ABCExceptions

3 Upper bounds for integer points

3.1 Fourier analysis

The following result uses basic Fourier analysis to bound the quantity defined in 2.0.3.

Proposition 3.1 Fourier analysis bound

Let \(d\geq 1\), \(\varepsilon {\gt}0\) and \(A\geq 1\) be fixed. Let

\[ X_1,\ldots , X_d, Y_1,\ldots , Y_d,Z_1,\ldots , Z_d\geq 1 \]

and put

\begin{equation} \label{eq:Delta} \Delta =\max _{1\leq i\leq d}(X_i Y_i Z_i). \end{equation}
3.1.1

Let \(\mathbf{c}=(c_1,c_2,c_3)\in \mathbb {Z}^3\) satisfy \(0{\lt}|c_1|,|c_2|,|c_3|\leq \Delta ^A\). Then

\begin{align*} B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \Delta ^\varepsilon \frac{\prod _{j\leq d} \big(X_j Y_j Z_j(Y_j+Z_j)\big)^{\frac{1}{2}}}{\max _{i{\gt}1}\prod _{j\equiv 0\bmod i}Z_j^{\frac{1}{2}}}. \end{align*}
Proof

By the orthogonality of characters, we have

\begin{align*} B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) & \leq \int _0^1 \sum _{x_j\sim X_j}\sum _{y_j\sim Y_i}\sum _{z_j\sim Z_j}e\Big(\alpha \Big(c_1\prod _{j\leq d}x_j^j+c_2\prod _{j\leq d}y_j^j-c_3\prod _{j\leq d}z_j^j\Big)\Big)\mathrm{d}{\alpha } \nonumber \\ & = \int _{0}^1 S_1(\alpha )S_2(\alpha )S_3(-\alpha )\, \mathrm{d}\alpha , \end{align*}

where

\begin{align*} & S_1(\alpha )=\sum _{x_1\sim X_1,\ldots , x_d\sim X_d}e(\alpha c_1x_1x_2^2\cdots x_d^d),\quad S_2(\alpha )=\sum _{y_1\sim Y_1,\ldots , y_d\sim Y_d}e(\alpha c_2y_1y_2^2\cdots y_d^d),\\ & S_3(\alpha )=\sum _{z_1\sim Z_1,\ldots , z_d\sim Z_d}e(\alpha c_3z_1z_2^2\cdots z_d^d). \end{align*}

Then Cauchy-Schwarz gives

\begin{align} \label{eq:Bc123} B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \leq \left(\int _{0}^1|S_1(\alpha )|^2\, \d\alpha \right)^{\frac{1}{2}}\left(\int _{0}^1|S_2(\alpha )|^2|S_3(\alpha )|^2\, \d\alpha \right)^{\frac{1}{2}} \ =: \sqrt{I_1\, I_2}. \end{align}

By Parseval’s identity and the divisor bound, we have

\begin{align} \label{eq:prop3.1I1}\begin{split} I_1 =\int _{0}^1|S_1(\alpha )|^2\, \d\alpha & = \sum _{x_j\sim X_j\forall j}\# \big\{ (x_1’,\ldots , x_d’)\colon x_j’\sim X_j\, \, \forall j,\, x_1x_2^2\cdots x_d^d=x_1’x_2’^2\cdots x_d’^d \big\} \\ & \ll \prod _{j\leq d} X_j^{1+\varepsilon }, \end{split}\end{align}

for any \(\varepsilon {\gt}0\). Using Cauchy-Schwarz again, for any \(i\le d\) we have

\begin{align*} |S_3(\alpha )|^2 = \bigg|\sum _{z_j\sim Z_j\, \forall j\leq d}e(\alpha c_3z_1\cdots z_d^d)\bigg|^2 & \leq T(\alpha )\prod _{j\not\equiv 0\bmod i}Z_{j}, \end{align*}

where

\[ T(\alpha )= \sum _{z_j\sim Z_j\, \forall j\not\equiv 0\bmod i}\bigg|\sum _{z_j\sim Z_j\, \forall j\equiv 0\bmod i}e(\alpha c_3z_1\cdots z_d^d)\bigg|^2. \]

Let \(ir\) be the largest multiple of \(i\) in \([1,d]\). Then

\begin{align} \label{eq:I2}\begin{split} I_2 & = \int _{0}^1|S_2(\alpha )|^2|S_3(\alpha )|^2\, \d\alpha \\ & \le \ \prod _{j\not\equiv 0\bmod i}Z_{j}\cdot \int _{0}^1|S_2(\alpha )|^2\, T(\alpha )\, \d\alpha \\ & \ = \prod _{j\not\equiv 0\bmod i}Z_{j} \cdot \tilde N, \end{split}\end{align}

where \(\tilde N\) is the number of

\[ (z_1,\ldots , z_d,z_i',\ldots , z_{ir}')\in \mathbb {N}^{d+r}, \quad (y_1,\ldots , y_d)\in \mathbb {N}^{d},\quad (y_1',\ldots , y_d')\in \mathbb {N}^d \]

such that

\[ y_j, y_j'\sim Y_j, \quad z_j \sim Z_j, \quad z_j'\sim Z_j \]

for all \(j\leq d\), and

\[ c_3\left(\prod _{j\equiv 0\bmod i}z_j^j-\prod _{j \equiv 0\bmod i}z_j'^j\right)\prod _{j\not\equiv 0\bmod i}z_{j}^j +c_2\prod _{j\leq d}y_j^j-c_2\prod _{j\leq d}y_j'^j=0. \]

Let us write

\begin{align} \label{eq:N} \tilde N=\tilde{N}_1+\tilde{N}_2, \end{align}

where \(\tilde{N}_1\) is the contribution to \(\tilde N\) from tuples with \(\prod _{j\equiv 0\bmod i}z_j^j= \prod _{j \equiv 0\bmod i}z_j'^j\), and \(\tilde{N}_2\) is the contribution of the complementary tuples.

Then by the divisor bound we have

\begin{align} \label{eq:N1} \tilde{N}_1= \# \Big\{ y_j, y_j’\sim Y_j,\, z_j \sim Z_j\, \forall j\leq d\, :\, \prod _{j}y_j^j=\prod _{j}y_j’^j\Big\} \ \ll \ \prod _j Z_j Y_j^{1+\varepsilon }, \end{align}

for any \(\varepsilon {\gt}0\). In order to bound \(\tilde{N}_2\), we first note that \(a-b\mid a^i-b^i\) for any integers \(a\neq b\) and \(i\geq 1\). Thus for any integers \(n\neq 0\) and \(i\geq 2\),

\begin{align*} \# \{ (a,b)\in \mathbb {Z}^2:a^i-b^i=n\} |\leq \tau (|n|)\max _{d\mid n}\# \{ b\in \mathbb {Z}:(b+d)^i-b^i=n\} \ll _{\varepsilon } |n|^{\varepsilon }. \end{align*}

This follows from the divisor bound and the fact that \((x+d)^i-x^i-n\) is a polynomial of degree \(i-1\). (Importantly, this argument fails when \(i=1\), since then the polynomial \((x+n)^i-x^i-n\) is identically \(0\).) Hence, on appealing to the divisor bound, we obtain

\begin{align} \label{eq:N2}\begin{split} \tilde{N}_2& \ll \prod _{j\leq d}Y_j^2\cdot \max _{0{\lt}|n|\leq \Delta ^{A+k^2}}\# \left\{ \begin{array}{l} (z_1,\ldots ,z_d,z_i’,\ldots , z_{ir}’)\in \mathbb {N}^{d+r}: z_j \sim Z_j\, z_j’ \sim Z_j’\, \forall j\\ c_3(\prod _{j\equiv 0\bmod i}z_j^j-\prod _{j\equiv 0\bmod i}z_j’^j)\prod _{j\not\equiv 0\bmod i}z_{j}^j=n \end{array}\right\} \\ & \ll \prod _{j\leq d}Y_j^2Z_j^{\varepsilon /2}\cdot \max _{0{\lt}|n|\leq \Delta ^{A+k^2}}\# \Big\{ (a,b,c)\in \mathbb {N}^{3}:\, \, c(a^i-b^i)=n \Big\} \\ & \ll \Delta ^\varepsilon \prod _{j\leq d}Y_j^{2}. \end{split}\end{align}

Combining 3.1.53.1.93.1.10 and 3.1.11, we deduce that

\begin{equation} \begin{split} \label{eq:prop3.1I2} I_2 & \ll \Delta ^\varepsilon \prod _{j\not\equiv 0\bmod i}Z_{j} \cdot \bigg(\prod _j Y_j Z_j + \prod _{j} Y_j^2 \bigg) \\ & \ll \Delta ^\varepsilon \prod _{j\equiv 0\bmod i}Z_j^{-1}\cdot \prod _{j}\Big(Y_j Z_j^{2} + Y_j^2Z_j\Big). \end{split}\end{equation}
3.1.15

Plugging 3.1.2 and 3.1.16 back into 3.1.1, we conclude that

\begin{align*} B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \leq \sqrt{I_1\, I_2}\ll \Delta ^\varepsilon \prod _{j\equiv 0\bmod i}Z_j^{-\frac{1}{2}}\cdot \prod _{j}(X_j Y_j Z_j(Y_j+Z_j))^{\frac{1}{2}}, \end{align*}

which is the desired bound.

3.2 Geometry of numbers

We can supplement Proposition 3.1 with the following bound, where \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\) is defined in 2.0.3.

Proposition 3.2 Geometry of numbers bound

Let \(d\geq 1\) and \(\varepsilon {\gt}0\) be fixed, and let

\[ X_1,\ldots , X_d, Y_1,\ldots , Y_d,Z_1,\ldots , Z_d\geq 1. \]

Let \(\mathbf{c}\in (c_1,c_2,c_3)\in \mathbb {Z}^3\) have non-zero and pairwise coprime coordinates. Then for \(\Delta \) as in 3.1.1,

\[ B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \hspace{-0.1cm} \Delta ^{\varepsilon } \hspace{-0.3cm} \min _{I,I',I''\subset [d]} \hspace{-0.1cm} \Big(\prod _{i\in I}X_i\prod _{i\in I'}Y_i\prod _{i\in I''}Z_i\Big)\bigg(1 + \frac{\prod _{i\notin I}X_i^i\prod _{i\notin I'}Y_i^i\prod _{i\notin I''}Z_i^i}{\max \{ |c_1|\prod _{i}X_i^i,\; |c_2|\prod _{i}Y_i^i,\; |c_3|\prod _{i}Z_i^i\} }\bigg). \]
Proof

Take any sets \(I,I',I''\subset [d]\). Let \((x_1,\ldots , x_d,y_1,\ldots , y_d,z_1,\ldots , z_d)\) be a tuple counted by \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\). We fix a choice of \( x_i,y_i,z_i \in \mathbb {Z} \) for all indices \(i\) in \(I,I',I''\), respectively, and define

\begin{alignat*}{3}& a_1=c_1\prod _{i\in I}x_i^i, \quad & & a_2=c_2\prod _{i\in I'}y_i^i, \quad & & a_3=c_3\prod _{i\in I''}z_i^i,\\ & x=\prod _{i\notin I}x_i^i, \qquad & & y=\prod _{i\notin I'}y_i^i, \qquad & & z=\prod _{i\notin I''}z_i^i,\\ & X=\prod _{i\notin I}X_i^i, \qquad & & Y=\prod _{i\notin I'}Y_i^i, \qquad & & Z=\prod _{i\notin I''}Z_i^i. \end{alignat*}

Then \(\gcd (a_1,a_2,a_3)=1\) and \(\gcd (x,y,z)=1\). According to Heath-Brown  [ 6 , Lemma 3 ] , the number of triples \((x,y,z)\) that contribute to \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\) is

\begin{align*} \ll \ 1 \ + \ & \frac{XYZ}{\max \big\{ |a_1|X,\, |a_2|Y,\, |a_3|Z\big\} }. \end{align*}

Moreover, by the divisor bound, any triple \((x,y,z)\) corresponds to \(O_{\varepsilon }(\Delta ^{\varepsilon })\) choices of \(x_i,y_j,z_r\) with \(i\not\in I, j\not\in I', r\not\in I''\). We arrive at the desired upper bound by summing over the choices \(x_i,y_i,z_i\in \mathbb {Z}\), with \(i\) in \(I,I',I''\).

Theorem 3.3
#

Let \(\gcd (a_1,a_2,a_3)=1\). Then for \(X_1,X_2,X_3{\gt}1\), we have

\begin{align*} \# \left((x_1,x_2,x_3)\in \mathbb {Z}^3 \; :\; \gcd (x_1,x_2,x_3)=1,\, |x_i| \le X_i ,\, a_1x_1+a_2x_2+a_3x_3=0\right) \ll \ 1 \ + \ & \frac{X_1X_2X_3}{\max _i\{ |a_i|X_i\} }. \end{align*}
Proof

3.3 Determinant method

In this section we will record a bound for \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\) in 2.0.3 that proceeds via the determinant method of Bombieri-Pila  [ 2 ] and Heath-Brown  [ 8 ] . See  [ 1 ] for a gentle introduction to the determinant method. We first record a basic fact about the irreducibility of certain polynomials.

Lemma 3.4

Let \(r\geq 1\) and let \(g\in \mathbb {C}[x]\) be a polynomial which has at least one root of multiplicity \(1\). Then the polynomial \(g(x)-y^r\) is absolutely irreducible.

Proof

We may assume a factorisation \(g(x)=l_1(x)^{e_1}\dots l_t(x)^{e_t}\), with pairwise non-proportional linear polynomials \(l_1,\dots ,l_t\in \mathbb {C}[x]\) and exponents \(e_1,\dots ,e_t\in \mathbb {N}\) such that \(e_1=1\). But \({\mathbb {C}}[x]\) is a unique factorisation domain and so we can apply Eisenstein’s criterion with the prime \(l_1\) in order to deduce that \(g(x)-y^r\) is irreducible over \( {\mathbb {C}}[y]\). It then follows that \(g(x)-y^r\) is irreducible over \( \mathbb {C}\), as claimed in the lemma.

Let \(p,q,r\) be positive integers and let \(a_1,a_2,a_3\in \mathbb {Z}_{\neq 0}\). We shall require a good upper bound for the counting function

\[ N(X,Y,Z)= \# \left\{ (x,y,z)\in \mathbb {Z}_{\neq 0}^3: \begin{array}{l} |x|\leq X, ~ |y|\leq Y,~ |z|\leq Z\\ \gcd (x,y)=\gcd (x,z)=\gcd (y,z)=1 \\ a_1x^p+a_2 y^q+a_3 z^r=0 \end{array} \right\} , \]

for given \(X,Y,Z\geq 1\). This is achieved in the following result.

Bombieri–Pila [ 2 , Theorem 4 ]

Theorem 3.5
#

Let \(f(x)\) be a \(C^\infty \) function on a closed subinterval of \([0,N]\), and suppose that\(F (x, f ) = 0\), where \(F(x, y) \in \mathbb {R}[x, y]\) is absolutely irreducible of degree \(d \ge 2\). Suppose that \(|f'(x)| \le 1\). Then

\begin{align*} \# \{ (x,f(x))\in \{ 1,\ldots ,N\} ^2 \} \ll _d (\log N)^{O(N)} N^{1/d} \end{align*}
Proof

Heath-Brown Theorem 15

Theorem 3.6
#

Let \(F\in \mathbb {Z}[x_1 ,\ldots , x_n]\) be an absolutely irreducible polynomial of degree \(d\), and let \(\varepsilon {\gt} 0\) and \(B_1,\ldots ,B_n \ge 1\) be given. Define

\begin{align*} B = \max _{(e_1,\ldots ,e_n)}\Big( \prod _{i\le n} B_i^{e_i} \Big) \end{align*}

where the maximum is taken over all integer \(n\)-tuples \((e_1,\ldots ,e_n)\) for which the corresponding monomial \(x_1^{e_1}\cdots x_n^{e_n}\) occurs in \(F\) with non-zero coefficient.

Then there exists \(D = D(n,d,\varepsilon )\) and an integer \(k\) with

\begin{align*} k \ll _{n,d,\varepsilon } T^\varepsilon (\log \| F\| )^{2n-3} \exp \Big((n-1)\Big(\frac{\prod _{i\le n}\log B_i}{\log B}\Big)^{1/(n-1)}\Big) \end{align*}

satisfying the following: There are \(k\) polynomials \(F_1,\ldots ,F_k\in \mathbb {Z}[x_1 ,\ldots , x_n]\) coprime to \(F\), with \(\deg F_i \le D\), such that every root of \(F(x_1 ,\ldots , x_n)=0\) with \(x_i \le B_i\) is also a root of \(F_j(x_1 ,\ldots , x_n)=0\), for some \(j\le k\).

Proof
Theorem 3.7

Given integers \(n,a_1,a_2\neq 0\), there are at most \(O_{\varepsilon ,D}(|na_1a_2 X_1 X_2|^{\varepsilon })\) many solutions \((x_1,x_2)\in \mathbb {Z}^2\) such that \(|x_i|\le X_i\) and

\begin{align*} n = a_1 x_1^2 + a_2 x_2^2. \end{align*}
Proof

Suppose \(n=a_1x_1^2+a_2x_2^2\). Then \(a_1n=a_1^2x_1^2+a_1a_2x_2^2\). Let \(D\) be the squarefree part of \(a_1a_2\). Then the number of solutions \((x_1,x_2)\) with \(|x_i|\leq X_i\) to this equations is at most the number of solutions \((m_1,m_2)\) to \(a_1n=m_1^2+Dm_2^2\) with \(|m_i|\leq X_ia_1a_2\). For any such solution, we have \(m_1+\sqrt{-D}m_2\mid a_1n\) in \(\mathbb {Q}(\sqrt{-D})\). The claim follows from the divisor bound in quadratic fields, in Lemma 3.8.

Lemma 3.8

Let \(\varepsilon {\gt}0\). Let \(D\ge 1\) be a squarefree integer, and set \(K=\mathbb {Q}(\sqrt{-D})\). Then for all \(\alpha \in K\), the number of ideals dividing \((\alpha )\) in \(K\) is \(O(N_\alpha ^\varepsilon )\).

Proof

Mimics the proof over the integers using the fundamental theorem of arithmetic, but with ideals [See link in comment]

Theorem 3.9

Given integers \(n,a_1,a_2\neq 0\) and \(p\ge 3\), there are at most \(O(p^{1+\omega (|n|)})\) many solutions \((x_1,x_2)\in \mathbb {Z}^2\) such that \(|x_i|\le X_i\) and

\begin{align*} n = a_1 x_1^p + a_2 x_2^p. \end{align*}
Proof
Definition 3.10
#

Let \(\omega (n)\) denote the number of distinct prime factors of an integer \(n\).

Lemma 3.11

For any \(n\ge 2\), we have \(\omega (n) \ll \log (3n)/\log \log (3n)\).

Proof
Lemma 3.12

Let \(\varepsilon {\gt}0\) and \(D\geq 1\) and assume that \(p,q,r\in [1,D]\) are integers. Then

\[ N(X,Y,Z)\ll _{\varepsilon ,D} Z \min \left(X^{\frac{1}{q}}, Y^{\frac{1}{p}} \right) (XY)^\varepsilon , \]

where the implied constant only depends on \(\varepsilon \) and \(D\). Furthermore, if \(p=q\geq 2\), then we have

\[ N(X,Y,Z)\ll _{\varepsilon ,D} Z (|a_1a_2a_3|XYZ)^\varepsilon . \]
Proof

We fix a choice of non-zero integer \(z\in [-Z,Z]\), of which there are \(O(Z)\). When \(z\) is fixed, the resulting equation defines a curve in \(\mathbb {A}^2\) and we can hope to apply work of Bombieri-Pila  [ 2 , Theorem 4 ] , which would show that the equation has \(O_{\varepsilon ,D}(\max (X,Y)^{\frac{1}{\max (p,q)}+\varepsilon })\) integer solutions in the region \(|x|\leq X\) and \(|y|\leq Y\), where the implied constant only depends on \(\varepsilon \) and \(D\). This is valid only when the curve is absolutely irreducible, which we claim is true when \(z\neq 0\). But, for fixed \(z\in \mathbb {Z}_{\neq 0}\) the polynomial \(a_2y^q+a_3z^r\) has non-zero discriminant as a polynomial in \(y\). Hence the claim follows from Lemma ??. Rather than appealing to Bombieri-Pila, however, we can get a sharper bound by using work of Heath-Brown  [ 8 , Theorem 15 ] . For fixed \(z\in \mathbb {Z}_{\neq 0}\) this gives the bound \(O_{\varepsilon ,D}(\min (X^{\frac{1}{q}},Y^{\frac{1}{p}})(XY)^\varepsilon )\) for the number of available \(x,y\).

Lemma 3.13

Let \(\varepsilon {\gt}0\) and \(D\geq 1\) and assume that \(p=q,r\in [1,D]\) are integers. Then

\[ N(X,Y,Z)\ll _{\varepsilon ,D} Z (|a_1a_2a_3|XYZ)^\varepsilon . \]

where the implied constant only depends on \(\varepsilon \) and \(D\).

Proof

Suppose now that \(p=q\geq 2\). Then, for given \(z\in \mathbb {Z}_{\neq 0}\), we are left with counting the number of integer solutions to the equation \(N=a_1x^p+a_2y^p\), with \(|x|,|y|\leq \max (X,Y)\), and where \(N=-a_3z^r\). For \(p=2\) this is a classical problem in quadratic forms. The bound \(O_{\varepsilon ,D}((|a_1a_2N|XY)^\varepsilon )\) follows from Heath-Brown  [ 7 , Theorem 3 ] , for example. For \(p\geq 3\) we obtain a Thue equation. According to work of Bombieri and Schmidt  [ 3 ] , there are at most \(O(p^{1+\omega (|N|)})\) solutions, for an absolute implied constant. Using the bound \(\omega (|N|)\ll \log (3|N|)/(\log \log (3|N|))\), this is \(O_{\varepsilon ,D}((|a_3|Z)^\varepsilon )\), which thereby completes the proof of the lemma.

Using these lemmas we can now supplement Propositions 3.1 and 3.2 with further bounds for \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\), as defined in 2.0.3.

Proposition 3.14

Let \(d\geq 1\), and let

\[ X_1,\ldots , X_d, Y_1,\ldots , Y_d,Z_1,\ldots , Z_d\geq 1. \]

Let \(\mathbf{c}\in (c_1,c_2,c_3)\in \mathbb {Z}_{\neq 0}^3\). Then for \(\Delta \) as in 3.1.1, we have

\[ B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \Delta ^\varepsilon \prod _{j\leq d} X_j Y_j Z_j \cdot \min _{p,q\geq 1}\left( (X_p Y_q)^{-1} \min \left( X_p^{\frac{1}{q}} , Y_q^{\frac{1}{p}}\right)\right). \]
Proof

Let \((x_1,\ldots , x_d,y_1,\ldots , y_d,z_1,\ldots , z_d)\) be a tuple counted by \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\). For any integers \(p,q\geq 1\), we fix all but \(x_p\) and \(y_q\) and apply the first part of Lemma 3.12. This gives

\[ B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \Delta ^\varepsilon \prod _{\substack {j\leq d\\ j\neq p }} X_j \prod _{\substack {j\leq d\\ j\neq q }} Y_j \prod _{\substack {j\leq d }} Z_j \cdot \min \left( X_p^{\frac{1}{q}} , Y_q^{\frac{1}{p}}\right) , \]

from which the statement of the lemma easily follows.

Proposition 3.15

Let \(d\geq 1\), and let

\[ X_1,\ldots , X_d, Y_1,\ldots , Y_d,Z_1,\ldots , Z_d\geq 1. \]

Let \(\mathbf{c}\in (c_1,c_2,c_3)\in \mathbb {Z}_{\neq 0}^3\). Then

\[ B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \Delta ^\varepsilon \prod _{j\leq d} X_j Y_j Z_j \cdot \min _{p\geq 2} \prod _{\substack {j\leq d \\ p\mid j}} (X_j Y_j)^{-1}, \]

where \(\Delta \) is given by 3.1.1.

Proof

Let \((x_1,\ldots , x_d,y_1,\ldots , y_d,z_1,\ldots , z_d)\) be a tuple counted by \(B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z})\). We collect together all indices which are multiples of \(p\) for any integer \(p\geq 2\). Applying the second part of Lemma 3.13 now gives

\[ B_d(\mathbf{c},\mathbf{X},\mathbf{Y},\mathbf{Z}) \ll \Delta ^\varepsilon \prod _{\substack {j\leq d\\ j\not\equiv 0 \bmod {p} }} X_j Y_j Z_j \prod _{\substack {j\leq d\\ j\equiv 0 \bmod {p} }} Z_j, \]

from which the statement easily follows.