Counting torsion points on subvarieties of the algebraic torus

Let $n\geq 1$ be an integer and $K$ be an algebraically closed field. We denote by $\mathbb{G}_{m}^{d}$ the $n$-dimensional algebraic torus with base field $K$. We will identify $\mathbb{G}_{m}^{n}(K)$ with $(K\setminus\{0\})^{n}$, the group of its $K$-points. We denote by $(\mathbb{G}_{m}^{n})_{\mathrm{tors}}$ the subgroup of points of finite order therein, which we call torsion points. A linear torus $G<\mathbb{G}_{m}^{n}$ is an irreducible algebraic subgroup and if $\boldsymbol{\zeta}\in(\mathbb{G}_{m}^{n})_{\mathrm{tors}}$ is a torsion point and $G$ is a linear torus, we call $\boldsymbol{\zeta}G$ a torsion coset. We often identify a subvariety of $\mathbb{G}_{m}^{n}$ with the set of its $K$-points.

Lukas Fink considered the following question in his master thesis [Fin08]. Let $X=\{(x,y)\in\mathbb{G}_{m}^{2}(\overline{\mathbb{F}}_{2}):x+y=1\}$ and $c(T)=\#\{\mathbf{x}\in X:\mathrm{ord}(\mathbf{x})\leq T\}$. What can we say about the asymptotic growth of $c(T)$ as $T\to\infty$? Considering the first coordinate yields $c(T)\leq\#\{x\in\mathbb{G}_{m}:\mathrm{ord}(x)\leq T\}\ll T^{2}$. He shows in Proposition 2.7 that $c(T)=o(T^{2})$. Fink’s result is therefore better than the trivial bound. Based on computations he conjectures that $c(T)\ll T$. Our first result, Theorem 1.1 below, includes a power saving over the trivial bound $c(T)\ll T^{2}$.

We consider the average rather than the number of torsion points of order exactly $n$ since the latter behaves erratic. In Fink’s example the set of $n\in\mathbb{N}$ such that there is no torsion point of order $n$ has natural density one. The number of points on Fink’s curve whose order divides $n$ is also bounded by a constant times $n^{1-1/560}$ if (4) in [Moh20] holds for the set of points $x\in\mathbb{F}_{2^{\varphi(n)}}$ such that $x^{n}=1=(x+1)^{n}$. On the other hand, this number is $n-1$ for all $n=2^{k}-1$.

An algebraic subset $X\subset\mathbb{G}_{m}^{n}(K)$ is the zero set of finitely many Laurent polynomials $P_{1},\dots,P_{r}\in K[X_{1}^{\pm 1},\dots,X_{n}^{\pm 1}]$ in $\mathbb{G}_{m}^{n}(K)$. The natural generalisation of the question of Fink is as follows. Take an algebraic set $X\subset\mathbb{G}_{m}^{n}(K)$, and denote by $X_{T}=\{\boldsymbol{\zeta}\in X\cap(\mathbb{G}_{m}^{n})_{\mathrm{tors}}:\mathrm{ord}(\boldsymbol{\zeta})\leq T\}$ the set of torsion points in $X$ of order bounded by $T$. How fast does $\#X_{T}$ grow?

In characteristic zero, the Manin-Mumford Conjecture for $\mathbb{G}_{m}^{n}$, a theorem of Laurent [Lau84] reduces this problem to the case where $X$ is a torsion coset as we will see below in Corollary 1.5. Altough many of our results are true in any characteristic, we highlight the case where $K$ is an algebraic closure of a finite field. In this case all points of $\mathbb{G}_{m}^{n}$ have finite order. To illustrate the kind of results we get and compare them to the work of Fink, we look at the curve defined by $x+y=1$:

To state the main result, we have to introduce more terminology. An algebraic subset $X\subset\mathbb{G}_{m}^{n}$ is called admissible if it is finite or does not contain a top dimensional torsion coset. The stabilizer $\mathrm{Stab}(X)$ of an algebraic subset $X\subset\mathbb{G}_{m}^{n}$ is the set of $\mathbf{g}\in\mathbb{G}_{m}^{n}$ such that $\mathbf{g}X\subset X$. It is well known that the stabilizer is an algebraic subgroup of $\mathbb{G}_{m}^{n}$. The main result is

As in the example there is an analogue of the trivial bound given by

So again we have a power saving with respect to the general bound.

If $C$ is a torsion coset, we can give an asymptotic formula whose main term depends only on the characteristic of the field, the dimension and the order $\mathrm{ord}(C)=\min\{\mathrm{ord}(\boldsymbol{\eta}):\boldsymbol{\eta}\in C\cap(\mathbb{G}_{m}^{n})_{\mathrm{tors}}\}$. In the theorem below $\zeta$ denotes the Riemann zeta function.

Theorem 1.4 implies that the exponent $d+1$ in Proposition 1.3 is optimal. We also see that torsion cosets are those subvarieties which in some sense contain the most torsion points if we compare among varieties of the same dimension.

Recall that in characteristic zero the Manin-Mumford Conjecture for $\mathbb{G}_{m}^{n}$ reduces the general case to torsion cosets and again we are able to give an asymptotic formula. Let $X\subset\mathbb{G}_{m}^{n}$ be an algebraic set. There exist torsion cosets $C_{1},\dots,C_{m}$ such that $\overline{X\cap(\mathbb{G}_{m}^{n})_{\mathrm{tors}}}=\bigcup_{i=1}^{m}C_{i}$ is the decomposition into irreducible components. Set

Then we have

What about lower bounds for $\#X_{T}$? Fink proved in Proposition 2.7 in [Fin08] that $c(T)\geq T/2-2$ for all $T\in\mathbb{N}$. Using the famous theorem of Lang-Weil, we can prove

Hence if $X$ is admissible, the correct exponent, if it exists, is in $[d,d+1-\frac{1}{d-\delta+1}]$, where $\delta=\dim(\mathrm{Stab}(X))$. In the example of Fink, numerical analysis suggests that the smallest exponent $d=1$ is possible.

Let us give two examples. First we look at the curve $C=\{(x,y)\in\mathbb{G}_{m}^{2}:x+y=1\}$ and prove Theorem 1.1. Let $T\geq 1$ and assume that $\boldsymbol{\zeta}=(\zeta,\xi)\in C_{T}$ is of order $N$. By Minkowski’s first theorem there exists $\mathbf{a}=(a,b)\in\mathbb{Z}^{2}\setminus\{0\}$ such that $\boldsymbol{\zeta}^{\mathbf{a}}:=\zeta^{a}\xi^{b}=1$ and $|\mathbf{a}|:=\max\{|a|,|b|\}\leq N^{1/2}$. Since $\xi=1-\zeta$ we have $\zeta^{a}(1-\zeta)^{b}=1$. After multiplying by suitable powers of $\zeta$ and $1-\zeta$ we see that $\zeta$ is a root of a nonzero polynomial of degree at most $|a|+|b|$. Hence the number of $\boldsymbol{\zeta}\in C$ such that $\boldsymbol{\zeta}^{\mathbf{a}}=1$ is at most $2|\mathbf{a}|$. Thus we have the bound $\#C_{T}\leq 2\sum_{|\mathbf{a}|\leq T^{1/2}}|\mathbf{a}|$. For a fixed $k\geq 1$ there are $4(2k+1)-4=8k$ vectors $\mathbf{a}\in\mathbb{Z}^{2}$ such that $|\mathbf{a}|=k$. Thus we find

Now let us consider $Y=\{(x,y,z)\in\mathbb{G}_{m}^{3}:x+y=1\}$. Note that Stab$(Y)=\{\mathbf{1}\}\times\mathbb{G}_{m}$ is one dimensional. Hence the exponent in Theorem 1.2 equals $5/2$. Then we observe that any $(\zeta_{1},\zeta_{2},\zeta_{3})\in Y\cap(\mathbb{G}_{m}^{3})_{\mathrm{tors}}$ is contained in the torsion coset $\{\zeta_{1}\}\times\{\zeta_{2}\}\times\mathbb{G}_{m}\subset Y.$ The order of such a torsion coset is given by $\mathrm{ord}(\zeta_{1},\zeta_{2})$. Thus any $\boldsymbol{\zeta}\in Y_{T}$ is contained in a torsion coset $\{\boldsymbol{\xi}\}\times\mathbb{G}_{m}$ for some $\boldsymbol{\xi}\in C_{T}$. Therefore we can bound $\#Y_{T}\leq\sum_{\boldsymbol{\xi}\in C_{T}}\#(\{\boldsymbol{\xi}\}\times\mathbb{G}_{m})_{T}$. Denote by $g$ the order of $\boldsymbol{\xi}\in C_{T}$. We claim that $\#(\{\boldsymbol{\xi}\}\times\mathbb{G}_{m})_{T}\leq T^{2}/g$. If $\zeta\in\mathbb{G}_{m}$ is of order $n$, the order of $(\boldsymbol{\xi},\zeta)$ is the least common multiple of $g$ and $n$. Thus $(\boldsymbol{\xi},\zeta)\in(\{\boldsymbol{\xi}\}\times\mathbb{G}_{m})_{T}$ if and only if $n\leq\gcd(g,n)T/g$. The number of elements of order $n$ in $\mathbb{G}_{m}$ is given by Euler’s totient $\varphi(n)$ except if the characteristic $p>0$ divides $n$. Since in positive characteristic the order is always coprime to $p$, there are no elements of order $n$, whenever $p$ divides $n$. Thus in any case there are at most $\varphi(n)$ elements of order $n$ in $\mathbb{G}_{m}$. We therefore have

If $d|g$ and $n$ is as in the inner sum, we can write $n=kd$ for some $1\leq k\leq T/g$. And since $\varphi(n)=n\prod_{p\mid n}(1-p^{-1})$ we see that $\varphi(ab)\leq a\varphi(b)$ for all $a,b\in\mathbb{N}$. It is also well known, that $\sum_{d|n}\varphi(d)=n$ for all $n\in\mathbb{N}$. Therefore we get

Let $a_{n}$ denote the number of elements of order $n$ in $C$. Then we can bound $\#Y_{T}\leq T^{2}\sum_{n=1}^{T}a_{n}/n$ and we know that $\sum_{n=1}^{N}a_{n}\leq 16N^{3/2}$ for every $N\geq 1$. To get sums of this kind, we write $1/n=1/T+\sum_{k=n}^{T-1}(\frac{1}{k}-\frac{1}{k+1})=1/T+\sum_{k=n}^{T-1}\frac{1}{k(k+1)}$ for all $n<T$. Thus we find

using that $\int x^{-1/2}dx=2x^{1/2}$ and hence $Y_{T}\leq 48T^{5/2}$.

Now let us consider a hypersurface $X\subset\mathbb{G}_{m}^{n}$. We divide $X$ into two parts: the union of all torsion cosets contained in $X$ of positive dimension and its complement $X^{*}$. To bound $\#X^{*}_{T}$ we generalise the proof for $C$: Instead of Minkowski’s first theorem we can use the second theorem to find short linearly independent vectors $\mathbf{a}_{1},\dots,\mathbf{a}_{n-1}\in\mathbb{Z}^{n}$ such that $\boldsymbol{\zeta}^{\mathbf{a}_{i}}=1$. The $\mathbf{x}\in\mathbb{G}_{m}^{n}$ such that $\mathbf{x}^{\mathbf{a}_{i}}=1$ for all $i=1,\dots,n-1$ are given by torsion cosets $\{\boldsymbol{\xi}t^{\mathbf{u}^{\top}}:t\in\mathbb{G}_{m}\}$ for some $\mathbf{u}\in\mathbb{Z}^{n}\setminus\{0\}$ and finitely many $\boldsymbol{\xi}\in(\mathbb{G}_{m}^{n})_{\mathrm{tors}}$; we refer to Section 2 for our notation. Let $P\in K[X_{1}^{\pm 1},\dots,X_{n}^{\pm 1}]$ be such that $X$ is the vanishing locus of $P$ in $\mathbb{G}_{m}^{n}$. Then $\boldsymbol{\xi}t^{\mathbf{u}^{\top}}$ is in $X$ if and only if $Q(t)=P(\boldsymbol{\xi}t^{\mathbf{u}^{\top}})=0$. If $Q$ is not the zero polynomial, the number of such $t$ is bounded in terms of the degree of $P$ and $|\mathbf{u}|$. This is always the case if $\boldsymbol{\xi}t^{\mathbf{u}^{\top}}\in X^{*}$. Here we use the fact that $X$ is a hypersurface. Summing over all possible $(n-1)$-tuples $\mathbf{a}_{1},\dots,\mathbf{a}_{n-1}$ gives us a bound $X^{*}_{T}\ll_{X}T^{n-1/n}$.

For the union of positive dimensional torsion cosets we show that if we consider the maximal cosets, then the number of corresponding algebraic groups is finite and every torsion coset in $X$ is contained in one of them. However in positive characteristic it is possible that for a fixed linear torus $G$ there are infinitely many torsion cosets $\boldsymbol{\zeta}G$ which are contained in $X$. So let us fix $G$. To control the number of cosets $\boldsymbol{\zeta}G$ which are contained in $X$, we map $\boldsymbol{\zeta}G$ to $\boldsymbol{\zeta}^{L}$ for some matrix $L$ such that $\mathbf{g}^{L}=\mathbf{1}$ for all $\mathbf{g}\in G$. Note that $\mathrm{ord}(\boldsymbol{\zeta}^{L})\leq\mathrm{ord}(\boldsymbol{\zeta}G)$. We can find equations for the image and use induction on $n$. Thus we have a bound for the number of cosets of $G$ contained in $X$. Together with a bound for cosets we get the claimed result for hypersurfaces.

The generalisation to arbitrary algebraic sets follows by projecting $X$ to some coordinates and doing algebraic geometry.

For Theorem 1.4 we can assume that $G=\{\boldsymbol{\zeta}\}\times\mathbb{G}_{m}^{d}$ for some $\boldsymbol{\zeta}\in(\mathbb{G}_{m}^{n-d})_{\mathrm{tors}}$ of order $g$, say. Since the order of $(\boldsymbol{\zeta,\xi})$ is $\mathrm{lcm}(g,\mathrm{ord}(\boldsymbol{\xi}))=\frac{g}{(g,\mathrm{ord}(\boldsymbol{\xi}))}\mathrm{ord}(\boldsymbol{\xi})$, we have to count the $\boldsymbol{\xi}\in(\mathbb{G}_{m}^{d})_{\mathrm{tors}}$ whose order is bounded and belongs to a fixed congruence class modulo $g$. With the Möbius inversion formula one can see that the number of elements of order $n$ is given by Jordan’s totient $J_{d}(n)=\mu\star I_{d}(n)$, where $\mu$ is the Möbius function and $I_{d}(n)=n^{d}$. Once we have the asymptotics of $J_{d}$ over arithmetic progressions, Theorem 1.4 is not hard to deduce. In positive characteristic the orders are always coprime to $p$. If $n$ is coprime to $p$, we still have $J_{d}(n)$ elements of order $n$. Thus the case $p>0$ follows from the case $p=0$.

The proofs of Corollary 1.5 and Theorem 1.6 are short and straight forward.

The algebraic subgroups of $\mathbb{G}_{m}^{n}$ play an important role in the results as well as in the proofs. Therefore we need a characterisation of them. This is already known in characteristic zero (Theorem 3.2.19 in [BG06]). In positive characteristic it is possibly known. But we could not pin down a suitable reference and have therefore added an appendix.

In section 3 we prove Theorem 1.4.

The next goal is to estimate $\#X^{*}_{T}$. The most technical part is to bound the degree of $Q(t)$. Some known preparations are done in the second appendix and used to prove the bound in section 4.

For the proof of Theorem 1.2 when $X$ is a hypersurface, we first deal with common roots and admissibility in section 5, then we establish an explicit bound when $X$ is a coset. In section 7 we give the proofs of Theorems 1.2 and 1.3 when $X$ is a hypersurface.

Finally we are able to prove Theorems 1.2 and 1.3 in full generality in the next section.

In the last two sections we establish Corollary 1.5 and Theorem 1.6.

I want to thank my advisor Philipp Habegger for good discussions and advice, and Pierre le Boudec for his help concerning the sum of Jordan’s totient in arithmetic progressions. This will be part of my PhD thesis. I have received funding from the Swiss National Science Foundation grant number 200020_184623.