Introduction
The Diophantine equation a b + b c + c a = N ab + bc + ca = N ab + b c + c a = N in three variables a a a , b b b and c c c is a classic ternary quadratic form studied extensively. For instance, the number of solutions in positive integers a a a , b b b and c c c of the equation, where N N N is a given positive integer, was investigated by Mordell [28 ] in 1923. In this paper, we will prove the following solvability result.
Theorem 1 Every natural number N ≥ 1849 N \ge1849 N ≥ 1849 can be represented as
N = a b + b c + c a (1) N = ab + bc + ca \tag*{(1)} N = ab + b c + c a (1) where a a a , b b b and c c c are three distinct positive integers.
In other words, Theorem 1 tells that every even number 2 N ≥ 3698 2N \ge3698 2 N ≥ 3698 is the surface area, i.e., the sum of areas of the six faces, of a right rectangular prism with three distinct edge lengths a a a , b b b and c c c , which has the value of 2 a b + 2 b c + 2 c a 2ab + 2bc + 2ca 2 ab + 2 b c + 2 c a .
The Diophantine equation a b + b c + c a = N ab + bc + ca = N ab + b c + c a = N in distinct positive integers a a a , b b b and c c c for a given positive integer N N N has the significance that its nonexistence of a solution is a characterization of a natural number N N N to be an Euler’s idoneal number (see the survey paper by Kani [22 p. 125] ). Idoneal numbers were discovered by Euler [11 ] around 1778 who used them to find large prime numbers. However, not every natural number is suitable for his method and hence Euler gave the name numeri idonei (i.e., suitable or convenient numbers) to those numbers that are suitable for testing the primality of large integers.
Euler listed the following 65 idoneal numbers:
Table 1.
Euler discovered that there are 65 idoneal numbers among all positive integers N ≤ 10000 N \le10000 N ≤ 10000 . Using the largest one 1848 in the list, Euler proved that 18518809 = 18480000 + 197 2 18518809 = 18480000 + 197^2 18518809 = 18480000 + 19 7 2 is prime by showing that it has a single representation in the form 1848 x 2 + y 2 1848x^2 + y^2 1848 x 2 + y 2 (see Weil [47 p. 225] ).
Idoneal numbers were later investigated by Gauss [14 ] in the Disquisitiones Arithmeticae . In the language of Gauss’s theory of quadratic forms, idoneal numbers correspond to discriminants where there is only one class of binary quadratic forms per genus. Gauss also identified the 65 idoneal numbers (see [14 Article 303] ). Euler and Gauss conjectured that this list of 65 idoneal numbers is complete, i.e., that there would not be any more idoneal number for all N > 10000 N > 10000 N > 10000 .
In 1973, Weinberger [48 ] showed that there are at most two other larger idoneal numbers if they exist, with help of a result due to D. H. Lehmer (see [48 p. 119] ). It is obvious that if a b + b c + c a = N ab + bc + ca = N ab + b c + c a = N , then ( 2 a ) ( 2 b ) + ( 2 b ) ( 2 c ) + ( 2 c ) ( 2 a ) = 4 N (2a)(2b) + (2b)(2c) + (2c)(2a) = 4N ( 2 a ) ( 2 b ) + ( 2 b ) ( 2 c ) + ( 2 c ) ( 2 a ) = 4 N . So, in case that 4 N 4N 4 N is one other larger idoneal number for some natural number N N N , then N N N is another. Therefore, there exist 65, 66, or 67 idoneal numbers.
By Theorem 1 and Euler’s list of 65 idoneal numbers, we obtain
Theorem 2 The number 1848 is the largest Euler’s idoneal number.
This theorem says that the list of 65 idoneal numbers is complete and hence the conjecture of Euler and Gauss is confirmed.
Outline of the proof of Theorem 1
In this paper, we will prove the following two propositions.
Proposition 1 Every squarefree positive integer N > 1365 N > 1365 N > 1365 with N ≡ 1 ( m o d 4 ) N \equiv1 \pmod{4} N ≡ 1 ( mod 4 ) can be represented as
N = a b + b c + c a (2) N = ab + bc + ca \tag*{(2)} N = ab + b c + c a (2) where a a a , b b b and c c c are three distinct positive integers.
Proposition 2 Every squarefree positive integer N > 462 N > 462 N > 462 with N ≡ 2 ( m o d 4 ) N \equiv2 \pmod{4} N ≡ 2 ( mod 4 ) can be represented as
N = a b + b c + c a (3) N = ab + bc + ca \tag*{(3)} N = ab + b c + c a (3) where a a a , b b b and c c c are three distinct positive integers.
In the work of Grube [16 ] published in 1874, Euler’s results on idoneal numbers were analyzed and extended by employing Gauss’s theory of binary quadratic forms. There are other definition and criterion of an idoneal number that are not the same as but equivalent to Euler’s original definition (see Weil [47 Chapter III, Section §X] ). Grube proved the following elementary criterion for Euler’s idoneal number (see also Frei [12 p. 57, Theorem 14] and Kani [22 p. 204, Proposition 17] ).
Lemma 1 (Grube’s criterion) A natural number N ≥ 3 N \ge3 N ≥ 3 is an idoneal number if and only if for every b = 1 , ⋯ , ⌊ N / 3 ⌋ b = 1, \cdots, \lfloor\sqrt{N/3}\rfloor b = 1 , ⋯ , ⌊ N /3 ⌋ and every factorization
N + b 2 = a c with c ≥ a ≥ 2 b (4) N + b^2 = ac \quad\text{with } c \ge a \ge2b \tag*{(4)} N + b 2 = a c with c ≥ a ≥ 2 b (4) we have that either a = c a = c a = c or a = 2 b a = 2b a = 2 b .
By putting b = x b = x b = x , a = x + y a = x+y a = x + y and c = x + z c = x+z c = x + z in Grube’s criterion, we see that a natural number N ≥ 3 N \ge3 N ≥ 3 is an idoneal number if and only if the Diophantine equation
x y + y z + z x = N xy + yz + zx = N x y + y z + z x = N
has no solution in distinct positive integers $0 < x < y < z$.
Based on Grube’s and Euler’s results as surveyed by Kani [22 Section 2.3] , the classification of large idoneal numbers is readily available as follows (see [22 p. 203, Corollary 14] ).
Proposition 3 If a natural number N > 1848 N > 1848 N > 1848 is an idoneal number, then N ≢ 3 ( m o d 4 ) N \not\equiv3 \pmod{4} N ≡ 3 ( mod 4 ) . Moreover, if N ≡ 1 , 2 ( m o d 4 ) N \equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , then N N N is squarefree, whereas if N ≡ 0 ( m o d 4 ) N \equiv0 \pmod{4} N ≡ 0 ( mod 4 ) , then M = N / 4 M = N/4 M = N /4 is a squarefree idoneal number with M ≡ 2 ( m o d 4 ) M \equiv2 \pmod{4} M ≡ 2 ( mod 4 ) .
For example, if N > 1848 N > 1848 N > 1848 and N = 4 L + 3 N = 4L + 3 N = 4 L + 3 for some positive integer L L L , then N N N can be expressed as N = x y + y z + z x N = xy + yz + zx N = x y + y z + z x with x = 1 x = 1 x = 1 , y = 3 y = 3 y = 3 , and z = L > 3 z = L > 3 z = L > 3 and thus N N N is not an idoneal number.
It follows easily from the above Propositions 1, 2 and 3 that every natural number N > 1848 N > 1848 N > 1848 cannot be an idoneal number and hence Theorem 1.
In the remainder of this paper, we are going to prove Propositions 1 and 2, which, together with the established classification result in Proposition 3, can imply Theorem 1.
An equivalent assertion of Propositions 1 and 2
In this section, we provide one assertion which is equivalent to Propositions 1 and 2 for every squarefree positive integer N > 10000 N > 10000 N > 10000 , thanks to Euler’s result in the range N ≤ 10000 N \le10000 N ≤ 10000 in which Propositions 1 and 2 are accordingly true.
Define r 3 ( n ) r_3(n) r 3 ( n ) to be the number of representations of a nonnegative integer n n n as a sum of three squares and let q q q be a nonzero complex number with ∣ q ∣ < 1 |q| < 1 ∣ q ∣ < 1 . Define the generating function
R 3 ( q ) = ∑ n = 0 ∞ r 3 ( n ) ( − q ) n = ( ∑ m ∈ Z ( − 1 ) m q m 2 ) 3 . R_3(q) = \sum_{n=0}^{\infty} r_3(n)(-q)^n = \left(\sum_{m\in\mathbb{Z}}(-1)^m q^{m^2}\right)^3. R 3 ( q ) = n = 0 ∑ ∞ r 3 ( n ) ( − q ) n = ( m ∈ Z ∑ ( − 1 ) m q m 2 ) 3 .
In his work regarding mock theta functions of Ramanujan, Andrews found the following representation for R 3 ( q ) R_3(q) R 3 ( q ) by using theory of q q q -series (see [1 p. 124, Eqn. (5.16)] )
∑ n = 0 ∞ ( − 1 ) n r 3 ( n ) q n = 1 + 4 ∑ n = 1 ∞ ( − 1 ) n q n 1 + q n − 2 ∑ n = 1 ∞ ∑ ∣ j ∣ < n ( − 1 ) j q n 2 − j 2 ( 1 − q n ) 1 + q n . (5) \sum_{n=0}^{\infty}(-1)^n r_3(n)q^n = 1 + 4\sum_{n=1}^{\infty}\frac{(-1)^nq^n}{1+q^n} - 2\sum_{n=1}^{\infty}\sum_{|j|<n}\frac{(-1)^j q^{n^2-j^2}(1-q^n)}{1+q^n}. \tag*{(5)} n = 0 ∑ ∞ ( − 1 ) n r 3 ( n ) q n = 1 + 4 n = 1 ∑ ∞ 1 + q n ( − 1 ) n q n − 2 n = 1 ∑ ∞ ∣ j ∣ < n ∑ 1 + q n ( − 1 ) j q n 2 − j 2 ( 1 − q n ) . (5)
It can be readily deduced from Andrews’s identity (5) to obtain the following identity due to Crandall ([8 p. 376, Eqn. (6.2)] ) that, for every positive integer N ≥ 3 N \ge3 N ≥ 3 ,
r 3 ( N ) = 6 ( − 1 ) N + 1 ∑ x y = N ( − 1 ) x + y + 4 ( − 1 ) N + 1 ∑ x y + y z + z x = N ( − 1 ) x + y + z (6) r_3(N) = 6(-1)^{N+1}\sum_{xy=N}(-1)^{x+y} + 4(-1)^{N+1}\sum_{xy+yz+zx=N}(-1)^{x+y+z} \tag*{(6)} r 3 ( N ) = 6 ( − 1 ) N + 1 x y = N ∑ ( − 1 ) x + y + 4 ( − 1 ) N + 1 x y + y z + z x = N ∑ ( − 1 ) x + y + z (6)
where every summation index integer x x x , y y y or z z z is positive.
Based on the Andrews–Crandall identity in the form of (6), Mortenson proved the following two results (see [29 p. 777] ).
Lemma 2 For every positive integer N ≥ 5 N \ge5 N ≥ 5 with N ≡ 1 ( m o d 4 ) N \equiv1 \pmod{4} N ≡ 1 ( mod 4 ) , we have
r 3 ( N ) = 6 ∑ d ∣ N 1 + 12 ∑ x ≠ y x 2 + 2 x y = N 1 + 24 ∑ x < y < z x y + y z + z x = N 1 (7) r_3(N) = 6\sum_{d\mid N}1 + 12\sum_{\substack{x\ne y \\ x^2+2xy=N}}1 + 24\sum_{\substack{x<y<z \\ xy+yz+zx=N}}1 \tag*{(7)} r 3 ( N ) = 6 d ∣ N ∑ 1 + 12 x = y x 2 + 2 x y = N ∑ 1 + 24 x < y < z x y + y z + z x = N ∑ 1 (7) where every summation index integer d d d , x x x , y y y or z z z is positive.
Lemma 3. For every positive integer N ≥ 6 N \ge6 N ≥ 6 with N ≡ 2 ( m o d 4 ) N \equiv2 \pmod{4} N ≡ 2 ( mod 4 ) , we have
r 3 ( N ) = 12 ∑ d ∣ ( N / 2 ) 1 + 24 ∑ x < y < z x y + y z + z x = N 1 (8) r_3(N) = 12 \sum_{d\mid(N/2)} 1 + 24 \sum_{\substack{x<y<z\\xy+yz+zx=N}} 1 \tag*{(8)} r 3 ( N ) = 12 d ∣ ( N /2 ) ∑ 1 + 24 x < y < z x y + y z + z x = N ∑ 1 (8) where every summation index integer d d d , x x x , y y y or z z z is positive.
From Lemmas 2 and 3, we have the following two special results.
Corollary 1. For every squarefree positive integer N ≥ 5 N \ge5 N ≥ 5 with N ≡ 1 ( m o d 4 ) N \equiv1 \pmod{4} N ≡ 1 ( mod 4 ) , we have
r 3 ( N ) = 12 ⋅ 2 ω ( N ) + 24 ∑ x < y < z x y + y z + z x = N 1 (9) r_3(N) = 12 \cdot2^{\omega(N)} + 24 \sum_{\substack{x<y<z\\xy+yz+zx=N}} 1 \tag*{(9)} r 3 ( N ) = 12 ⋅ 2 ω ( N ) + 24 x < y < z x y + y z + z x = N ∑ 1 (9) where every summation index integer x x x , y y y or z z z is positive and ω ( N ) \omega(N) ω ( N ) is the number of distinct prime factors of N N N .
Proof: Let N = p 1 ′ ⋯ p L ′ N = p'_1 \cdots p'_L N = p 1 ′ ⋯ p L ′ , where p 1 ′ , ⋯ , p L ′ p'_1,\cdots,p'_L p 1 ′ , ⋯ , p L ′ are L L L distinct odd primes. It is obvious that ∑ d ∣ N 1 = 2 L \sum_{d\mid N} 1 = 2^L ∑ d ∣ N 1 = 2 L . We now consider the second sum on the right-hand side of (7) .
If x 2 + 2 x y = N x^2 + 2xy = N x 2 + 2 x y = N , then N = x ( x + 2 y ) N = x(x + 2y) N = x ( x + 2 y ) and hence x ∣ N x \mid N x ∣ N and ( x + 2 y ) ∣ N (x + 2y) \mid N ( x + 2 y ) ∣ N . Therefore, x x x and x + 2 y x + 2y x + 2 y are both squarefree positive integers. We pick up a positive integer d ∣ N d \mid N d ∣ N , which must be squarefree, then d ≠ N / d d \ne N/d d = N / d . If d < N / d d < N/d d < N / d , we choose x = d x = d x = d and x + 2 y = N / d x + 2y = N/d x + 2 y = N / d , i.e., y = ( N / d − d ) / 2 y = (N/d - d)/2 y = ( N / d − d ) /2 . If d > N / d d > N/d d > N / d , we choose x = N / d x = N/d x = N / d and x + 2 y = d x + 2y = d x + 2 y = d , i.e., y = ( d − N / d ) / 2 y = (d - N/d)/2 y = ( d − N / d ) /2 . It is clear that x ≡ 1 , 3 ( m o d 4 ) x \equiv1,3 \pmod{4} x ≡ 1 , 3 ( mod 4 ) implying x 2 ≡ 1 ( m o d 4 ) x^2 \equiv1 \pmod{4} x 2 ≡ 1 ( mod 4 ) . Thus, 2 x y = N − x 2 2xy = N - x^2 2 x y = N − x 2 implies 2 x y ≡ 0 ( m o d 4 ) 2xy \equiv0 \pmod{4} 2 x y ≡ 0 ( mod 4 ) and hence 2 ∣ y 2 \mid y 2 ∣ y and x ≠ y x \ne y x = y . Noting the symmetry in ( d , N / d ) (d,N/d) ( d , N / d ) and ( N / d , d ) (N/d,d) ( N / d , d ) , we see that
∑ x ≠ y x 2 + 2 x y = N 1 = 2 L − 1 . \sum_{\substack{x\ne y\\x^2+2xy=N}} 1 = 2^{L-1}. x = y x 2 + 2 x y = N ∑ 1 = 2 L − 1 . Since ω ( N ) = L \omega(N) = L ω ( N ) = L , the identity (7) implies (9). Therefore, the proof is complete. □
Corollary 2. For every squarefree positive integer N ≥ 6 N \ge6 N ≥ 6 with N ≡ 2 ( m o d 4 ) N \equiv2 \pmod{4} N ≡ 2 ( mod 4 ) , we have
r 3 ( N ) = 12 ⋅ 2 ω ( N ) − 1 + 24 ∑ x < y < z x y + y z + z x = N 1 (10) r_3(N) = 12 \cdot2^{\omega(N)-1} + 24 \sum_{\substack{x<y<z\\xy+yz+zx=N}} 1 \tag*{(10)} r 3 ( N ) = 12 ⋅ 2 ω ( N ) − 1 + 24 x < y < z x y + y z + z x = N ∑ 1 (10) where every summation index integer x x x , y y y or z z z is positive.
Proof: Let N = 2 p 1 ′ ⋯ p L ′ N = 2p'_1 \cdots p'_L N = 2 p 1 ′ ⋯ p L ′ , where p 1 ′ , ⋯ , p L ′ p'_1,\cdots,p'_L p 1 ′ , ⋯ , p L ′ are L L L distinct odd primes. Then, N / 2 = p 1 ′ ⋯ p L ′ N/2 = p'_1 \cdots p'_L N /2 = p 1 ′ ⋯ p L ′ and it is obvious that
∑ d ∣ ( N / 2 ) 1 = 2 L . \sum_{d\mid(N/2)} 1 = 2^L. d ∣ ( N /2 ) ∑ 1 = 2 L . Noting that ω ( N ) = L + 1 \omega(N) = L + 1 ω ( N ) = L + 1 , the identity (8) implies (10). The proof is complete. □
Corollaries 1 and 2 together imply the following
Proposition 4. For every squarefree positive integer N > 10000 N > 10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , we have
r 3 ( N ) = 12 ⋅ 2 ω ( 4 N ) − 1 + 24 ∑ x < y < z x y + y z + z x = N 1 (11) r_3(N) = 12 \cdot2^{\omega(4N)-1} + 24 \sum_{\substack{x<y<z\\xy+yz+zx=N}} 1 \tag*{(11)} r 3 ( N ) = 12 ⋅ 2 ω ( 4 N ) − 1 + 24 x < y < z x y + y z + z x = N ∑ 1 (11) where every summation index integer x x x , y y y or z z z is positive and ω ( 4 N ) \omega(4N) ω ( 4 N ) is the number of distinct prime factors of 4 N 4N 4 N .
The determination of r 3 ( N ) r_3(N) r 3 ( N ) is a major topic in representations of integers as sums of squares (see Bateman [3 ] and Grosswald’s book [15 Chapter 4] ). According to Gauss’s theorem ([15 p. 51] ) and an equality regarding r 3 ( N ) r_3(N) r 3 ( N ) (see [15 Eqn. (4.8), p. 53] ), let h ( − 4 N ) h(-4N) h ( − 4 N ) stand for the number of classes of primitive binary quadratic form corresponding to the discriminant D = − 4 N D=-4N D = − 4 N , then, for every squarefree positive integer N > 10000 N > 10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) ,
r 3 ( N ) = 12 ⋅ h ( − 4 N ) . (12) r_3(N) = 12 \cdot h(-4N). \tag*{(12)} r 3 ( N ) = 12 ⋅ h ( − 4 N ) . (12)
Set the Kronecker symbol χ ( n ) = ( D n ) \chi(n) = \left(\frac{D}{n}\right) χ ( n ) = ( n D ) with D = − 4 N D=-4N D = − 4 N for every squarefree positive integer N > 10000 N > 10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . Define the corresponding L L L -series
L ( s , χ ) = ∑ n = 1 ∞ χ ( n ) n s L(s,\chi) = \sum_{n=1}^{\infty} \frac{\chi(n)}{n^s} L ( s , χ ) = n = 1 ∑ ∞ n s χ ( n )
for s > 0 s>0 s > 0 . According to Dirichlet’s theorem ([15 p. 56] ), if h ( − 4 N ) h(-4N) h ( − 4 N ) is the number of classes of primitive binary quadratic form a x 2 + 2 b x y + c y 2 ax^2+2bxy+cy^2 a x 2 + 2 b x y + c y 2 with N = a c − b 2 N=ac-b^2 N = a c − b 2 , then
h ( − 4 N ) = 2 N π L ( 1 , χ ) . (13) h(-4N) = \frac{2\sqrt{N}}{\pi}L(1,\chi). \tag*{(13)} h ( − 4 N ) = π 2 N L ( 1 , χ ) . (13)
There are connection and correspondence between the theory of classes of equivalent quadratic forms and the theory of ideals in quadratic fields (see Davenport’s book [9 pp. 51–53] ). According to the analytic class number formula for an arbitrary algebraic number field (see, e.g., Neukirch [31 p. 467] ), let h K h_K h K be the class number of an imaginary quadratic field K = Q ( D ) K=\mathbb{Q}(\sqrt{D}) K = Q ( D ) with D = − 4 N < 0 D=-4N<0 D = − 4 N < 0 for every squarefree positive integer N > 10000 N>10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N\equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , we have as a special case the following result (see, e.g., [27 p. 391] ),
h K = w ∣ D ∣ 2 π L ( 1 , χ ) (14) h_K = \frac{w\sqrt{|D|}}{2\pi}L(1,\chi) \tag*{(14)} h K = 2 π w ∣ D ∣ L ( 1 , χ ) (14)
where w w w denotes the number of roots of unity in the imaginary quadratic field K K K . Since w = 2 w=2 w = 2 for D < − 4 D<-4 D < − 4 , we see from (13) and (14) that
h K = h ( − 4 N ) = ∣ D ∣ π L ( 1 , χ ) . (15) h_K = h(-4N) = \frac{\sqrt{|D|}}{\pi}L(1,\chi). \tag*{(15)} h K = h ( − 4 N ) = π ∣ D ∣ L ( 1 , χ ) . (15)
Therefore, by using (12), from Proposition 4 we can deduce the following
Proposition 5. For every squarefree positive integer N > 10000 N > 10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , let h K h_K h K be the class number of the imaginary quadratic field K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) with D = − 4 N < 0 D = -4N < 0 D = − 4 N < 0 , we have
h K = 2 ω ( 4 N ) − 1 + 2 ∑ x < y < z x y + y z + z x = N 1 (16) h_K = 2^{\omega(4N)-1} + 2\sum_{\substack{x<y<z\\xy+yz+zx=N}}1 \tag*{(16)} h K = 2 ω ( 4 N ) − 1 + 2 x < y < z x y + y z + z x = N ∑ 1 (16) where every summation index integer x x x , y y y or z z z is positive. Consequently, we have h K ≥ 2 ω ( 4 N ) − 1 h_K \ge2^{\omega(4N)-1} h K ≥ 2 ω ( 4 N ) − 1 with equality if and only if the Diophantine equation x y + y z + z x = N xy+yz+zx=N x y + y z + z x = N has no solution in distinct positive integers $0 < x < y < z$ for the given number N N N .
It is now ready to demonstrate that Propositions 1 and 2 in the range N > 10000 N > 10000 N > 10000 is equivalent to the following assertion.
Proposition 6. For every squarefree positive integer N > 10000 N > 10000 N > 10000 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , let h K h_K h K be the class number of the imaginary quadratic field K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) with D = − 4 N < 0 D = -4N < 0 D = − 4 N < 0 , we have
h K ≠ 2 ω ( 4 N ) − 1 . (17) h_K \ne2^{\omega(4N)-1}. \tag*{(17)} h K = 2 ω ( 4 N ) − 1 . (17)
In fact, if h K ≠ 2 ω ( 4 N ) − 1 h_K \ne2^{\omega(4N)-1} h K = 2 ω ( 4 N ) − 1 , then, from Proposition 5, we have h K > 2 ω ( 4 N ) − 1 h_K > 2^{\omega(4N)-1} h K > 2 ω ( 4 N ) − 1 and hence N N N can be expressed as N = x y + y z + z x N = xy + yz + zx N = x y + y z + z x for some three distinct positive integers x x x , y y y and z z z . Conversely, if N N N can be written as N = x y + y z + z x N = xy + yz + zx N = x y + y z + z x with $0 < x < y < z$, it follows from (16) that h K > 2 ω ( 4 N ) − 1 h_K > 2^{\omega(4N)-1} h K > 2 ω ( 4 N ) − 1 and hence h K ≠ 2 ω ( 4 N ) − 1 h_K \ne2^{\omega(4N)-1} h K = 2 ω ( 4 N ) − 1 . Thus, the assertion of Proposition 6 is equivalent to Proposition 1 together with Proposition 2.
In the remainder of this paper, we are going to prove Proposition 6, which implies Propositions 1 and 2 and, in turn, Theorem 1 as shown in Section 2.
Reduction of range in Proposition 6 by existing numerical results for imaginary quadratic fields
Let N > 10000 N > 10000 N > 10000 be a squarefree positive integer with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . In this scenario, the imaginary quadratic field K = Q ( − N ) K = \mathbb{Q}(\sqrt{-N}) K = Q ( − N ) has the fundamental discriminant D = − 4 N < 0 D = -4N < 0 D = − 4 N < 0 and K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) .
Let I K I_K I K denote the multiplicative group of nonzero fractional ideals of K K K , and P K P_K P K its subgroup of nonzero principal fractional ideals of K K K . The quotient group I K / P K I_K/P_K I K / P K , denoted by C l K Cl_K C l K , is called the ideal class group of K K K , which is a finite abelian group. The order of the ideal class group C l K Cl_K C l K is called the class number of K K K , denoted by h K = # C l K h_K = \#Cl_K h K = # C l K .
On the contrary we assume that
# C l K = 2 ω ( 4 N ) − 1 . (18) \#Cl_K = 2^{\omega(4N)-1}. \tag*{(18)} # C l K = 2 ω ( 4 N ) − 1 . (18)
By Lagrange’s theorem in group theory, any subgroup H H H of the finite group C l K Cl_K C l K must divide the order of the group, namely, # H ∣ # C l K \#H \mid\#Cl_K # H ∣ # C l K , and hence # H = 2 m \#H = 2^m # H = 2 m for some nonnegative integer m m m . By the structure theorem for finite abelian groups (see, e.g., [20 ] ), the ideal class group C l K Cl_K C l K is isomorphic to a unique direct product of a collection of cyclic groups. Then, we have
C l K ≅ Z 2 m 1 × ⋯ × Z 2 m r Cl_K \cong\mathbb{Z}_{2^{m_1}} \times\cdots\times\mathbb{Z}_{2^{m_r}} C l K ≅ Z 2 m 1 × ⋯ × Z 2 m r
where r ≥ 1 r \ge1 r ≥ 1 and m 1 , … , m r m_1,\ldots,m_r m 1 , … , m r are nonnegative integers with 0 ≤ m 1 ≤ ⋯ ≤ m r 0 \le m_1 \le\cdots\le m_r 0 ≤ m 1 ≤ ⋯ ≤ m r and m r ≥ 1 m_r \ge1 m r ≥ 1 . The direct product can be rewritten as
C l K ≅ Z 2 a 1 × Z 2 2 a 2 ⋯ × Z 2 t a t (19) Cl_K \cong\mathbb{Z}_2^{a_1} \times\mathbb{Z}_{2^2}^{a_2} \cdots\times\mathbb{Z}_{2^t}^{a_t} \tag*{(19)} C l K ≅ Z 2 a 1 × Z 2 2 a 2 ⋯ × Z 2 t a t (19)
where t ≥ 1 t \ge1 t ≥ 1 and a 1 , … , a t a_1,\ldots,a_t a 1 , … , a t are nonnegative integers with a t ≥ 1 a_t \ge1 a t ≥ 1 . Here, a ℓ a_\ell a ℓ denotes the number of occurrences of Z 2 ℓ \mathbb{Z}_{2^\ell} Z 2 ℓ in the direct product for 1 ≤ ℓ ≤ t 1 \le\ell\le t 1 ≤ ℓ ≤ t . By genus theory, a 1 + ⋯ + a t = ω ( ∣ D ∣ ) − 1 a_1+\cdots+a_t=\omega(\lvert D\rvert)-1 a 1 + ⋯ + a t = ω (∣ D ∣) − 1 , where D = − 4 N D=-4N D = − 4 N is the discriminant of the field K K K (see [10 p. 222] ).
It follows from (18) and (19) that
2 ω ( 4 N ) − 1 = # C l K = 2 a 1 ⋅ 2 2 a 2 ⋯ 2 t a t ≥ 2 a 1 + ⋯ + a t ⋅ 2 ( t − 1 ) a t = 2 ω ( ∣ D ∣ ) − 1 ⋅ 2 ( t − 1 ) a t . 2^{\omega(4N)-1} = \#Cl_K = 2^{a_1}\cdot2^{2a_2}\cdots2^{ta_t} \ge2^{a_1+\cdots+a_t}\cdot2^{(t-1)a_t} = 2^{\omega(\lvert D\rvert)-1}\cdot2^{(t-1)a_t}. 2 ω ( 4 N ) − 1 = # C l K = 2 a 1 ⋅ 2 2 a 2 ⋯ 2 t a t ≥ 2 a 1 + ⋯ + a t ⋅ 2 ( t − 1 ) a t = 2 ω (∣ D ∣) − 1 ⋅ 2 ( t − 1 ) a t .
So, there must be t = 1 t=1 t = 1 and a t = ω ( ∣ D ∣ ) − 1 a_t=\omega(\lvert D\rvert)-1 a t = ω (∣ D ∣) − 1 . Hence, from (19) ,
C l K ≅ Z 2 ω ( ∣ D ∣ ) − 1 . (20) Cl_K \cong\mathbb{Z}_2^{\omega(\lvert D\rvert)-1}. \tag*{(20)} C l K ≅ Z 2 ω (∣ D ∣) − 1 . (20)
Due to the computational work by Mosunov and Jacobson [30 ] in 2016, we know that up to 2 40 2^{40} 2 40 the largest in its absolute value of the fundamental discriminant D D D with C l K Cl_K C l K isomorphic to Z 2 ℓ \mathbb{Z}_2^\ell Z 2 ℓ for some positive integer ℓ \ell ℓ is D = − 5460 D=-5460 D = − 5460 . Since (18) implying (20) and N = ∣ D ∣ / 4 N=\lvert D\rvert/4 N = ∣ D ∣ /4 , we see that Proposition 6 is true in the range up to N ≤ 2 38 N\le2^{38} N ≤ 2 38 .
The range has been further expanded owing to the computational work by Elsenhans, Klüners and Nicolae [10 ] in 2020. They showed that for imaginary quadratic fields K = Q ( D ) K=\mathbb{Q}(\sqrt{D}) K = Q ( D ) with discriminant D < 0 D<0 D < 0 up to ∣ D ∣ ≤ 3.1 ⋅ 10 20 \lvert D\rvert\le3.1\cdot10^{20} ∣ D ∣ ≤ 3.1 ⋅ 1 0 20 , the largest one in its absolute value with C l K Cl_K C l K isomorphic to Z 2 ℓ \mathbb{Z}_2^\ell Z 2 ℓ for some positive integer ℓ \ell ℓ is still D = − 5460 D=-5460 D = − 5460 . As a result, Proposition 6 is true in the range up to N ≤ 7.5 ⋅ 10 19 N\le7.5\cdot10^{19} N ≤ 7.5 ⋅ 1 0 19 .
Thanks to the existing numerical works in [30 ] and [10 ] , it suffices to prove Proposition 6 for every squarefree positive integer N > 7.5 ⋅ 10 19 N>7.5\cdot10^{19} N > 7.5 ⋅ 1 0 19 with N ≡ 1 , 2 ( m o d 4 ) N\equiv1,2\pmod{4} N ≡ 1 , 2 ( mod 4 ) . In the following two sections, we will prepare to develop an explicit lower bound for the class number h K h_K h K of the imaginary quadratic field K = Q ( − N ) K=\mathbb{Q}(\sqrt{-N}) K = Q ( − N ) for such N N N . Then, we will provide a proof of Proposition 6 and, in turn, Theorem 1.
An explicit Hecke’s theorem
Let N N N be any given squarefree positive integer with N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 and N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . Based on the class number formula (15) for the imaginary quadratic field K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) with D = − 4 N < 0 D = -4N < 0 D = − 4 N < 0 , an available lower bound for L ( 1 , χ ) L(1,\chi) L ( 1 , χ ) will induce a lower bound for the class number h K h_K h K of the quadratic field K K K , and vice versa.
In 1918 Hecke [18 ] demonstrated a fundamental link between the real zeros of Dirichlet L L L -functions for the quadratic Dirichlet characters and the class numbers of imaginary quadratic fields. An effective Hecke’s theorem was proved by Mahler [26 p. 302, Theorem 2] in 1934 as stated below (see also Friedlander and Iwaniec [13 p. 344] ).
Theorem 3 (Hecke’s theorem) Suppose that g g g is a positive constant and that the integer d > 1 d > 1 d > 1 is greater than a certain number d 1 d_1 d 1 which depends only on g g g . Then, if the L L L -series denoted by L d ( s ) L_d(s) L d ( s ) has no zero in the interval
1 − g log d ≤ s ≤ 1 , 1 - \frac{g}{\log d} \le s \le1, 1 − log d g ≤ s ≤ 1 , there exists a number G > 0 G > 0 G > 0 , depending only on g g g , such that the class number h h h of Q ( − d ) \mathbb{Q}(\sqrt{-d}) Q ( − d ) satisfies
h ≥ G d log d . h \ge G \frac{\sqrt{d}}{\log d}. h ≥ G log d d .
For the application in this paper, an explicit Hecke’s theorem is needed. There have been some results in the respect and, in particular, those proved by Tatuzawa [43 ] in 1951 and by Hoffstein [19 ] in 1980. In the following, we are going to establish an explicit Hecke’s theorem with a different and rigorous proof which employs algebraic number theory to obtain explicit constants best suited for the forthcoming application.
Let K K K be an algebraic number field with degree m = [ K : Q ] m = [K : \mathbb{Q}] m = [ K : Q ] over the field Q \mathbb{Q} Q of rational numbers. We denote by O K O_K O K the ring of integers of K K K and by Δ K \Delta_K Δ K the discriminant of O K O_K O K over the ring Z \mathbb{Z} Z of rational integers. If K K K has r r r real embeddings and t t t pairs of complex conjugate embeddings, then we have r + 2 t = m r + 2t = m r + 2 t = m . We let σ 1 , ⋯ , σ r : K ↪ R \sigma_1, \cdots, \sigma_r : K \hookrightarrow\mathbb{R} σ 1 , ⋯ , σ r : K ↪ R be the r r r real embeddings and σ r + 1 , σ ‾ r + 1 , ⋯ , σ r + t , σ ‾ r + t : K ↪ C \sigma_{r+1}, \overline{\sigma}_{r+1}, \cdots, \sigma_{r+t}, \overline{\sigma}_{r+t} : K \hookrightarrow\mathbb{C} σ r + 1 , σ r + 1 , ⋯ , σ r + t , σ r + t : K ↪ C the t t t pairs of complex conjugate embeddings of K K K . For each a ∈ K a \in K a ∈ K , the field norm N K / Q ( a ) ∈ Q N_{K/\mathbb{Q}}(a) \in\mathbb{Q} N K / Q ( a ) ∈ Q is defined to be (see, e.g., [42 p. 49] )
N K / Q ( a ) = σ 1 ( a ) ⋯ σ r ( a ) σ r + 1 ( a ) σ ‾ r + 1 ( a ) ⋯ σ r + t ( a ) σ ‾ r + t ( a ) . N_{K/\mathbb{Q}}(a) = \sigma_1(a) \cdots\sigma_r(a)\sigma_{r+1}(a)\overline{\sigma}_{r+1}(a) \cdots\sigma_{r+t}(a)\overline{\sigma}_{r+t}(a). N K / Q ( a ) = σ 1 ( a ) ⋯ σ r ( a ) σ r + 1 ( a ) σ r + 1 ( a ) ⋯ σ r + t ( a ) σ r + t ( a ) .
It is clear that, for a ∈ K a \in K a ∈ K and b ∈ K b \in K b ∈ K , we have N K / Q ( a b ) = N K / Q ( a ) N K / Q ( b ) N_{K/\mathbb{Q}}(ab) = N_{K/\mathbb{Q}}(a)N_{K/\mathbb{Q}}(b) N K / Q ( ab ) = N K / Q ( a ) N K / Q ( b ) . If a ≠ 0 a \ne0 a = 0 , then N K / Q ( a ) ≠ 0 N_{K/\mathbb{Q}}(a) \ne0 N K / Q ( a ) = 0 . If a ∈ O K a \in O_K a ∈ O K , then N K / Q ( a ) ∈ Z N_{K/\mathbb{Q}}(a) \in\mathbb{Z} N K / Q ( a ) ∈ Z . If a ∈ Q a \in\mathbb{Q} a ∈ Q , then N K / Q ( a ) = a [ K : Q ] = a m N_{K/\mathbb{Q}}(a) = a^{[K:\mathbb{Q}]} = a^m N K / Q ( a ) = a [ K : Q ] = a m .
Let a \mathfrak{a} a be a nonzero integral ideal of O K O_K O K . Then, the quotient ring O K / a O_K/\mathfrak{a} O K / a is finite. The norm of a \mathfrak{a} a is defined by N ( a ) = ∣ O K / a ∣ \mathcal{N}(\mathfrak{a}) = |O_K/\mathfrak{a}| N ( a ) = ∣ O K / a ∣ . If p \mathfrak{p} p is a nonzero prime ideal of O K O_K O K , then the residue ring O K / p O_K/\mathfrak{p} O K / p is a finite field, which is called the residue class field of O K O_K O K modulo p \mathfrak{p} p . The norm of p \mathfrak{p} p is N ( p ) = ∣ O K / p ∣ \mathcal{N}(\mathfrak{p}) = |O_K/\mathfrak{p}| N ( p ) = ∣ O K / p ∣ being a prime power. If a \mathfrak{a} a and b \mathfrak{b} b are nonzero integral ideals of O K O_K O K , then N ( a b ) = N ( a ) N ( b ) \mathcal{N}(\mathfrak{a}\mathfrak{b}) = \mathcal{N}(\mathfrak{a})\mathcal{N}(\mathfrak{b}) N ( ab ) = N ( a ) N ( b ) . Note that the norm of the nonzero principal ideal a = a O K \mathfrak{a} = aO_K a = a O K is equal to the absolute value of the field norm N K / Q ( a ) N_{K/\mathbb{Q}}(a) N K / Q ( a ) [42 Corollary 5.10] , i.e., if a ∈ O K a \in O_K a ∈ O K with a ≠ 0 a \ne0 a = 0 , then N ( a O K ) = ∣ N K / Q ( a ) ∣ \mathcal{N}(aO_K) = |N_{K/\mathbb{Q}}(a)| N ( a O K ) = ∣ N K / Q ( a ) ∣ . Therefore, if a ∈ O K a \in O_K a ∈ O K with a ≠ 0 a \ne0 a = 0 and b \mathfrak{b} b is a nonzero integral ideal of O K O_K O K , then N ( a b ) = N ( a O K b ) = N ( a O K ) N ( b ) = ∣ N K / Q ( a ) ∣ N ( b ) \mathcal{N}(a\mathfrak{b}) = \mathcal{N}(aO_K\mathfrak{b}) = \mathcal{N}(aO_K)\mathcal{N}(\mathfrak{b}) = |N_{K/\mathbb{Q}}(a)|\mathcal{N}(\mathfrak{b}) N ( a b ) = N ( a O K b ) = N ( a O K ) N ( b ) = ∣ N K / Q ( a ) ∣ N ( b ) .
The Dedekind zeta function ζ K ( s ) \zeta_K(s) ζ K ( s ) of s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 1 \operatorname{Re}(s) > 1 Re ( s ) > 1 for the number field K K K is defined by
ζ K ( s ) = ∑ a 1 N ( a ) s = ∏ p 1 1 − N ( p ) − s (21) \zeta_K(s) = \sum_{\mathfrak{a}} \frac{1}{\mathcal{N}(\mathfrak{a})^s} = \prod_{\mathfrak{p}} \frac{1}{1-\mathcal{N}(\mathfrak{p})^{-s}} \tag*{(21)} ζ K ( s ) = a ∑ N ( a ) s 1 = p ∏ 1 − N ( p ) − s 1 (21)
where a \mathfrak{a} a varies over the nonzero integral ideals of K K K and p \mathfrak{p} p runs through the nonzero prime ideals of K K K . The infinite product over all nonzero prime ideals p \mathfrak{p} p of O K O_K O K is the Euler product of the field K K K . It is known that the Dedekind zeta function ζ K ( s ) \zeta_K(s) ζ K ( s ) can be analytically continued to C ∖ { 1 } \mathbb{C}\setminus\{1\} C ∖ { 1 } and that it has a simple pole at s = 1 s = 1 s = 1 (see [31 p. 467] ).
In the following, we specify K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) with D = − 4 N < 0 D = -4N < 0 D = − 4 N < 0 for the given squarefree positive integer N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . The field K K K is an imaginary quadratic field with discriminant Δ K = D \Delta_K = D Δ K = D , thus Δ K \Delta_K Δ K is a fundamental discriminant . According to [5 Theorem 2.2.15] , the Kronecker symbol ( Δ K n ) \left(\frac{\Delta_K}{n}\right) ( n Δ K ) , n ∈ Z n \in\mathbb{Z} n ∈ Z , defines a real primitive Dirichlet character modulo ∣ Δ K ∣ = ∣ D ∣ = 4 N |\Delta_K| = |D| = 4N ∣ Δ K ∣ = ∣ D ∣ = 4 N . We note that if n = p n = p n = p , p p p an odd prime, then the Kronecker symbol ( Δ K n ) \left(\frac{\Delta_K}{n}\right) ( n Δ K ) has the same value as that of the Legendre symbol ( Δ K p ) \left(\frac{\Delta_K}{p}\right) ( p Δ K ) . Note that for an odd prime p p p and an integer a ∈ Z a \in\mathbb{Z} a ∈ Z , the Legendre symbol ( a p ) ∈ { − 1 , 0 , 1 } \left(\frac{a}{p}\right) \in\{-1, 0, 1\} ( p a ) ∈ { − 1 , 0 , 1 } is function of a a a and p p p defined as 0 0 0 when p ∣ a p \mid a p ∣ a and ± 1 \pm1 ± 1 depending on whether a a a is a quadratic residue modulo p p p when p ∤ a p \nmid a p ∤ a .
It is well known that, for a prime p ∈ Z p \in\mathbb{Z} p ∈ Z , the principal ideal p O K pO_K p O K decomposes in O K O_K O K as follows. If p ∣ ∣ Δ K ∣ p \mid|\Delta_K| p ∣ ∣ Δ K ∣ , then p p p ramifies in O K O_K O K and p O K = p 2 pO_K = \mathfrak{p}^2 p O K = p 2 where p \mathfrak{p} p is a prime ideal. In this case, ( Δ K p ) = 0 \left(\frac{\Delta_K}{p}\right) = 0 ( p Δ K ) = 0 . In particular, since 2 ∣ ∣ Δ K ∣ 2 \mid|\Delta_K| 2 ∣ ∣ Δ K ∣ , we have ( Δ K 2 ) = 0 \left(\frac{\Delta_K}{2}\right) = 0 ( 2 Δ K ) = 0 and 2 O K = p ^ 2 2O_K = \widehat{\mathfrak{p}}^2 2 O K = p 2 , where p ^ ∣ 2 \widehat{\mathfrak{p}} \mid2 p ∣ 2 is the prime ideal of O K O_K O K lying over the prime number 2 ∈ Z 2 \in\mathbb{Z} 2 ∈ Z .
If p p p is an odd prime and ( Δ K p ) = 1 \left(\frac{\Delta_K}{p}\right) = 1 ( p Δ K ) = 1 , then p p p splits completely in O K O_K O K and p O K = p 1 p 2 pO_K = \mathfrak{p}_1\mathfrak{p}_2 p O K = p 1 p 2 where p 1 \mathfrak{p}_1 p 1 and p 2 \mathfrak{p}_2 p 2 are two different prime ideals. If p p p is an odd prime and ( Δ K p ) = − 1 \left(\frac{\Delta_K}{p}\right) = -1 ( p Δ K ) = − 1 , then p p p is inert in O K O_K O K and p = p O K \mathfrak{p} = pO_K p = p O K itself is a prime ideal of O K O_K O K .
Let χ ( n ) = ( Δ K n ) \chi(n) = \left(\frac{\Delta_K}{n}\right) χ ( n ) = ( n Δ K ) be the real primitive Dirichlet character of K K K and L ( s , χ ) L(s,\chi) L ( s , χ ) be the corresponding Dirichlet L L L -function. According to the above decomposition rule and the Euler product in (21) , we can obtain
ζ K ( s ) = ζ ( s ) L ( s , χ ) (22) \zeta_K(s) = \zeta(s)L(s,\chi) \tag*{(22)} ζ K ( s ) = ζ ( s ) L ( s , χ ) (22)
for s ∈ C s \in\mathbb{C} s ∈ C with Re ( s ) > 1 \operatorname{Re}(s)>1 Re ( s ) > 1 , where ζ ( s ) \zeta(s) ζ ( s ) is the Riemann zeta function. The Riemann zeta function ζ ( s ) \zeta(s) ζ ( s ) is a meromorphic function on the whole complex plane, which is holomorphic everywhere except for a simple pole at s = 1 s=1 s = 1 with residue 1. Since χ \chi χ is nonprincipal and primitive, Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) is entire on the whole complex plane s ∈ C s \in\mathbb{C} s ∈ C . Thus, the above relation (22) can be employed as an alternative approach to providing the same analytic continuation of ζ K ( s ) \zeta_K(s) ζ K ( s ) defined in (21) to C ∖ { 1 } \mathbb{C}\setminus\{1\} C ∖ { 1 } which has a simple pole at s = 1 s=1 s = 1 with residue L ( 1 , χ ) L(1,\chi) L ( 1 , χ ) . It is known that L ( 1 , χ ) ≠ 0 L(1,\chi) \ne0 L ( 1 , χ ) = 0 (see [2 Theorem 6.20] ).
Remind that K = Q ( Δ K ) K=\mathbb{Q}(\sqrt{\Delta_K}) K = Q ( Δ K ) is an imaginary quadratic field with discriminant Δ K = − 4 N < 0 \Delta_K=-4N<0 Δ K = − 4 N < 0 . Let
ξ K ( s ) = ( ∣ Δ K ∣ 4 π 2 ) s / 2 Γ ( s ) ζ K ( s ) \xi_K(s)=\left(\frac{|\Delta_K|}{4\pi^2}\right)^{s/2}\Gamma(s)\zeta_K(s) ξ K ( s ) = ( 4 π 2 ∣ Δ K ∣ ) s /2 Γ ( s ) ζ K ( s )
where Γ ( s ) \Gamma(s) Γ ( s ) is the gamma function. It is known that ξ K ( s ) \xi_K(s) ξ K ( s ) satisfies the following functional equation (see Stark [41, p. 965])
ξ K ( 1 − s ) = ξ K ( s ) . \xi_K(1-s)=\xi_K(s). ξ K ( 1 − s ) = ξ K ( s ) .
Therefore, the functional equation for ζ K ( s ) \zeta_K(s) ζ K ( s ) is given by
Γ ( 1 − s ) ζ K ( 1 − s ) = ( ∣ Δ K ∣ 4 π 2 ) s − 1 / 2 Γ ( s ) ζ K ( s ) . (23) \Gamma(1-s)\zeta_K(1-s)=\left(\frac{|\Delta_K|}{4\pi^2}\right)^{s-1/2}\Gamma(s)\zeta_K(s). \tag*{(23)} Γ ( 1 − s ) ζ K ( 1 − s ) = ( 4 π 2 ∣ Δ K ∣ ) s − 1/2 Γ ( s ) ζ K ( s ) . (23)
We need to use the following lemma (see Ingham [21, p. 31]).
Lemma 4. If k k k is a positive integer, c > 0 c>0 c > 0 , y > 0 y>0 y > 0 , then
1 2 π i ∫ c − i ∞ c + i ∞ y s s ( s + 1 ) ⋯ ( s + k ) d s = { 0 , if y < 1 , 1 k ! ( 1 − 1 y ) k , if y ≥ 1. (24) \frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{y^s}{s(s+1)\cdots(s+k)}\,\mathrm{d}s=
\begin{cases}
0, & \text{if } y<1,\\
\frac{1}{k!}\left(1-\frac{1}{y}\right)^k, & \text{if } y\ge1.
\end{cases} \tag*{(24)} 2 π i 1 ∫ c − i ∞ c + i ∞ s ( s + 1 ) ⋯ ( s + k ) y s d s = ⎩ ⎨ ⎧ 0 , k ! 1 ( 1 − y 1 ) k , if y < 1 , if y ≥ 1. (24)
Lemma 4 implies the following result.
Lemma 5. If k k k is a positive integer with k ≥ 2 k\ge2 k ≥ 2 , c > 0 c>0 c > 0 , y > 0 y>0 y > 0 , then
1 2 π i ∫ c − i ∞ c + i ∞ y s s ( s + 2 ) ⋯ ( s + k ) d s = { 0 , if y < 1 , 1 k ! ( 1 − 1 y ) k − 1 ( 1 + k − 1 y ) , if y ≥ 1. (25) \frac{1}{2\pi i}\int_{c-i\infty}^{c+i\infty}\frac{y^s}{s(s+2)\cdots(s+k)}\,\mathrm{d}s=
\begin{cases}
0, & \text{if } y<1,\\
\frac{1}{k!}\left(1-\frac{1}{y}\right)^{k-1}\left(1+\frac{k-1}{y}\right), & \text{if } y\ge1.
\end{cases} \tag*{(25)} 2 π i 1 ∫ c − i ∞ c + i ∞ s ( s + 2 ) ⋯ ( s + k ) y s d s = ⎩ ⎨ ⎧ 0 , k ! 1 ( 1 − y 1 ) k − 1 ( 1 + y k − 1 ) , if y < 1 , if y ≥ 1. (25)
In fact, noticing that 1 s = 1 s ( s + 1 ) + 1 s + 1 \frac{1}{s}=\frac{1}{s(s+1)}+\frac{1}{s+1} s 1 = s ( s + 1 ) 1 + s + 1 1 and using the change of variable, for y ≥ 1 y\ge1 y ≥ 1 the integral in (25) equals
1 k ! ( 1 − 1 y ) k + 1 y ( k − 1 ) ! ( 1 − 1 y ) k − 1 = 1 k ! ( 1 − 1 y ) k − 1 ( 1 − 1 y + k y ) \frac{1}{k!}\left(1-\frac{1}{y}\right)^k+\frac{1}{y(k-1)!}\left(1-\frac{1}{y}\right)^{k-1}=\frac{1}{k!}\left(1-\frac{1}{y}\right)^{k-1}\left(1-\frac{1}{y}+\frac{k}{y}\right) k ! 1 ( 1 − y 1 ) k + y ( k − 1 )! 1 ( 1 − y 1 ) k − 1 = k ! 1 ( 1 − y 1 ) k − 1 ( 1 − y 1 + y k )
Since L ( 1 , χ ) ≠ 0 L(1,\chi)\ne0 L ( 1 , χ ) = 0 , there exists a number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) such that L ( s , χ ) ≠ 0 L(s,\chi)\ne0 L ( s , χ ) = 0 for s s s in the interval [ β , 1 ) [\beta,1) [ β , 1 ) . Note that the series ζ K ( s ) \zeta_K(s) ζ K ( s ) in (21) converges absolutely and uniformly in the domain Re ( s ) ≥ 1 + δ \operatorname{Re}(s)\ge1+\delta Re ( s ) ≥ 1 + δ for every δ > 0 \delta>0 δ > 0 and that ζ K ( s ) = ζ ( s ) L ( s , χ ) \zeta_K(s)=\zeta(s)L(s,\chi) ζ K ( s ) = ζ ( s ) L ( s , χ ) . For a given number y > 1 y>1 y > 1 , we consider the following complex integral
I = 1 2 π i ∫ 2 − β − i ∞ 2 − β + i ∞ ζ K ( s + β ) y s s ∏ ℓ = 2 9 ( s + ℓ ) d s = ∑ N ( a ) ≤ y 1 N ( a ) β 1 9 ! ( 1 − N ( a ) y ) 8 ( 1 + 8 N ( a ) y ) (26) \mathcal{I}=\frac{1}{2\pi i}\int_{2-\beta-i\infty}^{2-\beta+i\infty}\frac{\zeta_K(s+\beta)y^s}{s\prod_{\ell=2}^{9}(s+\ell)}\,ds=\sum_{\mathfrak{N}(a)\le y}\frac{1}{\mathfrak{N}(a)^\beta}\frac{1}{9!}\left(1-\frac{\mathfrak{N}(a)}{y}\right)^8\left(1+8\frac{\mathfrak{N}(a)}{y}\right) \tag*{(26)} I = 2 π i 1 ∫ 2 − β − i ∞ 2 − β + i ∞ s ∏ ℓ = 2 9 ( s + ℓ ) ζ K ( s + β ) y s d s = N ( a ) ≤ y ∑ N ( a ) β 1 9 ! 1 ( 1 − y N ( a ) ) 8 ( 1 + 8 y N ( a ) ) (26)
where we choose c = 2 − β > 1 c=2-\beta>1 c = 2 − β > 1 and k = 9 k=9 k = 9 in Lemma 5. It is clear that I \mathcal{I} I is a real number.
Let T ≥ 2 T\ge2 T ≥ 2 be any given truncation parameter and the complex variable s = σ + i t s=\sigma+it s = σ + i t . Define the function
f ( s ) = ζ K ( s + β ) y s s ∏ ℓ = 2 9 ( s + ℓ ) = ζ ( s + β ) L ( s + β , χ ) y s s ∏ ℓ = 2 9 ( s + ℓ ) (27) f(s)=\frac{\zeta_K(s+\beta)y^s}{s\prod_{\ell=2}^{9}(s+\ell)}=\frac{\zeta(s+\beta)L(s+\beta,\chi)y^s}{s\prod_{\ell=2}^{9}(s+\ell)} \tag*{(27)} f ( s ) = s ∏ ℓ = 2 9 ( s + ℓ ) ζ K ( s + β ) y s = s ∏ ℓ = 2 9 ( s + ℓ ) ζ ( s + β ) L ( s + β , χ ) y s (27)
in the rectangle R \mathcal{R} R with − 1 − β ≤ σ ≤ 2 − β -1-\beta\le\sigma\le2-\beta − 1 − β ≤ σ ≤ 2 − β and − T ≤ t ≤ T -T\le t\le T − T ≤ t ≤ T . It is seen that f ( s ) f(s) f ( s ) has two poles in R \mathcal{R} R at s = 0 s=0 s = 0 and s = 1 − β ∈ ( 0 , 1 ) s=1-\beta\in(0,1) s = 1 − β ∈ ( 0 , 1 ) .
Using Cauchy’s residue theorem, we obtain
1 2 π i { ∫ 2 − β − i T 2 − β + i T − ∫ − 1 − β + i T 2 − β + i T − ∫ − 1 − β − i T − 1 − β + i T + ∫ − 1 − β − i T 2 − β − i T } f ( s ) d s = Res s = 0 f ( s ) + Res s = 1 − β f ( s ) . (28) \frac{1}{2\pi i}\left\{\int_{2-\beta-iT}^{2-\beta+iT}-\int_{-1-\beta+iT}^{2-\beta+iT}-\int_{-1-\beta-iT}^{-1-\beta+iT}+\int_{-1-\beta-iT}^{2-\beta-iT}\right\}f(s)\,ds=\underset{s=0}{\operatorname{Res}}\,f(s)+\underset{s=1-\beta}{\operatorname{Res}}\,f(s). \tag*{(28)} 2 π i 1 { ∫ 2 − β − i T 2 − β + i T − ∫ − 1 − β + i T 2 − β + i T − ∫ − 1 − β − i T − 1 − β + i T + ∫ − 1 − β − i T 2 − β − i T } f ( s ) d s = s = 0 Res f ( s ) + s = 1 − β Res f ( s ) . (28)
We can prove that the second and fourth integrals in (28) will approach 0 as T → ∞ T\to\infty T → ∞ .
Lemma 6 For a given number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) and the function f ( s ) f(s) f ( s ) given in (27), we have
lim T → ∞ ∫ − 1 − β + i T 2 − β + i T f ( s ) d s = lim T → ∞ ∫ − 1 − β − i T 2 − β − i T f ( s ) d s = 0. (29) \lim_{T\to\infty}\int_{-1-\beta+iT}^{2-\beta+iT}f(s)\,ds=\lim_{T\to\infty}\int_{-1-\beta-iT}^{2-\beta-iT}f(s)\,ds=0. \tag*{(29)} T → ∞ lim ∫ − 1 − β + i T 2 − β + i T f ( s ) d s = T → ∞ lim ∫ − 1 − β − i T 2 − β − i T f ( s ) d s = 0. (29)
Proof. See Appendix A.
For the given number y > 1 y>1 y > 1 , we consider the following complex integral
J = 1 2 π i ∫ − 1 − β − i ∞ − 1 − β + i ∞ ζ K ( s + β ) y s s ∏ ℓ = 2 9 ( s + ℓ ) d s . (30) \mathcal{J}=\frac{1}{2\pi i}\int_{-1-\beta-i\infty}^{-1-\beta+i\infty}\frac{\zeta_K(s+\beta)y^s}{s\prod_{\ell=2}^{9}(s+\ell)}\,ds. \tag*{(30)} J = 2 π i 1 ∫ − 1 − β − i ∞ − 1 − β + i ∞ s ∏ ℓ = 2 9 ( s + ℓ ) ζ K ( s + β ) y s d s . (30)
Noting (26) , (29) and (30) and taking the limit T → ∞ T \to\infty T → ∞ in (28) , we can obtain
I = J + R e s s = 0 f ( s ) + R e s s = 1 − β f ( s ) . (31) \mathcal{I}=\mathcal{J}+\mathop{\mathrm{Res}}_{s=0}f(s)+\mathop{\mathrm{Res}}_{s=1-\beta}f(s). \tag*{(31)} I = J + Res s = 0 f ( s ) + Res s = 1 − β f ( s ) . (31)
Since the class number h K ≥ 1 h_K \ge1 h K ≥ 1 , we see from the class number formula (15) that
L ( 1 , χ ) = h K π ∣ D ∣ > 0. L(1,\chi)=h_K\frac{\pi}{\sqrt{\lvert D\rvert}}>0. L ( 1 , χ ) = h K ∣ D ∣ π > 0.
Hence, L ( s , χ ) > 0 L(s,\chi)>0 L ( s , χ ) > 0 for s s s in the interval [ β , 1 ) [\beta,1) [ β , 1 ) . Therefore, we have
R e s s = 0 f ( s ) = 1 9 ! ζ K ( β ) = 1 9 ! ζ ( β ) L ( β , χ ) < 0. \mathop{\mathrm{Res}}_{s=0}f(s)=\frac{1}{9!}\zeta_K(\beta)=\frac{1}{9!}\zeta(\beta)L(\beta,\chi)<0. Res s = 0 f ( s ) = 9 ! 1 ζ K ( β ) = 9 ! 1 ζ ( β ) L ( β , χ ) < 0.
Note that ζ ( s ) \zeta(s) ζ ( s ) has a simple pole at s = 1 s=1 s = 1 with residue 1. We get
R e s s = 1 − β f ( s ) = L ( 1 , χ ) 1 − β y 1 − β ∏ ℓ = 2 9 ( 1 − β + ℓ ) < y 1 − β 9 ! L ( 1 , χ ) 1 − β . \mathop{\mathrm{Res}}_{s=1-\beta}f(s)=\frac{L(1,\chi)}{1-\beta}\frac{y^{1-\beta}}{\prod_{\ell=2}^{9}(1-\beta+\ell)}<\frac{y^{1-\beta}}{9!}\frac{L(1,\chi)}{1-\beta}. Res s = 1 − β f ( s ) = 1 − β L ( 1 , χ ) ∏ ℓ = 2 9 ( 1 − β + ℓ ) y 1 − β < 9 ! y 1 − β 1 − β L ( 1 , χ ) .
We can deduce from (31) that the complex integral J \mathcal{J} J defined in (30) is a real number and that
I < J + y 1 − β 9 ! L ( 1 , χ ) 1 − β . (32) \mathcal{I}<\mathcal{J}+\frac{y^{1-\beta}}{9!}\frac{L(1,\chi)}{1-\beta}. \tag*{(32)} I < J + 9 ! y 1 − β 1 − β L ( 1 , χ ) . (32)
It is now ready to prove an explicit Hecke’s theorem as follows.
Proposition 7 Let N N N be any given squarefree positive integer with N > 7.5 ⋅ 10 19 N>7.5\cdot10^{19} N > 7.5 ⋅ 1 0 19 and N ≡ 1 , 2 ( m o d 4 ) N\equiv1,2\pmod{4} N ≡ 1 , 2 ( mod 4 ) and put D = − 4 N D=-4N D = − 4 N and q = 4 N q=4N q = 4 N . Let χ ( n ) = ( D n ) \chi(n)=\left(\frac{D}{n}\right) χ ( n ) = ( n D ) be the Kronecker symbol with the corresponding Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) . If L ( s , χ ) ≠ 0 L(s,\chi)\ne0 L ( s , χ ) = 0 for s s s in the interval [ β , 1 ) [\beta,1) [ β , 1 ) with 0 < β < 1 0<\beta<1 0 < β < 1 satisfying
1 − β < 0.015 1 log q , (33) 1-\beta<0.015\frac{1}{\log q}, \tag*{(33)} 1 − β < 0.015 log q 1 , (33) then
L ( 1 , χ ) > 2.43 ( 1 − β ) . (34) L(1,\chi)>2.43(1-\beta). \tag*{(34)} L ( 1 , χ ) > 2.43 ( 1 − β ) . (34)
Proof: We choose the number y = q = 4 N > 3 ⋅ 10 20 y=q=4N>3\cdot10^{20} y = q = 4 N > 3 ⋅ 1 0 20 . By using s = − 1 − β + i t s=-1-\beta+it s = − 1 − β + i t with − ∞ < t < ∞ -\infty<t<\infty − ∞ < t < ∞ , the complex integral J ∈ R \mathcal{J}\in\mathbb{R} J ∈ R given by (30) can be rewritten as
J = 1 2 π ∫ − ∞ ∞ ζ K ( − 1 + i t ) y − 1 − β + i t ( − 1 − β + i t ) ∏ ℓ = 2 9 ( − 1 − β + i t + ℓ ) d t = 1 2 π ∫ − ∞ ∞ f ( − 1 − β + i t ) d t . (35) \mathcal{J}=\frac{1}{2\pi}\int_{-\infty}^{\infty}\frac{\zeta_K(-1+it)y^{-1-\beta+it}}{(-1-\beta+it)\prod_{\ell=2}^{9}(-1-\beta+it+\ell)}\,dt=\frac{1}{2\pi}\int_{-\infty}^{\infty}f(-1-\beta+it)\,dt. \tag*{(35)} J = 2 π 1 ∫ − ∞ ∞ ( − 1 − β + i t ) ∏ ℓ = 2 9 ( − 1 − β + i t + ℓ ) ζ K ( − 1 + i t ) y − 1 − β + i t d t = 2 π 1 ∫ − ∞ ∞ f ( − 1 − β + i t ) d t . (35)
By using the functional equation (23) for ζ K ( s ) \zeta_K(s) ζ K ( s ) , we get
Γ ( − 1 + i t ) ζ K ( − 1 + i t ) = ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 − i t Γ ( 2 − i t ) ζ K ( 2 − i t ) (36) \Gamma(-1+it)\zeta_K(-1+it)=\left(\frac{\lvert\Delta_K\rvert}{4\pi^2}\right)^{3/2-it}\Gamma(2-it)\zeta_K(2-it) \tag*{(36)} Γ ( − 1 + i t ) ζ K ( − 1 + i t ) = ( 4 π 2 ∣ Δ K ∣ ) 3/2 − i t Γ ( 2 − i t ) ζ K ( 2 − i t ) (36)
where ∣ Δ K ∣ = 4 N = q |\Delta_K|=4N=q ∣ Δ K ∣ = 4 N = q . Hence,
∣ ζ K ( − 1 + i t ) ∣ = ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 ∣ Γ ( 2 − i t ) ∣ ∣ Γ ( − 1 + i t ) ∣ ∣ ζ K ( 2 − i t ) ∣ = ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 ∣ t ∣ ( 1 + t 2 ) ∣ ζ K ( 2 − i t ) ∣ . (37) |\zeta_K(-1+it)|=\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}\frac{|\Gamma(2-it)|}{|\Gamma(-1+it)|}|\zeta_K(2-it)|=\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}|t|(1+t^2)|\zeta_K(2-it)|. \tag*{(37)} ∣ ζ K ( − 1 + i t ) ∣ = ( 4 π 2 ∣ Δ K ∣ ) 3/2 ∣Γ ( − 1 + i t ) ∣ ∣Γ ( 2 − i t ) ∣ ∣ ζ K ( 2 − i t ) ∣ = ( 4 π 2 ∣ Δ K ∣ ) 3/2 ∣ t ∣ ( 1 + t 2 ) ∣ ζ K ( 2 − i t ) ∣. (37)
Note that ∣ ζ K ( 2 − i t ) ∣ ≤ ∣ ζ K ( 2 ) ∣ = ∣ ζ ( 2 ) ∣ ⋅ ∣ L ( 2 , χ ) ∣ ≤ ∣ ζ ( 2 ) ∣ 2 = π 4 / 36 |\zeta_K(2-it)|\leq|\zeta_K(2)|=|\zeta(2)|\cdot|L(2,\chi)|\leq|\zeta(2)|^2=\pi^4/36 ∣ ζ K ( 2 − i t ) ∣ ≤ ∣ ζ K ( 2 ) ∣ = ∣ ζ ( 2 ) ∣ ⋅ ∣ L ( 2 , χ ) ∣ ≤ ∣ ζ ( 2 ) ∣ 2 = π 4 /36 . We have
∣ f ( − 1 − β + i t ) ∣ ≤ 1 y 1 + β π 4 36 ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 ∣ t ∣ ( 1 + t 2 ) ∣ − 1 − β + i t ∣ ∏ ℓ = 2 ∞ ∣ − 1 − β + i t + ℓ ∣ |f(-1-\beta+it)|\leq\frac{1}{y^{1+\beta}}\frac{\pi^4}{36}\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}\frac{|t|(1+t^2)}{|-1-\beta+it|\prod_{\ell=2}^{\infty}|-1-\beta+it+\ell|} ∣ f ( − 1 − β + i t ) ∣ ≤ y 1 + β 1 36 π 4 ( 4 π 2 ∣ Δ K ∣ ) 3/2 ∣ − 1 − β + i t ∣ ∏ ℓ = 2 ∞ ∣ − 1 − β + i t + ℓ ∣ ∣ t ∣ ( 1 + t 2 )
and hence
∣ f ( − 1 − β + i t ) ∣ ≤ 1 y 1 + β π 4 36 ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 1 1 + t 2 . (38) |f(-1-\beta+it)|\leq\frac{1}{y^{1+\beta}}\frac{\pi^4}{36}\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}\frac{1}{1+t^2}. \tag*{(38)} ∣ f ( − 1 − β + i t ) ∣ ≤ y 1 + β 1 36 π 4 ( 4 π 2 ∣ Δ K ∣ ) 3/2 1 + t 2 1 . (38)
Thus, we obtain
∣ J ∣ ≤ 1 2 π ∫ − ∞ ∞ ∣ f ( − 1 − β + i t ) ∣ d t ≤ 1 2 1 y 1 + β π 4 36 ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 . (39) |\mathcal J|\leq\frac{1}{2\pi}\int_{-\infty}^{\infty}|f(-1-\beta+it)|dt\leq\frac{1}{2}\frac{1}{y^{1+\beta}}\frac{\pi^4}{36}\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}. \tag*{(39)} ∣ J ∣ ≤ 2 π 1 ∫ − ∞ ∞ ∣ f ( − 1 − β + i t ) ∣ d t ≤ 2 1 y 1 + β 1 36 π 4 ( 4 π 2 ∣ Δ K ∣ ) 3/2 . (39)
It follows from (26) that
I ≥ ∑ N ( a ) ≤ y 1 N ( a ) 1 9 ! ( 1 − 8 N ( a ) y ) ( 1 + 8 N ( a ) y ) ≥ ∑ N ( a ) ≤ 2 ⋅ 10 10 1 N ( a ) 1 9 ! ( 1 − 64 N ( a ) 2 y 2 ) . (40) \mathcal I\geq\sum_{\mathcal N(\mathfrak a)\leq y}\frac{1}{\mathcal N(\mathfrak a)}\frac{1}{9!}\left(1-8\frac{\mathcal N(\mathfrak a)}{y}\right)\left(1+8\frac{\mathcal N(\mathfrak a)}{y}\right)\geq\sum_{\mathcal N(\mathfrak a)\leq2\cdot10^{10}}\frac{1}{\mathcal N(\mathfrak a)}\frac{1}{9!}\left(1-64\frac{\mathcal N(\mathfrak a)^2}{y^2}\right). \tag*{(40)} I ≥ N ( a ) ≤ y ∑ N ( a ) 1 9 ! 1 ( 1 − 8 y N ( a ) ) ( 1 + 8 y N ( a ) ) ≥ N ( a ) ≤ 2 ⋅ 1 0 10 ∑ N ( a ) 1 9 ! 1 ( 1 − 64 y 2 N ( a ) 2 ) . (40)
Combining (32) with (39) and (40) , we obtain
L ( 1 , χ ) 1 − β > 9 ! y 1 − β I − 9 ! y 1 − β ∣ J ∣ ≥ 0.9999 y 1 − β ∑ N ( a ) ≤ 2 ⋅ 10 10 1 N ( a ) − 9 ! π 4 72 y 2 ( y 4 π 2 ) 3 / 2 . \frac{L(1,\chi)}{1-\beta}>\frac{9!}{y^{1-\beta}}\mathcal I-\frac{9!}{y^{1-\beta}}|\mathcal J|\geq\frac{0.9999}{y^{1-\beta}}\sum_{\mathcal N(\mathfrak a)\leq2\cdot10^{10}}\frac{1}{\mathcal N(\mathfrak a)}-\frac{9!\pi^4}{72y^2}\left(\frac{y}{4\pi^2}\right)^{3/2}. 1 − β L ( 1 , χ ) > y 1 − β 9 ! I − y 1 − β 9 ! ∣ J ∣ ≥ y 1 − β 0.9999 N ( a ) ≤ 2 ⋅ 1 0 10 ∑ N ( a ) 1 − 72 y 2 9 ! π 4 ( 4 π 2 y ) 3/2 .
Noticing that y 1 − β ≤ q 0.015 1 log q = e 0.015 y^{1-\beta}\leq q^{0.015\frac{1}{\log q}}=e^{0.015} y 1 − β ≤ q 0.015 l o g q 1 = e 0.015 by (33) , we get
L ( 1 , χ ) 1 − β > 0.985 ∑ N ( a ) ≤ 2 ⋅ 10 10 1 N ( a ) − 0.000001. (41) \frac{L(1,\chi)}{1-\beta}>0.985\sum_{\mathcal N(\mathfrak a)\leq2\cdot10^{10}}\frac{1}{\mathcal N(\mathfrak a)}-0.000001. \tag*{(41)} 1 − β L ( 1 , χ ) > 0.985 N ( a ) ≤ 2 ⋅ 1 0 10 ∑ N ( a ) 1 − 0.000001. (41)
For every positive integer a ≤ 10 5 a\leq10^5 a ≤ 1 0 5 , we choose a = a O K \mathfrak a=aO_K a = a O K and have N ( a ) = N ( a O K ) = ∣ N K / Q ( a ) ∣ = a [ K : Q ] = a 2 ≤ 10 10 \mathcal N(\mathfrak a)=\mathcal N(aO_K)=|\mathrm{N}_{K/\mathbb Q}(a)|=a^{[K:\mathbb Q]}=a^2\leq10^{10} N ( a ) = N ( a O K ) = ∣ N K / Q ( a ) ∣ = a [ K : Q ] = a 2 ≤ 1 0 10 . Moreover, we have seen that 2 O K = p ^ 2 2O_K=\hat{\mathfrak p}^{\,2} 2 O K = p ^ 2 , where p ^ ∣ 2 \hat{\mathfrak p}\mid2 p ^ ∣ 2 is the prime ideal of O K O_K O K lying over 2 ∈ Z 2\in\mathbb Z 2 ∈ Z , and hence N ( p ^ ) = 2 \mathcal N(\hat{\mathfrak p})=2 N ( p ^ ) = 2 . We choose a = a p ^ \mathfrak a=a\hat{\mathfrak p} a = a p ^ with N ( a ) = ∣ N K / Q ( a ) ∣ N ( p ^ ) = 2 a 2 ≤ 2 ⋅ 10 10 \mathcal N(\mathfrak a)=|\mathrm{N}_{K/\mathbb Q}(a)|\mathcal N(\hat{\mathfrak p})=2a^2\leq2\cdot10^{10} N ( a ) = ∣ N K / Q ( a ) ∣ N ( p ^ ) = 2 a 2 ≤ 2 ⋅ 1 0 10 . It is obvious that, for any two positive integers a 1 a_1 a 1 and a 2 a_2 a 2 , there is a 1 2 ≠ 2 a 2 2 a_1^2\ne2a_2^2 a 1 2 = 2 a 2 2 , since 2 \sqrt{2} 2 is irrational. Therefore, it follows from (41) that
L ( 1 , χ ) 1 − β > 0.985 ∑ n = 1 100000 ( 1 n 2 + 1 2 n 2 ) − 0.000001 > 2.43 \frac{L(1,\chi)}{1-\beta}>0.985\sum_{n=1}^{100000}\left(\frac{1}{n^2}+\frac{1}{2n^2}\right)-0.000001>2.43 1 − β L ( 1 , χ ) > 0.985 n = 1 ∑ 100000 ( n 2 1 + 2 n 2 1 ) − 0.000001 > 2.43
which implies (34) . The proof of Proposition 7 is complete.
An explicit zero-free interval for L ( s , χ ) L(s,\chi) L ( s , χ )
Let N N N be any given squarefree positive integer with N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 and N ≡ 1 , 2 ( m o d 4 ) N \equiv1,2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . The Kronecker symbol χ ( n ) = ( Δ K n ) \chi(n) = \left(\frac{\Delta_K}{n}\right) χ ( n ) = ( n Δ K ) is the real primitive Dirichlet character of the imaginary quadratic field K = Q ( Δ K ) K = \mathbb{Q}(\sqrt{\Delta_K}) K = Q ( Δ K ) with discriminant Δ K = − 4 N < 0 \Delta_K = -4N < 0 Δ K = − 4 N < 0 . The corresponding Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) of s ∈ C s \in\mathbb{C} s ∈ C is entire on the complex plane. We need to search for a number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) such that L ( s , χ ) ≠ 0 L(s,\chi) \ne0 L ( s , χ ) = 0 for s s s in the interval ( β , 1 ) (\beta,1) ( β , 1 ) .
Hecke’s conjecture on Dirichlet L L L -functions asserts that, for every real Dirichlet character χ \chi χ , we have L ( s , χ ) ≠ 0 L(s,\chi) \ne0 L ( s , χ ) = 0 for 0 < s < 1 0 < s < 1 0 < s < 1 (See Pan and Pan [34 p. 267] ). It is a long-standing conjecture in analytic number theory. As stated in Theorem 3, in 1918 Hecke [18 ] recognized the profound impact that nonexistence of real zeros of Dirichlet L L L -functions associated with real quadratic characters have on the arithmetic of quadratic fields.
A partial result in the respect was given by Conrey and Soundararajan [7 p. 2, Theorem 1] . Their result demonstrated that Hecke’s conjecture holds true for imaginary quadratic fields K = Q ( − 8 d ) K = \mathbb{Q}(\sqrt{-8d}) K = Q ( − 8 d ) for at least 20% of the odd square free integers d d d . A computational work was performed by Watkins [46 ] , who showed that L ( s , χ ) L(s,\chi) L ( s , χ ) has no positive real zero for each real odd Dirichlet character of modulus up to 3 ⋅ 10 8 3 \cdot10^8 3 ⋅ 1 0 8 .
In the other respect, if a real zero β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) very close to 1 exists, i.e., L ( β , χ ) = 0 L(\beta,\chi) = 0 L ( β , χ ) = 0 , it is called a Landau–Siegel zero or an exceptional zero. In 1935, Siegel [39 ] proved that for any given number ε > 0 \varepsilon> 0 ε > 0 and any real primitive character χ \chi χ of modulus q q q , if β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) is a real zero of L ( s , χ ) L(s,\chi) L ( s , χ ) , then 1 − β ≥ C ( ε ) q − ε 1-\beta\ge C(\varepsilon)q^{-\varepsilon} 1 − β ≥ C ( ε ) q − ε for some constant C ( ε ) > 0 C(\varepsilon) > 0 C ( ε ) > 0 . Thus, if 1 − β < C ( ε ) q − ε 1-\beta< C(\varepsilon)q^{-\varepsilon} 1 − β < C ( ε ) q − ε , then L ( s , χ ) ≠ 0 L(s,\chi) \ne0 L ( s , χ ) = 0 for β < s < 1 \beta< s < 1 β < s < 1 . Note that the constant C ( ε ) C(\varepsilon) C ( ε ) in Siegel’s result is ineffective , in that whose existence is proven but it cannot be explicitly calculated by the proof method. The Siegel–Walfisz theorem was obtained in 1937 (see Walfisz [45 ] ) as an application of Siegel’s result to distribution of primes in arithmetic progressions. With the Siegel–Walfisz theorem, in 1937 Vinogradov [44 ] found an unconditional proof of the three primes Goldbach conjecture for all sufficiently large odd integers. Vinogradov’s theorem states that every sufficiently large odd integer can be expressed as a sum of three primes. Hardy and Littlewood [17 ] had shown earlier in 1924 that the same result follows from the generalized Riemann hypothesis (GRH), which, however, remains unproven.
The most recent progress in Landau–Siegel zeros was presented in manuscript [32 ] titled “Uniform exclusion of Landau–Siegel zeros”. It claims that there is an absolute constant c > 0 c > 0 c > 0 such that every real zero β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) of every primitive nonprincipal real Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) of conductor q ≥ 3 q \ge3 q ≥ 3 , if it exists, satisfies ( 1 − β ) log q ≥ c (1-\beta)\log q \ge c ( 1 − β ) log q ≥ c . In other words, if 1 − β < c log q 1-\beta< \frac{c}{\log q} 1 − β < l o g q c , then L ( s , χ ) ≠ 0 L(s,\chi) \ne0 L ( s , χ ) = 0 for β < s < 1 \beta< s < 1 β < s < 1 . This assertion has remarkably improved Siegel’s result.
In the current scenario, we need to obtain an effective zero-free interval for L ( s , χ ) L(s,\chi) L ( s , χ ) with explicit constants, which will be utilized by the application for establishing Theorem 1. The arguments presented in [32 ] are the basis and will be employed appropriately. We will develop a rigorous and complete proof, which involves detailed derivations that do not appear in [32 ] . Moreover, effective constants will be explicitly calculated , which are best suited to the demands of the application.
In the following, we will focus on Dirichlet L L L -functions L ( s , χ ) L(s,\chi) L ( s , χ ) associated with the real primitive Dirichlet characters χ \chi χ of the imaginary quadratic fields K = Q ( − N ) K = \mathbb{Q}(\sqrt{-N}) K = Q ( − N ) with discriminant Δ K = − 4 N < 0 \Delta_K = -4N < 0 Δ K = − 4 N < 0 for all squarefree positive integers N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) . Let
q = ∣ Δ K ∣ = 4 N > 3 ⋅ 10 20 , q = \lvert\Delta_K\rvert= 4N > 3 \cdot10^{20}, q = ∣ Δ K ∣ = 4 N > 3 ⋅ 1 0 20 ,
then χ ( n ) = ( Δ K n ) ∈ { − 1 , 0 , 1 } \chi(n) = \left(\frac{\Delta_K}{n}\right) \in\{-1,0,1\} χ ( n ) = ( n Δ K ) ∈ { − 1 , 0 , 1 } is a real primitive Dirichlet character of modulus q q q .
It appears to be out of reach of the current approach to prove Hecke’s conjecture for the nonexistence of real zeros of Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) for every real Dirichlet character χ \chi χ in the whole interval 0 < s < 1 0 < s < 1 0 < s < 1 . Instead, we will investigate, under the assumption that a real zero β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) of L ( s , χ ) L(s,\chi) L ( s , χ ) exists, how far that β \beta β has to be away from 1. We assume that the number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) is the Landau–Siegel zero of L ( s , χ ) L(s,\chi) L ( s , χ ) satisfying
L ( β , χ ) = 0. L(\beta,\chi) = 0. L ( β , χ ) = 0.
It is known that the Landau–Siegel zero β \beta β is real and simple (see [27 p. 360, Theorem 11.3] ).
According to Stark’s lemma [40 p. 139, Lemma 3] (see also Stark [41, p. 965]), which can be derived from the Hadamard product of the Dedekind zeta-function of ζ K ( s ) \zeta_K(s) ζ K ( s ) , we have, for s ∈ C s \in\mathbb{C} s ∈ C ,
∑ ρ 1 s − ρ = 1 s − 1 + 1 2 log ∣ Δ K ∣ + 1 s − log π + Γ ′ ( s ) Γ ( s ) − log 2 + ζ K ′ ( s ) ζ K ( s ) (42) \sum_{\rho} \frac{1}{s-\rho} = \frac{1}{s-1} + \frac{1}{2}\log\lvert\Delta_K\rvert+ \frac{1}{s} - \log\pi+ \frac{\Gamma'(s)}{\Gamma(s)} - \log2 + \frac{\zeta_K'(s)}{\zeta_K(s)} \tag*{(42)} ρ ∑ s − ρ 1 = s − 1 1 + 2 1 log ∣ Δ K ∣ + s 1 − log π + Γ ( s ) Γ ′ ( s ) − log 2 + ζ K ( s ) ζ K ′ ( s ) (42)
where ρ \rho ρ runs through the zeros of ζ K ( s ) \zeta_K(s) ζ K ( s ) in the strip 0 < Re ( s ) < 1 0 < \operatorname{Re}(s) < 1 0 < Re ( s ) < 1 and the ρ \rho ρ and ρ ˉ \bar{\rho} ρ ˉ terms must be grouped together. For s = σ > 1 s = \sigma> 1 s = σ > 1 , we have
1 σ − ρ + 1 σ − ρ ˉ = 2 σ − Re ( ρ ) ∣ σ − ρ ∣ 2 > 0. \frac{1}{\sigma-\rho} + \frac{1}{\sigma-\bar{\rho}} = 2\frac{\sigma-\operatorname{Re}(\rho)}{\lvert\sigma-\rho\rvert^2} > 0. σ − ρ 1 + σ − ρ ˉ 1 = 2 ∣ σ − ρ ∣ 2 σ − Re ( ρ ) > 0.
Remind that ζ K ( s ) = ζ ( s ) L ( s , χ ) \zeta_K(s) = \zeta(s)L(s,\chi) ζ K ( s ) = ζ ( s ) L ( s , χ ) for s ∈ C s \in\mathbb{C} s ∈ C and hence ζ K ( β ) = 0 \zeta_K(\beta) = 0 ζ K ( β ) = 0 . We have, for σ > 1 \sigma> 1 σ > 1 ,
− ζ K ′ ( σ ) ζ K ( σ ) < − 1 σ − β + 1 σ − 1 + 1 2 log ∣ Δ K ∣ + 1 σ − log π + Γ ′ ( σ ) Γ ( σ ) − log 2. (43) -\frac{\zeta_K'(\sigma)}{\zeta_K(\sigma)} < -\frac{1}{\sigma-\beta} + \frac{1}{\sigma-1} + \frac{1}{2}\log\lvert\Delta_K\rvert+ \frac{1}{\sigma} - \log\pi+ \frac{\Gamma'(\sigma)}{\Gamma(\sigma)} - \log2. \tag*{(43)} − ζ K ( σ ) ζ K ′ ( σ ) < − σ − β 1 + σ − 1 1 + 2 1 log ∣ Δ K ∣ + σ 1 − log π + Γ ( σ ) Γ ′ ( σ ) − log 2. (43)
Note that Γ ′ ( s ) Γ ( s ) < Γ ′ ( 2 ) Γ ( 2 ) = 1 − γ \frac{\Gamma'(s)}{\Gamma(s)} < \frac{\Gamma'(2)}{\Gamma(2)} = 1-\gamma Γ ( s ) Γ ′ ( s ) < Γ ( 2 ) Γ ′ ( 2 ) = 1 − γ for 0 < s < 2 0<s<2 0 < s < 2 , where γ \gamma γ is Euler’s constant. Thus, we have, for 1 < σ < 2 1<\sigma<2 1 < σ < 2 ,
− ζ K ′ ( σ ) ζ K ( σ ) < − 1 σ − β + 1 σ − 1 + 1 2 log ∣ Δ K ∣ + 1 − log π + ( 1 − γ ) − log 2 -\frac{\zeta'_K(\sigma)}{\zeta_K(\sigma)} < -\frac{1}{\sigma-\beta}+\frac{1}{\sigma-1}+\frac{1}{2}\log|\Delta_K|+1-\log\pi+(1-\gamma)-\log2 − ζ K ( σ ) ζ K ′ ( σ ) < − σ − β 1 + σ − 1 1 + 2 1 log ∣ Δ K ∣ + 1 − log π + ( 1 − γ ) − log 2
and, consequently,
− ζ K ′ ( σ ) ζ K ( σ ) < − 1 σ − β + 1 σ − 1 + 1 2 log ∣ Δ K ∣ < 1 2 log ∣ Δ K ∣ + 1 − β ( σ − 1 ) 2 (44) -\frac{\zeta'_K(\sigma)}{\zeta_K(\sigma)} < -\frac{1}{\sigma-\beta}+\frac{1}{\sigma-1}+\frac{1}{2}\log|\Delta_K| < \frac{1}{2}\log|\Delta_K|+\frac{1-\beta}{(\sigma-1)^2} \tag*{(44)} − ζ K ( σ ) ζ K ′ ( σ ) < − σ − β 1 + σ − 1 1 + 2 1 log ∣ Δ K ∣ < 2 1 log ∣ Δ K ∣ + ( σ − 1 ) 2 1 − β (44)
It follows from (21) that, for σ > 1 \sigma>1 σ > 1 ,
− ζ K ′ ( σ ) ζ K ( σ ) = ∑ p log N ( p ) N ( p ) σ − 1 (45) -\frac{\zeta'_K(\sigma)}{\zeta_K(\sigma)}=\sum_{\mathfrak{p}}\frac{\log\mathfrak{N}(\mathfrak{p})}{\mathfrak{N}(\mathfrak{p})^\sigma-1} \tag*{(45)} − ζ K ( σ ) ζ K ′ ( σ ) = p ∑ N ( p ) σ − 1 log N ( p ) (45)
where p \mathfrak{p} p runs through the nonzero prime ideals of K K K . By using ζ K ( σ ) = ζ ( σ ) L ( σ , χ ) \zeta_K(\sigma)=\zeta(\sigma)L(\sigma,\chi) ζ K ( σ ) = ζ ( σ ) L ( σ , χ ) , we also have
− ζ K ′ ( σ ) ζ K ( σ ) = − ζ ′ ( σ ) ζ ( σ ) − L ′ ( σ , χ ) L ( σ , χ ) = ∑ n = 1 ∞ ( 1 + χ ( n ) ) Λ ( n ) n σ (46) -\frac{\zeta'_K(\sigma)}{\zeta_K(\sigma)}=-\frac{\zeta'(\sigma)}{\zeta(\sigma)}-\frac{L'(\sigma,\chi)}{L(\sigma,\chi)}=\sum_{n=1}^{\infty}(1+\chi(n))\frac{\Lambda(n)}{n^\sigma} \tag*{(46)} − ζ K ( σ ) ζ K ′ ( σ ) = − ζ ( σ ) ζ ′ ( σ ) − L ( σ , χ ) L ′ ( σ , χ ) = n = 1 ∑ ∞ ( 1 + χ ( n )) n σ Λ ( n ) (46)
for σ > 1 \sigma>1 σ > 1 , where Λ ( n ) \Lambda(n) Λ ( n ) is the von Mangoldt function defined as log p \log p log p if n = p k n=p^k n = p k for some prime p p p and integer k ≥ 1 k\geq1 k ≥ 1 and 0 0 0 otherwise.
We know that Galois extension K / Q K/\mathbb{Q} K / Q is uniquely determined by the set of prime numbers which split completely in K K K , as a consequence of Chebotarev’s density theorem (see Neukirch [31 Chapter VII, Section §13] ). For instance, the dominating terms in the summation (45) are in the prime ideals p ∣ p \mathfrak{p}\mid p p ∣ p where the prime p p p splits completely in K K K and hence N ( p ) = p \mathfrak{N}(\mathfrak{p})=p N ( p ) = p . If the prime p p p is inert in the imaginary quadratic field K K K and p ∣ p \mathfrak{p}\mid p p ∣ p , then χ ( p ) = ( Δ K p ) = − 1 \chi(p)=\left(\frac{\Delta_K}{p}\right)=-1 χ ( p ) = ( p Δ K ) = − 1 , p = p O K \mathfrak{p}=p\mathcal{O}_K p = p O K and N ( p ) = p 2 \mathfrak{N}(\mathfrak{p})=p^2 N ( p ) = p 2 . Furthermore, it is seen from (46) that the dominating terms in the summation are in the primes p p p with χ ( p ) = ( Δ K p ) = 1 \chi(p)=\left(\frac{\Delta_K}{p}\right)=1 χ ( p ) = ( p Δ K ) = 1 , which splits completely in K K K .
If we take σ = 1 + 2 / log X < 2 \sigma=1+2/\log X<2 σ = 1 + 2/ log X < 2 , where X ≥ q X\geq q X ≥ q is a real variable, then, from (44) and (46) , we get
1 2 log ∣ Δ K ∣ + 1 4 ( 1 − β ) ( log X ) 2 > ∑ p ( 1 + χ ( p ) ) log p p 1 p 2 / log X > 1 e 2 ∑ p ≤ X ( 1 + χ ( p ) ) log p p \frac{1}{2}\log|\Delta_K|+\frac{1}{4}(1-\beta)(\log X)^2>\sum_p(1+\chi(p))\frac{\log p}{p}\frac{1}{p^{2/\log X}}>\frac{1}{e^2}\sum_{p\leq X}(1+\chi(p))\frac{\log p}{p} 2 1 log ∣ Δ K ∣ + 4 1 ( 1 − β ) ( log X ) 2 > p ∑ ( 1 + χ ( p )) p log p p 2/ l o g X 1 > e 2 1 p ≤ X ∑ ( 1 + χ ( p )) p log p
where the infinite series ∑ p \sum_p ∑ p runs over the primes p ∈ Z p\in\mathbb{Z} p ∈ Z . Therefore, noticing that χ ( p ) ∈ { − 1 , 0 , 1 } \chi(p)\in\{-1,0,1\} χ ( p ) ∈ { − 1 , 0 , 1 } and that χ ( p ) = 0 \chi(p)=0 χ ( p ) = 0 if and only if p ∣ q p\mid q p ∣ q , we obtain
∑ p ≤ X χ ( p ) = 1 log p p + 1 2 ∑ p ≤ X p ∣ q log p p < e 2 4 log ∣ Δ K ∣ + e 2 8 ( 1 − β ) ( log X ) 2 (47) \sum_{\substack{p\leq X\\ \chi(p)=1}}\frac{\log p}{p}+\frac{1}{2}\sum_{\substack{p\leq X\\ p\mid q}}\frac{\log p}{p}<\frac{e^2}{4}\log|\Delta_K|+\frac{e^2}{8}(1-\beta)(\log X)^2 \tag*{(47)} p ≤ X χ ( p ) = 1 ∑ p log p + 2 1 p ≤ X p ∣ q ∑ p log p < 4 e 2 log ∣ Δ K ∣ + 8 e 2 ( 1 − β ) ( log X ) 2 (47)
for all X ≥ q X \ge q X ≥ q . Moreover, by noticing that q = ∣ Δ K ∣ = 4 N q = \lvert\Delta_K\rvert= 4N q = ∣ Δ K ∣ = 4 N and N N N is a squarefree positive integer, we have
∑ p ≤ X p ∣ q log p p = ∑ p ∣ q log p p = log 2 2 + ∑ p ∣ q p > 2 log p p ≤ log 2 2 + 1 3 log N < 1 3 log q . \sum_{\substack{p \le X\\p\mid q}} \frac{\log p}{p}
= \sum_{p\mid q} \frac{\log p}{p}
= \frac{\log2}{2} + \sum_{\substack{p\mid q\\p>2}} \frac{\log p}{p}
\le\frac{\log2}{2} + \frac{1}{3}\log N < \frac{1}{3}\log q. p ≤ X p ∣ q ∑ p log p = p ∣ q ∑ p log p = 2 log 2 + p ∣ q p > 2 ∑ p log p ≤ 2 log 2 + 3 1 log N < 3 1 log q .
It follows from (47) that, for all X ≥ q X \ge q X ≥ q ,
∑ p ≤ X χ ( p ) ≠ − 1 log p p = ∑ p ≤ X χ ( p ) = 1 log p p + ∑ p ≤ X p ∣ q log p p < ( 1 6 + e 2 4 ) log q + e 2 8 ( 1 − β ) ( log X ) 2 . (48) \sum_{\substack{p\le X\\\chi(p)\ne-1}}\frac{\log p}{p}
= \sum_{\substack{p\le X\\\chi(p)=1}}\frac{\log p}{p}
+ \sum_{\substack{p\le X\\p\mid q}}\frac{\log p}{p}
< \left(\frac{1}{6}+\frac{e^2}{4}\right)\log q+\frac{e^2}{8}(1-\beta)(\log X)^2. \tag*{(48)} p ≤ X χ ( p ) = − 1 ∑ p log p = p ≤ X χ ( p ) = 1 ∑ p log p + p ≤ X p ∣ q ∑ p log p < ( 6 1 + 4 e 2 ) log q + 8 e 2 ( 1 − β ) ( log X ) 2 . (48)
It is well known that ∑ p ≤ X log p p ∼ log X \sum_{p\le X}\frac{\log p}{p}\sim\log X ∑ p ≤ X p l o g p ∼ log X as X → ∞ X\to\infty X → ∞ . By (48) in the case that q q q is fixed or that log q \log q log q is relatively small to log X \log X log X for large X X X , if the real zero β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) were so close to 1 1 1 that ( 1 − β ) log X (1-\beta)\log X ( 1 − β ) log X is extremely small , then the primes p ≤ X p\le X p ≤ X , which are not inert in K K K , would be very rare . The nature of Landau–Siegel zeros extremely close to 1 1 1 in this respect had also been discussed by Friedlander and Iwaniec [13 ] .
The effect of Landau–Siegel zeros β \beta β extremely close to 1 1 1 causing the primes not inert in K K K to be very rare has been employed significantly in [32 ] to achieve a discovery that the real zero β \beta β has to be at least c / log q c/\log q c / log q away from 1 1 1 for an absolute constant c > 0 c>0 c > 0 , which is much greater than Siegel’s distance C ( ε ) / q ε C(\varepsilon)/q^\varepsilon C ( ε ) / q ε for an arbitrary positive number ε > 0 \varepsilon>0 ε > 0 with some constant C ( ε ) > 0 C(\varepsilon)>0 C ( ε ) > 0 depending on ε \varepsilon ε .
A basic strategy in [32 ] is to construct a nonzero algebraic integer , denoted by Δ \Delta Δ , in a biquadratic field L / K L/K L / K , which has all small enough primes inert in K K K as its prime factors. Then, one can obtain a larger lower bound for the absolute value of the field norm ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ when the primes not inert in K K K becomes more rare due to the real zero β \beta β being closer to 1 1 1 . On the other hand, the quantity ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ must have an upper bound, depending only on q q q , which arises from its definition in terms of real and complex embeddings, since Δ \Delta Δ is actually defined as the determinant of a full-rank square matrix with entries in O L O_L O L , the ring of integers of L L L , with predefined bounded Z \mathbb{Z} Z -coordinates in terms of q q q . Using the lower and upper bounds of ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ , we see that the rarity (or abundance) of primes p ≤ X p\le X p ≤ X which are not inert in K K K must have a bound from above (or below). By the nature of the Landau–Siegel zero β \beta β extremely close to 1 1 1 as shown in (48) , combing the available lower and upper bounds of ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ will yield a lower bound for ( 1 − β ) log X (1-\beta)\log X ( 1 − β ) log X . Hence, the distance 1 − β 1-\beta 1 − β between the real zero β \beta β and 1 1 1 will have a nontrivial lower bound, implying a remarkably enlarged zero-free interval ( β , 1 ) (\beta,1) ( β , 1 ) for L ( s , χ ) L(s,\chi) L ( s , χ ) .
The method of the field norm using the lower and upper bounds of its absolute value has been previously employed in 1986 by Lenstra [23 ] in theory of codes from geometry of numbers. Lenstra used this simple method to construct the code from algebraic number fields and provided some heuristic to show that the code would be of high quality in the sense that the minimum distance of the code can potentially be relatively large for a fixed rate of the code. The author of this paper has used the same method of the field norm in [24 ] to simplify the construction of codes and further analyzed the Lenstra codes, which have been demonstrated to be of high quality as expected.
However, it is crucial for the achievement in [32 ] to construct a specific nonzero algebraic integer Δ ∈ O L \Delta\in O_L Δ ∈ O L with a biquadratic field L / K L/K L / K , which has the property that all small enough primes inert in K K K are its prime factors.
In the following, we are going to present the construction of Δ \Delta Δ given in [32 ] and verify the properties of Δ \Delta Δ by providing accurate and accessible arguments in details for the purpose of rigor and completeness.
We introduce a biquadratic field as follows. For the given two quadratic fields
K = Q ( − N ) and F = Q ( 2 ) K = \mathbb{Q}(\sqrt{-N}) \text{ and } F = \mathbb{Q}(\sqrt{2}) K = Q ( − N ) and F = Q ( 2 )
we define a number field L L L as the composite field
L = K F = Q ( − N , 2 ) . L = KF = \mathbb{Q}(\sqrt{-N}, \sqrt{2}). L = K F = Q ( − N , 2 ) .
We need to find the Galois group of L L L for the calculation of the field norm N L / Q ( ⋅ ) N_{L/\mathbb{Q}}(\cdot) N L / Q ( ⋅ ) . The Galois group of K K K is Gal ( K / Q ) = { e 1 , σ 1 } \operatorname{Gal}(K/\mathbb{Q}) = \{e_1, \sigma_1\} Gal ( K / Q ) = { e 1 , σ 1 } , where e 1 e_1 e 1 is the identity element and σ 1 \sigma_1 σ 1 is defined by σ 1 ( − N ) = − − N \sigma_1(\sqrt{-N}) = -\sqrt{-N} σ 1 ( − N ) = − − N . The Galois group of F F F is Gal ( F / Q ) = { e 2 , τ 2 } \operatorname{Gal}(F/\mathbb{Q}) = \{e_2, \tau_2\} Gal ( F / Q ) = { e 2 , τ 2 } , where e 2 e_2 e 2 is the identity element and τ 2 \tau_2 τ 2 is defined by τ 2 ( 2 ) = − 2 \tau_2(\sqrt{2}) = -\sqrt{2} τ 2 ( 2 ) = − 2 . Since K ∩ F = Q K \cap F = \mathbb{Q} K ∩ F = Q , the Galois group of L L L is isomorphic to the direct product
Gal ( L / Q ) = { e , σ , τ , σ τ } ≅ Gal ( K / Q ) × Gal ( F / Q ) \operatorname{Gal}(L/\mathbb{Q}) = \{e, \sigma, \tau, \sigma\tau\} \cong\operatorname{Gal}(K/\mathbb{Q}) \times\operatorname{Gal}(F/\mathbb{Q}) Gal ( L / Q ) = { e , σ , τ , σ τ } ≅ Gal ( K / Q ) × Gal ( F / Q )
where e = ( e 1 , e 2 ) e = (e_1, e_2) e = ( e 1 , e 2 ) , σ = ( σ 1 , e 2 ) \sigma= (\sigma_1, e_2) σ = ( σ 1 , e 2 ) and τ = ( e 1 , τ 2 ) \tau= (e_1, \tau_2) τ = ( e 1 , τ 2 ) . Hence, σ τ = τ σ = ( σ 1 , τ 2 ) \sigma\tau= \tau\sigma= (\sigma_1, \tau_2) σ τ = τ σ = ( σ 1 , τ 2 ) with
σ ( − N ) = ( σ 1 , e 2 ) ∣ K ( − N ) = σ 1 ( − N ) = − − N \sigma(\sqrt{-N}) = (\sigma_1, e_2)|_K(\sqrt{-N}) = \sigma_1(\sqrt{-N}) = -\sqrt{-N} σ ( − N ) = ( σ 1 , e 2 ) ∣ K ( − N ) = σ 1 ( − N ) = − − N
and
τ ( 2 ) = ( e 1 , τ 2 ) ∣ F ( 2 ) = τ 2 ( 2 ) = − 2 . \tau(\sqrt{2}) = (e_1, \tau_2)|_F(\sqrt{2}) = \tau_2(\sqrt{2}) = -\sqrt{2}. τ ( 2 ) = ( e 1 , τ 2 ) ∣ F ( 2 ) = τ 2 ( 2 ) = − 2 .
It is clear that
σ ( 2 ) = ( σ 1 , e 2 ) ∣ F ( 2 ) = e 2 ( 2 ) = 2 \sigma(\sqrt{2}) = (\sigma_1, e_2)|_F(\sqrt{2}) = e_2(\sqrt{2}) = \sqrt{2} σ ( 2 ) = ( σ 1 , e 2 ) ∣ F ( 2 ) = e 2 ( 2 ) = 2
and
τ ( − N ) = ( e 1 , τ 2 ) ∣ K ( − N ) = e 1 ( − N ) = − N . \tau(\sqrt{-N}) = (e_1, \tau_2)|_K(\sqrt{-N}) = e_1(\sqrt{-N}) = \sqrt{-N}. τ ( − N ) = ( e 1 , τ 2 ) ∣ K ( − N ) = e 1 ( − N ) = − N .
An integral basis of K K K is { 1 , − N } \{1, \sqrt{-N}\} { 1 , − N } and hence O K = Z [ − N ] O_K = \mathbb{Z}[\sqrt{-N}] O K = Z [ − N ] . The field F F F has an integral basis of { 1 , 2 } \{1, \sqrt{2}\} { 1 , 2 } and therefore the ring of algebraic integers O F = Z [ 2 ] O_F = \mathbb{Z}[\sqrt{2}] O F = Z [ 2 ] . The determination of the ring of algebraic integers O L \mathcal{O}_L O L for the biquadratic field L L L had been explicitly provided by Williams [49 ] . It is seen that
R = Z [ − N , 2 ] ⊂ O L \mathcal{R} = \mathbb{Z}[\sqrt{-N},\sqrt{2}] \subset\mathcal{O}_L R = Z [ − N , 2 ] ⊂ O L
which is a proper subring of O L \mathcal{O}_L O L . Let W ≥ q W \ge q W ≥ q be a positive integer. We consider the following subset
B W = { n 1 + n 2 a + n 3 b + n 4 a b ∣ 0 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ R B_W = \{n_1+n_2a+n_3b+n_4ab \mid0 \le n_i \le W-1\text{ for }i=1,2,3,4\} \subset\mathcal{R} B W = { n 1 + n 2 a + n 3 b + n 4 ab ∣ 0 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ R
where n i ∈ Z n_i \in\mathbb{Z} n i ∈ Z for 1 ≤ i ≤ 4 1 \le i \le4 1 ≤ i ≤ 4 with
a = − N and b = 2 . a=\sqrt{-N}\text{ and }b=\sqrt{2}. a = − N and b = 2 .
It is noted that { 1 , a , b , a b } \{1,a,b,ab\} { 1 , a , b , ab } is the Q \mathbb{Q} Q -basis of the extension L / Q L/\mathbb{Q} L / Q of degree [ L : Q ] = 4 [L:\mathbb{Q}]=4 [ L : Q ] = 4 . Thus, the four numbers 1 1 1 , a a a , b b b and a b = − 2 N ab=\sqrt{-2N} ab = − 2 N are linearly independent over the rational field Q \mathbb{Q} Q .
Let n ⃗ = ( n 1 , n 2 , n 3 , n 4 ) T \vec{n}=(n_1,n_2,n_3,n_4)^{\mathsf{T}} n = ( n 1 , n 2 , n 3 , n 4 ) T be a column vector, where n i ∈ Z n_i\in\mathbb{Z} n i ∈ Z for 1 ≤ i ≤ 4 1\le i\le4 1 ≤ i ≤ 4 and T \mathsf{T} T denotes the transpose. We set
θ n ⃗ = n 1 + n 2 a + n 3 b + n 4 a b ∈ R . \theta_{\vec{n}}=n_1+n_2a+n_3b+n_4ab\in\mathcal{R}. θ n = n 1 + n 2 a + n 3 b + n 4 ab ∈ R .
It is clear that e ( θ n ⃗ ) = θ n ⃗ e(\theta_{\vec{n}})=\theta_{\vec{n}} e ( θ n ) = θ n ,
σ ( θ n ⃗ ) = n 1 + n 2 σ ( a ) + n 3 σ ( b ) + n 4 σ ( a ) σ ( b ) = n 1 − n 2 a + n 3 b − n 4 a b ∈ R \sigma(\theta_{\vec{n}})=n_1+n_2\sigma(a)+n_3\sigma(b)+n_4\sigma(a)\sigma(b)=n_1-n_2a+n_3b-n_4ab\in\mathcal{R} σ ( θ n ) = n 1 + n 2 σ ( a ) + n 3 σ ( b ) + n 4 σ ( a ) σ ( b ) = n 1 − n 2 a + n 3 b − n 4 ab ∈ R
and
τ ( θ n ⃗ ) = n 1 + n 2 τ ( a ) + n 3 τ ( b ) + n 4 τ ( a ) τ ( b ) = n 1 + n 2 a − n 3 b − n 4 a b ∈ R . \tau(\theta_{\vec{n}})=n_1+n_2\tau(a)+n_3\tau(b)+n_4\tau(a)\tau(b)=n_1+n_2a-n_3b-n_4ab\in\mathcal{R}. τ ( θ n ) = n 1 + n 2 τ ( a ) + n 3 τ ( b ) + n 4 τ ( a ) τ ( b ) = n 1 + n 2 a − n 3 b − n 4 ab ∈ R .
Moreover, we have
σ τ ( θ n ⃗ ) = n 1 + n 2 σ τ ( a ) + n 3 σ τ ( b ) + n 4 σ τ ( a ) σ τ ( b ) = n 1 − n 2 a − n 3 b + n 4 a b ∈ R . \sigma\tau(\theta_{\vec{n}})=n_1+n_2\sigma\tau(a)+n_3\sigma\tau(b)+n_4\sigma\tau(a)\sigma\tau(b)=n_1-n_2a-n_3b+n_4ab\in\mathcal{R}. σ τ ( θ n ) = n 1 + n 2 σ τ ( a ) + n 3 σ τ ( b ) + n 4 σ τ ( a ) σ τ ( b ) = n 1 − n 2 a − n 3 b + n 4 ab ∈ R .
Let Z ≥ 0 \mathbb{Z}_{\ge0} Z ≥ 0 denote the set of all nonnegative integers. Note that the subset B W B_W B W of R \mathcal{R} R has the cardinality of # B W = W 4 \#B_W=W^4 # B W = W 4 . For each α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathbb{Z}_{\ge0}^3 α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 , we define a row vector of dimension W 4 W^4 W 4 as follows
R α = ( θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 ) R_\alpha=\left(\theta_{\vec{n}}^{\alpha_1}\sigma(\theta_{\vec{n}})^{\alpha_2}\sigma\tau(\theta_{\vec{n}})^{\alpha_3}\right) R α = ( θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 )
where θ n ⃗ ∈ B W ⊂ R \theta_{\vec{n}}\in B_W\subset\mathcal{R} θ n ∈ B W ⊂ R . The W 4 W^4 W 4 algebraic integers θ n ⃗ \theta_{\vec{n}} θ n in B W B_W B W and correspondingly the W 4 W^4 W 4 entries in R α R_\alpha R α can be placed in a lexicographical order, namely, the alphabetical order. When n ⃗ = ( 0 , 0 , 0 , 0 ) T \vec{n}=(0,0,0,0)^{\mathsf{T}} n = ( 0 , 0 , 0 , 0 ) T , it is agreed that the first entry in the row vector R α R_\alpha R α is defined as 1 for α = ( 0 , 0 , 0 ) \alpha=(0,0,0) α = ( 0 , 0 , 0 ) and 0 for α ≠ ( 0 , 0 , 0 ) \alpha\ne(0,0,0) α = ( 0 , 0 , 0 ) . We can see that every entry of the vector R α R_\alpha R α is an algebraic integer in R ⊂ O L \mathcal{R}\subset O_L R ⊂ O L .
We choose the parameter H H H with the value
and define the following four numbers in terms of W W W
U = W 4 / 3 , T 1 = ⌈ 8 H 2 / 3 U ⌉ , T 2 = T 2 = ⌈ 8 H − 1 / 3 U ⌉ . (49) U=W^{4/3},\quad T_1=\lceil8H^{2/3}U\rceil,\quad T_2=T_2=\lceil8H^{-1/3}U\rceil. \tag*{(49)} U = W 4/3 , T 1 = ⌈ 8 H 2/3 U ⌉ , T 2 = T 2 = ⌈ 8 H − 1/3 U ⌉ . (49)
Let α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathbb{Z}_{\ge0}^3 α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 with α j ≤ T j \alpha_j\le T_j α j ≤ T j for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 . We define the matrix M \mathcal{M} M with
T = ( T 1 + 1 ) ( T 2 + 1 ) ( T 3 + 1 ) T=(T_1+1)(T_2+1)(T_3+1) T = ( T 1 + 1 ) ( T 2 + 1 ) ( T 3 + 1 )
row vectors of dimension W 4 W^4 W 4 as follows. The row vectors of M \mathcal{M} M are indexed by α = ( α 1 , α 2 , α 3 ) \alpha=(\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) with the alphabetical order and the row vector with index α \alpha α of M \mathcal{M} M equals R α R_\alpha R α . It is clear that T > 512 U 3 = 512 W 4 > W 4 T>512U^3=512W^4>W^4 T > 512 U 3 = 512 W 4 > W 4 , i.e., the number of row vectors of the T × W 4 T\times W^4 T × W 4 matrix M \mathcal{M} M is greater than the number of its column vectors.
The following lemma is a fundamental result which has essentially been given in [32 ] . Based on the arguments provided by [32 ] , we will present a detailed proof with extra arguments and derivations which are entailed but do not appear in [32 ] .
Lemma 7. The matrix M \mathcal{M} M has full column rank, i.e., the rank of M \mathcal{M} M equals W 4 W^4 W 4 , the number of its columns.
Proof: See Appendix B.
Note that every entry of the matrix M \mathcal{M} M is an algebraic integer in R ⊂ O L \mathcal{R}\subset O_L R ⊂ O L . The row rank and the column rank of any matrix are always equal. Lemma 7 guarantees that there exist W 4 W^4 W 4 row vectors of M \mathcal{M} M which are linearly independent in the number field L L L .
A specific nonzero algebraic integer Δ \Delta Δ in O L O_L O L
The row vectors R α R_\alpha R α of the matrix M \mathcal{M} M are indexed by α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathbb{Z}_{\ge0}^3 α = ( α 1 , α 2 , α 3 ) ∈ Z ≥ 0 3 with α j ≤ T j \alpha_j\le T_j α j ≤ T j for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 . We define the H H H -weight of the index α \alpha α as
wt ( α ) = α 1 + H α 2 + H α 3 \operatorname{wt}(\alpha)=\alpha_1+H\alpha_2+H\alpha_3 wt ( α ) = α 1 + H α 2 + H α 3
where H = 10 6 H=10^6 H = 1 0 6 is the chosen parameter.
We arrange the order of indices α \alpha α of the row vectors R α R_\alpha R α in the matrix M \mathcal{M} M by increasing wt ( α ) \operatorname{wt}(\alpha) wt ( α ) . In the case that wt ( α ) = wt ( α ^ ) \operatorname{wt}(\alpha)=\operatorname{wt}(\hat{\alpha}) wt ( α ) = wt ( α ^ ) for two different indices α ≠ α ^ \alpha\ne\hat{\alpha} α = α ^ , the arrangement to break tie resorts to using their alphabetical order. It is clear that the first index is α = ( 0 , 0 , 0 ) \alpha=(0,0,0) α = ( 0 , 0 , 0 ) and the corresponding row vector R α R_\alpha R α is an all-one vector. Starting from the first retained index α = ( 0 , 0 , 0 ) \alpha=(0,0,0) α = ( 0 , 0 , 0 ) , we consider the next index α \alpha α in the order of the H H H -weight. It will be retained if the row vector R α R_\alpha R α can increase the span of the earlier retained rows, namely, if the augmented submatrix consisting of the earlier retained rows appended with the row vector R α R_\alpha R α continues to have full row rank. According to Lemma 7, this procedure of selection will proceed in a unique way and end with W 4 W^4 W 4 retained rows of the matrix M \mathcal{M} M . Let P r i \mathcal{P}_{ri} P r i be the set of the retained indices α \alpha α , which is uniquely determined by the procedure with the H H H -weight.
It is clear that the ordered set P r i \mathcal{P}_{ri} P r i of retained indices has the cardinality of
# P r i = W 4 . \#\mathcal{P}_{ri}=W^4. # P r i = W 4 .
We denote by M r i \mathcal{M}_{ri} M r i the submatrix of the matrix M \mathcal{M} M consisting of the row vectors R α R_\alpha R α with α ∈ P r i \alpha\in\mathcal{P}_{ri} α ∈ P r i in the order of the H H H -weight. Remind that R α R_\alpha R α is of dimension W 4 W^4 W 4 . We see that M r i \mathcal{M}_{ri} M r i is a full-rank square matrix of size W 4 × W 4 W^4\times W^4 W 4 × W 4 , every entry of which is an algebraic integer in R ⊂ O L \mathcal{R}\subset O_L R ⊂ O L . The determinant of M r i \mathcal{M}_{ri} M r i is thus nonzero .
We define the nonzero algebraic integer
Δ = det ( M r i ) ∈ O L ∖ { 0 } (50) \Delta=\det(\mathcal{M}_{ri})\in O_L\setminus\{0\} \tag*{(50)} Δ = det ( M r i ) ∈ O L ∖ { 0 } (50)
which will play a vital role in the following process. The algebraic integer Δ \Delta Δ induces, via the field norm, a nonzero rational integer given by
N L / Q ( Δ ) ∈ Z ∖ { 0 } . (51) \mathrm{N}_{L/\mathbb{Q}}(\Delta)\in\mathbb{Z}\setminus\{0\}. \tag*{(51)} N L / Q ( Δ ) ∈ Z ∖ { 0 } . (51)
The following result on the set P r i \mathcal{P}_{ri} P r i of retained indices will be needed.
Lemma 8 We put
S 1 = ∑ α ∈ P r i α 1 and S 2 = ∑ α ∈ P r i ( α 2 + α 3 ) . S_1=\sum_{\alpha\in\mathcal{P}_{ri}}\alpha_1\quad\text{and}\quad S_2=\sum_{\alpha\in\mathcal{P}_{ri}}(\alpha_2+\alpha_3). S 1 = α ∈ P r i ∑ α 1 and S 2 = α ∈ P r i ∑ ( α 2 + α 3 ) . Then, we have
S 1 ≥ 1 135 W 4 H 2 / 3 U and S 2 ≤ 18 W 4 H − 1 / 3 U . S_1\ge\frac{1}{135}W^4H^{2/3}U\quad\text{and}\quad S_2\le18W^4H^{-1/3}U. S 1 ≥ 135 1 W 4 H 2/3 U and S 2 ≤ 18 W 4 H − 1/3 U .
Proof: Since U = W 4 / 3 U=W^{4/3} U = W 4/3 with W ≥ q > 3 ⋅ 10 20 W\ge q>3\cdot10^{20} W ≥ q > 3 ⋅ 1 0 20 , 0 ≤ α 2 ≤ T 2 = ⌈ 8 H − 1 / 3 U ⌉ 0\le\alpha_2\le T_2=\lceil8H^{-1/3}U\rceil 0 ≤ α 2 ≤ T 2 = ⌈ 8 H − 1/3 U ⌉ and 0 ≤ α 3 ≤ T 3 = ⌈ 8 H − 1 / 3 U ⌉ 0\le\alpha_3\le T_3=\lceil8H^{-1/3}U\rceil 0 ≤ α 3 ≤ T 3 = ⌈ 8 H − 1/3 U ⌉ , we have
S 2 ≤ 2 ⌈ 8 H − 1 / 3 U ⌉ ∑ α ∈ P r i 1 = 2 W 4 ⌈ 8 H − 1 / 3 U ⌉ ≤ 18 W 4 H − 1 / 3 U . S_2\le2\lceil8H^{-1/3}U\rceil\sum_{\alpha\in\mathcal{P}_{ri}}1=2W^4\lceil8H^{-1/3}U\rceil\le18W^4H^{-1/3}U. S 2 ≤ 2 ⌈ 8 H − 1/3 U ⌉ α ∈ P r i ∑ 1 = 2 W 4 ⌈ 8 H − 1/3 U ⌉ ≤ 18 W 4 H − 1/3 U .
For any fixed α 1 \alpha_1 α 1 with 0 ≤ α 1 ≤ T 1 = ⌈ 8 H 2 / 3 U ⌉ 0 \le\alpha_1 \le T_1 = \lceil8H^{2/3}U\rceil 0 ≤ α 1 ≤ T 1 = ⌈ 8 H 2/3 U ⌉ , there are at most
Q 0 = ( T 2 + 1 ) ( T 3 + 1 ) ≤ ( 8 H − 1 / 3 U + 2 ) 2 ≤ 66 H − 2 / 3 U 2 = 66 H − 2 / 3 W 8 / 3 Q_0 = (T_2 + 1)(T_3 + 1) \le(8H^{-1/3}U + 2)^2 \le66H^{-2/3}U^2 = 66H^{-2/3}W^{8/3} Q 0 = ( T 2 + 1 ) ( T 3 + 1 ) ≤ ( 8 H − 1/3 U + 2 ) 2 ≤ 66 H − 2/3 U 2 = 66 H − 2/3 W 8/3
retained indices α = ( α 1 , α 2 , α 3 ) \alpha= (\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) .
We rearrange elements in the set
{ α 1 ∣ α = ( α 1 , α 2 , α 3 ) ∈ P r i } \{\alpha_1 \mid\alpha= (\alpha_1,\alpha_2,\alpha_3) \in\mathcal{P}_{ri}\} { α 1 ∣ α = ( α 1 , α 2 , α 3 ) ∈ P r i }
by sorting in a nondecreasing order and get the following set
{ b 1 , b 2 , ⋯ , b W 4 } \{b_1,b_2,\cdots,b_{W^4}\} { b 1 , b 2 , ⋯ , b W 4 }
where W 4 = # P r i W^4 = \#\mathcal{P}_{ri} W 4 = # P r i . Therefore,
S 1 = ∑ α ∈ P r i α 1 = ∑ ℓ = 1 W 4 b ℓ . S_1 = \sum_{\alpha\in\mathcal{P}_{ri}} \alpha_1 = \sum_{\ell=1}^{W^4} b_\ell. S 1 = α ∈ P r i ∑ α 1 = ℓ = 1 ∑ W 4 b ℓ .
It is seen that 0 = b 1 ≤ b 2 ≤ ⋯ ≤ b W 4 ≤ T 1 0 = b_1 \le b_2 \le\cdots\le b_{W^4} \le T_1 0 = b 1 ≤ b 2 ≤ ⋯ ≤ b W 4 ≤ T 1 and each value of b ℓ b_\ell b ℓ can repeat itself at most Q 0 Q_0 Q 0 times. There exists a positive integer k k k such that
( k + 1 ) Q 0 ≤ W 4 < ( k + 2 ) Q 0 . (k+1)Q_0 \le W^4 < (k+2)Q_0. ( k + 1 ) Q 0 ≤ W 4 < ( k + 2 ) Q 0 .
Then, we have
S 1 ≥ ∑ m = 0 k m Q 0 = k ( k + 1 ) 2 Q 0 ≥ Q 0 2 ( W 4 Q 0 − 2 ) ( W 4 Q 0 − 1 ) ≥ 0.99 2 Q 0 W 4 Q 0 W 4 Q 0 S_1 \ge\sum_{m=0}^{k}mQ_0 = \frac{k(k+1)}{2}Q_0 \ge\frac{Q_0}{2}\left(\frac{W^4}{Q_0}-2\right)\left(\frac{W^4}{Q_0}-1\right) \ge\frac{0.99}{2}Q_0\frac{W^4}{Q_0}\frac{W^4}{Q_0} S 1 ≥ m = 0 ∑ k m Q 0 = 2 k ( k + 1 ) Q 0 ≥ 2 Q 0 ( Q 0 W 4 − 2 ) ( Q 0 W 4 − 1 ) ≥ 2 0.99 Q 0 Q 0 W 4 Q 0 W 4
and hence
S 1 ≥ 0.99 2 W 4 W 4 66 H − 2 / 3 W 8 / 3 = 0.99 132 W 4 H 2 / 3 U ≥ 1 135 W 4 H 2 / 3 U . S_1 \ge\frac{0.99}{2}W^4\frac{W^4}{66H^{-2/3}W^{8/3}} = \frac{0.99}{132}W^4H^{2/3}U \ge\frac{1}{135}W^4H^{2/3}U. S 1 ≥ 2 0.99 W 4 66 H − 2/3 W 8/3 W 4 = 132 0.99 W 4 H 2/3 U ≥ 135 1 W 4 H 2/3 U .
The proof of Lemma 8 is complete.
By Lemma 8, the ratio of S 2 S_2 S 2 to S 1 S_1 S 1 has an upper bound given by
S 2 S 1 ≤ 18 ⋅ 135 H = 2430 10 6 = 0.00243. (52) \frac{S_2}{S_1} \le\frac{18\cdot135}{H} = \frac{2430}{10^6} = 0.00243. \tag*{(52)} S 1 S 2 ≤ H 18 ⋅ 135 = 1 0 6 2430 = 0.00243. (52)
An upper bound for ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣
Ostrowski’s theorem for a number field says that every nontrivial absolute value on L L L is equivalent to either a p p p -adic nonarchimedean absolute value for a nonzero prime ideal p \mathfrak{p} p of O L \mathcal{O}_L O L or an archimedean absolute value associated to a real or complex-conjugate pair of embeddings of L L L (see [33 ] and [6 ] ).
For a given nonzero integer Δ ∈ O L ∖ { 0 } \Delta\in\mathcal{O}_L \setminus\{0\} Δ ∈ O L ∖ { 0 } , the archimedean absolute value v ∣ ∞ v\mid\infty v ∣ ∞ that corresponds to a real embedding σ : K ↪ R \sigma: K \hookrightarrow\mathbb{R} σ : K ↪ R is defined as ∣ Δ ∣ v = ∣ σ ( Δ ) ∣ |\Delta|_v = |\sigma(\Delta)| ∣Δ ∣ v = ∣ σ ( Δ ) ∣ . For archimedean v ∣ ∞ v\mid\infty v ∣ ∞ that corresponds to a complex embedding σ : K ↪ C \sigma: K \hookrightarrow\mathbb{C} σ : K ↪ C , ∣ Δ ∣ v = ∣ σ ( Δ ) ∣ 2 |\Delta|_v = |\sigma(\Delta)|^2 ∣Δ ∣ v = ∣ σ ( Δ ) ∣ 2 . For the p \mathfrak{p} p -adic nonarchimedean absolute value, we define
∣ Δ ∣ p = N ( p ) − v p ( Δ ) |\Delta|_{\mathfrak{p}} = \mathcal{N}(\mathfrak{p})^{-v_{\mathfrak{p}}(\Delta)} ∣Δ ∣ p = N ( p ) − v p ( Δ )
where v p ( Δ ) ∈ Z ≥ 0 v_{\mathfrak{p}}(\Delta) \in\mathbb{Z}_{\geq0} v p ( Δ ) ∈ Z ≥ 0 is the p \mathfrak{p} p -adic valuation of Δ \Delta Δ .
The archimedean and nonarchimedean absolute values on L L L can be tied together by a product formula as follows (see [6 ] ).
Lemma 9 For Δ ∈ O L ∖ { 0 } \Delta\in\mathcal{O}_L \setminus\{0\} Δ ∈ O L ∖ { 0 } , we have
∏ v ∣ Δ ∣ v = 1 \prod_v |\Delta|_v = 1 v ∏ ∣Δ ∣ v = 1 where the product runs over the absolute values v v v on L L L .
Suppose that the principal integral ideal Δ O L \Delta\mathcal{O}_L Δ O L is factored into prime ideals
Δ O L = p 1 a 1 ⋯ p r a r \Delta\mathcal{O}_L = \mathfrak{p}_1^{a_1} \cdots\mathfrak{p}_r^{a_r} Δ O L = p 1 a 1 ⋯ p r a r
where a j a_j a j is a positive integer and p j \mathfrak{p}_j p j ’s are distinct prime ideals for j = 1 , … , r j = 1,\ldots,r j = 1 , … , r . While the archimedean contribution to ∏ v ∣ Δ ∣ v \prod_v |\Delta|_v ∏ v ∣Δ ∣ v is ∣ N L / Q ( Δ ) ∣ |N_{L/\mathbb{Q}}(\Delta)| ∣ N L / Q ( Δ ) ∣ , the nonarchimedean contribution to ∏ v ∣ Δ ∣ v \prod_v |\Delta|_v ∏ v ∣Δ ∣ v is given by
∏ j = 1 r N ( p j ) − v p j ( Δ ) = ∏ j = 1 r N ( p j ) − a j . \prod_{j=1}^{r} \mathcal{N}(\mathfrak{p}_j)^{-v_{\mathfrak{p}_j}(\Delta)}
=
\prod_{j=1}^{r} \mathcal{N}(\mathfrak{p}_j)^{-a_j}. j = 1 ∏ r N ( p j ) − v p j ( Δ ) = j = 1 ∏ r N ( p j ) − a j .
From the product formula, it follows that
∏ v ∣ ∞ ∣ Δ ∣ v = ∣ N L / Q ( Δ ) ∣ = ∏ j = 1 r N ( p j ) a j . (53) \prod_{v\mid\infty} |\Delta|_v = |N_{L/\mathbb{Q}}(\Delta)| = \prod_{j=1}^{r} \mathcal{N}(\mathfrak{p}_j)^{a_j}. \tag*{(53)} v ∣ ∞ ∏ ∣Δ ∣ v = ∣ N L / Q ( Δ ) ∣ = j = 1 ∏ r N ( p j ) a j . (53)
We note that the product formula reduces to the fundamental theorem of arithmetic in the case that L = Q L = \mathbb{Q} L = Q .
In the method of the field norm, we use the left-hand (or right-hand) side of its absolute value in (53) to find an upper (or a lower) bound for ∣ N L / Q ( Δ ) ∣ |N_{L/\mathbb{Q}}(\Delta)| ∣ N L / Q ( Δ ) ∣ .
In the current scenario, the field L L L is Q ( − N , 2 ) \mathbb{Q}(\sqrt{-N}, \sqrt{2}) Q ( − N , 2 ) with Gal ( L / Q ) = { e , σ , τ , σ τ } \operatorname{Gal}(L/\mathbb{Q}) = \{e,\sigma,\tau,\sigma\tau\} Gal ( L / Q ) = { e , σ , τ , σ τ } . It is seen that { e , σ } \{e,\sigma\} { e , σ } and { τ , σ τ } \{\tau,\sigma\tau\} { τ , σ τ } are two pairs of complex embeddings. Therefore,
∣ N L / Q ( Δ ) ∣ = ∣ Δ ∣ 2 ∣ τ ( Δ ) ∣ 2 . |N_{L/\mathbb{Q}}(\Delta)| = |\Delta|^2 |\tau(\Delta)|^2. ∣ N L / Q ( Δ ) ∣ = ∣Δ ∣ 2 ∣ τ ( Δ ) ∣ 2 .
By using (50), it is clear that
τ ( Δ ) = τ ( det ( M r i ) ) = det ( τ ( M r i ) ) \tau(\Delta)=\tau(\det(\mathcal{M}_{ri}))=\det(\tau(\mathcal{M}_{ri})) τ ( Δ ) = τ ( det ( M r i )) = det ( τ ( M r i ))
and hence
∣ N L / Q ( Δ ) ∣ = ∣ det ( M r i ) ∣ 2 ∣ det ( τ ( M r i ) ) ∣ 2 . |N_{L/\mathbb{Q}}(\Delta)|=|\det(\mathcal{M}_{ri})|^2|\det(\tau(\mathcal{M}_{ri}))|^2. ∣ N L / Q ( Δ ) ∣ = ∣ det ( M r i ) ∣ 2 ∣ det ( τ ( M r i )) ∣ 2 .
We introduce the following subset
C W = { n 1 + n 2 a + n 3 b + n 4 a b ∣ − W + 1 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ R C_W=\{n_1+n_2a+n_3b+n_4ab\mid-W+1\le n_i\le W-1\text{ for }i=1,2,3,4\}\subset\mathbb{R} C W = { n 1 + n 2 a + n 3 b + n 4 ab ∣ − W + 1 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ R
where n i ∈ Z n_i\in\mathbb{Z} n i ∈ Z for 1 ≤ i ≤ 4 1\le i\le4 1 ≤ i ≤ 4 . We have B W ⊂ C W B_W\subset C_W B W ⊂ C W . If θ n ⃗ ∈ C W \theta_{\vec n}\in C_W θ n ∈ C W , then σ ( θ n ⃗ ) ∈ C W \sigma(\theta_{\vec n})\in C_W σ ( θ n ) ∈ C W and τ ( θ n ⃗ ) ∈ C W \tau(\theta_{\vec n})\in C_W τ ( θ n ) ∈ C W and thus σ τ ( θ n ⃗ ) ∈ C W \sigma\tau(\theta_{\vec n})\in C_W σ τ ( θ n ) ∈ C W .
Every entry in the matrix M r i \mathcal{M}_{ri} M r i has the form of θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 \theta_{\vec n}^{\alpha_1}\sigma(\theta_{\vec n})^{\alpha_2}\sigma\tau(\theta_{\vec n})^{\alpha_3} θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 where
θ n ⃗ ∈ B W ⊂ C W \theta_{\vec n}\in B_W\subset C_W θ n ∈ B W ⊂ C W
and α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i . Thus, every entry in the matrix τ ( M r i ) \tau(\mathcal{M}_{ri}) τ ( M r i ) has the form of
τ ( θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 ) = τ ( θ n ⃗ ) α 1 σ τ ( θ n ⃗ ) α 2 σ ( θ n ⃗ ) α 3 \tau\left(\theta_{\vec n}^{\alpha_1}\sigma(\theta_{\vec n})^{\alpha_2}\sigma\tau(\theta_{\vec n})^{\alpha_3}\right)=\tau(\theta_{\vec n})^{\alpha_1}\sigma\tau(\theta_{\vec n})^{\alpha_2}\sigma(\theta_{\vec n})^{\alpha_3} τ ( θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 ) = τ ( θ n ) α 1 σ τ ( θ n ) α 2 σ ( θ n ) α 3
where σ τ = τ σ \sigma\tau=\tau\sigma σ τ = τ σ and σ σ = τ τ = e \sigma\sigma=\tau\tau=e σ σ = τ τ = e in the Galois group Gal ( L / Q ) \operatorname{Gal}(L/\mathbb{Q}) Gal ( L / Q ) .
It is clear that, for any θ n ⃗ = n 1 + n 2 a + n 3 b + n 4 a b ∈ C W \theta_{\vec n}=n_1+n_2a+n_3b+n_4ab\in C_W θ n = n 1 + n 2 a + n 3 b + n 4 ab ∈ C W ,
∣ θ n ⃗ ∣ ≤ ( W − 1 ) ( 1 + ∣ a ∣ + ∣ b ∣ + ∣ a b ∣ ) = ( W − 1 ) ( 1 + N + 2 + 2 N ) |\theta_{\vec n}|\le(W-1)(1+|a|+|b|+|ab|)=(W-1)(1+\sqrt{N}+\sqrt{2}+\sqrt{2N}) ∣ θ n ∣ ≤ ( W − 1 ) ( 1 + ∣ a ∣ + ∣ b ∣ + ∣ ab ∣ ) = ( W − 1 ) ( 1 + N + 2 + 2 N )
and hence from q = 4 N > 3 ⋅ 10 20 q=4N>3\cdot10^{20} q = 4 N > 3 ⋅ 1 0 20 that
∣ θ n ⃗ ∣ ≤ W ( 1 + 2 ) ( 1 + N ) = W ( 1 + 2 ) ( 1 + q / 2 ) ≤ 1.3 W q = θ max . (54) |\theta_{\vec n}|\le W(1+\sqrt{2})(1+\sqrt{N})=W(1+\sqrt{2})(1+\sqrt{q/2})\le1.3W\sqrt{q}=\theta_{\max}. \tag*{(54)} ∣ θ n ∣ ≤ W ( 1 + 2 ) ( 1 + N ) = W ( 1 + 2 ) ( 1 + q /2 ) ≤ 1.3 W q = θ m a x . (54)
Therefore, every row vector indexed by α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i of the two matrices M r i \mathcal{M}_{ri} M r i and τ ( M r i ) \tau(\mathcal{M}_{ri}) τ ( M r i ) , which both have the same size of W 4 × W 4 W^4\times W^4 W 4 × W 4 , has the length at most
W 4 θ max α 1 + α 2 + α 3 = W 2 θ max α 1 + α 2 + α 3 . \sqrt{W^4}\theta_{\max}^{\alpha_1+\alpha_2+\alpha_3}=W^2\theta_{\max}^{\alpha_1+\alpha_2+\alpha_3}. W 4 θ m a x α 1 + α 2 + α 3 = W 2 θ m a x α 1 + α 2 + α 3 .
Hadamard’s inequality provides an upper bound on the determinant of a square matrix, whose entries are complex numbers, in terms of the lengths of its row vectors. According to Hadamard’s inequality, we have
max { ∣ det ( M r i ) ∣ , ∣ det ( τ ( M r i ) ) ∣ } ≤ ∏ α ∈ P r i W 2 θ max α 1 + α 2 + α 3 = W 2 W 4 θ max S 1 + S 2 . \max\{|\det(\mathcal{M}_{ri})|,|\det(\tau(\mathcal{M}_{ri}))|\}\le\prod_{\alpha\in\mathcal{P}_{ri}}W^2\theta_{\max}^{\alpha_1+\alpha_2+\alpha_3}=W^{2W^4}\theta_{\max}^{S_1+S_2}. max { ∣ det ( M r i ) ∣ , ∣ det ( τ ( M r i )) ∣ } ≤ α ∈ P r i ∏ W 2 θ m a x α 1 + α 2 + α 3 = W 2 W 4 θ m a x S 1 + S 2 .
Therefore, we obtain
∣ N L / Q ( Δ ) ∣ = ∣ det ( M r i ) ∣ 2 ∣ det ( τ ( M r i ) ) ∣ 2 ≤ ( W 2 W 4 θ max S 1 + S 2 ) 4 |N_{L/\mathbb{Q}(\Delta)|=|\det(\mathcal{M}_{ri})|^2|\det(\tau(\mathcal{M}_{ri}))|^2\le\left(W^{2W^4}\theta_{\max}^{S_1+S_2}\right)^4} ∣ N L / Q ( Δ ) ∣ = ∣ d e t ( M r i ) ∣ 2 ∣ d e t ( τ ( M r i )) ∣ 2 ≤ ( W 2 W 4 θ m a x S 1 + S 2 ) 4
and hence
log ∣ N L / Q ( Δ ) ∣ ≤ 4 log ( W 2 W 4 θ max S 1 + S 2 ) . \log|N_{L/\mathbb{Q}}(\Delta)|\le4\log\left(W^{2W^4}\theta_{\max}^{S_1+S_2}\right). log ∣ N L / Q ( Δ ) ∣ ≤ 4 log ( W 2 W 4 θ m a x S 1 + S 2 ) .
By noting (54), we can attain the following upper bound
1 4 log ∣ N L / Q ( Δ ) ∣ ≤ 2 W 4 log W + ( S 1 + S 2 ) ( log W + log q + log 1.3 ) . (55) \frac{1}{4}\log|N_{L/\mathbb{Q}}(\Delta)|\le2W^4\log W+(S_1+S_2)(\log W+\log\sqrt{q}+\log1.3). \tag*{(55)} 4 1 log ∣ N L / Q ( Δ ) ∣ ≤ 2 W 4 log W + ( S 1 + S 2 ) ( log W + log q + log 1.3 ) . (55)
A lower bound for ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣
In the following, we use the right-hand side in (53) to find a lower bound of ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ .
We consider all primes p > H p > H p > H which are inert in K K K . It is to be seen that if p ≤ α 1 p \le\alpha_1 p ≤ α 1 for some α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha= (\alpha_1,\alpha_2,\alpha_3) \in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i , then p p p must be a prime factor of Δ \Delta Δ in the sense that Δ = p t \Delta= pt Δ = pt for some algebraic integer t ∈ O L t \in O_L t ∈ O L .
Let p ∈ Z p \in\mathbb{Z} p ∈ Z with p > H p > H p > H and χ ( p ) = − 1 \chi(p)=-1 χ ( p ) = − 1 . Thus, p p p is an odd prime and p ∤ ∣ Δ K ∣ p \nmid\lvert\Delta_K\rvert p ∤ ∣ Δ K ∣ . Remind that a = − N a=\sqrt{-N} a = − N , b = 2 b=\sqrt{2} b = 2 and R = Z [ a , b ] \mathcal{R}=\mathbb{Z}[a,b] R = Z [ a , b ] . By Euler’s criterion, we have
χ ( p ) = ( Δ K p ) = ( − 4 N p ) = ( − N p ) ≡ ( − N ) p − 1 2 ( m o d p Z ) \chi(p)=\left(\frac{\Delta_K}{p}\right)=\left(\frac{-4N}{p}\right)=\left(\frac{-N}{p}\right)\equiv(-N)^{\frac{p-1}{2}}\pmod{p\mathbb{Z}} χ ( p ) = ( p Δ K ) = ( p − 4 N ) = ( p − N ) ≡ ( − N ) 2 p − 1 ( mod p Z )
and hence
a p = ( − N ) p − 1 2 a ≡ − a ( m o d p R ) (56) a^p=(-N)^{\frac{p-1}{2}}a\equiv-a\pmod{p\mathcal{R}} \tag*{(56)} a p = ( − N ) 2 p − 1 a ≡ − a ( mod p R ) (56)
In a similar fashion, we get
b p = 2 p − 1 2 b ≡ ( 2 p ) b ( m o d p R ) (57) b^p=2^{\frac{p-1}{2}}b\equiv\left(\frac{2}{p}\right)b\pmod{p\mathcal{R}} \tag*{(57)} b p = 2 2 p − 1 b ≡ ( p 2 ) b ( mod p R ) (57)
Since p p p is odd, there is ( 2 p ) ∈ { − 1 , 1 } \left(\frac{2}{p}\right)\in\{-1,1\} ( p 2 ) ∈ { − 1 , 1 } . In the first case that ( 2 p ) = 1 \left(\frac{2}{p}\right)=1 ( p 2 ) = 1 , by using (57) we have
b p ≡ b ( m o d p R ) (58) b^p\equiv b\pmod{p\mathcal{R}} \tag*{(58)} b p ≡ b ( mod p R ) (58)
Let θ n ⃗ = n 1 + n 2 a + n 3 b + n 4 a b ∈ R \theta_{\vec n}=n_1+n_2a+n_3b+n_4ab\in\mathcal{R} θ n = n 1 + n 2 a + n 3 b + n 4 ab ∈ R . Then, by using the multinomial theorem and Fermat’s little theorem, we obtain
θ n ⃗ p ≡ n 1 p + n 2 p a p + n 3 p b p + n 4 p a p b p ≡ n 1 + n 2 a p + n 3 b p + n 4 a p b p ( m o d p R ) \theta_{\vec n}^p\equiv n_1^p+n_2^pa^p+n_3^pb^p+n_4^pa^pb^p\equiv n_1+n_2a^p+n_3b^p+n_4a^pb^p\pmod{p\mathcal{R}} θ n p ≡ n 1 p + n 2 p a p + n 3 p b p + n 4 p a p b p ≡ n 1 + n 2 a p + n 3 b p + n 4 a p b p ( mod p R )
and hence, from (56) and (58) ,
θ n ⃗ p ≡ n 1 − n 2 a + n 3 b − n 4 a b ( m o d p R ) \theta_{\vec n}^p\equiv n_1-n_2a+n_3b-n_4ab\pmod{p\mathcal{R}} θ n p ≡ n 1 − n 2 a + n 3 b − n 4 ab ( mod p R )
i.e.,
θ n ⃗ p ≡ σ ( θ n ⃗ ) ( m o d p R ) . (59) \theta_{\vec n}^p\equiv\sigma(\theta_{\vec n})\pmod{p\mathcal{R}}. \tag*{(59)} θ n p ≡ σ ( θ n ) ( mod p R ) . (59)
For each α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i with α 1 ≥ p \alpha_1\ge p α 1 ≥ p , we have α 1 = p k + r \alpha_1=pk+r α 1 = p k + r with 0 ≤ r < p 0\le r<p 0 ≤ r < p and
k = ⌊ α 1 p ⌋ ≥ 1. k=\left\lfloor\frac{\alpha_1}{p}\right\rfloor\ge1. k = ⌊ p α 1 ⌋ ≥ 1.
Comparing with the row vector R α = ( θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 ) R_\alpha=(\theta_{\vec n}^{\alpha_1}\sigma(\theta_{\vec n})^{\alpha_2}\sigma\tau(\theta_{\vec n})^{\alpha_3}) R α = ( θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 ) in the matrix M r i \mathcal{M}_{ri} M r i , we define the following row vector
R ~ α = ( ( θ n ⃗ p − σ ( θ n ⃗ ) ) k θ n ⃗ r σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 ) (60) \widetilde{R}_\alpha=\left((\theta_{\vec n}^p-\sigma(\theta_{\vec n}))^k\theta_{\vec n}^r\sigma(\theta_{\vec n})^{\alpha_2}\sigma\tau(\theta_{\vec n})^{\alpha_3}\right) \tag*{(60)} R α = ( ( θ n p − σ ( θ n ) ) k θ n r σ ( θ n ) α 2 σ τ ( θ n ) α 3 ) (60)
if p ≤ α 1 p \le\alpha_1 p ≤ α 1 and define R ~ α = R α \widetilde{R}_{\alpha}=R_{\alpha} R α = R α if p > α 1 p>\alpha_1 p > α 1 to get a new matrix M ~ r i \widetilde{\mathcal{M}}_{ri} M r i with row vectors R ~ α \widetilde{R}_{\alpha} R α .
Since
( θ n ⃗ p − σ ( θ n ⃗ ) ) k θ n ⃗ r = ∑ m = 0 k ( k m ) ( − 1 ) m ( σ ( θ n ⃗ ) ) m θ n ⃗ p k + r − p m , \left(\theta_{\vec n}^{p}-\sigma(\theta_{\vec n})\right)^{k}\theta_{\vec n}^{r}=\sum_{m=0}^{k}\binom{k}{m}(-1)^{m}\left(\sigma(\theta_{\vec n})\right)^{m}\theta_{\vec n}^{pk+r-pm}, ( θ n p − σ ( θ n ) ) k θ n r = m = 0 ∑ k ( m k ) ( − 1 ) m ( σ ( θ n ) ) m θ n p k + r − p m ,
we have
( θ n ⃗ p − σ ( θ n ⃗ ) ) k θ n ⃗ r = θ n ⃗ α 1 + ∑ m = 1 k ( k m ) ( − 1 ) m θ n ⃗ α 1 − p m ( σ ( θ n ⃗ ) ) m . \left(\theta_{\vec n}^{p}-\sigma(\theta_{\vec n})\right)^{k}\theta_{\vec n}^{r}=\theta_{\vec n}^{\alpha_1}+\sum_{m=1}^{k}\binom{k}{m}(-1)^{m}\theta_{\vec n}^{\alpha_1-pm}\left(\sigma(\theta_{\vec n})\right)^{m}. ( θ n p − σ ( θ n ) ) k θ n r = θ n α 1 + m = 1 ∑ k ( m k ) ( − 1 ) m θ n α 1 − p m ( σ ( θ n ) ) m .
Hence, for each α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i with α 1 ≥ p \alpha_1\ge p α 1 ≥ p ,
R ~ α = R α + ∑ m = 1 k ( k m ) ( − 1 ) m R α − ( p m , − m , 0 ) (61) \widetilde{R}_{\alpha}=R_{\alpha}+\sum_{m=1}^{k}\binom{k}{m}(-1)^{m}R_{\alpha-(pm,-m,0)} \tag*{(61)} R α = R α + m = 1 ∑ k ( m k ) ( − 1 ) m R α − ( p m , − m , 0 ) (61)
which is the α \alpha α -indexed row vector in the matrix M ~ r i \widetilde{\mathcal{M}}_{ri} M r i .
It is claimed that det ( M ~ r i ) = det ( M r i ) \det(\widetilde{\mathcal{M}}_{ri})=\det(\mathcal{M}_{ri}) det ( M r i ) = det ( M r i ) .
In fact, for the first index α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathcal{P}_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i for which R ~ α ≠ R α \widetilde{R}_{\alpha}\ne R_{\alpha} R α = R α , it is clear that α ≠ ( 0 , 0 , 0 ) \alpha\ne(0,0,0) α = ( 0 , 0 , 0 ) and the H H H -weight wt ( α ) > 0 \operatorname{wt}(\alpha)>0 wt ( α ) > 0 . Note that the index α − ( p m , − m , 0 ) \alpha-(pm,-m,0) α − ( p m , − m , 0 ) of R α − ( p m , − m , 0 ) R_{\alpha-(pm,-m,0)} R α − ( p m , − m , 0 ) in (61) may or may not be in the ordered set P r i \mathcal{P}_{ri} P r i of retained indices.
For p > H p>H p > H , the index α − ( p m , − m , 0 ) = ( α 1 − p m , α 2 + m , α 3 ) \alpha-(pm,-m,0)=(\alpha_1-pm,\alpha_2+m,\alpha_3) α − ( p m , − m , 0 ) = ( α 1 − p m , α 2 + m , α 3 ) has the H H H -weight
wt ( α − ( p m , − m , 0 ) ) = wt ( ( α 1 − p m , α 2 + m , α 3 ) ) = wt ( α ) − ( p − H ) m < wt ( α ) (62) \operatorname{wt}(\alpha-(pm,-m,0))=\operatorname{wt}((\alpha_1-pm,\alpha_2+m,\alpha_3))=\operatorname{wt}(\alpha)-(p-H)m<\operatorname{wt}(\alpha) \tag*{(62)} wt ( α − ( p m , − m , 0 )) = wt (( α 1 − p m , α 2 + m , α 3 )) = wt ( α ) − ( p − H ) m < wt ( α ) (62)
for 1 ≤ m ≤ k 1\le m\le k 1 ≤ m ≤ k . According to the procedure of selection for the index set P r i \mathcal{P}_{ri} P r i , it follows from (62) that the row vector R α − ( p m , − m , 0 ) R_{\alpha-(pm,-m,0)} R α − ( p m , − m , 0 ) must be in the span of the retained row vectors before R α R_{\alpha} R α . Therefore, by using (61) and (62), we attain
R ~ α = R α + ∑ α ^ c α ^ R α ^ (63) \widetilde{R}_{\alpha}=R_{\alpha}+\sum_{\widehat{\alpha}}c_{\widehat{\alpha}}R_{\widehat{\alpha}} \tag*{(63)} R α = R α + α ∑ c α R α (63)
where the summation ∑ α ^ \sum_{\widehat{\alpha}} ∑ α runs over α ^ ∈ P r i \widehat{\alpha}\in\mathcal{P}_{ri} α ∈ P r i with wt ( α ^ ) < wt ( α ) \operatorname{wt}(\widehat{\alpha})<\operatorname{wt}(\alpha) wt ( α ) < wt ( α ) and the coefficients c α ^ ∈ L c_{\widehat{\alpha}}\in L c α ∈ L are obtained in the operation for the form of linear combination. Since α ∈ P r i \alpha\in\mathcal{P}_{ri} α ∈ P r i is the first index for which R ~ α ≠ R α \widetilde{R}_{\alpha}\ne R_{\alpha} R α = R α , all R α ^ R_{\widehat{\alpha}} R α in the summation (63) are row vectors in the matrix M ~ r i \widetilde{\mathcal{M}}_{ri} M r i , i.e., R α ^ = R ~ α ^ R_{\widehat{\alpha}}=\widetilde{R}_{\widehat{\alpha}} R α = R α . Applying the theory of determinants, while the row vector R ~ α \widetilde{R}_{\alpha} R α in the matrix M ~ r i \widetilde{\mathcal{M}}_{ri} M r i is replaced by R α R_{\alpha} R α , the determinant of the matrix remains unchanged . For this updated matrix after row replacement, we will repeat the process, but the determinant of the matrix keeping the same. At the end of the procedure, the matrix M ~ r i \widetilde{\mathcal{M}}_{ri} M r i reduces to the matrix M r i \mathcal{M}_{ri} M r i . So, we have proved that
det ( M ~ r i ) = det ( M r i ) = Δ . \det(\widetilde{\mathcal{M}}_{ri})=\det(\mathcal{M}_{ri})=\Delta. det ( M r i ) = det ( M r i ) = Δ.
It can be seen from (59) and (60) that every entry in the row vector R ~ α \widetilde{R}_{\alpha} R α belongs to p k R p^{k}R p k R . Applying the theory of determinants, we have
Δ = det ( M ~ r i ) ∈ p k R . \Delta=\det(\widetilde{\mathcal{M}}_{ri})\in p^{k}R. Δ = det ( M r i ) ∈ p k R .
In the second case that ( 2 p ) = − 1 \left(\frac{2}{p}\right)=-1 ( p 2 ) = − 1 , by using (57) we have
b p ≡ − b ( m o d p R ) (64) b^{p} \equiv-b \pmod{p\mathcal{R}} \tag*{(64)} b p ≡ − b ( mod p R ) (64)
It follows from (56) and (64) that
θ n ⃗ p = n 1 + n 2 a p + n 3 b p + n 4 a p b p ≡ n 1 − n 2 a − n 3 b + n 4 a b ( m o d p R ) \theta_{\vec{n}}^{p}=n_{1}+n_{2}a^{p}+n_{3}b^{p}+n_{4}a^{p}b^{p}\equiv n_{1}-n_{2}a-n_{3}b+n_{4}ab \pmod{p\mathcal{R}} θ n p = n 1 + n 2 a p + n 3 b p + n 4 a p b p ≡ n 1 − n 2 a − n 3 b + n 4 ab ( mod p R )
i.e.,
θ n ⃗ p ≡ σ τ ( θ n ⃗ ) ( m o d p R ) \theta_{\vec{n}}^{p}\equiv\sigma\tau(\theta_{\vec{n}}) \pmod{p\mathcal{R}} θ n p ≡ σ τ ( θ n ) ( mod p R )
We can use the same proof as that in the first case with corresponding replacements such as the row vector in (60) replaced by
R ~ α = ( ( θ n ⃗ p − σ τ ( θ n ⃗ ) ) k θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 ) (65) \widetilde{R}_{\alpha}=\left((\theta_{\vec{n}}^{p}-\sigma\tau(\theta_{\vec{n}}))^{k}\theta_{\vec{n}}^{\alpha_{1}}\sigma(\theta_{\vec{n}})^{\alpha_{2}}\sigma\tau(\theta_{\vec{n}})^{\alpha_{3}}\right) \tag*{(65)} R α = ( ( θ n p − σ τ ( θ n ) ) k θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 ) (65)
and the row vectors R α − ( p m , − m , 0 ) R_{\alpha-(pm,-m,0)} R α − ( p m , − m , 0 ) with indices α − ( p m , − m , 0 ) \alpha-(pm,-m,0) α − ( p m , − m , 0 ) in (61) replaced by the row vectors R α − ( p m , 0 , − m ) R_{\alpha-(pm,0,-m)} R α − ( p m , 0 , − m ) with indices α − ( p m , 0 , − m ) \alpha-(pm,0,-m) α − ( p m , 0 , − m ) .
In both cases of ( 2 p ) = 1 \left(\frac{2}{p}\right)=1 ( p 2 ) = 1 and − 1 -1 − 1 , respectively, we can achieve the following result. For each α = ( α 1 , α 2 , α 3 ) ∈ P r i \alpha=(\alpha_{1},\alpha_{2},\alpha_{3})\in P_{ri} α = ( α 1 , α 2 , α 3 ) ∈ P r i with α 1 ≥ p \alpha_{1}\geq p α 1 ≥ p ,
Δ ∈ p k R with k = ⌊ α 1 p ⌋ . \Delta\in p^{k}\mathcal{R}\text{ with }k=\left\lfloor\frac{\alpha_{1}}{p}\right\rfloor. Δ ∈ p k R with k = ⌊ p α 1 ⌋ .
Therefore, we have
Δ ∈ p E p R with E p = ∑ α ∈ P r i ⌊ α 1 p ⌋ . \Delta\in p^{E_{p}}\mathcal{R}\text{ with }E_{p}=\sum_{\alpha\in P_{ri}}\left\lfloor\frac{\alpha_{1}}{p}\right\rfloor. Δ ∈ p E p R with E p = α ∈ P r i ∑ ⌊ p α 1 ⌋ .
According to (50) , we have Δ ∈ O L \Delta\in\mathcal{O}_{L} Δ ∈ O L and Δ ≠ 0 \Delta\ne0 Δ = 0 . Note that R ⊂ O L \mathcal{R}\subset\mathcal{O}_{L} R ⊂ O L . We can write Δ = p E p t \Delta=p^{E_{p}}t Δ = p E p t with t ∈ R t\in\mathcal{R} t ∈ R and t ≠ 0 t\ne0 t = 0 . Thus, N L / Q ( t ) ∈ Z N_{L/\mathbb{Q}}(t)\in\mathbb{Z} N L / Q ( t ) ∈ Z is a nonzero integer and
∣ N L / Q ( Δ ) ∣ = ∣ N L / Q ( p E p ) ∣ ∣ N L / Q ( t ) ∣ = ( p E p ) [ L : Q ] ∣ N L / Q ( t ) ∣ = p 4 E p ∣ N L / Q ( t ) ∣ ≥ p 4 E p . \left|N_{L/\mathbb{Q}}(\Delta)\right|=\left|N_{L/\mathbb{Q}}(p^{E_{p}})\right|\left|N_{L/\mathbb{Q}}(t)\right|=(p^{E_{p}})^{[L:\mathbb{Q}]}\left|N_{L/\mathbb{Q}}(t)\right|=p^{4E_{p}}\left|N_{L/\mathbb{Q}}(t)\right|\geq p^{4E_{p}}. N L / Q ( Δ ) = N L / Q ( p E p ) N L / Q ( t ) = ( p E p ) [ L : Q ] N L / Q ( t ) = p 4 E p N L / Q ( t ) ≥ p 4 E p .
Consequently, p 4 E p ∣ ∣ N L / Q ( Δ ) ∣ p^{4E_{p}}\mid\left|N_{L/\mathbb{Q}}(\Delta)\right| p 4 E p ∣ N L / Q ( Δ ) for all primes p > H p>H p > H with χ ( p ) = − 1 \chi(p)=-1 χ ( p ) = − 1 , and hence the following lower bound
1 4 log ∣ N L / Q ( Δ ) ∣ ≥ ∑ p > H χ ( p ) = − 1 E p log p with E p = ∑ α ∈ P r i ⌊ α 1 p ⌋ . (66) \frac{1}{4}\log\left|N_{L/\mathbb{Q}}(\Delta)\right|\geq\sum_{\substack{p>H\\\chi(p)=-1}}E_{p}\log p\text{ with }E_{p}=\sum_{\alpha\in P_{ri}}\left\lfloor\frac{\alpha_{1}}{p}\right\rfloor. \tag*{(66)} 4 1 log N L / Q ( Δ ) ≥ p > H χ ( p ) = − 1 ∑ E p log p with E p = α ∈ P r i ∑ ⌊ p α 1 ⌋ . (66)
Estimation of the distance between the real zero β \beta β and 1
We have analyzed the effect of the Landau–Siegel zero β \beta β extremely close to 1 causing the primes not inert in the imaginary quadratic fields K = Q ( − N ) K=\mathbb{Q}(\sqrt{-N}) K = Q ( − N ) to be very rare, whose estimation has been given in (48) . In the following, the method of the field norm using the upper bound of its absolute value ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ given in (55) and the lower bound given in (66) will be employed to attain an estimation for the distance between the real zero β \beta β and 1.
Suppose that the Landau–Siegel zero β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) of L ( s , χ ) L(s,\chi) L ( s , χ ) exists, which is real and simple and satisfies L ( β , χ ) = 0 L(\beta,\chi)=0 L ( β , χ ) = 0 .
Let W = ⌊ q η ⌋ W=\lfloor q^{\eta}\rfloor W = ⌊ q η ⌋ with parameter η ≥ 16 \eta\ge16 η ≥ 16 to be determined later. Remind that U = W 4 / 3 > q U=W^{4/3}>q U = W 4/3 > q . We choose the real variable X = U X=U X = U in (48) and obtain
∑ p ≤ U χ ( p ) ≠ − 1 log p p < ( 1 6 + e 2 4 ) log q + e 2 8 ( 1 − β ) ( log U ) 2 . \sum_{\substack{p\le U\\ \chi(p)\ne-1}}\frac{\log p}{p}<\left(\frac{1}{6}+\frac{e^{2}}{4}\right)\log q+\frac{e^{2}}{8}(1-\beta)(\log U)^{2}. p ≤ U χ ( p ) = − 1 ∑ p log p < ( 6 1 + 4 e 2 ) log q + 8 e 2 ( 1 − β ) ( log U ) 2 .
The lower bound for ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}}(\Delta)\rvert ∣ N L / Q ( Δ )∣ given in (66) yields the following
1 4 log ∣ N L / Q ( Δ ) ∣ ≥ ∑ H < p ≤ U χ ( p ) = − 1 ∑ α ∈ P r i ⌊ α 1 p ⌋ log p ≥ S 1 ∑ H < p ≤ U χ ( p ) = − 1 log p p − W 4 ∑ p ≤ U log p \frac{1}{4}\log\lvert N_{L/\mathbb{Q}}(\Delta)\rvert\ge\sum_{\substack{H<p\le U\\ \chi(p)=-1}}\sum_{\alpha\in\mathrm{Pri}}\left\lfloor\frac{\alpha_{1}}{p}\right\rfloor\log p\ge S_{1}\sum_{\substack{H<p\le U\\ \chi(p)=-1}}\frac{\log p}{p}-W^{4}\sum_{p\le U}\log p 4 1 log ∣ N L / Q ( Δ )∣ ≥ H < p ≤ U χ ( p ) = − 1 ∑ α ∈ Pri ∑ ⌊ p α 1 ⌋ log p ≥ S 1 H < p ≤ U χ ( p ) = − 1 ∑ p log p − W 4 p ≤ U ∑ log p
where ⌊ r ⌋ ≥ r − 1 \lfloor r\rfloor\ge r-1 ⌊ r ⌋ ≥ r − 1 for r ∈ R r\in\mathbb{R} r ∈ R is used.
According to [37 p. 70, (3.24)] , we have ∑ p ≤ H log p p ≤ log H \sum_{p\le H}\frac{\log p}{p}\le\log H ∑ p ≤ H p l o g p ≤ log H . By [37 p. 70, (3.21)] and [37 p. 65, (2.11)] , we get ∑ p ≤ U log p p ≥ log U − 1.333 − 1 / ( 2 log U ) \sum_{p\le U}\frac{\log p}{p}\ge\log U-1.333-1/(2\log U) ∑ p ≤ U p l o g p ≥ log U − 1.333 − 1/ ( 2 log U ) . Therefore,
∑ H < p ≤ U χ ( p ) = − 1 log p p ≥ ∑ p ≤ U log p p − ∑ p ≤ U χ ( p ) ≠ − 1 log p p − ∑ p ≤ H log p p ≥ log U − 1.5 − ∑ p ≤ U χ ( p ) ≠ − 1 log p p − log H . \sum_{\substack{H<p\le U\\ \chi(p)=-1}}\frac{\log p}{p}\ge\sum_{p\le U}\frac{\log p}{p}-\sum_{\substack{p\le U\\ \chi(p)\ne-1}}\frac{\log p}{p}-\sum_{p\le H}\frac{\log p}{p}\ge\log U-1.5-\sum_{\substack{p\le U\\ \chi(p)\ne-1}}\frac{\log p}{p}-\log H. H < p ≤ U χ ( p ) = − 1 ∑ p log p ≥ p ≤ U ∑ p log p − p ≤ U χ ( p ) = − 1 ∑ p log p − p ≤ H ∑ p log p ≥ log U − 1.5 − p ≤ U χ ( p ) = − 1 ∑ p log p − log H .
According to [37 p. 71, (3.32)] and [36 p. 212] , we have ∑ p ≤ U log p ≤ 1.0163 U \sum_{p\le U}\log p\le1.0163U ∑ p ≤ U log p ≤ 1.0163 U . Combining the above inequalities, we obtain the following lower bound with explicit constants
1 4 log ∣ N L / Q ( Δ ) ∣ ≥ S 1 [ log U − log H − 1.5 − ( 1 / 6 + e 2 / 4 ) log q − ( e 2 / 8 ) ( 1 − β ) ( log U ) 2 ] − 1.0163 W 4 U \frac{1}{4}\log\lvert N_{L/\mathbb{Q}}(\Delta)\rvert\ge S_{1}\left[\log U-\log H-1.5-\left(1/6+e^{2}/4\right)\log q-\left(e^{2}/8\right)(1-\beta)(\log U)^{2}\right]-1.0163W^{4}U 4 1 log ∣ N L / Q ( Δ )∣ ≥ S 1 [ log U − log H − 1.5 − ( 1/6 + e 2 /4 ) log q − ( e 2 /8 ) ( 1 − β ) ( log U ) 2 ] − 1.0163 W 4 U
where H = 10 6 H=10^{6} H = 1 0 6 . On the other hand, from (55) we have the following upper bound
1 4 log ∣ N L / Q ( Δ ) ∣ ≤ 2 W 4 log W + ( S 1 + S 2 ) ( log W + log q + 0.2624 ) . \frac{1}{4}\log\lvert N_{L/\mathbb{Q}}(\Delta)\rvert\le2W^{4}\log W+(S_{1}+S_{2})(\log W+\log\sqrt{q}+0.2624). 4 1 log ∣ N L / Q ( Δ )∣ ≤ 2 W 4 log W + ( S 1 + S 2 ) ( log W + log q + 0.2624 ) .
Remind that S 1 ≥ 1 135 H 2 / 3 W 4 U S_{1}\ge\frac{1}{135}H^{2/3}W^{4}U S 1 ≥ 135 1 H 2/3 W 4 U and S 2 ≤ 18 H − 1 / 3 W 4 U S_{2}\le18H^{-1/3}W^{4}U S 2 ≤ 18 H − 1/3 W 4 U in Lemma 8, which satisfy S 2 / S 1 ≤ 0.00243 S_{2}/S_{1}\le0.00243 S 2 / S 1 ≤ 0.00243 given by (52) . Therefore, we obtain
log ∣ N L / Q ( Δ ) ∣ 4 S 1 log U ≤ 270 log W H 2 / 3 U log U + 1.00243 ( log W log U + log q 2 log U + 0.2624 log U ) . \frac{\log\lvert N_{L/\mathbb{Q}}(\Delta)\rvert}{4S_{1}\log U}\le270\frac{\log W}{H^{2/3}U\log U}+1.00243\left(\frac{\log W}{\log U}+\frac{\log q}{2\log U}+\frac{0.2624}{\log U}\right). 4 S 1 log U log ∣ N L / Q ( Δ )∣ ≤ 270 H 2/3 U log U log W + 1.00243 ( log U log W + 2 log U log q + log U 0.2624 ) .
Note that q > 3 ⋅ 10 20 q > 3 \cdot10^{20} q > 3 ⋅ 1 0 20 . It is clear that W = ⌊ q η ⌋ W = \lfloor q^\eta\rfloor W = ⌊ q η ⌋ with η ≥ 16 \eta\ge16 η ≥ 16 satisfies q 15 ≤ W ≤ q η q^{15} \le W \le q^\eta q 15 ≤ W ≤ q η . We have put U = W 4 / 3 U = W^{4/3} U = W 4/3 and H = 10 6 H = 10^6 H = 1 0 6 . Therefore,
log ∣ N L / Q ( Δ ) ∣ 4 S 1 log U ≤ 1.00243 3 4 + 1.00243 log q 2 log U + 0.001 < 0.7529 + 0.5013 log q log U . \frac{\log\lvert N_{L/\mathbb{Q}}(\Delta)\rvert}{4S_1\log U} \le1.00243\frac{3}{4}+\frac{1.00243\log q}{2\log U}+0.001 < 0.7529+0.5013\frac{\log q}{\log U}. 4 S 1 log U log ∣ N L / Q ( Δ )∣ ≤ 1.00243 4 3 + 2 log U 1.00243 log q + 0.001 < 0.7529 + 0.5013 log U log q .
Since S 1 ≥ 1 135 H 2 / 3 W 4 U S_1 \ge\frac{1}{135}H^{2/3}W^4U S 1 ≥ 135 1 H 2/3 W 4 U , we have
1.0163 W 4 U S 1 log U ≤ 1.0163 135 10000 log U < 0.0001. 1.0163\frac{W^4U}{S_1\log U}\le1.0163\frac{135}{10000\log U}<0.0001. 1.0163 S 1 log U W 4 U ≤ 1.0163 10000 log U 135 < 0.0001.
Combining the lower and upper bounds of ∣ N L / Q ( Δ ) ∣ \lvert N_{L/\mathbb{Q}(\Delta)\rvert} ∣ N L / Q ( Δ )∣ , we can attain
1 − log H log U − 1.5 log U − 0.7530 ≤ ( 1 6 + e 2 4 + 0.5013 ) log q log U + e 2 8 ( 1 − β ) log U . 1-\frac{\log H}{\log U}-\frac{1.5}{\log U}-0.7530\le\left(\frac{1}{6}+\frac{e^2}{4}+0.5013\right)\frac{\log q}{\log U}+\frac{e^2}{8}(1-\beta)\log U. 1 − log U log H − log U 1.5 − 0.7530 ≤ ( 6 1 + 4 e 2 + 0.5013 ) log U log q + 8 e 2 ( 1 − β ) log U .
It is estimated that log H / log U ≤ 6 / 400 = 0.015 \log H/\log U\le6/400=0.015 log H / log U ≤ 6/400 = 0.015 . Thus, we have
0.229 < ( 1 6 + e 2 4 + 0.5013 ) log q log U + e 2 8 ( 1 − β ) log U < 2.5153 log q log U + 0.9237 ( 1 − β ) log U . 0.229<\left(\frac{1}{6}+\frac{e^2}{4}+0.5013\right)\frac{\log q}{\log U}+\frac{e^2}{8}(1-\beta)\log U<2.5153\frac{\log q}{\log U}+0.9237(1-\beta)\log U. 0.229 < ( 6 1 + 4 e 2 + 0.5013 ) log U log q + 8 e 2 ( 1 − β ) log U < 2.5153 log U log q + 0.9237 ( 1 − β ) log U .
In W = ⌊ q η ⌋ W=\lfloor q^\eta\rfloor W = ⌊ q η ⌋ , we choose
η = 16.4751. \eta=16.4751. η = 16.4751.
Then, W = ⌊ q 16.4751 ⌋ ≥ q 16.4751 − 1 = q 16.4750 ( q 0.0001 − q − 16.4750 ) > q 16.4750 W=\lfloor q^{16.4751}\rfloor\ge q^{16.4751}-1=q^{16.4750}(q^{0.0001}-q^{-16.4750})>q^{16.4750} W = ⌊ q 16.4751 ⌋ ≥ q 16.4751 − 1 = q 16.4750 ( q 0.0001 − q − 16.4750 ) > q 16.4750 . Noting that log U = ( 4 / 3 ) log W ≤ ( 4 / 3 ) η log q \log U=(4/3)\log W\le(4/3)\eta\log q log U = ( 4/3 ) log W ≤ ( 4/3 ) η log q , we get
0.229 < 2.5153 3 4 ⋅ 16.4750 + 0.9237 4 3 η ( 1 − β ) log q 0.229<2.5153\frac{3}{4\cdot16.4750}+0.9237\frac{4}{3}\eta(1-\beta)\log q 0.229 < 2.5153 4 ⋅ 16.4750 3 + 0.9237 3 4 η ( 1 − β ) log q
and hence
( 1 − β ) log q > 3 ( 0.229 − 0.1146 ) 3.6948 η > 0.005638 , (1-\beta)\log q>\frac{3(0.229-0.1146)}{3.6948\eta}>0.005638, ( 1 − β ) log q > 3.6948 η 3 ( 0.229 − 0.1146 ) > 0.005638 ,
i.e.,
1 − β > 0.005638 1 log q . (67) 1-\beta>0.005638\frac{1}{\log q}. \tag*{(67)} 1 − β > 0.005638 log q 1 . (67)
We can summarize the above result with explicit constants in the form of a zero-free interval located in ( 0 , 1 ) (0,1) ( 0 , 1 ) for L ( s , χ ) L(s,\chi) L ( s , χ ) , which will be utilized in the next section.
Proposition 8 Let N N N be any given squarefree positive integer with N > 7.5 ⋅ 10 19 N>7.5\cdot10^{19} N > 7.5 ⋅ 1 0 19 and N ≡ 1 , 2 ( m o d 4 ) N\equiv1,2\pmod{4} N ≡ 1 , 2 ( mod 4 ) and put D = − 4 N D=-4N D = − 4 N and q = 4 N q=4N q = 4 N . Let χ ( n ) = ( D n ) \chi(n)=\left(\frac{D}{n}\right) χ ( n ) = ( n D ) be the real primitive Dirichlet character of K = Q ( D ) K=\mathbb{Q}(\sqrt{D}) K = Q ( D ) and L ( s , χ ) L(s,\chi) L ( s , χ ) be the corresponding Dirichlet L L L -function. If the number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) satisfies
1 − β ≤ 0.0056 1 log q (68) 1-\beta\le0.0056\frac{1}{\log q} \tag*{(68)} 1 − β ≤ 0.0056 log q 1 (68) then L ( s , χ ) ≠ 0 L(s,\chi)\ne0 L ( s , χ ) = 0 for s s s in the interval [ β , 1 ) [\beta,1) [ β , 1 ) .
Proof of Proposition 6
It is now ready to prove Proposition 6 for every squarefree positive integer N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 with N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) , which, in turn, proves Theorem 1 as stated in § 4 .
By using the class number formula given in (15) , it is clear that Proposition 7 together with Proposition 8 can imply the following result.
Theorem 4 Let N N N be any given squarefree positive integer with N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 and N ≡ 1 , 2 ( m o d 4 ) N \equiv1, 2 \pmod{4} N ≡ 1 , 2 ( mod 4 ) and put D = − 4 N D = -4N D = − 4 N and q = 4 N q = 4N q = 4 N . Let h K h_K h K be the class number of the imaginary quadratic field K = Q ( D ) K = \mathbb{Q}(\sqrt{D}) K = Q ( D ) and χ ( n ) = ( D n ) \chi(n) = \left(\frac{D}{n}\right) χ ( n ) = ( n D ) be the Kronecker symbol with the corresponding Dirichlet L L L -function L ( s , χ ) L(s,\chi) L ( s , χ ) . Then, we have
L ( 1 , χ ) > 0.013608 1 log q (69) L(1,\chi) > 0.013608 \frac{1}{\log q} \tag*{(69)} L ( 1 , χ ) > 0.013608 log q 1 (69) and
h K > 0.00433 q log q . (70) h_K > 0.00433 \frac{\sqrt{q}}{\log q}. \tag*{(70)} h K > 0.00433 log q q . (70)
In fact, from Proposition 8 we can choose β = 1 − 0.0056 / log q ∈ ( 0 , 1 ) \beta= 1 - 0.0056/\log q \in(0,1) β = 1 − 0.0056/ log q ∈ ( 0 , 1 ) , which satisfies (33) in applying Proposition 7.
We are going to prove that (17) in Proposition 6 holds, namely,
h K ≠ 2 ω ( 4 N ) − 1 h_K \ne2^{\omega(4N)-1} h K = 2 ω ( 4 N ) − 1
where ω ( 4 N ) \omega(4N) ω ( 4 N ) is the number of distinct prime factors of 4 N 4N 4 N .
If the squarefree positive integer N N N satisfies N ≡ 1 ( m o d 4 ) N \equiv1 \pmod{4} N ≡ 1 ( mod 4 ) , we can write
N = p 1 ′ ⋯ p ℓ ′ N = p'_1 \cdots p'_\ell N = p 1 ′ ⋯ p ℓ ′
where p 1 ′ , … , p ℓ ′ p'_1,\ldots,p'_\ell p 1 ′ , … , p ℓ ′ are ℓ \ell ℓ distinct odd primes with an even number of primes ≡ 3 ( m o d 4 ) \equiv3 \pmod{4} ≡ 3 ( mod 4 ) .
For any squarefree positive integer N N N with N ≡ 2 ( m o d 4 ) N \equiv2 \pmod{4} N ≡ 2 ( mod 4 ) , we can write
N = 2 p 1 ′ ⋯ p ℓ ′ N = 2p'_1 \cdots p'_\ell N = 2 p 1 ′ ⋯ p ℓ ′
where p 1 ′ , … , p ℓ ′ p'_1,\ldots,p'_\ell p 1 ′ , … , p ℓ ′ are ℓ \ell ℓ distinct odd primes. In both cases, we have ω ( 4 N ) = ℓ + 1 \omega(4N) = \ell+ 1 ω ( 4 N ) = ℓ + 1 . Therefore, we need to prove that
h K ≠ 2 ℓ . (71) h_K \ne2^\ell. \tag*{(71)} h K = 2 ℓ . (71)
Since N > 7.5 ⋅ 10 19 N > 7.5 \cdot10^{19} N > 7.5 ⋅ 1 0 19 , we have q = 4 N > 3 ⋅ 10 20 q = 4N > 3 \cdot10^{20} q = 4 N > 3 ⋅ 1 0 20 . By using (70) , we get
h K > 0.00433 q log q > 1590610. h_K > 0.00433 \frac{\sqrt{q}}{\log q} > 1590610. h K > 0.00433 log q q > 1590610.
In the case that 1 ≤ ℓ ≤ 20 1 \le\ell\le20 1 ≤ ℓ ≤ 20 , we have
2 ℓ ≤ 2 20 = 1048576 < h K . 2^{\ell} \le2^{20} = 1048576 < h_K. 2 ℓ ≤ 2 20 = 1048576 < h K .
For any given ℓ ≥ 21 \ell\ge21 ℓ ≥ 21 , we have
q = 4 N > 2 p 1 ′ ⋯ p ℓ ′ q = 4N > 2p'_1 \cdots p'_{\ell} q = 4 N > 2 p 1 ′ ⋯ p ℓ ′
where p 1 ′ , … , p ℓ ′ p'_1,\ldots,p'_{\ell} p 1 ′ , … , p ℓ ′ are ℓ \ell ℓ distinct odd primes.
The primorial p n # p_n\# p n # is defined as the product of the first n n n primes
p n # = ∏ k = 1 n p k p_n\# = \prod_{k=1}^{n} p_k p n # = k = 1 ∏ n p k
where p k p_k p k is the k k k th prime number. Therefore, q > p ℓ + 1 # q > p_{\ell+1}\# q > p ℓ + 1 # .
It is clear that the function x log x \frac{x}{\log x} l o g x x of real variable x > e x > e x > e is an increasing function. For ℓ ≥ 21 \ell\ge21 ℓ ≥ 21 , we have p ℓ + 1 # > p 21 ℓ − 19 p 20 # p_{\ell+1}\# > p_{21}^{\ell-19}p_{20}\# p ℓ + 1 # > p 21 ℓ − 19 p 20 # and hence
q log q ≥ p ℓ + 1 # log ( p ℓ + 1 # ) ≥ ( p 21 ) ℓ − 19 p 20 # ( ℓ − 19 ) log p 21 + log ( p 20 # ) . \frac{\sqrt{q}}{\log q} \ge\frac{\sqrt{p_{\ell+1}\#}}{\log(p_{\ell+1}\#)} \ge\frac{(\sqrt{p_{21}})^{\ell-19}\sqrt{p_{20}\#}}{(\ell-19)\log p_{21}+\log(p_{20}\#)}. log q q ≥ log ( p ℓ + 1 # ) p ℓ + 1 # ≥ ( ℓ − 19 ) log p 21 + log ( p 20 # ) ( p 21 ) ℓ − 19 p 20 # .
Therefore, by using (70) , we obtain for ℓ ≥ 21 \ell\ge21 ℓ ≥ 21
h K 2 ℓ > 0.00433 2 ℓ q log q ≥ 0.00433 2 19 ( p 21 / 2 ) ℓ − 19 p 20 # ( ℓ − 19 ) log p 21 + log ( p 20 # ) . \frac{h_K}{2^{\ell}} > \frac{0.00433}{2^{\ell}}\frac{\sqrt{q}}{\log q} \ge\frac{0.00433}{2^{19}}\frac{(\sqrt{p_{21}}/2)^{\ell-19}\sqrt{p_{20}\#}}{(\ell-19)\log p_{21}+\log(p_{20}\#)}. 2 ℓ h K > 2 ℓ 0.00433 log q q ≥ 2 19 0.00433 ( ℓ − 19 ) log p 21 + log ( p 20 # ) ( p 21 /2 ) ℓ − 19 p 20 # .
where p 20 = 71 p_{20}=71 p 20 = 71 and p 21 = 73 p_{21}=73 p 21 = 73 . Since ( ℓ − 19 ) log p 21 + log ( p 20 # ) ≤ ( ℓ − 19 ) log p 21 log ( p 20 # ) (\ell-19)\log p_{21}+\log(p_{20}\#) \le(\ell-19)\log p_{21}\log(p_{20}\#) ( ℓ − 19 ) log p 21 + log ( p 20 # ) ≤ ( ℓ − 19 ) log p 21 log ( p 20 # ) for ℓ ≥ 21 \ell\ge21 ℓ ≥ 21 , we have
h K 2 ℓ ≥ 0.00433 2 19 log p 21 p 20 # log ( p 20 # ) ( p 21 / 2 ) ℓ − 19 ℓ − 19 . \frac{h_K}{2^{\ell}} \ge\frac{0.00433}{2^{19}\log p_{21}}\frac{\sqrt{p_{20}\#}}{\log(p_{20}\#)}\frac{(\sqrt{p_{21}}/2)^{\ell-19}}{\ell-19}. 2 ℓ h K ≥ 2 19 log p 21 0.00433 log ( p 20 # ) p 20 # ℓ − 19 ( p 21 /2 ) ℓ − 19 .
Noting that the right-hand side is an increasing function for ℓ ≥ 21 \ell\ge21 ℓ ≥ 21 , we can attain
h K 2 ℓ ≥ 0.00433 2 19 log p 21 p 20 # log ( p 20 # ) ( p 21 / 2 ) 2 2 . \frac{h_K}{2^{\ell}} \ge\frac{0.00433}{2^{19}\log p_{21}}\frac{\sqrt{p_{20}\#}}{\log(p_{20}\#)}\frac{(\sqrt{p_{21}}/2)^2}{2}. 2 ℓ h K ≥ 2 19 log p 21 0.00433 log ( p 20 # ) p 20 # 2 ( p 21 /2 ) 2 .
It is easily checked that
0.00433 log p 21 ( p 21 / 2 ) 2 2 = 0.00433 log 73 ( 73 / 2 ) 2 2 > 0.0092 \frac{0.00433}{\log p_{21}}\frac{(\sqrt{p_{21}}/2)^2}{2} = \frac{0.00433}{\log73}\frac{(\sqrt{73}/2)^2}{2} > 0.0092 log p 21 0.00433 2 ( p 21 /2 ) 2 = log 73 0.00433 2 ( 73 /2 ) 2 > 0.0092
and
p 20 # log ( p 20 # ) > 3.8353 ⋅ 10 11 . \frac{\sqrt{p_{20}\#}}{\log(p_{20}\#)} > 3.8353 \cdot10^{11}. log ( p 20 # ) p 20 # > 3.8353 ⋅ 1 0 11 .
Thus,
h K 2 ℓ > 0.0092 2 19 ⋅ 3.8353 ⋅ 10 11 > 1. \frac{h_K}{2^{\ell}} > \frac{0.0092}{2^{19}} \cdot3.8353 \cdot10^{11} > 1. 2 ℓ h K > 2 19 0.0092 ⋅ 3.8353 ⋅ 1 0 11 > 1.
Therefore, we obtain h K > 2 ℓ h_K > 2^{\ell} h K > 2 ℓ and hence h K ≠ 2 ℓ h_K \ne2^{\ell} h K = 2 ℓ in (71) for all ℓ ≥ 1 \ell\ge1 ℓ ≥ 1 .
The proof of Proposition 6 is complete. □
Proof of Lemma 6
For T ≥ T 0 T \ge T_0 T ≥ T 0 , where T 0 ≥ 2 T_0 \ge2 T 0 ≥ 2 is a sufficiently large but fixed positive number, we consider the following integral
ρ ( T ) = ∫ − 1 − β + i T 2 − β + i T f ( s ) d s \rho(T) = \int_{-1-\beta+iT}^{2-\beta+iT} f(s)\,ds ρ ( T ) = ∫ − 1 − β + i T 2 − β + i T f ( s ) d s
where the number β ∈ ( 0 , 1 ) \beta\in(0,1) β ∈ ( 0 , 1 ) is fixed and the function f ( s ) f(s) f ( s ) is defined by
f ( s ) = ζ K ( s + β ) y s s ∏ ℓ = 2 9 ( s + ℓ ) = ζ ( s + β ) L ( s + β , χ ) y s s ∏ ℓ = 2 9 ( s + ℓ ) f(s) = \frac{\zeta_K(s+\beta)y^s}{s\prod_{\ell=2}^{9}(s+\ell)} = \frac{\zeta(s+\beta)L(s+\beta,\chi)y^s}{s\prod_{\ell=2}^{9}(s+\ell)} f ( s ) = s ∏ ℓ = 2 9 ( s + ℓ ) ζ K ( s + β ) y s = s ∏ ℓ = 2 9 ( s + ℓ ) ζ ( s + β ) L ( s + β , χ ) y s
for a given number y > 1 y > 1 y > 1 .
By using s = σ + i T s=\sigma+iT s = σ + i T with − 1 − β ≤ σ ≤ 2 − β -1-\beta\le\sigma\le2-\beta − 1 − β ≤ σ ≤ 2 − β , the complex integral ρ ( T ) \rho(T) ρ ( T ) can be rewritten as
ρ ( T ) = ∫ − 1 − β 2 − β f ( σ + i T ) d σ . \rho(T) = \int_{-1-\beta}^{2-\beta} f(\sigma+iT)\,d\sigma. ρ ( T ) = ∫ − 1 − β 2 − β f ( σ + i T ) d σ .
With a change of variable, we obtain
ρ ( T ) = ∫ − 1 2 f ( σ − β + i T ) d σ = ∫ − 1 2 ζ ( σ + i T ) L ( σ + i T , χ ) y σ − β + i T ( σ − β + i T ) ∏ ℓ = 2 9 ( σ − β + i T + ℓ ) d σ . \rho(T) = \int_{-1}^{2} f(\sigma-\beta+iT)\,d\sigma= \int_{-1}^{2} \frac{\zeta(\sigma+iT)L(\sigma+iT,\chi)y^{\sigma-\beta+iT}}{(\sigma-\beta+iT)\prod_{\ell=2}^{9}(\sigma-\beta+iT+\ell)}\,d\sigma. ρ ( T ) = ∫ − 1 2 f ( σ − β + i T ) d σ = ∫ − 1 2 ( σ − β + i T ) ∏ ℓ = 2 9 ( σ − β + i T + ℓ ) ζ ( σ + i T ) L ( σ + i T , χ ) y σ − β + i T d σ .
We are going to prove that lim T → ∞ ∣ ρ ( T ) ∣ = 0 \lim_{T\to\infty}|\rho(T)|=0 lim T → ∞ ∣ ρ ( T ) ∣ = 0 and hence lim T → ∞ ρ ( T ) = 0 \lim_{T\to\infty}\rho(T)=0 lim T → ∞ ρ ( T ) = 0 . It is clear that
∣ ρ ( T ) ∣ ≤ ∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ + ∫ 0 1 ∣ f ( σ − β + i T ) ∣ d σ + ∫ 1 2 ∣ f ( σ − β + i T ) ∣ d σ . (72) |\rho(T)| \le\int_{-1}^{0}|f(\sigma-\beta+iT)|\,d\sigma+\int_{0}^{1}|f(\sigma-\beta+iT)|\,d\sigma+\int_{1}^{2}|f(\sigma-\beta+iT)|\,d\sigma. \tag*{(72)} ∣ ρ ( T ) ∣ ≤ ∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ + ∫ 0 1 ∣ f ( σ − β + i T ) ∣ d σ + ∫ 1 2 ∣ f ( σ − β + i T ) ∣ d σ . (72)
In the case that 0 ≤ σ ≤ 1 0\le\sigma\le1 0 ≤ σ ≤ 1 , we can use the properties of ζ ( s ) \zeta(s) ζ ( s ) and L ( s , χ ) L(s,\chi) L ( s , χ ) , where s = σ + i T s=\sigma+iT s = σ + i T . Since T ≥ T 0 T\ge T_0 T ≥ T 0 , we have (see [34 p. 142, Theorem 4] )
∣ ζ ( σ + i T ) ∣ ≤ C 1 T ( 1 − σ ) / 2 log T |\zeta(\sigma+iT)| \le C_1T^{(1-\sigma)/2}\log T ∣ ζ ( σ + i T ) ∣ ≤ C 1 T ( 1 − σ ) /2 log T
for 0 ≤ σ ≤ 1 0\le\sigma\le1 0 ≤ σ ≤ 1 , where C 1 > 0 C_1>0 C 1 > 0 is an absolute constant. Moreover, since χ \chi χ is a real primitive character of modulus q ≥ 3 q\ge3 q ≥ 3 , it is known that (see [34 p. 269, Theorem 4] )
∣ L ( σ + i T , χ ) ∣ ≤ C 2 ( q T ) ( 1 − σ ) / 2 log ( q T ) |L(\sigma+iT,\chi)| \le C_2(qT)^{(1-\sigma)/2}\log(qT) ∣ L ( σ + i T , χ ) ∣ ≤ C 2 ( q T ) ( 1 − σ ) /2 log ( q T )
for 0 ≤ σ ≤ 1 0\le\sigma\le1 0 ≤ σ ≤ 1 , where C 2 > 0 C_2>0 C 2 > 0 is an absolute constant. Therefore,
∫ 0 1 ∣ f ( σ − β + i T ) ∣ d σ = ∫ 0 1 ∣ ζ ( σ + i T ) ∣ ∣ L ( σ + i T , χ ) ∣ y σ − β ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ d σ ≤ C 3 y 1 − β T (73) \int_{0}^{1}|f(\sigma-\beta+iT)|\,d\sigma= \int_{0}^{1}\frac{|\zeta(\sigma+iT)||L(\sigma+iT,\chi)|y^{\sigma-\beta}}{|\sigma-\beta+iT|\prod_{\ell=2}^{9}|\sigma-\beta+iT+\ell|}\,d\sigma\le C_3\frac{y^{1-\beta}}{T} \tag*{(73)} ∫ 0 1 ∣ f ( σ − β + i T ) ∣ d σ = ∫ 0 1 ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ ∣ ζ ( σ + i T ) ∣∣ L ( σ + i T , χ ) ∣ y σ − β d σ ≤ C 3 T y 1 − β (73)
where C 3 = C 3 ( q ) > 0 C_3=C_3(q)>0 C 3 = C 3 ( q ) > 0 is a constant.
In the case that 1 ≤ σ ≤ 2 1 \le\sigma\le2 1 ≤ σ ≤ 2 , we split it into two situations of 1 ≤ σ ≤ 3 / 2 1 \le\sigma\le3/2 1 ≤ σ ≤ 3/2 and 3 / 2 < σ ≤ 2 3/2 < \sigma\le2 3/2 < σ ≤ 2 . Applying the Phragmén–Lindelöf principle, Rademacher proved the following upper bound [35 p. 200]
∣ ζ K ( σ + i T ) ∣ ≤ 3 ∣ 1 + σ + i T ∣ ∣ 1 − σ − i T ∣ ( ∣ Δ K ∣ 4 π 2 ∣ 1 + σ + i T ∣ 2 ) 3 / 2 − σ 2 ζ ( 3 / 2 ) 2 |\zeta_K(\sigma+iT)| \le3\frac{|1+\sigma+iT|}{|1-\sigma-iT|}\left(\frac{|\Delta_K|}{4\pi^2}|1+\sigma+iT|^2\right)^{\frac{3/2-\sigma}{2}}\zeta(3/2)^2 ∣ ζ K ( σ + i T ) ∣ ≤ 3 ∣1 − σ − i T ∣ ∣1 + σ + i T ∣ ( 4 π 2 ∣ Δ K ∣ ∣1 + σ + i T ∣ 2 ) 2 3/2 − σ ζ ( 3/2 ) 2
for 1 ≤ σ ≤ 3 / 2 1 \le\sigma\le3/2 1 ≤ σ ≤ 3/2 . On the other hand, we have
∣ ζ K ( σ + i T ) ∣ ≤ ζ ( σ ) 2 ≤ ζ ( 3 / 2 ) 2 |\zeta_K(\sigma+iT)| \le\zeta(\sigma)^2 \le\zeta(3/2)^2 ∣ ζ K ( σ + i T ) ∣ ≤ ζ ( σ ) 2 ≤ ζ ( 3/2 ) 2
for 3 / 2 ≤ σ ≤ 2 3/2 \le\sigma\le2 3/2 ≤ σ ≤ 2 . Therefore,
∫ 1 2 ∣ f ( σ − β + i T ) ∣ d σ = ∫ 1 2 ∣ ζ K ( σ + i T ) ∣ y σ − β ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ d σ ≤ C 4 y 2 − β T (74) \int_1^2 |f(\sigma-\beta+iT)|\mathrm{d}\sigma= \int_1^2 \frac{|\zeta_K(\sigma+iT)|y^{\sigma-\beta}}{|\sigma-\beta+iT|\prod_{\ell=2}^{9}|\sigma-\beta+iT+\ell|}\mathrm{d}\sigma\le C_4\frac{y^{2-\beta}}{T} \tag*{(74)} ∫ 1 2 ∣ f ( σ − β + i T ) ∣ d σ = ∫ 1 2 ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ ∣ ζ K ( σ + i T ) ∣ y σ − β d σ ≤ C 4 T y 2 − β (74)
where C 4 = C 4 ( q ) > 0 C_4=C_4(q)>0 C 4 = C 4 ( q ) > 0 is a constant.
In the case that − 1 ≤ σ ≤ 0 -1 \le\sigma\le0 − 1 ≤ σ ≤ 0 , we apply the function equation (23) for ζ K ( s ) \zeta_K(s) ζ K ( s ) to obtain
∣ Γ ( 1 − σ − i T ) ∣ ∣ ζ K ( 1 − σ − i T ) ∣ = ( ∣ Δ K ∣ 4 π 2 ) σ − 1 / 2 ∣ Γ ( σ + i T ) ∣ ∣ ζ K ( σ + i T ) ∣ . |\Gamma(1-\sigma-iT)||\zeta_K(1-\sigma-iT)|=\left(\frac{|\Delta_K|}{4\pi^2}\right)^{\sigma-1/2}|\Gamma(\sigma+iT)||\zeta_K(\sigma+iT)|. ∣Γ ( 1 − σ − i T ) ∣∣ ζ K ( 1 − σ − i T ) ∣ = ( 4 π 2 ∣ Δ K ∣ ) σ − 1/2 ∣Γ ( σ + i T ) ∣∣ ζ K ( σ + i T ) ∣.
Thus, we have
∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ = ∫ − 1 0 ∣ ζ K ( σ + i T ) ∣ y σ − β ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ d σ = ∫ − 1 0 ( ∣ Δ K ∣ 4 π 2 ) 1 / 2 − σ ∣ Γ ( 1 − σ − i T ) ∣ ∣ ζ K ( 1 − σ − i T ) ∣ y σ − β ∣ Γ ( σ + i T ) ∣ ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ d σ = ∫ 1 2 ( ∣ Δ K ∣ 4 π 2 ) σ − 1 / 2 ∣ Γ ( σ − i T ) ∣ ∣ ζ K ( σ − i T ) ∣ y 1 − σ − β ∣ Γ ( 1 − σ + i T ) ∣ ∣ 1 − σ − β + i T ∣ ∏ ℓ = 2 9 ∣ 1 − σ − β + i T + ℓ ∣ d σ ≤ ( ∣ Δ K ∣ 4 π 2 ) 3 / 2 ∫ 1 2 ∣ Γ ( σ − i T ) ∣ ∣ ζ K ( σ − i T ) ∣ ∣ Γ ( 1 − σ + i T ) ∣ ∣ 1 − σ − β + i T ∣ ∏ ℓ = 2 9 ∣ 1 − σ − β + i T + ℓ ∣ d σ \begin{aligned}
\int_{-1}^{0}|f(\sigma-\beta+iT)|\mathrm{d}\sigma
&=\int_{-1}^{0}\frac{|\zeta_K(\sigma+iT)|y^{\sigma-\beta}}{|\sigma-\beta+iT|\prod_{\ell=2}^{9}|\sigma-\beta+iT+\ell|}\mathrm{d}\sigma\\
&=\int_{-1}^{0}\left(\frac{|\Delta_K|}{4\pi^2}\right)^{1/2-\sigma}\frac{|\Gamma(1-\sigma-iT)||\zeta_K(1-\sigma-iT)|y^{\sigma-\beta}}{|\Gamma(\sigma+iT)||\sigma-\beta+iT|\prod_{\ell=2}^{9}|\sigma-\beta+iT+\ell|}\mathrm{d}\sigma\\
&=\int_{1}^{2}\left(\frac{|\Delta_K|}{4\pi^2}\right)^{\sigma-1/2}\frac{|\Gamma(\sigma-iT)||\zeta_K(\sigma-iT)|y^{1-\sigma-\beta}}{|\Gamma(1-\sigma+iT)||1-\sigma-\beta+iT|\prod_{\ell=2}^{9}|1-\sigma-\beta+iT+\ell|}\mathrm{d}\sigma\\
&\le\left(\frac{|\Delta_K|}{4\pi^2}\right)^{3/2}\int_{1}^{2}\frac{|\Gamma(\sigma-iT)||\zeta_K(\sigma-iT)|}{|\Gamma(1-\sigma+iT)||1-\sigma-\beta+iT|\prod_{\ell=2}^{9}|1-\sigma-\beta+iT+\ell|}\mathrm{d}\sigma
\end{aligned} ∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ = ∫ − 1 0 ∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ ∣ ζ K ( σ + i T ) ∣ y σ − β d σ = ∫ − 1 0 ( 4 π 2 ∣ Δ K ∣ ) 1/2 − σ ∣Γ ( σ + i T ) ∣∣ σ − β + i T ∣ ∏ ℓ = 2 9 ∣ σ − β + i T + ℓ ∣ ∣Γ ( 1 − σ − i T ) ∣∣ ζ K ( 1 − σ − i T ) ∣ y σ − β d σ = ∫ 1 2 ( 4 π 2 ∣ Δ K ∣ ) σ − 1/2 ∣Γ ( 1 − σ + i T ) ∣∣1 − σ − β + i T ∣ ∏ ℓ = 2 9 ∣1 − σ − β + i T + ℓ ∣ ∣Γ ( σ − i T ) ∣∣ ζ K ( σ − i T ) ∣ y 1 − σ − β d σ ≤ ( 4 π 2 ∣ Δ K ∣ ) 3/2 ∫ 1 2 ∣Γ ( 1 − σ + i T ) ∣∣1 − σ − β + i T ∣ ∏ ℓ = 2 9 ∣1 − σ − β + i T + ℓ ∣ ∣Γ ( σ − i T ) ∣∣ ζ K ( σ − i T ) ∣ d σ
Remind that we have Rademacher’s upper bound
∣ ζ K ( σ − i T ) ∣ ≤ 3 ∣ 1 + σ − i T ∣ ∣ 1 − σ + i T ∣ ( ∣ Δ K ∣ 4 π 2 ∣ 1 + σ − i T ∣ 2 ) 3 / 2 − σ 2 ζ ( 3 / 2 ) 2 |\zeta_K(\sigma-iT)| \le3\frac{|1+\sigma-iT|}{|1-\sigma+iT|}\left(\frac{|\Delta_K|}{4\pi^2}|1+\sigma-iT|^2\right)^{\frac{3/2-\sigma}{2}}\zeta(3/2)^2 ∣ ζ K ( σ − i T ) ∣ ≤ 3 ∣1 − σ + i T ∣ ∣1 + σ − i T ∣ ( 4 π 2 ∣ Δ K ∣ ∣1 + σ − i T ∣ 2 ) 2 3/2 − σ ζ ( 3/2 ) 2
for 1 ≤ σ ≤ 3 / 2 1 \le\sigma\le3/2 1 ≤ σ ≤ 3/2 and the following bound
∣ ζ K ( σ − i T ) ∣ ≤ ζ ( σ ) 2 ≤ ζ ( 3 / 2 ) 2 |\zeta_K(\sigma-iT)| \le\zeta(\sigma)^2 \le\zeta(3/2)^2 ∣ ζ K ( σ − i T ) ∣ ≤ ζ ( σ ) 2 ≤ ζ ( 3/2 ) 2
for 3 / 2 ≤ σ ≤ 2 3/2 \le\sigma\le2 3/2 ≤ σ ≤ 2 .
By using Stirling’s formula for the gamma function Γ ( s ) \Gamma(s) Γ ( s ) , we can obtain (see [34 p. 55] )
∣ Γ ( σ + i t ) ∣ = 2 π e − π 2 ∣ t ∣ ∣ t ∣ σ − 1 / 2 ( 1 + O ( 1 / ∣ t ∣ ) ) |\Gamma(\sigma+it)|=\sqrt{2\pi}e^{-\frac{\pi}{2}|t|}|t|^{\sigma-1/2}(1+O(1/|t|)) ∣Γ ( σ + i t ) ∣ = 2 π e − 2 π ∣ t ∣ ∣ t ∣ σ − 1/2 ( 1 + O ( 1/∣ t ∣ ))
for ∣ t ∣ ≥ 1 |t| \ge1 ∣ t ∣ ≥ 1 and − 5 < σ < 5 -5 < \sigma< 5 − 5 < σ < 5 with an absolute implied constant. Thus, when T ≥ T 0 T \ge T_0 T ≥ T 0 with T 0 T_0 T 0 sufficiently large but fixed, there is
∣ Γ ( σ − i T ) ∣ ∣ Γ ( 1 − σ + i T ) ∣ = ∣ T ∣ 2 σ − 1 ( 1 + O ( 1 / ∣ T ∣ ) ) ≤ C 5 ∣ T ∣ 3 \frac{|\Gamma(\sigma-iT)|}{|\Gamma(1-\sigma+iT)|}=|T|^{2\sigma-1}(1+O(1/|T|)) \le C_5|T|^3 ∣Γ ( 1 − σ + i T ) ∣ ∣Γ ( σ − i T ) ∣ = ∣ T ∣ 2 σ − 1 ( 1 + O ( 1/∣ T ∣ )) ≤ C 5 ∣ T ∣ 3
for 1 ≤ σ ≤ 2 1 \le\sigma\le2 1 ≤ σ ≤ 2 , where C 5 > 0 C_5 > 0 C 5 > 0 is an absolute constant.
Therefore, combining the given upper bounds, we get
∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ ≤ C 6 1 T (75) \int_{-1}^{0}|f(\sigma-\beta+iT)|\,d\sigma\le C_6\frac{1}{T} \tag*{(75)} ∫ − 1 0 ∣ f ( σ − β + i T ) ∣ d σ ≤ C 6 T 1 (75)
for T ≥ T 0 T \ge T_0 T ≥ T 0 , where C 6 = C 6 ( q ) > 0 C_6=C_6(q)>0 C 6 = C 6 ( q ) > 0 is a constant.
Noting (73) , (74) and (75) , it follows from (72) that
∣ ρ ( T ) ∣ ≤ C 3 y 1 1 − β T + C 4 y 2 2 − β T + C 6 1 T |\rho(T)| \le C_3\frac{y_1^{1-\beta}}{T}+C_4\frac{y_2^{2-\beta}}{T}+C_6\frac{1}{T} ∣ ρ ( T ) ∣ ≤ C 3 T y 1 1 − β + C 4 T y 2 2 − β + C 6 T 1
for T ≥ T 0 T \ge T_0 T ≥ T 0 , which implies ∣ ρ ( T ) ∣ → 0 |\rho(T)| \to0 ∣ ρ ( T ) ∣ → 0 as T → ∞ T \to\infty T → ∞ . In a similar fashion, it can be shown that the following integral
∫ − 1 − β − i T 2 − β − i T f ( s ) d s → 0 \int_{-1-\beta-iT}^{2-\beta-iT}f(s)\,ds \to0 ∫ − 1 − β − i T 2 − β − i T f ( s ) d s → 0
when T → ∞ T \to\infty T → ∞ . The proof of Lemma 6 is complete. □
Proof of Lemma 7
In this section, we are going to demonstrate that the matrix M \mathcal{M} M given in Section 6 has full column rank.
Remind that the matrix M \mathcal{M} M is of size T × W 4 T \times W^4 T × W 4 , where W W W is a positive integer with W ≥ q > 3 ⋅ 10 20 W \ge q > 3\cdot10^{20} W ≥ q > 3 ⋅ 1 0 20 and T = ( T 1 + 1 ) ( T 2 + 1 ) ( T 3 + 1 ) T=(T_1+1)(T_2+1)(T_3+1) T = ( T 1 + 1 ) ( T 2 + 1 ) ( T 3 + 1 ) with
H = 10 6 , U = W 4 / 3 , T 1 = ⌈ 8 H 2 / 3 U ⌉ , T 2 = T 3 = ⌈ 8 H − 1 / 3 U ⌉ . H=10^6,\quad U=W^{4/3},\quad T_1=\lceil8H^{2/3}U\rceil,\quad T_2=T_3=\lceil8H^{-1/3}U\rceil. H = 1 0 6 , U = W 4/3 , T 1 = ⌈ 8 H 2/3 U ⌉ , T 2 = T 3 = ⌈ 8 H − 1/3 U ⌉ .
It is clear that the matrix M \mathcal{M} M has more rows than columns due to the following
T ≥ 8 H 2 / 3 U ( 8 H − 1 / 3 U ) 2 = 512 U 3 = 512 W 4 > W 4 . T \ge8H^{2/3}U(8H^{-1/3}U)^2=512U^3=512W^4>W^4. T ≥ 8 H 2/3 U ( 8 H − 1/3 U ) 2 = 512 U 3 = 512 W 4 > W 4 .
We have T j ≥ 8 H − 1 / 3 U = 8 ( W / H ) 1 / 3 W > 10 5 W > 3 ( W − 1 ) T_j \ge8H^{-1/3}U = 8(W/H)^{1/3}W > 10^5W > 3(W-1) T j ≥ 8 H − 1/3 U = 8 ( W / H ) 1/3 W > 1 0 5 W > 3 ( W − 1 ) for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 and
∏ j = 1 3 [ T j − 3 ( W − 1 ) + 1 ] > ∏ j = 1 3 ( T j + 1 − 3 W ) > T − 6 W ( T 1 T 2 + T 2 T 3 + T 3 T 1 ) \prod_{j=1}^{3}[T_j-3(W-1)+1] > \prod_{j=1}^{3}(T_j+1-3W) > T-6W(T_1T_2+T_2T_3+T_3T_1) j = 1 ∏ 3 [ T j − 3 ( W − 1 ) + 1 ] > j = 1 ∏ 3 ( T j + 1 − 3 W ) > T − 6 W ( T 1 T 2 + T 2 T 3 + T 3 T 1 )
and hence, from T 1 > T 2 = T 3 T_1>T_2=T_3 T 1 > T 2 = T 3 ,
∏ j = 1 3 [ T j − 3 ( W − 1 ) + 1 ] > T − 6 W ( 3 T 1 T 2 ) > T − ( 18 / 10 5 ) T 1 T 2 T 3 > 0.99 T > 500 W 4 . \prod_{j=1}^{3}[T_j-3(W-1)+1] > T-6W(3T_1T_2) > T-(18/10^5)T_1T_2T_3 > 0.99T > 500W^4. j = 1 ∏ 3 [ T j − 3 ( W − 1 ) + 1 ] > T − 6 W ( 3 T 1 T 2 ) > T − ( 18/1 0 5 ) T 1 T 2 T 3 > 0.99 T > 500 W 4 .
We put b j = T j − 3 ( W − 1 ) ≥ 1 b_j=T_j-3(W-1)\ge1 b j = T j − 3 ( W − 1 ) ≥ 1 for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 , which satisfies
∏ j = 1 3 ( b j + 1 ) > 500 W 4 > ( 4 W − 3 ) 4 . \prod_{j=1}^{3}(b_j+1)>500W^4>(4W-3)^4. j = 1 ∏ 3 ( b j + 1 ) > 500 W 4 > ( 4 W − 3 ) 4 .
We define the following subset of column indices
E W = { n ⃗ = ( n 1 , n 2 , n 3 , n 4 ) T ∣ 0 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ Z 4 E_W=\{\vec{n}=(n_1,n_2,n_3,n_4)^{\mathsf{T}}\mid0\le n_i\le W-1\text{ for }i=1,2,3,4\}\subset\mathbb{Z}^4 E W = { n = ( n 1 , n 2 , n 3 , n 4 ) T ∣ 0 ≤ n i ≤ W − 1 for i = 1 , 2 , 3 , 4 } ⊂ Z 4
We denote 0 ⃗ = ( 0 , 0 , 0 , 0 ) T ∈ E W \vec{0}=(0,0,0,0)^{\mathsf{T}}\in E_W 0 = ( 0 , 0 , 0 , 0 ) T ∈ E W as the index of the first column in the matrix M \mathcal{M} M .
The row indices are α = ( α 1 , α 2 , α 3 ) ∈ Z 3 \alpha=(\alpha_1,\alpha_2,\alpha_3)\in\mathbb{Z}^3 α = ( α 1 , α 2 , α 3 ) ∈ Z 3 with 0 ≤ α j ≤ T j 0\le\alpha_j\le T_j 0 ≤ α j ≤ T j for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 . Both row and column indices are, respectively, arranged in a lexicographical order, namely, the alphabetical order.
The entry of the matrix M \mathcal{M} M at row α = ( α 1 , α 2 , α 3 ) \alpha=(\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) and column n ⃗ ∈ E W \vec{n}\in E_W n ∈ E W is given by
M α , n ⃗ = θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 (76) \mathcal{M}_{\alpha,\vec{n}}=\theta_{\vec{n}}^{\alpha_1}\sigma(\theta_{\vec{n}})^{\alpha_2}\sigma\tau(\theta_{\vec{n}})^{\alpha_3} \tag*{(76)} M α , n = θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 (76)
with
θ n ⃗ = n 1 + n 2 a + n 3 b + n 4 a b ∈ R = Z [ a , b ] . \theta_{\vec{n}}=n_1+n_2a+n_3b+n_4ab\in R=\mathbb{Z}[a,b]. θ n = n 1 + n 2 a + n 3 b + n 4 ab ∈ R = Z [ a , b ] .
Remind that a = − N a=\sqrt{-N} a = − N and b = 2 b=\sqrt{2} b = 2 . It is agreed that for n ⃗ = 0 ⃗ \vec{n}=\vec{0} n = 0 with θ 0 ⃗ = 0 \theta_{\vec{0}}=0 θ 0 = 0 , M α , n ⃗ \mathcal{M}_{\alpha,\vec{n}} M α , n is defined as 1 1 1 for α = ( 0 , 0 , 0 ) \alpha=(0,0,0) α = ( 0 , 0 , 0 ) and 0 0 0 for α ≠ ( 0 , 0 , 0 ) \alpha\ne(0,0,0) α = ( 0 , 0 , 0 ) . We can see that every entry M α , n ⃗ \mathcal{M}_{\alpha,\vec{n}} M α , n of the matrix M \mathcal{M} M is an algebraic integer in R ⊂ O L R\subset O_L R ⊂ O L .
Let R α R_\alpha R α denote the W 4 W^4 W 4 -dimensional row vector of M \mathcal{M} M with row index α \alpha α and C n ⃗ C_{\vec{n}} C n the T T T -dimensional column vector of M \mathcal{M} M with column index n ⃗ ∈ E W \vec{n}\in E_W n ∈ E W . It is clear that the first row R ( 0 , 0 , 0 ) R_{(0,0,0)} R ( 0 , 0 , 0 ) in M \mathcal{M} M is an all-one vector and the first column C 0 ⃗ C_{\vec{0}} C 0 in M \mathcal{M} M is a standard unit vector with 1 1 1 in the first position.
On the contrary, we assume that the matrix M \mathcal{M} M does not have full column rank. As a consequence, all the column vectors C n ⃗ C_{\vec{n}} C n , n ⃗ ∈ E W \vec{n}\in E_W n ∈ E W , of the matrix M \mathcal{M} M must be linearly dependent in the number field L L L . Thus, there exist numbers v n ⃗ ∈ L v_{\vec{n}}\in L v n ∈ L , not all zeros, such that
∑ n ⃗ ∈ E W v n ⃗ C n ⃗ = 0 ∈ L T (77) \sum_{\vec{n}\in E_W}v_{\vec{n}}C_{\vec{n}}=\boldsymbol{0}\in L^T \tag*{(77)} n ∈ E W ∑ v n C n = 0 ∈ L T (77)
where 0 ∈ L T \mathbf{0} \in L^T 0 ∈ L T is the zero vector. We define the support subset
S = { n ⃗ ∈ E W ∣ v n ⃗ ≠ 0 } ⊆ E W ⊂ Z 4 (78) \mathcal{S}=\{\vec{n}\in E_W\mid v_{\vec{n}}\ne0\}\subseteq E_W\subset\mathbb{Z}^4 \tag*{(78)} S = { n ∈ E W ∣ v n = 0 } ⊆ E W ⊂ Z 4 (78)
which has the cardinality # S ≥ 2 \#\mathcal{S}\ge2 # S ≥ 2 , since every column vector of M \mathcal{M} M is a nonzero vector with 1 in the first position.
By noticing (76), we see from (77) that, for every row index α = ( α 1 , α 2 , α 3 ) \alpha=(\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) of the matrix M \mathcal{M} M ,
∑ n ⃗ ∈ S v n ⃗ θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 = 0. \sum_{\vec{n}\in\mathcal{S}}v_{\vec{n}}\theta_{\vec{n}}^{\alpha_1}\sigma(\theta_{\vec{n}})^{\alpha_2}\sigma_{\tau}(\theta_{\vec{n}})^{\alpha_3}=0. n ∈ S ∑ v n θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 = 0.
Therefore, we have
0 = ∑ α c α ∑ n ⃗ ∈ E W v n ⃗ θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 = ∑ n ⃗ ∈ E W v n ⃗ ∑ α c α θ n ⃗ α 1 σ ( θ n ⃗ ) α 2 σ τ ( θ n ⃗ ) α 3 (79) 0=\sum_{\alpha}c_{\alpha}\sum_{\vec{n}\in E_W}v_{\vec{n}}\theta_{\vec{n}}^{\alpha_1}\sigma(\theta_{\vec{n}})^{\alpha_2}\sigma_{\tau}(\theta_{\vec{n}})^{\alpha_3}=\sum_{\vec{n}\in E_W}v_{\vec{n}}\sum_{\alpha}c_{\alpha}\theta_{\vec{n}}^{\alpha_1}\sigma(\theta_{\vec{n}})^{\alpha_2}\sigma_{\tau}(\theta_{\vec{n}})^{\alpha_3} \tag*{(79)} 0 = α ∑ c α n ∈ E W ∑ v n θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 = n ∈ E W ∑ v n α ∑ c α θ n α 1 σ ( θ n ) α 2 σ τ ( θ n ) α 3 (79)
where the summation ∑ α \sum_{\alpha} ∑ α runs over the row indices α \alpha α of M \mathcal{M} M and c α ∈ C c_{\alpha}\in\mathbb{C} c α ∈ C are arbitrarily given complex numbers. Put the 3 × 4 3\times4 3 × 4 matrix
A = ( 1 a b a b 1 − a b − a b 1 − a − b a b ) (80) A=\begin{pmatrix}
1&a&b&ab\\
1&-a&b&-ab\\
1&-a&-b&ab
\end{pmatrix} \tag*{(80)} A = 1 1 1 a − a − a b b − b ab − ab ab (80)
we get the following equality, for n ⃗ ∈ E W ⊂ Z 4 \vec{n}\in E_W\subset\mathbb{Z}^4 n ∈ E W ⊂ Z 4 ,
A n ⃗ = ( θ n ⃗ , σ ( θ n ⃗ ) , σ τ ( θ n ⃗ ) ) T ∈ R 3 . A\vec{n}=(\theta_{\vec{n}},\sigma(\theta_{\vec{n}}),\sigma_{\tau}(\theta_{\vec{n}}))^{\mathsf{T}}\in\mathbb{R}^3. A n = ( θ n , σ ( θ n ) , σ τ ( θ n ) ) T ∈ R 3 .
We introduce the following arbitrary 3-variate polynomial arising from (79)
B ( z 1 , z 2 , z 3 ) = ∑ α c α z 1 α 1 z 2 α 2 z 3 α 3 = ∑ α 1 = 0 T 1 ∑ α 2 = 0 T 2 ∑ α 3 = 0 T 3 c ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 (81) B(z_1,z_2,z_3)=\sum_{\alpha}c_{\alpha}z_1^{\alpha_1}z_2^{\alpha_2}z_3^{\alpha_3}
=\sum_{\alpha_1=0}^{T_1}\sum_{\alpha_2=0}^{T_2}\sum_{\alpha_3=0}^{T_3}c_{(\alpha_1,\alpha_2,\alpha_3)}z_1^{\alpha_1}z_2^{\alpha_2}z_3^{\alpha_3} \tag*{(81)} B ( z 1 , z 2 , z 3 ) = α ∑ c α z 1 α 1 z 2 α 2 z 3 α 3 = α 1 = 0 ∑ T 1 α 2 = 0 ∑ T 2 α 3 = 0 ∑ T 3 c ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 (81)
with complex variables z 1 z_1 z 1 , z 2 z_2 z 2 and z 3 z_3 z 3 . It is defined that z j 0 = 1 z_j^0=1 z j 0 = 1 for j = 1 , 2 , 3 j=1,2,3 j = 1 , 2 , 3 in the ring of polynomials C [ z 1 , z 2 , z 3 ] \mathbb{C}[z_1,z_2,z_3] C [ z 1 , z 2 , z 3 ] . We call B ( z 1 , z 2 , z 3 ) B(z_1,z_2,z_3) B ( z 1 , z 2 , z 3 ) as a zero polynomial if all coefficients c α c_{\alpha} c α ’s are equal to zero. We have, from (79), that
∑ n ⃗ ∈ E W v n ⃗ B ( A n ⃗ ) = ∑ n ⃗ ∈ S v n ⃗ B ( A n ⃗ ) = 0 \sum_{\vec{n}\in E_W}v_{\vec{n}}B(A\vec{n})=\sum_{\vec{n}\in\mathcal{S}}v_{\vec{n}}B(A\vec{n})=0 n ∈ E W ∑ v n B ( A n ) = n ∈ S ∑ v n B ( A n ) = 0
for every nonzero polynomial B ( z 1 , z 2 , z 3 ) B(z_1,z_2,z_3) B ( z 1 , z 2 , z 3 ) with complex variables z 1 z_1 z 1 , z 2 z_2 z 2 and z 3 z_3 z 3 , given in (81), where ∑ α \sum_{\alpha} ∑ α is a summation over the row indices α = ( α 1 , α 2 , α 3 ) \alpha=(\alpha_1,\alpha_2,\alpha_3) α = ( α 1 , α 2 , α 3 ) of the matrix M \mathcal{M} M . In the following, we will construct a nonzero polynomial
B ^ ( z 1 , z 2 , z 3 ) = ∑ α 1 = 0 T 1 ∑ α 2 = 0 T 2 ∑ α 3 = 0 T 3 c ^ ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 \widehat{B}(z_1,z_2,z_3)=\sum_{\alpha_1=0}^{T_1}\sum_{\alpha_2=0}^{T_2}\sum_{\alpha_3=0}^{T_3}\widehat{c}_{(\alpha_1,\alpha_2,\alpha_3)}z_1^{\alpha_1}z_2^{\alpha_2}z_3^{\alpha_3} B ( z 1 , z 2 , z 3 ) = α 1 = 0 ∑ T 1 α 2 = 0 ∑ T 2 α 3 = 0 ∑ T 3 c ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3
with complex variables z 1 z_1 z 1 , z 2 z_2 z 2 and z 3 z_3 z 3 , whose coefficients c ^ ( α 1 , α 2 , α 3 ) \hat{c}_{(\alpha_1,\alpha_2,\alpha_3)} c ^ ( α 1 , α 2 , α 3 ) ’s are not all zeros, such that
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) ≠ 0 , (82) \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) \ne0, \tag*{(82)} n ∈ S ∑ v n B ^ ( A n ) = 0 , (82)
which gives a contradiction . As a result, the matrix M \mathcal{M} M must have full column rank.
We define
P = conv ( S ) ⊂ [ 0 , W − 1 ] 4 ⊂ R 4 and K = 4 P . \mathcal{P} = \operatorname{conv}(\mathcal{S}) \subset[0,W-1]^4 \subset\mathbb{R}^4 \quad\text{and} \quad\mathcal{K} = 4\mathcal{P}. P = conv ( S ) ⊂ [ 0 , W − 1 ] 4 ⊂ R 4 and K = 4 P .
Remind that b j = T j − 3 ( W − 1 ) ≥ 1 b_j = T_j - 3(W-1) \ge1 b j = T j − 3 ( W − 1 ) ≥ 1 for j = 1 , 2 , 3 j = 1,2,3 j = 1 , 2 , 3 . It is seen that
K ⊂ [ 0 , 4 ( W − 1 ) ] 4 ⊂ R 4 and # ( K ∩ Z 4 ) ≤ ( 4 W − 3 ) 4 < ∏ j = 1 3 ( b j + 1 ) . \mathcal{K} \subset[0,4(W-1)]^4 \subset\mathbb{R}^4 \quad\text{and} \quad\#(\mathcal{K}\cap\mathbb{Z}^4) \le(4W-3)^4 < \prod_{j=1}^{3}(b_j+1). K ⊂ [ 0 , 4 ( W − 1 ) ] 4 ⊂ R 4 and # ( K ∩ Z 4 ) ≤ ( 4 W − 3 ) 4 < j = 1 ∏ 3 ( b j + 1 ) .
Therefore, there exists a nonzero 3-variate polynomial
R ( z 1 , z 2 , z 3 ) = ∑ α 1 = 0 b 1 ∑ α 2 = 0 b 2 ∑ α 3 = 0 b 3 r ^ ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 (83) R(z_1,z_2,z_3)=\sum_{\alpha_1=0}^{b_1}\sum_{\alpha_2=0}^{b_2}\sum_{\alpha_3=0}^{b_3}\hat{r}_{(\alpha_1,\alpha_2,\alpha_3)}z_1^{\alpha_1}z_2^{\alpha_2}z_3^{\alpha_3} \tag*{(83)} R ( z 1 , z 2 , z 3 ) = α 1 = 0 ∑ b 1 α 2 = 0 ∑ b 2 α 3 = 0 ∑ b 3 r ^ ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 (83)
with complex variables z 1 z_1 z 1 , z 2 z_2 z 2 and z 3 z_3 z 3 and r ^ ( α 1 , α 2 , α 3 ) ∈ C \hat{r}_{(\alpha_1,\alpha_2,\alpha_3)}\in\mathbb{C} r ^ ( α 1 , α 2 , α 3 ) ∈ C such that
R ( A k ⃗ ) = 0 for all k ⃗ ∈ K ∩ Z 4 (84) R(A\vec{k})=0 \quad\text{for all } \vec{k}\in\mathcal{K}\cap\mathbb{Z}^4 \tag*{(84)} R ( A k ) = 0 for all k ∈ K ∩ Z 4 (84)
since A k ⃗ ∈ C 3 A\vec{k}\in\mathbb{C}^3 A k ∈ C 3 and # { A k ⃗ ∣ k ⃗ ∈ K ∩ Z 4 } ≤ # ( K ∩ Z 4 ) \#\{A\vec{k}\mid\vec{k}\in\mathcal{K}\cap\mathbb{Z}^4\}\le\#(\mathcal{K}\cap\mathbb{Z}^4) # { A k ∣ k ∈ K ∩ Z 4 } ≤ # ( K ∩ Z 4 ) .
The 3 × 4 3\times4 3 × 4 matrix A A A given by (80) has full row rank. Thus, for any vector z ∈ C 3 z\in\mathbb{C}^3 z ∈ C 3 , there is a vector x ∈ C 4 x\in\mathbb{C}^4 x ∈ C 4 such that A x = z Ax=z A x = z . As a consequence, the 4-variate polynomial R ( A x ) R(Ax) R ( A x ) with complex variables x = ( x 1 , x 2 , x 3 , x 4 ) T x=(x_1,x_2,x_3,x_4)^{\mathsf{T}} x = ( x 1 , x 2 , x 3 , x 4 ) T is nonzero .
By using the fundamental theorem of algebra and the method of mathematical induction, we can readily prove that the nonzero 4-variate polynomial R ( A x ) R(Ax) R ( A x ) cannot vanish on the entire integer lattice Z 4 \mathbb{Z}^4 Z 4 . Noting (84), we see that there exists an m ⃗ ∈ Z 4 \vec{m}\in\mathbb{Z}^4 m ∈ Z 4 but m ⃗ ∉ K \vec{m}\notin\mathcal{K} m ∈ / K such that R ( A m ⃗ ) ≠ 0 R(A\vec{m})\ne0 R ( A m ) = 0 .
It is clear that K \mathcal{K} K is a nonempty, closed and convex subset of R 4 \mathbb{R}^4 R 4 . The projection of a point m ⃗ ∈ R 4 \vec{m}\in\mathbb{R}^4 m ∈ R 4 onto the set K \mathcal{K} K , denoted by Proj K ( m ⃗ ) \operatorname{Proj}_{\mathcal{K}}(\vec{m}) Proj K ( m ) , is defined as the solution (which must exist and be unique) to the mathematical program of min x ∈ K ∥ m ⃗ − x ∥ \min_{x\in\mathcal{K}}\|\vec{m}-x\| min x ∈ K ∥ m − x ∥ , where ∥ x ∥ = x T x \|x\|=\sqrt{x^{\mathsf{T}}x} ∥ x ∥ = x T x denotes the Euclidean norm or 2-norm of the column vector x ∈ R 4 x\in\mathbb{R}^4 x ∈ R 4 . Moreover, we denote the distance between the point m ⃗ \vec{m} m and the closed convex subset K \mathcal{K} K by dist ( m ⃗ , K ) = min x ∈ K ∥ m ⃗ − x ∥ = ∥ m ⃗ − Proj K ( m ⃗ ) ∥ \operatorname{dist}(\vec{m},\mathcal{K})=\min_{x\in\mathcal{K}}\|\vec{m}-x\|=\|\vec{m}-\operatorname{Proj}_{\mathcal{K}}(\vec{m})\| dist ( m , K ) = min x ∈ K ∥ m − x ∥ = ∥ m − Proj K ( m ) ∥ . For x ∈ R 4 x\in\mathbb{R}^4 x ∈ R 4 and y ∈ R 4 y\in\mathbb{R}^4 y ∈ R 4 , the inner product of x x x and y y y is given by
⟨ x , y ⟩ = x T y \langle x,y\rangle=x^{\mathsf{T}}y ⟨ x , y ⟩ = x T y
which implies that ⟨ x , y ⟩ = ⟨ y , x ⟩ \langle x,y\rangle=\langle y,x\rangle ⟨ x , y ⟩ = ⟨ y , x ⟩ and ⟨ x , x ⟩ = ∥ x ∥ 2 \langle x,x\rangle=\|x\|^2 ⟨ x , x ⟩ = ∥ x ∥ 2 . The projection operator Proj K ( ⋅ ) \operatorname{Proj}_{\mathcal{K}}(\cdot) Proj K ( ⋅ ) has the following property (see, e.g., [25 Lemma 1] ). If x ∈ K x\in\mathcal{K} x ∈ K , then, for any m ⃗ ∈ R 4 \vec{m}\in\mathbb{R}^4 m ∈ R 4 ,
⟨ m ⃗ − Proj K ( m ⃗ ) , x − Proj K ( m ⃗ ) ⟩ ≤ 0. \left\langle\vec{m}-\operatorname{Proj}_{\mathcal{K}}(\vec{m}),x-\operatorname{Proj}_{\mathcal{K}}(\vec{m})\right\rangle\le0. ⟨ m − Proj K ( m ) , x − Proj K ( m ) ⟩ ≤ 0.
We can choose a specific element from the set of all m ⃗ ∈ Z 4 ∖ K \vec{m} \in\mathbb{Z}^{4} \setminus\mathcal{K} m ∈ Z 4 ∖ K satisfying
R ( A m ⃗ ) ≠ 0 R(A\vec{m}) \ne0 R ( A m ) = 0
which has the shortest distance dist ( m ⃗ , K ) \operatorname{dist}(\vec{m},\mathcal{K}) dist ( m , K ) from K \mathcal{K} K , which is still denoted by m ⃗ ∈ Z 4 \vec{m} \in\mathbb{Z}^{4} m ∈ Z 4 . We define two points p ⃗ ∈ R 4 \vec{p} \in\mathbb{R}^{4} p ∈ R 4 and h ⃗ ∈ R 4 \vec{h} \in\mathbb{R}^{4} h ∈ R 4 as follows
p ⃗ = Proj K ( m ⃗ ) and h ⃗ = m ⃗ − p ⃗ (85) \vec{p} = \operatorname{Proj}_{\mathcal{K}}(\vec{m}) \text{ and } \vec{h} = \vec{m} - \vec{p} \tag*{(85)} p = Proj K ( m ) and h = m − p (85)
which has the property that, for all x ∈ K = 4 P ⊂ R 4 x \in\mathcal{K} = 4\mathcal{P} \subset\mathbb{R}^{4} x ∈ K = 4 P ⊂ R 4 ,
⟨ h ⃗ , x − p ⃗ ⟩ = ⟨ m ⃗ − p ⃗ , x − p ⃗ ⟩ ≤ 0. \langle\vec{h},x-\vec{p}\rangle= \langle\vec{m}-\vec{p},x-\vec{p}\rangle\le0. ⟨ h , x − p ⟩ = ⟨ m − p , x − p ⟩ ≤ 0.
For all v ⃗ ∈ P ⊂ R 4 \vec{v} \in\mathcal{P} \subset\mathbb{R}^{4} v ∈ P ⊂ R 4 , we have 4 v ⃗ ∈ K = 4 P 4\vec{v} \in\mathcal{K} = 4\mathcal{P} 4 v ∈ K = 4 P and therefore ⟨ h ⃗ , 4 v ⃗ − p ⃗ ⟩ ≤ 0 \langle\vec{h},4\vec{v}-\vec{p}\rangle\le0 ⟨ h , 4 v − p ⟩ ≤ 0 . Hence,
⟨ h ⃗ , v ⃗ ⟩ ≤ ⟨ h ⃗ , p ⃗ / 4 ⟩ (86) \langle\vec{h},\vec{v}\rangle\le\langle\vec{h},\vec{p}/4\rangle \tag*{(86)} ⟨ h , v ⟩ ≤ ⟨ h , p /4 ⟩ (86)
where p ⃗ / 4 ∈ P \vec{p}/4 \in\mathcal{P} p /4 ∈ P due to p ⃗ ∈ K \vec{p} \in\mathcal{K} p ∈ K . We introduce the following supporting face
G = { v ⃗ ∈ P ∣ ⟨ h ⃗ , v ⃗ ⟩ = max w ⃗ ∈ P ⟨ h ⃗ , w ⃗ ⟩ } ⊂ R 4 (87) \mathcal{G} = \{\vec{v} \in\mathcal{P} \mid\langle\vec{h},\vec{v}\rangle= \max_{\vec{w}\in\mathcal{P}}\langle\vec{h},\vec{w}\rangle\} \subset\mathbb{R}^{4} \tag*{(87)} G = { v ∈ P ∣ ⟨ h , v ⟩ = w ∈ P max ⟨ h , w ⟩} ⊂ R 4 (87)
which has dimensions at most three . It is clear from (86) that p ⃗ / 4 ∈ G \vec{p}/4 \in\mathcal{G} p /4 ∈ G .
According to Carathéodory’s convexity theorem [4 ] , if the convex hull of a set of points has ℓ \ell ℓ dimensions, then any point in the convex hull can be written as a convex combination of at most ℓ + 1 \ell+ 1 ℓ + 1 vertices.
Since p ⃗ / 4 ∈ P ⊂ R 4 \vec{p}/4 \in\mathcal{P} \subset\mathbb{R}^{4} p /4 ∈ P ⊂ R 4 , there exist k k k vertices v ⃗ 1 , ⋯ , v ⃗ k ∈ S ⊂ P = conv ( S ) \vec{v}_{1}, \cdots, \vec{v}_{k} \in\mathcal{S} \subset\mathcal{P} = \operatorname{conv}(\mathcal{S}) v 1 , ⋯ , v k ∈ S ⊂ P = conv ( S ) with 1 ≤ k ≤ 5 1 \le k \le5 1 ≤ k ≤ 5 such that
p ⃗ / 4 = ∑ ℓ = 1 k λ ℓ v ⃗ ℓ \vec{p}/4 = \sum_{\ell=1}^{k}\lambda_{\ell}\vec{v}_{\ell} p /4 = ℓ = 1 ∑ k λ ℓ v ℓ
where λ ℓ > 0 \lambda_{\ell} > 0 λ ℓ > 0 for 1 ≤ ℓ ≤ k 1 \le\ell\le k 1 ≤ ℓ ≤ k and ∑ ℓ = 1 k λ ℓ = 1 \sum_{\ell=1}^{k}\lambda_{\ell} = 1 ∑ ℓ = 1 k λ ℓ = 1 . We can prove that it is enough for k k k to be at most four. In fact, if k = 5 k = 5 k = 5 , then, by using (86) ,
⟨ h ⃗ , p ⃗ / 4 ⟩ = ⟨ h ⃗ , ∑ ℓ = 1 5 λ ℓ v ⃗ ℓ ⟩ = ∑ ℓ = 1 5 λ ℓ ⟨ h ⃗ , v ⃗ ℓ ⟩ ≤ ∑ ℓ = 1 5 λ ℓ ⟨ h ⃗ , p ⃗ / 4 ⟩ = ⟨ h ⃗ , p ⃗ / 4 ⟩ \langle\vec{h},\vec{p}/4\rangle= \left\langle\vec{h},\sum_{\ell=1}^{5}\lambda_{\ell}\vec{v}_{\ell}\right\rangle= \sum_{\ell=1}^{5}\lambda_{\ell}\langle\vec{h},\vec{v}_{\ell}\rangle\le\sum_{\ell=1}^{5}\lambda_{\ell}\langle\vec{h},\vec{p}/4\rangle= \langle\vec{h},\vec{p}/4\rangle ⟨ h , p /4 ⟩ = ⟨ h , ℓ = 1 ∑ 5 λ ℓ v ℓ ⟩ = ℓ = 1 ∑ 5 λ ℓ ⟨ h , v ℓ ⟩ ≤ ℓ = 1 ∑ 5 λ ℓ ⟨ h , p /4 ⟩ = ⟨ h , p /4 ⟩
which implies that, for 1 ≤ ℓ ≤ 5 1 \le\ell\le5 1 ≤ ℓ ≤ 5 ,
⟨ h ⃗ , v ⃗ ℓ ⟩ = ⟨ h ⃗ , p ⃗ / 4 ⟩ . \langle\vec{h},\vec{v}_{\ell}\rangle= \langle\vec{h},\vec{p}/4\rangle. ⟨ h , v ℓ ⟩ = ⟨ h , p /4 ⟩ .
Therefore, by (87) , we have v ⃗ ℓ ∈ G \vec{v}_{\ell} \in\mathcal{G} v ℓ ∈ G for 1 ≤ ℓ ≤ 5 1 \le\ell\le5 1 ≤ ℓ ≤ 5 . It follows that
p ⃗ / 4 ∈ conv ( { v ⃗ 1 , v ⃗ 2 , v ⃗ 3 , v ⃗ 4 , v ⃗ 5 } ) ⊂ G \vec{p}/4 \in\operatorname{conv}(\{\vec{v}_{1},\vec{v}_{2},\vec{v}_{3},\vec{v}_{4},\vec{v}_{5}\}) \subset\mathcal{G} p /4 ∈ conv ({ v 1 , v 2 , v 3 , v 4 , v 5 }) ⊂ G
which has dimensions at most three. By using Carathéodory’s theorem again, p ⃗ / 4 \vec{p}/4 p /4 can be written as a convex combination of at most four vertices in { v ⃗ 1 , v ⃗ 2 , v ⃗ 3 , v ⃗ 4 , v ⃗ 5 } \{\vec{v}_1,\vec{v}_2,\vec{v}_3,\vec{v}_4,\vec{v}_5\} { v 1 , v 2 , v 3 , v 4 , v 5 } . Therefore, combining the above results, there exist k k k vertices u ⃗ 1 , … , u ⃗ k ∈ S ∩ G \vec{u}_1,\ldots,\vec{u}_k \in S \cap G u 1 , … , u k ∈ S ∩ G with 1 ≤ k ≤ 4 1 \le k \le4 1 ≤ k ≤ 4 such that
p ⃗ / 4 = ∑ ℓ = 1 k λ ℓ u ⃗ ℓ \vec{p}/4 = \sum_{\ell=1}^{k} \lambda_{\ell}\vec{u}_{\ell} p /4 = ℓ = 1 ∑ k λ ℓ u ℓ
where λ 1 ≥ ⋯ ≥ λ k > 0 \lambda_1 \ge\cdots\ge\lambda_k > 0 λ 1 ≥ ⋯ ≥ λ k > 0 and ∑ ℓ = 1 k λ ℓ = 1 \sum_{\ell=1}^{k}\lambda_{\ell} = 1 ∑ ℓ = 1 k λ ℓ = 1 . Thus, there must be λ 1 ≥ 1 / 4 \lambda_1 \ge1/4 λ 1 ≥ 1/4 . We define the point
u ⃗ = u ⃗ 1 ∈ S ∩ G (88) \vec{u} = \vec{u}_1 \in S \cap G \tag*{(88)} u = u 1 ∈ S ∩ G (88)
which has the property of
p ⃗ − u ⃗ = 4 ∑ ℓ = 1 k λ ℓ u ⃗ ℓ − u ⃗ = ( 4 λ 1 − 1 ) u ⃗ 1 + ∑ ℓ = 2 k 4 λ ℓ u ⃗ ℓ ∈ 3 P . \vec{p} - \vec{u} = 4\sum_{\ell=1}^{k}\lambda_{\ell}\vec{u}_{\ell} - \vec{u} = (4\lambda_1 - 1)\vec{u}_1 + \sum_{\ell=2}^{k}4\lambda_{\ell}\vec{u}_{\ell} \in3P. p − u = 4 ℓ = 1 ∑ k λ ℓ u ℓ − u = ( 4 λ 1 − 1 ) u 1 + ℓ = 2 ∑ k 4 λ ℓ u ℓ ∈ 3 P .
In the case that S ∖ G ≠ ∅ S \setminus G \ne\varnothing S ∖ G = ∅ , for every n ⃗ ∈ S ∖ G ⊂ P \vec{n} \in S \setminus G \subset P n ∈ S ∖ G ⊂ P , by noting (87) we have
⟨ h ⃗ , u ⃗ − n ⃗ ⟩ = ⟨ h ⃗ , u ⃗ ⟩ − ⟨ h ⃗ , n ⃗ ⟩ > 0. \langle\vec{h},\vec{u}-\vec{n}\rangle= \langle\vec{h},\vec{u}\rangle- \langle\vec{h},\vec{n}\rangle> 0. ⟨ h , u − n ⟩ = ⟨ h , u ⟩ − ⟨ h , n ⟩ > 0.
We define two points p ⃗ 1 \vec{p}_1 p 1 and m ⃗ 1 \vec{m}_1 m 1 by
p ⃗ 1 = p ⃗ − u ⃗ + n ⃗ ∈ K = 4 P and m ⃗ 1 = m ⃗ − u ⃗ + n ⃗ ∈ Z 4 . \vec{p}_1 = \vec{p} - \vec{u} + \vec{n} \in K = 4P \quad\text{and}\quad\vec{m}_1 = \vec{m} - \vec{u} + \vec{n} \in\mathbb{Z}^4. p 1 = p − u + n ∈ K = 4 P and m 1 = m − u + n ∈ Z 4 .
It is clear that m ⃗ 1 = p ⃗ 1 + h ⃗ \vec{m}_1 = \vec{p}_1 + \vec{h} m 1 = p 1 + h , where h ⃗ = m ⃗ − p ⃗ = m ⃗ − Proj K ( m ⃗ ) \vec{h} = \vec{m} - \vec{p} = \vec{m} - \operatorname{Proj}_K(\vec{m}) h = m − p = m − Proj K ( m ) . For any 0 < τ < 1 0 < \tau< 1 0 < τ < 1 , we have τ p ⃗ + ( 1 − τ ) p ⃗ 1 ∈ K \tau\vec{p} + (1-\tau)\vec{p}_1 \in K τ p + ( 1 − τ ) p 1 ∈ K and
∥ m ⃗ 1 − [ τ p ⃗ + ( 1 − τ ) p ⃗ 1 ] ∥ 2 = ∥ h ⃗ − τ ( u ⃗ − n ⃗ ) ∥ 2 = ∥ h ⃗ ∥ 2 − 2 τ ⟨ h ⃗ , u ⃗ − n ⃗ ⟩ + τ 2 ∥ u ⃗ − n ⃗ ∥ 2 . \left\|\vec{m}_1 - [\tau\vec{p} + (1-\tau)\vec{p}_1]\right\|^2 = \left\|\vec{h} - \tau(\vec{u}-\vec{n})\right\|^2 = \|\vec{h}\|^2 - 2\tau\langle\vec{h},\vec{u}-\vec{n}\rangle+ \tau^2\|\vec{u}-\vec{n}\|^2. ∥ m 1 − [ τ p + ( 1 − τ ) p 1 ] ∥ 2 = h − τ ( u − n ) 2 = ∥ h ∥ 2 − 2 τ ⟨ h , u − n ⟩ + τ 2 ∥ u − n ∥ 2 .
Let η = 2 ⟨ h ⃗ , u ⃗ − n ⃗ ⟩ / ∥ u ⃗ − n ⃗ ∥ 2 > 0 \eta= 2\langle\vec{h},\vec{u}-\vec{n}\rangle/\|\vec{u}-\vec{n}\|^2 > 0 η = 2 ⟨ h , u − n ⟩ /∥ u − n ∥ 2 > 0 . Then, for any 0 < τ < min { 1 , η } 0 < \tau< \min\{1,\eta\} 0 < τ < min { 1 , η } , we get
( dist ( m ⃗ 1 , K ) ) 2 ≤ ∥ m ⃗ 1 − [ τ p ⃗ + ( 1 − τ ) p ⃗ 1 ] ∥ 2 < ∥ h ⃗ ∥ 2 . \left(\operatorname{dist}(\vec{m}_1,K)\right)^2 \le\left\|\vec{m}_1 - [\tau\vec{p} + (1-\tau)\vec{p}_1]\right\|^2 < \|\vec{h}\|^2. ( dist ( m 1 , K ) ) 2 ≤ ∥ m 1 − [ τ p + ( 1 − τ ) p 1 ] ∥ 2 < ∥ h ∥ 2 .
Hence, dist ( m ⃗ 1 , K ) < ∥ h ⃗ ∥ = ∥ m ⃗ − Proj K ( m ⃗ ) ∥ = dist ( m ⃗ , K ) \operatorname{dist}(\vec{m}_1,K) < \|\vec{h}\| = \|\vec{m} - \operatorname{Proj}_K(\vec{m})\| = \operatorname{dist}(\vec{m},K) dist ( m 1 , K ) < ∥ h ∥ = ∥ m − Proj K ( m ) ∥ = dist ( m , K ) . According to the property of the chosen point m ⃗ \vec{m} m , we obtain
R ( A n ⃗ + A ( m ⃗ − u ⃗ ) ) = R ( A m ⃗ 1 ) = 0 for n ⃗ ∈ S ∖ G . (89) R(A\vec{n} + A(\vec{m}-\vec{u})) = R(A\vec{m}_1) = 0 \quad\text{for}\quad\vec{n} \in S \setminus G. \tag*{(89)} R ( A n + A ( m − u )) = R ( A m 1 ) = 0 for n ∈ S ∖ G . (89)
Remind that # S ≥ 2 \#S \ge2 # S ≥ 2 and the point u ⃗ ∈ S ∩ G \vec{u} \in S \cap G u ∈ S ∩ G . It is clear that
S = ( S ∩ G ) ∪ ( S ∖ G ) . S = (S \cap G) \cup(S \setminus G). S = ( S ∩ G ) ∪ ( S ∖ G ) .
In the case that # ( S ∩ G ) ≥ 2 \#(S \cap G) \ge2 # ( S ∩ G ) ≥ 2 , there exists at least one n ⃗ ∈ S ∩ G \vec{n} \in S \cap G n ∈ S ∩ G with n ⃗ ≠ 0 ⃗ \vec{n} \ne\vec{0} n = 0 . Every n ⃗ ∈ S ∩ G ⊂ Z 4 \vec{n} \in S \cap G \subset\mathbb{Z}^4 n ∈ S ∩ G ⊂ Z 4 locates in the affine space G ⊂ R 4 G \subset\mathbb{R}^4 G ⊂ R 4 of dimensions at most three.
By using Gaussian elimination, the data-fitting linear equations arising from the integer coordinates of all n ⃗ ∈ S ∩ G \vec{n} \in\mathcal{S} \cap\mathcal{G} n ∈ S ∩ G can be solved in the rational field Q \mathbb{Q} Q via exact arithmetic. Therefore, we can find four rational numbers r ~ 1 \tilde{r}_1 r ~ 1 , r ~ 2 \tilde{r}_2 r ~ 2 , r ~ 3 \tilde{r}_3 r ~ 3 and r ~ 4 \tilde{r}_4 r ~ 4 , which are not all zeros, and the fifth number k ~ ∈ Q \tilde{k} \in\mathbb{Q} k ~ ∈ Q such that every n ⃗ = ( n 1 , n 2 , n 3 , n 4 ) T ∈ S ∩ G \vec{n} = (n_1,n_2,n_3,n_4)^{\mathsf{T}} \in\mathcal{S} \cap\mathcal{G} n = ( n 1 , n 2 , n 3 , n 4 ) T ∈ S ∩ G satisfies the following affine equation
r ~ 1 n 1 + r ~ 2 n 2 + r ~ 3 n 3 + r ~ 4 n 4 = k ~ . \tilde{r}_1n_1+\tilde{r}_2n_2+\tilde{r}_3n_3+\tilde{r}_4n_4=\tilde{k}. r ~ 1 n 1 + r ~ 2 n 2 + r ~ 3 n 3 + r ~ 4 n 4 = k ~ .
Multiplying both sides of the equation by the largest denominator of the nonzero numbers in { r ~ 1 , r ~ 2 , r ~ 3 , r ~ 4 , k ~ } ⊂ Q \{\tilde{r}_1,\tilde{r}_2,\tilde{r}_3,\tilde{r}_4,\tilde{k}\} \subset\mathbb{Q} { r ~ 1 , r ~ 2 , r ~ 3 , r ~ 4 , k ~ } ⊂ Q , the above equation is equivalent to
r 1 n 1 + r 2 n 2 + r 3 n 3 + r 4 n 4 = k r_1n_1+r_2n_2+r_3n_3+r_4n_4=k r 1 n 1 + r 2 n 2 + r 3 n 3 + r 4 n 4 = k
where r 1 , r 2 , r 3 , r 4 , k ∈ Z r_1,r_2,r_3,r_4,k \in\mathbb{Z} r 1 , r 2 , r 3 , r 4 , k ∈ Z with r 1 2 + r 2 2 + r 3 2 + r 4 2 > 0 r_1^2+r_2^2+r_3^2+r_4^2>0 r 1 2 + r 2 2 + r 3 2 + r 4 2 > 0 . We can define the following affine hyperplane
H x = { x ∈ C 4 ∣ r 1 x 1 + r 2 x 2 + r 3 x 3 + r 4 x 4 = k } . (90) \mathcal{H}_x=\{x\in\mathbb{C}^4\mid r_1x_1+r_2x_2+r_3x_3+r_4x_4=k\}. \tag*{(90)} H x = { x ∈ C 4 ∣ r 1 x 1 + r 2 x 2 + r 3 x 3 + r 4 x 4 = k } . (90)
where x = ( x 1 , x 2 , x 3 , x 4 ) T x=(x_1,x_2,x_3,x_4)^{\mathsf{T}} x = ( x 1 , x 2 , x 3 , x 4 ) T is a column vector. Therefore,
S ∩ G ⊂ H x . \mathcal{S}\cap\mathcal{G}\subset\mathcal{H}_x. S ∩ G ⊂ H x .
The 3 × 4 3\times4 3 × 4 matrix A A A given by (80) has the following four 3 × 3 3\times3 3 × 3 submatrices by deleting one of the four columns of the matrix
( a b a b − a b − a b − a − b a b ) , ( 1 b a b 1 b − a b 1 − b a b ) , ( 1 a a b 1 − a − a b 1 − a a b ) and ( 1 a b 1 − a b 1 − a − b ) \begin{pmatrix}
a & b & ab\\
-a & b & -ab\\
-a & -b & ab
\end{pmatrix}
,\quad
\begin{pmatrix}
1 & b & ab\\
1 & b & -ab\\
1 & -b & ab
\end{pmatrix}
,\quad
\begin{pmatrix}
1 & a & ab\\
1 & -a & -ab\\
1 & -a & ab
\end{pmatrix}
\quad\text{and}\quad
\begin{pmatrix}
1 & a & b\\
1 & -a & b\\
1 & -a & -b
\end{pmatrix} a − a − a b b − b ab − ab ab , 1 1 1 b b − b ab − ab ab , 1 1 1 a − a − a ab − ab ab and 1 1 1 a − a − a b b − b
which have nonzero determinants of 4 a 2 b 2 4a^2b^2 4 a 2 b 2 , − 4 a b 2 -4ab^2 − 4 a b 2 , − 4 a 2 b -4a^2b − 4 a 2 b and 4 a b 4ab 4 ab , respectively. Without loss of generality, we can assume that
and will work with the fourth 3 × 3 3\times3 3 × 3 submatrix. Note that if r 4 = 0 r_4=0 r 4 = 0 but r j ≠ 0 r_j\ne0 r j = 0 for some j ∈ { 1 , 2 , 3 } j\in\{1,2,3\} j ∈ { 1 , 2 , 3 } , we can work with the j j j th 3 × 3 3\times3 3 × 3 submatrix in a similar fashion. We introduce the following matrix and its inverse
Γ = ( 1 0 0 ∣ 0 0 1 0 ∣ 0 0 0 1 ∣ 0 r 1 r 2 r 3 ∣ r 4 ) and Γ − 1 = ( 1 0 0 ∣ 0 0 1 0 ∣ 0 0 0 1 ∣ 0 − r 1 / r 4 − r 2 / r 4 − r 3 / r 4 ∣ 1 / r 4 ) . (91) \Gamma=
\begin{pmatrix}
1&0&0&\mid&0\\
0&1&0&\mid&0\\
0&0&1&\mid&0\\
\hline
r_1&r_2&r_3&\mid&r_4
\end{pmatrix}
\quad\text{and}\quad
\Gamma^{-1}=
\begin{pmatrix}
1&0&0&\mid&0\\
0&1&0&\mid&0\\
0&0&1&\mid&0\\
\hline
-r_1/r_4&-r_2/r_4&-r_3/r_4&\mid&1/r_4
\end{pmatrix}. \tag*{(91)} Γ = 1 0 0 r 1 0 1 0 r 2 0 0 1 r 3 ∣ ∣ ∣ ∣ 0 0 0 r 4 and Γ − 1 = 1 0 0 − r 1 / r 4 0 1 0 − r 2 / r 4 0 0 1 − r 3 / r 4 ∣ ∣ ∣ ∣ 0 0 0 1/ r 4 . (91)
Then, under the invertible linear transformation of y = Γ x y = \Gamma x y = Γ x , the image of the affine hyperplane H x \mathcal{H}_x H x given by (90) becomes the following affine hyperplane
H y = { y ∈ C 4 ∣ y 4 = k } (92) \mathcal{H}_y = \{y \in\mathbb{C}^{4} \mid y_{4} = k\} \tag*{(92)} H y = { y ∈ C 4 ∣ y 4 = k } (92)
where y = ( y 1 , y 2 , y 3 , y 4 ) T y = (y_{1}, y_{2}, y_{3}, y_{4})^{\mathsf{T}} y = ( y 1 , y 2 , y 3 , y 4 ) T is a column vector. We can define the following linear transformation
z = A Γ − 1 y z = A\Gamma^{-1}y z = A Γ − 1 y
where z = ( z 1 , z 2 , z 3 ) T z = (z_{1}, z_{2}, z_{3})^{\mathsf{T}} z = ( z 1 , z 2 , z 3 ) T is a column vector and A Γ − 1 A\Gamma^{-1} A Γ − 1 is a 3 × 4 3 \times4 3 × 4 matrix with full row rank.
It can be computed that
z = C ( y 1 , y 2 , y 3 ) T + k r 4 ( a b , − a b , a b ) T z = C(y_{1}, y_{2}, y_{3})^{\mathsf{T}} + \frac{k}{r_{4}}(ab,-ab,ab)^{\mathsf{T}} z = C ( y 1 , y 2 , y 3 ) T + r 4 k ( ab , − ab , ab ) T
where C C C is a 3 × 3 3 \times3 3 × 3 matrix given by
C = ( 1 a b 1 − a b 1 − a − b ) + ( a b − a b a b ) ( − r 1 / r 4 , − r 2 / r 4 , − r 3 / r 4 ) . C =
\begin{pmatrix}
1 & a & b \\
1 & -a & b \\
1 & -a & -b
\end{pmatrix}
+
\begin{pmatrix}
ab \\
-ab \\
ab
\end{pmatrix}
(-r_{1}/r_{4},-r_{2}/r_{4},-r_{3}/r_{4}). C = 1 1 1 a − a − a b b − b + ab − ab ab ( − r 1 / r 4 , − r 2 / r 4 , − r 3 / r 4 ) .
It is easily calculated that
1 + ( − r 1 / r 4 , − r 2 / r 4 , − r 3 / r 4 ) ( 1 a b 1 − a b 1 − a − b ) − 1 ( a b − a b a b ) = 1 + r 3 r 4 a − r 2 r 4 b − r 1 r 4 a b 1 + (-r_{1}/r_{4},-r_{2}/r_{4},-r_{3}/r_{4})
\begin{pmatrix}
1 & a & b \\
1 & -a & b \\
1 & -a & -b
\end{pmatrix}^{-1}
\begin{pmatrix}
ab \\
-ab \\
ab
\end{pmatrix}
= 1 + \frac{r_{3}}{r_{4}}a - \frac{r_{2}}{r_{4}}b - \frac{r_{1}}{r_{4}}ab 1 + ( − r 1 / r 4 , − r 2 / r 4 , − r 3 / r 4 ) 1 1 1 a − a − a b b − b − 1 ab − ab ab = 1 + r 4 r 3 a − r 4 r 2 b − r 4 r 1 ab
which must be nonzero since { 1 , a , b , a b } \{1, a, b, ab\} { 1 , a , b , ab } is the Q \mathbb{Q} Q -basis of the extension L / Q L/\mathbb{Q} L / Q of degree [ L : Q ] = 4 [L : \mathbb{Q}] = 4 [ L : Q ] = 4 . By using the Sherman–Morrison formula [38 ] , we see that the 3 × 3 3 \times3 3 × 3 matrix C C C is invertible and its inverse C − 1 C^{-1} C − 1 can be given explicitly. Therefore, we obtain
( y 1 , y 2 , y 3 ) T = C − 1 ( z 1 , z 2 , z 3 ) T − k r 4 C − 1 ( a b , − a b , a b ) T . (y_{1},y_{2},y_{3})^{\mathsf{T}} = C^{-1}(z_{1},z_{2},z_{3})^{\mathsf{T}} - \frac{k}{r_{4}}C^{-1}(ab,-ab,ab)^{\mathsf{T}}. ( y 1 , y 2 , y 3 ) T = C − 1 ( z 1 , z 2 , z 3 ) T − r 4 k C − 1 ( ab , − ab , ab ) T .
Hence, we can define the following three linear polynomials
L j ( z ) = y j = c j 1 z 1 + c j 2 z 2 + c j 3 z 3 + d j for j = 1 , 2 , 3 (93) L_{j}(z) = y_{j} = c_{j1}z_{1} + c_{j2}z_{2} + c_{j3}z_{3} + d_{j} \text{ for } j = 1,2,3 \tag*{(93)} L j ( z ) = y j = c j 1 z 1 + c j 2 z 2 + c j 3 z 3 + d j for j = 1 , 2 , 3 (93)
where the numbers c j 1 , c j 2 , c j 3 , d j ∈ L c_{j1}, c_{j2}, c_{j3}, d_{j} \in L c j 1 , c j 2 , c j 3 , d j ∈ L for j = 1 , 2 , 3 j = 1,2,3 j = 1 , 2 , 3 .
It is clear that the invertible linear transformation y = Γ x y = \Gamma x y = Γ x implies that y j = x j y_{j} = x_{j} y j = x j for j = 1 , 2 , 3 j = 1,2,3 j = 1 , 2 , 3 and that z = A Γ − 1 y = A x z = A\Gamma^{-1}y = Ax z = A Γ − 1 y = A x . By (93), the linear polynomials L j ( z ) L_{j}(z) L j ( z ) ’s have the property that for all x = ( x 1 , x 2 , x 3 , x 4 ) T ∈ H x x = (x_{1},x_{2},x_{3},x_{4})^{\mathsf{T}} \in\mathcal{H}_{x} x = ( x 1 , x 2 , x 3 , x 4 ) T ∈ H x ,
L j ( A x ) = L j ( z ) = y j = x j for j = 1 , 2 , 3. L_{j}(Ax) = L_{j}(z) = y_{j} = x_{j} \text{ for } j = 1,2,3. L j ( A x ) = L j ( z ) = y j = x j for j = 1 , 2 , 3.
In particular, every n ⃗ = ( n 1 , n 2 , n 3 , n 4 ) ⊤ ∈ S ∩ G ⊂ H x \vec{n} = (n_{1}, n_{2}, n_{3}, n_{4})^{\top} \in\mathcal{S} \cap\mathcal{G} \subset\mathcal{H}_{x} n = ( n 1 , n 2 , n 3 , n 4 ) ⊤ ∈ S ∩ G ⊂ H x satisfies
L j ( A n ⃗ ) = n j for j = 1 , 2 , 3. L_{j}(A\vec{n}) = n_{j}\ \text{for } j = 1,2,3. L j ( A n ) = n j for j = 1 , 2 , 3.
Remind that u ⃗ = ( u 1 , u 2 , u 3 , u 4 ) ⊤ ∈ S ∩ G ⊂ H x \vec{u} = (u_{1}, u_{2}, u_{3}, u_{4})^{\top} \in\mathcal{S} \cap\mathcal{G} \subset\mathcal{H}_{x} u = ( u 1 , u 2 , u 3 , u 4 ) ⊤ ∈ S ∩ G ⊂ H x is given by (88) . If n ⃗ ∈ S ∩ G \vec{n} \in\mathcal{S} \cap\mathcal{G} n ∈ S ∩ G but n ⃗ ≠ u ⃗ \vec{n} \ne\vec{u} n = u , then
( n 1 , n 2 , n 3 ) ≠ ( u 1 , u 2 , u 3 ) . (n_{1},n_{2},n_{3}) \ne(u_{1},u_{2},u_{3}). ( n 1 , n 2 , n 3 ) = ( u 1 , u 2 , u 3 ) .
Otherwise, by using (90) and r 4 ≠ 0 r_{4} \ne0 r 4 = 0 , we have n 4 = u 4 n_{4} = u_{4} n 4 = u 4 and hence n ⃗ = u ⃗ \vec{n} = \vec{u} n = u .
We set the Lagrange-type interpolating polynomial of z 1 z_{1} z 1 , z 2 z_{2} z 2 and z 3 z_{3} z 3 as follows
Q ( z ) = ∏ 1 ≤ j ≤ 4 j ≠ 4 ∏ 0 ≤ a ≤ W − 1 a ≠ u j L j ( z ) − a u j − a Q(z) = \prod_{\substack{1 \le j \le4 \\ j \ne4}} \prod_{\substack{0 \le a \le W-1 \\ a \ne u_{j}}} \frac{L_{j}(z)-a}{u_{j}-a} Q ( z ) = 1 ≤ j ≤ 4 j = 4 ∏ 0 ≤ a ≤ W − 1 a = u j ∏ u j − a L j ( z ) − a
where u ⃗ = ( u 1 , u 2 , u 3 , u 4 ) ⊤ ∈ S ∩ G ⊂ E W \vec{u} = (u_{1},u_{2},u_{3},u_{4})^{\top} \in\mathcal{S} \cap\mathcal{G} \subset E_{W} u = ( u 1 , u 2 , u 3 , u 4 ) ⊤ ∈ S ∩ G ⊂ E W . It is clear that Q ( z ) Q(z) Q ( z ) is a nonzero 3-variate polynomial of total degree at most 3 ( W − 1 ) 3(W-1) 3 ( W − 1 ) and it has the following property
Q ( A n ⃗ ) = { 1 , if n ⃗ = u ⃗ , 0 , if n ⃗ ∈ S ∩ G but n ⃗ ≠ u ⃗ . (94) Q(A\vec{n}) =
\begin{cases}
1, & \text{if } \vec{n} = \vec{u},\\
0, & \text{if } \vec{n} \in\mathcal{S} \cap\mathcal{G}\text{ but } \vec{n} \ne\vec{u}.
\end{cases} \tag*{(94)} Q ( A n ) = { 1 , 0 , if n = u , if n ∈ S ∩ G but n = u . (94)
It is now ready to construct a nonzero polynomial B ^ ( z ) = B ^ ( z 1 , z 2 , z 3 ) \hat{B}(z) = \hat{B}(z_{1},z_{2},z_{3}) B ^ ( z ) = B ^ ( z 1 , z 2 , z 3 ) as follows. We note that # ( S ∩ G ) + # ( S ∖ G ) = # S ≥ 2 \#(\mathcal{S} \cap\mathcal{G}) + \#(\mathcal{S} \setminus\mathcal{G}) = \#\mathcal{S} \ge2 # ( S ∩ G ) + # ( S ∖ G ) = # S ≥ 2 and the specific point u ⃗ ∈ S ∩ G \vec{u} \in\mathcal{S} \cap\mathcal{G} u ∈ S ∩ G .
If S ∖ G = ∅ \mathcal{S} \setminus\mathcal{G} = \varnothing S ∖ G = ∅ , then # ( S ∩ G ) = # S ≥ 2 \#(\mathcal{S} \cap\mathcal{G}) = \#\mathcal{S} \ge2 # ( S ∩ G ) = # S ≥ 2 . We can set B ^ ( z ) = Q ( z ) \hat{B}(z) = Q(z) B ^ ( z ) = Q ( z ) , which satisfies
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) = ∑ n ⃗ ∈ S ∩ G v n ⃗ Q ( A n ⃗ ) = v u ⃗ Q ( A u ⃗ ) = v u ⃗ ≠ 0. \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) = \sum_{\vec{n}\in\mathcal{S}\cap\mathcal{G}} v_{\vec{n}}Q(A\vec{n}) = v_{\vec{u}}Q(A\vec{u}) = v_{\vec{u}} \ne0. n ∈ S ∑ v n B ^ ( A n ) = n ∈ S ∩ G ∑ v n Q ( A n ) = v u Q ( A u ) = v u = 0.
If S ∖ G ≠ ∅ \mathcal{S} \setminus\mathcal{G} \ne\varnothing S ∖ G = ∅ and S ∩ G = { u ⃗ } \mathcal{S} \cap\mathcal{G} = \{\vec{u}\} S ∩ G = { u } , we can set B ^ ( z ) = R ( z + A ( m ⃗ − u ⃗ ) ) \hat{B}(z) = R(z + A(\vec{m} - \vec{u})) B ^ ( z ) = R ( z + A ( m − u )) . It is clear that
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) = ∑ n ⃗ ∈ S ∖ G v n ⃗ B ^ ( A n ⃗ ) + v u ⃗ B ^ ( A u ⃗ ) = ∑ n ⃗ ∈ S ∖ G v n ⃗ R ( A m ⃗ 1 ) + v u ⃗ R ( A m ⃗ ) \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) = \sum_{\vec{n}\in\mathcal{S}\setminus\mathcal{G}} v_{\vec{n}}\hat{B}(A\vec{n}) + v_{\vec{u}}\hat{B}(A\vec{u}) = \sum_{\vec{n}\in\mathcal{S}\setminus\mathcal{G}} v_{\vec{n}}R(A\vec{m}_{1}) + v_{\vec{u}}R(A\vec{m}) n ∈ S ∑ v n B ^ ( A n ) = n ∈ S ∖ G ∑ v n B ^ ( A n ) + v u B ^ ( A u ) = n ∈ S ∖ G ∑ v n R ( A m 1 ) + v u R ( A m )
where m ⃗ 1 = n ⃗ + m ⃗ − u ⃗ ∈ Z 4 \vec{m}_{1} = \vec{n} + \vec{m} - \vec{u} \in\mathbb{Z}^{4} m 1 = n + m − u ∈ Z 4 . Hence, from (89) ,
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) = v u ⃗ R ( A m ⃗ ) ≠ 0. \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) = v_{\vec{u}}R(A\vec{m}) \ne0. n ∈ S ∑ v n B ^ ( A n ) = v u R ( A m ) = 0.
If S ∖ G ≠ ∅ \mathcal{S} \setminus\mathcal{G} \ne\varnothing S ∖ G = ∅ and # ( S ∩ G ) ≥ 2 \#(\mathcal{S} \cap\mathcal{G}) \ge2 # ( S ∩ G ) ≥ 2 , we set B ^ ( z ) = R ( z + A ( m ⃗ − u ⃗ ) ) Q ( z ) \hat{B}(z) = R(z + A(\vec{m} - \vec{u}))Q(z) B ^ ( z ) = R ( z + A ( m − u )) Q ( z ) . Then, we have
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) = ∑ n ⃗ ∈ S ∖ G v n ⃗ R ( A m ⃗ 1 ) Q ( A n ⃗ ) + ∑ n ⃗ ∈ S ∩ G v n ⃗ R ( A m ⃗ 1 ) Q ( A n ⃗ ) \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) = \sum_{\vec{n}\in\mathcal{S}\setminus\mathcal{G}} v_{\vec{n}}R(A\vec{m}_{1})Q(A\vec{n}) + \sum_{\vec{n}\in\mathcal{S}\cap\mathcal{G}} v_{\vec{n}}R(A\vec{m}_{1})Q(A\vec{n}) n ∈ S ∑ v n B ^ ( A n ) = n ∈ S ∖ G ∑ v n R ( A m 1 ) Q ( A n ) + n ∈ S ∩ G ∑ v n R ( A m 1 ) Q ( A n )
where m ⃗ 1 = n ⃗ + m ⃗ − u ⃗ ∈ Z 4 \vec{m}_{1} = \vec{n} + \vec{m} - \vec{u} \in\mathbb{Z}^{4} m 1 = n + m − u ∈ Z 4 . By using (89) and (94) , we obtain
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) = ∑ n ⃗ ∈ S ∩ G v n ⃗ R ( A m ⃗ 1 ) Q ( A n ⃗ ) = v u ⃗ R ( A m ⃗ ) Q ( A u ⃗ ) = v u ⃗ R ( A m ⃗ ) ≠ 0. \sum_{\vec{n}\in\mathcal{S}} v_{\vec{n}}\hat{B}(A\vec{n}) = \sum_{\vec{n}\in\mathcal{S}\cap\mathcal{G}} v_{\vec{n}}R(A\vec{m}_{1})Q(A\vec{n}) = v_{\vec{u}}R(A\vec{m})Q(A\vec{u}) = v_{\vec{u}}R(A\vec{m}) \ne0. n ∈ S ∑ v n B ^ ( A n ) = n ∈ S ∩ G ∑ v n R ( A m 1 ) Q ( A n ) = v u R ( A m ) Q ( A u ) = v u R ( A m ) = 0.
Therefore, we find a nonzero polynomial of three complex variables z 1 z_1 z 1 , z 2 z_2 z 2 and z 3 z_3 z 3
B ^ ( z 1 , z 2 , z 3 ) = Q ( z ) R ( z − A ( u ⃗ − m ⃗ ) ) = Q ( z ) ∑ α 1 = 0 b 1 ∑ α 2 = 0 b 2 ∑ α 3 = 0 b 3 r ^ ( α 1 , α 2 , α 3 ) ( z 1 − s 1 ) α 1 ( z 2 − s 2 ) α 2 ( z 3 − s 3 ) α 3 = ∑ α 1 = 0 T 1 ∑ α 2 = 0 T 2 ∑ α 3 = 0 T 3 c ^ ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3 \begin{aligned}
\widehat{B}(z_1,z_2,z_3)
&= Q(z)R(z-A(\vec{u}-\vec{m})) \\
&= Q(z)\sum_{\alpha_1=0}^{b_1}\sum_{\alpha_2=0}^{b_2}\sum_{\alpha_3=0}^{b_3}\widehat{r}_{(\alpha_1,\alpha_2,\alpha_3)}(z_1-s_1)^{\alpha_1}(z_2-s_2)^{\alpha_2}(z_3-s_3)^{\alpha_3} \\
&= \sum_{\alpha_1=0}^{T_1}\sum_{\alpha_2=0}^{T_2}\sum_{\alpha_3=0}^{T_3}\widehat{c}_{(\alpha_1,\alpha_2,\alpha_3)}z_1^{\alpha_1}z_2^{\alpha_2}z_3^{\alpha_3}
\end{aligned} B ( z 1 , z 2 , z 3 ) = Q ( z ) R ( z − A ( u − m )) = Q ( z ) α 1 = 0 ∑ b 1 α 2 = 0 ∑ b 2 α 3 = 0 ∑ b 3 r ( α 1 , α 2 , α 3 ) ( z 1 − s 1 ) α 1 ( z 2 − s 2 ) α 2 ( z 3 − s 3 ) α 3 = α 1 = 0 ∑ T 1 α 2 = 0 ∑ T 2 α 3 = 0 ∑ T 3 c ( α 1 , α 2 , α 3 ) z 1 α 1 z 2 α 2 z 3 α 3
with the shift A ( u ⃗ − m ⃗ ) = ( s 1 , s 2 , s 3 ) T A(\vec{u}-\vec{m})=(s_1,s_2,s_3)^{\mathsf{T}} A ( u − m ) = ( s 1 , s 2 , s 3 ) T , which satisfies
∑ n ⃗ ∈ S v n ⃗ B ^ ( A n ⃗ ) ≠ 0. \sum_{\vec{n}\in S}v_{\vec{n}}\widehat{B}(A\vec{n})\ne0. n ∈ S ∑ v n B ( A n ) = 0.
Thus, the proof of Lemma 7 is complete. □ \square □ Louisiana State University, Baton Rouge, LA 70803
E-mail: xbliang@lsu.edu