1 Introduction and the main result
Given an integer $Q\ge 1$ , we consider the congruence subgroup
where
We are interested in counting matrices $A\in \Gamma _0(Q) $ with entries of size at most
The question is a natural generalization of the classical counting result of Newman [Reference Newman10] concerning matrices $A \in \operatorname {SL}_2({\mathbb Z})$ with
and of Krieg [Reference Krieg9] who counts matrices $A \in \operatorname {SL}_2({\mathbb Z})$ with respect to the $L^\infty $ -norm as (1.1). We note that both of these results correspond to $Q=1$ .
We note that while we can also use the $L^2$ -norm as in (1.2) to measure the “size” of $A\in \operatorname {SL}_2({\mathbb Z})$ , for us it is more convenient to use the $L^\infty $ -norm as in (1.1). However, our main purpose to have an asymptotic formula in a broad range of uniformity with respect to the size of Q compared to X.
Let
The question of investigating the cardinality $\# \Gamma _0(Q, X)$ has been raised in [Reference Bulinski, Ostafe and Shparlinski3], where it is also shown that for $Q\le X$ , we have
We are interested in obtaining an asymptotic formula for the cardinality $\# \Gamma _0(Q,X)$ in a broad range of Q and X. Furthermore, our bound on error term relies on some results of Ustinov [Reference Ustinov14], which go beyond standard techniques.
We first give an asymptotic formula for $\# \Gamma _0(Q, X) $ with the main term expressed via sums of some standard arithmetic functions. For this, we also define
where
where as usual $\varphi (k)$ denotes the Euler function.
Theorem 1.1 Uniformly over an integer $Q\ge 1$ and a positive real $X\ge Q$ , we have
Next, we study the function $F(Q,X)$ . As indicated to us by one the referees, the sum $F_2(Q,X)$ has already been computed in [Reference Suryanarayana13]. When Q is fixed a much more general result is given in [Reference Shapiro11, Theorem 5.5A.1]. We have not, however, been able to locate references for an asymptotic formula for $F_1(Q,X)$ with the desired level of uniformity in Q, so we derive one in this paper (see 4.4). For this, we first recall the definition of the Dedekind function
Theorem 1.2 Uniformly over an integer $Q\ge 1$ and a positive real $X\ge Q$ , we have
Combining Theorems 1.1 and 1.2, we obtain the following asymptotic formula.
Corollary 1.3 Uniformly over an integer $Q\ge 1$ and a positive real $X\ge Q$ ,
We remark that the appearance of the Dedekind function $\psi (Q)$ in the denominator of the asymptotic formula for $\# \Gamma _0(Q, X)$ in Corollary 1.3 is not surprising as function itself appears in as the index of $\Gamma _0(Q)$ in $\operatorname {SL}_2({\mathbb Z})$ , that is,
(see [Reference Iwaniec8, Proposition 2.5]).
Elementary estimates easily show that $\psi (Q) = Q^{1+o(1)}$ . Thus, Corollary 1.3 is nontrivial in an essentially full range of Q and X, namely for $Q \le X^{1-\varepsilon }$ for a fixed $\varepsilon> 0$ .
2 Preparations
2.1 Notation and some elementary estimates
We recall that the notations $U = O(V)$ , $U \ll V$ and $ V\gg U$ are equivalent to $|U|\leqslant c V$ for some positive constant c, which throughout this work, is absolute.
Furthermore, we write $U \asymp V$ to express that $V \ll U \ll V$ .
We also write $U= V^{o(1)}$ if for all $\varepsilon>0$ , there exists a constant $c(\varepsilon )>0$ such that $ |U| \leq c(\varepsilon ) V^\varepsilon $ as $V\to \infty $ .
The letter p always denotes a prime number.
For an integer $k \ge 1$ , we denote by $\mu (k)$ , $\tau (k),$ and $\varphi (k)$ , the Möbius function, the number of integer positive divisors, and the Euler function of k, respectively, for which we use the well-known bound
as $k \to \infty $ (see [Reference Hardy and Wright6, Theorems 317 and 328]).
As usual, we define
For positive integers u and v, using the Möbius function $\mu (e)$ and the inclusion–exclusion principle to detect the co-primality condition and then interchanging the order of summation, we obtain
(see [Reference Hardy and Wright6, Equation (16.1.3)]).
2.2 Modular hyperbolas
Here, we need some results on the distribution of points on the modular hyperbola
where $q \ge 1$ is an arbitrary integer.
We start with a very well-known case counting the number $N(q;U,V)$ of solutions in a rectangular domain $(u,v) \in [1, U]\times [1,V]$ . For example, such a result has been recorded in [Reference Shparlinski12, Theorem 13] (we note that the restriction $U,V\le q$ is not really necessary).
Lemma 2.1 For any $U,V\ge 1$ , we have
Next, we recall a result of Ustinov [Reference Ustinov14] on the number $T_f(q; Z,U)$ of points $(u,v)$ on the modular hyperbola (2.3) with variables run through a domain of the form
where f is a positive function with a continuous second derivative.
Namely, a special case of [Reference Ustinov14], where we have also used (2.1) to estimate various divisor sums, can be formulated as follows.
Let
and let
Lemma 2.2 Assume that the function $f:{\mathbb R}\to {\mathbb R}_{\geq 0}$ has a continuous second derivative on $[Z,Z+U]$ such that for some $L>0 $ , we have
Then we have the estimate
For other results on the distribution of points on modular hyperbolas, we refer to the survey [Reference Shparlinski12] and also more recent works [Reference Baier1, Reference Browning and Haynes2, Reference Chan4, Reference Garaev and Shparlinski5, Reference Humphries7, Reference Ustinov15].
3 Proof of Theorem 1.1
3.1 Separating contributions to the main term and to the error term
It is easy to see that there are only $O(X)$ matrices in $\operatorname {SL}_2({\mathbb Z}; X)$ with $abcd=0$ . We now consider the following eight sets for different choices of the signs of a, $c,$ and d:
with $\alpha , \gamma , \delta \in \{-1,1\}$ .
Now observe that $\Gamma _0(Q,X)$ is preserved under the bijections
and
This means
and
for all pairs $\alpha ,\gamma \in \{-1,1\}$ .
Thus
3.2 Preliminary counting of $\Gamma _0^{1, 1, 1 }(Q,X)$
Writing $cQ$ instead of c, we need to count the number of solutions to the equation
We first do this for a fixed c and then sum up over all $c \le X/Q$ .
First, we consider the values $a \le cQ$ . We note that setting
for a solution $(a,d)$ to the congruence
we have $b \le X$ . Hence, we see from Lemma 2.1 (and then recalling that $cQ \le X$ ) that for every $c \in [1, X/Q]$ , there are
such matrices
Next, we count the contribution $G_{2}(c)$ from matrices $A\in \Gamma _0^{1,1,1}(Q,X)$ with $a> cQ$ . To do this, we recall the notation of Section 2.2 and then parameterize this set using a modular hyperbola as follows.
Lemma 3.1 Fix $1 \leq c \leq X/Q$ , $0 < U \leq X - cQ$ and define
Then the map
given by
is well-defined, injective and its image is exactly the set of those $A\in \Gamma _0^{1,1,1}(Q,X)$ with $cQ < a \leq cQ + U$ and bottom left entry equal to $cQ$ .
Proof For $(x,y) \in {\mathcal T}_{f_c}(cQ,cQ,U)$ , we have that $(xy-1)/cQ \in \mathbb {Z}$ and
which is equivalent to
As $x> cQ \geq 1$ and $y>0$ , this is actually equivalent to
We also need to check that $1 \leq y \leq X$ . This follows since
Thus, indeed, $(x,y)$ is mapped to an element of $\Gamma _0^{1,1,1}(Q,X)$ with the desired properties. Conversely, suppose that $A \in \Gamma _0^{1,1,1}(Q,X)$ with $a>cQ$ and bottom left entry equal to $cQ$ . As $ad \equiv 1 \ \ \pmod {cQ}$ , we have $1 \leq x,y \leq X$ such that
Also by definition (the lower bound holds as $x>cQ\geq 1$ )
which means
and so indeed $(x,y) \in \mathcal {T}(f_c, cQ, U)$ .
We partition the interval $(cQ,X]$ into $I \ll \log X$ dyadic intervals of the form $(Z_i, Z_i+U_i]$ with
(in fact $U_i = Z_i$ , except maybe for $i=I$ ) and note that
We now write
where $f_c(x)$ is as in Lemma 3.1.
Next, for each $i =1, \ldots , I$ , we use Lemma 2.2 with $q = cQ$ and use that
for $x \in (Z_i, Z_i+U_i]$ . Therefore, we conclude that
where
Combing the main terms $M_{i}(c)$ , $i =1, \ldots , I$ , together and recalling (3.4), we obtain
where
and
Recalling (3.3) and using $cQ \le X$ , we obtain
which after the substitution in (3.6) yields
3.3 Asymptotic formula for $\Gamma _0^{1, 1, 1 }(Q,X)$
From the equations (3.2) and (3.7), we obtain
where
and
We also note that
Change the order of summation, we write
Hence, recalling (2.2), we derive that
Thus, we see from (3.8) that
3.4 Counting $\Gamma ^{-1,1,1}(Q,X) $
Recalling (3.1), we see that it remains to count $\Gamma _0^{-1, 1, 1 }(Q,X)$ . One can use a similar argument, but in fact, we show that
where the error term $\mathbf {E} = O(X^{5/3+o(1)} Q^{-1} )$ is the same as obtained above.
Thus, we wish to count matrices of the form
where $xy \equiv 1 \ \ \pmod {cQ}$ , $-X\leq x \leq -1$ , $1 \leq y \leq X$ , $1 \leq cQ \leq X$ and $-X \leq (xy-1)/cQ \leq -1$ .
Without loss of generality, we can assume that $X \not \in {\mathbb Z}$ . Then, we consider the following two cases.
Case I: $x> -cQ$ . Note that for any $x,y$ with $xy \equiv 1 \ \ \pmod {cQ}$ , $-cQ < x \leq -1$ and $1 \leq y \leq X$ , we have
and so
Thus, indeed, the corresponding A is in $\Gamma _0^{-1,1,1}(Q,X)$ . Note that since $0<x + cQ \leq cQ$ and $-X \leq (xy-1)/cQ + y \leq X $ , we have that
So in fact, the number of such matrices A is exactly $G_1(c)$ as computed in (3.2) in the $\Gamma _0^{1,1,1}(Q,X)$ case.
Case II: $-X<x \leq -cQ$ . Let
We now need an analogue of Lemma 3.1. While the argument is very similar to that of the proof of Lemma 3.1, there are some differences, so we prefer to present it in full detail.
Lemma 3.2 Fix $1 \leq c \leq X/Q$ , $0 < U \leq X - cQ$ . Then the map
given by
is well-defined, injective and its image is exactly the set of those $A\in \Gamma _0^{-1,1,1}(Q,X)$ with $-X < x \leq -X + U$ and bottom left entry equal to $cQ$ .
Proof Let $(x,y) \in \mathcal {T}_{\widetilde {f}_c}(cQ, -X, U) $ . Thus, by definition,
As $x<-cQ$ , we have that
and so indeed $y \leq X$ . Moreover, as $x<0$ , we have
So indeed this mapping has range inside $\Gamma _0^{-1,1,1}(Q,X)$ . Conversely, suppose
is in $\Gamma _0^{-1,1,1}(Q,X)$ with $-X <x \leq -X + U$ . Then $-X \leq b \leq 0$ is an integer, thus ${xy \equiv 1 \ \ \pmod {cQ}}$ and
Thus, as $x<0,$ we have
Thus
and so indeed $(x,y) \in \mathcal {T}_{\widetilde {f}_c}(cQ, -X, U)$ as desired.
We now fix c with $1 \leq c \leq X/Q$ and observe now that by Lemma 3.2, for any $Z \in [-X,0)$ and $0 < U \leq |Z|$ , we have that $T_{\widetilde {f}_c}(cQ, Z, U)$ has the main term
where we recall $f_c(x) = (cQX-1)/x$ as used in Lemma 3.1. But this is precisely the same main term as for $T_{f_c}(cQ, |Z| - U, U)$ except for the boundary terms ( $x = -Z-U, -Z$ ) which contribute only $O(X)$ (uniformly in Q as $|(cQX-1)/x \leq (cQX-1)/cQ \leq X$ ). Thus, recalling (3.4), (3.5), and (3.6), we see that for each admissible c, we obtain the contribution to $\# \Gamma ^{-1,1,1}(Q,X)$ , which is asymptotic to $G_2(c)$ . Now observe that $\widetilde {f}_c(x) = - f_c(x)$ and so $|\widetilde {f}^{\prime \prime }_c(x)| = |f^{\prime \prime }_c(-x)|$ which means that the error terms we obtain from applying Lemma 2.2 to $\widetilde {f}_c$ are the same as those obtained for $f_c$ (we have $x \in [-X, -cQ]$ and before we had $x \in [cQ, X]$ ). Thus, if we sum over c and proceed as before, we see that the error term is at most $O\left (\mathbf {E} + X\right )$ which implies (3.10).
4 Proof of Theorem 1.2
4.1 Approximating $F_1(Q,X)$
For convenience, we let
So
We now define the function
Lemma 4.1 We have
Proof Observe that for any integer $n \ge 1$ ,
Hence
Thus, we derive
which completes the proof.
We now see from Lemma 4.1 that
Using that
we write
Note that
Thus, we see from (4.2) that
and so by (4.1), we derive
4.2 Approximating $F_2(Q,X)$
We can now easily recover an estimate for $F_2(Q,X)$ originally derived in [Reference Suryanarayana13]. We do this for the sake of completeness as [Reference Suryanarayana13] is not easily available. Let
be the characteristic function of the set of integer multiplies of an integer $d\ne 0$ . Then
We can now use (4.3) and then the multiplicativity of $\psi (d)$ to obtain
since
where $\omega (Q)$ is the number of prime divisors of Q.
A simple computation shows that
Therefore
Therefore, using that $2^{\omega (Q)} \le \tau (Q) = Q^{o(1)}$ , we obtain
whence $\psi (Q) \ge Q$ .
5 Comments
We presented our result, Corollary 1.3 as a direct consequence of Theorems 1.1 and 1.2 of very different nature with error terms of different strength. This makes it apparent that Theorem 1.1 is the bottleneck to further improvements of Corollary 1.3.
The methods of this work can also be used for counting elements of bounded norm of other congruence subgroup such as
and
One can also adjust our approach to counting matrices of restricted size with respect to other natural matrix norms.
Acknowledgments
The authors would like to thank Régis de la Bretèche for his suggestion which has led to the proof of Theorem 1.2 with the current error term. The authors are also very grateful to the referees for the very careful reading of the manuscript and very useful comments.