14, at x ≈ 12.4431, and the other after 14, at x = c2/x∗ ≈ 225/(12.4431)2 ≈ 18.0824 as shown in Fig.(4).
(a) Plot of u7(15, 7, x). (b) Roots of u7(15, 7, x) in the complex plane.
2 n n n 2 Figure 4: Polynomial un(c, α, x) = [c + x(x − 2α)] − [c − x ] evaluated for c = 15, α = 7 and n = 7 [26].
Remark 1. For n = 2, a quadratic equation of the form (α + h)2 − α(α + h) = 0 results, which has two real roots h = −α corresponding to a (degenerate) triangle with zero area, and h = 0, corresponding to a right triangle. As expected, a second solution occurs when x = α + h = c2/α, or, c2 = α(α + h), which implies that α + h forms a right triangle with sides c and b, which can be seen by applying the cosine law, b2 = c2 + c4/α2 − 2c(c2/α)(α/c), which results in c2 + b2 = (c2/α)2. Geometrically, the solution x = α + h = c2/α simply corresponds to scaling the original right triangle with sides (α, b, c), by the factor c/α, to get a right triangle (c, bc/α, c2/α).
4. Main Results
We will state and prove two propositions about the irreducibility of polynomials sn(c, α, x), and the rationality of their roots. Before we state the propositions, we define some terminology and symbols, which we will use in addition to the terminology described in Sect. 2.1.
4.1. Preliminaries
Let us denote S = {sn(c, α, x)[c, α ∈ R>0 ∧ n ∈ Z>0]}. Then S ⊂ R[x], the ring of polynomials over R. When c, α ∈ Q>0, this subset of S contains only polynomials with rational coefficients, which will be denoted by SQ, and SQ ⊂ Q[x]. Also, SZ ⊂ SQ contains only polynomials with integer coefficients. Further, let M be the set of all odd primes, with elements symbolized by m, and SM be the subset of polynomials sm(c, α, x) ∈ SZ with Fermat index an element of M. Note that SM ⊂ SZ ⊂ Z[x]. Let the algebraic set σn = {(x | sn(c, α, x) = 0)[sn(c, α, x) ∈ S]} ∈ σ, where σ is the set of all such algebraic sets, and ρ ⊂ σ be the set of algebraic sets ρn = {(x | sn(c, α, x) = 0)[sn(c, α, x) ∈ SQ]}. Further, let ρZ>k ⊂ ρ be the set of algebraic sets of polynomials sn(c, α, x) ∈ SZ with integer coefficients, and integer Fermat indices greater than k, while ρM ⊂ ρZ>2 is the set of algebraic sets of polynomials sm(c, α, x) ∈ SM with integer coefficients and Fermat indices being odd prime numbers m, which are the elements of M.
Proposition 1. Let Tη(c, α) = (a, b, c) ∈ F be a primitive integer triangle with c the longest side, α the projection of c on a, and m an odd prime number. Let α = p/q, where p and q are positive integers, and C = qc. Then, if C, p and C − p 6≡ 0 (mod m), sm(c, α, x) is either irreducible in Q[x], or reducible to a 12 On Triangles and an Extended Fermat Equation Giri Prabhakar product of at most two polynomials with rational coefficients and of equal degree m − 1, both of which are irreducible in Q[x].
Proposition 2. The algebraic sets of the polynomial sn(c, α, x) have no rational numbers for n > 2, hence {ρ − ρ2} ∩ Q = ∅ The main results of this paper are the proofs of propositions 1 and 2. We present below a few theorems and lemmas that will be used in our proofs.
Lemma 8. (Gauss [17]) A nonconstant polynomial is irreducible in Z[x] only if it is irreducible in Q[x] and primitive in Z[x]. The rational root theorem is a consequence of Gauss’s Lemma. Given an nth degree polynomial Q(x) with integer coefficients, if Q(x) is reducible in Q[x] and has a factor (x − q/p), then from Gauss’s Lemma Q(x) is also reducible in Z[x], to (px − q)g(x), where p multiplies into the coefficient of xn−1 in g(x), and q multiplies into the constant term [27]. Therefore we have Theorem 9. (Rational Root [27, 28]) Let a polynomial with integer coefficients be defined as following,
n n−1 Q(x) = anx + an−1x + ··· + a1x + a0.
If Q(x) has a rational root of the form p/q, p is a factor of a0, while q is a factor of an. Any integer root of Q(x) must be a factor of a0 only. These conditions hold irrespective of the sign of the root. Theorem 10. (Schoenemann-Eisenstein, [29–31]) Suppose we have the following polynomial with integer coefficients. n n−1 Q(x) = anx + an−1x + ··· + a1x + a0 If there exists a prime number m such that the following three conditions all apply:
(i) m divides each ai for i 6= n,
(ii) m does not divide an, and
2 (iii) m does not divide a0, then Q(x) is irreducible over the rational numbers. Theorem 11. (Fermat [32]) If p is a prime and a is any integer not divisible by p, ap−1 ≡ 1 (mod p). Remark 2. We will invoke two well-known results that have been long established in the course of the mathematical developments towards the proof of Fermat’s last theorem: firstly, that if the theorem is true for all odd primes (m as per our notation) and 4, then it is true for all positive integers [3]. We also note that the case n = 4 was proved long before the case n = m was proved by Wiles [33]. Secondly, two cases have been historically been formulated, and only these cases hold [34]: Case I, in which none of a, b and c are divisible by m, and Case II, in which exactly one of a, b and c are divisible by m.
4.2. Proof of Proposition 1 We will begin by proposing and proving an extension of the Schoenemann-Eisenstein theorem. Theorem 12. Suppose we have the following polynomial with integer coefficients.
n n−1 Q(x) = anx + an−1x + ··· + a1x + a0
If there exists a prime number m, and an integer 0 ≤ k ≤ n such that the following three conditions all apply:
(i) m divides each ai for i 6= k 13 On Triangles and an Extended Fermat Equation Giri Prabhakar
(ii) m does not divide ak, and
2 (iii) m does not divide a0 and an, then Q(x) is either irreducible over the rational numbers, or can be reduced to a product of at most two polynomials with rational coeffcients, one of degree k and the other of degree n − k, both of which are irreducible over the rational numbers. Proof. We adopt a method based upon the approach in [31]. We assume Q(x) is primitive but reducible in R[x] so that as a consequence of Gauss’s Lemma (8), without loss of generality, we let Q(x) = g(x)h(x), where g(x) and h(x) are two polynomials with integer coefficients:
u u−1 u−2 2 g(x) = gux + gu−1x + gu−2x + ··· + g2x + g1x + g0, v v−1 v−2 2 (16) h(x) = hvx + hv−1x + hv−2x + ··· + h2x + h1x + h0, so that u + v = n. We begin by assuming v ≥ u, and first consider which coefficients of h(x) and g(x) must necessarily be nonzero. We claim that these are the coefficients h0, g0, hv, gu, and at least one of hk, gk. Clearly hv, h0 and gv, g0 have to be nonzero since an = hvgu and a0 = h0g0. If both the following hold: hk = 0, gk = 0, then for ak 6= 0 to hold, there must be at least two integers i0 6= k and j0 6= k such that i0 + j0 = k and hi0 gj0 6= 0. But we know hi0 gj0 6≡ 0 (mod m), since ak 6≡ 0 (mod m). This 2 means that both of hi0 , gj0 6≡ 0 (mod m). Since we must have an = hvgu 6≡ 0 (mod m ), without loss of generality, let us assume that gu 6≡ 0 (mod m). This means that the product hi0 gu 6≡ 0 (mod m).
However, we know that ai0+u ≡ 0 (mod m). Therefore, there must be at least one index i1 such that hi1 gj1 + hi0 gu ≡ 0 (mod m) =⇒ hi1 gj1 6≡ 0 (mod m). Note that i0 + u = i1 + j1 > k. Now if we assumed that i1 < i0, then j1 > u, which is not possible. Therefore, i1 > i0. We repeat the same argument for index i1, and in this manner, generate a sequence of indices i0 < i1 < i2 < . . . . Clearly, this cannot happen indefinitely since the largest possible index of h(x) is v. This means that there will be at least one index in h(x), for which no higher index in can be found in h(x), along with a corresponding index jn in g(x), that renders ain+jn ≡ 0 (mod m). This implies that there must exist some ain+jn 6≡ 0 (mod m), where in + jn > k. This contradicts our assumption. Hence, it is not possible that both hk = 0 and gk = 0. 2 Since a0, an 6≡ 0 (mod m ), m must be a factor of only one of hv or gu and h0 or g0 respectively. Let us first assume that m is a factor of hv, and not gu. Now an−1 = hv−1gu + hvgu−1 ≡ 0 (mod m) =⇒ hv−1 ≡ 0 (mod m). Extending this argument to an−2, an−3, . . . , au we have hn−2, hn−3, .., h0 ≡ 0 (mod m). Pk Noting that u > k, we see that ak = i=0 higk−i ≡ 0 (mod m), which cannot be possible since we assumed ak 6≡ 0 (mod m). Moreover, the choice of hv ≡ 0 (mod m) and gu 6≡ 0 (mod m) means that ∀ n ≥ j ≥ u, aj ≡ 0 (mod m) necessarily implies hv, hv−1, . . . , hv+j−n ≡ 0 (mod m). Since ak+u ≡ 0 (mod m), we must also have hk ≡ 0 (mod m). Then the only possibility to ensure ak 6≡ 0 (mod m) is that gkh0 6≡ 0 (mod m). This means that we must have g0 ≡ 0 (mod m) whereas h0 6≡ 0 (mod m). Then, a1 = h1g0 + h0g1 =⇒ g1 ≡ 0 (mod m). This argument is extended to a2, a3, . . . , ak−1, and necessarily implies that g0, g1, . . . , gk−1 ≡ 0 (mod m). Then gk 6≡ 0 (mod m) ensures that ak 6≡ 0 (mod m), and thereafter, we ensure that ak+1, ak+2, · · · ≡ 0 (mod m) by setting gk+1, gk+2, · · · ≡ 0 (mod m) and noting that h1, h2, . . . , hv ≡ 0 (mod m). However, this leaves guh0 6≡ 0(mod m) =⇒ au 6≡ 0(mod m), which could be resolved if h0 ≡ 0 (mod m), since by definition gu 6≡ 0 (mod m), but this would render ak ≡ 0 (mod m) because h0gk ≡ 0 (mod m), and thus cannot be allowed. Assuming u < k eliminates the term gk, and therefore cannot meet the criterion ak 6≡ 0 (mod m). Hence the only possibility is u = k, since au = ak 6≡ 0(mod m) satisfying both conditions simultaneously. The same arguments are applicable upon initially choosing gu ≡ 0 (mod m) and hv 6≡ 0 (mod m). Note that these arguments hold even if only hv, gu, h0, g0, and at least one of hk, gk are nonzero, thus addressing every possible occurrence of zeros as coefficients under the given conditions. We have shown that the only possibility for Q(x) = h(x)g(x) is when the degrees of h(x) and g(x) are n − k and k respectively, given v ≥ u, which is an assumption without loss of generality. Further, assuming without loss of generality that g(x) is the kth degree polynomial, we note in this case, that g(x) is already in a form that satisfies the criteria in Theorem 10, since gu = gk 6≡ 0 (mod m), all other coefficients are 14 On Triangles and an Extended Fermat Equation Giri Prabhakar
2 divisible by m, and g0 6≡ 0 (mod m ). Upon setting x = 1/y and multiplying throughout h(x) by the term yv, h(x) can also be transformed into a polynomial h(x) = H(y)/yv, where H(y) is irreducible as it satisfies the criteria in Theorem 10, and hence h(x) is irreducible. The same can also be seen with g(x) as a special case of the proof of Theorem 12 given above for k = 0, and h(x) as a special case of the above proof with k = n, since in each case the second polynomial factor will need to be an n − k = 0th degree polynomial. Therefore, h(x) and g(x) cannot be further reduced.
Every polynomial in SM arises from a triangle with a rational value of cos θ, so that cos θ = α/c = p/qc = p/C where p and q are coprime. Moreover, we assume C, p 6≡ 0 (mod m), therefore m is coprime to both p and C.
2(m−1) m−1 m sm(C, p, x) = mC (x − 2p) + x (x − 2p) m−1 X m + C2(m−k)xk−1(x − 2p)k + 2Cmxm−1 − x2m−1, k k=2 (17) which upon some simplification becomes
2(m−1) 2 2m−3 m m m−1 m m−1 −sm(C, p, x) = 2mpx − 2m(m − 1)p x + ··· + 2 p x − 2C x m−1 X m − C2(m−k)xk−1(x − 2p)k − mC2(m−1)x + 2mC2(m−1)p. k k=2 (18)