COUNTING RATIONAL POINTS OF A GRASSMANNIAN

arXiv:1908.01245v5 [math.NT] 16 Dec 2020

SEUNGKI KIM

Abstract. We prove an estimate on the number of rational points on the Grassman-

nian variety of bounded twisted height, refining the classical result of Thunder ([21])
over the rational field: most importantly, our formula counts all points. Among the
consequences are a couple of new implications on the classical subject of counting
rational points on flag varieties.

1. Introduction
1.1. Main result. For L ⊆ Rn a lattice of full rank, d ∈ {1, 2, . . . , n − 1} and H > 0, let
P (L, d, H) be the number of primitive rank d sublattices of L of determinant less than
or equal to H. The purpose of this paper is to investigate the quantitative behavior of
P (L, d, H). The earliest result of this kind goes back to the mid-twentieth century, due
to W. Schmidt ([13]):
Theorem 1.1 (Schmidt [13]). Let
  d
1 n Y V (n − i + 1) ζ(i)
a(n, d) = · ,
n d i=1 V (i) ζ(n − i + 1)
 
1 1
b(n, d) = max , ,
d n−d
where V (i) := π i/2 /Γ(i/2 + 1) is the volume of the unit ball in Ri and ζ(s) is the Riemann
zeta function, except that we understand ζ(1) = 1 for convenience. Then
P (Zn , d, H) = a(n, d)H n + O(H n−b(n,d) ).
P (L, d, H) may also be interpreted as the number of the rational points on the Grass-
mannian variety Gr(n, d) whose twisted height by L is less than or equal to H — see
Thunder ([20], [21]) for more on the notion of the twisted height, in which it is first in-
troduced. From this perspective, Thunder ([21]) proved a vast generalization of Theorem
1.1 above, extending it to any lattice and to any number field K (where OK -modules play
the role of lattices). His result, from the 1990’s, remains state-of-the-art to this day. We
state his result in case K = Q:
Theorem 1.2 (Thunder [21]). In addition to the notations in Theorem 1.1, define
L(|−i) = L ∩ spanR (v1 , . . . , vn−i ),
where v1 , . . . , vn are choices of linearly independent vectors in L such that kvi k = λi (L).
Let PL(|−d) (L, d, H) be the number of rank d sublattices of L of determinant ≤ H whose
intersection with L(|−d) is trivial. Then

Hn H n−b(n,d)
 
PL(|−d) (L, d, H) = a(n, d) +O ,
(det L)d (det L)d−b(n,d)(det L(|−d) )b(n,d)
1
2 SEUNGKI KIM

where the implicit constant is dependent only on n.

An important feature of Theorem 1.2 is that it reflects the “skewness” of the lattice, i.e.
it describes the dependence of the error term on the successive minima λ1 (L), . . . , λn (L)
of L (note det L(|−i) ∼ λ1 λ2 . . . λn−i by Minkowski’s second theorem).
However, Thunder does not provide an estimate for P (L, d, H), remarking that it would
“be a cumbersome task.” In the present paper, we introduce a method that circumvent
this difficulty, and prove
Theorem 1.3. Continue with the notations in Theorems 1.1 and 1.2 above. Then for all
H > 0,
 
n
H X
(1.1) P (L, d, H) = a(n, d) +O bγ (L)H γ  ,
 
(det L)d γ∈Q
0≤γ<n

where the implied constant depends only on n, and the sum on the right is finite but the
same element of Q may appear multiple times. Each bγ is a reciprocal of a product of
det L(|−i) ’s. In particular, the leading error term may be written as
H n−b(n,d)
(1.2) ,
(det L)d−b(n,d) (det L(|−d) )b(n,d)
as in Theorem 1.2. Also bγ (cL) = c−dγ bγ (L) for any c > 0, so that the formula (1.1)
is scale-invariant i.e. it remains unchanged if L and H are replaced by cL and cd H,
respectively.
Remark. One could of course put bγ (L) = (λ1 (L))−dγ , if an explicit formula is desired.
For the most refined expression, one may chase the proof to unearth a recursive formula
for bγ — cf. Section 1.3 below. However, it seems to be a rather laborious task which
does not yield a pretty formula.
Before we go on to discuss a few applications of Theorem 1.3, let us present a few of
its variants that may also be of use.
Corollary 1.1. Let N (L, d, H) be the number of (not necessarily primitive) rank d sub-
lattices of L of determinant ≤ H. Also let
d
Y
c(n, d) = a(n, d) ζ(n − i + 1).
i=1
Then similarly to Theorem 1.3, we have
 
n
H X
N (L, d, H) = c(n, d) +O bγ (L)H γ  ,
 
(det L)d γ∈Q
0≤γ<n

where the highest-degree term in the sum equals (1.2) for d ≤ n − 2. For d = n − 1, the
secondary term has degree n − 1 + η for any η > 0.
Corollary 1.2. Choose a primitive sublattice S ⊆ L of rank ≤ n − d. Then the number
PS (L, d, H) of primitive rank d sublattices of L whose intersection with S is trivial satisfies
the estimate  
n
H X
PS (L, d, H) = a(n, d) +O bγ (L)H γ  ,
 
(det L)d γ∈Q
0≤γ<n
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 3

where again the implied constant depends on n only, and the sum over γ is finite, with
the leading degree n − b(n, d).
The analogous statement applies for NS (L, d, H), with its expected definition.
1.2. Applications. Below we demonstrate a few immediate applications of Theorem 1.3
and the techniques used in its proof. Its main strength lies in the fact that it counts
all the sublattices, and that it provides information regardless of how skewed the given
lattice is, in particular relative to H. In comparison, its precedent Theorem 1.2 misses
the sublattices contained in L(|−d) , and thus it does not say anything about the lattices
for which det L/ det L(|−d) > H.
We expect there to be more uses of Theorem 1.3; for instance, see a recent work of Le
Boudec ([7]), which employs the d = 1 case (due to Schmidt ([13]), see (2.1) below) as the
main device.
1.2.1. Rational points of flag varieties. It is natural to expect that a counting formula on
Grassmannians should yield a counting formula for general flag varieties. Indeed, Thunder
([21]) derives such a formula as a relatively simple application of Theorem 1.2. We present
its simplest case to initiate the discussion:
Theorem 1.4 (Thunder ([21])). Let L be a lattice of rank n, and suppose H is sufficiently
large. Then the number of rational flags Se ⊆ Sd ⊆ L of type (e, d) (hence rk Si = i) such
(|−e)
that Si ∩ L(|−i) = 0 for i ∈ {e, d}, and Se ∩ Sd = 0, whose height twisted by L is at
most H is
 
aH H H
(1.3) log +O ,
(det L)d (det L)d (det L)d−b(n,d)(n−d)/n (det L(|−d) )b(n,d)
where a is some explicit constant depending only on n, d, e, and the implicit constant in
the error depends only on n.
In this context, the height of the flag Se ⊆ Sd is the quantity (det Se )d (det Sd )n−e ; see
e.g. Thunder ([21]) for details.
In comparison, we can derive from Theorem 1.3 the following
Corollary 1.3. Let L be a lattice of rank n, and H be sufficiently large. Then the number
(|−e)
of rational flags Se ⊆ Sd ⊆ L of type (e, d) such that Se ∩ Sd = 0 whose height twisted
by L is at most H is
 
aH H X
(1.4) log d n−e + O  bγ (L)H γ  ,
 
(det L)d εe εd γ∈Q
0≤γ≤1

where εe = min{det X : X ⊆ L, rk X = e} and likewise for εd , a is the same as in (1.3),

the implicit constant depends on n only, and bγ (L)’s are appropriate inverse products of
the successive minima of L.
Furthermore, the largest γ is 1, and the next largest is either
 
b(n, d) b(d, e)(n − e) 1 2b(d, e) 1 − b(d, e)
1− , 1− , or 1 − 1− + ;
n nd n d n−e
if d ≤ n/2, it is always 1 − b(n, d)/n = 1 − 1/dn.
In order to keep the proof relatively short and simple, we had to keep some of the
assumptions made by Theorem 1.4. Still, it has a couple of new implications that may
be of interest. First, it gives an insight into the size of the gap between Theorem 1.4 and
4 SEUNGKI KIM

an ideal counting formula that would count all the rational flags: it must be at least of
size O(H log εde εdn−e /(det L)d ), if not more. Depending on one’s perspective, this gap is
either negligible — which seems to have been the common wisdom, see e.g. [9] — or not
so small. Below we will describe a situation in which it turns out to be not small.
The second implication has to do with the error term in the well-known theorem of
Franke, Manin, and Tschinkel ([3]) on the number of rational points on flag varieties. In
the corollary to Theorem 5 therein, which says that the number of rational points on a
flag variety V of (“untwisted”) height bounded by H is
Hp(log H) + o(H),
where p is a polynomial of degree rk Pic(V ) − 1, they conjecture that the error term is of
size O(H 1−ε ) with ε = (dim V )−1 . On the other hand, when V is a Grassmannian, the
literature (for instance Schmidt [13], and Thunder [20] [21]) suggests that ε = b(n, d)/n,
as their analyses seem fairly sharp. Our Corollary 1.3 extends this to flag varieties of
type (e, d), suggesting that we have ε = b(n, d)/n again, at least when d ≤ n/2. In case
e ≥ n/2, one may be able to estimate ε = 1/(n − e)n by duality. But in general the nature
of ε appears rather complicated.
One can of course prove an analogue of Corollary 1.3 for a flag variety of any type that
is strong enough to yield these same implications, by essentially the same argument. On
the other hand, we expect the ideal formula that would count all points on a flag variety
to require another substantial amount of effort along the lines of the present paper. One
potential approach would be to investigate if every bγ in (1.1) contains sufficiently many
factors of det L, which would allow our error-bounding techniques to apply immediately.

1.2.2. Mean value theorems over lattices. In fact, the author’s original motivation for
proving Theorem 1.3 is to extend a family of mean value theorems such as the Siegel inte-
gral formula ([18]) and the Rogers integral formula ([11]) to the counting of d-dimensional
sublattices, in the hope that we could extend the statistical study of the vectors of ran-
dom lattices (see e.g. [5], [16], [19]) to that of the rational points on Grassmannians with
random twisted height. To illustrate what a mean value theorem looks like, we cite a
special case of Rogers’ formula below. It is a generalization of the celebrated formula by
Siegel ([18]), which corresponds to the case k = 1.

Theorem 1.5 (Rogers [11]). For k < n, and let f : (Rn )k → R be a Borel measurable
and compactly supported function. Then
(1.5)
Z Z Z
X
f (x1 , . . . , xk )dµ(L) = ... f (x1 , . . . , xk )dx1 . . . dxk ,
SL(n,Z)\SL(n,R) x1 ,...,xk ∈L Rn Rn
independent

where dµ is the normalized Haar measure on SL(n, R).

There are many variants of Theorem 1.5, in which the sum in the left-hand side of (1.5)
is subject to various conditions, or in which k ≥ n — see [15] for a list of these variants.
Recently, in another work ([6]), the author generalized Theorem 1.5 to Grassmannians:
Theorem 1.6 (Kim [6]). Define
(
1 if det A ≤ H
fH (A) =
0 otherwise.
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 5

Also suppose 1 ≤ k ≤ n − 1, 1 ≤ d1 , . . . , dk ≤ n − 1 with d1 + . . . + dk ≤ n − 1. Then

Z X X Y k
... fH1 (A1 ) . . . fHk (Ak )dµn (L) = a(n, di )Hin .
Xn A ∈Gr(L,d ) Ak ∈Gr(L,dk ) i=1
1 1
A1 ,...,Ak independent

We note that Thunder ([22]) proved the k = 1 case of this result.

There exist several approaches to formulas of this kind. The method introduced in [6]
uses Theorem 1.3, or more specifically Corollary 1.2, as a crucial ingredient. In case k = 2
for example, we first estimate
X X X
(1.6) fH1 (det A)fH2 (det B) = fH1 (det A) fH2 (det B),
A,B∈Gr(L,d) A B
A∩B={0} A∩B={0}

and then we show how to translate a counting formula into a mean-value formula, by
applying a discrete analogue of Theorem 1 in Rogers ([11]), which has some semblance to
the Hecke equidistribution. In fact, Corollary 1.2 is designed with precisely this application
in mind, namely to deal with the inner sum on the right-hand side of (1.6).
For general flag varieties, it turns out that the corresponding mean value formula
diverges to infinity, even for k = 1: this is a corollary to Theorem 5 of [6]. The machinery
of [6] suggests that this has to do with the fact that the main term of the pertinent
counting formula depends on the skewness of L, as we learn from Corollary 1.3 above.
1.3. Method of proof. All previous works on this topic ([13], [20], [21]) count “upwards,”
i.e. they construct the d-dimensional sublattice from either a (d−1)-dimensional sublattice
or a d-dimensional sublattice lying in an (n − 1)-dimensional ambient space. Our main
idea is to take the dual approach, and count “downwards” instead: we project all the
d-dimensional sublattices to a hyperplane, and count the cardinality of each fiber. This
lets us bypass some of the technical difficulties that arise when counting upwards.
To elaborate, we prove Theorem 1.3 by the following inductive procedure that resembles
the Pascal’s triangle method of computing the binomial coefficients. In case d = 1 or
d = n − 1, the formulas are well-known. Otherwise, let L̄ be the projection of L onto the
orthogonal complement of a shortest nonzero vector of L. Then we have
H
(1.7) P (L, d, H) = P (L̄, d − 1, ) + Φ(P (L̄, d, H)),
λ1 (L)
where Φ can be regarded as a certain integral transformation. For a sublattice B ⊆ L of
rank d, let us say B is of d-type (α1 , . . . , αn ) — “d” stands for “dual” — if the projection
of B onto spanR (v1 , . . . , vn−i+1 )⊥ has rank αi . Then the first term on the right-hand side
of (1.7) is counting the sublattices of d-types (∗, . . . , ∗, d − 1, d), and the second term is
counting those of d-types (∗, . . . , ∗, d, d).
In comparison, Theorem 1.2 counts precisely the sublattices of d-type (1, 2, . . . , d, . . . , d).
The upward counting method forces one to count each d-type separately, which is pre-
cisely what Thunder refers to as being “cumbersome.” The downward method completely
resolves this difficulty.
Most of this paper is devoted to explicitly writing out and estimating Φ(P (L̄, d, H)).
Many parts of the computation can be done by slightly refining the methods of Schmidt
([13]) or Thunder ([21]). However, the fact that L can be arbitrarily skewed presents a new
difficulty, especially when bounding the error terms. We manage this by comparing the
successive minima of both L and the sublattices being counted to H 1/d : if λi+1 −λi ≫ H 1/d
for any i, we exploit this gap to finesse the desired error bound, and if there is no such i, it
6 SEUNGKI KIM

actually becomes much easier to handle. This forms another main technical contribution
of the present paper.

1.4. Definitions and notations. Unless mentioned otherwise:

• The lowercase letter p denotes a prime.
• As in the statement of Theorem 1.2, L(|−i) denotes the primitive (n−i)-dimensional
sublattice containing v1 , . . . , vn−i ∈ L, which are linearly independent with kvi k =
λi (L). For convenience, we also sometimes write L(i) := L(|−(n−i)) .
• By abuse of language, we identify a basis {v1 , . . . , vd } of a lattice M ⊆ Rn with
the d × n matrix whose i-th row equals vi , and refer to this matrix as M as well.
When we make this abuse, either the basis of M is chosen in the context, or the
discussion is independent of the choice of a basis.
• By the same token, if a matrix M is given, we identify it with the lattice spanned
by its row vectors, which we also denote by M .
• A d × n integral matrix X ∈ Matd×n (Z) is primitive if X can be completed to
an element of GL(n, Z). When d = 1, this agrees with the standard notion of a
primitive vector. We denote the set of all primitive d × n matrices by Matpr d×n (Z).
• We write Γ = GL(d, Z). For a lattice L of rank n, we write Gr(L, d) = Γ\(Matpr d×n (Z)·
L). √
• For a non-square matrix X, we define det X = det XX tr . For E ∈ Gr(L, d),
pr
det E = det Y , where Y ∈ Matd×n (Z) · L is any representative of E.
• Following Schmidt ([13]), if M ⊆ Rn is a lattice of rank m, we define the polar
lattice M P of M by M P = {w ∈ R ⊗ M : hv, wi ∈ Z, ∀v ∈ M }. If S ∈ Gr(M, d),
we define its orthogonal lattice S ⊥ ∈ Gr(M P , m − d) by S ⊥ = {w ∈ M P : hv, wi =
0, ∀v ∈ M }.
• The (i, j)-entry of a matrix is denoted by the lowercase of the name of the matrix
indexed by ij. For example, if A is a d× n matrix, then A = (aij ) 1≤i≤d . Similarly,
1≤j≤n
if x ∈ Rn , then the i-th entry of x is denoted by xi .
• Later, given a d × (n − 1) matrix A and a d × 1 vector v, we need to consider the
d × n matrix B whose i-th row equals (ai1 , . . . , ai,n−1 , vi ). In this case, we denote
B = (A; v).
• For two quantities f and g, f ≪ g means f < Cg, where C is a positive constant
possibly depending on d and n but no other variables. f ∼ g meansQf ≪ g and
g ≪ f . For example, Minkowski’s second theorem says that det L ∼ λi (L).
For two matrices A and B with d rows, A ∼ B means they differ by the left
multiplication by an element of Γ, i.e. they represent the same element in the
Grassmannian.

1.5. Organization. In Section 2, we state the known formulas for P (L, 1, H) and P (L, n−
1, H). In Section 3, we set up the induction argument, establishing the precise version of
(1.7). Sections 4 and 5 are devoted to the main and error term estimates, respectively.
Section 6 collects all the computations and concludes the proof of Theorem 1.3. The
corollaries are all proved in Section 7.

1.6. Acknowledgment. Part of this work is supported by NSF grant CNS-2034176. The
author thanks Lillian Pierce and Jeffery Thunder for helpful comments and suggestions.
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 7

2. Base cases
In case d = 1, Theorem 1.3 is precisely Theorem 4 in [21] (also Lemma 2 of [13]), which
states that
n
!
Hn X H n−i
(2.1) P (L, 1, H) = a(n, 1) +O .
det L i=1
det L(|−i)

Below in Lemma 3.6, we provide a proof of an extension of (2.1) to an affine lattice,

which we will need later.
In case d = n − 1, we apply the duality theorem (see [21]) to (2.1), which says that, for
a sublattice S ⊆ L and its orthogonal lattice S ⊥ ⊆ LP ,
det S
det S ⊥ =
det L
holds, and thus
H
(2.2) P (L, d, H) = P (LP , n − d, ).
det L
Therefore (2.1) implies
n
!
Hn X H n−i
P (L, n − 1, H) = a(n, n − 1) + O .
(det L)n det LP i=1
(det L)n−i det(LP )(|−i)

By the well-known facts that det L · det LP = 1 and λi (L)λn−i (LP ) ≥ 1 (see e.g. [10]),
we have
(2.3) det(LP )(|−i) ≫ det L(|−(n−i)) / det L,
so we can rewrite the above as
n
!
Hn X H n−i
P (L, n − 1, H) = a(n, n − 1) · + O .
(det L)n−1 i=1
det L(|−(n−i)) · (det L)n−1−i

3. Division into two parts

3.1. Preliminaries. For 2 ≤ d ≤ n − 2, we will divide P (L, d, H) into two parts, and
deal with them one at a time. We induct on n, assuming that P has been computed for
all lattices of rank < n.
Fix a basis {v1 , . . . , vn } of L. Define L̄ = L/hvn i, and identify it with the projection
of L onto the subspace of Rn orthogonal to vn i.e. we think of L̄ as a subset of Rn . Let
v̄i be the component of vi orthogonal to vn , so that vi = v̄i + ai vn for some ai ∈ R and
L̄ = spanZ (v̄1 , . . . , v̄n−1 ).
We write
P (L, d, H) = P 1 (L, d, H) + P 2 (L, d, H)
where P 1 (L, d, H) equals the number of rank d sublattices of L of height ≤ H such that
its projection to L̄ is also of rank d, and P 2 (L, d, H) equals the number of those whose
projection is of rank d − 1. Equivalently, P 1 counts sublattices whose R-span does not
contain vn , and P 2 counts those that does.
It helps to think of X ∈ Gr(L, d) explicitly as a coset ΓM L, for some M = (cij ) 1≤i≤d ∈
1≤j≤n

Matpr
d×n (Z). Also, let L̃ be the n×n matrix whose i-th row vector equals v̄i for 1 ≤ i ≤ n−1,
8 SEUNGKI KIM

and vn for i = n, so that

 
1 a1
 1 a2 
..
 
L=
 ..  L̃.

 . . 
 1 an−1 
1
Then we can also write X in the form Γ(C; c + c′ )L̃, where C = (cij ) 1≤i≤d is the first
1≤j≤n−1
d × (n − 1) submatrix of M , and c = (c1n , . . . , cdn )tr and c′ = ( j aj c1j , . . . , j aj cdj )tr
P P

are vectors in Rd .

3.2. Computing P 2 (L, d, H). Consider first the case rank C = d − 1, so that X con-
tributes to P 2 . We may assume that M is a Hermite normal form, so that C is too.
Because M is primitive, so is C, and the d-th entry of the vectors c and c′ must be equal
to 1 and 0 respectively. This forces each of the other entries of c + c′ to have only one
choice modulo the left action of Γ. Thus
H
(3.1) P 2 (L, d, H) = P (L̄, d − 1, ).
kvn k
3.3. Some lemmas. Working with P 1 is much more involved. Most of the remainder of
this paper is devoted to this task. The goal of this section is to derive the expression (3.5)
for P 1 that is amenable to computation.
We start by recalling the standard choice of the representatives of the right cosets
of Γ in the double coset ΓaΓ, where a ∈ Matd×d (Z) has determinant k > 0. Such a
representative, say h = (hij )1≤i,j≤d , is a lower diagonal matrix with determinant k, with
the condition that 0 ≤ hji < hii for all j > i. Of course, Γh ⊆ ΓaΓ if and only if a and h
have the same invariant factors.
Lemma 3.1. Given a d × n matrix (C; c) with rank C = d, there exists a unique triple
(h, B, b), where h is one of the right coset representatives described above, B is a d×(n−1)
primitive Hermite normal form of rank d, and d ∈ Zn , such that (C; c) ∼ (hB; b).
Proof. By the theory of the Smith normal form, we have (C; c) ∼ (aB0 ; b0 ) where a is an
invariant factor matrix — that is, a = diag(a1 , . . . , ad ) with ai |ai+1 — B0 is a primitive
d × (n − 1) matrix of full rank, and b0 ∈ Zd . Write B0 = γB, where B is the Hermite
normal form of B0 and γ ∈ Γ. Then there exists γ ′ ∈ Γ and h a coset representative of
ΓaΓ such that γ ′ h = aγ. Therefore, writing b = γ ′−1 b0 , we have (C; c) ∼ (hB, b).
Suppose we have another triple (h′ , B ′ , b′ ) such that (hB, b) ∼ (h′ B ′ , b′ ). This is pos-
sible only if the row vectors of B and B ′ generate the same lattice. Since both B and B ′
are in the Hermite normal form, B = B ′ . This in turn implies h = h′ and b = b′ . 
Lemma 3.2. Again given a d × n matrix (C; c), write C = γaB, where γ ∈ Γ, a =
diag(a1 , . . . , ad ) is an invariant factor matrix, and B is primitive. Thus (C; c) ∼ (aB; γ −1 c) =
(aB; b), where b := γ −1 c.
Then (aB; b) is primitive if and only if a1 = . . . = ad−1 = 1 and bd is coprime to ad .
Proof. Without loss of generality, we may assume B to be the matrix which has 1’s in
the diagonal and 0’s elsewhere. (aB, b) is imprimitive if and only if there P exist integers
= 1, . . . , d, ri not all zero, such that (r1 , . . . , rd , 0, . . . , 0, i bi ri /ai ) ∈ Zn ,
0 ≤ ri < ai for iP
or equivalently i bi ri /ai ∈ Z.
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 9

Suppose ad−1 6= 1. We claim that, for any bd−1 and bd , bd−1 rd−1 /ad−1 + bd rd /ad ∈ Z
for a nontrivial choice of the r’s. There exists a prime p such that p|ad−1 and p|ad , so it
suffices to find a nontrivial solution to the expression bd−1 rd−1 + bd rd ≡ 0(mod p). But
this is clearly possible.
Next suppose ad−1 = 1. We are led to consider the condition bd rd /ad ∈ Z. This is
impossible if and only if (bd , ad ) = 1, which completes the proof. 
Lemma 3.3. Write e(pα ) = diag(1, . . . , 1, pα ). Then the necessary and sufficient condi-
tion for h ∈ Matd×d (Z) to be one of the standard form right coset representatives of Γ in
α ai
P )Γ is as follows: h is a lower triangular matrix with hii = p , where ai ≥ 0 and
Γe(p
ai = α, 0 ≤ hji < hii for j > i, and in addition if i < j are two indices such that
ai , aj ≥ 1 and ai+1 = . . . = aj−1 = 0 — i.e. all diagonal entries between hii and hjj are
trivial — then (hji , p) = 1.
Proof. Let h be a coset representative of some double coset of a matrix of determinant
pα , in the form that we chose in the beginning of this section. Then all but the last
condition are automatically satisfied. For the last condition, choose the three smallest
indices i < j < k for which ai , aj , ak > 0. We consider the 3 × 3 matrix
 a 
p i
(3.2) hji paj .
ak
hki hkj p
We will show that this matrix has invariant factors (1, 1, pai +aj +ak ) if and only if hji
and hkj are coprime to p. Then the proof is complete because we can repeatedly apply
this argument to h to compute the invariant factors of h.
If hji and p are coprime, there exist integers x, y such that yhji − xpai = 1, so that the
matrix
hji pai 0
 
x y 0
0 0 1
has determinant 1. Multiplying this on the left of (3.2), we have
0 pai +aj
 
0
 1 ypaj 0 ,
hki hkj p ak
which, upon multiplying by suitable elements of Γ from both sides, becomes
 
1 0 0
0 pai +aj 0 .
0 hkj − yp hki pak
aj

If furthermore hkj is coprime to p, then so is hkj − ypaj hki , so we can use the same
trick to see that (3.2) has invariant factors (1, 1, pai +aj +ak ) indeed.
Now go back to (3.2) and consider the case hji = cpb ; we can assume 1 ≤ b < aj and
(c, p) = 1. We restrict our attention to the 2 × 2 upper-left corner submatrix of (3.2), and
temporarily use ≈ to denote the equivalence under the left and right multiplication by Γ.
Then, by a similar argument as earlier, for an appropriate integer y,
 a
0 pai +aj −b
  a −b  b
0 pai +aj −b
   b   
p i p i p p
= ≈ ≈ ,
cpb paj c p aj 1 1 ypaj 1 pb 0
so pb appears as one of the invariant factors.

10 SEUNGKI KIM

Lemma 3.4. Write e(k) = diag(1, . . . , 1, k), as in the previous lemma. Then the number
of the right cosets of Γ in Γe(k)Γ equals
Y
p(α−1)(d−1) (1 + p + . . . + pd−1 ).
p|k
pα kk

Proof. From the general theory of Hecke operators (see Chapter 3 of Shimura [17]), it
suffices to prove the lemma for the case k = pα . We proceed by induction on α.
In case α = 1, there exist pd−i coset representatives which has aii = p and ajj = 1 for
all j 6= i. This exhausts all the representatives of Γe(p)Γ, so the lemma holds true in this
case.
For the general case, it suffices to match, to each representative h of Γe(pα−1 )Γ, pd−1
representatives of Γe(pα )Γ, different for each h. Suppose j is the smallest number for
which hjj is a power of p. Then modifying hjj to phjj and hkj (k > j) to hkj + ck hjj , for
any choice of 0 ≤ ck < p, yields a representative of Γe(pα )Γ, accounting for pd−j out of
pd−1 total. Also, for each i < j, replacing hii (= 1) by p, a choice of each hki (k 6= j) from
{0, . . . , p− 1} and of hji from {1, . . . , p− 1} (hji cannot be 0 by the previous lemma) yields
a representative of Γe(pα )Γ, and there are pd−i−1 (p − 1) of this kind. Therefore, for each
h there is a total of pd−j + pd−j (p − 1) + pd−j+1 (p − 1) + . . . + pd−2 (p − 1) = pd−1 coset
representatives of Γe(pα )Γ constructed in this manner, as desired. It remains to show
that these representatives do not overlap with those constructed from a different choice of
h. But this is immediate since, given a representative of Γe(pα )Γ, one can read off which
representative of Γe(pα−1 )Γ it came from, by discarding the first factor of p that appears
in its diagonal. 

3.4. A computable expression for P 1 (L, d, H). For X ∈ Gr(L, d), define fH (X) = 1
if detL X ≤ H and 0 otherwise. Also, as in the statement of Lemma 3.4 write e(k) :=
diag(1, . . . , 1, k). Thanks to Lemmas 3.1, 3.2 and 3.4, we can rewrite P 1 (L, d, H) as
X XX X
(3.3) fH ((hB; b) L) ,
B∈Gr(Zn−1 ,d) k≥1 h b∈Zd
(hB;b) prim.

where the sum over h is taken over all coset representatives of Γe(k)Γ in the standard
form.
Fix h, k, B for a moment, and consider the innermost summation in (3.3). For some
B ′ ∼ B, it is equal to (cf. Lemma 3.2)

X
fH ((e(k)B ′ ; b) L)
b∈Zd
(k,bd )=1
X X
= µ(l) fH ((e(k)B ′ ; e(l)b) L)
l|k b∈Zd
X X  
= µ(l) fH (e(k)B ′ ; e(l)b + e(k)t) L̃
l|k b∈Zd
X X
fH e(k)B ′ L̄ + (e(l)b + e(k)t)vn ,


(3.4) = µ(l)
l|k b∈Zd
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 11

where µ is the Möbius function, and we wrote

P ′ 
j aj b1j
t=
 .. 
P . ′

j aj bdj
for short. Note that vn is a row vector, whereas b and t are column vectors.
Temporarily write A = e(k)B ′ L̄ and B = (e(l)b + e(k)t)vn . We will use the matrix
determinant lemma to compute the height of A + B. To proceed, we need the following
lemma, which implies that the inverse of AAtr is given by AP (AP )tr .
Lemma 3.5. Let Y be a full-rank d × n matrix whose i-th row equals yi ∈ Rn . Let
z1 , . . . , zd ∈ Rn such that they form the basis of the polar lattice spanned by y1 , . . . , yd and
that hzi , yj i = δij . Let Z be the d × n matrix whose i-th row equals zi . Then the inverse
of Y Y T is given by ZZ T .
Proof. Complete Y to an invertible n × n matrix Ȳ = YY ′ , such that the rows of Y ′ are

orthogonal to the rows of Y . Similarly complete Z to Z̄ = ZZ′ , so that the rows of Z̄

form the dual basis to that formed by the rows of Ȳ . Then the rows of Z ′ are orthogonal
to the rows of Z as well.
Since Z̄ and Ȳ T are
inverses of T T

each other, we have Ȳ Ȳ Z̄ Z̄ = I. By abuse of
Y ′ 0
language, write Y = 0 , Y = Y ′ , and similarly with Z. Then
Ȳ Ȳ T Z̄ Z̄ T = (Y + Y ′ )(Y T Z + Y ′T Z ′ )(Z T + Z ′T ) = Y Y T ZZ T + Y ′ Y ′T Z ′ Z ′T ,
and observe that the first term on the right is zero outside the first d × d submatrix, and
the second term is zero outside the “last” (n − d) × (n − d) submatrix. This completes
the proof. 
We return to computing the height of A + B: it is equal to the square root of
det(AAtr ) 1 + B tr (AAtr )−1 B

= det(AAtr ) 1 + B tr (AP (AP )tr )B

 2 
= k 2 det(B ′ L̄)2 1 + kvn k2 (e(l)b + e(k)t)tr e(k −1 )(B ′ L̄)P .

For convenience, we define

s
1 H2
K(B) = −1
kvn k k 2 det(B L̄)2
if H ≥ k det(B L̄), and set K(B) = 0 otherwise. Then (3.4) becomes

number of vectors (nonzero, if k 6= 1) in e(l/k)(B ′ L̄)P

X  
µ(l) · .
whose translates by t has length ≤ K(B ′ )
l|k
The lemma below ensures that the translation of the vectors by t does not present any
extra difficulty in our estimate of this sum.
Lemma 3.6. Let Λ ∈ Rd be a lattice of rank d, and t ∈ Rd . Temporarily denote by N (r)
the number of points v ∈ Λ + t with kvk ≤ r. Then
d
!
V (d)rd X rd−i
N (r) = +O ,
det Λ i=1
det Λ(|−i)
where the implicit constant depends on d only.
12 SEUNGKI KIM

Proof. This is Lemma 2 in [13] for an affine lattice. The proof is almost exactly the same,
which we reproduce here for completeness.
We proceed by induction on d. The base case d = 1 is clear. Now assume the lemma
for d − 1. By adjusting det Λ, we may assume r = 1.
First consider the case λd ≤ 1. Let xi ∈ Λ, i ∈ {1, . . . , d}, be a vector with kxi k = λi ,
and consider the parallelepiped spanned by x1 , . . . , xd . Its diameter is ≤ λ1 + . . . + λd ≤
dλd , and it contains a fundamental parallelepiped F of Λ, which also has diameter ≤ dλd .
Write B(s) for the ball in Rn at the origin of radius s. Then since B(max(0, 1 − dλd)) ⊆
(Λ + t) ∩ B(1) + F ⊆ B(1 + dλd ), we have
|N (r) det Λ − V (d)| ≤ V (d)((1 + dλd )d − max(0, 1 − dλd )d )
≤ V (d)(2dλd )d d,
and thus      
N (r) − V (d)  = O λd 1
 
= O ,
 det Λ  det Λ det Λ(|−i)
where the second equality follows from the Minkowski’s second theorem.
It remains to consider the case λd > 1. Then (Λ + t) ∩ B(1) lies in at most two
translates
P of Λ(|−1) inthe direction of λd . Thus the induction hypothesis implies N (r) =
d (|−i)
O i=1 1/ det Λ . Also we have
 
1 λd 1
< =O
det Λ det Λ det Λ(|−1)
as above. This completes the proof.

It follows that (3.4) equals
d
!!
X V (d)K(B ′ )d X K(B ′ )d−i
µ(l) ′ P
+ O .
det(e(l/k)(B L̄) ) i=1
det(e(l/k)(B ′ L̄)P )(|−i)
l|k

e(l/k)(B ′ L̄)P = (e(k/l)B ′ L̄)P , and det((e(k/l)B ′ L̄)P )(|−i) ≫ det(e(k/l)B ′ L̄)(|−(d−i) / det(e(k/l)B ′ L̄)
by (2.3). Also, det(e(k/l)B ′ L̄)(|−(d−i)) ≫ det(B ′ L̄)(|−(d−i)) , so the above sum can be
rewritten as
d
!!
X k V (d)K(B)d X K(B)d−i det(B L̄)
µ(l) +O
l det(B L̄)P i=1
det(B L̄)(|−(d−i))
l|k

(note that B and B ′ are interchangeable in this line).

Summing up all our work in this section, we deduce that (3.3) equals
d
!!
X X Y
(α−1)(d−1) d−1
X k V (d)K(B)d X K(B)d−i det(B L̄)
p (1 + p + . . . + p ) µ(l) P
+ O
l det(B L̄) i=1
det(B L̄)(|−(d−i))
B∈Gr(Zn−1 ,d) k≥1 p|k l|k
pα kk

(3.5)
d
!!
X Y X X K(B)d−i det(B L̄)
= p(α−1)(d−1) (1 + p + . . . + pd−1 )ϕ(k)V (d) K(B)d det(B L̄) + O .
i=1
det(B L̄)(|−(d−i))
k≥1 p|k B∈Gr(Zn−1 ,d)
pα kk

µ(l) kl is the Euler totient.

P
Here ϕ(k) = l|k
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 13

The remainder of this paper is devoted to computing (3.5). Because K(B) depends on
k, we cannot deal with the constant factor just yet. However, we will later use
Lemma 3.7. For m > d + 1,
X Y ζ(m − d)
p(α−1)(d−1) (1 + p + . . . + pd−1 ) · ϕ(k)k −m = .
ζ(m)
k≥1
p|k
pα kk

Proof. We can write the expression under question multiplicatively as

 
  Y
X Y 1 X
p−(m−d)α 1 − d = 1 + (1 − p−d )p−i(m−d)  ,
p p
k≥1p|k i≥1
pα kk

which that becomes

 
Y X X
 p−i(m−d) − p−m p−i(m−d) 
p i≥0 i≥0
Y
= (1 − p−m )(1 − pm−d )−1
p
ζ(m − d)
= .
ζ(m)


4. Main term of (3.5)

In this section, we estimate the intended main term of (3.5), namely
X
(4.1) K(B)d det(B L̄),
B∈Gr(Zn−1 ,d)

for each k ≥ 1 and 2 ≤ d ≤ n − 2. We may also assume H ≥ k minB det(B L̄), since
otherwise (4.1) is equal to 0. Our approach is essentially that of Schmidt [13], who uses
summation by parts. We improve it somewhat by adopting the language of the Riemann-
Stieltjes integral, in order to simplify the computation and to derive pretty error terms.
Rewrite (4.1) as
 d2
H2

1 X
2
det(B L̄) −k ,
kvn kd k d n−1
det(B L̄)2
B∈Gr(Z ,d)

so that the problem comes down to estimating

X
Q(k, H) := ψ(det(B L̄))
B∈Gr(Zn−1 ,d)

where ψ(t) = t((H/t)2 − k 2 )d/2 for 0 < t ≤ H/k, and ψ(t) = 0 otherwise. It is easy
to check that ψ(t) is a twice differentiable function on 0 < t ≤ H/k, with ψ ′ (t) =
−((d − 1)(H/t)2 + k 2 )((H/t)2 − k 2 )(d/2−1) ≤ 0.
Choose a δ > 0 with δ ≤ minB det(B L̄). Write H/k = (α+s)δ with α ∈ [0, 1) and s ∈ Z.
Also, let P1 (t) be the number of elements B ∈ Gr(Zn−1 , d) such that t < det(B L̄) ≤ t + δ
, and P2 (t) = P1 (t − δ). Then for i = 1, 2,
14 SEUNGKI KIM

 
s−1
X
(−1)i Q(k, H) − ψ((α + j)δ)Pi ((α + j)δ) ≥ 0.
j=0

Write R1 (t) for the number of B ∈ Gr(Zn−1 , d) such that det(B L̄) ≤ t + δ, and
R2 (t) = R1 (t − δ) (= P (L̄, d − 1, t), of course). Since ψ((a + s)δ) = 0, by the summation
by parts,
 s−1
X 
(−1)i Q(k, H) − Ri ((α + j)δ)(ψ((α + j)δ) − ψ((α + j + 1)δ)) ≥ 0.
j=0

Thus we have bounded Q(k, H) from both sides by certain Riemann-Stieltjes sums.
The remaining issue is that of convergence as δ → 0. First, observe that, since Ri ’s are
supported strictly away from zero by ε = minB det(B L̄), we may assume the same of
ψ, so that ψ is of bounded variation. Second, Ri are clearly not continuous, but by the
induction hypothesis on n, we know it is bounded from both sides by a polynomial in t;
e.g.
 
tn−1  X
R2 (t) = a(n − 1, d) +O c γ tγ 

det(L̄)d
γ∈Q
0≤γ<n−1

where cγ = bγ (L̄) is as in Theorem 1.3. As for R1 (t), strictly speaking it is bounded by

a polynomial in (t + δ); but the ensuing technicality is easy to deal with, e.g. choose a
δ ′ > 0 independent of δ, and bound R1 (t) by a polynomial in (t + δ ′ ), then take δ ′ → 0 at
the very end. We have shown that
 
Z H/k H/k
a(n − 1, d)
Z
 X
(4.2) Q(k, H) = −tn−1 ψ ′ (t)dt + O  cγ −tγ ψ ′ (t)dt .

(det L̄)d
ε γ∈Q ε
0≤γ<n−1

Since the same argument will be used repeatedly later in this paper, we summarize our
discussion so far in the form of a lemma:

Lemma 4.1. Assume Theorem 1.3 for n = m, and let M be a full-rank lattice in Rm .
Suppose ψ is a decreasing twice differentiable function supported on [a, b]. Then
 
Z b Z b
X a(m, d) X
ψ(det B) = −tm ψ ′ (t)dt + O  bγ −tγ ψ ′ (t)dt .
(det M )d a a
B∈Gr(M,d) γ∈Q∩[0,m)

We return to estimating (4.2). Recall ε = minB det(B L̄) ∼ di=1 λi (L̄). In (4.2), for
Q
the integrals inside the O-notation, there is no harm in replacing ε with 0 if γ > d − 1.
For the main term, we can do the same at the cost of
Z ε Z ε
1 n−1 ′ 1 H d εn−d Hd
d
−t ψ (t)dt ≪ d
H d tn−d−1 dt ∼ d
≪ d−1 .
(det L̄) 0 (det L̄) 0 (det L̄) ε

Now the main term contributes

 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 15

Z H/k
−tn−1 ψ ′ (t)dt
0
H/k Z H/k
= −tn−1 ψ(t) + (n − 1) tn−2 ψ(t)dt

0 0
H/k  d2
H2
Z 
n−1
= (n − 1) t − k2 dt
0 t2
Z 1
d
= (n − 1)H n k −n+d xn−d−1 (1 − x2 ) 2 dt
0
(n − 1)V (n)
= H n k −n+d .
(n − d)V (n − d)V (d)
For the last equality, we used the identity on the beta function (see e.g. [2])
Z 1
B(a, b) = 2 x2a−1 (1 − x2 )b−1 dx.
0
Similarly, the secondary term i.e. the case γ = n − 1 − b(n − 1, d) gives
 
O cγ H n−b(n−1,d) k −n+d+b(n−1,d) .
In general, each integral corresponding to γ > d − 1 is
O cγ H γ+1 k d−γ−1

and those corresponding to γ < d − 1 is

O cγ H d ε−d+γ+1 .

The case γ = d − 1 contributes

 
γ+1 H
= O cγ H γ+1+η (kε)−η


O cγ H log
kε
for any η > 0.
In conclusion, we proved that (4.1) equals
!
a(n, d) ζ(n) n
X
(H/k) + O c′γ (H/k)γ ,
det(L)d ζ(n − d)V (d) γ

where each c′γ is a reciprocal of products of λi (L̄)’s and kvn k, so that c′γ (H/k)γ is invariant
under scaling of L.

5. Error term of (3.5)

In this section, we work on the intended error term of (3.5), namely
X K(B)d−i det(B L̄)
(5.1)
n−1
det(B L̄)(|−(d−i))
B∈Gr(Z ,d)

for 1 ≤ i ≤ d. Rewrite (5.1) as 1/(kvn k)d−i times

 d−i
H2

1 X det(B L̄) 2
− k2 ,
k d−i det(B L̄)(|−(d−i)) det(B L̄)2
B∈Gr(Zn−1 ,d)
16 SEUNGKI KIM

which we simplify and bound from above by

X fH/k (B)
(5.2) (H/k)d−i d−i−1
.
(det B) det B (|−(d−i))
B∈Gr(L̄,d)

Our analysis of (5.2) depends on the “skewedness” of B and L̄. We will first explain
how to deal with (5.2) in case all λi (L̄) is of size (H/k)1/d — i.e. L̄ is not too skewed —
and then work out the general case.
In addition, for the rest of this section, we assume k = 1 for simplicity. To restore the
general case, one could simply replace H by H/k.

5.1. When L̄ is “not skewed”. Assume λn−1 (L) ≤ 2n−1 H 1/d . For each 0 ≤ d′ ≤ d, we
restrict the sum (5.2) to those B for which d′ is the lowest number such that
′ ′
(5.3) λd′ (B) ≤ 2d H 1/d and λd′ +1 (B) − λd′ (B) > 2d H 1/d ,
where we interpret λ0 = 0 and λd+1 = ∞. Then we can bound (5.2) by a constant times
′ X fH (B)
(5.4) H (d +1)(1−i/d)
,
(d′ ) d−i
B∈Gr(L̄,d) det B

where the sum is over all B that satisfies (5.3).

The idea is that, because we are assuming λn−1 (L) ≤ 2n−1 H 1/d , we can proceed as in
Section 9 of Schmidt ([13]). We reproduce his argument here for completeness.
Lemma 5.1. Let L̄ be an m(= n − 1 in our context) dimensional lattice. Fix a Bd′ ∈
(|−j)
Gr(L̄, d′ ), and let j = d − d′ . Then the number of Bd ∈ Gr(L̄, d) such that Bd = Bd ′ ,
λd′ +1 (Bd ) ≫ λm (L̄) and det Bd ≤ H is
 j  m−d′
det Bd′ H
≪ ,
det L̄ det Bd′
where the implicit constant here depends only on n, d, and the implied constants on the
bound relating λm (L̄) and λd′ +1 (Bd ).
′
Proof. We may assume det Bd′ ≪ H d /d , because by Minkowski’s second
′ ′
det Bd′ ∼ λ1 (Bd′ ) . . . λd′ (Bd′ ) ≤ (λ1 (Bd ) . . . λd (Bd ))d /d ≪ H d /d .
We proceed by induction on j. Suppose first that j = 1. Let π : L̄ ⊗ R → L̄ ⊗ R be the
orthogonal projection onto the orthogonal complement of Bd′ ⊗ R. Then π(L̄) (∼ = L̄/Bd′ )
is a rank m − d′ lattice of determinant det L̄/ det Bd′ , and π(Bd′ +1 ) is a 1-dimensional
primitive sublattice spanned by a vector whose length is det Bd′ +1 / det Bd′ . Therefore,
the number of Bd′ +j is bounded by the number of primitive vectors of π(L̄) of length
≤ H/ det Bd′ .
If F is a fundamental domain of L̄, then π(F) is a fundamental domain of π(L̄). Since
we can choose an F of diameter λ1 (L̄) + . . . + λm (L̄) ≤ mλm (L̄) and π is a contraction,
π(F) has diameter ≤ mλm (L̄). So the number of vectors of π(L̄) of length ≤ H/ det Bd′
is bounded by a constant times
 m−d′  m−d′
det Bd′ H det Bd′ H
(5.5) + mλm (L̄) ∼ .
det L̄ det Bd′ det L̄ det Bd′
This works because H/ det Bd′ ≥ det Bd / det Bd′ ∼ λd′ +1 (Bd ) ≫ λm (L̄).
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 17

For a general j, by inductive hypothesis what we need to estimate is

X  det Bd′ +1 j−1  m−1−d′
H
(5.6)
det L̄ det Bd′ +1
B d′ +1

(|−1)
where the sum is over all Bd′ +1 ∈ Gr(L̄, d′ +1) such that Bd′ +1 = Bd′ and λd′ +1 (Bd′ +1 ) ≫
λm (L̄). In addition, Bd′ +1 must satisfy det Bd′ +1 ≪ det Bd′ (H/ det Bd′ )1/j =: h say, since
λd′ +1 (Bd ) ≪ (H/ det Bd′ )1/j .
(|−1)
From the (proof of) case j = 1, the number of Bd′ +1 with Bd′ +1 = Bd′ , λd′ +1 (Bd′ +1 ) ≫
λm (L̄), and det Bd′ +1 ≤ t is
′
tm−d t
≪ ′ if det Bd′ ≥ (const) · λm (L̄),
det L̄ det Bdm−1−d
′

det Bd′ ′
≪ λm (L̄)m−d otherwise.
det L̄
But t ≥ det Bd′ +1 ∼ det Bd′ · λd′ +1 (Bd′ +1 ) ≫ det Bd′ · λm (L̄), so we may disregard the
latter possibility.
Therefore, we can apply Lemma 4.1, the Riemann-Stieltjes argument in the previous
section, and deduce that (5.6) is bounded by a constant times
′
Z h m−d′ −1
tm−d −m+1+d′ +j H
(5.7) m−1−d′
· t dt,
0 det L̄ det Bd′ (det L̄)j−1
which turns out to be equal to a constant times
 j  m−d′
det Bd′ H
,
det L̄ det Bd′
as desired.


We proceed to estimating (5.4). Thanks to Lemma 5.1, we can bound it by

 ′ n−1−d′
f(const)H d′ /d (Bd′ ) det Bd′ d−d

(d′ +1)(1−i/d)
X H
H
′
(det Bd′ )d−i det L̄ det Bd′
Bd′ ∈Gr(L̄,d )
n−(1+d′ )i/d
H X
= f(const)H d′ /d (Bd′ )(det Bd′ )−n+1+i .
(det L̄)d−d′
Bd′ ∈Gr(L̄,d′ )

This can be handled again as in the previous section using Lemma 4.1, yielding terms
of H-degree at most n − i/d satisfying all the miscellaneous conditions that we need, such
as the scaling invariance.

5.2. The skewed case. Now assume that 0 ≤ l < n − 1 is the lowest number such that
λl (L̄) ≤ 2l H 1/d and λl+1 (L̄) − λl (L̄) > 2l H 1/d .
As earlier, we again restrict the sum (5.2) to those B for which 0 ≤ d′ ≤ d is the lowest
number such that
′ ′
λd′ (B) ≤ 2d H 1/d and λd′ +1 (B) − λd′ (B) > 2d H 1/d ,
18 SEUNGKI KIM

′
and write Bd′ = B (d ) . Then we must have d′ ≤ l and Bd′ ⊆ L̄(l) . Writing L̄l = L̄(l) , it is
possible to decompose
L̄ = L̄l ⊕ M,
where M is an n − 1 − l dimensional lattice chosen as follows: take an LLL basis (see
[8]) {x1 , . . . , xn−1 } of L̄, so that kxi k ∼ λi (L̄) and span{x1 , . . . , xl } = L̄l . Then we
let M = span{xl+1 , . . . , xn−1 }. Also, let M̄ to be the orthogonal projection of M onto
L̄⊥
l ⊆ L̄ ⊗ R. An important fact we will use later is that λ1 (M̄ ) ≫ H
1/d
by construction.
We further restrict (5.2) to those B for which rk B∩L̄l = r for a fixed r ∈ {d′ , . . . , min(l, d)},
and call Br = B ∩ L̄l . We also let A ⊆ M̄ be the projection of B onto M̄ . Clearly
det B = det Br det A, and since det A ≫ H (d−r)/d we have det Br ≪ H r/d .
Our considerations so far lead us to bound (5.2) by
′ X f(const)H r/d (Br ) X
H (d +1)(1−i/d) d−i
fH/ det Br (A).
(det Bd′ )
Br ∈Gr(L̄l ,r) A∈Gr(M̄ ,d−r)

Using the induction hypothesis on our main theorem, and the fact that λ1 (M̄ ) ≫ H 1/d ,
we can rewrite the inner sum so that this becomes
γ
f(const)H r/d (Br ) X

′ X H
H (d +1)(1−i/d) H −γ(d−r)/d.
(det Bd′ )d−i det Br
Br ∈Gr(L̄l ,r) γ≤n−1−l

Let us look at one γ at a time, and consider

r ′ i
X f(const)H r/d (Br ) 1
H d γ+(d +1)(1− d )
(det Bd′ )d−i (det Br )γ
Br ∈Gr(L̄l ,r)

d′
γ+(d′ +1)(1− di )
X f(const)H r/d (Br )
=H d .
(det Bd′ )d−i+γ
Br ∈Gr(L̄l ,r)

By Lemma 5.1 and arguing similarly to the “not skewed” case, we obtain that this is
d′ ′ i
X f≪H d′ /d (Bd′ ) X
= H d γ+(d +1)(1− d ) d−i+γ
f≪H r/d (Br )
′
(det Bd′ ) B ∈Gr(L̄ ,r)
Bd′ ∈Gr(L̄l ,d ) r l
(d′ )
Br =B ′
d
r−d′  l−d′
f≪H d′ /d (Bd′ ) H r/d

d′
γ+(d′ +1)(1− di )
X det Bd′
≪H d
(det Bd′ )d−i+γ det L̄l det Bd′
Bd′ ∈Gr(L̄l ,d′ )
d′
γ+ rd (l−d′ )+(d′ +1)(1− di )
H d X
(5.8) = f≪H d′ /d (Bd′ )(det Bd′ )−d+i−γ+r−l .
(det L̄(|−(n−1−l)) )r−d′
Bd′ ∈Gr(L̄l ,d′ )

It remains to apply Lemma 4.1, and make sure the H-degree of this expression is strictly
below n. Here we only discuss the terms of the highest degrees, as the rest can be dealt
with in a similar fashion.
If −d + i − γ + r < 0, estimating the sum in (5.8) does not yield any additional power of
H, because it would be dominated by O((det L̄(l) )−d−i+γ+r ). In this case, the H-degree
of (5.8) is bounded by
d′ r i i id′
γ + (l − d′ ) + (d′ + 1)(1 − ) ≤ n − − ,
d d d d d
because d′ ≤ r ≤ d and γ + l ≤ n − 1.
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 19

If −d + i − γ + r = 0, the sum is of size O(log H), in which case we can say that, for a
small η > 0, the H-degree is ≤ n − i/d − id′ /d + η if d′ 6= 0, and is ≤ n − 1 − i/d + η if
d′ = 0. Finally, if −d + i − γ + r > 0, the H-degree of (5.8) equals
rl i
+1− ,
d d
which attains its maximum n − i/d only if r = d and l = n − 1 — but recall that we are
assuming l < n − 1 here.

6. Summary, and a proof of Theorem 1.3

6.1. A polynomial expression for P (L, d, H). Summing up all our work so far, we
have that

(6.1)    
H/ε  
X Y 1   a(n, d) ζ(n)  X
P 1 (L, d, H) = pαd 1 − d   H n k −n + O  cγ H γ k −γ 


p (det L)d ζ(n − d)
k=1 p|k γ∈Q
pα kk 0≤γ<n

where ε = minB∈Gr(Zn−1 ,d) det(B L̄), and each cγ is a reciprocal of products of λi (L̄)’s and
kvn k so that cγ H γ is invariant under scaling of L. In this section, we will estimate the
sum (6.1), and then make a choice of vn ∈ L so that the dependence on λi (L̄)’s turns into
dependence on λi (L)’s. This will prove our main theorem.
We treat (6.1) one monomial at a time. The highest degree term contributes
 
H/ε    
X Y
αd 1  a(n, d) ζ(n) n −n
p 1− d  H k .
p (det L)d ζ(n − d)

k=1 p|k
pα kk

The corresponding infinite sum, by Lemma 3.7, equals

a(n, d) n
H ,
(det L)d
the desired main term. It remains to bound the tail, which we can, up to a constant
factor, approximate as
1 X
d
H n k d−n ,
(det L)
k>H/ε

which is of size
H d+1 εn−d−1
.
(det L)d
We need to show that εn−d−1 /(det L)d is bounded by a reciprocal of a product of
λi (L)’s. Since ε ∼ di=1 λi (L̄) and λi (L̄) ≤ λi+1 (L) (quick proof: project a dimension
Q
Qd
(i + 1) subspace of Rn onto the orthogonal complement of vn ), we have ε ∼ i=1 λi+1 (L).
So εn−d−1Qnis a product of λi (L)n−d−1 , for each iQ= 2, . . . d + 1. On the other hand,
d d n
(det L) ∼ j=1 λj (L) , which contains the factor j=d+2 λj (L) d times. For any i ≤
n Qd+1
d + 1, λi (L)n−d−1 / j=d+2 λj (L) ≤ 1, so εn−d−1 /(det L)d ≪ j=1 λj (L)−d , as desired.
Q
20 SEUNGKI KIM

We return to other monomials in (6.1). For γ > d + 1, the sum under consideration is
H/ε  
γ
X Y
αd 1
cγ H p 1 − d k −γ ,
p
k=1 p|k
pα kk

which we can bound by the infinite sum and apply Lemma 3.7, obtaining O(cγ H γ ). For
γ < d + 1, the sum is of size
H/ε
X cγ H d+1
cγ H γ k d−γ ≈ ,
εd−γ+1
k=1

and for γ = d + 1, it is
H/ε
X H cγ H γ+η
cγ H γ k −1 ≈ cγ H γ log ≪
ε εη
k=1

for any η > 0. Hence, together with the expression (3.1) of P 2 , we conclude that
 
a(n, d) n X
P (L, d, H) = H +O bγ H γ 
 
(det L)d γ∈Q
0≤γ<n

where each bγ is a product of reciprocals of λi (L)’s, λi (L̄)’s, and kvn k, so that bγ H γ is

invariant under scaling of L. The following lemma shows that we can replace bγ by a
product of λi (L)−1 ’s only, so that it makes sense to write bγ = bγ (L):

Lemma 6.1. Recall that L̄ is the orthogonal projection of L onto the complement of a
vector vn ∈ L If we choose vn to be a shortest nonzero vector of L, then λi−1 (L̄) ∼ λi (L)
for all i = 2, . . . , n.

Proof. Let {w1 , . . . , wn } be an LLL basis (see [8]) of L containing vn = w1 . Then, writing
w̄i for the projection of wi to the complement of vn , {w̄2 , . . . , w̄n } is an LLL basis of L̄.
Therefore, by Proposition 1.12 of [8], kwi k ∼ λi (L) and kw̄i k ∼ λi−1 (L̄).
On the other hand, by the definition of an LLL basis, kw̄i k2 = kwi k2 − µ2 kw1 k2 for
some |µ| ≤ 1/2. This immediately implies kw̄i k ≤ kwi k, and also, since kw1 k ≤ kwi k, we
have kw̄i k ≫ kwi k, completing the proof. 

6.2. The primary error term, d ≤ n/2. Finally, we provide an estimate on the primary
error term of P (L, d, H), again assuming kvn k = λ1 (L). We temporarily assume d ≤ n/2,
and argue the cases d > n/2 by duality. Tracing back our estimates so far, there are two
candidates for the primary error term: one is from the estimate of the “main part” (4.1),
which contributes
bn−1−b(n−1,d) (L̄) n−b(n−1,d)
 
(6.2) O H ,
kvn kd
and the other is from the estimate of the “error part” (5.1) in case i = 1, which contributes

H n−b(n,d)
 
O ,
(det L)d−1 det L̄
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 21

but by rewriting everything in terms of λi (L)’s with help of Lemma 6.1, we find that this
is bounded by
H n−b(n,d)
 
(6.3) O .
(det L)d−b(n,d)(det L(|−d) )b(n,d)
The reason we use this slightly inferior bound is that this possesses a convenient symmetry
under duality, as we will see below.
We claim by induction that the main error term has degree n − b(n, d), and that we
can take
1
bn−1/d (L) = .
(det L)d−b(n,d) (det L̄)b(n,d)
In the base case n = 4, d = 2, it is clear that (6.3) is the primary error term. For the
induction step, we need to show that (6.2) is no greater than (6.3). If d = n/2, (6.2) is of
degree strictly less than n − b(n, d), and we are done. If d < n/2, then by the fact that
kvn k = λ1 (L) and Lemma 6.1,
kvn kd (det L̄)d−1/d (det L̄(|−d) )1/d ∼ (det L)d−1/d (det L(|−d) )1/d ,
which shows that (6.2) has the same size as (6.3), completing the proof of the claim.

6.3. The primary error term, d > n/2. Write d′ = n − d for short. We think
of P (LP , d′ , H) as consisting of two parts, one that counts the sublattices of d-type
(1, 2, . . . , d′ , . . . , d′ ) and the other that counts the rest, and then apply the duality theorem
to the former. Our method makes it clear that the contribution from the latter is bounded
by terms of H-degree at most n − 1. As for the former, either from Theorem 3 of Thunder
′
([21]) — since those sublattices are precisely the ones whose intersection with (LP )(|−d )
is trivial — or by an appropriate adaptation of our argument so far — in which case our
computation simplifies immensely — the number of such lattices is
′
!
a(n, d′ ) n H n−b(n,d )
H +O .
(det LP )d′ (det LP )d′ −b(n,d′ ) (det(LP )(|−d′ ) )b(n,d′ )
To this part alone we apply the duality theorem (2.2), which yields the main error term
of

H n−b(n,d)
(det L)n−b(n,d) (det LP )n−d−b(n,d) (det(LP )|−(n−d) )b(n,d)
H n−b(n,d)
=
(det L)d−b(n,d)(det L(|−d))b(n,d)
by the relation (2.3), as desired.

7. Proofs of Corollaries to Theorem 1.3

7.1. Formula for N (L, d, H). An asymptotic formula on N (L, d, H) can be derived easily
from that of P (L, d, H) by a standard Möbius inversion, as in Schmidt ([13]). As in [13],
define σd (m) inductively by
σ1 (m) = 1,
X
σd (m) = rk−1 σk−1 (m/r).
r|n
22 SEUNGKI KIM

It is shown in [13] that σd (m) equals the number of index m sublattices of a rank d
lattice, and that
σd (m) ≪ (m log log m)d−1 ,
∞
X d
Y
σd (m)/mn = ζ(n + 1 − i)
m=1 i=1

for d ≤ n − 1. From the latter it follows that

H/ε
X
N (L, d, H) = P (L, d, H/m)σd (m)
m=1
 
H/ε H/ε
a(n, d) X X X
= (H/m)n σd (m) + O  bγ (L)(H/m)γ σd (m) ,
 
d
(det L) m=1 γ∈Q m=1
0≤γ<n

where ε := minX∈Gr(L,d) detL X. If we bound the tail of each summation over m, the
proof of Corollary will be completed. The required properties of the coefficients b′γ (L) can
be checked straightforwardly, so we omit the proof.
For the main term, we have
X X H d+η
(H/m)n σd (m) ≪ md−n−1+η H n ≈ d−n+η
ε
m>H/ε m>H/ε

for any η > 0.

PH/ε
P∞ In the error term, for γ > d we can safely replace the sum m=1 by the infinite sum
m=1 . For γ ≤ d, we see that
H/ε H/ε  d−γ+η
X X H
σd (m)m−γ ≪ md−1−γ+η ≈
m=1 m=1
ε
for any η > 0. If d < n − 1, η can be set small enough, so that the secondary term has
H-degree n − b(n, d). If d = n − 1, the secondary term has degree n − 1 + η.
Remark. One may wonder what the formula for N (L, n, H) would be. In this case, the
skewness of L induces no subtlety at all, and simply
 n  n−1+η !
H H
N (L, n, H) = c · +O
det L det L
for any η > 0.
7.2. Formula for PS (L, d, H). Let S ⊆ L be a sublattice of rank e ≤ n − d. By choosing
the basis {v1 , . . . , vn } of L so that {vn−e+1 , . . . , vn } is a basis of S, and applying the
division idea in Section 3 repeatedly, we obtain an estimate of PS (L, d, H) analogous to
that of P (L, d, H) in (1.1), with the coefficients bγ being a product of reciprocals of λi (L)
and λi (L/S). However, the reciprocal of λi (L/S) could be arbitrarily large, which may
cause difficulties in some applications of Theorem 1.3. For instance, suppose one wants
to compute
X X X
fH1 (A)fH2 (B) = fH1 (A) fH2 (B),
A,B∈Gr(L,d) A B
A∩B={0} A∩B={0}
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 23

which is exactly (1.6) in the introduction. Here one is eventually led to sum the multiples
of the reciprocals of λi (L/A) over sublattices A of height bounded by H1 . It seems to be
a nontrivial task to show that such a sum is asymptotically small — a potential approach
may involve a version of the equidistribution result in [4] and the estimate in Theorem 5
of [14].
Fortunately, with minor modifications to our proof of Theorem 1.3, it is possible to
provide a formula for PS (L, d, H) independent of S, avoiding the above complication
altogether. In this section, we point out where the modifications are.
Consider first the base cases d = 1 or n − 1. If d = 1, PS (L, 1, H) = P (L, 1, H) −
P (S, 1, H), and bounding the contribution from P (S, 1, H) in terms of L using λi (S) ≥
λi (L) (because S ⊆ L), we obtain the same type of estimate as in (2.1). In case d = n − 1,
we must have rk S = 1, and thus for B ∈ Gr(L, n − 1), B ∩ S = {0} if and only if
B ⊥ ∩ S ⊥ = {0}; hence the proof follows from the d = 1 case and the duality theorem.
For other values of d, we proceed by induction on n, and split PS = PS1 + PS2 as in
Section 3 above. For PS2 , we simply bound it by P 2 . As for PS1 , observe that, analogously
to (3.3), we can write
X XX X
PS1 (L, d, H) = fH ((hB; b) L) .
B∈Gr(Zn−1 ,d) k≥1 h b∈Zd
(hB;b) prim.
(hB;b)L∩S={0}

The idea is that the main contribution of the above sum comes from those B with
B L̄ ∩ S̄ = {0}, where S̄ is the projection of S onto L̄. Since B L̄ ∩ S̄ = {0} implies
(hB; b)L ∩ S = {0}, we can further subdivide
X X
PS1 = (. . .) + (. . .)
B∈Gr(Zn−1 ,d) B∈Gr(Zn−1 ,d)
B L̄∩S̄={0} B L̄∩S̄6={0}

= PS1,1 + O(PS1,2 ).
More precisely,
PS1,1 =
X XX X
fH ((hB; b) L) ,
B∈Gr(Zn−1 ,d) k≥1 h b∈Zd
B L̄∩S̄={0} (hB;b) prim.

PS1,2 =
X XX X
fH ((hB; b) L) .
B∈Gr(Zn−1 ,d) k≥1 h b∈Zd
B L̄∩S̄6={0} (hB;b) prim.

To estimate these sums, we proceed by the exact same argument that led us to Theorem
1.3. That is, estimating PS1,1 amounts to integrating the summand against PS̄ (L̄, d, H),
and for PS1,2 it is P (L̄, d, H) − PS̄ (L̄, d, H). The former computation works out exactly
the same way, but as for the latter, since P (L̄, d, H) − PS̄ (L̄, d, H) = O(H n−1−b(n−1,d) )
by induction hypothesis its contribution is at most O(H n−b(n−1,d) ).

7.3. Flag varieties of type (e, d). Let L ⊆ Rn be a lattice, and let 1 ≤ e < d < n. Our
goal is to estimate the sum
X
P (W, e, (H/(det W )n−e )1/d )
W ∈Gr(n,d)

H 1−b/d
 
X a(d, e)H
(7.1) = +O .
(det W )n (det W )n−b(n+d−e)/d (det W (|−e) )b
W ∈Gr(n,d)
24 SEUNGKI KIM

Here b = b(d, e). Note that there is the natural constraint

(7.2) (det W )n−e (det W (e) )d ≤ H.

Estimating the sum over the main term is very similar to the computation in Section
4, so we will be brief. The “main-main” term of (7.1) comes from the integral
Z (H/εde )1/(n−e) ′ d 1/(n−e)
a(n, d)tn a(n, d)a(d, e)nH (H/εe )

a(d, e)H dt
Z
d
− n
dt = d
εd (det L) t (det L) εd t
n H H
= a(n, d)a(d, e) d
log d n−e ,
n − e (det L) εe εd
where εe = min{det X : X ⊆ L, dim X = e} and likewise for εd . The smaller terms can be
computed similarly, and it turns out the largest error term is of order O(H) as expected,
and the second largest is of order O(H 1−b(n,d)/(n−e) ).
The harder part of (7.1) is the sum over the error term, namely
X H 1−b/d
(7.3) .
(det W )n−b(n+d−e)/d (det W (|−e) )b
W ∈Gr(n,d)

To bound this, we employ our method in Section 5 above. For brevity, we only show
how to compute the first two largest H-degree terms, and suppress the λi (L) factors.
As in Section 5, for each 0 ≤ d′ ≤ d, we restrict the sum in (7.3) to those W for which
′
d is the smallest number such that
′ ′
λd′ (W ) ≤ 2d H 1/nd and λd′ +1 (W ) − λd′ (W ) > 2d H 1/nd .
Then we can bound (7.3) by
b b ′ X 1 X 1
(7.4) H 1− d + nd (d −d+e) .
(det Wd′ )b (det W )n−b(n+d−e)/d
Wd′ ∈Gr(n,d′ ) W ∈Gr(n,d)
′
W (d ) =W ′
d

In order to work on the inner sum, we first determine the range of det W . By Minkowski’s
′ ′
second, we have (det W (d ) )d/d ≪ det W . On the other hand, again by Minkowski’s sec-
′ ′
ond we have det W (d ) H (e−d )/nd ≤ det W (e) , so (7.2) implies
′ ′
(7.5) (det W )n−e (det W (d ) )d H (e−d )/n ≪ H
n−e+d′ ′ d
⇒ det W ≪ H (n−e)n (det W (d ) )− n−e .
If d′ = 0, the outer sum of (7.4) is vacuous, and the inner sum can be computed as
in the main term estimate, yielding O(H + H 1−b(n+d−e)/nd + H 1−b(n,d)/n ) up to lower
H-degree terms. Similarly, we obtain the same bound in case d′ = d.
So assume 1 ≤ d′ ≤ d − 1. We will apply Lemma 5.1. To do so, we need to check
that λd′ +1 (W ) ≫ λn (L), which is true provided λn (L) ≪ H 1/nd . Thus, assuming H is
sufficiently large, the inner sum of (7.4) is bounded by a constant times
n−e+d′ ′ d
−
H (n−e)n (det W (d ) ) n−e ′
1 xn−d −1
Z
dx.
(det L)d−d′ (det Wd′ )n−d (det Wd′ )d/d′ xn−b(n+d−e)/d
We divide into cases according to whether b(n + d − e)/d − d′ < 0 or not:
 COUNTING RATIONAL POINTS OF A GRASSMANNIAN 25

(i) If b(n + d − e)/d − d′ < 0, then (7.4) is

b b ′ ∞ ′
H 1− d + nd (d −d+e) 1 xn−d −1
X Z
≪ dx
(det L)d−d′ (det Wd′ )n−d+b (det Wd′ )d/d′ xn−b(n+d−e)/d
Wd′ ∈Gr(n,d′ )
b b ′
H 1− d + nd (d −d+e) X 1
≪ .
(det L)d−d′ (det Wd′ )n−b(n+d−d′ −e)/d′
Wd′ ∈Gr(n,d′ )
′
At this point, observe that det Wd′ ≪ (det W )d /d , so (7.5) implies det Wd′ ≪
d′ /dn
H . Therefore (7.4) is, up to the largest H-degree term,
′ /dn
b b ′ Hd
H 1− d + nd (d −d+e) xn−1
Z
≪ dx
(det L)d εd′ xn−b(n+d−d′ −e)/d′
b b ′
≪ O(H + H 1− d + nd (d −d+e) )
as desired.
(ii) If b(n + d − e)/d − d′ > 0, then we instead have
n−e+d′ ′ d
−
b b ′ H (n−e)n (det W (d ) ) n−e ′
H 1− d + nd (d −d+e) 1 xn−d −1
X Z
≪ dx
(det L)d−d′ (det Wd′ )n−d+b 0 xn−b(n+d−e)/d
Wd′ ∈Gr(n,d′ )
d′ d′ b ′
H (n−e)n ( d (n+d−e)−d )
b
H 1− n (1− d ) X
≪ .
(det L)d−d′ d
(det Wd′ )n−d+b+ n−e ( d (n+d−e)−d )
b ′
Wd′ ∈Gr(n,d′ )

Replacing the sum by the integral, this is

d′ 2b d′ −b
O(H + H 1− n (1− d + n−e ) ),
and since 1 − 2b/d + (d′ − b)/(n − e) ≥ 0, we are done.
(iii) In the rare, yet possible, case that b(n + d − e)/d − d′ = 0, we proceed similarly and
by (7.4) by
b b ′ d′
H 1− d + nd (d −d+e) X 1 H (n−e)n
log
(det L)d−d′ (det Wd′ )n−d+b (det Wd′ ) n−e
d

Wd′ ∈Gr(n,d′ )
b b ′ 1
H 1− d + nd (d −d+e) X 1 Hn
≪ log d′
(det L)d−d′ (det Wd′ )n−d+b (det Wd′ ) d
Wd′ ∈Gr(n,d′ )
′ /dn
b b ′ Hd
H 1− d + nd (d −d+e) H 1/n
Z  
≪ xd−b−1 log dx
(det L)d εd′ xd/d′
b b ′ H d′ /dn
H 1− d + nd (d −d+e)
 
1 1/n d−b d 1 d−b 1 (d−b)
= log H ·x − ′ x log x − x
(det L)d d−b d d−b (d − b)2 εd′
b b ′
= O(H + H 1− d + nd (d −d+e) log H).
In addition, in all three cases above, the contribution from the leading error term of
′ ′ ′
P (L, d′ , H d /dn ) is of size O(H 1−b(n,d )d /dn ).
This completes the error estimate. We note that the related computation in Thunder
([21]), lines 5-6 on p.185, contains a minor error: if d − e = 1, the integral there diverges.
26 SEUNGKI KIM

In summary, we estimated (7.1) to be

 
n H H X
a(n, d)a(d, e) log d n−e + O  bγ (L)H γ  ,
n − e (det L)d εe εd γ∈[0,1]∩Q

where bγ (L)’s are appropriate inverse products of λi (L)’s, and the implied constant de-
pends on n only. The largest γ is 1, and the second largest is one of
 
b(n, d) b(d, e)(n − e) 1 2b(d, e) 1 − b(d, e)
1− , 1− , or 1 − 1− + .
n nd n d n−e
When d ≤ n/2, it is always 1 − b(n, d)/n = 1 − 1/dn, but otherwise it may be either of
the other two.

References
[1] W. Banaszczyk. New bounds in some transference theorems in the geometry of numbers. Math. Ann.
296 (1993), no. 4, 625-635.
[2] P. J. Davis. 6. Gamma function and related functions, in M. Abramowitz and I. Stegun, Handbook
of Mathematical Functions with Formulas, Graphs, and Mathematical Tables, New York: Dover
Publications, 1972.
[3] J. Franke, Y. Manin, and Y. Tschinkel. Rational points of bounded height on Fano varieties. Invent.
Math. 95 (1989), no. 2, 421-435.
[4] T. Horesh and Y. Karasik. Equidistribution of primitive vectors in Zn . arXiv:1903.01560v2.
[5] S. Kim. Random lattice vectors in a set of size O(n). Int. Math. Res. Not. (2020), 2020(5): 1385-1416.
[6] S. Kim. Mean value formulas on sublattices and flags of the random lattice. arXiv:2005.10874v1.
[7] P. Le Boudec. Height of rational points on random Fano hypersurfaces. arXiv:2006.02288v1.
[8] A. K. Lenstra, H. W. Lenstra, Jr., and L. Lovász. Factoring polynomials with rational coefficients.
Math. Ann. 261 (1982), no. 4, 515-534.
[9] M. Morishita and T. Watanabe. Adele geometry of numbers. Advanced Studies in Pure Mathematics
30, 2001, Class Field Theory - Its Centenary and Prospect, 509-536.
[10] O. Regev. Dual lattices (lecture notes). Available at https://cims.nyu.edu/~ regev/teaching/lattices_fall_2004/ln/DualLattice
[11] C. A. Rogers, Mean values over the space of lattices. Acta Math. 94 (1955), 249-287.
[12] S. H. Schanuel. Heights in number fields. Bull. Amer. Math. Soc. 70 (1964), 262-263.
[13] W. M. Schmidt. Asymptotic formulae for point lattices of bounded determinant and subspaces of
bounded height. Duke Math. J. 35 (1968), 327-339.
[14] W. M. Schmidt. The distribution of sublattices of Zm . Mh. Math. 125, 37-81 (1998).
[15] W. M. Schmidt. Masstheorie in der Geometrie der Zahlen. Acta Math. 102, no. 3-4 (1959): 159-224.
[16] U. Shapira and B. Weiss. A volume estimate for the set of stable lattices. Comptes Rendus
Mathématique 352, no.11 (2014), pp.875-879.
[17] G. Shimura. Introduction to the arithmetic theory of automorphic functions. Princeton University
Press, Princeton, N.J., 1971.
[18] C. L. Siegel. A mean value theorem in geometry of numbers. Ann. of Math. (2) 46, (1945). 340-347.
[19] A. Södergren and A. Strömbergsson. On the generalized circle problem for a random lattice in large
dimension. Adv. Math. 345 (2019), 1042-1074.
[20] J. L. Thunder. An asymptotic estimate for heights of algebraic subspaces. Trans. Amer. Math. Soc.
331 (1992), no. 1, 395-424.
[21] J. L. Thunder. Asymptotic estimates for rational points of bounded height on flag varieties. Compo-
sitio Math. 88 (1993), no. 2, 155-186.
[22] J. L. Thunder. Higher-dimensional analogs of Hermite’s constant. Michigan Mathematics Journal 45,
no. 2 (1998): 301-314.