Introduction
A set of positive integers is a Diophantine tuple if the product of any two of its elements, increased by 1, is a perfect square. Every Diophantine triple { a , b , c } \{a,b,c\} { a , b , c } with a < b < c a < b < c a < b < c , where a b + 1 = r 2 ab+1=r^2 ab + 1 = r 2 , a c + 1 = s 2 ac+1=s^2 a c + 1 = s 2 and b c + 1 = t 2 bc+1=t^2 b c + 1 = t 2 , extends to the Diophantine quadruple { a , b , c , d + } \{a,b,c,d_+\} { a , b , c , d + } , where
d + = d + ( a , b , c ) = a + b + c + 2 a b c + 2 r s t , d_+ = d_+(a,b,c) = a+b+c+2abc+2rst, d + = d + ( a , b , c ) = a + b + c + 2 ab c + 2 r s t ,
as observed by Arkin, Hoggatt and Straus [3 ] and, independently, by Gibbs (see [15 ] ). An integer d > c d>c d > c such that { a , b , c , d } \{a,b,c,d\} { a , b , c , d } is a Diophantine quadruple is an extension of the triple; it is regular if d = d + ( a , b , c ) d=d_+(a,b,c) d = d + ( a , b , c ) and irregular otherwise. The regularity conjecture, which goes back to the same authors, asserts that every Diophantine quadruple is regular, that is, that every extension of a Diophantine triple is regular (see [18 ] ). It implies that there is no Diophantine quintuple: Dujella proved that there are only finitely many [14 ] , and He, Togbé and Ziegler that there are none [19 ] . For a fixed triple, Fujita and Miyazaki [18 ] showed that there are at most eleven extensions, and Cipu, Fujita and Miyazaki [11 ] that there are at most eight. In a recent preprint He and Togbé [20 ] give a proof that there are only finitely many irregular Diophantine quadruples. The conjecture itself is open.
This paper is about the extensions of Euler triples , the Diophantine triples { a , b , c } \{a,b,c\} { a , b , c } with a < b < c a<b<c a < b < c and
c = a + b + 2 r , r = a b + 1 c=a+b+2r,\qquad r=\sqrt{ab+1} c = a + b + 2 r , r = ab + 1
(Definition 2.3). The name is that of He, Togbé and Ziegler [19 ] ; Cipu, Fujita and Mignotte [7 ] and Dujella [15 ] call these triples regular, a word that in this paper qualifies only extensions. By a lemma of Jones [21 ] (see [19 ] ), a Diophantine triple with a < b < c a<b<c a < b < c either is an Euler triple or satisfies c > 4 a b c>4ab c > 4 ab . So each Diophantine pair { a , b } \{a,b\} { a , b } with a < b a<b a < b lies in exactly one Euler triple with third element larger than b b b , and among the Diophantine triples that contain the pair as their two smallest elements it has the least third element. For an Euler triple, s = a + r s=a+r s = a + r , t = b + r t=b+r t = b + r and d + ( a , b , c ) = 4 r s t d_+(a,b,c)=4rst d + ( a , b , c ) = 4 r s t (Lemma 2.4), and we write
α = r + a b , β = s + a c , γ = t + b c \alpha=r+\sqrt{ab},\qquad\beta=s+\sqrt{ac},\qquad\gamma=t+\sqrt{bc} α = r + ab , β = s + a c , γ = t + b c
for its units. Cipu, Fujita and Miyazaki [11 ] showed that an Euler triple has at most three extensions d > d + ( a , b , c ) d>d_+(a,b,c) d > d + ( a , b , c ) .
Index systems and the main results
Let d d d be an irregular extension of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , that is, an integer d > c d>c d > c with d ≠ d + ( a , b , c ) d\ne d_+(a,b,c) d = d + ( a , b , c ) such that { a , b , c , d } \{a,b,c,d\} { a , b , c , d } is a Diophantine quadruple, and write a d + 1 = x 2 ad+1=x^2 a d + 1 = x 2 , b d + 1 = y 2 bd+1=y^2 b d + 1 = y 2 and c d + 1 = z 2 cd+1=z^2 c d + 1 = z 2 with x , y , z > 0 x,y,z>0 x , y , z > 0 . An index system of d d d (Definition 2.5) is a quadruple ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) , with λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } and integers l , m , n ≥ 0 l,m,n\geq0 l , m , n ≥ 0 , such that
y a + x b = ( a + b ) α 2 l , z a + x c = ( λ a + c ) β 2 m , y\sqrt{a}+x\sqrt{b}=(\sqrt{a}+\sqrt{b})\alpha^{2l},\qquad z\sqrt{a}+x\sqrt{c}=(\lambda\sqrt{a}+\sqrt{c})\beta^{2m}, y a + x b = ( a + b ) α 2 l , z a + x c = ( λ a + c ) β 2 m ,
z b + y c = ( λ b + c ) γ 2 n . z\sqrt{b}+y\sqrt{c}=(\lambda\sqrt{b}+\sqrt{c})\gamma^{2n}. z b + y c = ( λ b + c ) γ 2 n .
By Proposition 2.11, every extension has exactly one index system, and by Lemma 2.14 the index system of an irregular extension satisfies 3 ≤ n < m < l < 2 n 3\leq n<m<l<2n 3 ≤ n < m < l < 2 n ; we call m m m the index of the extension. It locates z z z as a common term of the Pell sequences of the pairs { a , c } \{a,c\} { a , c } and { b , c } \{b,c\} { b , c } , at the positions m m m and n n n (Lemma 2.8). Each of our results compares a lower bound for the index, coming from the arithmetic of the three representations, with an upper bound coming from Diophantine approximation. The first result is a universal lower bound.
Index floor (Theorem 3.6 ). Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then c < 4 m 2 ( m + λ ) 2 < 8 m 4 c < 4m^2(m+\lambda)^2 < 8m^4 c < 4 m 2 ( m + λ ) 2 < 8 m 4 .
Equivalently, m > ( c / 8 ) 1 / 4 m > (c/8)^{1/4} m > ( c /8 ) 1/4 : the index grows at least like the fourth root of c c c . The previous lower bound for the index of an Euler triple, 32 m 5 > c log c 32m^5 > c\log c 32 m 5 > c log c for c ≥ 10 7 c \ge10^7 c ≥ 1 0 7 [20 ] , gives m > ( c log c / 32 ) 1 / 5 m > (c\log c/32)^{1/5} m > ( c log c /32 ) 1/5 , a fifth root; Theorem 3.6 raises the exponent of c c c from 1 / 5 1/5 1/5 to 1 / 4 1/4 1/4 and needs no condition on c c c . We do not know whether the exponent 1 / 4 1/4 1/4 is optimal. The index floor is the heart of the paper: it is the lower bound behind the pair-coordinate criterion, the effective bound and the three-parameter family below.
Two of the units α , β , γ \alpha,\beta,\gamma α , β , γ are close when b / a b/a b / a is large, and when b − a b-a b − a is small compared with a c \sqrt{ac} a c . In both cases the three representations force regularity.
Close Pell units (Theorem 4.5 ). Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with b ≥ 3355 2 a b \ge3355^2a b ≥ 335 5 2 a or a c ≥ 3075 ( b − a ) \sqrt{ac} \ge3075(b-a) a c ≥ 3075 ( b − a ) . Then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Consequently an Euler triple with an irregular extension satisfies 1 + 1 1538 < b / a < 3355 2 1+\frac{1}{1538}<b/a<3355^2 1 + 1538 1 < b / a < 335 5 2 (Corollary 4.6 ). The second condition holds whenever b − a ≤ a / 1538 b-a \le a/1538 b − a ≤ a /1538 ; for Euler triples with a > 3.8 ⋅ 10 7 a>3.8\cdot10^7 a > 3.8 ⋅ 1 0 7 this range contains the range b < a + 4 a b<a+4\sqrt{a} b < a + 4 a of the close-pair theorem of Filipin, Fujita and Togbé [17 ] . The constants come from one numerical inequality in Laurent’s bound for linear forms in two logarithms [22 ] , which they satisfy with a margin below one half of one percent. In their concluding remarks Cipu, Fujita and Mignotte [7 ] explain that their two-parameter families became tractable through sharp bounds for the difference of the integer coefficients of the logarithms, and write that “it remains for future works to obtain similar bounds for general triples, not necessarily given parametrically”. For Euler triples such bounds are the gap inequalities
m > ( l − m ) R x , n > ( m − n ) R z , where R x = b a log α , R z = a c b − a log β m>(l-m)R_x,\qquad n>(m-n)R_z,\qquad\text{where}\quad R_x=\sqrt{\frac{b}{a}}\log\alpha,\qquad R_z=\frac{\sqrt{ac}}{b-a}\log\beta m > ( l − m ) R x , n > ( m − n ) R z , where R x = a b log α , R z = b − a a c log β
(Lemma 2.18 ), and Theorem 4.5 follows from them and Laurent’s theorem.
The third criterion uses coordinates adapted to the hypergeometric method. A pair representation of sign σ ∈ { − 1 , 1 } \sigma\in\{-1,1\} σ ∈ { − 1 , 1 } of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } is a quadruple ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) of positive integers with a = p 1 q 1 a=p_1q_1 a = p 1 q 1 , b = p 2 q 2 b=p_2q_2 b = p 2 q 2 and p 1 q 2 − p 2 q 1 = 2 σ p_1q_2-p_2q_1=2\sigma p 1 q 2 − p 2 q 1 = 2 σ (Definition 5.8). Every Euler triple has one, and its pair coordinates ( p 1 , q 1 ) (p_1,q_1) ( p 1 , q 1 ) , ( p 2 , q 2 ) (p_2,q_2) ( p 2 , q 2 ) and ( p 3 , q 3 ) = ( p 1 + p 2 , q 1 + q 2 ) (p_3,q_3)=(p_1+p_2,q_1+q_2) ( p 3 , q 3 ) = ( p 1 + p 2 , q 1 + q 2 ) satisfy c = p 3 q 3 c=p_3q_3 c = p 3 q 3 (Lemma 5.10 ).
Pair coordinates (Theorem 5.19 ). Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with a pair representation such that p 1 q 3 ≥ 800 p 3 2 p_1q_3\ge800p_3^2 p 1 q 3 ≥ 800 p 3 2 , or such that p 1 q 3 ≥ 42 p 3 2 p_1q_3\ge42p_3^2 p 1 q 3 ≥ 42 p 3 2 and c ≥ 4.5 ⋅ 10 12 c\ge4.5\cdot10^{12} c ≥ 4.5 ⋅ 1 0 12 . Then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Cipu, Fujita and Mignotte [10 ] proved the regularity of a three-parameter family of Euler triples by the simultaneous approximations of Rickert [26 ] and the criterion of Bennett [6 ] , under a growth condition on the parameters. Their approximation measure extends to every pair representation with p 1 q 3 ≥ 168 p_1q_3\ge168 p 1 q 3 ≥ 168 (Theorem 5.14 ). Together with the index floor, which takes the place of their congruence lower bound for the index, it gives Theorem 5.19 ; on their family it removes the growth condition.
The fourth criterion is arithmetic. For an Euler triple put
Δ a b = gcd ( a , b ) gcd ( a , b , 4 ) , Δ b c = gcd ( b , c ) gcd ( b , c , 4 ) , Δ a c = gcd ( a , c ) gcd ( a , c , 4 ) , \Delta_{ab}=\frac{\gcd(a,b)}{\gcd(a,b,4)},\qquad\Delta_{bc}=\frac{\gcd(b,c)}{\gcd(b,c,4)},\qquad\Delta_{ac}=\frac{\gcd(a,c)}{\gcd(a,c,4)}, Δ ab = g cd( a , b , 4 ) g cd( a , b ) , Δ b c = g cd( b , c , 4 ) g cd( b , c ) , Δ a c = g cd( a , c , 4 ) g cd( a , c ) ,
G r c = gcd ( r , c ) gcd ( r , c , 2 ) , G t a = gcd ( t , a ) gcd ( t , a , 2 ) , G s b = gcd ( s , b ) gcd ( s , b , 2 ) , G_{rc}=\frac{\gcd(r,c)}{\gcd(r,c,2)},\qquad G_{ta}=\frac{\gcd(t,a)}{\gcd(t,a,2)},\qquad G_{sb}=\frac{\gcd(s,b)}{\gcd(s,b,2)}, G r c = g cd( r , c , 2 ) g cd( r , c ) , G t a = g cd( t , a , 2 ) g cd( t , a ) , G s b = g cd( s , b , 2 ) g cd( s , b ) ,
and call Π 1 = Δ a b G r c \Pi_1 = \Delta_{ab}G_{rc} Π 1 = Δ ab G r c , Π 2 = Δ b c G t a \Pi_2 = \Delta_{bc}G_{ta} Π 2 = Δ b c G t a and Π 3 = Δ a c G s b \Pi_3 = \Delta_{ac}G_{sb} Π 3 = Δ a c G s b its coprime divisors (Definition 6.1 ); the two factors of each product are coprime (Lemma 6.2 ). The index combinations of an index system are m − n m-n m − n , l + λ m l+\lambda m l + λm and l + λ n l+\lambda n l + λn .
Coprime divisors (Theorem 6.3 ). Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the coprime divisors Π 1 , Π 2 , Π 3 \Pi_1,\Pi_2,\Pi_3 Π 1 , Π 2 , Π 3 and the gap ratio R z R_z R z above. Then
Π 1 ∣ m − n , Π 2 ∣ l + λ m , Π 3 ∣ l + λ n , and m > Π 1 ( 1 + R z ) . \Pi_1 \mid m-n,\qquad\Pi_2 \mid l+\lambda m,\qquad\Pi_3 \mid l+\lambda n,\qquad\text{and}\qquad m>\Pi_1(1+R_z). Π 1 ∣ m − n , Π 2 ∣ l + λm , Π 3 ∣ l + λn , and m > Π 1 ( 1 + R z ) .
The application is to Euler triples with two elements in a fixed ratio, for which the coprime divisors grow linearly with c c c . An Euler triple has ratio type j : k j:k j : k , for coprime integers 1 ≤ j < k 1\leq j<k 1 ≤ j < k , if two of its elements v < w v<w v < w satisfy j w = k v jw=kv j w = k v (Definition 8.5 ); for each j : k j:k j : k and each choice of the two positions, these triples come from the positive solutions of the Pell equation u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 (Lemma 8.6 ).
Fixed ratios (Theorem 8.7 ). Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple of ratio type j : k j:k j : k . If { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension, then
c 6 ( j + k ) − 0.15 < 2.4638 ⋅ 10 12 ( log c ) 2 log 19.46 c 6 ( j + k ) . \frac{c}{6(j+k)}-0.15<2.4638\cdot10^{12}(\log c)^2\log\frac{19.46c}{6(j+k)}. 6 ( j + k ) c − 0.15 < 2.4638 ⋅ 1 0 12 ( log c ) 2 log 6 ( j + k ) 19.46 c .
If j k ≤ 300 jk\leq300 j k ≤ 300 , then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Earlier results of this kind concern single ratios; they are listed in the table at the head of Section 8.
Finally, all irregular extensions of Euler triples are confined to an explicit range.
Effective bound (Theorem 7.9 ). Let { a , b , c , d } \{a,b,c,d\} { a , b , c , d } be a Diophantine quadruple with a < b < c < d a<b<c<d a < b < c < d and c = a + b + 2 a b + 1 c=a+b+2\sqrt{ab}+1 c = a + b + 2 ab + 1 . If d ≠ d + ( a , b , c ) d\ne d_+(a,b,c) d = d + ( a , b , c ) , then c < 4 ⋅ 10 48 c<4\cdot10^{48} c < 4 ⋅ 1 0 48 .
This improves the bound c < 10 98 c<10^{98} c < 1 0 98 of [20 ] . Near the new bound the index floor forces the index m m m to be of order 10 12 10^{12} 1 0 12 .
As applications we settle two subfamilies of the three-parameter family of Cipu, Fujita and Mignotte [10 ] . With F − 1 = 1 F_{-1}=1 F − 1 = 1 , F 0 = 0 F_0=0 F 0 = 0 and F k + 2 = F k + 1 + F k F_{k+2}=F_{k+1}+F_k F k + 2 = F k + 1 + F k , integers ν , A , K ≥ 1 \nu,A,K\geq1 ν , A , K ≥ 1 and a sign ε ∈ { − 1 , 1 } \varepsilon\in\{-1,1\} ε ∈ { − 1 , 1 } , put ϕ k = F k A + F k − 1 \phi_k=F_kA+F_{k-1} ϕ k = F k A + F k − 1 and f k = ϕ k ( ϕ k K + 2 ε F k ) f_k=\phi_k(\phi_kK+2\varepsilon F_k) f k = ϕ k ( ϕ k K + 2 ε F k ) ; the family triple of the parameter point ( ν , A , K , ε ) (\nu,A,K,\varepsilon) ( ν , A , K , ε ) is { f ν , f ν + 1 , f ν + 2 } \{f_\nu,f_{\nu+1},f_{\nu+2}\} { f ν , f ν + 1 , f ν + 2 } (Convention 8.1), an Euler triple for K ≥ 3 K\geq3 K ≥ 3 (Lemma 8.2 ). We prove that the family triple has no irregular extension when ν A ≥ 2 \nu A\geq2 ν A ≥ 2 and either K ≥ 643 K\geq643 K ≥ 643 , or K ≥ 128 K\geq128 K ≥ 128 and c ≥ 4.5 ⋅ 10 12 c\geq4.5\cdot10^{12} c ≥ 4.5 ⋅ 1 0 12 (Corollary 8.3), and when K = 1 K=1 K = 1 , A ∈ { 2 , 3 } A\in\{2,3\} A ∈ { 2 , 3 } and ν ≥ 2 \nu\geq2 ν ≥ 2 (Theorem 8.4 ). For ( A , ε ) = ( 3 , − 1 ) (A,\varepsilon)=(3,-1) ( A , ε ) = ( 3 , − 1 ) the family triples are the Hoggatt–Bergum triples { F 2 ν + 2 , F 2 ν + 4 , F 2 ν + 6 } \{F_{2\nu+2},F_{2\nu+4},F_{2\nu+6}\} { F 2 ν + 2 , F 2 ν + 4 , F 2 ν + 6 } , and for extensions d > c d>c d > c Theorem 8.4 recovers Dujella’s theorem on them [12 ] ; these triples are also covered by the results of Filipin, Fujita and Togbé [17 ] on the pairs { F 2 k , F 2 k + 2 } \{F_{2k},F_{2k+2}\} { F 2 k , F 2 k + 2 } and of Park and Lee [25 ] on the pairs { F 2 k , F 2 k + 4 } \{F_{2k},F_{2k+4}\} { F 2 k , F 2 k + 4 } . The table at the head of Section 8 compares these applications, and the other families settled here, with what was known.
Methods and plan of the paper
The lower bounds are arithmetic. Expanding the three Pell sequences at their common terms to second order, by the congruences of Dujella [13 ] , and to third order, following He and Togbé [20 ] , gives congruences for l , m , n l,m,n l , m , n modulo a a a , b b b and c c c , and integer combinations of them in which the leading terms cancel give the index floor. Reducing the multiplication matrices modulo the gcds of two elements, or of a root and an element, gives the divisibilities behind the coprime divisors, and the divisibility of a quadratic index form in pair coordinates gives a further lower bound in the pair-coordinate criterion; for close units the gap inequalities play the part of the lower bound. The upper bounds are analytic. Those from linear forms in logarithms start from the two linear forms
Λ x = 2 l log α − 2 m log β + log D λ , Λ z = 2 m log β − 2 n log γ + log E λ \Lambda_x = 2l \log\alpha- 2m \log\beta+ \log D_\lambda,\qquad\Lambda_z = 2m \log\beta- 2n \log\gamma+ \log E_\lambda Λ x = 2 l log α − 2 m log β + log D λ , Λ z = 2 m log β − 2 n log γ + log E λ
of Definition 2.16 , with explicit algebraic numbers D λ D_\lambda D λ , E λ E_\lambda E λ , which satisfy 0 < Λ x < 4.01 β − 4 m 0 < \Lambda_x < 4.01\beta^{-4m} 0 < Λ x < 4.01 β − 4 m and 0 < Λ z < 2 β − 4 n 0 < \Lambda_z < 2\beta^{-4n} 0 < Λ z < 2 β − 4 n (Lemma 2.18 ). They are bounded from below by Laurent’s theorem on linear forms in two logarithms [22 ] for close units, by Matveev’s theorem [23 ] for the a priori bound (Proposition 7.3 ), and by the theorem of Mignotte and Voutier [24 ] on linear forms in three logarithms for the effective bound. For pair coordinates the upper bound comes from the approximants of Rickert [26 ] , with Bennett’s criterion [6 ] and the prime product of Cipu and Fujita [9 ] . Individual triples below the analytic thresholds are closed by the reduction of Baker and Davenport [4 ] in the form of Dujella and Pethő [16 ] (Proposition B.1). Several steps of the proofs are finite computations: numerical inequalities certified in interval arithmetic with outward rounding, exhaustive checks over explicit finite sets, and these reductions. Each is stated where it is used, with what it checks and how.
Section 2 contains the common setting: Euler triples, their index systems and the two linear forms. Its results are used throughout, above all the existence and uniqueness of the index system (Proposition 2.11) and the ordering of the indices (Lemma 2.14), on which every later section rests. The index floor, the criteria for close units, pair coordinates and coprime divisors, and the effective bound are proved in Sections 3 to 7, in this order. After Section 2, Sections 3, 4 and 6 can be read independently of one another; Section 5 uses the index floor, and Section 7 uses the index floor and the heights and Laurent’s theorem of Section 4. Section 8 proves the results on fixed ratios and on the three-parameter family, with a table comparing them with the literature. Appendix A contains the proof for the three-parameter family with 643 ≤ K ≤ 2401 643 \le K \le2401 643 ≤ K ≤ 2401 , and Appendix B the reduction for a single triple.
Euler triples, index systems and the two linear forms
This section sets up the objects of the paper and proves the facts on which the later sections rest: every extension of an Euler triple has exactly one index system (Proposition 2.11), and for an irregular extension the indices are ordered (Lemma 2.14) and the two linear forms Λ x \Lambda_x Λ x and Λ z \Lambda_z Λ z are exponentially small in the index (Lemma 2.18 ). Most of it is routine or quoted. The one step that carries an idea is the comparison of two expressions of one number (Lemma 2.17 ), which gives both linear forms their smallness.
Euler triples and their extensions
Definition 2.1. A Diophantine tuple is a finite set of distinct positive integers in which the product of any two distinct elements, increased by 1, is a perfect square. A Diophantine tuple with three elements is a Diophantine triple , and one with four elements is a Diophantine quadruple .
Definition 2.2. Let { a , b , c } \{a,b,c\} { a , b , c } be a Diophantine triple (Definition 2.1 ) with a < b < c a < b < c a < b < c , and put
d ± ( a , b , c ) = a + b + ( 2 a b + 1 ) c ± 2 ( a b + 1 ) ( a c + 1 ) ( b c + 1 ) . d_{\pm}(a,b,c) = a+b+(2ab+1)c \pm2\sqrt{(ab+1)(ac+1)(bc+1)}. d ± ( a , b , c ) = a + b + ( 2 ab + 1 ) c ± 2 ( ab + 1 ) ( a c + 1 ) ( b c + 1 ) . An extension of { a , b , c } \{a,b,c\} { a , b , c } is an integer d > c d>c d > c such that { a , b , c , d } \{a,b,c,d\} { a , b , c , d } is a Diophantine quadruple. The extension d d d is regular if d = d + ( a , b , c ) d=d_+(a,b,c) d = d + ( a , b , c ) and irregular otherwise. A Diophantine quadruple
{ a , b , c , d } \{a,b,c,d\} { a , b , c , d } with a < b < c < d a < b < c < d a < b < c < d is regular or irregular according as d d d is a regular or an irregular extension of { a , b , c } \{a,b,c\} { a , b , c } .
Definition 2.3. An Euler triple is a set { a , b , c } \{a,b,c\} { a , b , c } of positive integers a < b < c a < b < c a < b < c such that
for a positive integer r r r and
c = a + b + 2 r . c=a+b+2r. c = a + b + 2 r . Its roots are r r r , s = a + r s=a+r s = a + r and t = b + r t=b+r t = b + r . Its units are
α = r + a b , β = s + a c , γ = t + b c , \alpha=r+\sqrt{ab},\qquad\beta=s+\sqrt{ac},\qquad\gamma=t+\sqrt{bc}, α = r + ab , β = s + a c , γ = t + b c , and its multiplication matrices are
A = ( r b a r ) , B = ( s c a s ) , C = ( t c b t ) . A=\begin{pmatrix}r&b\\a&r\end{pmatrix},\qquad B=\begin{pmatrix}s&c\\a&s\end{pmatrix},\qquad C=\begin{pmatrix}t&c\\b&t\end{pmatrix}. A = ( r a b r ) , B = ( s a c s ) , C = ( t b c t ) .
Lemma 2.4. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with roots r , s , t r,s,t r , s , t , units α , β , γ \alpha,\beta,\gamma α , β , γ and multiplication matrices A , B , C \mathbf{A},\mathbf{B},\mathbf{C} A , B , C (Definition 2.3 ). Then:
a c + 1 = s 2 ac+1=s^2 a c + 1 = s 2 and b c + 1 = t 2 bc+1=t^2 b c + 1 = t 2 ; in particular { a , b , c } \{a,b,c\} { a , b , c } is a Diophantine triple;
c r − s t = r s − a t = r t − b s = 1 cr-st=rs-at=rt-bs=1 cr − s t = r s − a t = r t − b s = 1 , and d + ( a , b , c ) = 4 r s t = 4 r ( c r − 1 ) d_{+}(a,b,c)=4rst=4r(cr-1) d + ( a , b , c ) = 4 r s t = 4 r ( cr − 1 ) ;
r , s , t r,s,t r , s , t are pairwise coprime, and gcd ( a , r ) = gcd ( a , s ) = gcd ( b , t ) = 1 \gcd(a,r)=\gcd(a,s)=\gcd(b,t)=1 g cd( a , r ) = g cd( a , s ) = g cd( b , t ) = 1 ;
b − a ≥ 2 b-a\ge2 b − a ≥ 2 , b ≥ 3 b\ge3 b ≥ 3 , a < r < b a<r<b a < r < b and 4 a < c ≤ 2 ( a + b ) 4a<c\le2(a+b) 4 a < c ≤ 2 ( a + b ) ;
none of a b ab ab , a c ac a c , b c bc b c is a perfect square; det A = det B = det C = 1 \det\mathbf{A}=\det\mathbf{B}=\det\mathbf{C}=1 det A = det B = det C = 1 , and α − 1 = r − a b \alpha^{-1}=r-\sqrt{ab} α − 1 = r − ab , β − 1 = s − a c \beta^{-1}=s-\sqrt{ac} β − 1 = s − a c , γ − 1 = t − b c \gamma^{-1}=t-\sqrt{bc} γ − 1 = t − b c ; for integers u , v u,v u , v ,
( u a + v b ) α = u 1 a + v 1 b , ( u a + v c ) β = u 2 a + v 2 c , (u\sqrt{a}+v\sqrt{b})\alpha=u_1\sqrt{a}+v_1\sqrt{b},\qquad(u\sqrt{a}+v\sqrt{c})\beta=u_2\sqrt{a}+v_2\sqrt{c}, ( u a + v b ) α = u 1 a + v 1 b , ( u a + v c ) β = u 2 a + v 2 c , ( u b + v c ) γ = u 3 b + v 3 c , (u\sqrt{b}+v\sqrt{c})\gamma=u_3\sqrt{b}+v_3\sqrt{c}, ( u b + v c ) γ = u 3 b + v 3 c , where
( u 1 v 1 ) = A ( u v ) , ( u 2 v 2 ) = B ( u v ) , ( u 3 v 3 ) = C ( u v ) ; \begin{pmatrix}u_1\\v_1\end{pmatrix}=\mathbf{A}\begin{pmatrix}u\\v\end{pmatrix},\qquad\begin{pmatrix}u_2\\v_2\end{pmatrix}=\mathbf{B}\begin{pmatrix}u\\v\end{pmatrix},\qquad\begin{pmatrix}u_3\\v_3\end{pmatrix}=\mathbf{C}\begin{pmatrix}u\\v\end{pmatrix}; ( u 1 v 1 ) = A ( u v ) , ( u 2 v 2 ) = B ( u v ) , ( u 3 v 3 ) = C ( u v ) ; 6.
B ( − 1 1 ) = ( t r ) , B − 1 ( 1 1 ) = ( − t r ) , C ( − 1 1 ) = ( s r ) and C − 1 ( 1 1 ) = ( − s r ) . \mathbf{B}\begin{pmatrix}-1\\1\end{pmatrix}=\begin{pmatrix}t\\r\end{pmatrix},\qquad\mathbf{B}^{-1}\begin{pmatrix}1\\1\end{pmatrix}=\begin{pmatrix}-t\\r\end{pmatrix},\qquad\mathbf{C}\begin{pmatrix}-1\\1\end{pmatrix}=\begin{pmatrix}s\\r\end{pmatrix}\qquad\text{and}\qquad\mathbf{C}^{-1}\begin{pmatrix}1\\1\end{pmatrix}=\begin{pmatrix}-s\\r\end{pmatrix}. B ( − 1 1 ) = ( t r ) , B − 1 ( 1 1 ) = ( − t r ) , C ( − 1 1 ) = ( s r ) and C − 1 ( 1 1 ) = ( − s r ) .
Proof. Throughout, c = a + b + 2 r c=a+b+2r c = a + b + 2 r , s = a + r s=a+r s = a + r , t = b + r t=b+r t = b + r and r 2 = a b + 1 r^2=ab+1 r 2 = ab + 1 . We first prove items 1 and 2. One has
s 2 = a 2 + 2 a r + a b + 1 = a ( a + b + 2 r ) + 1 = a c + 1 , s^2=a^2+2ar+ab+1=a(a+b+2r)+1=ac+1, s 2 = a 2 + 2 a r + ab + 1 = a ( a + b + 2 r ) + 1 = a c + 1 ,
and in the same way t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 ; with a b + 1 = r 2 ab+1=r^2 ab + 1 = r 2 this shows that { a , b , c } \{a,b,c\} { a , b , c } is a Diophantine triple (Definition 2.1 ). Next,
c r − s t = r ( a + b ) + 2 r 2 − a b − r ( a + b ) − r 2 = r 2 − a b = 1 , cr-st=r(a+b)+2r^2-ab-r(a+b)-r^2=r^2-ab=1, cr − s t = r ( a + b ) + 2 r 2 − ab − r ( a + b ) − r 2 = r 2 − ab = 1 ,
r s − a t = r a + r 2 − a b − a r = 1 , r t − b s = r b + r 2 − a b − b r = 1. rs-at=ra+r^2-ab-ar=1,\qquad rt-bs=rb+r^2-ab-br=1. r s − a t = r a + r 2 − ab − a r = 1 , r t − b s = r b + r 2 − ab − b r = 1.
By Definition 2.2 and item 1, d + ( a , b , c ) = a + b + ( 2 a b + 1 ) c + 2 r s t d_{+}(a,b,c)=a+b+(2ab+1)c+2rst d + ( a , b , c ) = a + b + ( 2 ab + 1 ) c + 2 r s t . Using a + b = c − 2 r a+b=c-2r a + b = c − 2 r , 2 a b + 1 = 2 r 2 − 1 2ab+1=2r^2-1 2 ab + 1 = 2 r 2 − 1 and c r − 1 = s t cr-1=st cr − 1 = s t ,
d + ( a , b , c ) = c − 2 r + ( 2 r 2 − 1 ) c + 2 r s t = 2 r ( c r − 1 + s t ) = 4 r s t = 4 r ( c r − 1 ) . d_{+}(a,b,c)=c-2r+(2r^2-1)c+2rst=2r(cr-1+st)=4rst=4r(cr-1). d + ( a , b , c ) = c − 2 r + ( 2 r 2 − 1 ) c + 2 r s t = 2 r ( cr − 1 + s t ) = 4 r s t = 4 r ( cr − 1 ) .
We next prove item 3. A common divisor of r r r and s s s divides r t − b s = 1 rt-bs=1 r t − b s = 1 ; a common divisor of r r r and t t t , or of s s s and t t t , divides r s − a t = 1 rs-at=1 r s − a t = 1 ; a common divisor of a a a and r r r divides r s − a t = 1 rs-at=1 r s − a t = 1 . Finally s 2 − a c = 1 s^2-ac=1 s 2 − a c = 1 and t 2 − b c = 1 t^2-bc=1 t 2 − b c = 1 give gcd ( a , s ) = gcd ( b , t ) = 1 \gcd(a,s)=\gcd(b,t)=1 g cd( a , s ) = g cd( b , t ) = 1 .
We next prove item 4. If b = a + 1 b=a+1 b = a + 1 , then a 2 < a b + 1 = a 2 + a + 1 < ( a + 1 ) 2 a^2<ab+1=a^2+a+1<(a+1)^2 a 2 < ab + 1 = a 2 + a + 1 < ( a + 1 ) 2 , so a b + 1 ab+1 ab + 1 is not a square; hence b − a ≥ 2 b-a\ge2 b − a ≥ 2 and b ≥ 3 b\ge3 b ≥ 3 . Since r 2 = a b + 1 > a 2 r^2=ab+1>a^2 r 2 = ab + 1 > a 2 , r > a r>a r > a ; since a ≤ b − 1 a\le b-1 a ≤ b − 1 , r 2 ≤ b 2 − b + 1 < b 2 r^2\le b^2-b+1<b^2 r 2 ≤ b 2 − b + 1 < b 2 , so r < b r<b r < b . Then c = a + b + 2 r > a + a + 2 a = 4 a c=a+b+2r>a+a+2a=4a c = a + b + 2 r > a + a + 2 a = 4 a . Finally ( a + b ) 2 − 4 r 2 = ( b − a ) 2 − 4 ≥ 0 (a+b)^2-4r^2=(b-a)^2-4\ge0 ( a + b ) 2 − 4 r 2 = ( b − a ) 2 − 4 ≥ 0 , so 2 r ≤ a + b 2r\le a+b 2 r ≤ a + b and c ≤ 2 ( a + b ) c\le2(a+b) c ≤ 2 ( a + b ) .
We next prove item 5. Since r > a ≥ 1 r>a\ge1 r > a ≥ 1 , r ≥ 2 r\ge2 r ≥ 2 and ( r − 1 ) 2 < r 2 − 1 = a b < r 2 (r-1)^2<r^2-1=ab<r^2 ( r − 1 ) 2 < r 2 − 1 = ab < r 2 , so a b ab ab is not a square; likewise a c = s 2 − 1 ac=s^2-1 a c = s 2 − 1 and b c = t 2 − 1 bc=t^2-1 b c = t 2 − 1 with s , t ≥ 3 s,t\ge3 s , t ≥ 3 . The determinants are r 2 − a b r^2-ab r 2 − ab , s 2 − a c s^2-ac s 2 − a c , t 2 − b c t^2-bc t 2 − b c , all equal to 1 by r 2 = a b + 1 r^2=ab+1 r 2 = ab + 1 and item 1, and ( r + a b ) ( r − a b ) = r 2 − a b = 1 (r+\sqrt{ab})(r-\sqrt{ab})=r^2-ab=1 ( r + ab ) ( r − ab ) = r 2 − ab = 1 , and likewise for β , γ \beta,\gamma β , γ . Expanding,
( u a + v b ) ( r + a b ) = ( r u + b v ) a + ( a u + r v ) b , (u\sqrt{a}+v\sqrt{b})(r+\sqrt{ab})=(ru+bv)\sqrt{a}+(au+rv)\sqrt{b}, ( u a + v b ) ( r + ab ) = ( r u + b v ) a + ( a u + r v ) b ,
( u a + v c ) ( s + a c ) = ( s u + c v ) a + ( a u + s v ) c , (u\sqrt{a}+v\sqrt{c})(s+\sqrt{ac})=(su+cv)\sqrt{a}+(au+sv)\sqrt{c}, ( u a + v c ) ( s + a c ) = ( s u + c v ) a + ( a u + s v ) c ,
( u b + v c ) ( t + b c ) = ( t u + c v ) b + ( b u + t v ) c . (u\sqrt{b}+v\sqrt{c})(t+\sqrt{bc})=(tu+cv)\sqrt{b}+(bu+tv)\sqrt{c}. ( u b + v c ) ( t + b c ) = ( t u + c v ) b + ( b u + t v ) c .
which are the matrix products of item 5. Since a b ab ab is not a square, a \sqrt{a} a and b \sqrt{b} b are linearly independent over Q \mathbb{Q} Q , so the coordinates in the first display are unique; likewise for the pairs a , c \sqrt{a},\sqrt{c} a , c and b , c \sqrt{b},\sqrt{c} b , c .
We finally prove item 6. Since c − s = b + r = t c-s=b+r=t c − s = b + r = t , s − a = r s-a=r s − a = r , c − t = a + r = s c-t=a+r=s c − t = a + r = s and t − b = r t-b=r t − b = r ,
B ( − 1 1 ) = ( c − s s − a ) = ( t r ) , B − 1 ( 1 1 ) = ( s − c − a s ) ( 1 1 ) = ( − t r ) . \mathbf{B}\begin{pmatrix}-1\\1\end{pmatrix}
=\begin{pmatrix}c-s\\s-a\end{pmatrix}
=\begin{pmatrix}t\\r\end{pmatrix},
\qquad
\mathbf{B}^{-1}\begin{pmatrix}1\\1\end{pmatrix}
=\begin{pmatrix}s&-c\\-a&s\end{pmatrix}
\begin{pmatrix}1\\1\end{pmatrix}
=\begin{pmatrix}-t\\r\end{pmatrix}. B ( − 1 1 ) = ( c − s s − a ) = ( t r ) , B − 1 ( 1 1 ) = ( s − a − c s ) ( 1 1 ) = ( − t r ) .
C ( − 1 1 ) = ( c − t t − b ) = ( s r ) , C − 1 ( 1 1 ) = ( t − c − b t ) ( 1 1 ) = ( − s r ) . \mathbf{C}\begin{pmatrix}-1\\1\end{pmatrix}
=\begin{pmatrix}c-t\\t-b\end{pmatrix}
=\begin{pmatrix}s\\r\end{pmatrix},
\qquad
\mathbf{C}^{-1}\begin{pmatrix}1\\1\end{pmatrix}
=\begin{pmatrix}t&-c\\-b&t\end{pmatrix}
\begin{pmatrix}1\\1\end{pmatrix}
=\begin{pmatrix}-s\\r\end{pmatrix}. C ( − 1 1 ) = ( c − t t − b ) = ( s r ) , C − 1 ( 1 1 ) = ( t − b − c t ) ( 1 1 ) = ( − s r ) .
where B − 1 \mathbf{B}^{-1} B − 1 and C − 1 \mathbf{C}^{-1} C − 1 are the adjugates because the determinants are 1. □
The index system of an extension
An extension d d d of an Euler triple gives one solution of the Pell equation of each pair of the triple, and the index system records where these solutions lie in the orbits of the units.
Definition 2.5. Let d d d be an extension (Definition 2.2) of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } with units α , β , γ \alpha,\beta,\gamma α , β , γ (Definition 2.3), and let x , y , z x,y,z x , y , z be the positive integers with
a d + 1 = x 2 , b d + 1 = y 2 , c d + 1 = z 2 . ad+1=x^2,\qquad bd+1=y^2,\qquad cd+1=z^2. a d + 1 = x 2 , b d + 1 = y 2 , c d + 1 = z 2 . An index system of d d d is a quadruple ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) , where λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } and l , m , n ≥ 0 l,m,n\ge0 l , m , n ≥ 0 are integers, such that
y a + x b = ( a + b ) α 2 l , z a + x c = ( λ a + c ) β 2 m , z b + y c = ( λ b + c ) γ 2 n . y\sqrt{a}+x\sqrt{b}=(\sqrt{a}+\sqrt{b})\alpha^{2l},\qquad
z\sqrt{a}+x\sqrt{c}=(\lambda\sqrt{a}+\sqrt{c})\beta^{2m},\qquad
z\sqrt{b}+y\sqrt{c}=(\lambda\sqrt{b}+\sqrt{c})\gamma^{2n}. y a + x b = ( a + b ) α 2 l , z a + x c = ( λ a + c ) β 2 m , z b + y c = ( λ b + c ) γ 2 n . The integer m m m is the index of the index system. The Pell equation of the pair { a , b } \{a,b\} { a , b } is the equation a y 2 − b x 2 = a − b ay^2-bx^2=a-b a y 2 − b x 2 = a − b in the unknowns x , y x,y x , y ; likewise the Pell equation of { a , c } \{a,c\} { a , c } is a z 2 − c x 2 = a − c az^2-cx^2=a-c a z 2 − c x 2 = a − c , and that of { b , c } \{b,c\} { b , c } is b z 2 − c y 2 = b − c bz^2-cy^2=b-c b z 2 − c y 2 = b − c .
By Proposition 2.11 below, every extension has exactly one index system, so we speak of the index system and the index of an extension. The pair ( x , y ) (x,y) ( x , y ) solves the Pell equation of { a , b } \{a,b\} { a , b } , since a y 2 − b x 2 = a ( b d + 1 ) − b ( a d + 1 ) = a − b ay^2-bx^2=a(bd+1)-b(ad+1)=a-b a y 2 − b x 2 = a ( b d + 1 ) − b ( a d + 1 ) = a − b ; likewise ( x , z ) (x,z) ( x , z ) and ( y , z ) (y,z) ( y , z ) solve those of { a , c } \{a,c\} { a , c } and { b , c } \{b,c\} { b , c } . The second and third displays say that z = V m λ = W n λ z=V_m^\lambda=W_n^\lambda z = V m λ = W n λ (Lemma 2.8), with the Pell sequences of the following convention; in the notation of Dujella [13 ] and of Fujita and Miyazaki [18 ] this reads z = v 2 m = w 2 n z=v_{2m}=w_{2n} z = v 2 m = w 2 n .
Convention 2.6. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with multiplication matrices B , C \mathbf{B},\mathbf{C} B , C (Definition 2.3 ). For λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } and integers j ≥ 0 j \ge0 j ≥ 0 , the Pell sequences V j λ V_j^\lambda V j λ and W j λ W_j^\lambda W j λ are the first coordinates of the integer vectors
B 2 j ( λ 1 ) and C 2 j ( λ 1 ) , \mathbf{B}^{2j}\begin{pmatrix}\lambda\\1\end{pmatrix}
\qquad\text{and}\qquad
\mathbf{C}^{2j}\begin{pmatrix}\lambda\\1\end{pmatrix}, B 2 j ( λ 1 ) and C 2 j ( λ 1 ) , respectively.
Lemma 2.7. Let P , Q , S P,Q,S P , Q , S be positive integers with P Q + 1 = S 2 PQ+1=S^2 P Q + 1 = S 2 , put η P Q = S + P Q \eta_{PQ}=S+\sqrt{PQ} η P Q = S + P Q , and let u 0 , v 0 ∈ Z u_0,v_0\in\mathbb{Z} u 0 , v 0 ∈ Z . For j ≥ 0 j\ge0 j ≥ 0 let u j , v j ∈ Z u_j,v_j\in\mathbb{Z} u j , v j ∈ Z be defined by
u j P + v j Q = ( u 0 P + v 0 Q ) η P Q j . u_j\sqrt{P}+v_j\sqrt{Q}=(u_0\sqrt{P}+v_0\sqrt{Q})\eta_{PQ}^{j}. u j P + v j Q = ( u 0 P + v 0 Q ) η P Q j . The even subsequences satisfy X i + 1 = ( 4 P Q + 2 ) X i − X i − 1 X_{i+1}=(4PQ+2)X_i-X_{i-1} X i + 1 = ( 4 P Q + 2 ) X i − X i − 1 for i ≥ 1 i\ge1 i ≥ 1 , where X i = u 2 i X_i=u_{2i} X i = u 2 i or X i = v 2 i X_i=v_{2i} X i = v 2 i .
For every integer i ≥ 0 i\ge0 i ≥ 0 ,
u 2 i ≡ u 0 + 2 Q i ( P i u 0 + S v 0 ) ( m o d 8 P Q 2 ) , u_{2i}\equiv u_0+2Qi(P^iu_0+Sv_0)\pmod{8PQ^2}, u 2 i ≡ u 0 + 2 Q i ( P i u 0 + S v 0 ) ( mod 8 P Q 2 ) , v 2 i ≡ v 0 + 2 P i ( Q i v 0 + S u 0 ) ( m o d 8 P 2 Q ) . v_{2i}\equiv v_0+2Pi(Q^iv_0+Su_0)\pmod{8P^2Q}. v 2 i ≡ v 0 + 2 P i ( Q i v 0 + S u 0 ) ( mod 8 P 2 Q ) .
Proof. This is the second-order congruence of Dujella (see [15 ] ), in a form symmetric in the two square roots; we give the induction.
Since P Q = S 2 − 1 PQ=S^2-1 P Q = S 2 − 1 with S ≥ 2 S\ge2 S ≥ 2 , P Q PQ P Q is not a square, so P \sqrt{P} P and Q \sqrt{Q} Q are linearly independent over Q \mathbb{Q} Q and u j , v j u_j,v_j u j , v j are well defined; explicitly u j + 1 = S u j + Q v j u_{j+1}=Su_j+Qv_j u j + 1 = S u j + Q v j and v j + 1 = P u j + S v j v_{j+1}=Pu_j+Sv_j v j + 1 = P u j + S v j .
We first prove item 1. Since η P Q − 1 = S − P Q \eta_{PQ}^{-1}=S-\sqrt{PQ} η P Q − 1 = S − P Q , one has η P Q + η P Q − 1 = 2 ( 2 P Q + 1 ) = 4 P Q + 2 \eta_{PQ}+\eta_{PQ}^{-1}=2(2PQ+1)=4PQ+2 η P Q + η P Q − 1 = 2 ( 2 P Q + 1 ) = 4 P Q + 2 , hence
η P Q 2 ( i + 1 ) = ( 4 P Q + 2 ) η P Q 2 i − η P Q 2 ( i − 1 ) . \eta_{PQ}^{2(i+1)}=(4PQ+2)\eta_{PQ}^{2i}-\eta_{PQ}^{2(i-1)}. η P Q 2 ( i + 1 ) = ( 4 P Q + 2 ) η P Q 2 i − η P Q 2 ( i − 1 ) .
Multiplying by u 0 P + v 0 Q u_0\sqrt{P}+v_0\sqrt{Q} u 0 P + v 0 Q and comparing the coefficients of P \sqrt{P} P and Q \sqrt{Q} Q gives item 1.
We next prove item 2. Put T = P Q T=PQ T = P Q and write η P Q 2 i = U i + K i T \eta_{PQ}^{2i}=U_i+K_i\sqrt{T} η P Q 2 i = U i + K i T with U i , K i ∈ Z U_i,K_i\in\mathbb{Z} U i , K i ∈ Z . Expanding ( u 0 P + v 0 Q ) ( U i + K i P Q ) (u_0\sqrt{P}+v_0\sqrt{Q})(U_i+K_i\sqrt{PQ}) ( u 0 P + v 0 Q ) ( U i + K i P Q ) gives
u 2 i = u 0 U i + Q v 0 K i , v 2 i = v 0 U i + P u 0 K i . (1) u_{2i}=u_0U_i+Qv_0K_i,\qquad v_{2i}=v_0U_i+Pu_0K_i. \tag*{(1)} u 2 i = u 0 U i + Q v 0 K i , v 2 i = v 0 U i + P u 0 K i . (1)
From η P Q 2 i + 2 = η P Q 2 i ( ( 2 T + 1 ) + 2 S T ) \eta_{PQ}^{2i+2}=\eta_{PQ}^{2i}((2T+1)+2S\sqrt{T}) η P Q 2 i + 2 = η P Q 2 i (( 2 T + 1 ) + 2 S T ) one gets U i + 1 = ( 2 T + 1 ) U i + 2 S T K i U_{i+1}=(2T+1)U_i+2STK_i U i + 1 = ( 2 T + 1 ) U i + 2 S T K i and K i + 1 = ( 2 T + 1 ) K i + 2 S U i K_{i+1}=(2T+1)K_i+2SU_i K i + 1 = ( 2 T + 1 ) K i + 2 S U i , with U 0 = 1 U_0=1 U 0 = 1 , K 0 = 0 K_0=0 K 0 = 0 . By induction K i = S J i K_i=SJ_i K i = S J i with J i ∈ Z J_i\in\mathbb{Z} J i ∈ Z , where J 0 = 0 J_0=0 J 0 = 0 and, using S 2 = T + 1 S^2=T+1 S 2 = T + 1 ,
J i + 1 = ( 2 T + 1 ) J i + 2 U i , U i + 1 = ( 2 T + 1 ) U i + 2 T ( T + 1 ) J i . J_{i+1}=(2T+1)J_i+2U_i,\qquad U_{i+1}=(2T+1)U_i+2T(T+1)J_i. J i + 1 = ( 2 T + 1 ) J i + 2 U i , U i + 1 = ( 2 T + 1 ) U i + 2 T ( T + 1 ) J i .
We claim that
U i ≡ 1 + 2 i 2 T ( m o d 8 T 2 ) , J i ≡ 2 i ( m o d 8 T ) . (2) U_i\equiv1+2i^2T\pmod{8T^2},\qquad J_i\equiv2i\pmod{8T}. \tag*{(2)} U i ≡ 1 + 2 i 2 T ( mod 8 T 2 ) , J i ≡ 2 i ( mod 8 T ) . (2)
This holds for i = 0 i=0 i = 0 . If U i = 1 + 2 i 2 T + 8 T 2 g U_i=1+2i^2T+8T^2g U i = 1 + 2 i 2 T + 8 T 2 g and J i = 2 i + 8 T h J_i=2i+8Th J i = 2 i + 8 T h with g , h ∈ Z g,h\in\mathbb{Z} g , h ∈ Z , then
J i + 1 = 2 ( i + 1 ) + 4 i ( i + 1 ) T + 8 T ( h + 2 T h + 2 T g ) , J_{i+1}=2(i+1)+4i(i+1)T+8T(h+2Th+2Tg), J i + 1 = 2 ( i + 1 ) + 4 i ( i + 1 ) T + 8 T ( h + 2 T h + 2 T g ) ,
U i + 1 = 1 + 2 ( i + 1 ) 2 T + 4 i ( i + 1 ) T 2 + 8 T 2 ( g + 2 T g + 2 h + 2 T h ) , U_{i+1}=1+2(i+1)^2T+4i(i+1)T^2+8T^2(g+2Tg+2h+2Th), U i + 1 = 1 + 2 ( i + 1 ) 2 T + 4 i ( i + 1 ) T 2 + 8 T 2 ( g + 2 T g + 2 h + 2 T h ) ,
and 8 ∣ 4 i ( i + 1 ) 8\mid4i(i+1) 8 ∣ 4 i ( i + 1 ) , which proves (2) for i + 1. i+1. i + 1. By (1) and (2),
u 2 i − u 0 − 2 Q i ( P i u 0 + S v 0 ) = u 0 ( U i − 1 − 2 i 2 P Q ) + Q S v 0 ( J i − 2 i ) , u_{2i}-u_0-2Qi(Piu_0+Sv_0)=u_0(U_i-1-2i^2PQ)+QSv_0(J_i-2i), u 2 i − u 0 − 2 Q i ( P i u 0 + S v 0 ) = u 0 ( U i − 1 − 2 i 2 P Q ) + QS v 0 ( J i − 2 i ) ,
where the first term is divisible by 8 T 2 = 8 P 2 Q 2 8T^2=8P^2Q^2 8 T 2 = 8 P 2 Q 2 and the second by 8 Q T = 8 P Q 2 8QT=8PQ^2 8 QT = 8 P Q 2 . Likewise
v 2 i − v 0 − 2 P i ( Q i v 0 + S u 0 ) = v 0 ( U i − 1 − 2 i 2 P Q ) + P S u 0 ( J i − 2 i ) ≡ 0 ( m o d 8 P 2 Q ) . v_{2i}-v_0-2Pi(Q_iv_0+Su_0)=v_0(U_i-1-2i^2PQ)+PSu_0(J_i-2i)\equiv0\pmod{8P^2Q}. v 2 i − v 0 − 2 P i ( Q i v 0 + S u 0 ) = v 0 ( U i − 1 − 2 i 2 P Q ) + P S u 0 ( J i − 2 i ) ≡ 0 ( mod 8 P 2 Q ) .
Lemma 2.8. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with roots r , s , t r,s,t r , s , t , units α , β , γ \alpha,\beta,\gamma α , β , γ and multiplication matrices A , B , C \mathbf{A},\mathbf{B},\mathbf{C} A , B , C (Definition 2.3), and let λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } .
For every integer k ≥ 0 k\ge0 k ≥ 0 , the vectors A k ( 1 1 ) \mathbf{A}^k\binom{1}{1} A k ( 1 1 ) , B k ( λ 1 ) \mathbf{B}^k\binom{\lambda}{1} B k ( 1 λ ) and C k ( λ 1 ) \mathbf{C}^k\binom{\lambda}{1} C k ( 1 λ ) are the coordinate vectors of ( a + b ) α k (\sqrt{a}+\sqrt{b})\alpha^k ( a + b ) α k with respect to a , b \sqrt{a},\sqrt{b} a , b , of ( λ a + c ) β k (\lambda\sqrt{a}+\sqrt{c})\beta^k ( λ a + c ) β k with respect to a , c \sqrt{a},\sqrt{c} a , c , and of ( λ b + c ) γ k (\lambda\sqrt{b}+\sqrt{c})\gamma^k ( λ b + c ) γ k with respect to b , c \sqrt{b},\sqrt{c} b , c . Equivalently, they are the vectors ( u k v k ) \binom{u_k}{v_k} ( v k u k ) of Lemma 2.7 for ( P , Q , S , u 0 , v 0 ) = ( a , b , r , 1 , 1 ) (P,Q,S,u_0,v_0)=(a,b,r,1,1) ( P , Q , S , u 0 , v 0 ) = ( a , b , r , 1 , 1 ) , ( a , c , s , λ , 1 ) (a,c,s,\lambda,1) ( a , c , s , λ , 1 ) and ( b , c , t , λ , 1 ) (b,c,t,\lambda,1) ( b , c , t , λ , 1 ) , respectively.
Let d d d be an extension of { a , b , c } \{a,b,c\} { a , b , c } , let x , y , z x,y,z x , y , z be as in Definition 2.5, and let l , m , n ≥ 0 l,m,n\ge0 l , m , n ≥ 0 be integers. The first, second and third displays of Definition 2.5 hold if and only if, respectively,
( y x ) = A 2 l ( 1 1 ) , ( z x ) = B 2 m ( λ 1 ) , ( z y ) = C 2 n ( λ 1 ) . \binom{y}{x}=\mathbf{A}^{2l}\binom{1}{1},\qquad
\binom{z}{x}=\mathbf{B}^{2m}\binom{\lambda}{1},\qquad
\binom{z}{y}=\mathbf{C}^{2n}\binom{\lambda}{1}. ( x y ) = A 2 l ( 1 1 ) , ( x z ) = B 2 m ( 1 λ ) , ( y z ) = C 2 n ( 1 λ ) . In particular, if ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) is an index system of d d d , then z = V m λ = W n λ z=V_m^\lambda=W_n^\lambda z = V m λ = W n λ , with the Pell sequences of Convention 2.6.
For every integer j ≥ 1 j\ge1 j ≥ 1 ,
A 2 j + 2 ( 1 1 ) = ( 4 a b + 2 ) A 2 j ( 1 1 ) − A 2 j − 2 ( 1 1 ) , \mathbf{A}^{2j+2}\binom{1}{1}=(4ab+2)\mathbf{A}^{2j}\binom{1}{1}-\mathbf{A}^{2j-2}\binom{1}{1}, A 2 j + 2 ( 1 1 ) = ( 4 ab + 2 ) A 2 j ( 1 1 ) − A 2 j − 2 ( 1 1 ) , B 2 j + 2 ( λ 1 ) = ( 4 a c + 2 ) B 2 j ( λ 1 ) − B 2 j − 2 ( λ 1 ) , \mathbf{B}^{2j+2}\binom{\lambda}{1}=(4ac+2)\mathbf{B}^{2j}\binom{\lambda}{1}-\mathbf{B}^{2j-2}\binom{\lambda}{1}, B 2 j + 2 ( 1 λ ) = ( 4 a c + 2 ) B 2 j ( 1 λ ) − B 2 j − 2 ( 1 λ ) , C 2 j + 2 ( λ 1 ) = ( 4 b c + 2 ) C 2 j ( λ 1 ) − C 2 j − 2 ( λ 1 ) . \mathbf{C}^{2j+2}\binom{\lambda}{1}=(4bc+2)\mathbf{C}^{2j}\binom{\lambda}{1}-\mathbf{C}^{2j-2}\binom{\lambda}{1}. C 2 j + 2 ( 1 λ ) = ( 4 b c + 2 ) C 2 j ( 1 λ ) − C 2 j − 2 ( 1 λ ) . In particular the Pell sequences satisfy V j + 1 λ = ( 4 a c + 2 ) V j λ − V j − 1 λ V_{j+1}^\lambda=(4ac+2)V_j^\lambda-V_{j-1}^\lambda V j + 1 λ = ( 4 a c + 2 ) V j λ − V j − 1 λ and W j + 1 λ = ( 4 b c + 2 ) W j λ − W j − 1 λ W_{j+1}^\lambda=(4bc+2)W_j^\lambda-W_{j-1}^\lambda W j + 1 λ = ( 4 b c + 2 ) W j λ − W j − 1 λ for j ≥ 1 j\ge1 j ≥ 1 , and their initial terms are
V 0 λ = W 0 λ = λ , V 1 λ = λ ( 2 a c + 1 ) + 2 s c , W 1 λ = λ ( 2 b c + 1 ) + 2 t c . V_0^\lambda=W_0^\lambda=\lambda,\qquad V_1^\lambda=\lambda(2ac+1)+2sc,\qquad W_1^\lambda=\lambda(2bc+1)+2tc. V 0 λ = W 0 λ = λ , V 1 λ = λ ( 2 a c + 1 ) + 2 sc , W 1 λ = λ ( 2 b c + 1 ) + 2 t c . Both coordinates of B 2 j ( λ 1 ) \mathbf{B}^{2j}\binom{\lambda}{1} B 2 j ( 1 λ ) and both coordinates of C 2 j ( λ 1 ) \mathbf{C}^{2j}\binom{\lambda}{1} C 2 j ( 1 λ ) are strictly increasing in j ≥ 0 j\ge0 j ≥ 0 and positive for j ≥ 1 j\ge1 j ≥ 1 .
Proof. Item 1 is item 5 of Lemma 2.4, applied k k k times; items 2 and 3 follow from it, the recurrences by item 1 of Lemma 2.7 and the monotonicity by induction.
We first prove item 1. By Lemma 2.4, item 5, a b ab ab is not a square, so a \sqrt{a} a and b \sqrt{b} b are linearly independent over Q \mathbb{Q} Q and an element u a + v b u\sqrt{a}+v\sqrt{b} u a + v b with u , v ∈ Z u,v\in\mathbb{Z} u , v ∈ Z has a unique coordinate vector ( u v ) \binom{u}{v} ( v u ) ; by the same item, multiplication by α \alpha α sends the element with coordinate vector ( u v ) \binom{u}{v} ( v u ) to the element with coordinate vector A ( u v ) \mathbf{A}\binom{u}{v} A ( v u ) . Induction on k k k shows that A k ( 1 1 ) \mathbf{A}^k\binom{1}{1} A k ( 1 1 ) is the coordinate vector of ( a + b ) α k (\sqrt{a}+\sqrt{b})\alpha^k ( a + b ) α k . In the same way, since a c ac a c and b c bc b c are not squares (same item), B k ( λ 1 ) \mathbf{B}^k\binom{\lambda}{1} B k ( 1 λ ) and C k ( λ 1 ) \mathbf{C}^k\binom{\lambda}{1} C k ( 1 λ ) are the coordinate vectors of ( λ a + c ) β k (\lambda\sqrt{a}+\sqrt{c})\beta^k ( λ a + c ) β k and ( λ b + c ) γ k (\lambda\sqrt{b}+\sqrt{c})\gamma^k ( λ b + c ) γ k . Finally a b + 1 = r 2 ab+1=r^2 ab + 1 = r 2 (Definition 2.3), a c + 1 = s 2 ac+1=s^2 a c + 1 = s 2 and b c + 1 = t 2 bc+1=t^2 b c + 1 = t 2 (Lemma 2.4, item 1), so ( P , Q , S ) = ( a , b , r ) (P,Q,S)=(a,b,r) ( P , Q , S ) = ( a , b , r ) , ( a , c , s ) (a,c,s) ( a , c , s ) and ( b , c , t ) (b,c,t) ( b , c , t ) satisfy the hypothesis of Lemma 2.7, with η P Q = α \eta_{PQ}=\alpha η P Q = α , β \beta β and γ \gamma γ ; and its defining identity u k P + v k Q = ( u 0 P + v 0 Q ) η P Q k u_k\sqrt{P}+v_k\sqrt{Q}=(u_0\sqrt{P}+v_0\sqrt{Q})\eta_{PQ}^k u k P + v k Q = ( u 0 P + v 0 Q ) η P Q k says that ( u k v k ) \binom{u_k}{v_k} ( v k u k ) is the coordinate vector just described.
We next prove item 2. By item 1, the right side ( a + b ) α 2 l (\sqrt{a}+\sqrt{b})\alpha^{2l} ( a + b ) α 2 l of the first display of Definition 2.5 has coordinate vector A 2 l ( 1 1 ) \mathbf{A}^{2l}\binom{1}{1} A 2 l ( 1 1 ) , and its left side y a + x b y\sqrt{a}+x\sqrt{b} y a + x b has coordinate vector ( y x ) \binom{y}{x} ( x y ) . Since coordinates with respect to a , b \sqrt{a},\sqrt{b} a , b are unique, the first display holds if and only if ( y x ) = A 2 l ( 1 1 ) \binom{y}{x}=\mathbf{A}^{2l}\binom{1}{1} ( x y ) = A 2 l ( 1 1 ) . In the same way the second display holds if and only if ( x z ) = B 2 m ( λ 1 ) \binom{x}{z}=\mathbf{B}^{2m}\binom{\lambda}{1} ( z x ) = B 2 m ( 1 λ ) , and the third if and only if ( y z ) = C 2 n ( λ 1 ) \binom{y}{z}=\mathbf{C}^{2n}\binom{\lambda}{1} ( z y ) = C 2 n ( 1 λ ) . If ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) is an index system of d d d , the first coordinates of the last two identities are z = V m λ z=V_m^\lambda z = V m λ and z = W n λ z=W_n^\lambda z = W n λ , by Convention 2.6.
We next prove the recurrences of item 3. By item 1, the two coordinates of A 2 j ( 1 1 ) \mathbf{A}^{2j}\binom{1}{1} A 2 j ( 1 1 ) , of B 2 j ( λ 1 ) \mathbf{B}^{2j}\binom{\lambda}{1} B 2 j ( 1 λ ) and of C 2 j ( λ 1 ) \mathbf{C}^{2j}\binom{\lambda}{1} C 2 j ( 1 λ ) are the terms u 2 j u_{2j} u 2 j and v 2 j v_{2j} v 2 j of Lemma 2.7 for P Q = a b PQ=ab P Q = ab , a c ac a c and b c bc b c , respectively. By item 1 of Lemma 2.7, each of them satisfies X j + 1 = ( 4 P Q + 2 ) X j − X j − 1 X_{j+1}=(4PQ+2)X_j-X_{j-1} X j + 1 = ( 4 P Q + 2 ) X j − X j − 1 for j ≥ 1 j\geq1 j ≥ 1 , which gives the three vector recurrences; their first coordinates for B \mathbf{B} B and C \mathbf{C} C are the recurrences of V j λ V_j^\lambda V j λ and W j λ W_j^\lambda W j λ (Convention 2.6).
We next compute the initial terms. Since B 0 \mathbf{B}^0 B 0 and C 0 \mathbf{C}^0 C 0 are the identity matrix, V 0 λ = W 0 λ = λ V_0^\lambda=W_0^\lambda=\lambda V 0 λ = W 0 λ = λ . Next B ( λ 1 ) = ( λ s + c λ a + s ) \mathbf{B}\binom{\lambda}{1}=\binom{\lambda s+c}{\lambda a+s} B ( 1 λ ) = ( λa + s λ s + c ) , and, by s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 ,
B 2 ( λ 1 ) = ( s ( λ s + c ) + c ( λ a + s ) a ( λ s + c ) + s ( λ a + s ) ) = ( λ ( 2 a c + 1 ) + 2 s c 1 + 2 a ( c + λ s ) ) ; \mathbf{B}^2\binom{\lambda}{1}
=
\binom{s(\lambda s+c)+c(\lambda a+s)}{a(\lambda s+c)+s(\lambda a+s)}
=
\binom{\lambda(2ac+1)+2sc}{1+2a(c+\lambda s)}; B 2 ( 1 λ ) = ( a ( λ s + c ) + s ( λa + s ) s ( λ s + c ) + c ( λa + s ) ) = ( 1 + 2 a ( c + λ s ) λ ( 2 a c + 1 ) + 2 sc ) ;
in the same way, by t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 , C 2 ( λ 1 ) = ( λ ( 2 b c + 1 ) + 2 t c 1 + 2 b ( c + λ t ) ) \mathbf{C}^2\binom{\lambda}{1}=\binom{\lambda(2bc+1)+2tc}{1+2b(c+\lambda t)} C 2 ( 1 λ ) = ( 1 + 2 b ( c + λ t ) λ ( 2 b c + 1 ) + 2 t c ) . The first coordinates are V 1 λ V_1^\lambda V 1 λ and W 1 λ W_1^\lambda W 1 λ .
We finally prove the monotonicity. Let ( X j ) j ≥ 0 (X_j)_{j\geq0} ( X j ) j ≥ 0 be one of the four coordinate sequences of B 2 j ( λ 1 ) \mathbf{B}^{2j}\binom{\lambda}{1} B 2 j ( 1 λ ) and C 2 j ( λ 1 ) \mathbf{C}^{2j}\binom{\lambda}{1} C 2 j ( 1 λ ) . By the recurrences, X j + 1 − X j = 4 a c X j + ( X j − X j − 1 ) X_{j+1}-X_j=4acX_j+(X_j-X_{j-1}) X j + 1 − X j = 4 a c X j + ( X j − X j − 1 ) for j ≥ 1 j\geq1 j ≥ 1 in the first two cases, and X j + 1 − X j = 4 b c X j + ( X j − X j − 1 ) X_{j+1}-X_j=4bcX_j+(X_j-X_{j-1}) X j + 1 − X j = 4 b c X j + ( X j − X j − 1 ) in the last two. Moreover X 1 > X 0 X_1>X_0 X 1 > X 0 and X 1 > 0 X_1>0 X 1 > 0 : the second coordinates start at 1 1 1 and the first at λ \lambda λ , and, since c − s = b + r > 0 c-s=b+r>0 c − s = b + r > 0 , c − t = a + r > 0 c-t=a+r>0 c − t = a + r > 0 and s − a = t − b = r s-a=t-b=r s − a = t − b = r (Definition 2.3),
1 + 2 a ( c + λ s ) ≥ 1 + 2 a ( c − s ) > 1 , 1 + 2 b ( c + λ t ) ≥ 1 + 2 b ( c − t ) > 1 , 1+2a(c+\lambda s)\geq1+2a(c-s)>1,\qquad1+2b(c+\lambda t)\geq1+2b(c-t)>1, 1 + 2 a ( c + λ s ) ≥ 1 + 2 a ( c − s ) > 1 , 1 + 2 b ( c + λ t ) ≥ 1 + 2 b ( c − t ) > 1 ,
V 1 λ , W 1 λ ≥ 2 c r − 1 > 1 ≥ λ . V_1^\lambda,\ W_1^\lambda\geq2cr-1>1\geq\lambda. V 1 λ , W 1 λ ≥ 2 cr − 1 > 1 ≥ λ .
If X j > 0 X_j>0 X j > 0 and X j > X j − 1 X_j>X_{j-1} X j > X j − 1 for some j ≥ 1 j\geq1 j ≥ 1 , the identity above gives X j + 1 > X j > 0 X_{j+1}>X_j>0 X j + 1 > X j > 0 . By induction, ( X j ) j ≥ 0 (X_j)_{j\geq0} ( X j ) j ≥ 0 is strictly increasing, and X j > 0 X_j>0 X j > 0 for j ≥ 1 j\geq1 j ≥ 1 .
We quote two results on the fundamental solutions of these Pell equations; they are the inputs of Proposition 2.11.
Theorem 2.9. Let d d d be an extension of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } with roots r , s , t r,s,t r , s , t and multiplication matrices B , C \mathbf{B},\mathbf{C} B , C , and let x , y , z x,y,z x , y , z be as in Definition 2.5. There are integers k 1 , k 2 ≥ 0 k_1,k_2\geq0 k 1 , k 2 ≥ 0 and integers z 1 , x 1 , z 2 , y 2 z_1,x_1,z_2,y_2 z 1 , x 1 , z 2 , y 2 with x 1 , y 2 > 0 x_1,y_2>0 x 1 , y 2 > 0 such that
( z x ) = B k 1 ( z 1 x 1 ) , ( z y ) = C k 2 ( z 2 y 2 ) , \binom{z}{x}=\mathbf{B}^{k_1}\binom{z_1}{x_1},\qquad
\binom{z}{y}=\mathbf{C}^{k_2}\binom{z_2}{y_2}, ( x z ) = B k 1 ( x 1 z 1 ) , ( y z ) = C k 2 ( y 2 z 2 ) , and, according to the parities of ( k 1 , k 2 ) (k_1,k_2) ( k 1 , k 2 ) :
even/even : ( ∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( 1 , 1 ; 1 , 1 ) (|z_1|,x_1;|z_2|,y_2)=(1,1;1,1) ( ∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( 1 , 1 ; 1 , 1 ) , with z 1 = z 2 z_1=z_2 z 1 = z 2 ;
odd/even : ( ∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( t , r ; 1 , 1 ) (\lvert z_1\rvert, x_1; \lvert z_2\rvert, y_2) = (t, r; 1, 1) (∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( t , r ; 1 , 1 ) , with z 1 z 2 < 0 z_1z_2 < 0 z 1 z 2 < 0 ;
even/odd : ( ∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( 1 , 1 ; s , r ) (\lvert z_1\rvert, x_1; \lvert z_2\rvert, y_2) = (1, 1; s, r) (∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( 1 , 1 ; s , r ) , with z 1 z 2 < 0 z_1z_2 < 0 z 1 z 2 < 0 ;
odd/odd : ( ∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( t , r ; s , r ) (\lvert z_1\rvert, x_1; \lvert z_2\rvert, y_2) = (t, r; s, r) (∣ z 1 ∣ , x 1 ; ∣ z 2 ∣ , y 2 ) = ( t , r ; s , r ) , with z 1 z 2 > 0 z_1z_2 > 0 z 1 z 2 > 0 .
Proof. This is [11 Theorem 2.1] , specialised to an Euler triple; that theorem completes [18 Theorem 1.3] by the values of the second coordinates. In the source the fundamental solutions are written ( z 0 , x 0 ) (z_0,x_0) ( z 0 , x 0 ) and ( z 1 , y 1 ) (z_1,y_1) ( z 1 , y 1 ) and the exponents m , n m,n m , n ; its matrices of multiplication by β \beta β and γ \gamma γ are B \mathbf{B} B and C \mathbf{C} C (Lemma 2.4, item 5); and the values c r − s t cr-st cr − s t , r s − a t rs-at r s − a t , r t − b s rt-bs r t − b s that occur in its four cases are all equal to 1 (Lemma 2.4, item 2). With these values the two alternatives of the even/even case coincide, and the four cases become the four cases above.
Lemma 2.10
Lemma 2.10. Let P < Q P < Q P < Q be positive integers with P Q + 1 = S 2 PQ + 1 = S^2 P Q + 1 = S 2 for an integer S S S , and put η P Q = S + P Q \eta_{PQ} = S + \sqrt{PQ} η P Q = S + P Q . Every solution in positive integers ( X , Y ) (X,Y) ( X , Y ) of P Y 2 − Q X 2 = P − Q PY^2 - QX^2 = P - Q P Y 2 − Q X 2 = P − Q satisfies
Y P + X Q = ( Y 0 P + X 0 Q ) η P Q k Y\sqrt{P} + X\sqrt{Q} = (Y_0\sqrt{P} + X_0\sqrt{Q})\eta_{PQ}^{k} Y P + X Q = ( Y 0 P + X 0 Q ) η P Q k
for an integer k ≥ 0 k \ge0 k ≥ 0 and an integer solution ( X 0 , Y 0 ) (X_0,Y_0) ( X 0 , Y 0 ) of the same equation with
1 ≤ X 0 ≤ P ( Q − P ) 2 ( S − 1 ) , 1 ≤ ∣ Y 0 ∣ ≤ ( S − 1 ) ( Q − P ) 2 P . 1 \le X_0 \le\sqrt{\frac{P(Q-P)}{2(S-1)}}, \qquad1 \le\lvert Y_0\rvert\le\sqrt{\frac{(S-1)(Q-P)}{2P}}. 1 ≤ X 0 ≤ 2 ( S − 1 ) P ( Q − P ) , 1 ≤ ∣ Y 0 ∣ ≤ 2 P ( S − 1 ) ( Q − P ) .
In particular Y 0 2 < Q 2 / 4 Y_0^2 < Q^2/4 Y 0 2 < Q 2 /4 .
Proof. This is [19 Lemma 8] (numbering of the arXiv version 2), going back to [13 ] . For the last inequality, S 2 − 1 = P Q S^2 - 1 = PQ S 2 − 1 = P Q gives ( S − 1 ) / P = Q / ( S + 1 ) (S-1)/P = Q/(S+1) ( S − 1 ) / P = Q / ( S + 1 ) , so Y 0 2 ≤ Q ( Q − P ) / ( 2 ( S + 1 ) ) Y_0^2 \le Q(Q-P)/(2(S+1)) Y 0 2 ≤ Q ( Q − P ) / ( 2 ( S + 1 )) ; since S ≥ 2 S \ge2 S ≥ 2 and Q − P < Q Q-P < Q Q − P < Q , this is less than Q 2 / 4 Q^2/4 Q 2 /4 .
Proposition 2.11
Proposition 2.11. Every extension of an Euler triple has exactly one index system.
Proof. Existence comes from the classification of the fundamental solutions (Theorem 2.9 ), which gives the second and third displays of Definition 2.5, and from the bounded fundamental solutions of the Pell equation of { a , b } \{a,b\} { a , b } (Lemma 2.10), which give the first; uniqueness follows by comparing two index systems in the three displays. For d > d + ( a , b , c ) d > d_+(a,b,c) d > d + ( a , b , c ) existence is known [18 Introduction] ; see also [10 ] for the three-parameter family. The proof below is valid for every extension.
Let d d d be an extension of the Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with roots r , s , t r,s,t r , s , t , units α , β , γ \alpha,\beta,\gamma α , β , γ , multiplication matrices B , C \mathbf{B},\mathbf{C} B , C , and x , y , z x,y,z x , y , z as in Definition 2.5.
We first obtain the second and third displays of Definition 2.5. By Theorem 2.9 there are integers k 1 , k 2 ≥ 0 k_1,k_2 \ge0 k 1 , k 2 ≥ 0 and fundamental solutions with
( z x ) = B k 1 ( z 1 x 1 ) \begin{pmatrix} z \\ x \end{pmatrix} = \mathbf{B}^{k_1}\begin{pmatrix} z_1 \\ x_1 \end{pmatrix} ( z x ) = B k 1 ( z 1 x 1 )
and
( z y ) = C k 2 ( z 2 y 2 ) , \begin{pmatrix} z \\ y \end{pmatrix} = \mathbf{C}^{k_2}\begin{pmatrix} z_2 \\ y_2 \end{pmatrix}, ( z y ) = C k 2 ( z 2 y 2 ) ,
in one of four parity cases. Let ϵ ∈ { − 1 , 1 } \epsilon\in\{-1,1\} ϵ ∈ { − 1 , 1 } be the sign of z 1 z_1 z 1 . By the one-step identities of Lemma 2.4, item 6, the four cases convert as follows.
Even/even, k 1 = 2 m 0 k_1 = 2m_0 k 1 = 2 m 0 , k 2 = 2 n 0 k_2 = 2n_0 k 2 = 2 n 0 : ( z 1 , x 1 ) = ( z 2 , y 2 ) = ( ϵ , 1 ) (z_1,x_1) = (z_2,y_2) = (\epsilon,1) ( z 1 , x 1 ) = ( z 2 , y 2 ) = ( ϵ , 1 ) . Take λ = ϵ \lambda= \epsilon λ = ϵ , m = m 0 m = m_0 m = m 0 , n = n 0 n = n_0 n = n 0 .
Odd/odd, k 1 = 2 m 0 + 1 k_1 = 2m_0 + 1 k 1 = 2 m 0 + 1 , k 2 = 2 n 0 + 1 k_2 = 2n_0 + 1 k 2 = 2 n 0 + 1 : ( z 1 , x 1 ) = ( ϵ t , r ) (z_1,x_1) = (\epsilon t,r) ( z 1 , x 1 ) = ( ϵ t , r ) , ( z 2 , y 2 ) = ( ϵ s , r ) (z_2,y_2) = (\epsilon s,r) ( z 2 , y 2 ) = ( ϵs , r ) . If ϵ = 1 \epsilon= 1 ϵ = 1 , then
( t r ) = B − 1 ( 1 1 ) and ( s r ) = C − 1 ( 1 1 ) \begin{pmatrix} t \\ r \end{pmatrix} = \mathbf{B}^{-1}\begin{pmatrix} 1 \\ 1 \end{pmatrix}
\quad\text{and}\quad
\begin{pmatrix} s \\ r \end{pmatrix} = \mathbf{C}^{-1}\begin{pmatrix} 1 \\ 1 \end{pmatrix} ( t r ) = B − 1 ( 1 1 ) and ( s r ) = C − 1 ( 1 1 )
give
( z x ) = B 2 m 0 + 2 ( − 1 1 ) , ( z y ) = C 2 n 0 + 2 ( − 1 1 ) ; \begin{pmatrix} z \\ x \end{pmatrix} = \mathbf{B}^{2m_0+2}\begin{pmatrix} -1 \\ 1 \end{pmatrix},
\qquad
\begin{pmatrix} z \\ y \end{pmatrix} = \mathbf{C}^{2n_0+2}\begin{pmatrix} -1 \\ 1 \end{pmatrix}; ( z x ) = B 2 m 0 + 2 ( − 1 1 ) , ( z y ) = C 2 n 0 + 2 ( − 1 1 ) ;
take λ = − 1 \lambda= -1 λ = − 1 , m = m 0 + 1 m = m_0 + 1 m = m 0 + 1 , n = n 0 + 1 n = n_0 + 1 n = n 0 + 1 . If ϵ = − 1 \epsilon= -1 ϵ = − 1 , then
( − t r ) = B − 1 ( 1 1 ) and ( − s r ) = C − 1 ( 1 1 ) \begin{pmatrix} -t \\ r \end{pmatrix} = \mathbf{B}^{-1}\begin{pmatrix} 1 \\ 1 \end{pmatrix}
\quad\text{and}\quad
\begin{pmatrix} -s \\ r \end{pmatrix} = \mathbf{C}^{-1}\begin{pmatrix} 1 \\ 1 \end{pmatrix} ( − t r ) = B − 1 ( 1 1 ) and ( − s r ) = C − 1 ( 1 1 )
give
( z x ) = B 2 m 0 ( 1 1 ) , ( z y ) = C 2 n 0 ( 1 1 ) ; \begin{pmatrix} z \\ x \end{pmatrix} = \mathbf{B}^{2m_0}\begin{pmatrix} 1 \\ 1 \end{pmatrix},
\qquad
\begin{pmatrix} z \\ y \end{pmatrix} = \mathbf{C}^{2n_0}\begin{pmatrix} 1 \\ 1 \end{pmatrix}; ( z x ) = B 2 m 0 ( 1 1 ) , ( z y ) = C 2 n 0 ( 1 1 ) ;
take λ = 1 \lambda= 1 λ = 1 , m = m 0 m = m_0 m = m 0 , n = n 0 n = n_0 n = n 0 .
Even/odd, k 1 = 2 m 0 k_1 = 2m_0 k 1 = 2 m 0 , k 2 = 2 n 0 + 1 k_2 = 2n_0 + 1 k 2 = 2 n 0 + 1 : ( z 1 , x 1 ) = ( ϵ , 1 ) (z_1,x_1) = (\epsilon,1) ( z 1 , x 1 ) = ( ϵ , 1 ) , ( z 2 , y 2 ) = ( − ϵ s , r ) (z_2,y_2) = (-\epsilon s,r) ( z 2 , y 2 ) = ( − ϵs , r ) . If ϵ = 1 \epsilon= 1 ϵ = 1 : λ = 1 \lambda= 1 λ = 1 , m = m 0 m = m_0 m = m 0 , n = n 0 n = n_0 n = n 0 . If ϵ = − 1 \epsilon= -1 ϵ = − 1 : λ = − 1 \lambda= -1 λ = − 1 , m = m 0 m = m_0 m = m 0 , n = n 0 + 1 n = n_0 + 1 n = n 0 + 1 .
Odd/even, k 1 = 2 m 0 + 1 k_1 = 2m_0 + 1 k 1 = 2 m 0 + 1 , k 2 = 2 n 0 k_2 = 2n_0 k 2 = 2 n 0 : ( z 1 , x 1 ) = ( ε t , r ) (z_1, x_1) = (\varepsilon t, r) ( z 1 , x 1 ) = ( εt , r ) , ( z 2 , y 2 ) = ( − ε , 1 ) (z_2, y_2) = (-\varepsilon, 1) ( z 2 , y 2 ) = ( − ε , 1 ) . If ε = 1 \varepsilon= 1 ε = 1 : λ = − 1 \lambda= -1 λ = − 1 , m = m 0 + 1 m = m_0 + 1 m = m 0 + 1 , n = n 0 n = n_0 n = n 0 . If ε = − 1 \varepsilon= -1 ε = − 1 : λ = 1 \lambda= 1 λ = 1 , m = m 0 m = m_0 m = m 0 , n = n 0 n = n_0 n = n 0 .
In every case
( z x ) = B 2 m ( λ 1 ) \binom{z}{x} = \mathbf{B}^{2m}\binom{\lambda}{1} ( x z ) = B 2 m ( 1 λ )
and
( z y ) = C 2 n ( λ 1 ) \binom{z}{y} = \mathbf{C}^{2n}\binom{\lambda}{1} ( y z ) = C 2 n ( 1 λ )
with the same λ \lambda λ and m , n ≥ 0 m,n \geq0 m , n ≥ 0 , which by item 2 of Lemma 2.8 are the second and third displays.
We next obtain the first display. The positive integers x , y x,y x , y satisfy a y 2 − b x 2 = a ( b d + 1 ) − b ( a d + 1 ) = a − b ay^2 - bx^2 = a(bd+1) - b(ad+1) = a-b a y 2 − b x 2 = a ( b d + 1 ) − b ( a d + 1 ) = a − b . By Lemma 2.10 with ( P , Q , S ) = ( a , b , r ) (P,Q,S) = (a,b,r) ( P , Q , S ) = ( a , b , r ) , so that η P Q = α \eta_{PQ} = \alpha η P Q = α , there are an integer k ≥ 0 k \geq0 k ≥ 0 and integers X 0 ≥ 1 X_0 \geq1 X 0 ≥ 1 , Y 0 Y_0 Y 0 with a Y 0 2 − b X 0 2 = a − b aY_0^2 - bX_0^2 = a-b a Y 0 2 − b X 0 2 = a − b , Y 0 2 < b 2 / 4 Y_0^2 < b^2/4 Y 0 2 < b 2 /4 and
y a + x b = ( Y 0 a + X 0 b ) α k . y\sqrt{a}+x\sqrt{b}=(Y_0\sqrt{a}+X_0\sqrt{b})\alpha^k. y a + x b = ( Y 0 a + X 0 b ) α k .
Write α k = p k + q k a b \alpha^k=p_k+q_k\sqrt{ab} α k = p k + q k ab . Then y = Y 0 p k + b X 0 q k y=Y_0p_k+bX_0q_k y = Y 0 p k + b X 0 q k , and in the binomial expansion of ( r + a b ) k (r+\sqrt{ab})^k ( r + ab ) k every term of p k p_k p k except r k r^k r k contains a positive power of a b ab ab ; so y ≡ Y 0 r k ( m o d b ) y\equiv Y_0r^k\pmod b y ≡ Y 0 r k ( mod b ) . On the other hand, writing γ 2 n = u + v b c \gamma^{2n}=u+v\sqrt{bc} γ 2 n = u + v b c , the third display gives y = u + λ b v y=u+\lambda bv y = u + λb v , and in the same way u ≡ t 2 n = ( b c + 1 ) n ≡ 1 ( m o d b ) u\equiv t^{2n}=(bc+1)^n\equiv1\pmod b u ≡ t 2 n = ( b c + 1 ) n ≡ 1 ( mod b ) . Hence Y 0 r k ≡ 1 ( m o d b ) Y_0r^k\equiv1\pmod b Y 0 r k ≡ 1 ( mod b ) , and since r 2 = a b + 1 ≡ 1 ( m o d b ) r^2=ab+1\equiv1\pmod b r 2 = ab + 1 ≡ 1 ( mod b ) (multiply by r r r when k k k is odd),
Y 0 ≡ 1 ( m o d b ) ( k even ) , Y 0 ≡ r ( m o d b ) ( k odd ) . Y_0\equiv1\pmod b\quad(k\text{ even}),\qquad Y_0\equiv r\pmod b\quad(k\text{ odd}). Y 0 ≡ 1 ( mod b ) ( k even ) , Y 0 ≡ r ( mod b ) ( k odd ) .
By Lemma 2.4 , item 4, b ≥ 3 b\geq3 b ≥ 3 and 1 ≤ r < b 1\leq r<b 1 ≤ r < b ; and the integers in ( − b 2 , b 2 ) \left(-\frac b2,\frac b2\right) ( − 2 b , 2 b ) are pairwise incongruent modulo b b b .
If k k k is even, Y 0 = 1 Y_0=1 Y 0 = 1 , hence b X 0 2 = b bX_0^2=b b X 0 2 = b and X 0 = 1 X_0=1 X 0 = 1 ; take l = k / 2 l=k/2 l = k /2 .
If k k k is odd, Y 0 ∈ { r , r − b } Y_0\in\{r,r-b\} Y 0 ∈ { r , r − b } . The value Y 0 = r Y_0=r Y 0 = r gives b X 0 2 = a r 2 − a + b = b ( a 2 + 1 ) bX_0^2=ar^2-a+b=b(a^2+1) b X 0 2 = a r 2 − a + b = b ( a 2 + 1 ) , impossible since a 2 < a 2 + 1 < ( a + 1 ) 2 a^2<a^2+1<(a+1)^2 a 2 < a 2 + 1 < ( a + 1 ) 2 . So Y 0 = r − b Y_0=r-b Y 0 = r − b , and b X 0 2 = a ( r − b ) 2 − a + b = b ( r − a ) 2 bX_0^2=a(r-b)^2-a+b=b(r-a)^2 b X 0 2 = a ( r − b ) 2 − a + b = b ( r − a ) 2 by a b + 1 = r 2 ab+1=r^2 ab + 1 = r 2 ; as r > a r>a r > a , X 0 = r − a X_0=r-a X 0 = r − a . Then
Y 0 a + X 0 b = ( r − b ) a + ( r − a ) b = ( a + b ) ( r − a b ) = ( a + b ) α − 1 , Y_0\sqrt{a}+X_0\sqrt{b}=(r-b)\sqrt{a}+(r-a)\sqrt{b}=(\sqrt{a}+\sqrt{b})(r-\sqrt{ab})=(\sqrt{a}+\sqrt{b})\alpha^{-1}, Y 0 a + X 0 b = ( r − b ) a + ( r − a ) b = ( a + b ) ( r − ab ) = ( a + b ) α − 1 ,
so the first display holds with l = ( k − 1 ) / 2 ≥ 0 l=(k-1)/2\geq0 l = ( k − 1 ) /2 ≥ 0 .
We finally prove uniqueness. Let ( λ 1 , l 1 , m 1 , n 1 ) (\lambda_1,l_1,m_1,n_1) ( λ 1 , l 1 , m 1 , n 1 ) and ( λ 2 , l 2 , m 2 , n 2 ) (\lambda_2,l_2,m_2,n_2) ( λ 2 , l 2 , m 2 , n 2 ) be index systems of d d d ; we show first that λ 1 = λ 2 \lambda_1=\lambda_2 λ 1 = λ 2 . By the second display, ( λ 1 a + c ) β 2 m 1 = ( λ 2 a + c ) β 2 m 2 (\lambda_1\sqrt a+\sqrt c)\beta^{2m_1}=(\lambda_2\sqrt a+\sqrt c)\beta^{2m_2} ( λ 1 a + c ) β 2 m 1 = ( λ 2 a + c ) β 2 m 2 . If λ 1 ≠ λ 2 \lambda_1\ne\lambda_2 λ 1 = λ 2 then, up to exchanging the two systems, λ 1 = 1 \lambda_1=1 λ 1 = 1 , λ 2 = − 1 \lambda_2=-1 λ 2 = − 1 , and
β 2 ( m 2 − m 1 ) = c + a c − a = c + a c − a + 2 c − a a c . \beta^{2(m_2-m_1)}=\frac{\sqrt c+\sqrt a}{\sqrt c-\sqrt a}=\frac{c+a}{c-a}+\frac{2}{c-a}\sqrt{ac}. β 2 ( m 2 − m 1 ) = c − a c + a = c − a c + a + c − a 2 a c .
The left side lies in Z [ a c ] \mathbb{Z}[\sqrt{ac}] Z [ a c ] , because β ± 1 = s ± a c \beta^{\pm1}=s\pm\sqrt{ac} β ± 1 = s ± a c (Lemma 2.4 , item 5), and a c \sqrt{ac} a c is irrational; so ( c − a ) ∣ 2 (c-a)\mid2 ( c − a ) ∣ 2 . But c > 4 a c>4a c > 4 a (Lemma 2.4 , item 4), so c − a > 3 a ≥ 3 c-a>3a\geq3 c − a > 3 a ≥ 3 , a contradiction.
With λ 1 = λ 2 \lambda_1=\lambda_2 λ 1 = λ 2 , after cancelling the factors λ 1 a + c \lambda_1\sqrt a+\sqrt c λ 1 a + c , λ 1 b + c \lambda_1\sqrt b+\sqrt c λ 1 b + c and a + b \sqrt a+\sqrt b a + b , which are nonzero as a < b < c a<b<c a < b < c , the second display gives β 2 ( m 1 − m 2 ) = 1 \beta^{2(m_1-m_2)}=1 β 2 ( m 1 − m 2 ) = 1 , so m 1 = m 2 m_1=m_2 m 1 = m 2 because β > 1 \beta>1 β > 1 . The third display gives γ 2 ( n 1 − n 2 ) = 1 \gamma^{2(n_1-n_2)}=1 γ 2 ( n 1 − n 2 ) = 1 , so n 1 = n 2 n_1=n_2 n 1 = n 2 , and the first gives α 2 ( l 1 − l 2 ) = 1 \alpha^{2(l_1-l_2)}=1 α 2 ( l 1 − l 2 ) = 1 , so l 1 = l 2 l_1=l_2 l 1 = l 2 .
We next identify the index system of the regular extension. It is used to exclude small indices in Lemma 2.14, and its common term V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 reappears in the finite computations of § 7 and Appendix § 8 .
Lemma 2.12. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple. Then d + ( a , b , c ) d_+(a,b,c) d + ( a , b , c ) is an extension of { a , b , c } \{a,b,c\} { a , b , c } , and its index system is ( − 1 , 1 , 1 , 1 ) (-1,1,1,1) ( − 1 , 1 , 1 , 1 ) . In particular V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 , and this common term is the number z z z of d + ( a , b , c ) d_+(a,b,c) d + ( a , b , c ) .
Proof. Let r , s , t r,s,t r , s , t be the roots of { a , b , c } \{a,b,c\} { a , b , c } and d = d + ( a , b , c ) d=d_+(a,b,c) d = d + ( a , b , c ) . By Lemma 2.4, item 2, d = 4 r s t d=4rst d = 4 r s t and s t = c r − 1 st=cr-1 s t = cr − 1 , so d = 4 r ( c r − 1 ) > c d=4r(cr-1)>c d = 4 r ( cr − 1 ) > c , as r > a ≥ 1 r>a\ge1 r > a ≥ 1 (item 4 of Lemma 2.4). By Definition 2.3 and items 2 and 6 of Lemma 2.4 (r s − a t = r t − b s = 1 rs-at=rt-bs=1 r s − a t = r t − b s = 1 , B ( 1 − 1 ) = ( t r ) \mathbf{B}\binom{1}{-1}=\binom{t}{r} B ( − 1 1 ) = ( r t ) , C ( − 1 1 ) = ( s r ) \mathbf{C}\binom{-1}{1}=\binom{s}{r} C ( 1 − 1 ) = ( r s ) ),
A 2 ( 1 1 ) = A ( t s ) = ( 1 + 2 b s 1 + 2 a t ) , B 2 ( − 1 1 ) = B ( t r ) = ( c r + s t 1 + 2 a t ) , \mathbf{A}^{2}\binom{1}{1}=\mathbf{A}\binom{t}{s}=\binom{1+2bs}{1+2at},\qquad
\mathbf{B}^{2}\binom{-1}{1}=\mathbf{B}\binom{t}{r}=\binom{cr+st}{1+2at}, A 2 ( 1 1 ) = A ( s t ) = ( 1 + 2 a t 1 + 2 b s ) , B 2 ( 1 − 1 ) = B ( r t ) = ( 1 + 2 a t cr + s t ) ,
C 2 ( − 1 1 ) = C ( s r ) = ( c r + s t 1 + 2 b s ) . \mathbf{C}^{2}\binom{-1}{1}=\mathbf{C}\binom{s}{r}=\binom{cr+st}{1+2bs}. C 2 ( 1 − 1 ) = C ( r s ) = ( 1 + 2 b s cr + s t ) .
Put x = 1 + 2 a t x=1+2at x = 1 + 2 a t , y = 1 + 2 b s y=1+2bs y = 1 + 2 b s and z = c r + s t z=cr+st z = cr + s t . Since r s = 1 + a t rs=1+at r s = 1 + a t , r t = 1 + b s rt=1+bs r t = 1 + b s and c r − s t = 1 cr-st=1 cr − s t = 1 ,
x 2 = 1 + 4 a t r s = a d + 1 , y 2 = 1 + 4 b s r t = b d + 1 , z 2 = ( c r − s t ) 2 + 4 c r s t = c d + 1 , x^2=1+4atrs=ad+1,\qquad y^2=1+4bsrt=bd+1,\qquad z^2=(cr-st)^2+4crst=cd+1, x 2 = 1 + 4 a t r s = a d + 1 , y 2 = 1 + 4 b sr t = b d + 1 , z 2 = ( cr − s t ) 2 + 4 cr s t = c d + 1 ,
so, by item 1 of Lemma 2.4, d d d is an extension of { a , b , c } \{a,b,c\} { a , b , c } and x , y , z x,y,z x , y , z are its numbers of Definition 2.5. By item 2 of Lemma 2.8, the three identities above are the three displays of Definition 2.5 for ( λ , l , m , n ) = ( − 1 , 1 , 1 , 1 ) (\lambda,l,m,n)=(-1,1,1,1) ( λ , l , m , n ) = ( − 1 , 1 , 1 , 1 ) , which by Proposition 2.11 is therefore the index system of d d d . Finally z = V 1 − 1 = W 1 − 1 z=V_1^{-1}=W_1^{-1} z = V 1 − 1 = W 1 − 1 by Convention 2.6.
The index ordering and the two linear forms
For an irregular extension the indices are ordered, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n , and the triple is not small, b > 4000 b>4000 b > 4000 ; the second fact is a result of Cipu, Filipin and Fujita [8 ] , which we quote.
Theorem 2.13. Let { a , b , c } \{a,b,c\} { a , b , c } be a Diophantine triple with a < b < c a<b<c a < b < c , and let d d d be an extension of { a , b , c } \{a,b,c\} { a , b , c } with d > d + ( a , b , c ) d>d_+(a,b,c) d > d + ( a , b , c ) . Then b > 4000 b>4000 b > 4000 .
Proof. This is a result of Cipu, Filipin and Fujita [8 ] , in the form quoted in [18 Lemma 2.1] : if a < b < c < d + ( a , b , c ) < d a<b<c<d_+(a,b,c)<d a < b < c < d + ( a , b , c ) < d , then b > 21000 b>21000 b > 21000 when b < 2 a b<2a b < 2 a , b > 130000 b>130000 b > 130000 when 2 a ≤ b ≤ 12 a 2a\le b\le12a 2 a ≤ b ≤ 12 a , and b > 4000 b>4000 b > 4000 when b > 12 a b>12a b > 12 a . The hypothesis c < d + ( a , b , c ) c<d_+(a,b,c) c < d + ( a , b , c ) holds because d + ( a , b , c ) > ( 2 a b + 1 ) c d_+(a,b,c)>(2ab+1)c d + ( a , b , c ) > ( 2 ab + 1 ) c (Definition 2.2).
Lemma 2.14. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then
3 ≤ n < m < l < 2 n , m ≥ 4 , l ≥ 5 , 3\le n<m<l<2n,\qquad m\ge4,\qquad l\ge5, 3 ≤ n < m < l < 2 n , m ≥ 4 , l ≥ 5 , and d > d + ( a , b , c ) d>d_+(a,b,c) d > d + ( a , b , c ) and b > 4000 b>4000 b > 4000 .
Proof. Each of the inequalities n < m n<m n < m , m < l m<l m < l and l < 2 n l<2n l < 2 n compares the same coordinate, z z z , x x x or y y y , along two sequences built from the three displays of Definition 2.5. The two sequences satisfy second-order recurrences with different coefficients; the one with the larger coefficient stays ahead (the comparison principle below), so it reaches the common value at the smaller index.
We first set up the sequences. By items 1 and 5 of Lemma 2.4, s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 , t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 , and multiplication by α , β , γ \alpha,\beta,\gamma α , β , γ acts on coordinates through A , B , C \mathbf{A},\mathbf{B},\mathbf{C} A , B , C . For k ≥ 0 k\ge0 k ≥ 0 write A k ( 1 1 ) = ( u k v k ) \mathbf{A}^{k}\binom{1}{1}=\binom{u_k}{v_k} A k ( 1 1 ) = ( v k u k ) , i.e. ( a + b ) α k = u k a + v k b (\sqrt{a}+\sqrt{b})\alpha^k=u_k\sqrt{a}+v_k\sqrt{b} ( a + b ) α k = u k a + v k b ; then u 0 = v 0 = 1 u_0=v_0=1 u 0 = v 0 = 1 , u k + 1 = r u k + b v k u_{k+1}=ru_k+bv_k u k + 1 = r u k + b v k , v k + 1 = a u k + r v k v_{k+1}=au_k+rv_k v k + 1 = a u k + r v k , so u k , v k u_k,v_k u k , v k are positive and strictly increasing in k k k . For j ≥ 0 j\ge0 j ≥ 0 write
B 2 j ( λ 1 ) = ( V j λ ξ j ) , C 2 j ( λ 1 ) = ( W j λ ψ j ) , \mathbf{B}^{2j}\binom{\lambda}{1}=\binom{V_j^\lambda}{\xi_j},\qquad
\mathbf{C}^{2j}\binom{\lambda}{1}=\binom{W_j^\lambda}{\psi_j}, B 2 j ( 1 λ ) = ( ξ j V j λ ) , C 2 j ( 1 λ ) = ( ψ j W j λ ) ,
with the Pell sequences V j λ V_j^\lambda V j λ , W j λ W_j^\lambda W j λ of Convention 2.6. By item 2 of Lemma 2.8 , the three displays say
x = v 2 l = ξ m , y = u 2 l = ψ n , z = V m λ = W n λ . x=v_{2l}=\xi_m,\qquad y=u_{2l}=\psi_n,\qquad z=V_m^\lambda=W_n^\lambda. x = v 2 l = ξ m , y = u 2 l = ψ n , z = V m λ = W n λ .
By item 3 of Lemma 2.8 , the sequences ( v 2 j ) j (v_{2j})_j ( v 2 j ) j , ( V j λ ) j (V_j^\lambda)_j ( V j λ ) j and ( ξ j ) j (\xi_j)_j ( ξ j ) j , ( W j λ ) j (W_j^\lambda)_j ( W j λ ) j and ( ψ j ) j (\psi_j)_j ( ψ j ) j satisfy X j + 1 = T X j − X j − 1 X_{j+1}=TX_j-X_{j-1} X j + 1 = T X j − X j − 1 (j ≥ 1 j\ge1 j ≥ 1 ) with T = 4 a b + 2 T=4ab+2 T = 4 ab + 2 , 4 a c + 2 4ac+2 4 a c + 2 , 4 b c + 2 4bc+2 4 b c + 2 respectively; and ( u 4 j ) j (u_{4j})_j ( u 4 j ) j satisfies it with T = T α : = α 4 + α − 4 = 2 ( 8 r 4 − 8 r 2 + 1 ) T=T_\alpha:=\alpha^4+\alpha^{-4}=2(8r^4-8r^2+1) T = T α := α 4 + α − 4 = 2 ( 8 r 4 − 8 r 2 + 1 ) , since α 4 \alpha^4 α 4 and α − 4 \alpha^{-4} α − 4 are the roots of X 2 − T α X + 1 X^2-T_\alpha X+1 X 2 − T α X + 1 . The initial values of ( V j λ ) (V_j^\lambda) ( V j λ ) and ( W j λ ) (W_j^\lambda) ( W j λ ) are given in item 3 of Lemma 2.8 ; those of the other sequences are v 0 = ξ 0 = ψ 0 = u 0 = 1 v_0=\xi_0=\psi_0=u_0=1 v 0 = ξ 0 = ψ 0 = u 0 = 1 and
v 2 = 1 + 2 a ( b + r ) , ξ 1 = 1 + 2 a ( c + λ s ) , v_2=1+2a(b+r),\qquad\xi_1=1+2a(c+\lambda s), v 2 = 1 + 2 a ( b + r ) , ξ 1 = 1 + 2 a ( c + λ s ) ,
ψ 1 = 1 + 2 b ( c + λ t ) , u 4 = 8 r 4 − 8 r 2 + 1 + 4 b r ( 2 r 2 − 1 ) . \psi_1=1+2b(c+\lambda t),\qquad u_4=8r^4-8r^2+1+4br(2r^2-1). ψ 1 = 1 + 2 b ( c + λ t ) , u 4 = 8 r 4 − 8 r 2 + 1 + 4 b r ( 2 r 2 − 1 ) .
We use the following comparison principle. Let X j + 1 = T 1 X j − X j − 1 X_{j+1}=T_1X_j-X_{j-1} X j + 1 = T 1 X j − X j − 1 and Y j + 1 = T 2 Y j − Y j − 1 Y_{j+1}=T_2Y_j-Y_{j-1} Y j + 1 = T 2 Y j − Y j − 1 (j ≥ 1 j\ge1 j ≥ 1 ) with T 1 > T 2 ≥ 2 T_1>T_2\ge2 T 1 > T 2 ≥ 2 , and let j 0 ≥ 1 j_0\ge1 j 0 ≥ 1 be such that δ j = X j − Y j \delta_j=X_j-Y_j δ j = X j − Y j satisfies δ j 0 > δ j 0 − 1 ≥ 0 \delta_{j_0}>\delta_{j_0-1}\ge0 δ j 0 > δ j 0 − 1 ≥ 0 and X j > 0 X_j>0 X j > 0 for j ≥ j 0 j\ge j_0 j ≥ j 0 . Then X j > Y j X_j>Y_j X j > Y j for all j ≥ j 0 j\ge j_0 j ≥ j 0 . Indeed δ j + 1 − δ j = ( T 2 − 2 ) δ j + ( δ j − δ j − 1 ) + ( T 1 − T 2 ) X j \delta_{j+1}-\delta_j=(T_2-2)\delta_j+(\delta_j-\delta_{j-1})+(T_1-T_2)X_j δ j + 1 − δ j = ( T 2 − 2 ) δ j + ( δ j − δ j − 1 ) + ( T 1 − T 2 ) X j , so by induction δ j + 1 > δ j > 0 \delta_{j+1}>\delta_j>0 δ j + 1 > δ j > 0 for j ≥ j 0 j\ge j_0 j ≥ j 0 . By item 3 of Lemma 2.8 , the sequences ( ξ j ) (\xi_j) ( ξ j ) , ( V j λ ) (V_j^\lambda) ( V j λ ) , ( W j λ ) (W_j^\lambda) ( W j λ ) , ( ψ j ) (\psi_j) ( ψ j ) are strictly increasing and positive for j ≥ 1 j\ge1 j ≥ 1 .
We next exclude small indices. If m = 0 m=0 m = 0 or n = 0 n=0 n = 0 then z = λ z=\lambda z = λ and c d = z 2 − 1 = 0 cd=z^2-1=0 c d = z 2 − 1 = 0 , impossible. By Lemma 2.12 , V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 is the number z + z_+ z + of the regular extension. If λ = − 1 \lambda=-1 λ = − 1 and m = 1 m=1 m = 1 or n = 1 n=1 n = 1 , then z = z + z=z_+ z = z + , so d = ( z 2 − 1 ) / c = d + ( a , b , c ) d=(z^2-1)/c=d_+(a,b,c) d = ( z 2 − 1 ) / c = d + ( a , b , c ) , contradicting irregularity. Hence m , n ≥ 1 m,n\ge1 m , n ≥ 1 , and m , n ≥ 2 m,n\ge2 m , n ≥ 2 when λ = − 1 \lambda=-1 λ = − 1 . Put j 0 = 1 j_0=1 j 0 = 1 if λ = 1 \lambda=1 λ = 1 and j 0 = 2 j_0=2 j 0 = 2 if λ = − 1 \lambda=-1 λ = − 1 ; then m , n ≥ j 0 m,n\ge j_0 m , n ≥ j 0 .
We next prove m > n m>n m > n . Here W 1 λ − V 1 λ = 2 c ( b − a ) ( 1 + λ ) W_1^\lambda-V_1^\lambda=2c(b-a)(1+\lambda) W 1 λ − V 1 λ = 2 c ( b − a ) ( 1 + λ ) because t − s = b − a t-s=b-a t − s = b − a . If λ = 1 \lambda=1 λ = 1 , W 1 λ − V 1 λ = 4 c ( b − a ) > 0 = W 0 λ − V 0 λ W_1^\lambda-V_1^\lambda=4c(b-a)>0=W_0^\lambda-V_0^\lambda W 1 λ − V 1 λ = 4 c ( b − a ) > 0 = W 0 λ − V 0 λ . If λ = − 1 \lambda=-1 λ = − 1 , W 1 λ = V 1 λ W_1^\lambda=V_1^\lambda W 1 λ = V 1 λ and W 2 λ − V 2 λ = 4 ( b − a ) c W 1 λ > 0 W_2^\lambda-V_2^\lambda=4(b-a)cW_1^\lambda>0 W 2 λ − V 2 λ = 4 ( b − a ) c W 1 λ > 0 . The comparison principle with T 1 = 4 b c + 2 > T 2 = 4 a c + 2 T_1=4bc+2>T_2=4ac+2 T 1 = 4 b c + 2 > T 2 = 4 a c + 2 gives W j λ > V j λ W_j^\lambda>V_j^\lambda W j λ > V j λ for j ≥ j 0 j\ge j_0 j ≥ j 0 , in particular W n λ > V n λ W_n^\lambda>V_n^\lambda W n λ > V n λ . If m ≤ n m\le n m ≤ n , then z = V m λ ≤ V n λ < W n λ = z z=V_m^\lambda\le V_n^\lambda<W_n^\lambda=z z = V m λ ≤ V n λ < W n λ = z , since ( V j λ ) (V_j^\lambda) ( V j λ ) is increasing. So m > n m>n m > n .
We next prove l > m l>m l > m . Here ξ 1 − v 2 = 2 a ( c + λ s − b − r ) = 2 a s ( 1 + λ ) \xi_1-v_2=2a(c+\lambda s-b-r)=2as(1+\lambda) ξ 1 − v 2 = 2 a ( c + λ s − b − r ) = 2 a s ( 1 + λ ) . If λ = 1 \lambda=1 λ = 1 this is positive; if λ = − 1 \lambda=-1 λ = − 1 , ξ 1 = v 2 \xi_1=v_2 ξ 1 = v 2 and ξ 2 − v 4 = 4 a ( c − b ) v 2 > 0 \xi_2-v_4=4a(c-b)v_2>0 ξ 2 − v 4 = 4 a ( c − b ) v 2 > 0 . The comparison principle with T 1 = 4 a c + 2 > T 2 = 4 a b + 2 T_1=4ac+2>T_2=4ab+2 T 1 = 4 a c + 2 > T 2 = 4 ab + 2 gives ξ j > v 2 j \xi_j>v_{2j} ξ j > v 2 j for j ≥ j 0 j\ge j_0 j ≥ j 0 , in particular ξ m > v 2 m \xi_m>v_{2m} ξ m > v 2 m . If l ≤ m l\le m l ≤ m , then x = v 2 l ≤ v 2 m < ξ m = x x=v_{2l}\le v_{2m}<\xi_m=x x = v 2 l ≤ v 2 m < ξ m = x . So l > m l>m l > m .
We next prove l < 2 n l<2n l < 2 n . First T α > 4 b c + 2 T_\alpha>4bc+2 T α > 4 b c + 2 : this is 16 r 2 ( r 2 − 1 ) > 4 b c 16r^2(r^2-1)>4bc 16 r 2 ( r 2 − 1 ) > 4 b c , i.e. 4 a r 2 > c 4ar^2>c 4 a r 2 > c , true as 4 a r 2 > 4 b > a + b + 2 r = c 4ar^2>4b>a+b+2r=c 4 a r 2 > 4 b > a + b + 2 r = c (by a < r < b a<r<b a < r < b , Lemma 2.4 , item 4). Second u 4 > ψ 1 u_4>\psi_1 u 4 > ψ 1 : since ψ 1 ≤ 1 + 2 b ( c + t ) = 1 + 2 b ( a + 2 b + 3 r ) \psi_1\le1+2b(c+t)=1+2b(a+2b+3r) ψ 1 ≤ 1 + 2 b ( c + t ) = 1 + 2 b ( a + 2 b + 3 r ) ,
u 4 − ψ 1 ≥ 8 r 2 ( r 2 − 1 ) + 2 b ( r ( 4 r 2 − 5 ) − a − 2 b ) > 0 , u_4-\psi_1\ge8r^2(r^2-1)+2b(r(4r^2-5)-a-2b)>0, u 4 − ψ 1 ≥ 8 r 2 ( r 2 − 1 ) + 2 b ( r ( 4 r 2 − 5 ) − a − 2 b ) > 0 ,
because r ( 4 r 2 − 5 ) = r ( 4 a b − 1 ) ≥ 2 ( 4 b − 1 ) > a + 2 b r(4r^2-5)=r(4ab-1)\ge2(4b-1)>a+2b r ( 4 r 2 − 5 ) = r ( 4 ab − 1 ) ≥ 2 ( 4 b − 1 ) > a + 2 b . With u 0 = ψ 0 u_0=\psi_0 u 0 = ψ 0 , the comparison principle with T 1 = T α > T 2 = 4 b c + 2 T_1=T_\alpha>T_2=4bc+2 T 1 = T α > T 2 = 4 b c + 2 and j 0 = 1 j_0=1 j 0 = 1 gives u 4 j > ψ j u_{4j}>\psi_j u 4 j > ψ j for j ≥ 1 j\ge1 j ≥ 1 . If l ≥ 2 n l\ge2n l ≥ 2 n , then y = u 2 l ≥ u 4 n > ψ n = y y=u_{2l}\ge u_{4n}>\psi_n=y y = u 2 l ≥ u 4 n > ψ n = y . So l < 2 n l<2n l < 2 n .
Next, n < m < l < 2 n n<m<l<2n n < m < l < 2 n places two distinct integers strictly between n n n and 2 n 2n 2 n , so n ≥ 3 n\ge3 n ≥ 3 , m ≥ n + 1 ≥ 4 m\ge n+1\ge4 m ≥ n + 1 ≥ 4 and l ≥ m + 1 ≥ 5 l\ge m+1\ge5 l ≥ m + 1 ≥ 5 .
We finally prove d > d + ( a , b , c ) d>d_+(a,b,c) d > d + ( a , b , c ) and b > 4000 b>4000 b > 4000 . Since ( V j λ ) (V_j^\lambda) ( V j λ ) is increasing, m ≥ 2 m\ge2 m ≥ 2 and V 1 1 − V 1 − 1 = 2 ( 2 a c + 1 ) > 0 V_1^1-V_1^{-1}=2(2ac+1)>0 V 1 1 − V 1 − 1 = 2 ( 2 a c + 1 ) > 0 , we have z = V m λ > V 1 λ > V 1 − 1 = z + z=V_m^\lambda>V_1^\lambda>V_1^{-1}=z_+ z = V m λ > V 1 λ > V 1 − 1 = z + . Hence d = ( z 2 − 1 ) / c > ( z + 2 − 1 ) / c = d + ( a , b , c ) d=(z^2-1)/c>(z_+^2-1)/c=d_+(a,b,c) d = ( z 2 − 1 ) / c > ( z + 2 − 1 ) / c = d + ( a , b , c ) . By Theorem 2.13 , applied to the Diophantine triple { a , b , c } \{a,b,c\} { a , b , c } (Lemma 2.4 , item 1), b > 4000 b>4000 b > 4000 .
We now turn to the two linear forms. Each of x x x and z z z is a coordinate of two of the three representations, and the closed forms below write it in two ways.
Lemma 2.15. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be an index system of an extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } with units α , β , γ \alpha,\beta,\gamma α , β , γ , and let x , y , z x,y,z x , y , z be as in Definition 2.5. Then
2 x = ( 1 + a b ) α 2 l + ( 1 − a b ) α − 2 l = ( 1 + λ a c ) β 2 m + ( 1 − λ a c ) β − 2 m , 2x=\left(1+\sqrt{\frac{a}{b}}\right)\alpha^{2l}+\left(1-\sqrt{\frac{a}{b}}\right)\alpha^{-2l}=\left(1+\lambda\sqrt{\frac{a}{c}}\right)\beta^{2m}+\left(1-\lambda\sqrt{\frac{a}{c}}\right)\beta^{-2m}, 2 x = ( 1 + b a ) α 2 l + ( 1 − b a ) α − 2 l = ( 1 + λ c a ) β 2 m + ( 1 − λ c a ) β − 2 m , 2 z = ( c a + λ ) β 2 m + ( λ − c a ) β − 2 m = ( c b + λ ) γ 2 n + ( λ − c b ) γ − 2 n . 2z=\left(\sqrt{\frac{c}{a}}+\lambda\right)\beta^{2m}+\left(\lambda-\sqrt{\frac{c}{a}}\right)\beta^{-2m}=\left(\sqrt{\frac{c}{b}}+\lambda\right)\gamma^{2n}+\left(\lambda-\sqrt{\frac{c}{b}}\right)\gamma^{-2n}. 2 z = ( a c + λ ) β 2 m + ( λ − a c ) β − 2 m = ( b c + λ ) γ 2 n + ( λ − b c ) γ − 2 n . Proof. By Lemma 2.4, item 5, ( u a + v b ) α = u 1 a + v 1 b (u\sqrt{a}+v\sqrt{b})\alpha=u_1\sqrt{a}+v_1\sqrt{b} ( u a + v b ) α = u 1 a + v 1 b with ( u 1 v 1 ) = A ( u v ) \binom{u_1}{v_1}=\mathbf{A}\binom{u}{v} ( v 1 u 1 ) = A ( v u ) for integers u , v u,v u , v ; since α − 1 = r − a b \alpha^{-1}=r-\sqrt{ab} α − 1 = r − ab , the same expansion gives ( u a − v b ) α − 1 = u 1 a − v 1 b (u\sqrt{a}-v\sqrt{b})\alpha^{-1}=u_1\sqrt{a}-v_1\sqrt{b} ( u a − v b ) α − 1 = u 1 a − v 1 b . By induction on k ≥ 0 k\geq0 k ≥ 0 , for integers u 0 , v 0 u_0,v_0 u 0 , v 0 the identity ( u 0 a + v 0 b ) α k = u a + v b (u_0\sqrt{a}+v_0\sqrt{b})\alpha^k=u\sqrt{a}+v\sqrt{b} ( u 0 a + v 0 b ) α k = u a + v b implies ( u 0 a − v 0 b ) α − k = u a − v b (u_0\sqrt{a}-v_0\sqrt{b})\alpha^{-k}=u\sqrt{a}-v\sqrt{b} ( u 0 a − v 0 b ) α − k = u a − v b , the coordinates being unique because a b ab ab is not a square. The same holds for β \beta β with a , c \sqrt{a},\sqrt{c} a , c and B \mathbf{B} B , and for γ \gamma γ with b , c \sqrt{b},\sqrt{c} b , c and C \mathbf{C} C . So the three displays of Definition 2.5 imply
y a − x b = ( a − b ) α − 2 l , z a − x c = ( λ a − c ) β − 2 m , y\sqrt{a}-x\sqrt{b}=(\sqrt{a}-\sqrt{b})\alpha^{-2l},\qquad z\sqrt{a}-x\sqrt{c}=(\lambda\sqrt{a}-\sqrt{c})\beta^{-2m}, y a − x b = ( a − b ) α − 2 l , z a − x c = ( λ a − c ) β − 2 m , z b − y c = ( λ b − c ) γ − 2 n . z\sqrt{b}-y\sqrt{c}=(\lambda\sqrt{b}-\sqrt{c})\gamma^{-2n}. z b − y c = ( λ b − c ) γ − 2 n . Subtracting the first two of these identities from the corresponding displays and dividing by b \sqrt{b} b and by c \sqrt{c} c gives the two expressions of 2 x 2x 2 x ; adding the last two to the corresponding displays and dividing by a \sqrt{a} a and by b \sqrt{b} b gives the two expressions of 2 z 2z 2 z .
Definition 2.16. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with units α , β , γ \alpha,\beta,\gamma α , β , γ (Definition 2.3). For λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } put
D λ = b c + a c b c + λ a b , E λ = b c + λ a b a c + λ a b , κ λ = 2 c ( b − a ) b ( c + λ a ) 2 . D_\lambda=\frac{\sqrt{bc}+\sqrt{ac}}{\sqrt{bc}+\lambda\sqrt{ab}},\qquad E_\lambda=\frac{\sqrt{bc}+\lambda\sqrt{ab}}{\sqrt{ac}+\lambda\sqrt{ab}},\qquad\kappa_\lambda=\frac{2c(b-a)}{b(\sqrt{c}+\lambda\sqrt{a})^2}. D λ = b c + λ ab b c + a c , E λ = a c + λ ab b c + λ ab , κ λ = b ( c + λ a ) 2 2 c ( b − a ) . The linear forms of an index system ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) (Definition 2.5) of an extension of { a , b , c } \{a,b,c\} { a , b , c } are
Λ x = 2 l log α − 2 m log β + log D λ , Λ z = 2 m log β − 2 n log γ + log E λ , \Lambda_x=2l\log\alpha-2m\log\beta+\log D_\lambda,\qquad\Lambda_z=2m\log\beta-2n\log\gamma+\log E_\lambda, Λ x = 2 l log α − 2 m log β + log D λ , Λ z = 2 m log β − 2 n log γ + log E λ , and the gap ratios of { a , b , c } \{a,b,c\} { a , b , c } are
R x = b a log α , R z = a c b − a log β . R_x=\sqrt{\frac{b}{a}}\log\alpha,\qquad R_z=\frac{\sqrt{ac}}{b-a}\log\beta. R x = a b log α , R z = b − a a c log β . The numbers D λ D_\lambda D λ and E λ E_\lambda E λ are the ratios of the leading coefficients of the two expressions of x x x and of z z z in Lemma 2.15, so that Λ x \Lambda_x Λ x and Λ z \Lambda_z Λ z are the logarithms of the ratios of the leading terms. The number κ λ \kappa_\lambda κ λ is the constant of the bound Λ z < κ λ β − 4 m \Lambda_z<\kappa_\lambda\beta^{-4m} Λ z < κ λ β − 4 m of Lemma 2.18, item 5, which is the form in which the reduction of Appendix B uses Λ z \Lambda_z Λ z ; the gap ratios are the factors of the gap inequalities, item 3 there. The two comparisons are one argument, which we prove once, for two abstract expressions; it is the one technical step of this section.
Lemma 2.17. Let μ 1 > μ 2 > 0 \mu_1>\mu_2>0 μ 1 > μ 2 > 0 , ν 1 \nu_1 ν 1 and ν 2 \nu_2 ν 2 be real numbers with ν 1 < μ 2 \nu_1<\mu_2 ν 1 < μ 2 and μ 1 + ν 1 = μ 2 + ν 2 \mu_1+\nu_1=\mu_2+\nu_2 μ 1 + ν 1 = μ 2 + ν 2 , and put e i = μ i ν i e_i=\mu_i\nu_i e i = μ i ν i for i = 1 , 2 i=1,2 i = 1 , 2 . Let u 1 ≥ 1 u_1\geq1 u 1 ≥ 1 and u 2 > 1 u_2>1 u 2 > 1 be real numbers with
μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 , \mu_1u_1+\nu_1u_1^{-1}=\mu_2u_2+\nu_2u_2^{-1}, μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 , and put X i = μ i u i X_i=\mu_i u_i X i = μ i u i . Then e 1 < e 2 e_1<e_2 e 1 < e 2 and X 2 2 > e 2 X_2^2>e_2 X 2 2 > e 2 , and:
u 1 < u 2 u_1 < u_2 u 1 < u 2 and X 1 > X 2 X_1 > X_2 X 1 > X 2 , that is, 0 < log ( X 1 / X 2 ) < log ( μ 1 / μ 2 ) 0 < \log(X_1/X_2) < \log(\mu_1/\mu_2) 0 < log ( X 1 / X 2 ) < log ( μ 1 / μ 2 ) ;
log ( X 1 / X 2 ) < e 2 − e 1 X 2 2 − e 2 \log(X_1/X_2) < \frac{e_2-e_1}{X_2^2-e_2} log ( X 1 / X 2 ) < X 2 2 − e 2 e 2 − e 1 ;
if e 2 ≤ 0 e_2 \le0 e 2 ≤ 0 and 2 ( e 2 − e 1 ) ≤ X 1 2 2(e_2-e_1) \le X_1^2 2 ( e 2 − e 1 ) ≤ X 1 2 , then log ( X 1 / X 2 ) < 2 ( e 2 − e 1 ) X 1 2 \log(X_1/X_2) < \frac{2(e_2-e_1)}{X_1^2} log ( X 1 / X 2 ) < X 1 2 2 ( e 2 − e 1 ) .
Proof. We compare the two expressions, which agree at u 1 = u 2 = 1 u_1=u_2=1 u 1 = u 2 = 1 , in three ways: at equal arguments the first is the larger, which gives the upper side of item 1; at equal leading terms it is the smaller, which gives the lower side; and one exact identity between the leading terms gives the size of log ( X 1 / X 2 ) \log(X_1/X_2) log ( X 1 / X 2 ) , items 2 and 3.
Put σ = μ 1 + ν 1 = μ 2 + ν 2 \sigma=\mu_1+\nu_1=\mu_2+\nu_2 σ = μ 1 + ν 1 = μ 2 + ν 2 , and, for i = 1 , 2 i=1,2 i = 1 , 2 , a real number e e e and real numbers u , X > 0 u,X>0 u , X > 0 ,
F i ( u ) = μ i u + ν i u − 1 , h e ( X ) = X + e X − 1 . F_i(u)=\mu_i u+\nu_i u^{-1},\qquad h_e(X)=X+eX^{-1}. F i ( u ) = μ i u + ν i u − 1 , h e ( X ) = X + e X − 1 .
Then F i ( u ) = h e i ( μ i u ) F_i(u)=h_{e_i}(\mu_i u) F i ( u ) = h e i ( μ i u ) , because e i ( μ i u ) − 1 = ν i u − 1 e_i(\mu_i u)^{-1}=\nu_i u^{-1} e i ( μ i u ) − 1 = ν i u − 1 ; moreover F 1 ( 1 ) = F 2 ( 1 ) = σ F_1(1)=F_2(1)=\sigma F 1 ( 1 ) = F 2 ( 1 ) = σ , and the hypothesis reads F 1 ( u 1 ) = F 2 ( u 2 ) F_1(u_1)=F_2(u_2) F 1 ( u 1 ) = F 2 ( u 2 ) , that is, h e 1 ( X 1 ) = h e 2 ( X 2 ) h_{e_1}(X_1)=h_{e_2}(X_2) h e 1 ( X 1 ) = h e 2 ( X 2 ) . Since μ 1 , μ 2 > 0 \mu_1,\mu_2>0 μ 1 , μ 2 > 0 and u 1 , u 2 ≥ 1 u_1,u_2\ge1 u 1 , u 2 ≥ 1 , we have X 1 , X 2 > 0 X_1,X_2>0 X 1 , X 2 > 0 .
We first prove e 1 < e 2 e_1<e_2 e 1 < e 2 . Since ν i = σ − μ i \nu_i=\sigma-\mu_i ν i = σ − μ i , one has e i = σ μ i − μ i 2 e_i=\sigma\mu_i-\mu_i^2 e i = σ μ i − μ i 2 , so
e 2 − e 1 = ( μ 1 2 − μ 2 2 ) − σ ( μ 1 − μ 2 ) = ( μ 1 − μ 2 ) ( μ 1 + μ 2 − σ ) = ( μ 1 − μ 2 ) ( μ 2 − ν 1 ) , e_2-e_1=(\mu_1^2-\mu_2^2)-\sigma(\mu_1-\mu_2)=(\mu_1-\mu_2)(\mu_1+\mu_2-\sigma)=(\mu_1-\mu_2)(\mu_2-\nu_1), e 2 − e 1 = ( μ 1 2 − μ 2 2 ) − σ ( μ 1 − μ 2 ) = ( μ 1 − μ 2 ) ( μ 1 + μ 2 − σ ) = ( μ 1 − μ 2 ) ( μ 2 − ν 1 ) ,
and both factors are positive, as μ 1 > μ 2 \mu_1>\mu_2 μ 1 > μ 2 and ν 1 < μ 2 \nu_1<\mu_2 ν 1 < μ 2 .
We next prove u 1 < u 2 u_1<u_2 u 1 < u 2 . Since ν 1 − ν 2 = μ 2 − μ 1 \nu_1-\nu_2=\mu_2-\mu_1 ν 1 − ν 2 = μ 2 − μ 1 ,
F 1 ( u ) − F 2 ( u ) = ( μ 1 − μ 2 ) ( u − u − 1 ) > 0 ( u > 1 ) . F_1(u)-F_2(u)=(\mu_1-\mu_2)(u-u^{-1})>0\qquad(u>1). F 1 ( u ) − F 2 ( u ) = ( μ 1 − μ 2 ) ( u − u − 1 ) > 0 ( u > 1 ) .
For u ≥ 1 u\ge1 u ≥ 1 one has ν 1 u − 2 ≤ max { ν 1 , 0 } < μ 1 \nu_1u^{-2}\le\max\{\nu_1,0\}<\mu_1 ν 1 u − 2 ≤ max { ν 1 , 0 } < μ 1 , because ν 1 < μ 2 < μ 1 \nu_1<\mu_2<\mu_1 ν 1 < μ 2 < μ 1 ; hence F 1 ′ ( u ) = μ 1 − ν 1 u − 2 > 0 F_1'(u)=\mu_1-\nu_1u^{-2}>0 F 1 ′ ( u ) = μ 1 − ν 1 u − 2 > 0 , and F 1 F_1 F 1 is strictly increasing on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) . If u 1 ≥ u 2 u_1\ge u_2 u 1 ≥ u 2 , then, as u 2 > 1 u_2>1 u 2 > 1 , F 1 ( u 1 ) ≥ F 1 ( u 2 ) > F 2 ( u 2 ) F_1(u_1)\ge F_1(u_2)>F_2(u_2) F 1 ( u 1 ) ≥ F 1 ( u 2 ) > F 2 ( u 2 ) , which contradicts F 1 ( u 1 ) = F 2 ( u 2 ) F_1(u_1)=F_2(u_2) F 1 ( u 1 ) = F 2 ( u 2 ) . So u 1 < u 2 u_1<u_2 u 1 < u 2 .
We next prove X 2 2 > e 2 X_2^2>e_2 X 2 2 > e 2 . Suppose that X 2 2 ≤ e 2 X_2^2\le e_2 X 2 2 ≤ e 2 . Then e 2 > 0 e_2>0 e 2 > 0 , so ν 2 > 0 \nu_2>0 ν 2 > 0 ; and u 2 2 ≤ ν 2 / μ 2 u_2^2\le\nu_2/\mu_2 u 2 2 ≤ ν 2 / μ 2 , because X 2 2 = μ 2 2 u 2 2 X_2^2=\mu_2^2u_2^2 X 2 2 = μ 2 2 u 2 2 and e 2 = μ 2 ν 2 e_2=\mu_2\nu_2 e 2 = μ 2 ν 2 . Thus 1 < u 2 ≤ ( ν 2 / μ 2 ) 1 / 2 1<u_2\le(\nu_2/\mu_2)^{1/2} 1 < u 2 ≤ ( ν 2 / μ 2 ) 1/2 . For 0 < u < ( ν 2 / μ 2 ) 1 / 2 0<u<(\nu_2/\mu_2)^{1/2} 0 < u < ( ν 2 / μ 2 ) 1/2 one has F 2 ′ ( u ) = μ 2 − ν 2 u − 2 < 0 F_2'(u)=\mu_2-\nu_2u^{-2}<0 F 2 ′ ( u ) = μ 2 − ν 2 u − 2 < 0 , so F 2 F_2 F 2 is strictly decreasing on ( 0 , ( ν 2 / μ 2 ) 1 / 2 ] (0,(\nu_2/\mu_2)^{1/2}] ( 0 , ( ν 2 / μ 2 ) 1/2 ] , and F 2 ( u 2 ) < F 2 ( 1 ) = σ F_2(u_2)<F_2(1)=\sigma F 2 ( u 2 ) < F 2 ( 1 ) = σ . On the other hand F 1 ( u 1 ) ≥ F 1 ( 1 ) = σ F_1(u_1)\ge F_1(1)=\sigma F 1 ( u 1 ) ≥ F 1 ( 1 ) = σ , because u 1 ≥ 1 u_1\ge1 u 1 ≥ 1 and F 1 F_1 F 1 is increasing on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) . Hence F 2 ( u 2 ) < F 1 ( u 1 ) F_2(u_2)<F_1(u_1) F 2 ( u 2 ) < F 1 ( u 1 ) , a contradiction.
We next prove X 1 > X 2 X_1>X_2 X 1 > X 2 . For X > 0 X>0 X > 0 one has h e ′ ( X ) = 1 − e X − 2 h_e'(X)=1-eX^{-2} h e ′ ( X ) = 1 − e X − 2 . Since u 1 ≥ 1 u_1\ge1 u 1 ≥ 1 and μ 1 > ν 1 \mu_1>\nu_1 μ 1 > ν 1 , X 1 2 ≥ μ 1 2 > μ 1 ν 1 = e 1 X_1^2\ge\mu_1^2>\mu_1\nu_1=e_1 X 1 2 ≥ μ 1 2 > μ 1 ν 1 = e 1 ; so h e 1 ′ ( X ) > 0 h_{e_1}'(X)>0 h e 1 ′ ( X ) > 0 for X ≥ X 1 X\ge X_1 X ≥ X 1 , and h e 1 h_{e_1} h e 1 is strictly increasing on [ X 1 , ∞ ) [X_1,\infty) [ X 1 , ∞ ) . If X 1 ≤ X 2 X_1\le X_2 X 1 ≤ X 2 , then
h e 1 ( X 1 ) ≤ h e 1 ( X 2 ) < h e 2 ( X 2 ) , h_{e_1}(X_1)\le h_{e_1}(X_2)<h_{e_2}(X_2), h e 1 ( X 1 ) ≤ h e 1 ( X 2 ) < h e 2 ( X 2 ) ,
the second inequality because e 1 < e 2 e_1<e_2 e 1 < e 2 and X 2 > 0 X_2>0 X 2 > 0 ; this contradicts h e 1 ( X 1 ) = h e 2 ( X 2 ) h_{e_1}(X_1)=h_{e_2}(X_2) h e 1 ( X 1 ) = h e 2 ( X 2 ) . So X 1 > X 2 X_1>X_2 X 1 > X 2 . Since log ( X 1 / X 2 ) = log ( μ 1 / μ 2 ) + log ( u 1 / u 2 ) \log(X_1/X_2)=\log(\mu_1/\mu_2)+\log(u_1/u_2) log ( X 1 / X 2 ) = log ( μ 1 / μ 2 ) + log ( u 1 / u 2 ) , the inequalities u 1 < u 2 u_1<u_2 u 1 < u 2 and X 1 > X 2 X_1>X_2 X 1 > X 2 say that 0 < log ( X 1 / X 2 ) < log ( μ 1 / μ 2 ) 0<\log(X_1/X_2)<\log(\mu_1/\mu_2) 0 < log ( X 1 / X 2 ) < log ( μ 1 / μ 2 ) , which is item 1.
We next derive the exact identity on which items 2 and 3 rest. The relation h e 1 ( X 1 ) = h e 2 ( X 2 ) h_{e_1}(X_1)=h_{e_2}(X_2) h e 1 ( X 1 ) = h e 2 ( X 2 ) reads X 1 − X 2 = e 2 X 2 − 1 − e 1 X 1 − 1 X_1-X_2=e_2X_2^{-1}-e_1X_1^{-1} X 1 − X 2 = e 2 X 2 − 1 − e 1 X 1 − 1 . Multiplying by X 1 X 2 X_1X_2 X 1 X 2 gives ( X 1 − X 2 ) X 1 X 2 = e 2 X 1 − e 1 X 2 (X_1-X_2)X_1X_2=e_2X_1-e_1X_2 ( X 1 − X 2 ) X 1 X 2 = e 2 X 1 − e 1 X 2 , and subtracting e 2 ( X 1 − X 2 ) e_2(X_1-X_2) e 2 ( X 1 − X 2 ) from both sides gives
( X 1 − X 2 ) ( X 1 X 2 − e 2 ) = ( e 2 − e 1 ) X 2 . (X_1-X_2)(X_1X_2-e_2)=(e_2-e_1)X_2. ( X 1 − X 2 ) ( X 1 X 2 − e 2 ) = ( e 2 − e 1 ) X 2 .
We next prove item 2. Since X 1 > X 2 > 0 X_1 > X_2 > 0 X 1 > X 2 > 0 and X 2 2 > e 2 X_2^2 > e_2 X 2 2 > e 2 , we have X 1 X 2 − e 2 > X 2 2 − e 2 > 0 X_1X_2-e_2 > X_2^2-e_2 > 0 X 1 X 2 − e 2 > X 2 2 − e 2 > 0 . Since X 1 / X 2 > 1 X_1/X_2 > 1 X 1 / X 2 > 1 and the logarithm lies strictly below its tangent line at 1, the identity gives
log X 1 X 2 < X 1 − X 2 X 2 = e 2 − e 1 X 1 X 2 − e 2 < e 2 − e 1 X 2 2 − e 2 . \log\frac{X_1}{X_2}<\frac{X_1-X_2}{X_2}=\frac{e_2-e_1}{X_1X_2-e_2}<\frac{e_2-e_1}{X_2^2-e_2}. log X 2 X 1 < X 2 X 1 − X 2 = X 1 X 2 − e 2 e 2 − e 1 < X 2 2 − e 2 e 2 − e 1 .
the last step because e 2 − e 1 > 0 e_2-e_1 > 0 e 2 − e 1 > 0 .
We finally prove item 3. Let e 2 ≤ 0 e_2 \le0 e 2 ≤ 0 and 2 ( e 2 − e 1 ) ≤ X 1 2 2(e_2-e_1) \le X_1^2 2 ( e 2 − e 1 ) ≤ X 1 2 . Since 0 < X 2 < X 1 0 < X_2 < X_1 0 < X 2 < X 1 and e 2 ≤ 0 e_2 \le0 e 2 ≤ 0 , we have e 2 X 2 − 1 ≤ e 2 X 1 − 1 e_2X_2^{-1} \le e_2X_1^{-1} e 2 X 2 − 1 ≤ e 2 X 1 − 1 , so
X 1 − X 2 = e 2 X 2 − 1 − e 1 X 1 − 1 ≤ ( e 2 − e 1 ) X 1 − 1 . X_1-X_2=e_2X_2^{-1}-e_1X_1^{-1}\le(e_2-e_1)X_1^{-1}. X 1 − X 2 = e 2 X 2 − 1 − e 1 X 1 − 1 ≤ ( e 2 − e 1 ) X 1 − 1 .
Hence δ = 1 − X 2 / X 1 \delta=1-X_2/X_1 δ = 1 − X 2 / X 1 satisfies 0 < δ ≤ ( e 2 − e 1 ) / X 1 2 ≤ 1 2 0 < \delta\le(e_2-e_1)/X_1^2 \le\frac{1}{2} 0 < δ ≤ ( e 2 − e 1 ) / X 1 2 ≤ 2 1 . For 0 < δ < 1 0 < \delta< 1 0 < δ < 1 one has − log ( 1 − δ ) = ∑ k ≥ 1 δ k / k < ∑ k ≥ 1 δ k = δ / ( 1 − δ ) -\log(1-\delta)=\sum_{k\ge1}\delta^k/k < \sum_{k\ge1}\delta^k=\delta/(1-\delta) − log ( 1 − δ ) = ∑ k ≥ 1 δ k / k < ∑ k ≥ 1 δ k = δ / ( 1 − δ ) , and δ / ( 1 − δ ) ≤ 2 δ \delta/(1-\delta)\le2\delta δ / ( 1 − δ ) ≤ 2 δ because δ ≤ 1 2 \delta\le\frac{1}{2} δ ≤ 2 1 . Therefore
log X 1 X 2 = − log ( 1 − δ ) < δ 1 − δ ≤ 2 δ ≤ 2 ( e 2 − e 1 ) X 1 2 . \log\frac{X_1}{X_2}=-\log(1-\delta)<\frac{\delta}{1-\delta}\le2\delta\le\frac{2(e_2-e_1)}{X_1^2}. log X 2 X 1 = − log ( 1 − δ ) < 1 − δ δ ≤ 2 δ ≤ X 1 2 2 ( e 2 − e 1 ) .
Lemma 2.18. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the linear forms, the gap ratios and the numbers D λ D_\lambda D λ , E λ E_\lambda E λ , κ λ \kappa_\lambda κ λ of Definition 2.16. Then:
l log α < m log β < n log γ l\log\alpha< m\log\beta< n\log\gamma l log α < m log β < n log γ ;
a / c < log ( β / α ) < a / b \sqrt{a/c}<\log(\beta/\alpha)<\sqrt{a/b} a / c < log ( β / α ) < a / b , b − a b c < log ( γ / β ) < b − a a c \frac{b-a}{\sqrt{bc}}<\log(\gamma/\beta)<\frac{b-a}{\sqrt{ac}} b c b − a < log ( γ / β ) < a c b − a , and log ( γ / α ) < b / a \log(\gamma/\alpha)<\sqrt{b/a} log ( γ / α ) < b / a ;
m > ( l − m ) R x m>(l-m)R_x m > ( l − m ) R x and n > ( m − n ) R z n>(m-n)R_z n > ( m − n ) R z ;
0 < Λ x < min { log D λ , 4.01 β − 4 m } 0<\Lambda_x<\min\{\log D_\lambda,\ 4.01\beta^{-4m}\} 0 < Λ x < min { log D λ , 4.01 β − 4 m } ;
0 < Λ z < min { log E λ , κ λ β − 4 m , 2 β − 4 n } 0<\Lambda_z<\min\{\log E_\lambda,\ \kappa_\lambda\beta^{-4m},\ 2\beta^{-4n}\} 0 < Λ z < min { log E λ , κ λ β − 4 m , 2 β − 4 n } .
Proof. Items 1, 4 and 5 compare the two expressions of x x x and of z z z given by Lemma 2.15, by Lemma 2.17; items 2 and 3 are elementary. By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n and m ≥ 4 m\ge4 m ≥ 4 . By items 1 and 4 of Lemma 2.4, s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 , t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 and c > 4 a c>4a c > 4 a ; so a / c < 1 2 \sqrt{a/c}<\frac{1}{2} a / c < 2 1 , and β = s + a c > 2 a c \beta=s+\sqrt{ac}>2\sqrt{ac} β = s + a c > 2 a c .
We first compare the two expressions of x x x . By Lemma 2.15, 2 x = μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 2x=\mu_1u_1+\nu_1u_1^{-1}=\mu_2u_2+\nu_2u_2^{-1} 2 x = μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 with
( μ 1 , ν 1 , u 1 ) = ( 1 + a / b , 1 − a / b , α 2 l ) , ( μ 2 , ν 2 , u 2 ) = ( 1 + λ a / c , 1 − λ a / c , β 2 m ) . (\mu_1,\nu_1,u_1)=\left(1+\sqrt{a/b},1-\sqrt{a/b},\alpha^{2l}\right),\qquad
(\mu_2,\nu_2,u_2)=\left(1+\lambda\sqrt{a/c},1-\lambda\sqrt{a/c},\beta^{2m}\right). ( μ 1 , ν 1 , u 1 ) = ( 1 + a / b , 1 − a / b , α 2 l ) , ( μ 2 , ν 2 , u 2 ) = ( 1 + λ a / c , 1 − λ a / c , β 2 m ) .
The hypotheses of Lemma 2.17 hold: μ 1 + ν 1 = μ 2 + ν 2 = 2 \mu_1+\nu_1=\mu_2+\nu_2=2 μ 1 + ν 1 = μ 2 + ν 2 = 2 ; μ 1 > μ 2 ≥ 1 − a / c > 1 2 \mu_1>\mu_2\ge1-\sqrt{a/c}>\frac{1}{2} μ 1 > μ 2 ≥ 1 − a / c > 2 1 and ν 1 < μ 2 \nu_1<\mu_2 ν 1 < μ 2 , because a / c < a / b \sqrt{a/c}<\sqrt{a/b} a / c < a / b ; and u 1 ≥ 1 u_1\ge1 u 1 ≥ 1 , u 2 > 1 u_2>1 u 2 > 1 . Dividing the numerator and the denominator of D λ D_\lambda D λ by b c \sqrt{bc} b c gives μ 1 / μ 2 = D λ \mu_1/\mu_2=D_\lambda μ 1 / μ 2 = D λ , so log ( μ 1 u 1 / ( μ 2 u 2 ) ) = Λ x \log(\mu_1u_1/(\mu_2u_2))=\Lambda_x log ( μ 1 u 1 / ( μ 2 u 2 )) = Λ x . Item 1 of Lemma 2.17 gives α 2 l < β 2 m \alpha^{2l}<\beta^{2m} α 2 l < β 2 m and 0 < Λ x < log D λ 0<\Lambda_x<\log D_\lambda 0 < Λ x < log D λ . Here e 1 = 1 − a / b e_1=1-a/b e 1 = 1 − a / b and e 2 = 1 − a / c e_2=1-a/c e 2 = 1 − a / c , so 0 < e 2 − e 1 < 1 0<e_2-e_1<1 0 < e 2 − e 1 < 1 and e 2 < 1 e_2<1 e 2 < 1 ; with μ 2 u 2 > 1 2 β 2 m \mu_2u_2>\frac{1}{2}\beta^{2m} μ 2 u 2 > 2 1 β 2 m , item 2 gives
Λ x < 1 1 4 β 4 m − 1 = 4 β 4 m − 4 < 4.01 β − 4 m , \Lambda_x<\frac{1}{\frac{1}{4}\beta^{4m}-1}=\frac{4}{\beta^{4m}-4}<4.01\beta^{-4m}, Λ x < 4 1 β 4 m − 1 1 = β 4 m − 4 4 < 4.01 β − 4 m ,
the last inequality because β 4 m ≥ β 16 > ( 4 a c ) 8 > 1604 \beta^{4m} \ge\beta^{16} > (4ac)^8 > 1604 β 4 m ≥ β 16 > ( 4 a c ) 8 > 1604 .
We next compare the two expressions of z z z . By Lemma 2.15 , 2 z = μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 2z=\mu_1u_1+\nu_1u_1^{-1}=\mu_2u_2+\nu_2u_2^{-1} 2 z = μ 1 u 1 + ν 1 u 1 − 1 = μ 2 u 2 + ν 2 u 2 − 1 with
( μ 1 , ν 1 , u 1 ) = ( c / a + λ , λ − c / a , β 2 m ) , ( μ 2 , ν 2 , u 2 ) = ( c / b + λ , λ − c / b , γ 2 n ) . (\mu_1,\nu_1,u_1)=\left(\sqrt{c/a}+\lambda,\lambda-\sqrt{c/a},\beta^{2m}\right),\qquad
(\mu_2,\nu_2,u_2)=\left(\sqrt{c/b}+\lambda,\lambda-\sqrt{c/b},\gamma^{2n}\right). ( μ 1 , ν 1 , u 1 ) = ( c / a + λ , λ − c / a , β 2 m ) , ( μ 2 , ν 2 , u 2 ) = ( c / b + λ , λ − c / b , γ 2 n ) .
The hypotheses hold: μ 1 + ν 1 = μ 2 + ν 2 = 2 λ \mu_1+\nu_1=\mu_2+\nu_2=2\lambda μ 1 + ν 1 = μ 2 + ν 2 = 2 λ ; μ 1 > μ 2 > 0 \mu_1>\mu_2>0 μ 1 > μ 2 > 0 and ν 1 < 0 < μ 2 \nu_1<0<\mu_2 ν 1 < 0 < μ 2 , because a < b < c a<b<c a < b < c ; and u 1 , u 2 > 1 u_1,u_2>1 u 1 , u 2 > 1 . Here μ 1 / μ 2 = E λ \mu_1/\mu_2=E_\lambda μ 1 / μ 2 = E λ , so log ( μ 1 u 1 / ( μ 2 u 2 ) ) = Λ z \log(\mu_1u_1/(\mu_2u_2))=\Lambda_z log ( μ 1 u 1 / ( μ 2 u 2 )) = Λ z , and item 1 gives β 2 m < γ 2 n \beta^{2m}<\gamma^{2n} β 2 m < γ 2 n and 0 < Λ z < log E λ 0<\Lambda_z<\log E_\lambda 0 < Λ z < log E λ . Here e 1 = 1 − c / a e_1=1-c/a e 1 = 1 − c / a and e 2 = 1 − c / b < 0 e_2=1-c/b<0 e 2 = 1 − c / b < 0 , so e 2 − e 1 = c ( b − a ) / ( a b ) e_2-e_1=c(b-a)/(ab) e 2 − e 1 = c ( b − a ) / ( ab ) and 2 ( e 2 − e 1 ) / ( μ 1 u 1 ) 2 = κ λ β − 4 m 2(e_2-e_1)/(\mu_1u_1)^2=\kappa_\lambda\beta^{-4m} 2 ( e 2 − e 1 ) / ( μ 1 u 1 ) 2 = κ λ β − 4 m . Since c + λ a ≥ c − a > 1 2 c \sqrt{c}+\lambda\sqrt{a}\ge\sqrt{c}-\sqrt{a}>\frac12\sqrt{c} c + λ a ≥ c − a > 2 1 c , we have κ λ < 8 ( b − a ) / b < 8 \kappa_\lambda<8(b-a)/b<8 κ λ < 8 ( b − a ) / b < 8 , while β 4 > 16 a 2 ≥ 16 \beta^4>16a^2\ge16 β 4 > 16 a 2 ≥ 16 ; so κ λ β − 4 m < 1 \kappa_\lambda\beta^{-4m}<1 κ λ β − 4 m < 1 , and item 3 gives Λ z < κ λ β − 4 m \Lambda_z<\kappa_\lambda\beta^{-4m} Λ z < κ λ β − 4 m . Finally m ≥ n + 1 m\ge n+1 m ≥ n + 1 gives κ λ β − 4 m < 8 β − 4 β − 4 n < 2 β − 4 n \kappa_\lambda\beta^{-4m}<8\beta^{-4}\beta^{-4n}<2\beta^{-4n} κ λ β − 4 m < 8 β − 4 β − 4 n < 2 β − 4 n .
The two comparisons give α 2 l < β 2 m < γ 2 n \alpha^{2l}<\beta^{2m}<\gamma^{2n} α 2 l < β 2 m < γ 2 n , which is item 1, and items 4 and 5.
We next prove item 2 by one mean-value step. As α = r + r 2 − 1 \alpha=r+\sqrt{r^2-1} α = r + r 2 − 1 , β = s + s 2 − 1 \beta=s+\sqrt{s^2-1} β = s + s 2 − 1 and γ = t + t 2 − 1 \gamma=t+\sqrt{t^2-1} γ = t + t 2 − 1 , one has log α = f ( r ) \log\alpha=f(r) log α = f ( r ) , log β = f ( s ) \log\beta=f(s) log β = f ( s ) , log γ = f ( t ) \log\gamma=f(t) log γ = f ( t ) for f ( u ) = log ( u + u 2 − 1 ) f(u)=\log(u+\sqrt{u^2-1}) f ( u ) = log ( u + u 2 − 1 ) , whose derivative ( u 2 − 1 ) − 1 / 2 (u^2-1)^{-1/2} ( u 2 − 1 ) − 1/2 is strictly decreasing on ( 1 , ∞ ) (1,\infty) ( 1 , ∞ ) . By the mean value theorem on [ r , s ] [r,s] [ r , s ] , [ s , t ] [s,t] [ s , t ] and [ r , t ] [r,t] [ r , t ] , where s − r = a s-r=a s − r = a , t − s = b − a t-s=b-a t − s = b − a and t − r = b t-r=b t − r = b ,
a s 2 − 1 < log β α < a r 2 − 1 , b − a t 2 − 1 < log γ β < b − a s 2 − 1 , log γ α < b r 2 − 1 , \frac{a}{\sqrt{s^2-1}}<\log\frac{\beta}{\alpha}<\frac{a}{\sqrt{r^2-1}},\qquad
\frac{b-a}{\sqrt{t^2-1}}<\log\frac{\gamma}{\beta}<\frac{b-a}{\sqrt{s^2-1}},\qquad
\log\frac{\gamma}{\alpha}<\frac{b}{\sqrt{r^2-1}}, s 2 − 1 a < log α β < r 2 − 1 a , t 2 − 1 b − a < log β γ < s 2 − 1 b − a , log α γ < r 2 − 1 b ,
and r 2 − 1 = a b \sqrt{r^2-1}=\sqrt{ab} r 2 − 1 = ab , s 2 − 1 = a c \sqrt{s^2-1}=\sqrt{ac} s 2 − 1 = a c , t 2 − 1 = b c \sqrt{t^2-1}=\sqrt{bc} t 2 − 1 = b c give item 2.
We next prove item 3, by an argument adapted from Cipu, Fujita and Mignotte [10 ] . By items 1 and 2, ( l − m ) log α < m log ( β / α ) < m a / b (l-m)\log\alpha<m\log(\beta/\alpha)<m\sqrt{a/b} ( l − m ) log α < m log ( β / α ) < m a / b , which is m > ( l − m ) R x m>(l-m)R_x m > ( l − m ) R x . Likewise ( m − n ) log β < n log ( γ / β ) < n ( b − a ) / a c (m-n)\log\beta<n\log(\gamma/\beta)<n(b-a)/\sqrt{ac} ( m − n ) log β < n log ( γ / β ) < n ( b − a ) / a c , which is n > ( m − n ) R z n>(m-n)R_z n > ( m − n ) R z .
Items 4 and 5 are the input of every bound by linear forms in logarithms below, in Sections 4 and 7 and in Appendix B, and item 3, the gap inequalities, is the lower side of the criterion for close units. The lower bound for the index that holds for every irregular extension is the subject of the next section.
A quartic index floor
This section proves the quartic index floor (Theorem 3.6 ). The exponent 4 has a simple source: the index polynomials of the convention below are quadratic in l , m , n l,m,n l , m , n , the second-order congruences make the order-two quotients integers, with 0 < k c < H 0<k_c<H 0 < k c < H (Lemma 3.2 ), and each of the two bounds of the section multiplies two quadratic quantities, k c k_c k c and L 3 L_3 L 3 in the bracket for c c c (Lemma 3.4 ), and the index polynomials and k c k_c k c in the bound for a + b a+b a + b that the cofactor triples give (Lemma 3.5 ). The bracket also needs the third-order congruence modulo c c c of Lemma 3.3 .
Convention 3.1. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with roots r , s , t r,s,t r , s , t (Definition 2.3 ), and let λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } and l , m , n l,m,n l , m , n be integers; when ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) is an index system (Definition 2.5 ), the quantities below are evaluated at it. The index polynomials are
L 1 = l 2 − m 2 , L 2 = l 2 − n 2 , L 3 = m 2 − n 2 , L_1=l^2-m^2,\qquad L_2=l^2-n^2,\qquad L_3=m^2-n^2, L 1 = l 2 − m 2 , L 2 = l 2 − n 2 , L 3 = m 2 − n 2 , M 1 = 2 m 2 + λ m − l , M 2 = 2 n 2 + λ n − l , M_1=2m^2+\lambda m-l,\qquad M_2=2n^2+\lambda n-l, M 1 = 2 m 2 + λm − l , M 2 = 2 n 2 + λn − l , H 1 = m ( m + λ ) , H 2 = n ( n + λ ) , H = H 1 + H 2 , H_1=m(m+\lambda),\qquad H_2=n(n+\lambda),\qquad H=H_1+H_2, H 1 = m ( m + λ ) , H 2 = n ( n + λ ) , H = H 1 + H 2 , B m = 2 m 2 + λ ( m + n ) , B n = 2 n 2 + λ ( m + n ) , B_m=2m^2+\lambda(m+n),\qquad B_n=2n^2+\lambda(m+n), B m = 2 m 2 + λ ( m + n ) , B n = 2 n 2 + λ ( m + n ) ,
and the order-two quotients are the rational numbers
k a = b L 1 − r M 1 a , k b = a L 2 − r M 2 b , k c = a H + r B n c . k_a = \frac{bL_1-rM_1}{a}, \qquad k_b = \frac{aL_2-rM_2}{b}, \qquad k_c = \frac{aH+rB_n}{c}. k a = a b L 1 − r M 1 , k b = b a L 2 − r M 2 , k c = c a H + r B n .
Lemma 3.2. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the index polynomials and the order-two quotients k a , k b , k c k_a,k_b,k_c k a , k b , k c of Convention 3.1. Then:
the index polynomials satisfy
( b l 2 + r l ) − ( c m 2 + λ s m ) = b L 1 − r M 1 − a H 1 , (bl^2+rl)-(cm^2+\lambda sm)=bL_1-rM_1-aH_1, ( b l 2 + r l ) − ( c m 2 + λ s m ) = b L 1 − r M 1 − a H 1 , ( a l 2 + r l ) − ( c n 2 + λ t n ) = a L 2 − r M 2 − b H 2 , (al^2+rl)-(cn^2+\lambda tn)=aL_2-rM_2-bH_2, ( a l 2 + r l ) − ( c n 2 + λ t n ) = a L 2 − r M 2 − b H 2 , a m 2 + λ s m − b n 2 − λ t n = a H + r B n − c H 2 ; am^2+\lambda sm-bn^2-\lambda tn=aH+rB_n-cH_2; a m 2 + λ s m − b n 2 − λ t n = a H + r B n − c H 2 ; k a , k b , k c k_a,k_b,k_c k a , k b , k c are integers, k a < 0 k_a<0 k a < 0 , k b < 0 k_b<0 k b < 0 and 0 < k c < H 0<k_c<H 0 < k c < H .
Proof. Item 1 consists of polynomial identities. Integrality comes from comparing, for each of x x x , y y y and z z z , the second-order congruences of the two Pell equations that contain it; the signs of k a k_a k a and k b k_b k b come from the index inequalities of Lemma 2.18 , and the bounds for k c k_c k c from the index ordering. Throughout, c = a + b + 2 r c=a+b+2r c = a + b + 2 r , s = a + r s=a+r s = a + r , t = b + r t=b+r t = b + r and r 2 = a b + 1 r^2=ab+1 r 2 = ab + 1 (Definition 2.3 ), so that a c + 1 = s 2 ac+1=s^2 a c + 1 = s 2 and b c + 1 = t 2 bc+1=t^2 b c + 1 = t 2 (item 1 of Lemma 2.4 ), and the index polynomials are those of Convention 3.1. By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n , so l ≤ 2 n − 1 ≤ 2 m − 3 l\le2n-1\le2m-3 l ≤ 2 n − 1 ≤ 2 m − 3 and m ≤ 2 n − 2 m\le2n-2 m ≤ 2 n − 2 .
We first prove item 1. Substituting c = a + b + 2 r c=a+b+2r c = a + b + 2 r , s = a + r s=a+r s = a + r and t = b + r t=b+r t = b + r and expanding the index polynomials, the two sides of each identity become the same polynomial in a , b , r a,b,r a , b , r , with λ , l , m , n \lambda,l,m,n λ , l , m , n as parameters.
We next prove that k a , k b , k c k_a,k_b,k_c k a , k b , k c are integers. By items 1 and 2 of Lemma 2.8 , y a + x b y\sqrt{a}+x\sqrt{b} y a + x b , z a + x c z\sqrt{a}+x\sqrt{c} z a + x c and z b + y c z\sqrt{b}+y\sqrt{c} z b + y c are the terms of index 2 l 2l 2 l , 2 m 2m 2 m , 2 n 2n 2 n of the orbits of Lemma 2.7 with ( P , Q , S , u 0 , v 0 ) = ( a , b , r , 1 , 1 ) (P,Q,S,u_0,v_0)=(a,b,r,1,1) ( P , Q , S , u 0 , v 0 ) = ( a , b , r , 1 , 1 ) , ( a , c , s , λ , 1 ) (a,c,s,\lambda,1) ( a , c , s , λ , 1 ) and ( b , c , t , λ , 1 ) (b,c,t,\lambda,1) ( b , c , t , λ , 1 ) respectively. Item 2 of Lemma 2.7 , applied to x x x in the first two orbits, to y y y in the first and the third, and to z z z in the last two, gives
x ≡ 1 + 2 a ( b l 2 + r l ) ≡ 1 + 2 a ( c m 2 + λ s m ) ( m o d 8 a 2 ) , x\equiv1+2a(bl^2+rl)\equiv1+2a(cm^2+\lambda sm)\pmod{8a^2}, x ≡ 1 + 2 a ( b l 2 + r l ) ≡ 1 + 2 a ( c m 2 + λ s m ) ( mod 8 a 2 ) ,
y ≡ 1 + 2 b ( a l 2 + r l ) ≡ 1 + 2 b ( c n 2 + λ t n ) ( m o d 8 b 2 ) , y\equiv1+2b(al^2+rl)\equiv1+2b(cn^2+\lambda tn)\pmod{8b^2}, y ≡ 1 + 2 b ( a l 2 + r l ) ≡ 1 + 2 b ( c n 2 + λ t n ) ( mod 8 b 2 ) ,
z ≡ λ + 2 c ( λ a m 2 + s m ) ≡ λ + 2 c ( λ b n 2 + t n ) ( m o d 8 c 2 ) . z\equiv\lambda+2c(\lambda am^2+sm)\equiv\lambda+2c(\lambda bn^2+tn)\pmod{8c^2}. z ≡ λ + 2 c ( λa m 2 + s m ) ≡ λ + 2 c ( λb n 2 + t n ) ( mod 8 c 2 ) .
Cancelling 2 a 2a 2 a , 2 b 2b 2 b , 2 c 2c 2 c and multiplying the last congruence by λ \lambda λ ,
4 a ∣ ( b l 2 + r l ) − ( c m 2 + λ s m ) , 4 b ∣ ( a l 2 + r l ) − ( c n 2 + λ t n ) , 4 c ∣ a m 2 + λ s m − b n 2 − λ t n . 4a\mid(bl^2+rl)-(cm^2+\lambda sm), \qquad4b\mid(al^2+rl)-(cn^2+\lambda tn), \qquad4c\mid am^2+\lambda sm-bn^2-\lambda tn. 4 a ∣ ( b l 2 + r l ) − ( c m 2 + λ s m ) , 4 b ∣ ( a l 2 + r l ) − ( c n 2 + λ t n ) , 4 c ∣ a m 2 + λ s m − b n 2 − λ t n .
By item 1, these three differences are
b L 1 − r M 1 − a H 1 , a L 2 − r M 2 − b H 2 , a H + r B n − c H 2 , bL_1-rM_1-aH_1, \qquad aL_2-rM_2-bH_2, \qquad aH+rB_n-cH_2, b L 1 − r M 1 − a H 1 , a L 2 − r M 2 − b H 2 , a H + r B n − c H 2 ,
so a ∣ b L 1 − r M 1 a\mid bL_1-rM_1 a ∣ b L 1 − r M 1 , b ∣ a L 2 − r M 2 b\mid aL_2-rM_2 b ∣ a L 2 − r M 2 and c ∣ a H + r B n c\mid aH+rB_n c ∣ a H + r B n , that is, k a , k b , k c ∈ Z k_a,k_b,k_c\in\mathbb{Z} k a , k b , k c ∈ Z .
We next bound log α \log\alpha log α . By Lemma 2.14, b > 4000 b>4000 b > 4000 , so r 2 = a b + 1 > 4000 r^2=ab+1>4000 r 2 = ab + 1 > 4000 and r ≥ 64 r\ge64 r ≥ 64 . Since α = r + a b > r \alpha=r+\sqrt{ab}>r α = r + ab > r ,
log α > log 64 > 3. \log\alpha>\log64>3. log α > log 64 > 3.
We next prove k a < 0 k_a < 0 k a < 0 and k b < 0 k_b < 0 k b < 0 , by one argument. By item 3 of Lemma 2.18 , with the gap ratio R x R_x R x of Definition 2.16 , m > ( l − m ) R x = ( l − m ) b / a log α m > (l-m)R_x = (l-m)\sqrt{b/a}\log\alpha m > ( l − m ) R x = ( l − m ) b / a log α , so ( l − m ) log α < m a / b (l-m)\log\alpha< m\sqrt{a/b} ( l − m ) log α < m a / b ; and l + m < 3 m l+m < 3m l + m < 3 m , as l < 2 n < 2 m l < 2n < 2m l < 2 n < 2 m . Hence
b L 1 = b ( l − m ) ( l + m ) < 3 a b log α m 2 < r m 2 < r M 1 , bL_1 = b(l-m)(l+m) < \frac{3\sqrt{ab}}{\log\alpha}m^2 < rm^2 < rM_1, b L 1 = b ( l − m ) ( l + m ) < log α 3 ab m 2 < r m 2 < r M 1 ,
since a b < r \sqrt{ab} < r ab < r , log α > 3 \log\alpha> 3 log α > 3 and M 1 ≥ 2 m 2 − m − ( 2 m − 3 ) = 2 m 2 − 3 m + 3 > m 2 M_1 \ge2m^2-m-(2m-3)=2m^2-3m+3>m^2 M 1 ≥ 2 m 2 − m − ( 2 m − 3 ) = 2 m 2 − 3 m + 3 > m 2 ; so k a < 0 k_a<0 k a < 0 . By items 1 and 2 of Lemma 2.18 , log α < n log γ < n ( log α + b / a ) \log\alpha< n\log\gamma< n(\log\alpha+\sqrt{b/a}) log α < n log γ < n ( log α + b / a ) , so ( l − n ) log α < n b / a (l-n)\log\alpha< n\sqrt{b/a} ( l − n ) log α < n b / a ; and l + n < 3 n l+n<3n l + n < 3 n , as l < 2 n l<2n l < 2 n . Hence
a L 2 = a ( l − n ) ( l + n ) < 3 a b log α n 2 < r n 2 < r M 2 , aL_2 = a(l-n)(l+n) < \frac{3\sqrt{ab}}{\log\alpha}n^2 < rn^2 < rM_2, a L 2 = a ( l − n ) ( l + n ) < log α 3 ab n 2 < r n 2 < r M 2 ,
since M 2 ≥ 2 n 2 − n − ( 2 n − 1 ) = 2 n 2 − 3 n + 1 > n 2 M_2\ge2n^2-n-(2n-1)=2n^2-3n+1>n^2 M 2 ≥ 2 n 2 − n − ( 2 n − 1 ) = 2 n 2 − 3 n + 1 > n 2 , as n ≥ 3 n\ge3 n ≥ 3 ; so k b < 0 k_b<0 k b < 0 .
We finally prove 0 < k c < H 0<k_c<H 0 < k c < H . Clearly H > 0 H>0 H > 0 , and B n ≥ 2 n 2 − ( 2 n − 2 ) − n = 2 n 2 − 3 n + 2 > 0 B_n\ge2n^2-(2n-2)-n=2n^2-3n+2>0 B n ≥ 2 n 2 − ( 2 n − 2 ) − n = 2 n 2 − 3 n + 2 > 0 , so k c > 0 k_c>0 k c > 0 . Since 2 H − B n = B m = 2 m 2 + λ ( m + n ) > 0 2H-B_n=B_m=2m^2+\lambda(m+n)>0 2 H − B n = B m = 2 m 2 + λ ( m + n ) > 0 , we have a H + r B n < ( a + 2 r ) H < c H aH+rB_n<(a+2r)H<cH a H + r B n < ( a + 2 r ) H < cH , so k c < H k_c<H k c < H .
Lemma 3.3. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the order-two quotient k c k_c k c and the index polynomial H 2 H_2 H 2 of Convention 3.1, and put
A z = a ( m 3 − m ) ( 2 s + λ a m ) − b ( n 3 − n ) ( 2 t + λ b n ) . A_z=a(m^3-m)(2s+\lambda am)-b(n^3-n)(2t+\lambda bn). A z = a ( m 3 − m ) ( 2 s + λam ) − b ( n 3 − n ) ( 2 t + λbn ) . Then A z + 3 λ ( k c − H 2 ) ≡ 0 ( m o d 24 c ) A_z+3\lambda(k_c-H_2)\equiv0\pmod{24c} A z + 3 λ ( k c − H 2 ) ≡ 0 ( mod 24 c ) .
Proof. Following [20 ] , we expand the common term z z z to third order along each of the two Pell equations that contain it, modulo 16 c 3 16c^3 16 c 3 , and compare. By item 2 of Lemma 3.2 , k c k_c k c is an integer. For q ≥ 0 q\ge0 q ≥ 0 put
f a ( q ) = λ + 2 c ( λ a q 2 + s q ) + 2 3 a c 2 ( q 3 − q ) ( 2 s + λ a q ) , f_a(q)=\lambda+2c(\lambda aq^2+sq)+\frac{2}{3}ac^2(q^3-q)(2s+\lambda aq), f a ( q ) = λ + 2 c ( λa q 2 + s q ) + 3 2 a c 2 ( q 3 − q ) ( 2 s + λa q ) ,
an integer since 6 ∣ q 3 − q 6\mid q^3-q 6 ∣ q 3 − q , and define f b ( q ) f_b(q) f b ( q ) in the same way with ( b , t ) (b,t) ( b , t ) in place of ( a , s ) (a,s) ( a , s ) . We first prove V q λ ≡ f a ( q ) V^\lambda_q\equiv f_a(q) V q λ ≡ f a ( q ) and W q λ ≡ f b ( q ) ( m o d 16 c 3 ) W^\lambda_q\equiv f_b(q)\pmod{16c^3} W q λ ≡ f b ( q ) ( mod 16 c 3 ) for all q ≥ 0 q\ge0 q ≥ 0 , with the Pell sequences of Convention 2.6. One has f a ( 0 ) = λ = V 0 0 f_a(0)=\lambda=V^0_0 f a ( 0 ) = λ = V 0 0 and f a ( 1 ) = λ ( 2 a c + 1 ) + 2 s c = V 1 λ f_a(1)=\lambda(2ac+1)+2sc=V^\lambda_1 f a ( 1 ) = λ ( 2 a c + 1 ) + 2 sc = V 1 λ , by item 3 of Lemma 2.8 . Identically in a , c , s a,c,s a , c , s ,
f a ( q + 2 ) − ( 4 a c + 2 ) f a ( q + 1 ) + f a ( q ) = − 2 3 a 2 c 3 q ( q + 1 ) ( q + 2 ) ( 2 s + λ a ( q + 1 ) ) ≡ 0 ( m o d 16 c 3 ) , f_a(q+2)-(4ac+2)f_a(q+1)+f_a(q)=-\frac{2}{3}a^2c^3q(q+1)(q+2)(2s+\lambda a(q+1))\equiv0\pmod{16c^3}, f a ( q + 2 ) − ( 4 a c + 2 ) f a ( q + 1 ) + f a ( q ) = − 3 2 a 2 c 3 q ( q + 1 ) ( q + 2 ) ( 2 s + λa ( q + 1 )) ≡ 0 ( mod 16 c 3 ) ,
because 6 ∣ q ( q + 1 ) ( q + 2 ) 6\mid q(q+1)(q+2) 6 ∣ q ( q + 1 ) ( q + 2 ) . By item 3 of Lemma 2.8 , the sequence ( V q λ ) q (V^\lambda_q)_q ( V q λ ) q satisfies V q + 2 λ = ( 4 a c + 2 ) V q + 1 λ − V q λ V^\lambda_{q+2}=(4ac+2)V^\lambda_{q+1}-V^\lambda_q V q + 2 λ = ( 4 a c + 2 ) V q + 1 λ − V q λ . Induction on q q q gives V q λ ≡ f a ( q ) ( m o d 16 c 3 ) V^\lambda_q\equiv f_a(q)\pmod{16c^3} V q λ ≡ f a ( q ) ( mod 16 c 3 ) . The same argument with ( b , t ) (b,t) ( b , t ) in place of ( a , s ) (a,s) ( a , s ) gives W q λ ≡ f b ( q ) ( m o d 16 c 3 ) W^\lambda_q\equiv f_b(q)\pmod{16c^3} W q λ ≡ f b ( q ) ( mod 16 c 3 ) .
We next compare the two expansions at the common term. By item 2 of Lemma 2.8 , z = V m λ = W n λ z=V^\lambda_m=W^\lambda_n z = V m λ = W n λ . By item 1 of Lemma 3.2 , a H + r B n − c H 2 = a m 2 + λ s m − b n 2 − λ t n aH+rB_n-cH_2=am^2+\lambda sm-bn^2-\lambda tn a H + r B n − c H 2 = a m 2 + λ s m − b n 2 − λ t n , and a H + r B n = c k c aH+rB_n=ck_c a H + r B n = c k c ; hence λ ( a m 2 − b n 2 ) + s m − t n = λ c ( k c − H 2 ) \lambda(am^2-bn^2)+sm-tn=\lambda c(k_c-H_2) λ ( a m 2 − b n 2 ) + s m − t n = λ c ( k c − H 2 ) . Therefore
0 = V m λ − W n λ ≡ f a ( m ) − f b ( n ) = 2 c ( λ ( a m 2 − b n 2 ) + s m − t n ) + 2 3 c 2 A z = 2 3 c 2 ( A z + 3 λ ( k c − H 2 ) ) ( m o d 16 c 3 ) . \begin{aligned}
0&=V^\lambda_m-W^\lambda_n\equiv f_a(m)-f_b(n)\\
&=2c\bigl(\lambda(am^2-bn^2)+sm-tn\bigr)+\frac{2}{3}c^2A_z\\
&=\frac{2}{3}c^2\bigl(A_z+3\lambda(k_c-H_2)\bigr)\pmod{16c^3}.
\end{aligned} 0 = V m λ − W n λ ≡ f a ( m ) − f b ( n ) = 2 c ( λ ( a m 2 − b n 2 ) + s m − t n ) + 3 2 c 2 A z = 3 2 c 2 ( A z + 3 λ ( k c − H 2 ) ) ( mod 16 c 3 ) .
Since 6 6 6 divides m 3 − m m^3-m m 3 − m and n 3 − n n^3-n n 3 − n , 3 ∣ A z 3\mid A_z 3 ∣ A z , and dividing by 2 c 2 2c^2 2 c 2 gives A z / 3 + λ ( k c − H 2 ) ≡ 0 ( m o d 8 c ) A_z/3+\lambda(k_c-H_2)\equiv0\pmod{8c} A z /3 + λ ( k c − H 2 ) ≡ 0 ( mod 8 c ) , that is A z + 3 λ ( k c − H 2 ) ≡ 0 ( m o d 24 c ) A_z+3\lambda(k_c-H_2)\equiv0\pmod{24c} A z + 3 λ ( k c − H 2 ) ≡ 0 ( mod 24 c ) .
The bracket below is the main step. Its lower end Ψ ( m , n ) \Psi(m,n) Ψ ( m , n ) is shown to be positive when m ≤ 3 2 n m \le\frac{3}{2}n m ≤ 2 3 n , and the end of its proof shows by two examples that this condition cannot be replaced by m < 2 n m < 2n m < 2 n ; the range m > 3 2 n m > \frac{3}{2}n m > 2 3 n is covered by the bound for a + b a+b a + b that follows it, since c ≤ 2 ( a + b ) c \le2(a+b) c ≤ 2 ( a + b ) (Lemma 2.4 ).
Lemma 3.4. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the index polynomials and the order-two quotient k c k_c k c of Convention 3.1. Put
Ψ ( u , v ) = − u 4 + u 2 ( 3 v 2 + λ v + 1 ) + 2 λ u v ( 2 v + λ ) + v 2 ( 2 v 2 + 3 λ v + 1 ) , P z = 3 k c L 3 + Ψ ( m , n ) . \Psi(u,v)=-u^4+u^2(3v^2+\lambda v+1)+2\lambda uv(2v+\lambda)+v^2(2v^2+3\lambda v+1), \qquad P_z=3k_cL_3+\Psi(m,n). Ψ ( u , v ) = − u 4 + u 2 ( 3 v 2 + λ v + 1 ) + 2 λ uv ( 2 v + λ ) + v 2 ( 2 v 2 + 3 λ v + 1 ) , P z = 3 k c L 3 + Ψ ( m , n ) . Then c ∣ P z c\mid P_z c ∣ P z and
Ψ ( m , n ) < P z < Ψ ( n , m ) < 4 m 2 ( m + λ ) 2 . \Psi(m,n)<P_z<\Psi(n,m)<4m^2(m+\lambda)^2. Ψ ( m , n ) < P z < Ψ ( n , m ) < 4 m 2 ( m + λ ) 2 . If moreover m ≤ 3 2 n m\le\frac{3}{2}n m ≤ 2 3 n , then Ψ ( m , n ) > 0 \Psi(m,n)>0 Ψ ( m , n ) > 0 and c ≤ P z < 4 m 2 ( m + λ ) 2 c\le P_z<4m^2(m+\lambda)^2 c ≤ P z < 4 m 2 ( m + λ ) 2 .
Proof. The divisibility comes from the third-order congruence of Lemma 3.3 , multiplied by L 3 L_3 L 3 and reduced modulo c c c ; the bracket from writing P z P_z P z as a convex combination of Ψ ( m , n ) \Psi(m,n) Ψ ( m , n ) and Ψ ( n , m ) \Psi(n,m) Ψ ( n , m ) with weight k c / H k_c/H k c / H . By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n ; by item 2 of Lemma 3.2 , k c k_c k c is an integer with 0 < k c < H 0<k_c<H 0 < k c < H .
We first prove c ∣ P z c\mid P_z c ∣ P z . Write b = c − 2 s + a b=c-2s+a b = c − 2 s + a and t = c − s t=c-s t = c − s , where s 2 = a c + 1 ≡ 1 ( m o d c ) s^2=ac+1\equiv1\pmod c s 2 = a c + 1 ≡ 1 ( mod c ) (item 1 of Lemma 2.4 ), and note that c k c = a H + r B n = a ( H − B n ) + ( a + r ) B n = a L 3 + s B n ck_c=aH+rB_n=a(H-B_n)+(a+r)B_n=aL_3+sB_n c k c = a H + r B n = a ( H − B n ) + ( a + r ) B n = a L 3 + s B n . Substituting b ≡ a − 2 s b\equiv a-2s b ≡ a − 2 s , t ≡ − s t\equiv-s t ≡ − s and s 2 ≡ 1 ( m o d c ) s^2\equiv1\pmod c s 2 ≡ 1 ( mod c ) in A z A_z A z , with A z A_z A z as in Lemma 3.3 , gives
A z ≡ λ a 2 L 3 ( m 2 + n 2 − 1 ) + 2 a s Θ − 4 ( n 3 − n ) ( 1 + λ n ) ( m o d c ) , A_z\equiv\lambda a^2L_3(m^2+n^2-1)+2as\Theta-4(n^3-n)(1+\lambda n)\pmod c, A z ≡ λ a 2 L 3 ( m 2 + n 2 − 1 ) + 2 a s Θ − 4 ( n 3 − n ) ( 1 + λn ) ( mod c ) ,
Θ = ( m 3 − m ) + ( n 3 − n ) + 2 λ n ( n 3 − n ) , \Theta=(m^3-m)+(n^3-n)+2\lambda n(n^3-n), Θ = ( m 3 − m ) + ( n 3 − n ) + 2 λn ( n 3 − n ) ,
since ( a − 2 s ) ( n 3 − n ) ( − 2 s + λ ( a − 2 s ) n ) ≡ ( n 3 − n ) ( λ a 2 n − 2 a s − 4 λ a s n + 4 + 4 λ n ) (a-2s)(n^3-n)(-2s+\lambda(a-2s)n)\equiv(n^3-n)(\lambda a^2n-2as-4\lambda asn+4+4\lambda n) ( a − 2 s ) ( n 3 − n ) ( − 2 s + λ ( a − 2 s ) n ) ≡ ( n 3 − n ) ( λ a 2 n − 2 a s − 4 λa s n + 4 + 4 λn ) . Multiplying by λ L 3 \lambda L_3 λ L 3 and using a L 3 ≡ − s B n aL_3\equiv-sB_n a L 3 ≡ − s B n and s 2 ≡ 1 s^2\equiv1 s 2 ≡ 1 : a 2 L 3 2 ( m 2 + n 2 − 1 ) ≡ B n 2 ( m 2 + n 2 − 1 ) a^2L_3^2(m^2+n^2-1)\equiv B_n^2(m^2+n^2-1) a 2 L 3 2 ( m 2 + n 2 − 1 ) ≡ B n 2 ( m 2 + n 2 − 1 ) , 2 λ a s L 3 Θ ≡ − 2 λ B n Θ 2\lambda asL_3\Theta\equiv-2\lambda B_n\Theta 2 λa s L 3 Θ ≡ − 2 λ B n Θ , and − 4 λ L 3 ( n 3 − n ) ( 1 + λ n ) = − 4 L 3 ( λ ( n 3 − n ) + n 4 − n 2 ) -4\lambda L_3(n^3-n)(1+\lambda n)=-4L_3(\lambda(n^3-n)+n^4-n^2) − 4 λ L 3 ( n 3 − n ) ( 1 + λn ) = − 4 L 3 ( λ ( n 3 − n ) + n 4 − n 2 ) . Together with 3 L 3 ( k c − H 2 ) 3L_3(k_c-H_2) 3 L 3 ( k c − H 2 ) this gives
λ L 3 ( A z + 3 λ ( k c − H 2 ) ) = B n 2 ( m 2 + n 2 − 1 ) − 2 λ B n Θ − 4 L 3 ( λ ( n 3 − n ) + n 4 − n 2 ) + 3 L 3 ( k c − H 2 ) = P z ( m o d c ) , \lambda L_3\bigl(A_z+3\lambda(k_c-H_2)\bigr)=B_n^2(m^2+n^2-1)-2\lambda B_n\Theta-4L_3\bigl(\lambda(n^3-n)+n^4-n^2\bigr)+3L_3(k_c-H_2)=P_z\pmod c, λ L 3 ( A z + 3 λ ( k c − H 2 ) ) = B n 2 ( m 2 + n 2 − 1 ) − 2 λ B n Θ − 4 L 3 ( λ ( n 3 − n ) + n 4 − n 2 ) + 3 L 3 ( k c − H 2 ) = P z ( mod c ) ,
the last equality being an identity of polynomials in m , n , k c m,n,k_c m , n , k c for each λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } . Since 24 c ∣ A z + 3 λ ( k c − H 2 ) 24c\mid A_z+3\lambda(k_c-H_2) 24 c ∣ A z + 3 λ ( k c − H 2 ) by Lemma 3.3 , c ∣ P z c\mid P_z c ∣ P z .
We next prove the bracket. Grouping terms by degree and using λ 2 = 1 \lambda^2=1 λ 2 = 1 ,
Ψ ( n , m ) − Ψ ( m , n ) = 3 ( m 4 − n 4 ) + 3 λ ( m 3 + m 2 n − m n 2 − n 3 ) = 3 ( m 2 − n 2 ) ( m 2 + n 2 + λ ( m + n ) ) = 3 H L 3 . \begin{aligned}
\Psi(n,m)-\Psi(m,n)=3(m^4-n^4)+3\lambda(m^3+m^2n-mn^2-n^3) \\
&=3(m^2-n^2)(m^2+n^2+\lambda(m+n))=3HL_3.
\end{aligned} Ψ ( n , m ) − Ψ ( m , n ) = 3 ( m 4 − n 4 ) + 3 λ ( m 3 + m 2 n − m n 2 − n 3 ) = 3 ( m 2 − n 2 ) ( m 2 + n 2 + λ ( m + n )) = 3 H L 3 .
Therefore P z = ( 1 − k c H ) Ψ ( m , n ) + k c H Ψ ( n , m ) P_z=\left(1-\frac{k_c}{H}\right)\Psi(m,n)+\frac{k_c}{H}\Psi(n,m) P z = ( 1 − H k c ) Ψ ( m , n ) + H k c Ψ ( n , m ) . Since 0 < k c < H 0<k_c<H 0 < k c < H and L 3 > 0 L_3>0 L 3 > 0 , Ψ ( m , n ) < Ψ ( n , m ) \Psi(m,n)<\Psi(n,m) Ψ ( m , n ) < Ψ ( n , m ) and Ψ ( m , n ) < P z < Ψ ( n , m ) \Psi(m,n)<P_z<\Psi(n,m) Ψ ( m , n ) < P z < Ψ ( n , m ) . Writing n = m − e n=m-e n = m − e ,
Ψ ( n , m ) = 4 m 2 ( m + λ ) 2 − e ( 2 m ( m + λ ) ( m + 2 λ ) + ( 3 m 2 − λ m − 1 ) e − 4 m e 2 + e 3 ) . \Psi(n,m)=4m^2(m+\lambda)^2-e\bigl(2m(m+\lambda)(m+2\lambda)+(3m^2-\lambda m-1)e-4me^2+e^3\bigr). Ψ ( n , m ) = 4 m 2 ( m + λ ) 2 − e ( 2 m ( m + λ ) ( m + 2 λ ) + ( 3 m 2 − λm − 1 ) e − 4 m e 2 + e 3 ) .
Because n < m < 2 n n<m<2n n < m < 2 n , we have 1 ≤ e < m / 2 1\le e<m/2 1 ≤ e < m /2 , so 4 m e < 2 m 2 4me<2m^2 4 m e < 2 m 2 and the bracket exceeds 2 m ( m − 1 ) ( m − 2 ) + ( m 2 − m − 1 ) e > 0 2m(m-1)(m-2)+(m^2-m-1)e>0 2 m ( m − 1 ) ( m − 2 ) + ( m 2 − m − 1 ) e > 0 , as m ≥ 4 m\ge4 m ≥ 4 . Hence Ψ ( n , m ) < 4 m 2 ( m + λ ) 2 \Psi(n,m)<4m^2(m+\lambda)^2 Ψ ( n , m ) < 4 m 2 ( m + λ ) 2 .
We finally treat the case m ≤ 3 2 n m\le\frac{3}{2}n m ≤ 2 3 n . For real u ≥ n u\ge n u ≥ n ,
∂ 2 Ψ ∂ u 2 ( u , n ) = − 12 u 2 + 2 ( 3 n 2 + λ n + 1 ) ≤ − 6 n 2 + 2 n + 2 < 0 , (3) \frac{\partial^2\Psi}{\partial u^2}(u,n)=-12u^2+2(3n^2+\lambda n+1)\le-6n^2+2n+2<0, \tag*{(3)} ∂ u 2 ∂ 2 Ψ ( u , n ) = − 12 u 2 + 2 ( 3 n 2 + λn + 1 ) ≤ − 6 n 2 + 2 n + 2 < 0 , (3)
so u ↦ Ψ ( u , n ) u \mapsto\Psi(u,n) u ↦ Ψ ( u , n ) is strictly concave on [ n , 3 2 n ] [n,\frac{3}{2}n] [ n , 2 3 n ] . At the endpoints Ψ ( n , n ) = 4 n 2 ( n + λ ) 2 > 0 \Psi(n,n)=4n^2(n+\lambda)^2>0 Ψ ( n , n ) = 4 n 2 ( n + λ ) 2 > 0 and 16 Ψ ( 3 2 n , n ) = n 2 ( 59 n 2 + 180 λ n + 100 ) > 0 16\Psi(\frac{3}{2}n,n)=n^2(59n^2+180\lambda n+100)>0 16Ψ ( 2 3 n , n ) = n 2 ( 59 n 2 + 180 λn + 100 ) > 0 , the last because 59 n 2 − 180 n + 100 > 0 59n^2-180n+100>0 59 n 2 − 180 n + 100 > 0 for n ≥ 3 n\ge3 n ≥ 3 . Hence Ψ ( m , n ) > 0 \Psi(m,n)>0 Ψ ( m , n ) > 0 for n < m ≤ 3 2 n n<m\le\frac{3}{2}n n < m ≤ 2 3 n , and then P z > Ψ ( m , n ) > 0 P_z>\Psi(m,n)>0 P z > Ψ ( m , n ) > 0 is a positive multiple of c c c , so c ≤ P z < 4 m 2 ( m + λ ) c\le P_z<4m^2(m+\lambda) c ≤ P z < 4 m 2 ( m + λ ) . For the positivity of Ψ ( m , n ) \Psi(m,n) Ψ ( m , n ) , the condition m ≤ 3 2 n m\le\frac{3}{2}n m ≤ 2 3 n cannot be replaced by m < 2 n m<2n m < 2 n : Ψ ( 50 , 26 ) = − 7344 \Psi(50,26)=-7344 Ψ ( 50 , 26 ) = − 7344 for λ = 1 \lambda=1 λ = 1 and Ψ ( 18 , 10 ) = − 432 \Psi(18,10)=-432 Ψ ( 18 , 10 ) = − 432 for λ = − 1 \lambda=-1 λ = − 1 .
Lemma 3.5. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then a + b ≤ F λ ( m , n ) a+b\le F_\lambda(m,n) a + b ≤ F λ ( m , n ) , where
F 1 ( m , n ) = 4 m 2 n 2 + m 3 + m 2 n − m 2 + m n 2 − 2 m + n 3 − n 2 , F_1(m,n)=4m^2n^2+m^3+m^2n-m^2+mn^2-2m+n^3-n^2, F 1 ( m , n ) = 4 m 2 n 2 + m 3 + m 2 n − m 2 + m n 2 − 2 m + n 3 − n 2 , F − 1 ( m , n ) = 4 m 2 n 2 − 2 m 3 − 2 m 2 n − 2 m n 2 + 2 m n + m − 2 n 3 + n . F_{-1}(m,n)=4m^2n^2-2m^3-2m^2n-2mn^2+2mn+m-2n^3+n. F − 1 ( m , n ) = 4 m 2 n 2 − 2 m 3 − 2 m 2 n − 2 m n 2 + 2 mn + m − 2 n 3 + n .
Proof. The two cofactor triples ( Q a , Q b , Q r ) (Q_a,Q_b,Q_r) ( Q a , Q b , Q r ) and ( Q a ′ , Q b ′ , Q r ′ ) (Q'_a,Q'_b,Q'_r) ( Q a ′ , Q b ′ , Q r ′ ) defined below are cross products of two integer vectors orthogonal to ( a , b , r ) (a,b,r) ( a , b , r ) , hence integer multiples of ( a , b , r ) (a,b,r) ( a , b , r ) , with positive multipliers g , g ′ g,g' g , g ′ ; then a + b ≤ g ′ a + g b a+b\le g'a+gb a + b ≤ g ′ a + g b , and the identity for g ′ a + g b g'a+gb g ′ a + g b , with 0 < k c < H 0<k_c<H 0 < k c < H , bounds it by a polynomial of degree 4 in m , n m,n m , n . With the index polynomials and the order-two quotients k a , k b , k c k_a,k_b,k_c k a , k b , k c of Convention 3.1, the cofactor triples are given by
Q a = k b ( B n − 2 k c ) + M 2 k c , Q b = L 2 B n + M 2 H − ( 2 L 2 + M 2 ) k c , Q r = k b ( k c − H ) + L 2 k c , Q_a=k_b(B_n-2k_c)+M_2k_c,\qquad Q_b=L_2B_n+M_2H-(2L_2+M_2)k_c,\qquad Q_r=k_b(k_c-H)+L_2k_c, Q a = k b ( B n − 2 k c ) + M 2 k c , Q b = L 2 B n + M 2 H − ( 2 L 2 + M 2 ) k c , Q r = k b ( k c − H ) + L 2 k c ,
Q a ′ = ( 2 L 1 + M 1 ) k c − L 1 B n , Q b ′ = M 1 ( H − k c ) + k a ( 2 k c − B n ) , Q r ′ = ( − k a ) k c + L 1 ( H − k c ) . Q'_a=(2L_1+M_1)k_c-L_1B_n,\qquad Q'_b=M_1(H-k_c)+k_a(2k_c-B_n),\qquad Q'_r=(-k_a)k_c+L_1(H-k_c). Q a ′ = ( 2 L 1 + M 1 ) k c − L 1 B n , Q b ′ = M 1 ( H − k c ) + k a ( 2 k c − B n ) , Q r ′ = ( − k a ) k c + L 1 ( H − k c ) .
By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n , so L 1 , L 2 , L 3 , M 1 , M 2 , H , B n L_1,L_2,L_3,M_1,M_2,H,B_n L 1 , L 2 , L 3 , M 1 , M 2 , H , B n are positive integers; by item 2 of Lemma 3.2 , k a , k b , k c ∈ Z k_a,k_b,k_c\in\mathbb{Z} k a , k b , k c ∈ Z with k a < 0 k_a<0 k a < 0 , k b < 0 k_b<0 k b < 0 , 0 < k c < H 0<k_c<H 0 < k c < H . In particular Q a , Q b , Q r , Q a ′ , Q b ′ , Q r ′ Q_a,Q_b,Q_r,Q'_a,Q'_b,Q'_r Q a , Q b , Q r , Q a ′ , Q b ′ , Q r ′ are integers.
We first prove that the two cofactor triples are positive integer multiples of ( a , b , r ) (a,b,r) ( a , b , r ) . By the definitions of k a , k b , k c k_a,k_b,k_c k a , k b , k c and c = a + b + 2 r c=a+b+2r c = a + b + 2 r , the vector ( a , b , r ) (a,b,r) ( a , b , r ) is orthogonal to each of
v 1 = ( k a , − L 1 , M 1 ) , v 2 = ( − L 2 , k b , M 2 ) , v 3 = ( k c − H , k c , 2 k c − B n ) , \mathbf{v}_1=(k_a,-L_1,M_1),\qquad\mathbf{v}_2=(-L_2,k_b,M_2),\qquad\mathbf{v}_3=(k_c-H,k_c,2k_c-B_n), v 1 = ( k a , − L 1 , M 1 ) , v 2 = ( − L 2 , k b , M 2 ) , v 3 = ( k c − H , k c , 2 k c − B n ) ,
and ( Q a , Q b , Q r ) = − ( v 2 × v 3 ) (Q_a,Q_b,Q_r)=-(\mathbf{v}_2\times\mathbf{v}_3) ( Q a , Q b , Q r ) = − ( v 2 × v 3 ) , ( Q a ′ , Q b ′ , Q r ′ ) = − ( v 1 × v 3 ) (Q'_a,Q'_b,Q'_r)=-(\mathbf{v}_1\times\mathbf{v}_3) ( Q a ′ , Q b ′ , Q r ′ ) = − ( v 1 × v 3 ) . Direct expansion using a k a = b L 1 − r M 1 ak_a=bL_1-rM_1 a k a = b L 1 − r M 1 , b k b = a L 2 − r M 2 bk_b=aL_2-rM_2 b k b = a L 2 − r M 2 and c k c = a H + r B n ck_c=aH+rB_n c k c = a H + r B n gives
b Q r = r Q b , a Q b = b Q a , a Q r ′ = r Q a ′ , b Q a ′ = a Q b ′ . bQ_r=rQ_b,\qquad aQ_b=bQ_a,\qquad aQ'_r=rQ'_a,\qquad bQ'_a=aQ'_b. b Q r = r Q b , a Q b = b Q a , a Q r ′ = r Q a ′ , b Q a ′ = a Q b ′ .
Since k a < 0 k_a<0 k a < 0 , k b < 0 k_b<0 k b < 0 , 0 < k c < H 0<k_c<H 0 < k c < H and L 1 , L 2 > 0 L_1,L_2>0 L 1 , L 2 > 0 , both summands of Q r = ( − k b ) ( H − k c ) + L 2 k c Q_r=(-k_b)(H-k_c)+L_2k_c Q r = ( − k b ) ( H − k c ) + L 2 k c and of Q r ′ = ( − k a ) k c + L 1 ( H − k c ) Q'_r=(-k_a)k_c+L_1(H-k_c) Q r ′ = ( − k a ) k c + L 1 ( H − k c ) are positive, so Q r , Q r ′ > 0 Q_r,Q'_r>0 Q r , Q r ′ > 0 . As r 2 − a b = 1 r^2-ab=1 r 2 − ab = 1 , gcd ( a , r ) = gcd ( b , r ) = 1 \gcd(a,r)=\gcd(b,r)=1 g cd( a , r ) = g cd( b , r ) = 1 ; hence r ∣ Q r r\mid Q_r r ∣ Q r (from b Q r = r Q b bQ_r=rQ_b b Q r = r Q b ) and r ∣ Q r ′ r\mid Q'_r r ∣ Q r ′ (from a Q r ′ = r Q a ′ aQ'_r=rQ'_a a Q r ′ = r Q a ′ ), and with g = Q r / r ≥ 1 g=Q_r/r\ge1 g = Q r / r ≥ 1 , g ′ = Q r ′ / r ≥ 1 g'=Q'_r/r\ge1 g ′ = Q r ′ / r ≥ 1 the four identities give ( Q a , Q b , Q r ) = g ( a , b , r ) (Q_a,Q_b,Q_r)=g(a,b,r) ( Q a , Q b , Q r ) = g ( a , b , r ) and ( Q a ′ , Q b ′ , Q r ′ ) = g ′ ( a , b , r ) (Q'_a,Q'_b,Q'_r)=g'(a,b,r) ( Q a ′ , Q b ′ , Q r ′ ) = g ′ ( a , b , r ) .
We next derive the identity for g ′ a + g b g'a+gb g ′ a + g b . We have 2 L 1 + M 1 = 2 l 2 − l + λ m 2L_1+M_1=2l^2-l+\lambda m 2 L 1 + M 1 = 2 l 2 − l + λm and 2 L 2 + M 2 = 2 l 2 − l + λ n 2L_2+M_2=2l^2-l+\lambda n 2 L 2 + M 2 = 2 l 2 − l + λn , so ( 2 L 1 + M 1 ) − ( 2 L 2 + M 2 ) = λ ( m − n ) (2L_1+M_1)-(2L_2+M_2)=\lambda(m-n) ( 2 L 1 + M 1 ) − ( 2 L 2 + M 2 ) = λ ( m − n ) , while L 2 − L 1 = L 3 L_2-L_1=L_3 L 2 − L 1 = L 3 . Adding Q a ′ = g ′ a Q'_a=g'a Q a ′ = g ′ a and Q b = g b Q_b=gb Q b = g b cancels l 2 l^2 l 2 and gives
g ′ a + g b = L 3 B n + M 2 H + λ ( m − n ) k c . g'a+gb=L_3B_n+M_2H+\lambda(m-n)k_c. g ′ a + g b = L 3 B n + M 2 H + λ ( m − n ) k c .
We finally prove a + b ≤ F λ ( m , n ) a+b\le F_\lambda(m,n) a + b ≤ F λ ( m , n ) . Since g , g ′ ≥ 1 g,g'\ge1 g , g ′ ≥ 1 , a + b ≤ g ′ a + g b a+b\le g'a+gb a + b ≤ g ′ a + g b . Since l ≥ m + 1 l\ge m+1 l ≥ m + 1 , M 2 ≤ 2 n 2 + λ n − m − 1 M_2\le2n^2+\lambda n-m-1 M 2 ≤ 2 n 2 + λn − m − 1 ; and 1 ≤ k c ≤ H − 1 1\le k_c\le H-1 1 ≤ k c ≤ H − 1 . When λ = 1 \lambda=1 λ = 1 , using ( m − n ) k c ≤ ( m − n ) ( H − 1 ) (m-n)k_c\le(m-n)(H-1) ( m − n ) k c ≤ ( m − n ) ( H − 1 ) ,
g ′ a + g b ≤ L 3 B n + ( 2 n 2 − 1 ) H − ( m − n ) = F 1 ( m , n ) . (4) g'a+gb\le L_3B_n+(2n^2-1)H-(m-n)=F_1(m,n). \tag*{(4)} g ′ a + g b ≤ L 3 B n + ( 2 n 2 − 1 ) H − ( m − n ) = F 1 ( m , n ) . (4)
When λ = − 1 \lambda=-1 λ = − 1 , using − ( m − n ) k c < 0 -(m-n)k_c<0 − ( m − n ) k c < 0 and M 2 ≤ 2 n 2 − n − m − 1 M_2\le2n^2-n-m-1 M 2 ≤ 2 n 2 − n − m − 1 ,
g ′ a + g b < L 3 B n + ( 2 n 2 − n − m − 1 ) H = F − 1 ( m , n ) . g'a+gb<L_3B_n+(2n^2-n-m-1)H=F_{-1}(m,n). g ′ a + g b < L 3 B n + ( 2 n 2 − n − m − 1 ) H = F − 1 ( m , n ) .
Theorem 3.6. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then
c < 4 m 2 ( m + λ ) 2 < 8 m 4 . c<4m^2(m+\lambda)^2<8m^4. c < 4 m 2 ( m + λ ) 2 < 8 m 4 .
Proof. The first inequality comes from the quartic bracket of Lemma 3.4 when m ≤ 3 2 n m\le\frac{3}{2}n m ≤ 2 3 n , and from the quartic bound for a + b a+b a + b of Lemma 3.5 otherwise; the second from m ≥ 4 m\ge4 m ≥ 4 . By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n and m ≥ 4 m\ge4 m ≥ 4 , so m ≤ 2 n − 2 m\le2n-2 m ≤ 2 n − 2 .
If m ≤ 3 2 n m\le\frac{3}{2}n m ≤ 2 3 n , then Lemma 3.4 gives c ≤ P z < 4 m 2 ( m + λ ) 2 c\le P_z<4m^2(m+\lambda)^2 c ≤ P z < 4 m 2 ( m + λ ) 2 .
Suppose now m > 3 2 n m>\frac{3}{2}n m > 2 3 n , so n < 2 3 m n<\frac{2}{3}m n < 3 2 m . By item 4 of Lemma 2.4 , c ≤ 2 ( a + b ) c\le2(a+b) c ≤ 2 ( a + b ) , and by Lemma 3.5 , a + b ≤ F λ ( m , n ) a+b\le F_\lambda(m,n) a + b ≤ F λ ( m , n ) .
When λ = 1 \lambda=1 λ = 1 : 4 m 2 n 2 < 16 9 m 4 4m^2n^2<\frac{16}{9}m^4 4 m 2 n 2 < 9 16 m 4 and m 3 + m 2 n + m n 2 + n 3 < ( 1 + 2 3 + 4 9 + 8 27 ) m 3 = 65 27 m 3 m^3+m^2n+mn^2+n^3<\left(1+\frac{2}{3}+\frac{4}{9}+\frac{8}{27}\right)m^3=\frac{65}{27}m^3 m 3 + m 2 n + m n 2 + n 3 < ( 1 + 3 2 + 9 4 + 27 8 ) m 3 = 27 65 m 3 , while the remaining terms of F 1 ( m , n ) F_1(m,n) F 1 ( m , n ) are negative, so
c ≤ 2 F 1 ( m , n ) < 32 9 m 4 + 130 27 m 3 < 4 m 4 + 8 m 3 < 4 m 2 ( m + 1 ) 2 . c\le2F_1(m,n)<\frac{32}{9}m^4+\frac{130}{27}m^3<4m^4+8m^3<4m^2(m+1)^2. c ≤ 2 F 1 ( m , n ) < 9 32 m 4 + 27 130 m 3 < 4 m 4 + 8 m 3 < 4 m 2 ( m + 1 ) 2 .
When λ = − 1 \lambda=-1 λ = − 1 : − 2 m 2 n + 2 m n ≤ 0 -2m^2n+2mn\le0 − 2 m 2 n + 2 mn ≤ 0 , m − 2 m n 2 < 0 m-2mn^2<0 m − 2 m n 2 < 0 and n − 2 n 3 < 0 n-2n^3<0 n − 2 n 3 < 0 , so F − 1 ( m , n ) < 4 m 2 n 2 − 2 m 3 < 16 9 m 4 − 2 m 3 F_{-1}(m,n)<4m^2n^2-2m^3<\frac{16}{9}m^4-2m^3 F − 1 ( m , n ) < 4 m 2 n 2 − 2 m 3 < 9 16 m 4 − 2 m 3 . Moreover 3 2 n < m ≤ 2 n − 2 \frac{3}{2}n<m\le2n-2 2 3 n < m ≤ 2 n − 2 with n ≥ 3 n\ge3 n ≥ 3 forces m ≥ 8 m\ge8 m ≥ 8 : for n ≤ 4 n\le4 n ≤ 4 no integer m m m satisfies 3 2 n < m ≤ 2 n − 2 \frac{3}{2}n<m\le2n-2 2 3 n < m ≤ 2 n − 2 , and for n ≥ 5 n\ge5 n ≥ 5 , m > 15 2 m>\frac{15}{2} m > 2 15 . Hence
4 m 2 ( m − 1 ) 2 − c > 4 m 2 ( m − 1 ) 2 − 2 ( 16 9 m 4 − 2 m 3 ) = 4 9 m 2 ( m 2 − 9 m + 9 ) > 0. 4m^2(m-1)^2-c>4m^2(m-1)^2-2\left(\frac{16}{9}m^4-2m^3\right)=\frac{4}{9}m^2(m^2-9m+9)>0. 4 m 2 ( m − 1 ) 2 − c > 4 m 2 ( m − 1 ) 2 − 2 ( 9 16 m 4 − 2 m 3 ) = 9 4 m 2 ( m 2 − 9 m + 9 ) > 0.
Here m 2 − 9 m + 9 = m ( m − 9 ) + 9 ≥ 1 m^2-9m+9=m(m-9)+9\ge1 m 2 − 9 m + 9 = m ( m − 9 ) + 9 ≥ 1 , as m ≥ 8 m\ge8 m ≥ 8 . So c < 4 m 2 ( m + λ ) 2 c<4m^2(m+\lambda)^2 c < 4 m 2 ( m + λ ) 2 in all cases.
Finally, m ≥ 4 m\ge4 m ≥ 4 gives m 2 − 2 m − 1 > 0 m^2-2m-1>0 m 2 − 2 m − 1 > 0 , that is ( m + 1 ) 2 < 2 m 2 (m+1)^2<2m^2 ( m + 1 ) 2 < 2 m 2 ; hence 4 m 2 ( m + λ ) 2 ≤ 4 m 2 ( m + 1 ) 2 < 8 m 4 4m^2(m+\lambda)^2\le4m^2(m+1)^2<8m^4 4 m 2 ( m + λ ) 2 ≤ 4 m 2 ( m + 1 ) 2 < 8 m 4 .
The next section does not use it: for close units the gap inequalities take its place.
Close Pell units force regularity
This section proves the criterion for close Pell units (Theorem 4.5 ) and the ratio window that follows from it (Corollary 4.6 ). Both conditions of the criterion rest on one mechanism, applied to two pairs of units. When b / a b/a b / a is large, β / α \beta/\alpha β / α is close to 1 1 1 and Λ x \Lambda_x Λ x is a linear form in log ( β / α ) \log(\beta/\alpha) log ( β / α ) and log ( α 2 ( l − m ) D λ ) \log(\alpha^{2(l-m)}D_\lambda) log ( α 2 ( l − m ) D λ ) ; when b − a b-a b − a is small compared with a c \sqrt{ac} a c , γ / β \gamma/\beta γ / β is close to 1 1 1 and Λ z \Lambda_z Λ z is a linear form in log ( γ / β ) \log(\gamma/\beta) log ( γ / β ) and log ( β 2 ( m − n ) E λ ) \log(\beta^{2(m-n)}E_\lambda) log ( β 2 ( m − n ) E λ ) . The mechanism is proved once, for two abstract numbers (Lemma 4.4 ), and the criterion checks its conditions in the two cases. All the numbers involved lie in the real biquadratic field K = Q ( a b , a c ) K=\mathbb{Q}(\sqrt{ab},\sqrt{ac}) K = Q ( ab , a c ) , where we first compute their sizes and heights.
Definition 4.1. For a nonzero algebraic number η \eta η of degree D = [ Q ( η ) : Q ] \mathcal{D}=[\mathbb{Q}(\eta):\mathbb{Q}] D = [ Q ( η ) : Q ] , let a η ∏ j = 1 D ( X − η ( j ) ) a_\eta\prod_{j=1}^{\mathcal{D}}(X-\eta^{(j)}) a η ∏ j = 1 D ( X − η ( j ) ) , with a η > 0 a_\eta>0 a η > 0 , be its primitive irreducible polynomial over Z \mathbb{Z} Z . Its absolute logarithmic height is
h ( η ) = 1 D ( log a η + ∑ j = 1 D log max { 1 , ∣ η ( j ) ∣ } ) . h(\eta)=\frac{1}{\mathcal{D}}\left(\log a_\eta+\sum_{j=1}^{\mathcal{D}}\log\max\{1,|\eta^{(j)}|\}\right). h ( η ) = D 1 ( log a η + j = 1 ∑ D log max { 1 , ∣ η ( j ) ∣ } ) .
Lemma 4.2. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with D λ , E λ D_\lambda,E_\lambda D λ , E λ as in Definition 2.16 , and put K = Q ( a b , a c ) \mathcal{K}=\mathbb{Q}(\sqrt{ab},\sqrt{ac}) K = Q ( ab , a c ) . Then:
1 2 log c < log α < log c − log 2 \frac{1}{2}\log c < \log\alpha< \log c-\log2 2 1 log c < log α < log c − log 2 , 1 2 log c + log 2 < log β < log c \frac{1}{2}\log c+\log2 < \log\beta< \log c 2 1 log c + log 2 < log β < log c , and log β − log α < log 2 + 4 / c \log\beta-\log\alpha< \log2+4/c log β − log α < log 2 + 4/ c ;
K \mathcal{K} K is a real biquadratic field that contains α , β , γ , D λ , E λ \alpha,\beta,\gamma,D_\lambda,E_\lambda α , β , γ , D λ , E λ , and for ( e 1 , e 2 ) ∈ { − 1 , 1 } 2 (e_1,e_2)\in\{-1,1\}^2 ( e 1 , e 2 ) ∈ { − 1 , 1 } 2 it has an embedding with α ↦ α e 1 \alpha\mapsto\alpha^{e_1} α ↦ α e 1 , β ↦ β e 2 \beta\mapsto\beta^{e_2} β ↦ β e 2 , γ ↦ γ e 1 e 2 \gamma\mapsto\gamma^{e_1e_2} γ ↦ γ e 1 e 2 ;
for integers i , j i,j i , j , h ( β i α j ) = 1 2 max ( ∣ i ∣ log β , ∣ j ∣ log α ) \mathrm{h}(\beta^i\alpha^j)=\frac{1}{2}\max(|i|\log\beta,|j|\log\alpha) h ( β i α j ) = 2 1 max ( ∣ i ∣ log β , ∣ j ∣ log α ) , in particular h ( α ) = 1 2 log α \mathrm{h}(\alpha)=\frac{1}{2}\log\alpha h ( α ) = 2 1 log α and h ( β / α ) = 1 2 log β \mathrm{h}(\beta/\alpha)=\frac{1}{2}\log\beta h ( β / α ) = 2 1 log β ; and h ( γ / β ) = 1 2 log γ \mathrm{h}(\gamma/\beta)=\frac{1}{2}\log\gamma h ( γ / β ) = 2 1 log γ ; moreover β / α \beta/\alpha β / α and γ / β \gamma/\beta γ / β have degree 4 4 4 over Q \mathbb{Q} Q ;
0 < log D λ < min { log 4 , 2 a / b } 0<\log D_\lambda<\min\{\log4,2\sqrt{a/b}\} 0 < log D λ < min { log 4 , 2 a / b } , N K / Q ( D λ ) = ( b ( c − b ) b ( c − a ) ) 2 ∈ ( 0 , 1 ) N_{\mathcal{K}/\mathbb{Q}}(D_\lambda)=\left(\frac{b(c-b)}{b(c-a)}\right)^2\in(0,1) N K / Q ( D λ ) = ( b ( c − a ) b ( c − b ) ) 2 ∈ ( 0 , 1 ) , and h ( D λ ) < log c \mathrm{h}(D_\lambda)<\log c h ( D λ ) < log c ;
0 < log E λ < 2 ( b − a ) / a c 0<\log E_\lambda<2(b-a)/\sqrt{ac} 0 < log E λ < 2 ( b − a ) / a c , N K / Q ( E λ ) = ( b ( c − a ) a ( c − b ) ) 2 > 1 N_{\mathcal{K}/\mathbb{Q}}(E_\lambda)=\left(\frac{b(c-a)}{a(c-b)}\right)^2>1 N K / Q ( E λ ) = ( a ( c − b ) b ( c − a ) ) 2 > 1 , and h ( E λ ) < log β \mathrm{h}(E_\lambda)<\log\beta h ( E λ ) < log β if b ≤ 3 2 a b\leq\frac{3}{2}a b ≤ 2 3 a ;
The numbers α \alpha α , β / α \beta/\alpha β / α and D λ D_\lambda D λ exceed 1 1 1 and are multiplicatively independent, and so do γ / β \gamma/\beta γ / β and β 2 ( m − n ) E λ \beta^{2(m-n)}E_\lambda β 2 ( m − n ) E λ .
Proof. The size estimates in items 1, 4 and 5 are elementary consequences of the identities of an Euler triple; the norms, heights, degrees and independence statements come from the four embeddings of K \mathcal{K} K of item 2, which invert α \alpha α and β \beta β independently. We use the identities of Lemma 2.4 throughout: s = a + r s=a+r s = a + r , t = b + r t=b+r t = b + r , s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 , t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 , a < r < b a<r<b a < r < b and 4 a < c ≤ 2 ( a + b ) 4a<c\leq2(a+b) 4 a < c ≤ 2 ( a + b ) ; as c = a + b + 2 r c=a+b+2r c = a + b + 2 r , these give 2 r ≤ a + b 2r\leq a+b 2 r ≤ a + b , c ≥ 4 r c\geq4r c ≥ 4 r and c < 4 b c<4b c < 4 b , and c > 4 a c>4a c > 4 a gives a < 1 2 c \sqrt{a}<\frac{1}{2}\sqrt{c} a < 2 1 c . By Lemma 2.14, n < m n<m n < m .
We first prove item 1. Since c ≥ 4 r c\geq4r c ≥ 4 r , α < 2 r ≤ c / 2 \alpha<2r\leq c/2 α < 2 r ≤ c /2 , so log α < log c − log 2 \log\alpha<\log c-\log2 log α < log c − log 2 . Next α 2 = 2 a b + 1 + 2 r a b \alpha^2=2ab+1+2r\sqrt{ab} α 2 = 2 ab + 1 + 2 r ab , so α 2 − c = ( 2 a b − a − b ) + 1 + 2 r ( a b − 1 ) > 0 \alpha^2-c=(2ab-a-b)+1+2r(\sqrt{ab}-1)>0 α 2 − c = ( 2 ab − a − b ) + 1 + 2 r ( ab − 1 ) > 0 and log α > 1 2 log c \log\alpha>\frac{1}{2}\log c log α > 2 1 log c . Since s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 ,
2 c ≤ 2 a c < β < 2 s = 2 a + 2 r = c − ( b − a ) < c , 2\sqrt{c}\leq2\sqrt{ac}<\beta<2s=2a+2r=c-(b-a)<c, 2 c ≤ 2 a c < β < 2 s = 2 a + 2 r = c − ( b − a ) < c ,
so 1 2 log c + log 2 < log β < log c \frac{1}{2}\log c+\log2<\log\beta<\log c 2 1 log c + log 2 < log β < log c . Finally α = 2 r − α − 1 > 2 r − r − 1 \alpha=2r-\alpha^{-1}>2r-r^{-1} α = 2 r − α − 1 > 2 r − r − 1 and s / r = 1 + a / r < 2 s/r=1+a/r<2 s / r = 1 + a / r < 2 , so, as 4 r 2 = 4 a b + 4 > 2 ( a + b ) ≥ c 4r^2=4ab+4>2(a+b)\geq c 4 r 2 = 4 ab + 4 > 2 ( a + b ) ≥ c ,
log β − log α < log 2 s 2 r − r − 1 = log s r − log ( 1 − 1 2 r 2 ) < log 2 + 1 r 2 < log 2 + 4 c , \log\beta-\log\alpha<\log\frac{2s}{2r-r^{-1}}=\log\frac{s}{r}-\log\left(1-\frac{1}{2r^2}\right)<\log2+\frac{1}{r^2}<\log2+\frac{4}{c}, log β − log α < log 2 r − r − 1 2 s = log r s − log ( 1 − 2 r 2 1 ) < log 2 + r 2 1 < log 2 + c 4 ,
using − log ( 1 − x ) ≤ x / ( 1 − x ) ≤ 2 x -\log(1-x)\leq x/(1-x)\leq2x − log ( 1 − x ) ≤ x / ( 1 − x ) ≤ 2 x for 0 < x ≤ 1 2 0<x\leq\frac{1}{2} 0 < x ≤ 2 1 .
We next prove item 2. The integers a b ab ab , a c ac a c , b c bc b c are not squares (Lemma 2.4 , item 5), and a b ⋅ a c = a 2 ⋅ b c ab\cdot ac=a^2\cdot bc ab ⋅ a c = a 2 ⋅ b c ; hence K \mathcal{K} K is a real biquadratic field with the four embeddings σ e 1 , e 2 : a b ↦ e 1 a b \sigma_{e_1,e_2}:\sqrt{ab}\mapsto e_1\sqrt{ab} σ e 1 , e 2 : ab ↦ e 1 ab , a c ↦ e 2 a c \sqrt{ac}\mapsto e_2\sqrt{ac} a c ↦ e 2 a c , and b c = a b a c / a \sqrt{bc}=\sqrt{ab}\sqrt{ac}/a b c = ab a c / a lies in K \mathcal{K} K and is sent to e 1 e 2 b c e_1e_2\sqrt{bc} e 1 e 2 b c . The units α = r + a b \alpha=r+\sqrt{ab} α = r + ab , β = s + a c \beta=s+\sqrt{ac} β = s + a c and γ = t + b c \gamma=t+\sqrt{bc} γ = t + b c lie in K \mathcal{K} K , and α − 1 = r − a b \alpha^{-1}=r-\sqrt{ab} α − 1 = r − ab , β − 1 = s − a c \beta^{-1}=s-\sqrt{ac} β − 1 = s − a c , γ − 1 = t − b c \gamma^{-1}=t-\sqrt{bc} γ − 1 = t − b c (Lemma 2.4 , item 5); so σ e 1 , e 2 \sigma_{e_1,e_2} σ e 1 , e 2 sends α ↦ α e 1 \alpha\mapsto\alpha^{e_1} α ↦ α e 1 , β ↦ β e 2 \beta\mapsto\beta^{e_2} β ↦ β e 2 and γ ↦ γ e 1 e 2 \gamma\mapsto\gamma^{e_1e_2} γ ↦ γ e 1 e 2 . The numbers D λ D_\lambda D λ and E λ E_\lambda E λ are quotients of sums of ± a b \pm\sqrt{ab} ± ab , ± a c \pm\sqrt{ac} ± a c , b c \sqrt{bc} b c , so they lie in K \mathcal{K} K as well.
We next prove item 3. By item 2, the conjugates of the unit β i α j \beta^i\alpha^j β i α j are exp ( ± i log β ± j log α ) \exp(\pm i\log\beta\pm j\log\alpha) exp ( ± i log β ± j log α ) with independent signs, so
4 h ( β i α j ) = ∣ i log β + j log α ∣ + ∣ i log β − j log α ∣ = 2 max ( ∣ i ∣ log β , ∣ j ∣ log α ) . 4\mathrm{h}(\beta^i\alpha^j)=|i\log\beta+j\log\alpha|+|i\log\beta-j\log\alpha|=2\max(|i|\log\beta,|j|\log\alpha). 4 h ( β i α j ) = ∣ i log β + j log α ∣ + ∣ i log β − j log α ∣ = 2 max ( ∣ i ∣ log β , ∣ j ∣ log α ) .
With β > α > 1 \beta>\alpha>1 β > α > 1 (as s > r s>r s > r and a c > a b \sqrt{ac}>\sqrt{ab} a c > ab ) this gives h ( α ) = 1 2 log α \mathrm{h}(\alpha)=\frac{1}{2}\log\alpha h ( α ) = 2 1 log α and h ( β / α ) = 1 2 log β \mathrm{h}(\beta/\alpha)=\frac{1}{2}\log\beta h ( β / α ) = 2 1 log β . Likewise, by item 2 the four conjugates of the unit γ / β \gamma/\beta γ / β are ( γ / β ) ± 1 (\gamma/\beta)^{\pm1} ( γ / β ) ± 1 and ( β γ ) ± 1 (\beta\gamma)^{\pm1} ( β γ ) ± 1 ; as γ > β > 1 \gamma>\beta>1 γ > β > 1 (because t > s t>s t > s and b c > a c \sqrt{bc}>\sqrt{ac} b c > a c ), h ( γ / β ) = 1 4 ( log ( γ / β ) + log ( β γ ) ) = 1 2 log γ \mathrm{h}(\gamma/\beta)=\frac{1}{4}(\log(\gamma/\beta)+\log(\beta\gamma))=\frac{1}{2}\log\gamma h ( γ / β ) = 4 1 ( log ( γ / β ) + log ( β γ )) = 2 1 log γ . For the degrees, the images β e 2 α − e 1 \beta^{e_2}\alpha^{-e_1} β e 2 α − e 1 of β / α \beta/\alpha β / α under the four embeddings of item 2 are β / α \beta/\alpha β / α , α β \alpha\beta α β , ( α β ) − 1 (\alpha\beta)^{-1} ( α β ) − 1 , α / β \alpha/\beta α / β , which are distinct because β > α > 1 \beta> \alpha> 1 β > α > 1 ; and the images γ e 1 e 2 β − e 2 \gamma^{e_1e_2}\beta^{-e_2} γ e 1 e 2 β − e 2 of γ / β \gamma/\beta γ / β are γ / β \gamma/\beta γ / β , β / γ \beta/\gamma β / γ , ( β γ ) − 1 (\beta\gamma)^{-1} ( β γ ) − 1 , β γ \beta\gamma β γ (for ( e 1 , e 2 ) = ( 1 , 1 ) , ( 1 , − 1 ) , ( − 1 , 1 ) , ( − 1 , − 1 ) (e_1,e_2)=(1,1),(1,-1),(-1,1),(-1,-1) ( e 1 , e 2 ) = ( 1 , 1 ) , ( 1 , − 1 ) , ( − 1 , 1 ) , ( − 1 , − 1 ) ), which are distinct because γ > β > 1 \gamma> \beta> 1 γ > β > 1 . So β / α \beta/\alpha β / α and γ / β \gamma/\beta γ / β have degree 4.
We next prove item 4. Dividing numerator and denominator in Definition 2.16 by b c \sqrt{bc} b c gives D λ = ( 1 + a / b ) / ( 1 + λ a / c ) D_\lambda=(1+\sqrt{a/b})/(1+\lambda\sqrt{a/c}) D λ = ( 1 + a / b ) / ( 1 + λ a / c ) . Since c − ( a + b ) 2 = 2 ( r − a b ) > 0 c-(\sqrt{a}+\sqrt{b})^2=2(r-\sqrt{ab})>0 c − ( a + b ) 2 = 2 ( r − ab ) > 0 , one has a / c < a / ( a + b ) \sqrt{a/c}<\sqrt{a}/(\sqrt{a}+\sqrt{b}) a / c < a / ( a + b ) , i.e. 1 − a / c > ( 1 + a / b ) − 1 1-\sqrt{a/c}>(1+\sqrt{a/b})^{-1} 1 − a / c > ( 1 + a / b ) − 1 . Hence 1 < D λ ≤ D − 1 < ( 1 + a / b ) 2 1< D_\lambda\le D_{-1}<(1+\sqrt{a/b})^2 1 < D λ ≤ D − 1 < ( 1 + a / b ) 2 , where D λ > 1 D_\lambda>1 D λ > 1 because a / b > a / c \sqrt{a/b}>\sqrt{a/c} a / b > a / c . So 0 < log D λ < 2 log ( 1 + a / b ) 0<\log D_\lambda<2\log(1+\sqrt{a/b}) 0 < log D λ < 2 log ( 1 + a / b ) , which is less than 2 a / b 2\sqrt{a/b} 2 a / b and than log 4 \log4 log 4 . The numerator N x = b c + a c N_x=\sqrt{bc}+\sqrt{ac} N x = b c + a c and the denominator M x = b c + λ a b M_x=\sqrt{bc}+\lambda\sqrt{ab} M x = b c + λ ab of D λ D_\lambda D λ are algebraic integers, and by item 2
σ e 1 , e 2 N x = e 2 ( e 1 b c + a c ) , ∣ σ e 1 , e 2 N x ∣ = b c + e 1 a c , \sigma_{e_1,e_2}N_x=e_2(e_1\sqrt{bc}+\sqrt{ac}),\qquad|\sigma_{e_1,e_2}N_x|=\sqrt{bc}+e_1\sqrt{ac}, σ e 1 , e 2 N x = e 2 ( e 1 b c + a c ) , ∣ σ e 1 , e 2 N x ∣ = b c + e 1 a c ,
σ e 1 , e 2 M x = e 1 ( e 2 b c + λ a b ) , ∣ σ e 1 , e 2 M x ∣ = b c + λ e 2 a b . \sigma_{e_1,e_2}M_x=e_1(e_2\sqrt{bc}+\lambda\sqrt{ab}),\qquad|\sigma_{e_1,e_2}M_x|=\sqrt{bc}+\lambda e_2\sqrt{ab}. σ e 1 , e 2 M x = e 1 ( e 2 b c + λ ab ) , ∣ σ e 1 , e 2 M x ∣ = b c + λ e 2 ab .
The inner factors e 1 b c + a c e_1\sqrt{bc}+\sqrt{ac} e 1 b c + a c and e 2 b c + λ a b e_2\sqrt{bc}+\lambda\sqrt{ab} e 2 b c + λ ab have signs e 1 e_1 e 1 and e 2 e_2 e 2 , so σ e 1 , e 2 N x \sigma_{e_1,e_2}N_x σ e 1 , e 2 N x and σ e 1 , e 2 M x \sigma_{e_1,e_2}M_x σ e 1 , e 2 M x both have sign e 1 e 2 e_1e_2 e 1 e 2 . Hence every conjugate of D λ D_\lambda D λ is positive, N K / Q ( N x ) = ( b c − a c ) 2 N_{K/\mathbb{Q}}(N_x)=(bc-ac)^2 N K / Q ( N x ) = ( b c − a c ) 2 and N K / Q ( M x ) = ( b c − a b ) 2 N_{K/\mathbb{Q}}(M_x)=(bc-ab)^2 N K / Q ( M x ) = ( b c − ab ) 2 , which gives the norm; it lies in ( 0 , 1 ) (0,1) ( 0 , 1 ) because b ( c − a ) − c ( b − a ) = a ( c − b ) > 0 b(c-a)-c(b-a)=a(c-b)>0 b ( c − a ) − c ( b − a ) = a ( c − b ) > 0 . Since a c > a b ac>ab a c > ab , the four values of max ( ∣ σ e 1 , e 2 N x ∣ , ∣ σ e 1 , e 2 M x ∣ ) \max(|\sigma_{e_1,e_2}N_x|,|\sigma_{e_1,e_2}M_x|) max ( ∣ σ e 1 , e 2 N x ∣ , ∣ σ e 1 , e 2 M x ∣ ) are b c + a c \sqrt{bc}+\sqrt{ac} b c + a c (for e 1 = 1 e_1=1 e 1 = 1 , twice), b c + a b \sqrt{bc}+\sqrt{ab} b c + ab and b c − a b \sqrt{bc}-\sqrt{ab} b c − ab (for e 1 = − 1 e_1=-1 e 1 = − 1 ). By the product formula, 4 h ( N x / M x ) 4h(N_x/M_x) 4 h ( N x / M x ) is the sum of log max ( ∥ N x ∥ , ∥ M x ∥ ) \log\max(\|N_x\|,\|M_x\|) log max ( ∥ N x ∥ , ∥ M x ∥ ) over the normalized absolute values ∥ ⋅ ∥ \|\cdot\| ∥ ⋅ ∥ of K \mathcal{K} K ; the finite places contribute at most 0 because N x , M x N_x,M_x N x , M x are algebraic integers, and the four real places give
4 h ( D λ ) ≤ 2 log ( b c + a c ) + log ( b c − a b ) = log ( c ( a + b ) 2 ) + log ( b ( c − a ) ) < 4 log c , 4h(D_\lambda)\le2\log(\sqrt{bc}+\sqrt{ac})+\log(bc-ab)=\log\left(c(\sqrt{a}+\sqrt{b})^2\right)+\log(b(c-a))<4\log c, 4 h ( D λ ) ≤ 2 log ( b c + a c ) + log ( b c − ab ) = log ( c ( a + b ) 2 ) + log ( b ( c − a )) < 4 log c ,
using ( a + b ) 2 < c (\sqrt{a}+\sqrt{b})^2<c ( a + b ) 2 < c and b ( c − a ) < c 2 b(c-a)<c^2 b ( c − a ) < c 2 .
We next prove item 5. Since b c > a c > a b \sqrt{bc}>\sqrt{ac}>\sqrt{ab} b c > a c > ab , E λ − 1 = ( b c − a c ) / ( a c + λ a b ) > 0 E_\lambda-1=(\sqrt{bc}-\sqrt{ac})/(\sqrt{ac}+\lambda\sqrt{ab})>0 E λ − 1 = ( b c − a c ) / ( a c + λ ab ) > 0 . Put u = a u=\sqrt{a} u = a , v = b v=\sqrt{b} v = b and w = c w=\sqrt{c} w = c , so that u < v u<v u < v and u + v < w < 2 v u+v<w<2v u + v < w < 2 v (as ( a + b ) 2 < c < 4 b (\sqrt{a}+\sqrt{b})^2<c<4b ( a + b ) 2 < c < 4 b ), and E λ = v ( w + λ u ) u ( w + λ v ) E_\lambda=\frac{v(w+\lambda u)}{u(w+\lambda v)} E λ = u ( w + λ v ) v ( w + λ u ) . For λ = 1 \lambda=1 λ = 1 , E 1 < v / u E_1<v/u E 1 < v / u , so log E 1 < v − u u ≤ 2 ( v 2 − u 2 ) u w \log E_1<\frac{v-u}{u}\le\frac{2(v^2-u^2)}{uw} log E 1 < u v − u ≤ u w 2 ( v 2 − u 2 ) , the last step because w < 2 v < 2 ( u + v ) w<2v<2(u+v) w < 2 v < 2 ( u + v ) . For λ = − 1 \lambda=-1 λ = − 1 , using log ( 1 + ρ ) < ρ \log(1+\rho)<\rho log ( 1 + ρ ) < ρ for ρ > 0 \rho>0 ρ > 0 ,
log E − 1 = log v u + log ( 1 + v − u w − v ) < ( v − u ) ( 1 u + 1 w − v ) ≤ 2 ( v 2 − u 2 ) u w , \log E_{-1}=\log\frac{v}{u}+\log\left(1+\frac{v-u}{w-v}\right)<(v-u)\left(\frac{1}{u}+\frac{1}{w-v}\right)\le\frac{2(v^2-u^2)}{uw}, log E − 1 = log u v + log ( 1 + w − v v − u ) < ( v − u ) ( u 1 + w − v 1 ) ≤ u w 2 ( v 2 − u 2 ) ,
where the last inequality is equivalent to w 2 − ( u + 3 v ) w + 2 v ( u + v ) ≤ 0 w^2-(u+3v)w+2v(u+v)\le0 w 2 − ( u + 3 v ) w + 2 v ( u + v ) ≤ 0 , that is, to ( w − u − v ) ( w − 2 v ) ≤ 0 (w-u-v)(w-2v)\le0 ( w − u − v ) ( w − 2 v ) ≤ 0 . Since v 2 − u 2 = b − a v^2-u^2=b-a v 2 − u 2 = b − a and u w = a c uw=\sqrt{ac} u w = a c , in both cases 0 < log E λ < 2 ( b − a ) / a c 0<\log E_\lambda<2(b-a)/\sqrt{ac} 0 < log E λ < 2 ( b − a ) / a c .
The numerator N z = b c + λ a b N_z=\sqrt{bc}+\lambda\sqrt{ab} N z = b c + λ ab and the denominator M z = a c + λ a b M_z=\sqrt{ac}+\lambda\sqrt{ab} M z = a c + λ ab of E λ E_\lambda E λ are algebraic integers, and by item 2 ∣ σ e 1 , e 2 N z ∣ = b c + λ e 2 a b |\sigma_{e_1,e_2}N_z|=\sqrt{bc}+\lambda e_2\sqrt{ab} ∣ σ e 1 , e 2 N z ∣ = b c + λ e 2 ab and ∣ σ e 1 , e 2 M z ∣ = a c + λ e 1 e 2 a b |\sigma_{e_1,e_2}M_z|=\sqrt{ac}+\lambda e_1e_2\sqrt{ab} ∣ σ e 1 , e 2 M z ∣ = a c + λ e 1 e 2 ab , both positive. Hence N K / Q ( N z ) = ( b c − a b ) 2 = b 2 ( c − a ) 2 N_{K/\mathbb{Q}}(N_z)=(bc-ab)^2=b^2(c-a)^2 N K / Q ( N z ) = ( b c − ab ) 2 = b 2 ( c − a ) 2 and N K / Q ( M z ) = ( a c − a b ) 2 = a 2 ( c − b ) 2 N_{K/\mathbb{Q}}(M_z)=(ac-ab)^2=a^2(c-b)^2 N K / Q ( M z ) = ( a c − ab ) 2 = a 2 ( c − b ) 2 ; since b ( c − a ) − a ( c − b ) = c ( b − a ) > 0 b(c-a)-a(c-b)=c(b-a)>0 b ( c − a ) − a ( c − b ) = c ( b − a ) > 0 , the norm of E λ E_\lambda E λ exceeds 1. For e 1 = 1 e_1=1 e 1 = 1 the larger of ∣ σ e 1 , e 2 N z ∣ |\sigma_{e_1,e_2}N_z| ∣ σ e 1 , e 2 N z ∣ and ∣ σ e 1 , e 2 M z ∣ |\sigma_{e_1,e_2}M_z| ∣ σ e 1 , e 2 M z ∣ is ∣ σ e 1 , e 2 N z ∣ |\sigma_{e_1,e_2}N_z| ∣ σ e 1 , e 2 N z ∣ , and the product of its two values (e 2 = ± 1 e_2=\pm1 e 2 = ± 1 ) is b c − a b = b ( c − a ) bc-ab=b(c-a) b c − ab = b ( c − a ) ; for e 1 = − 1 e_1=-1 e 1 = − 1 both are at most b c + a b = b ( c + a ) \sqrt{bc}+\sqrt{ab}=\sqrt{b}(\sqrt{c}+\sqrt{a}) b c + ab = b ( c + a ) . By the product formula, the finite places contributing at most 0,
4 h ( E λ ) ≤ log ( b ( c − a ) ) + log ( b ( c + a ) 2 ) . 4h(E_\lambda)\le\log(b(c-a))+\log\left(b(\sqrt{c}+\sqrt{a})^2\right). 4 h ( E λ ) ≤ log ( b ( c − a )) + log ( b ( c + a ) 2 ) .
When b ≤ 3 2 a b\le\frac{3}{2}a b ≤ 2 3 a , one has b ( c − a ) < b c ≤ 3 2 a c b(c-a)<bc\le\frac{3}{2}ac b ( c − a ) < b c ≤ 2 3 a c and, as ( c + a ) 2 < 9 4 c (\sqrt{c}+\sqrt{a})^2<\frac{9}{4}c ( c + a ) 2 < 4 9 c , b ( c + a ) 2 < 9 4 b c ≤ 27 8 a c b(\sqrt{c}+\sqrt{a})^2<\frac{9}{4}bc\le\frac{27}{8}ac b ( c + a ) 2 < 4 9 b c ≤ 8 27 a c ; so 4 h ( E λ ) < log ( 81 16 ( a c ) 2 ) < log β 4 4h(E_\lambda)<\log\left(\frac{81}{16}(ac)^2\right)<\log\beta^4 4 h ( E λ ) < log ( 16 81 ( a c ) 2 ) < log β 4 , because β = s + a c > 2 a c \beta=s+\sqrt{ac}>2\sqrt{ac} β = s + a c > 2 a c . Hence h ( E λ ) < log β h(E_\lambda)<\log\beta h ( E λ ) < log β .
We finally prove item 6. Here α > 1 \alpha>1 α > 1 , β / α > 1 \beta/\alpha>1 β / α > 1 (as s > r s>r s > r and a c > a b \sqrt{ac}>\sqrt{ab} a c > ab ) and D λ > 1 D_\lambda>1 D λ > 1 (item 4). If α i ( β / α ) j D λ k = 1 \alpha^i(\beta/\alpha)^jD_\lambda^k=1 α i ( β / α ) j D λ k = 1 with integers i , j , k i,j,k i , j , k , taking norms gives N K / Q ( D λ ) k = 1 N_{K/\mathbb{Q}}(D_\lambda)^k=1 N K / Q ( D λ ) k = 1 , as the units α , β \alpha,\beta α , β have norm 1 by item 2; so k = 0 k=0 k = 0 by item 4. Then β j α i − j = 1 \beta^j\alpha^{i-j}=1 β j α i − j = 1 , and the embedding σ 1 , − 1 \sigma_{1,-1} σ 1 , − 1 gives β − j α i − j = 1 \beta^{-j}\alpha^{i-j}=1 β − j α i − j = 1 , whence β 2 j = 1 \beta^{2j}=1 β 2 j = 1 , j = 0 j=0 j = 0 and i = 0 i=0 i = 0 . Next γ / β > 1 \gamma/\beta>1 γ / β > 1 because t > s t>s t > s and b c > a c \sqrt{bc}>\sqrt{ac} b c > a c , and β 2 ( m − n ) E λ > 1 \beta^{2(m-n)}E_{\lambda}>1 β 2 ( m − n ) E λ > 1 because m > n m>n m > n and E λ > 1 E_{\lambda}>1 E λ > 1 (item 5). If ( γ / β ) i ( β 2 ( m − n ) E λ ) j = 1 (\gamma/\beta)^i(\beta^{2(m-n)}E_{\lambda})^j=1 ( γ / β ) i ( β 2 ( m − n ) E λ ) j = 1 with integers i , j i,j i , j , taking norms gives N K / Q ( E λ ) j = 1 N_{\mathbb{K}/\mathbb{Q}}(E_{\lambda})^j=1 N K / Q ( E λ ) j = 1 , as the units β , γ \beta,\gamma β , γ have norm 1 by item 2; so j = 0 j=0 j = 0 by item 5, and then i = 0 i=0 i = 0 because γ / β > 1 \gamma/\beta>1 γ / β > 1 .
The bound h ( D λ ) < log c h(D_{\lambda})<\log c h ( D λ ) < log c of item 4 is essentially sharp: for the Euler triples { 3 , 5 , 16 } \{3,5,16\} { 3 , 5 , 16 } and { 1 , 120 , 143 } \{1,120,143\} { 1 , 120 , 143 } an exact computation gives h ( D λ ) ≈ 0.875 log c h(D_{\lambda})\approx0.875\log c h ( D λ ) ≈ 0.875 log c and h ( D λ ) ≈ 0.991 log c h(D_{\lambda})\approx0.991\log c h ( D λ ) ≈ 0.991 log c , for both values of λ \lambda λ . It enters the criterion through the height of α 2 ( l − m ) D λ \alpha^{2(l-m)}D_{\lambda} α 2 ( l − m ) D λ in the case of large b / a b/a b / a .
The lower bound that we play against the smallness of the forms is Laurent’s theorem on linear forms in two logarithms [22 ] , quoted in the form we use.
Theorem 4.3. Let η 1 , η 2 > 1 \eta_1,\eta_2>1 η 1 , η 2 > 1 be real algebraic numbers that are multiplicatively independent, put D = [ Q ( η 1 , η 2 ) : Q ] \mathcal{D}=[\mathbb{Q}(\eta_1,\eta_2):\mathbb{Q}] D = [ Q ( η 1 , η 2 ) : Q ] , let b 1 , b 2 b_1,b_2 b 1 , b 2 be positive integers, and
Λ L = b 2 log η 2 − b 1 log η 1 . \Lambda_L=b_2\log\eta_2-b_1\log\eta_1. Λ L = b 2 log η 2 − b 1 log η 1 . Let ρ > 1 \rho>1 ρ > 1 and 1 3 ≤ μ ≤ 1 \frac{1}{3}\le\mu\le1 3 1 ≤ μ ≤ 1 be real, and put σ L = 1 + 2 μ − μ 2 2 \sigma_L=\frac{1+2\mu-\mu^2}{2} σ L = 2 1 + 2 μ − μ 2 and λ L = σ L log ρ \lambda_L=\sigma_L\log\rho λ L = σ L log ρ . With the height h h h of Definition 4.1, let A 1 , A 2 A_1,A_2 A 1 , A 2 be real numbers with
A i ≥ max { 1 , ( ρ − 1 ) log η i + 2 D h ( η i ) } ( i = 1 , 2 ) , A 1 A 2 ≥ λ L 2 , A_i\ge\max\{1,(\rho-1)\log\eta_i+2\mathcal{D}h(\eta_i)\}\quad(i=1,2),\qquad A_1A_2\ge\lambda_L^2, A i ≥ max { 1 , ( ρ − 1 ) log η i + 2 D h ( η i )} ( i = 1 , 2 ) , A 1 A 2 ≥ λ L 2 , and let h L h_L h L be a real number with
h L ≥ max { D ( log ( b 1 A 2 + b 2 A 1 ) + log λ L + 1.75 ) + 0.06 , λ L , D log 2 2 } . h_L\ge\max\left\{\mathcal{D}\left(\log\left(\frac{b_1}{A_2}+\frac{b_2}{A_1}\right)+\log\lambda_L+1.75\right)+0.06,\ \lambda_L,\ \frac{\mathcal{D}\log2}{2}\right\}. h L ≥ max { D ( log ( A 2 b 1 + A 1 b 2 ) + log λ L + 1.75 ) + 0.06 , λ L , 2 D log 2 } . Put
H = h L λ L + 1 σ L , ω L = 2 ( 1 + 1 + 1 4 H 2 ) , θ L = 1 + 1 4 H 2 + 1 2 H , \mathcal{H}=\frac{h_L}{\lambda_L}+\frac{1}{\sigma_L},\qquad\omega_L=2\left(1+\sqrt{1+\frac{1}{4\mathcal{H}^2}}\right),\qquad\theta_L=\sqrt{1+\frac{1}{4\mathcal{H}^2}}+\frac{1}{2\mathcal{H}}, H = λ L h L + σ L 1 , ω L = 2 ( 1 + 1 + 4 H 2 1 ) , θ L = 1 + 4 H 2 1 + 2 H 1 , C = μ λ L 3 σ L ( ω L 6 + 1 2 ω L 2 9 + 8 λ L ω L 5 / 4 θ L 1 / 4 3 A 1 A 2 H + 4 3 ( 1 A 1 + 1 A 2 ) λ L ω L H ) 2 , C ′ = C σ L ω L θ L λ L 3 μ . C=\frac{\mu}{\lambda_L^3\sigma_L}\left(\frac{\omega_L}{6}+\frac{1}{2}\sqrt{\frac{\omega_L^2}{9}+\frac{8\lambda_L\omega_L^{5/4}\theta_L^{1/4}}{3\sqrt{A_1A_2}\sqrt{\mathcal{H}}}}+\frac{4}{3}\left(\frac{1}{A_1}+\frac{1}{A_2}\right)\frac{\lambda_L\omega_L}{\mathcal{H}}\right)^2,\qquad C'=\sqrt{\frac{C\sigma_L\omega_L\theta_L}{\lambda_L^3\mu}}. C = λ L 3 σ L μ 6 ω L + 2 1 9 ω L 2 + 3 A 1 A 2 H 8 λ L ω L 5/4 θ L 1/4 + 3 4 ( A 1 1 + A 2 1 ) H λ L ω L 2 , C ′ = λ L 3 μ C σ L ω L θ L . Then
log ∣ Λ L ∣ ≥ − C ( h L + log ρ ) 2 A 1 A 2 − ω L θ L ( h L + log ρ ) − log ( C ′ ( h L + log ρ ) 2 A 1 A 2 ) . \log|\Lambda_L|\ge-C(h_L+\log\rho)^2A_1A_2-\sqrt{\omega_L\theta_L}(h_L+\log\rho)-\log\left(C'(h_L+\log\rho)^2A_1A_2\right). log ∣ Λ L ∣ ≥ − C ( h L + log ρ ) 2 A 1 A 2 − ω L θ L ( h L + log ρ ) − log ( C ′ ( h L + log ρ ) 2 A 1 A 2 ) .
Proof. This is [22 Theorem 2, p. 327] , in the case of two real numbers greater than 1 with their real logarithms; there D = [ Q ( η 1 , η 2 ) : Q ] / [ R ( η 1 , η 2 ) : R ] \mathcal{D}=[\mathbb{Q}(\eta_1,\eta_2):\mathbb{Q}]/[\mathbb{R}(\eta_1,\eta_2):\mathbb{R}] D = [ Q ( η 1 , η 2 ) : Q ] / [ R ( η 1 , η 2 ) : R ] , which is the degree in this case, and λ L / σ L = log ρ \lambda_L/\sigma_L=\log\rho λ L / σ L = log ρ .
The next lemma is the crux of the section. It concerns two numbers η 1 , η 2 \eta_1,\eta_2 η 1 , η 2 , the first close to 1 and the second of height linear in an integer g g g , and it excludes a very small form log η 2 − 2 k log η 1 \log\eta_2-2k\log\eta_1 log η 2 − 2 k log η 1 when k > g T ξ k>gT\xi k > g T ξ ; its two sets of constants serve the two cases of the criterion.
Lemma 4.4. Let η 1 , η 2 > 1 \eta_1,\eta_2>1 η 1 , η 2 > 1 be real algebraic numbers that are multiplicatively independent, with [ Q ( η 1 , η 2 ) : Q ] = 4 [\mathbb{Q}(\eta_1,\eta_2):\mathbb{Q}]=4 [ Q ( η 1 , η 2 ) : Q ] = 4 , and let
( T , ξ 0 , h 0 , C ) = ( 3355 , log 6710 , 2 , 4.01 ) or ( T , ξ 0 , h 0 , C ) = ( 3075 , log 12300 , 1 , 2 ) . (T,\xi_0,h_0,C)=(3355,\log6710,2,4.01)\qquad\text{or}\qquad(T,\xi_0,h_0,C)=(3075,\log12300,1,2). ( T , ξ 0 , h 0 , C ) = ( 3355 , log 6710 , 2 , 4.01 ) or ( T , ξ 0 , h 0 , C ) = ( 3075 , log 12300 , 1 , 2 ) . There do not exist a real number ξ > ξ 0 \xi>\xi_0 ξ > ξ 0 and integers g , k ≥ 1 g,k\ge1 g , k ≥ 1 such that
log η 1 < 1 T , h ( η 1 ) < 1 2 ( ξ + 1 T ) , log η 2 < 2 g ξ + 2 T , h ( η 2 ) < ( g + h 0 ) ξ , k > g T ξ , \log\eta_1<\frac{1}{T},\qquad h(\eta_1)<\frac{1}{2}\left(\xi+\frac{1}{T}\right),\qquad\log\eta_2<2g\xi+\frac{2}{T},\qquad h(\eta_2)<(g+h_0)\xi,\qquad k>gT\xi, log η 1 < T 1 , h ( η 1 ) < 2 1 ( ξ + T 1 ) , log η 2 < 2 g ξ + T 2 , h ( η 2 ) < ( g + h 0 ) ξ , k > g T ξ ,
and
0 < log η 2 − 2 k log η 1 < C e − 4 k ξ . 0 < \log\eta_2 - 2k \log\eta_1 < C e^{-4k\xi}. 0 < log η 2 − 2 k log η 1 < C e − 4 k ξ .
Proof. We play the upper bound C e − 4 k ξ C e^{-4k\xi} C e − 4 k ξ of the last condition against Laurent’s lower bound (Theorem 4.3). The first four conditions allow Laurent’s parameters A 1 A_1 A 1 and A 2 A_2 A 2 of order ξ \xi ξ and g ξ g\xi g ξ , so that the lower bound is roughly exponential in g ξ 2 g\xi^2 g ξ 2 times the square of log ( k / ( g ξ ) ) \log(k/(g\xi)) log ( k / ( g ξ )) , while the upper bound is exponential in k ξ k\xi k ξ ; and the condition k > g T ξ k > gT\xi k > g T ξ makes k / ( g ξ ) k/(g\xi) k / ( g ξ ) too large for both bounds to hold.
Suppose that ξ > ξ 0 \xi> \xi_0 ξ > ξ 0 and g , k ≥ 1 g, k \ge1 g , k ≥ 1 satisfy all the displayed conditions, and put Λ L = log η 2 − 2 k log η 1 \Lambda_L = \log\eta_2 - 2k \log\eta_1 Λ L = log η 2 − 2 k log η 1 . We derive a contradiction from Theorem 4.3 , applied with b 1 = 2 k b_1 = 2k b 1 = 2 k , b 2 = 1 b_2 = 1 b 2 = 1 , D = [ Q ( η 1 , η 2 ) : Q ] = 4 \mathcal{D} = [\mathbb{Q}(\eta_1,\eta_2) : \mathbb{Q}] = 4 D = [ Q ( η 1 , η 2 ) : Q ] = 4 , ρ = 37 \rho= 37 ρ = 37 and μ = 63 100 \mu= \frac{63}{100} μ = 100 63 , so that σ L = 1 + 2 μ − μ 2 2 = 0.93155 \sigma_L = \frac{1+2\mu-\mu^2}{2} = 0.93155 σ L = 2 1 + 2 μ − μ 2 = 0.93155 and λ L = σ L log 37 = 3.36375 … \lambda_L = \sigma_L \log37 = 3.36375\ldots λ L = σ L log 37 = 3.36375 … .
We first fix the parameters A 1 , A 2 , h L A_1,A_2,h_L A 1 , A 2 , h L and check the hypotheses of Theorem 4.3 . Put ( χ 1 , χ 2 ) = ( 4.0014 , 16.0025 ) (\chi_1,\chi_2) = (4.0014, 16.0025) ( χ 1 , χ 2 ) = ( 4.0014 , 16.0025 ) in the first case and ( χ 1 , χ 2 ) = ( 4.0015 , 8.0025 ) (\chi_1,\chi_2) = (4.0015, 8.0025) ( χ 1 , χ 2 ) = ( 4.0015 , 8.0025 ) in the second, and
A 1 = χ 1 ξ , A 2 = ( 80 g + χ 2 ) ξ . A_1 = \chi_1\xi,\qquad A_2 = (80g+\chi_2)\xi. A 1 = χ 1 ξ , A 2 = ( 80 g + χ 2 ) ξ .
Laurent’s condition with ρ = 37 \rho= 37 ρ = 37 , D = 4 \mathcal{D} = 4 D = 4 reads A i ≥ 36 log η i + 8 h ( η i ) A_i \ge36\log\eta_i + 8h(\eta_i) A i ≥ 36 log η i + 8 h ( η i ) , with the height h h h of Definition 4.1; and since ξ > ξ 0 \xi> \xi_0 ξ > ξ 0 , the first four conditions give
36 log η 1 + 8 h ( η 1 ) < 4 ξ + 40 T < ( 4 + 40 T ξ 0 ) ξ < A 1 , 36\log\eta_1 + 8h(\eta_1) < 4\xi+ \frac{40}{T} < \left(4+\frac{40}{T\xi_0}\right)\xi< A_1, 36 log η 1 + 8 h ( η 1 ) < 4 ξ + T 40 < ( 4 + T ξ 0 40 ) ξ < A 1 ,
36 log η 2 + 8 h ( η 2 ) < ( 80 g + 8 h 0 ) ξ + 72 T < ( 80 g + 8 h 0 + 72 T ξ 0 ) ξ < A 2 , 36\log\eta_2 + 8h(\eta_2) < (80g+8h_0)\xi+ \frac{72}{T} < \left(80g+8h_0+\frac{72}{T\xi_0}\right)\xi< A_2, 36 log η 2 + 8 h ( η 2 ) < ( 80 g + 8 h 0 ) ξ + T 72 < ( 80 g + 8 h 0 + T ξ 0 72 ) ξ < A 2 ,
and A 1 , A 2 > 1 A_1,A_2 > 1 A 1 , A 2 > 1 , A 1 A 2 > 320 ξ 0 2 > λ L 2 A_1A_2 > 320\xi_0^2 > \lambda_L^2 A 1 A 2 > 320 ξ 0 2 > λ L 2 . Put δ 0 = 1 / ( χ 1 ξ 0 ) \delta_0 = 1/(\chi_1\xi_0) δ 0 = 1/ ( χ 1 ξ 0 ) , ℓ ∗ = log λ L + 1.75 + 1 4 ( 0.06 + log 37 ) = 3.88078 … \ell_* = \log\lambda_L + 1.75 + \frac{1}{4}(0.06+\log37) = 3.88078\ldots ℓ ∗ = log λ L + 1.75 + 4 1 ( 0.06 + log 37 ) = 3.88078 … , and
Y = 2 k ( 80 g + χ 2 ) ξ , ℓ Y = log ( Y + δ 0 ) + ℓ ∗ , Y = \frac{2k}{(80g+\chi_2)\xi},\qquad\ell_Y = \log(Y+\delta_0)+\ell_*, Y = ( 80 g + χ 2 ) ξ 2 k , ℓ Y = log ( Y + δ 0 ) + ℓ ∗ ,
h L = 4 ℓ Y − log 37 = 4 ( log ( Y + δ 0 ) + log λ L + 1.75 ) + 0.06. h_L = 4\ell_Y-\log37 = 4\bigl(\log(Y+\delta_0)+\log\lambda_L+1.75\bigr)+0.06. h L = 4 ℓ Y − log 37 = 4 ( log ( Y + δ 0 ) + log λ L + 1.75 ) + 0.06.
Since b 1 / A 2 + b 2 / A 1 = Y + 1 / ( χ 1 ξ ) < Y + δ 0 b_1/A_2+b_2/A_1 = Y+1/(\chi_1\xi) < Y+\delta_0 b 1 / A 2 + b 2 / A 1 = Y + 1/ ( χ 1 ξ ) < Y + δ 0 , the number h L h_L h L satisfies the first lower bound required of it in Theorem 4.3 . Next, the condition k > g T ξ k > gT\xi k > g T ξ , which enters only here, and 80 g + χ 2 ≤ ( 80 + χ 2 ) g 80g+\chi_2 \le(80+\chi_2)g 80 g + χ 2 ≤ ( 80 + χ 2 ) g give
Y > Y 0 : = 2 T 80 + χ 2 , Y > Y_0 := \frac{2T}{80+\chi_2}, Y > Y 0 := 80 + χ 2 2 T ,
that is
Y 0 = 6710 96.0025 = 69.89401 … or Y 0 = 6150 88.0025 = 69.88437 … , Y_0 = \frac{6710}{96.0025}=69.89401\ldots\quad\text{or}\quad Y_0 = \frac{6150}{88.0025}=69.88437\ldots, Y 0 = 96.0025 6710 = 69.89401 … or Y 0 = 88.0025 6150 = 69.88437 … ,
so ℓ Y > ℓ Y 0 > 8.12 \ell_Y > \ell_{Y_0} > 8.12 ℓ Y > ℓ Y 0 > 8.12 and h L > 28 h_L > 28 h L > 28 , which exceeds λ L \lambda_L λ L and D log 2 2 = 2 log 2 \frac{\mathcal{D}\log2}{2}=2\log2 2 D l o g 2 = 2 log 2 . As λ L / σ L = log 37 \lambda_L/\sigma_L = \log37 λ L / σ L = log 37 , the quantity H \mathcal{H} H of Theorem 4.3 is H = ( h L + log 37 ) / λ L = 4 ℓ Y / λ L \mathcal{H} = (h_L+\log37)/\lambda_L = 4\ell_Y/\lambda_L H = ( h L + log 37 ) / λ L = 4 ℓ Y / λ L , and since Λ L > 0 \Lambda_L > 0 Λ L > 0 the theorem gives
log Λ L ≥ − 16 C ′ ℓ Y 2 A 1 A 2 − 4 ω L θ L ℓ Y − log ( 16 C ′ ℓ Y 2 A 1 A 2 ) . (5) \log\Lambda_L \ge-16C' \ell_Y^2 A_1A_2 - 4\sqrt{\omega_L\theta_L}\,\ell_Y - \log(16C'\ell_Y^2A_1A_2). \tag*{(5)} log Λ L ≥ − 16 C ′ ℓ Y 2 A 1 A 2 − 4 ω L θ L ℓ Y − log ( 16 C ′ ℓ Y 2 A 1 A 2 ) . (5)
We next replace the auxiliary functions of (1) by constants. The functions ω L \omega_L ω L and θ L \theta_L θ L of Theorem 4.3 decrease as H \mathcal{H} H increases, so C C C decreases as H \mathcal{H} H , A 1 A_1 A 1 or A 2 A_2 A 2 increases, and so does C ′ C' C ′ . We have H ≥ H 0 : = 4 ℓ Y 0 / λ L \mathcal{H} \ge\mathcal{H}_0 := 4\ell_{Y_0}/\lambda_L H ≥ H 0 := 4 ℓ Y 0 / λ L , A 1 ≥ A 1 , 0 : = χ 1 ξ 0 A_1 \ge A_{1,0} := \chi_1\xi_0 A 1 ≥ A 1 , 0 := χ 1 ξ 0 and A 2 ≥ A 2 , 0 : = ( 80 + χ 2 ) ξ 0 A_2 \ge A_{2,0} := (80+\chi_2)\xi_0 A 2 ≥ A 2 , 0 := ( 80 + χ 2 ) ξ 0 . Let ω L , 0 , θ L , 0 , C 0 , C 0 ′ \omega_{L,0},\theta_{L,0},C_0,C'_0 ω L , 0 , θ L , 0 , C 0 , C 0 ′ denote the values of ω L , θ L , C , C ′ \omega_L,\theta_L,C,C' ω L , θ L , C , C ′ at ( H 0 , A 1 , 0 , A 2 , 0 ) (\mathcal{H}_0,A_{1,0},A_{2,0}) ( H 0 , A 1 , 0 , A 2 , 0 ) ; then ω L θ L ≤ ω L , 0 θ L , 0 \omega_L\theta_L \le\omega_{L,0}\theta_{L,0} ω L θ L ≤ ω L , 0 θ L , 0 , C ≤ C 0 C \le C_0 C ≤ C 0 and C ′ ≤ C 0 ′ C' \le C'_0 C ′ ≤ C 0 ′ . We now turn (1) and the upper bound for Λ L \Lambda_L Λ L into an inequality Y < Θ ( Y ) Y < \Theta(Y) Y < Θ ( Y ) . By hypothesis log Λ L < log C − 4 k ξ \log\Lambda_L < \log C - 4k\xi log Λ L < log C − 4 k ξ . Combining this with (1) and the three bounds just obtained, and dividing by 2 A 2 2 ξ = 2 Ξ 2A_2^2\xi= 2\Xi 2 A 2 2 ξ = 2Ξ with Ξ = ( 80 g + χ 2 ) ξ 2 \Xi= (80g + \chi_2)\xi^2 Ξ = ( 80 g + χ 2 ) ξ 2 , we get, since 4 k ξ / ( 2 Ξ ) = Y 4k\xi/(2\Xi) = Y 4 k ξ / ( 2Ξ ) = Y and A 1 A 2 / ( 2 A 2 ξ ) = χ 1 / 2 A_1A_2/(2A_2\xi) = \chi_1/2 A 1 A 2 / ( 2 A 2 ξ ) = χ 1 /2 ,
Y < 8 χ 1 C 0 ℓ Y 2 + 2 ω L , 0 θ L , 0 ℓ Y Ξ + log ( 16 C 0 ′ ℓ Y 2 χ 1 Ξ ) + log C 2 Ξ . Y < 8\chi_1 C_0\ell_Y^2 + \frac{2\sqrt{\omega_{L,0}\theta_{L,0}}\ell_Y}{\Xi} + \frac{\log(16C'_0\ell_Y^2\chi_1\Xi)+\log C}{2\Xi}. Y < 8 χ 1 C 0 ℓ Y 2 + Ξ 2 ω L , 0 θ L , 0 ℓ Y + 2Ξ log ( 16 C 0 ′ ℓ Y 2 χ 1 Ξ ) + log C .
Put Ξ 0 = ( 80 + χ 2 ) ξ 0 2 \Xi_0 = (80+\chi_2)\xi_0^2 Ξ 0 = ( 80 + χ 2 ) ξ 0 2 , so that Ξ ≥ Ξ 0 \Xi\geq\Xi_0 Ξ ≥ Ξ 0 . The function u ↦ log ( Q u ) / u u \mapsto\log(Qu)/u u ↦ log ( Q u ) / u decreases for Q u ≥ e Qu \geq e Q u ≥ e , and Q = 16 C 0 ′ ℓ Y 2 χ 1 Q = 16C'_0\ell_Y^2\chi_1 Q = 16 C 0 ′ ℓ Y 2 χ 1 satisfies Q Ξ 0 > 2 ⋅ 10 6 Q\Xi_0 > 2\cdot10^6 Q Ξ 0 > 2 ⋅ 1 0 6 ; moreover log C > 0 \log C > 0 log C > 0 . Hence each of the last two terms is largest at Ξ = Ξ 0 \Xi= \Xi_0 Ξ = Ξ 0 , and
Y < Θ ( Y ) : = ψ 2 ℓ Y 2 + ψ 1 ℓ Y + ψ 0 log ℓ Y + ψ ∗ , Y < \Theta(Y) := \psi_2\ell_Y^2+\psi_1\ell_Y+\psi_0\log\ell_Y+\psi_*, Y < Θ ( Y ) := ψ 2 ℓ Y 2 + ψ 1 ℓ Y + ψ 0 log ℓ Y + ψ ∗ ,
ψ 2 = 8 χ 1 C 0 , ψ 1 = 2 ω L , 0 θ L , 0 Ξ 0 , ψ 0 = 1 Ξ 0 , ψ ∗ = log ( 16 C C 0 ′ A 1 , 0 A 2 , 0 ) 2 Ξ 0 , \psi_2 = 8\chi_1C_0,\qquad\psi_1 = \frac{2\sqrt{\omega_{L,0}\theta_{L,0}}}{\Xi_0},\qquad\psi_0 = \frac{1}{\Xi_0},\qquad\psi_* = \frac{\log(16CC'_0A_{1,0}A_{2,0})}{2\Xi_0}, ψ 2 = 8 χ 1 C 0 , ψ 1 = Ξ 0 2 ω L , 0 θ L , 0 , ψ 0 = Ξ 0 1 , ψ ∗ = 2 Ξ 0 log ( 16 C C 0 ′ A 1 , 0 A 2 , 0 ) ,
where ℓ Y = log ( Y + δ 0 ) + ℓ ∗ \ell_Y = \log(Y+\delta_0)+\ell_* ℓ Y = log ( Y + δ 0 ) + ℓ ∗ is now read as a function of the real variable Y ≥ Y 0 Y \geq Y_0 Y ≥ Y 0 .
We next show that Θ ( Y ) / Y \Theta(Y)/Y Θ ( Y ) / Y is strictly decreasing for Y ≥ Y 0 Y \geq Y_0 Y ≥ Y 0 . The coefficients ψ 2 , ψ 1 , ψ 0 \psi_2,\psi_1,\psi_0 ψ 2 , ψ 1 , ψ 0 are positive, and ψ ∗ > 0 \psi_* > 0 ψ ∗ > 0 by the evaluation below. Since d ℓ Y / d Y = 1 / ( Y + δ 0 ) < 1 / Y d\ell_Y/dY = 1/(Y+\delta_0) < 1/Y d ℓ Y / d Y = 1/ ( Y + δ 0 ) < 1/ Y and ℓ Y ≥ ℓ Y 0 > 2 \ell_Y \geq\ell_{Y_0} > 2 ℓ Y ≥ ℓ Y 0 > 2 ,
Y Θ ′ ( Y ) < 2 ψ 2 ℓ Y + ψ 1 + ψ 0 ℓ Y ≤ ψ 2 ℓ Y 2 + ψ 1 ℓ Y + ψ 0 log ℓ Y < Θ ( Y ) , Y\Theta'(Y) < 2\psi_2\ell_Y+\psi_1+\frac{\psi_0}{\ell_Y} \leq\psi_2\ell_Y^2+\psi_1\ell_Y+\psi_0\log\ell_Y < \Theta(Y), Y Θ ′ ( Y ) < 2 ψ 2 ℓ Y + ψ 1 + ℓ Y ψ 0 ≤ ψ 2 ℓ Y 2 + ψ 1 ℓ Y + ψ 0 log ℓ Y < Θ ( Y ) ,
because 2 ℓ Y ≤ ℓ Y 2 2\ell_Y \leq\ell_Y^2 2 ℓ Y ≤ ℓ Y 2 , 1 ≤ ℓ Y 1 \leq\ell_Y 1 ≤ ℓ Y and 1 / ℓ Y < log ℓ Y 1/\ell_Y < \log\ell_Y 1/ ℓ Y < log ℓ Y for ℓ Y ≥ 2 \ell_Y \geq2 ℓ Y ≥ 2 . Hence d d Y ( Θ ( Y ) / Y ) = ( Y Θ ′ ( Y ) − Θ ( Y ) ) / Y 2 < 0 \frac{d}{dY}(\Theta(Y)/Y) = (Y\Theta'(Y)-\Theta(Y))/Y^2 < 0 d Y d ( Θ ( Y ) / Y ) = ( Y Θ ′ ( Y ) − Θ ( Y )) / Y 2 < 0 .
Finally we evaluate at Y 0 Y_0 Y 0 . All constants above are explicit, and their values were computed in interval arithmetic with 256-bit endpoints, every comparison below holding for the full intervals. In the first case H 0 = 9.66560 … \mathcal{H}_0 = 9.66560\ldots H 0 = 9.66560 … , ω L , 0 = 4.00267 … \omega_{L,0} = 4.00267\ldots ω L , 0 = 4.00267 … , θ L , 0 = 1.05306 … \theta_{L,0} = 1.05306\ldots θ L , 0 = 1.05306 … , C 0 = 0.0329567 … C_0 = 0.0329567\ldots C 0 = 0.0329567 … , C 0 ′ = 0.0734636 … C'_0 = 0.0734636\ldots C 0 ′ = 0.0734636 … , ψ 2 = 1.054986 … \psi_2 = 1.054986\ldots ψ 2 = 1.054986 … , ψ 1 = 5.5089 … ⋅ 10 − 4 \psi_1 = 5.5089\ldots\cdot10^{-4} ψ 1 = 5.5089 … ⋅ 1 0 − 4 , ψ 0 = 1.3416 … ⋅ 10 − 4 \psi_0 = 1.3416\ldots\cdot10^{-4} ψ 0 = 1.3416 … ⋅ 1 0 − 4 , ψ ∗ = 7.9515 … ⋅ 10 − 4 \psi_* = 7.9515\ldots\cdot10^{-4} ψ ∗ = 7.9515 … ⋅ 1 0 − 4 , and
Θ ( Y 0 ) = 69.70554 … < 69.89401 … = Y 0 . \Theta(Y_0) = 69.70554\ldots< 69.89401\ldots= Y_0. Θ ( Y 0 ) = 69.70554 … < 69.89401 … = Y 0 .
In the second case H 0 = 9.66541 … \mathcal{H}_0 = 9.66541\ldots H 0 = 9.66541 … , ω L , 0 = 4.00267 … \omega_{L,0} = 4.00267\ldots ω L , 0 = 4.00267 … , θ L , 0 = 1.05306 … \theta_{L,0} = 1.05306\ldots θ L , 0 = 1.05306 … , C 0 = 0.0329086 … C_0 = 0.0329086\ldots C 0 = 0.0329086 … , C 0 ′ = 0.0734100 … C'_0 = 0.0734100\ldots C 0 ′ = 0.0734100 … , ψ 2 = 1.053472 … \psi_2 = 1.053472\ldots ψ 2 = 1.053472 … , ψ 1 = 5.2611 … ⋅ 10 − 4 \psi_1 = 5.2611\ldots\cdot10^{-4} ψ 1 = 5.2611 … ⋅ 1 0 − 4 , ψ 0 = 1.2812 … ⋅ 10 − 4 \psi_0 = 1.2812\ldots\cdot10^{-4} ψ 0 = 1.2812 … ⋅ 1 0 − 4 , ψ ∗ = 7.1772 … ⋅ 10 − 4 \psi_* = 7.1772\ldots\cdot10^{-4} ψ ∗ = 7.1772 … ⋅ 1 0 − 4 , and
Θ ( Y 0 ) = 69.60241 … < 69.88437 … = Y 0 . \Theta(Y_0) = 69.60241\ldots< 69.88437\ldots= Y_0. Θ ( Y 0 ) = 69.60241 … < 69.88437 … = Y 0 .
In both cases Q Ξ 0 > 2 ⋅ 10 6 Q\Xi_0 > 2\cdot10^6 Q Ξ 0 > 2 ⋅ 1 0 6 at ℓ Y = ℓ Y 0 \ell_Y = \ell_{Y_0} ℓ Y = ℓ Y 0 , which gives it for every larger ℓ Y \ell_Y ℓ Y . Since Θ ( Y ) / Y \Theta(Y)/Y Θ ( Y ) / Y decreases, Θ ( Y ) < Y \Theta(Y) < Y Θ ( Y ) < Y for every Y ≥ Y 0 Y \geq Y_0 Y ≥ Y 0 , which contradicts Y < Θ ( Y ) Y < \Theta(Y) Y < Θ ( Y ) at our Y > Y 0 Y > Y_0 Y > Y 0 .
Theorem 4.5. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with
b ≥ 3355 2 a or a c ≥ 3075 ( b − a ) . b \geq3355^2a \qquad\text{or}\qquad\sqrt{ac} \geq3075\,(b-a). b ≥ 335 5 2 a or a c ≥ 3075 ( b − a ) . Then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Proof. Suppose that d d d is an irregular extension of { a , b , c } \{a,b,c\} { a , b , c } , and let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be its index system (Proposition 2.11), with the linear forms Λ x , Λ z \Lambda_x,\Lambda_z Λ x , Λ z , the numbers D λ , E λ D_\lambda,E_\lambda D λ , E λ and the gap ratios R x , R z R_x,R_z R x , R z of Definition 2.16 . By Lemma 2.14, n < m < l n < m < l n < m < l . In each of the two cases we exhibit numbers η 1 , η 2 , ξ , g , k \eta_1,\eta_2,\xi,g,k η 1 , η 2 , ξ , g , k that satisfy all the conditions which Lemma 4.4 excludes; the first case uses the linear form Λ x \Lambda_x Λ x , the second the linear form Λ z \Lambda_z Λ z . Throughout, h h h is the height of Definition 4.1, and we use h ( u v ) ≤ h ( u ) + h ( v ) h(uv) \le h(u) + h(v) h ( uv ) ≤ h ( u ) + h ( v ) .
By item 2 of Lemma 4.2, the numbers α \alpha α , β \beta β , γ \gamma γ , D λ D_\lambda D λ , E λ E_\lambda E λ lie in the real biquadratic field K = Q ( a b , a c ) \mathcal{K} = \mathbb{Q}(\sqrt{ab},\sqrt{ac}) K = Q ( ab , a c ) .
The case b ≥ 3355 2 a b \ge3355^2a b ≥ 335 5 2 a . Put
η 1 = β α , η 2 = α 2 ( l − m ) D λ , ξ = log α , g = l − m ≥ 1 , k = m , \eta_1 = \frac{\beta}{\alpha}, \qquad\eta_2 = \alpha^{2(l-m)}D_\lambda, \qquad\xi= \log\alpha, \qquad g = l-m \ge1, \qquad k = m, η 1 = α β , η 2 = α 2 ( l − m ) D λ , ξ = log α , g = l − m ≥ 1 , k = m ,
and ( T , ξ 0 , h 0 , C ) = ( 3355 , log 6710 , 2 , 4.01 ) (T,\xi_0,h_0,C) = (3355,\log6710,2,4.01) ( T , ξ 0 , h 0 , C ) = ( 3355 , log 6710 , 2 , 4.01 ) . We check the conditions of Lemma 4.4.
By Lemma 4.2 (item 6), α \alpha α , β / α \beta/\alpha β / α and D λ D_\lambda D λ exceed 1 and are multiplicatively independent; hence η 1 , η 2 > 1 \eta_1,\eta_2 > 1 η 1 , η 2 > 1 , and η 1 i η 2 j = 1 \eta_1^i\eta_2^j = 1 η 1 i η 2 j = 1 with integers i , j i,j i , j forces i = 2 g j = j = 0 i = 2gj = j = 0 i = 2 g j = j = 0 , so η 1 , η 2 \eta_1,\eta_2 η 1 , η 2 are multiplicatively independent. By Lemma 4.2 (item 3), η 1 \eta_1 η 1 has degree 4, so [ Q ( η 1 , η 2 ) : Q ] = 4 [\mathbb{Q}(\eta_1,\eta_2):\mathbb{Q}] = 4 [ Q ( η 1 , η 2 ) : Q ] = 4 as Q ( η 1 , η 2 ) ⊆ K \mathbb{Q}(\eta_1,\eta_2) \subseteq\mathcal{K} Q ( η 1 , η 2 ) ⊆ K .
By Lemma 2.18 (item 2), log η 1 < a / b ≤ 1 / T \log\eta_1 < \sqrt{a/b} \le1/T log η 1 < a / b ≤ 1/ T ; by Lemma 4.2 (item 3), h ( η 1 ) = 1 2 log β = 1 2 ( ξ + log η 1 ) < 1 2 ( ξ + 1 / T ) h(\eta_1) = \frac{1}{2}\log\beta= \frac{1}{2}(\xi+ \log\eta_1) < \frac{1}{2}(\xi+ 1/T) h ( η 1 ) = 2 1 log β = 2 1 ( ξ + log η 1 ) < 2 1 ( ξ + 1/ T ) .
By Lemma 4.2 (item 4), log η 2 = 2 g ξ + log D λ < 2 g ξ + 2 a / b ≤ 2 g ξ + 2 / T \log\eta_2 = 2g\xi+ \log D_\lambda< 2g\xi+ 2\sqrt{a/b} \le2g\xi+ 2/T log η 2 = 2 g ξ + log D λ < 2 g ξ + 2 a / b ≤ 2 g ξ + 2/ T . By items 3, 4 and 1 of Lemma 4.2, h ( η 2 ) ≤ h ( α 2 g ) + h ( D λ ) < g ξ + log c < ( g + 2 ) ξ h(\eta_2) \le h(\alpha^{2g}) + h(D_\lambda) < g\xi+ \log c < (g+2)\xi h ( η 2 ) ≤ h ( α 2 g ) + h ( D λ ) < g ξ + log c < ( g + 2 ) ξ , since log c < 2 log α \log c < 2\log\alpha log c < 2 log α .
log η 2 − 2 k log η 1 = 2 l log α − 2 m log β + log D λ = Λ x \log\eta_2 - 2k\log\eta_1 = 2l\log\alpha- 2m\log\beta+ \log D_\lambda= \Lambda_x log η 2 − 2 k log η 1 = 2 l log α − 2 m log β + log D λ = Λ x , and by Lemma 2.18 (item 4) 0 < Λ x < 4.01 β − 4 m < 4.01 e − 4 k ξ 0 < \Lambda_x < 4.01\beta^{-4m} < 4.01e^{-4k\xi} 0 < Λ x < 4.01 β − 4 m < 4.01 e − 4 k ξ , as β > α \beta> \alpha β > α .
This contradicts Lemma 4.4.
The case a c ≥ 3075 ( b − a ) \sqrt{ac} \ge3075(b-a) a c ≥ 3075 ( b − a ) . Put
η 1 = γ β , η 2 = β 2 ( m − n ) E λ , ξ = log β , g = m − n ≥ 1 , k = n , \eta_1 = \frac{\gamma}{\beta}, \qquad\eta_2 = \beta^{2(m-n)}E_\lambda, \qquad\xi= \log\beta, \qquad g = m-n \ge1, \qquad k = n, η 1 = β γ , η 2 = β 2 ( m − n ) E λ , ξ = log β , g = m − n ≥ 1 , k = n ,
and ( T , ξ 0 , h 0 , C ) = ( 3075 , log 12300 , 1 , 2 ) (T,\xi_0,h_0,C) = (3075,\log12300,1,2) ( T , ξ 0 , h 0 , C ) = ( 3075 , log 12300 , 1 , 2 ) . We first note that b < 3 2 a b < \frac{3}{2}a b < 2 3 a : by Lemma 2.4 (item 4), c ≤ 2 ( a + b ) < 4 b c \le2(a+b) < 4b c ≤ 2 ( a + b ) < 4 b , so a c < 2 b \sqrt{ac} < 2b a c < 2 b and b − a ≤ a c / 3075 < 2 b / 3075 b-a \le\sqrt{ac}/3075 < 2b/3075 b − a ≤ a c /3075 < 2 b /3075 , whence a > 3073 3075 b > 2 3 b a > \frac{3073}{3075}b > \frac{2}{3}b a > 3075 3073 b > 3 2 b . We check the conditions of Lemma 4.4.
Since b − a ≥ 2 b-a \ge2 b − a ≥ 2 (Lemma 2.4, item 4) and s > a c s > \sqrt{ac} s > a c , we have β > 2 a c ≥ 6150 ( b − a ) ≥ 12300 \beta> 2\sqrt{ac} \ge6150(b-a) \ge12300 β > 2 a c ≥ 6150 ( b − a ) ≥ 12300 , so ξ > ξ 0 \xi> \xi_0 ξ > ξ 0 .
By Lemma 4.2 (item 6), η 1 , η 2 > 1 \eta_1,\eta_2 > 1 η 1 , η 2 > 1 and they are multiplicatively independent. By Lemma 4.2 (item 3), η 1 \eta_1 η 1 has degree 4, so [ Q ( η 1 , η 2 ) : Q ] = 4 [\mathbb{Q}(\eta_1,\eta_2):\mathbb{Q}] = 4 [ Q ( η 1 , η 2 ) : Q ] = 4 as Q ( η 1 , η 2 ) ⊆ K \mathbb{Q}(\eta_1,\eta_2) \subseteq\mathcal{K} Q ( η 1 , η 2 ) ⊆ K .
By Lemma 2.18 (item 2), log η 1 < ( b − a ) / a c ≤ 1 / T \log\eta_1 < (b-a)/\sqrt{ac} \le1/T log η 1 < ( b − a ) / a c ≤ 1/ T ; by Lemma 4.2 (item 3), h ( η 1 ) = 1 2 log γ = 1 2 ( ξ + log η 1 ) < 1 2 ( ξ + 1 / T ) h(\eta_1) = \frac{1}{2}\log\gamma= \frac{1}{2}(\xi+ \log\eta_1) < \frac{1}{2}(\xi+ 1/T) h ( η 1 ) = 2 1 log γ = 2 1 ( ξ + log η 1 ) < 2 1 ( ξ + 1/ T ) .
By Lemma 4.2 (item 5), log η 2 = 2 g ξ + log E λ < 2 g ξ + 2 ( b − a ) / a c ≤ 2 g ξ + 2 / T \log\eta_2 = 2g\xi+ \log E_\lambda< 2g\xi+ 2(b-a)/\sqrt{ac} \le2g\xi+ 2/T log η 2 = 2 g ξ + log E λ < 2 g ξ + 2 ( b − a ) / a c ≤ 2 g ξ + 2/ T . By items 3 and 5 of Lemma 4.2, the latter applicable as b ≤ 3 2 a b \le\frac{3}{2}a b ≤ 2 3 a , h ( η 2 ) ≤ h ( β 2 g ) + h ( E λ ) < g ξ + ξ h(\eta_2) \le h(\beta^{2g}) + h(E_\lambda) < g\xi+ \xi h ( η 2 ) ≤ h ( β 2 g ) + h ( E λ ) < g ξ + ξ .
This contradicts Lemma 4.4 . Hence { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Corollary 4.6. If an Euler triple { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension, then 1 + 1 1538 < b a < 3355 2 1+\frac{1}{1538}<\frac{b}{a}<3355^{2} 1 + 1538 1 < a b < 335 5 2 .
Proof. By Theorem 4.5 , b < 3355 2 a b<3355^{2}a b < 335 5 2 a and a c < 3075 ( b − a ) \sqrt{ac}<3075(b-a) a c < 3075 ( b − a ) . By Lemma 2.4 (item 4), c > 4 a c>4a c > 4 a , so a c > 2 a \sqrt{ac}>2a a c > 2 a . Hence 2 a < 3075 ( b − a ) 2a<3075(b-a) 2 a < 3075 ( b − a ) , that is b / a > 1 + 2 3075 > 1 + 1 1538 b/a>1+\frac{2}{3075}>1+\frac{1}{1538} b / a > 1 + 3075 2 > 1 + 1538 1 .
The heights of Lemma 4.2 and Laurent’s theorem are used again in § 6 .
Regularity from pair coordinates
This section proves the pair-coordinate criterion (Theorem 5.19 ). The key idea is that an Euler triple, written in pair coordinates, falls into the setting of Rickert’s simultaneous approximations to square roots [26 ] . In a pair representation a = p 1 q 1 a=p_{1}q_{1} a = p 1 q 1 , b = p 2 q 2 b=p_{2}q_{2} b = p 2 q 2 with p 1 q 2 − p 2 q 1 = ± 2 p_{1}q_{2}-p_{2}q_{1}=\pm2 p 1 q 2 − p 2 q 1 = ± 2 , the numbers a / c \sqrt{a/c} a / c and b / c \sqrt{b/c} b / c , after rescaling, become square roots 1 + a j / N \sqrt{1+a_{j}/N} 1 + a j / N with integer shifts a j a_{j} a j and an integer N N N . The hypergeometric method then gives them an effective simultaneous approximation measure (Theorem 5.14 ), with an exponent below 2 exactly when 40 p 3 2 < ω p 1 q 3 40p_{3}^{2}<\omega p_{1}q_{3} 40 p 3 2 < ω p 1 q 3 , while the common term z z z of an irregular extension approximates them with exponent 2. Comparing the two exponents caps the index m m m (Lemma 5.18 ); under the hypotheses of the criterion the cap is a multiple of log c \log c log c , against the floors for m m m given by the quartic index floor (Theorem 3.6 ) and, in the first case, by Lemma 5.17 .
The section has three parts: Subsection 5.1 states the method for arbitrary integer shifts, Subsection 5.2 applies it to a pair representation, and Subsection 5.3 bounds the index and proves the criterion. The weight lies in Theorem 5.14 and Lemma 5.18 ; the rest is the machinery of the method, stated once in general form, and bookkeeping in pair coordinates. From the earlier sections the section uses the index system and its common term z z z (Proposition 2.11, Lemma 2.15 ), the ordering of Lemma 2.14, the order-two quotient k c k_{c} k c of Convention 3.1, and the quartic index floor.
Rickert’s approximants and Bennett’s criterion
We state the hypergeometric method for arbitrary distinct integer shifts a 0 , a 1 , a 2 a_{0},a_{1},a_{2} a 0 , a 1 , a 2 ; nothing in this subsection involves a Diophantine triple. Rickert’s approximants [26 ] are polynomials p i j p_{ij} p ij of degree at most k k k whose linear forms Λ i = ∑ j p i j ϑ j \Lambda_{i}=\sum_{j}p_{ij}\vartheta_{j} Λ i = ∑ j p ij ϑ j in the three square roots ϑ j = ( 1 + a j x ) 1 / 2 \vartheta_{j}=(1+a_{j}x)^{1/2} ϑ j = ( 1 + a j x ) 1/2 vanish to order 3 k 3k 3 k at x = 0 x=0 x = 0 , as the proof of Lemma 5.3 shows. Three properties make them usable. Their determinant does not vanish (Lemma 5.3, after Rickert and Bennett), and both this fact and the bounds of Lemma 5.12 rest on one residue representation (Lemma 5.2 ). Their denominators are controlled: the coefficient products are integers divisible by every prime of the Cip̄u–Fujita prime product [9 ] (Lemma 5.5 ), which exceeds 1.6 k / ( 4.09 ⋅ 10 13 ) 1.6^{k}/(4.09\cdot10^{13}) 1. 6 k / ( 4.09 ⋅ 1 0 13 ) (Lemma 5.6 ). And Bennett’s criterion [6 ] (Lemma 5.7) turns integer approximants with entries of size P k P^{k} P k and linear forms of size L − k L^{-k} L − k into an approximation measure with exponent 1 + log P / log L 1+\log P/\log L 1 + log P / log L .
Convention 5.1. Rickert’s approximants ([26 ] , in the normalisation of [10 ] ). Let the shifts a 0 , a 1 , a 2 a_{0},a_{1},a_{2} a 0 , a 1 , a 2 be distinct integers, M = max l ∣ a l ∣ M=\max_{l}|a_{l}| M = max l ∣ a l ∣ , and
F ( z ) = ∏ l = 0 2 ( z − a l ) , k i l = k + δ i l ( 0 ≤ i , l ≤ 2 ) , F(z)=\prod_{l=0}^{2}(z-a_{l}), \qquad k_{il}=k+\delta_{il}\quad(0\leq i,l\leq2), F ( z ) = l = 0 ∏ 2 ( z − a l ) , k i l = k + δ i l ( 0 ≤ i , l ≤ 2 ) ,
where k ≥ 1 k \ge1 k ≥ 1 is an integer and δ i l \delta_{il} δ i l the Kronecker delta. For 0 ≤ i , j ≤ 2 0 \le i,j \le2 0 ≤ i , j ≤ 2 the Rickert polynomial p i j ( x ) = p i j ( k ) ( x ) p_{ij}(x)=p_{ij}^{(k)}(x) p ij ( x ) = p ij ( k ) ( x ) is
p i j ( x ) = ∑ h 0 + h 1 + h 2 = k i j − 1 ( k + 1 2 h j ) x h j ( 1 + a j x ) k − h j ∏ l ≠ j ( − k i l h l ) ( a j − a l ) − ( k i l + h l ) . p_{ij}(x)=\sum_{h_0+h_1+h_2=k_{ij}-1}\binom{k+\frac{1}{2}}{h_j}x^{h_j}(1+a_jx)^{k-h_j}\prod_{l\ne j}\binom{-k_{il}}{h_l}(a_j-a_l)^{-(k_{il}+h_l)}. p ij ( x ) = h 0 + h 1 + h 2 = k ij − 1 ∑ ( h j k + 2 1 ) x h j ( 1 + a j x ) k − h j l = j ∏ ( h l − k i l ) ( a j − a l ) − ( k i l + h l ) .
the sum over all non-negative integers h 0 , h 1 , h 2 h_0,h_1,h_2 h 0 , h 1 , h 2 with h 0 + h 1 + h 2 = k i j − 1 h_0+h_1+h_2=k_{ij}-1 h 0 + h 1 + h 2 = k ij − 1 . For a real number x x x with ∣ x ∣ M < 1 |x|M<1 ∣ x ∣ M < 1 the approximated numbers are
ϑ j = ϑ j ( x ) = ( 1 + a j x ) 1 / 2 > 0 ( j = 0 , 1 , 2 ) , \vartheta_j=\vartheta_j(x)=(1+a_jx)^{1/2}>0\qquad(j=0,1,2), ϑ j = ϑ j ( x ) = ( 1 + a j x ) 1/2 > 0 ( j = 0 , 1 , 2 ) ,
and the linear forms are Λ i ( x ) = ∑ j = 0 2 p i j ( x ) ϑ j ( x ) \Lambda_i(x)=\sum_{j=0}^{2}p_{ij}(x)\vartheta_j(x) Λ i ( x ) = ∑ j = 0 2 p ij ( x ) ϑ j ( x ) ( i = 0 , 1 , 2 ) (i=0,1,2) ( i = 0 , 1 , 2 ) . The index k k k is suppressed when it is fixed. Throughout, ( 1 + z x ) 1 / 2 (1+zx)^{1/2} ( 1 + z x ) 1/2 denotes the principal branch, analytic in the disc ∣ z ∣ < 1 / ∣ x ∣ |z|<1/|x| ∣ z ∣ < 1/∣ x ∣ .
The coefficient products of order k k k are the rational numbers
T i j ( k ) ( h ) = 2 2 k − 1 ( k + 1 2 h j ) ∏ l ≠ j ( − k i l h l ) , \mathcal{T}_{ij}^{(k)}(h)=2^{2k-1}\binom{k+\frac{1}{2}}{h_j}\prod_{l\ne j}\binom{-k_{il}}{h_l}, T ij ( k ) ( h ) = 2 2 k − 1 ( h j k + 2 1 ) l = j ∏ ( h l − k i l ) ,
for i , j ∈ { 0 , 1 , 2 } i,j\in\{0,1,2\} i , j ∈ { 0 , 1 , 2 } and every triple h = ( h 0 , h 1 , h 2 ) h=(h_0,h_1,h_2) h = ( h 0 , h 1 , h 2 ) of non-negative integers with h 0 + h 1 + h 2 = k i j − 1 h_0+h_1+h_2=k_{ij}-1 h 0 + h 1 + h 2 = k ij − 1 . When all of them are integers, G k G_k G k denotes their greatest common divisor.
Lemma 5.2. Let a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 be distinct integers, let k ≥ 1 k\ge1 k ≥ 1 be an integer, and let p i j = p i j ( k ) p_{ij}=p_{ij}^{(k)} p ij = p ij ( k ) be Rickert’s polynomials of order k k k for these shifts, with M M M , F F F and k i l k_{il} k i l as in Convention 5.1. Put
g i ( z ) = ∏ l = 0 2 ( z − a l ) − k i l = ( ( z − a i ) F ( z ) k ) − 1 ( i = 0 , 1 , 2 ) . g_i(z)=\prod_{l=0}^{2}(z-a_l)^{-k_{il}}=\left((z-a_i)F(z)^k\right)^{-1}\qquad(i=0,1,2). g i ( z ) = l = 0 ∏ 2 ( z − a l ) − k i l = ( ( z − a i ) F ( z ) k ) − 1 ( i = 0 , 1 , 2 ) . Then for all 0 ≤ i , j ≤ 2 0\le i,j\le2 0 ≤ i , j ≤ 2 :
in the ring Q [ [ x ] ] \mathbb{Q}[[x]] Q [[ x ]] of formal power series, with ϑ j = ∑ e ≥ 0 ( 1 / 2 e ) a j e x e \vartheta_j=\sum_{e\ge0}\binom{1/2}{e}a_j^e x^e ϑ j = ∑ e ≥ 0 ( e 1/2 ) a j e x e , the coefficient of x u x^u x u in p i j ( x ) ϑ j p_{ij}(x)\vartheta_j p ij ( x ) ϑ j is ( k + 1 / 2 u ) Res z = a j z u g i ( z ) \binom{k+1/2}{u}\operatorname{Res}_{z=a_j}z^u g_i(z) ( u k + 1/2 ) Res z = a j z u g i ( z ) for every integer u ≥ 0 u\ge0 u ≥ 0 ; in particular p i j ( 0 ) = Res z = a j g i ( z ) p_{ij}(0)=\operatorname{Res}_{z=a_j}g_i(z) p ij ( 0 ) = Res z = a j g i ( z ) ;
for every real x x x with ∣ x ∣ M < 1 |x|M<1 ∣ x ∣ M < 1 ,
p i j ( x ) ϑ j ( x ) = Res z = a j ψ i ( z ) , ψ i ( z ) = ( 1 + z x ) k + 1 / 2 g i ( z ) . p_{ij}(x)\vartheta_j(x)=\operatorname{Res}_{z=a_j}\psi_i(z),\qquad\psi_i(z)=(1+zx)^{k+1/2}g_i(z). p ij ( x ) ϑ j ( x ) = Res z = a j ψ i ( z ) , ψ i ( z ) = ( 1 + z x ) k + 1/2 g i ( z ) .
Proof. The residue form is Rickert’s [26 ] , who defines the approximants by contour integrals and derives the formula of Convention 5.1 from them, and Bennett [6 ] estimates the approximants through the same integrals; we include a proof for completeness. In view of the formal expansion ( 1 + z x ) k + 1 / 2 = ∑ u ≥ 0 ( k + 1 / 2 u ) z u x u (1+zx)^{k+1/2}=\sum_{u\ge0}\binom{k+1/2}{u}z^ux^u ( 1 + z x ) k + 1/2 = ∑ u ≥ 0 ( u k + 1/2 ) z u x u , item 1 is the formal counterpart of item 2; we prove the two items separately, item 1 in Q [ [ x ] ] \mathbb{Q}[[x]] Q [[ x ]] and item 2 for real x x x . For a function ϕ \phi ϕ meromorphic near a a a , Res z = a ϕ \operatorname{Res}_{z=a}\phi Res z = a ϕ is the coefficient of ( z − a ) − 1 (z-a)^{-1} ( z − a ) − 1 in its Laurent expansion at a a a .
We first express the coefficients of p i j p_{ij} p ij by residues. For e ≥ 0 e\ge0 e ≥ 0 put ρ i j ( e ) = Res z = a j ( z − a j ) e g i ( z ) \rho_{ij}(e)=\operatorname{Res}_{z=a_j}(z-a_j)^e g_i(z) ρ ij ( e ) = Res z = a j ( z − a j ) e g i ( z ) . Since g i = ( z − a j ) − k i j ∏ l ≠ j ( z − a l ) − k i l g_i=(z-a_j)^{-k_{ij}}\prod_{l\ne j}(z-a_l)^{-k_{il}} g i = ( z − a j ) − k ij ∏ l = j ( z − a l ) − k i l and
∏ l ≠ j ( z − a l ) − k i l = ∏ l ≠ j ∑ h l ≥ 0 ( − k i l h l ) ( a j − a l ) − ( k i l + h l ) ( z − a j ) h l \prod_{l\ne j}(z-a_l)^{-k_{il}}=\prod_{l\ne j}\sum_{h_l\ge0}\binom{-k_{il}}{h_l}(a_j-a_l)^{-(k_{il}+h_l)}(z-a_j)^{h_l} l = j ∏ ( z − a l ) − k i l = l = j ∏ h l ≥ 0 ∑ ( h l − k i l ) ( a j − a l ) − ( k i l + h l ) ( z − a j ) h l
near z = a j z=a_j z = a j , we get ρ i j ( e ) = ∑ h l ≥ 0 ( l ≠ j ) ∑ l ≠ j h l = k i j − 1 − e ∏ l ≠ j ( − k i l ) h l ( a j − a l ) − ( k i l + h l ) \rho_{ij}(e)=\sum_{\substack{h_l\geq0\ (l\ne j)\\ \sum_{l\ne j}h_l=k_{ij}-1-e}}\prod_{l\ne j}(-k_{il})_{h_l}(a_j-a_l)^{-(k_{il}+h_l)} ρ ij ( e ) = ∑ h l ≥ 0 ( l = j ) ∑ l = j h l = k ij − 1 − e ∏ l = j ( − k i l ) h l ( a j − a l ) − ( k i l + h l ) , the sum over h l ≥ 0 h_l\geq0 h l ≥ 0 (l ≠ j l\ne j l = j ) with ∑ l ≠ j h l = k i j − 1 − e \sum_{l\ne j}h_l=k_{ij}-1-e ∑ l = j h l = k ij − 1 − e , for 0 ≤ e ≤ k i j − 1 0\leq e\leq k_{ij}-1 0 ≤ e ≤ k ij − 1 , and ρ i j ( e ) = 0 \rho_{ij}(e)=0 ρ ij ( e ) = 0 for e ≥ k i j e\geq k_{ij} e ≥ k ij . Comparing with the formula of Convention 5.1 (with e = h j e=h_j e = h j , and k i j − 1 ≤ k k_{ij}-1\leq k k ij − 1 ≤ k ),
p i j ( x ) = ∑ e = 0 k ( k + 1 2 e ) ρ i j ( e ) x e ( 1 + a j x ) k − e , p i j ( 0 ) = ρ i j ( 0 ) . p_{ij}(x)=\sum_{e=0}^{k}\binom{k+\frac{1}{2}}{e}\rho_{ij}(e)x^e(1+a_jx)^{k-e},\qquad p_{ij}(0)=\rho_{ij}(0). p ij ( x ) = e = 0 ∑ k ( e k + 2 1 ) ρ ij ( e ) x e ( 1 + a j x ) k − e , p ij ( 0 ) = ρ ij ( 0 ) .
We next prove item 1. (For ∣ x ∣ M < 1 |x|M<1 ∣ x ∣ M < 1 the series ϑ j \vartheta_j ϑ j converges to the ϑ j \vartheta_j ϑ j of Convention 5.1.) For a rational number w w w put ( 1 + a j x ) w : = ∑ e ( w e ) a j e x e (1+a_jx)^w:=\sum_e\binom{w}{e}a_j^ex^e ( 1 + a j x ) w := ∑ e ( e w ) a j e x e ; Vandermonde’s identity gives ( 1 + a j x ) w ( 1 + a j x ) w ′ = ( 1 + a j x ) w + w ′ (1+a_jx)^w(1+a_jx)^{w'}=(1+a_jx)^{w+w'} ( 1 + a j x ) w ( 1 + a j x ) w ′ = ( 1 + a j x ) w + w ′ , and for w ∈ Z ≥ 0 w\in\mathbb{Z}_{\geq0} w ∈ Z ≥ 0 the series is the polynomial ( 1 + a j x ) w (1+a_jx)^w ( 1 + a j x ) w . Hence p i j ( x ) ϑ j = ∑ e ( k + 1 / 2 e ) ρ i j ( e ) x e ( 1 + a j x ) k + 1 / 2 − e p_{ij}(x)\vartheta_j=\sum_e\binom{k+1/2}{e}\rho_{ij}(e)x^e(1+a_jx)^{k+1/2-e} p ij ( x ) ϑ j = ∑ e ( e k + 1/2 ) ρ ij ( e ) x e ( 1 + a j x ) k + 1/2 − e , and its coefficient of x u x^u x u (u ≥ 0 u\geq0 u ≥ 0 ) is
∑ e = 0 u ( k + 1 2 e ) ( k + 1 2 − e u − e ) a j u − e ρ i j ( e ) = ( k + 1 2 u ) ∑ e = 0 u ( u e ) a j u − e ρ i j ( e ) = ( k + 1 2 u ) Res z = a j z u g i ( z ) , \begin{aligned}
\sum_{e=0}^{u}\binom{k+\frac{1}{2}}{e}\binom{k+\frac{1}{2}-e}{u-e}a_j^{u-e}\rho_{ij}(e)
=
\binom{k+\frac{1}{2}}{u}\sum_{e=0}^{u}\binom{u}{e}a_j^{u-e}\rho_{ij}(e) \\
&=\binom{k+\frac{1}{2}}{u}\operatorname{Res}_{z=a_j}z^u g_i(z),
\end{aligned} e = 0 ∑ u ( e k + 2 1 ) ( u − e k + 2 1 − e ) a j u − e ρ ij ( e ) = ( u k + 2 1 ) e = 0 ∑ u ( e u ) a j u − e ρ ij ( e ) = ( u k + 2 1 ) Res z = a j z u g i ( z ) ,
using ( w e ) ( w − e u − e ) = ( w u ) ( u e ) \binom{w}{e}\binom{w-e}{u-e}=\binom{w}{u}\binom{u}{e} ( e w ) ( u − e w − e ) = ( u w ) ( e u ) , ρ i j ( e ) = 0 \rho_{ij}(e)=0 ρ ij ( e ) = 0 for e > k e>k e > k , and z u = ∑ e ( u e ) a j u − e ( z − a j ) e z^u=\sum_e\binom{u}{e}a_j^{u-e}(z-a_j)^e z u = ∑ e ( e u ) a j u − e ( z − a j ) e . As ρ i j ( 0 ) = Res z = a j g i ( z ) \rho_{ij}(0)=\operatorname{Res}_{z=a_j}g_i(z) ρ ij ( 0 ) = Res z = a j g i ( z ) , this proves item 1.
We finally prove item 2. Let x x x be real with ∣ x ∣ M < 1 |x|M<1 ∣ x ∣ M < 1 , so that ψ i \psi_i ψ i is analytic in the disc ∣ z ∣ < 1 / ∣ x ∣ |z|<1/|x| ∣ z ∣ < 1/∣ x ∣ except for poles at a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 . Near a j a_j a j , ψ i ( z ) = χ ( z ) ( z − a j ) − k i j \psi_i(z)=\chi(z)(z-a_j)^{-k_{ij}} ψ i ( z ) = χ ( z ) ( z − a j ) − k ij with χ ( z ) = ( 1 + z x ) k + 1 / 2 ∏ l ≠ j ( z − a l ) − k i l \chi(z)=(1+zx)^{k+1/2}\prod_{l\ne j}(z-a_l)^{-k_{il}} χ ( z ) = ( 1 + z x ) k + 1/2 ∏ l = j ( z − a l ) − k i l analytic at a j a_j a j , so the residue is χ ( e ) ( a j ) / e ! \chi^{(e)}(a_j)/e! χ ( e ) ( a j ) / e ! , e = k i j − 1 e=k_{ij}-1 e = k ij − 1 . By Leibniz’s rule, χ ( e ) / e ! \chi^{(e)}/e! χ ( e ) / e ! is the sum over h j + h l 1 + h l 2 = e h_j+h_{l_1}+h_{l_2}=e h j + h l 1 + h l 2 = e , where { l 1 , l 2 } = { 0 , 1 , 2 } ∖ { j } \{l_1,l_2\}=\{0,1,2\}\setminus\{j\} { l 1 , l 2 } = { 0 , 1 , 2 } ∖ { j } , of the products of the h h h -th derivatives divided by h ! h! h ! , and 1 h ! d h d z h ( 1 + z x ) k + 1 / 2 = ( k + 1 / 2 h ) x h ( 1 + z x ) k − h ( 1 + z x ) 1 / 2 \frac{1}{h!}\frac{d^h}{dz^h}(1+zx)^{k+1/2}=\binom{k+1/2}{h}x^h(1+zx)^{k-h}(1+zx)^{1/2} h ! 1 d z h d h ( 1 + z x ) k + 1/2 = ( h k + 1/2 ) x h ( 1 + z x ) k − h ( 1 + z x ) 1/2 , 1 h ! d h d z h ( z − a l ) − k i l = ( − k i l h ) ( z − a l ) − k i l − h \frac{1}{h!}\frac{d^h}{dz^h}(z-a_l)^{-k_{il}}=\binom{-k_{il}}{h}(z-a_l)^{-k_{il}-h} h ! 1 d z h d h ( z − a l ) − k i l = ( h − k i l ) ( z − a l ) − k i l − h . Evaluating at z = a j z=a_j z = a j gives exactly ϑ j p i j ( x ) \vartheta_jp_{ij}(x) ϑ j p ij ( x ) by the formula of Convention 5.1.
Lemma 5.3. Let a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 be distinct integers, let k ≥ 1 k\geq1 k ≥ 1 be an integer, and let p i j ( x ) p_{ij}(x) p ij ( x ) be Rickert’s polynomials of order k k k for these shifts (Convention 5.1). Then there is a rational number C k ≠ 0 C_k\ne0 C k = 0 such that
det ( p i j ( x ) ) 0 ≤ i , j ≤ 2 = C k x 3 k . \det(p_{ij}(x))_{0\leq i,j\leq2}=C_kx^{3k}. det ( p ij ( x ) ) 0 ≤ i , j ≤ 2 = C k x 3 k . In particular det ( p i j ( x ) ) ≠ 0 \det(p_{ij}(x))\ne0 det ( p ij ( x )) = 0 for every real x ≠ 0 x\ne0 x = 0 .
Proof. The nonvanishing of this determinant is due to Rickert [26 ] , and Bennett [5 Section 2] states the identity with an explicit value of C k C_k C k , for any number of shifts. We include a proof for completeness. For a rational function ϕ ∈ Q ( z ) \phi\in\mathbb{Q}(z) ϕ ∈ Q ( z ) , Res z = a ϕ \operatorname{Res}_{z=a}\phi Res z = a ϕ is the coefficient of ( z − a ) − 1 (z-a)^{-1} ( z − a ) − 1 in its Laurent expansion at a a a , and [ z − 1 ] ∞ ϕ [z^{-1}]_\infty\phi [ z − 1 ] ∞ ϕ is the coefficient of z − 1 z^{-1} z − 1 in its expansion in powers of 1 / z 1/z 1/ z . We use the residue theorem for rational functions: ∑ a Res z = a ϕ = [ z − 1 ] ∞ ϕ \sum_a\operatorname{Res}_{z=a}\phi=[z^{-1}]_\infty\phi ∑ a Res z = a ϕ = [ z − 1 ] ∞ ϕ , the sum over the poles a a a of ϕ \phi ϕ ; this follows from the partial fraction decomposition, since [ z − 1 ] ∞ ( z − a ) − e [z^{-1}]_\infty(z-a)^{-e} [ z − 1 ] ∞ ( z − a ) − e is 1 for e = 1 e=1 e = 1 and 0 for e ≥ 2 e\geq2 e ≥ 2 , and polynomials contribute nothing. With F F F and k i l k_{il} k i l as in Convention 5.1, let
g i ( z ) = ∏ l = 0 2 ( z − a l ) − k i l = ( ( z − a i ) F ( z ) k ) − 1 ( i = 0 , 1 , 2 ) g_i(z)=\prod_{l=0}^{2}(z-a_l)^{-k_{il}}=\left((z-a_i)F(z)^k\right)^{-1}\qquad(i=0,1,2) g i ( z ) = l = 0 ∏ 2 ( z − a l ) − k i l = ( ( z − a i ) F ( z ) k ) − 1 ( i = 0 , 1 , 2 )
as in Lemma 5.2, and put ρ i j = Res z = a j g i ( z ) \rho_{ij}=\operatorname{Res}_{z=a_j}g_i(z) ρ ij = Res z = a j g i ( z ) . Each p i j p_{ij} p ij is a polynomial in x x x of degree at most k k k with rational coefficients, since every term of its defining sum is a rational multiple of x h j ( 1 + a j x ) k − h j x^{h_j}(1+a_jx)^{k-h_j} x h j ( 1 + a j x ) k − h j with h j ≤ k i j − 1 ≤ k h_j\leq k_{ij}-1\leq k h j ≤ k ij − 1 ≤ k . By Lemma 5.2, item 1, p i j ( 0 ) = ρ i j p_{ij}(0)=\rho_{ij} p ij ( 0 ) = ρ ij , and in the ring Q [ [ x ] ] \mathbb{Q}[[x]] Q [[ x ]] of formal power series, with ϑ j = ∑ e ≥ 0 ( 1 / 2 e ) a j e x e \vartheta_j=\sum_{e\geq0}\binom{1/2}{e}a_j^e x^e ϑ j = ∑ e ≥ 0 ( e 1/2 ) a j e x e , the coefficient of x u x^u x u (u ≥ 0 u\geq0 u ≥ 0 ) in p i j ( x ) ϑ j p_{ij}(x)\vartheta_j p ij ( x ) ϑ j is ( k + 1 / 2 u ) Res z = a j z u g i ( z ) \binom{k+1/2}{u}\operatorname{Res}_{z=a_j}z^u g_i(z) ( u k + 1/2 ) Res z = a j z u g i ( z ) .
We next show that, in Q [ [ x ] ] \mathbb{Q}[[x]] Q [[ x ]] , Λ i : = ∑ j p i j ( x ) ϑ j ≡ u k x 3 k ( m o d x 3 k + 1 ) \Lambda_i:=\sum_j p_{ij}(x)\vartheta_j\equiv u_kx^{3k}\pmod{x^{3k+1}} Λ i := ∑ j p ij ( x ) ϑ j ≡ u k x 3 k ( mod x 3 k + 1 ) for each i i i , where u k = ( k + 1 / 2 k ) u_k=\binom{k+1/2}{k} u k = ( k k + 1/2 ) . Summing these coefficients over j j j and applying the residue theorem, the coefficient of x u x^u x u in Λ i \Lambda_i Λ i is ( k + 1 / 2 u ) [ z − 1 ] ∞ z u g i ( z ) \binom{k+1/2}{u}[z^{-1}]_{\infty}z^u g_i(z) ( u k + 1/2 ) [ z − 1 ] ∞ z u g i ( z ) . Since g i ( z ) = z − 3 k − 1 ∏ l ( 1 − a l / z ) − k i = z − 3 k − 1 ( 1 + ∑ e ≥ 1 t e z − e ) g_i(z)=z^{-3k-1}\prod_l(1-a_l/z)^{-k_i}=z^{-3k-1}(1+\sum_{e\geq1}t_ez^{-e}) g i ( z ) = z − 3 k − 1 ∏ l ( 1 − a l / z ) − k i = z − 3 k − 1 ( 1 + ∑ e ≥ 1 t e z − e ) with rational t e t_e t e , the expansion of z u g i z^u g_i z u g i at infinity has only powers z u − 3 k − 1 − e z^{u-3k-1-e} z u − 3 k − 1 − e (e ≥ 0 e\geq0 e ≥ 0 ), with coefficient 1 1 1 for e = 0 e=0 e = 0 ; so [ z − 1 ] ∞ z u g i [z^{-1}]_{\infty}z^u g_i [ z − 1 ] ∞ z u g i is 0 0 0 for u < 3 k u<3k u < 3 k and 1 1 1 for u = 3 k u=3k u = 3 k .
We next reduce the determinant to u k E k u_k\mathcal{E}_k u k E k , where E k \mathcal{E}_k E k is the determinant of the matrix with rows ( ρ i 0 , 1 , ρ i 2 ) (\rho_{i0},1,\rho_{i2}) ( ρ i 0 , 1 , ρ i 2 ) , i = 0 , 1 , 2 i=0,1,2 i = 0 , 1 , 2 . Let p 0 , p 1 , p 2 \mathbf{p}_0,\mathbf{p}_1,\mathbf{p}_2 p 0 , p 1 , p 2 be the columns of ( p i j ( x ) ) (p_{ij}(x)) ( p ij ( x )) and E ( x ) = det ( p i j ( x ) ) ∈ Q [ x ] \mathcal{E}(x)=\det(p_{ij}(x))\in\mathbb{Q}[x] E ( x ) = det ( p ij ( x )) ∈ Q [ x ] , of degree at most 3 k 3k 3 k because each entry has degree at most k k k . In Q [ [ x ] ] \mathbb{Q}[[x]] Q [[ x ]] , ϑ 1 \vartheta_1 ϑ 1 is a unit (constant term 1 1 1 ); adding ϑ 0 / ϑ 1 \vartheta_0/\vartheta_1 ϑ 0 / ϑ 1 times column 0 and ϑ 2 / ϑ 1 \vartheta_2/\vartheta_1 ϑ 2 / ϑ 1 times column 2 to column 1 does not change the determinant and turns column 1 into ( Λ i / ϑ 1 ) i (\Lambda_i/\vartheta_1)_i ( Λ i / ϑ 1 ) i . By the previous step ( Λ i ) i = x 3 k ( u k e + x g ) (\Lambda_i)_i=x^{3k}(u_k\mathbf{e}+x\mathbf{g}) ( Λ i ) i = x 3 k ( u k e + x g ) with e = ( 1 , 1 , 1 ) T \mathbf{e}=(1,1,1)^{\mathsf T} e = ( 1 , 1 , 1 ) T and g ∈ Q [ [ x ] ] 3 \mathbf{g}\in\mathbb{Q}[[x]]^3 g ∈ Q [[ x ] ] 3 , so
E ( x ) = ϑ 1 − 1 x 3 k ( u k det [ p 0 , e , p 2 ] + x det [ p 0 , g , p 2 ] ) ≡ u k E k x 3 k ( m o d x 3 k + 1 ) . \mathcal{E}(x)=\vartheta_1^{-1}x^{3k}\left(u_k\det[\mathbf{p}_0,\mathbf{e},\mathbf{p}_2]+x\det[\mathbf{p}_0,\mathbf{g},\mathbf{p}_2]\right)\equiv u_k\mathcal{E}_kx^{3k}\pmod{x^{3k+1}}. E ( x ) = ϑ 1 − 1 x 3 k ( u k det [ p 0 , e , p 2 ] + x det [ p 0 , g , p 2 ] ) ≡ u k E k x 3 k ( mod x 3 k + 1 ) .
because at x = 0 x=0 x = 0 the columns p 0 , p 2 \mathbf{p}_0,\mathbf{p}_2 p 0 , p 2 are ( ρ i 0 ) i , ( ρ i 2 ) i (\rho_{i0})_i,(\rho_{i2})_i ( ρ i 0 ) i , ( ρ i 2 ) i and ϑ 1 ( 0 ) = 1 \vartheta_1(0)=1 ϑ 1 ( 0 ) = 1 . As deg E ≤ 3 k \deg\mathcal{E}\leq3k deg E ≤ 3 k , E ( x ) = u k E k x 3 k \mathcal{E}(x)=u_k\mathcal{E}_kx^{3k} E ( x ) = u k E k x 3 k . Moreover u k = ( 3 k ) ! ( 3 k ) ! ∏ e = 0 3 k − 1 ( k + 1 2 − e ) ≠ 0 u_k=\frac{(3k)!}{(3k)!}\prod_{e=0}^{3k-1}(k+\frac12-e)\ne0 u k = ( 3 k )! ( 3 k )! ∏ e = 0 3 k − 1 ( k + 2 1 − e ) = 0 , since no factor vanishes.
We finally prove E k ≠ 0 \mathcal{E}_k\ne0 E k = 0 . Each row of ( ρ i j ) (\rho_{ij}) ( ρ ij ) sums to zero: ∑ j ρ i j = [ z − 1 ] ∞ g i = 0 \sum_j\rho_{ij}=[z^{-1}]_{\infty}g_i=0 ∑ j ρ ij = [ z − 1 ] ∞ g i = 0 , as g i g_i g i has order 3 k + 1 ≥ 2 3k+1\geq2 3 k + 1 ≥ 2 at infinity. Suppose E k = 0 \mathcal{E}_k=0 E k = 0 . Then there is ( η 0 , η 1 , η 2 ) ∈ Q 3 ∖ { 0 } (\eta_0,\eta_1,\eta_2)\in\mathbb{Q}^3\setminus\{0\} ( η 0 , η 1 , η 2 ) ∈ Q 3 ∖ { 0 } with ∑ i η i ρ i 0 = ∑ i η i ρ i 2 = ∑ i η i = 0 \sum_i\eta_i\rho_{i0}=\sum_i\eta_i\rho_{i2}=\sum_i\eta_i=0 ∑ i η i ρ i 0 = ∑ i η i ρ i 2 = ∑ i η i = 0 , and by the row sums also ∑ i η i ρ i 1 = 0 \sum_i\eta_i\rho_{i1}=0 ∑ i η i ρ i 1 = 0 . Put
g ( z ) = ∑ i η i g i ( z ) = f ( z ) F ( z ) k + 1 , f ( z ) = ∑ i η i ∏ l ≠ i ( z − a l ) . g(z)=\sum_i\eta_i g_i(z)=\frac{f(z)}{F(z)^{k+1}},\qquad f(z)=\sum_i\eta_i\prod_{l\ne i}(z-a_l). g ( z ) = i ∑ η i g i ( z ) = F ( z ) k + 1 f ( z ) , f ( z ) = i ∑ η i l = i ∏ ( z − a l ) .
Then f ( a i ) = η i ∏ l ≠ i ( a i − a l ) f(a_i)=\eta_i\prod_{l\ne i}(a_i-a_l) f ( a i ) = η i ∏ l = i ( a i − a l ) , so f ≠ 0 f\ne0 f = 0 ; its coefficient of z 2 z^2 z 2 is ∑ i η i = 0 \sum_i\eta_i=0 ∑ i η i = 0 , so deg f ≤ 1 \deg f\leq1 deg f ≤ 1 ; and Res z = a j g = ∑ i η i ρ i j = 0 \operatorname{Res}_{z=a_j}g=\sum_i\eta_i\rho_{ij}=0 Res z = a j g = ∑ i η i ρ ij = 0 for j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 . Since deg f < deg F k + 1 \deg f<\deg F^{k+1} deg f < deg F k + 1 , partial fractions give g = ∑ j ∑ e = 1 k + 1 ξ j e ( z − a j ) − e g=\sum_j\sum_{e=1}^{k+1}\xi_{je}(z-a_j)^{-e} g = ∑ j ∑ e = 1 k + 1 ξ j e ( z − a j ) − e with ξ j 1 = Res z = a j g = 0 \xi_{j1}=\operatorname{Res}_{z=a_j}g=0 ξ j 1 = Res z = a j g = 0 . Hence I = ∑ j ∑ e = 1 k + 1 ξ j e 1 − e ( z − a j ) 1 − e I=\sum_j\sum_{e=1}^{k+1}\frac{\xi_{je}}{1-e}(z-a_j)^{1-e} I = ∑ j ∑ e = 1 k + 1 1 − e ξ j e ( z − a j ) 1 − e satisfies I ′ = g I'=g I ′ = g , and J = F k I J=F^kI J = F k I is a polynomial. At infinity I = ∑ e ≥ 1 ζ e z − e I=\sum_{e\geq1}\zeta_ez^{-e} I = ∑ e ≥ 1 ζ e z − e and I ′ = − ∑ e ≥ 1 e ζ e z − e − 1 I'=-\sum_{e\geq1}e\zeta_ez^{-e-1} I ′ = − ∑ e ≥ 1 e ζ e z − e − 1 , while g = f / F k + 1 g=f/F^{k+1} g = f / F k + 1 has only powers z − e z^{-e} z − e with e ≥ 3 k + 3 − deg f ≥ 3 k + 2 e\geq3k+3-\deg f\geq3k+2 e ≥ 3 k + 3 − deg f ≥ 3 k + 2 ; therefore ζ e = 0 \zeta_e=0 ζ e = 0 for e ≤ 3 k e\leq3k e ≤ 3 k , that is I = ∑ e ≥ 3 k + 1 ζ e z − e I=\sum_{e\geq3k+1}\zeta_ez^{-e} I = ∑ e ≥ 3 k + 1 ζ e z − e . As F k = z 3 k ( 1 + O ( z − 1 ) ) F^k=z^{3k}(1+O(z^{-1})) F k = z 3 k ( 1 + O ( z − 1 )) , the polynomial J = F k I J=F^kI J = F k I has only negative powers of z z z at infinity, so J = 0 J=0 J = 0 . Then I = 0 I=0 I = 0 , g = I ′ = 0 g=I'=0 g = I ′ = 0 and f = 0 f=0 f = 0 , a contradiction. Hence E k ≠ 0 \mathcal{E}_k\ne0 E k = 0 , and C k = u k E k ≠ 0 C_k=u_k\mathcal{E}_k\ne0 C k = u k E k = 0 .
Definition 5.4. For an integer k ≥ 1 k\geq1 k ≥ 1 the Cipu–Fujita prime product P ( k ) \mathcal{P}(k) P ( k ) ([9 ] , proof of Theorem 2.2) is the product of all primes p p p with
p > 2 k + 1 , gcd ( p , 2 k ) = 1 , { k − 1 p } > 3 4 , p>\sqrt{2k+1},\qquad\gcd(p,2k)=1,\qquad\left\{\frac{k-1}{p}\right\}>\frac{3}{4}, p > 2 k + 1 , g cd( p , 2 k ) = 1 , { p k − 1 } > 4 3 ,
where { u } = u − ⌊ u ⌋ \{u\}=u-\lfloor u\rfloor { u } = u − ⌊ u ⌋ ; the empty product is 1.
Lemma 5.5. Let k ≥ 1 k\geq1 k ≥ 1 be an integer, and let T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) be the coefficient products of order k k k (Convention 5.1).
For 0 ≤ h ≤ k 0\leq h\leq k 0 ≤ h ≤ k , 2 2 k − 1 ( k + 1 / 2 h ) ∈ Z 2^{2k-1}\binom{k+1/2}{h}\in\mathbb{Z} 2 2 k − 1 ( h k + 1/2 ) ∈ Z . In particular every T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) is an integer, so G k G_k G k is defined.
Let p p p be an odd prime with p ∤ k p\nmid k p ∤ k and { ( k − 1 ) / p } > 3 4 \{(k-1)/p\}>\frac{3}{4} {( k − 1 ) / p } > 4 3 . Then p p p divides every T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) .
P ( k ) P(k) P ( k ) (Definition 5.4) divides G k G_k G k ; in particular G k ≥ P ( k ) G_k \ge P(k) G k ≥ P ( k ) .
Proof. Item 1 is a lemma of Rickert [26 ] , as quoted by Cipu, Fujita and Mignotte [10 ] ; items 2 and 3 are the term-wise form of the divisibility used in [9 proof of Theorem 2.2] , which cites [5 Lemma 4.1] . We include the proofs. Throughout, v p v_p v p denotes the p p p -adic valuation on Q × \mathbb{Q}^{\times} Q × , and for an integer u u u and q ≥ 1 q \ge1 q ≥ 1 we write u m o d q u \bmod q u mod q for the least non-negative residue.
We first record the valuation of the half-integral binomial coefficients. For 0 ≤ h ≤ k 0 \le h \le k 0 ≤ h ≤ k ,
( k + 1 2 h ) = 1 h ! ∏ e = 0 h − 1 ( k + 1 2 − e ) = ∏ e = 0 h − 1 ( 2 k + 1 − 2 e ) 2 h h ! . \binom{k+\frac{1}{2}}{h}=\frac{1}{h!}\prod_{e=0}^{h-1}\left(k+\frac{1}{2}-e\right)=\frac{\prod_{e=0}^{h-1}(2k+1-2e)}{2^h h!}. ( h k + 2 1 ) = h ! 1 e = 0 ∏ h − 1 ( k + 2 1 − e ) = 2 h h ! ∏ e = 0 h − 1 ( 2 k + 1 − 2 e ) .
a quotient whose numerator is a product of positive odd integers. Let p p p be an odd prime and, for f ≥ 1 f \ge1 f ≥ 1 , let w f ∈ { 0 , … , p f − 1 } w_f \in\{0,\ldots,p^f-1\} w f ∈ { 0 , … , p f − 1 } be defined by 2 w f ≡ 2 k + 1 ( m o d p f ) 2w_f \equiv2k+1 \pmod{p^f} 2 w f ≡ 2 k + 1 ( mod p f ) . Then p f ∣ 2 k + 1 − 2 e p^f \mid2k+1-2e p f ∣ 2 k + 1 − 2 e if and only if e ≡ w f ( m o d p f ) e \equiv w_f \pmod{p^f} e ≡ w f ( mod p f ) , and writing h = g p f + η h=g p^f+\eta h = g p f + η with 0 ≤ η < p f 0 \le\eta< p^f 0 ≤ η < p f , the interval 0 ≤ e ≤ h − 1 0 \le e \le h-1 0 ≤ e ≤ h − 1 contains exactly g + [ η > w f ] g+[\eta>w_f] g + [ η > w f ] such e e e (here [ ⋅ ] [\cdot] [ ⋅ ] is 1 if the condition holds and 0 otherwise). By Legendre’s formula v p ( h ! ) = ∑ f ≥ 1 ⌊ h / p f ⌋ v_p(h!)=\sum_{f\ge1}\lfloor h/p^f\rfloor v p ( h !) = ∑ f ≥ 1 ⌊ h / p f ⌋ , so
v p ( ( k + 1 2 h ) ) = ∑ f ≥ 1 [ ( h m o d p f ) > w f ] ≥ [ ( h m o d p ) > w 1 ] ≥ 0. (6) v_p\left(\binom{k+\frac{1}{2}}{h}\right)=\sum_{f\ge1}\left[(h\bmod p^f)>w_f\right]\ge[(h\bmod p)>w_1]\ge0. \tag*{(6)} v p ( ( h k + 2 1 ) ) = f ≥ 1 ∑ [ ( h mod p f ) > w f ] ≥ [( h mod p ) > w 1 ] ≥ 0. (6)
We next prove item 1. By (6) , v p ( 2 2 k − 1 ( k + 1 / 2 h ) ) ≥ 0 v_p\left(2^{2k-1}\binom{k+1/2}{h}\right)\ge0 v p ( 2 2 k − 1 ( h k + 1/2 ) ) ≥ 0 for every odd prime p p p . For p = 2 p=2 p = 2 , the numerator is odd, so v 2 ( 2 2 k − 1 ( k + 1 / 2 h ) ) = 2 k − 1 − h − v 2 ( h ! ) v_2\left(2^{2k-1}\binom{k+1/2}{h}\right)=2k-1-h-v_2(h!) v 2 ( 2 2 k − 1 ( h k + 1/2 ) ) = 2 k − 1 − h − v 2 ( h !) ; this is 2 k − 1 ≥ 1 2k-1\ge1 2 k − 1 ≥ 1 if h = 0 h=0 h = 0 , and for h ≥ 1 h\ge1 h ≥ 1 it is at least 2 k − 1 − h − ( h − 1 ) = 2 ( k − h ) ≥ 0 2k-1-h-(h-1)=2(k-h)\ge0 2 k − 1 − h − ( h − 1 ) = 2 ( k − h ) ≥ 0 , since v 2 ( h ! ) = h − ( number of binary digits 1 of h ) ≤ h − 1 v_2(h!)=h-(\text{number of binary digits 1 of }h)\le h-1 v 2 ( h !) = h − ( number of binary digits 1 of h ) ≤ h − 1 . Hence 2 2 k − 1 ( k + 1 / 2 h ) ∈ Z 2^{2k-1}\binom{k+1/2}{h}\in\mathbb{Z} 2 2 k − 1 ( h k + 1/2 ) ∈ Z . In T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) one has h j ≤ k i j − 1 ≤ k h_j\le k_{ij}-1\le k h j ≤ k ij − 1 ≤ k , and ( − v h l ) = ( − 1 ) h l ( v + h l − 1 h l ) ∈ Z \binom{-v}{h_l}=(-1)^{h_l}\binom{v+h_l-1}{h_l}\in\mathbb{Z} ( h l − v ) = ( − 1 ) h l ( h l v + h l − 1 ) ∈ Z for every integer v ≥ 1 v\ge1 v ≥ 1 ; so T i j ( k ) ( h ) ∈ Z T_{ij}^{(k)}(h)\in\mathbb{Z} T ij ( k ) ( h ) ∈ Z .
We next record the valuation of the ordinary factors. For integers v ≥ 1 v\ge1 v ≥ 1 and h ≥ 0 h\ge0 h ≥ 0 , ( − v h ) = ( − 1 ) h ( v + h − 1 h ) \binom{-v}{h}=(-1)^h\binom{v+h-1}{h} ( h − v ) = ( − 1 ) h ( h v + h − 1 ) , and by Legendre’s formula
v p ( ( v + h − 1 h ) ) = ∑ f ≥ 1 ( ⌊ v − 1 + h p f ⌋ − ⌊ v − 1 p f ⌋ − ⌊ h p f ⌋ ) . v_p\left(\binom{v+h-1}{h}\right)=\sum_{f\ge1}\left(\left\lfloor\frac{v-1+h}{p^f}\right\rfloor-\left\lfloor\frac{v-1}{p^f}\right\rfloor-\left\lfloor\frac{h}{p^f}\right\rfloor\right). v p ( ( h v + h − 1 ) ) = f ≥ 1 ∑ ( ⌊ p f v − 1 + h ⌋ − ⌊ p f v − 1 ⌋ − ⌊ p f h ⌋ ) .
Each summand is ≥ 0 \ge0 ≥ 0 , and the summand f = 1 f=1 f = 1 equals [ ( v − 1 ) m o d p + h m o d p ≥ p ] [(v-1)\bmod p+h\bmod p\ge p] [( v − 1 ) mod p + h mod p ≥ p ] . Hence
v p ( ( − v h ) ) ≥ [ ( v − 1 ) m o d p + h m o d p ≥ p ] ≥ 0. (7) v_p\left(\binom{-v}{h}\right)\ge[(v-1)\bmod p+h\bmod p\ge p]\ge0. \tag*{(7)} v p ( ( h − v ) ) ≥ [( v − 1 ) mod p + h mod p ≥ p ] ≥ 0. (7)
We now prove item 2. Put ρ = ( k − 1 ) m o d p \rho=(k-1)\bmod p ρ = ( k − 1 ) mod p , so that { ( k − 1 ) / p } = ρ / p \{(k-1)/p\}=\rho/p {( k − 1 ) / p } = ρ / p and the hypothesis reads 4 ρ > 3 p 4\rho>3p 4 ρ > 3 p . Since p ∤ k p\nmid k p ∤ k , ρ ≠ p − 1 \rho\ne p-1 ρ = p − 1 , so ρ ≤ p − 2 \rho\le p-2 ρ ≤ p − 2 and k m o d p = ρ + 1 k\bmod p=\rho+1 k mod p = ρ + 1 . Since 2 k + 1 = 2 ( k − 1 ) + 3 ≡ 2 ρ + 3 ( m o d p ) 2k+1=2(k-1)+3\equiv2\rho+3\pmod p 2 k + 1 = 2 ( k − 1 ) + 3 ≡ 2 ρ + 3 ( mod p ) and 3 p < 2 ρ ≤ 2 p − 4 \frac{3}{p}<2\rho\le2p-4 p 3 < 2 ρ ≤ 2 p − 4 , the integer 2 ρ + 3 − p 2\rho+3-p 2 ρ + 3 − p is even (as p p p is odd) and lies in [ 1 , p − 1 ] [1,p-1] [ 1 , p − 1 ] , so
w 1 = 2 ρ + 3 − p 2 = ρ − p − 3 2 . w_1=\frac{2\rho+3-p}{2}=\rho-\frac{p-3}{2}. w 1 = 2 2 ρ + 3 − p = ρ − 2 p − 3 .
Fix i i i , j j j and h = ( h 0 , h 1 , h 2 ) h=(h_0,h_1,h_2) h = ( h 0 , h 1 , h 2 ) with h 0 + h 1 + h 2 = k i j − 1 h_0+h_1+h_2=k_{ij}-1 h 0 + h 1 + h 2 = k ij − 1 , and write h ˉ l = h l m o d p \bar h_l=h_l\bmod p h ˉ l = h l mod p . Since p p p is odd, v p ( T i j ( k ) ( h ) ) = v p ( ( k + 1 / 2 h j ) ) + ∑ l ≠ j v p ( ( − k i l h l ) ) v_p(T_{ij}^{(k)}(h))=v_p\left(\binom{k+1/2}{h_j}\right)+\sum_{l\ne j}v_p\left(\binom{-k_{il}}{h_l}\right) v p ( T ij ( k ) ( h )) = v p ( ( h j k + 1/2 ) ) + ∑ l = j v p ( ( h l − k i l ) ) , a sum of non-negative terms by (6) and (7) . Suppose that p ∤ T i j ( k ) ( h ) p\nmid T_{ij}^{(k)}(h) p ∤ T ij ( k ) ( h ) , that is v p ( T i j ( k ) ( h ) ) = 0 v_p(T_{ij}^{(k)}(h))=0 v p ( T ij ( k ) ( h )) = 0 . Then (6) gives h ˉ j ≤ w 1 \bar h_j\le w_1 h ˉ j ≤ w 1 . For l ≠ j l\ne j l = j , k i l − 1 ∈ { k − 1 , k } k_{il}-1\in\{k-1,k\} k i l − 1 ∈ { k − 1 , k } , so ( k i l − 1 ) m o d p ∈ { ρ , ρ + 1 } (k_{il}-1)\bmod p\in\{\rho,\rho+1\} ( k i l − 1 ) mod p ∈ { ρ , ρ + 1 } is at least ρ \rho ρ , and (7) gives h ˉ l ≤ p − 1 − ρ \bar h_l\le p-1-\rho h ˉ l ≤ p − 1 − ρ . Therefore
0 ≤ h ˉ 0 + h ˉ 1 + h ˉ 2 ≤ w 1 + 2 ( p − 1 − ρ ) = 3 p − 1 2 − ρ < ρ , 0\le\bar h_0+\bar h_1+\bar h_2\le w_1+2(p-1-\rho)=\frac{3p-1}{2}-\rho<\rho, 0 ≤ h ˉ 0 + h ˉ 1 + h ˉ 2 ≤ w 1 + 2 ( p − 1 − ρ ) = 2 3 p − 1 − ρ < ρ ,
where the last inequality is 3 p − 1 < 4 ρ 3p-1<4\rho 3 p − 1 < 4 ρ . On the other hand h ˉ 0 + h ˉ 1 + h ˉ 2 ≡ h 0 + h 1 + h 2 = k i j − 1 ( m o d p ) \bar{h}_0+\bar{h}_1+\bar{h}_2\equiv h_0+h_1+h_2=k_{ij}-1\pmod{p} h ˉ 0 + h ˉ 1 + h ˉ 2 ≡ h 0 + h 1 + h 2 = k ij − 1 ( mod p ) , and k i j − 1 ∈ { k − 1 , k } k_{ij}-1\in\{k-1,k\} k ij − 1 ∈ { k − 1 , k } is congruent to ρ \rho ρ or to ρ + 1 \rho+1 ρ + 1 , both in [ 0 , p − 1 ] [0,p-1] [ 0 , p − 1 ] . As 0 ≤ h ˉ 0 + h ˉ 1 + h ˉ 2 < ρ ≤ p − 2 0\leq\bar{h}_0+\bar{h}_1+\bar{h}_2<\rho\leq p-2 0 ≤ h ˉ 0 + h ˉ 1 + h ˉ 2 < ρ ≤ p − 2 , the sum h ˉ 0 + h ˉ 1 + h ˉ 2 \bar{h}_0+\bar{h}_1+\bar{h}_2 h ˉ 0 + h ˉ 1 + h ˉ 2 is its own least residue and would equal ρ \rho ρ or ρ + 1 \rho+1 ρ + 1 , which is impossible. Hence p ∣ T i j ( k ) ( h ) p\mid T_{ij}^{(k)}(h) p ∣ T ij ( k ) ( h ) .
We finally prove item 3. A prime p p p counted in P ( k ) \mathcal{P}(k) P ( k ) satisfies p < 4 3 ( k − 1 ) p<\frac{4}{3}(k-1) p < 3 4 ( k − 1 ) : this is clear if p ≤ k − 1 p\leq k-1 p ≤ k − 1 (then k ≥ 2 k\geq2 k ≥ 2 ), and if p > k − 1 p>k-1 p > k − 1 then { ( k − 1 ) / p } = ( k − 1 ) / p > 3 4 \{(k-1)/p\}=(k-1)/p>\frac{3}{4} {( k − 1 ) / p } = ( k − 1 ) / p > 4 3 . So P ( k ) \mathcal{P}(k) P ( k ) is a finite product of distinct primes. Every such prime satisfies gcd ( p , 2 k ) = 1 \gcd(p,2k)=1 g cd( p , 2 k ) = 1 , so it is odd and does not divide k k k , and it satisfies { ( k − 1 ) / p } > 3 4 \{(k-1)/p\}>\frac{3}{4} {( k − 1 ) / p } > 4 3 ; by item 2 it divides every T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) . Hence P ( k ) \mathcal{P}(k) P ( k ) divides every T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) , hence their greatest common divisor G k G_k G k , which is positive; so G k ≥ P ( k ) G_k\geq\mathcal{P}(k) G k ≥ P ( k ) .
Lemma 5.6. For every integer k ≥ 1 k\geq1 k ≥ 1 ,
P ( k ) > 1.6 k 4.09 ⋅ 10 13 , \mathcal{P}(k)>\frac{1.6^k}{4.09\cdot10^{13}}, P ( k ) > 4.09 ⋅ 1 0 13 1. 6 k , where P ( k ) \mathcal{P}(k) P ( k ) is the Cipu–Fujita prime product (Definition 5.4).
Proof. This is proved in [9 ] , proof of Theorem 2.2. With J l ( k ) J_l(k) J l ( k ) the open interval ( ( k − 1 ) / l , 4 ( k − 1 ) / ( 4 l − 1 ) ) ((k-1)/l,4(k-1)/(4l-1)) (( k − 1 ) / l , 4 ( k − 1 ) / ( 4 l − 1 )) , display (2.1) there bounds P ( k ) \mathcal{P}(k) P ( k ) from below by the product of the primes p ∈ J l ( k ) p\in J_l(k) p ∈ J l ( k ) with gcd ( p , 2 k ) = 1 \gcd(p,2k)=1 g cd( p , 2 k ) = 1 , over 1 ≤ l ≤ ⌊ ( k − 1 ) / 2 k − 1 ⌋ 1\leq l\leq\lfloor(k-1)/\sqrt{2k-1}\rfloor 1 ≤ l ≤ ⌊( k − 1 ) / 2 k − 1 ⌋ . The logarithm of this product is shown there to exceed 0.4706 k > k log 1.6 0.4706k>k\log1.6 0.4706 k > k log 1.6 for k ≥ 903683 k\geq903683 k ≥ 903683 (explicit prime-number estimates, applied to the intervals with l ≤ 13 l\leq13 l ≤ 13 ), and to exceed k log 1.6 − 31.342 k\log1.6-31.342 k log 1.6 − 31.342 for 1 ≤ k ≤ 903682 1\leq k\leq903682 1 ≤ k ≤ 903682 (computer search; the minimum is attained at k = 607 k=607 k = 607 ). Since e 31.342 < 4.09 ⋅ 10 13 e^{31.342}<4.09\cdot10^{13} e 31.342 < 4.09 ⋅ 1 0 13 , this gives the claim; display (2.2) there states the resulting bound for an integer that is a multiple of P ( k ) \mathcal{P}(k) P ( k ) , but both derivations bound the right side of (2.1), hence P ( k ) \mathcal{P}(k) P ( k ) .
:::
Lemma 5.7. Let θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 be real numbers and θ 0 = 1 \theta_0=1 θ 0 = 1 . Assume that there are positive real numbers ℓ ∗ , p ∗ , L , P \ell_*,p_*,L,P ℓ ∗ , p ∗ , L , P with L > 1 L>1 L > 1 such that for every positive integer k k k there are integers p i j k p_{ijk} p ij k (0 ≤ i , j ≤ 2 0\leq i,j\leq2 0 ≤ i , j ≤ 2 ) with det ( p i j k ) 0 ≤ i , j ≤ 2 ≠ 0 \det(p_{ijk})_{0\leq i,j\leq2}\ne0 det ( p ij k ) 0 ≤ i , j ≤ 2 = 0 ,
∣ p i j k ∣ ≤ p ∗ P k ( 0 ≤ i , j ≤ 2 ) , ∣ ∑ j = 0 2 p i j k θ j ∣ ≤ ℓ ∗ L − k ( 0 ≤ i ≤ 2 ) . |p_{ijk}|\leq p_*P^k\quad(0\leq i,j\leq2),\qquad\left|\sum_{j=0}^{2}p_{ijk}\theta_j\right|\leq\ell_*L^{-k}\quad(0\leq i\leq2). ∣ p ij k ∣ ≤ p ∗ P k ( 0 ≤ i , j ≤ 2 ) , j = 0 ∑ 2 p ij k θ j ≤ ℓ ∗ L − k ( 0 ≤ i ≤ 2 ) . Then for all integers A 0 , A 1 , A 2 A_0,A_1,A_2 A 0 , A 1 , A 2 with A 0 > 0 A_0>0 A 0 > 0 ,
max { ∣ θ 1 − A 1 A 0 ∣ , ∣ θ 2 − A 2 A 0 ∣ } > C A 0 − τ , τ = 1 + log P log L , C − 1 = 4 p ∗ P ( max { 1 , 2 ℓ ∗ } ) τ − 1 . \max\left\{\left|\theta_1-\frac{A_1}{A_0}\right|,\left|\theta_2-\frac{A_2}{A_0}\right|\right\}>CA_0^{-\tau},\qquad\tau=1+\frac{\log P}{\log L},\qquad C^{-1}=4p_*P\left(\max\{1,2\ell_*\}\right)^{\tau-1}. max { θ 1 − A 0 A 1 , θ 2 − A 0 A 2 } > C A 0 − τ , τ = 1 + log L log P , C − 1 = 4 p ∗ P ( max { 1 , 2 ℓ ∗ } ) τ − 1 . Proof. This is Bennett’s criterion [6 ] , as stated by Cipu, Fujita and Mignotte [10 ] ; the constants p , l p,l p , l , the exponent λ \lambda λ , the numerators p 1 , p 2 p_1,p_2 p 1 , p 2 and the denominator q q q of the sources are written p ∗ , ℓ ∗ , τ , A 1 , A 2 p_*,\ell_*,\tau,A_1,A_2 p ∗ , ℓ ∗ , τ , A 1 , A 2 and A 0 A_0 A 0 here. The statement is invariant under a permutation of the index j j j , since only the set of columns and the linear forms enter; this is used when the columns are indexed by shifts a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 of Convention 5.1 with a 1 = 0 a_1=0 a 1 = 0 , so that the approximated number ϑ 1 = 1 \vartheta_1=1 ϑ 1 = 1 stands in the middle column.
An approximation measure for a pair representation
We now place an Euler triple in this setting. A pair representation factors the two smaller elements so that the cross products are p 1 q 2 = r + σ p_1q_2=r+\sigma p 1 q 2 = r + σ and p 2 q 1 = r − σ p_2q_1=r-\sigma p 2 q 1 = r − σ , and the sums p 3 = p 1 + p 2 p_3=p_1+p_2 p 3 = p 1 + p 2 and q 3 = q 1 + q 2 q_3=q_1+q_2 q 3 = q 1 + q 2 factor the largest, c = p 3 q 3 c=p_3q_3 c = p 3 q 3 ; every Euler triple has one of sign + 1 +1 + 1 , its canonical quadruple (Lemma 5.10 ).
Definition 5.8. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with root r r r (Definition 2.3 ). A pair representation of { a , b , c } \{a,b,c\} { a , b , c } of sign σ ∈ { − 1 , 1 } \sigma\in\{-1,1\} σ ∈ { − 1 , 1 } is a quadruple ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) of positive integers with
a = p 1 q 1 , b = p 2 q 2 , p 1 q 2 − p 2 q 1 = 2 σ . a=p_1q_1,\qquad b=p_2q_2,\qquad p_1q_2-p_2q_1=2\sigma. a = p 1 q 1 , b = p 2 q 2 , p 1 q 2 − p 2 q 1 = 2 σ . The canonical quadruple of { a , b , c } \{a,b,c\} { a , b , c } is the quadruple of rational numbers
p 1 = gcd ( a , r + 1 ) , q 1 = a p 1 , q 2 = r + 1 p 1 , p 2 = r − 1 q 1 . p_1=\gcd(a,r+1),\qquad q_1=\frac{a}{p_1},\qquad q_2=\frac{r+1}{p_1},\qquad p_2=\frac{r-1}{q_1}. p 1 = g cd( a , r + 1 ) , q 1 = p 1 a , q 2 = p 1 r + 1 , p 2 = q 1 r − 1 . The pair coordinates of a pair representation are the vectors
( p 1 , q 1 ) , ( p 2 , q 2 ) , ( p 3 , q 3 ) = ( p 1 + p 2 , q 1 + q 2 ) . (p_1,q_1),\qquad(p_2,q_2),\qquad(p_3,q_3)=(p_1+p_2,q_1+q_2). ( p 1 , q 1 ) , ( p 2 , q 2 ) , ( p 3 , q 3 ) = ( p 1 + p 2 , q 1 + q 2 ) .
Definition 5.9. Let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of sign σ \sigma σ of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with p 3 = p 1 + p 2 p_3=p_1+p_2 p 3 = p 1 + p 2 and q 3 = q 1 + q 2 q_3=q_1+q_2 q 3 = q 1 + q 2 (Definition 5.8). Its hypergeometric parameter and its shifts are
N = 1 2 p 1 p 2 q 3 , a 0 = − σ p 2 , a 1 = 0 , a 2 = σ p 1 , N=\frac{1}{2}p_1p_2q_3,\qquad a_0=-\sigma p_2,\qquad a_1=0,\qquad a_2=\sigma p_1, N = 2 1 p 1 p 2 q 3 , a 0 = − σ p 2 , a 1 = 0 , a 2 = σ p 1 , and its rescaled roots are the positive real numbers
θ 1 = p 3 p 1 a c , θ 2 = p 3 p 2 b c . \theta_1=\frac{p_3}{p_1}\sqrt{\frac{a}{c}},\qquad\theta_2=\frac{p_3}{p_2}\sqrt{\frac{b}{c}}. θ 1 = p 1 p 3 c a , θ 2 = p 2 p 3 c b . These data are chosen so that the rescaled roots are numbers that Rickert’s approximants approximate. By item 3 of the next lemma, θ 1 = 1 + a 0 / N \theta_1=\sqrt{1+a_0/N} θ 1 = 1 + a 0 / N and θ 2 = 1 + a 2 / N \theta_2=\sqrt{1+a_2/N} θ 2 = 1 + a 2 / N , while 1 + a 1 / N = 1 \sqrt{1+a_1/N}=1 1 + a 1 / N = 1 ; so ( θ 1 , 1 , θ 2 ) (\theta_1,1,\theta_2) ( θ 1 , 1 , θ 2 ) are the approximated numbers ϑ 0 , ϑ 1 , ϑ 2 \vartheta_0,\vartheta_1,\vartheta_2 ϑ 0 , ϑ 1 , ϑ 2 of Convention 5.1 at x = 1 / N x=1/N x = 1/ N . The same item gives 2 ( N + a j ) / Δ j = q 1 , − q 3 , q 2 2(N+a_j)/\Delta_j=q_1,-q_3,q_2 2 ( N + a j ) / Δ j = q 1 , − q 3 , q 2 for j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 , which is the hypothesis under which Lemma 5.11 makes the scaled approximants integers.
Lemma 5.10. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with roots r , s , t r,s,t r , s , t (Definition 2.3 ), and use the notation of Definition 5.8 and Definition 5.9.
The canonical quadruple of { a , b , c } \{a,b,c\} { a , b , c } is a pair representation of sign + 1 +1 + 1 . Every pair representation of sign σ \sigma σ satisfies
p 1 q 2 = r + σ , p 2 q 1 = r − σ . p_1q_2=r+\sigma,\qquad p_2q_1=r-\sigma. p 1 q 2 = r + σ , p 2 q 1 = r − σ . For every pair representation of sign σ \sigma σ ,
c = p 3 q 3 , s = p 1 q 3 − σ = p 3 q 1 + σ , t = p 3 q 2 − σ = p 2 q 3 + σ . c=p_3q_3,\qquad s=p_1q_3-\sigma=p_3q_1+\sigma,\qquad t=p_3q_2-\sigma=p_2q_3+\sigma. c = p 3 q 3 , s = p 1 q 3 − σ = p 3 q 1 + σ , t = p 3 q 2 − σ = p 2 q 3 + σ .
For every pair representation of sign σ \sigma σ , the hypergeometric parameter N N N is a positive integer and
N + a 0 = 1 2 p 2 p 3 q 1 , N + a 1 = 1 2 p 1 p 2 q 3 , N + a 2 = 1 2 p 1 p 3 q 2 ; N+a_0=\frac{1}{2}p_2p_3q_1,\qquad N+a_1=\frac{1}{2}p_1p_2q_3,\qquad N+a_2=\frac{1}{2}p_1p_3q_2; N + a 0 = 2 1 p 2 p 3 q 1 , N + a 1 = 2 1 p 1 p 2 q 3 , N + a 2 = 2 1 p 1 p 3 q 2 ;
moreover, with Δ j = ∏ l ≠ j ( a j − a l ) \Delta_j=\prod_{l\ne j}(a_j-a_l) Δ j = ∏ l = j ( a j − a l ) , one has Δ 0 = p 2 p 3 \Delta_0=p_2p_3 Δ 0 = p 2 p 3 , Δ 1 = − p 1 p 2 \Delta_1=-p_1p_2 Δ 1 = − p 1 p 2 , Δ 2 = p 1 p 3 \Delta_2=p_1p_3 Δ 2 = p 1 p 3 , so that
2 ( N + a 0 ) Δ 0 = q 1 , 2 ( N + a 1 ) Δ 1 = − q 3 , 2 ( N + a 2 ) Δ 2 = q 2 . \frac{2(N+a_0)}{\Delta_0}=q_1,\qquad\frac{2(N+a_1)}{\Delta_1}=-q_3,\qquad\frac{2(N+a_2)}{\Delta_2}=q_2. Δ 0 2 ( N + a 0 ) = q 1 , Δ 1 2 ( N + a 1 ) = − q 3 , Δ 2 2 ( N + a 2 ) = q 2 .
Consequently
θ 1 = 1 + a 0 N , θ 2 = 1 + a 2 N , c a = p 3 2 p 1 2 ⋅ p 1 q 3 p 1 q 3 − 2 σ . \theta_1=\sqrt{1+\frac{a_0}{N}},\qquad\theta_2=\sqrt{1+\frac{a_2}{N}},\qquad\frac{c}{a}=\frac{p_3^2}{p_1^2}\cdot\frac{p_1q_3}{p_1q_3-2\sigma}. θ 1 = 1 + N a 0 , θ 2 = 1 + N a 2 , a c = p 1 2 p 3 2 ⋅ p 1 q 3 − 2 σ p 1 q 3 .
For every pair representation of sign σ \sigma σ , p 1 ≤ p 2 p_1\le p_2 p 1 ≤ p 2 and q 1 ≤ q 2 q_1\le q_2 q 1 ≤ q 2 .
Proof. We first prove item 1. By Definition 2.3 and item 4 of Lemma 2.4, a ≥ 1 a\ge1 a ≥ 1 and b ≥ 3 b\ge3 b ≥ 3 , so r 2 = a b + 1 ≥ 4 r^2=ab+1\ge4 r 2 = ab + 1 ≥ 4 and r ≥ 2 r\ge2 r ≥ 2 . Put p 1 = gcd ( a , r + 1 ) p_1=\gcd(a,r+1) p 1 = g cd( a , r + 1 ) , q 1 = a / p 1 q_1=a/p_1 q 1 = a / p 1 , q 2 = ( r + 1 ) / p 1 q_2=(r+1)/p_1 q 2 = ( r + 1 ) / p 1 ; these are positive integers with gcd ( q 1 , q 2 ) = 1 \gcd(q_1,q_2)=1 g cd( q 1 , q 2 ) = 1 . From a b = r 2 − 1 = ( r − 1 ) ( r + 1 ) ab=r^2-1=(r-1)(r+1) ab = r 2 − 1 = ( r − 1 ) ( r + 1 ) we get p 1 q 1 b = ( r − 1 ) p 1 q 2 p_1q_1b=(r-1)p_1q_2 p 1 q 1 b = ( r − 1 ) p 1 q 2 , hence q 1 b = ( r − 1 ) q 2 q_1b=(r-1)q_2 q 1 b = ( r − 1 ) q 2 , and gcd ( q 1 , q 2 ) = 1 \gcd(q_1,q_2)=1 g cd( q 1 , q 2 ) = 1 gives q 1 ∣ r − 1 q_1\mid r-1 q 1 ∣ r − 1 . Thus p 2 = ( r − 1 ) / q 1 p_2=(r-1)/q_1 p 2 = ( r − 1 ) / q 1 is a positive integer, b = ( r − 1 ) q 2 / q 1 = p 2 q 2 b=(r-1)q_2/q_1=p_2q_2 b = ( r − 1 ) q 2 / q 1 = p 2 q 2 , and p 1 q 2 − p 2 q 1 = ( r + 1 ) − ( r − 1 ) = 2 p_1q_2-p_2q_1=(r+1)-(r-1)=2 p 1 q 2 − p 2 q 1 = ( r + 1 ) − ( r − 1 ) = 2 . So the canonical quadruple is a pair representation of sign + 1 +1 + 1 . Finally let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) have sign σ \sigma σ and put u = p 1 q 2 u=p_1q_2 u = p 1 q 2 , v = p 2 q 1 v=p_2q_1 v = p 2 q 1 . Then u , v > 0 u,v>0 u , v > 0 , u v = a b = r 2 − 1 uv=ab=r^2-1 uv = ab = r 2 − 1 and u − v = 2 σ u-v=2\sigma u − v = 2 σ , so ( u + v ) 2 = ( u − v ) 2 + 4 u v = 4 r 2 (u+v)^2=(u-v)^2+4uv=4r^2 ( u + v ) 2 = ( u − v ) 2 + 4 uv = 4 r 2 , u + v = 2 r u+v=2r u + v = 2 r , and u = r + σ u=r+\sigma u = r + σ , v = r − σ v=r-\sigma v = r − σ .
We next prove item 2. By Definition 2.3, c = a + b + 2 r c=a+b+2r c = a + b + 2 r , s = a + r s=a+r s = a + r and t = b + r t=b+r t = b + r . By item 1, p 3 q 3 = ( p 1 + p 2 ) ( q 1 + q 2 ) = a + b + p 1 q 2 + p 2 q 1 = a + b + 2 r = c p_3q_3=(p_1+p_2)(q_1+q_2)=a+b+p_1q_2+p_2q_1=a+b+2r=c p 3 q 3 = ( p 1 + p 2 ) ( q 1 + q 2 ) = a + b + p 1 q 2 + p 2 q 1 = a + b + 2 r = c , p 1 q 3 − σ = a + ( r + σ ) − σ = s p_1q_3-\sigma=a+(r+\sigma)-\sigma=s p 1 q 3 − σ = a + ( r + σ ) − σ = s , p 3 q 1 + σ = a + ( r − σ ) + σ = s p_3q_1+\sigma=a+(r-\sigma)+\sigma=s p 3 q 1 + σ = a + ( r − σ ) + σ = s , p 3 q 2 − σ = ( r + σ ) + b − σ = t p_3q_2-\sigma=(r+\sigma)+b-\sigma=t p 3 q 2 − σ = ( r + σ ) + b − σ = t and p 2 q 3 + σ = ( r − σ ) + b + σ = t p_2q_3+\sigma=(r-\sigma)+b+\sigma=t p 2 q 3 + σ = ( r − σ ) + b + σ = t .
We now prove item 3. We show that N N N is an integer. Since p 1 q 2 − p 2 q 1 = 2 σ p_1q_2-p_2q_1=2\sigma p 1 q 2 − p 2 q 1 = 2 σ is even, p 1 q 2 ≡ p 2 q 1 ( m o d 2 ) p_1q_2\equiv p_2q_1\pmod{2} p 1 q 2 ≡ p 2 q 1 ( mod 2 ) . If p 1 p 2 p_1p_2 p 1 p 2 is even, p 1 p 2 q 3 p_1p_2q_3 p 1 p 2 q 3 is even. If p 1 p_1 p 1 and p 2 p_2 p 2 are both odd, then q 1 ≡ q 2 ( m o d 2 ) q_1\equiv q_2\pmod{2} q 1 ≡ q 2 ( mod 2 ) , so q 3 = q 1 + q 2 q_3=q_1+q_2 q 3 = q 1 + q 2 is even. In both cases N = 1 2 p 1 p 2 q 3 ∈ Z N=\frac{1}{2}p_1p_2q_3\in\mathbb{Z} N = 2 1 p 1 p 2 q 3 ∈ Z , and N > 0 N>0 N > 0 . Using p 1 q 2 − 2 σ = p 2 q 1 p_1q_2-2\sigma=p_2q_1 p 1 q 2 − 2 σ = p 2 q 1 and p 2 q 1 + 2 σ = p 1 q 2 p_2q_1+2\sigma=p_1q_2 p 2 q 1 + 2 σ = p 1 q 2 from item 1,
N − σ p 2 = 1 2 p 2 ( p 1 q 1 + p 1 q 2 − 2 σ ) = 1 2 p 2 ( p 1 q 1 + p 2 q 1 ) = 1 2 p 2 p 3 q 1 , N-\sigma p_2=\frac{1}{2}p_2(p_1q_1+p_1q_2-2\sigma)=\frac{1}{2}p_2(p_1q_1+p_2q_1)=\frac{1}{2}p_2p_3q_1, N − σ p 2 = 2 1 p 2 ( p 1 q 1 + p 1 q 2 − 2 σ ) = 2 1 p 2 ( p 1 q 1 + p 2 q 1 ) = 2 1 p 2 p 3 q 1 ,
N + σ p 1 = 1 2 p 1 ( p 2 q 1 + p 2 q 2 + 2 σ ) = 1 2 p 1 ( p 1 q 2 + p 2 q 2 ) = 1 2 p 1 p 3 q 2 . N+\sigma p_1=\frac{1}{2}p_1(p_2q_1+p_2q_2+2\sigma)=\frac{1}{2}p_1(p_1q_2+p_2q_2)=\frac{1}{2}p_1p_3q_2. N + σ p 1 = 2 1 p 1 ( p 2 q 1 + p 2 q 2 + 2 σ ) = 2 1 p 1 ( p 1 q 2 + p 2 q 2 ) = 2 1 p 1 p 3 q 2 .
The shift differences are a 0 − a 1 = − σ p 2 a_0-a_1=-\sigma p_2 a 0 − a 1 = − σ p 2 , a 0 − a 2 = − σ p 3 a_0-a_2=-\sigma p_3 a 0 − a 2 = − σ p 3 , a 1 − a 2 = − σ p 1 a_1-a_2=-\sigma p_1 a 1 − a 2 = − σ p 1 , so Δ 0 = ( − σ p 2 ) ( − σ p 3 ) = p 2 p 3 \Delta_0=(-\sigma p_2)(-\sigma p_3)=p_2p_3 Δ 0 = ( − σ p 2 ) ( − σ p 3 ) = p 2 p 3 , Δ 1 = ( σ p 2 ) ( − σ p 1 ) = − p 1 p 2 \Delta_1=(\sigma p_2)(-\sigma p_1)=-p_1p_2 Δ 1 = ( σ p 2 ) ( − σ p 1 ) = − p 1 p 2 , Δ 2 = ( σ p 3 ) ( σ p 1 ) = p 1 p 3 \Delta_2=(\sigma p_3)(\sigma p_1)=p_1p_3 Δ 2 = ( σ p 3 ) ( σ p 1 ) = p 1 p 3 , and the three quotients are q 1 q_1 q 1 , − q 3 -q_3 − q 3 , q 2 q_2 q 2 . Using a = p 1 q 1 a=p_1q_1 a = p 1 q 1 , b = p 2 q 2 b=p_2q_2 b = p 2 q 2 and c = p 3 q 3 c=p_3q_3 c = p 3 q 3 (item 2),
θ 1 2 = p 3 2 p 1 2 ⋅ p 1 q 1 p 3 q 3 = p 2 p 3 q 1 p 1 p 2 q 3 = N + a 0 N , θ 2 2 = p 3 2 p 2 2 ⋅ p 2 q 2 p 3 q 3 = p 1 p 3 q 2 p 1 p 2 q 3 = N + a 2 N , \theta_1^2=\frac{p_3^2}{p_1^2}\cdot\frac{p_1q_1}{p_3q_3}=\frac{p_2p_3q_1}{p_1p_2q_3}=\frac{N+a_0}{N},\qquad
\theta_2^2=\frac{p_3^2}{p_2^2}\cdot\frac{p_2q_2}{p_3q_3}=\frac{p_1p_3q_2}{p_1p_2q_3}=\frac{N+a_2}{N}, θ 1 2 = p 1 2 p 3 2 ⋅ p 3 q 3 p 1 q 1 = p 1 p 2 q 3 p 2 p 3 q 1 = N N + a 0 , θ 2 2 = p 2 2 p 3 2 ⋅ p 3 q 3 p 2 q 2 = p 1 p 2 q 3 p 1 p 3 q 2 = N N + a 2 ,
and θ 1 , θ 2 > 0 \theta_1,\theta_2>0 θ 1 , θ 2 > 0 , which gives the two square roots. Finally p 3 q 1 = p 1 q 3 − 2 σ p_3q_1=p_1q_3-2\sigma p 3 q 1 = p 1 q 3 − 2 σ by item 2, so c / a = p 3 q 3 / ( p 1 q 1 ) = p 3 2 q 3 / ( p 1 ( p 1 q 3 − 2 σ ) ) c/a=p_3q_3/(p_1q_1)=p_3^2q_3/(p_1(p_1q_3-2\sigma)) c / a = p 3 q 3 / ( p 1 q 1 ) = p 3 2 q 3 / ( p 1 ( p 1 q 3 − 2 σ )) .
We finally prove item 4. From item 1, p 1 p 2 ⋅ q 2 q 1 = r + σ r − σ \frac{p_1}{p_2}\cdot\frac{q_2}{q_1}=\frac{r+\sigma}{r-\sigma} p 2 p 1 ⋅ q 1 q 2 = r − σ r + σ and p 1 p 2 ⋅ q 1 q 2 = a b \frac{p_1}{p_2}\cdot\frac{q_1}{q_2}=\frac{a}{b} p 2 p 1 ⋅ q 2 q 1 = b a , whence
p 1 2 p 2 2 = a ( r + σ ) b ( r − σ ) , q 1 2 q 2 2 = a ( r − σ ) b ( r + σ ) . \frac{p_1^2}{p_2^2}=\frac{a(r+\sigma)}{b(r-\sigma)},\qquad\frac{q_1^2}{q_2^2}=\frac{a(r-\sigma)}{b(r+\sigma)}. p 2 2 p 1 2 = b ( r − σ ) a ( r + σ ) , q 2 2 q 1 2 = b ( r + σ ) a ( r − σ ) .
By Lemma 2.4, item 4, b − a ≥ 2 b-a \ge2 b − a ≥ 2 . Now a ( r + 1 ) ≤ b ( r − 1 ) a(r+1) \le b(r-1) a ( r + 1 ) ≤ b ( r − 1 ) is equivalent to a + b ≤ r ( b − a ) a+b \le r(b-a) a + b ≤ r ( b − a ) , which holds because both sides are positive and r 2 ( b − a ) 2 − ( a + b ) 2 = ( a b + 1 ) ( b − a ) 2 − ( a + b ) 2 = a b ( ( b − a ) 2 − 4 ) ≥ 0 r^2(b-a)^2-(a+b)^2=(ab+1)(b-a)^2-(a+b)^2=ab((b-a)^2-4)\ge0 r 2 ( b − a ) 2 − ( a + b ) 2 = ( ab + 1 ) ( b − a ) 2 − ( a + b ) 2 = ab (( b − a ) 2 − 4 ) ≥ 0 . Also a ( r − 1 ) < b ( r + 1 ) a(r-1)<b(r+1) a ( r − 1 ) < b ( r + 1 ) , as a < b a<b a < b . For σ = 1 \sigma=1 σ = 1 this gives p 1 ≤ p 2 p_1\le p_2 p 1 ≤ p 2 and q 1 < q 2 q_1<q_2 q 1 < q 2 ; for σ = − 1 \sigma=-1 σ = − 1 it gives p 1 < p 2 p_1<p_2 p 1 < p 2 and q 1 ≤ q 2 q_1\le q_2 q 1 ≤ q 2 .
Lemma 5.11. Let a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 be distinct integers, let k ≥ 1 k\ge1 k ≥ 1 , and let p i j = p i j ( k ) p_{ij}=p_{ij}^{(k)} p ij = p ij ( k ) and G k G_k G k be as in Convention 5.1. Let N N N be a nonzero rational number with N + a j ≠ 0 N+a_j\ne0 N + a j = 0 (j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 ), put Δ j = ∏ l ≠ j ( a j − a l ) \Delta_j=\prod_{l\ne j}(a_j-a_l) Δ j = ∏ l = j ( a j − a l ) and Δ = ∣ ( a 0 − a 1 ) ( a 0 − a 2 ) ( a 1 − a 2 ) ∣ \Delta=\lvert(a_0-a_1)(a_0-a_2)(a_1-a_2)\rvert Δ = ∣( a 0 − a 1 ) ( a 0 − a 2 ) ( a 1 − a 2 )∣ , and assume that
2 ( N + a j ) Δ j ∈ Z ( j = 0 , 1 , 2 ) . \frac{2(N+a_j)}{\Delta_j}\in\mathbb{Z}\qquad(j=0,1,2). Δ j 2 ( N + a j ) ∈ Z ( j = 0 , 1 , 2 ) . Then for all i , j ∈ { 0 , 1 , 2 } i,j\in\{0,1,2\} i , j ∈ { 0 , 1 , 2 } ,
p ^ i j : = ( 8 Δ N ) k 2 G k p i j ( 1 N ) ∈ Z . \widehat{p}_{ij}:=\frac{(8\Delta N)^k}{2G_k}p_{ij}\left(\frac{1}{N}\right)\in\mathbb{Z}. p ij := 2 G k ( 8Δ N ) k p ij ( N 1 ) ∈ Z . In particular, if N N N and a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 are the hypergeometric parameter and the shifts of a pair representation of an Euler triple (Definition 5.9), the hypothesis holds and Δ = p 1 p 2 p 3 \Delta=p_1p_2p_3 Δ = p 1 p 2 p 3 , so that
( 8 p 1 p 2 p 3 N ) k 2 G k p i j ( 1 N ) ∈ Z ( k ≥ 1 , 0 ≤ i , j ≤ 2 ) . \frac{(8p_1p_2p_3N)^k}{2G_k}p_{ij}\left(\frac{1}{N}\right)\in\mathbb{Z}\qquad(k\ge1,\ 0\le i,j\le2). 2 G k ( 8 p 1 p 2 p 3 N ) k p ij ( N 1 ) ∈ Z ( k ≥ 1 , 0 ≤ i , j ≤ 2 ) .
Proof. By Lemma 5.5, item 1, the coefficient products T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) are integers, so G k G_k G k is defined. Write Y j = 2 ( N + a j ) / Δ j ∈ Z Y_j=2(N+a_j)/\Delta_j\in\mathbb{Z} Y j = 2 ( N + a j ) / Δ j ∈ Z .
We first establish the closed form of a single term. Fix i i i , j j j and h = ( h 0 , h 1 , h 2 ) h=(h_0,h_1,h_2) h = ( h 0 , h 1 , h 2 ) with h 0 + h 1 + h 2 = k i j − 1 h_0+h_1+h_2=k_{ij}-1 h 0 + h 1 + h 2 = k ij − 1 , and write { l 1 , l 2 } = { 0 , 1 , 2 } ∖ { j } \{l_1,l_2\}=\{0,1,2\}\setminus\{j\} { l 1 , l 2 } = { 0 , 1 , 2 } ∖ { j } . Since ( 1 + a j / N ) k − h j N − h j = ( N + a j ) k − h j N − k (1+a_j/N)^{k-h_j}N^{-h_j}=(N+a_j)^{k-h_j}N^{-k} ( 1 + a j / N ) k − h j N − h j = ( N + a j ) k − h j N − k and Δ k = ± ( a l 1 − a l 2 ) k ∏ l ≠ j ( a j − a l ) k \Delta^k=\pm(a_{l_1}-a_{l_2})^k\prod_{l\ne j}(a_j-a_l)^k Δ k = ± ( a l 1 − a l 2 ) k ∏ l = j ( a j − a l ) k ,
2 k Δ k N k ⋅ ( 1 + a j / N ) k − h j N − h j ∏ l ≠ j ( a j − a l ) k i l + h l = ± 2 k ( N + a j ) k − h j ( a l 1 − a l 2 ) k ∏ l ≠ j ( a j − a l ) k − k i l − h l . 2^k\Delta^kN^k\cdot\frac{(1+a_j/N)^{k-h_j}N^{-h_j}}{\prod_{l\ne j}(a_j-a_l)^{k i_l+h_l}}
=\pm2^k(N+a_j)^{k-h_j}(a_{l_1}-a_{l_2})^k\prod_{l\ne j}(a_j-a_l)^{k-k i_l-h_l}. 2 k Δ k N k ⋅ ∏ l = j ( a j − a l ) k i l + h l ( 1 + a j / N ) k − h j N − h j = ± 2 k ( N + a j ) k − h j ( a l 1 − a l 2 ) k l = j ∏ ( a j − a l ) k − k i l − h l .
By hypothesis 2 ( N + a j ) = Y j Δ j = Y j ∏ l ≠ j ( a j − a l ) 2(N+a_j)=Y_j\Delta_j=Y_j\prod_{l\ne j}(a_j-a_l) 2 ( N + a j ) = Y j Δ j = Y j ∏ l = j ( a j − a l ) , so 2 k ( N + a j ) k − h j = 2 h j Y j k − h j ∏ l ≠ j ( a j − a l ) k − h j 2^k(N+a_j)^{k-h_j}=2^{h_j}Y_j^{k-h_j}\prod_{l\ne j}(a_j-a_l)^{k-h_j} 2 k ( N + a j ) k − h j = 2 h j Y j k − h j ∏ l = j ( a j − a l ) k − h j , and collecting the exponents of a j − a l a_j-a_l a j − a l gives
2 k Δ k N k ⋅ ( 1 + a j / N ) k − h j N − h j ∏ l ≠ j ( a j − a l ) k i l + h l = ± 2 h j Y j k − h j ( a l 1 − a l 2 ) k ∏ l ≠ j ( a j − a l ) k − h j − h l − δ i l . 2^k\Delta^kN^k\cdot\frac{(1+a_j/N)^{k-h_j}N^{-h_j}}{\prod_{l\ne j}(a_j-a_l)^{k i_l+h_l}}
=\pm2^{h_j}Y_j^{k-h_j}(a_{l_1}-a_{l_2})^k\prod_{l\ne j}(a_j-a_l)^{k-h_j-h_l-\delta_{il}}. 2 k Δ k N k ⋅ ∏ l = j ( a j − a l ) k i l + h l ( 1 + a j / N ) k − h j N − h j = ± 2 h j Y j k − h j ( a l 1 − a l 2 ) k l = j ∏ ( a j − a l ) k − h j − h l − δ i l .
We next show that the right side is an integer. Since h j + h l = k i j − 1 − h l ′ h_j+h_l=k_{ij}-1-h_{l'} h j + h l = k ij − 1 − h l ′ for the third index l ′ l' l ′ , we have k − h j − h l − δ i l = k − k i j + 1 + h l ′ − δ i l = 1 − δ i j − δ i l + h l ′ ≥ 0 k-h_j-h_l-\delta_{il}=k-k_{ij}+1+h_{l'}-\delta_{il}=1-\delta_{ij}-\delta_{il}+h_{l'}\ge0 k − h j − h l − δ i l = k − k ij + 1 + h l ′ − δ i l = 1 − δ ij − δ i l + h l ′ ≥ 0 , because j ≠ l j\ne l j = l so at most one of δ i j , δ i l \delta_{ij},\delta_{il} δ ij , δ i l equals 1. Also k − h j ≥ k − ( k i j − 1 ) ≥ 0 k-h_j\ge k-(k_{ij}-1)\ge0 k − h j ≥ k − ( k ij − 1 ) ≥ 0 . Hence every factor on the right is an integer.
We now prove the integrality of p ^ i j \widehat{p}_{ij} p ij . Multiplying the formula of Convention 5.1 for p i j ( 1 / N ) p_{ij}(1/N) p ij ( 1/ N ) by 2 3 k − 1 Δ k N k = 2 2 k − 1 ⋅ 2 k Δ k N k 2^{3k-1}\Delta^kN^k=2^{2k-1}\cdot2^k\Delta^kN^k 2 3 k − 1 Δ k N k = 2 2 k − 1 ⋅ 2 k Δ k N k gives
2 3 k − 1 Δ k N k p i j ( 1 / N ) = ∑ h T i j ( k ) ( h ) ⋅ [ 2 k Δ k N k ( 1 + a j / N ) k − h j N − h j ∏ l ≠ j ( a j − a l ) k i l + h l ] . 2^{3k-1}\Delta^kN^kp_{ij}(1/N)
=\sum_h T_{ij}^{(k)}(h)\cdot\left[2^k\Delta^kN^k\frac{(1+a_j/N)^{k-h_j}N^{-h_j}}{\prod_{l\ne j}(a_j-a_l)^{k i_l+h_l}}\right]. 2 3 k − 1 Δ k N k p ij ( 1/ N ) = h ∑ T ij ( k ) ( h ) ⋅ [ 2 k Δ k N k ∏ l = j ( a j − a l ) k i l + h l ( 1 + a j / N ) k − h j N − h j ] .
In each summand T i j ( k ) ( h ) T_{ij}^{(k)}(h) T ij ( k ) ( h ) is an integer divisible by G k G_k G k , and the bracket is an integer by the first two steps. Hence the sum is an integer divisible by G k G_k G k , and p ^ i j = 2 3 k − 1 Δ k N k p i j ( 1 / N ) / G k ∈ Z \hat{p}_{ij}=2^{3k-1}\Delta^kN^kp_{ij}(1/N)/G_k\in\mathbb{Z} p ^ ij = 2 3 k − 1 Δ k N k p ij ( 1/ N ) / G k ∈ Z , as ( 8 Δ N ) k / 2 = 2 3 k − 1 Δ k N k (8\Delta N)^{k/2}=2^{3k-1}\Delta^kN^k ( 8Δ N ) k /2 = 2 3 k − 1 Δ k N k .
Finally we verify the hypothesis for a pair representation. By Lemma 5.10, item 3, N N N is a positive integer, Δ 0 = p 2 p 3 \Delta_0=p_2p_3 Δ 0 = p 2 p 3 , Δ 1 = − p 1 p 2 \Delta_1=-p_1p_2 Δ 1 = − p 1 p 2 , Δ 2 = p 1 p 3 \Delta_2=p_1p_3 Δ 2 = p 1 p 3 , and 2 ( N + a j ) / Δ j 2(N+a_j)/\Delta_j 2 ( N + a j ) / Δ j equals q 1 q_1 q 1 , − q 3 -q_3 − q 3 , q 2 q_2 q 2 for j = 0 , 1 , 2 j=0,1,2 j = 0 , 1 , 2 , which are integers; moreover N + a 0 = 1 2 p 2 p 3 q 1 N+a_0=\frac{1}{2}p_2p_3q_1 N + a 0 = 2 1 p 2 p 3 q 1 , N + a 1 = N N+a_1=N N + a 1 = N and N + a 2 = 1 2 p 1 p 3 q 2 N+a_2=\frac{1}{2}p_1p_3q_2 N + a 2 = 2 1 p 1 p 3 q 2 are positive, and the shifts − σ p 2 -\sigma p_2 − σ p 2 , 0 0 0 , σ p 1 \sigma p_1 σ p 1 are distinct as p 1 , p 2 > 0 p_1,p_2>0 p 1 , p 2 > 0 . So the hypothesis holds, and Δ = ∣ Δ 0 Δ 1 Δ 2 ∣ 1 / 2 = p 1 p 2 p 3 \Delta=|\Delta_0\Delta_1\Delta_2|^{1/2}=p_1p_2p_3 Δ = ∣ Δ 0 Δ 1 Δ 2 ∣ 1/2 = p 1 p 2 p 3 .
Lemma 5.12. Let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of an Euler triple, with hypergeometric parameter N N N , shifts a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 and rescaled roots ϑ 1 , ϑ 2 \vartheta_1,\vartheta_2 ϑ 1 , ϑ 2 (Definition 5.9), and assume N ≥ 3 p 2 N\geq3p_2 N ≥ 3 p 2 . Put μ = p 2 / N ≤ 1 3 \mu=p_2/N\leq\frac{1}{3} μ = p 2 / N ≤ 3 1 , and let p i j = p i j ( k ) p_{ij}=p_{ij}^{(k)} p ij = p ij ( k ) , ϑ j \vartheta_j ϑ j , Λ i \Lambda_i Λ i be Rickert’s approximants (Convention 5.1) for these shifts at x = 1 / N x=1/N x = 1/ N . Then ( ϑ 0 , ϑ 1 , ϑ 2 ) = ( θ 1 , 1 , θ 2 ) (\vartheta_0,\vartheta_1,\vartheta_2)=(\theta_1,1,\theta_2) ( ϑ 0 , ϑ 1 , ϑ 2 ) = ( θ 1 , 1 , θ 2 ) , and for every integer k ≥ 1 k\geq1 k ≥ 1 and all 0 ≤ i , j ≤ 2 0\leq i,j\leq2 0 ≤ i , j ≤ 2 :
∣ Λ i ( 1 N ) ∣ ≤ 27 64 ( 1 − μ ) − 1 { 27 4 ( 1 − μ ) 2 N 3 } − k ; \left|\Lambda_i\left(\frac{1}{N}\right)\right|\leq\frac{27}{64}(1-\mu)^{-1}\left\{\frac{27}{4}(1-\mu)^2N^3\right\}^{-k}; Λ i ( N 1 ) ≤ 64 27 ( 1 − μ ) − 1 { 4 27 ( 1 − μ ) 2 N 3 } − k ;
∣ p i j ( 1 N ) ∣ ≤ ( 1 + 3 2 μ 1 − μ ) 1 / 2 ( 4 p 1 2 p 2 ) k . \left|p_{ij}\left(\frac{1}{N}\right)\right|\leq\left(\frac{1+\frac{3}{2}\mu}{1-\mu}\right)^{1/2}\left(\frac{4}{p_1^2p_2}\right)^k. p ij ( N 1 ) ≤ ( 1 − μ 1 + 2 3 μ ) 1/2 ( p 1 2 p 2 4 ) k .
Proof. The two bounds are those of Cipu, Fujita and Mignotte [10 ] , which follow the arguments of Bennett [6 ] , except that the prefactor in item 2 is taken here for both signs of the representation and for the shift a 0 a_0 a 0 , whereas the source gives its value for a 2 > 0 a_2>0 a 2 > 0 only; we give the proof. Throughout, x = 1 / N x=1/N x = 1/ N , σ \sigma σ is the sign of the representation and M = max l ∣ a l ∣ = p 2 M=\max_l|a_l|=p_2 M = max l ∣ a l ∣ = p 2 , since p 1 ≤ p 2 p_1\leq p_2 p 1 ≤ p 2 (Lemma 5.10, item 4); so x M = μ < 1 xM=\mu<1 x M = μ < 1 . By Lemma 5.10, item 3, ϑ 0 = ( 1 + a 0 / N ) 1 / 2 = θ 1 \vartheta_0=(1+a_0/N)^{1/2}=\theta_1 ϑ 0 = ( 1 + a 0 / N ) 1/2 = θ 1 and ϑ 2 = ( 1 + a 2 / N ) 1 / 2 = θ 2 \vartheta_2=(1+a_2/N)^{1/2}=\theta_2 ϑ 2 = ( 1 + a 2 / N ) 1/2 = θ 2 , and ϑ 1 = 1 \vartheta_1=1 ϑ 1 = 1 as a 1 = 0 a_1=0 a 1 = 0 . Put
ψ i ( z ) = ( 1 + z x ) k + 1 / 2 ( z − a i ) − 1 F ( z ) − k , F ( z ) = ∏ l ( z − a l ) , \psi_i(z)=(1+zx)^{k+1/2}(z-a_i)^{-1}F(z)^{-k},\qquad F(z)=\prod_l(z-a_l), ψ i ( z ) = ( 1 + z x ) k + 1/2 ( z − a i ) − 1 F ( z ) − k , F ( z ) = l ∏ ( z − a l ) ,
which is analytic in ∣ z ∣ < N |z|<N ∣ z ∣ < N except for poles at a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 .
We first note that p i j ( x ) ϑ j p_{ij}(x)\vartheta_j p ij ( x ) ϑ j is the residue of ψ i \psi_i ψ i at a j a_j a j , by Lemma 5.2, item 2 (there ψ i = ( 1 + z x ) k + 1 / 2 g i \psi_i=(1+zx)^{k+1/2}g_i ψ i = ( 1 + z x ) k + 1/2 g i with g i = ( ( z − a i ) F k ) − 1 g_i=((z-a_i)F^k)^{-1} g i = (( z − a i ) F k ) − 1 , and ∣ x ∣ M = μ < 1 |x|M=\mu<1 ∣ x ∣ M = μ < 1 ), so that, by the residue theorem,
Λ i ( x ) = 1 2 π − 1 ∮ ψ i ( z ) d z \Lambda_i(x)=\frac{1}{2\pi\sqrt{-1}}\oint\psi_i(z)\,dz Λ i ( x ) = 2 π − 1 1 ∮ ψ i ( z ) d z
over the counter-clockwise circle ∣ z ∣ = T |z|=T ∣ z ∣ = T , for any M < T < N M<T<N M < T < N .
We next prove item 1. The integrand ψ i \psi_i ψ i extends analytically to C ∖ ( ( − ∞ , − N ] ∪ { a 0 , a 1 , a 2 } ) \mathbb{C}\setminus((-\infty,-N]\cup\{a_0,a_1,a_2\}) C ∖ (( − ∞ , − N ] ∪ { a 0 , a 1 , a 2 }) , with ∣ ψ i ( z ) ∣ = O ( ∣ z ∣ − 2 k − 1 / 2 ) |\psi_i(z)|=O(|z|^{-2k-1/2}) ∣ ψ i ( z ) ∣ = O ( ∣ z ∣ − 2 k − 1/2 ) at infinity, so the circle of radius T → ∞ T\to\infty T → ∞ contributes nothing and the contour can be collapsed onto the cut: ∮ ψ i d z = ∫ − N − ∞ ( ψ i + ( z ) − ψ i − ( z ) ) d z \oint\psi_i\,dz=\int_{-N}^{-\infty}(\psi_i^+(z)-\psi_i^-(z))\,dz ∮ ψ i d z = ∫ − N − ∞ ( ψ i + ( z ) − ψ i − ( z )) d z , where ψ i ± \psi_i^\pm ψ i ± are the boundary values from the upper and lower half-planes (the small circle around the branch point − N -N − N contributes nothing, as ψ i \psi_i ψ i is bounded there). For z = − ξ z=-\xi z = − ξ , ξ > N \xi>N ξ > N , one has ( 1 + z x ) 1 / 2 = ± − 1 ( ξ / N − 1 ) 1 / 2 (1+zx)^{1/2}=\pm\sqrt{-1}(\xi/N-1)^{1/2} ( 1 + z x ) 1/2 = ± − 1 ( ξ / N − 1 ) 1/2 from above and below, so ψ i + ( − ξ ) − ψ i − ( − ξ ) = 2 − 1 ( 1 − ξ / N ) k ( ξ / N − 1 ) 1 / 2 / ( ( − ξ − a i ) F ( − ξ ) k ) \psi_i^+(-\xi)-\psi_i^-(-\xi)=2\sqrt{-1}(1-\xi/N)^{k}(\xi/N-1)^{1/2}/((-\xi-a_i)F(-\xi)^k) ψ i + ( − ξ ) − ψ i − ( − ξ ) = 2 − 1 ( 1 − ξ / N ) k ( ξ / N − 1 ) 1/2 / (( − ξ − a i ) F ( − ξ ) k ) . Since ∣ ξ + a l ∣ ≥ ξ − M |\xi+a_l|\geq\xi-M ∣ ξ + a l ∣ ≥ ξ − M for all l l l and ξ ≥ N > M \xi\geq N>M ξ ≥ N > M ,
∣ Λ i ( x ) ∣ ≤ 1 π ∫ N ∞ ( ξ / N − 1 ) k + 1 / 2 ( ξ − M ) 3 k + 1 d ξ = N − 3 k π ∫ 0 ∞ v k + 1 / 2 ( v + 1 − μ ) 3 k + 1 d v = N − 3 k ( 1 − μ ) − 2 k + 1 / 2 π B ( k + 3 2 , 2 k − 1 2 ) , \begin{aligned}
|\Lambda_i(x)|\leq\frac{1}{\pi}\int_N^\infty\frac{(\xi/N-1)^{k+1/2}}{(\xi-M)^{3k+1}}\,d\xi
=\frac{N^{-3k}}{\pi}\int_0^\infty\frac{v^{k+1/2}}{(v+1-\mu)^{3k+1}}\,dv \\
&=\frac{N^{-3k}(1-\mu)^{-2k+1/2}}{\pi}B\left(k+\frac{3}{2},2k-\frac{1}{2}\right),
\end{aligned} ∣ Λ i ( x ) ∣ ≤ π 1 ∫ N ∞ ( ξ − M ) 3 k + 1 ( ξ / N − 1 ) k + 1/2 d ξ = π N − 3 k ∫ 0 ∞ ( v + 1 − μ ) 3 k + 1 v k + 1/2 d v = π N − 3 k ( 1 − μ ) − 2 k + 1/2 B ( k + 2 3 , 2 k − 2 1 ) ,
by the substitutions ξ = N ( 1 + v ) \xi= N(1+v) ξ = N ( 1 + v ) , v = ( 1 − μ ) w v=(1-\mu)w v = ( 1 − μ ) w and ∫ 0 ∞ w u 1 − 1 ( 1 + w ) − u 1 − u 2 d w = B ( u 1 , u 2 ) \int_{0}^{\infty}w^{u_{1}-1}(1+w)^{-u_{1}-u_{2}}\,\mathrm{d}w=B(u_{1},u_{2}) ∫ 0 ∞ w u 1 − 1 ( 1 + w ) − u 1 − u 2 d w = B ( u 1 , u 2 ) . We claim B ( k + 3 2 , 2 k − 1 2 ) ≤ 27 64 ( 4 27 ) k B(k+\frac{3}{2},2k-\frac{1}{2})\leq\frac{27}{64}\left(\frac{4}{27}\right)^{k} B ( k + 2 3 , 2 k − 2 1 ) ≤ 64 27 ( 27 4 ) k for k ≥ 1 k\geq1 k ≥ 1 . At k = 1 k=1 k = 1 , B ( 5 2 , 3 2 ) = Γ ( 5 2 ) Γ ( 3 2 ) / Γ ( 4 ) = π 16 B(\frac{5}{2},\frac{3}{2})=\Gamma(\frac{5}{2})\Gamma(\frac{3}{2})/\Gamma(4)=\frac{\pi}{16} B ( 2 5 , 2 3 ) = Γ ( 2 5 ) Γ ( 2 3 ) /Γ ( 4 ) = 16 π , with equality. The ratio of consecutive values is
B ( k + 5 2 , 2 k + 3 2 ) B ( k + 3 2 , 2 k − 1 2 ) = ( k + 3 2 ) ( 2 k + 1 2 ) ( 2 k − 1 2 ) ( 3 k + 3 ) ( 3 k + 2 ) ( 3 k + 1 ) < 4 27 . \frac{B(k+\frac{5}{2},2k+\frac{3}{2})}{B(k+\frac{3}{2},2k-\frac{1}{2})}
=\frac{(k+\frac{3}{2})(2k+\frac{1}{2})(2k-\frac{1}{2})}{(3k+3)(3k+2)(3k+1)}<\frac{4}{27}. B ( k + 2 3 , 2 k − 2 1 ) B ( k + 2 5 , 2 k + 2 3 ) = ( 3 k + 3 ) ( 3 k + 2 ) ( 3 k + 1 ) ( k + 2 3 ) ( 2 k + 2 1 ) ( 2 k − 2 1 ) < 27 4 .
because 4 ( 3 k + 3 ) ( 3 k + 2 ) ( 3 k + 1 ) − 27 ( k + 3 2 ) ( 2 k + 1 2 ) ( 2 k − 1 2 ) = 54 k 2 + 555 4 k + 273 8 > 0 4(3k+3)(3k+2)(3k+1)-27(k+\frac{3}{2})(2k+\frac{1}{2})(2k-\frac{1}{2})=54k^{2}+\frac{555}{4}k+\frac{273}{8}>0 4 ( 3 k + 3 ) ( 3 k + 2 ) ( 3 k + 1 ) − 27 ( k + 2 3 ) ( 2 k + 2 1 ) ( 2 k − 2 1 ) = 54 k 2 + 4 555 k + 8 273 > 0 . Hence, as ( 1 − μ ) 3 / 2 ≤ 1 (1-\mu)^{3/2}\leq1 ( 1 − μ ) 3/2 ≤ 1 ,
∣ Λ i ( x ) ∣ ≤ 27 64 ( 1 − μ ) − 2 k − 1 ( 4 27 ) k N − 3 k = 27 64 ( 1 − μ ) − 1 { 27 4 ( 1 − μ ) 2 N 3 } − k . |\Lambda_i(x)|\leq\frac{27}{64}(1-\mu)^{-2k-1}\left(\frac{4}{27}\right)^kN^{-3k}=\frac{27}{64}(1-\mu)^{-1}\left\{\frac{27}{4}(1-\mu)^2N^3\right\}^{-k}. ∣ Λ i ( x ) ∣ ≤ 64 27 ( 1 − μ ) − 2 k − 1 ( 27 4 ) k N − 3 k = 64 27 ( 1 − μ ) − 1 { 4 27 ( 1 − μ ) 2 N 3 } − k .
We now prove item 2. Fix j j j and let Γ j \Gamma_j Γ j be the circle ∣ z − a j ∣ = ρ |z-a_j|=\rho ∣ z − a j ∣ = ρ , ρ = p 1 / 2 \rho=p_1/2 ρ = p 1 /2 . It separates a j a_j a j from the other shifts (their distances to a j a_j a j are among p 1 , p 2 , p 3 p_1,p_2,p_3 p 1 , p 2 , p 3 , all ≥ 2 ρ \geq2\rho ≥ 2 ρ ), and lies in ∣ z ∣ ≤ p 2 + p 1 / 2 < N |z|\leq p_2+p_1/2<N ∣ z ∣ ≤ p 2 + p 1 /2 < N . By the residue representation, ∣ p i j ( x ) ∣ ϑ j ≤ ρ ⋅ max Γ j ∣ 1 + z x ∣ k + 1 / 2 / ( min Γ j ∣ z − a i ∣ ⋅ min Γ j ∣ F ( z ) ∣ ) |p_{ij}(x)|\vartheta_j\leq\rho\cdot\max_{\Gamma_j}|1+zx|^{k+1/2}/(\min_{\Gamma_j}|z-a_i|\cdot\min_{\Gamma_j}|F(z)|) ∣ p ij ( x ) ∣ ϑ j ≤ ρ ⋅ max Γ j ∣1 + z x ∣ k + 1/2 / ( min Γ j ∣ z − a i ∣ ⋅ min Γ j ∣ F ( z ) ∣ ) , and min Γ j ∣ z − a i ∣ ≥ ρ \min_{\Gamma_j}|z-a_i|\geq\rho min Γ j ∣ z − a i ∣ ≥ ρ (equality if i = j i=j i = j ; otherwise ∣ z − a i ∣ ≥ ∣ a j − a i ∣ − ρ ≥ p 1 − ρ = ρ |z-a_i|\geq|a_j-a_i|-\rho\geq p_1-\rho=\rho ∣ z − a i ∣ ≥ ∣ a j − a i ∣ − ρ ≥ p 1 − ρ = ρ ). Also max Γ j ∣ 1 + z x ∣ ≤ 1 + ( ∣ a j ∣ + ρ ) / N = : Υ j \max_{\Gamma_j}|1+zx|\leq1+(|a_j|+\rho)/N=:\Upsilon_j max Γ j ∣1 + z x ∣ ≤ 1 + ( ∣ a j ∣ + ρ ) / N =: Υ j . For Z j : = min Γ j ∣ F ( z ) ∣ Z_j:=\min_{\Gamma_j}|F(z)| Z j := min Γ j ∣ F ( z ) ∣ , write z = a j + ρ e − 1 ϕ z=a_j+\rho e^{\sqrt{-1}\phi} z = a j + ρ e − 1 ϕ ; the two factors ∣ z − a l ∣ |z-a_l| ∣ z − a l ∣ , l ≠ j l\neq j l = j , are ( X l 2 + ρ 2 + 2 ϵ l X l ρ cos ϕ ) 1 / 2 (X_l^2+\rho^2+2\epsilon_lX_l\rho\cos\phi)^{1/2} ( X l 2 + ρ 2 + 2 ϵ l X l ρ cos ϕ ) 1/2 with X l = ∣ a j − a l ∣ X_l=|a_j-a_l| X l = ∣ a j − a l ∣ and ϵ l = sgn ( a j − a l ) \epsilon_l=\operatorname{sgn}(a_j-a_l) ϵ l = sgn ( a j − a l ) . For j = 0 j=0 j = 0 and j = 2 j=2 j = 2 the two signs ϵ l \epsilon_l ϵ l coincide (a 0 a_0 a 0 and a 2 a_2 a 2 are the extreme shifts), so both factors are minimal at the same ϕ \phi ϕ , giving Z 0 = ρ ( p 2 − ρ ) ( p 3 − ρ ) Z_0=\rho(p_2-\rho)(p_3-\rho) Z 0 = ρ ( p 2 − ρ ) ( p 3 − ρ ) and Z 2 = ρ ( p 1 − ρ ) ( p 3 − ρ ) Z_2=\rho(p_1-\rho)(p_3-\rho) Z 2 = ρ ( p 1 − ρ ) ( p 3 − ρ ) . For j = 1 j=1 j = 1 the signs differ and the product of the squared factors is ( p 1 2 + ρ 2 − 2 p 1 ρ u ) ( p 2 2 + ρ 2 + 2 p 2 ρ u ) (p_1^2+\rho^2-2p_1\rho u)(p_2^2+\rho^2+2p_2\rho u) ( p 1 2 + ρ 2 − 2 p 1 ρ u ) ( p 2 2 + ρ 2 + 2 p 2 ρ u ) , u = cos ϕ ∈ [ − 1 , 1 ] u=\cos\phi\in[-1,1] u = cos ϕ ∈ [ − 1 , 1 ] , a concave quadratic in u u u , minimal at an endpoint; the values there are ( p 1 − ρ ) 2 ( p 2 + ρ ) 2 (p_1-\rho)^2(p_2+\rho)^2 ( p 1 − ρ ) 2 ( p 2 + ρ ) 2 and ( p 1 + ρ ) 2 ( p 2 − ρ ) 2 (p_1+\rho)^2(p_2-\rho)^2 ( p 1 + ρ ) 2 ( p 2 − ρ ) 2 , and the first is the smaller since ( p 1 + ρ ) ( p 2 − ρ ) − ( p 1 − ρ ) ( p 2 + ρ ) = 2 ρ ( p 2 − p 1 ) ≥ 0 (p_1+\rho)(p_2-\rho)-(p_1-\rho)(p_2+\rho)=2\rho(p_2-p_1)\geq0 ( p 1 + ρ ) ( p 2 − ρ ) − ( p 1 − ρ ) ( p 2 + ρ ) = 2 ρ ( p 2 − p 1 ) ≥ 0 . So Z 1 = ρ ( p 1 − ρ ) ( p 2 + ρ ) Z_1=\rho(p_1-\rho)(p_2+\rho) Z 1 = ρ ( p 1 − ρ ) ( p 2 + ρ ) . With ρ = p 1 / 2 \rho=p_1/2 ρ = p 1 /2 :
Z 1 = Z 2 = p 1 2 ( p 1 + 2 p 2 ) 8 , Z 0 = p 1 ( 2 p 2 − p 1 ) ( 2 p 2 + p 1 ) 8 , Z_1=Z_2=\frac{p_1^2(p_1+2p_2)}{8},\qquad Z_0=\frac{p_1(2p_2-p_1)(2p_2+p_1)}{8}, Z 1 = Z 2 = 8 p 1 2 ( p 1 + 2 p 2 ) , Z 0 = 8 p 1 ( 2 p 2 − p 1 ) ( 2 p 2 + p 1 ) ,
Υ 1 = 1 + p 1 2 N , Υ 2 = 1 + 3 p 1 2 N , Υ 0 = 1 + 2 p 2 + p 1 2 N . \Upsilon_1=1+\frac{p_1}{2N},\qquad\Upsilon_2=1+\frac{3p_1}{2N},\qquad\Upsilon_0=1+\frac{2p_2+p_1}{2N}. Υ 1 = 1 + 2 N p 1 , Υ 2 = 1 + 2 N 3 p 1 , Υ 0 = 1 + 2 N 2 p 2 + p 1 .
Hence ∣ p i j ( x ) ∣ ≤ ( Υ j / ϑ j 2 ) 1 / 2 ( Υ j / Z j ) k |p_{ij}(x)|\leq(\Upsilon_j/\vartheta_j^2)^{1/2}(\Upsilon_j/Z_j)^k ∣ p ij ( x ) ∣ ≤ ( Υ j / ϑ j 2 ) 1/2 ( Υ j / Z j ) k . For j = 1 , 2 j=1,2 j = 1 , 2 , Υ j ≤ 1 + 3 p 1 2 N \Upsilon_j\leq1+\frac{3p_1}{2N} Υ j ≤ 1 + 2 N 3 p 1 and Υ j / Z j ≤ 8 ( 1 + 3 p 1 / ( 2 N ) ) p 1 2 ( p 1 + 2 p 2 ) \Upsilon_j/Z_j\leq\frac{8(1+3p_1/(2N))}{p_1^2(p_1+2p_2)} Υ j / Z j ≤ p 1 2 ( p 1 + 2 p 2 ) 8 ( 1 + 3 p 1 / ( 2 N )) . For j = 0 j=0 j = 0 the same bound holds because
Υ 0 Z 0 ≤ 8 ( 1 + 3 p 1 2 N ) p 1 2 ( p 1 + 2 p 2 ) ⟺ p 1 ( 1 + 2 p 2 + p 1 2 N ) ≤ ( 2 p 2 − p 1 ) ( 1 + 3 p 1 2 N ) \frac{\Upsilon_0}{Z_0}\leq\frac{8(1+\frac{3p_1}{2N})}{p_1^2(p_1+2p_2)}
\Longleftrightarrow
p_1\left(1+\frac{2p_2+p_1}{2N}\right)\leq(2p_2-p_1)\left(1+\frac{3p_1}{2N}\right) Z 0 Υ 0 ≤ p 1 2 ( p 1 + 2 p 2 ) 8 ( 1 + 2 N 3 p 1 ) ⟺ p 1 ( 1 + 2 N 2 p 2 + p 1 ) ≤ ( 2 p 2 − p 1 ) ( 1 + 2 N 3 p 1 )
⟺ − 2 ( p 2 − p 1 ) ( 1 + p 1 N ) ≤ 0. \Longleftrightarrow-2(p_2-p_1)\left(1+\frac{p_1}{N}\right)\leq0. ⟺ − 2 ( p 2 − p 1 ) ( 1 + N p 1 ) ≤ 0.
The prefactors: ϑ 1 = 1 \vartheta_1=1 ϑ 1 = 1 , ϑ 2 2 = 1 + σ p 1 / N ≥ 1 − p 1 / N \vartheta_2^2=1+\sigma p_1/N\geq1-p_1/N ϑ 2 2 = 1 + σ p 1 / N ≥ 1 − p 1 / N , ϑ 0 2 = 1 − σ p 2 / N ≥ 1 − μ \vartheta_0^2=1-\sigma p_2/N\geq1-\mu ϑ 0 2 = 1 − σ p 2 / N ≥ 1 − μ , so Υ j / ϑ j 2 ≤ 1 + 3 p 1 / ( 2 N ) 1 − p 1 / N ≤ 1 + 3 2 μ 1 − μ \Upsilon_j/\vartheta_j^2\leq\frac{1+3p_1/(2N)}{1-p_1/N}\leq\frac{1+\frac{3}{2}\mu}{1-\mu} Υ j / ϑ j 2 ≤ 1 − p 1 / N 1 + 3 p 1 / ( 2 N ) ≤ 1 − μ 1 + 2 3 μ for j = 1 , 2 j=1,2 j = 1 , 2 (the function y ↦ ( 1 + 3 2 y ) / ( 1 − y ) y\mapsto(1+\frac{3}{2}y)/(1-y) y ↦ ( 1 + 2 3 y ) / ( 1 − y ) is increasing and p 1 / N ≤ μ p_1/N\leq\mu p 1 / N ≤ μ ) and Υ 0 / ϑ 0 2 ≤ 1 + 3 2 μ 1 − μ \Upsilon_0/\vartheta_0^2\leq\frac{1+\frac{3}{2}\mu}{1-\mu} Υ 0 / ϑ 0 2 ≤ 1 − μ 1 + 2 3 μ since 2 p 2 + p 1 ≤ 3 p 2 2p_2+p_1\leq3p_2 2 p 2 + p 1 ≤ 3 p 2 . Hence
∣ p i j ( 1 N ) ∣ ≤ ( 1 + 3 2 μ 1 − μ ) 1 / 2 ( 8 ( 1 + 3 p 1 2 N ) p 1 2 ( p 1 + 2 p 2 ) ) k . \left|p_{ij}\left(\frac{1}{N}\right)\right|\leq\left(\frac{1+\frac{3}{2}\mu}{1-\mu}\right)^{1/2}\left(\frac{8(1+\frac{3p_1}{2N})}{p_1^2(p_1+2p_2)}\right)^k. p ij ( N 1 ) ≤ ( 1 − μ 1 + 2 3 μ ) 1/2 ( p 1 2 ( p 1 + 2 p 2 ) 8 ( 1 + 2 N 3 p 1 ) ) k .
Finally N ≥ 3 p 2 N\geq3p_2 N ≥ 3 p 2 gives 8 ( 1 + 3 p 1 2 N ) p 2 ≤ 4 ( p 1 + 2 p 2 ) 8(1+\frac{3p_1}{2N})p_2\leq4(p_1+2p_2) 8 ( 1 + 2 N 3 p 1 ) p 2 ≤ 4 ( p 1 + 2 p 2 ) , which is equivalent to 3 p 1 p 2 N ≤ p 1 \frac{3p_1p_2}{N}\leq p_1 N 3 p 1 p 2 ≤ p 1 ; this is item 2.
Convention 5.13. The measure constant of the hypergeometric method is ω = 1.3175 \omega= 1.3175 ω = 1.3175 .
The value of ω \omega ω is fitted to the hypothesis p 1 q 3 ≥ 168 p_1q_3 \ge168 p 1 q 3 ≥ 168 of the next theorem. For a pair representation the ratio μ = p 2 / N \mu= p_2/N μ = p 2 / N of Lemma 5.12 equals 2 / ( p 1 q 3 ) 2/(p_1q_3) 2/ ( p 1 q 3 ) , so this hypothesis reads μ ≤ 1 84 \mu\le\frac{1}{84} μ ≤ 84 1 . The linear forms of the scaled approximants decay like L 0 − k L_0^{-k} L 0 − k with L 0 = 1.35 ( 1 − μ ) 2 N 2 / Δ L_0 = 1.35(1-\mu)^2N^2/\Delta L 0 = 1.35 ( 1 − μ ) 2 N 2 /Δ , where Δ = p 1 p 2 p 3 \Delta= p_1p_2p_3 Δ = p 1 p 2 p 3 and 1.35 = 27 20 1.35 = \frac{27}{20} 1.35 = 20 27 is the factor 27 4 \frac{27}{4} 4 27 of Lemma 5.12 divided by the factor 8 / 1.6 = 5 8/1.6 = 5 8/1.6 = 5 that the scaling of Lemma 5.11 leaves after the bound for the prime product. The constant ω \omega ω lies just below 1.35 ( 1 − 1 84 ) 2 = 1.3180 … 1.35(1-\frac{1}{84})^2 = 1.3180\ldots 1.35 ( 1 − 84 1 ) 2 = 1.3180 … , so for every such representation L 0 L_0 L 0 may be replaced by ω N 2 / Δ \omega N^2/\Delta ω N 2 /Δ , the quantity in the denominator of the exponent τ \tau τ .
Theorem 5.14. Let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of an Euler triple, with hypergeometric parameter N N N and rescaled roots θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 (Definition 5.9), and assume that p 1 q 3 ≥ 168 p_1q_3 \ge168 p 1 q 3 ≥ 168 . With ω \omega ω as in Convention 5.13, put
τ = 1 + log ( 20 p 3 N / p 1 ) log ( ω N 2 / ( p 1 p 2 p 3 ) ) = 1 + log ( 10 p 2 p 3 q 3 ) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) \tau= 1 + \frac{\log(20p_3N/p_1)}{\log(\omega N^2/(p_1p_2p_3))} = 1 + \frac{\log(10p_2p_3q_3)}{\log(\omega p_1p_2q_3^2/(4p_3))} τ = 1 + log ( ω N 2 / ( p 1 p 2 p 3 )) log ( 20 p 3 N / p 1 ) = 1 + log ( ω p 1 p 2 q 3 2 / ( 4 p 3 )) log ( 10 p 2 p 3 q 3 ) Then τ < 2 \tau< 2 τ < 2 if and only if 40 p 3 2 < ω p 1 q 3 40p_3^2 < \omega p_1q_3 40 p 3 2 < ω p 1 q 3 , and in that case, for all integers A 0 , A 1 , A 2 A_0,A_1,A_2 A 0 , A 1 , A 2 with A 0 > 0 A_0 > 0 A 0 > 0 ,
max { ∣ θ 1 − A 1 A 0 ∣ , ∣ θ 2 − A 2 A 0 ∣ } > ( 2.9 ⋅ 10 28 p 3 p 1 N ) − 1 A 0 − τ . \max\left\{\left|\theta_1-\frac{A_1}{A_0}\right|,\left|\theta_2-\frac{A_2}{A_0}\right|\right\} > \left(2.9\cdot10^{28}\frac{p_3}{p_1}N\right)^{-1}A_0^{-\tau}. max { θ 1 − A 0 A 1 , θ 2 − A 0 A 2 } > ( 2.9 ⋅ 1 0 28 p 1 p 3 N ) − 1 A 0 − τ .
Proof. We apply Bennett’s criterion (Lemma 5.7 ) to Rickert’s approximants at x = 1 / N x = 1/N x = 1/ N , scaled to integers. Their entries grow at most like P k P^k P k and their linear forms decay at least like L − k L^{-k} L − k , where P = 20 p 3 N / p 1 P = 20p_3N/p_1 P = 20 p 3 N / p 1 and L = ω N 2 / ( p 1 p 2 p 3 ) L = \omega N^2/(p_1p_2p_3) L = ω N 2 / ( p 1 p 2 p 3 ) ; so the exponent of the measure is τ = 1 + log P / log L \tau= 1 + \log P/\log L τ = 1 + log P / log L , and it is below 2 2 2 exactly when P < L P < L P < L , that is, when 40 p 3 2 < ω p 1 q 3 40p_3^2 < \omega p_1q_3 40 p 3 2 < ω p 1 q 3 . This application follows Cipu, Fujita and Mignotte [10 ] : the proof is that of their approximation measure, with the pair coordinates ( p 1 , p 2 , p 3 ) (p_1,p_2,p_3) ( p 1 , p 2 , p 3 ) and q 3 q_3 q 3 in place of the parameters of the three-parameter family, and their constant 2.8605 ⋅ 10 28 2.8605\cdot10^{28} 2.8605 ⋅ 1 0 28 becomes 2.9 ⋅ 10 28 2.9\cdot10^{28} 2.9 ⋅ 1 0 28 because the prefactor of the bound for p i j p_{ij} p ij in Lemma 5.12 is taken for both signs of the representation and for the shift a 0 a_0 a 0 , on the whole range p 1 q 3 ≥ 168 p_1q_3 \ge168 p 1 q 3 ≥ 168 . We use the shifts a 0 , a 1 , a 2 a_0,a_1,a_2 a 0 , a 1 , a 2 and Rickert’s approximants p i j = p i j ( k ) p_{ij}=p_{ij}^{(k)} p ij = p ij ( k ) of Convention 5.1 at x = 1 / N x = 1/N x = 1/ N , whose approximated numbers are ( ϑ 0 , ϑ 1 , ϑ 2 ) = ( θ 1 , 1 , θ 2 ) (\vartheta_0,\vartheta_1,\vartheta_2) = (\theta_1,1,\theta_2) ( ϑ 0 , ϑ 1 , ϑ 2 ) = ( θ 1 , 1 , θ 2 ) (Lemma 5.12 ), Δ = p 1 p 2 p 3 \Delta= p_1p_2p_3 Δ = p 1 p 2 p 3 and μ = p 2 / N = 2 / ( p 1 q 3 ) ≤ 1 / 84 \mu= p_2/N = 2/(p_1q_3) \le1/84 μ = p 2 / N = 2/ ( p 1 q 3 ) ≤ 1/84 ; in particular N ≥ 84 p 2 ≥ 3 p 2 N \ge84p_2 \ge3p_2 N ≥ 84 p 2 ≥ 3 p 2 , so Lemma 5.12 applies.
We first produce the integer matrices. By Lemma 5.11 , for every k ≥ 1 k \ge1 k ≥ 1 the numbers p ^ i j = ( 8 Δ N ) k 2 G k p i j ( 1 / N ) \widehat{p}_{ij} = \frac{(8\Delta N)^k}{2G_k}p_{ij}(1/N) p ij = 2 G k ( 8Δ N ) k p ij ( 1/ N ) are integers, where G k G_k G k is the greatest common divisor of the coefficient products of order k k k (Convention 5.1). By Lemma 5.5 , item 3, and Lemma 5.6 , G k ≥ P ( k ) > 1.6 k / ( 4.09 ⋅ 10 13 ) G_k \ge\mathcal{P}(k) > 1.6^k/(4.09\cdot10^{13}) G k ≥ P ( k ) > 1. 6 k / ( 4.09 ⋅ 1 0 13 ) , where P ( k ) \mathcal{P}(k) P ( k ) is the prime product of Definition 5.4; hence
( 8 Δ N ) k 2 G k ≤ 4.09 ⋅ 10 13 2 ( 8 Δ N 1.6 ) k = 2.045 ⋅ 10 13 ( 5 Δ N ) k . \frac{(8\Delta N)^k}{2G_k} \le\frac{4.09\cdot10^{13}}{2}\left(\frac{8\Delta N}{1.6}\right)^k = 2.045\cdot10^{13}(5\Delta N)^k. 2 G k ( 8Δ N ) k ≤ 2 4.09 ⋅ 1 0 13 ( 1.6 8Δ N ) k = 2.045 ⋅ 1 0 13 ( 5Δ N ) k .
The matrix ( p i j ( 1 / N ) ) i , j (p_{ij}(1/N))_{i,j} ( p ij ( 1/ N ) ) i , j is nonsingular by Lemma 5.3, since 1 / N ≠ 0 1/N \ne0 1/ N = 0 ; hence det ( p ^ i j ) = ( ( 8 Δ N ) k / ( 2 G k ) ) 3 det ( p i j ( 1 / N ) ) ≠ 0 \det(\widehat{p}_{ij}) = ((8\Delta N)^k/(2G_k))^3\det(p_{ij}(1/N)) \ne0 det ( p ij ) = (( 8Δ N ) k / ( 2 G k ) ) 3 det ( p ij ( 1/ N )) = 0 .
We next verify the two bounds of Lemma 5.7 for the integers p i j k = p ^ i j p_{ijk}=\widehat{p}_{ij} p ij k = p ij . By Lemma 5.12 , item 2,
∣ p ^ i j ∣ ≤ 2.045 ⋅ 10 13 ( 1 + 3 2 μ 1 − μ ) 1 / 2 ( 5 Δ N ⋅ 4 p 1 2 p 2 ) k = p ∗ P k , P = 20 p 3 N p 1 , (8) |\widehat{p}_{ij}| \le2.045\cdot10^{13}\left(\frac{1+\frac{3}{2}\mu}{1-\mu}\right)^{1/2}\left(5\Delta N\cdot\frac{4}{p_1^2p_2}\right)^k = p_*P^k,\qquad P=\frac{20p_3N}{p_1}, \tag*{(8)} ∣ p ij ∣ ≤ 2.045 ⋅ 1 0 13 ( 1 − μ 1 + 2 3 μ ) 1/2 ( 5Δ N ⋅ p 1 2 p 2 4 ) k = p ∗ P k , P = p 1 20 p 3 N , (8)
with p ∗ = 2.045 ⋅ 10 13 ( ( 1 + 3 2 μ ) / ( 1 − μ ) ) 1 / 2 ≤ 2.045 ⋅ 10 13 ⋅ ( 85.5 / 83 ) 1 / 2 < 2.0756 ⋅ 10 13 p_* = 2.045 \cdot10^{13}\left((1+\frac{3}{2}\mu)/(1-\mu)\right)^{1/2} \le2.045 \cdot10^{13} \cdot(85.5/83)^{1/2} < 2.0756 \cdot10^{13} p ∗ = 2.045 ⋅ 1 0 13 ( ( 1 + 2 3 μ ) / ( 1 − μ ) ) 1/2 ≤ 2.045 ⋅ 1 0 13 ⋅ ( 85.5/83 ) 1/2 < 2.0756 ⋅ 1 0 13 . By Lemma 5.12 , item 1,
∣ ∑ j p ^ i j ϑ j ∣ ≤ 2.045 ⋅ 10 13 ⋅ 27 64 ( 1 − μ ) − 1 ( 5 Δ N ⋅ 4 27 ( 1 − μ ) 2 N 3 ) k = ℓ ∗ L 0 − k , L 0 = 1.35 ( 1 − μ ) 2 N 2 Δ . \left|\sum_j \widehat{p}_{ij}\vartheta_j\right| \le2.045 \cdot10^{13} \cdot\frac{27}{64}(1-\mu)^{-1}\left(\frac{5\Delta N \cdot4}{27(1-\mu)^2N^3}\right)^k = \ell_*L_0^{-k}, \qquad L_0=\frac{1.35(1-\mu)^2N^2}{\Delta}. j ∑ p ij ϑ j ≤ 2.045 ⋅ 1 0 13 ⋅ 64 27 ( 1 − μ ) − 1 ( 27 ( 1 − μ ) 2 N 3 5Δ N ⋅ 4 ) k = ℓ ∗ L 0 − k , L 0 = Δ 1.35 ( 1 − μ ) 2 N 2 .
with ℓ ∗ = 2.045 ⋅ 10 13 ⋅ 27 64 ( 1 − μ ) − 1 ≤ 2.045 ⋅ 10 13 ⋅ 27 64 ⋅ 84 83 < 8.7313 ⋅ 10 12 \ell_* = 2.045 \cdot10^{13}\cdot\frac{27}{64}(1-\mu)^{-1} \le2.045 \cdot10^{13}\cdot\frac{27}{64}\cdot\frac{84}{83}<8.7313\cdot10^{12} ℓ ∗ = 2.045 ⋅ 1 0 13 ⋅ 64 27 ( 1 − μ ) − 1 ≤ 2.045 ⋅ 1 0 13 ⋅ 64 27 ⋅ 83 84 < 8.7313 ⋅ 1 0 12 . The hypotheses of Lemma 5.7 hold a fortiori with p ∗ p_* p ∗ , ℓ ∗ \ell_* ℓ ∗ replaced by the upper bounds 2.0756 ⋅ 10 13 2.0756\cdot10^{13} 2.0756 ⋅ 1 0 13 and 8.7313 ⋅ 10 12 8.7313\cdot10^{12} 8.7313 ⋅ 1 0 12 , which we use from now on. Since 1.35 ( 1 − 1 84 ) 2 > 1.31804 > ω 1.35(1-\frac{1}{84})^2>1.31804>\omega 1.35 ( 1 − 84 1 ) 2 > 1.31804 > ω , we have L 0 ≥ L = ω N 2 / Δ L_0\ge L=\omega N^2/\Delta L 0 ≥ L = ω N 2 /Δ , so the bound holds with L L L in place of L 0 L_0 L 0 . In pair coordinates N 2 / Δ = p 1 p 2 q 3 2 / ( 4 p 3 ) N^2/\Delta=p_1p_2q_3^2/(4p_3) N 2 /Δ = p 1 p 2 q 3 2 / ( 4 p 3 ) and p 3 N / p 1 = p 2 p 3 q 3 / 2 p_3N/p_1=p_2p_3q_3/2 p 3 N / p 1 = p 2 p 3 q 3 /2 , which gives the second expression of τ \tau τ . Moreover L = ω p 1 p 2 q 3 2 / ( 4 p 3 ) ≥ ω ⋅ 168 ⋅ q 3 ⋅ p 2 p 3 ≥ 21 ω q 3 > 1 L=\omega p_1p_2q_3^2/(4p_3)\ge\omega\cdot168\cdot q_3\cdot\frac{p_2}{p_3}\ge21\omega q_3>1 L = ω p 1 p 2 q 3 2 / ( 4 p 3 ) ≥ ω ⋅ 168 ⋅ q 3 ⋅ p 3 p 2 ≥ 21 ω q 3 > 1 , using p 1 q 3 ≥ 168 p_1q_3\ge168 p 1 q 3 ≥ 168 and p 2 ≥ p 3 / 2 p_2\ge p_3/2 p 2 ≥ p 3 /2 (as p 1 ≤ p 2 p_1\le p_2 p 1 ≤ p 2 by Lemma 5.10 , item 4); and P > 1 P>1 P > 1 . Hence τ = 1 + log P / log L > 1 \tau=1+\log P/\log L>1 τ = 1 + log P / log L > 1 , and the decisive comparison is that of P P P with L L L : τ < 2 \tau<2 τ < 2 if and only if P < L P<L P < L , that is 20 p 3 N / p 1 < ω N 2 / ( p 1 p 2 p 3 ) 20p_3N/p_1<\omega N^2/(p_1p_2p_3) 20 p 3 N / p 1 < ω N 2 / ( p 1 p 2 p 3 ) , that is 20 p 2 p 3 2 < ω N = 1 2 ω p 1 p 2 q 3 20p_2p_3^2<\omega N=\frac{1}{2}\omega p_1p_2q_3 20 p 2 p 3 2 < ω N = 2 1 ω p 1 p 2 q 3 , that is 40 p 3 2 < ω p 1 q 3 40p_3^2<\omega p_1q_3 40 p 3 2 < ω p 1 q 3 .
We finally apply Lemma 5.7 , with the columns indexed by the shifts as explained there: for all integers A 0 , A 1 , A 2 A_0,A_1,A_2 A 0 , A 1 , A 2 with A 0 > 0 A_0>0 A 0 > 0 , max i ∣ θ i − A i / A 0 ∣ > C A 0 − τ \max_i|\theta_i-A_i/A_0|>CA_0^{-\tau} max i ∣ θ i − A i / A 0 ∣ > C A 0 − τ with C − 1 = 4 p ∗ P ( max { 1 , 2 ℓ ∗ } ) τ − 1 = 4 p ∗ P ( 2 ℓ ∗ ) τ − 1 C^{-1}=4p_*P(\max\{1,2\ell_*\})^{\tau-1}=4p_*P(2\ell_*)^{\tau-1} C − 1 = 4 p ∗ P ( max { 1 , 2 ℓ ∗ } ) τ − 1 = 4 p ∗ P ( 2 ℓ ∗ ) τ − 1 . If τ < 2 \tau<2 τ < 2 , then 1 < τ < 2 1<\tau<2 1 < τ < 2 gives ( 2 ℓ ∗ ) τ − 1 < 2 ℓ ∗ (2\ell_*)^{\tau-1}<2\ell_* ( 2 ℓ ∗ ) τ − 1 < 2 ℓ ∗ and
C − 1 < 8 p ∗ ℓ ∗ P ≤ 8 ⋅ 2.0756 ⋅ 10 13 ⋅ 8.7313 ⋅ 10 12 ⋅ 20 p 3 p 1 N < 2.9 ⋅ 10 28 p 3 p 1 N . C^{-1}<8p_*\ell_*P\le8\cdot2.0756\cdot10^{13}\cdot8.7313\cdot10^{12}\cdot20\frac{p_3}{p_1}N<2.9\cdot10^{28}\frac{p_3}{p_1}N. C − 1 < 8 p ∗ ℓ ∗ P ≤ 8 ⋅ 2.0756 ⋅ 1 0 13 ⋅ 8.7313 ⋅ 1 0 12 ⋅ 20 p 1 p 3 N < 2.9 ⋅ 1 0 28 p 1 p 3 N .
The numerical constants of this proof are certified in interval arithmetic with 256-bit precision.
The index in pair coordinates
We return to an irregular extension d d d with index system ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) . The next two lemmas, both after Cipu, Fujita and Mignotte [10 ] , are the inputs of the index inequality Lemma 5.18 : the approximation of the rescaled roots by the common term z z z (Lemma 5.15 ), and the growth of z z z with m m m (Lemma 5.16 ). Lemma 5.17 adds a lower bound for m m m that is special to pair coordinates; it is used in the first case of Theorem 5.19 .
Lemma 5.15. Let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with rescaled roots θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 (Definition 5.9). Let x , y , z x,y,z x , y , z be positive integers with
a z 2 − c x 2 = a − c , b z 2 − c y 2 = b − c . az^2-cx^2=a-c,\qquad bz^2-cy^2=b-c. a z 2 − c x 2 = a − c , b z 2 − c y 2 = b − c . Then
max { ∣ θ 1 − p 3 x p 1 z ∣ , ∣ θ 2 − p 3 y p 2 z ∣ } < p 3 c 2 p 1 a z − 2 . \max\left\{\left|\theta_1-\frac{p_3x}{p_1z}\right|,\left|\theta_2-\frac{p_3y}{p_2z}\right|\right\}<\frac{p_3\sqrt{c}}{2p_1\sqrt{a}}z^{-2}. max { θ 1 − p 1 z p 3 x , θ 2 − p 2 z p 3 y } < 2 p 1 a p 3 c z − 2 .
Proof. This is a lemma of Cipu, Fujita and Mignotte [10 ] in pair coordinates, and the proof is that of the source. By Definition 5.9, θ 1 = p 3 p 1 a / c \theta_1=\frac{p_3}{p_1}\sqrt{a/c} θ 1 = p 1 p 3 a / c and θ 2 = p 3 p 2 b / c \theta_2=\frac{p_3}{p_2}\sqrt{b/c} θ 2 = p 2 p 3 b / c . Since c x 2 − a z 2 = c − a > 0 cx^2-az^2=c-a>0 c x 2 − a z 2 = c − a > 0 , we have x c > z a x\sqrt{c}>z\sqrt{a} x c > z a , hence
∣ θ 1 − p 3 x p 1 z ∣ = p 3 p 1 z c ∣ z a − x c ∣ = p 3 p 1 z c ⋅ c − a z a + x c < p 3 p 1 z c ⋅ c 2 z a = p 3 c 2 p 1 a z − 2 . \left|\theta_1-\frac{p_3x}{p_1z}\right|=\frac{p_3}{p_1z\sqrt{c}}\left|z\sqrt{a}-x\sqrt{c}\right|=\frac{p_3}{p_1z\sqrt{c}}\cdot\frac{c-a}{z\sqrt{a}+x\sqrt{c}}<\frac{p_3}{p_1z\sqrt{c}}\cdot\frac{c}{2z\sqrt{a}}=\frac{p_3\sqrt{c}}{2p_1\sqrt{a}}z^{-2}. θ 1 − p 1 z p 3 x = p 1 z c p 3 z a − x c = p 1 z c p 3 ⋅ z a + x c c − a < p 1 z c p 3 ⋅ 2 z a c = 2 p 1 a p 3 c z − 2 .
In the same way, y c > z b y\sqrt{c}>z\sqrt{b} y c > z b and
∣ θ 2 − p 3 y p 2 z ∣ = p 3 p 2 z c ⋅ c − b z b + y c < p 3 c 2 p 2 b z − 2 ≤ p 3 c 2 p 1 a z − 2 , \left|\theta_2-\frac{p_3y}{p_2z}\right|=\frac{p_3}{p_2z\sqrt{c}}\cdot\frac{c-b}{z\sqrt{b}+y\sqrt{c}}<\frac{p_3\sqrt{c}}{2p_2\sqrt{b}}z^{-2}\le\frac{p_3\sqrt{c}}{2p_1\sqrt{a}}z^{-2}, θ 2 − p 2 z p 3 y = p 2 z c p 3 ⋅ z b + y c c − b < 2 p 2 b p 3 c z − 2 ≤ 2 p 1 a p 3 c z − 2 ,
because p 2 ≥ p 1 p_2\ge p_1 p 2 ≥ p 1 (Lemma 5.10 , item 4) and b > a b>a b > a .
Lemma 5.16. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then
2 z > β 2 m > ( 4 a c ) m . 2z>\beta^{2m}>(4ac)^m. 2 z > β 2 m > ( 4 a c ) m .
Proof. The argument is that of Cipu, Fujita and Mignotte [10 ] , without the factor 0.9999 and without a lower bound on c c c . Here z z z is the positive integer with c d + 1 = z 2 cd+1=z^2 c d + 1 = z 2 and β = s + a c \beta=s+\sqrt{ac} β = s + a c (Definition 2.5, Definition 2.3).
We first bound z z z from below. By Lemma 2.15 , multiplied by a \sqrt{a} a , and since λ ≥ − 1 \lambda\geq-1 λ ≥ − 1 ,
2 z a = ( c + λ a ) β 2 m + ( λ a − c ) β − 2 m ≥ ( c − a ) β 2 m − ( c + a ) β − 2 m . 2z\sqrt{a}=(\sqrt{c}+\lambda\sqrt{a})\beta^{2m}+(\lambda\sqrt{a}-\sqrt{c})\beta^{-2m}\geq(\sqrt{c}-\sqrt{a})\beta^{2m}-(\sqrt{c}+\sqrt{a})\beta^{-2m}. 2 z a = ( c + λ a ) β 2 m + ( λ a − c ) β − 2 m ≥ ( c − a ) β 2 m − ( c + a ) β − 2 m .
We next show that ( c − 2 a ) β 4 m > c + a (\sqrt{c}-2\sqrt{a})\beta^{4m}>\sqrt{c}+\sqrt{a} ( c − 2 a ) β 4 m > c + a . By Lemma 2.4 , item 4, b ≥ a + 2 b\geq a+2 b ≥ a + 2 and r ≥ a + 1 r\geq a+1 r ≥ a + 1 , so c − 4 a = b + 2 r − 3 a ≥ 4 c-4a=b+2r-3a\geq4 c − 4 a = b + 2 r − 3 a ≥ 4 , and c > 4 a c>4a c > 4 a gives 2 a < c 2\sqrt{a}<\sqrt{c} 2 a < c ; hence
c − 2 a = c − 4 a c + 2 a > 4 2 c = 2 c . \sqrt{c}-2\sqrt{a}=\frac{c-4a}{\sqrt{c}+2\sqrt{a}}>\frac{4}{2\sqrt{c}}=\frac{2}{\sqrt{c}}. c − 2 a = c + 2 a c − 4 a > 2 c 4 = c 2 .
Since s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 , s > a c s>\sqrt{ac} s > a c and β > 2 a c \beta>2\sqrt{ac} β > 2 a c ; as m ≥ 1 m\geq1 m ≥ 1 (Lemma 2.14), β 4 m ≥ β 4 > 16 a 2 c 2 ≥ 16 c 2 \beta^{4m}\geq\beta^4>16a^2c^2\geq16c^2 β 4 m ≥ β 4 > 16 a 2 c 2 ≥ 16 c 2 . Therefore ( c − 2 a ) β 4 m > 32 c 3 / 2 > 2 c > c + a (\sqrt{c}-2\sqrt{a})\beta^{4m}>32c^{3/2}>2\sqrt{c}>\sqrt{c}+\sqrt{a} ( c − 2 a ) β 4 m > 32 c 3/2 > 2 c > c + a .
Combining, ( c + a ) β − 2 m < ( c − 2 a ) β 2 m (\sqrt{c}+\sqrt{a})\beta^{-2m}<(\sqrt{c}-2\sqrt{a})\beta^{2m} ( c + a ) β − 2 m < ( c − 2 a ) β 2 m , so 2 z a > ( c − a ) β 2 m − ( c − 2 a ) β 2 m = a β 2 m 2z\sqrt{a}>(\sqrt{c}-\sqrt{a})\beta^{2m}-(\sqrt{c}-2\sqrt{a})\beta^{2m}=\sqrt{a}\,\beta^{2m} 2 z a > ( c − a ) β 2 m − ( c − 2 a ) β 2 m = a β 2 m , that is 2 z > β 2 m 2z>\beta^{2m} 2 z > β 2 m . Finally β 2 m > ( 2 a c ) 2 m = ( 4 a c ) m \beta^{2m}>(2\sqrt{ac})^{2m}=(4ac)^m β 2 m > ( 2 a c ) 2 m = ( 4 a c ) m .
Lemma 5.17. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , and let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of { a , b , c } \{a,b,c\} { a , b , c } (Definition 5.8). Then the integer
Q = 2 ( p 1 m 2 + p 2 n 2 ) + λ p 3 ( m + n ) Q=2(p_1m^2+p_2n^2)+\lambda p_3(m+n) Q = 2 ( p 1 m 2 + p 2 n 2 ) + λ p 3 ( m + n ) is divisible by q 3 q_3 q 3 and satisfies 0 < Q < 2 p 3 m ( m + 1 ) 0<Q<2p_3m(m+1) 0 < Q < 2 p 3 m ( m + 1 ) ; in particular q 3 < 2 p 3 m ( m + 1 ) q_3<2p_3m(m+1) q 3 < 2 p 3 m ( m + 1 ) .
Proof. The divisibility is the integrality of the order-two quotient k c = ( a H + r B n ) / c k_c=(aH+rB_n)/c k c = ( a H + r B n ) / c of Convention 3.1 written in pair coordinates; here H = H 1 + H 2 H=H_1+H_2 H = H 1 + H 2 and B n = 2 n 2 + λ ( m + n ) B_n=2n^2+\lambda(m+n) B n = 2 n 2 + λ ( m + n ) are index polynomials, and k c k_c k c is an integer by item 2 of Lemma 3.2 . Let σ \sigma σ be the sign of the pair representation.
We first prove that q 3 ∣ Q q_3\mid Q q 3 ∣ Q . By Lemma 5.10 , items 1 and 2, a = p 1 q 1 a=p_1q_1 a = p 1 q 1 , c = p 3 q 3 c=p_3q_3 c = p 3 q 3 , p 1 q 2 = r + σ p_1q_2=r+\sigma p 1 q 2 = r + σ , p 3 q 1 = p 1 q 3 − 2 σ p_3q_1=p_1q_3-2\sigma p 3 q 1 = p 1 q 3 − 2 σ and p 3 q 2 = p 2 q 3 + 2 σ p_3q_2=p_2q_3+2\sigma p 3 q 2 = p 2 q 3 + 2 σ . Consequently
p 3 r = p 3 ( p 1 q 2 − σ ) = p 1 ( p 2 q 3 + 2 σ ) − σ p 3 = p 1 p 2 q 3 + σ ( p 1 − p 2 ) , p_3r=p_3(p_1q_2-\sigma)=p_1(p_2q_3+2\sigma)-\sigma p_3=p_1p_2q_3+\sigma(p_1-p_2), p 3 r = p 3 ( p 1 q 2 − σ ) = p 1 ( p 2 q 3 + 2 σ ) − σ p 3 = p 1 p 2 q 3 + σ ( p 1 − p 2 ) ,
and
p 3 2 q 3 k c = p 3 ( a H + r B n ) = p 1 ( p 1 q 3 − 2 σ ) H + ( p 1 p 2 q 3 + σ ( p 1 − p 2 ) ) B n = q 3 p 1 ( p 1 H + p 2 B n ) − σ ( 2 p 1 H − ( p 1 − p 2 ) B n ) . \begin{aligned}
p_3^2q_3k_c=p_3(aH+rB_n)=p_1(p_1q_3-2\sigma)H+(p_1p_2q_3+\sigma(p_1-p_2))B_n \\
&=q_3p_1(p_1H+p_2B_n)-\sigma(2p_1H-(p_1-p_2)B_n).
\end{aligned} p 3 2 q 3 k c = p 3 ( a H + r B n ) = p 1 ( p 1 q 3 − 2 σ ) H + ( p 1 p 2 q 3 + σ ( p 1 − p 2 )) B n = q 3 p 1 ( p 1 H + p 2 B n ) − σ ( 2 p 1 H − ( p 1 − p 2 ) B n ) .
Since 2 H − B n = 2 m 2 + 2 λ m + 2 n 2 + 2 λ n − 2 n 2 − λ ( m + n ) = 2 m 2 + λ ( m + n ) 2H-B_n=2m^2+2\lambda m+2n^2+2\lambda n-2n^2-\lambda(m+n)=2m^2+\lambda(m+n) 2 H − B n = 2 m 2 + 2 λm + 2 n 2 + 2 λn − 2 n 2 − λ ( m + n ) = 2 m 2 + λ ( m + n ) , we have
2 p 1 H − ( p 1 − p 2 ) B n = p 1 ( 2 m 2 + λ ( m + n ) ) + p 2 ( 2 n 2 + λ ( m + n ) ) = Q . 2p_1H-(p_1-p_2)B_n=p_1(2m^2+\lambda(m+n))+p_2(2n^2+\lambda(m+n))=Q. 2 p 1 H − ( p 1 − p 2 ) B n = p 1 ( 2 m 2 + λ ( m + n )) + p 2 ( 2 n 2 + λ ( m + n )) = Q .
Hence σ Q = q 3 ( p 1 ( p 1 H + p 2 B n ) − p 3 2 k c ) \sigma Q=q_3(p_1(p_1H+p_2B_n)-p_3^2k_c) σ Q = q 3 ( p 1 ( p 1 H + p 2 B n ) − p 3 2 k c ) , and as σ = ± 1 \sigma=\pm1 σ = ± 1 and all quantities in the bracket are integers, q 3 ∣ Q q_3\mid Q q 3 ∣ Q .
We next prove the two bounds. By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\leq n<m<l<2n 3 ≤ n < m < l < 2 n . Write Q = p 1 ( 2 m 2 + λ ( m + n ) ) + p 2 ( 2 n 2 + λ ( m + n ) ) Q=p_1(2m^2+\lambda(m+n))+p_2(2n^2+\lambda(m+n)) Q = p 1 ( 2 m 2 + λ ( m + n )) + p 2 ( 2 n 2 + λ ( m + n )) . Both brackets are positive: for λ = 1 \lambda=1 λ = 1 trivially, and for λ = − 1 \lambda=-1 λ = − 1 because m + n < 2 m ≤ 2 m 2 m+n<2m\leq2m^2 m + n < 2 m ≤ 2 m 2 and m + n < 3 n < 2 n 2 m+n<3n<2n^2 m + n < 3 n < 2 n 2 (as m < 2 n m<2n m < 2 n and n ≥ 3 n\geq3 n ≥ 3 ). So Q > 0 Q>0 Q > 0 . Since n < m n<m n < m and λ ≤ 1 \lambda\leq1 λ ≤ 1 , Q ≤ 2 p 1 m 2 + 2 p 2 n 2 + p 3 ( m + n ) < 2 p 3 m 2 + 2 p 3 m = 2 p 3 m ( m + 1 ) Q\leq2p_1m^2+2p_2n^2+p_3(m+n)<2p_3m^2+2p_3m=2p_3m(m+1) Q ≤ 2 p 1 m 2 + 2 p 2 n 2 + p 3 ( m + n ) < 2 p 3 m 2 + 2 p 3 m = 2 p 3 m ( m + 1 ) .
Finally Q Q Q is a positive multiple of q 3 q_3 q 3 , so q 3 ≤ Q < 2 p 3 m ( m + 1 ) q_3\leq Q<2p_3m(m+1) q 3 ≤ Q < 2 p 3 m ( m + 1 ) .
Lemma 5.18. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , and let ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) be a pair representation of { a , b , c } \{a,b,c\} { a , b , c } with p 1 q 3 ≥ 168 p_1q_3 \ge168 p 1 q 3 ≥ 168 and 40 p 3 2 < ω p 1 q 3 40p_3^2 < \omega p_1q_3 40 p 3 2 < ω p 1 q 3 , with ω \omega ω as in Convention 5.13. Then
m log ( 4 a c ) < log 2 + log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) log ( ω p 1 q 3 / ( 40 p 3 2 ) ) ⋅ log ( 7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 ) . m\log(4ac)<\log2+\frac{\log(\omega p_1p_2q_3^2/(4p_3))}{\log(\omega p_1q_3/(40p_3^2))}\cdot\log(7.3\cdot10^{27}p_2^3p_3^3q_3). m log ( 4 a c ) < log 2 + log ( ω p 1 q 3 / ( 40 p 3 2 )) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 )) ⋅ log ( 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 ) .
Proof. The common term z z z gives a simultaneous rational approximation of quality z − 2 z^{-2} z − 2 to the rescaled roots ( θ 1 , θ 2 ) (\theta_1,\theta_2) ( θ 1 , θ 2 ) (Lemma 5.15 ), while the measure of Theorem 5.14 has exponent τ < 2 \tau<2 τ < 2 ; this bounds log z \log z log z from above, and since z z z grows exponentially in m m m (Lemma 5.16 ), it bounds m m m .
Let N N N be the hypergeometric parameter and θ 1 , θ 2 \theta_1,\theta_2 θ 1 , θ 2 the rescaled roots of the pair representation (Definition 5.9), and let σ \sigma σ be its sign. Let τ \tau τ be as in Theorem 5.14 ; by that theorem and the hypothesis 40 p 3 2 < ω p 1 q 3 40p_3^2<\omega p_1q_3 40 p 3 2 < ω p 1 q 3 , τ < 2 \tau<2 τ < 2 . Explicitly, in pair coordinates
τ = 1 + log ( 10 p 2 p 3 q 3 ) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) , 2 − τ = log ( ω p 1 q 3 / ( 40 p 3 2 ) ) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) > 0 , \tau=1+\frac{\log(10p_2p_3q_3)}{\log(\omega p_1p_2q_3^2/(4p_3))},\qquad2-\tau=\frac{\log(\omega p_1q_3/(40p_3^2))}{\log(\omega p_1p_2q_3^2/(4p_3))}>0, τ = 1 + log ( ω p 1 p 2 q 3 2 / ( 4 p 3 )) log ( 10 p 2 p 3 q 3 ) , 2 − τ = log ( ω p 1 p 2 q 3 2 / ( 4 p 3 )) log ( ω p 1 q 3 / ( 40 p 3 2 )) > 0 ,
as 10 p 2 p 3 q 3 > 1 10p_2p_3q_3>1 10 p 2 p 3 q 3 > 1 and ω p 1 q 3 / ( 40 p 3 2 ) > 1 \omega p_1q_3/(40p_3^2)>1 ω p 1 q 3 / ( 40 p 3 2 ) > 1 .
We first bound z z z from above. Let x , y , z x,y,z x , y , z be the positive integers with a d + 1 = x 2 ad+1=x^2 a d + 1 = x 2 , b d + 1 = y 2 bd+1=y^2 b d + 1 = y 2 , c d + 1 = z 2 cd+1=z^2 c d + 1 = z 2 (Definition 2.5 ); then a z 2 − c x 2 = a ( c d + 1 ) − c ( a d + 1 ) = a − c az^2-cx^2=a(cd+1)-c(ad+1)=a-c a z 2 − c x 2 = a ( c d + 1 ) − c ( a d + 1 ) = a − c and likewise b z 2 − c y 2 = b − c bz^2-cy^2=b-c b z 2 − c y 2 = b − c . Apply Theorem 5.14 with ( A 1 , A 2 , A 0 ) = ( p 2 p 3 x , p 1 p 3 y , p 1 p 2 z ) (A_1,A_2,A_0)=(p_2p_3x,p_1p_3y,p_1p_2z) ( A 1 , A 2 , A 0 ) = ( p 2 p 3 x , p 1 p 3 y , p 1 p 2 z ) , so that A 1 / A 0 = p 3 x / ( p 1 z ) A_1/A_0=p_3x/(p_1z) A 1 / A 0 = p 3 x / ( p 1 z ) and A 2 / A 0 = p 3 y / ( p 2 z ) A_2/A_0=p_3y/(p_2z) A 2 / A 0 = p 3 y / ( p 2 z ) , and compare with Lemma 5.15 :
( 2.9 ⋅ 10 28 p 3 p 1 N ) − 1 ( p 1 p 2 z ) − τ < max i ∣ θ i − A i A 0 ∣ < p 3 c 2 p 1 a z − 2 . \left(2.9\cdot10^{28}\frac{p_3}{p_1}N\right)^{-1}(p_1p_2z)^{-\tau}<\max_i\left|\theta_i-\frac{A_i}{A_0}\right|<\frac{p_3\sqrt{c}}{2p_1\sqrt{a}}z^{-2}. ( 2.9 ⋅ 1 0 28 p 1 p 3 N ) − 1 ( p 1 p 2 z ) − τ < i max θ i − A 0 A i < 2 p 1 a p 3 c z − 2 .
Since τ < 2 \tau<2 τ < 2 and p 1 p 2 ≥ 1 p_1p_2\ge1 p 1 p 2 ≥ 1 , ( p 1 p 2 ) τ ≤ ( p 1 p 2 ) 2 (p_1p_2)^\tau\le(p_1p_2)^2 ( p 1 p 2 ) τ ≤ ( p 1 p 2 ) 2 , and with N = 1 2 p 1 p 2 q 3 N=\frac{1}{2}p_1p_2q_3 N = 2 1 p 1 p 2 q 3 ,
z 2 − τ < 2.9 ⋅ 10 28 p 3 p 1 ⋅ p 1 p 2 q 3 2 ⋅ p 1 2 p 2 2 ⋅ p 3 c 2 p 1 a = 2.9 ⋅ 10 28 4 p 1 p 2 3 p 3 2 q 3 c a . z^{2-\tau}<2.9\cdot10^{28}\frac{p_3}{p_1}\cdot\frac{p_1p_2q_3}{2}\cdot p_1^2p_2^2\cdot\frac{p_3\sqrt{c}}{2p_1\sqrt{a}}=\frac{2.9\cdot10^{28}}{4}p_1p_2^3p_3^2q_3\sqrt{\frac{c}{a}}. z 2 − τ < 2.9 ⋅ 1 0 28 p 1 p 3 ⋅ 2 p 1 p 2 q 3 ⋅ p 1 2 p 2 2 ⋅ 2 p 1 a p 3 c = 4 2.9 ⋅ 1 0 28 p 1 p 2 3 p 3 2 q 3 a c .
By Lemma 5.10 , item 3, c / a = p 3 p 1 ( p 1 q 3 p 1 q 3 − 2 σ ) 1 / 2 ≤ p 3 p 1 ( 168 166 ) 1 / 2 \sqrt{c/a}=\frac{p_3}{p_1}\left(\frac{p_1q_3}{p_1q_3-2\sigma}\right)^{1/2}\le\frac{p_3}{p_1}\left(\frac{168}{166}\right)^{1/2} c / a = p 1 p 3 ( p 1 q 3 − 2 σ p 1 q 3 ) 1/2 ≤ p 1 p 3 ( 166 168 ) 1/2 , as p 1 q 3 ≥ 168 p_1q_3\ge168 p 1 q 3 ≥ 168 and u ↦ u / ( u − 2 ) u\mapsto u/(u-2) u ↦ u / ( u − 2 ) decreases (for σ = − 1 \sigma=-1 σ = − 1 the factor u / ( u + 2 ) u/(u+2) u / ( u + 2 ) is below 1). Hence
z 2 − τ < 2.9 ⋅ 10 28 4 ( 168 166 ) 1 / 2 p 2 3 p 3 3 q 3 < 7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 = : Ξ . z^{2-\tau}<\frac{2.9\cdot10^{28}}{4}\left(\frac{168}{166}\right)^{1/2}p_2^3p_3^3q_3<7.3\cdot10^{27}p_2^3p_3^3q_3=:\Xi. z 2 − τ < 4 2.9 ⋅ 1 0 28 ( 166 168 ) 1/2 p 2 3 p 3 3 q 3 < 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 =: Ξ.
Taking logarithms (Ξ > 1 \Xi>1 Ξ > 1 , 2 − τ > 0 2-\tau>0 2 − τ > 0 ),
log z < log Ξ 2 − τ = log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) log ( ω p 1 q 3 / ( 40 p 3 2 ) ) log Ξ . \log z<\frac{\log\Xi}{2-\tau}=\frac{\log(\omega p_1p_2q_3^2/(4p_3))}{\log(\omega p_1q_3/(40p_3^2))}\log\Xi. log z < 2 − τ log Ξ = log ( ω p 1 q 3 / ( 40 p 3 2 )) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 )) log Ξ.
We next bound z z z from below. By Lemma 5.16 , ( 4 a c ) m < 2 z (4ac)^m<2z ( 4 a c ) m < 2 z , so m log ( 4 a c ) < log z + log 2 m\log(4ac)<\log z+\log2 m log ( 4 a c ) < log z + log 2 . Combining the two bounds gives the claim.
Theorem 5.19. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with a pair representation such that
p 1 q 3 ≥ 800 p 3 2 , or such that p 1 q 3 ≥ 42 p 3 2 and c ≥ 4.5 ⋅ 10 12 . p_1q_3\ge800p_3^2,\qquad\text{or such that}\qquad p_1q_3\ge42p_3^2\quad\text{and}\quad c\ge4.5\cdot10^{12}. p 1 q 3 ≥ 800 p 3 2 , or such that p 1 q 3 ≥ 42 p 3 2 and c ≥ 4.5 ⋅ 1 0 12 . Then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Proof. In each case the upper bound for the index m m m given by Lemma 5.18 contradicts the lower bounds for m m m given by Lemma 5.17 and Theorem 3.6 (first case) or by Theorem 3.6 alone (second case).
Suppose that { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension d d d , let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be its index system (Proposition 2.11), let σ \sigma σ be the sign of the pair representation, and let ω \omega ω be as in Convention 5.13. We first collect what both cases use. By Lemma 5.10 , item 4, 1 ≤ p 1 ≤ 1 2 p 3 1\leq p_1\leq\frac{1}{2}p_3 1 ≤ p 1 ≤ 2 1 p 3 , so p 3 ≥ 2 p_3\geq2 p 3 ≥ 2 . In both cases p 1 q 3 ≥ 42 p 3 2 p_1q_3\geq42p_3^2 p 1 q 3 ≥ 42 p 3 2 , hence p 1 q 3 ≥ 42 ⋅ 4 = 168 p_1q_3\geq42\cdot4=168 p 1 q 3 ≥ 42 ⋅ 4 = 168 and 40 p 3 2 < 42 ω p 3 2 ≤ ω p 1 q 3 40p_3^2<42\omega p_3^2\leq\omega p_1q_3 40 p 3 2 < 42 ω p 3 2 ≤ ω p 1 q 3 (as 42 ω > 40 42\omega>40 42 ω > 40 ); so Lemma 5.18 applies and gives
m log ( 4 a c ) < R : = log 2 + log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) log ( ω p 1 q 3 / ( 40 p 3 2 ) ) ⋅ log ( 7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 ) . m\log(4ac)<\mathcal{R}:=\log2+\frac{\log\left(\omega p_1p_2q_3^2/(4p_3)\right)}{\log\left(\omega p_1q_3/(40p_3^2)\right)}\cdot\log\left(7.3\cdot10^{27}p_2^3p_3^3q_3\right). m log ( 4 a c ) < R := log 2 + log ( ω p 1 q 3 / ( 40 p 3 2 ) ) log ( ω p 1 p 2 q 3 2 / ( 4 p 3 ) ) ⋅ log ( 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 ) .
The three logarithms in R \mathcal{R} R are positive: ω p 1 q 3 / ( 40 p 3 2 ) > 1 \omega p_1q_3/(40p_3^2)>1 ω p 1 q 3 / ( 40 p 3 2 ) > 1 as just shown, 7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 > 1 7.3\cdot10^{27}p_2^3p_3^3q_3>1 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 > 1 , and ω p 1 p 2 q 3 2 / ( 4 p 3 ) = 10 p 2 p 3 q 3 ⋅ ω p 1 q 3 / ( 40 p 3 2 ) > 1 \omega p_1p_2q_3^2/(4p_3)=10p_2p_3q_3\cdot\omega p_1q_3/(40p_3^2)>1 ω p 1 p 2 q 3 2 / ( 4 p 3 ) = 10 p 2 p 3 q 3 ⋅ ω p 1 q 3 / ( 40 p 3 2 ) > 1 . Hence R \mathcal{R} R can only increase when the two logarithms of the numerator are replaced by upper bounds and the logarithm of the denominator by a positive lower bound. On the other side, by Theorem 3.6 , c < 8 m 4 c<8m^4 c < 8 m 4 , that is
2 m > ( 2 c ) 1 / 4 . 2m>(2c)^{1/4}. 2 m > ( 2 c ) 1/4 .
and by Lemma 5.17 , q 3 < 2 p 3 m ( m + 1 ) q_3<2p_3m(m+1) q 3 < 2 p 3 m ( m + 1 ) , that is
m ( m + 1 ) > q 3 2 p 3 . m(m+1)>\frac{q_3}{2p_3}. m ( m + 1 ) > 2 p 3 q 3 .
We now treat the case p 1 q 3 ≥ 800 p 3 2 p_1q_3\geq800p_3^2 p 1 q 3 ≥ 800 p 3 2 . Put u : = q 3 / ( 2 p 3 ) u:=q_3/(2p_3) u := q 3 / ( 2 p 3 ) . Since p 1 ≤ 1 2 p 3 p_1\leq\frac{1}{2}p_3 p 1 ≤ 2 1 p 3 , u ≥ p 1 q 3 / p 3 2 ≥ 800 u\geq p_1q_3/p_3^2\geq800 u ≥ p 1 q 3 / p 3 2 ≥ 800 , and c = p 3 q 3 = 2 u p 3 2 c=p_3q_3=2up_3^2 c = p 3 q 3 = 2 u p 3 2 (Lemma 5.10 , item 2). Moreover p 3 q 1 = p 1 q 3 − 2 σ ≥ 800 p 3 2 − 2 p_3q_1=p_1q_3-2\sigma\geq800p_3^2-2 p 3 q 1 = p 1 q 3 − 2 σ ≥ 800 p 3 2 − 2 (Lemma 5.10 , item 2), so
a = p 1 q 1 ≥ q 1 ≥ 800 p 3 − 2 p 3 ≥ 799.5 p 3 , 4 a c ≥ 8 ⋅ 799.5 u p 3 3 = 6396 u p 3 3 . a=p_1q_1\geq q_1\geq800p_3-\frac{2}{p_3}\geq799.5p_3,\qquad4ac\geq8\cdot799.5up_3^3=6396up_3^3. a = p 1 q 1 ≥ q 1 ≥ 800 p 3 − p 3 2 ≥ 799.5 p 3 , 4 a c ≥ 8 ⋅ 799.5 u p 3 3 = 6396 u p 3 3 .
The two lower bounds for m m m read
m > u + 1 4 − 1 2 and m > 1 2 ( 2 c ) 1 / 4 = ( u p 3 2 4 ) 1 / 4 . m>\sqrt{u+\frac{1}{4}}-\frac{1}{2}
\qquad\text{and}\qquad
m>\frac{1}{2}(2c)^{1/4}=\left(\frac{up_3^2}{4}\right)^{1/4}. m > u + 4 1 − 2 1 and m > 2 1 ( 2 c ) 1/4 = ( 4 u p 3 2 ) 1/4 .
On the other hand, p 1 p 2 ≤ 1 4 p 3 2 p_1p_2\leq\frac{1}{4}p_3^2 p 1 p 2 ≤ 4 1 p 3 2 , p 2 < p 3 p_2<p_3 p 2 < p 3 , q 3 = 2 u p 3 q_3=2up_3 q 3 = 2 u p 3 and log ( ω p 1 q 3 / ( 40 p 3 2 ) ) ≥ log ( 800 ω / 40 ) = log ( 20 ω ) > 0 \log(\omega p_1q_3/(40p_3^2))\geq\log(800\omega/40)=\log(20\omega)>0 log ( ω p 1 q 3 / ( 40 p 3 2 )) ≥ log ( 800 ω /40 ) = log ( 20 ω ) > 0 , so the bound m log ( 4 a c ) < R m\log(4ac)<\mathcal{R} m log ( 4 a c ) < R yields
m log ( 6396 u p 3 3 ) ≤ m log ( 4 a c ) < log 2 + log ( ω u 2 p 3 4 / 4 ) log ( 20 ω ) log ( 1.46 ⋅ 10 28 u p 3 7 ) = : R ∗ ( u , p 3 ) . m\log(6396up_3^3)\leq m\log(4ac)<\log2+\frac{\log(\omega u^2p_3^4/4)}{\log(20\omega)}\log(1.46\cdot10^{28}up_3^7)=:\mathcal{R}^*(u,p_3). m log ( 6396 u p 3 3 ) ≤ m log ( 4 a c ) < log 2 + log ( 20 ω ) log ( ω u 2 p 3 4 /4 ) log ( 1.46 ⋅ 1 0 28 u p 3 7 ) =: R ∗ ( u , p 3 ) .
We next show that the contradiction need only be checked at u = 800 u=800 u = 800 . Fix p 3 ≥ 2 p_3\geq2 p 3 ≥ 2 , put ξ = log u \xi=\log u ξ = log u , η = log p 3 \eta=\log p_3 η = log p 3 and w 0 = 1 / log ( 20 ω ) < 0.306 w_0=1/\log(20\omega)<0.306 w 0 = 1/ log ( 20 ω ) < 0.306 , and consider
g 1 ( u ) = ( u + 1 4 − 1 2 ) log ( 6396 u p 3 3 ) − R ∗ ( u , p 3 ) , g_1(u)=\left(\sqrt{u+\frac{1}{4}}-\frac{1}{2}\right)\log(6396up_3^3)-\mathcal{R}^*(u,p_3), g 1 ( u ) = ( u + 4 1 − 2 1 ) log ( 6396 u p 3 3 ) − R ∗ ( u , p 3 ) ,
g 2 ( u ) = ( u p 3 2 4 ) 1 / 4 log ( 6396 u p 3 3 ) − R ∗ ( u , p 3 ) . g_2(u)=\left(\frac{up_3^2}{4}\right)^{1/4}\log(6396up_3^3)-\mathcal{R}^*(u,p_3). g 2 ( u ) = ( 4 u p 3 2 ) 1/4 log ( 6396 u p 3 3 ) − R ∗ ( u , p 3 ) .
Here R ∗ ( u , p 3 ) = log 2 + w 0 ( 2 ξ + 3 η + log ( ω / 4 ) ) ( ξ + 7 η + log ( 1.46 ⋅ 10 28 ) ) \mathcal{R}^*(u,p_3)=\log2+w_0(2\xi+3\eta+\log(\omega/4))(\xi+7\eta+\log(1.46\cdot10^{28})) R ∗ ( u , p 3 ) = log 2 + w 0 ( 2 ξ + 3 η + log ( ω /4 )) ( ξ + 7 η + log ( 1.46 ⋅ 1 0 28 )) , so
u ∂ u R ∗ ( u , p 3 ) = w 0 ( 4 ξ + 17 η + 2 log ( 1.46 ⋅ 10 28 ) + log ( ω / 4 ) ) < 1.23 ξ + 5.21 η + 39.4. u\partial_u\mathcal{R}^*(u,p_3)=w_0(4\xi+17\eta+2\log(1.46\cdot10^{28})+\log(\omega/4))<1.23\xi+5.21\eta+39.4. u ∂ u R ∗ ( u , p 3 ) = w 0 ( 4 ξ + 17 η + 2 log ( 1.46 ⋅ 1 0 28 ) + log ( ω /4 )) < 1.23 ξ + 5.21 η + 39.4.
On the other hand, log ( 6396 u p 3 3 ) = ξ + 3 η + log 6396 > ξ + 3 η + 8.76 \log(6396up_{3}^{3})=\xi+3\eta+\log6396>\xi+3\eta+8.76 log ( 6396 u p 3 3 ) = ξ + 3 η + log 6396 > ξ + 3 η + 8.76 . For u ≥ 800 u\ge800 u ≥ 800 we have u / ( 2 u + 1 / 4 ) > 14.1 u/(2\sqrt{u+1/4})>14.1 u / ( 2 u + 1/4 ) > 14.1 , so
u ∂ u [ ( u + 1 4 − 1 2 ) log ( 6396 u p 3 3 ) ] > 14.1 ( ξ + 3 η + 8.76 ) > 1.23 ξ + 5.21 η + 39.4. u\,\partial_u\left[\left(\sqrt{u+\frac{1}{4}}-\frac{1}{2}\right)\log(6396up_{3}^{3})\right]>14.1(\xi+3\eta+8.76)>1.23\xi+5.21\eta+39.4. u ∂ u [ ( u + 4 1 − 2 1 ) log ( 6396 u p 3 3 ) ] > 14.1 ( ξ + 3 η + 8.76 ) > 1.23 ξ + 5.21 η + 39.4.
For u ≥ 800 u\ge800 u ≥ 800 and p 3 ≥ 55 p_{3}\ge55 p 3 ≥ 55 we have ( u p 3 2 / 4 ) 1 / 4 > 27.8 (up_{3}^{2}/4)^{1/4}>27.8 ( u p 3 2 /4 ) 1/4 > 27.8 , so
u ∂ u [ ( u p 3 2 4 ) 1 / 4 log ( 6396 u p 3 3 ) ] = ( u p 3 2 4 ) 1 / 4 ( ξ + 3 η + log 6396 4 + 1 ) > 27.8 ( ξ + 3 η + 8.76 4 + 1 ) > 1.23 ξ + 5.21 η + 39.4. \begin{aligned}
u\,\partial_u\left[\left(\frac{up_{3}^{2}}{4}\right)^{1/4}\log(6396up_{3}^{3})\right]=\left(\frac{up_{3}^{2}}{4}\right)^{1/4}\left(\frac{\xi+3\eta+\log6396}{4}+1\right) \\
&>27.8\left(\frac{\xi+3\eta+8.76}{4}+1\right)>1.23\xi+5.21\eta+39.4.
\end{aligned} u ∂ u [ ( 4 u p 3 2 ) 1/4 log ( 6396 u p 3 3 ) ] = ( 4 u p 3 2 ) 1/4 ( 4 ξ + 3 η + log 6396 + 1 ) > 27.8 ( 4 ξ + 3 η + 8.76 + 1 ) > 1.23 ξ + 5.21 η + 39.4.
Hence g 1 g_{1} g 1 is strictly increasing on [ 800 , ∞ ) [800,\infty) [ 800 , ∞ ) for every p 3 ≥ 2 p_{3}\ge2 p 3 ≥ 2 , and g 2 g_{2} g 2 is strictly increasing on [ 800 , ∞ ) [800,\infty) [ 800 , ∞ ) for every p 3 ≥ 55 p_{3}\ge55 p 3 ≥ 55 . The contradiction to be derived is g 1 ( u ) > 0 g_{1}(u)>0 g 1 ( u ) > 0 for 2 ≤ p 3 ≤ 55 2\le p_{3}\le55 2 ≤ p 3 ≤ 55 and g 2 ( u ) > 0 g_{2}(u)>0 g 2 ( u ) > 0 for p 3 ≥ 55 p_{3}\ge55 p 3 ≥ 55 , so it suffices to establish it at u = 800 u=800 u = 800 .
We now treat u = 800 u=800 u = 800 and 2 ≤ p 3 ≤ 55 2\le p_{3}\le55 2 ≤ p 3 ≤ 55 , with η = log p 3 ≥ log 2 \eta=\log p_{3}\ge\log2 η = log p 3 ≥ log 2 . Using m > 800.25 − 0.5 m>\sqrt{800.25}-0.5 m > 800.25 − 0.5 , the difference
f 1 ( η ) : = ( 800.25 − 0.5 ) ( log ( 5 116 800 ) + 3 η ) − log 2 − ( log ( 160 000 ω ) + 3 η ) ( log ( 1.168 ⋅ 10 31 ) + 7 η ) log ( 20 ω ) f_{1}(\eta):=(\sqrt{800.25}-0.5)(\log(5\,116\,800)+3\eta)-\log2-\frac{(\log(160\,000\omega)+3\eta)(\log(1.168\cdot10^{31})+7\eta)}{\log(20\omega)} f 1 ( η ) := ( 800.25 − 0.5 ) ( log ( 5 116 800 ) + 3 η ) − log 2 − log ( 20 ω ) ( log ( 160 000 ω ) + 3 η ) ( log ( 1.168 ⋅ 1 0 31 ) + 7 η )
is a strictly concave quadratic polynomial in η \eta η , since its second derivative is − 42 / log ( 20 ω ) < 0 -42/\log(20\omega)<0 − 42/ log ( 20 ω ) < 0 . Interval evaluation at the endpoints, in 256-bit arithmetic, gives f 1 ( log 2 ) > 151.5836 f_{1}(\log2)>151.5836 f 1 ( log 2 ) > 151.5836 and f 1 ( log 55 ) > 23.5355 f_{1}(\log55)>23.5355 f 1 ( log 55 ) > 23.5355 , so by concavity f 1 ( log p 3 ) > 0 f_{1}(\log p_{3})>0 f 1 ( log p 3 ) > 0 for all p 3 ∈ [ 2 , 55 ] p_{3}\in[2,55] p 3 ∈ [ 2 , 55 ] .
We next treat u = 800 u=800 u = 800 and p 3 ≥ 55 p_{3}\ge55 p 3 ≥ 55 . Using m > ( 200 p 3 2 ) 1 / 4 = 200 1 / 4 e η / 2 m>(200p_{3}^{2})^{1/4}=200^{1/4}e^{\eta/2} m > ( 200 p 3 2 ) 1/4 = 20 0 1/4 e η /2 , the function
f 2 ( η ) : = 200 1 / 4 e η / 2 ( log ( 5 116 800 ) + 3 η ) − R ∗ ( 800 , e η ) f_{2}(\eta):=200^{1/4}e^{\eta/2}(\log(5\,116\,800)+3\eta)-R^{*}(800,e^{\eta}) f 2 ( η ) := 20 0 1/4 e η /2 ( log ( 5 116 800 ) + 3 η ) − R ∗ ( 800 , e η )
satisfies f 2 ( log 55 ) > 26.3014 f_{2}(\log55)>26.3014 f 2 ( log 55 ) > 26.3014 , by the same interval evaluation. Its derivative is positive for η ≥ η 0 : = log 55 \eta\ge\eta_{0}:=\log55 η ≥ η 0 := log 55 . Indeed, with w 1 = log ( 5 116 800 ) w_{1}=\log(5\,116\,800) w 1 = log ( 5 116 800 ) , ξ 0 = log 800 \xi_{0}=\log800 ξ 0 = log 800 , C 1 = 2 ξ 0 + log ( ω / 4 ) C_{1}=2\xi_{0}+\log(\omega/4) C 1 = 2 ξ 0 + log ( ω /4 ) and C 2 = ξ 0 + log ( 1.46 ⋅ 10 28 ) C_{2}=\xi_{0}+\log(1.46\cdot10^{28}) C 2 = ξ 0 + log ( 1.46 ⋅ 1 0 28 ) ,
f 2 ′ ( η ) = 200 1 / 4 e η / 2 ( w 1 + 3 η 2 + 3 ) − w 0 ( 42 η + 3 C 2 + 7 C 1 ) . f_{2}'(\eta)=200^{1/4}e^{\eta/2}\left(\frac{w_{1}+3\eta}{2}+3\right)-w_{0}(42\eta+3C_{2}+7C_{1}). f 2 ′ ( η ) = 20 0 1/4 e η /2 ( 2 w 1 + 3 η + 3 ) − w 0 ( 42 η + 3 C 2 + 7 C 1 ) .
By convexity e η / 2 ≥ e η 0 / 2 ( 1 + v / 2 ) e^{\eta/2}\ge e^{\eta_{0}/2}(1+v/2) e η /2 ≥ e η 0 /2 ( 1 + v /2 ) with v = η − η 0 ≥ 0 v=\eta-\eta_{0}\ge0 v = η − η 0 ≥ 0 , so the first term is at least
200 1 / 4 55 ( 1 + v 2 ) ( w 1 + 3 η 0 2 + 3 + 3 v 2 ) > 466 + 275 v , 200^{1/4}\sqrt{55}\left(1+\frac{v}{2}\right)\left(\frac{w_{1}+3\eta_{0}}{2}+3+\frac{3v}{2}\right)>466+275v, 20 0 1/4 55 ( 1 + 2 v ) ( 2 w 1 + 3 η 0 + 3 + 2 3 v ) > 466 + 275 v ,
while the second is less than 145 + 13 v 145+13v 145 + 13 v . Thus f 2 ( log p 3 ) > 0 f_{2}(\log p_{3})>0 f 2 ( log p 3 ) > 0 for all p 3 ≥ 55 p_{3}\ge55 p 3 ≥ 55 . In both ranges the two lower bounds for m m m contradict the upper bound m log ( 4 a c ) < R m\log(4ac)<\mathcal{R} m log ( 4 a c ) < R , which settles the first case.
We finally treat the case p 1 q 3 ≥ 42 p 3 2 p_{1}q_{3}\ge42p_{3}^{2} p 1 q 3 ≥ 42 p 3 2 and c ≥ 4.5 ⋅ 10 12 c\ge4.5\cdot10^{12} c ≥ 4.5 ⋅ 1 0 12 , by an argument of Cipu, Fujita and Mignotte [10 ] in pair coordinates, with their congruence lower bound for m m m replaced by the quartic index floor of Theorem 3.6 . We first bound R \mathcal{R} R by a function of c c c alone. By Lemma 5.10 , item 2, c = p 3 q 3 c=p_{3}q_{3} c = p 3 q 3 ; moreover p 1 p 2 ≤ 1 4 p 3 2 p_{1}p_{2}\le\frac{1}{4}p_{3}^{2} p 1 p 2 ≤ 4 1 p 3 2 by the AM–GM inequality, and p 3 ≥ 2 p_{3}\ge2 p 3 ≥ 2 . Hence
ω p 1 p 2 q 3 2 4 p 3 ≤ ω p 3 q 3 2 16 = ω c 2 16 p 3 ≤ ω c 2 32 ; ω p 1 q 3 2 40 p 3 2 ≥ 42 ω 40 > 1. \frac{\omega p_{1}p_{2}q_{3}^{2}}{4p_{3}}\le\frac{\omega p_{3}q_{3}^{2}}{16}=\frac{\omega c^{2}}{16p_{3}}\le\frac{\omega c^{2}}{32};\qquad\frac{\omega p_{1}q_{3}^{2}}{40p_{3}^{2}}\ge\frac{42\omega}{40}>1. 4 p 3 ω p 1 p 2 q 3 2 ≤ 16 ω p 3 q 3 2 = 16 p 3 ω c 2 ≤ 32 ω c 2 ; 40 p 3 2 ω p 1 q 3 2 ≥ 40 42 ω > 1.
Moreover p 1 ≤ 1 2 p 3 p_{1}\le\frac{1}{2}p_{3} p 1 ≤ 2 1 p 3 gives 42 p 3 2 ≤ p 1 q 3 ≤ 1 2 p 3 q 3 = 1 2 c 42p_{3}^{2}\le p_{1}q_{3}\le\frac{1}{2}p_{3}q_{3}=\frac{1}{2}c 42 p 3 2 ≤ p 1 q 3 ≤ 2 1 p 3 q 3 = 2 1 c , so p 3 ≤ ( c / 84 ) 1 / 2 p_{3}\le(c/84)^{1/2} p 3 ≤ ( c /84 ) 1/2 , and with p 2 < p 3 p_{2}<p_{3} p 2 < p 3 , q 3 = c / p 3 q_{3}=c/p_{3} q 3 = c / p 3 ,
7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 < 7.3 ⋅ 10 27 p 3 5 c ≤ 7.3 ⋅ 10 27 84 5 / 2 c 7 / 2 . 7.3\cdot10^{27}p_{2}^{3}p_{3}^{3}q_{3}<7.3\cdot10^{27}p_{3}^{5}c\le\frac{7.3\cdot10^{27}}{84^{5/2}}c^{7/2}. 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 < 7.3 ⋅ 1 0 27 p 3 5 c ≤ 8 4 5/2 7.3 ⋅ 1 0 27 c 7/2 .
On the left, 2 m > ( 2 c ) 1 / 4 2m > (2c)^{1/4} 2 m > ( 2 c ) 1/4 , and log ( 4 a c ) ≥ log ( 4 c ) \log(4ac) \ge\log(4c) log ( 4 a c ) ≥ log ( 4 c ) as a ≥ 1 a \ge1 a ≥ 1 . Put ζ = log c \zeta= \log c ζ = log c , A = 7.3 ⋅ 10 27 / 84 5 / 2 A = 7.3 \cdot10^{27}/84^{5/2} A = 7.3 ⋅ 1 0 27 /8 4 5/2 and, for real ζ \zeta ζ ,
ϕ ( ζ ) = 2 1 / 4 2 e ζ / 4 ( ζ + log 4 ) − log 2 − ( 2 ζ + log ω 32 ) ( 7 2 ζ + log A ) log 42 ω 40 . \phi(\zeta) = \frac{2^{1/4}}{2}e^{\zeta/4}(\zeta+\log4)-\log2-\frac{(2\zeta+\log\frac{\omega}{32})(\frac{7}{2}\zeta+\log A)}{\log\frac{42\omega}{40}}. ϕ ( ζ ) = 2 2 1/4 e ζ /4 ( ζ + log 4 ) − log 2 − log 40 42 ω ( 2 ζ + log 32 ω ) ( 2 7 ζ + log A ) .
Since 1 2 ( 2 c ) 1 / 4 log ( 4 c ) = 2 1 / 4 2 e ζ / 4 ( ζ + log 4 ) \frac{1}{2}(2c)^{1/4}\log(4c)=\frac{2^{1/4}}{2}e^{\zeta/4}(\zeta+\log4) 2 1 ( 2 c ) 1/4 log ( 4 c ) = 2 2 1/4 e ζ /4 ( ζ + log 4 ) , log ( ω c 2 / 32 ) = 2 ζ + log ω 32 \log(\omega c^2/32)=2\zeta+\log\frac{\omega}{32} log ( ω c 2 /32 ) = 2 ζ + log 32 ω and log ( A c 7 / 2 ) = 7 2 ζ + log A \log(Ac^{7/2})=\frac{7}{2}\zeta+\log A log ( A c 7/2 ) = 2 7 ζ + log A , these bounds turn m log ( 4 a c ) < R m\log(4ac)<\mathcal{R} m log ( 4 a c ) < R into ϕ ( log c ) < 0 \phi(\log c)<0 ϕ ( log c ) < 0 .
We now show that ϕ ≥ 0 \phi\ge0 ϕ ≥ 0 on [ ζ 1 , ∞ ) [\zeta_1,\infty) [ ζ 1 , ∞ ) , where ζ 1 = log ( 4.5 ⋅ 10 12 ) \zeta_1=\log(4.5\cdot10^{12}) ζ 1 = log ( 4.5 ⋅ 1 0 12 ) ; since log c ≥ ζ 1 \log c\ge\zeta_1 log c ≥ ζ 1 , this contradicts ϕ ( log c ) < 0 \phi(\log c)<0 ϕ ( log c ) < 0 . Differentiating twice,
ϕ ′ ′ ( ζ ) = 2 1 / 4 2 e ζ / 4 ( ζ + log 4 16 + 1 2 ) − 14 log 42 ω 40 . \phi''(\zeta)=\frac{2^{1/4}}{2}e^{\zeta/4}\left(\frac{\zeta+\log4}{16}+\frac{1}{2}\right)-\frac{14}{\log\frac{42\omega}{40}}. ϕ ′′ ( ζ ) = 2 2 1/4 e ζ /4 ( 16 ζ + log 4 + 2 1 ) − log 40 42 ω 14 .
and the first term has derivative 2 1 / 4 2 e ζ / 4 ( ζ + log 4 64 + 3 16 ) > 0 \frac{2^{1/4}}{2}e^{\zeta/4}\left(\frac{\zeta+\log4}{64}+\frac{3}{16}\right)>0 2 2 1/4 e ζ /4 ( 64 ζ + l o g 4 + 16 3 ) > 0 for ζ > − 12 − log 4 \zeta>-12-\log4 ζ > − 12 − log 4 . As ζ 1 > 29 > − 12 − log 4 \zeta_1>29>-12-\log4 ζ 1 > 29 > − 12 − log 4 , ϕ ′ ′ \phi'' ϕ ′′ is increasing on [ ζ 1 , ∞ ) [\zeta_1,\infty) [ ζ 1 , ∞ ) . Hence, if ϕ ′ ′ ( ζ 1 ) ≥ 0 \phi''(\zeta_1)\ge0 ϕ ′′ ( ζ 1 ) ≥ 0 , then ϕ ′ \phi' ϕ ′ is non-decreasing on [ ζ 1 , ∞ ) [\zeta_1,\infty) [ ζ 1 , ∞ ) ; and if moreover ϕ ( ζ 1 ) ≥ 0 \phi(\zeta_1)\ge0 ϕ ( ζ 1 ) ≥ 0 and ϕ ′ ( ζ 1 ) ≥ 0 \phi'(\zeta_1)\ge0 ϕ ′ ( ζ 1 ) ≥ 0 , then for ζ ≥ ζ 1 \zeta\ge\zeta_1 ζ ≥ ζ 1 , ϕ ( ζ ) = ϕ ( ζ 1 ) + ∫ ζ 1 ζ ϕ ′ ≥ ϕ ( ζ 1 ) + ( ζ − ζ 1 ) ϕ ′ ( ζ 1 ) ≥ 0 \phi(\zeta)=\phi(\zeta_1)+\int_{\zeta_1}^{\zeta}\phi'\ge\phi(\zeta_1)+(\zeta-\zeta_1)\phi'(\zeta_1)\ge0 ϕ ( ζ ) = ϕ ( ζ 1 ) + ∫ ζ 1 ζ ϕ ′ ≥ ϕ ( ζ 1 ) + ( ζ − ζ 1 ) ϕ ′ ( ζ 1 ) ≥ 0 . In interval arithmetic with 256-bit precision, ϕ ( ζ 1 ) > 115.1 \phi(\zeta_1)>115.1 ϕ ( ζ 1 ) > 115.1 , ϕ ′ ( ζ 1 ) > 5924 \phi'(\zeta_1)>5924 ϕ ′ ( ζ 1 ) > 5924 and ϕ ′ ′ ( ζ 1 ) > 2041 \phi''(\zeta_1)>2041 ϕ ′′ ( ζ 1 ) > 2041 . Hence ϕ ( log c ) ≥ 0 \phi(\log c)\ge0 ϕ ( log c ) ≥ 0 , a contradiction, which settles the second case. So { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Since p 1 q 3 = p 3 q 1 + 2 σ p_1q_3=p_3q_1+2\sigma p 1 q 3 = p 3 q 1 + 2 σ (Lemma 5.10, item 2), the hypotheses p 1 q 3 ≥ 800 p 3 2 p_1q_3\ge800p_3^2 p 1 q 3 ≥ 800 p 3 2 and p 1 q 3 ≥ 42 p 3 2 p_1q_3\ge42p_3^2 p 1 q 3 ≥ 42 p 3 2 say, up to the term 2 σ 2\sigma 2 σ , that q 1 ≥ 800 p 3 q_1\ge800p_3 q 1 ≥ 800 p 3 and q 1 ≥ 42 p 3 q_1\ge42p_3 q 1 ≥ 42 p 3 : the smallest element a = p 1 q 1 a=p_1q_1 a = p 1 q 1 has a factor q 1 q_1 q 1 that is large compared with p 1 + p 2 p_1+p_2 p 1 + p 2 . This is the case in the three-parameter family of § 7 , for which item 3 of Lemma 8.2 gives p 1 q 3 ≥ K − 2 3 p 3 2 p_1q_3\ge\frac{K-2}{3}p_3^2 p 1 q 3 ≥ 3 K − 2 p 3 2 . The criterion is applied there, and Lemma 5.18 once more in Appendix § 8 .
Coprime divisors of the index combinations
This section proves Theorem 6.3 , a lower bound for the index of another kind than the quartic floor. The quartic floor (Theorem 3.6 ) bounds m m m from below by a power of c c c . Theorem 6.3 shows that three coprime divisors of the triple, built from greatest common divisors of two elements or of an element and a root, divide the index combinations m − n m-n m − n , l + λ m l+\lambda m l + λm and l + λ n l+\lambda n l + λn ; through the gap inequality n > ( m − n ) R z n>(m-n)R_z n > ( m − n ) R z (Lemma 2.18, item 3) the first of them bounds m m m from below. Its one application, to the triples with two elements in a fixed ratio, is in Subsection 8.2. The whole argument lies in the proof of the theorem, which works with the matrix form of an index system (Lemma 2.8, item 2); the lemma before it is a two-line coprimality check.
Definition 6.1. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with roots r , s , t r,s,t r , s , t (Definition 2.3). Its gcd divisors are
Δ a b = gcd ( a , b ) gcd ( a , b , 4 ) , Δ b c = gcd ( b , c ) gcd ( b , c , 4 ) , Δ a c = gcd ( a , c ) gcd ( a , c , 4 ) , \Delta_{ab}=\frac{\gcd(a,b)}{\gcd(a,b,4)},\qquad\Delta_{bc}=\frac{\gcd(b,c)}{\gcd(b,c,4)},\qquad\Delta_{ac}=\frac{\gcd(a,c)}{\gcd(a,c,4)}, Δ ab = g cd( a , b , 4 ) g cd( a , b ) , Δ b c = g cd( b , c , 4 ) g cd( b , c ) , Δ a c = g cd( a , c , 4 ) g cd( a , c ) , its root divisors are
G r c = gcd ( r , c ) gcd ( r , c , 2 ) , G t a = gcd ( t , a ) gcd ( t , a , 2 ) , G s b = gcd ( s , b ) gcd ( s , b , 2 ) , G_{rc}=\frac{\gcd(r,c)}{\gcd(r,c,2)},\qquad G_{ta}=\frac{\gcd(t,a)}{\gcd(t,a,2)},\qquad G_{sb}=\frac{\gcd(s,b)}{\gcd(s,b,2)}, G r c = g cd( r , c , 2 ) g cd( r , c ) , G t a = g cd( t , a , 2 ) g cd( t , a ) , G s b = g cd( s , b , 2 ) g cd( s , b ) ,
and its coprime divisors are the products
Π 1 = Δ a b G r c , Π 2 = Δ b c G t a , Π 3 = Δ a c G s b . \Pi_{1}=\Delta_{ab}G_{rc},\qquad\Pi_{2}=\Delta_{bc}G_{ta},\qquad\Pi_{3}=\Delta_{ac}G_{sb}. Π 1 = Δ ab G r c , Π 2 = Δ b c G t a , Π 3 = Δ a c G s b .
The three index combinations of an index system ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) of an extension of { a , b , c } \{a,b,c\} { a , b , c } (Definition 2.5 ) are numbered
1 : m − n , 2 : l + λ m , 3 : l + λ n , 1:m-n,\qquad2:l+\lambda m,\qquad3:l+\lambda n, 1 : m − n , 2 : l + λm , 3 : l + λn ,
and objects attached to them carry the same number and are listed in this order. The pair of elements and the element–root pair attached to them are { a , b } \{a,b\} { a , b } and { r , c } \{r,c\} { r , c } for 1, { b , c } \{b,c\} { b , c } and { t , a } \{t,a\} { t , a } for 2, { a , c } \{a,c\} { a , c } and { s , b } \{s,b\} { s , b } for 3.
Each index combination receives two divisors, one from a pair of elements and one from an element and a root. The factors gcd ( ⋅ , 4 ) \gcd(\mathord{\cdot},4) g cd( ⋅ , 4 ) and gcd ( ⋅ , 2 ) \gcd(\mathord{\cdot},2) g cd( ⋅ , 2 ) are removed because the congruences in the proof of Theorem 6.3 hold for 4 and 2 times the combination, respectively. The two divisors attached to one combination are coprime by the next lemma, so that with each of them their product Π i \Pi_i Π i divides the combination.
Lemma 6.2. For every Euler triple { a , b , c } \{a,b,c\} { a , b , c } , the gcd divisors and the root divisors (Definition 6.1 ) satisfy
gcd ( Δ a b , G r c ) = gcd ( Δ b c , G t a ) = gcd ( Δ a c , G s b ) = 1. \gcd(\Delta_{ab},G_{rc})=\gcd(\Delta_{bc},G_{ta})=\gcd(\Delta_{ac},G_{sb})=1. g cd( Δ ab , G r c ) = g cd( Δ b c , G t a ) = g cd( Δ a c , G s b ) = 1.
Proof. By Definition 6.1 , Δ a b ∣ a \Delta_{ab}\mid a Δ ab ∣ a and G r c ∣ r G_{rc}\mid r G r c ∣ r , Δ b c ∣ b \Delta_{bc}\mid b Δ b c ∣ b and G t a ∣ t G_{ta}\mid t G t a ∣ t , Δ a c ∣ a \Delta_{ac}\mid a Δ a c ∣ a and G s b ∣ s G_{sb}\mid s G s b ∣ s . By Lemma 2.4 , item 3, gcd ( a , r ) = gcd ( b , t ) = gcd ( a , s ) = 1 \gcd(a,r)=\gcd(b,t)=\gcd(a,s)=1 g cd( a , r ) = g cd( b , t ) = g cd( a , s ) = 1 . Hence a common divisor of Δ a b \Delta_{ab} Δ ab and G r c G_{rc} G r c divides gcd ( a , r ) = 1 \gcd(a,r)=1 g cd( a , r ) = 1 , a common divisor of Δ b c \Delta_{bc} Δ b c and G t a G_{ta} G t a divides gcd ( b , t ) = 1 \gcd(b,t)=1 g cd( b , t ) = 1 , and a common divisor of Δ a c \Delta_{ac} Δ a c and G s b G_{sb} G s b divides gcd ( a , s ) = 1 \gcd(a,s)=1 g cd( a , s ) = 1 .
Theorem 6.3. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , with the coprime divisors Π 1 , Π 2 , Π 3 \Pi_1,\Pi_2,\Pi_3 Π 1 , Π 2 , Π 3 of Definition 6.1 and the gap ratio R z R_z R z of Definition 2.16 . Then
Π 1 ∣ m − n , Π 2 ∣ l + λ m , Π 3 ∣ l + λ n , and m > Π 1 ( 1 + R z ) . \Pi_1\mid m-n,\qquad\Pi_2\mid l+\lambda m,\qquad\Pi_3\mid l+\lambda n,\qquad\text{and}\qquad m>\Pi_1(1+R_z). Π 1 ∣ m − n , Π 2 ∣ l + λm , Π 3 ∣ l + λn , and m > Π 1 ( 1 + R z ) .
Proof. We reduce the three displays of Definition 2.5 modulo each of the six integers gcd ( a , b ) \gcd(a,b) g cd( a , b ) , gcd ( b , c ) \gcd(b,c) g cd( b , c ) , gcd ( a , c ) \gcd(a,c) g cd( a , c ) , gcd ( r , c ) \gcd(r,c) g cd( r , c ) , gcd ( t , a ) \gcd(t,a) g cd( t , a ) and gcd ( s , b ) \gcd(s,b) g cd( s , b ) : modulo each of them the squares of the multiplication matrices are unipotent or − I -I − I , so the displays become linear congruences in l , m , n l,m,n l , m , n , which give the three divisibilities; the gap inequality then gives the lower bound for m m m . Congruences modulo factors of c c c are used in [10 ] ; here the moduli are these six greatest common divisors. Throughout, c = a + b + 2 r c=a+b+2r c = a + b + 2 r , s = a + r s=a+r s = a + r , t = b + r t=b+r t = b + r and r 2 = a b + 1 r^2=ab+1 r 2 = ab + 1 , and s 2 = a c + 1 s^2=ac+1 s 2 = a c + 1 , t 2 = b c + 1 t^2=bc+1 t 2 = b c + 1 by Lemma 2.4 , item 1. Let A A A , B B B , C C C be the multiplication matrices of { a , b , c } \{a,b,c\} { a , b , c } (Definition 2.3 ), and let x , y , z x,y,z x , y , z be as in Definition 2.5 .
We first write the index system in matrix form: by item 2 of Lemma 2.8 ,
( y x ) = A 2 l ( 1 1 ) , ( z x ) = B 2 m ( λ 1 ) , ( z y ) = C 2 n ( λ 1 ) . \begin{pmatrix}y\\x\end{pmatrix}
=A^{2l}\begin{pmatrix}1\\1\end{pmatrix},
\qquad
\begin{pmatrix}z\\x\end{pmatrix}
=B^{2m}\begin{pmatrix}\lambda\\1\end{pmatrix},
\qquad
\begin{pmatrix}z\\y\end{pmatrix}
=C^{2n}\begin{pmatrix}\lambda\\1\end{pmatrix}. ( y x ) = A 2 l ( 1 1 ) , ( z x ) = B 2 m ( λ 1 ) , ( z y ) = C 2 n ( λ 1 ) .
We next record the squares of the multiplication matrices and the powers of special matrices modulo an integer. By r 2 = a b + 1 r^{2}=ab+1 r 2 = ab + 1 , s 2 = a c + 1 s^{2}=ac+1 s 2 = a c + 1 and t 2 = b c + 1 t^{2}=bc+1 t 2 = b c + 1 ,
A 2 = ( 2 a b + 1 2 r b 2 r a 2 a b + 1 ) , B 2 = ( 2 a c + 1 2 s c 2 s a 2 a c + 1 ) , C 2 = ( 2 b c + 1 2 t c 2 t b 2 b c + 1 ) . \mathbf{A}^{2}=\begin{pmatrix}2ab+1 & 2rb \\ 2ra & 2ab+1\end{pmatrix},\qquad\mathbf{B}^{2}=\begin{pmatrix}2ac+1 & 2sc \\ 2sa & 2ac+1\end{pmatrix},\qquad\mathbf{C}^{2}=\begin{pmatrix}2bc+1 & 2tc \\ 2tb & 2bc+1\end{pmatrix}. A 2 = ( 2 ab + 1 2 r a 2 r b 2 ab + 1 ) , B 2 = ( 2 a c + 1 2 s a 2 sc 2 a c + 1 ) , C 2 = ( 2 b c + 1 2 t b 2 t c 2 b c + 1 ) .
Modulo any integer, a matrix congruent to ( 1 u 0 1 ) \begin{pmatrix}1&u\\0&1\end{pmatrix} ( 1 0 u 1 ) , to ( 1 0 u 1 ) \begin{pmatrix}1&0\\u&1\end{pmatrix} ( 1 u 0 1 ) or to − I -\mathbf{I} − I has k k k -th power congruent to ( 1 k u 0 1 ) \begin{pmatrix}1&ku\\0&1\end{pmatrix} ( 1 0 k u 1 ) , to ( 1 0 k u 1 ) \begin{pmatrix}1&0\\ku&1\end{pmatrix} ( 1 k u 0 1 ) or to ( − 1 ) k I (-1)^{k}\mathbf{I} ( − 1 ) k I , for every k ≥ 0 k\geq0 k ≥ 0 . Modulo each of the six greatest common divisors above, the squares A 2 , B 2 , C 2 \mathbf{A}^{2},\mathbf{B}^{2},\mathbf{C}^{2} A 2 , B 2 , C 2 are of one of these three shapes, so the three matrix equations above become linear in l , m , n l,m,n l , m , n ; we compare a coordinate that two of them share. All congruences below are modulo the integer g g g named at the start of the paragraph.
Modulo g = gcd ( a , b ) g=\gcd(a,b) g = g cd( a , b ) : r 2 ≡ 1 r^{2}\equiv1 r 2 ≡ 1 , c ≡ 2 r c\equiv2r c ≡ 2 r and s ≡ t ≡ r s\equiv t\equiv r s ≡ t ≡ r , so B 2 ≡ ( 1 2 s c 0 1 ) \mathbf{B}^{2}\equiv\begin{pmatrix}1&2sc\\0&1\end{pmatrix} B 2 ≡ ( 1 0 2 sc 1 ) and C 2 ≡ ( 1 2 t c 0 1 ) \mathbf{C}^{2}\equiv\begin{pmatrix}1&2tc\\0&1\end{pmatrix} C 2 ≡ ( 1 0 2 t c 1 ) . The first coordinates of the second and third equations give z ≡ λ + 2 m s c ≡ λ + 2 n t c z\equiv\lambda+2msc\equiv\lambda+2ntc z ≡ λ + 2 m sc ≡ λ + 2 n t c , hence 0 ≡ 2 r c ( m − n ) ≡ 4 r 2 ( m − n ) ≡ 4 ( m − n ) 0\equiv2rc(m-n)\equiv4r^{2}(m-n)\equiv4(m-n) 0 ≡ 2 r c ( m − n ) ≡ 4 r 2 ( m − n ) ≡ 4 ( m − n ) .
Modulo g = gcd ( b , c ) g=\gcd(b,c) g = g cd( b , c ) : r 2 ≡ 1 r^{2}\equiv1 r 2 ≡ 1 , a ≡ − 2 r a\equiv-2r a ≡ − 2 r and s ≡ − r s\equiv-r s ≡ − r , so A 2 ≡ ( 1 0 2 r a 1 ) \mathbf{A}^{2}\equiv\begin{pmatrix}1&0\\2ra&1\end{pmatrix} A 2 ≡ ( 1 2 r a 0 1 ) and B 2 ≡ ( 1 0 2 s a 1 ) \mathbf{B}^{2}\equiv\begin{pmatrix}1&0\\2sa&1\end{pmatrix} B 2 ≡ ( 1 2 s a 0 1 ) . The second coordinates of the first two equations give x ≡ 1 + 2 l r a ≡ 1 + 2 λ m s a x\equiv1+2lra\equiv1+2\lambda msa x ≡ 1 + 2 l r a ≡ 1 + 2 λm s a , hence 0 ≡ 2 a ( l r − λ m s ) ≡ 2 a r ( l + λ m ) ≡ − 4 r 2 ( l + λ m ) ≡ − 4 ( l + λ m ) 0\equiv2a(lr-\lambda ms)\equiv2ar(l+\lambda m)\equiv-4r^{2}(l+\lambda m)\equiv-4(l+\lambda m) 0 ≡ 2 a ( l r − λm s ) ≡ 2 a r ( l + λm ) ≡ − 4 r 2 ( l + λm ) ≡ − 4 ( l + λm ) .
Modulo g = gcd ( a , c ) g=\gcd(a,c) g = g cd( a , c ) : r 2 ≡ 1 r^{2}\equiv1 r 2 ≡ 1 , b ≡ − 2 r b\equiv-2r b ≡ − 2 r and t ≡ − r t\equiv-r t ≡ − r , so A 2 ≡ ( 1 2 r b 0 1 ) \mathbf{A}^{2}\equiv\begin{pmatrix}1&2rb\\0&1\end{pmatrix} A 2 ≡ ( 1 0 2 r b 1 ) and C 2 ≡ ( 1 0 2 t b 1 ) \mathbf{C}^{2}\equiv\begin{pmatrix}1&0\\2tb&1\end{pmatrix} C 2 ≡ ( 1 2 t b 0 1 ) . The coordinate y y y of the first and third equations gives y ≡ 1 + 2 l r b ≡ 1 + 2 λ n t b y\equiv1+2lrb\equiv1+2\lambda ntb y ≡ 1 + 2 l r b ≡ 1 + 2 λn t b , hence 0 ≡ 2 b ( l r − λ n t ) ≡ 2 b r ( l + λ n ) ≡ − 4 r 2 ( l + λ n ) ≡ − 4 ( l + λ n ) 0\equiv2b(lr-\lambda nt)\equiv2br(l+\lambda n)\equiv-4r^{2}(l+\lambda n)\equiv-4(l+\lambda n) 0 ≡ 2 b ( l r − λn t ) ≡ 2 b r ( l + λn ) ≡ − 4 r 2 ( l + λn ) ≡ − 4 ( l + λn ) .
Modulo g = gcd ( r , c ) g=\gcd(r,c) g = g cd( r , c ) : b ≡ − a b\equiv-a b ≡ − a and a b ≡ − 1 ab\equiv-1 ab ≡ − 1 , so a 2 ≡ 1 a^{2}\equiv1 a 2 ≡ 1 , s ≡ a s\equiv a s ≡ a and t ≡ − a t\equiv-a t ≡ − a . Then A 2 ≡ − I \mathbf{A}^{2}\equiv-\mathbf{I} A 2 ≡ − I , so x ≡ y ≡ ( − 1 ) l x\equiv y\equiv(-1)^{l} x ≡ y ≡ ( − 1 ) l , while B 2 = C 2 ≡ ( 1 0 2 1 ) \mathbf{B}^{2}=\mathbf{C}^{2}\equiv\begin{pmatrix}1&0\\2&1\end{pmatrix} B 2 = C 2 ≡ ( 1 2 0 1 ) , so x ≡ 1 + 2 λ m x\equiv1+2\lambda m x ≡ 1 + 2 λm and y ≡ 1 + 2 λ n y\equiv1+2\lambda n y ≡ 1 + 2 λn . Hence 2 λ m ≡ 2 λ n 2\lambda m\equiv2\lambda n 2 λm ≡ 2 λn , that is 2 ( m − n ) ≡ 0 2(m-n)\equiv0 2 ( m − n ) ≡ 0 , since λ = ± 1 \lambda=\pm1 λ = ± 1 .
Modulo g = gcd ( t , a ) g=\gcd(t,a) g = g cd( t , a ) : r ≡ − b r\equiv-b r ≡ − b , b 2 ≡ r 2 ≡ 1 b^{2}\equiv r^{2}\equiv1 b 2 ≡ r 2 ≡ 1 , c ≡ − b c\equiv-b c ≡ − b and s ≡ − b s\equiv-b s ≡ − b . Then C 2 ≡ − I \mathbf{C}^{2}\equiv-\mathbf{I} C 2 ≡ − I , so z ≡ λ y z\equiv\lambda y z ≡ λ y ; A 2 ≡ ( 1 − 2 0 1 ) \mathbf{A}^{2}\equiv\begin{pmatrix}1&-2\\0&1\end{pmatrix} A 2 ≡ ( 1 0 − 2 1 ) gives y ≡ 1 − 2 l y\equiv1-2l y ≡ 1 − 2 l ; and B 2 ≡ ( 1 2 0 1 ) \mathbf{B}^{2}\equiv\begin{pmatrix}1&2\\0&1\end{pmatrix} B 2 ≡ ( 1 0 2 1 ) gives z ≡ λ + 2 m z\equiv\lambda+2m z ≡ λ + 2 m . Hence λ + 2 m ≡ λ − 2 λ l \lambda+2m\equiv\lambda-2\lambda l λ + 2 m ≡ λ − 2 λ l , that is 2 ( m + λ l ) ≡ 0 2(m+\lambda l)\equiv0 2 ( m + λ l ) ≡ 0 , and multiplying by λ \lambda λ , 2 ( l + λ m ) ≡ 0 2(l+\lambda m)\equiv0 2 ( l + λm ) ≡ 0 .
Modulo g = gcd ( s , b ) g=\gcd(s,b) g = g cd( s , b ) : r ≡ − a r\equiv-a r ≡ − a , a 2 ≡ r 2 ≡ 1 a^{2}\equiv r^{2}\equiv1 a 2 ≡ r 2 ≡ 1 , c ≡ − a c\equiv-a c ≡ − a and t ≡ − a t\equiv-a t ≡ − a . Then B 2 ≡ − I \mathbf{B}^{2}\equiv-\mathbf{I} B 2 ≡ − I , so z ≡ λ x z\equiv\lambda x z ≡ λ x ; A 2 ≡ ( 1 0 − 2 1 ) \mathbf{A}^{2}\equiv\begin{pmatrix}1&0\\-2&1\end{pmatrix} A 2 ≡ ( 1 − 2 0 1 ) gives x ≡ 1 − 2 l x\equiv1-2l x ≡ 1 − 2 l ; and C 2 ≡ ( 1 2 0 1 ) \mathbf{C}^{2}\equiv\begin{pmatrix}1&2\\0&1\end{pmatrix} C 2 ≡ ( 1 0 2 1 ) gives z ≡ λ + 2 n z\equiv\lambda+2n z ≡ λ + 2 n . Hence λ + 2 n ≡ λ − 2 λ l \lambda+2n\equiv\lambda-2\lambda l λ + 2 n ≡ λ − 2 λ l , that is 2 ( n + λ l ) ≡ 0 2(n+\lambda l)\equiv0 2 ( n + λ l ) ≡ 0 , and multiplying by λ \lambda λ , 2 ( l + λ n ) ≡ 0 2(l+\lambda n)\equiv0 2 ( l + λn ) ≡ 0 .
We next pass to the divisors. The six congruences just proved say that
gcd ( a , b ) ∣ 4 ( m − n ) , gcd ( b , c ) ∣ 4 ( l + λ m ) , gcd ( a , c ) ∣ 4 ( l + λ n ) , \gcd(a,b)\mid4(m-n),\qquad\gcd(b,c)\mid4(l+\lambda m),\qquad\gcd(a,c)\mid4(l+\lambda n), g cd( a , b ) ∣ 4 ( m − n ) , g cd( b , c ) ∣ 4 ( l + λm ) , g cd( a , c ) ∣ 4 ( l + λn ) ,
gcd ( r , c ) ∣ 2 ( m − n ) , gcd ( t , a ) ∣ 2 ( l + λ m ) , gcd ( s , b ) ∣ 2 ( l + λ n ) . \gcd(r,c)\mid2(m-n),\qquad\gcd(t,a)\mid2(l+\lambda m),\qquad\gcd(s,b)\mid2(l+\lambda n). g cd( r , c ) ∣ 2 ( m − n ) , g cd( t , a ) ∣ 2 ( l + λm ) , g cd( s , b ) ∣ 2 ( l + λn ) .
If g ∣ 4 k g\mid4k g ∣ 4 k for integers g ≥ 1 g\geq1 g ≥ 1 and k k k , then g gcd ( g , 4 ) ∣ k \frac{g}{\gcd(g,4)}\mid k g c d ( g , 4 ) g ∣ k , and if g ∣ 2 k g\mid2k g ∣ 2 k , then g gcd ( g , 2 ) ∣ k \frac{g}{\gcd(g,2)}\mid k g c d ( g , 2 ) g ∣ k . Since gcd ( a , b , 4 ) = gcd ( gcd ( a , b ) , 4 ) \gcd(a,b,4)=\gcd(\gcd(a,b),4) g cd( a , b , 4 ) = g cd( g cd( a , b ) , 4 ) and gcd ( r , c , 2 ) = gcd ( gcd ( r , c ) , 2 ) \gcd(r,c,2)=\gcd(\gcd(r,c),2) g cd( r , c , 2 ) = g cd( g cd( r , c ) , 2 ) , and likewise for the other four, it follows by Definition 6.1 that Δ a b \Delta_{ab} Δ ab and G r c G_{rc} G r c divide m − n m-n m − n , Δ b c \Delta_{bc} Δ b c and G t a G_{ta} G t a divide l + λ m l+\lambda m l + λm , and Δ a c \Delta_{ac} Δ a c and G s b G_{sb} G s b divide l + λ n l+\lambda n l + λn . The two divisors in each pair are coprime by Lemma 6.2, so their products, the coprime divisors, satisfy
Π 1 ∣ m − n , Π 2 ∣ l + λ m , Π 3 ∣ l + λ n . \Pi_{1}\mid m-n,\qquad\Pi_{2}\mid l+\lambda m,\qquad\Pi_{3}\mid l+\lambda n. Π 1 ∣ m − n , Π 2 ∣ l + λm , Π 3 ∣ l + λn .
We finally prove m > Π 1 ( 1 + R z ) m > \Pi_1(1 + R_z) m > Π 1 ( 1 + R z ) . By Lemma 2.14, $3 \le n < m < l < 2n$. In particular m − n ≥ 1 m-n \ge1 m − n ≥ 1 , and as Π 1 ∣ m − n \Pi_1 \mid m-n Π 1 ∣ m − n , m − n ≥ Π 1 m-n \ge\Pi_1 m − n ≥ Π 1 . By the gap inequality of Lemma 2.18 , item 3, n > ( m − n ) R z n > (m-n)R_z n > ( m − n ) R z , where R z = a c b − a log β > 0 R_z = \frac{\sqrt{ac}}{b-a}\log\beta> 0 R z = b − a a c log β > 0 is the gap ratio of Definition 2.16 . Hence
m = n + ( m − n ) > ( m − n ) ( 1 + R z ) ≥ Π 1 ( 1 + R z ) . m = n + (m-n) > (m-n)(1+R_z) \ge\Pi_1(1+R_z). m = n + ( m − n ) > ( m − n ) ( 1 + R z ) ≥ Π 1 ( 1 + R z ) .
An effective bound for irregular extensions
This section proves the effective bound (Theorem 7.9 ), c < 4 ⋅ 10 48 c < 4 \cdot10^{48} c < 4 ⋅ 1 0 48 . Its ceilings for the index come from lower bounds for linear forms in three logarithms, applied to the form Λ x = 2 l log α − 2 m log β + log D λ \Lambda_x = 2l\log\alpha- 2m\log\beta+ \log D_\lambda Λ x = 2 l log α − 2 m log β + log D λ of Definition 2.16 , which is positive and smaller than 4.01 β − 4 m 4.01\beta^{-4m} 4.01 β − 4 m (Lemma 2.18 , item 4); the sizes and heights of α \alpha α , β \beta β and D λ D_\lambda D λ are those of Lemma 4.2 . These bounds enter at two levels of precision. In Subsection 7.1, Matveev’s theorem [23 ] gives the a priori bound for the index (Proposition 7.3 ), valid for every irregular extension, and against the quartic index floor (Theorem 3.6 ) it gives c < 10 75 c < 10^{75} c < 1 0 75 (Proposition 7.5 ). In Subsection 7.2, the theorem of Mignotte and Voutier [24 ] gives caps for m m m , from m < 7 ⋅ 10 12 m < 7 \cdot10^{12} m < 7 ⋅ 1 0 12 down to m < 10 12 m < 10^{12} m < 1 0 12 , on four windows of c c c (Proposition 7.8). This proposition is the heart of the section; its main step excludes an exceptional linear relation between the coefficients of Λ x \Lambda_x Λ x , with Laurent’s theorem (Theorem 4.3 ) from Section 4. The effective bound then follows by feeding each cap back into the quartic floor.
The a priori bound
Convention 7.1. The a priori function is defined, for real c > 1 c > 1 c > 1 and u ≥ 1 u \ge1 u ≥ 1 , by
Φ c ( u ) = u − 0.15 − K E ( log c ) 2 log ( 19.46 u ) , K E = 2.4638 ⋅ 10 12 . \Phi_c(u) = u - 0.15 - K_E(\log c)^2\log(19.46u), \qquad K_E = 2.4638 \cdot10^{12}. Φ c ( u ) = u − 0.15 − K E ( log c ) 2 log ( 19.46 u ) , K E = 2.4638 ⋅ 1 0 12 .
Its constants K E K_E K E , 0.15 and 19.46 are those of the a priori bound for the index, obtained from Matveev’s theorem applied to the linear form Λ x \Lambda_x Λ x (Definition 2.16 ).
The function Φ c \Phi_c Φ c is the a priori bound in the form in which it is used: a single value u 0 ≥ 1 u_0 \ge1 u 0 ≥ 1 with Φ c ( u 0 ) ≥ 0 \Phi_c(u_0) \ge0 Φ c ( u 0 ) ≥ 0 caps the index by 2 m < u 0 2m < u_0 2 m < u 0 (Proposition 7.3 ). It is used in this form in this section, in Section 8 and in Appendix B.
Theorem 7.2. Let L \mathcal{L} L be a real number field of degree D D D , let η 1 , η 2 , η 3 ∈ L \eta_1, \eta_2, \eta_3 \in\mathcal{L} η 1 , η 2 , η 3 ∈ L be positive and multiplicatively independent, and let b 1 , b 2 , b 3 ∈ Z b_1, b_2, b_3 \in\mathbb{Z} b 1 , b 2 , b 3 ∈ Z with b 3 ≠ 0 b_3 \ne0 b 3 = 0 and
Λ M = b 1 log η 1 + b 2 log η 2 + b 3 log η 3 . \Lambda_M = b_1\log\eta_1 + b_2\log\eta_2 + b_3\log\eta_3. Λ M = b 1 log η 1 + b 2 log η 2 + b 3 log η 3 . Let A 1 , A 2 , A 3 A_1, A_2, A_3 A 1 , A 2 , A 3 be real numbers with A i ≥ max { D h ( η i ) , ∣ log η i ∣ , 0.16 } A_i \ge\max\{D h(\eta_i), \lvert\log\eta_i\rvert, 0.16\} A i ≥ max { D h ( η i ) , ∣ log η i ∣ , 0.16 } (Definition 4.1), and put B M = max { 1 , ∣ b 1 ∣ A 1 / A 3 , ∣ b 2 ∣ A 2 / A 3 , ∣ b 3 ∣ } B_M = \max\{1, \lvert b_1\rvert A_1/A_3, \lvert b_2\rvert A_2/A_3, \lvert b_3\rvert\} B M = max { 1 , ∣ b 1 ∣ A 1 / A 3 , ∣ b 2 ∣ A 2 / A 3 , ∣ b 3 ∣} . Then
log ∣ Λ M ∣ > − C M ( 3 ) C 0 W 0 D 2 A 1 A 2 A 3 , \log\lvert\Lambda_M\rvert> -C_M(3) C_0 W_0 D^2 A_1A_2A_3, log ∣ Λ M ∣ > − C M ( 3 ) C 0 W 0 D 2 A 1 A 2 A 3 , where C M ( 3 ) = 16 3 ! e 3 ⋅ 9 ⋅ 5 ⋅ 16 4 ⋅ 3 e 2 < 6.441 ⋅ 10 8 C_M(3) = \frac{16}{3!}e^3 \cdot9 \cdot5 \cdot16^4 \cdot\frac{3e}{2} < 6.441 \cdot10^8 C M ( 3 ) = 3 ! 16 e 3 ⋅ 9 ⋅ 5 ⋅ 1 6 4 ⋅ 2 3 e < 6.441 ⋅ 1 0 8 , C 0 = log ( e 20.235.5 D 2 log ( e D ) ) C_0 = \log(e^{20.235.5D^2\log(eD)}) C 0 = log ( e 20.235.5 D 2 l o g ( eD ) ) and W 0 = log ( 1.5 e B M D log ( e D ) ) W_0 = \log(1.5e B_M D\log(eD)) W 0 = log ( 1.5 e B M D log ( eD )) .
Proof. This is Matveev’s theorem [23 ] for three logarithms in a real field; the condition A i ≥ 0.16 A_i \ge0.16 A i ≥ 0.16 only restricts the admissible choices of A i A_i A i . With A j = max { D h ( η j ) , ∣ log η j ∣ } A_j = \max\{D h(\eta_j), \lvert\log\eta_j\rvert\} A j = max { D h ( η j ) , ∣ log η j ∣} it is stated in [18 Proposition 6.2] , with the constant 11796480 e 4 D 2 log ( 3 5.5 e 20.235 D 2 log ( e D ) ) 11796480 e^4D^2\log(3^{5.5}e^{20.235}D^2\log(eD)) 11796480 e 4 D 2 log ( 3 5.5 e 20.235 D 2 log ( eD )) , which is C M ( 3 ) C 0 D 2 C_M(3)C_0D^2 C M ( 3 ) C 0 D 2 because 11796480 = 16 6 ⋅ 9 ⋅ 5 ⋅ 16 4 ⋅ 3 2 11796480 = \frac{16}{6} \cdot9 \cdot5 \cdot16^4 \cdot\frac{3}{2} 11796480 = 6 16 ⋅ 9 ⋅ 5 ⋅ 1 6 4 ⋅ 2 3 .
Proposition 7.3. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , and let Φ c \Phi_c Φ c be as in Convention 7.1. Then Φ c ( 2 m ) < 0 \Phi_c(2m)<0 Φ c ( 2 m ) < 0 ; consequently 2 m < u 0 2m<u_0 2 m < u 0 for every real u 0 ≥ 1 u_0\ge1 u 0 ≥ 1 with Φ c ( u 0 ) ≥ 0 \Phi_c(u_0)\ge0 Φ c ( u 0 ) ≥ 0 .
Proof. We apply Matveev’s theorem (Theorem 7.2) to the linear form Λ x \Lambda_x Λ x of Definition 2.16 .
We first collect the inputs. By Lemma 2.14, m ≥ 4 m\ge4 m ≥ 4 and c > b > 4000 c>b>4000 c > b > 4000 . By Lemma 2.18 , l log α < m log β l\log\alpha<m\log\beta l log α < m log β (item 1) and 0 < Λ x < 4.01 β − 4 m 0<\Lambda_x<4.01\beta^{-4m} 0 < Λ x < 4.01 β − 4 m (item 4). By Lemma 4.2 : log α < log c \log\alpha<\log c log α < log c and 1 2 log c + log 2 < log β < log c \frac{1}{2}\log c+\log2<\log\beta<\log c 2 1 log c + log 2 < log β < log c (item 1); K \mathcal{K} K is a real biquadratic field (item 2); h ( α ) = 1 2 log α h(\alpha)=\frac{1}{2}\log\alpha h ( α ) = 2 1 log α and h ( β ) = 1 2 log β h(\beta)=\frac{1}{2}\log\beta h ( β ) = 2 1 log β (item 3); 0 < log D λ < log 4 0<\log D_\lambda<\log4 0 < log D λ < log 4 and h ( D λ ) < log c h(D_\lambda)<\log c h ( D λ ) < log c (item 4); and α \alpha α , β / α \beta/\alpha β / α , D λ D_\lambda D λ exceed 1 and are multiplicatively independent (item 6).
We next choose the parameters of Theorem 7.2 . Take L = K \mathcal{L}=\mathcal{K} L = K , which contains α \alpha α , β \beta β and D λ D_\lambda D λ and has degree D = 4 D=4 D = 4 ; take η 1 = α \eta_1=\alpha η 1 = α , η 2 = β \eta_2=\beta η 2 = β , η 3 = D λ \eta_3=D_\lambda η 3 = D λ and b 1 = 2 l b_1=2l b 1 = 2 l , b 2 = − 2 m b_2=-2m b 2 = − 2 m , b 3 = 1 b_3=1 b 3 = 1 , so that Λ M = Λ x \Lambda_M=\Lambda_x Λ M = Λ x . The η i \eta_i η i are positive, and multiplicatively independent because they generate the same group as α , β / α , D λ \alpha,\beta/\alpha,D_\lambda α , β / α , D λ ; b 3 ≠ 0 b_3\ne0 b 3 = 0 . We may take
A 1 = 2 log α , A 2 = 2 log β , A 3 = 4 log c . A_1=2\log\alpha,\qquad A_2=2\log\beta,\qquad A_3=4\log c. A 1 = 2 log α , A 2 = 2 log β , A 3 = 4 log c .
since 4 h ( α ) = 2 log α 4h(\alpha)=2\log\alpha 4 h ( α ) = 2 log α , 4 h ( β ) = 2 log β 4h(\beta)=2\log\beta 4 h ( β ) = 2 log β , 4 h ( D λ ) < 4 log c 4h(D_\lambda)<4\log c 4 h ( D λ ) < 4 log c , log D λ < log 4 < 4 log c \log D_\lambda<\log4<4\log c log D λ < log 4 < 4 log c , and all three exceed 0.16 because log c > log 4000 \log c>\log4000 log c > log 4000 and 2 log α > log c 2\log\alpha>\log c 2 log α > log c (item 1 of Lemma 4.2 ). As l log α < m log β l\log\alpha<m\log\beta l log α < m log β , log β < log c \log\beta<\log c log β < log c and m ≥ 4 m\ge4 m ≥ 4 ,
B M = max { 1 , l log α log c , m log β log c } ≤ m , B_M=\max\left\{1,\frac{l\log\alpha}{\log c},\frac{m\log\beta}{\log c}\right\}\le m, B M = max { 1 , log c l log α , log c m log β } ≤ m ,
so W 0 = log ( 1.5 e B M D log ( e D ) ) ≤ log ( 0.75 e D log ( e D ) ⋅ 2 m ) < log ( 19.46 ⋅ 2 m ) W_0=\log(1.5eB_MD\log(eD))\le\log(0.75eD\log(eD)\cdot2m)<\log(19.46\cdot2m) W 0 = log ( 1.5 e B M D log ( eD )) ≤ log ( 0.75 eD log ( eD ) ⋅ 2 m ) < log ( 19.46 ⋅ 2 m ) .
We now prove that Φ c ( 2 m ) < 0 \Phi_c(2m)<0 Φ c ( 2 m ) < 0 . Put C ^ = C M ( 3 ) C 0 D 2 \widehat C=C_M(3)C_0D^2 C = C M ( 3 ) C 0 D 2 . Theorem 7.2 gives log Λ x > − C ^ W 0 A 1 A 2 A 3 = − 16 C ^ W 0 log α log β log c \log\Lambda_x>-\widehat C W_0A_1A_2A_3=-16\widehat C W_0\log\alpha\log\beta\log c log Λ x > − C W 0 A 1 A 2 A 3 = − 16 C W 0 log α log β log c , and Λ x < 4.01 β − 4 m \Lambda_x<4.01\beta^{-4m} Λ x < 4.01 β − 4 m gives log Λ x < log 4.01 − 4 m log β \log\Lambda_x<\log4.01-4m\log\beta log Λ x < log 4.01 − 4 m log β . Hence 4 m log β < log 4.01 + 16 C ^ W 0 log α log β log c 4m\log\beta<\log4.01+16\widehat C W_0\log\alpha\log\beta\log c 4 m log β < log 4.01 + 16 C W 0 log α log β log c ; dividing by 2 log β 2\log\beta 2 log β and using log α < log c \log\alpha<\log c log α < log c and 2 log β > log 4000 + 2 log 2 2\log\beta>\log4000+2\log2 2 log β > log 4000 + 2 log 2 ,
2 m < log 4.01 log 4000 + 2 log 2 + 8 C ^ ( log c ) 2 log ( 19.46 ⋅ 2 m ) < 0.15 + K E ( log c ) 2 log ( 19.46 ⋅ 2 m ) , 2m<\frac{\log4.01}{\log4000+2\log2}+8\widehat C(\log c)^2\log(19.46\cdot2m)<0.15+K_E(\log c)^2\log(19.46\cdot2m), 2 m < log 4000 + 2 log 2 log 4.01 + 8 C ( log c ) 2 log ( 19.46 ⋅ 2 m ) < 0.15 + K E ( log c ) 2 log ( 19.46 ⋅ 2 m ) ,
that is, Φ c ( 2 m ) < 0 \Phi_c(2m)<0 Φ c ( 2 m ) < 0 . The constants of Φ c \Phi_c Φ c enter through three numerical bounds, 8 C ^ < K E 8\widehat C<K_E 8 C < K E , log 4.01 / ( log 4000 + 2 log 2 ) < 0.15 \log4.01/(\log4000+2\log2)<0.15 log 4.01/ ( log 4000 + 2 log 2 ) < 0.15 and 0.75 e D log ( e D ) < 19.46 0.75eD\log(eD)<19.46 0.75 eD log ( eD ) < 19.46 , which were verified in interval arithmetic with outward rounding.
For the second assertion, let u 0 ≥ 1 u_0\ge1 u 0 ≥ 1 with Φ c ( u 0 ) ≥ 0 \Phi_c(u_0)\ge0 Φ c ( u 0 ) ≥ 0 . Since log ( 19.46 u 0 ) > 1 \log(19.46u_0)>1 log ( 19.46 u 0 ) > 1 , u 0 > K E ( log c ) 2 log ( 19.46 u 0 ) > K E ( log c ) 2 u_0>K_E(\log c)^2\log(19.46u_0)>K_E(\log c)^2 u 0 > K E ( log c ) 2 log ( 19.46 u 0 ) > K E ( log c ) 2 , and Φ c ′ ( u ) = 1 − K E ( log c ) 2 / u > 0 \Phi_c'(u)=1-K_E(\log c)^2/u>0 Φ c ′ ( u ) = 1 − K E ( log c ) 2 / u > 0 for u ≥ u 0 u\ge u_0 u ≥ u 0 ; so Φ c \Phi_c Φ c is strictly increasing on [ u 0 , ∞ ) [u_0,\infty) [ u 0 , ∞ ) . If 2 m ≥ u 0 2m\ge u_0 2 m ≥ u 0 , then Φ c ( 2 m ) ≥ Φ c ( u 0 ) ≥ 0 \Phi_c(2m)\ge\Phi_c(u_0)\ge0 Φ c ( 2 m ) ≥ Φ c ( u 0 ) ≥ 0 , contradicting the first assertion. Hence 2 m < u 0 2m<u_0 2 m < u 0 . □ \square □
To use the cap one needs a point at which Φ c \Phi_c Φ c is positive for all large c c c at once. The next lemma gives such points along curves u = ι c θ u=\iota c^\theta u = ι c θ , from a check at a single value c 0 c_0 c 0 . It is developed once and used twice: with ( ι , θ , c 0 ) = ( 2 1 / 4 , 1 4 , 10 75 ) (\iota,\theta,c_0)=(2^{1/4},\frac14,10^{75}) ( ι , θ , c 0 ) = ( 2 1/4 , 4 1 , 1 0 75 ) , against the quartic floor, in Proposition 7.5 ; and with ( ι , θ , c 0 ) = ( 1 / 1806 , 1 , 10 21 ) (\iota,\theta,c_0)=(1/1806,1,10^{21}) ( ι , θ , c 0 ) = ( 1/1806 , 1 , 1 0 21 ) , against the floor from coprime divisors, which is linear in c c c , in Theorem 8.7 .
Lemma 7.4. Let Φ c \Phi_c Φ c and K E K_E K E be as in Convention 7.1. Let ι > 0 \iota>0 ι > 0 , θ > 0 \theta>0 θ > 0 and c 0 > 1 c_0>1 c 0 > 1 satisfy ι c 0 θ ≥ 1 \iota c_0^\theta\ge1 ι c 0 θ ≥ 1 , θ log c 0 ≥ 2 + ∣ log ( 19.46 ι ) ∣ \theta\log c_0\ge2+\lvert\log(19.46\iota)\rvert θ log c 0 ≥ 2 + ∣ log ( 19.46 ι )∣ ,
Φ c 0 ( ι c 0 θ ) > 0 and θ ι c 0 θ > K E log c 0 ( 2 log ( 19.46 ι c 0 θ ) + θ log c 0 ) . \Phi_{c_0}(\iota c_0^\theta)>0\qquad\text{and}\qquad\theta\iota c_0^\theta>K_E\log c_0\left(2\log(19.46\iota c_0^\theta)+\theta\log c_0\right). Φ c 0 ( ι c 0 θ ) > 0 and θ ι c 0 θ > K E log c 0 ( 2 log ( 19.46 ι c 0 θ ) + θ log c 0 ) . Then Φ c ( ι c θ ) > 0 \Phi_c(\iota c^\theta)>0 Φ c ( ι c θ ) > 0 for every real c ≥ c 0 c\ge c_0 c ≥ c 0 .
Proof. Put ℓ = log ( 19.46 ι ) \ell=\log(19.46\iota) ℓ = log ( 19.46 ι ) and w = θ log c w=\theta\log c w = θ log c , so that c θ = e w ≥ 1 c^{\theta}=e^w\geq1 c θ = e w ≥ 1 for c ≥ c 0 c\geq c_0 c ≥ c 0 and log ( 19.46 c θ ) = ℓ + w \log(19.46c^\theta)=\ell+w log ( 19.46 c θ ) = ℓ + w . Then Φ c ( c θ ) = f ( w ) \Phi_c(c^\theta)=f(w) Φ c ( c θ ) = f ( w ) , where
f ( w ) = ι e w − P ( w ) , P ( w ) = 0.15 + K E θ 2 w 2 ( ℓ + w ) . f(w)=\iota e^w-P(w),\qquad P(w)=0.15+\frac{K_E}{\theta^2}w^2(\ell+w). f ( w ) = ι e w − P ( w ) , P ( w ) = 0.15 + θ 2 K E w 2 ( ℓ + w ) .
Put w 0 = θ log c 0 w_0=\theta\log c_0 w 0 = θ log c 0 . It suffices to show that f ( w ) > 0 f(w)>0 f ( w ) > 0 for w ≥ w 0 w\geq w_0 w ≥ w 0 . We have
P ′ ( w ) = K E θ 2 ( 3 w 2 + 2 ℓ w ) , P ′ ′ ( w ) = K E θ 2 ( 6 w + 2 ℓ ) , P'(w)=\frac{K_E}{\theta^2}(3w^2+2\ell w),\qquad P''(w)=\frac{K_E}{\theta^2}(6w+2\ell), P ′ ( w ) = θ 2 K E ( 3 w 2 + 2 ℓ w ) , P ′′ ( w ) = θ 2 K E ( 6 w + 2 ℓ ) ,
θ 2 K E ( P ′ ( w ) − P ′ ′ ( w ) ) = 3 w ( w − 2 ) + 2 ℓ ( w − 1 ) . \frac{\theta^2}{K_E}\left(P'(w)-P''(w)\right)=3w(w-2)+2\ell(w-1). K E θ 2 ( P ′ ( w ) − P ′′ ( w ) ) = 3 w ( w − 2 ) + 2 ℓ ( w − 1 ) .
For w ≥ w 0 ≥ 2 + ∣ ℓ ∣ w\geq w_0\geq2+|\ell| w ≥ w 0 ≥ 2 + ∣ ℓ ∣ the last quantity is nonnegative: if ℓ ≥ 0 \ell\geq0 ℓ ≥ 0 both terms are nonnegative, and if ℓ < 0 \ell<0 ℓ < 0 then 3 w ( w − 2 ) ≥ 3 w ∣ ℓ ∣ ≥ 2 ∣ ℓ ∣ ( w − 1 ) 3w(w-2)\geq3w|\ell|\geq2|\ell|(w-1) 3 w ( w − 2 ) ≥ 3 w ∣ ℓ ∣ ≥ 2∣ ℓ ∣ ( w − 1 ) . Hence P ′ ′ ≤ P ′ P''\leq P' P ′′ ≤ P ′ on [ w 0 , ∞ ) [w_0,\infty) [ w 0 , ∞ ) .
Let g = f ′ = ι e w − P ′ g=f'=\iota e^w-P' g = f ′ = ι e w − P ′ . On [ w 0 , ∞ ) [w_0,\infty) [ w 0 , ∞ ) we have g ′ = ι e w − P ′ ′ ≥ g g'=\iota e^w-P''\geq g g ′ = ι e w − P ′′ ≥ g , so ( e − w g ) ′ ≥ 0 (e^{-w}g)'\geq0 ( e − w g ) ′ ≥ 0 and g ( w ) ≥ g ( w 0 ) e w − w 0 g(w)\geq g(w_0)e^{w-w_0} g ( w ) ≥ g ( w 0 ) e w − w 0 . Moreover
θ g ( w 0 ) = θ ι c 0 θ − K E log c 0 ( 3 w 0 + 2 ℓ ) = θ ι c 0 θ − K E log c 0 ( 2 log ( 19.46 c 0 θ ) + θ log c 0 ) > 0 \theta g(w_0)=\theta\iota c_0^\theta-K_E\log c_0(3w_0+2\ell)=\theta\iota c_0^\theta-K_E\log c_0\left(2\log(19.46c_0^\theta)+\theta\log c_0\right)>0 θ g ( w 0 ) = θ ι c 0 θ − K E log c 0 ( 3 w 0 + 2 ℓ ) = θ ι c 0 θ − K E log c 0 ( 2 log ( 19.46 c 0 θ ) + θ log c 0 ) > 0
by hypothesis. Hence g > 0 g>0 g > 0 on [ w 0 , ∞ ) [w_0,\infty) [ w 0 , ∞ ) , so f f f is increasing there, and f ( w ) ≥ f ( w 0 ) = Φ c 0 ( c 0 θ ) > 0 f(w)\geq f(w_0)=\Phi_{c_0}(c_0^\theta)>0 f ( w ) ≥ f ( w 0 ) = Φ c 0 ( c 0 θ ) > 0 for w ≥ w 0 w\geq w_0 w ≥ w 0 . □ \square □
Proposition 7.5. Let d d d be an irregular extension of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } . Then c < 10 75 c<10^{75} c < 1 0 75 .
Proof. We compare the quartic index floor with the a priori bound along the curve u = 2 1 / 4 c 1 / 4 u=2^{1/4}c^{1/4} u = 2 1/4 c 1/4 . Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of d d d (Proposition 2.11), let Φ c \Phi_c Φ c and K E K_E K E be as in Convention 7.1, and suppose that c ≥ 10 75 c\geq10^{75} c ≥ 1 0 75 .
By Theorem 3.6 , c < 8 m 4 c<8m^4 c < 8 m 4 , hence
2 m > 2 ( c 8 ) 1 / 4 = ι c 1 / 4 , ι = 2 1 / 4 . 2m>2\left(\frac{c}{8}\right)^{1/4}=\iota c^{1/4},\qquad\iota=2^{1/4}. 2 m > 2 ( 8 c ) 1/4 = ι c 1/4 , ι = 2 1/4 .
We apply Lemma 7.4 with ι = 2 1 / 4 \iota=2^{1/4} ι = 2 1/4 , θ = 1 4 \theta=\frac{1}{4} θ = 4 1 and c 0 = 10 75 c_0=10^{75} c 0 = 1 0 75 , and check its four hypotheses. First, ι c 0 θ = 2 1 / 4 10 75 / 4 > 6.68 ⋅ 10 18 ≥ 1 \iota c_0^\theta=2^{1/4}10^{75/4}>6.68\cdot10^{18}\geq1 ι c 0 θ = 2 1/4 1 0 75/4 > 6.68 ⋅ 1 0 18 ≥ 1 . Second, θ log c 0 = 75 4 log 10 > 43.17 \theta\log c_0=\frac{75}{4}\log10>43.17 θ log c 0 = 4 75 log 10 > 43.17 , while 0 < log ( 19.46 ι ) < 3.15 0<\log(19.46\iota)<3.15 0 < log ( 19.46 ι ) < 3.15 , so θ log c 0 ≥ 2 + ∣ log ( 19.46 ι ) ∣ \theta\log c_0\geq2+|\log(19.46\iota)| θ log c 0 ≥ 2 + ∣ log ( 19.46 ι ) ∣ . Third, Φ c 0 ( ι c 0 θ ) > 3.28 ⋅ 10 18 > 0 \Phi_{c_0}(\iota c_0^\theta)>3.28\cdot10^{18}>0 Φ c 0 ( ι c 0 θ ) > 3.28 ⋅ 1 0 18 > 0 . Fourth, θ ι c 0 θ > 1.67 ⋅ 10 18 \theta\iota c_0^\theta>1.67\cdot10^{18} θ ι c 0 θ > 1.67 ⋅ 1 0 18 , whereas K E log c 0 ( 2 log ( 19.46 ι c 0 θ ) + θ log c 0 ) < 5.78 ⋅ 10 16 K_E\log c_0\left(2\log(19.46\iota c_0^\theta)+\theta\log c_0\right)<5.78\cdot10^{16} K E log c 0 ( 2 log ( 19.46 ι c 0 θ ) + θ log c 0 ) < 5.78 ⋅ 1 0 16 . These four values were verified in 256-bit interval arithmetic with outward rounding.
Hence Lemma 7.4 gives Φ c ( ι c 1 / 4 ) > 0 \Phi_c(\iota c^{1/4})>0 Φ c ( ι c 1/4 ) > 0 . Since ι c 1 / 4 ≥ 1 \iota c^{1/4}\geq1 ι c 1/4 ≥ 1 , Proposition 7.3 with u 0 = ι c 1 / 4 u_0=\iota c^{1/4} u 0 = ι c 1/4 gives 2 m < ι c 1 / 4 2m<\iota c^{1/4} 2 m < ι c 1/4 , which contradicts the floor above. Therefore c < 10 75 c<10^{75} c < 1 0 75 . □ \square □
The value 10 75 10^{75} 1 0 75 is not the limit of this argument: the function c ↦ Φ c ( 2 1 / 4 c 1 / 4 ) c\mapsto\Phi_c(2^{1/4}c^{1/4}) c ↦ Φ c ( 2 1/4 c 1/4 ) changes sign near log 10 c = 73.74 \log_{10}c=73.74 log 10 c = 73.74 , so the same comparison gives c < 10 73.75 c<10^{73.75} c < 1 0 73.75 . We keep 10 75 10^{75} 1 0 75 because it is the upper end of the first window of the index caps.
Index caps on windows of c c c
Below 10 75 10^{75} 1 0 75 we use the sharper lower bound of Mignotte and Voutier for linear forms in three logarithms (Theorem 7.6 ). It is quoted with its factorial term b \mathbf{b} b read from the definition: the displayed statement of Lemma 3.5(b) of [24 ] transposes two coefficients, as the proof of Theorem 7.6 explains. In Proposition 7.8 it is applied to Λ x \Lambda_x Λ x in the basis ( α , D λ , β / α ) (\alpha,D_\lambda,\beta/\alpha) ( α , D λ , β / α ) rather than ( α , β , D λ ) (\alpha,\beta,D_\lambda) ( α , β , D λ ) . The reason is the unit β / α \beta/\alpha β / α : its height is 1 2 log β \frac{1}{2}\log\beta 2 1 log β (Lemma 4.2 , item 3), but its logarithm is less than log 2 + 4 / c \log2 + 4/c log 2 + 4/ c (item 1), so its parameter ( ρ − 1 ) log ( β / α ) + 8 h ( β / α ) (\rho- 1)\log(\beta/\alpha) + 8h(\beta/\alpha) ( ρ − 1 ) log ( β / α ) + 8 h ( β / α ) in Theorem 7.6 is about 4 log β 4\log\beta 4 log β , where β \beta β itself would need ( ρ + 3 ) log β (\rho+ 3)\log\beta ( ρ + 3 ) log β , that is 28 log β 28\log\beta 28 log β for the value ρ = 25 \rho= 25 ρ = 25 used there. The Liouville bound Lemma 7.7 serves one step of that proof: it shows that the two logarithms of the centred linear form in two logarithms have opposite signs.
Theorem 7.6. Let η 1 , η 2 , η 3 > 1 \eta_1,\eta_2,\eta_3 > 1 η 1 , η 2 , η 3 > 1 be real algebraic numbers that are multiplicatively independent, put D = [ Q ( η 1 , η 2 , η 3 ) : Q ] D = [\mathbb{Q}(\eta_1,\eta_2,\eta_3) : \mathbb{Q}] D = [ Q ( η 1 , η 2 , η 3 ) : Q ] , let b 1 , b 2 , b 3 b_1,b_2,b_3 b 1 , b 2 , b 3 be positive integers with gcd ( b 1 , b 2 , b 3 ) = 1 \gcd(b_1,b_2,b_3) = 1 g cd( b 1 , b 2 , b 3 ) = 1 , and
Λ M V = b 1 log η 1 + b 2 log η 2 − b 3 log η 3 , 0 < ∣ Λ M V ∣ < 2 π / w , \Lambda_{\mathrm{MV}} = b_1\log\eta_1 + b_2\log\eta_2 - b_3\log\eta_3,\qquad0 < \lvert\Lambda_{\mathrm{MV}}\rvert< 2\pi/w, Λ MV = b 1 log η 1 + b 2 log η 2 − b 3 log η 3 , 0 < ∣ Λ MV ∣ < 2 π / w , where w w w is the largest order of a root of unity in Q ( η 1 , η 2 , η 3 ) \mathbb{Q}(\eta_1,\eta_2,\eta_3) Q ( η 1 , η 2 , η 3 ) . Let K 0 ≥ 3 K_0 \ge3 K 0 ≥ 3 , ℓ ≥ 5 \ell\ge5 ℓ ≥ 5 , R M V , S , T R_{\mathrm{MV}},S,T R MV , S , T be positive integers, put N ∗ = K 0 ( K 0 + 1 ) ℓ / 2 N_* = K_0(K_0 + 1)\ell/2 N ∗ = K 0 ( K 0 + 1 ) ℓ /2 , assume R M V S T ≥ N ∗ R_{\mathrm{MV}}ST \ge N_* R MV S T ≥ N ∗ , and put g = 1 4 − N ∗ 12 R M V S T g = \frac{1}{4} - \frac{N_*}{12R_{\mathrm{MV}}ST} g = 4 1 − 12 R MV S T N ∗ . Put δ 1 = gcd ( b 1 , b 3 ) \delta_1 = \gcd(b_1,b_3) δ 1 = g cd( b 1 , b 3 ) , δ 2 = gcd ( b 2 , b 3 ) \delta_2 = \gcd(b_2,b_3) δ 2 = g cd( b 2 , b 3 ) and
b = ( R M V − 1 ) b 3 + ( T − 1 ) b 1 2 δ 1 ⋅ ( S − 1 ) b 3 + ( T − 1 ) b 2 2 δ 2 ⋅ ( ∏ j = 1 K 0 − 1 ( j ! ) K 0 − j ) − 12 K 0 ( K 0 − 1 ) ( K 0 + 1 ) . \mathfrak b =
\frac{(R_{\mathrm{MV}}-1)b_3+(T-1)b_1}{2\delta_1}
\cdot
\frac{(S-1)b_3+(T-1)b_2}{2\delta_2}
\cdot
\left(\prod_{j=1}^{K_0-1}(j!)^{K_0-j}\right)^{-\frac{12}{K_0(K_0-1)(K_0+1)}}. b = 2 δ 1 ( R MV − 1 ) b 3 + ( T − 1 ) b 1 ⋅ 2 δ 2 ( S − 1 ) b 3 + ( T − 1 ) b 2 ⋅ ( j = 1 ∏ K 0 − 1 ( j ! ) K 0 − j ) − K 0 ( K 0 − 1 ) ( K 0 + 1 ) 12 . Let R 1 , R 2 , R 3 , S 1 , S 2 , S 3 , T 1 , T 2 , T 3 R_1,R_2,R_3,S_1,S_2,S_3,T_1,T_2,T_3 R 1 , R 2 , R 3 , S 1 , S 2 , S 3 , T 1 , T 2 , T 3 be positive integers with R M V > R 1 + R 2 + R 3 R_{\mathrm{MV}} > R_1+R_2+R_3 R MV > R 1 + R 2 + R 3 , S > S 1 + S 2 + S 3 S > S_1+S_2+S_3 S > S 1 + S 2 + S 3 and T > T 1 + T 2 + T 3 T > T_1+T_2+T_3 T > T 1 + T 2 + T 3 . Let ρ ≥ 2 \rho\ge2 ρ ≥ 2 be real, and let A 1 , A 2 , A 3 A_1,A_2,A_3 A 1 , A 2 , A 3 be real numbers with A i ≥ ( ρ − 1 ) log η i + 2 D h ( η i ) A_i \ge(\rho- 1)\log\eta_i + 2D h(\eta_i) A i ≥ ( ρ − 1 ) log η i + 2 D h ( η i ) (Definition 4.1), such that
( K 0 ℓ 2 + ℓ 2 − 0.37 K 0 − 2 ) log ρ ≥ ( D + 1 ) log N ∗ + g ℓ ( A 1 R M V + A 2 S + A 3 T ) + 2 D ( K 0 − 1 ) 3 log b . \left(\frac{K_0\ell}{2}+\frac{\ell}{2}-0.37K_0-2\right)\log\rho
\ge(D+1)\log N_*+g\ell(A_1R_{\mathrm{MV}}+A_2S+A_3T)+\frac{2D(K_0-1)}{3}\log\mathfrak b. ( 2 K 0 ℓ + 2 ℓ − 0.37 K 0 − 2 ) log ρ ≥ ( D + 1 ) log N ∗ + g ℓ ( A 1 R MV + A 2 S + A 3 T ) + 3 2 D ( K 0 − 1 ) log b . Put V = ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) \mathcal V = \sqrt{(R_1+1)(S_1+1)(T_1+1)} V = ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) , and suppose that for some real χ > 0 \chi> 0 χ > 0
( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > K 0 M , M = max { R 1 + S 1 + 1 , S 1 + T 1 + 1 , R 1 + T 1 + 1 , χ V } , (R_1+1)(S_1+1)(T_1+1)>K_0\mathcal M,\qquad\mathcal M=\max\{R_1+S_1+1,S_1+T_1+1,R_1+T_1+1,\chi\mathcal V\}, ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > K 0 M , M = max { R 1 + S 1 + 1 , S 1 + T 1 + 1 , R 1 + T 1 + 1 , χ V } , # { η 1 j 1 η 2 j 2 η 3 j 3 : 0 ≤ j 1 ≤ R 1 , 0 ≤ j 2 ≤ S 1 , 0 ≤ j 3 ≤ T 1 } > ℓ , \#\{\eta_1^{j_1}\eta_2^{j_2}\eta_3^{j_3}:0\le j_1\le R_1,\ 0\le j_2\le S_1,\ 0\le j_3\le T_1\}>\ell, # { η 1 j 1 η 2 j 2 η 3 j 3 : 0 ≤ j 1 ≤ R 1 , 0 ≤ j 2 ≤ S 1 , 0 ≤ j 3 ≤ T 1 } > ℓ , # { η 1 j 1 η 2 j 2 η 3 j 3 : 0 ≤ j 1 ≤ R 2 , 0 ≤ j 2 ≤ S 2 , 0 ≤ j 3 ≤ T 2 } > 2 K 0 ℓ , \#\{\eta_1^{j_1}\eta_2^{j_2}\eta_3^{j_3}:0\le j_1\le R_2,\ 0\le j_2\le S_2,\ 0\le j_3\le T_2\}>2K_0\ell, # { η 1 j 1 η 2 j 2 η 3 j 3 : 0 ≤ j 1 ≤ R 2 , 0 ≤ j 2 ≤ S 2 , 0 ≤ j 3 ≤ T 2 } > 2 K 0 ℓ , ( R 2 + 1 ) ( S 2 + 1 ) ( T 2 + 1 ) > K 0 2 , ( R 3 + 1 ) ( S 3 + 1 ) ( T 3 + 1 ) > 3 K 0 2 ℓ . (R_2+1)(S_2+1)(T_2+1)>K_0^2,\qquad(R_3+1)(S_3+1)(T_3+1)>3K_0^2\ell. ( R 2 + 1 ) ( S 2 + 1 ) ( T 2 + 1 ) > K 0 2 , ( R 3 + 1 ) ( S 3 + 1 ) ( T 3 + 1 ) > 3 K 0 2 ℓ . Then
Λ M V ′ = ∣ Λ M V ∣ ⋅ ℓ T e ℓ T ∣ Λ M V ∣ / ( 2 b 3 ) 2 b 3 > ρ − K 0 ℓ , \Lambda'_{\mathrm{MV}}=\lvert\Lambda_{\mathrm{MV}}\rvert\cdot\frac{\ell T e^{\ell T\lvert\Lambda_{\mathrm{MV}}\rvert/(2b_3)}}{2b_3}>\rho^{-K_0\ell}, Λ MV ′ = ∣ Λ MV ∣ ⋅ 2 b 3 ℓ T e ℓ T ∣ Λ MV ∣ / ( 2 b 3 ) > ρ − K 0 ℓ , or one of the following holds:
(C1) b 1 ≤ max { R 1 , R 2 } b_1\le\max\{R_1,R_2\} b 1 ≤ max { R 1 , R 2 } , b 2 ≤ max { S 1 , S 2 } b_2\le\max\{S_1,S_2\} b 2 ≤ max { S 1 , S 2 } and b 3 ≤ max { T 1 , T 2 } b_3\le\max\{T_1,T_2\} b 3 ≤ max { T 1 , T 2 } ;
(C2) there are integers u 1 , u 2 , u 3 u_1,u_2,u_3 u 1 , u 2 , u 3 with gcd ( u 1 , u 2 , u 3 ) = 1 \gcd(u_1,u_2,u_3)=1 g cd( u 1 , u 2 , u 3 ) = 1 and u 1 b 1 + u 2 b 2 + u 3 b 3 = 0 u_1b_1+u_2b_2+u_3b_3=0 u 1 b 1 + u 2 b 2 + u 3 b 3 = 0 , such that
∣ u 1 ∣ ≤ ( S 1 + 1 ) ( T 1 + 1 ) M − max { S 1 , T 1 } , ∣ u 2 ∣ ≤ ( R 1 + 1 ) ( T 1 + 1 ) M − max { R 1 , T 1 } , ∣ u 3 ∣ ≤ ( R 1 + 1 ) ( S 1 + 1 ) M − max { R 1 , S 1 } . \lvert u_1\rvert\le\frac{(S_1+1)(T_1+1)}{\mathcal M-\max\{S_1,T_1\}},\qquad
\lvert u_2\rvert\le\frac{(R_1+1)(T_1+1)}{\mathcal M-\max\{R_1,T_1\}},\qquad
\lvert u_3\rvert\le\frac{(R_1+1)(S_1+1)}{\mathcal M-\max\{R_1,S_1\}}. ∣ u 1 ∣ ≤ M − max { S 1 , T 1 } ( S 1 + 1 ) ( T 1 + 1 ) , ∣ u 2 ∣ ≤ M − max { R 1 , T 1 } ( R 1 + 1 ) ( T 1 + 1 ) , ∣ u 3 ∣ ≤ M − max { R 1 , S 1 } ( R 1 + 1 ) ( S 1 + 1 ) .
Proof. This is [24 Theorem 2.1] in its real case (numbering of the arXiv version 3), with the conventions and notation of its Sections 2.1 and 2.2. The parameters written here K 0 K_0 K 0 , ℓ \ell ℓ , N ∗ N_* N ∗ , R M V R_{\mathrm{MV}} R MV , A i A_i A i , δ 1 \delta_1 δ 1 , δ 2 \delta_2 δ 2 are K K K , L L L , N N N , R R R , a i a_i a i , d 1 d_1 d 1 , d 2 d_2 d 2 there, and D D D is the quotient [ Q ( η 1 , η 2 , η 3 ) : Q ] / [ R ( η 1 , η 2 , η 3 ) : R ] [\mathbb{Q}(\eta_1,\eta_2,\eta_3):\mathbb{Q}]/[\mathbb{R}(\eta_1,\eta_2,\eta_3):\mathbb{R}] [ Q ( η 1 , η 2 , η 3 ) : Q ] / [ R ( η 1 , η 2 , η 3 ) : R ] of its Liouville inequality, which in the real case is the degree. The number b \mathfrak b b is the quantity b b b of its definition (2.5), written out by means of (2.3) and (2.4), since
b 3 ′ η 0 = ( R M V − 1 ) b 3 + ( T − 1 ) b 1 2 δ 1 b'_3\eta_0=\frac{(R_{\mathrm{MV}}-1)b_3+(T-1)b_1}{2\delta_1} b 3 ′ η 0 = 2 δ 1 ( R MV − 1 ) b 3 + ( T − 1 ) b 1
and
b 3 ′ ′ ζ 0 = ( S − 1 ) b 3 + ( T − 1 ) b 2 2 δ 2 . b''_3\zeta_0=\frac{(S-1)b_3+(T-1)b_2}{2\delta_2}. b 3 ′′ ζ 0 = 2 δ 2 ( S − 1 ) b 3 + ( T − 1 ) b 2 .
For real η i > 1 \eta_i>1 η i > 1 the lower bound for a i a_i a i in the source, ρ ∣ log η i ∣ − log ∣ η i ∣ + 2 D h ( η i ) \rho\lvert\log\eta_i\rvert-\log\lvert\eta_i\rvert+2D h(\eta_i) ρ ∣ log η i ∣ − log ∣ η i ∣ + 2 D h ( η i ) , is the one stated here. Two remarks on the source. First, b \mathfrak b b is to be read from (2.5): the displayed statement of its Lemma 3.5(b) transposes b 2 b_2 b 2 and b 3 b_3 b 3 in the second fraction, a misprint, as the last display of the proof of that lemma and the entries ( s 3 j ′ ′ + t 3 j ′ ′ ) (s''_{3j}+t''_{3j}) ( s 3 j ′′ + t 3 j ′′ ) of its interpolation determinant show. Second, if Q ( η 1 , η 2 , η 3 ) \mathbb{Q}(\eta_1,\eta_2,\eta_3) Q ( η 1 , η 2 , η 3 ) is real then w = 2 w=2 w = 2 , and the condition on Λ M V \Lambda_{\mathrm{MV}} Λ MV reads 0 < ∣ Λ M V ∣ < π 0<\lvert\Lambda_{\mathrm{MV}}\rvert<\pi 0 < ∣ Λ MV ∣ < π . ∅
Lemma 7.7. Let η ≠ 1 \eta\ne1 η = 1 be a positive real algebraic number with [ Q ( η ) : Q ] ≤ 4 [\mathbb{Q}(\eta) : \mathbb{Q}] \le4 [ Q ( η ) : Q ] ≤ 4 . Then
log ∣ log η ∣ > − 4 h ( η ) − 4. \log|\log\eta| > -4h(\eta)-4. log ∣ log η ∣ > − 4 h ( η ) − 4. Proof. We first bound ∣ η − 1 ∣ |\eta-1| ∣ η − 1∣ from below by evaluating the primitive minimal polynomial of η \eta η at 1. Let D = [ Q ( η ) : Q ] ≤ 4 D=[\mathbb{Q}(\eta):\mathbb{Q}]\le4 D = [ Q ( η ) : Q ] ≤ 4 and let P ( X ) = a η ∏ i = 1 D ( X − η i ) ∈ Z [ X ] P(X)=a_{\eta}\prod_{i=1}^{D}(X-\eta_i)\in\mathbb{Z}[X] P ( X ) = a η ∏ i = 1 D ( X − η i ) ∈ Z [ X ] be the primitive minimal polynomial of η \eta η , with a η > 0 a_{\eta}>0 a η > 0 and η 1 = η \eta_1=\eta η 1 = η . By Definition 4.1, a η ∏ i = 1 D max ( 1 , ∣ η i ∣ ) = e D h ( η ) a_{\eta}\prod_{i=1}^{D}\max(1,|\eta_i|)=e^{Dh(\eta)} a η ∏ i = 1 D max ( 1 , ∣ η i ∣ ) = e D h ( η ) . Since P P P is irreducible and η ≠ 1 \eta\ne1 η = 1 , the integer P ( 1 ) P(1) P ( 1 ) is nonzero, so
1 ≤ ∣ P ( 1 ) ∣ = a η ∣ 1 − η ∣ ∏ i = 2 D ∣ 1 − η i ∣ ≤ ∣ 1 − η ∣ ⋅ a η ∏ i = 2 D 2 max ( 1 , ∣ η i ∣ ) ≤ 2 D − 1 e D h ( η ) ∣ η − 1 ∣ , 1\le|P(1)|=a_{\eta}|1-\eta|\prod_{i=2}^{D}|1-\eta_i|\le|1-\eta|\cdot a_{\eta}\prod_{i=2}^{D}2\max(1,|\eta_i|)\le2^{D-1}e^{Dh(\eta)}|\eta-1|, 1 ≤ ∣ P ( 1 ) ∣ = a η ∣1 − η ∣ i = 2 ∏ D ∣1 − η i ∣ ≤ ∣1 − η ∣ ⋅ a η i = 2 ∏ D 2 max ( 1 , ∣ η i ∣ ) ≤ 2 D − 1 e D h ( η ) ∣ η − 1∣ , using ∣ 1 − η i ∣ ≤ 2 max ( 1 , ∣ η i ∣ ) |1-\eta_i|\le2\max(1,|\eta_i|) ∣1 − η i ∣ ≤ 2 max ( 1 , ∣ η i ∣ ) and max ( 1 , ∣ η ∣ ) ≥ 1 \max(1,|\eta|)\ge1 max ( 1 , ∣ η ∣ ) ≥ 1 . As D ≤ 4 D\le4 D ≤ 4 and h ( η ) ≥ 0 h(\eta)\ge0 h ( η ) ≥ 0 , this gives
log ∣ η − 1 ∣ ≥ − 4 h ( η ) − 3 log 2. \log|\eta-1|\ge-4h(\eta)-3\log2. log ∣ η − 1∣ ≥ − 4 h ( η ) − 3 log 2. We now pass from ∣ η − 1 ∣ |\eta-1| ∣ η − 1∣ to ∣ log η ∣ |\log\eta| ∣ log η ∣ in three cases. If 1 < η ≤ 2 1<\eta\le2 1 < η ≤ 2 , then log η ≥ η − 1 η ≥ η − 1 2 \log\eta\ge\frac{\eta-1}{\eta}\ge\frac{\eta-1}{2} log η ≥ η η − 1 ≥ 2 η − 1 , so log log η ≥ log ( η − 1 ) − log 2 ≥ − 4 h ( η ) − 4 log 2 > − 4 h ( η ) − 4 \log\log\eta\ge\log(\eta-1)-\log2\ge-4h(\eta)-4\log2>-4h(\eta)-4 log log η ≥ log ( η − 1 ) − log 2 ≥ − 4 h ( η ) − 4 log 2 > − 4 h ( η ) − 4 . If 0 < η < 1 0<\eta<1 0 < η < 1 , then ∣ log η ∣ = − log η ≥ 1 − η = ∣ η − 1 ∣ |\log\eta|=-\log\eta\ge1-\eta=|\eta-1| ∣ log η ∣ = − log η ≥ 1 − η = ∣ η − 1∣ , so log ∣ log η ∣ ≥ − 4 h ( η ) − 3 log 2 > − 4 h ( η ) − 4 \log|\log\eta|\ge-4h(\eta)-3\log2>-4h(\eta)-4 log ∣ log η ∣ ≥ − 4 h ( η ) − 3 log 2 > − 4 h ( η ) − 4 . If η > 2 \eta>2 η > 2 , then log ∣ log η ∣ > log log 2 > − 0.37 > − 4 ≥ − 4 h ( η ) − 4 \log|\log\eta|>\log\log2>-0.37>-4\ge-4h(\eta)-4 log ∣ log η ∣ > log log 2 > − 0.37 > − 4 ≥ − 4 h ( η ) − 4 . This proves the bound in all cases.
Proposition 7.8. Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of an Euler triple { a , b , c } \{a,b,c\} { a , b , c } , and suppose that c ≥ 10 48.6 c\ge10^{48.6} c ≥ 1 0 48.6 .
If c < 10 75 c<10^{75} c < 1 0 75 , then m < 7 ⋅ 10 12 m<7\cdot10^{12} m < 7 ⋅ 1 0 12 .
If c < 10 52 c<10^{52} c < 1 0 52 , then m < 3.45 ⋅ 10 12 m<3.45\cdot10^{12} m < 3.45 ⋅ 1 0 12 .
If c < 10 51 c<10^{51} c < 1 0 51 , then m < 1.1 ⋅ 10 12 m<1.1\cdot10^{12} m < 1.1 ⋅ 1 0 12 .
If c < 10 48.77 c<10^{48.77} c < 1 0 48.77 , then m < 10 12 m<10^{12} m < 1 0 12 .
Proof. We apply the theorem of Mignotte and Voutier (Theorem 7.6) to the linear form Λ x \Lambda_x Λ x of Definition 2.16 in the basis ( α , D λ , β / α ) (\alpha,D_{\lambda},\beta/\alpha) ( α , D λ , β / α ) , and exclude its exceptional alternative (C2) by a linear form in two logarithms centred at the relation it provides, together with Laurent’s theorem (Theorem 4.3) applied separately for each denominator. The argument is the same in four rows, which differ only in their parameters; each row starts from the cap proved by the row before, and the first from the a priori bound (Proposition 7.3).
We first fix the parameters. Put L 0 = 48.6 log 10 L_0=48.6\log10 L 0 = 48.6 log 10 , ρ = 25 \rho=25 ρ = 25 , M M V = 5 2 M_{\mathrm{MV}}=\frac{5}{2} M MV = 2 5 , ρ + = ρ + 3 = 28 \rho_+=\rho+3=28 ρ + = ρ + 3 = 28 , ρ − = ρ − 1 = 24 \rho_-=\rho-1=24 ρ − = ρ − 1 = 24 , and
Table 1.
In each row we assume L 0 ≤ log c < U L_0 \le\log c < U L 0 ≤ log c < U and prove 2 m < Q 2m < Q 2 m < Q ; rows 1, 2, 3 and 4 prove items 1, 2, 3 and 4.
We next show 2 m < Q 0 2m < Q_0 2 m < Q 0 in each row. In row 1, with Φ c \Phi_c Φ c and K E K_E K E as in Convention 7.1, since log c < U \log c < U log c < U and log ( 19.46 Q 0 ) > 0 \log(19.46Q_0) > 0 log ( 19.46 Q 0 ) > 0 ,
Φ c ( Q 0 ) ≥ Q 0 − 0.15 − K E U 2 log ( 19.46 Q 0 ) > 0 , \Phi_c(Q_0) \ge Q_0 - 0.15 - K_E U^2 \log(19.46Q_0) > 0, Φ c ( Q 0 ) ≥ Q 0 − 0.15 − K E U 2 log ( 19.46 Q 0 ) > 0 ,
because K E U 2 log ( 19.46 Q 0 ) + 0.15 < 3.3535 ⋅ 10 18 K_E U^2 \log(19.46Q_0) + 0.15 < 3.3535 \cdot10^{18} K E U 2 log ( 19.46 Q 0 ) + 0.15 < 3.3535 ⋅ 1 0 18 ; so Proposition 7.3 with u 0 = Q 0 ≥ 1 u_0 = Q_0 \ge1 u 0 = Q 0 ≥ 1 gives 2 m < Q 0 2m < Q_0 2 m < Q 0 . In rows 2, 3 and 4, Q 0 Q_0 Q 0 is the value of Q Q Q in rows 1, 2 and 3, whose windows contain the window of the row; so 2 m < Q 0 2m < Q_0 2 m < Q 0 follows from those rows, which are proved first.
Suppose 2 m ≥ Q 2m \ge Q 2 m ≥ Q ; we derive a contradiction in each row. By item 4 of Lemma 2.18 , Λ x < 4.01 β − 4 m \Lambda_x < 4.01\beta^{-4m} Λ x < 4.01 β − 4 m , and log 4.01 < 1.39 \log4.01 < 1.39 log 4.01 < 1.39 , so
log Λ x < − 4 m log β + 1.39. (9) \log\Lambda_x < -4m\log\beta+ 1.39. \tag*{(9)} log Λ x < − 4 m log β + 1.39. (9)
We next set up the linear form in three logarithms. By Definition 2.16 , Λ x = 2 l log α − 2 m log β + log D λ = ( 2 l − 2 m ) log α + log D λ − 2 m log ( β / α ) \Lambda_x = 2l\log\alpha- 2m\log\beta+ \log D_\lambda= (2l - 2m)\log\alpha+ \log D_\lambda- 2m\log(\beta/\alpha) Λ x = 2 l log α − 2 m log β + log D λ = ( 2 l − 2 m ) log α + log D λ − 2 m log ( β / α ) , and l > m l > m l > m by Lemma 2.14. We apply Theorem 7.6 with ( η 1 , η 2 , η 3 ) = ( α , D λ , β / α ) (\eta_1,\eta_2,\eta_3) = (\alpha,D_\lambda,\beta/\alpha) ( η 1 , η 2 , η 3 ) = ( α , D λ , β / α ) and ( b 1 , b 2 , b 3 ) = ( 2 l − 2 m , 1 , 2 m ) (b_1,b_2,b_3) = (2l - 2m,1,2m) ( b 1 , b 2 , b 3 ) = ( 2 l − 2 m , 1 , 2 m ) . The b i b_i b i are positive integers with gcd ( b 1 , b 2 , b 3 ) = 1 \gcd(b_1,b_2,b_3) = 1 g cd( b 1 , b 2 , b 3 ) = 1 . By item 6 of Lemma 4.2 , the η i \eta_i η i exceed 1 and are multiplicatively independent, and they generate the real biquadratic field K K K of item 2 (already α \alpha α and β = α ⋅ β / α \beta= \alpha\cdot\beta/\alpha β = α ⋅ β / α do), so D = 4 D = 4 D = 4 ; as K K K is real, the condition on Λ M V \Lambda_{\mathrm{MV}} Λ MV in Theorem 7.6 reads 0 < Λ x < π 0 < \Lambda_x < \pi 0 < Λ x < π , which holds as Λ x < 1 \Lambda_x < 1 Λ x < 1 by item 4 of Lemma 2.18 . Put
A 1 = ρ + log α , A 2 = 8 log c + ρ − log 4 , A 3 = ρ + log β − ρ − log α = 4 log β + ρ − log ( β / α ) . A_1 = \rho_+\log\alpha,\qquad A_2 = 8\log c + \rho_-\log4,\qquad A_3 = \rho_+\log\beta- \rho_-\log\alpha= 4\log\beta+ \rho_-\log(\beta/\alpha). A 1 = ρ + log α , A 2 = 8 log c + ρ − log 4 , A 3 = ρ + log β − ρ − log α = 4 log β + ρ − log ( β / α ) .
These are admissible: by item 3 of Lemma 4.2 , h ( α ) = 1 2 log α h(\alpha) = \frac{1}{2}\log\alpha h ( α ) = 2 1 log α and h ( β / α ) = 1 2 log β h(\beta/\alpha) = \frac{1}{2}\log\beta h ( β / α ) = 2 1 log β , so A 1 = ( ρ − 1 ) log α + 8 h ( α ) A_1 = (\rho_-1)\log\alpha+ 8h(\alpha) A 1 = ( ρ − 1 ) log α + 8 h ( α ) and A 3 = ( ρ − 1 ) log ( β / α ) + 8 h ( β / α ) A_3 = (\rho_-1)\log(\beta/\alpha) + 8h(\beta/\alpha) A 3 = ( ρ − 1 ) log ( β / α ) + 8 h ( β / α ) ; and by item 4, ( ρ − 1 ) log D λ + 8 h ( D λ ) < ρ − log 4 + 8 log c = A 2 (\rho_-1)\log D_\lambda+ 8h(D_\lambda) < \rho_-\log4 + 8\log c = A_2 ( ρ − 1 ) log D λ + 8 h ( D λ ) < ρ − log 4 + 8 log c = A 2 . Item 1 of Lemma 4.2 gives
A 1 > ρ + L 0 2 , A 2 ≥ 8 L 0 + ρ − log 4 , A 3 > 4 log β > 2 L 0 + 4 log 2 , A_1 > \frac{\rho_+L_0}{2},\qquad A_2 \ge8L_0 + \rho_-\log4,\qquad A_3 > 4\log\beta> 2L_0 + 4\log2, A 1 > 2 ρ + L 0 , A 2 ≥ 8 L 0 + ρ − log 4 , A 3 > 4 log β > 2 L 0 + 4 log 2 ,
so A − : = min i A i > 2 L 0 + 4 log 2 A_- := \min_i A_i > 2L_0 + 4\log2 A − := min i A i > 2 L 0 + 4 log 2 and Ω : = A 1 A 2 A 3 > Ω − \Omega:= A_1A_2A_3 > \Omega_- Ω := A 1 A 2 A 3 > Ω − , the product of these three lower bounds. Also A 1 < A 1 + : = ρ + ( U − log 2 ) A_1 < A_1^+ := \rho_+(U - \log2) A 1 < A 1 + := ρ + ( U − log 2 ) , A 2 < A 2 + : = 8 U + ρ − log 4 A_2 < A_2^+ := 8U + \rho_-\log4 A 2 < A 2 + := 8 U + ρ − log 4 , and A 1 + A 2 + A 3 = 4 log α + ρ + log β + 8 log c + ρ − log 4 < Σ + : = ( 12 + ρ + ) U + ρ − log 4 A_1 + A_2 + A_3 = 4\log\alpha+ \rho_+\log\beta+ 8\log c + \rho_-\log4 < \Sigma_+ := (12 + \rho_+)U + \rho_-\log4 A 1 + A 2 + A 3 = 4 log α + ρ + log β + 8 log c + ρ − log 4 < Σ + := ( 12 + ρ + ) U + ρ − log 4 .
We next bound the product Ω \Omega Ω . Put ε = 4 ⋅ 10 − 48.6 \varepsilon= 4\cdot10^{-48.6} ε = 4 ⋅ 1 0 − 48.6 , so that log β − log α < log 2 + ε \log\beta- \log\alpha< \log2 + \varepsilon log β − log α < log 2 + ε by item 1 (as 4 / c ≤ ε 4/c \le\varepsilon 4/ c ≤ ε ). Then ( log α ) 2 / log β > ( log α ) 2 / ( log α + log 2 + ε ) > log α − log 2 − ε (\log\alpha)^2/\log\beta> (\log\alpha)^2/(\log\alpha+\log2+\varepsilon) > \log\alpha- \log2 - \varepsilon ( log α ) 2 / log β > ( log α ) 2 / ( log α + log 2 + ε ) > log α − log 2 − ε , so A 3 log α / log β = ρ + log α − ρ − ( log α ) 2 / log β < 4 log α + ρ − ( log 2 + ε ) < 4 log c + ( ρ − − 4 ) log 2 + ρ − ε A_3\log\alpha/\log\beta= \rho_+\log\alpha- \rho_-(\log\alpha)^2/\log\beta< 4\log\alpha+ \rho_-(\log2+\varepsilon) < 4\log c + (\rho_- - 4)\log2 + \rho_-\varepsilon A 3 log α / log β = ρ + log α − ρ − ( log α ) 2 / log β < 4 log α + ρ − ( log 2 + ε ) < 4 log c + ( ρ − − 4 ) log 2 + ρ − ε , using log α < log c − log 2 \log\alpha< \log c - \log2 log α < log c − log 2 . Hence, with Υ ( Y ) = ρ + ( 8 Y + ρ − log 4 ) ( 4 Y + ( ρ − − 4 ) log 2 + ρ − ε ) \Upsilon(Y) = \rho_+(8Y + \rho_-\log4)(4Y + (\rho_- - 4)\log2 + \rho_-\varepsilon) Υ ( Y ) = ρ + ( 8 Y + ρ − log 4 ) ( 4 Y + ( ρ − − 4 ) log 2 + ρ − ε ) for real Y > 0 Y > 0 Y > 0 ,
Ω log β < Υ ( log c ) ≤ Υ ( U ) , Ω < Ω + : = Υ ( U ) U . (10) \frac{\Omega}{\log\beta} < \Upsilon(\log c) \le\Upsilon(U),\qquad\Omega< \Omega_+ := \Upsilon(U)U. \tag*{(10)} log β Ω < Υ ( log c ) ≤ Υ ( U ) , Ω < Ω + := Υ ( U ) U . (10)
We now choose the remaining parameters of Theorem 7.6 and check its side conditions. Following Section 5.2 of [24 ] , put K 0 = ⌊ M M V ℓ Ω ⌋ K_0 = \lfloor M_{\mathrm{MV}}\ell\Omega\rfloor K 0 = ⌊ M MV ℓ Ω ⌋ , c 1 = ( χ M M V ℓ ) 2 / 3 c_1 = (\chi M_{\mathrm{MV}}\ell)^{2/3} c 1 = ( χ M MV ℓ ) 2/3 , c 2 = ( M M V ℓ ) 2 / 3 c_2 = (M_{\mathrm{MV}}\ell)^{2/3} c 2 = ( M MV ℓ ) 2/3 , c 3 = ( 3 M M V 2 ) 1 / 3 c_3 = (3M_{\mathrm{MV}}^2)^{1/3} c 3 = ( 3 M MV 2 ) 1/3 , R i = ⌈ c i A 2 A 3 ⌉ R_i = \lceil c_iA_2A_3\rceil R i = ⌈ c i A 2 A 3 ⌉ , S i = ⌈ c i A 1 A 3 ⌉ S_i = \lceil c_iA_1A_3\rceil S i = ⌈ c i A 1 A 3 ⌉ , T i = ⌈ c i A 1 A 2 ⌉ T_i = \lceil c_iA_1A_2\rceil T i = ⌈ c i A 1 A 2 ⌉ for i = 1 , 2 , 3 i = 1,2,3 i = 1 , 2 , 3 , R M V = 1 + ∑ R i R_{\mathrm{MV}} = 1 + \sum R_i R MV = 1 + ∑ R i , S = 1 + ∑ S i S = 1 + \sum S_i S = 1 + ∑ S i , T = 1 + ∑ T i T = 1 + \sum T_i T = 1 + ∑ T i , and C ∗ = c 1 + c 2 + c 3 C_* = c_1 + c_2 + c_3 C ∗ = c 1 + c 2 + c 3 . Then R i + 1 > c i A 2 A 3 R_i + 1 > c_iA_2A_3 R i + 1 > c i A 2 A 3 , and similarly for S i , T i S_i,T_i S i , T i . The side checks c 1 2 > 2 M M V ℓ / A − c_1^2 > 2M_{\mathrm{MV}}\ell/A_- c 1 2 > 2 M MV ℓ / A − , χ c 1 A − > 1 \chi\sqrt{c_1}A_- > 1 χ c 1 A − > 1 and M M V Ω − > 2 M_{\mathrm{MV}}\Omega_- > 2 M MV Ω − > 2 hold in every row. First, ν 2 = ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > c 1 3 Ω 2 = ( χ M M V ℓ Ω ) 2 ≥ ( χ K 0 ) 2 \nu^2 = (R_1 + 1)(S_1 + 1)(T_1 + 1) > c_1^3\Omega^2 = (\chi M_{\mathrm{MV}}\ell\Omega)^2 \ge(\chi K_0)^2 ν 2 = ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > c 1 3 Ω 2 = ( χ M MV ℓ Ω ) 2 ≥ ( χ K 0 ) 2 , so ν 2 > K 0 χ ν \nu^2 > K_0\chi\nu ν 2 > K 0 χ ν ; and
( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) R 1 + S 1 + 1 > T 1 + 1 ( R 1 + 1 ) − 1 + ( S 1 + 1 ) − 1 > c 1 2 Ω A 1 − 1 + A 2 − 1 ≥ c 1 2 A − Ω 2 > M M V ℓ Ω ≥ K 0 , \begin{aligned}
\frac{(R_1+1)(S_1+1)(T_1+1)}{R_1+S_1+1}
>
\frac{T_1+1}{(R_1+1)^{-1}+(S_1+1)^{-1}}
>
\frac{c_1^2\Omega}{A_1^{-1}+A_2^{-1}} \\
&\ge\frac{c_1^2A_-\Omega}{2} > M_{\mathrm{MV}}\ell\Omega\ge K_0,
\end{aligned} R 1 + S 1 + 1 ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > ( R 1 + 1 ) − 1 + ( S 1 + 1 ) − 1 T 1 + 1 > A 1 − 1 + A 2 − 1 c 1 2 Ω ≥ 2 c 1 2 A − Ω > M MV ℓ Ω ≥ K 0 ,
and likewise for the other two pairs; so ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > K 0 M (R_1 + 1)(S_1 + 1)(T_1 + 1) > K_0\mathcal{M} ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) > K 0 M . By the multiplicative independence of the η i \eta_i η i , the two cardinalities in Theorem 7.6 equal ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) (R_1 + 1)(S_1 + 1)(T_1 + 1) ( R 1 + 1 ) ( S 1 + 1 ) ( T 1 + 1 ) and ( R 2 + 1 ) ( S 2 + 1 ) ( T 2 + 1 ) (R_2 + 1)(S_2 + 1)(T_2 + 1) ( R 2 + 1 ) ( S 2 + 1 ) ( T 2 + 1 ) ; and c 1 Ω 2 > ℓ c_1\Omega^2 > \ell c 1 Ω 2 > ℓ , c 2 2 Ω 2 = ( M M V ℓ Ω ) 2 > max ( 2 K 0 ℓ , K 0 2 ) c_2^2\Omega^2 = (M_{\mathrm{MV}}\ell\Omega)^2 > \max(2K_0\ell,K_0^2) c 2 2 Ω 2 = ( M MV ℓ Ω ) 2 > max ( 2 K 0 ℓ , K 0 2 ) (as M M V Ω > 2 M_{\mathrm{MV}}\Omega> 2 M MV Ω > 2 ), c 3 Ω 2 = 3 M M V 2 ℓ 2 Ω 2 ≥ 3 K 0 2 ℓ c_3\Omega^2 = 3M_{\mathrm{MV}}^2\ell^2\Omega^2 \geq3K_0^2\ell c 3 Ω 2 = 3 M MV 2 ℓ 2 Ω 2 ≥ 3 K 0 2 ℓ give the remaining four conditions. Finally R M V S T > c 3 Ω 2 > N ∗ = K 0 ( K 0 + 1 ) ℓ / 2 R_{\mathrm{MV}}ST > c_3\Omega^2 > N_* = K_0(K_0 + 1)\ell/2 R MV S T > c 3 Ω 2 > N ∗ = K 0 ( K 0 + 1 ) ℓ /2 , K 0 ≥ 3 K_0 \geq3 K 0 ≥ 3 and ℓ ≥ 5 \ell\geq5 ℓ ≥ 5 .
We next bound the factorial term b \mathfrak{b} b of Theorem 7.6. Here δ 1 = gcd ( 2 ℓ − 2 m , 2 m ) ≥ 1 \delta_1 = \gcd(2\ell- 2m,2m) \geq1 δ 1 = g cd( 2 ℓ − 2 m , 2 m ) ≥ 1 and δ 2 = gcd ( 1 , 2 m ) = 1 \delta_2 = \gcd(1,2m) = 1 δ 2 = g cd( 1 , 2 m ) = 1 , and the last factor of b \mathfrak{b} b is at most e 11 / 3 K 0 − 2 e^{11/3}K_0^{-2} e 11/3 K 0 − 2 by [24 Lemma 3.5(a)] . With ( b 1 , b 2 , b 3 ) = ( 2 ℓ − 2 m , 1 , 2 m ) (b_1,b_2,b_3) = (2\ell- 2m,1,2m) ( b 1 , b 2 , b 3 ) = ( 2 ℓ − 2 m , 1 , 2 m ) this gives
log b ≤ log 2 ( R M V − 1 ) m + ( T − 1 ) ( 2 ℓ − 2 m ) 2 + log 2 ( S − 1 ) m + ( T − 1 ) 2 − 2 log K 0 + 11 3 . \log\mathfrak{b} \leq\log\frac{2(R_{\mathrm{MV}} - 1)m + (T - 1)(2\ell- 2m)}{2} + \log\frac{2(S - 1)m + (T - 1)}{2} - 2\log K_0 + \frac{11}{3}. log b ≤ log 2 2 ( R MV − 1 ) m + ( T − 1 ) ( 2 ℓ − 2 m ) + log 2 2 ( S − 1 ) m + ( T − 1 ) − 2 log K 0 + 3 11 .
Since R M V − 1 ≤ C ∗ A 2 A 3 R_{\mathrm{MV}} - 1 \leq C_*A_2A_3 R MV − 1 ≤ C ∗ A 2 A 3 , S − 1 ≤ C ∗ A 1 A 3 S - 1 \leq C_*A_1A_3 S − 1 ≤ C ∗ A 1 A 3 and T − 1 ≤ C ∗ A 1 A 2 T - 1 \leq C_*A_1A_2 T − 1 ≤ C ∗ A 1 A 2 , the two numerators are at most C ∗ Ω ( 2 m A 1 + 2 ℓ − 2 m A 3 ) C_*\Omega\left(\frac{2m}{A_1} + \frac{2\ell-2m}{A_3}\right) C ∗ Ω ( A 1 2 m + A 3 2 ℓ − 2 m ) and C ∗ Ω ( 2 m A 2 + 1 A 3 ) C_*\Omega\left(\frac{2m}{A_2} + \frac{1}{A_3}\right) C ∗ Ω ( A 2 2 m + A 3 1 ) . As ℓ log α < m log β \ell\log\alpha< m\log\beta ℓ log α < m log β (item 1 of Lemma 2.18), 2 ℓ − 2 m < 2 m log ( β / α ) / log α 2\ell- 2m < 2m\log(\beta/\alpha)/\log\alpha 2 ℓ − 2 m < 2 m log ( β / α ) / log α ; with A 3 > 4 log β A_3 > 4\log\beta A 3 > 4 log β , log α > 1 2 log c \log\alpha> \frac{1}{2}\log c log α > 2 1 log c , log β > 1 2 log c \log\beta> \frac{1}{2}\log c log β > 2 1 log c and 2 m < Q 0 2m < Q_0 2 m < Q 0 ,
2 m A 1 + 2 ℓ − 2 m A 3 < 2 m log α ( 1 ρ + + log 2 + ε 4 log β ) < 2 Q 0 L 0 ( 1 ρ + + log 2 + ε 2 L 0 ) = : N 1 , \frac{2m}{A_1} + \frac{2\ell- 2m}{A_3} < \frac{2m}{\log\alpha}\left(\frac{1}{\rho_+} + \frac{\log2 + \varepsilon}{4\log\beta}\right) < \frac{2Q_0}{L_0}\left(\frac{1}{\rho_+} + \frac{\log2 + \varepsilon}{2L_0}\right) =: N_1, A 1 2 m + A 3 2 ℓ − 2 m < log α 2 m ( ρ + 1 + 4 log β log 2 + ε ) < L 0 2 Q 0 ( ρ + 1 + 2 L 0 log 2 + ε ) =: N 1 ,
2 m A 2 + 1 A 3 < Q 0 8 L 0 + ρ − log 4 + 1 2 L 0 + 4 log 2 = : N 2 . \frac{2m}{A_2} + \frac{1}{A_3} < \frac{Q_0}{8L_0 + \rho_-\log4} + \frac{1}{2L_0 + 4\log2} =: N_2. A 2 2 m + A 3 1 < 8 L 0 + ρ − log 4 Q 0 + 2 L 0 + 4 log 2 1 =: N 2 .
With K 0 ≥ M M V ℓ Ω − 1 K_0 \geq M_{\mathrm{MV}}\ell\Omega- 1 K 0 ≥ M MV ℓ Ω − 1 we obtain
log b < B : = 2 log C ∗ M M V ℓ + log N 1 N 2 4 + 11 3 − 2 log ( 1 − 1 M M V ℓ Ω − ) . \log\mathfrak{b} < \mathcal{B} := 2\log\frac{C_*}{M_{\mathrm{MV}}\ell} + \log\frac{N_1N_2}{4} + \frac{11}{3} - 2\log\left(1 - \frac{1}{M_{\mathrm{MV}}\ell\Omega_-}\right). log b < B := 2 log M MV ℓ C ∗ + log 4 N 1 N 2 + 3 11 − 2 log ( 1 − M MV ℓ Ω − 1 ) .
We next reduce the main condition of Theorem 7.6 to one numerical inequality. With D = 4 \mathcal{D}=4 D = 4 it reads ( K 0 ℓ 2 + ℓ 2 − 0.37 K 0 − 2 ) log ρ ≥ 5 log N ∗ + g ℓ ( A 1 R M V + A 2 S + A 3 T ) + 8 ( K 0 − 1 ) 3 log b \left(\frac{K_0\ell}{2} + \frac{\ell}{2} - 0.37K_0 - 2\right)\log\rho\geq5\log N_* + g\ell(A_1R_{\mathrm{MV}} + A_2S + A_3T) + \frac{8(K_0-1)}{3}\log\mathfrak{b} ( 2 K 0 ℓ + 2 ℓ − 0.37 K 0 − 2 ) log ρ ≥ 5 log N ∗ + g ℓ ( A 1 R MV + A 2 S + A 3 T ) + 3 8 ( K 0 − 1 ) log b . Divide by K 0 K_0 K 0 . The left side becomes at least ( ℓ 2 − 0.37 ) log ρ \left(\frac{\ell}{2} - 0.37\right)\log\rho ( 2 ℓ − 0.37 ) log ρ , as ℓ ≥ 4 \ell\geq4 ℓ ≥ 4 . On the right, 5 log N ∗ K 0 < 5 log N + K − \frac{5\log N_*}{K_0} < \frac{5\log N_+}{K_-} K 0 5 l o g N ∗ < K − 5 l o g N + with K − = M M V ℓ Ω − − 1 K_- = M_{\mathrm{MV}}\ell\Omega_- - 1 K − = M MV ℓ Ω − − 1 , K + = M M V ℓ Ω + K_+ = M_{\mathrm{MV}}\ell\Omega_+ K + = M MV ℓ Ω + and N + = K + ( K + + 1 ) ℓ / 2 N_+ = K_+(K_+ + 1)\ell/2 N + = K + ( K + + 1 ) ℓ /2 ; next g ≤ 1 4 g \leq\frac{1}{4} g ≤ 4 1 and A 1 R M V + A 2 S + A 3 T ≤ 3 C ∗ Ω + Σ + A_1R_{\mathrm{MV}} + A_2S + A_3T \leq3C_*\Omega+ \Sigma_+ A 1 R MV + A 2 S + A 3 T ≤ 3 C ∗ Ω + Σ + , and ( 3 C ∗ Ω + Σ + ) / ( M M V ℓ Ω − 1 ) (3C_*\Omega+ \Sigma_+)/(M_{\mathrm{MV}}\ell\Omega- 1) ( 3 C ∗ Ω + Σ + ) / ( M MV ℓ Ω − 1 ) decreases with Ω \Omega Ω ; finally 8 ( K 0 − 1 ) 3 K 0 log b < 8 3 B \frac{8(K_0-1)}{3K_0}\log\mathfrak{b} < \frac{8}{3}\mathcal{B} 3 K 0 8 ( K 0 − 1 ) log b < 3 8 B . Hence the main condition follows from
( ℓ 2 − 0.37 ) log ρ > 5 log N + K − + ℓ 4 3 C ∗ Ω − + Σ + M M V ℓ Ω − − 1 + 8 3 B . (11) \left(\frac{\ell}{2} - 0.37\right)\log\rho> \frac{5\log N_+}{K_-} + \frac{\ell}{4}\frac{3C_*\Omega_- + \Sigma_+}{M_{\mathrm{MV}}\ell\Omega_- - 1} + \frac{8}{3}\mathcal{B}. \tag*{(11)} ( 2 ℓ − 0.37 ) log ρ > K − 5 log N + + 4 ℓ M MV ℓ Ω − − 1 3 C ∗ Ω − + Σ + + 3 8 B . (11)
which holds in every row, with the margins shown in the table below.
We now exclude the main alternative of Theorem 7.6. Put T 0 = ℓ ( 1 + C ∗ A 1 + A 2 + ) T_0 = \ell(1 + C_*A_1^+A_2^+) T 0 = ℓ ( 1 + C ∗ A 1 + A 2 + ) , so that ℓ T ≤ T 0 \ell T \leq T_0 ℓ T ≤ T 0 ; in every row T 0 < Q / 2 T_0 < Q/2 T 0 < Q /2 . As 2 m ≥ Q 2m \geq Q 2 m ≥ Q and Λ x < 1 \Lambda_x < 1 Λ x < 1 , Λ M V ′ = Λ x ℓ T 4 m e ℓ T Λ x / ( 4 m ) < 1 4 Λ x < Λ x \Lambda'_{\mathrm{MV}} = \Lambda_x\frac{\ell T}{4m}e^{\ell T\Lambda_x/(4m)} < \frac{1}{4}\Lambda_x < \Lambda_x Λ MV ′ = Λ x 4 m ℓ T e ℓ T Λ x / ( 4 m ) < 4 1 Λ x < Λ x . The inequality Λ M V ′ > ρ − K 0 ℓ \Lambda'_{\mathrm{MV}} > \rho^{-K_0\ell} Λ MV ′ > ρ − K 0 ℓ then gives log Λ x > − M M V ℓ 2 Ω log ρ \log\Lambda_x > -M_{\mathrm{MV}}\ell^2\Omega\log\rho log Λ x > − M MV ℓ 2 Ω log ρ , and with (1), (2) and log β > L 0 2 + log 2 \log\beta> \frac{L_0}{2} + \log2 log β > 2 L 0 + log 2 ,
2 m < M M V ℓ 2 log ρ 2 ⋅ Ω log β + 1.39 2 log β < M M V ℓ 2 log ρ 2 γ ( U ) + 1.39 L 0 + 2 log 2 = : cap . 2m < \frac{M_{\mathrm{MV}}\ell^2\log\rho}{2}\cdot\frac{\Omega}{\log\beta} + \frac{1.39}{2\log\beta} < \frac{M_{\mathrm{MV}}\ell^2\log\rho}{2}\gamma(U) + \frac{1.39}{L_0 + 2\log2} =: \operatorname{cap}. 2 m < 2 M MV ℓ 2 log ρ ⋅ log β Ω + 2 log β 1.39 < 2 M MV ℓ 2 log ρ γ ( U ) + L 0 + 2 log 2 1.39 =: cap .
In every row cap < Q \operatorname{cap} < Q cap < Q , which contradicts 2 m ≥ Q 2m \geq Q 2 m ≥ Q .
Alternative (C1) gives 2 m = b 3 ≤ max ( T 1 , T 2 ) ≤ max ( c 1 , c 2 ) A 1 A 2 < max ( c 1 , c 2 ) A 1 + A 2 + 2m = b_3 \leq\max(T_1,T_2) \leq\max(c_1,c_2)A_1A_2 < \max(c_1,c_2)A_1^+A_2^+ 2 m = b 3 ≤ max ( T 1 , T 2 ) ≤ max ( c 1 , c 2 ) A 1 A 2 < max ( c 1 , c 2 ) A 1 + A 2 + , and the last number is below 7 ⋅ 10 8 < Q 7\cdot10^8 < Q 7 ⋅ 1 0 8 < Q in every row; this is impossible.
It remains to exclude alternative (C2); this is the main step of the proof. Let u 1 , u 2 , u 3 u_1,u_2,u_3 u 1 , u 2 , u 3 be the integers of (C2): gcd ( u 1 , u 2 , u 3 ) = 1 \gcd(u_1,u_2,u_3) = 1 g cd( u 1 , u 2 , u 3 ) = 1 , u 1 ( 2 ℓ − 2 m ) + u 2 + 2 u 3 m = 0 u_1(2\ell- 2m) + u_2 + 2u_3m = 0 u 1 ( 2 ℓ − 2 m ) + u 2 + 2 u 3 m = 0 , ∣ u 1 ∣ ≤ ( S 1 + 1 ) ( T 1 + 1 ) M − max ( S 1 , T 1 ) |u_1| \leq\frac{(S_1+1)(T_1+1)}{\mathcal{M}-\max(S_1,T_1)} ∣ u 1 ∣ ≤ M − m a x ( S 1 , T 1 ) ( S 1 + 1 ) ( T 1 + 1 ) , and the two analogous bounds hold. Since M ≥ χ V > χ c 1 3 / 2 Ω \mathcal{M} \ge\chi\mathcal{V} > \chi c_1^{3/2}\Omega M ≥ χ V > χ c 1 3/2 Ω , ( S 1 + 1 ) ( T 1 + 1 ) ≤ c 1 2 A 1 2 A 2 A 3 ( 1 + 1 c 1 A 2 2 ) 2 (S_1+1)(T_1+1) \le c_1^2A_1^2A_2A_3\left(1+\frac{1}{c_1A_2^2}\right)^2 ( S 1 + 1 ) ( T 1 + 1 ) ≤ c 1 2 A 1 2 A 2 A 3 ( 1 + c 1 A 2 2 1 ) 2 and max ( S 1 , T 1 ) ≤ c 1 A 1 max ( A 2 , A 3 ) \max(S_1,T_1) \le c_1A_1\max(A_2,A_3) max ( S 1 , T 1 ) ≤ c 1 A 1 max ( A 2 , A 3 ) , and similarly for u 2 , u 3 u_2,u_3 u 2 , u 3 , we get ∣ u i ∣ ≤ F A i |u_i| \le FA_i ∣ u i ∣ ≤ F A i with
F = c 1 χ ( 1 + 1 / ( c 1 A − 2 ) ) 2 1 − 1 / ( χ c 1 A − ) . F=\frac{\sqrt{c_1}}{\chi}\frac{\left(1+1/(c_1A_-^2)\right)^2}{1-1/(\chi\sqrt{c_1}A_-)}. F = χ c 1 1 − 1/ ( χ c 1 A − ) ( 1 + 1/ ( c 1 A − 2 ) ) 2 .
If u 1 = 0 u_1=0 u 1 = 0 , then u 3 ≠ 0 u_3\ne0 u 3 = 0 by coprimality and 2 m ≤ ∣ u 2 ∣ < F A 2 + 2m\le|u_2|<FA_2^+ 2 m ≤ ∣ u 2 ∣ < F A 2 + ; in every row F A 2 + < Q FA_2^+<Q F A 2 + < Q , which is impossible. So u 1 ≠ 0 u_1\ne0 u 1 = 0 ; let δ = gcd ( u 1 , u 1 − u 3 ) \delta=\gcd(u_1,u_1-u_3) δ = g cd( u 1 , u 1 − u 3 ) , which divides u 2 u_2 u 2 , and put q = ∣ u 1 ∣ / δ q=|u_1|/\delta q = ∣ u 1 ∣/ δ , a = sgn ( u 1 ) ( u 1 − u 3 ) / δ a=\operatorname{sgn}(u_1)(u_1-u_3)/\delta a = sgn ( u 1 ) ( u 1 − u 3 ) / δ and C r e l = − sgn ( u 1 ) u 2 / δ C_{\rm rel}=-\operatorname{sgn}(u_1)u_2/\delta C rel = − sgn ( u 1 ) u 2 / δ . Then
2 a m − 2 q l + C r e l = 0 , gcd ( a , q ) = 1 , 2am-2ql+C_{\rm rel}=0,\qquad\gcd(a,q)=1, 2 am − 2 q l + C rel = 0 , g cd( a , q ) = 1 ,
1 ≤ q ≤ q max : = ⌊ F A 1 + ⌋ , ∣ C r e l ∣ ≤ C max : = ⌊ F A 2 + ⌋ , 1\le q\le q_{\max}:=\lfloor FA_1^+\rfloor,\qquad|C_{\rm rel}|\le C_{\max}:=\lfloor FA_2^+\rfloor, 1 ≤ q ≤ q m a x := ⌊ F A 1 + ⌋ , ∣ C rel ∣ ≤ C m a x := ⌊ F A 2 + ⌋ ,
and a > 0 a>0 a > 0 because 2 a m = 2 q l − C r e l ≥ 2 l − ∣ C r e l ∣ > 0 2am=2ql-C_{\rm rel}\ge2l-|C_{\rm rel}|>0 2 am = 2 q l − C rel ≥ 2 l − ∣ C rel ∣ > 0 .
We rewrite − Λ x -\Lambda_x − Λ x as a linear form in two logarithms, centred at this relation. Choose p 0 ∈ Z p_0\in\mathbb{Z} p 0 ∈ Z with ∣ p 0 ∣ ≤ q / 2 |p_0|\le q/2 ∣ p 0 ∣ ≤ q /2 and a p 0 ≡ − C r e l ( m o d q ) ap_0\equiv-C_{\rm rel}\pmod q a p 0 ≡ − C rel ( mod q ) , and put k 0 = ( a p 0 + C r e l ) / q k_0=(ap_0+C_{\rm rel})/q k 0 = ( a p 0 + C rel ) / q and t = ( 2 m − p 0 ) / q t=(2m-p_0)/q t = ( 2 m − p 0 ) / q ; these are integers, t > 0 t>0 t > 0 and 2 l = k 0 + a t 2l=k_0+at 2 l = k 0 + a t . With
ζ = β q α − a , ξ = D λ − 1 β p 0 α − k 0 , \zeta=\beta^q\alpha^{-a},\qquad\xi=D_\lambda^{-1}\beta^{p_0}\alpha^{-k_0}, ζ = β q α − a , ξ = D λ − 1 β p 0 α − k 0 ,
one has exactly − Λ x = t log ζ + log ξ -\Lambda_x=t\log\zeta+\log\xi − Λ x = t log ζ + log ξ , log ζ = C r e l log α + q log D λ − q Λ x 2 m \log\zeta=\frac{C_{\rm rel}\log\alpha+q\log D_\lambda-q\Lambda_x}{2m} log ζ = 2 m C rel l o g α + q l o g D λ − q Λ x and log ξ = p 0 q log ζ − C r e l log α q − log D λ \log\xi=\frac{p_0}{q}\log\zeta-\frac{C_{\rm rel}\log\alpha}{q}-\log D_\lambda log ξ = q p 0 log ζ − q C rel l o g α − log D λ . Using log α < log c < U \log\alpha<\log c<U log α < log c < U (item 1 of Lemma 4.2 ), 0 < Λ x < log D λ < log 4 0<\Lambda_x<\log D_\lambda<\log4 0 < Λ x < log D λ < log 4 and h ( D λ ) < log c < U h(D_\lambda)<\log c<U h ( D λ ) < log c < U (item 4 of Lemma 2.18 and item 4 of Lemma 4.2 ), 2 m ≥ Q 2m\ge Q 2 m ≥ Q , the height formula of item 3 of Lemma 4.2 , a log α = q log β − log ζ a\log\alpha=q\log\beta-\log\zeta a log α = q log β − log ζ and ∣ k 0 ∣ ≤ a 2 + ∣ C r e l ∣ q |k_0|\le\frac a2+\frac{|C_{\rm rel}|}{q} ∣ k 0 ∣ ≤ 2 a + q ∣ C rel ∣ ,
∣ log ζ ∣ < Z q : = C max U + q log 4 Q , ∣ log ξ ∣ < X q : = log 4 + Z q 2 + C max U 2 q , t < t q : = Q 0 q + 1 2 , |\log\zeta|<Z_q:=\frac{C_{\max}U+q\log4}{Q},\qquad|\log\xi|<X_q:=\log4+\frac{Z_q}{2}+\frac{C_{\max}U}{2q},\qquad t<t_q:=\frac{Q_0}{q}+\frac12, ∣ log ζ ∣ < Z q := Q C m a x U + q log 4 , ∣ log ξ ∣ < X q := log 4 + 2 Z q + 2 q C m a x U , t < t q := q Q 0 + 2 1 ,
and h ( ζ ) = 1 2 max ( q log β , a log α ) ≤ H ζ ( log β ) h(\zeta)=\frac12\max(q\log\beta,a\log\alpha)\le H_\zeta(\log\beta) h ( ζ ) = 2 1 max ( q log β , a log α ) ≤ H ζ ( log β ) and h ( ξ ) ≤ H ξ ( log β ) h(\xi)\le H_\xi(\log\beta) h ( ξ ) ≤ H ξ ( log β ) , where for real Y Y Y
H ζ ( Y ) = 1 2 ( q Y + Z q ) , H ξ ( Y ) = U + q Y + Z q 4 + C max U 2 q ; H_\zeta(Y)=\frac12(qY+Z_q),\qquad H_\xi(Y)=U+\frac{qY+Z_q}{4}+\frac{C_{\max}U}{2q}; H ζ ( Y ) = 2 1 ( q Y + Z q ) , H ξ ( Y ) = U + 4 q Y + Z q + 2 q C m a x U ;
both increase with Y Y Y . By item 2 of Lemma 4.2 , the four embeddings of K \mathcal{K} K send the unit ζ \zeta ζ to the four numbers β ± q α ± a \beta^{\pm q}\alpha^{\pm a} β ± q α ± a , which are distinct by item 6, so ζ ≠ 1 \zeta\ne1 ζ = 1 generates K \mathcal{K} K ; and N K / Q ( ξ ) = N K / Q ( D λ ) − 1 ≠ ± 1 N_{\mathcal{K}/\mathbb{Q}}(\xi)=N_{\mathcal{K}/\mathbb{Q}}(D_\lambda)^{-1}\ne\pm1 N K / Q ( ξ ) = N K / Q ( D λ ) − 1 = ± 1 by item 4, while units have norm ± 1 \pm1 ± 1 , so ζ \zeta ζ and ξ \xi ξ are multiplicatively independent. If log ζ \log\zeta log ζ and log ξ \log\xi log ξ had the same sign, then Λ x ≥ ∣ log ξ ∣ \Lambda_x\ge|\log\xi| Λ x ≥ ∣ log ξ ∣ , and Lemma 7.7 , applied to the positive number ξ ≠ 1 \xi\ne1 ξ = 1 of degree at most four, would give log Λ x > − 4 H ξ ( log β ) − 4 ≥ − 4 H ξ ( U ) − 4 \log\Lambda_x>-4H_\xi(\log\beta)-4\ge-4H_\xi(U)-4 log Λ x > − 4 H ξ ( log β ) − 4 ≥ − 4 H ξ ( U ) − 4 ; but in every row 4 H ξ ( U ) + 4 < 2 Q ( L 0 2 + log 2 ) − 1.39 4H_\xi(U)+4<2Q\left(\frac{L_0}{2}+\log2\right)-1.39 4 H ξ ( U ) + 4 < 2 Q ( 2 L 0 + log 2 ) − 1.39 for every integer 1 ≤ q ≤ q max 1\le q\le q_{\max} 1 ≤ q ≤ q m a x , which contradicts (1). So the two logarithms have opposite signs. Replacing ζ \zeta ζ or ξ \xi ξ by its inverse, Λ x = b 2 log η 2 − b 1 log η 1 \Lambda_x=b_2\log\eta_2-b_1\log\eta_1 Λ x = b 2 log η 2 − b 1 log η 1 with { η 1 , η 2 } = { ζ ± 1 , ξ ± 1 } \{\eta_1,\eta_2\}=\{\zeta^{\pm1},\xi^{\pm1}\} { η 1 , η 2 } = { ζ ± 1 , ξ ± 1 } both greater than 1, the coefficient t t t on the number from ζ \zeta ζ and 1 on the number from ξ \xi ξ .
We finally apply Theorem 4.3 separately for each denominator q q q . Put B − = L 0 2 + log 2 B^-=\frac{L_0}{2}+\log2 B − = 2 L 0 + log 2 and B + = U B^+=U B + = U , and split [ B − , B + ] [B^-,B^+] [ B − , B + ] , which contains log β \log\beta log β , into sixteen equal subintervals [ B j , B j + 1 ] [B_j,B_{j+1}] [ B j , B j + 1 ] , 0 ≤ j ≤ 15 0\le j\le15 0 ≤ j ≤ 15 . For fixed q q q and j j j with log β ∈ [ B j , B j + 1 ] \log\beta\in[B_j,B_{j+1}] log β ∈ [ B j , B j + 1 ] , apply Theorem 4.3 with D = 4 \mathcal{D}=4 D = 4 , ρ = ρ q : = q + 2 \rho=\rho_q:=q+2 ρ = ρ q := q + 2 , μ = 0.63 \mu=0.63 μ = 0.63 , σ L = 1 + 2 μ − μ 2 2 \sigma_L=\frac{1+2\mu-\mu^2}{2} σ L = 2 1 + 2 μ − μ 2 , λ L = σ L log ρ q \lambda_L=\sigma_L\log\rho_q λ L = σ L log ρ q , with the numbers A ζ ≥ 8 H ζ ( B j + 1 ) + ( ρ q − 1 ) Z q + 1 A_\zeta\ge8H_\zeta(B_{j+1})+(\rho_q-1)Z_q+1 A ζ ≥ 8 H ζ ( B j + 1 ) + ( ρ q − 1 ) Z q + 1 and A ξ ≥ 8 H ξ ( B j + 1 ) + ( ρ q − 1 ) X q + 1 A_\xi\ge8H_\xi(B_{j+1})+(\rho_q-1)X_q+1 A ξ ≥ 8 H ξ ( B j + 1 ) + ( ρ q − 1 ) X q + 1 in the roles of the A i A_i A i of the numbers coming from ζ \zeta ζ and ξ \xi ξ , and with
h L ≥ max { 4 ( log ( 1 A ζ + t q A ξ ) + log λ L + 1.75 ) + 0.06 , λ L , 2 log 2 } . (12) h_L\ge\max\left\{4\left(\log\left(\frac{1}{A_\zeta}+\frac{t_q}{A_\xi}\right)+\log\lambda_L+1.75\right)+0.06,\ \lambda_L,\ 2\log2\right\}. \tag*{(12)} h L ≥ max { 4 ( log ( A ζ 1 + A ξ t q ) + log λ L + 1.75 ) + 0.06 , λ L , 2 log 2 } . (12)
each taken as the upper endpoint of its interval enclosure. These choices are admissible, since h ( η − 1 ) = h ( η ) h(\eta^{-1})=h(\eta) h ( η − 1 ) = h ( η ) and t < t q t<t_q t < t q . With H , ω L , θ L , C , C ′ \mathcal{H},\omega_L,\theta_L,\mathcal{C},\mathcal{C}' H , ω L , θ L , C , C ′ as in Theorem 4.3 (for A 1 A 2 = A ζ A ξ A_1A_2=A_\zeta A_\xi A 1 A 2 = A ζ A ξ ), the theorem gives log Λ x ≥ − E q , j \log\Lambda_x\geq-E_{q,j} log Λ x ≥ − E q , j , where
E q , j = C ( h L + log ρ q ) 2 A ζ A ξ + ω L θ L ( h L + log ρ q ) + log ( C ′ ( h L + log ρ q ) 2 A ζ A ξ ) E_{q,j}=\mathcal{C}(h_L+\log\rho_q)^2A_\zeta A_\xi+\sqrt{\omega_L\theta_L}(h_L+\log\rho_q)+\log\left(\mathcal{C}'(h_L+\log\rho_q)^2A_\zeta A_\xi\right) E q , j = C ( h L + log ρ q ) 2 A ζ A ξ + ω L θ L ( h L + log ρ q ) + log ( C ′ ( h L + log ρ q ) 2 A ζ A ξ )
For every integer 1 ≤ q ≤ q max 1\leq q\leq q_{\max} 1 ≤ q ≤ q m a x and every j j j , in every row, A ζ A ξ > λ L 2 A_\zeta A_\xi>\lambda_L^2 A ζ A ξ > λ L 2 and E q , j < 2 Q B j − 1.39 E_{q,j}<2QB_j-1.39 E q , j < 2 Q B j − 1.39 . Since 4 m log β ≥ 2 Q B j 4m\log\beta\geq2QB_j 4 m log β ≥ 2 Q B j , this contradicts (1).
The numerical values used above were computed in interval arithmetic of at least 160 bits with outward rounding, with A − A_- A − replaced by its lower bound 2 L 0 + 4 log 2 2L_0+4\log2 2 L 0 + 4 log 2 in the side checks and in F F F , which is legitimate because the side checks hold a fortiori and F F F decreases as A − A_- A − grows; each entry is an endpoint of an enclosure.
Table 2.
The largest ratio E q , j / ( 2 Q B j − 1.39 ) E_{q,j}/(2QB_j-1.39) E q , j / ( 2 Q B j − 1.39 ) is below 0.8476, 0.1490, 0.6476, 0.5902 in rows 1 to 4, and occurs at q = q max q=q_{\max} q = q m a x , j = 15 j=15 j = 15 . In every row the side checks hold, the bound in (C1) is below 7 ⋅ 10 8 7\cdot10^8 7 ⋅ 1 0 8 , F A + 2 < Q FA_+^2<Q F A + 2 < Q , and H ξ ( U ) < 1.63 ⋅ 10 6 H_\xi(U)<1.63\cdot10^6 H ξ ( U ) < 1.63 ⋅ 1 0 6 for 1 ≤ q ≤ q max 1\leq q\leq q_{\max} 1 ≤ q ≤ q m a x .
All alternatives of Theorem 7.6 are excluded, so 2 m < Q 2m<Q 2 m < Q in each row. Rows 1, 2, 3 and 4 give 2 m < 1.4 ⋅ 10 13 2m<1.4\cdot10^{13} 2 m < 1.4 ⋅ 1 0 13 , 2 m < 6.9 ⋅ 10 12 2m<6.9\cdot10^{12} 2 m < 6.9 ⋅ 1 0 12 , 2 m < 2.2 ⋅ 10 12 2m<2.2\cdot10^{12} 2 m < 2.2 ⋅ 1 0 12 and 2 m < 2 ⋅ 10 12 2m<2\cdot10^{12} 2 m < 2 ⋅ 1 0 12 , which are items 1, 2, 3 and 4.
Theorem 7.9. Let { a , b , c , d } \{a,b,c,d\} { a , b , c , d } be a Diophantine quadruple with a < b < c < d a<b<c<d a < b < c < d and c = a + b + 2 a b + 1 c=a+b+2\sqrt{ab+1} c = a + b + 2 ab + 1 . If d ≠ d + ( a , b , c ) d\neq d_+(a,b,c) d = d + ( a , b , c ) , then c < 4 ⋅ 10 48 c<4\cdot10^{48} c < 4 ⋅ 1 0 48 .
Proof. The proof is a descent: starting from c < 10 75 c<10^{75} c < 1 0 75 , each index cap of Proposition 7.8 gives, through the quartic index floor (Theorem 3.6 ), a smaller bound for c c c , until c < 4 ⋅ 10 48 c<4\cdot10^{48} c < 4 ⋅ 1 0 48 .
We first put the hypotheses in the language of index systems. Since { a , b , c , d } \{a,b,c,d\} { a , b , c , d } is a Diophantine quadruple, a b + 1 ab+1 ab + 1 is a square, so r = a b + 1 r=\sqrt{ab+1} r = ab + 1 is a positive integer and c = a + b + 2 r c=a+b+2r c = a + b + 2 r : the set { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple (Definition 2.3 ). It is a Diophantine triple and d > c d>c d > c , so d d d is an extension of { a , b , c } \{a,b,c\} { a , b , c } , irregular because d ≠ d + ( a , b , c ) d\neq d_+(a,b,c) d = d + ( a , b , c ) (Definition 2.2 ). Let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be its index system (Proposition 2.11).
We next record the form in which the floor is used. By Theorem 3.6 , c < 4 m 2 ( m + λ ) ≤ 4 m 2 ( m + 1 ) 2 c<4m^2(m+\lambda)\leq4m^2(m+1)^2 c < 4 m 2 ( m + λ ) ≤ 4 m 2 ( m + 1 ) 2 . Hence, for every positive integer Q Q Q ,
m < Q ⟹ c < 4 ( Q − 1 ) 2 Q 2 < 4 Q 4 , (13) m<Q\Longrightarrow c<4(Q-1)^2Q^2<4Q^4, \tag*{(13)} m < Q ⟹ c < 4 ( Q − 1 ) 2 Q 2 < 4 Q 4 , (13)
since then m ≤ Q − 1 m\leq Q-1 m ≤ Q − 1 and m ( m + 1 ) ≤ ( Q − 1 ) Q m(m+1)\leq(Q-1)Q m ( m + 1 ) ≤ ( Q − 1 ) Q .
We now descend. By Proposition 7.5 , c < 10 75 c<10^{75} c < 1 0 75 . If c < 10 48.6 c<10^{48.6} c < 1 0 48.6 , then c < 4 ⋅ 10 48 c<4\cdot10^{48} c < 4 ⋅ 1 0 48 , as 10 0.6 < 3.99 10^{0.6}<3.99 1 0 0.6 < 3.99 . Let c ≥ 10 48.6 c\geq10^{48.6} c ≥ 1 0 48.6 . Item 1 of Proposition 7.8 gives m < 7 ⋅ 10 12 m<7\cdot10^{12} m < 7 ⋅ 1 0 12 , so by (1)
c < 4 ( 7 ⋅ 10 12 ) 4 = 9.604 ⋅ 10 51 < 10 52 . c<4(7\cdot10^{12})^4=9.604\cdot10^{51}<10^{52}. c < 4 ( 7 ⋅ 1 0 12 ) 4 = 9.604 ⋅ 1 0 51 < 1 0 52 .
Item 2 of Proposition 7.8 now gives m < 3.45 ⋅ 10 12 m<3.45\cdot10^{12} m < 3.45 ⋅ 1 0 12 , so by (1)
c < 4 ( 3.45 ⋅ 10 12 ) 4 < 10 51 . c<4(3.45\cdot10^{12})^4<10^{51}. c < 4 ( 3.45 ⋅ 1 0 12 ) 4 < 1 0 51 .
Item 3 of Proposition 7.8 now gives m < 1.1 ⋅ 10 12 m < 1.1 \cdot10^{12} m < 1.1 ⋅ 1 0 12 , so by (1)
c < 4 ( 1.1 ⋅ 10 12 ) 4 < 10 48.77 . c < 4(1.1 \cdot10^{12})^4 < 10^{48.77}. c < 4 ( 1.1 ⋅ 1 0 12 ) 4 < 1 0 48.77 .
Item 4 of Proposition 7.8 now gives m < 10 12 m < 10^{12} m < 1 0 12 , and (1) gives
c < 4 ( 10 12 − 1 ) 2 ⋅ 10 24 < 4 ⋅ 10 48 . c < 4(10^{12}-1)^2 \cdot10^{24} < 4 \cdot10^{48}. c < 4 ( 1 0 12 − 1 ) 2 ⋅ 1 0 24 < 4 ⋅ 1 0 48 .
Here 10 0.6 < 3.99 10^{0.6} < 3.99 1 0 0.6 < 3.99 and 4 ( 1.1 ⋅ 10 12 ) 4 < 10 48.77 4(1.1 \cdot10^{12})^4 < 10^{48.77} 4 ( 1.1 ⋅ 1 0 12 ) 4 < 1 0 48.77 were verified in interval arithmetic with outward rounding; the other three comparisons are exact integer comparisons.
Every step of this descent uses the clause c < 4 m 2 ( m + λ ) 2 c < 4m^2(m+\lambda)^2 c < 4 m 2 ( m + λ ) 2 of the quartic floor, not only c < 8 m 4 c < 8m^4 c < 8 m 4 . From m < Q m < Q m < Q the weaker floor gives only c < 8 ( Q − 1 ) 4 c < 8(Q-1)^4 c < 8 ( Q − 1 ) 4 , which for Q = 7 ⋅ 10 12 Q = 7 \cdot10^{12} Q = 7 ⋅ 1 0 12 , 3.45 ⋅ 10 12 3.45 \cdot10^{12} 3.45 ⋅ 1 0 12 , 1.1 ⋅ 10 12 1.1 \cdot10^{12} 1.1 ⋅ 1 0 12 and 10 12 10^{12} 1 0 12 exceeds 10 52 10^{52} 1 0 52 , 10 51 10^{51} 1 0 51 , 10 48.77 10^{48.77} 1 0 48.77 and 4 ⋅ 10 48 4 \cdot10^{48} 4 ⋅ 1 0 48 , respectively; so none of the four steps would close, and the last would give only c < 8 ⋅ 10 48 c < 8 \cdot10^{48} c < 8 ⋅ 1 0 48 . The effective bound is used once more, in Section 8, to confine the family with K = 1 K = 1 K = 1 to finitely many triples.
Families of Euler triples that extend only regularly
We apply the pair-coordinate criterion (Theorem 5.19 ), the coprime divisors (Theorem 6.3 ) and the effective bound (Theorem 7.9 ) to two families of Euler triples: the three-parameter Fibonacci family of Cipu, Fujita and Mignotte [10 ] (Subsection 8.1), and the triples with two elements in a fixed ratio (Subsection 8.2). The following table lists the families of Euler triples whose regularity follows from our results, with what was known before.
Table 1 (PDF p. 60).
Here P k P_k P k denotes the k k k -th Pell number. Any further family all of whose members satisfy b ≥ 3355 2 a b \ge3355^2a b ≥ 335 5 2 a , or a c ≥ 3075 ( b − a ) \sqrt{ac} \ge3075(b-a) a c ≥ 3075 ( b − a ) , is covered by Theorem 4.5 directly. The proofs are completed in the appendices: Appendix A treats the range 643 ≤ K ≤ 2401 643 \le K \le2401 643 ≤ K ≤ 2401 of the three-parameter family, and Appendix B gives the reduction that settles the finitely many single triples left below the analytic bounds.
The three-parameter Fibonacci family
Convention 8.1. The three-parameter Fibonacci family is the family of [10 ] in the case ∣ ε ∣ = 1 |\varepsilon| = 1 ∣ ε ∣ = 1 , defined as follows. Let F − 1 = 1 F_{-1} = 1 F − 1 = 1 , F 0 = 0 F_0 = 0 F 0 = 0 and F k + 2 = F k + 1 + F k F_{k+2} = F_{k+1} + F_k F k + 2 = F k + 1 + F k . For integers ν ≥ 1 \nu\ge1 ν ≥ 1 , A ≥ 1 A \ge1 A ≥ 1 , K ≥ 1 K \ge1 K ≥ 1 and a sign ε ∈ { − 1 , 1 } \varepsilon\in\{-1,1\} ε ∈ { − 1 , 1 } put, for k ≥ 0 k \ge0 k ≥ 0 ,
ϕ k = F k A + F k − 1 , ψ k = ϕ k K + 2 ε F k , f k = ϕ k ψ k , \phi_k = F_k A + F_{k-1}, \qquad\psi_k = \phi_k K + 2\varepsilon F_k, \qquad f_k = \phi_k\psi_k, ϕ k = F k A + F k − 1 , ψ k = ϕ k K + 2 ε F k , f k = ϕ k ψ k ,
and X = F ν A X = F_\nu A X = F ν A . The quadruple ( ν , A , K , ε ) (\nu, A, K, \varepsilon) ( ν , A , K , ε ) is a parameter point , and the set { f ν , f ν + 1 , f ν + 2 } \{f_\nu, f_{\nu+1}, f_{\nu+2}\} { f ν , f ν + 1 , f ν + 2 } is its family triple ; when the family triple is an Euler triple, its elements are a = f ν a = f_\nu a = f ν , b = f ν + 1 b = f_{\nu+1} b = f ν + 1 , c = f ν + 2 c = f_{\nu+2} c = f ν + 2 .
Lemma 8.2. Let ( ν , A , K , ε ) (\nu, A, K, \varepsilon) ( ν , A , K , ε ) be a parameter point (Convention 8.1), and put
r = ϕ ν ϕ ν + 1 K + ε ( F ν ϕ ν + 1 + F ν + 1 ϕ ν ) . r = \phi_\nu\phi_{\nu+1}K + \varepsilon(F_\nu\phi_{\nu+1} + F_{\nu+1}\phi_\nu). r = ϕ ν ϕ ν + 1 K + ε ( F ν ϕ ν + 1 + F ν + 1 ϕ ν ) . f ν f ν + 1 + 1 = r 2 f_\nu f_{\nu+1} + 1 = r^2 f ν f ν + 1 + 1 = r 2 and f ν + 2 = f ν + f ν + 1 + 2 r f_{\nu+2} = f_\nu+ f_{\nu+1} + 2r f ν + 2 = f ν + f ν + 1 + 2 r .
If K ≥ 3 K \ge3 K ≥ 3 , or if K = 1 K = 1 K = 1 , A ≥ 2 A \ge2 A ≥ 2 and ν ≥ 2 \nu\ge2 ν ≥ 2 , then 0 < f ν < f ν + 1 0 < f_\nu< f_{\nu+1} 0 < f ν < f ν + 1 and r > 0 r > 0 r > 0 . Hence the family triple { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple with root r r r , and ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) (\phi_\nu,\psi_\nu,\phi_{\nu+1},\psi_{\nu+1}) ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) is a pair representation of it of sign ε ( − 1 ) ν \varepsilon(-1)^\nu ε ( − 1 ) ν , with pair coordinates ( p 3 , q 3 ) = ( ϕ ν + 2 , ψ ν + 2 ) (p_3,q_3) = (\phi_{\nu+2},\psi_{\nu+2}) ( p 3 , q 3 ) = ( ϕ ν + 2 , ψ ν + 2 ) .
If K ≥ 3 K \ge3 K ≥ 3 , then this pair representation satisfies p 1 q 3 ≥ K − 2 3 p 3 2 p_1q_3 \ge\frac{K-2}{3}p_3^2 p 1 q 3 ≥ 3 K − 2 p 3 2 .
Proof. We first record an identity for a symmetric bilinear form. For ξ = ( ξ 1 , ξ 2 ) \xi= (\xi_1,\xi_2) ξ = ( ξ 1 , ξ 2 ) and η = ( η 1 , η 2 ) \eta= (\eta_1,\eta_2) η = ( η 1 , η 2 ) in Z 2 \mathbb{Z}^2 Z 2 put
⟨ ξ , η ⟩ = K ξ 1 η 1 + ε ( ξ 1 η 2 + ξ 2 η 1 ) , [ ξ , η ] = ξ 1 η 2 − ξ 2 η 1 . \langle\xi,\eta\rangle= K\xi_1\eta_1 + \varepsilon(\xi_1\eta_2 + \xi_2\eta_1), \qquad[\xi,\eta] = \xi_1\eta_2 - \xi_2\eta_1. ⟨ ξ , η ⟩ = K ξ 1 η 1 + ε ( ξ 1 η 2 + ξ 2 η 1 ) , [ ξ , η ] = ξ 1 η 2 − ξ 2 η 1 .
Since ε 2 = 1 \varepsilon^2 = 1 ε 2 = 1 ,
⟨ ξ , η ⟩ 2 − ⟨ ξ , ξ ⟩ ⟨ η , η ⟩ = [ ξ , η ] 2 . \langle\xi,\eta\rangle^2 - \langle\xi,\xi\rangle\langle\eta,\eta\rangle= [\xi,\eta]^2. ⟨ ξ , η ⟩ 2 − ⟨ ξ , ξ ⟩ ⟨ η , η ⟩ = [ ξ , η ] 2 .
We next prove item 1. Put e k = ( ϕ k , F k ) e_k = (\phi_k,F_k) e k = ( ϕ k , F k ) for k ≥ 0 k \ge0 k ≥ 0 . Since F j + 2 = F j + 1 + F j F_{j+2} = F_{j+1} + F_j F j + 2 = F j + 1 + F j for j ≥ − 1 j \ge-1 j ≥ − 1 , we have ϕ k + 2 = ϕ k + 1 + ϕ k \phi_{k+2} = \phi_{k+1} + \phi_k ϕ k + 2 = ϕ k + 1 + ϕ k , hence e k + 2 = e k + 1 + e k e_{k+2} = e_{k+1} + e_k e k + 2 = e k + 1 + e k and ψ k + 2 = ψ k + 1 + ψ k \psi_{k+2} = \psi_{k+1} + \psi_k ψ k + 2 = ψ k + 1 + ψ k , for k ≥ 0 k \ge0 k ≥ 0 . Moreover f k = ϕ k ψ k = ⟨ e k , e k ⟩ f_k = \phi_k\psi_k = \langle e_k,e_k\rangle f k = ϕ k ψ k = ⟨ e k , e k ⟩ and r = ⟨ e ν , e ν + 1 ⟩ r = \langle e_\nu,e_{\nu+1}\rangle r = ⟨ e ν , e ν + 1 ⟩ . By Cassini’s identity F k − 1 F k + 1 − F k 2 = ( − 1 ) k F_{k-1}F_{k+1} - F_k^2 = (-1)^k F k − 1 F k + 1 − F k 2 = ( − 1 ) k , valid for k ≥ 0 k \ge0 k ≥ 0 with F − 1 = 1 F_{-1} = 1 F − 1 = 1 ,
[ e k , e k + 1 ] = ( F k A + F k − 1 ) F k + 1 − F k ( F k + 1 A + F k ) = ( − 1 ) k ( k ≥ 0 ) . [e_k,e_{k+1}] = (F_kA + F_{k-1})F_{k+1} - F_k(F_{k+1}A + F_k) = (-1)^k \qquad(k \ge0). [ e k , e k + 1 ] = ( F k A + F k − 1 ) F k + 1 − F k ( F k + 1 A + F k ) = ( − 1 ) k ( k ≥ 0 ) .
So the identity above gives r 2 − f ν f ν + 1 = 1 r^2 - f_\nu f_{\nu+1} = 1 r 2 − f ν f ν + 1 = 1 , and bilinearity gives f ν + 2 = ⟨ e ν + e ν + 1 , e ν + e ν + 1 ⟩ = f ν + f ν + 1 + 2 r f_{\nu+2} = \langle e_\nu+ e_{\nu+1},e_\nu+ e_{\nu+1}\rangle= f_\nu+ f_{\nu+1} + 2r f ν + 2 = ⟨ e ν + e ν + 1 , e ν + e ν + 1 ⟩ = f ν + f ν + 1 + 2 r .
We now show that ⟨ e k , e j ⟩ > 0 \langle e_k,e_j\rangle> 0 ⟨ e k , e j ⟩ > 0 for the pairs of indices needed below. For k ≥ 0 k \ge0 k ≥ 0 we have F k ≥ 0 F_k \ge0 F k ≥ 0 , ϕ k ≥ 1 \phi_k \ge1 ϕ k ≥ 1 and F k ≤ ϕ k F_k \le\phi_k F k ≤ ϕ k , since ϕ 0 = 1 \phi_0 = 1 ϕ 0 = 1 , ϕ 1 = A \phi_1 = A ϕ 1 = A and ϕ k = F k A + F k − 1 ≥ F k \phi_k = F_kA + F_{k-1} \ge F_k ϕ k = F k A + F k − 1 ≥ F k for k ≥ 1 k \ge1 k ≥ 1 . If K ≥ 3 K \ge3 K ≥ 3 , then for all k , j ≥ 0 k,j \ge0 k , j ≥ 0
⟨ e k , e j ⟩ ≥ K ϕ k ϕ j − ( F k ϕ j + F j ϕ k ) ≥ ϕ k ϕ j ( K − 2 ) > 0. \langle e_k,e_j\rangle\ge K\phi_k\phi_j - (F_k\phi_j + F_j\phi_k) \ge\phi_k\phi_j(K - 2) > 0. ⟨ e k , e j ⟩ ≥ K ϕ k ϕ j − ( F k ϕ j + F j ϕ k ) ≥ ϕ k ϕ j ( K − 2 ) > 0.
Let K = 1 K = 1 K = 1 and A ≥ 2 A \ge2 A ≥ 2 . For k ≥ 0 k \ge0 k ≥ 0 , ϕ k − F k = ( A − 1 ) F k + F k − 1 ≥ F k + F k − 1 = F k + 1 \phi_k - F_k = (A - 1)F_k + F_{k-1} \ge F_k + F_{k-1} = F_{k+1} ϕ k − F k = ( A − 1 ) F k + F k − 1 ≥ F k + F k − 1 = F k + 1 . If ε = 1 \varepsilon= 1 ε = 1 , then ⟨ e k , e j ⟩ ≥ ϕ k ϕ j ≥ 1 \langle e_k,e_j\rangle\ge\phi_k\phi_j \ge1 ⟨ e k , e j ⟩ ≥ ϕ k ϕ j ≥ 1 for k , j ≥ 0 k,j \ge0 k , j ≥ 0 . If ε = − 1 \varepsilon= -1 ε = − 1 , then for k , j ≥ 0 k,j \ge0 k , j ≥ 0
⟨ e k , e j ⟩ = ( ϕ k − F k ) ( ϕ j − F j ) − F k F j ≥ F k + 1 F j + 1 − F k F j , \langle e_k,e_j\rangle= (\phi_k - F_k)(\phi_j - F_j) - F_kF_j \ge F_{k+1}F_{j+1} - F_kF_j, ⟨ e k , e j ⟩ = ( ϕ k − F k ) ( ϕ j − F j ) − F k F j ≥ F k + 1 F j + 1 − F k F j ,
and the right side is positive unless k = j = 1 k = j = 1 k = j = 1 , because F i + 1 ≥ 1 F_{i+1} \ge1 F i + 1 ≥ 1 and F i + 1 ≥ F i F_{i+1} \ge F_i F i + 1 ≥ F i for i ≥ 0 i \ge0 i ≥ 0 , with equality in the second only for i = 1 i = 1 i = 1 . So, for K = 1 K = 1 K = 1 and A ≥ 2 A \ge2 A ≥ 2 , ⟨ e k , e j ⟩ > 0 \langle e_k,e_j\rangle> 0 ⟨ e k , e j ⟩ > 0 for all k , j ≥ 0 k,j \ge0 k , j ≥ 0 with ( k , j ) ≠ ( 1 , 1 ) (k,j) \ne(1,1) ( k , j ) = ( 1 , 1 ) .
We next prove item 2. The pairs ( ν , ν ) (\nu,\nu) ( ν , ν ) , ( ν + 1 , ν + 1 ) (\nu+ 1,\nu+ 1) ( ν + 1 , ν + 1 ) , ( ν , ν + 1 ) (\nu,\nu+ 1) ( ν , ν + 1 ) and ( ν − 1 , ν + 2 ) (\nu- 1,\nu+ 2) ( ν − 1 , ν + 2 ) have nonnegative entries, as ν ≥ 1 \nu\ge1 ν ≥ 1 , and none of them is ( 1 , 1 ) (1,1) ( 1 , 1 ) when ν ≥ 2 \nu\ge2 ν ≥ 2 ; so under either hypothesis the previous paragraph gives f ν > 0 f_\nu> 0 f ν > 0 , f ν + 1 > 0 f_{\nu+1} > 0 f ν + 1 > 0 , r > 0 r > 0 r > 0 and, since e ν + 1 − e ν = e ν − 1 e_{\nu+1} - e_\nu= e_{\nu-1} e ν + 1 − e ν = e ν − 1 and e ν + 1 + e ν = e ν + 2 e_{\nu+1} + e_\nu= e_{\nu+2} e ν + 1 + e ν = e ν + 2 ,
f ν + 1 − f ν = ⟨ e ν + 1 − e ν , e ν + 1 + e ν ⟩ = ⟨ e ν − 1 , e ν + 2 ⟩ > 0. f_{\nu+1} - f_\nu= \langle e_{\nu+1} - e_\nu,e_{\nu+1} + e_\nu\rangle= \langle e_{\nu-1},e_{\nu+2}\rangle> 0. f ν + 1 − f ν = ⟨ e ν + 1 − e ν , e ν + 1 + e ν ⟩ = ⟨ e ν − 1 , e ν + 2 ⟩ > 0.
By item 1, f ν f ν + 1 + 1 = r 2 f_{\nu}f_{\nu+1}+1=r^2 f ν f ν + 1 + 1 = r 2 with r r r a positive integer and f ν + 2 = f ν + f ν + 1 + 2 r > f ν + 1 f_{\nu+2}=f_{\nu}+f_{\nu+1}+2r>f_{\nu+1} f ν + 2 = f ν + f ν + 1 + 2 r > f ν + 1 ; so { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple with root r r r . Put ( p 1 , q 1 , p 2 , q 2 ) = ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) (p_1,q_1,p_2,q_2)=(\phi_{\nu},\psi_{\nu},\phi_{\nu+1},\psi_{\nu+1}) ( p 1 , q 1 , p 2 , q 2 ) = ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) . These are positive integers, since ϕ ν , ϕ ν + 1 ≥ 1 \phi_{\nu},\phi_{\nu+1}\ge1 ϕ ν , ϕ ν + 1 ≥ 1 and ψ ν = f ν / ϕ ν , ψ ν + 1 = f ν + 1 / ϕ ν + 1 \psi_{\nu}=f_{\nu}/\phi_{\nu},\psi_{\nu+1}=f_{\nu+1}/\phi_{\nu+1} ψ ν = f ν / ϕ ν , ψ ν + 1 = f ν + 1 / ϕ ν + 1 are positive. We have a = p 1 q 1 a=p_1q_1 a = p 1 q 1 , b = p 2 q 2 b=p_2q_2 b = p 2 q 2 and
p 1 q 2 − p 2 q 1 = 2 ε ( ϕ ν F ν + 1 − ϕ ν + 1 F ν ) = 2 ε [ e ν , e ν + 1 ] = 2 ε ( − 1 ) ν p_1q_2-p_2q_1=2\varepsilon(\phi_{\nu}F_{\nu+1}-\phi_{\nu+1}F_{\nu})=2\varepsilon[e_{\nu},e_{\nu+1}]=2\varepsilon(-1)^{\nu} p 1 q 2 − p 2 q 1 = 2 ε ( ϕ ν F ν + 1 − ϕ ν + 1 F ν ) = 2 ε [ e ν , e ν + 1 ] = 2 ε ( − 1 ) ν
so ( p 1 , q 1 , p 2 , q 2 ) (p_1,q_1,p_2,q_2) ( p 1 , q 1 , p 2 , q 2 ) is a pair representation of sign ε ( − 1 ) ν \varepsilon(-1)^{\nu} ε ( − 1 ) ν . Its pair coordinates are p 3 = ϕ ν + ϕ ν + 1 = ϕ ν + 2 p_3=\phi_{\nu}+\phi_{\nu+1}=\phi_{\nu+2} p 3 = ϕ ν + ϕ ν + 1 = ϕ ν + 2 and q 3 = ψ ν + ψ ν + 1 = ψ ν + 2 q_3=\psi_{\nu}+\psi_{\nu+1}=\psi_{\nu+2} q 3 = ψ ν + ψ ν + 1 = ψ ν + 2 .
We finally prove item 3. As K ≥ 3 K\ge3 K ≥ 3 , item 2 applies. Moreover ϕ ν − 1 ≤ ϕ ν \phi_{\nu-1}\le\phi_{\nu} ϕ ν − 1 ≤ ϕ ν : for ν = 1 \nu=1 ν = 1 this is 1 ≤ A 1\le A 1 ≤ A , and for ν ≥ 2 \nu\ge2 ν ≥ 2 , ϕ ν − ϕ ν − 1 = ϕ ν − 2 ≥ 1 \phi_{\nu}-\phi_{\nu-1}=\phi_{\nu-2}\ge1 ϕ ν − ϕ ν − 1 = ϕ ν − 2 ≥ 1 . Hence ϕ ν + 2 = 2 ϕ ν + ϕ ν − 1 ≤ 3 ϕ ν \phi_{\nu+2}=2\phi_{\nu}+\phi_{\nu-1}\le3\phi_{\nu} ϕ ν + 2 = 2 ϕ ν + ϕ ν − 1 ≤ 3 ϕ ν , and since F ν + 2 ≤ ϕ ν + 2 F_{\nu+2}\le\phi_{\nu+2} F ν + 2 ≤ ϕ ν + 2 ,
p 1 q 3 = ϕ ν ( ϕ ν + 2 K + 2 ε F ν + 2 ) ≥ ϕ ν ϕ ν + 2 ( K − 2 ) ≥ K − 2 3 ϕ ν + 2 2 = K − 2 3 p 3 2 p_1q_3=\phi_{\nu}\left(\phi_{\nu+2}K+2\varepsilon F_{\nu+2}\right)\ge\phi_{\nu}\phi_{\nu+2}(K-2)\ge\frac{K-2}{3}\phi_{\nu+2}^{2}=\frac{K-2}{3}p_3^{2} p 1 q 3 = ϕ ν ( ϕ ν + 2 K + 2 ε F ν + 2 ) ≥ ϕ ν ϕ ν + 2 ( K − 2 ) ≥ 3 K − 2 ϕ ν + 2 2 = 3 K − 2 p 3 2
Item 3 is the link with the pair-coordinate criterion: in the family, its hypothesis on p 1 q 3 / p 3 2 p_1q_3/p_3^2 p 1 q 3 / p 3 2 becomes a condition on K K K alone. The proof of the next corollary takes the range 643 ≤ K ≤ 2401 643\le K\le2401 643 ≤ K ≤ 2401 from Appendix § 8 .
Corollary 8.3. Let ( ν , A , K , ε ) (\nu,A,K,\varepsilon) ( ν , A , K , ε ) be a parameter point of the three-parameter Fibonacci family with ν A ≥ 2 \nu A\ge2 ν A ≥ 2 , and let { a , b , c } \{a,b,c\} { a , b , c } be its family triple. If
K ≥ 643 , or K ≥ 128 and c ≥ 4.5 ⋅ 10 12 , K\ge643,\qquad\text{or}\qquad K\ge128\ \text{ and }\ c\ge4.5\cdot10^{12}, K ≥ 643 , or K ≥ 128 and c ≥ 4.5 ⋅ 1 0 12 , then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Proof. Under either hypothesis K ≥ 128 K\ge128 K ≥ 128 , so by Lemma 8.2 , items 2 and 3, { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple with a pair representation satisfying p 1 q 3 ≥ K − 2 3 p 3 2 p_1q_3\ge\frac{K-2}{3}p_3^2 p 1 q 3 ≥ 3 K − 2 p 3 2 . Hence p 1 q 3 ≥ 42 p 3 2 p_1q_3\ge42p_3^2 p 1 q 3 ≥ 42 p 3 2 , and p 1 q 3 ≥ 800 p 3 2 p_1q_3\ge800p_3^2 p 1 q 3 ≥ 800 p 3 2 if K ≥ 2402 K\ge2402 K ≥ 2402 ; so, by Theorem 5.19 , { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension if K ≥ 2402 K\ge2402 K ≥ 2402 (first case) or if c ≥ 4.5 ⋅ 10 12 c\ge4.5\cdot10^{12} c ≥ 4.5 ⋅ 1 0 12 (second case). Otherwise 643 ≤ K ≤ 2401 643\le K\le2401 643 ≤ K ≤ 2401 , and the claim is Theorem A.5 .
For K = 1 K=1 K = 1 item 3 of Lemma 8.2 does not apply, and the next theorem is proved through the effective bound instead: together with b > 4000 b>4000 b > 4000 , it confines ν \nu ν , for each of the four pairs ( A , ε ) (A,\varepsilon) ( A , ε ) , to a finite range, whose 427 triples are settled by the reduction of Appendix § 10 .
Theorem 8.4. Let A ∈ { 2 , 3 } A\in\{2,3\} A ∈ { 2 , 3 } , ε ∈ { − 1 , 1 } \varepsilon\in\{-1,1\} ε ∈ { − 1 , 1 } and ν ≥ 2 \nu\ge2 ν ≥ 2 , and let { a , b , c } \{a,b,c\} { a , b , c } be the family triple of the parameter point ( ν , A , 1 , ε ) (\nu,A,1,\varepsilon) ( ν , A , 1 , ε ) , that is,
a = ϕ ν ( ϕ ν + 2 ε F ν ) , b = ϕ ν + 1 ( ϕ ν + 1 + 2 ε F ν + 1 ) , c = ϕ ν + 2 ( ϕ ν + 2 + 2 ε F ν + 2 ) , a=\phi_{\nu}(\phi_{\nu}+2\varepsilon F_{\nu}),\qquad b=\phi_{\nu+1}(\phi_{\nu+1}+2\varepsilon F_{\nu+1}),\qquad c=\phi_{\nu+2}(\phi_{\nu+2}+2\varepsilon F_{\nu+2}), a = ϕ ν ( ϕ ν + 2 ε F ν ) , b = ϕ ν + 1 ( ϕ ν + 1 + 2 ε F ν + 1 ) , c = ϕ ν + 2 ( ϕ ν + 2 + 2 ε F ν + 2 ) , with ϕ k = F k + 2 \phi_k=F_{k+2} ϕ k = F k + 2 if A = 2 A=2 A = 2 and ϕ k = L k + 1 \phi_k=L_{k+1} ϕ k = L k + 1 if A = 3 A=3 A = 3 , where L k L_k L k denotes the k k k -th Lucas number. Then { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple, and it has no irregular extension.
Proof. For ( A , ε ) = ( 3 , − 1 ) (A,\varepsilon)=(3,-1) ( A , ε ) = ( 3 , − 1 ) one has ϕ k − 2 F k = L k + 1 − 2 F k = F k + 1 \phi_k-2F_k=L_{k+1}-2F_k=F_{k+1} ϕ k − 2 F k = L k + 1 − 2 F k = F k + 1 and L k + 1 F k + 1 = F 2 k + 2 L_{k+1}F_{k+1}=F_{2k+2} L k + 1 F k + 1 = F 2 k + 2 , so the family triples are the Hoggatt–Bergum triples { F 2 ν + 2 , F 2 ν + 4 , F 2 ν + 6 } \{F_{2\nu+2},F_{2\nu+4},F_{2\nu+6}\} { F 2 ν + 2 , F 2 ν + 4 , F 2 ν + 6 } (with k = ν + 1 k=\nu+1 k = ν + 1 ); for them the theorem is contained in Dujella’s theorem [12 ] : if { F 2 k , F 2 k + 2 , F 2 k + 4 , d } \{F_{2k},F_{2k+2},F_{2k+4},d\} { F 2 k , F 2 k + 2 , F 2 k + 4 , d } is a Diophantine quadruple, then d = 4 F 2 k + 1 F 2 k + 2 F 2 k + 3 d=4F_{2k+1}F_{2k+2}F_{2k+3} d = 4 F 2 k + 1 F 2 k + 2 F 2 k + 3 , which is d + { F 2 k , F 2 k + 2 , F 2 k + 4 } d+\{F_{2k},F_{2k+2},F_{2k+4}\} d + { F 2 k , F 2 k + 2 , F 2 k + 4 } . Filipin, Fujita and Togbé [17 ] proved the stronger statement that every Diophantine quadruple containing the pair { F 2 k , F 2 k + 2 } \{F_{2k},F_{2k+2}\} { F 2 k , F 2 k + 2 } is regular. For 2 ≤ ν ≤ 10 2\le\nu\le10 2 ≤ ν ≤ 10 the theorem is contained in the result of Cipiu, Fujita and Mignotte [10 ] that the family triples with 1 ≤ ν ≤ 10 1 \le\nu\le10 1 ≤ ν ≤ 10 and 1 ≤ A ≤ 10 1 \le A \le10 1 ≤ A ≤ 10 have no irregular extension, for every K K K . The argument below treats the four cases and every ν ≥ 2 \nu\ge2 ν ≥ 2 together.
We first record the Euler property and a monotonicity. By Lemma 8.2 , item 2 (with K = 1 K = 1 K = 1 , A ≥ 2 A \ge2 A ≥ 2 , ν ≥ 2 \nu\ge2 ν ≥ 2 ), { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple. The same item, applied to every parameter point ( k , A , 1 , ε ) (k,A,1,\varepsilon) ( k , A , 1 , ε ) with k ≥ 2 k \ge2 k ≥ 2 , gives f k < f k + 1 f_k < f_{k+1} f k < f k + 1 ; so ( f k ) k ≥ 2 (f_k)_{k \ge2} ( f k ) k ≥ 2 is strictly increasing, and b = f ν + 1 b = f_{\nu+1} b = f ν + 1 and c = f ν + 2 c = f_{\nu+2} c = f ν + 2 increase with ν \nu ν .
Suppose that { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension d d d . We next confine ν \nu ν to a finite range. By Lemma 2.14, applied to the index system of d d d (Proposition 2.11), b > 4000 b > 4000 b > 4000 . Let ν 1 = 6 , 8 , 6 , 7 \nu_1 = 6, 8, 6, 7 ν 1 = 6 , 8 , 6 , 7 for ( A , ε ) = ( 2 , 1 ) , ( 2 , − 1 ) , ( 3 , 1 ) , ( 3 , − 1 ) (A,\varepsilon) = (2,1), (2,-1), (3,1), (3,-1) ( A , ε ) = ( 2 , 1 ) , ( 2 , − 1 ) , ( 3 , 1 ) , ( 3 , − 1 ) , respectively. For ν = ν 1 \nu= \nu_1 ν = ν 1 we have b = 2040 , 1869 , 3431 , 2584 b = 2040, 1869, 3431, 2584 b = 2040 , 1869 , 3431 , 2584 , and for ν = ν 1 + 1 \nu= \nu_1 + 1 ν = ν 1 + 1 we have b = 5335 , 4896 , 8968 , 6765 b = 5335, 4896, 8968, 6765 b = 5335 , 4896 , 8968 , 6765 ; since b b b increases with ν \nu ν , we get ν > ν 1 \nu> \nu_1 ν > ν 1 . On the other hand { a , b , c , d } \{a,b,c,d\} { a , b , c , d } is a Diophantine quadruple with a < b < c < d a < b < c < d a < b < c < d , c = a + b + 2 a b + 1 c = a + b + 2\sqrt{ab + 1} c = a + b + 2 ab + 1 and d ≠ d + ( a , b , c ) d \ne d_+(a,b,c) d = d + ( a , b , c ) , so c < 4 ⋅ 10 48 c < 4 \cdot10^{48} c < 4 ⋅ 1 0 48 by Theorem 7.9 . Let ν E = 114 , 116 , 113 , 115 \nu_E = 114, 116, 113, 115 ν E = 114 , 116 , 113 , 115 for the four cases, respectively; at ν = ν E \nu= \nu_E ν = ν E the element c c c is about 7.39 ⋅ 10 48 , 6.78 ⋅ 10 48 , 4.75 ⋅ 10 48 , 9.37 ⋅ 10 48 7.39 \cdot10^{48}, 6.78 \cdot10^{48}, 4.75 \cdot10^{48}, 9.37 \cdot10^{48} 7.39 ⋅ 1 0 48 , 6.78 ⋅ 1 0 48 , 4.75 ⋅ 1 0 48 , 9.37 ⋅ 1 0 48 , so c ≥ 4 ⋅ 10 48 c \ge4 \cdot10^{48} c ≥ 4 ⋅ 1 0 48 for every ν ≥ ν E \nu\ge\nu_E ν ≥ ν E , as c c c increases with ν \nu ν . Hence ν 1 < ν < ν E \nu_1 < \nu< \nu_E ν 1 < ν < ν E .
We finally treat the remaining triples, by a finite computation. For each of the 10 7 , 10 7 , 10 6 , 10 7 10^7, 10^7, 10^6, 10^7 1 0 7 , 1 0 7 , 1 0 6 , 1 0 7 triples with ν 1 < ν < ν E \nu_1 < \nu< \nu_E ν 1 < ν < ν E , let M ≥ 1 M \ge1 M ≥ 1 be an integer just above the zero of Φ c \Phi_c Φ c in [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) , for which Φ c ( M ) > 0 \Phi_c(M) > 0 Φ c ( M ) > 0 was verified. For each λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } a number J λ ≤ 3 J_\lambda\le3 J λ ≤ 3 satisfying item 1 or item 2 of Proposition B.1 was found, and the exact check of that proposition (no exponent 2 m ≤ J λ 2m \le J_\lambda 2 m ≤ J λ gives an irregular extension) succeeds. For λ = − 1 \lambda= -1 λ = − 1 , the common term V 1 − 1 = W 1 − 1 V_1^{-1} = W_1^{-1} V 1 − 1 = W 1 − 1 , the number z z z of the regular extension (Lemma 2.12 ), makes the number μ − 1 \mu_{-1} μ − 1 of Proposition B.1 extremely close to 2 − 2 log β / log γ 2 - 2 \log\beta/ \log\gamma 2 − 2 log β / log γ ; in the cases where item 1 was not found, item 2 holds with ( m 0 , n 0 ) = ( 2 , 2 ) (m_0,n_0) = (2,2) ( m 0 , n 0 ) = ( 2 , 2 ) . All these evaluations were carried out in outward-rounded interval arithmetic with 170 decimal digits. By Proposition B.1, none of these triples has an irregular extension, which contradicts the existence of d d d . Hence { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Euler triples of small ratio type
The second family is defined by an arithmetic condition, two elements in a fixed ratio, and it is the application of Theorem 6.3 announced in § 5 . The next lemma describes these triples by a Pell equation, so that those below any bound for c c c are finite in number and can be listed.
Definition 8.5. Let j < k j < k j < k be coprime positive integers. An Euler triple { a , b , c } \{a,b,c\} { a , b , c } (Definition 2.3 ) has ratio type j : k j:k j : k if two of its elements v < w v < w v < w satisfy
Lemma 8.6. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with root r r r , of ratio type j : k j:k j : k (Definition 8.5 ) realised by elements v < w v < w v < w with j w = k v jw = kv j w = k v , and let g = gcd ( v , w ) g = \gcd(v,w) g = g cd( v , w ) . Then:
if ( v , w ) = ( b , c ) (v,w) = (b,c) ( v , w ) = ( b , c ) then k < 4 j k < 4j k < 4 j , and if ( v , w ) = ( a , c ) (v,w) = (a,c) ( v , w ) = ( a , c ) then k > 4 j k > 4j k > 4 j ;
there is a positive integer u u u with u 2 − j k g 2 = 1 u^2 - jkg^2 = 1 u 2 − j k g 2 = 1 , and
( v , w ) = ( a , b ) : r = u , c = ( j + k ) g + 2 u ; ( v , w ) = ( b , c ) : r = u − j g , a = ( j + k ) g − 2 u ; ( v , w ) = ( a , c ) : r = u − j g , b = ( j + k ) g − 2 u . \begin{aligned}
(v,w) = (a,b)&:\quad r = u,\quad c = (j+k)g + 2u;\\
(v,w) = (b,c)&:\quad r = u-jg,\quad a = (j+k)g - 2u;\\
(v,w) = (a,c)&:\quad r = u-jg,\quad b = (j+k)g - 2u.
\end{aligned} ( v , w ) = ( a , b ) ( v , w ) = ( b , c ) ( v , w ) = ( a , c ) : r = u , c = ( j + k ) g + 2 u ; : r = u − j g , a = ( j + k ) g − 2 u ; : r = u − j g , b = ( j + k ) g − 2 u .
In particular j k jk j k is not a square, and every Euler triple of ratio type j : k j:k j : k realised by a given pair of positions arises in this way from a positive solution ( u , g ) (u,g) ( u , g ) of u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 , that is, from u + g j k = ( u ∗ + g ∗ j k ) i u+g\sqrt{jk}=(u_*+g_*\sqrt{jk})^i u + g j k = ( u ∗ + g ∗ j k ) i with an integer i ≥ 1 i\ge1 i ≥ 1 , where ( u ∗ , g ∗ ) (u_*,g_*) ( u ∗ , g ∗ ) is the least positive solution.
Proof. Recall a b + 1 = r 2 ab+1=r^2 ab + 1 = r 2 and c = a + b + 2 r c=a+b+2r c = a + b + 2 r (Definition 2.3 ).
We first prove item 1. By Lemma 2.4 , item 4, a < r < b a<r<b a < r < b and c > 4 a c>4a c > 4 a ; hence b < c = a + b + 2 r < 4 b b<c=a+b+2r<4b b < c = a + b + 2 r < 4 b . If ( v , w ) = ( b , c ) (v,w)=(b,c) ( v , w ) = ( b , c ) then k / j = c / b < 4 k/j=c/b<4 k / j = c / b < 4 ; if ( v , w ) = ( a , c ) (v,w)=(a,c) ( v , w ) = ( a , c ) then k / j = c / a > 4 k/j=c/a>4 k / j = c / a > 4 .
We next prove item 2. Since j w = k v jw=kv j w = k v and gcd ( j , k ) = 1 \gcd(j,k)=1 g cd( j , k ) = 1 , j j j divides v v v ; writing v = j h v=jh v = j h gives w = k h w=kh w = k h , and gcd ( v , w ) = h gcd ( j , k ) = h \gcd(v,w)=h\gcd(j,k)=h g cd( v , w ) = h g cd( j , k ) = h , so v = j g v=jg v = j g and w = k g w=kg w = k g .
If ( v , w ) = ( a , b ) (v,w)=(a,b) ( v , w ) = ( a , b ) , then r 2 = a b + 1 = j k g 2 + 1 r^2=ab+1=jkg^2+1 r 2 = ab + 1 = j k g 2 + 1 , so u = r u=r u = r works and c = a + b + 2 r = ( j + k ) g + 2 u c=a+b+2r=(j+k)g+2u c = a + b + 2 r = ( j + k ) g + 2 u .
If ( v , w ) = ( b , c ) (v,w)=(b,c) ( v , w ) = ( b , c ) , then a = c − b − 2 r = ( k − j ) g − 2 r a=c-b-2r=(k-j)g-2r a = c − b − 2 r = ( k − j ) g − 2 r and
r 2 = a b + 1 = j g ( ( k − j ) g − 2 r ) + 1 , equivalently ( r + j g ) 2 = j k g 2 + 1. r^2=ab+1=jg((k-j)g-2r)+1,\qquad\text{equivalently}\qquad(r+jg)^2=jkg^2+1. r 2 = ab + 1 = j g (( k − j ) g − 2 r ) + 1 , equivalently ( r + j g ) 2 = j k g 2 + 1.
Put u = r + j g ≥ 1 u=r+jg\ge1 u = r + j g ≥ 1 . Then u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 , r = u − j g r=u-jg r = u − j g and a = ( k − j ) g − 2 u + 2 j g = ( j + k ) g − 2 u a=(k-j)g-2u+2jg=(j+k)g-2u a = ( k − j ) g − 2 u + 2 j g = ( j + k ) g − 2 u .
If ( v , w ) = ( a , c ) (v,w)=(a,c) ( v , w ) = ( a , c ) , then b = c − a − 2 r = ( k − j ) g − 2 r b=c-a-2r=(k-j)g-2r b = c − a − 2 r = ( k − j ) g − 2 r and r 2 = a b + 1 = j g ( ( k − j ) g − 2 r ) + 1 r^2=ab+1=jg((k-j)g-2r)+1 r 2 = ab + 1 = j g (( k − j ) g − 2 r ) + 1 , which is the same relation as before. Hence u = r + j g u=r+jg u = r + j g satisfies u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 , and b = ( j + k ) g − 2 u b=(j+k)g-2u b = ( j + k ) g − 2 u .
We finally prove the last assertion. Since g ≥ 1 g\ge1 g ≥ 1 and u 2 = j k g 2 + 1 u^2=jkg^2+1 u 2 = j k g 2 + 1 , the number j k jk j k is not a square: otherwise, with j k = e 2 jk=e^2 j k = e 2 , ( u − e g ) ( u + e g ) = 1 (u-eg)(u+eg)=1 ( u − e g ) ( u + e g ) = 1 with u + e g u+eg u + e g a positive integer, which forces u + e g = 1 u+eg=1 u + e g = 1 and so g = 0 g=0 g = 0 . Conversely, the displayed formulas express the triple through ( j , k , u , g ) (j,k,u,g) ( j , k , u , g ) , so every Euler triple of the given type realised by the given positions arises from a positive solution ( u , g ) (u,g) ( u , g ) of u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 . Since j k jk j k is not a square, the positive solutions of u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 are exactly the pairs with u + g j k = ( u ∗ + g ∗ j k ) i u+g\sqrt{jk}=(u_*+g_*\sqrt{jk})^i u + g j k = ( u ∗ + g ∗ j k ) i , i ≥ 1 i\ge1 i ≥ 1 , where ( u ∗ , g ∗ ) (u_*,g_*) ( u ∗ , g ∗ ) is the least positive solution; this is the classical description of the solutions of Pell’s equation (see [15 ] ). ▫
Theorem 8.7. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple of ratio type j : k j:k j : k . If { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension, then
c 6 ( j + k ) − 0.15 < 2.4638 ⋅ 10 12 ( log c ) 2 log 19.46 c 6 ( j + k ) . \frac{c}{6(j+k)}-0.15<2.4638\cdot10^{12}(\log c)^2\log\frac{19.46c}{6(j+k)}. 6 ( j + k ) c − 0.15 < 2.4638 ⋅ 1 0 12 ( log c ) 2 log 6 ( j + k ) 19.46 c . If j k ≤ 300 jk\le300 j k ≤ 300 , then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Proof. We bound m m m from below linearly in c c c by the coprime divisors (Theorem 6.3 ) and compare this bound with the a priori bound (Proposition 7.3 ); this proves the first assertion and, for c ≥ 10 21 c\ge10^{21} c ≥ 1 0 21 , the second, and the triples with j k ≤ 300 jk\le300 j k ≤ 300 and c < 10 21 c<10^{21} c < 1 0 21 , which are finitely many, are treated by a finite computation.
Suppose that { a , b , c } \{a,b,c\} { a , b , c } has an irregular extension d d d , and let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of d d d (Proposition 2.11). Let v < w v<w v < w be elements realising the ratio type (Definition 8.5 ), and g = gcd ( v , w ) g=\gcd(v,w) g = g cd( v , w ) , so that v = j g v=jg v = j g and w = k g w=kg w = k g because gcd ( j , k ) = 1 \gcd(j,k)=1 g cd( j , k ) = 1 .
We first prove 2 m > c / ( 6 ( j + k ) ) 2m>c/(6(j+k)) 2 m > c / ( 6 ( j + k )) . By Theorem 6.3 , Π 1 ∣ m − n \Pi_1\mid m-n Π 1 ∣ m − n , Π 2 ∣ l + λ m \Pi_2\mid l+\lambda m Π 2 ∣ l + λm , Π 3 ∣ l + λ n \Pi_3\mid l+\lambda n Π 3 ∣ l + λn and m > Π 1 ( 1 + R z ) m>\Pi_1(1+R_z) m > Π 1 ( 1 + R z ) . By Lemma 2.14, 3 ≤ n < m < l < 2 n 3\le n<m<l<2n 3 ≤ n < m < l < 2 n . As 3 ≤ n < m 3\le n<m 3 ≤ n < m and Π 1 ∣ m − n \Pi_1\mid m-n Π 1 ∣ m − n , we have m ≥ n + Π 1 ≥ 3 + Δ a b m\ge n+\Pi_1\ge3+\Delta_{ab} m ≥ n + Π 1 ≥ 3 + Δ ab . As l > m > n l>m>n l > m > n and l < 2 n < 2 m l<2n<2m l < 2 n < 2 m , we have 1 ≤ l + λ m ≤ l + m < 3 m 1\le l+\lambda m\le l+m<3m 1 ≤ l + λm ≤ l + m < 3 m and 2 ≤ l + λ n ≤ l + n < 3 m 2\le l+\lambda n\le l+n<3m 2 ≤ l + λn ≤ l + n < 3 m , so the divisibilities Π 2 ∣ l + λ m \Pi_2\mid l+\lambda m Π 2 ∣ l + λm and Π 3 ∣ l + λ n \Pi_3\mid l+\lambda n Π 3 ∣ l + λn give Π 2 < 3 m \Pi_2<3m Π 2 < 3 m and Π 3 < 3 m \Pi_3<3m Π 3 < 3 m . Since each gcd divisor divides the corresponding coprime divisor (Definition 6.1 ),
m > max ( Δ a b + 2 , 1 3 Δ b c , 1 3 Δ a c ) , Δ v w = gcd ( v , w ) gcd ( v , w , 4 ) ≥ g 4 . m>\max\left(\Delta_{ab}+2,\frac{1}{3}\Delta_{bc},\frac{1}{3}\Delta_{ac}\right),\qquad\Delta_{vw}=\frac{\gcd(v,w)}{\gcd(v,w,4)}\ge\frac{g}{4}. m > max ( Δ ab + 2 , 3 1 Δ b c , 3 1 Δ a c ) , Δ v w = g cd( v , w , 4 ) g cd( v , w ) ≥ 4 g .
If ( v , w ) = ( a , b ) (v,w)=(a,b) ( v , w ) = ( a , b ) , then c ≤ 2 ( a + b ) = 2 ( j + k ) g c\le2(a+b)=2(j+k)g c ≤ 2 ( a + b ) = 2 ( j + k ) g (Lemma 2.4 , item 4); hence m > g / 4 ≥ c / ( 8 ( j + k ) ) m>g/4\ge c/(8(j+k)) m > g /4 ≥ c / ( 8 ( j + k )) .
If ( v , w ) ∈ { ( b , c ) , ( a , c ) } (v,w)\in\{(b,c),(a,c)\} ( v , w ) ∈ {( b , c ) , ( a , c )} , then c = k g c=kg c = k g and m > g / 12 = c / ( 12 k ) m>g/12=c/(12k) m > g /12 = c / ( 12 k ) . In all cases 2 m > c / ( 6 ( j + k ) ) 2m>c/(6(j+k)) 2 m > c / ( 6 ( j + k )) . We next deduce the first assertion. Put Y = c / ( 6 ( j + k ) ) Y=c/(6(j+k)) Y = c / ( 6 ( j + k )) , and let Φ c \Phi_c Φ c and K E K_E K E be as in Convention 7.1. If Y ≥ 1 Y\ge1 Y ≥ 1 , then Φ c ( Y ) < 0 \Phi_c(Y)<0 Φ c ( Y ) < 0 , since otherwise Proposition 7.3 , applied with u 0 = Y u_0=Y u 0 = Y , would give 2 m < Y 2m<Y 2 m < Y ; and Φ c ( Y ) < 0 \Phi_c(Y)<0 Φ c ( Y ) < 0 is the inequality of the first assertion. Otherwise 1 12 < Y < 1 \frac{1}{12}<Y<1 12 1 < Y < 1 , because c ≥ w = k g ≥ k c\ge w=kg\ge k c ≥ w = k g ≥ k and j < k j<k j < k ; then the left side of the first assertion is less than 1 1 1 , while its right side is at least K E ( log c ) 3 19.46 12 > 1 K_E(\log c)^3\frac{19.46}{12}>1 K E ( log c ) 3 12 19.46 > 1 .
We now prove the second assertion for c ≥ 10 21 c\ge10^{21} c ≥ 1 0 21 . Let j k ≤ 300 jk\le300 j k ≤ 300 ; since ( j − 1 ) ( k − 1 ) ≥ 0 (j-1)(k-1)\ge0 ( j − 1 ) ( k − 1 ) ≥ 0 , we have j + k ≤ j k + 1 ≤ 301 j+k\le jk+1\le301 j + k ≤ j k + 1 ≤ 301 , so Y ≥ c / 1806 Y\ge c/1806 Y ≥ c /1806 . We apply Lemma 7.4 with ι = 1 / 1806 \iota=1/1806 ι = 1/1806 , θ = 1 \theta=1 θ = 1 and c 0 = 10 21 c_0=10^{21} c 0 = 1 0 21 . Its hypotheses hold: ι c 0 ≥ 1 \iota c_0\ge1 ι c 0 ≥ 1 ; log c 0 > 48.3 \log c_0>48.3 log c 0 > 48.3 and ∣ log ( 19.46 / 1806 ) ∣ < 4.6 \lvert\log(19.46/1806)\rvert<4.6 ∣ log ( 19.46/1806 )∣ < 4.6 ; an evaluation in outward-rounded interval arithmetic gives Φ c 0 ( c 0 / 1806 ) > 3.0 ⋅ 10 17 \Phi_{c_0}(c_0/1806)>3.0\cdot10^{17} Φ c 0 ( c 0 /1806 ) > 3.0 ⋅ 1 0 17 ; and c 0 / 1806 > 5.5 ⋅ 10 17 c_0/1806>5.5\cdot10^{17} c 0 /1806 > 5.5 ⋅ 1 0 17 exceeds K E log c 0 ( 2 log ( 19.46 c 0 / 1806 ) + log c 0 ) K_E\log c_0\left(2\log(19.46c_0/1806)+\log c_0\right) K E log c 0 ( 2 log ( 19.46 c 0 /1806 ) + log c 0 ) , which is less than 1.7 ⋅ 10 16 1.7\cdot10^{16} 1.7 ⋅ 1 0 16 . Hence Φ c ( c / 1806 ) > 0 \Phi_c(c/1806)>0 Φ c ( c /1806 ) > 0 for every c ≥ 10 21 c\ge10^{21} c ≥ 1 0 21 , and Proposition 7.3 , applied with u 0 = c / 1806 ≥ 1 u_0=c/1806\ge1 u 0 = c /1806 ≥ 1 , gives 2 m < c / 1806 ≤ Y 2m<c/1806\le Y 2 m < c /1806 ≤ Y , which contradicts 2 m > Y 2m>Y 2 m > Y . So no Euler triple of ratio type j : k j:k j : k with j k ≤ 300 jk\le300 j k ≤ 300 and c ≥ 10 21 c\ge10^{21} c ≥ 1 0 21 has an irregular extension.
Finally we treat c < 10 21 c<10^{21} c < 1 0 21 , by a finite computation. By Lemma 8.6 , j k jk j k is not a square, and each Euler triple of ratio type j : k j:k j : k arises from a positive solution of u 2 − j k g 2 = 1 u^2-jkg^2=1 u 2 − j k g 2 = 1 for one of the three pairs of positions. The triples with j k ≤ 300 jk\le300 j k ≤ 300 and c < 10 21 c<10^{21} c < 1 0 21 are therefore obtained from finitely many powers of the least positive solutions of these equations; there are 8 906 such triples. For each of them let M M M be the least integer M ≥ 1 M\ge1 M ≥ 1 with Φ c ( M ) ≥ 0 \Phi_c(M)\ge0 Φ c ( M ) ≥ 0 ; by Proposition 7.3 , an irregular extension would have 2 m < M 2m<M 2 m < M . For 1 048 of the triples, twice one of the lower bounds for m m m above, or, when ( v , w ) = ( a , b ) (v,w)=(a,b) ( v , w ) = ( a , b ) , twice the lower bound
Δ a b ( 1 + R z ) = Δ a b + a c gcd ( a , b , 4 ) ( k − j ) log β ≤ Π 1 ( 1 + R z ) < m \Delta_{ab}(1+R_z)=\Delta_{ab}+\frac{\sqrt{ac}}{\gcd(a,b,4)(k-j)}\log\beta\le\Pi_1(1+R_z)<m Δ ab ( 1 + R z ) = Δ ab + g cd( a , b , 4 ) ( k − j ) a c log β ≤ Π 1 ( 1 + R z ) < m
given by Theorem 6.3 (here b − a = ( k − j ) g = gcd ( a , b , 4 ) Δ a b ( k − j ) b-a=(k-j)g=\gcd(a,b,4)\Delta_{ab}(k-j) b − a = ( k − j ) g = g cd( a , b , 4 ) Δ ab ( k − j ) , and R z R_z R z is the gap ratio of Definition 2.16 ), is at least M M M ; so these triples have no irregular extension. For each of the remaining 7 858 triples and for both signs λ \lambda λ , a number J λ ≤ 11 J_\lambda\le11 J λ ≤ 11 satisfying item 1 or item 2 of Proposition B.1 was found, and the exact check of that proposition (no exponent 2 m ≤ J λ 2m\le J_\lambda 2 m ≤ J λ gives an irregular extension) succeeds; so these triples have no irregular extension either. Item 1 is the reduction of [16 Lemma 5(a)] . For λ = − 1 \lambda=-1 λ = − 1 , the common term V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 , which is the number z z z of the regular extension (Lemma 2.12 ), makes the number μ − 1 \mu_{-1} μ − 1 of Proposition B.1 extremely close to 2 − 2 log β / log γ 2-2\log\beta/\log\gamma 2 − 2 log β / log γ , so that item 1 may fail with denominators q q q of the size of M M M ; in the cases where item 1 was not found, item 2 holds with ( m 0 , n 0 ) = ( 2 , 2 ) (m_0,n_0)=(2,2) ( m 0 , n 0 ) = ( 2 , 2 ) . All these evaluations were carried out in outward-rounded interval arithmetic with 150 decimal digits. Hence no Euler triple of ratio type j : k j:k j : k with j k ≤ 300 jk\le300 j k ≤ 300 has an irregular extension.
The first assertion holds for every ratio type. Its left side grows linearly in c c c and its right side like ( log c ) 3 (\log c)^3 ( log c ) 3 , so for each j : k j:k j : k it fails for all large c c c , and by Lemma 8.6 only finitely many triples of ratio type j : k j:k j : k lie below that point: each ratio type is reduced to a finite computation. For j k ≤ 300 jk\le300 j k ≤ 300 the proof excludes all c ≥ 10 21 c\ge10^{21} c ≥ 1 0 21 in this way, and settles the triples below that bound one by one.
The three-parameter family for 643 ≤ K ≤ 2401 643\le K\le2401 643 ≤ K ≤ 2401
This appendix proves Theorem A.5 , which completes Corollary 8.3 in the range 643 ≤ K ≤ 2401 643\le K\le2401 643 ≤ K ≤ 2401 . It applies the results of Section 5 once more, to the pair representation of the family (Lemma 8.2 ), now with every quantity bounded in terms of the two parameters X = F v A X=F_vA X = F v A and K K K . The quartic floor (Theorem 3.6 ) and the index inequality Lemma 5.18 become a floor and a ceiling for $2m (Lemma A.2 ), the ceiling being the index function U ( X , K ) U(X,K) U ( X , K ) defined first. Lemma A.3 compares the two, and Lemma A.4 is an exhaustive search over the 1 681 604 parameter points with F ν A ≤ 144 F_{\nu}A \le144 F ν A ≤ 144 and 643 ≤ K ≤ 2401 643 \le K \le2401 643 ≤ K ≤ 2401 . The plan that combines them is stated at the head of the proof of Theorem A.5 .
Definition A.1
Definition A.1. With ω \omega ω as in Convention 5.13 and the exact constants
u 1 = 2.47 ⋅ 10 32 , u 2 = 2.72766 ω , u 3 = 0.498 ω 60 , u_{1}=2.47\cdot10^{32},\qquad u_{2}=2.72766\sqrt{\omega},\qquad u_{3}=\frac{0.498\omega}{60}, u 1 = 2.47 ⋅ 1 0 32 , u 2 = 2.72766 ω , u 3 = 60 0.498 ω ,
the index function of the three-parameter Fibonacci family (Convention 8.1) is the function of real X ≥ 1 X\ge1 X ≥ 1 and K ≥ 643 K\ge643 K ≥ 643 given by
U ( X , K ) = 2 log ( u 1 X 7 K ) log ( u 2 X 3 / 2 K ) log ( 4 X 2 K ) log ( u 3 K ) . U(X,K)=\frac{2\log(u_{1}X^{7}K)\log(u_{2}X^{3/2}K)}{\log(4X^{2}K)\log(u_{3}K)}. U ( X , K ) = log ( 4 X 2 K ) log ( u 3 K ) 2 log ( u 1 X 7 K ) log ( u 2 X 3/2 K ) .
These numerals are the constants of the ceiling for the index in Lemma A.2 : u 1 u_{1} u 1 is 1.46 ⋅ 10 28 ⋅ 10 125 ( 5 3 + 2 643 ) 1.46\cdot10^{28}\cdot10^{125\left(\frac{5}{3}+\frac{2}{643}\right)} 1.46 ⋅ 1 0 28 ⋅ 1 0 125 ( 3 5 + 643 2 ) rounded up, where 1.46 ⋅ 10 28 1.46\cdot10^{28} 1.46 ⋅ 1 0 28 is twice the constant 7.3 ⋅ 10 27 7.3\cdot10^{27} 7.3 ⋅ 1 0 27 of Lemma 5.18 ; u 2 / ω u_{2}/\sqrt{\omega} u 2 / ω is 0.498 30 0.498\sqrt{30} 0.498 30 rounded up; and u 3 = 0.498 ω / 60 u_{3}=0.498\omega/60 u 3 = 0.498 ω /60 .
Lemma A.2. Let ( ν , A , K , ε ) (\nu,A,K,\varepsilon) ( ν , A , K , ε ) be a parameter point (Convention 8.1) with ν A ≥ 2 \nu A\ge2 ν A ≥ 2 and K ≥ 643 K\ge643 K ≥ 643 , and let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of an irregular extension d d d of its family triple { a , b , c } \{a,b,c\} { a , b , c } . Then, with the index function U U U of Definition A.1,
2 m > ( 2 c ) 1 / 4 ≥ ( 8 X 2 ( K − 2 ) ) 1 / 4 2m>(2c)^{1/4}\ge\left(8X^{2}(K-2)\right)^{1/4} 2 m > ( 2 c ) 1/4 ≥ ( 8 X 2 ( K − 2 ) ) 1/4 ;
2 m < U ( X , K ) 2m<U(X,K) 2 m < U ( X , K ) .
Proof. Item 1 is Theorem 3.6 with a lower bound for c c c ; item 2 is the index inequality of Lemma 5.18 for the pair representation of the family, with each quantity in it bounded in terms of X X X and K K K .
We first fix the pair representation and record elementary bounds. By Lemma 8.2 , item 2, { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple and ( p 1 , q 1 , p 2 , q 2 ) = ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) (p_{1},q_{1},p_{2},q_{2})=(\phi_{\nu},\psi_{\nu},\phi_{\nu+1},\psi_{\nu+1}) ( p 1 , q 1 , p 2 , q 2 ) = ( ϕ ν , ψ ν , ϕ ν + 1 , ψ ν + 1 ) is a pair representation of it, with pair coordinates ( p 3 , q 3 ) = ( ϕ ν + 2 , ψ ν + 2 ) (p_{3},q_{3})=(\phi_{\nu+2},\psi_{\nu+2}) ( p 3 , q 3 ) = ( ϕ ν + 2 , ψ ν + 2 ) ; so a = f ν = p 1 q 1 a=f_{\nu}=p_{1}q_{1} a = f ν = p 1 q 1 and c = f ν + 2 = p 3 q 3 c=f_{\nu+2}=p_{3}q_{3} c = f ν + 2 = p 3 q 3 . Since ν ≥ 1 \nu\ge1 ν ≥ 1 ,
F ν ≤ F ν + 1 ≤ 2 F ν , 2 F ν ≤ F ν + 2 ≤ 3 F ν , F ν + 3 ≤ min { 5 F ν , 3 F ν + 1 , 5 3 F ν + 2 } , F_{\nu}\le F_{\nu+1}\le2F_{\nu},\qquad2F_{\nu}\le F_{\nu+2}\le3F_{\nu},\qquad F_{\nu+3}\le\min\left\{5F_{\nu},3F_{\nu+1},\frac{5}{3}F_{\nu+2}\right\}, F ν ≤ F ν + 1 ≤ 2 F ν , 2 F ν ≤ F ν + 2 ≤ 3 F ν , F ν + 3 ≤ min { 5 F ν , 3 F ν + 1 , 3 5 F ν + 2 } ,
and F k ≤ ϕ k ≤ F k + 1 A F_{k}\le\phi_{k}\le F_{k+1}A F k ≤ ϕ k ≤ F k + 1 A for k ≥ 0 k\ge0 k ≥ 0 . Hence
X ≤ ϕ ν , ϕ ν + 1 ≤ 3 X , 2 X + F ν + 1 ≤ ϕ ν + 2 ≤ min { 5 X , 3 ϕ ν } , p 1 p 2 p 3 ≤ 30 X 3 . X\le\phi_{\nu},\qquad\phi_{\nu+1}\le3X,\qquad2X+F_{\nu+1}\le\phi_{\nu+2}\le\min\{5X,3\phi_{\nu}\},\qquad p_{1}p_{2}p_{3}\le30X^{3}. X ≤ ϕ ν , ϕ ν + 1 ≤ 3 X , 2 X + F ν + 1 ≤ ϕ ν + 2 ≤ min { 5 X , 3 ϕ ν } , p 1 p 2 p 3 ≤ 30 X 3 .
Moreover q 3 = ϕ ν + 2 K + 2 ε F ν + 2 q_{3}=\phi_{\nu+2}K+2\varepsilon F_{\nu+2} q 3 = ϕ ν + 2 K + 2 ε F ν + 2 with 0 < F ν + 2 ≤ ϕ ν + 2 0<F_{\nu+2}\le\phi_{\nu+2} 0 < F ν + 2 ≤ ϕ ν + 2 and A K ≥ 643 AK\ge643 A K ≥ 643 , so
ϕ ν + 2 ( K − 2 ) ≤ q 3 ≤ F ν + 3 A K + 2 F ν + 2 ≤ ( 5 3 + 2 643 ) F ν + 2 A K ≤ 3 ( 5 3 + 2 643 ) X K . \phi_{\nu+2}(K-2)\le q_{3}\le F_{\nu+3}AK+2F_{\nu+2}\le\left(\frac{5}{3}+\frac{2}{643}\right)F_{\nu+2}AK\le3\left(\frac{5}{3}+\frac{2}{643}\right)XK. ϕ ν + 2 ( K − 2 ) ≤ q 3 ≤ F ν + 3 A K + 2 F ν + 2 ≤ ( 3 5 + 643 2 ) F ν + 2 A K ≤ 3 ( 3 5 + 643 2 ) X K .
We now prove item 1. By Theorem 3.6 , c < 8 m 4 c<8m^{4} c < 8 m 4 , that is 2 m > ( 2 c ) 1 / 4 2m>(2c)^{1/4} 2 m > ( 2 c ) 1/4 . By the bounds above, c = p 3 q 3 ≥ ϕ ν + 2 2 ( K − 2 ) ≥ 4 X 2 ( K − 2 ) c=p_{3}q_{3}\ge\phi_{\nu+2}^{2}(K-2)\ge4X^{2}(K-2) c = p 3 q 3 ≥ ϕ ν + 2 2 ( K − 2 ) ≥ 4 X 2 ( K − 2 ) , which gives the second inequality.
We next apply Lemma 5.18. Since K ≥ 643 K \ge643 K ≥ 643 , Lemma 8.2, item 3, gives p 1 q 3 ≥ K − 2 3 p 3 2 ≥ 213 p 3 2 p_{1}q_{3} \ge\frac{K-2}{3}p_{3}^{2} \ge213p_{3}^{2} p 1 q 3 ≥ 3 K − 2 p 3 2 ≥ 213 p 3 2 ; as p 3 ≥ 2 X ≥ 2 p_{3} \ge2X \ge2 p 3 ≥ 2 X ≥ 2 , this gives p 1 q 3 ≥ 852 > 168 p_{1}q_{3} \ge852 > 168 p 1 q 3 ≥ 852 > 168 and ω p 1 q 3 ≥ 213 ω p 3 2 > 40 p 3 2 \omega p_{1}q_{3} \ge213\omega p_{3}^{2} > 40p_{3}^{2} ω p 1 q 3 ≥ 213 ω p 3 2 > 40 p 3 2 . Put
P = 10 p 2 p 3 q 3 , L = ω p 1 p 2 q 3 2 4 p 3 , Ξ = 7.3 ⋅ 10 27 p 2 3 p 3 3 q 3 , g = log L log ( L / P ) . P = 10p_{2}p_{3}q_{3}, \qquad L = \frac{\omega p_{1}p_{2}q_{3}^{2}}{4p_{3}}, \qquad\Xi= 7.3 \cdot10^{27}p_{2}^{3}p_{3}^{3}q_{3}, \qquad g = \frac{\log L}{\log(L/P)}. P = 10 p 2 p 3 q 3 , L = 4 p 3 ω p 1 p 2 q 3 2 , Ξ = 7.3 ⋅ 1 0 27 p 2 3 p 3 3 q 3 , g = log ( L / P ) log L .
so that L / P = ω p 1 q 3 / ( 40 p 3 2 ) > 1 L/P = \omega p_{1}q_{3}/(40p_{3}^{2}) > 1 L / P = ω p 1 q 3 / ( 40 p 3 2 ) > 1 . Then Lemma 5.18 gives m log ( 4 a c ) < log 2 + g log Ξ m\log(4ac) < \log2 + g\log\Xi m log ( 4 a c ) < log 2 + g log Ξ . Since P > 1 P > 1 P > 1 and L / P > 1 L/P > 1 L / P > 1 we have g > 1 g > 1 g > 1 , and Ξ > 1 \Xi> 1 Ξ > 1 ; hence
m log ( 4 a c ) < g log 2 + g log Ξ = g log ( 2 Ξ ) . m\log(4ac) < g\log2 + g\log\Xi= g\log(2\Xi). m log ( 4 a c ) < g log 2 + g log Ξ = g log ( 2Ξ ) .
We bound g g g . With the hypergeometric parameter N = 1 2 p 1 p 2 q 3 N = \frac{1}{2}p_{1}p_{2}q_{3} N = 2 1 p 1 p 2 q 3 and S 3 = p 1 p 2 p 3 S_{3} = p_{1}p_{2}p_{3} S 3 = p 1 p 2 p 3 one has L = ω N 2 / S 3 L = \omega N^{2}/S_{3} L = ω N 2 / S 3 and L / P = ω N / ( 20 p 2 p 3 2 ) L/P = \omega N/(20p_{2}p_{3}^{2}) L / P = ω N / ( 20 p 2 p 3 2 ) . For fixed p 1 , p 2 , p 3 p_{1},p_{2},p_{3} p 1 , p 2 , p 3 , the function
N ′ ↦ log ( ω N ′ 2 / S 3 ) log ( ω N ′ / ( 20 p 2 p 3 2 ) ) N' \mapsto\frac{\log(\omega N'^{2}/S_{3})}{\log(\omega N'/(20p_{2}p_{3}^{2}))} N ′ ↦ log ( ω N ′ / ( 20 p 2 p 3 2 )) log ( ω N ′ 2 / S 3 )
of real N ′ > 0 N' > 0 N ′ > 0 is decreasing where its denominator is positive, since there its derivative has the sign of log ( ω p 1 / ( 400 p 2 p 3 3 ) ) < 0 \log(\omega p_{1}/(400p_{2}p_{3}^{3})) < 0 log ( ω p 1 / ( 400 p 2 p 3 3 )) < 0 .
Since q 3 ≥ p 3 ( K − 2 ) q_{3} \ge p_{3}(K-2) q 3 ≥ p 3 ( K − 2 ) and K 2 − 1 ≥ 0.498 K \frac{K}{2}-1 \ge0.498K 2 K − 1 ≥ 0.498 K (as K ≥ 500 K \ge500 K ≥ 500 ), N ≥ 1 2 S 3 ( K − 2 ) ≥ N 0 : = 0.498 S 3 K N \ge\frac{1}{2}S_{3}(K-2) \ge N_{0} := 0.498S_{3}K N ≥ 2 1 S 3 ( K − 2 ) ≥ N 0 := 0.498 S 3 K . At N ′ = N 0 N' = N_{0} N ′ = N 0 the denominator is log ( 0.498 ω p 1 K / ( 20 p 3 ) ) ≥ log ( 0.498 ω K / 60 ) = log ( u 3 K ) \log(0.498\omega p_{1}K/(20p_{3})) \ge\log(0.498\omega K/60) = \log(u_{3}K) log ( 0.498 ω p 1 K / ( 20 p 3 )) ≥ log ( 0.498 ω K /60 ) = log ( u 3 K ) , by p 3 ≤ 3 p 1 p_{3} \le3p_{1} p 3 ≤ 3 p 1 , and u 3 K ≥ 643 u 3 > 7 u_{3}K \ge643u_{3} > 7 u 3 K ≥ 643 u 3 > 7 . So the denominator is positive on [ N 0 , ∞ ) [N_{0},\infty) [ N 0 , ∞ ) , and, as 0.498 2 ω S 3 K 2 > 1 0.498^{2}\omega S_{3}K^{2} > 1 0.49 8 2 ω S 3 K 2 > 1 ,
g ≤ log ( ω N 0 2 / S 3 ) log ( u 3 K ) = log ( 0.498 2 ω S 3 K 2 ) log ( u 3 K ) ≤ 2 log ( u 2 X 3 / 2 K ) log ( u 3 K ) , g \le\frac{\log(\omega N_{0}^{2}/S_{3})}{\log(u_{3}K)} = \frac{\log(0.498^{2}\omega S_{3}K^{2})}{\log(u_{3}K)} \le\frac{2\log(u_{2}X^{3/2}K)}{\log(u_{3}K)}, g ≤ log ( u 3 K ) log ( ω N 0 2 / S 3 ) = log ( u 3 K ) log ( 0.49 8 2 ω S 3 K 2 ) ≤ log ( u 3 K ) 2 log ( u 2 X 3/2 K ) ,
by S 3 ≤ 30 X 3 S_{3} \le30X^{3} S 3 ≤ 30 X 3 and 0.498 30 < 2.72766 0.498\sqrt{30} < 2.72766 0.498 30 < 2.72766 .
We bound Ξ \Xi Ξ . By the bounds above, p 2 2 p 3 3 q 3 ≤ ( 3 X ) 3 ( 5 X ) 3 ⋅ 3 ( 5 3 + 2 643 ) X K = 10125 ( 5 3 + 2 643 ) X 7 K p_{2}^{2}p_{3}^{3}q_{3} \le(3X)^{3}(5X)^{3}\cdot3\left(\frac{5}{3}+\frac{2}{643}\right)XK = 10125\left(\frac{5}{3}+\frac{2}{643}\right)X^{7}K p 2 2 p 3 3 q 3 ≤ ( 3 X ) 3 ( 5 X ) 3 ⋅ 3 ( 3 5 + 643 2 ) X K = 10125 ( 3 5 + 643 2 ) X 7 K , so
2 Ξ ≤ 1.46 ⋅ 10 28 ⋅ 10125 ( 5 3 + 2 643 ) X 7 K < 2.47 ⋅ 10 32 X 7 K = u 1 X 7 K . 2\Xi\le1.46\cdot10^{28}\cdot10125\left(\frac{5}{3}+\frac{2}{643}\right)X^{7}K < 2.47\cdot10^{32}X^{7}K = u_{1}X^{7}K. 2Ξ ≤ 1.46 ⋅ 1 0 28 ⋅ 10125 ( 3 5 + 643 2 ) X 7 K < 2.47 ⋅ 1 0 32 X 7 K = u 1 X 7 K .
We bound 4 a c 4ac 4 a c from below. Write Y = ϕ ν ϕ ν + 2 Y = \phi_{\nu}\phi_{\nu+2} Y = ϕ ν ϕ ν + 2 . Then a c = f ν f ν + 2 = ( p 1 q 3 ) ( p 3 q 1 ) ac = f_{\nu}f_{\nu+2} = (p_{1}q_{3})(p_{3}q_{1}) a c = f ν f ν + 2 = ( p 1 q 3 ) ( p 3 q 1 ) with
p 1 q 3 = Y K + 2 ϵ ϕ ν F ν + 2 , p 3 q 1 = Y K + 2 ϵ ϕ ν + 2 F ν . p_{1}q_{3} = YK + 2\epsilon\phi_{\nu}F_{\nu+2}, \qquad p_{3}q_{1} = YK + 2\epsilon\phi_{\nu+2}F_{\nu}. p 1 q 3 = Y K + 2 ϵ ϕ ν F ν + 2 , p 3 q 1 = Y K + 2 ϵ ϕ ν + 2 F ν .
By the bounds above, Y ≥ X ( 2 X + F ν + 1 ) Y \ge X(2X+F_{\nu+1}) Y ≥ X ( 2 X + F ν + 1 ) , 2 ϕ ν F ν + 2 ≤ 2 F ν + 1 A ⋅ 3 F ν = 6 F ν + 1 X 2\phi_{\nu}F_{\nu+2} \le2F_{\nu+1}A\cdot3F_{\nu} = 6F_{\nu+1}X 2 ϕ ν F ν + 2 ≤ 2 F ν + 1 A ⋅ 3 F ν = 6 F ν + 1 X and 2 ϕ ν + 2 F ν ≤ 2 F ν + 3 A F ν ≤ 6 F ν + 1 X 2\phi_{\nu+2}F_{\nu} \le2F_{\nu+3}AF_{\nu} \le6F_{\nu+1}X 2 ϕ ν + 2 F ν ≤ 2 F ν + 3 A F ν ≤ 6 F ν + 1 X . So each factor is at least 2 X 2 K + F ν + 1 X ( K − 6 ) ≥ 2 X 2 K 2X^{2}K + F_{\nu+1}X(K-6) \ge2X^{2}K 2 X 2 K + F ν + 1 X ( K − 6 ) ≥ 2 X 2 K , hence 4 a c ≥ 16 X 4 K 2 4ac \ge16X^{4}K^{2} 4 a c ≥ 16 X 4 K 2 , that is log ( 4 a c ) ≥ 2 log ( 4 X 2 K ) > 0 \log(4ac) \ge2\log(4X^{2}K) > 0 log ( 4 a c ) ≥ 2 log ( 4 X 2 K ) > 0 .
We conclude. Combining the three bounds,
2 m log ( 4 X 2 K ) ≤ m log ( 4 a c ) < g log ( 2 Ξ ) ≤ 2 log ( u 2 X 3 / 2 K ) log ( u 3 K ) log ( u 1 X 7 K ) , 2m\log(4X^{2}K) \le m\log(4ac) < g\log(2\Xi) \le\frac{2\log(u_{2}X^{3/2}K)}{\log(u_{3}K)}\log(u_{1}X^{7}K), 2 m log ( 4 X 2 K ) ≤ m log ( 4 a c ) < g log ( 2Ξ ) ≤ log ( u 3 K ) 2 log ( u 2 X 3/2 K ) log ( u 1 X 7 K ) ,
that is, 2 m < U ( X , K ) 2m < U(X,K) 2 m < U ( X , K ) .
Lemma A.3. Let U U U be the index function of the three-parameter family (Definition A.1).
For every real X ≥ 1 X \ge1 X ≥ 1 , the function K ↦ U ( X , K ) K \mapsto U(X,K) K ↦ U ( X , K ) is positive and decreasing on [ 643 , ∞ ) [643,\infty) [ 643 , ∞ ) .
For all real X ≥ 145 X \ge145 X ≥ 145 and K ≥ 643 K \ge643 K ≥ 643 , ( 8 X 2 ( K − 2 ) ) 1 / 4 > U ( X , K ) (8X^2(K-2))^{1/4} > \mathcal{U}(X,K) ( 8 X 2 ( K − 2 ) ) 1/4 > U ( X , K ) .
For all real X X X and K K K with 1 ≤ X ≤ 144 1 \le X \le144 1 ≤ X ≤ 144 and K ≥ 643 K \ge643 K ≥ 643 , U ( X , K ) ≤ U ( 144 , 643 ) < 100.6 \mathcal{U}(X,K) \le\mathcal{U}(144,643) < 100.6 U ( X , K ) ≤ U ( 144 , 643 ) < 100.6 .
Proof. We first prove item 1. For constants C 1 > C 2 > 0 C_1 > C_2 > 0 C 1 > C 2 > 0 , the function K ↦ log ( C 1 K ) / log ( C 2 K ) K \mapsto\log(C_1K)/\log(C_2K) K ↦ log ( C 1 K ) / log ( C 2 K ) is positive where C 2 K > 1 C_2K > 1 C 2 K > 1 , and there its derivative has the sign of log ( C 2 / C 1 ) < 0 \log(C_2/C_1) < 0 log ( C 2 / C 1 ) < 0 . Write U = 2 Q 1 Q 2 \mathcal{U} = 2Q_1Q_2 U = 2 Q 1 Q 2 with
Q 1 ( X , K ) = log ( u 1 X 7 K ) log ( 4 X 2 K ) , Q 2 ( X , K ) = log ( u 2 X 3 / 2 K ) log ( u 3 K ) . Q_1(X,K) = \frac{\log(u_1X^7K)}{\log(4X^2K)}, \qquad Q_2(X,K) = \frac{\log(u_2X^{3/2}K)}{\log(u_3K)}. Q 1 ( X , K ) = log ( 4 X 2 K ) log ( u 1 X 7 K ) , Q 2 ( X , K ) = log ( u 3 K ) log ( u 2 X 3/2 K ) .
For X ≥ 1 X \ge1 X ≥ 1 and K ≥ 643 K \ge643 K ≥ 643 we have u 1 X 7 > 4 X 2 u_1X^7 > 4X^2 u 1 X 7 > 4 X 2 , 4 X 2 K > 1 4X^2K > 1 4 X 2 K > 1 , u 2 X 3 / 2 ≥ u 2 > u 3 u_2X^{3/2} \ge u_2 > u_3 u 2 X 3/2 ≥ u 2 > u 3 and u 3 K ≥ 643 u 3 > 7 u_3K \ge643u_3 > 7 u 3 K ≥ 643 u 3 > 7 . So both factors are positive and decreasing in K K K on [ 643 , ∞ ) [643,\infty) [ 643 , ∞ ) , and so is U ( X , ⋅ ) \mathcal{U}(X,\cdot) U ( X , ⋅ ) .
We next prove item 2. For p > q > 0 p > q > 0 p > q > 0 and C 1 , C 2 > 0 C_1,C_2 > 0 C 1 , C 2 > 0 , the derivative of the function X ↦ log ( C 1 X p ) / log ( C 2 X q ) X \mapsto\log(C_1X^p)/\log(C_2X^q) X ↦ log ( C 1 X p ) / log ( C 2 X q ) , where C 2 X q > 1 C_2X^q > 1 C 2 X q > 1 , has the sign of log ( C 2 p / C 1 q ) \log(C_2^p/C_1^q) log ( C 2 p / C 1 q ) . With C 1 = 643 u 1 C_1 = 643u_1 C 1 = 643 u 1 , C 2 = 4 ⋅ 643 C_2 = 4 \cdot643 C 2 = 4 ⋅ 643 , p = 7 p = 7 p = 7 , q = 2 q = 2 q = 2 this sign is that of log ( 4 7 ⋅ 643 5 / u 1 2 ) = − 107.13 … < 0 \log(4^7 \cdot643^5/u_1^2) = -107.13\ldots< 0 log ( 4 7 ⋅ 64 3 5 / u 1 2 ) = − 107.13 … < 0 , so Q 1 ( ⋅ , 643 ) Q_1(\cdot,643) Q 1 ( ⋅ , 643 ) decreases on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) . Hence, by item 1, for X ≥ 145 X \ge145 X ≥ 145 and K ≥ 643 K \ge643 K ≥ 643 ,
U ( X , K ) ≤ U ( X , 643 ) ≤ 2 Q 1 ( 145 , 643 ) Q 2 ( X , 643 ) = : H ( X ) . \mathcal{U}(X,K) \le\mathcal{U}(X,643) \le2Q_1(145,643)Q_2(X,643) =: H(X). U ( X , K ) ≤ U ( X , 643 ) ≤ 2 Q 1 ( 145 , 643 ) Q 2 ( X , 643 ) =: H ( X ) .
Moreover,
( 8 X 2 ( K − 2 ) ) 1 / 4 ≥ ( 8 ⋅ 641 X 2 ) 1 / 4 = 5128 1 / 4 X 1 / 2 = : L ( X ) . (8X^2(K-2))^{1/4} \ge(8 \cdot641X^2)^{1/4} = 5128^{1/4}X^{1/2} =: L(X). ( 8 X 2 ( K − 2 ) ) 1/4 ≥ ( 8 ⋅ 641 X 2 ) 1/4 = 512 8 1/4 X 1/2 =: L ( X ) .
Since Q 2 ( X , 643 ) = log ( 643 u 2 X 3 / 2 ) / log ( 643 u 3 ) Q_2(X,643) = \log(643u_2X^{3/2})/\log(643u_3) Q 2 ( X , 643 ) = log ( 643 u 2 X 3/2 ) / log ( 643 u 3 ) ,
X ( L − H ) ′ ( X ) = 1 2 5128 1 / 4 X 1 / 2 − 3 Q 1 ( 145 , 643 ) log ( 643 u 3 ) , X(L-H)'(X) = \tfrac{1}{2}5128^{1/4}X^{1/2} - \frac{3Q_1(145,643)}{\log(643u_3)}, X ( L − H ) ′ ( X ) = 2 1 512 8 1/4 X 1/2 − log ( 643 u 3 ) 3 Q 1 ( 145 , 643 ) ,
which is increasing in X X X and positive at X = 145 X = 145 X = 145 (50.949 … > 10.011 … 50.949\ldots> 10.011\ldots 50.949 … > 10.011 … ). So L − H L-H L − H is increasing on [ 145 , ∞ ) [145,\infty) [ 145 , ∞ ) , and L ( 145 ) = ( 8 ⋅ 641 ⋅ 145 2 ) 1 / 4 = 101.899 … > H ( 145 ) = U ( 145 , 643 ) = 100.595 … L(145) = (8 \cdot641 \cdot145^2)^{1/4} = 101.899\ldots> H(145) = \mathcal{U}(145,643) = 100.595\ldots L ( 145 ) = ( 8 ⋅ 641 ⋅ 14 5 2 ) 1/4 = 101.899 … > H ( 145 ) = U ( 145 , 643 ) = 100.595 … . Therefore ( 8 X 2 ( K − 2 ) ) 1 / 4 ≥ L ( X ) > H ( X ) ≥ U ( X , K ) (8X^2(K-2))^{1/4} \ge L(X) > H(X) \ge\mathcal{U}(X,K) ( 8 X 2 ( K − 2 ) ) 1/4 ≥ L ( X ) > H ( X ) ≥ U ( X , K ) .
We finally prove item 3. By item 1, U ( X , K ) ≤ U ( X , 643 ) \mathcal{U}(X,K) \le\mathcal{U}(X,643) U ( X , K ) ≤ U ( X , 643 ) , so it suffices to show that X ↦ U ( X , 643 ) X \mapsto\mathcal{U}(X,643) X ↦ U ( X , 643 ) is increasing on [ 1 , ∞ ) [1,\infty) [ 1 , ∞ ) . With w = log X ≥ 0 w = \log X \ge0 w = log X ≥ 0 and the positive constants e 1 = log ( 643 u 1 ) e_1 = \log(643u_1) e 1 = log ( 643 u 1 ) , e 2 = log 2572 e_2 = \log2572 e 2 = log 2572 , e 3 = log ( 643 u 2 ) e_3 = \log(643u_2) e 3 = log ( 643 u 2 ) ,
U ( X , 643 ) = 2 log ( 643 u 3 ) ⋅ ( e 1 + 7 w ) ( e 3 + 3 2 w ) e 2 + 2 w , \mathcal{U}(X,643) = \frac{2}{\log(643u_3)} \cdot\frac{(e_1+7w)(e_3+\frac{3}{2}w)}{e_2+2w}, U ( X , 643 ) = log ( 643 u 3 ) 2 ⋅ e 2 + 2 w ( e 1 + 7 w ) ( e 3 + 2 3 w ) ,
and the derivative in w w w of the last factor has the sign of
( 7 ( e 3 + 3 2 w ) + 3 2 ( e 1 + 7 w ) ) ( e 2 + 2 w ) − 2 ( e 1 + 7 w ) ( e 3 + 3 2 w ) = ( 7 e 3 + 3 2 e 1 ) e 2 − 2 e 1 e 3 + 21 e 2 w + 21 w 2 , \left(7\left(e_3+\frac{3}{2}w\right)+\frac{3}{2}(e_1+7w)\right)(e_2+2w)-2(e_1+7w)\left(e_3+\frac{3}{2}w\right) = \left(7e_3+\frac{3}{2}e_1\right)e_2-2e_1e_3+21e_2w+21w^2, ( 7 ( e 3 + 2 3 w ) + 2 3 ( e 1 + 7 w ) ) ( e 2 + 2 w ) − 2 ( e 1 + 7 w ) ( e 3 + 2 3 w ) = ( 7 e 3 + 2 3 e 1 ) e 2 − 2 e 1 e 3 + 21 e 2 w + 21 w 2 ,
which is positive for w ≥ 0 w \ge0 w ≥ 0 , since its value at w = 0 w = 0 w = 0 is 139.64 … 139.64\ldots 139.64 … . Hence U ( X , K ) ≤ U ( 144 , 643 ) = 100.562 … < 100.6 \mathcal{U}(X,K) \le\mathcal{U}(144,643) = 100.562\ldots< 100.6 U ( X , K ) ≤ U ( 144 , 643 ) = 100.562 … < 100.6 for 1 ≤ X ≤ 144 1 \le X \le144 1 ≤ X ≤ 144 and K ≥ 643 K \ge643 K ≥ 643 .
All numerical values in this proof were evaluated in interval arithmetic with the exact constants of Definition A.1.
The threshold 145 of item 2 is not optimal: its argument already closes from X = 141 X = 141 X = 141 .
Lemma A.4. Let ( ν , A , K , ε ) (\nu,A,K,\varepsilon) ( ν , A , K , ε ) be a parameter point (Convention 8.1) with ν A ≥ 2 \nu A \ge2 ν A ≥ 2 , F ν A ≤ 144 F_\nu A \le144 F ν A ≤ 144 and 643 ≤ K ≤ 2401 643 \le K \le2401 643 ≤ K ≤ 2401 . Then the Pell sequences (Convention 2.6) of its family triple satisfy V m λ = W n λ \mathcal{V}^{\lambda}_m = \mathcal{W}^{\lambda}_n V m λ = W n λ , with λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } and integers 1 ≤ m , n ≤ 50 1 \le m,n \le50 1 ≤ m , n ≤ 50 , only for ( λ , m , n ) = ( − 1 , 1 , 1 ) (\lambda,m,n)=(-1,1,1) ( λ , m , n ) = ( − 1 , 1 , 1 ) .
Proof. The proof is a finite computation. There are 478 pairs ( ν , A ) (\nu,A) ( ν , A ) with ν A ≥ 2 \nu A \ge2 ν A ≥ 2 and F ν A ≤ 144 F_{\nu A} \le144 F ν A ≤ 144 , all with ν ≤ 12 \nu\le12 ν ≤ 12 , hence 478 ⋅ 1759 ⋅ 2 = 1 681 604 478 \cdot1759 \cdot2 = 1\,681\,604 478 ⋅ 1759 ⋅ 2 = 1 681 604 parameter points as in the statement. At each of them the family triple { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple with root r r r (Lemma 8.2, item 2), and its other roots are s = a + r s=a+r s = a + r and t = b + r t=b+r t = b + r . At each point, a , b , c , r a,b,c,r a , b , c , r were computed in exact integer arithmetic and the identities of Lemma 8.2, item 1, were checked; then V j λ V_j^\lambda V j λ and W j λ W_j^\lambda W j λ were computed for both λ \lambda λ and 0 ≤ j ≤ 50 0\le j\le50 0 ≤ j ≤ 50 by the recurrences and initial terms of item 3 of Lemma 2.8, modulo the prime 2 31 − 1 2^{31}-1 2 31 − 1 . A non-congruence V m λ ≢ W n λ V_m^\lambda\not\equiv W_n^\lambda V m λ ≡ W n λ modulo 2 31 − 1 2^{31}-1 2 31 − 1 proves V m λ ≠ W n λ V_m^\lambda\ne W_n^\lambda V m λ = W n λ . Besides ( λ , m , n ) = ( − 1 , 1 , 1 ) (\lambda,m,n)=(-1,1,1) ( λ , m , n ) = ( − 1 , 1 , 1 ) , which gives a congruence at every point, there are fifteen congruences with 1 ≤ m , n ≤ 50 1\le m,n\le50 1 ≤ m , n ≤ 50 , at four parameter points; each of them was recomputed in exact integer arithmetic, and none is an equality. This proves the claim.
The exception is a genuine common term: by Lemma 2.12, V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 is the number z z z of the regular extension d + ( a , b , c ) d_+(a,b,c) d + ( a , b , c ) .
Theorem A.5. Let ( ν , A , K , ε ) (\nu,A,K,\varepsilon) ( ν , A , K , ε ) be a parameter point (Convention 8.1) with ν A ≥ 2 \nu A\ge2 ν A ≥ 2 and 643 ≤ K ≤ 2401 643\le K\le2401 643 ≤ K ≤ 2401 . Then its family triple has no irregular extension.
Proof. The floor and the ceiling for the index (Lemma A.2) are incompatible when X ≥ 145 X\ge145 X ≥ 145 ; when X ≤ 144 X\le144 X ≤ 144 the ceiling leaves only indices m ≤ 50 m\le50 m ≤ 50 , which the search of Lemma A.4 excludes. By Lemma 8.2, item 2, the family triple { a , b , c } \{a,b,c\} { a , b , c } is an Euler triple. Suppose that it has an irregular extension d d d , and let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of d d d (Proposition 2.11). By Lemma A.2,
8 X 2 ( K − 2 ) 1 / 4 < 2 m < U ( X , K ) . 8X^{2}(K-2)^{1/4}<2m< U(X,K). 8 X 2 ( K − 2 ) 1/4 < 2 m < U ( X , K ) . Since X X X is a positive integer, either X ≥ 145 X\ge145 X ≥ 145 or X ≤ 144 X\le144 X ≤ 144 . If X ≥ 145 X\ge145 X ≥ 145 , this contradicts Lemma A.3, item 2.
Let X ≤ 144 X\le144 X ≤ 144 . By Lemma A.3, item 3, 2 m < U ( X , K ) < 100.6 2m<U(X,K)<100.6 2 m < U ( X , K ) < 100.6 , so m ≤ 50 m\le50 m ≤ 50 ; and Lemma 2.14 gives 3 ≤ n < m 3\le n<m 3 ≤ n < m . By item 2 of Lemma 2.8, V m λ = z = W n λ V_m^\lambda=z=W_n^\lambda V m λ = z = W n λ with 3 ≤ n < m ≤ 50 3\le n<m\le50 3 ≤ n < m ≤ 50 . This contradicts Lemma A.4, since m ≠ 1 m\ne1 m = 1 .
Hence { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
A Baker–Davenport reduction for a single Euler triple
For a single Euler triple the a priori bound leaves finitely many indices: 2 m < M 2m<M 2 m < M for every integer M M M with Φ c ( M ) ≥ 0 \Phi_c(M)\ge0 Φ c ( M ) ≥ 0 (Proposition 7.3). By the method of Baker and Davenport [4 ] , in the inhomogeneous form of Dujella and Pethő [16 ] , the linear form Λ z \Lambda_z Λ z , divided by log γ \log\gamma log γ , reduces them to 2 m ≤ J λ 2m\le J_\lambda 2 m ≤ J λ for a small J λ J_\lambda J λ , and these are checked directly; Proposition B.1 states the resulting criterion. Its second alternative serves the sign λ = − 1 \lambda=-1 λ = − 1 , for which the common term V 1 − 1 = W 1 − 1 V_1^{-1}=W_1^{-1} V 1 − 1 = W 1 − 1 of the regular extension can make the number μ − 1 \mu_{-1} μ − 1 extremely close to 2 − 2 log β / log γ 2-2\log\beta/\log\gamma 2 − 2 log β / log γ , so that the first alternative does not apply with denominators of the size of M M M . The criterion is used in Theorem 8.4 and Theorem 8.7.
Proposition B.1. Let { a , b , c } \{a,b,c\} { a , b , c } be an Euler triple with units β , γ \beta,\gamma β , γ and multiplication matrix B \mathbf{B} B (Definition 2.3), and let M ≥ 1 M\ge1 M ≥ 1 be an integer with Φ c ( M ) ≥ 0 \Phi_c(M)\ge0 Φ c ( M ) ≥ 0 , where Φ c \Phi_c Φ c is as in Convention 7.1. For λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } let E λ E_\lambda E λ and κ λ \kappa_\lambda κ λ be as in Definition 2.16, and put
ξ = log β log γ , μ λ = log E λ log γ . \xi=\frac{\log\beta}{\log\gamma},\qquad\mu_\lambda=\frac{\log E_\lambda}{\log\gamma}. ξ = log γ log β , μ λ = log γ log E λ .
Suppose that for each λ ∈ { − 1 , 1 } \lambda\in\{-1,1\} λ ∈ { − 1 , 1 } there is an integer J λ ≥ 0 J_\lambda\ge0 J λ ≥ 0 for which one of the following holds.
There are integers p , q p,q p , q with q > 0 q>0 q > 0 such that η = ∥ q μ λ ∥ − M ∣ q ξ − p ∣ > 0 \eta=\lVert q\mu_{\lambda}\rVert-M|q\xi-p|>0 η = ∥ q μ λ ∥ − M ∣ q ξ − p ∣ > 0 , where ∥ ⋅ ∥ \lVert\cdot\rVert ∥ ⋅ ∥ denotes the distance to the nearest integer, and q κ λ ≤ η β 2 J λ log γ q\kappa_{\lambda}\leq\eta\beta^{2J_{\lambda}}\log\gamma q κ λ ≤ η β 2 J λ log γ .
There are integers m 0 ≥ 0 m_{0}\geq0 m 0 ≥ 0 and n 0 n_{0} n 0 with m 0 < J λ m_{0}<J_{\lambda} m 0 < J λ such that
∣ m 0 ξ − n 0 + μ λ ∣ + κ λ log γ β − 2 J λ < min { ∣ q ξ − p ∣ : p / q a convergent of ξ , 1 ≤ q < M } . |m_{0}\xi-n_{0}+\mu_{\lambda}|+\frac{\kappa_{\lambda}}{\log\gamma}\beta^{-2J_{\lambda}}<\min\{|q\xi-p|:p/q\text{ a convergent of }\xi,\ 1\leq q<M\}. ∣ m 0 ξ − n 0 + μ λ ∣ + log γ κ λ β − 2 J λ < min { ∣ q ξ − p ∣ : p / q a convergent of ξ , 1 ≤ q < M } .
If, for each λ \lambda λ and each integer m ≥ 0 m\geq0 m ≥ 0 with 2 m ≤ J λ 2m\leq J_{\lambda} 2 m ≤ J λ , the second coordinate x x x of B ( 1 ) 2 m ( λ ) \mathbf{B}^{2m}_{(1)}(\lambda) B ( 1 ) 2 m ( λ ) does not give an integer d = ( x 2 − 1 ) / a d=(x^{2}-1)/a d = ( x 2 − 1 ) / a with d > c d>c d > c , d ≠ d + ( a , b , c ) d\neq d_{+}(a,b,c) d = d + ( a , b , c ) and b d + 1 bd+1 b d + 1 a square, then { a , b , c } \{a,b,c\} { a , b , c } has no irregular extension.
Proof. Suppose that d d d is an irregular extension of { a , b , c } \{a,b,c\} { a , b , c } , let ( λ , l , m , n ) (\lambda,l,m,n) ( λ , l , m , n ) be the index system of d d d (Proposition 2.11), and let x , y , z x,y,z x , y , z be as in Definition 2.5 . By item 2 of Lemma 2.8 , ( z x ) = B ( 1 ) 2 m ( λ ) \binom{z}{x}=\mathbf{B}^{2m}_{(1)}(\lambda) ( x z ) = B ( 1 ) 2 m ( λ ) ; so x x x is the second coordinate of B ( 1 ) 2 m ( λ ) \mathbf{B}^{2m}_{(1)}(\lambda) B ( 1 ) 2 m ( λ ) . By Proposition 7.3 , applied with u 0 = M u_{0}=M u 0 = M , we have 2 m < M 2m<M 2 m < M .
We show that 2 m ≤ J λ 2m\leq J_{\lambda} 2 m ≤ J λ . Suppose 2 m > J λ 2m>J_{\lambda} 2 m > J λ . Item 5 of Lemma 2.18 , divided by log γ \log\gamma log γ , gives
0 < 2 m ξ − 2 n + μ λ < κ λ log γ β − 4 m < κ λ log γ β − 2 J λ . 0<2m\xi-2n+\mu_{\lambda}<\frac{\kappa_{\lambda}}{\log\gamma}\beta^{-4m}<\frac{\kappa_{\lambda}}{\log\gamma}\beta^{-2J_{\lambda}}. 0 < 2 m ξ − 2 n + μ λ < log γ κ λ β − 4 m < log γ κ λ β − 2 J λ .
In case 1 we use the reduction of Baker and Davenport [4 ] , in the inhomogeneous form of [16 Lemma 5(a)] . Since q μ λ − q ( 2 m ξ − 2 n + μ λ ) + 2 m ( q ξ − p ) q\mu_{\lambda}-q(2m\xi-2n+\mu_{\lambda})+2m(q\xi-p) q μ λ − q ( 2 m ξ − 2 n + μ λ ) + 2 m ( q ξ − p ) is an integer and 2 m < M 2m<M 2 m < M ,
∥ q μ λ ∥ ≤ q ( 2 m ξ − 2 n + μ λ ) + 2 m ∣ q ξ − p ∣ < q κ λ log γ β − 4 m + M ∣ q ξ − p ∣ , \lVert q\mu_{\lambda}\rVert\leq q(2m\xi-2n+\mu_{\lambda})+2m|q\xi-p|<\frac{q\kappa_{\lambda}}{\log\gamma}\beta^{-4m}+M|q\xi-p|, ∥ q μ λ ∥ ≤ q ( 2 m ξ − 2 n + μ λ ) + 2 m ∣ q ξ − p ∣ < log γ q κ λ β − 4 m + M ∣ q ξ − p ∣ ,
so q κ λ β − 4 m > η log γ q\kappa_{\lambda}\beta^{-4m}>\eta\log\gamma q κ λ β − 4 m > η log γ , that is η β 4 m log γ < q κ λ ≤ η β 2 J λ log γ \eta\beta^{4m}\log\gamma<q\kappa_{\lambda}\leq\eta\beta^{2J_{\lambda}}\log\gamma η β 4 m log γ < q κ λ ≤ η β 2 J λ log γ , which contradicts 2 m > J λ 2m>J_{\lambda} 2 m > J λ . In case 2, 2 m − m 0 ≥ 1 2m-m_{0}\geq1 2 m − m 0 ≥ 1 and
∣ ( 2 m − m 0 ) ξ − ( 2 n − n 0 ) ∣ ≤ ∣ 2 m ξ − 2 n + μ λ ∣ + ∣ m 0 ξ − n 0 + μ λ ∣ < κ λ log γ β − 2 J λ + ∣ m 0 ξ − n 0 + μ λ ∣ . |(2m-m_{0})\xi-(2n-n_{0})|\leq|2m\xi-2n+\mu_{\lambda}|+|m_{0}\xi-n_{0}+\mu_{\lambda}|<\frac{\kappa_{\lambda}}{\log\gamma}\beta^{-2J_{\lambda}}+|m_{0}\xi-n_{0}+\mu_{\lambda}|. ∣ ( 2 m − m 0 ) ξ − ( 2 n − n 0 ) ∣ ≤ ∣2 m ξ − 2 n + μ λ ∣ + ∣ m 0 ξ − n 0 + μ λ ∣ < log γ κ λ β − 2 J λ + ∣ m 0 ξ − n 0 + μ λ ∣.
The number ξ \xi ξ is irrational: β Q = γ P \beta^{Q}=\gamma^{P} β Q = γ P with integers P , Q ≥ 1 P,Q\geq1 P , Q ≥ 1 is impossible, since β Q ∈ Q ( a c ) ∖ Q \beta^{Q}\in\mathbb{Q}(\sqrt{ac})\setminus\mathbb{Q} β Q ∈ Q ( a c ) ∖ Q and γ P ∈ Q ( b c ) \gamma^{P}\in\mathbb{Q}(\sqrt{bc}) γ P ∈ Q ( b c ) , and these two fields are distinct because a b ab ab is not a square (Lemma 2.4 , item 5). Let p i / q i p_{i}/q_{i} p i / q i be the convergent of ξ \xi ξ with q i ≤ 2 m − m 0 < q i + 1 q_{i}\leq2m-m_{0}<q_{i+1} q i ≤ 2 m − m 0 < q i + 1 . By the best-approximation property of convergents, ∣ ( 2 m − m 0 ) ξ − ( 2 n − n 0 ) ∣ ≥ ∣ q i ξ − p i ∣ |(2m-m_{0})\xi-(2n-n_{0})|\geq|q_{i}\xi-p_{i}| ∣ ( 2 m − m 0 ) ξ − ( 2 n − n 0 ) ∣ ≥ ∣ q i ξ − p i ∣ , and 1 ≤ q i ≤ 2 m − m 0 < M 1\leq q_{i}\leq2m-m_{0}<M 1 ≤ q i ≤ 2 m − m 0 < M ; this contradicts the hypothesis of case 2.
Hence 2 m ≤ J λ 2m\leq J_{\lambda} 2 m ≤ J λ , and d = ( x 2 − 1 ) / a d=(x^{2}-1)/a d = ( x 2 − 1 ) / a with x x x the second coordinate of B ( 1 ) 2 m ( λ ) \mathbf{B}^{2m}_{(1)}(\lambda) B ( 1 ) 2 m ( λ ) , d > c d>c d > c , d ≠ d + ( a , b , c ) d\neq d_{+}(a,b,c) d = d + ( a , b , c ) (as d d d is irregular, Definition 2.2 ) and b d + 1 = y 2 bd+1=y^{2} b d + 1 = y 2 . This contradicts the hypothesis. □
Acknowledgements
This paper was prepared with the assistance of Aleph, an AI research assistant for mathematics, using an ensemble of AI models from Anthropic (Claude), OpenAI (ChatGPT) and Google (Gemini) together with bespoke AI models. The authors take full responsibility for the content of the paper and for all mathematical results. (A. Dujella) Department of Mathematics, University of Zagreb, Croatia E-mail address: duje@math.hr
(S. Slijepčević) Department of Mathematics, University of Zagreb, Croatia Cantab Pi, Cambridge, UK E-mail address: sinisa@math.hr, sinisa@cantabpi.com