The next step is to count the number of integral points on the universal torsor subject to certain bounds, given by lifting the height function to the torsor, and satisfying the coprimality conditions. For three of the ten variables on the torsor, this summation is done in Section  5 by elementary methods from analytic number theory. The summation over the last seven variables, completing the proof of Theorem  1 , is carried out in Section  6 .

During the final preparation of this paper, I was informed by M. Joyce (Brown University) that similar results towards Manin’s conjecture for certain cubic surfaces will appear in his thesis.

Acknowledgments.

Most of this paper was written while I was visiting B. Hassett at Rice University in February 2005. I am very grateful to him for the invitation and stimulating discussions. I thank T. D. Browning for helpfully explaining the details of his paper [?] . I thank S. Elsenhans and E. Kappos for many useful suggestions. This research has been partially supported by the Studienstiftung des Deutschen Volkes.

2 Manin’s Conjecture

In this section, we calculate the invariants appearing in Manin’s conjecture and its refinement by Peyre. We use the notation from [?] .

Lemma 2. Manin’s conjecture predicts in case of $S$  as defined in  1.1 : $$NU,H\left(B\right)\sim c_{S,H}B(logB{)}^6,$$  where $c_{S,H}=\alpha \left(\tilde{S}\right)\beta \left(\tilde{S}\right){\omega }_H\left(\tilde{S}\right)$  with

$$\begin{array}{cccccc}\alpha \left(\tilde{S}\right)& :=\frac{1}{6!{\prod }_i{\lambda }_i}=\frac{1}{6!\cdot 2\cdot 3\cdot 4\cdot 3\cdot 4\cdot 5\cdot 6}=\frac{1}{6220800},& \beta \left(\tilde{S}\right)& :=1,& {\omega }_H\left(\tilde{S}\right)& :={\omega }_{\infty }\cdot {\omega }_0,\end{array}$$

and

$$\begin{array}{cccccccc}\lambda & =({\lambda }_1,{\lambda }_2,{\lambda }_3,{\lambda }_{\ell },{\lambda }_4,{\lambda }_5,{\lambda }_6):=(2,3,4,3,4,5,6),& {\omega }_{\infty }& :=6\int \int {\int }_{\left\{\right(u,v,w)\in {\mathbb{R}}^3|\left|t{v}^3\right|⩽1,|t^2+u^3|⩽1,0⩽v⩽1,\left|u{v}^4\right|⩽1\}}1dtdudv,& {\omega }_0& :={\prod }_p{\omega }_p,& {\omega }_p& :={\left(1-\frac{1}{p}\right)}^7\left(1+\frac{7}{p}+\frac{1}{p^2}\right).\end{array}$$

• Proof. The line $\ell$  and all exceptional divisors $F_1,...,F_6$  of the desingularization $\tilde{S}\to S$  are defined over $\mathbb{Q}$  . Thus, $rkPic\left(\tilde{S}\right)=7$  . By [?] , the effective cone is generated by the classes of $\ell$  and the exceptional divisors. The dual cone of nef divisors is also simplicial (i.e., the number of generators is equal to $rkPic\left(\tilde{S}\right)$  ), and $$-K\tilde{S}=2F_1+3F_2+4F_3+3\ell +4F_4+5F_5+6F_6.$$  Note that $\lambda =-K\tilde{S}\in Pic\left(\tilde{S}\right)\sim ={\mathbb{Z}}^7$  , in the basis $\{F_1,F_2,F_3,\ell ,F_4,F_5,F_6\}$  .
  Using the definition, we calculate $$\alpha \left(\tilde{S}\right)=Vol\left\{(t_1,t_2,t_3,t_{\ell },t_4,t_5,t_6)\in {\mathbb{R}}_{⩾0}^7|{\sum }_{i\in \{1,2,3,\ell ,4,5,6\}}{\lambda }_i{t}_i=1\right\}=\frac{1}{6!\cdot {\prod }_i{\lambda }_i}.$$  The surface $S$  is split over $\mathbb{Q}$  , so that $$\beta \left(\tilde{S}\right)=\#H^1(\mathbb{Q},Pic(\tilde{S}\left)\right)=1.$$  By definition, $${\omega }_H\left(\tilde{S}\right)={lim}_{s\to 1}\left(\right(s-1{)}^{rkPic\left(\tilde{S}\right)}L(s,Pic(\tilde{S}\left)\right)){\omega }_{\infty }{\prod }_p\frac{{\omega }_p}{L_p(1,Pic(\tilde{S}\left)\right)},$$  where, in our case, $${lim}_{s\to 1}\left(\right(s-1{)}^{rkPic\left(\tilde{S}\right)}L(s,Pic(\tilde{S}\left)\right)={lim}_{s\to 1}\left(\right(s-1{)}^7\zeta \left(s{)}^7\right)=1$$  and $$L_p(1,Pic(\tilde{S}){)}^{-1}=(1-1/p{)}^7.$$  We use Peyre’s method [?]  to compute ${\omega }_{\infty }$  and parametrize the points by writing $x_1$  as a function of $x_0,x_2,x_3$  . Since $\mathbf{x}=-\mathbf{x}$  in $\mathbb{P}3$  , we may assume $x_2⩾0$  . Since $\frac{d}{d{x}_1}f=x_2^2$  , the Leray form ${\omega }_L\left(\tilde{S}\right)$  is given by $x_2^{-2}d{x}_{0}d{x}_{3}d{x}_2$  , and we obtain ${\omega }_{\infty }\left(\tilde{S}\right)$  from $$\int \int {\int }_{\left\{\right|x_0|⩽1,|x_2^{-2}(x_2{x}_0^2+x_3^3)|⩽1,0⩽x_2⩽1,|x_3|⩽1\}}{x}_2^{-2}d{x}_{0}d{x}_{3}d{x}_2,$$  using the transformations $$x_0=t{x}_2^{1/2},x_3=u{x}_2^{2/3},x_2=v^6.$$  The calculation of ${\omega }_p$  is done as in Lemma 1 of [?]  and we omit it here.

3 The Universal Torsor

The universal torsor $\mathcal{T}=Spec(\mathbb{Q}[{\xi }_1,{\xi }_2,{\xi }_3,{\xi }_{\ell },{\xi }_4,{\xi }_5,{\xi }_6,{\tau }_1,{\tau }_2,{\tau }_{\ell }]/(T({\xi }_i,{\tau }_i)\left)\right)$  is given by the equation

$$\begin{array}{c}T({\xi }_i,{\tau }_i)=\tau \ell \xi 3\ell \xi 24\xi 5+\tau 22\xi 2+\tau 31\xi 21\xi 3=0\end{array}$$

(3.1)

and the map $\Psi :\mathcal{T}\to S=Spec\left(k\right[x_0,...,x_3]/(f(\mathbf{x})\left)\right)$  defined by

$$\begin{array}{c}\begin{array}{cccccccc}{\Psi }^{*}\left(x_0\right)& =\xi (1,2,2,0,1,2,3){\tau }_2& {\Psi }^{*}\left(x_1\right)& ={\tau }_{\ell }& {\Psi }^{*}\left(x_2\right)& =\xi (2,3,4,3,4,5,6)& {\Psi }^{*}\left(x_3\right)& =\xi (2,2,3,1,2,3,4){\tau }_1\end{array}\end{array}$$

(3.2)

where we use the notation $\xi (n_1,n_2,n_3,n_{\ell },n_4,n_5,n_6):={\xi }_1^{n_1}{\xi }_2^{n_2}{\xi }_3^{n_3}{\xi }_{\ell }^{n_{\ell }}{\xi }_4^{n_4}{\xi }_5^{n_5}{\xi }_6^{n_6}$  .

Note that ${\Psi }^{*}\left(x_2\right)={\xi }^{\lambda }$  with $\lambda \in {\mathbb{Z}}^7$  as in Lemma  2 .

We want to establish a bijection between rational points on the surface $S$  and integral points on the torsor $\mathcal{T}$  which are subject to certain coprimality conditions.

More precisely, the coprimality conditions can be summarized in the following table, where a “ $-$  ” means that the two variables are coprime, and a “ $\times$  ” that they may have common factors. For a variable combined with itself, “ $-$  ” means that each prime occurs at most once, and “ $\times$  ” means that it may occur more often.

$$\begin{array}{ccccccccccc}& {\xi }_1& {\xi }_2& {\xi }_3& {\xi }_{\ell }& {\xi }_4& {\xi }_5& {\xi }_6& {\tau }_1& {\tau }_2& {\tau }_{\ell }\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_1& \times & -& \times & -& -& -& \times & \times & -& -\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_2& -& -& -& -& -& -& \times & -& \times & -\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_3& \times & -& -& -& -& -& \times & -& -& -\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_{\ell }& -& -& -& \times & \times & \times & \times & -& -& \times \end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_4& -& -& -& \times & -& -& \times & -& -& -\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_5& -& -& -& \times & -& -& \times & -& -& -\end{array}$$

$$\begin{array}{ccccccccccc}{\xi }_6& \times & \times & \times & \times & \times & \times & \times & -& \times & -\end{array}$$

$$\begin{array}{ccccccccccc}{\tau }_1& \times & -& -& -& -& -& -& \times & \times & \times \end{array}$$

$$\begin{array}{ccccccccccc}{\tau }_2& -& \times & -& -& -& -& \times & \times & \times & \times \end{array}$$

$$\begin{array}{ccccccccccc}{\tau }_{\ell }& -& -& -& \times & -& -& -& \times & \times & \times \end{array}$$

Later, we will refer to the

$$\begin{array}{c}\text{coprimality conditions between}{\xi }_1,...,{\xi }_6\text{as given in the table}\end{array}$$

(3.3)

Because of the torsor equation $T$  , we can write the coprimality conditions for ${\tau }_i$  equivalently as

$$\begin{array}{c}gcd(\tau 1,\xi 2\xi 3\xi \ell \xi 4\xi 5\xi 6)=1\end{array}$$

(3.4)

and

$$\begin{array}{c}gcd(\tau 2,\xi 1\xi 3)=1,gcd(\tau \ell ,\xi 4\xi 5\xi 6)=1.\end{array}$$

(3.5)

The goal of this section is the following result:

Proposition 3. The map $\Psi$  induces a bijection between $${\mathcal{T}}_1:=\left\{\right({\xi }_i,{\tau }_i)\in \mathcal{T}(\mathbb{Z}\left)\right|\text{<ref><sref>eq:hcf\_xi</sref><pref>3.3</pref></ref>,<ref><sref>eq:hcf\_tone</sref><pref>3.4</pref></ref>,<ref><sref>eq:hcf\_ttwo\_tell</sref><pref>3.5</pref></ref> hold},{\xi }_i>0\}$$  and $U(\mathbb{Q})\subset S(\mathbb{Q})$  .

The proof of this is split into two parts. First, we establish a similar bijection with slightly different coprimality conditions:

Lemma 4. Let ${\mathcal{T}}_2$  be set of all $({\xi }_i,{\tau }_i)\in \mathcal{T}(\mathbb{Z})$  such that ${\xi }_i>0$  and the coprimality conditions described by the table hold, except that the conditions

$$\begin{array}{c}gcd({\xi }_3,\tau 1)=1\text{and}gcd({\xi }_6,\tau 1)=1\end{array}$$

(3.6)

in the table are replaced by

$$\begin{array}{c}\left|\mu \right({\xi }_1\left)\right|=1\text{and}gcd({\xi }_1,{\xi }_3)=1.\end{array}$$

(3.7)

Then the map $\Psi$  induces a bijection between ${\mathcal{T}}_2$  and $U(\mathbb{Q})\subset S(\mathbb{Q})$  .

• Proof. Using the method of [?] , we now show that the coprimality conditions lead to a bijection.
  We go through a series of coprimality considerations and replace the original variables by products of new ones which fulfill certain conditions. When doing this, the new variables will be uniquely determined.

  Since $\mathbf{x}=-\mathbf{x}$  , and $x_2=0$  is equivalent to $\mathbf{x}\in \ell$  , we can write each $\mathbf{x}\in U(\mathbb{Q})$  uniquely such that $x_i\in \mathbb{Z}$  , $x_2>0$  , and $gcd\left(x_i\right)=1$  .

  • ∙ Note that $x_2|x_3^3$  . Write $x_2=y_1{y}_2^2{y}_3^3$  with $y_i\in \mathbb{Z}>0$  , where each triple occurrence of a prime factor of $x_3$  is put in $y_3$  and each double occurrence in $y_2$  , so that $y_1,y_2,y_3$  are unique if we assume $\left|\mu \right(y_1{y}_2\left)\right|=1$  . Then $x_3=y_1{y}_2{y}_{3}z$  must hold for a suitable $z\in \mathbb{Z}$  . Substituting into $f$  and dividing by $y_1{y}_2^2{y}_3^3$  gives $$f_1(x_0,x_1,y_1,y_2,y_3,z)=x_1{y}_1{y}_2^2{y}_3^3+x_0^2+y_1^2{y}_2{z}^3=0.$$ 
  • ∙ Now $y_1{y}_2|x_0^2$  , and since $\left|\mu \right(y_1{y}_2\left)\right|=1$  , we have $y_1{y}_2|x_0$  . Write $x_0=y_1{y}_{2}w$  , where $w\in \mathbb{Z}$  . Substituting and dividing by $y_1{y}_2$  , we obtain $$f_2(x_1,y_1,y_2,y_3,z,w)=x_1{y}_2{y}_3^3+w^2{y}_1{y}_2+y_1{z}^3=0.$$ 
  • ∙ Since $y_2|y_1{z}^3$  and $\left|\mu \right(y_1{y}_2\left)\right|=1$  , we must have $y_2|z$  . Write $z=y_2{z}^{\prime }$  , where $z^{\prime }\in \mathbb{Z}$  , and obtain, after dividing by $y_2$  , the relation $$f_3(x_1,y_1,y_2,y_3,w,z^{\prime })=x_1{y}_3^3+w^2{y}_1+y_1{y}_2^2{z^{\prime }}^3=0.$$ 
  • ∙ Since $y_1$  divides our original variables $x_0,x_2,x_3$  , it cannot divide $x_1$  .
    Together with $\left|\mu \right(y_1\left)\right|=1$  , the fact $y_1|x_1{y}_3^3$  implies $y_1|y_3$  . Write $y_3=y_1{y}_3^{\prime }$  , where $y_3^{\prime }\in \mathbb{Z}>0$  and obtain $$f_4(x_1,y_1,y_2,w,z^{\prime },y_3^{\prime })=x_1{y}_1^2{y}_{3^{\prime }}^3+w^2+y_2^2{z^{\prime }}^3=0$$ 
  • ∙ Let $a=gcd(y_3^{\prime },z^{\prime })\in \mathbb{Z}>0$  and write $y_3^{\prime }=a{y}_3^{\prime \prime }$  , where $y_3^{\prime \prime }\in \mathbb{Z}>0$  and $z^{\prime }=a{z}^{\prime \prime }$  , where $z^{\prime \prime }\in \mathbb{Z}$  . This gives $$f_5(x_1,y_1,y_2,w,z^{\prime \prime },y_3^{\prime \prime },a)=x_1{y}_1^2{{y_3}^{\prime \prime }}^3{a}^3+w^2+y_2^2{z}^{\prime }{a}^3=0.$$ 
  • ∙ Now $a^3|w^2$  . Writing $a={\xi }_6^2{\xi }_2$  , where ${\xi }_2,{\xi }_6\in \mathbb{Z}>0$  with $\left|\mu \right({\xi }_2\left)\right|=1$  , gives $w=w^{\prime }{\xi }_6^3{\xi }_2^2$  , where $w^{\prime }\in \mathbb{Z}$  leading to the equation $$f_6(x_1,y_1,y_2,z^{\prime \prime },y_3^{\prime \prime },w^{\prime },{\xi }_2,{\xi }_6)=x_1{y}_1^2{{y_3}^{\prime \prime }}^3+{w^{\prime }}^2{\xi }_2+y_2^2{z^{\prime \prime }}^3=0.$$ 
  • ∙ Let ${\xi }_5=gcd(y_3^{\prime \prime },w^{\prime })\in \mathbb{Z}>0$  and write $y_3^{\prime \prime }=\xi \ell {\xi }_5$  , where $\xi \ell \in \mathbb{Z}>0$  and $w^{\prime }=w^{\prime \prime }{\xi }_5$  , with $w^{\prime \prime }\in \mathbb{Z}$  . Then $$f_7(x_1,y_1,y_2,z^{\prime \prime },w^{\prime \prime },{\xi }_2,\xi \ell ,{\xi }_5,{\xi }_6)=x_1{y}_1^2\xi {\ell }^3{\xi }_5^3+{w^{\prime \prime }}^2{\xi }_2{\xi }_5^2+y_2^2{z^{\prime \prime }}^3=0.$$ 
  • ∙ Since $gcd(y_3^{\prime \prime },z^{\prime \prime })=1$  , also $gcd(\xi \ell {\xi }_5,z^{\prime \prime })=1$  . Therefore, ${\xi }_5^2|y_2^2$  , which means ${\xi }_5|y_2$  , and we write $y_2={\xi }_1{\xi }_5$  , with ${\xi }_1\in \mathbb{Z}>0$  . We obtain $$f_8(x_1,y_1,z^{\prime \prime },w^{\prime \prime },{\xi }_1,{\xi }_2,\xi \ell ,{\xi }_5,{\xi }_6)=x_1{y}_1^2\xi {\ell }^3{\xi }_5+{w^{\prime \prime }}^2{\xi }_2+{\xi }_1^2{z^{\prime \prime }}^3=0.$$ 
  • ∙ Let ${\xi }_3=gcd(w^{\prime \prime },y_1)\in \mathbb{Z}>0$  . Since $\left|\mu \right(y_1{y}_2\left)\right|=1$  , we have $gcd({\xi }_1,{\xi }_3)=1$  . Therefore, ${\xi }_3|{z^{\prime \prime }}^3$  and even ${\xi }_3|z^{\prime \prime }$  . Write $w^{\prime \prime }=\tau 2{\xi }_3$  , where $\tau 2\in \mathbb{Z}$  , $y_1={\xi }_4{\xi }_3$  , where ${\xi }_4\in \mathbb{Z}>0$  , and $z^{\prime \prime }=\tau 1{\xi }_3$  , where $\tau 1\in \mathbb{Z}$  . Replacing $x_1=\tau \ell$  , where $\tau \ell \in \mathbb{Z}$  , we get $$f_9({\xi }_1,{\xi }_2,{\xi }_3,\xi \ell ,{\xi }_4,{\xi }_5,{\xi }_6,\tau 1,\tau 2,\tau \ell )=\tau \ell \xi {\ell }^3{\xi }_4^2{\xi }_5+{\tau }^{2}2{\xi }_2+{\tau }^{3}1{\xi }_1^2{\xi }_3=0.$$  This is the torsor equation $T({\xi }_i,{\tau }_i)$  as in  3.1 .

  The substitutions lead to $x_0,...,x_3$  in terms of ${\xi }_i,{\tau }_i$  as in  3.2 . Conversely, it is easy to check that each $({\xi }_i,{\tau }_i)$  satisfying $T$  is mapped by $\Psi$  to a point $\mathbf{x}\in S(\mathbb{Q})$  . Note that ${\xi }_i\in \mathbb{Z}>0$  and ${\tau }_i\in \mathbb{Z}$  . Furthermore, the coprimality conditions we introduced impose the following conditions on ${\xi }_i,{\tau }_i$  :

  $$\begin{array}{cccc}& \left|\mu \right(y_1{y}_2\left)\right|=\left|\mu \right({\xi }_1{\xi }_3{\xi }_4{\xi }_5\left)\right|=1,\left|\mu \right({\xi }_2\left)\right|=1,gcd(\tau 2,{\xi }_4)=1,& & gcd(y_3^{\prime \prime },z^{\prime \prime })=gcd(\xi \ell {\xi }_5,\tau 1{\xi }_3)=1,gcd(\xi \ell ,w^{\prime \prime })=gcd(\xi \ell ,\tau 2{\xi }_3)=1.\end{array}$$

  The condition $gcd\left(x_i\right)=1$  is equivalent to $gcd(\tau \ell ,{\xi }_1{\xi }_2{\xi }_3{\xi }_4{\xi }_5{\xi }_6)=1$  .

  We obtain $gcd({\xi }_2,{\xi }_3)=1$  in the following manner: If $p|{\xi }_2,{\xi }_3$  for some prime $p$  , then $p|\tau \ell \xi 3\ell \xi 24\xi 5$  by the torsor equation $T$  . On the other hand, a divisor of ${\xi }_3$  cannot divide any of the factors by the coprimality conditions we found.

  Similarly, we conclude $gcd({\xi }_3,\tau 2)=gcd({\xi }_1,\tau 2)=gcd({\xi }_2,{\xi }_5)=gcd({\xi }_5,\tau 2)=1$  .

  Finally, if a prime $p$  divides two of ${\xi }_2,{\xi }_4,\tau 1$  , we see using $T$  that $p$  must divide all of them. Since $\left|\mu \right({\xi }_2\left)\right|=1$  and $p^2|\tau \ell \xi 3\ell \xi 24\xi 5+\tau 31\xi 21\xi 3$  , we conclude $p|\tau 2$  which is impossible since $gcd(\tau 2,{\xi }_4)=1$  . Therefore, ${\xi }_2,{\xi }_4,\tau 1$  must be pairwise coprime. In the same way we derive that no two of ${\xi }_1,{\xi }_2,\xi \ell$  have a common factor.

  It is easy to check that we cannot derive any other coprimality condition because we could construct a solution to $T$  violating it.

  Note that the conditions on $({\xi }_i,{\tau }_i)$  are exactly the ones given in the definition of ${\mathcal{T}}_2$  . Since in every step the newly introduced variables are uniquely determined, we have established a bijection between $U(\mathbb{Q})$  and ${\mathcal{T}}_2$  .

The following Lemma is the second step towards the proof of Proposition  3 :

Lemma 5. There is a bijection between ${\mathcal{T}}_1$  and ${\mathcal{T}}_2$  .

• Proof. Given a point $({\xi }_i^{\prime },{\tau }_j^{\prime })\in {\mathcal{T}}_2$  violating  3.6 , we could replace a common prime factor $p$  of ${\xi }_3^{\prime },{\xi }_6^{\prime }$  and ${\tau }_1^{\prime }$  by powers of $p$  as factors of ${\xi }_1^{\prime }$  and possibly ${\xi }_3^{\prime }$  such that  3.6 holds. This way, we obtain a point $({\xi }_i,{\tau }_j)\in {\mathcal{T}}_1$  . This should be done in a way such that $\Psi$  maps $({\xi }_i,{\tau }_j)$  and $({\xi }_i^{\prime },{\tau }_j^{\prime })$  to the same point $\mathbf{x}\in U(\mathbb{Q})$  , and such that we have an inverse map, taking care of the conditions  3.7 .
  Let $({\xi }_i,{\tau }_j)\in {\mathcal{T}}_1$  and $({\xi }_i^{\prime },{\tau }_j^{\prime })\in {\mathcal{T}}_2$  . Decompose the coordinates into their prime factors: Let $${\xi }_i={\prod }_p{p}^{n_{ip}},{\tau }_j={\prod }_p{p}^{m_{jp}}\text{and}{\xi }_i^{\prime }={\prod }_p{p}^{n_{ip}^{\prime }},{\tau }_j^{\prime }={\prod }_p{p}^{m_{jp}^{\prime }},$$  where $i\in \{1,2,3,\ell ,4,5,6\}$  and $j\in \{1,2,\ell \}$  . Note that  3.6 translates to $(n_{ip},m_{jp})$  fulfilling $n_{3p}=n_{6p}=0$  or $m_{1p}=0$  , and that  3.7 means that $(n_{ip}^{\prime },m_{jp}^{\prime })$  must fulfill $n_{1p}^{\prime }+n_{3p}^{\prime }⩽1$  . Furthermore, $n_{3p},n_{3p}^{\prime }\in \{0,1\}$  always holds.

  Define the map

  $$\begin{array}{cccc}{\Phi }^{\prime }:& {\mathcal{T}}_2& \to & {\mathcal{T}}_1\end{array}$$

  $$\begin{array}{cccc}& ({\xi }_i^{\prime },{\tau }_i^{\prime })& \mapsto & ({\xi }_i,{\tau }_j),\end{array}$$

  where $n_{ip}:=n_{ip}^{\prime }$  for $i\in \{2,\ell ,4,5\}$  and $m_{jp}:=m_{jp}^{\prime }$  for $j\in \{2,\ell \}$  , and the values of $n_{1p},n_{3p},n_{6p},m_{1p}$  depend on the size of $n_{6p}^{\prime }$  compared to $m_{1p}^{\prime }$  , whether $m_{1p}^{\prime }$  is even or odd, and whether $n_{3p}^{\prime }$  is $0$  or $1$  :

  • ∙ If $m_{1p}^{\prime }=2k+1,n_{6p}^{\prime }⩾k+1,n_{3p}^{\prime }=0$  , then $$(n_{1p},n_{3p},n_{6p},m_{1p}):=(n_{1p}^{\prime }+3k+1,1,n_{6p}^{\prime }-k-1,m_{1p}^{\prime }-2k-1);$$ 
  • ∙ if $m_{1p}^{\prime }=2k+1,n_{6p}^{\prime }⩾k+1,n_{3p}^{\prime }=1$  or $n_{6p}^{\prime }=k,m_{1p}^{\prime }>2k,n_{3p}^{\prime }=1$  , then $$(n_{1p},n_{3p},n_{6p},m_{1p}):=(n_{1p}^{\prime }+3k+2,0,n_{6p}^{\prime }-k,m_{1p}^{\prime }-2k-1);$$ 
  • ∙ otherwise, with $n_{6p}^{\prime }=k,m_{1p}^{\prime }>2k,n_{3p}^{\prime }=0$  or $m_{1p}^{\prime }=2k,n_{6p}^{\prime }⩾k$  : $$(n_{1p},n_{3p},n_{6p},m_{1p}):=(n_{1p}^{\prime }+3k,n_{3p}^{\prime },n_{6p}^{\prime }-k,m_{1p}^{\prime }-2k).$$ 

  Conversely, define

  $$\begin{array}{cccc}\Phi :& {\mathcal{T}}_1& \to & {\mathcal{T}}_2\end{array}$$

  $$\begin{array}{cccc}& ({\xi }_i,{\tau }_i)& \mapsto & ({\xi }_i^{\prime },{\tau }_j^{\prime }),\end{array}$$

  where $n_{ip}^{\prime }:=n_{ip}$  for $i\in \{2,\ell ,4,5\}$  and $m_{jp}^{\prime }:=m_{jp}$  for $j\in \{2,\ell \}$  , and the values of $n_{1p}^{\prime },n_{3p}^{\prime },n_{6p}^{\prime },m_{1p}^{\prime }$  depend on $n_{1p}$  modulo $3$  and whether $n_{3p}$  is $0$  or $1$  :

  • ∙ If $n_{1p}\in \{3k+1,3k+2\}$  and $n_{3p}=1$  , then $$(n_{1p}^{\prime },n_{3p}^{\prime },n_{6p}^{\prime },m_{1p}^{\prime }):=(n_{1p}-3k-1,0,n_{6p}+k+1,m_{1p}+2k+1);$$ 
  • ∙ if $n_{1p}=3k+2$  and $n_{3p}=0$  , then $$(n_{1p}^{\prime },n_{3p}^{\prime },n_{6p}^{\prime },m_{1p}^{\prime }):=(n_{1p}-3k-2,1,n_{6p}+k,m_{1p}+2k+1);$$ 
  • ∙ otherwise, with $n_{1p}\in \{3k,3k+1\}$  : $$(n_{1p}^{\prime },n_{3p}^{\prime },n_{6p}^{\prime },m_{1p}^{\prime }):=(n_{1p}-3k,n_{3p},n_{6p}+k,m_{1p}+2k).$$ 

  It is not difficult to check that $\Phi$  and ${\Phi }^{\prime }$  are well-defined, that $({\xi }_i,{\tau }_j)\in \mathcal{T}(\mathbb{Z})$  and $({\xi }_i^{\prime },{\tau }_j^{\prime })\in \mathcal{T}(\mathbb{Z})$  correspond to the same point $\mathbf{x}\in U(\mathbb{Q})$  under the map $\Psi$  , and that $\Phi$  and ${\Phi }^{\prime }$  are inverse to each other.

Together, Lemma  4 and Lemma  5 prove Proposition  3 .

4 Congruences

We will use the following results from Chapter 3 of [?]  on the number of solutions of linear and quadratic congruences.

Let $\eta (a;q)$  be the number of positive integers $n⩽q$  such that $n^2\equiv a(modq)$  .

Then by equation (3.1) of [?] , we have for any $q\in \mathbb{Z}>0$ 

$$\begin{array}{c}\eta (a;q)⩽2^{\omega \left(q\right)},\end{array}$$

(4.1)

where $\omega \left(q\right)$  is the number of distinct prime factors of $q$  . Let $\vartheta$  be an arithmetic function such that $${\sum }_{d=1}^{\infty }\left|\right(\vartheta *\mu \left)\right(d\left)\right|<\infty ,$$  where $\vartheta *\mu$  is the usual Dirichlet convolution.

Lemma 6. Let $a,q\in \mathbb{Z}$  such that $q>0$  and $gcd(a,q)=1$  . Then $${\sum }_{n⩽tn\equiv a(modq)}\vartheta \left(n\right)=\frac{t}{q}{\sum }_{d=1gcd(d,q)=1}^{\infty }(\vartheta *\mu )\left(d\right)+O\left({\sum }_{d=1}^{\infty }\left|\right(\vartheta *\mu \left)\right(d\left)\right|\right).$$ 

• Proof. This is the case $\kappa =0$  of Lemma 2 of [?] .

Let $\psi \left(t\right)=\left\{t\right\}-1/2$  where $\left\{t\right\}$  is the fractional part of $t\in \mathbb{R}$  . Let $\tilde{\psi }\left(t\right)=\psi \left(t\right)+1$  for $t\in \mathbb{Z}$  and $\tilde{\psi }\left(t\right)=\psi \left(t\right)$  otherwise.

Lemma 7. Let $a,q\in \mathbb{Z}$  , where $q>0$  and $gcd(a,q)=1$  . Let $b_1,b_2\in \mathbb{R}$  with $b_1⩽b_2$  . Then $$\#\left\{n\right|b_1⩽n⩽b_2,n\equiv a(modq)\}=\frac{b_2-b_1}{q}+r(b_1,b_2;a,q),$$  where $$r(b_1,b_2;a,q)=\tilde{\psi }\left(\frac{b_1-a}{q}\right)-\psi \left(\frac{b_2-a}{q}\right).$$ 

• Proof. This is a slight generalization of Lemma 3 of [?] .

Lemma 8. Let $\epsilon >0$  and $t⩾0$  . Let $a,q\in \mathbb{Z}$  such that $q>0$  and $gcd(a,q)=1$  . Then $${\sum }_{0⩽\varrho <qgcd(\varrho ,q)=1}\psi \left(\frac{t-a{\varrho }^2}{q}\right)\ll \epsilon q^{1/2+\epsilon }\text{and}{\sum }_{0⩽\varrho <qgcd(\varrho ,q)=1}\tilde{\psi }\left(\frac{t-a{\varrho }^2}{q}\right)\ll \epsilon q^{1/2+\epsilon }.$$ 

• Proof. For $\psi$  , this is Lemma 5 of [?] .
  Note that if $t\equiv a{\varrho }_i^2(modq)$  for $i\in \{1,2\}$  , then ${\varrho }_1\equiv \pm {\varrho }_2(modq)$  , which implies that there are at most two different values for $\varrho$  with $0⩽\varrho <q$  such that $(t-a{\varrho }^2)/q$  is integral. Therefore, the sum for $\tilde{\psi }$  differs from the one for $\psi$  by at most $2$  .

5 Summations

Note that ${\tau }_{\ell }$  is determined uniquely by $T$  and the other variables once a certain congruence is fulfilled. Therefore, our strategy is first to compute the number of possible $\tau 2$  depending on $\tau 1,{\xi }_i$  such that there exists a unique $\tau \ell$  satisfying $T$  . By summing over $\tau 1$  , the number of possible ${\tau }_i$  is then computed depending on ${\xi }_i$  . The summation over the variables ${\xi }_i$  is finally handled using the height zeta function.

Let $$X_1=\left(B\xi \right(-2,0,-1,3,2,1,0){)}^{1/3},X_2=(B\xi (0,-1,0,3,2,1,0){)}^{1/2},$$  and $$X_0=\left(B^{-1}\xi \right(2,3,4,3,4,5,6){)}^{1/6}.$$  Then the height conditions $\left|x_i\right|⩽B$  lift to

$$\begin{array}{c}\left|\left(\frac{{\tau }_2}{X_2}\right)X_0^3\right|⩽1,\left|{\left(\frac{{\tau }_2}{X_2}\right)}^2+{\left(\frac{{\tau }_1}{X_1}\right)}^3\right|⩽1,\end{array}$$

(5.1)

and

$$\begin{array}{c}\left|X_0\right|⩽1,\end{array}$$

(5.2)

and

$$\begin{array}{c}\left|\left(\frac{{\tau }_1}{X_1}\right)X_0^4\right|⩽1,\end{array}$$

(5.3)

respectively.

Using Proposition  3 , we can now relate the counting of rational points of bounded height on $U\subset S$  to a count on the torsor.

Lemma 9. We have $$NU,H\left(B\right)=\#\left\{({\xi }_i,{\tau }_i)\in \mathcal{T}(\mathbb{Z})|\begin{array}{cc}& \text{<ref><sref>eq:hcf\_xi</sref><pref>3.3</pref></ref>,<ref><sref>eq:hcf\_tone</sref><pref>3.4</pref></ref>,<ref><sref>eq:hcf\_ttwo\_tell</sref><pref>3.5</pref></ref>,<ref><sref>eq:height\_ttwo</sref><pref>5.1</pref></ref>,<ref><sref>eq:height\_xi</sref><pref>5.2</pref></ref>,<ref><sref>eq:height\_tone</sref><pref>5.3</pref></ref> hold},\\ & {\xi }_i>0\end{array}\right\}.$$ 

5.1 Summation over $\tau 2$  and $\tau \ell$ 

Let ${\xi }_i,{\tau }_1$  satisfy the coprimality conditions  3.3 ,  3.4 and the height conditions  5.2 and  5.3 . Let $N=N({\xi }_i,{\tau }_1)$  be the number of pairs $(\tau 2,\tau \ell )$  such that  3.5 ,  5.1 and the torsor equation $T$  are fulfilled. Then a Mobius inversion gives $$N={\sum }_{k\ell |\xi 4\xi 5\xi 6}\mu (k\ell )N_{k\ell },$$  where $N_{k\ell }$  has the same definition as $N$  except that $gcd(\tau \ell ,\xi 4\xi 5\xi 6)=1$  is replaced by the condition $k\ell |\tau \ell$  , and $T$  is replaced by $$T_{k\ell }({\xi }_i,{\tau }_i,k\ell )=k\ell \tau \ell \xi 3\ell \xi 24\xi 5+\tau 22\xi 2+\tau 31\xi 21\xi 3=0.$$  Note that $\tau 2$  together with $T_{k\ell }$  defines $\tau \ell$  uniquely once a certain congruence is fulfilled. Therefore, $$N_{k\ell }=\#\left\{\tau 2\right|gcd(\tau 2,\xi 1\xi 3)=1,\text{<ref><sref>eq:height\_ttwo</sref><pref>5.1</pref></ref> holds},-\tau 22\xi 2\equiv \tau 31\xi 21\xi 3(modk\ell \xi 3\ell \xi 24\xi 5)\}.$$  Note that $$gcd(\tau 31\xi 21\xi 3,k\ell \xi 3\ell \xi 24\xi 5)=gcd(\tau 31\xi 21\xi 3,\xi 4\xi 5\xi 6,\tau 22\xi 2)=gcd(\tau 1,{\xi }_6,\tau 22\xi 2)=1$$  and $$gcd(\xi 2,k\ell \xi 3\ell \xi 24\xi 5)=gcd(\xi 2,k\ell )=gcd(\xi 2,\xi 4\xi 5\xi 6,\tau 31\xi 21\xi 3)=1.$$  Therefore, it is enough to sum over all $k\ell |\xi 4\xi 5\xi 6$  with $gcd(k\ell ,\tau 1{\xi }_1{\xi }_2{\xi }_3)=1$  , and since $k\ell |\xi 4\xi 5\xi 6$  implies $gcd(k\ell ,\tau 1)=1$  , we reduce to $gcd(k\ell ,{\xi }_1{\xi }_2{\xi }_3)=1$  .

This implies that there is a unique integer $\varrho$  satisfying $0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5$  and $gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1$  such that $$\tau 2\equiv \varrho \tau 1\xi 1(modk\ell \xi 3\ell \xi 24\xi 5)\text{and}-\varrho 2\xi 2\equiv \tau 1\xi 3(modk\ell \xi 3\ell \xi 24\xi 5).$$  We have $$N={\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(k\ell ,{\xi }_1{\xi }_2{\xi }_3)=1}\mu (k\ell ){\sum }_{0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5-\varrho 2\xi 2\equiv \tau 1\xi 3(modk\ell \xi 3\ell \xi 24\xi 5)gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1}{N}_{k\ell }\left(\varrho \right)$$  where $$N_{k\ell }\left(\varrho \right)=\#\left\{\tau 2\right|gcd(\tau 2,\xi 1\xi 3)=1,\text{<ref><sref>eq:height\_ttwo</sref><pref>5.1</pref></ref> holds},\tau 2\equiv \varrho \tau 1\xi 1(modk\ell \xi 3\ell \xi 24\xi 5)\}.$$  We also know that $gcd(\varrho \tau 1{\xi }_1,k\ell \xi 3\ell \xi 24\xi 5)=1$  . Now we can apply Lemma  6 to the characteristic function $\chi \left(n\right)=\{\begin{array}{cc}1,& \text{if}gcd(n,\xi 1\xi 3)=1\\ 0,& \text{else.}\end{array}$  Since $${\sum }_{d=1gcd(d,k\ell \xi 3\ell \xi 24\xi 5)=1}^{\infty }\frac{(\chi *\mu )\left(d\right)}{d}={\prod }_{p|\xi 1\xi 3p\mathbb{\nmid }k\ell \xi 3\ell \xi 24\xi 5}\left(1-\frac{1}{p}\right)={\prod }_{p|\xi 1\xi 3}\left(1-\frac{1}{p}\right)={\phi }^{*}\left(\xi 1\xi 3\right),$$  where we use $gcd(\xi 1\xi 3,k\ell \xi 3\ell \xi 24\xi 5)=1$  and the notation ${\phi }^{*}\left(n\right):=\phi \left(n\right)/n$  as in (5.10) of [?] , we conclude $$N_{k\ell }\left(\varrho \right)=\frac{{\phi }^{*}\left(\xi 1\xi 3\right)X_2}{k\ell \xi 3\ell \xi 24\xi 5}g1(\tau 1/X_1,X_0)+O\left(2^{\omega \left(\xi 1\xi 3\right)}\right),$$  where $X_{2}g1(\tau 1/X_1,X_0)$  gives the total length of the intervals in which $\tau 2$  must lie by  5.1 , with

$$\begin{array}{c}g1(u,v)={\int }_{\{t\in \mathbb{R}|\left|t{v}^3\right|⩽1,|t^2+u^3|⩽1\}}1dt.\end{array}$$

(5.4)

By equation  4.1 , the number of integers $\varrho$  with $0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5$  such that $gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1$  and $-\varrho 2\xi 2\equiv \tau 1\xi 3(modk\ell \xi 3\ell \xi 24\xi 5)$  is at most $$\eta ({\xi }_2{\xi }_3\tau 1;k\ell \xi 3\ell \xi 24\xi 5)⩽2^{\omega (k\ell \xi 3\ell \xi 24\xi 5)}⩽2^{\omega \left({\xi }_{\ell }{\xi }_4{\xi }_5{\xi }_6\right)}.$$  This gives as the first step towards the proof of Theorem  1 :

Lemma 10. $$N=\frac{X_2}{\xi 3\ell \xi 24\xi 5}g1(\tau 1/X_1,X_0)\Sigma ({\xi }_i,{\tau }_1)+O\left(2^{\omega \left(\xi 1\xi 3\right)+\omega \left({\xi }_{\ell }{\xi }_4{\xi }_5{\xi }_6\right)+\omega \left(\xi 4\xi 5\xi 6\right)}\right),$$  where $$\Sigma ({\xi }_i,{\tau }_1)={\phi }^{*}\left(\xi 1\xi 3\right){\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(k\ell ,{\xi }_1{\xi }_2{\xi }_3)=1}\frac{\mu (k\ell )}{k\ell }{\sum }_{0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5-\varrho 2\xi 2\equiv \tau 1\xi 3(modk\ell \xi 3\ell \xi 24\xi 5)gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1}1$$ 

Now we show that the error term suffices for Theorem  1 by summing it over all the ${\xi }_i,\tau 1$  which satisfy the height conditions  5.2 and  5.3 ; we can ignore the coprimality conditions  3.3 ,  3.4 ,  3.5 . We obtain:

$$\begin{array}{cccccccccc}& {\sum }_{{\xi }_i}{\sum }_{\tau 1}{2}^{\omega \left(\xi 1\xi 3\right)+\omega \left({\xi }_{\ell }{\xi }_4{\xi }_5{\xi }_6\right)+\omega \left(\xi 4\xi 5\xi 6\right)}& \ll & {\sum }_{{\xi }_i}{2}^{\omega \left(\xi 1\xi 3\right)+\omega \left({\xi }_{\ell }{\xi }_4{\xi }_5{\xi }_6\right)+\omega \left(\xi 4\xi 5\xi 6\right)}\frac{X_1}{X_0^4}& =& {\sum }_{{\xi }_i}{2}^{\omega \left(\xi 1\xi 3\right)+\omega \left({\xi }_{\ell }{\xi }_4{\xi }_5{\xi }_6\right)+\omega \left(\xi 4\xi 5\xi 6\right)}\frac{B}{\xi (2,2,3,1,2,3,4)}& \ll & B(logB{)}^2{\sum }_{{\xi }_i,i\ne \ell }\frac{2^{\omega \left(\xi 1\xi 3\right)+\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left(\xi 4\xi 5\xi 6\right)}}{\xi (2,2,3,0,2,3,4)}& \ll & B(logB{)}^2\end{array}$$

For $\xi \ell$  , we have used the estimate $${\sum }_{n⩽x}{2}^{\omega \left(n\right)}\ll x(logx)$$  together with partial summation.

Therefore, we only need to consider the main term when summing over ${\tau }_1,{\xi }_i$  in order to prove Theorem  1 .

5.2 Summation over $\tau 1$ 

For fixed ${\xi }_2,...,{\xi }_6$  satisfying  3.3 and  5.2 , we sum over all $\tau 1$  satisfying the coprimality condition  3.4 and the height condition  5.3 . Let $$N^{\prime }=N^{\prime }\left({\xi }_i\right)=\frac{X_2}{\xi 3\ell \xi 24\xi 5}{\sum }_{\tau 1,\text{<ref><sref>eq:height\_tone</sref><pref>5.3</pref></ref> holds}gcd(\tau 1,\xi 2\xi 3\xi \ell \xi 4\xi 5\xi 6)=1}g1(\tau 1/X_1,X_0)\Sigma ({\xi }_i,\tau 1).$$  First, we find an asymptotic formula for $$\mathcal{N}(b_1,b_2)={\phi }^{*}\left(\xi 1\xi 3\right){\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(k\ell ,{\xi }_1{\xi }_2{\xi }_3)=1}\frac{\mu (k\ell )}{k\ell }{\sum }_{0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1}{N}_{k\ell }^{\prime }(\varrho ,b_1,b_2),$$  where $$N_{k\ell }^{\prime }(\varrho ,b_1,b_2)=\#\left\{\tau 1\in [b_1,b_2]|\begin{array}{cc}& gcd(\tau 1,\xi 2\xi 3\xi \ell \xi 4\xi 5\xi 6)=1,\\ & -\varrho 2\xi 2\equiv \tau 1\xi 3(modk\ell \xi 3\ell \xi 24\xi 5)\end{array}\right\}.$$  We have $gcd({\varrho }^2{\xi }_2,k\ell \xi 3\ell \xi 24\xi 5)=1$  . Therefore we can replace the condition $gcd(\tau 1,\xi 2\xi 3\xi \ell \xi 4\xi 5\xi 6)=1$  by $gcd(\tau 1,{\xi }_2{\xi }_3{\xi }_6)=1$  in the definition of $N_{k\ell }^{\prime }(\varrho ,b_1,b_2)$  and perform another Mobius inversion to obtain

$$\begin{array}{cccc}\mathcal{N}(b_1,b_2)=& {\phi }^{*}\left(\xi 1\xi 3\right){\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(k\ell ,{\xi }_1{\xi }_2{\xi }_3)=1}\frac{\mu (k\ell )}{k\ell }{\sum }_{k_1|{\xi }_2{\xi }_3{\xi }_{6}gcd(k_1,k\ell \xi 3\ell \xi 24\xi 5)=1}\mu \left(k_1\right){\sum }_{0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1}& & N_{k\ell ,k_1}^{\prime }(\varrho ,b_1,b_2),\end{array}$$

where $$N_{k\ell ,k_1}^{\prime }(\varrho ,b_1,b_2)=\#\{\tau 1\in [b_1/k_1,b_2/k_1\left]\right|-{\varrho }^2{\xi }_2\equiv k_1\tau 1{\xi }_3(modk\ell \xi 3\ell \xi 24\xi 5)\}.$$  Note that we must only sum over the $k_1$  with $gcd(k_1,k\ell \xi 3\ell \xi 24\xi 5)$  because of $gcd({\varrho }^2{\xi }_2,k\ell \xi 3\ell \xi 24\xi 5)=1$  .

Let $a=a({\xi }_i,k_1,k\ell )$  be the unique integer such that $0<a⩽k\ell \xi 3\ell \xi 24\xi 5$  and $$-{\xi }_2\equiv k_{1}a{\xi }_3(modk\ell \xi 3\ell \xi 24\xi 5).$$  Then $-{\varrho }^2{\xi }_2\equiv k_1\tau 1{\xi }_3(modk\ell \xi 3\ell \xi 24\xi 5)$  if and only if $\tau 1\equiv a{\varrho }^2(modk\ell \xi 3\ell \xi 24\xi 5)$  .

Since $gcd({\xi }_2,k\ell \xi 3\ell \xi 24\xi 5)=1$  , we have $gcd(a,k\ell \xi 3\ell \xi 24\xi 5)=1$  . Using Lemma  7 , we conclude $$N_{k\ell ,k_1}^{\prime }(\varrho ,b_1,b_2)=\frac{b_2-b_1}{k_{1}k\ell \xi 3\ell \xi 24\xi 5}+r(b_1/k_1,b_2/k_1,a{\varrho }^2,k\ell \xi 3\ell \xi 24\xi 5)$$  where, by definition of $r$  , $$r(b_1/k_1,b_2/k_1,a{\varrho }^2,k\ell \xi 3\ell \xi 24\xi 5)=\tilde{\psi }\left(\frac{b_1/k_1-a{\varrho }^2}{k\ell \xi 3\ell \xi 24\xi 5}\right)-\psi \left(\frac{b_2/k_1-a{\varrho }^2}{k\ell \xi 3\ell \xi 24\xi 5}\right).$$  Let $$\vartheta =\vartheta \left({\xi }_i\right)=\{\begin{array}{cc}{\phi }^{*}\left(\xi 1\xi 3\right){\phi }^{*}(\xi 2\xi 3\xi \ell \xi 4\xi 5\xi 6)\frac{{\phi }^{*}\left(\xi 4\xi 5\xi 6\right)}{{\phi }^{*}(gcd(\xi 4\xi 5\xi 6,{\xi }_1{\xi }_2{\xi }_3\left)\right)},& \text{<ref><sref>eq:hcf\_xi</sref><pref>3.3</pref></ref> holds,}\\ 0,& \text{otherwise.}\end{array}$$  Then for any $b_1<b_2$  , we have $$\mathcal{N}(b_1,b_2)=\vartheta \left(\xi \right)\cdot (b_2-b_1)+\mathcal{ℛ}(b_1,b_2)$$  where

$$\begin{array}{cccc}\mathcal{ℛ}(b_1,b_2)=& {\phi }^{*}\left(\xi 1\xi 3\right){\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(\tau \ell ,\xi 4\xi 5\xi 6)=1}\frac{\mu (k\ell )}{k\ell }{\sum }_{k_1|{\xi }_2{\xi }_3{\xi }_{6}gcd(k_1,k\ell \xi 3\ell \xi 24\xi 5)=1}\mu \left(k_1\right){\sum }_{0<\varrho ⩽k\ell \xi 3\ell \xi 24\xi 5gcd(\varrho ,k\ell \xi 3\ell \xi 24\xi 5)=1}& & r(b_1/k_1,b_2/k_1,a{\varrho }^2,k\ell \xi 3\ell \xi 24\xi 5).\end{array}$$

By partial summation, we obtain $$N^{\prime }\left({\xi }_i\right)=\frac{\vartheta \left({\xi }_i\right)X_1{X}_2}{\xi 3\ell \xi 24\xi 5}g2\left(X_0\right)+R^{\prime }\left({\xi }_i\right)$$  with

$$\begin{array}{c}g2\left(v\right)={\int }_{\{u\in \mathbb{R}|\left|u{v}^4\right|⩽1\}}g1(u,v)du\end{array}$$

(5.5)

and $$R^{\prime }\left({\xi }_i\right)=\frac{-X_2}{\xi 3\ell \xi 24\xi 5}{\int }_{-X_0^{-4}}^{X_0^{-4}}\left(D_{1}g1\right)(u,X_0)\mathcal{ℛ}(-X_1/X_0^4,X_{1}u)du$$  where $D_{1}g1$  is the derivation of $g1$  with respect to the first variable.

Lemma 11. For any ${\xi }_i$  as in  3.3 ,  5.2 , we have $$N^{\prime }\left({\xi }_i\right)=\frac{\vartheta \left({\xi }_i\right)X_1{X}_2}{\xi 3\ell \xi 24\xi 5}g2\left(X_0\right)+R^{\prime }\left({\xi }_i\right)$$  where the error term $R^{\prime }\left({\xi }_i\right)$  satisfies $${\sum }_{{\xi }_i,\text{<ref><sref>eq:hcf\_xi</sref><pref>3.3</pref></ref>,<ref><sref>eq:height\_xi</sref><pref>5.2</pref></ref> holds}}{R}^{\prime }\left({\xi }_i\right)=O(BlogB).$$ 

• Proof. By Lemma  8 , we have

  $$\begin{array}{cccccc}\mathcal{ℛ}(b_1,b_2)& \ll \epsilon {\phi }^{*}\left(\xi 1\xi 3\right){\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(\tau \ell ,\xi 4\xi 5\xi 6)=1}\frac{\left|\mu \right(k\ell \left)\right|}{k\ell }{\sum }_{k_1|{\xi }_2{\xi }_3{\xi }_{6}gcd(k_1,k\ell \xi 3\ell \xi 24\xi 5)=1}\left|\mu \right(k_1\left)\right|(k\ell \xi 3\ell \xi 24\xi 5{)}^{1/2+\epsilon }& & ⩽{\sum }_{k\ell |\xi 4\xi 5\xi 6gcd(\tau \ell ,\xi 4\xi 5\xi 6)=1}\left|\mu \right(k\ell \left)\right|{\sum }_{k_1|{\xi }_2{\xi }_3{\xi }_{6}gcd(k_1,k\ell \xi 3\ell \xi 24\xi 5)=1}\left|\mu \right(k_1\left)\right|(\xi 3\ell \xi 24\xi 5{)}^{1/2+\epsilon }& & ⩽2^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}(\xi 3\ell \xi 24\xi 5{)}^{1/2+\epsilon }\end{array}$$

  Therefore,

  $$\begin{array}{cccc}{R}^{\prime }\left({\xi }_i\right)& \ll \epsilon \frac{X_2}{\xi 3\ell \xi 24\xi 5}{\int }_{-X_0^{-4}}^{X_0^{-4}}\left(D_{1}g1\right)(u,X_0)2^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}(\xi 3\ell \xi 24\xi 5{)}^{1/2+\epsilon }du& & \ll \frac{X_2}{(\xi 3\ell \xi 24\xi 5{)}^{1/2-\epsilon }}{2}^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}\end{array}$$

  Summing this over all ${\xi }_i⩽B$  , we get using  5.2 

  $$\begin{array}{cccccccc}{\sum }_{{\xi }_i⩽B\text{<ref><sref>eq:hcf\_xi</sref><pref>3.3</pref></ref>,<ref><sref>eq:height\_xi</sref><pref>5.2</pref></ref> hold}}{R}^{\prime }\left({\xi }_i\right)& \ll \epsilon {\sum }_{{\xi }_i⩽B}\frac{X_2}{X_0^3(\xi 3\ell \xi 24\xi 5{)}^{1/2-\epsilon }}{2}^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}& & ={\sum }_{{\xi }_i⩽B}\frac{B}{\xi (1,2,2,3/2-3\epsilon ,2-2\epsilon ,5/2-\epsilon ,3)}{2}^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}& & \ll {\sum }_{{\xi }_i,i\ne 1}\frac{BlogB}{\xi (0,2,2,3/2-3\epsilon ,2-2\epsilon ,5/2-\epsilon ,3)}{2}^{\omega \left({\xi }_4{\xi }_5{\xi }_6\right)+\omega \left({\xi }_2{\xi }_3{\xi }_6\right)}& & \ll BlogB\Lambda \end{array}$$

5.3 Summation over ${\xi }_i$ 

Define $$\Delta \left(n\right)=B^{-5/6}{\sum }_{{\xi }_i,\xi (2,3,4,3,4,5,6)=n}\frac{\vartheta \left({\xi }_i\right)X_1{X}_2}{\xi 3\ell \xi 24\xi 5}.$$  We sum $N^{\prime }\left({\xi }_i\right)$  in Lemma  11 over the seven variables ${\xi }_i$  such that the coprimality conditions  3.3 and the height condition  5.2 hold. Note that the definition of $\vartheta \left({\xi }_i\right)$  ensures that the main term of $N^{\prime }\left({\xi }_i\right)$  is zero if  3.3 is not satisfied. In view of Lemma  9 , this implies:

Lemma 12. We have $$NU,H\left(B\right)=B^{5/6}{\sum }_{n⩽B}\Delta \left(n\right)g2\left(\right(n/B{)}^{1/6})+O(B(logB{)}^2).$$ 

6 Proof of Manin’s Conjecture for $S$ 

Our argument is similar to [?] . We need to estimate $$M\left(t\right):={\sum }_{n⩽t}\Delta \left(t\right)$$  for $t>1$  . Therefore, we consider the Dirichlet series $F\left(s\right):={\sum }_{n=1}^{\infty }\Delta \left(n\right)n^{-s}$  .

Observing $$\frac{X_1{X}_2}{\xi 3\ell \xi 24\xi 5}=\frac{B^{5/6}\left(\xi \right(2,3,4,3,4,5,6){)}^{1/6}}{\xi (1,1,1,1,1,1,1)},$$  we get $$F(s+1/6)={\sum }_{{\xi }_i}\frac{\vartheta \left({\xi }_i\right)}{{\xi }_1^{2s+1}{\xi }_2^{3s+1}{\xi }_3^{4s+1}{\xi }_{\ell }^{3s+1}{\xi }_4^{4s+1}{\xi }_5^{5s+1}{\xi }_6^{6s+1}},$$  and writing $F(s+1/6)={\prod }_p{F}_p(s+1/6)$  as a product of local factors, we obtain:

$$\begin{array}{cccccc}{F}_p(s+1/6)=& 1+\frac{(1-1/p{)}^2}{(p^{{\lambda }_{6}s+1}-1)}(\frac{p^{{\lambda }_{1}s+1}}{p^{{\lambda }_{1}s+1}-1}+\frac{p^{{\lambda }_{1}s+1}{p}^{{\lambda }_{6}s+1}}{p^{{\lambda }_{3}s+1}(p^{{\lambda }_{1}s+1}-1)}& & +\frac{p^{{\lambda }_{6}s+1}}{(1-1/p)p^{{\lambda }_{2}s+1}}+\frac{1}{p^{{\lambda }_{\ell }s+1}-1}+\frac{p^{{\lambda }_{\ell }s+1}{p}^{{\lambda }_{6}s+1}}{p^{{\lambda }_{4}s+1}(p^{{\lambda }_{\ell }s+1}-1)}& & +\frac{p^{{\lambda }_{\ell }s+1}{p}^{{\lambda }_{6}s+1}}{p^{{\lambda }_{5}s+1}(p^{{\lambda }_{\ell }s+1}-1)})+\frac{1-1/p}{p^{{\lambda }_{1}s+1}-1}+\frac{1-1/p}{p^{{\lambda }_{\ell }s+1}-1}\end{array}$$

for any prime $p$  .

Since $1/p^{{\lambda }_{i}s+1}=O_{\epsilon }(1/p^{1/2+\epsilon })$  for $s\in \mathcal{\mathscr{H}}:=\{s\in \mathbb{C}|\Re e\left(s\right)⩾-1/12+\epsilon \}$  and $i\in \{1,2,3,\ell ,4,5,6\}$  , we have $$F_p(s+1/6)=1+{\sum }_i\frac{1}{p^{{\lambda }_{i}s+1}}+O_{\epsilon }\left(\frac{1}{p^{1+\epsilon }}\right)$$  for $s\in \mathcal{\mathscr{H}}$  , and defining $$E\left(s\right):={\prod }_i\zeta ({\lambda }_{i}s+1)=\zeta (2s+1)\zeta (3s+1{)}^2\zeta (4s+1{)}^2\zeta (5s+1)\zeta (6s+1),$$  we have $$\frac{1}{E_p\left(s\right)}=1-{\sum }_i\frac{1}{p^{{\lambda }_{i}s+1}}+O_{\epsilon }\left(\frac{1}{p^{1+\epsilon }}\right)$$  for $s\in \mathcal{\mathscr{H}}$  . Define $$G\left(s\right):=F(s+1/6)/E\left(s\right)$$  for $\Re e\left(s\right)>0$  . Then $G$  has a holomorphic and bounded continuation to $\mathcal{\mathscr{H}}$  .

Note that $$G\left(0\right)={\prod }_p{\left(1-\frac{1}{p}\right)}^7\left(1+\frac{7}{p}+\frac{1}{p^2}\right),$$  and that for $s\to 0$  , we have $$E\left(s\right)=\frac{1}{{\prod }_i{\lambda }_i}{s}^{-7}+O\left(s^{-6}\right).$$  Therefore, the residue of $F\left(s\right)t^s/s$  at $s=1/6$  is $$Res\left(t\right)=\frac{6G\left(0\right)t^{1/6}{Q}_1(logt)}{6!\cdot {\prod }_i{\lambda }_i}$$  for some monic polynomial $Q_1$  of degree $6$  .

Lemma 13. $M\left(t\right)={\omega }_0\alpha \left(\tilde{S}\right)\cdot 6t^{1/6}{Q}_1(logt)+O_{\epsilon }\left(t^{1/6-1/24+\epsilon }\right)$  .

• Proof. Integrating Perron’s formula for $M\left(t\right)$  over $t$  , we have $${\int }_0^{t}M\left(u\right)du=\frac{1}{2\pi i}{\int }_{1/6+\epsilon -i\infty }^{1/6+\epsilon +i\infty }F\left(s\right)\frac{t^{s+1}}{s(s+1)}ds$$  for $t>1$  and $\epsilon >0$  .
  We apply Cauchy’s residue theorem to the rectangle with vertices $$1/12+\epsilon -iT,1/12+\epsilon +iT,1/6+\epsilon +iT,1/6+\epsilon -iT,$$  for some $T>1$  , where $\epsilon >0$  is sufficiently small.

  By the convexity bound $$\zeta (1+\sigma +i\tau )\ll \epsilon |\tau {|}^{-\sigma /3+\epsilon }$$  for any $\sigma \in [-1/2,0)$  , we have

  $$\begin{array}{c}F(1/6+\sigma +i\tau )\ll E(\sigma +i\tau )\ll \epsilon |\tau {|}^{-9\sigma +\epsilon }\end{array}$$

  (6.1)

  for any $\sigma \in [-1/12+\epsilon ,0)$  , using $\left({\sum }_i{\lambda }_i\right)/3=9$  and that $G(\sigma +i\tau )$  is bounded.

  For the ray going down from $1/6+\epsilon -iT$  , we get

  $$\begin{array}{cccccc}\left|{\int }_{1/6+\epsilon -i\infty }^{1/6+\epsilon -iT}\frac{F\left(s\right)t^{s+1}}{s(s+1)}ds\right|& ⩽{\int }_{-\infty }^T\frac{\left|F\right(1/6+\epsilon +i\sigma \left)\right|\left|t^{7/6+\epsilon +i\sigma }\right|}{\left|\right(1/6+\epsilon +i\sigma \left)\right(7/6+\epsilon +i\sigma )}d\sigma & & \ll t^{7/6+\epsilon }{\int }_{-\infty }^T\frac{1}{|\sigma {|}^2}d\sigma & & \ll t^{7/6+\epsilon }{T}^{-1}\end{array}$$

  where we use that $F\left(s\right)$  is bounded for $\Re e\left(s\right)⩾1/6+\epsilon$  . Integrating from $1/6+\epsilon +iT$  to $1/6+\epsilon +i\infty$  gives the same result.

  For the lower edge, we estimate

  $$\begin{array}{cccc}\left|{\int }_{1/12+\epsilon -iT}^{1/6+\epsilon -iT}\frac{F\left(s\right)t^{s+1}}{s(s+1)}ds\right|& ⩽{\int }_{-1/12+\epsilon }^{\epsilon }\frac{\left|F\right(1/6+\sigma -iT\left)\right|\left|t^{7/6+\sigma -iT}\right|}{\left|\right(1/6+\sigma -iT\left)\right(7/6+\sigma -iT\left)\right|}d\sigma & & \ll \epsilon \frac{T^{9/12+\epsilon }{t}^{7/6+\epsilon }}{T^2},\end{array}$$

  because  6.1 gives a bound for $-1/12+\epsilon ⩽\sigma ⩽-\epsilon$  , $F\left(s\right)$  being continuous gives a bound in an $\epsilon$  -neighborhood of $1/6-iT$  , and the length of the integration interval is $1/12$  . For the upper edge, we obtain the same bound. For the edge on the left, we have

  $$\begin{array}{cccccc}\left|{\int }_{1/12+\epsilon -iT}^{1/12+\epsilon +iT}\frac{F\left(s\right)t^{s+1}}{s(s+1)}ds\right|& ⩽{\int }_{-T}^T\frac{\left|F\right(1/12+\epsilon +i\sigma \left)\right|\left|t^{13/12+\epsilon +i\sigma }\right|}{\left|\right(1/12+\epsilon +i\sigma \left)\right(13/12+\epsilon +i\sigma )}d\sigma & & \ll \epsilon {\int }_{-T}^T\frac{|\sigma {|}^{9/12+\epsilon }{t}^{13/12+\epsilon }}{(1+|\sigma |{)}^2}d\sigma & & \ll t^{13/12+\epsilon }\end{array}$$

  since the integral over $\sigma$  is bounded independently of $T$  , and using  6.1 again.

  Taking $T=t$  , we have proved $${\int }_0^{t}M\left(u\right)du={\int }_0^{t}Res\left(u\right)du+O_{\epsilon }\left(t^{13/12+\epsilon }\right).$$  But now $$\frac{1}{H}{\int }_{t-H}^{t}M\left(u\right)du⩽M\left(t\right)⩽\frac{1}{H}{\int }_t^{t+H}M\left(u\right)du,$$  and for $H⩽t/3$  , both integrals are equal to $$Res\left(t\right)+O_{\epsilon }\left(H{t}^{5/6}\right(logt{)}^6+H^{-1}{t}^{13/12+\epsilon }).$$  The proof of the Lemma is completed by choosing $H=t^{23/24}$  and noting that ${\omega }_0=G\left(0\right)$  and $\alpha \left(\tilde{S}\right)=(6!{\prod }_i{\lambda }_i{)}^{-1}$  by the definitions of ${\omega }_0$  and $\alpha \left(\tilde{S}\right)$  in Lemma  2 .

By partial summation we conclude

$$\begin{array}{cccccc}& {\sum }_{n⩽B}\Delta \left(n\right)g2\left(\right(n/B{)}^{1/6})& & ={\omega }_0\alpha \left(\tilde{S}\right)\cdot 6{\int }_0^{B}g2(u^{1/6}/B^{1/6})\frac{d}{du}\left(u^{1/6}{Q}_1\right(logu\left)\right)du+O_{\epsilon }\left(B^{1/6-1/24+\epsilon }\right)& & =B^{1/6}{\omega }_0\alpha \left(\tilde{S}\right)\cdot 6{\int }_0^{1}g2\left(v\right)Q_2(logB+6logv)dv+O_{\epsilon }\left(B^{1/6-1/24+\epsilon }\right)\end{array}$$

for some monic polynomial $Q_2$  of degree $6$  . Considering definitions  5.4 and  5.5 , note that $${\omega }_{\infty }=6{\int }_{\{v\in \mathbb{R}|0⩽v⩽1\}}g2\left(v\right)dv.$$  Together with Lemma  12 , this completes the proof of Theorem  1 .