arXiv:2305.19782v1 [math.

NT] 31 May 2023

HOMOGENEOUS FORMS
INEQUALITIES

faustin adiceam* & oscar marmon 1

 1

To all our students

abstract
Let F1 pxq, . . . , Fp pxq be a set of p ě 1 real homogeneous forms in n ě 2 variables,
all of the same degree. Given a compact set K Ă Rn and a nondecreasing map T ÞÑ
bpT q ě 1, consider the region of the space determined by the inequalities

|Fi pxq| ď bpT q for all i “ 1, . . . , p when x P T ¨ K.

The following intertwined results are established :

‚ The asymptotic behavior as T tends to infinity of the volume of this region is

determined under various regularity assumptions on the set K. The growth rate
of the volume is furthermore related to the largest root of the Sato-Bernstein
polynomial associated to a homogeneous form derived from F1 pxq, . . . , Fp pxq, and
to its multiplicity;

‚ In relation with this result, a classical open problem in the theory of D-modules
is settled : it is known that the largest root of the Sato-Bernstein polynomial of
a multivariate polynomial coincides with the opposite of the smallest pole of the
complex meromorphic distribution attached to it. The corresponding multiplicity
and order are proved to also coincide;

‚ In the case that the p homogeneous forms under consideration are twisted by
composing them with unimodular matrices, the solubility of the above set of ine-
qualities in integer lattice points with bounded height is established in the form
of a metric statement depending on the convergence of a volume sum. This ef-
fective, metric and uniform generalisation of the Oppenheim Conjecture to the
case of a system of homogeneous forms answers a question raised by Athreya
and Margulis (2018). So does the related statement determining the asymptotic
behavior, generic in a metric sense, of the number of integer solutions to this
system of twisted inequalities;

‚ In the deterministic case where the forms are fixed, an upper bound is established
for the function counting the number of integer lattice points satisfying the sys-
tem of inequalities. The main term in this bound is related to the volume of the
region of the space under consideration and is sharp under suitable assumptions.
The error term is furthermore shown to admit a power saving provided that a
quantitative measure of flatness emerging from geometric tomography is large
enough. This settles a conjecture stated by Sarnak (1997).
2
CONTENTS

Notation 5

1 Introduction 7
1.1 The General Setup . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.2 Volume, Homogeneous Forms and Globally Semianalytic Domains . . . . 8
1.3 Order and Multiplicity of the Log–Canonical Threshold . . . . . . . . . . 12
1.4 Generic Unimodular Distorsions of a Family of Homogeneous Forms . . 13
1.5 The Number of Solutions to Homogeneous Forms Inequalities . . . . . . 18
1.6 Structure of the Memoir . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23

2 Volume Estimates 25
2.1 Local and Global Sato–Bernstein Polynomials . . . . . . . . . . . . . . . . 25
2.2 A Consequence of a Tauberian Theorem . . . . . . . . . . . . . . . . . . . . 28
2.3 Local and Global Real Log–Canonical Thresholds . . . . . . . . . . . . . . 49
2.4 Parameters of the Volume Estimate . . . . . . . . . . . . . . . . . . . . . . . 59

3 Log–Canonical Threshold and the Sato–Bernstein Polynomial 63

4 Counting Solutions to Generic Homogeneous Forms Inequalities 73

5 Deterministic Counting in Large Domains 81

5.1 Smooth Counting . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 83
5.2 Non–stationary Phase Analysis and Estimate of the Error Term . . . . . . 88
5.3 Analysis of the Leading Term . . . . . . . . . . . . . . . . . . . . . . . . . . 90

6 Deterministic Counting in Thin Domains 93

6.1 Non–Stationary Phase Analysis and Reduction to a Finite Sum . . . . . . 96
6.2 Geometric Tomography on Semi–Algebraic Sets . . . . . . . . . . . . . . . 98
6.2.1 Uniform Fourier Decay of the Gel’fand–Leray Function . . . . . . . 98
6.2.2 Counting Intervals of Monotonicity through O–Minimality . . . . 100
6.3 Completion of the Proof of the Main Theorem . . . . . . . . . . . . . . . . 103
6.3.1 The Counting Bound . . . . . . . . . . . . . . . . . . . . . . . . . . . 104
6.3.2 The Case of Smooth Varieties . . . . . . . . . . . . . . . . . . . . . . 107
6.4 Determining the Measure of Flatness . . . . . . . . . . . . . . . . . . . . . . 111

3
 CONTENTS 4

References 119

*Laboratoire d’analyse et de mathématiques appliquées, Université Paris-Est Créteil, Créteil, France.

[email protected]
1 Centre for Mathematical Sciences, Lund University, Box 118, 221 00 Lund, Sweden.
[email protected]
N OTAT I O N S

notations in euclidean spaces :

‚ Vectors are always denoted with bold characters (e.g., x, y, z...).

‚ The Euclidean norm is denoted by } ¨ }.

‚ Given r ą 0, Bn prq stands for the closed Euclidean ball centered at the origin with
radius r in dimension n ě 1, and Bn for Bn p1q.

algebraic notations :
‚ The set K stands for either the field of reals R or the field of complex numbers C.

‚ The notation N refers to the set of (strictly) positive integers and N0 to the set of
nonnegative integers (i.e. N0 “ N Y t0u ).

‚ Given a set of p ě 1 homogeneous forms Fpxq “ pF1 pxq, . . . , Fp pxqq, their common
set of zeros over K is denoted by ZK pFq; that is,

ZK pFq “ ty P Kn : @i “ 1, . . . , p, Fi pxq “ 0u .

‚ Given a subset A Ă Kn , the above notation is extended by setting ZA pFq “

ZK pFq X A.

analytic notations :
‚ The partial derivative with respect to the ith coordinate in dimension n (where
1 ď i ď n) is denoted by Bi . The gradient operator is then ∇ “ pB1 , . . . , Bn q.

‚ When k “ pk1 , . . . , kn q is an n-tuple of nonnegative integers, define its size by

|k| “ ni“1 ki and the differential operator B k by B k “ ni“1 Biki .
ř ś

‚ Similarly, given x P Rn , xk is shorthand notation for ni“1 xki i .

ś

‚ Given a subset A Ă Rn , its characteristic function is denoted by χA . Similarly,

given a property P depending on a vector y P Rn , χtPpyqu is the boolean function
equal to 1 when Ppyq holds and 0 otherwise.

‚ When k P N0 Y t8u, Ck pKn q stands for the set of k times continuously diffe-
rentiable functions over Kn and Ckc pKn q for the subset of such functions with
compact support; the support of an element ψ in Ckc pKn q is then denoted by
Supp ψ.

5
 CONTENTS 6

‚ Given two functions f and g depending on a variable y P Kn , the notation

fpyq ! gpyq is used to mean the existence of a real c ą 0, referred to as the
implicit constant, such that |fpyq| ď c ¨ |gpyq| for all admissible values of the vari-
able y. This is equivalent to the notation fpyq “ O pgpyqq. Also, the relation
fpyq — gpyq means that the two conditions fpyq ! gpyq and gpyq ! fpyq are
simultaneously met.

‚ The Lebesgue measure in Rn is denoted by Voln .

‚ Given x P R, set epxq “ expp2iπxq.

miscellaneous notations :

‚ The following total order over pairs of nonnegative reals is used in Chapters 2
and 3: given a, b, c, d ě 0,

pa, bq ď pc, dq ðñ pa ă cq _ ppa “ cq ^ pb ě dqq .

Explicitly, this corresponds to the order induced by the asymptotic growth of the
map x ÞÑ x´a ¨ plog xqb at infinity. Set also pa, bq ě pc, dq to mean pc, dq ď pa, bq.

‚ The notational use of the hat symbol has two different meanings which shall not
cause confusion as they are employed in distinct places : in Chapter 3 and in
the corresponding introductory Section 1.3 in Chapter 1, it refers to the complex
analogue of a real quantity. In Chapters 5 and 6 however, it refers to the Fourier
transform of a function.

‚ In the final Chapter 6, X stands for the topological closure of a set X Ă Rn .

1 INTRODUCTION

1.1 the general setup

Let p ě 1, n ě 2 and d ě 2 be integers. Fix a p–tuple Fpxq “ pF1 pxq, . . . , Fp pxqq of non–
constant real homogeneous forms Fi pxq of degree d in n variables. Let K be a subset of
Rn . Throughout, it is assumed that
pH1 q : the set K Ă Rn is compact and is equal to the closure of its interior.
Given reals a, b ą 0 and a unimodular matrix g P SLn pRq, define the set

SF pK, g, a, bq “ tx P a ¨ K : }pF ˝ gq pxq} ď bu . (1.1)

Here, a ¨ K “ tax : x P Ku and } ¨ } denotes the Euclidean norm. Thus,

g
f p
fÿ
}pF ˝ gq pxq} “ e F2i pgxq.
i“1

The overarching undertaken goal is the study of the behaviour of the lattice point
counting function
NF pK, g, a, bq “ # pSF pK, g, a, bq X Zn q (1.2)
under various assumptions on the parameters a and b, on the set K and on the uni-
modular matrix g.

Let x P K. The trivial inequalities

ad ¨ αK pF ˝ gq ď }pF ˝ gq pa ¨ xq} ď ad ¨ βK pF ˝ gq ,

where

αK pF ˝ gq “ min }pF ˝ gq pyq} and βK pF ˝ gq “ max }pF ˝ gq pyq} ,

yPK yPK

show that the set SF pK, g, a, bq is of interest only when

ad ¨ αK pF ˝ gq ď b ă ad ¨ βK pF ˝ gq . (1.3)

Of singular interest is the case when a “ T is a large parameter and when b “ bpT q
is taken as a function of this parameter. Then, for the sake of simplicity of notations,
set

SF1 pK, g, bpT qq “ SF pK, g, T , bpT qq and NF1 pK, g, bpT qq “ NF pK, g, T , bpT qq,

7
 volume, homogeneous forms and globally semianalytic domains 8

where, in view of the right–hand side inequality in (1.3), it is always assumed that the
limit condition
bpT q
cpT q :“ −Ñ 0 (1.4)
T d T Ñ8
holds. The left–hand side inequality in (1.3) then shows that the problem is nontrivial
only when

pH2 q : the set K Ă Rn intersects non trivially the algebraic variety ZR pFq.

Here,
ZR pFq “ tx P Rn : Fi pxq “ 0 for all 1 ď i ď pu . (1.5)
Conventionally, the dependence on the matrix g and on the set K in the above va-
rious sets and quantities is dropped when g “ In is the identity matrix and/or when
K “ Bn is the closed Euclidean ball centered at the origin.

The benchmark for most of the results is the generic case where the set of homoge-
neous forms Fpxq has smooth complete intersection over K. This is understood in the
sense that for all x P K, the gradient vectors ∇F1 pxq, . . . , ∇Fp pxq are linearly indepen-
dent. Equivalently, this is saying that the map
p
ľ
x P K ÞÑ ∇Fi pxq (1.6)
i“1

does not vanish.

1.2 volume determined by homogeneous forms over

globally semianalytic domains
The matrix g P SLn pRq is first seen as a fixed parameter. Without loss of generality,
upon redefining the forms under consideration, it can thus be taken as the identity.
With this in mind, define the set

SF pK, bpT qq “ SF1 pK, In , bpT qq (1.7)

and the counting function

NF pK, bpT qq “ NF1 pK, In , bpT qq. (1.8)

Recall also that both the topological conditions pH1 q and pH2 q and the limit condi-
tion (1.4) are assumed to hold.

It should be expected that, in favorable circumstances made precise in the following

sections, the counting function NF pK, bpT qq should be comparable to the volume of the
 volume, homogeneous forms and globally semianalytic domains 9

domain SF pK, bpT qq. Denoting by Voln the n–dimensional Lebesgue volume, the goal
in this first section is thus to determine the behavior of the quantity Voln pSF pK, bpT qqq
as the parameter T tends to infinity. For the stated results to be effective upon resol-
ving singularities, it is convenient to make the further general assumption that the set
K should be a globally semianalytic domain (in the sense that it is defined by a finite
number of inequalities involving analytic maps — see Section 2.1 for details). The
growth of the quantity Voln pSF pK, bpT qqq is then estimated in three cases realising to-
gether a trade–off between, on the one hand, the accuracy of the bounds that can be
obtained and, on the other, the generality of the assumptions under which they hold.
Specifically, the asymptotic growth of Voln pSF pK, bpT qqq is expressed as a function of
the properties of the zeta distribution ζF attached to the p–tuple Fpxq.

The distribution ζF is defined for any complex number s with non–positive real
part and any element ψ lying in the space of compactly supported smooth functions
C8 n
c pR q by setting ż
ψpxq
xζF psq, ψy “ s ¨ dx. (1.9)
Rn }Fpxq}
As recalled in Sections 2.2 and 2.3, it can be extended meromorphically to the entire
complex plane.

In Case (1), the most general considered, an upper bound is obtained for the
quantity Voln pSF pK, bpT qqq as a function of the smallest pole rF pψq of the function
s ÞÑ xζF psq, ψy, and of its order mF pψq. This is assuming only that the set K meets the
above conditions pH1 q and pH2 q and that the smooth test function ψ is nonnegative
and bounds from above the characteristic function of the set K.

In Case (2), the set K is furthermore assumed to be generic enough so that the
singularities of the algebraic variety ZR pFq should not be located on its boundary. This
is formalised in the following assumption :

pH3 q : there exist a real number rF pKq ě 0, an integer mF pKq ě 1, a compact

set C contained in the interior of K and an open set U containing K satisfying the
following property : the smallest poles and the corresponding orders of the functions
s ÞÑ xζF psq , ψy remain constant, respectively equal to rF pKq and mF pKq, for any
nonnegative map ψ in C8 n
c pR q positive over C with support contained in U.

In particular, for any smooth test function ψ meeting this condition,

C Ă Supp ψ Ă U. (1.10)

As a matter of fact, Assumption pH3 q turns out to be always verified in the generic
case where the homogeneous forms defining Fpxq have smooth complete intersection
over any given set K (see Proposition 2.18 in Chapter 2).
 volume, homogeneous forms and globally semianalytic domains 10

Under the three conditions pH1 q ´ pH3 q assumed to hold in this second case, Theo-
rem 1.1 below provides the exact asymptotic order of Voln pSF pK, bpT qqq as a function
of the pair prP pKq, mP pKqq; that is, its sharp asymptotic growth up to multiplicative
constants.

In Case (3), which is perhaps the most natural in view of the assumption of the
homogeneity of the forms, the set K is assumed to be star-shaped with respect to the
origin and to contain the origin in its interior. Then, the precise asymptotic behavior
of the volume of the domain SF pK, bpT qq is determined in Theorem 1.1 below. It is
expressed as a function of the smallest pole rP ą 0 and of the corresponding order
mP ě 1 of the above-defined zeta distribution ζF attached to the p–tuple Fpxq.

From the above discussion, the assumptions of Case (2) clearly contain those of
Case (1). Also, from the homogeneity of the polynomials under consideration, the
condition pH2 q required in the first two cases is immediately verified in Case (3) since
the origin then lies in the set K. As established in Lemma 2.7 of Chapter 2, it also turns
out that the condition pH3 q assumed in Case (2) always hold under the assumptions
of Case (3). This succession of observations show that the various cases considered
provide an increasing degree of refinement in the following sense :

Case (1) ð= Case (2) ð= Case (3).

Correspondingly, the volume estimates stated in Theorem 1.1 below become sharper
as the cases are more refined.

Theorem 1.1 below also shows that the locations and orders of the poles rF and
rF pKq are closely related to the Sato–Bernstein polynomial BF psq of the homogeneous
form }Fpxq}2 . The Sato–Bernstein theory and the related theory of D–modules are
reviewed in Section 2.1. It suffices here to mention that the Sato–Bernstein polynomial
is a univariate, monic, split polynomial with negative rational roots that is uniquely
associated to any given non–constant (real or complex) multivariate polynomial. It can
furthermore be computed explicitly.

Theorem 1.1 (Volume growth of domains determined by homogeneous forms inequali-

ties over globally semianalytic sets). Let the set K be globally semianalytic and let it satisfy
the assumptions pH1 q and pH2 q. Let the limit condition (1.4), whereby the quantity cpT q is
defined, hold.

Then, the volume of the set SF pK, bpT qq can be estimated with an increasing degree of
precision as follows :

(1) – assume that ψ ě 0 is a smooth test function bounding from above the characteristic
function of the set K. Denote by rF pψq the smallest real pole of the meromorphic map s ÞÑ
 volume, homogeneous forms and globally semianalytic domains 11

xζF psq, ψy defined in (1.9) and by mF pψq ě 1 its order. Then, the pair prF pψq, mF pψqq is
well-defined and for T ě 1,

Voln pSF pK, bpT qqq ! T n ¨ cpT qrF pψq ¨ |log pcpT qq|mF pψq´1 . (1.11)

(2) – assume that pH3 q is verified, and let prF pKq, mF pKqq be the pair introduced in pH3 q.
Then, rF pKq is a strictly positive rational number. Moreover, as the parameter T tends to
infinity, it holds that

Voln pSF pK, bpT qqq — T n ¨ cpT qrF pKq ¨ |log pcpT qq|mF pKq´1 . (1.12)

Here, one has that prF pKq, mF pKqq “ pp, 1q when the set of homogeneous forms Fpxq has
smooth complete intersection over K. Furthermore, under this smoothness assumption,
the condition pH3 q can be guaranteed under pH1 q provided that pH2 q is strenghtened as
follows : the algebraic variety ZR pFq should intersect the interior of the set K.

(3) – assume that the set K is star-shaped with respect to the origin and that it contains
the origin in its interior. Denote by rF the smallest real pole of the meromorphic distri-
bution ζF and by mF ě 1 its order. Then, rF is a well-defined rational number lying in
the interval p0, n{ds. Furthermore, there exists a constant γF pKq ą 0 such that, as the
parameter T tends to infinity,

Voln pSF pK, bpT qqq “ pγF pKq ` op1qq ¨ T n ¨ cpT qrF ¨ |log pcpT qq|mF ´1 . (1.13)

The pairs prF pKq, mF pKqq and prF , mF q involved in the above estimates can moreover be de-
termined by resolution of singularities in a finite number of steps. Also, the rationals ´rF pKq{2
and ´rF {2 are roots of the Sato–Bernstein polynomial BF psq with multiplicity at most mF pKq
and mF , respectively.

It must be emphasised that the error term in the volume estimate (1.13) can be
made explicit depending on invariants attached to the homogeneous form }Fpxq}2 —
see Section 2.2 in Chapter 2 for details.

Specialising Case (3) to the situation where bpT q is constant equal to 1 and where
K “ r´1, 1sn is the closed cube centered at the origin with sidelength 2, one obtains a
statement interesting on its own :

Corollary 1.2 (Finiteness of the volume of domains determined by homogeneous forms

inequalities). The set of points x P Rn determined by the system of homogeneous forms in-
equalities
|F1 pxq| ď 1, |F2 pxq| ď 1, . . . , |Fp pxq| ď 1
has a finite n–dimensional volume if, and only if, the distribution ζF has a simple pole at n{d
which is, furthermore, its smallest real pole.
 order and multiplicity of the log–canonical threshold 12

1.3 order and multiplicity of the log–canonical

threshold
The conclusion of Theorem 1.1 prompts a finer analysis of the relation between the
roots of the Sato–Bernstein polynomial of a multivariate polynomial and the poles of
the corresponding zeta function. Let then Ppzq P C rzs be a (non–necessarily homo-
geneous) polynomial in n variables z “ pz1 , . . . , zn q with associated Sato–Bernstein
polynomial BP psq P Qrss. The roots of BP psq happen to be closely related to the poles
of the complex zeta distribution ζpP associated to Ppzq, which is defined for all complex
numbers s with non–positive real parts and all elements ψ in the space of complex-
valued compactly supported smooth functions C8 n
c pC q by the formula
A E ż
ζP psq, ψ “
p |Ppzq|´2s ¨ ψpz, zq ¨ dz ^ dz. (1.14)
Cn

As justified in Chapter 3, the distribution ζpP also admits a meromorphic extension to

the entire complex plane.

Malgrange [65] proved that the roots of BP psq are determined by the geometry of
the algebraic variety ZC pPq “ tz P Cn : Ppzq “ 0u in the following sense : if ρ P Q is a
root of BP psq, then expp2iπρq is an eigenvalue of the local monodromy of Ppzq at some
point of ZC pPq; conversely, all eigenvalues are obtained this way. It can then be shown
that, if expp2iπρq is an eigenvalue of monodromy, then ρ is a pole of the complex zeta
distribution ζpP associated to Ppzq; here again, each eigenvalue arises this way — see [8]
and [64]. For detailed studies on the location of the poles of ζpP , the reader is referred
to [60] and [62].

The multiplicity of the roots of the Sato–Bernstein polynomial and their relation to
the zeta distribution are not as well understood. The only known general result seems
to be due to Saito and is concerned with the roots of B̃P psq “ BP psq{ps ` 1q, which
is always a non–constant rational polynomial when Ppzq is singular over C (see Sec-
tion 2.1 in Chapter 2 for a justification of this claim). Denoting by ρ̂P the largest root of
B̃P psq, it is indeed proved in [68] that any root ρ̂ of this polynomial lies in the interval
rρ̂P ´ n, ρ̂P s Ă r´n ´ 1, 0s and has multiplicity at most n ` ρ̂P ` ρ̂ ` 1 ď n ` 2ρ̂P ` 1 ď
n ` 1. In some cases where the polynomial Ppzq is not “too singular” (in a suitable
sense), the multiplicity of the roots of its Sato–Bernstein polynomial can also be de-
duced from an explicit formula for BP psq — see [18] and [19] for further details.

The statement below characterises the multiplicity of the largest root r̂P ą 0 of
BP p´sq (recall that all roots of BP psq are rational and negative). It is known, see [55,
§10.6], that this root coincides with the smallest real pole of the distribution ζpP . It
is referred to as the (complex) log–canonical threshold of the polynomial Ppzq. More
precisely, the following statement elucidates the relationship between the multiplicity
 generic unimodular distorsions of a family of homogeneous forms 13

of the log–canonical threshold as a root of the polynomial BP p´sq and its order as a
pole of the distribution ζpP . It answers a classical problem in the theory of D–modules.
Theorem 1.3 (Order and multiplicity of the log–canonical threshold). Let Ppzq P Crzs
be a non–constant polynomial and let r̂P be its log–canonical threshold. Then, the multiplicity
of ´r̂P as the largest root of the Sato–Bernstein polynomial Bp psq is also the order of r̂P as the
smallest real pole of the meromorphic distribution ζpP .

1.4 generic unimodular distorsions of a family of

homogeneous forms
Beyond its instrinsic interest, some of the main applications of the volume estimate
obtained in Theorem 1.1 are to determine, on the one hand, the generic existence of
solutions to Diophantine inequalities defined by the system of homogeneous forms
Fpxq and, on the other, the generic asymptotic behavior of variants, natural in this con-
text, of the counting function (1.2). A property is here said to be generic if its holds
for almost all unimodular transformations g when SLn pRq is equipped with its Haar
measure.

More precisely, given g P SLn pRq, T ě 1 and reals a ă b, set

´ ¯
SF pg, a, b, T q “ }pF ˝ gq}´1 pra, bsq X Bn pT q
r

“ tx P Bn pT q : a ă }pF ˝ gq pxq} ď bu (1.15)

“ SF pBn , g, T , bq z SF pBn , g, T , aq , (1.16)
(1.1)

where Bn pT q denotes the closed Euclidean ball centered at the origin with radius T ě 1,
and where Bn “ Bn p1q. (The reader should not be confused by the fact that the rôle
of the variable a in the above relation (1.16) differs from the one it plays in the defini-
tion (1.1).)

Athreya and Margulis [4] gave impetus to the study of the Diophantine corollaries
that can be derived from the volume growth of the set r SF pg, a, b, T q. Specifically, their
focus was on the case where d “ 2 and p “ 1, and where Fpxq “ F1 pxq is a quadratic
form in n ě 3 variables — denote it by Qpxq for convenience – such that for some
constant cQ ą 0 and all reals a ă b, it holds that
´ ¯ ´ ¯
Voln r SQ pg, a, b, T q “ cQ ¨ pb ´ aq ¨ T n´2 ` o T n´2 (1.17)

as T tends to infinity. From [33, Lemma 3.8], this is a full measure condition on the
set of quadratic forms. Athreya and Margulis show in [4, Theorem 1.1] that, under
these assumptions, for every δ ą 0 and for almost all g P SLn pRq, there are constants
 generic unimodular distorsions of a family of homogeneous forms 14

` ˘
κQ pgq ě 1 and εQ pgq ą 0 such that for all ε P 0, εQ pgq , there exists a nonzero integer
x P Zn satisfying the inequalities

|pQ ˝ gq pxq| ă ε and }x} ď κQ pgq ¨ ε´p1{pn´2q`δq . (1.18)

This can be rephrased as follows : almost all unimodular twists of a quadratic form for
which the volume estimate (1.13) holds with parameters pr, mq “ p1, 1q when K “ Bn
satisfies a uniform Diophantine approximation property with exponent 1{pn ´ 2q ` δ
for any δ ą 0 (in the sense that the system (1.18) admits a nontrivial solution for all
ε ą 0 small enough). The last section of [4] raises the problem of generalising this
uniform Diophantine approximation property to any homogeneous form.

Athreya and Margulis’ result is motivated by the problem of proving quantitative

forms of the Oppenheim Conjecture. Recall that in its simplest form, this conjecture,
solved by Margulis in [66], claims that for any indefinite quadratic form Qpxq in n ě 3
variables which is not a real multiple of a form with rational coefficients, the origin is
an accumulation point of the set Q pZn q Ă R. From a metric point of view, the problem
of providing a quantitative estimate for this statement in the case of a generic indefinite
quadratic form (in a metric sense which varies depending on the context) has been un-
dertaken by many authors. The existing literature in this area is therefore vast. For the
purpose of the present work, it is relevant to mention the contribution by Bourgain [17]
which is concerned with ternary diagonal forms which are generic in a suitable sense.
In [37], Ghosh, Gorodnik and Nevo use spectral methods to obtain general results of
this type which imply the case n “ 3 of the above uniform approximation result (1.18).
They also provide a heuristic to show that a homogeneous form of degree d in n ą d
variables, considered as generic in a suitable sense, should admit an exponent of uni-
form approximation of the form 1{pn ´ dq ` δ for any δ ą 0. For more background on
such metric and effective statements in the context of the Oppenheim Conjecture, the
reader is referred to the works by Gosh, Kelmer and Yu [39], by Buterus, Götze and
Hille [13] and to the references within.

As far as the problem raised by Athreya and Margulis of generalising the uniform
approximation result (1.18) to the case of unimodular twists of any homogeneous form
is concerned, two notable contributions should be mentioned. The first one, due to
Kelmer and Yu [53], proves the sought approximation property in the case of the inde-
finite diagonal homogeneous form
p
ÿ n
ÿ
F1 pxq “ x2d
i ´ x2d
i , (1.19)
i“1 i“p`1

where 1 ď p ď n ´ 1. More precisely, the particular case of [53, Corollary 2] relevant to

this discussion establishes the required uniform Diophantine approximation property
with exponent 1{pn ´ 2dq ` δ for any δ ą 0 provided that n ą 2d. The proof relies on
 generic unimodular distorsions of a family of homogeneous forms 15

the fact that for this homogeneous form, the volume estimate (1.13) holds with expo-
nents pr, mq “ p1, 1q when K “ Bn .

The second contribution that should be mentioned is that of Kleinbock and Sken-
deri [54, Theorem 1.3] which, when specialised to the case of homogeneous forms,
establishes the following claim : a generic unimodular twists of a system of homoge-
neous forms Fpxq satisfies a uniform Diophantine approximation property at a given
scale (i.e. for a certain function of ε ą 0 on the right–hand side of the second inequality
in (1.18)) provided that a series converges. This series depends on the scaling function
and is expressed as the volume of sets of the form r SF pg, a, b, T q, where a and b are
taken as functions of T . However, Kleinbock and Skenderi do not determine the volu-
me of such sets.

The sharp volume estimate obtained in Case (3) of Theorem 1.1 yields the following
solution to the problem raised by Athreya and Margulis in the case of any system of
homogeneous forms. This solution contains as particular cases the above–mentioned
results by Kelmer and Yu [53] and by Kleinbock and Skenderi [54].

Theorem 1.4 (Uniform Diophantine approximation property of generic unimodular

twists of systems of homogeneous forms). In the notation of Theorem 1.1, let

pr, mq “ prF , mF q P Q` ˆ N

be the pair such that the volume estimate (1.13) holds for the p–tuple Fpxq “ pF1 pxq, . . . , Fp pxqq.
Let f : R` Ñ p1, 8q be a non–decreasing function tending to infinity at infinity and satisfying
the growth condition ` ˘
f 2j`1
lim sup ă 8 (1.20)
jÑ`8 f p2j q
and also the limit condition
2j
−Ñ 0.
d jÑ8
(1.21)
f p2j q
Assume furthermore that
8
ÿ 2jr
ă 8.
´d ˇm´1
ˇ ´ ¯ˇ
n´rd ˇ
j“0 f p2j q ¨ ˇlog 2j ¨ f p2j q ˇ

Then, for almost every g P SLn pRq, there exists constants κF pg, fq ě 1 and εF pg, fq P p0, 1q
such that for any 0 ă ε ă εF pg, fq, the system of Diophantine inequalities
´ ¯
´1
}pF ˝ gq pmq} ă ε and 1 ď }m} ď κF pg, fq ¨ f ε

admits a solution in m P Zn .
 generic unimodular distorsions of a family of homogeneous forms 16

The growth condition (1.20) means that the function f is required not to admit
"abrupt" variations. As for the limit condition (1.21), it is not a restrictive one inas-
much as it is the analogue, in this context, of the limit condition (1.4) meant to avoid
the cases where the considered inequalities are trivially satisfied.

Theorem 1.4 supersedes the above–mentioned heuristic by Gosh, Gorodnik and

Nevo [37] providing an exponent of uniform approximation of the form 1{pn ´ dq ` δ
for any δ ą 0 when n ą d in the following sense : it shows that the underlying
approximation property relies on the values of the pair pr, mq for which the volume
estimate (1.13) holds. The particular case when pr, mq “ p1, 1q then recovers the
heuristic bound proposed by Gosh, Gorodnik and Nevo [37].

Fixing once again g P SLn pRq, and reals T ě 1 and a ă b, define now the counting
function
´´ ¯ ¯
NF pg, a, b, T q “ # SF pg, a, b, T q X Z
r r n

“ # tm P Zn X Bn pT q : a ă }pF ˝ gq pmq} ď bu . (1.22)

Thus, N r F pg, a, b, T q is the lattice points counting function associated to the difference
set SF pBn , g, T , bq z SF pBn , g, T , aq defined from (1.1).

Another problem raised by Athreya and Margulis in [4] is to determine the generic
asymptotic behavior of the function N r F pg, a, b, T q as the parameter T tends to infinity. In
the case of a quadratic form Qpxq in n ě 3 variables satisfying the volume growth (1.17),
Theorem 1.2 in [4] shows that for all δ ą 0 and almost every g P SLn pRq, there exists a
constant cQ,g ą 0 such that
´ ´ ¯¯
N
r F pg, a, b, T q “ pb ´ aq ¨ cQ,g ` O T ´pn´5q{2`δ ¨ T n´2 (1.23)

as T tends to infinity. This has been extended in [38] to the case of shifts of quadratic
forms. More generally, the reader is referred to [14, 39] and to the references within
for the state of the art in metric counting problems related to quadratic forms.

When the homogenous form has degree d ě 3, the only result towards a solution
to this counting problem the authors are aware of is due to Kelmer and Yu [53] (and
its slight extension to a system of closely related forms by Bandi, Ghosh and Han [7]).
It deals with the case of the homogeneous form F1 pxq defined in (1.19) when n ą 2d.
Specifically, Kelmer and Yu are concerned with the related but slightly different pro-
blem when the bounds a “ apT q and b “ bpT q of the real intervals depend on the pa-
rameter T and satisfy the shrinking target property that`bpT q “ apT q ` c ¨ `T ´γ for some˘˘
c, γ ą 0. Then, [53, Theorem 1] states that for every β P 0, 2 pn ´ d ´ γq { n2 ` n ` 4
and for almost all g P SLn pRq,
´ ´ ¯¯
N
r F g, T , apT q, apT q ` c ¨ T ´γ “ c ¨ T n´2d´γ ¨ cF ,g ` O T ´β ¨ T n´2 ,
` ˘
1 1
(1.24)
 generic unimodular distorsions of a family of homogeneous forms 17

where cF1 ,g ą 0.

The proof of both counting estimates (1.23) and (1.24) share two common features :
on the one hand, they crucially rely on the fact that the corresponding algebraic
domains r SQ pg, a, b, T q and r
SF1 pg, a, b, T q are “large enough” in the sense that their
volumes grow at least polynomially. Equivalently, when considering the volume
estimate (1.13) in the case that K “ Bn and bpT q “ 1, this is saying that the pairs pr, mq
corresponding to each of these forms are both such that the exponent r does not take
the extremal value n{d, where d denotes the degree of either forms Qpxq or F1 pxq (in
other words, one requires that 0 ă r ă n{d). On the other hand, the power savings in
the error terms in (1.23) and (1.24) strongly rely on the fact that, in both cases, it holds
that m “ 1 (in fact, as mentioned above, pr, mq “ p1, 1q for the two forms Qpxq and
F1 pxq under consideration).

The following statement generalises these observations and settles the counting
problem raised by Athreya and Margulis in the more general case of systems of homo-
geneous forms in n ě 3 variables.

Theorem 1.5 (Generic asymptotic behavior of the counting function of unimodular

twists of systems of homogeneous forms). Assume that n ě 3 and fix real numbers 0 ď
a ă b. In the notation of Theorem 1.1, let pr, mq “ prF , mF q P Q` ˆ N be the pair such that
the volume estimate (1.13) holds for the p–tuple Fpxq “ pF1 pxq, . . . , Fp pxqq. Then, under the
assumption that r ă n{d, there exists, for almost all g P SLn pRq, a constant γF,g ą 0 such that
˜ ˇ ˆ ˙ˇm´1 ¸
b ˇ ´ a ¯ˇm´1
N
` ˘ n´rd r
ˇ ˇ r
r F pg, a, b, T q “ γF,g ` op1q ¨ T ¨ b ¨ ˇˇlog ´ a ¨ ˇlog d ˇ (1.25)
ˇ ˇ ˇ
Td ˇ T

as the parameter T tends to infinity.

As a matter of fact, this theorem is deduced in Chapter 4 upon specialising a much

more general statement to the case of the homogeneous form PF pxq “ }Fpxq}2 . This
more general statement deals with any real homogeneous form Ppxq and provides an
explicit form for the error term in (1.25) as a function of various invariants attached
to Ppxq. It suffices to say here that in the case m “ 1, it enables one to recover the
power saving estimates obtained in (1.23) and (1.24) with a comparable strength.
When m ě 2 however, the presence of logarithmic factors in the volume estimate (1.13)
imposes that the induced saving can only be of the order of a power of a logarithm.

Various other problems related to metric twists of algebraic functions have been
considered in the literature. To cite but a few relevant to this work, Vanderkam [77]
studies the metric behavior of a counting function similar to (1.22) when considering
all coefficients of a polynomial as variables behaving independently. The paper [42]
also surveys the state of the art in the related problem of counting integral points on
random algebraic curves. Finally, in [1] is considered the problem of approximating
 the number of solutions to homogeneous forms inequalities 18

values taken by a polynomial twisted by a random nonlinear transformation, namely

a vertical shift.

1.5 the number of solutions to homogeneous

forms inequalities
Let K Ă Rn be a set meeting the assumptions pH1 q, pH2 q and pH3 q (recall here that
pH3 q is necessarily satisfied when K is star-shaped with respect to the origin and that
it contains the origin in its interior). Given a parameter α ą 0, consider the case where

bpT q “ T d´α (1.26)

in the definitions of the set SF pK, bpT qq in (1.7) and of the corresponding counting
function NF pK, bpT qq in (1.8). Note that the limit condition (1.4) then holds. In order to
single out the rôle of the parameter α in the forthcoming statements, define
´ ¯ ! )
S:F pK, T , αq “ SF K, T d´α “ x P T ¨ K : }Fpxq} ď T d´α (1.27)
(1.7)

and ´ ¯ ´ ¯
NF: pK, T , αq “ NF K, T d´α “ # S:F pK, T , αq X Zn . (1.28)
(1.8)

The focus on a power function such as in (1.26) is motivated by the following

deterministic counting result due to Sarnak : assuming that the zero set ZK pFq :“
tx P K : Fpxq “ 0u does not lie in a linear subspace of dimension n ´ p and that
the smooth complete intersection condition determined by the non-vanishing of the
map (1.6) over the set K holds, it is established in [70, Appendix 1] that the estimate

NF: pK, T , αq ! T n´pα ` T n´p´δ (1.29)

is verified for some δ ą 0 provided that the set K is "nice". Sarnak does not make
explicit the meaning of the latter condition; his proof nevertheless certainly works if,
say, K meets the assumptions pH1 q and pH2 q and if, furthermore, it is convex and has
piecewise smooth boundary. Moreover, the arguments of the proof show that one can
then take the value δ “ 1{pndq.

The fact that there should exist a power saving factor δ ą 0 in the error term of
the estimate (1.29) has a particularly interesting consequence : as shown in [73, p. 177]
(with the choice of α “ 1{2 in the above notation), the right-hand side of (1.29) then
bounds nontrivially the number of solutions to the system of Diophantine equations
Fpmq “ 0 when m P T ¨ K. Inspired by this implication, Sarnak conjectures at the end
of his proof that an estimate of the form (1.29) should hold without the assumption of
smoothness in the following sense : assuming that the zero set ZK pFq is non empty, not
 the number of solutions to homogeneous forms inequalities 19

contained in a linear space of dimension n ´ p and that the set K is again "nice", there
should exist δ ą 0 such that
´ ¯
NF: pK, T , αq ! Voln S:F pK, T , αq ` T n´p´δ . (1.30)

Interpreting the assumption that the set K is "nice" in the sense that the conditions
pH1 q and pH2 q are met, Case (2) in Theorem 1.1 confirms that the above upper bound
indeed reduces to the inequality (1.29) in the case of smooth complete intersection.

It nevertheless turns out that a power saving in the error term such as in (1.30) is
not unconditionnally true in the general case just assuming that the zero set ZK pFq is
not empty, not contained in a linear subspace of dimension n ´ p and that the set K is
"nice" (or regular) enough. To see this, consider the case where p “ 1 and where the
homogeneous form under consideration is
n
ź
Fn pxq :“ xi , (1.31)
i“1

which has degree d “ n. The zero set of Fn pxq is here the union of the coordinate
hyperplanes. Taking K “ Kn to be the unit cube r0, 1sn and choosing any α ą 0, the
counting function NF:n pK, T , αq is closely related to the generalised divisor summatory
function ÿ
∆n ptq “ 1
1ďFn pxqďt

in the sense that

0 ď ∆n T n´α ` n ¨ pT ` 1qn´1 ´ NF:n pK, T , αq

` ˘

npn ` 1q ´ ¯
ď ¨ pT ` 1qn´2 ` n ¨ ∆n´1 T n´1´α . (1.32)
2

Indeed, this follows upon considering the contributions to NF:n pK, T , αq of the
coordinate hyperplanes separately and, as far as the upper bound is concerned, upon
furthermore noticing this elementary fact : the number of integer solutions to the ine-
quality 1 ď Fn pxq ď T n´α when xi ą T for some coordinate 1 ď i ď n is bounded above
by n times the number of integer solutions to the inequality 1 ď Fn´1 pyq ď T n´1´α ,
where y P Rn´1 .

Now, it is known that there exists a polynomial Qn ptq of degree n ´ 1 such that
´ ¯
1´ε
∆n ptq “ t ¨ Qn plog tq ` O t

for some ε ą 0. (Determining an optimal exponent for the error term is the Piltz
Divisor Problem, which reduces to the better known Dirichlet Divisor Problem when
 the number of solutions to homogeneous forms inequalities 20

n “ 2.) More precisely, the coefficients of the polynomials Qn ptq can be evaluated from
the formula ´ ¯
Qn plog tq “ Ress“1 ts´1 ¨ ζn psq ¨ s´1
and from the Laurent series expansion of the Riemann zeta function
8
1 ÿ p´1q
ζpsq “ ` ¨ γk ¨ ps ´ 1qk
s ´ 1 k“0 k!

valid in a neighbourhood of s “ 1. Here, pγk qkě0 denotes the sequence of the Stieltjes
constants. Explicit calculations of the coefficients of Qn ptq are carried out, for instance,
in [56, Theorem 1]. In particular, one obtains that

t2 ´
2
¯
Q2 ptq “ t ` p2γ ´ 1q and Q3 ptq “ ` p3γ ´ 1q ¨ t ` 3γ ´ 3γ ` 3γ1 ` 1 ,
2
where γ « 0.577 is the Euler–Mascheroni constant and where γ1 « ´0.073. Since
elementary calculations show that the volume of the region S:Fn pK, T , αq grows, up to
a multiplicative constant, as T n´α ¨ plog T qn´1 (this is saying that prFn , mFn q “ p1, nq in
the notations of Theorem 1.1), the estimates (1.32) imply that there can be no power
saving in the error term when n “ 3 and, more generally, in the generic case where
the coefficients do not cancel out suitably.

This counterexample makes clear the nature of the obstacle to overcome for
Sarnak’s estimate (1.30) to hold : in the presence of singularities (such as the origin in
the case of the homogeneous form (1.31)), logarithmic terms appearing in the volume
term need not be compensated by a power saving in the error term. The presence
of such logarithmic terms should be seen as an indication of the fact that the zero
set ZK pFq is in some sense "too flat" (whithout necessarily being contained in a linear
space as illustrated by the case of the homogeneous form (1.31)). In the particular case
of a planar curve, this link between flatness and behavior of the counting function of
Diophantine inequalities has been observed and formalised (in a related but slightly
different context) by Vaughan and Velani [78] and then by Huang [44].

With this in mind, the final goal of this paper is to determine, under the assump-
tions pH1 q, pH2 q and pH3 q, conditions related to the "the level of flatness" of the zero
set ZK pFq under which Sarnak’s claim holds. As this goal is achieved working in the
context of semialgebraic geometry, it is natural to assume furthermore that the set
K is semialgebraic, meaning that it is defined as a finite union of sets that can be
represented as finitely many equalities and inequalities involving polynomials. The
statement of the results furthermore depends on the choice of a compact semialgebraic
set K containing K in its interior and being contained in the open set U introduced
in (1.10). This is because to make the values of the power savings effective whenever
they hold, one indeed needs to consider the behaviour of the set of homogeneous
 the number of solutions to homogeneous forms inequalities 21

forms Fpxq over a compact neighbourhood of K.

Define the dimension dim pZK pFqq of the algebraic variety ZR pFq over the set K
as the dimension of the tangent vector space at any non singular point of ZR pFq X K
(this is well–defined). Letting codim pZK pFqq “ n ´ dim pZK pFqq, set for the sake of
simplicity of notation

τF pKq “ dim pZK pFqq and τpF pKq “ codim pZK pFqq .

For instance, if the variety ZR pFq has smooth complete intersection over K and
that the compact neighbourhood K is chosen small enough around K so that
this property is preserved over K, then τpF pKq “ p. The exponent in the error
term in (1.29) valid in the smooth case can then be interpreted as τF pKq ´ δ. This
is this form of the exponent that is generalised in what follows to the non–smooth case.

The existence of a power saving error term is expressed as a condition depending

on a measure of "the level of flatness" of the zero set ZK pFq defined as follows : given
a unit vector v P Sn´1 and a real σ, set first

vK pσq :“ tx P Rn : v ¨ x “ σu .

The flatness of the zero set ZK pFq along such an affine space is measured by the volume
section
µF,K pv, σ, εq “ Voln´1 x P K X vK pσq : }Fpxq} ď ε ,
`␣ (˘

where ε ą 0. The "biggest" such volume section is then defined as

MF pK, εq “ sup sup µF,K pv, σ, εq .

vPSn´1 σPR

From this, the measure of flatness relevant to the problem under consideration is de-
fined as ˆ ˙
log MF pK, εq
qF pKq “ lim inf . (1.33)
εÑ0` log ε
This is a measure of the "biggest level of flatness" that can be achieved when inter-
secting the sublevel set tx P K : }Fpxq} ď εu with affine hyperplanes in the following
sense : the smaller it is, the more flat the intersection of the sublevel set in some di-
rection. This measure of flatness is a well–defined real number under the assumption
that ZK pFq ‰ H. It is furthermore shown in Chapter 6 (see the last section therein)
that the real qF pKq can then be estimated effectively in a finite number of steps (in a
suitable sense). The statement of the main result in this section (Theorem 1.6 below)
relies on a comparison between the value of qF pKq and the volume growth exponent
rF pKq present in the statement of Theorem 1.1 and part of the assumption pH3 q.

Specifically, the following provides a solution to Sarnak’s problem generalising the

smooth case :
 the number of solutions to homogeneous forms inequalities 22

Theorem 1.6 (Counting solutions to homogeneous forms inequalities). Recall that the
conditions pH1 q, pH2 q and pH3 q are assumed to hold and that the above defined sets K Ă K
are assumed to be semialgebraic. Fix a parameter α ą 0 and assume that the measure of
flatness (1.33) is big enough in the sense that
" *
τF pKq
qF pKq ą max rF pKq ´ 1, n ´ 1 ´ rF pKq ¨ . (1.34)
τpF pKq
Then, there exists a real δpK, Kq ą 0 depending on the semialgebraic sets K and K such that
the following estimate with power saving error term holds :
´ ¯
NF: pK, T , αq ! Voln S:F pK, T , αq ` T τF pKq´δF pK,Kq . (1.35)

Furthermore, the inequality (1.34) is satisfied when the following two conditions are simultane-
ously met : the set of polynomials Fpxq has smooth complete intersection over K and the zero
set ZK pFq is not contained in a linear subspace of dimension n ´ p, where n ´ p “ τF pKq.
It should be noted that in the case of the counterexample (1.31) to Sarnak’s claim,
the inequality (1.34) is indeed not met since in any semialgebraic compact neighbour-
hood K` ` n pFn pK`
n of Kn “ r0, 1s , one easily checks that τ
`
n q “ 1, that rFn pKn q “ n ´ 1 and
`
that qFn pKn q “ 0 (this last relation follows from the fact that the quantity MFn pK` n , εq
is constant independent of ε ą 0 as can be seen upon considering the volume of the
sections of the sublevel set |Fn pxq| ă ε with coordinate hyperplanes).

In fact, Theorem 1.6 as stated above is an extremely abridged version of a

class of statements providing a much more general picture of the behaviour of the
´ function¯ NF pK, T , αq. Thus, it is established in Chapter 5 that the volume
:
counting
Voln S:F pK, T , αq determines the exact asymptotic order of this counting function as
T tends to infinity in the case that α P p0, 1q (i.e. when the algebraic set S:F pK, T , αq is
"large enough") and also that the counting function is, up to a multiplicative constant,
bounded above by this volume in the case α “ 1. These estimates are further refined
when α P p0, 1q under the stronger assumptions of Case (3) stated in the above
Section 1.2 : an asymptotic expansion with power saving error term is then established
for the quantity NF: pK, T , αq as T tends to infinity.

In the complementary case when α ą 1 (i.e. when the algebraic set S:F pK, T , αq is
"small"), it is proved in Chapter 6 that an estimate of the form (1.35) can still hold in
the degenerate case when qF pKq ď rF pKq ´ 1 under an explicit condition (which is, of
course, still not satisfied by the counterexample (1.31)). Also, in all instances where
an estimate with power saving error term holds, a range of admissible values of the
exponent δF pK, Kq ą 0 is explicitly determined.

Theorem 1.6 complements and extends a number of previously known results.

Thus, in the case of a single positive homogeneous function Fpxq (not necessarily
 structure of the memoir 23

assumed to be a polynomial), Randol [67] establishes a counting bound with power

saving error term depending on the decay of a Fourier operator acting on the
boundary surface tF “ 1u, which is supposed to be regular enough (in a suitable
sense). This has been slightly generalised by Colin de Verdière [26] who parallelly con-
siders the behavior of the counting bound when rotating randomly the domain tF ď 1u.

In the general case where one considers the problem of counting rational points
near manifolds (Theorem 1.6 being concerned with the particular case where the
manifold is a homogeneous algebraic variety), the most powerful results have been
obtained under various assumptions of non degeneracy (which often amount to im-
posing constraints on the behavior of the local curvature). In this respect, Beresnevich
et al. [11] show that, provided that a certain matrix of second order derivatives of a
prametrisation of a twice continuously differentiable manifold admits a determinant
bounded below by a strictly positive constant, one can recover a counting bound with
a power saving error term. Under the stronger assumption that the Gaussian curvature
remains strictly positive, Huang [45] bounds from above the counting function of
the number of rational points near a manifold by the volume of a neighbourhood
of this manifold (which takes a particularly simple form under the considered
assumption). This establishes the heuristics that the number of rational points in a
"nice enough domain" should be given by the volume of this domain provided that
it is "large enough". In the case that one considers a domain with analytic boundary,
Greenblatt [40] proves a counting bound related to Theorem 1.6 by the use of suitable
resolutions of singularities. This nevertheless comes with an additional restriction not
present in the above Theorem 1.6 : in his work, the boundary of the domain must be
locally the graph of an analytic function which separates the interior of the domain
from its complement in exactly two pieces.

Finally, it should be noted that although Theorem 1.6 holds in the binary case
of n “ 2 variables, explicit and sharper results are then known if one considers a
single homogeneous form with integer coefficents (one is then dealing with a Thue
inequality). To see recent progress in this well–studied topic, see the works by Fouvry
and Waldschmidt [35] , by Stewart & Xiao [74] and the references within.

1.6 structure of the memoir

The following table summarises the sections where the statements presented in the
introduction are established. As almost all of these have not been stated in full
generality for the sake of the coherence of the exposition, the table also provides the
statements either generalising them or relating them to other aspects of the problem
they are concerned with.
 structure of the memoir 24

Statement Proof Related extensions

Corollary 2.9
Proposition 2.17
Theorem 1.1 Sections 2.2 & 2.3
Proposition 2.18
Theorem 3.6
Theorem 1.3 Chapter 3
Theorem 1.4 Chapter 4
Theorem 1.5 Chapter 4 Theorem 4.2
Chapter 5 when 0 ă α ď 1 Theorem 5.1
Theorem 1.6
& Chapter 6 when α ą 1 Theorem 6.1

acknowledgments This memoir is the conclusion of a long project during which

the authors exchanged ideas with many people. In this respect, they would like to
thank Profs. Victor Beresnevich, Raf Cluckers, Pietro Corvaja, Roger Heath–Brown,
Tobias Kaiser, Mircea Mustaţă and Mihnea Popa, and also Quentin Guignard and
Gareth O. Jones.

The first-named author would especially like to thank Dr. Detta Dickinson who
quite unexpectedly initiated him to this area of research in 2011 when he started his
PhD under her supervision : as far as he is concerned, the problem she suggested him
to work on somehow gave impetus to this project.
2 V O L U M E E S T I M AT E S

The goal of this chapter is to establish Theorem 1.1. The set K denotes either the field
of reals R or the field of complex numbers C. Furthermore, Ppyq is a non–constant
polynomial of degree q ě 1 with coefficients in K which will be specialised when
needed to the homogeneous form }Fpyq}2 .

The following form of Hironaka’s Theorem on simultaneous resolution of singula-

rities stated in [5, p.147] (when l “ 1) and in [79, Theorem 2.8] (when l ě 2) will be
used repeatedly :

Theorem 2.1 (Simultaneous Resolution of Singularities, Hironaka, 1964). Let l ě 1 be

an integer and let f1 , . . . , fl be analytic maps defined over an open set U Ă Kn containing the
origin. Assume that given 1 ď i ď l, the map fi is non–constant and satisfies fi p0q “ 0. Then,
there exists an open set W Ă U containing the origin, an n–dimensional K–analytic manifold
M and a K–analytic map g : M Ñ W such that the following hold :

‚ the map g is proper (that is, the preimage of a compact set is compact) and surjective;

‚ the map g is an analytic isomorphism from MzM0 to WzW0 , where

l
ď l
ď
M0 “ ty P M : fi pg pyqq “ 0u and W0 “ tx P W : fi pxq “ 0u ;
i“1 i“1

‚ given any point y0 P M0 , there exists local coordinates y “ py1 , . . . , yn q P M centered

at y0 such that for all 1 ď i ď l,

fi pgpyqq “ αi pyq ¨ yki . (2.1)

Furthermore, the Jacobian of the change of variables x “ gpyq is such that

Jacg pyq “ βpyq ¨ yh . (2.2)

In these relations, α1 is a constant map equal to ˘1; α2 , . . . , αn and β are non–vanishing

anaytic maps, and ki and h are n–tuples of nonnegative integers.

2.1 local and global sato–bernstein polynomials

This section is devoted to recalls on and extensions of the elements of the Sato–
Bernstein theory needed for the purpose of this work. Classical references where the

25
 local and global sato–bernstein polynomials 26

known claims stated hereafter are proved include [46] and [50].

Given a non–constant polynomial Ppyq P K rys, there exists a univariate nonzero

polynomial bpsq P Krss and a differential operator D py, s, Bq with coefficients in K
such that the following identity holds formally :

D py, s, Bq ¨ Ppyqs`1 “ bpsqPpyqs . (2.3)

Furthermore, the operator D py, s, Bq can be chosen as a (non–cummutative) poly-

nomial in the variables y “ py1 , . . . , yn q, in an additional variable s and in the set of
partial derivatives B “ pB1 , . . . , Bn q. Clearly, the set of all those polynomials bpsq P Krss
for which a differential operator satisfying (2.3) exists forms an ideal in the ring Krss.
The (global) Sato–Bernstein polynomial BP psq of Ppyq is then defined as the monic genera-
tor of this ideal; in other words, it is the monic polynomial of smallest degree satisfying
the relation (2.3) for some differential operator D. The following fundamental result
due to Kashiwara [51] implies that BP psq has rational coefficients.

Theorem 2.2 (Kashiwara, 1976). The Sato–Bernstein polynomial BP psq of a polynomial

Ppyq P Krys is a split polynomial all of whose roots are rational and negative.

For instance, in the case that Ppyq “ P1 pyq “ ni“1 y2i , one obtains when taking the
ř

Laplacian as the differential operator that

n
ÿ ´ n¯
Bi2 P1 pyqs`1 “ 4 ps ` 1q ¨ s ` ¨ P1 pyqs .
i“1
2

It is then easy to deduce from` this ˘relation that the corresponding Sato–Bernstein
polynomial is BP1 psq “ ps ` 1q ¨ s ` n2 .

The proof of Kashiwara’s Theorem 2.2 shows that computing a Sato–Bernstein poly-
nomial can be achieved by resolving singularities (although more efficient algorithms
are known — see, e.g., [16]). In view of this, the case where

Ppyq “ P2 pyq “ yk

is a monomial with integral powers k “ pk1 , . . . , kn q P Nn0 z t0u is of particular interest.

As established in [46, p.49], the relation
¨ ˛
ki
n ź
ź
B k P2 pyqs`1 “ ˝ pki s ` jq‚¨ P2 pyqs
i“1 j“1

implies that the Sato–Bernstein polynomial of P2 pyq is

ki ˆ
n ź ˙
ź j
BP2 psq “ s` .
i“1 j“1
ki
 local and global sato–bernstein polynomials 27

In the general case, specialising the identity (2.3) to the case where bpsq “ BP psq
and s “ ´1 reveals that BP psq is always divisible by the factor s ` 1. Furthermore,
BP psq “ s ` 1 if and only if ZC p∇Pq “ H; that is, if and only if the polynomial Ppyq
admits no singular point in C (see [46, p.48 (v)] for a proof of the direct implication
and [20] for the converse).

The proof of Theorem 1.1 (and also that of Theorem 1.3 in Chapter 3) requires
the concept of a local Sato–Bernstein polynomial BK P,y0 pY, sq at a point y0 lying in the
topological closure of a domain Y which, when K “ C, will always be chosen as
Y “ Cn . Then, for the sake of simplicity of notation, set BC n C
P,y0 pC , sq “ BP,y0 psq. In
the case where K “ R, the domain Y Ă Rn shall be assumed to be contained in the
topological closure of its interior and also to be globally semianalytic in the sense that
it is defined as a finite union of sets of the form
tx P Rn : fi pxq ˝i 0 for all 1 ď i ď lu . (2.4)
Here, for each 1 ď i ď l (with l ě 1 an integer), the symbol ˝i P tě, ąu stands for an
inequality and fi : Rn Ñ Rn is an analytic map.

The definition of the local polynomial BK P,y0 pY, sq relies on Lemma 2.3 below. Before
stating it, given an open set U Ă K intersecting non–trivially the interior of Y, denote
n

by BKP pU X Y, sq an auxiliary Sato–Bernstein polynomial. It is defined from the restric-

tion of Ppyq to the set U X Y as the monic polynomial of least degree satisfying the
identity (2.3) for all y P U X Y when the differential operator D py, s, Bq is allowed to
be a polynomial in s and B with coefficients which are power series in y P Kn conver-
gent in U X Y. Since the global Bernstein–Sato polynomial BP satisfies such a relation,
BKP pU X Y, sq divides BP psq. In particular, the conclusions of Kashiwara’s Theorem 2.2
remain true for the polynomial BK P pU X Y, sq.

Lemma 2.3. Keep the above notation and let y0 be a point in the topological closure of the set Y.
Then, there exists an open neighbourhood Uy0 of y0 such that for any other open neighbourhood
Vy0 , the polynomial BK K
P pUy0 X Y, sq divides the polynomial BP pVy0 X Y, sq.

Proof. Consider a neighbourhood U1 of y0 and assume that it does not satisfies the
conclusion of the lemma. Then, there exists another neighbourhood U2 of y0 such
that BK K
P pU1 X Y, sq does not divide BP pU2 X Y, sq. Since the Sato–Bernstein identities
on U1 X Y and U2 X Y can be restricted to the intersection U1 X U2 X Y, the polynomial
BK K
P pU1 X U2 X Y, sq is a proper factor of the polynomial BP pU1 X Y, sq.

If the neigbourhood U1 X U2 does not satisfy the property stated in the lemma,
the process can be iterated : there exists a neighbourhood U3 of y0 such that the
polynomial BK K
P pU1 X U2 X U3 X Y, sq is a proper factor of BP pU1 X Y, sq.

The number of possible iterations in this process is finite since the number of proper
factors of the polynomial BKP pU1 X Y, sq is finite. If the process stabilises after r ě 1
 a consequence of a tauberian theorem 28

steps, then the neighbourhood of y0 defined as Uy0 “ Xri“1 Ui meets the required
property.

Given y0 lying in the closure of Y, let Uy0 be a neighbourhood of y0 satisfying the

conclusion of Lemma 2.3. Recall that when K “ R, the set Y Ă Rn is assumed to be
globally semianalytic and to be contained in the topological closure of its interior. The
local Sato–Bernstein polynomial BKP,y0 pY, sq with respect to the domain Y Ă K at the point
n

y0 is then
BK K
P,y0 pY, sq “ BP pUy0 X Y, sq P Qrss.

It is thus also a split polynomial with negative rational roots dividing the global Sato–
Bernstein polynomial BP psq. The (global) Sato–Bernstein polynomial BKP pY, sq with respect
to the domain Y Ă K isn

BK K
P pY, sq “ lcm BP,y0 pY, sq P Qrss.
y0 PY

In view of Lemma 2.3, it is the lowest common multiple of finitely many local Sato–
Bernstein polynomials when the closure of Y is compact.

In the case when K “ C, it is known, see [41, Lemma 2.5.2], that the local and global
Sato–Bernstein polynomials are related by the formula

BP psq “ lcm BC
P,z0 psq. (2.5)
z0 PCn

This relation does not necessarily hold anymore when K “ R and Y “ Rn .

Given z0 P Cn and x0 P Rn , let from now on and for the sake of simplicity of
notation
Bp P,z psq “ BC psq and BP,x0 pY, sq “ BR (2.6)
0 P,z0 P,x0 pY, sq

be the local Sato–Bernstein polynomials at z0 P Cn and x0 P Rn , respectively. Let also

BP pY, sq “ BR
P pY, sq (2.7)

be the global Sato–Bernstein polynomial with respect to the domain Y.

2.2 a consequence of a tauberian theorem

Take K “ R and let K Ă Rn be a set satisfying the conditions pH1 q and pH2 q. Assume
in this section that the polynomial Ppxq P Rrxs is homogeneous of degree q ě 1. Theo-
rem 1.1 is established in the case of the polynomial Ppxq upon estimating the volume
of the set
SP pK, bpT qq “ tx P T ¨ K : |P pxq| ď bpT qu (2.8)
 a consequence of a tauberian theorem 29

under the successive assumptions of Cases (1), (2) and (3). In all cases, the limit con-
dition (1.4) is assumed to be verified with the exponent q instead of d; namely, it is
assumed throughout that
bpT q
cpT q :“ −Ñ 0. (2.9)
T q T Ñ8
The zeta distribution associated to the polynomial Ppxq is here defined by setting
ż
xζP psq , ψy “ |Ppxq|´s ¨ ψpxq ¨ dx (2.10)
Rn

for any complex number s with non–positive real part and any ψ in C8 n
c pR q.

Denote by O` the collection of all those non–empty open sets obtained by taking
the interiors of the connected components of the region

tx P Rn : Ppxq ą 0u . (2.11)

It is then known that O` has a finite cardinality (see [28, Theorem 2.23]) and that each
element of O` is a semialgebraic domain (i.e. a domain which is a finite union of sets
such as (2.4) when the maps fi are polynomials for all 1 ď i ď l) — see [9, Proposition
3.1] for a proof of this claim. These properties also hold for the collection of sets O´
defined as above when replacing Ppxq with ´Ppxq in (2.11), and therefore also for their
union
O “ O` Y O´ . (2.12)
From the homogeneity of the polynomial Ppxq, the volume of the domain (2.8) can
then be expanded as
ż
Voln pSP pK, bpT qqq “ χt|Ppxq|ďbpT qu ¨ χtxPT ¨Ku ¨ dx
Rn ż

“ T n χt|Ppxq|ďcpT qu ¨ dx (2.13)
(2.9) K
ÿ ż
n
“ T χΩ pxq ¨ χt|Ppxq|ďcpT qu ¨ dx. (2.14)
ΩPO K

The explicit form of the volume in (2.13) will be used to determine its asymptotic
order of growth in Case (2). The decomposition (2.14) will be required to establish the
precise asymptotic behavior of the volume in Case (3) at a level of generality needed to
also prove related statements in Chapter 4. This level of generality essentially requires
the determination of the asymptotic behavior of each of the integrals appearing in the
sum (2.14) when the parameter T tends to infinity. This is achieved by introducing the
zeta distributions induced by the corresponding decomposition.

To do so and to fix the notations, given Ω P O, assume that

@x P Ω, PΩ pxq ą 0, (2.15)
 a consequence of a tauberian theorem 30

where PΩ pxq is either the polynomial Ppxq or its opposite. Extend the map x ÞÑ PΩ pxq
to Rn by setting PΩ pxq “ 0 for all x R Ω. The restricted zeta distribution ζP pΩ, ¨ q is
then defined by setting for all ψ in C8c pR q and all s P C with Repsq ď 0
n

ż
xζP pΩ, sq , ψy “ PΩ pxq´s ¨ ψpxq ¨ dx. (2.16)
Rn

Clearly, it is analytic in the domain of complex numbers with non-positive real parts.
Furthermore, it holds that ÿ
ζP “ ζP pΩ, ¨ q. (2.17)
ΩPO
Given Ω P O, whenever well-defined, let

prP , mP q and prP pΩq, mP pΩqq (2.18)

be the pairs made from the smallest real poles rP and rP pΩq of the distributions ζP
and ζP pΩ, ¨ q, respectively, and from the corresponding respective orders mP ě 1
and mP pΩq ě 1. If either of these quantities does not exist, set conventionally
prP , mP q “ p8, ´q and prP pΩq, mP pΩqq “ p8, ´q (in particular, the integers determining
the orders are then left undefined).

Recall here that a meromorphic distribution, say ζ, has a pole at r P C of order m ě

m
1 if for all test functions ψ in C8 n
c pR q, the map s ÞÑ ps ´ rq ¨ xζpsq, ψy is holomorphic
and if m is the smallest integer with this property.

Lemma 2.4. Let Ω P O. The distribution ζP pΩ, ¨ q defined by (2.16) extends meromorphically
to the entire complex plane. Furthermore, its poles, whenever they exist, lie in a finite number
of arithmetic progressions contained in the interval p0, 8q. The same therefore holds for the
distribution ζP .

Let BP psq P Qrss be the (global) Sato–Bernstein polynomial of the homogeneous

form Ppxq. Before justifying the claims made in Lemma 2.4, denote by
ÿ
D px, s, Bq “ aα px, sqB α (2.19)
αPNn
0

a differential operator associated to BP psq, where the sum has finite support and where
aα px, sq is a polynomial. Define also the dual operator
ÿ
D˚ px, s, Bq “ p´1q|α| B α paα px, sq ¨ q . (2.20)
αPNn
0

Given ψ in C8
c pR q and s P C such that Repsq ě ´1, it satisfies the integration by parts
n

formula
ż ż
D px, s, Bq PΩ pxq s`1
¨ ψpxq ¨ dx “ PΩ pxqs`1 ¨ D˚ px, s, Bq ψpxq ¨ dx.
Rn Rn
 a consequence of a tauberian theorem 31

If Repsq ď 0, the defining identity (2.3) satisfied by the polynomial BP psq then implies
that

BP p´sq ¨ xζP pΩ, sq , ψy “ xζP pΩ, s ´ 1q , D˚ px, ´s, Bq ¨ ψpxqy . (2.21)

Proof of Lemma 2.4. The argument is well–known and is reproduced here for the sake
of completeness as it is short and as it will be referred to often in what follows. Let
s be a complex number such that Repsq ď 0 and let ψ be in C8 n
c pR q : since in the
equation (2.21), the right–hand side is well–defined whenever Repsq ď 1, dividing this
relation by BP p´sq and proceeding recursively shows the existence of the meromorphic
continuation over the entire complex plane. This relation also implies that the real
poles of the distribution ζP pΩ, ¨ q, whenever they exist, are roots of the polynomial
BP p´sq modulo one. This completes the proof of Lemma 2.4 since the last claim is a
direct consequence of the decomposition (2.17).

Alternative proofs of this (well–known) result based on resolutions of singulari-

ties can also be found in the classical works by Atiyah [5] and by Bernstein and
Gel’fand [15].
The following lemma relates the properties of the zeta distributions under con-
sideration to the volume of the domains appearing in the expansion (2.13) and in
the decomposition (2.14). The first two claims in the statement are particular cases
adapted to the present situation of a Tauberian Theorem due to Chambert–Loir and
Tschinkel, see [21, Appendix A, Theorem A.1].

In order to state the lemma, given ψ in C8 n

c pR q , denote by

prP pψq, mP pψqq and prP pΩ, ψq, mP pΩ, ψqq (2.22)

the pairs made from the smallest real poles rP pψq ě 0 and rP pΩ, ψq ě 0 of the meromor-
phic functions s P C ÞÑ xζP psq , ψy and s P C ÞÑ xζP pΩ, sq , ψy, respectively, and from the
corresponding orders mP pψq ě 1 and mP pΩ, ψq ě 1, respectively. When either of these
quantities is not well-defined, set here again conventionally prP pψq, mP pψqq “ p8, ´q
and prP pΩ, ψq, mP pΩ, ψqq “ p8, ´q. Also, when c is a real lying in the interval p0, 1q, ψ
a nonnegative map in C8 n
c pR q and Ω a set in O, let
ż ż
µP pψ, cq “ χt|Ppxq|ďcu ¨ ψpxq ¨ dx and µP pΩ, ψ, cq “ χtPΩ pxqďcu ¨ ψpxq ¨ dx.
Rn Rn
(2.23)
Lemma 2.5. Let c P p0, 1q be a real and let ψ be a non–negative map in C8 n
c pR q.

Assume first that rP pψq P p0, 8q and thus that mP pψq ě 1. Then, there exists a univariate
monic polynomial Rψ pxq P Rrxs of degree mP pψq ´ 1 and some ε ą 0 such that, as c tends to
zero, ´ ¯
µP pψ, cq “ ΘP pψq ¨ crP pψq ¨ Rψ p|log c|q ` O crP pψq`ε . (2.24)
 a consequence of a tauberian theorem 32

Here,
1
ΘP pψq “ lim pσ ´ rP pψqqmP pψq xζP psq , ψy ą 0.
mP pψq! σÑrP pψq

Fix from now on Ω P O. Assume that rP pΩ, ψq P p0, 8q and thus that mP pΩ, ψq ě 1.
Then, there exists a univariate monic polynomial RΩ ψ pxq P Rrxs of degree mP pΩ, ψq ´ 1 and
some ε ą 0 such that, as c tends to zero,
´ ¯
rP pΩ,ψq Ω rP pΩ,ψq`ε
µP pΩ, ψ, cq “ ΘP pΩ, ψq ¨ c ¨ Rψ p|log c|q ` O c . (2.25)

Here,
1
ΘP pΩ, ψq “ lim pσ ´ rP pΩ, ψqqmP pΩ,ψq xζP pΩ, sq , ψy ą 0.
mP pΩ, ψq! σÑrP pΩ,ψq

Assume finally that F is a family of nonnegative maps in C8 n

c pR q satisfying the following
two properties :

(A1) [uniform boundedness of the support] there exists a compact set K Ă Rn such that
Supp ψ Ă K for all ψ P F;

(A2) [uniform boundedness of the range] there exists φF in C8 c pR q such that for all ψ P F,
n

it holds that 0 ď ψ ď φF and, with the above notation, that prP pΩ, ψq , mP pΩ, ψqq “
prP pΩ, φF q , mP pΩ, φF qq (in particular, the pair prP pΩ, ψq , mP pΩ, ψqq remains con-
stant over the choice of ψ P F).

Then, the following uniformity claims on the parameters appearing in the expansion (2.25) hold
when the map ψ varies in F :

(V1) the leading coefficient ΘP pΩ, ψq is uniformly bounded above by a constant depending on
the choice of the smooth map φF ;

(V2) there exist integers M, N ě 1 depending on the polynomial Ppxq and on the compact
set K and, given a value of ε ą 0 and a choice of φF as above, there exists a constant
Γ pF, φF , εq ą 0 such
! that the following holds : )the implicit constant in the error term can
be taken as 2 max Γ pF, φF , εq, pCN pF, KqqM provided that the quantity
ˇ ˇ
ˇ B |k| ψ ˇ
CN pF, Kq :“ sup max max ˇ k pxqˇ (2.26)
ˇ ˇ
ψPF 0ď|k|ďN xPK ˇ Bx ˇ

is finite. In particular, as soon as maxxPK ψpxq “ 1 for some ψ P F, this implicit

constant can be taken as pCN pF, KqqM up to a multiplicative constant depending on
Ppxq, on F, on K, on a given a value of ε ą 0 and on the choice of φF ;
 a consequence of a tauberian theorem 33

(V3) there exist integers M, N ě 1 depending on the polynomial Ppxq and on the compact set
K such that the coefficients of the univariate polynomial ΘP pΩ, ψq ¨ RΩ
ψ pxq are, up to a
multiplicative constant depending on K, upper bounded by the above defined quantity
pCN pF, KqqM , provided it is finite.

With obvious modifications (consisting in dropping all notational dependencies on the set
Ω), the uniformity claims (V1)–(V3) also hold under the assumptions (A1) and (A2) for the
expansion (2.24).

ψ pxq P Rrxs can be made explicit

All the coefficients of the polynomials Rψ pxq, RΩ
— see, for instance, [76, Chap. II.5]. Also, the main interest of the above statement
is that the growths of the functions c ÞÑ µP pψ, cq and c ÞÑ µP pΩ, ψ, cq when c tends
to zero are expressed as functions of the smallest poles of the zeta functions xζP , ψy
and xζP pΩ, ¨ q , ψy, respectively, and of their orders. Such growth rates can indeed
otherwise be obtained from resolutions of singularities not relying on the properties of
the zeta functions under consideration (albeit the error terms are then not so explicit)
— see, e.g., [10, Proposition 7.2] for details.

Proof of Lemma 2.5. Let Ω P O. It is enough to establish the relation (2.25) as (2.24)
follows upon reproducing verbatim the same proof after removing the references to the
set Ω in the successive relations.

For the sake of the simplicity of the notation, set in this proof pr, mq “
prP pΩ, ψq, mP pΩ, ψqq and recall that it is assumed that r P p0, 8q.

Then, with evident notational modifications, the volume estimate (2.25) is Theorem
A.1 in [21, Appendix A] applied to the case where the map f therein is taken as x ÞÑ
ş where the mesure µ therein is defined for any Borel set A Ă R by setting
PΩ pxq and n

µpAq “ A ψpxq ¨ dx. According to this statement, the relation (2.25) holds provided that
two assumptions are met :

(a) the meromorphic function s P C Ñ Þ xζP pΩ, sq , ψy has no pole in the half–plane
Repsq ď r ´ δ for some value of δ P p0, 1q (which determines the value of ε in the
statement);

(b) this function has moderate growth rate on the vertical strip r ´ δ ` iR in the sense
that there exists κ ą 0 such that for any τ P R,

|xζP pΩ, r ´ δ ` iτq , ψy| ! p1 ` |τ|qκ . (2.27)

Assumption (a) follows immediately from Lemma 2.4, which gives the location of the
poles of the meromorphic function xζP pΩ, ¨ q , ψy as a subset of a finite number of
arithmetic progressions. As for (b), fix δ ą 0 satisfying the first assumption and take it
smaller if needed to ensure that the Sato–Bernstein polynomial BP psq does not vanish
at ´r ` δ (and therefore on the vertical strip ´r ` δ ` iR, since its roots are rational).
 a consequence of a tauberian theorem 34

Then, denoting by D˚ px, s, Bq the dual operator associated to BP psq and expanding it
as in (2.20), one obtains from the equation (2.21) that

|BP p´r ` δ ´ iτq| ¨ |xζP pΩ, r ´ δ ` iτq , ψy|

“ |xζP pΩ, r ´ δ ´ 1 ` iτq , D˚ px, ´r ` δ ´ iτ, Bq ¨ ψpxqy|
ÿ ˇˇA Eˇ
ˇ ζR r δ 1 iτq , α
δ iτq ψpxqq
ˇ
ď P pΩ, ´ ´ ` B pa α px, ´r ` ´ ¨ ˇ
αPNn
0
ÿ
ď xζP pΩ, r ´ δ ´ 1q , |B α paα px, ´r ` δ ´ iτq ¨ ψpxqq|y ,
αPNn
0

where the sum has a finite support. The moderate growth assumption is then easily
seen to be implied by the properties that infτPR |BP p´r ` δ ´ iτq| ą 0, that aα px, sq is a
polynomial in s and that r ´ δ ´ 1 is not a pole of the distribution ζP pΩ, ¨ q.

As for the uniformity claims, as above, it is here again enough to establish them
in the case of the expansion (2.25). Note then first that the assumption (A2) implies
that µP pΩ, ψ, cq ď µP pΩ, φF , cq. Letting c tend to zero, one infers from the same
assumption that
ΘP pΩ, ψq ď ΘP pΩ, φF q ,
whence (V1).

It is easier to establish the claim (V3) before (V2). To this end, the proof of [21,
Appendix A, Theorem A.1] shows that the coefficients of the polynomial ΘP pΩ, ψq ¨
RΩ
ψ pxq depend polynomially on the constants r and κ and also on the coefficients of the
(well–defined) univariate polynomial R rΩ pxq of degree m ´ 1 determined by the residue
ψ
relation
ˆ ˙
´r rΩ plog Bq “ Ress“r B´s ¨ xζP pΩ, sq , ψy
B ¨R ψ
sk`1
« ff
1 dm´1 ´ ¯
“ ¨ lim ps ´ rqm ¨ s´k´1 ¨ B´s ¨ xζP pΩ, sq , ψy .
pm ´ 1q! sÑr dsm´1
(2.28)

rΩ pxq,
Here, k is again an integer larger than κ ą 0. To isolate the coefficients of R ψ
expand xζP pΩ, sq , ψy as a Laurent series around s “ r :
m
ÿ ar,l pΩ, ψq
xζP pΩ, sq , ψy “ ` xhP,r pΩ, sq , ψy , (2.29)
l“1 ps ´ rql

where hP,r pΩ, ¨ q is a distribution holomorphic in a neighbourhood of s “ r. Cauchy’s

Integral Formula implies that the maps ψ P C8 n
c pR q ÞÑ ar,l pΩ, ψq (1 ď l ď m) deter-
mining the coefficients of the Laurent expansion are well–defined distributions (see [36,
 a consequence of a tauberian theorem 35

Appendix A.2.3] for details). This is saying that these linear functionals are continu-
ous, thereby guaranteeing the existence of an integer N ě 1 and of an implicit constant
both depending only on the compact set K such that for all ψ in C8 n
c pR q with support
contained in K, ˇ ˇ
ˇ B |k| ψ ˇ
|ar,l pΩ, ψq| ! max max ˇ k pxqˇ . (2.30)
ˇ ˇ
0ď|k|ďN xPK ˇ Bx ˇ
Inserting the expansion (2.29) into the equation (2.28), the coefficients of the
polynomial R rΩ pxq are seen to be linear combinations of elements in the set
ψ
t1, ar,1 pΩ, ψq , . . . , ar,m pΩ, ψqu . The inequality (2.30) then establishes (V3).

As for the claim (V2), it follows from the proof of [21, Appendix A, Theorem A.1]
that the implicit constant in the error term in (2.25) can be taken as the sum of two
quantities :

‚ a multiple, depending on the exponent r “ rP pΩ, ψq, on the order m “ mP pΩ, ψq

and on a small enough parameter δ ą 0 (determining the value of ϵ ą 0 in (2.25))
of the factor ˇż ż ˇ
ˇ ds ψpxq ˇ
ˇ
k`1
¨ s
¨ dxˇ.
r´δ`iR s Rn PΩ pxq
ˇ ˇ
Here, k is any integer larger than the exponent κ ą 0 appearing in (2.27). This
factor is clearly bounded above by a multiple (depending on r, δ and κ) of the
nonnegative quantity xζP pΩ, r ´ δq , ψy which, in turn, is bounded above by the
constant xζP pΩ, r ´ δq , φF y uniformly in ψ P F. This constant is well-defined
from the assumption (A2);

‚ the coefficients of the polynomials ΘP pΩ, ψq ¨ RΩ

ψ pxq (this dependency is captured
in the error term denoted by O a´ε
r pB q in the proof of [21, Appendix A, Lemma
A.5]). As established in (V3) and with the notation therein, these coefficients are,
up to a multiplicative constant depending on K, bounded above by the quantity
pCN pF, KqqM , whenever it is finite.

These two observations prove the first part of the claim (V2). As far as the second one
is concerned, it is enough to note that, under the assumption that maxxPK ψpxq “ 1 for
some ψ P F, it is immediate that CN pF, Kq ě 1. As a consequence,
! )
2 max Γ pF, φF , εq, pCN pF, KqqM ď 2 max t1, Γ pF, φF , εqu ¨ pCN pF, KqqM .

This completes the proof of Lemma 2.5.

The goal is now to deduce from Lemma 2.5 sharp volume estimates for the set (2.8)
defined from the polynomial Ppxq in the three cases considered in Theorem 1.1. These
estimates are established under the assumption that the largest poles of the various
meromorphic distributions under consideration are well-defined. The proofs of the
 a consequence of a tauberian theorem 36

existence and of the determination of these poles are the subject of the next Section 2.3
(see in particular Proposition 2.16 therein).

Consider first Case (1). If ψ is a map in C8 n

c pR q satisfying the inequalities

0 ď χK ď ψ
which is such that the pair prP pψq, mP pψqq introduced in (2.22) is well-defined with
rP pψq ą 0, then an immediate consequence of the expansion (2.13) and of the rela-
tion (2.24) in the above Lemma 2.5 is that
Voln pSP pK, bpT qqq ! T n ¨ cpT qrP pψq ¨ |log pcpT qq|mP pψq´1 .
This is enough to establish the volume estimate (1.11) in Theorem 1.1 upon special-
ising this inequality to the case where Ppxq is the homogeneous polynomial }Fpxq}2
of degree q “ 2d and upon noticing that the poles and orders of the distribution
ζF defined in (1.9) then coincide with those of the distribution ζP p ¨ {2q defined by (2.10).

Consider now Case (2), thus assuming that the condition pH3 q additionally holds.
Denote by prP pKq, mP pKqq the pair provided by pH3 q in the case of the polynomial Ppxq
(recall that the assumption pH3 q relates the analytic properties of the distribution ζP to
the topological properties of the set K). The following result, which turns out to be an
easy consequence of Lemma 2.5, then provides the sharp order of asymptotic growth
of the volume of the set SP pK, bpT qq. Upon specialising it as above to the case of the
polynomial Ppxq “ }Fpxq}2 , it establishes the relation (1.12) in Theorem 1.1.
Theorem 2.6 (Volume Estimate in Case (2)). Let the set K Ă Rn satisfy the three assump-
tions pH1 q ´ pH3 q. Assume that the pair prP pKq, mP pKqq is such that rP pKq ą 0 and that the
limit condition (2.9), where the quantity cpT q is defined, holds. Then for T ě 1,

Voln pSP pK, bpT qqq — T n ¨ cpT qrP pKq ¨ |log cpT q|mP pKq´1 .
For the sake of simplicity of the terminology, say from now on that a smooth test
function ψ whose support satisfies the inclusions (1.10) while being strictly positive
over the set C meets the support restriction condition (1.10).
Proof of Theorem 2.6. Let ψ1 , ψ2 be two maps in C8 n
c pR q meeting the support restriction
condition (1.10) such that
0 ď ψ1 ď χK ď ψ2 .
Under the assumption pH3 q, it holds that
prP pKq, mP pKqq “ prP pψ1 q, mP pψ1 qq “ prP pψ2 q, mP pψ2 qq , (2.31)
where, given ψ in C8 n
c pR q, the pair prP pψq, mP pψqq is defined in (2.22). Furthermore,
the expansion (2.13) implies that
Voln pSP pα, bpT qqq
µP pψ1 , cpT qq ď ď µP pψ2 , cpT qq,
Tn
 a consequence of a tauberian theorem 37

where the quantities µP pψ1 , cpT qq and µP pψ2 , cpT qq are explicitly given in (2.23). The
conclusion of the statement is then a consequence of the relation (2.31) and of Equa-
tion (2.24) in Lemma 2.5.

Consider finally Case (3); that is, assume from now on that K is a body star-shaped
with respect to the origin containing the origin in its interior. With a view towards the
metric counting results proved in Chapter 4, the volume estimate (1.13) is established
at a level of generality bigger than what is needed in Case (3) of Theorem 1.1. To this
end, fix a non–empty subcollection of sets O 1 contained in the collection O defined
in (2.12). While it is obvious that the condition pH2 q is verified in Case (3) since the
set K then contains the origin in its interior (recall that the polynomial Ppxq is assumed
to be homogeneous), Lemma 2.7 below makes it clear that pH3 q also holds under this
same assumption on K. With this in mind, define, with a slight abuse of notations,
` ˘ ÿ
ζP O 1 , ¨ “ ζP pΩ, ¨ q, (2.32)
ΩPO 1

where the restricted distributions ζP pΩ, ¨ q are defined in (2.16).

It should be clear that Lemma 2.4 also applies to the distribution ζP pO 1 , ¨ q which
therefore extends to a meromorphic distribution defined over the complex plane. Let
then rP pO 1 q ě 0 be its smallest real pole (assuming it exists) and let mP pO 1 q ě 1
denote the corresponding order.

If ψ is a nonnegative map in C8 n
c pR q which does not vanish at the origin, it is not
hard to see that, given Ω P O 1 , the pair prP pΩ, ψq, mP pΩ, ψqq introduced in (2.22) is
well–defined under the assumption of the homogeneity of the polynomial Ppxq, and
that the same holds for the pair prP pψq, mP pψqq also introduced in (2.22) (this is also
established in much more generality in Proposition 2.16 in Section 2.3 below). An
immediate consequence of the following lemma is that the assumption pH3 q is ne-
cessarily met when the set K contains the origin in its interior (and thus under the
assumptions of Case (3)).

Lemma 2.7. Let ψ ě 0 be in C8 n

c pR q such that ψp0q ą 0. Then, under the assumption of the
homogeneity of the polynomial Ppxq, it holds on the one hand that

rP pΩq “ rP pΩ, ψq and mP pΩq “ mP pΩ, ψq, (2.33)

and on the other that

rP “ rP pψq and mP “ mP pψq. (2.34)

In these relations, the restricted pair prP pΩq, mP pΩqq and the global pair prP , mP q are both
defined in (2.18).
 a consequence of a tauberian theorem 38

To justify the claim that pH3 q is satisfied when the set K contains the origin in its
interior, it is enough to define, in the notation of the support restriction condition (1.10),
the set C to be any compact neighbourhood of the origin contained in K and the set
U to be the full space Rn . In view of the above lemma, it then suffices to set the pair
prP pKq, mP pKqq as prP , mP q independently of the choice of such a set K.

Proof. It suffices to establish the relations (2.33) as those in (2.34) follow from this first
case upon reproducing verbatim the proof after removing the notational dependence
on Ω of the various quantities under consideration.

By definition, one has that rP pΩ, ψq ě rP pΩq and, in the case where equality holds,
that mP pΩ, ψq ď mP pΩq.

As for the reverse inequalities, let φ be in C8 n

c pR q such that the meromorphic map

s P C ÞÑ xζP pΩ, sq , φy (2.35)

achieves its smallest real pole at rP pΩq with an order mP pΩq. Clearly, when σ ă rpΩq
and when ξ is a map in C8 n
c pR q which has the same support as φ and which bounds
from above |φ|, it holds that

|xζP pΩ, σq , φy| ď xζP pΩ, σq , ξy . (2.36)

Letting σ tend to rP pΩq´ , the meromorphic distribution on the right–hand side

appears to have a smallest pole at most equal to rP pΩq, and therefore equal to this
quantity from its definition. Furthermore, one then also obtains that its order at this
pole is at least mP pΩq, and therefore also equal to this integer from its definition. This
shows that one may assume without loss of generality that φ ě 0 and, even if it means
renormalising it, that 0 ď φ ă 1.

Fix again σ ă rP pΩq and let λ ą 0 be a parameter large enough so that

λ´1 ¨ Supppφq :“ tx{λ P Rn : x P Supppφqu is contained in a small enough neighbour-
hood of the origin where ψ remains strictly positive. Upon setting φλ : y P Rn ÞÑ
φpλyq, the change of variables y “ x{λ in the integral defining the map (2.35) yields

xζP pΩ, σq , φy “ λn´qσ ¨ xζP pΩ, σq , φλ y , (2.37)

where it is recalled that q ě 1 is the degree of the polynomial Ppxq. This shows that
the meromorphic distribution s ÞÑ xζP pΩ, σq , φλ y is such that

prP pΩ, φλ q, mP pΩ, φλ qq “ prP pΩ, φq, mP pΩ, φqq “ prP pΩq, mP pΩqq . (2.38)

Also,

xζP pΩ, σq , ψy “ xζP pΩ, σq , ψ ¨ φλ y ` xζP pΩ, σq , ψ ¨ p1 ´ φλ qy , (2.39)

 a consequence of a tauberian theorem 39

where all the terms remain positive for σ ă rP pΩq from the relation (2.16) defining the
distribution ζP pΩ, ¨ q. Since the test function ψ ¨ φλ is, up to strictly positive multi-
plicative constants, bounded from above and below by φλ on its support (this follows
from the assumption made on Supppφλ q “ λ´1 ¨ Supppφq), the smallest pole and the
corresponding order of the function xζP pΩ, ¨ q , ψ ¨ φλ y are also given by (2.38). Upon
letting σ tend to rP pΩq´ , it is then easily deduced that the decomposition (2.39) implies
the reverse inequalities rP pΩq ě rP pΩ, ψq and mP pΩq ď mP pΩ, ψq. This establishes the
lemma.
The following notational reductions shall be convenient to either adopt or keep in
mind :
‚ when O 1 “ O, the definition of ζP pO, ¨ q in (2.32) coincide with that of ζP in (2.10);
‚ when O 1 “ O˘ , set ζ˘ ˘
P “ ζP pO , ¨ q;

‚ when O 1 “ O and when O 1 “ O˘ , let respectively

` ˘ ˘˘ ` ` ˘˘ ` ˘˘
prP , mP q “ prP pOq , mP pOqq and rP , mP “ rP O , mP O˘ . (2.40)

The precise asymptotic behavior of the quantity Voln pSP pK, bpT qqq in Case (3) is a
corollary of the following general theorem. To state it, note that if g P SLn pRq, then
the map Ω P O 1 ÞÑ g´1 ¨ Ω P g´1 ¨ O 1 induces a bijection between connected sets where
the polynomial Ppxq keeps a constant sign and connected sets where the polynomial
pP ˝ gq pxq satisfies the same property.
Theorem 2.8 (Volume estimate in Case (3) under a unimodular distorsion). Assume
that the set K is star-shaped with respect to the origin, that it meets the condition pH1 q and
that it contains the origin in its interior. Let furthermore the limit condition (2.9), where the
quantity cpT q is defined, hold. Let g P SLn pRq and let˘ O 1 be a nonempty subcollection of sets
` ´1
1
in O such that there exists Ω P O for which g ¨ Ω X K ‰ H.

Then, there exist strictly positive constants δP pO 1 , Kq and γP,g pO 1 , Kq such that, as the
parameter T tends to infinity,
˜ ¸
ď ! )
Voln x P T ¨ K : 0 ă pP ˝ gqpg´1 ¨Ωq pxq ă bpT q “
ΩPO 1
ˆ ˆ´ ´ ¯¯´δP pO 1 ,Kq ˙˙ 1 1
¨ T n ¨ cpT qrP pO q ¨ |log cpT q|mP pO q´1 .
` 1 ˘ 1 ´1
γP,g O , K ` O fP O , cpT q
(2.41)
In this relation, the pair prP pO 1 q , mP pO 1 qq is well–defined and is such that rP pO 1 q ą 0; the
restricted polynomials pP ˝ gqpg´1 ¨Ωq pxq are defined as in (2.15) and, given c P p0, 1q,
$
&c
’ if for all Ω P O 1 , it holds that mP pΩq “ 1 whenever
` 1 ˘
fP O , c “ rP pΩq “ rP pO 1 q ;
’
logpcq otherwise.
%
 a consequence of a tauberian theorem 40

Finally, given any compact set B Ă SLn pRq, the implicit constant as well as the leading
coefficient in the volume estimate (2.41) can be chosen uniformly over all polynomials in the
family tpP ˝ gq pxqugPB .

Recall that from Lemma 2.7, pH3 q is verified under the assumptions of Case (3).
The above statement can then be slightly refined in the following sense : given
g P SLn pRq, one can take fP pO 1 , cq “ c in the asymptotic relation (2.41) when-
ever for any nonnegative map ψ in C8 n
c pR q satisfying the support condition (1.10)
present in pH3 q, the non–leading coefficients of the monic univariate polynomials
ψ pxq “ Rg,ψ pxq P Rrxs appearing in the statement of Lemma 2.5 vanish for all Ω in
RΩ Ω

the collection O 1 (there is one such univariate polynomial associated to each restricted
polynomial pP ˝ gqpg´1 ¨Ωq pxq as Ω varies in O 1 ). In other words, this is requiring
that RΩ
g,ψ pxq “ x
m´1 for all such sets Ω and all such maps ψ ě 0. See the proof of

Lemma 2.13 below for a justification of this claim.

Taking B Ă SLn pRq to be reduced to the identity element and O 1 “ O, Theorem 2.8
admits the following immediate corollary. As previously, upon specialising this corol-
lary to the case where Ppxq is the homogeneous polynomial }Fpxq}2 of degree q “ 2d
and upon noticing that the poles and orders of the distribution ζF defined in (1.9)
then coincide with those of the distribution ζP p ¨ {2q, it implies (in a stronger form) the
asymptotic expansion (1.13) part of Theorem 1.1.

Corollary 2.9 (Volume estimate in Case (3)). Assume that the set K meets the condi-
tion pH1 q, that it is star-shaped with respect to the origin and that it contains the origin
in its interior. Let furthermore the limit condition (2.9), where the quantity cpT q is defined, hold.

Then, there exists constants δP pKq, γP pKq ą 0 such that

ˆ ˆ´ ´ ¯¯´δP pKq ˙˙
Voln pSP pK, bpT qqq “ γP pKq ` O fP cpT q ´1
¨ T n ¨ cpT qrP ¨ |log cpT q|mP ´1

as the parameter T tends to infinity. Here, rP ą 0 and, given c P p0, 1q,

$
&c
’ if for all Ω P O, it holds that mP pΩq “ 1 whenever
fP pcq “ rP pΩq “ rP ;
’
logpcq otherwise.
%

Before beginning the proof of Theorem 2.8, another corollary is derived from this
statement upon specialising it to the case where O 1 is either of the collections of sets
O˘ and where K “ Bn is the closed unit Euclidean ball. This corollary plays a crucial
role in the proofs of the metric counting results established in Chapter 4.

Corollary 2.10. Let B be a compact subset of SLn pRq and let g P B. Assume that ` ˘the ˘
limit
˘
condition (2.9), where the quantity cpT q is defined, holds and assume that the pair rP , mP is
 a consequence of a tauberian theorem 41

well–defined.

Then, there exist constants δ˘ ˘

P , γP,g ą 0 such that, as the parameter T tends to infinity,

Voln ptx P Bn pT q : 0 ă ˘ pP ˝ gq pxq ă bpT quq

ˆ ˆ´ ´ ¯¯´δ˘ ˙˙ ˘ ˘
P
˘
“ γP,g ` O fP cpT q˘ ´1
¨ T n ¨ cpT qrP ¨ |log cpT q|mP ´1 .
(2.42)

Here, given c P p0, 1q,

#
˘ c if for all Ω P O˘ , it holds that mP pΩq “ 1 whenever rP pΩq “ r˘
P;
fP pcq “
logpcq otherwise.

Furthermore, the leading coefficient γ˘

P,g and the implicit constant in the relation (2.42) are
bounded above uniformly over all polynomials in the family tpP ˝ gq pxqugPB .

The first step in the proof of Theorems 2.8 is to give an explicit expres-
1 q , m pO 1 qq and to relate it to the corresponding pair
`sion for
` ´1the 1pair
˘ prP pO
` ´1 P ˘˘
rpP˝gq g ¨ O , mpP˝gq g ¨ O 1 associated to the polynomial pP ˝ gq pxq. This is
achieved in the following lemma.

Lemma 2.11. Let g be in SLn pRq and let O 1 be a nonempty subcollection of O.

` ` ˘ ` ˘˘
Then, the pair rpP˝gq g´1 ¨ O 1 , mpP˝gq g´1 ¨ O 1 is well-defined if, and only if, so is the
pair prP pO 1 q , mP pO 1 qq, in which case
´ ¯ ` ˘
rpP˝gq g´1 ¨ O 1 “ rP O 1 “ min rP pΩq (2.43)
ΩPO 1

and ´ ¯ ` ˘
mpP˝gq g ¨ O “ mP O 1 “
´1 1
max mP pΩq. (2.44)
ΩPO 1
rP pΩq“rP pO 1 q

Proof. Given @a map`ψ in C8 n

c pR ˘q and
D a set
@ Ω in O 1 , `an obvious
˘D change a variables
´1 ´1
shows that ζpP˝gq g ¨ pΩq , ¨ , ψ “ ζP pΩ, ¨ q, ψ g ¨ . As a consequence,
` ` ˘ ` ˘˘
rpP˝gq g´1 ¨ pΩq , mpP˝gq g´1 ¨ pΩq “ prP pΩq, mP pΩqq whenever either of these pairs
exists. It is therefore enough to establish the second equations, in (2.43) and in (2.44)
respectively, to prove the lemma.

To this end, note that the definition (2.32) of the distribution ζP pO 1 , ¨ q immediately
implies the inequalities
` ˘ ` ˘
rP O 1 ě min rP pΩq and mP O 1 ď max mP pΩq. (2.45)
ΩPO 1 ΩPO 1
rP pΩq“rP pO 1 q
 a consequence of a tauberian theorem 42

The reverse inequalities are obtained upon noticing that the smallest poles of the dis-
tributions ζP pΩ, ¨ q and their respective orders cannot cancel out as Ω varies in O 1 .
To see this, fix Ω P O 1 and consider ψΩ in C8 n
c pR q for which the meromorphic map
s P C ÞÑ xζP pΩ, sq , ψΩ y has a pole at rP pΩq of order mP pΩq ě 1. Let φ ě 0 be a map
in C8 n
c pR q bounding from above |ψΩ |. Assume that s “ σ ă rP pΩq is real. From the
Triangle Inequality,
prP pΩq ´ σqmP pΩq ¨ |xζP pΩ, σq , ψΩ y| ď prP pΩq ´ σqmP pΩq ¨ xζP pΩ, σq , φy . (2.46)
Upon taking the limit σ Ñ rP pΩq´ , it follows that rP pΩq ě rP pΩ, φq and that, in the
case where equality holds, mP pΩq ď mP pΩ, φq. Since, by the definition of the pole
rP pΩq and of its order mP pΩq, it holds that rP pΩ, φq ě rP pΩq and that, in the case of
equality, mP pΩ, φq ď mP pΩq, one obtains that prP pΩ, φq, mP pΩ, φqq “ prP pΩq, mP pΩqq.
(The argument is similar to the one used in (2.36).)

Also, the coefficient of prP pΩq ´ sq´mP pΩq in the Laurent expansion of the map s P
C ÞÑ xζP pΩ, sq , φy at rP pΩq is positive whenever this map is well-defined (since it is
then clearly positive when s “ σ ă rP pΩq is real). Choose φ ě 0 bounding from
above the absolute values of all smooth functions ψΩ when Ω varies over all those
sets achieving the minimum in (2.43). Then, the sum over these sets Ω of the Laurent
coefficients of prP pΩq ´ sq´mP pΩq of the meromorphic maps xζP pΩ, ¨ q , φy does not va-
nish, which shows that the converse inequalities in (2.45) also hold. This completes the
proof.

` Lemma
` ´1 2.11˘ reduces calculations of the pairs prP pO 1 q , mP pO 1 qq and
` ´1the 1 ˘˘
1
rpP˝gq g ¨ O , mpP˝gq g ¨ O to that of a finite number or pairs prP pΩq, mP pΩqq
1
when Ω P O . The determination of the latter quantities can be achieved thanks to the
above Lemma 2.7.

Proof of Theorem 2.8. The proof of the claim that the pair prP pO 1 q , mP pO 1 qq is well-
defined and that its first component is strictly positive is an immediate consequence
of the above Lemma 2.11 and of Proposition 2.16 from Section 2.3 below. The goal is
here to establish the remaining claims in the statement of Theorem 2.8.

In view of the result to establish, even if it means discarding elements from the
collection O 1 , it may be assumed without loss of generality that for each Ω P O 1 , it
` ´1 ˘
holds that g ¨ Ω X K ‰ H. It is then enough to prove Theorem 2.8 in the case that
O 1 contains a unique element Ω for which the pair prP pΩq , mP pΩqq is well–defined.
Indeed, assume that for some δP pΩ, Kq ą 0,
´! )¯
Voln x P T ¨ K : 0 ă pP ˝ gqpg´1 ¨Ωq pxq ă bpT q “
ˆ ˆ´ ´ ¯¯´δP pΩ,Kq ˙˙
γP,g pΩ, Kq ` O fP Ω, cpT q´1 ¨ T n ¨ cpT qrP pΩq ¨ |log cpT q|mP pΩq´1
(2.47)
 a consequence of a tauberian theorem 43

with the required uniformity claims, where γP,g pΩ, Kq ą 0 and where for c P p0, 1q,
#
c if mP pΩq “ 1;
fP pΩ, cq “ (2.48)
logpcq otherwise.

From the disjointness of the supports of the restricted polynomials pP ˝ gqpg´1 ¨Ωq pxq,
˜ ¸
ď ! )
Voln x P T ¨ K : 0 ă pP ˝ gqpg´1 ¨Ωq pxq ă bpT q
ΩPO 1
ÿ ´! )¯
“ Voln x P T ¨ K : 0 ă pP ˝ gqpg´1 ¨Ωq pxq ă bpT q .
ΩPO 1

Combining the above two relations, it follows from Lemma 2.11 that Theorem 2.8 is
obtained upon setting
` ˘ ÿ
γP,g O 1 , K “ γP,g pΩ, Kq,
ΩPO 1
prP pΩq,mP pΩqq“prP pO 1 q,mP pO 1 qq
` ˘
δP O 1 , K “ min δP pΩ, Kq
ΩPO 1
rP pΩq“rP pO 1 q

and
fP pΩ, cq “ min fP pΩ, cq .
ΩPO 1
δP pΩ,Kq“δP pO 1 ,Kq

` ˘
In order to establish the relation (2.47), consider two sequences ψ´ k kě1 and
ψk kě1 of elements in Cc pR q satisfying the following two properties :
` `˘ 8 n

` ˘ ` `˘
(i) the sequence ψ´ k kě1 is pointwise increasing and the sequence ψk kě1 point-
wise decreasing;

(ii) for all k ě 1, it holds that χKp1´1{kq ď ψ´ `

k ď χK ď ψk ď χKp1`1{kq , where,
given λ ą 0, Kpλq denotes the dilate of the set K by the factor λ; in other words,
Kpλq “ tλ ¨ x : x P Ku.

In the above, for any given integer k ě 1, the inequalities χKp1´1{kq ď χK ď χKp1`1{kq
hold under the assumption that the set K is star-shaped with respect to the origin.
When g P SLn pRq, k ě 1 and x P Rn , set furthermore
´ ¯
ψ˘g,k pxq “ ψ˘
k g ´1
¨ x , where Supp ψ˘ g,k Ă K p2 ¨ κg q . (2.49)
 a consequence of a tauberian theorem 44

Here and throughout, κg ą 0 denotes the operator norm of the linear map induced
by g (with respect to the Euclidean norm). In the notations introduced in the defining
relations (2.23) and in Lemma 2.5, given c P p0, 1q and k ě 1, set first
´ ¯ ż
˘ ´1 ˘
νP,g pΩ, k, cq “ µpP˝gq g ¨ Ω, ψk , c “ χtpP˝gqpxqďcu ¨ ψ˘ k pxq ¨ dx
´1
g ¨Ω
´ ¯ ż
˘
“ µP Ω, ψg,k , c “ χtPΩ pxqďcu ¨ ψ˘
g,k pxq ¨ dx, (2.50)
Rn

then ż
νP,g pΩ, K, cq “ χtpP˝gqpxqďcu ¨ dx
KXpg´1 ¨Ωq

and finally ´ ¯ ´ ¯
Θ˘
P,g pΩ, kq “ Θ pP˝gq g´1
¨ Ω, ψ˘
k “ Θ P Ω, ψ˘
g,k .

Thus, by an elementary change of variables relying on the homogeneity of the polyno-

mial Ppxq, the volume appearing in (2.47) can be expressed as
´! )¯
Voln x P T ¨ K : 0 ă pP ˝ gqpg´1 ¨Ωq pxq ă bpT q “ T n ¨ νP,g pΩ, K, cpT qq (2.51)

with the quantity cpT q defined in (2.9).

Clearly, given k ě 1, the above assumptions (i) and (ii) imply that

ν´ ´
P,g pΩ, k, cq ď νP,g pΩ, k ` 1, cq ď νP,g pΩ, K, cq
ď ν` `
P,g pΩ, k ` 1, cq ď νP,g pΩ, k, cq . (2.52)

Since from Lemma 2.5 and Lemma 2.11 (applied to the case when O 1 “ tΩu), it
holds that ˜ ¸
ν˘P,g pΩ, k, cq
lim m pΩq´1
“ Θ˘
P,g pΩ, kq , (2.53)
cÑ0 ` r
cP pΩq ¨ |log c| P

´ ¯
one deduces from the inequalities (2.52) that the two sequences Θ˘ P,g pΩ, kq are
kě1
both monotonic and bounded. They therefore converge to respective values Θ˘
P,g pΩ, Kq
which satisfy the following relations for any k ě 1 :

0 ă Θ´ ´ ´
P,g pΩ, 1q ď ΘP,g pΩ, kq ď ΘP,g pΩ, Kq
ď Θ` ` `
P,g pΩ, Kq ď ΘP,g pΩ, kq ď ΘP,g pΩ, 1q . (2.54)

Lemma 2.12. Let g P SLn pRq with operator norm κg ą 0. With the above notations, there
exists ε ą 0 such that when c P p0, 1q,
ˇ
ˇ `
ˇ ν`
P pΩ, 1, cq
ˇνP,g pΩ, k, cq ´ ν´ k, cq λ ε, gq , (2.55)
ˇ
P,g pΩ, ˇ ! P pΩ, ¨
k
 a consequence of a tauberian theorem 45

where
n´q¨rP pΩq
¨ p1 ` |log κg |qmP pΩq´1
␣ (
λP pΩ, ε, gq “ κg ¨ max 1, κ´ϵ
g (2.56)
and where ν` `
P pΩ, 1, cq is shorthand notation for νP,g pΩ, 1, cq when g is the identity. In parti-
cular, one obtains that
ˇ
ˇ `
ˇ λP pΩ, ε, gq
ˇΘP,g pΩ, kq ´ Θ´ kq (2.57)
ˇ
P,g pΩ, ˇ !
k
and also that
Θ´ `
P,g pΩ, Kq “ ΘP,g pΩ, Kq p“ ΘP,g pΩ, Kq , sayq. (2.58)
In the above relations, the implicit constants depend only on the sets Ω and K and on the
polynomial Ppxq.

Proof. From the inequalities in (ii) above and from the assumption of the homogeneity
of the polynomial Ppxq, it follows from an elementary change of variables that
ˇ ˇ ż
ˇ ` ´ * ¨ dx
ˇνP,g pΩ, k, cq ´ νP,g pΩ, k, cqˇ ď χ"
ˇ
pKp1`1{kqzKp1´1{kqq pP˝gq ´1 pxqďc
pg ¨Ωq
ˆ ˙n ż
1 * ¨ dy
“ 1` ¨ χ"
k K pP˝gq ´1 pyqďc¨p1`1{kq´q
pg ¨Ωq
ˆ ˙n ż
1 * ¨ dy.
´ 1´ ¨ χ"
k K pP˝gq ´1 pyqďc¨p1´1{kq´q
pg ¨Ωq

Expanding the nth powers multiplying the integrals, rewriting the whole expression
as a polynomial in 1{k and isolating the constant term in this polynomial, one obtains
that
ˇ ˇ ż
ˇ ` ´ * ¨ dx
ˇνP,g pΩ, k, cq ´ νP,g pΩ, k, cqˇ ! χ"
ˇ
c¨p1`1{kq ´q ă pP˝gq pxq ď c¨p1´1{kq ´q
K pg´1 ¨Ωq
ż
1 * ¨ dx
` ¨ χ"
k K pP˝gq g´1 ¨Ω pxq ď 2q ¨c
p q
ż
! χtc¨p1`1{kq´q ă PΩ pyq ď c¨p1´1{kq´q u ¨ dy
g¨K
ż
1
` ¨ χ q ¨ dy.
k g¨K tPΩ pyq ď 2 ¨cu
 a consequence of a tauberian theorem 46

Since g ¨ K Ă K pκg q, this yields

ˇ ˇ ż
ˇ ` ´
ˇνP,g pΩ, k, cq ´ νP,g pΩ, k, cqˇ ! χtc¨p1`1{kq´q ă PΩ pyq ď c¨p1´1{kq´q u ¨ dy
ˇ
Kpκg q
ż
1
` ¨ χtPΩ pyq ď 2q ¨cu ¨ dy
k Kpκg q
ˆż
! κng ¨ χ! ´q ´q
) ¨ dy
K c¨pκg p1`1{kqq ă PΩ pyq ď c¨pκg p1`1{kqq
ż ˙
1
` ¨ χtPΩ pyq ď c¨pκg {2q´q u ¨ dy .
k K

To bound from above this quantity, note that the inequality χK ď ψ` 1 implies one the
one hand that
ż
χ! ´q dy
)
´q
K c¨pκg p1`1{kqq ă PΩ pyq ď c¨pκg p1´1{kqq
ż
`
ď χ! ´q ¨ ψ1 pyq ¨ dy
)
´q
Rn c¨pκg p1`1{kqq ă PΩ pyq ď c¨pκg p1´1{kqq

and, on the other, that

ż ż
χtPΩ pyq ď c¨pκ{2q´q u ¨ dy ď χtPΩ pyq ď c¨pκg {2q´q u ¨ ψ`
1 pyq ¨ dy.
K Rn

Consequently, it follows from the definition of ν` P pΩ, 1, cq that

˜ ˜ ˆ ˆ ˙˙´q ¸
ˇ
ˇ `
ˇ
n 1
ˇνP,g pΩ, k, cq ´ ν´
P,g pΩ, k, cqˇ ! κg ¨ νP
`
Ω, 1, c ¨ κg 1 ´
ˇ
´
k
˜ ˙˙´q ¸ ¸
ˆ ˆ ` ´q q
1 ν pΩ, 1, c ¨ pκg {2q
ν`
P Ω, 1, c ¨ κg 1 ` ` P .
k k

Since the set K contains the origin in its interior and that χK ď ψ` 1 , one infers from
Lemma 2.7 that prP pΩ, ψq , mP pΩ, ψqq “ prP pΩq , mP pΩqq. One then deduces from the
above inequality and from the asymptotic expansion (2.25) that there exists ε ą 0 such
that
ˇ ˇ
n´q¨rP pΩq
¨ p1 ` |log κg |qmP pΩq´1 ˆ
ˇ ` ␣ (
ˇνP,g pΩ, k, cq ´ ν´ k, cqˇ ! κg ¨ max 1, κ´ε
ˇ
P,g pΩ, g
˜ˆ ˙´qrP pΩq ˆ ˙´qrP pΩq ¸
1 1 1
ν`P pΩ, 1, cq 1´ ´ 1` `
k k k
ν` pΩ, 1, cq
! λP pΩ, ε, gq ¨ P ¨
(2.56) k

This establishes (2.55). Dividing through by crP pΩq ¨ |log c|mP pΩq´1 and calling on the
limit relation (2.53), one obtains the inequality (2.57). Letting k tend to infinity then
proves the identity (2.58).
 a consequence of a tauberian theorem 47

Set from now on

γP,g pΩ, Kq “ ΘP,g pΩ, Kq , (2.59)
where the quantity ΘP,g pΩ, Kq is defined in (2.58). Furthermore, given integers k, L ě 1,
let ˇ ˇ
ˇ B |l| Ψ` ˇ
k
CL pkq “ max maxn ˇ pxqˇ .
ˇ ˇ
0ď|l|ďL xPR ˇ Bx l ˇ
Lemma 2.13. Let g belong to a compact subset B of SLn pRq. Then, there exist a constant
δP pΩ, Kq ą 0 and integers M, N ě 1 such that for all k ě 1
ˇ ˇ
ˇ νP,g pΩ, K, cq ˇ CN pkqM 1
γ Kq ` ,
ˇ ˇ
ˇ m
´ P,g pΩ, ˇ ! δ
ˇ crP pΩq ¨ |log c| P pΩq´1 ˇ pfP pΩ, c´1 qq P pΩ,Kq k

where the map fP pΩ, ¨ q is defined in (2.48). Furthermore, the implicit constant depends only
on the polynomial Ppxq, on the set Ω and on the choice of the compact subset B.
Proof. Let k ě 1 and c P p0, 1q. Then,
ˇ ˇ
ˇ ˇ `
ˇνP pΩ, K, cq ´ νP,g pΩ, k, cqˇ
ˇ ˇ
ˇ νP,g pΩ, K, cq ˇ
´ γP,g pΩ, Kqˇ
ˇ ˇ
m
ď
crP pΩq ¨ |log c|mP pΩq´1
ˇ pΩq´1
ˇ crP pΩq ¨ |log c| P ˇ (2.59)
ˇ ˇ
ˇ ν`
P,g pΩ, k, cq `
ˇ
´ ΘP,g pΩ, kqˇ
ˇ ˇ
`ˇ m pΩq´1
ˇ crP pΩq ¨ |log c| P ˇ
ˇ ˇ
` ˇΘ `P,g pΩ, kq ´ ΘP,g pΩ, Kqˇ .
ˇ ˇ

Each of the terms on the right–hand side of this inequality is estimated separately.

First, note that

ˇ
ˇ `
ˇ ˇ
ˇ `
ˇ 1
ˇΘP,g pΩ, kq ´ ΘP,g pΩ, Kqˇ ˇΘP,g pΩ, kq ´ Θ´ kq , (2.60)
ˇ ˇ
ď P,g pΩ, ˇ !
(2.54) (2.57) k
where the implicit constant can be chosen uniformly over the compact subset B of
SLn pRq.

Then, it follows from the relations (2.52) that

ˇ ˇ ˇ ˇ
ˇνP,g pΩ, K, cq ´ ν`
ˇ ` ´
k, cq ν k, cq ν k, cq
ˇ ˇ ˇ
P,g pΩ, ˇ ď ˇ P,g pΩ, ´ P,g pΩ, ˇ

in such a way that

ˇ ˇ
`
ˇνP,g pΩ, K, cq ´ νP,g pΩ, k, cqˇ
ˇ ˇ
ν`
P pΩ, 1, cq 1
mP pΩq´1
! mP pΩq´1
¨
crP pΩq ¨ |log c| (2.55) crP pΩq ¨ |log c| k
1
! ¨ (2.61)
k
 a consequence of a tauberian theorem 48

This last inequality is a consequence of Lemma 2.5 applied to the smooth map ψ`
1 . All
implicit constants are, again, uniform over B Ă SLn pRq.

Finally, apply the uniform estimates stated in the second part of Lemma 2.5 with
the following assumptions :
! )
‚ the family F is taken as F “ Ψg,k
`
;
gPB

‚ the compact set K Ă Rn is defined as K p2κq, where κ ą 0 is an upper bound for

the operator norms of all g P B.

1 px{p2κqq for all x P R .

‚ the smooth map φF is defined by setting φF pxq “ Ψ` n

It is then a consequence of the assumption (ii) above, of the relations (2.49) and of
Lemma 2.7 that the smooth map φF satisfies the assumption (A2) in Lemma 2.5. Fur-
thermore, from repeated applications of the chain rule, it should be clear that for any
multi–index l P Nn , any g P B and any x P Rn ,
ˇ ˇ ˇ ˇ
ˇ B |l| Ψ` ˇ ˇ B |j| Ψ` ˇ
g,k k
pxqˇ ! max maxn ˇ
ˇ ˇ ˇ ˇ
pxqˇ
ˇ Bxl 0ď|j|ď|l| xPR ˇ Bxj
ˇ
ˇ ˇ

with an implicit constant depending on the compact set B only. Recalling the explicit
definition of the quantity ν` P,g pΩ, k, T q in (2.50), Lemma 2.5 then implies the existence
of integers M, N ě 1 such that for some implicit constant depending on B,
ˇ `
ˇ
ˇ νP,g pΩ, k, cq `
ˇ CN pkqM
Θ kq (2.62)
ˇ ˇ
´ P,g pΩ, ! ¨
ˇ crP pΩq ¨ |log c|mP pΩq´1 δ pΩ,Kq
ˇ ˇ
ˇ pfP pΩ, c´1 qq P

To define precisely the exponent δP pΩ, Kq and the choice of the function fP pΩ, ¨ q
in this relation, let 0 ď j`
g,k pΩq ď mP pΩq ´ 2 denote the exponent associated to the
second largest nonzero coefficient of the real, univariate and monic polynomial RΩ
ψ`
rxs
g,k
introduced in Lemma 2.5. If there exists an integer k0 ě 1 such that this exponent is
well-defined for all k ě k0 , set fP pΩ, cq “ logpcq and
! )
δP pΩ, Kq “ min mP pΩq ´ 1 ´ j` g,k pΩq ě 1.
kěk0

Otherwise (this is the case when mP pΩq “ 1), set fP pΩ, cq “ c and δP pΩ, Kq “ ε, where
ε ą 0 is the quantity introduced in the asymptotic relation (2.25).

Adding up the three inequalities (2.60), (2.61) and (2.62) then concludes the proof
of the lemma.
 local and global real log–canonical thresholds 49

The completion of the `proof˘ of Theorem 2.8 relies on a suitable choice of the se-
quences of smooth maps ψk kě1 in Cc pR q. A classical convolution argument de-
˘ 8 n

tailed in [43, Theorem 1.4.1, p.25] shows that there exist such sequences satisfying the
above assumptions (i) and (ii) with the additional property that for all k ě 1,
ˇ ˇ
ˇ B |l| ψ˘ ˇ
k
maxn ˇ pxqˇ ! k|l| .
ˇ ˇ
xPR ˇ Bx l ˇ

Here, the implicit constant depends only on the multi–index l P Nn and on the di-
mension n ě 1. With the notation of Lemma 2.13, this yields the existence of integers
M, N ě 1 such that for all k ě 1,
ˇ ˇ
ˇ ν P,g pΩ, K, cq ˇ kNM 1
γ Kq ,
ˇ ˇ
´ P,g pΩ, ! `
ˇ crP pΩq ¨ |log c|mP pΩq´1 δ pΩ,Kq
ˇ ˇ
ˇ pfP pΩ, K, c´1 qq P k

where δP pΩ, Kq is strictly positive and where the multiplicative constant is uniform
over all compact sets containing the matrix g P SLn pRq. Let c “ cpT q be as in the
` ` ˘˘δ pΩ,Kq{p1`MNq
relation (2.9). Choose then k “ kpT q as the integer part of fP Ω, cpT q´1 P
in such a way that
ˆ´ ´ ¯¯´δP pΩ,Kq{p1`MNq ˙
νP,g pΩ, K, cpT qq ´1
“ γP,g pΩ, Kq ` O fP Ω, cpT q
cpT qrP pΩq ¨ |log cpT q|mP pΩq´1

with the error term satisfying the uniformity property required in the statement of
Theorem 2.8. From the equation (2.51), this completes the proof of Theorem 2.8 upon
redefining the value of the exponent δP pΩ, Kq as δP pΩ, Kq { p1 ` MNq.

2.3 local and global real log–canonical thresh-

olds
Throughout this section, the set Y Ă Rn refers either to a globally semianalytic set K
meeting the assumptions pH1 q ´ pH3 q (with respect to the polynomial Ppxq P Rrxs as
far as pH3 q is concerned) or else to a set Ω Ă Rn belonging to the collection O defined
in (2.12). Then, Y is a finite union of sets such as (2.4), where the maps fi (1 ď i ď l)
are all analytic (and even polynomials when Y “ Ω). Set also
#
ζP pΩ, ¨ q when Y “ Ω;
ζP pY, ¨ q “
ζP when Y “ K,

where the meromorphic distribution ζP pΩ, ¨ q is defined at all complex numbers with
non–positive real parts by (2.16). When Y “ K, the dependency on K in the above
notation is expressed by the fact that the distribution ζP will be tested against maps ψ
 local and global real log–canonical thresholds 50

in C8 n
c pR q meeting the support restriction condition (1.10) part of the assumption pH3 q.
Let then #
prP pKq, mP pKqq when Y “ K;
prP pYq, mP pYqq “
prP pΩq, mP pΩqq when Y “ Ω.
Here, the quantities rP pKq and mP pKq are defined as part of pH3 q, and rP pΩq, whenever
well-defined, denotes the smallest real pole of the meromorphic distribution ζP pΩ, ¨ q,
the integer mP pΩq being its order. Let finally BP pY, sq denote the Sato–Bernstein
polynomial defined in (2.6).

The goal of this section is to establish two claims : on the one hand, that the
quantity rP pYq is a strictly positive rational number that can be determined by
resolutions of singularities in a finite number of steps under the assumption pH2 q;
on the other, that rP pYq is a root of the polynomial BP pY, ´sq whose multiplicity is
at most its order mP pYq as a pole of ζP pY, ¨ q. The same claim is proved to hold
for the global meromorphic distribution ζP upon considering its smallest pole rP ,
the corresponding multiplicity mP and the global Sato-Bernstein polynomial BP psq.
Finally, the specialisation to the case where the polynomial Ppxq is the homogeneous
form }Fpxq}2 will complete the proof of Theorem 1.1.

To establish the aforementioned claims, define, whenever it makes sense, the (global)
real log–canonical threshold of the map determined by Ppxq over the domain Ω with respect to
the test function ψ as the pair

R-LCTP pΩ, ψq “ prP pΩ, ψq, mP pΩ, ψqq , (2.63)

where, here again, rP pΩ, ψq denotes the smallest real pole of the map xζP pΩ, ¨ q , ψy
and where mP pΩ, ψq is its order. If not well–defined, let RLCTP pΩ, ψq “ p8, ´q; the
quantity mP pΩ, ψq is in particular left undefined (this happens, for instance, when the
variety ZR pPq does not intersect the support of the test function ψ). Similarly, given a
test function ψ, let
R-LCTP pK, ψq “ prP pψq, mP pψqq , (2.64)
where rP pψq denotes the smallest real pole of the map xζP , ψy and mP pψq its order
(adopting the same convention as above when these quantities do not exist). When
ψ meets the support restriction (1.10), R-LCTP pK, ψq equals the pair prP pKq, mP pKqq,
which is well-defined by assumption.

Given a subset Z Ă Rn with positive measure, the notation R-LCTP pZ|ψq is used,
provided it makes sense, to refer to the şpair given by the smallest real pole of the zeta
function defined for Repsq ď 0 by s ÞÑ Z |Ppxq|´s ¨ ψpxqdx, and by its order. It is here
again set to be p8, ´q if not well-defined.

The following lemma is concerned with the determination of the real log–canonical
threshold in the case where the polynomial Ppxq is a monomial.
 local and global real log–canonical thresholds 51

Lemma 2.14. Assume that ψ is an element in C8 n `

c pR q such that ψp0q ą 0. Let Yn “ p0, 8q
n

be the nonnegative orthant in Rn and let k “ pk1 , . . . , kn q P Nn0 and h “ ph1 , . . . , hn q P Nn0
be n–tuples of non–negative integers. Then,
´ ¯
R-LCTxk Yn |x ¨ ψ “ pr, mq ,
` h

where
" *
hi ` 1 hi ` 1
r “ min and m “ # 1ďiďn : “ r .
1ďiďn ki ki
In these relations, a ratio is conventionally taken to be infinite if the denominator vanishes, and
the quantity m is left undefined whenever r “ 8.
Proof. The claim regarding the value of the rational r is the specific calculation worked
out in the proof of [6, Lemma 7.3]. The value of the integer m follows from [6, Lemma
7.5, (1)].
Lemma 2.14 extends to the case of any polynomial Ppxq by resolving singularities.
This is the main content of the following statement.
Lemma 2.15. Let x0 be a point in the closure of the set Y Ă Rn . Then, there exists a neigh-
bourhood Yx0 of x0 in Rn such that the following properties hold :
(W1) for any two functions φx0 , φx1 0 ě 0 in C8 n
c pR q with supports contained in `Yx0 and
taking strictly positive values at x0 , one has that R-LCTP pY, φx0 q “ R-LCTP Y, φx1 0 .
˘

Let
R-LCTP,x0 pYq “ rP,x0 pYq, mP,x0 pYq
` ˘
(2.65)
denote this common value.

(W2) When Ppx0 q “ 0, the pair R-LCTP,x0 pYq can be determined by resolving singularities
and the quantity rP,x0 pYq is then a strictly positive rational number. If Ppx0 q ‰ 0, the
quantity rP,x0 pYq is undefined (and is thus conventionally set to equal 8).

(W3) let ζP,x0 pY, ¨ q be the restriction of the distribution ζP pY, ¨ q to the open neighbourhood
Yx0 : it is defined for all φx0 in@ C8 n
c pR q with support
D contained in Yx0 and all s P C
such that Repsq ď 0 by setting ζP,x0 pY, sq , φx0 “ xζP pY, sq , φx0 y. Then, assuming
that Ppx0 q “ 0, the rational rP,x0 pYq ą 0 introduced in (2.65) is the smallest real pole of
the local distribution ζP,x0 pY, ¨ q and mP,x0 pYq is its order.

(W4) Whenever finite, rP,x0 pYq is also a root of the local Sato–Bernstein polynomial
BP,x0 pY, ´sq defined in (2.6) with multiplicity at most mP,x0 pYq.
The quantity R-LCTP,x0 pYq introduced in (2.65) defines the (local) real log–canonical
threshold of the map x ÞÑ Ppxq at the point x0 of the closure of the domain Y. One of the
main features of the proof of the lemma is to provide an explicit formula for each of
the components of this pair. Related expressions are known for the first component
rP,x0 pYq, whenever well–defined — see, e.g., [55, §8.5 & §10.7] and [69].
 local and global real log–canonical thresholds 52

Proof. If Ppx0 q ‰ 0, there exists a neighbourhood Yx0 of x0 such that Ppxq does not
vanish on Yx0 . It is then immediate that R-LCTP,x0 pYq “ p8, ´q and the lemma is
trivially true in this case. Assume therefore that Ppx0 q “ 0.

Recall that Y is a finite union of sets of the form (2.4), where the maps fi (1 ď i ď l)
are analytic. Apply Hironaka’s Theorem 2.1 on simultaneous resolution of singulari-
ties to the set of all maps x ÞÑ fi px0 ` xq (1 ď i ď l) and x ÞÑ Ppx0 ` xq vanishing at the
origin. It provides the existence of an analytic morphism g defined on a real manifold
M and taking values in a neighbourhood W of x0 in Rn such that Ppgpyqq and the
relevant fi pgpyqq’s take the monomial form (2.1) on some chart My0 Ă M. These charts
My0 are furthermore centered around points y0 of the preimage g´1 ptx0 uq. Since
g is proper, this preimage is compact and therefore admits a finite open subcover
tMy0 uy0 PIpx0 q , where Ipx0 q is a finite index set.
´Ť ¯
The first step is to show that g y0 PIpx0 q My0 contains an open neighbourhood
p1q
Yx0 of x0 in Rn . Indeed,
! ) assume this ´Ť is not the case.
¯ Then, there exists a bounded
pjq
sequence of points x0 in Wzg y0 PIpx0 q My0 with limit x0 . Since g is surjective,
!jě1 ) ´ ¯
pjq pjq pjq
there exists a sequence y0 in M such that g y0 “ x0 for all j ě 1. The
! ) jě1
pjq
sequence x0 being bounded, the properness of the map g implies that the
! )jě1
pjq
sequence y0 lies in a compact set. Upon considering a subsequence, it can be
jě1
assumed to converge to a point y˚0 , which remains outside the set y0 PIpx0 q My0 as it is
Ť
` ˘
open. However, the relation g y˚0 “ x0 shows that y˚0 lies in the preimage g´1 ptx0 uq,
which is covered by tMy0 uy0 PIpx0 q . This contradiction establishes the claim.

Consider now a maximal subset ´ Jpx0 q of

¯ the index set Ipx0 q such that for all y0 P
p1q
Jpx0 q, the chart My0 “ My0 X g
1 ´1 Yx0 X Y has positive measure. Let tπy0 uy0 PJpx0 q be
a partition of unity subordinate to the collection of open sets My1 0 y PJpx q such that
␣ (
0 0
each πy0 takes a @positive value atD y0 . Let rP,x0 pY, φx0 q denote the smallest real pole
of the map σ ÞÑ ζP,x0 pY, σq , φx0 , assuming for the time being that this quantity is
p1q
well–defined. Then, given a smooth test function φx0 with support in Yx0 , assuming
that the considered partition of unity has been chosen so that it sums to the constant
function 1 on its domain of definition intersected with g´1 pSupp φx0 q, it holds that for
any σ ă rP,x0 pY, φx0 q,
@ D ÿ ż ˇ ˇ
ζP,x0 pY, σq , φx0 “ |Ppg pyqq|´σ ¨ φx0 pg pyqq ¨ ˇJacg pyqˇ ¨ πy0 pyq ¨ dy.
y0 PJpx0 q My0
1

(2.66)
For each y0 P Jpx0 q, the boundary constraints fi pgpyqq ě 0 and Ppgpyqq ě 0 are mono-
mial inequalities. The set My1 0 is thus a finite union of closed orthants which intersect
 local and global real log–canonical thresholds 53

any neighbourhood of y0 . As a consequence, each of the above integrals can be re-

duced to a finite sum of, say, Kpy0 q ě 1 integrals of the form
ż
yh´σk ¨ ψpyq ¨ dy, (2.67)
`
Yn

where Yn` “ p0, 8qn , where ψp0q ą 0 and where for 1 ď j ď Kpy0 q, k “ kj py0 q P Nn0
and h “ hj py0 q P Nn0 . Here, the exponents k and h depend neither on the smooth func-
tion ψ nor on the choice of the orthant indexing the finite sum. The real log–canonical
threshold related to each of the integrals (2.67) is then determined from Lemma 2.14,
and so is R-LCTP pY, φx0 q in view of the decomposition (2.66). Since the resulting for-
mula is independent of the choice of the test function φx0 , the ´claim (W1) in the state-
¯
p1q pnq
ment follows. Explicitly, if, in the above notations, kj py0 q “ kj py0 q, . . . , kj py0 q
´ ¯
p1q pnq
and hj py0 q “ hj py0 q, . . . , hj py0 q , then R-LCTP,x0 pYq “ rP,x0 pYq, mP,x0 pYq , where
` ˘

plq
hj py0 q ` 1
rP,x0 pYq “ min min min plq
y0 PJpx0 q 1ďjďKpy0 q 1ďlďn kj

and
$ ,
plq
& hj py0 q ` 1 .
mP,x0 pYq “ max max # 1ďlďn : plq
“ rP,x0 pY, ψq .
y0 PJpx0 q 1ďjďKpy0 q % k -
j

These formulae, and the fact that they do not depend on the choice of the resolution
of singularities, hold provided that the existence of R-LCTP,x0 pY, φx0 q is justified when
φx0 is a smooth test function with support in any small enough neighbourhood of x0 .
This follows from the fact that the local zeta distribution ζP,x0 pY, ¨ q admits a meromor-
phic continuation to the whole complex plane. This, in turn, is a consequence of the
following local version of the identity (2.21), which is deduced from the existence of
the local Sato–Bernstein polynomial BP,x0 pY, sq : keeping the same notation as in the
p2q
aforementioned identity, there exists a neighbourhood Yx0 of x0 in Rn such that for
p2q
any smooth test function φx0 supported in Yx0 , it holds that

BP,x0 pY, ´sq ¨ ζP,x0 pY, sq , φx0 “ ζP,x0 pY, s ´ 1q , D˚ px, ´s, Bq φx0 pxq .
@ D @ D
(2.68)

To complete the proof of (W1) and to establish (W2), it is therefore enough to set
p1q p2q
Yx0 “ Yx0 X Yx0 .

As for (W3) and (W4), they are also consequences of (2.68). Indeed, to establish
first the claim (W3), let φ̃x0 be any smooth test function supported in Yx0 such that the
 local and global real log–canonical thresholds 54

@ D
meromorphic map s ÞÑ ζP,x0 pY, sq , φ̃x0 achieves the smallest real pole, say r ą 0, of
the zeta distribution ζP,x0 pY, ¨ q. Assume that the smooth test function φx0 supported
in Yx0 bounds from above |φ̃x0 |. Then, given σ ă r, the inequality
ˇ@ Dˇ @ D
ˇ ζP,x pY, σq , φ̃x ˇ ď ζP,x pY, σq , φx (2.69)
0 0 0 0

shows that r is also the smallest pole of the meromorphic map defined on the
right–hand side, and that the orders coincide (the argument is similar to the one
used in the relation (2.46)). From (W1), this establishes that the smallest pole of the
distribution ζP,x0 pY, ¨ q is rP,x0 pYq and that it has order mP,x0 pYq, whence (W3).

Finally, (W4) is a consequence of the following observation : since the right–hand

side of (2.68) is holomorphic at rP,x0 pYq (this is a consequence the definition of rP,x0 pYq
as the smallest real pole of the distribution ζP,x0 pY, ¨ q), for any given smooth test
function
@ φx0 supported D in Yx0 , the order of the pole rP,x0 pYq of the meromorphic map
σ ÞÑ ζP,x0 pY, σq , φx0 cannot be larger than the multiplicity of rP,x0 pYq as a root of the
polynomial BP,x0 pY, ´sq.

This completes the proof of the Lemma 2.15.

The relation between the global real log–canonical thresholds defined in (2.63) and
in (2.64) and the local ones introduced in Lemma 2.15 is made clear in the following
proposition. In order to state it, introduce an order over pairs of nonnegative reals by
setting, given a, b, c, d ě 0,

pa, bq ď pc, dq ðñ pa ă cq _ ppa “ cq ^ pb ě dqq . (2.70)

Explicitly, this corresponds to the ordering induced by the asymptotic growth of the
map x ÞÑ x´a ¨ plog xqb at infinity. Set also pa, bq ě pc, dq to mean pc, dq ď pa, bq.

Recall that when a smooth test function ψ meets the support restriction condi-
ton (1.10), it takes, by assumption, strictly positive values on the set C introduced
in this condition.
Proposition 2.16. Keep the notation of Lemma 2.15. When Y “ Ω, recall that the real polyno-
mial Ppxq is assumed to be homogeneous and when Y “ K, recall that the set K is assumed to
be semianalytic and to meet the conditions pH1 q ´ pH3 q. The following claims then hold true :
(X1) Let ψ ě 0 be a smooth test function such that ψp0q ą 0 when Y “ Ω and such that
support restriction condition (1.10) is met when Y “ K. Then,

R-LCTP pY, ψq “ minn R-LCTP,x0 pYq “ min R-LCTP,x0 pYq, (2.71)

x0 PR x0 PZR pPq

where the minimum is taken with respect to the order induced by (2.70). This pair can
furthermore be computed by resolving singularities in a finite number of steps and its
first component is a strictly positive rational number.
 local and global real log–canonical thresholds 55

(X2) The pair R-LCTP pY, ψq is independent of the choice of ψ ě 0 provided that ψp0q ą 0
when Y “ Ω and provided that support restriction condition (1.10) is met when Y “ K.
Denote the common value by R-LCTP pYq “ prP pYq, mP pYqq. Then, rP pYq is the smallest
real pole of the distribution ζP pY, ¨ q and mP pYq is its order, under the following additional
requirement when Y “ K : only those poles obtained when testing the distribution against
smooth maps satisfying the support restriction condition (1.10) must be considered.

(X3) The pole rP pYq is a root of the global Sato–Bernstein polynomial BP pY, ´sq with respect
to the domain Y as defined in (2.7). Its multiplicity as a root is, furthermore, at most its
order mP pYq as a pole.

The proof of the statement makes it clear that the above defined pair R-LCTP pY, ψq
belongs to the set Qą0 ˆ Ně1 if, and only if, the support of ψ intersects non trivially the
algebraic variety ZR pPq. This is what is guaranteed by the assumption of homogeneity
when Y “ Ω and ψp0q ą 0 and by the condition pH2 q when Y “ K. In the latter case, it
is immediate that the definition of the pair R-LCTP pKq ensuing from the above point
(X2) coincides with the pair prP pKq, mP pKqq introduced in the assumption pH3 q.

Let O be the finite collection of sets defined in (2.12). Define the real log–canonical
threshold of the polynomial Ppxq as

R´LCTP “ min R-LCTP pΩq. (2.72)

ΩPO

From Lemma 2.11 and from the above proposition,

R-LCTP “ prP , mP q P Qą0 ˆ N, (2.73)

where rP is the smallest real pole of the zeta distribution ζP and mP its order.

Proof. The second relation in (2.71) follows from the fact that the local real-log canonical
threshold is p8, ´q at a point where the polynomial Ppxq does not vanish. To prove the
first equation, assign to each x0 lying in the support of the test function ψ ě 0 an
open neighbourhood Yx0 in Rn satisfying all the conclusions of Lemma 2.15. Extract a
cover tYx0 ux0 PSpψq of the support of ψ , where the index set Spψq has a finite cardinality.
Consider a smooth partition of unity tπx0 ux0 PSpψq subordinate to this cover such that
πx0 px0 q ą 0 for each x0 P Spψq and such that this partition of unity sums to the constant
function 1 in a small enough neighbourhood of Supp ψ. Then,
ÿ
xζP pY, sq , ψy “ xζP pY, sq , ψ ¨ πx0 y (2.74)
x0 PSpψq

whenever s P C is not a pole of any of the maps in the sum of the right–hand side. As
in the proof of Lemma 2.11, this decomposition implies that

R-LCTP pY, ψq “ min R-LCTP,x0 pY, ψ ¨ πx0 q . (2.75)

x0 PSpψq
 local and global real log–canonical thresholds 56

In this relation, the claim (W1) in Lemma 2.15 implies that R-LCTP,x0 pY, ψ ¨ πx0 q “
R-LCTP,x0 pYq whenever ψ px0 q ą 0. Dealing with the case where ψ px0 q “ 0 requires
more work. Assume therefore that for a given x0 P Spψq, it holds that ψ px0 q “ 0. The
argument differs depending on whether Y “ Ω or Y “ K.

r be a nonnegative map in C8 pRn q such

Consider first the case where Y “ K. Let ψ c
that :
(a) with the notations of the support restriction condition (1.10), the support of ψ
r
is contained both in the open neighbourhood U of K and in the small enough
neighbourhood of Supp ψ where the partition of unity tπx0 ux0 PSpψq sums to the
constant function 1;

(b) it holds that 0 ď ψ ď ψ

r and that ψ
r is strictly positive in the interior of its support.

It is then clear that for any x0 P Spψq,

A E
0 ď xζP pK, σq , ψ ¨ πx0 y ď ζP pK, σq , ψ
r ¨ πx
0

as long as the real σ remains smaller than the smallest real pole of the left-hand side.
As a consequence,
´ ¯
R-LCTP pK, ψ ¨ πx0 q ě R-LCTP K, ψr ¨ πx .
0
(2.76)

Since ψpx
r 0 q ą 0 from the above assumption (b), the claim (W1) in Lemma 2.15 implies
that the right-hand side is precisely R-LCTP,x0 pKq. Given that the map ψ meets the
support restriction condition (1.10), the assumption pH3 q yields

R-LCTP pKq “ R-LCTP pK, ψq “ min R-LCTP,x0 pK, ψ ¨ πx0 q

(2.75) x0 PSpψq
´ ¯
ě min R-LCTP,x0 K, ψr ¨ πx
0
(2.76) x0 PSpψq
“ min R-LCTP,x0 pKq . (2.77)
x0 PSpψq

From the above assumption (a), upon testing the distribution in (2.74) against ψ
r instead
of ψ, the equation (2.75) remains true with ψ
r inside the real log-canonical thresholds.
From the assumption pH3 q again, one thus has that
´ ¯
R-LCTP pKq “ R-LCTP pK, ψq r “ min R-LCTP,x K, ψ
0
r ¨ πx .
0
x0 PSpψq

Combined with the equation (2.77), this finally establishes that when Y “ K, it indeed
holds that
R-LCTP pK, ψq “ min R-LCTP,x0 pKq .
x0 PSpψq
 local and global real log–canonical thresholds 57

Showing that the same relation is valid in the case where Y “ Ω requires a different
and more elaborate argument. To this end, the first goal is to show that there exists x
such that ψ pxq ą 0 satisfying the property that
R-LCTP,x0 pΩ, ψ ¨ πx0 q ě R-LCTP,x pΩ, φx q . (2.78)
In this relation, it is required that φx ě 0 should be a test function supported in a
small enough neighbourhood of x such that φx pxq ą 0. This can be obtained from
the homogeneity of the polynomial Ppxq. Indeed, under this assumption, for any real
σ ă rP,x0 pΩq,
@ D
xζP pΩ, σq , ψ ¨ πx0 y ! xζP pΩ, σq , πx0 y ! ζP pΩ, σq , πx0 ,λ . (2.79)
The second inequality is here a consequence of the change of variables y “ x{λ upon
setting πx0 ,λ : x P Rn ÞÑ πx0 pλxq. The parameter λ is chosen to be large enough
so that ψpx0 {λq ą 0. Setting x “ x0 {λ and φx “ πx0 ,λ , the inequality between the
first and the last terms in (2.79) then implies the one in (2.78). In fact, one may even
assume that φx is supported in an arbitrarily small neighbourhood of x. To see this, it
is enough to note that if πx1 0 ě 0 is any smooth test function taking a strictly positive
value at x0 , then the claim (W1) in Lemma 2.15 implies@ that the smallest Dreal poles
of the meromorphic maps s ÞÑ xζP pΩ, σq , πx0 y and s ÞÑ ζP pΩ, σq , πx0 ¨ πx1 0 coincide,
and so do their respective multiplicities. Therefore, for any real σ strictly less than the
common value rP,x0 pΩq taken by these smallest poles,
@ D A E
xζP pΩ, σq , πx0 y ! ζP pΩ, σq , πx0 ¨ πx1 0 ! ζP pΩ, σq , πx0 ,λ ¨ πx1 0 ,λ ,

where πx1 0 ,λ is defined in the same way as πx0 ,λ . One may then set φx “ πx0 ,λ ¨ πx1 0 ,λ
upon choosing adequately the support of πx1 0 .

To conclude the proof of the relation (2.71) when Y “ Ω, consider a small enough
neighbourhood Ωx of x on which ψ remains positive and where the conclusions of
Lemma 2.15 hold. Working with a partition of unity tπx0 ux0 PSpψq Y tπx u subordinate to
the cover tΩx0 ux0 PSpψq Y tΩx u of the support of ψ, the same argument as the one used
to obtain (2.75) implies that
" *
R-LCTP pΩ, ψq “ min R-LCTP,x pΩ, ψ ¨ πx q , min R-LCTP,x0 pΩ, ψ ¨ πx0 q .
x0 PSpψq

Combining this with (2.75) thus yields that

R-LCTP,x pΩ, ψ ¨ πx q ě R-LCTP pΩ, ψq. (2.80)
Since, by assumption, πx pxq ą 0, ψpxq ą 0 and φx pxq ą 0, one infers from the claim
(W1) in Lemma 2.15 and from the relations (2.78) and (2.80) that, at any point x0 where
ψpx0 q “ 0,
R-LCTP,x0 pΩ, ψ ¨ πx0 q ě R-LCTP,x pΩq ě R-LCTP pΩ, ψq .
 local and global real log–canonical thresholds 58

The minimum in the equation (2.75) is thus achieved at those points where ψ takes a
positive value. At such points, the local log–canonical threshold can be determined by
resolution of singularities from Lemma 2.15. This establishes (X1).

As for (X2), that the pair R-LCTP pY, ψq should be independent of the choice of the
test function ψ ě 0 under the assumptions of the statement is immediate from the
relations (2.71).

Assume that for such a map ψ, it holds that R-LCTP pY, ψq “ R-LCTP pYq and
denote this pair by R-LCTP pYq “ prP pYq, mP pYqq. Recall that the argument based on
the Triangle Inequality in (2.69) shows that one may assume without loss of generality
that the log-canonical thresholds of the various zeta distributions under consideration
are achieved when tested against a nonnegative map in C8 n
c pR q. Then, when Y “ K,
it is clear that the assumption pH3 q implies that rP pKq is the smallest real pole of the
distribution ζP pY, ¨ q when it is tested against all nonnegative smooth test functions
satisfying the support restriction condition (1.10), and that mP pKq is the corresponding
order.

When Y “ Ω, it follows from the homogeneity of the polynomial Ppxq that the
pair R-LCTP pΩq is also the log-canonical threshold of the restriction of the distribution
ζP pΩ, ¨ q to any neighbourhood of the origin (this is, here again, a simple consequence
of the relation (2.37)). That rP pΩq should be the smallest real pole of the distribution
ζP pΩ, ¨ q and that mP pKq should be the corresponding order is then easily deduced
from the claim (W3) in Lemma 2.15. This establishes (X2).

To prove (X3), fix ψ ě 0 in C8 n

c pR q such that ψp0q ą 0 when Y “ Ω and such that
the support restriction (1.10) is met when Y “ K. From the claim (X2), the meromorphic
map s ÞÑ xζP pY, sq, ψy achieves the log-canonical threshold R-LCTP pYq “ prP pYq, mP pYqq.
Note then that there exists a differential operator, say DY px, s, Bq, such that
BP pY, ´sq ¨ xζP pY, sq , ψy “ xζP pY, s ´ 1q , D˚Y px, ´s, Bq ψy , (2.81)
where the polynomial BP pY, sq is defined in (2.7) and where D˚Y px, s, Bq denotes here
also the dual operator. To see this, let tσx0 ux0 PIpY q be a locally finite partition of unity
subordinate to the open cover tUx0 ux0 PIpY q of the topological closure Y of Y, where the
sets Ux0 result from Lemma 2.3. For each x0 P I Y , there exists a differential operator
` ˘

DY,x0 px, s, Bq acting on Y X Ux0 such that

BP,x0 pY, sq ¨ Ppxqs “ DY,x0 px, s, Bq Ppxqs`1 . (2.82)
Here, the local polynomial BP,x0 pY, sq defined in (2.6) divides the global one BP pY, sq;
consequently, there exists some polynomial TP,x0 pY, sq with rational coefficients such
 parameters of the volume estimate 59

that BP pY, sq “ TP,x0 pY, sq ¨ BP,x0 pY, sq. Choose the partition of unity tσx0 ux0 PIpY q in
such a way that it sums to the constant function 1 in a small enough neighbourhood
of Y, and assume without loss of generality that the support of ψ is contained in this
neighbourhood. It then follows from the relation (2.82) that
ÿ
BP pY, ´sq ¨ xζP pY, sq , ψy “ TP,x0 pY, ´sq ¨ BP,x0 pY, ´sq ¨ xζP pY, sq , σx0 ¨ ψy
x0 PIpY q
ÿ A E
“ TP,x0 pY, ´sq ζP pY, s ´ 1q , D˚Y,x0 px, ´s, Bq pσx0 ¨ ψq .
x0 PIpY q

In view of the formula defining the dual operator in (2.20), it is thus enough to let
ÿ
DY px, s, Bq “ TP,x0 pY, sq ¨ σx0 pxq ¨ DY,x0 px, s, Bq
x0 PIpY q

so that the identity (2.81) holds.

Set then s “ σ to be a real number and let σ ă rP pYq tend to rP pYq in the
identity (2.81) : since the right–hand side is holomorphic at σ “ rP pYq, the claim (X3)
follows.

This completes the proof of the Proposition 2.16.

2.4 parameters of the volume estimate under ad-

ditional regularity and homogeneity assump-
tions
The goal in this concluding section of the chapter is to establish the few remaining
claims to complete the proof of the main Theorem 1.1. To begin with, the following
proposition establishes under more general assumptions the claim on the localisation
of the smallest real pole of the zeta distribution ζF made in Theorem 1.1.

Proposition 2.17. Let O 1 be a subcollection of the collection of sets O defined in (2.12). Assume
that the corresponding distribution ζP pO 1 , ¨ q introduced in (2.32) is well–defined. Then, under
the assumption of the homogeneity of the polynomial Ppxq, it holds that the smallest real pole
rP pO 1 q of ζP pO 1 , ¨ q lies in the interval p0, n{qs, where q denotes the degree of Ppxq.

Proof. From Lemma 2.11, it is enough to establish the statement for the quantity rP pΩq
for any fixed Ω P O 1 such that the distribution ζP pΩ, ¨ q is well-defined. To this end,
consider a nonnegative, smooth and compactly supported function ξ defined over the
real line and taking a positive value at the origin. Define a radial function ψ in C8
c pRq
 parameters of the volume estimate 60

by setting ψ pxq “ ξ p}x}q for all x P Rn . Then, denoting by τn´1 the induced Lebesgue
measure on the n–dimensional sphere Sn´1 Ă Rn , it holds that for any real σ ă rP pΩq,
ż
ψpxq
xζP pΩ, σq , ψy “ σ
¨ dx
Rn PΩ pxq
ˆż ˙ ˆż ˙
n´1´qσ ´σ
“ ξprq ¨ r ¨ dr ¨ PΩ puq ¨ dτn´1 puq .
R Sn´1

Let σ0 ą 0 be the smallest real pole of the second factor on the right–hand side of this
relation. Then, clearly, rP pΩq ď min tn{q, σ0 u ď n{q under the assumption that ξp0q ą
0. The sought conclusion then follows from the point (X2) in Proposition 2.16.
Recall that Fpxq “ pF1 pxq, ¨ ¨ ¨ , Fp pxqq is assumed to be a set of p ě 1 homogeneous
forms in n ě 2 variables with common degree d ě 2. Set
p
ÿ
2
PF pxq “ }Fpxq} “ Fi pxq2 . (2.83)
i“1

Proposition 2.18 (The real log-canonical threshold in the case of a smooth complete
intersection). With the above notations, assume that the set of homogeneous forms Fpxq has a
smooth complete intersection over a domain K Ă Rn satisfying the conditions pH1 q and pH2 q.
Suppose furthermore that, if the condition pH3 q is not assumed, then the algebraic variety
ZR pFq defined in (1.5) intersects non trivially the interior of K.
Then, the condition pH3 q holds in all cases. Furthermore, the real log-canonical threshold of
the polynomial PF pxq over the domain K is well-defined and equals
´p ¯
R-LCTPF pKq “ ,1 .
2
Proof. In the case that it is just assumed that ZR pFq intersects non trivially the interior
of K, define in the notation of the support restriction condition (1.10) the open set U as
an open neighbourhood of K where the map (1.6) does not vanish. Let also the set C be
any compact subset of the interior of K with an interior intersecting ZR pFq non trivially.
Fix then a smooth test function ψ ě 0 meeting the support restriction condition (1.10)
for this choice of the sets C and U. The goal is to show that the smallest real pole of
the map @ D
s ÞÑ ζPF psq, ψ (2.84)
and the corresponding order do not depend on the choice of ψ, and also that they
equal the values p{2 and 1, respectively.

In the case that the condition pH3 q is assumed, it may furthermore be imposed
without loss of generality, even if it means making it slightly bigger, that the support
of ψ ě 0 intersects non trivially the variety ZR pFq (this can be done since the condition
pH2 q is assumed). In view of the point (X2) in Proposition 2.16, the goal is then to
prove that the real log-canonical threshold is again the pair pp{2, 1q by establishing this
 parameters of the volume estimate 61

claim for the above–defined map (2.84).

To prove these claims, note that the smooth complete intersection assumption im-
plies that the function
` ˘
x “ px1 , . . . , xn q ÞÑ F1 pxq, . . . , Fp pxq, xp`1 , . . . , xn
›2
realises a C1 -diffeomorphism between K and its image (its Jacobian is ˘ › pi“1 ∇Fi pxq› ).
›Ź

As a consequence, the corresponding change of variables yields that for any real σ
strictly smaller than the smallest real pole of the map (2.84),
ż ż
@ D ψpxq ψ
r F pyq
ζPF pσq, ψ “ ¨ dx “ ¨ dy,
Rn }PF pxq}2σ (2.83) Rp }y}2σ

where ψ r F is a smooth nonnegative test function taking a strictly positive value at the
origin (this follows from the assumption that the support of ψ intersects non trivially
the variety ZR pFq from the above discussion). Thus, working in polar coordinates, one
obtains the existence of a nonnegative map ξFψ in C8 c pRq taking a strictly positive value
at the origin such that
ż
ξFψ prq ¨ rp´1´2σ ¨ dr.
@ D
ζPF pσq, ψ “
R

It is clear that the smallest real pole of the map defined by the right-hand side is
p{2 and that it has order 1, regardless of the choice of the smooth compactly supported
map ξFψ ě 0 such that ξFψ p0q ą 0. This completes the proof of the proposition.
Completion of the proof of Theorem 1.1. The final goal in this chapter is to establish the
few remaining claims in the theorem which have not already been proved. To this
end, let PF pxq be as in (2.83). Denote by ζPF the corresponding distribution as defined
in (2.10) and recall that ζF denotes the distribution introduced in (1.9). It is clear from
these definitions that when tested against any smooth compactly supported function,
the smallest pole of the former distribution is obtained by dividing by 2 the smallest
pole of the latter, and that the orders coincide.

In view of this observation, it is immediate from Proposition 2.18 that in the case
of a smooth complete intersection, the real log canonical threshold of the distribution
ζF is, in the notations of the theorem, the pair pp, 1q.

Similarly, to prove that the pairs prF pKq, mF pKqq and prF , mF q can be determined by
resolving singularities in a finite number of steps and that their first components are
strictly positive rationals, it is enough to prove this statement for the real log-canonical
thresholds R-LCTPF pKq and R-LCTPF (i.e. for the polynomial PF pxq). The claim then
immediately follows from the points (X1) and (X2) in Proposition 2.16 as far as
R-LCTPF pKq is concerned. As for R-LCTPF , one needs to additionally consider the
 parameters of the volume estimate 62

equation (2.72) to reach the same conclusion.

To prove that the rationals ´rF pKq{2 “ ´rPF pKq and ´rF {2 “ ´rPF are roots of the
Sato-Bernstein polynomial BF psq with respective multiplicities at most mF pKq “ mPF pKq
and mF “ mPF , apply the point (X3) in Proposition 2.16 in the case of the homogeneous
form PF pxq. Then,
` ˘
‚ the claim for the pair ´rPF pKq, mPF pKq follows from the observation that the
global Sato-Bernstein polynomial BPF pK, sq with respect to the domain K divides
the Sato-Bernstein polynomial BPF psq of the homogeneous form PF pxq;
` ˘
‚ in the case of the pair ´rPF , mPF , it is implied by the same division property
satisfied by the polynomial BP pΩ, sq, where Ω P O, with the additional need to
include the conclusion of Lemma 2.11 .

To complete the proof of Theorem 1.1, it remains to notice that Proposition 2.17
applied to the homogeneous form PF pxq of degree q “ 2d and the relation rF {2 “ rPF
yield that the rational rF lies in the interval p0, n{ds.
3 LO G – C A N O N I C A L T H R E S H O L D, R O OT S
O F T H E S ATO – B E R N S T E I N
P O LY N O M I A L A N D R E L A T E D
M U LT I P L I C I T I E S

The goal of this section is to establish Theorem 1.3. Let Ppzq P Crzs be a (non–
necessarily homogeneous) complex polynomial in n–variables z “ pz1 , . . . , zn q. Define
its (complex) log–canonical threshold as the pair

C-LCTP “ pr̂P , m̂P q ,

where r̂P is the smallest real pole of the complex zeta distribution ζpP defined in (1.14)
and where m̂P is its order (in accordance with the previous section, this slightly extends
the definition of the complex log–canonical threshold given in §1.3 of the Introduction,
where only the quantity r̂P was considered). The existence of the pair C-LCTP follows
from the fact that ζpP admits a meromorphic extension to the entire complex plane. To
see this, let Bp psq P Qrss be the Sato–Bernstein polynomial of Ppzq and let Dpz, s, Bq
be a differential operator such that the identity (2.3) holds with Bp psq. If Dpz, s, Bq is
expanded as in (2.19), define its conjugate by setting, with obvious notations,
ÿ α
Dpz, s, Bq “ aα pz, sq ¨ B .
αPNn
0

It is then easily seen that the operator Dpz, s, Bq commutes with its conjugate Dpz, s, Bq.
Moreover,
´ ¯ ´ s`1
¯
Dpz, s, BqPpzqs`1 ¨ Dpz, s, BqPpzq “ Dpz, s, BqDpz, s, Bq |Ppzq|2ps`1q ,
` ˘

which implies that for any ψ in C8 n

c pC q,
A E A ´ ¯ E
˚
|BP psq|2 ¨ ζpP psq , ψ “ ζP ps ´ 1q , Dpz, s, BqD pz, s, Bq ψ . (3.1)

The meromorphic continuation of the distribution ζpP then follows in the same way as
in the proof of Lemma 2.4.

It is known, see, e.g., [55, §§10.6 & 10.7] and [69, §3], that the smallest pole r̂P is a
well–defined rational number such that

0 ă r̂P ď 1. (3.2)

63
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 64

In the case that Ppzq is a polyomial with real coefficients, it always holds that

rP ě r̂P ,

where rP is the first component of the real log–canonical threshold defined in (2.73).
See [69] for a justification of this claim, where it is also shown that the inequality may
be strict, and that it may also happen that rP ą 1 in the case that ZR pPq is an algebraic
variety of dimension strictly less than n ´ 1. It is nevertheless the case that the (global)
complex and real Sato–Bernstein polynomials of Ppzq which, in the notation of (2.5)
and (2.7), are respectively BP psq and BP pRn , sq, coincide. To see this, note first that
the complexification of a relation such as (2.3) valid for real variables shows that
BP psq divides BP pRn , sq. Conversely, if such a relation is true for complex variables,
then restricting to real variables and taking the real part of each side of the equation
shows that BP pRn , sq divides BP psq when the polynomial Ppzq is assumed to have real
coefficients, whence the claim.

From the just established relation BP psq “ BP pRn , sq and from the fact that rP can
be strictly larger than r̂P , one deduces there can be no real analogue of Theorem 1.3
when considering the real zeta function of a polynomial Ppxq P Rrxs.

Revert to the general case where Ppzq P Crzs. The first step in the proof of
Theorem 1.3 is the following complex analogue of Lemma 2.14. Given Z Ă Cn
and an element ψ in C8 c pC q, define, whenever it exists, C-LCTP pZ|ψq as the pair
n

pr̂pψq, m̂pψqq, where r̂pψq is the smallest real pole of the map defined for Repsq ď 0
by s ÞÑ Z |Ppzq|´2s ¨ ψpz, zq ¨ dz ^ dz and where m̂pψq is its order. Let C-LCTP pZ|ψq “
ş

p8, ´q if this is undefined. When Z “ Cn , set for the sake of simplicity of notation
C-LCTP pCn , ψq “ C-LCTP pψq.
Lemma 3.1. Assume that ψ is an element in C8 n
c pC q not vanishing in some neighbourhood
of the origin. Let k “ pk1 , . . . , kn q P Nn0 and h “ ph1 , . . . , hn q P Nn0 be n–tuples of non–
negative integers. Then, with the same convention as in Lemma 2.14 (and with a slight abuse
of notations), ´ ¯
C-LCT|zk | |z2h | ¨ ψ “ pr̂, m̂q ,
where
" *
hi ` 1 hi ` 1
r̂ “ min and m̂ “ # 1 ď i ď n : “ r .
1ďiďn ki ki
Proof. Decompose each coordinate zi (1 ď i ď n) of the point z P Cn zt0u in polar form
as zi “ ρi eiθi , where ρi ą 0 and θi P r0, 2πq. Set Yn` “ p0, 8qn , ρ “ pρ1 , . . . , ρn q P Yn`
and θ “ pθ1 , . . . , θn q P r0, 2πqn and note that dz ^ dz “ ρ ¨ dρ ^ dθ. Then, for s P C
such that the integrals exist,
ż ˇ ˇ ż
ˇ k ˇ´2s ˇ 2h ˇ
ˇ ˇ
ˇz ˇ ¨ ˇz ˇ ¨ ψpz, zq ¨ dz ^ dz “ ρ2h`1´2sk ¨ ψ̃pρq ¨ dρ. (3.3)
Cn `
Yn
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 65

Here, 1 “ p1, . . . , 1q P Nn and ψ̃ is a complex valued bounded, smooth function sup-

ported in the intersection of a compact neighbourhood of the origin with the closure
of Yn` . Furthermore, the map ψ̃ does not vanish when ρ P Yn` has small enough norm.
Thus, taking the minimum with respect to the order introduced in (2.70),
´ ¯
C-LCT|zk | |zh | ¨ ψ
! ´ ` ˘¯ ´ ` ˘¯)
“ min R-LCTρ2k Yn |ρ ` 2h`1
¨ Re ψ̃ , R-LCTρ2k Yn |ρ
` 2h`1
¨ Im ψ̃ ,
` ˘ ` ˘
where either Re ψ̃ p0q ‰ 0 or Im ψ̃ p0q ‰ 0 . The claim is therefore reduced to
Lemma 2.14.

By analogy withA the real E E local (complex) zeta distribution ζP,z0 at

case, Adefine the p
z0 P Cn by setting ζpP,z0 , φz0 “ ζpP , φz0 for any smooth function φz0 compactly
supported in a small enough open neigbourhood of z0 . This neighbourhood is chosen
such that a resolution of singularities map is well-defined over it when Ppz0 q “ 0
(cf. Theorem 2.1) and such that the polynomial Ppzq does not vanish over it when
Ppz0 q ‰ 0.

Let furthermore
C-LCTP,z0 “
` ˘
r̂P,z0 , m̂P,z0
denote the (complex) log–canonical threshold of the polynomial Ppzq at z0 : it is here again
the data of the smallest real pole r̂P,z0 of ζpP,z0 and of its order m̂P,z0 , provided it is
well–defined.

Lemma 3.2. Let z0 P Cn . If Ppz0 q “ 0, let ĝ : M p ÑW p be an analytic morphism obtained by

applying a resolution of singularities to Ppzq at z0 , where M p is a complex manifold and where
W
p Ă C is a neighbourhood of z0 . Then,
n

(Y1) when Ppz0 q ‰ 0, the pair C-LCTP,z0 “ r̂P,z0 , m̂P,z0 is not defined, and thus set to
` ˘

p8, ´q. If, however, Ppz0 q “ 0, it !

is well–defined,
) in which case the preimage ĝ´1 ptz0 uq
is covered by finitely many charts M py
0
.
y0 PÎpz0 q

Assuming from now on that Ppz0 q “ 0, let Ĵpz0 q Ă Îpz0 q be a maximal index set such that
for each y0 P Ĵpz0 q, the set M
p y has positive measure. Then,
0

(Y2) for each index y0 P Ĵpz0 q, there exists an integer K̂py0 q ě 1 such that for any φz0 smooth
and
A compactly E supported in a small enough neighbourhood of z0 , the meromorphic map
ζpP,z , φz can be decomposed as the sum of K̂py0 q integrals of the form
0 0

ż
ˇ kj py0 q ˇ´2s ˇ 2hj py0 q ˇ py0 q
ˇ ˇ ˇ ˇ
ˇy ˇ ¨ ˇy ˇ ¨ ψj py, yq ¨ dy ^ dy.
Cn
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 66

py q
Here, the smooth maps ψj 0 are supported in a neighbourhood of y0 for 1 ď j ď K̂py0 q.
Furthermore, the integer n–tuples
´ ¯ ´ ¯
p1q pnq p1q pnq
kj py0 q “ kj py0 q, . . . , kj py0 q and hj py0 q “ hj py0 q, . . . , hj py0 q

are obtained from the monomial forms (2.1) and (2.2) of the resolution of singularities ĝ
restricted to M
py .
0

(Y3) moreover, it holds that

plq
hj py0 q ` 1
r̂P,z0 “ min min min plq
(3.4)
y0 PĴpx0 q 1ďjďK̂py0 q 1ďlďn kj

and
$ ,
plq
& hj py0 q ` 1 .
m̂P,z0 “ max max # 1ďlďn : plq
“ r̂z0 pPq (3.5)
y0 PĴpx0 q 1ďjďK̂py0 q % k j
-

with the same conventions as in Lemma 2.14 to deal with the case of vanishing denomi-
nators.

The formula for r̂P,z0 above is known — see, e.g., [55, §10.7]. The main novelty is
the expression for the multiplicity m̂P,z0 .

Proof. Note first that the local distribution ζpP,z0 can be meromorphically continued to
the entire complex plane since the global distribution ζpP can be so. The lemma then
turns out to be a rephrasing of the results established as part of the proof of Lemma 2.15
upon considering a complex resolution of singularities. The only change leading to the
specific forms of the above statements (Y2) and (Y3) is that the equation (2.66) now
becomes
A E
ζpP,z0 pσq, φz0
ÿ ż ´ ¯ ˇ ˇ2
“ |Ppg pyqq|´2σ ¨ φz0 gpyq, gpyq ¨ ˇJacg pyqˇ ¨ πy0 pyq ¨ dy ^ dy,
M
py
y0 PĴpx0 q 0

ˇ ˇ2
where ˇJacg pyqˇ is the (real) Jacobian of the map of resolution of singularities conside-
red as a smooth map of real 2n–dimensional manifolds. This is enough to justify (Y2).
The calculation of the complex log–canonical threshold in (Y3) is then easily seen to
be reduced to the determination of the minimum over all˘ indices 1 ď j ď K̂py0 q and
over all y0 P Ĵpz0 q of the pair C-LCT|zkj py0 q | |z
` 2h py
j 0 q | ¨ ψ for some smooth function

ψ satisfying the assumptions of Lemma 3.1 (here again with respect to the order
introduced in (2.70)).

This concludes the proof of Lemma 3.2.

 chapter 3. log–canonical threshold and the sato–bernstein polynomial 67

The following is a local version of Theorem 1.3 which is needed to prove it in full
generality.

` Let z0 P ˘C be such that Ppz0 q “ 0. Then, the log–canonical threshold

Proposition 3.3. n

C-LCTP,z0 “ r̂P,z0 , m̂P,z0 is such that ´r̂P,z0 is the largest root of the local polynomial
BP,z0 psq with multiplicity exactly m̂P,z0 .

Proof. The statement is established with the help of two lemmata, the first one showing
that the multiplicity under consideration is at least m̂P,z0 and the second one that it is
at most this same quantity.
Lemma 3.4. `Under the assumptions of the proposition, the local log–canonical threshold
C-LCTP,z0 “ r̂P,z0 , m̂P,z0 is such that ´r̂P,z0 is the largest root of the local Sato–Bernstein
˘

polynomial BP,z0 psq with multiplicity at least m̂P,z0 .

Proof. Let Dz0 pz, s, Bq be a differential operator defined in a neighbourhood Zz0 Ă Cn

of z0 such that for all z P Zz0 ,

BP,z0 psqPpzqs “ Dz0 pz, s, Bq ¨ Ppzqs`1 . (3.6)

A E
Let ψ be in C8 c pCn q such that the meromorphic map s ÞÑ ζ pP,z psq, ψ achieves the
0
pair C-LCTP,z0 . Since the dual operator D˚z0 pz, s, Bq is polynomial in s and analytic in z,
the real
max ˇD˚z0 pz, ´σ, Bqψ pz, zqˇ
` ˘ ˇ ˇ
M ψ, r̂P,z0 “
|σ´r̂P,z0 |ď1
zPSupp ψ

is well–defined. Fixing φ ě 0 in C8 n
` ˘
c pC q` such that φ pz, ˘ zq ě M ψ, r̂P,z 0
for all z P
Supp ψ, one obtains that for all reals σ P r̂P,z0 ´ 1, r̂Pz0 ,
ˇ A Eˇ ˇż ´ ¯´σ ˇ
ˇ ´σ
ˇ
ˇBP,z0 p´σq ¨ ζP,z0 pσq, ψ ˇ “ ˇ BP,z0 p´σq ¨ Ppzq ¨ Ppzq ¨ ψ pz, zq ¨ dz ^ dzˇˇ
ˇ p ˇ ˇ
ˇCż
n
ˇ
¨ Dz0 pz, ´σ, Bqψ pz, zq ¨ dz ^ dzˇˇ
ˇ ´2σ ˚
ˇ
“ ˇˇ Ppzq ¨ |Ppzq|
(3.6)
ż C
n

ď |Ppzq|´2pσ´1{2q φ pz, zq ¨ dz ^ dz
BC
n
ˆ ˙ F
1
“ ζpP,z0 σ ´ ,φ .
2

From the definition of r̂P,z0 as the smallest real pole of the local zeta distribution, this
last quantity remains bounded as σ tends to r̂P,z0 , which is easily seen to imply the
claim.

Lemma 3.5. `Under the assumptions of the proposition, the local log–canonical threshold
C-LCTP,z0 “ r̂P,z0 , m̂P,z0 is such that ´r̂P,z0 is the largest root of the local Sato–Bernstein
˘

polynomial BP,z0 psq with multiplicity at most m̂P,z0 .

 chapter 3. log–canonical threshold and the sato–bernstein polynomial 68

Proof. The proof relies on Lichtin’s refinement of Kashiwara’s Theorem 2.2 as found
in [58].

Consider a resolution of singularity ĝ of the polynomial Ppzq at z “ z0 as in

Lemma 3.2, and keep the notations of this statement. Set

ω̂ : y P M
p ÞÑ pP ˝ ĝq pyq (3.7)

and let Bω̂ psq denote the Sato–Bernstein polynomial of the analytic map ω̂ (it is
defined as in the polynomial case by allowing the differential operators Dpy, s, Bq to be
analytic in the variable y; furthermore, it still satisfies the conclusions of Kashiwara’s
Theorem 2.2). It then follows from [58, Equation (4.6)] that there exists an integer
a ě 0 such that BP,z0 psq divides the product am“0 Bω̂ ps ` mq.
ś

It is known that a local version of the inequality (3.2) holds; namely, that ´r̂P,z0 is
a rational number in r´1, 0q which is, furthermore, the largest root of the polynomial
BP,z0 psq — see [55, §§10.6 & 10.7] and [69, §3]. Since the roots of the polynomial Bω̂ psq
are negative, this implies that the multiplicity of ´r̂P,z0 as a root of BP,z0 psq is bounded
above by its muliplicity as a root of the factor Bω̂ psq in the above product (which factor
is obtained when m “ 0).
! )
Assume that Ny0 p ĂM p is a finite collection of charts covering an open neigh-
y0
bourhood of the preimage ĝ´1 ptz0 uq. Assume also that the restriction of the map ω̂
to each Np y admits a monomial form such as in (2.1) and (2.2) with exponents, say,
` 0
kpy0 q “ kp1q py0 q, . . . , kpnq py0 q and hpy0 q “ hp1q py0 q, . . . , hpnq py0 q in Nn0 . Then, as
˘ ` ˘

stated in [58, Equation (4.13)], the Sato–Bernstein polynomial of such a restriction is

piq
n kź
py0 q
˜ ¸
ź hpiq py0 q ` b
s` , (3.8)
i“1 b“1
kpiq py0 q

where a factor is conventionally taken as 1 when kpiq py0 q “ 0. Furthermore — this

follows from [58, Point (3) p.301], the polynomial Bω̂ psq (of the "unrestricted" map ω̂
defined in (3.7)) divides any nonzero element lying in the intersection of the ideals
generated by the polynomials (3.8) as y0 varies. (It should be noted here that for the
purpose of his work, it is enough for Lichtin to choose in the aforementioned reference
the product of all such polynomials as a nonzero element in this intersection. The
present proof, however, requires to take below their lowest common multiple.)
! )
Specialise these claims to the case where the collection of charts N py
0
is the
! ) y0
neighbourhood of the preimage ĝ´1 ptz0 uq induced by the charts M py
0
in
y0 PÎpz0 q
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 69

Lemma 3.2 (where each chart M p y is itself a union of at most K̂py0 q ě 1 subsets).
0
One thus obtains that the polynomial Bω̂ psq divides
$ piq ¨ ˛ ,
’
&ź n kj ź py0 q piq
hj py0 q ` b
/
.
lcm ˝ s` piq
‚ : y0 P Îpz0 q, 1 ď j ď K̂py0 q . (3.9)
’
%i“1 b“1 kj py0 q /
-

The equation (3.4) shows that the pole r̂P,z0 is the opposite of the largest root of
the polynomial (3.9) (which largest root is in particular attained when b “ 1). The
multiplicity of this root is clearly given by the equation (3.5) expressing the order of
the quantity r̂P,z0 considered as a pole of the local zeta distribution. As a consequence,
the equation (3.5) also provides an upper bound for the multiplicity of ´r̂z0 as the
root of Bω̂ psq, and therefore as a root of the local Sato–Bernstein polynomial BP,z0 psq.

This establishes Lemma 3.5.

Proposition 3.3 follows from Lemmata 3.4 and 3.5.

Proof of Theorem 1.3. Let z0 P Cn . Consider an open neighbourhood Zz0 Ă Cn of

z0 obtained as the intersection of a neighbourhood where the local Sato–Bernstein
polynomial B p P,z psq exists and of a neighbourhood supporting the test functions
0
against which the above defined local zeta distribution ζpP,z0 is evaluated.

Then, on the one hand, it follows from the equation (2.5) that there exists z0 P Cn
such that the largest root of the local polynomial BP,z0 psq and its multiplicity coincide
with those of the global polynomial BP psq (recall here that the relation (2.5) is only
valid when the ground field is K “ C).

On the other, the global zeta distribution ζpP can be decomposed with the help
of a locally finite partition of unity into a sum of local zeta distributions ζpP,z0 , each
evaluated against test functions supported in the above defined open sets Zz0 . From
Proposition 3.3 and from the point (Y1) in Lemma 3.2, the log–canonical thresholds
of all these local distributions at points z0 of the algebraic variety ZC pPq determine
the largest roots of the corresponding local Sato–Bernstein polynomials and their
multiplicities.

The theorem then immediately follows upon combining the above two points.

The local version of Theorem 1.3 obtained in Proposition 3.3 implies the following
neat statement relating, on the one hand the largest root of the local Sato–Bernstein
polynomial and its multiplicity and, on the other, the growth rate of the weighted
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 70

volume of the complex domain around a point in the variety ZC pPq. In order to state
the result, given a vector z “ pz1 , . . . , zn q P Cn , set (with a slight abuse of notation)
g
fn
fÿ
}z} “ e |zi |2 .
i“1

Theorem 3.6. Let Ppzq P Crzs be polynomial in n ě 2 variables and let z0 P Cn . Let
r̂P,z0 P p0, 1s be the opposite of the largest root of the local Sato–Bernstein polynomial BP,z0 psq.
Denote by m̂P,z0 ě 1 the multiplicity of this root. Then, there exists δP pz0 q ą 0 such that for
any map ψz0 ě 0 in C8 n
c pC q taking a positive value at z0 and supported in the ball determined
by the inequality }z ´ z0 } ă δP pz0 q, it holds that
ż ´ ¯
χt|Ppzq|ďcu ¨ ψz0 pz, zq ¨ dz ^ dz “ γP,z0 pψz0 q ¨ cr̂P,z0 ¨ Rψz0 p|log c|q ` O cr̂P,z0 `ε
Cn
(3.10)
for some ε ą 0 as the parameter c tends to zero. In this relation, γP,z0 pψz0 q ą 0 is a positive
constant and Rψz0 pxq P Rrxs is a univariate monic polynomial with degree m̂P,z0 ´ 1.

Upon considering an appropriate partition of unity, this statement and its proof
can be extended to the case where the volume estimates are concerned with a union
of finitely many balls with radii δP pz0 q ą 0 centered at points z0 lying in the com-
plex variety ZC pPq. The growth rate of the volume estimates is then a function of the
largest root of the lowest common multiple of the corresponding local Sato–Bernstein
polynomials, and of its multiplicity.

Proof. The proof of the statement is sketched as it follows upon repeating various
calculations done previously.

Let δP pz0 q ą 0 be chosen so that the local Sato–Bernstein polynomial BP,z0 psq exists
over the complex ball of radius δP pz0 q centered at z0 and such that the local zeta
distribution ζpP,z0 is well-defined when tested against smooth functions supported in
this same ball.

As in Lemma 2.5, the relation (3.10) is a consequence of Chambert–Loir’s and

Tschinkel’s Tauberian Theorem [21, Appendix A, Theorem A.1] ş applied in the follo-
wing form : define a measure µψz0 in C by setting µψz0 pAq “ A ψz0 pz, zq ¨ dz ^ dz for
n

A A Ă C . EThe assumption that ψz0 pz0 q ą 0 implies that the meromorphic

any Borel set n

map s ÞÑ ζpP,z psq, ψ achieves the pair C-LCTP,z (the argument follows from the
0 0
complex analogue of the point (W1) in Lemma 2.15, which analogue is a direct
consequence of Lemma 3.2). The aforementioned Tauberian Theorem then yields
that the relation (3.10) holds provided that the assumptions (a) and (b) stated in the
proof of Lemma 2.5 are satisfied for the local complex zeta function ζpP,z0 . This is
indeed easily verified in the same way as in this proof upon working with the complex
 chapter 3. log–canonical threshold and the sato–bernstein polynomial 71

analogue (3.1) of the Sato–Bernstein relation.

This is enough to establish Theorem 3.6 in view of Proposition 3.3, which states
that the largest root of the Sato–Bernstein polynomial BP,z0 psq is the opposite of the
smallest real pole of ζpP,z0 , the order and the multiplicity coinciding.
chapter 3. log–canonical threshold and the sato–bernstein polynomial 72
4 COUNTING SOLUTIONS TO GENERIC
HOMOGENEOUS FORMS
INEQUALITIES
The goal in this chapter is to establish Theorems 1.4 and 1.5. To this end, given an
integer n ě 2, denote by µn the Haar probability measure on the space Xn of unimo-
vdular lattices in Rn identified with the homogeneous space SLn pRq{SLn pZq through
the map g P SLn pRq{SLn pZq ÞÑ g ¨ Zn P Xn . Recall that Voln denotes the n–dimensional
Lebesgue measure. The proofs rely on the sharp volume estimates obtained in Chap-
ter 2 and on the following lemma.

Lemma 4.1 (Athreya & Margulis, 2009). Let A Ă Rn be a Lebesgue measurable set such
that Voln pAq ą 0. Then,
cn
µn ptΛ P Xn : Λ X A “ Huq ď
Voln pAq
for some constant cn ą 0 depending only on the dimension n ě 2.

Proof. See [3, Theorem 2.2].

For the sake of simplicity of notation, given real numbers a, b, set

SF pa, bq “ SF pBn , In , a, bq ,

where Bn denotes the unit Eulidean ball and where the set on the right-hand side is
defined in (1.1). When a “ T is a (large) parameter and when b “ apT q is a function of
this parameter, this is also the set SF pBn , bpT qq defined in (1.7).

Proof of Theorem 1.4. Given reals T , b ą 0, from the homogeneity of degree d of the
forms composing the p–tuple Fpxq, it holds that

Voln SF pT , bq “ Voln ptx P Bn pT q : 0 ă }F pxq} ă buq

` ˘
ż
“ χt}Fpxq}ăbu ¨ χt}x}ďT u ¨ dx
Rn ż

“ bn{d ¨ χt}Fpxq}ă1u ¨ χt}x}ďb´1{d ¨T u ¨ dx

Rn ´ ´ ¯¯
“ b n{d
¨ Voln SF b ´1{d
T, 1 . (4.1)

Thus, defining for j ě 0 the sets

! ´ ¯ )
AF pjq “ x P Bn fp2j q : 0 ă }Fpxq} ă 2´j ,

73
chapter 4. counting solutions to generic homogeneous forms inequalities 74

Lemma 4.1 implies that

2jr
µn ptΛ P Xn : Λ X AF pjq “ Huq ! ˇ ´ ¯ˇm´1 ¨
(1.13) & (4.1) j n´rd ˇ j d ´j
f p2 q ¨ ˇlog f p2 q ¨ 2 ˇ
ˇ

By assumption, the right–hand side is the general term of a convergent series. As a

consequence, the Borel–Cantelli Lemma yields that for µn –almost every Λ P Xn , there
exists an integer jpΛq ě 0 such that for all j ě jpΛq, one has that Λ X AF pjq ‰ H. Since
the probability measure µn descends from the Haar measure on SLn pRq, this is easily
seen to imply the existence, for almost all g P SLn pRq, of an integer jg ě 1 and of a
vector m P Zn z t0u such that for all j ě jg ,
´ ¯
}pF ˝ gq pmq} ă 2 ´j
and }g ¨ m} ď f 2j . (4.2)
` ˘
Fix ε P 0, 2´jg `1 and let j ě jg be the integer such that 2´j ď ε ă 2´j`1 . From
the growth condition (1.20) and the assumption that the function f is non–decreasing,
one infers the existence of an integer jf ě 0 and of a constant κf ě 1 such that for all
j ě j pg, fq “ max tjf , jg u, the inequalities (4.2) yield that
´ ¯
´1
}pF ˝ gq pmq} ă ε and }g ¨ m} ď κf ¨ f ε .

Define κg´1 ą 0 as the operator norm of the linear map determined by g´1 in such a
´ ¯´1
way that }m} ď κg´1 ¨ }g ¨ m}. Then, upon setting κF pg, fq “ κf ¨ κg´1 , one obtains
´ ¯
´1
}pF ˝ gq pmq} ă ε and }m} ď κF pg, fq ¨ f ε .

The proof is then complete upon defining the quantity εF pg, fq in the statement of
Theorem 1.4 as εF pg, fq “ 2´jpg,fq`1 .

As for Theorem 1.5, it is a particular case of the following more general result. To
state it, given a real number x, set x` “ max tx, 0u ě 0 and x´ “ max t´x, 0u ě 0 so
that x “ x` ´ x´ and |x| “ x` ` x´ .

Theorem 4.2. Let Ppxq P Rrxs be a homogeneous polynomial of degree q ě 1 in n ě 3

variables. Given real numbers a ă b, a parameter T ě 1 and a matrix g P SLn pRq, set

N
r P pg, a, b, T q “ # tm P Zn X Bn pT q : a ă pP ˝ gq pmq ď bu .
chapter 4. counting solutions to generic homogeneous forms inequalities 75

Assume that whenever b˘ ą a˘ , it holds that n ą q ¨ r˘ P , where the quantities rP are

˘

defined in the equation (2.40). Then, for almost all g P SLn pRq, there exists exponents η˘ P ą0
depending only on the polynomial Ppxq such that, as T tends to infinity,
˜ ˜ ˆ ˙ ` ¸¸ ˆ ˙ ` ˇ ˆ ˙ˇm` ´1
N
r P pg, a, b, T q
` ` b`
´ηP
b` rP ˇˇ b` ˇˇ P
“ γ P,g ` O fP ¨ ¨ log
Tn Tq Tq ˇ Tq ˇ
ˆ ˆ ´ ¯ ` ˙˙ ´ ¯ ` ˇ ´ a ¯ˇm` ´1
` ` a`
´ηP a` rP ˇ ` ˇ P
´ γP,g ` O fP q
¨ q
¨ ˇlog
Tq
ˇ
T T
˜ ˜ ˆ ˙ ´ ¸¸ ˆ ˙ ´ ˇ ˆ ˙ˇm´ ´1
´ηP rP ˇ
b ´ b ´ ˇlog b´ ˇ
ˇ P
´ γ´ P,g ` O f´
P ¨ ¨
Tq Tq ˇ Tq ˇ
ˆ ˆ ´ ¯ ´ ˙˙ ´ ¯ ` ˇ ´ a ¯ˇm´ ´1
a´ ´ηP a´ rP ˇ ´ ˇ P
` γP,g ` O f´
´
P q
¨ q
¨ log . (4.3)
Tq
ˇ ˇ
T T

In this relation, the functions f˘ ˘

P and the constants γP,g ě 0 are those defined in Corollary 2.10.
Furthermore, the implicit constants are allowed to depend on g.
Admissible ranges for the exponents η˘P ą 0 can easily be derived from the proof
of the theorem. They have not been recorded here as they depend on a rather tedious
distinction of cases depending on the form of the maps f˘
P and on the signs of the real
numbers a and b.

Deduction of Theorem 1.5 from Theorem 4.2. To see why the above statement implies The-
orem 1.5, take Ppxq “ PF pxq “ }Fpxq}2 . Then, the volume estimate (2.42) in Corol-
lary 2.10 reduces to an estimate for the quantity Voln pSF pBn , bpT qqq when q “´ 2d and is ¯
` ˘
only relevant when all signs are positive. Furthermore, the pair r` P , m `
P “ r`
PF , m `
PF
therein becomes the real log-canonical threshold, defined in (2.73), of the polynomial
PF pxq. This pair also equals prF {2, mF q, where rF ą 0 is the smallest real pole of the
distribution ζF defined in (1.9) and where mF ě 1 is its order. From Theorem 1.1, the
pair prF , mF q is therefore the one appearing in the volume estimate (1.13). To obtain
Theorem 1.5, it then remains to notice that, for this choice of the homogeneous form
PF pxq, all quantities appearing with a negative sign in the above Theorem 4.2 can be
taken as zero.
Proof of Theorem 4.2. The proof relies on the refinement by Kelmer and Yu [53, Theo-
rem 1] of Theorem 1.2 by Athreya and Margulis in [4]. The latter is concerned with
the case of unimodular twists of indefinite quadratic forms satisfying the volume
estimate (1.17) and the former with unimodular twists of the diagonal homogeneous
forms defined in (1.19). The additional — and essential – ingredient introduced here
to deal with the most general case of any homogeneous form is the sharp volume
estimate obtained in Corollary 2.10, the form of which requires a much finer analysis
of the error terms.
chapter 4. counting solutions to generic homogeneous forms inequalities 76

Given g P SLn pRq, set as in (1.15)

´ ¯
SP pg, a, b, T q “ pP ˝ gq´1 pra, bsq X Bn pT q “ tx P Bn pT q : a ă pP ˝ gq pxq ď bu
r

and note that

´ ¯
N SP pg, a, b, T q X Zn
r P pg, a, b, T q “ # r
´´ ¯ ¯
“ # P´1 ppa, bsq X pg ¨ Bn pT qq X pg ¨ Zn q
“ # tm P g ¨ Zn X g ¨ Bn pT q : a ă Ppxq ď bu .

It is enough to establish Theorem 4.2 when the transformation g is assumed to lie in

any given compact subset B of SLn pRq. Let
! ˇ ´ ¯ˇ )
BP,B pa, b, T q “ g P B : ˇNP pg, a, b, T q ´ Voln SP pg, a, b, T q ˇ ě kP pg, a, b, T q ,
ˇr r ˇ

where T ÞÑ kP pg, a, b, T q is some increasing function to be defined. Throughout this

proof, all implicit multiplicative constants are allowed to depend on B.

Given an integer j ě 1, let Oj be the p1{jq–neighbourhood of the identity in SLn pRq

induced by the operator norm (with respect to the Euclidean norm). It is proved in [53,
Lemma 2.1] that there exists a finite set Ij Ă K with cardinality

#Ij ! jpn`2qpn´1q{2 (4.4)

h ¨ Oj . Fixing j ě 1 and h P Ij , let

Ť ` ˘
such that B Ă hPIj
´ ¯ č
p1q
SP ph, a, b, jq “ pP ˝ hq ppa, bsq
r ´1
Bn px1 pj, σqq

and ´ ¯ č
p2q
SP ph, a, b, jq “
r pP ˝ hq´1 ppa, bsq Bn px2 pj, σqq ,
where σ ą 0 is a parameter to be adjusted later and where
ˆ ˙ ˆ ˙
1 1
x1 pj, σq “ 1 ´ ¨j σ
and x2 pj, σq “ 1 ` ¨ pj ` 1qσ .
j j
Thus, whenever jσ ď T ă pj ` 1qσ and g P h ¨ Oj , it holds that
´ ¯
p1q p2q
rSP ph, a, b, jq Ă P´1 ppa, bsq X pg ¨ Bn pT qq Ă r SP ph, a, b, jq. (4.5)

Denote by µr n the Haar measure on SLn pRq. As detailed in [4, p.12], it follows from
Roger’s second moment formula for the Siegel transform combined with Chebyshev’s
inequality that, whenever n ě 3, given any measurable set A Ă Rn and any real
parameter M ě 1, it holds that
Voln pAq
µ
r n pMB pA, Mqq ! , (4.6)
M2
chapter 4. counting solutions to generic homogeneous forms inequalities 77

where
MB pA, Mq “ tg P B : |# ppg ¨ Zn q X Aq ´ Voln pAq| ą Mu
and where the implicit constant depends only on the dimension n. Moreover, [53,
Theorem 6] establishes that for any j ě 1,

ď 2
ďď ´ ¯
plq
BP,B pa, b, T q Ă SP ph, a, b, jq, k
M r rP ph, a, b, jq , (4.7)
jσ ďT ăpj`1qσ hPIj l“1

where
´´ ¯I´ ¯¯
p2q p1q
rP ph, a, b, jq “ kP ph, a, b, jσ q ´ Voln
k SP ph, a, b, jq
r rSP ph, a, b, jq . (4.8)

Keeping the notations of Corollary 2.10, an elementary change of variables as in the

proof of Theorem 1.4 (see the equation (4.1)) yields that
´ ¯
p1q
Voln r SP ph, a, b, jq
“
x1 pj, σqn
˜ ˜ˆ ˆ ˙˙´δ` ¸¸ ˜ˆ ˙r` ˇ ˆ ˙ˇm` ´1 ¸
b `
P b `
P ˇ b ` ˇ P
γ`P,h ` O f`
P ¨ ¨ ˇlog ˇ
x1 pj, σq x1 pj, σqq ˇ x1 pj, σqq ˇ
˜ ˜ˆ ˆ ˙˙´δ` ¸¸ ˜ˆ ˙r` ˇ ˆ ˙ˇm` ´1 ¸
a `
P a `
P ˇ a ` ˇ P
´ γ` P,h ` O f `
P ¨ ¨ ˇlog ˇ
x1 pj, σq x1 pj, σqq ˇ x1 pj, σqq ˇ
˜ ˜ˆ ˆ ˙˙´δ´ ¸¸ ˜ˆ ˙r´ ˇ ˆ ˙ˇm´ ´1 ¸
b ´
P b ´
P ˇ b ´ ˇ P
´ γ´ P,h ` O f´
P ¨ q
¨ ˇˇlog q
ˇ
x1 pj, σq x1 pj, σq x1 pj, σq ˇ
˜ ˜ˆ ˆ ˙˙´δ´ ¸¸ ˜ˆ ˙r´ ˇ ˆ ˙ˇm´ ´1 ¸
a ´
P a ´
P ˇ a ´ ˇ P
` γ´ P,h ` O f´P ¨ ¨ ˇlog ˇ ,
x1 pj, σq x1 pj, σqq ˇ x1 pj, σqq ˇ

where the implicit

´ constants ¯are uniform over the compact set B. The same formula
p2q
S ph, a, b, jq upon replacing x1 pj, σq with x2 pj, σq. As a consequence,
holds for Voln r P
a tedious but not difficult asymptotic expansion calculation leads one to
´´ ¯I´ ¯¯
p2q p1q
Voln r SP ph, a, b, jq SP ph, a, b, jq
r
ˆ ` ˙
r` ˘ ` ) σpn´q¨r` q
!
rP ` σ ´δP m` ´1
! b` ´ a` ¨ max j´1 , f`
P
P pj q ¨j P ¨ plog jq P

ˆ ´ ˙
r´ ˘ ´ ) σpn´q¨r´ q
!
rP ` σ ´δP m´ ´1
` b´ ´ a´ ¨ max j´1 , f´
P
P pj q ¨j P ¨ plog jq P . (4.9)

In this bound, the implicit constant may depend on a and on b.

The proof is now split into several cases :

chapter 4. counting solutions to generic homogeneous forms inequalities 78

r` r` r´ r´
‚ Case A : b`P ą a`P and b´P “ a´P , and either (i) : f` `
P pT q “ T or (ii) : fP pT q “ log T ;

r` r` r´ r´
‚ Case B : b`P “ a`P and b´P ą a´P , and either (i) : f´ ´
P pT q “ T or (ii) : fP pT q “ log T ;

r` r` r´ r´
‚ Case C : b`P ą a`P and b´P ą a´P .

It is enough to establish the sought estimate (4.3) in Case (A). Indeed, the proof
and the result in Case (B) then follow upon switching signs and Case (C) is obtained
by adding the appropriate estimates obtained in the first two cases.

r` r` r´ r´
Assume therefore that (A) : b`P ą a`P and b´P “ a´P , and also that (i) : f`
P pT q “ T
hold. Letting
σ ě 1{δ` P, (4.10)
one obtains
´´ ¯I´ ¯¯ ` `
! jσpn´q¨rP q´1 ¨ plog jqmP ´1 .
p2q p1q
Voln SP ph, a, b, jq
r rSP ph, a, b, jq

Then, the inequality (4.6) combined with the relations (4.4), (4.5) and (4.7) yields that
¨ ˛ ´ ¯
p2q
ď Voln SP ph, a, b, jq
r
µ
rn ˝ BP,B pa, b, T q‚ ! jpn`2qpn´1q{2 ¨ ´ ¯2 ,
σ
j ďT ăpj`1qσ kP ph, a, b, jq
r

where ´ ¯ `
`
Voln SP ph, a, b, jq ! jσpn´q¨rP q ¨ plog jqmP ´1 .
rp2q

In these inequalities, the implicit constants are, from Corollary 2.10, uniform over the
compact set B. Define kP ph, a, b, jq “ jγ for some exponent γ ą 0 satisfying the inequa-
lities
1
n ´ q ¨ r` P ´ σ ă γ ă n ´ q ¨ rP .
`
(4.11)

Then, the equation (4.8) implies that

¨ ˛
` `
ď jpn`2qpn´1q{2 ¨ jσpn´q¨rP q ¨ plog jqmP ´1
µ
rn ˝ BP,B pa, b, T q ! ´
‚ ´ ` `
¯¯2 ,
γσ σpn´q¨r q´1 m ´1
σ
j ďT ăpj`1q σ j `O j P ¨ plog jq P

` `
! jpn`2qpn´1q{2`σpn´q¨rP ´2γq ¨ plog jqmP ´1 .
(4.11)

This is the general term of a convergent series whenever

n ´ q ¨ r`
P npn ` 1q
γ ą ` ¨ (4.12)
2 4σ
chapter 4. counting solutions to generic homogeneous forms inequalities 79

Under the assumption that n ´ q ¨ r` P ą 0, taking into account the constraints (4.10),
(4.11) and (4.12), this occurs for any
ˆ ˙ " *
` 1 ` 1 npn ` 1q ` 4
γ P n ´ q ¨ rP ´ , n ´ q ¨ rP , where σ0 “ max ` , . (4.13)
σ0 δP 2 ¨ pn ´ q ¨ r`
Pq
From the Borel–Cantelli Lemma, the upper limit set formed by the family
pBP,B pa, b, T qqT ě1 as T tends to infinity has then zero µ r n –measure. In other words,
for almost all g P B,
ˇ ´ ¯ˇ
ˇNP pg, a, b, T q ´ Voln SP pg, a, b, T q ˇ ! T γ
ˇr r ˇ

with an implicit constant depending on the choice of g and on the reals a and b. To
complete the proof of Theorem 4.2 in Case A.(i), it is then enough to notice that the
above estimate and Corollary 2.10 imply that
ˇ ˜ˆ ˙ ` ˇ ˆ ˙ˇm` ´1 ´ ¯ ` ˇ ¸ˇ
rP ˇ rP ˇ ´ a ¯ˇm` ´1 ˇ
ˇ b ` b `
P a ` ` P
ˇNP pg, a, b, T q ´ γ`
ˇ
¨ Tn ¨ ¨ ˇˇlog ¨ ˇlog
ˇr ˇ ˇ
P,g q q
ˇ ´ q q ˇ ˇ
ˇ T T ˇ T T ˇ
ˇ ´ ¯ˇ
ď ˇN P pg, a, b, T q ´ Voln SP pg, a, b, T q ˇ `
ˇr r ˇ
ˇ ˜ˆ ˙ ` ˇ ˆ ˙ˇm` ´1 ´ ¯ ` ˇ ¸ ˇ
rP ˇ r m `
ˇ b` b` ˇ P a` P ˇ ´ a` ˇ P
¯ˇ ´1 ´ ¯ ˇ
´ Voln SP pg, a, b, T q ˇ
ˇ ` n
ˇ
ˇγP,g ¨ T ¨ ¨ ˇlog ¨ ˇlog
ˇ
q
ˇ
q
´ q q ˇ r
ˇ T T ˇ T T ˇ
` ` `
! T γ ` T n´q¨rP ´δP ¨ plog T q´1`mP . (4.14)
␣ (
Theorem 4.2 thus follows in this case upon choosing any η` `
P ă min n ´ q ¨ rP ´ γ, δP
`

when γ satisfies the conditions (4.13); that is, upon choosing any η`
P P p0, 1{σ0 q.

r` r`
Assume now that the remaining case holds, namely that (A) : b`P ą a`P and
r´ r´
b´P “ a´P , and also that (ii) : f`
P pT q “ log T . Then, the error term (4.9) becomes
´´ ¯I´ ¯¯ ` ` `
! jσpn´q¨rP q ¨ plog jqmP ´1´δP
p2q p1q
Voln SP ph, a, b, jq
r SP ph, a, b, jq
r

` ` `
for any choice of the parameter σ ą 0. Set kP ph, a, b, jq “ jn´q¨rP ¨ plog jqmP ´1´ηP for
some
0 ă η`P ă δP .
`
(4.15)
Then, as above, one obtains
¨ ˛
ď
µ
rn ˝ BP,B pa, b, T q‚
jσ ďT ăpj`1qσ
` `
jpn`2qpn´1q{2 ¨ jσpn´q¨rP q ¨ plog jqmP ´1
! ´
` ` `
´ ` ` `
¯¯2 ,
n´q¨r m ´1´η n´q¨r m ´1´δ
j P ¨ plog jq P P `O j P ¨ plog jq P P

` ` `
! jpn`2qpn´1q{2´σpn´q¨rP q ¨ plog jq´mP `1´2ηP .
chapter 4. counting solutions to generic homogeneous forms inequalities 80

Under the assumption that n ´ q ¨ r` P ą 0, this is the general term of a convergent

series for σ large enough. From the Borel–Cantelli Lemma, this yields that for almost
all g P B,
ˇ ´ ¯ˇ ` ` `
SP pg, a, bT q ˇ ! T n´q¨rP ¨ plog T qmP ´1´ηP
ˇNP pg, a, b, T q ´ Voln r
ˇr ˇ

with an implicit constant depending on the choice of g and on the reals a and b. The
proof is then concluded as in (4.14). The upshot is that Theorem 4.2 is obtained in this
case upon choosing any exponent η` P in the range determined by (4.15).

This completes the proof of Theorem 4.2.

5 DETERMINISTIC COUNTING IN LARGE
DOMAINS
Let K be a globally semianalytic subset of Rn meeting the assumptions pH1 q and
pH2 q stated in the Introduction. The goal of this section is to establish a more precise
version of Theorem 1.6 in the case that the domain S:F pK, T , αq defined in (1.27) is
"large enough" in the sense that 0 ă α ď 1. Indeed, when, moreover, the assumption
pH3 q holds (one then deals with what is referred to as Case (2) in Section 1.2 of the
Introduction), it is then possible to determine the precise asymptotic order for the
counting function NF: pK, T , αq defined in (1.28) when 0 ă α ă 1, and to also obtain an
upper bound for this function when α “ 1. When one assumes instead the stronger
conditon that the set K is star-shaped with respect to the origin and that it contains the
origin in its interior (one then deals with what is referred to as Case (3) in Section 1.2
of the Introduction), it becomes possible to get an asymptotic expansion for NF: pK, T , αq
when 0 ă α ă 1.

To state the main result proved in this chapter, let Ppxq P R rxs denote a homoge-
neous form of degree q ě 2. By analogy with the above notations, let

S:P pK, T , αq “ x P T ¨ K : |P pxq| ď T q´α

␣ (
(5.1)
and ´ ¯
NP: pK, T , αq “ # Zn X S:P pK, T , αq . (5.2)
Theorem 1.6 in the case where 0 ă α ď 1 is then a consequence of Case (1’) in
the statement below : it indeed suffices to apply it to the polynomial Ppxq “ }F pxq}2
after substituting the parameters p2d, 2αq for pq, αq in the above definition (5.1). The
resulting conclusion generalises that of Theorem 1.6 inasmuch as it is valid without
assuming the flatness condition (1.34).
Theorem 5.1 (Asymptotic behaviour of the counting function of homogeneous form
inequalities in large domains). Assume that the set K Ă Rn is globally semianalytic and
that it satisfies the assumptions pH1 q and pH2 q.

Then, the counting function NP: pK, T , αq can be estimated as follows :

(1’) – assume that the condition pH3 q holds. Then, as the parameter T tends to infinity, one has
that ´ ¯
NP: pK, T , αq — Voln S:P pK, T , αq
provided that α P p0, 1q. When α “ 1, it holds that
´ ¯
NP: pK, T , 1q ! Voln S:P pK, T , 1q . (5.3)

81
 chapter 5. deterministic counting in large domains 82

(2’) – assume the stronger condition that the set K is star-shaped with respect to the origin and
that it contains the origin in its interior. Then, provided that α P p0, 1q, one has that as
the parameter T tends to infinity,
ˆ ˆ ˙˙
1 ´ ¯
NP pK, T , αq “ 1 ` O
:
¨ Vol n S:
P pK, T , αq
Tδ

for some δ ą 0.

In this statement, the volumes of the sets S:P pK, T , αq (where 0 ă α ď 1) are
respectively given by Cases (2) and (3) of Theorem 1.1 stated in the Introduction.

The rest of this chapter is devoted to the proof of Theorem 5.1, starting with the
inequality (5.3).

Proof of the volume estimate (5.3). Assume that the conditions pH1 q ´ pH3 q are met. An
elementary packing argument, a variant of which appears in [70, Appendix 1], shows
that the sought estimate holds for any value of

α P r0, 1s. (5.4)

Indeed, fix η P p0, 1{4q and assume that m P Zn X S:P pK, T , αq. Let u P Rn be a vector
with Euclidean norm at most η. Consider then the Taylor expansion
q
ÿ 1 ´ ¯
P pm ` uq “ Ppmq ` ¨ pu ¨ ∇qk P pmq,
k“1
k!

where u ¨ ∇ is short-hand notation for the differential operator ni“1 ui ¨ Bi when

ř

u “ pu1 , . . . , un q and ∇ “ pB1 , . . . , Bn q, and where the power applied to it must be

understood as a formal multiplication process.
´ ¯
For any given integer 1 ď k ď q, the map x ÞÑ pu ¨ ∇qk P pxq is homogeneous of
degree q ´ k. As a consequence, upon decomposing the integer vector m as m “ T ¨ y,
where y P K, one obtains that

|P pm ` uq| ď T q´α ` η ¨ cP pKq ¨ T q´1 ď p1 ` η ¨ cP pKqq ¨ T q´α

(5.4)

with a constant cP pKq ą 0 determined by the sup norms of the polynomial Ppxq and of
its partial derivatives over the set K. This shows that the pairwise disjoint Euclidean
balls with`radius η centered at the integer points in S:P pK, T , αq are contained in, say,
the set SP Kpηq , p1 ` η ¨ cP pKqq ¨ T q´α defined explicitly in (2.8). Here, Kpηq denotes the
˘

η-neighbourhood of the set K. Thus,

´ ´ ¯¯
NP pK, T , αq ď Voln SP K , p1 ` η ¨ cP pKqq ¨ T
: pηq q´α
. (5.5)
 smooth counting 83

If η ą 0 is chosen small enough, it is immediate that the set Kpηq also satisfies the
assumption pH3 q with the same compact set C and the same open set U as those
introduced in the support restriction condition
` ` (1.10)˘ for K. Consequently,
` pηq ˘˘ with the no-
tations of pH3 q, one deduces that the pair rP K pηq , mP K equals prP pKq, mP pKqq.
It then follows from Case (2) in Theorem 1.1 that
´ ´ ¯¯ ´ ¯
Voln SP Kpηq , p1 ` η ¨ cP pKqq ¨ T q´α ! Voln SP K, T q´α “ Voln S:P pK, T , αq .
` ` ˘˘

In view of the inequality (5.5), this completes the proof.

Establishing the remaining statements part of Theorem 5.1 requires more work. To
do so, assume from now on and until the end of the chapter that α P p0, 1q.

5.1 smooth counting

It is convenient to work in the first place with a smooth version of the counting
function NP: pK, T , αq. To this end, fix a smooth map ψ ě 0 in C8 n
c pR q whose support
will be specified later to approximate suitably the set K. The proof of the remaining
cases in Theorem 5.1 requires to also approximate the characteristic function of an
interval by a function whose Fourier transform behaves "nicely". This is a well–studied
topic where the extremal functions introduced by Beurling and Selberg are often used
(see the discussion and the references in [27] for more details). In the present situa-
tion however, it is more relevant to work with the map obtained from the following
(particular case of a) result due to Colzani, Gigante and Travaglini in [27, Theorem 1.1].

Before stating it, recall that a function θ : R` Ñ R` has fast decay at infinity if for
any β ą 0, there exists a constant cβ ą 0 such that for all t ě 0, it holds that
cβ
0 ď θptq ď ¨ (5.6)
p1 ` tqβ
Also, a holomorphic function F is said to be of exponential type R ą 0 if
log |Fpzq|
R “ lim sup max ¨
rÑ8 |z|“r |z|
Equivalently, this is saying that for any ε ą 0, there exists a constant κpεq ą 0 such that
|Fpzq| ď κpεq ¨ epR`εq¨|z| .
Theorem 5.2 (Colzani, Gigante & Travaglini, 2011). Let c P p0, 1q. Set I pcq “ r´c, cs.
Then, there exists a positive function θ with fast decay at infinity such that for any R ą 0, there
exist integrable entire functions ApR,cq and BpR,cq satisfying the following properties : these two
functions are of exponential type R and have their Fourier transforms supported in the interval
r´R, Rs; furthermore, for all x P R,
ApR,cq pxq ď χIpcq pxq ď BpR,cq pxq (5.7)
 smooth counting 84

and
0 ď BpR,cq pxq ´ ApR,cq pxq ď θ pR ¨ ||x| ´ c|q . (5.8)

ˇ The main point ˇof this statement is that it allows for an easy control of the gap
ˇBpR,cq pxq ´ ApR,cq pxqˇ by the distance from x P R to the boundary of the interval I pcq :
ˇ ˇ
roughly speaking, the quantity ˇBpR,cq pxq ´ ApR,cq pxqˇ is approximately equal to 1 at
points x within distance 1{R of the interval Ipcq and essentially zero at larger distances.

Let T ě 1. The proof of Theorem 5.1 mainly reduces to the estimate for Nr P pψ, T , αq
obtained in Proposition 5.3 below, which takes into account the support restriction
condition (1.10) induced by the assumption pH3 q. To state it, fix a compact set K
contained in the open set U and containing the compact set K in its interior, where the
set U is defined as part of the condition (1.10). Assume then that the map ψ is such
that
C Ă Supp ψ Ă K, 0 ď ψ ď 1 and max ψpxq “ 1, (5.9)
xPK
where the set C is also defined in (1.10). Take
T
c “ T ´α and R “ , where ∆P pKq “ max }p∇Pq pxq} ,
2 ¨ ∆P pKq xPK
(5.10)
and define EpT ,αq as being either of the maps ApT {p2∆P pKqq, T ´α q or BpT {p2∆P pKqq, T ´α q in
Theorem 5.2 (i.e. the maps obtained with the above choice for the parameters R and c).

Let then ÿ pP,ψq

N
r P pψ, T , αq “ GT ,α pkq , (5.11)
kPZn
where, given x P Rn ,
´x¯ ´ ´ x ¯¯
pP,ψq
GT ,α pxq “ ψ ¨ EpT ,αq P . (5.12)
T T
Given c P p0, 1q, set furthermore, as in (2.23),
ż
µP pψ, cq “ χt|Ppxq|ăcu ¨ ψpxq ¨ dx. (5.13)
Rn
It is useful to record right here the asymptotic behaviour at the origin of the map
c ą 0 ÞÑ µP pψ, cq. This can be inferred from Lemma 2.5 upon choosing, with the
notation therein, the compact set K as the one defined above and the collection of
maps F as the singleton tψu. Indeed, note first that the assumptions of Lemma 2.5
are met from the last claim in the point (X1) of Proposition 2.16. From this point
(X1) again and from the conclusions of the lemma, one then obtains the existence of
a rational ρ ą 0 and of an integer m ě 0, both independent of ψ as chosen above,
satisfying the estimate
µP pψ, cq “ ΘP pψq ` hψ pcq ¨ cρ ¨ |log c|m , where ˇhψ pcqˇ ď pm ` 1q ¨ pCN pψ, KqqM ¨ ηpcq .
` ˘ ˇ ˇ

(5.14)
In this relation,
 smooth counting 85

‚ the integers M, N ě 1 depend only on K and on the polynomial Ppxq;

‚ CN pψ, Kq is the constant defined in (2.26), namely

ˇ ˇ
ˇ B |k| ψ ˇ
CN pψ, Kq “ max max ˇ k pxqˇ . (5.15)
ˇ ˇ
0ď|k|ďN xPK ˇ Bx ˇ

The last relation in (5.9) implies that CN pψ, Kq ě 1;

‚ from the uniformity claim (V3) in Lemma 2.5,

0 ă ΘP pψq ď pCN pψ, KqqM ; (5.16)

‚ the quantity η pcq tends to 0 as c Ñ 0` uniformly in the choice of ψ meeting the re-
lations (5.9) under the assumption pH3 q. Indeed, the factor pm ` 1q ¨ pCN pψ, KqqM ¨
ηpcq in (5.14) is obtained upon adding upper bounds for two maps depending on
c:

– on the one hand the map

c ÞÑ ΘP pψq ¨ Rψ p|log c|q ´ |log c|m { |log c|m ,

` ˘
(5.17)

where Rψ is, with the present notations, the degree m monic polynomial ap-
pearing in (2.24); furthermore, from the uniformity claim (V3) in Lemma 2.5,
each of the m coefficients obtained when expanding this expression is, up to
a multiplicative constant depending on K, bounded above by pCN pψ, KqqM ;
– on the other hand, the map determined by the error term present in (2.24),
which is of the form O pcε q for some fixed ε ą 0. Here, from the uniformity
claim (V2) in Lemma 2.5 and under pH3 q, the implicit constant can be taken
up to a multiplicative constant depending on K and on ε, and uniformly in
any ψ meeting the conditions stated in (5.9), as CN pψ, Kq.

Upon adding the resulting form of the error term in the latter case with the
upper bound obtained for the map (5.17), this yields the required shape pm ` 1q ¨
pCN pψ, KqqM ¨ ηpcq with the sought uniformity claim on η.

Proposition 5.3. Assume that ψ is a map in C8 n

c pR q satisfying the relations (5.9) and ta-
king strictly positive values over the set C defined therein. Let α P p0, 1q. Then, under the
assumption of the homogeneity of the polynomial Ppxq, there exist integers M, N ě 1 such that
ˇ ˘ˇ
N n
ˇ ! CN pψ, KqM ¨ T n´δ ¨ µP ψ, T ´α
` ´α ˇ
` ˘
T , αq T µ ψ, T (5.18)
ˇr
ˇ P pψ, ´ ¨ P

for any δ P p0, 1 ´ αq. In this inequality, the constant CN pψ, Kq is the one defined in (5.15).
Furthermore, the implicit constant, the integers M and N and the choice of the exponent δ do
not depend on ψ.
 smooth counting 86

Deduction of Theorem 5.1 from Proposition 5.3 when 0 ă α ă 1. Several steps in this
deduction already appeared in Section 2.2 so that the argument is here sketched
highlighting the parts which are specific to this setup.

Consider first Case (1’) of Theorem 5.1. Fix 0 ă α ă 1 and let ψ1 , ψ2 be two maps
in C8 n
c pR q satisfying the conditions (5.9) and the inequalities

0 ď ψ1 ď χK ď ψ2 .

Under the assumption pH3 q, the relation (2.24) part of Lemma 2.5 and the expan-
sion (2.13) yield that
´ ¯
T n ¨ µP ψ1 , T ´α — Voln S:P pK, T , αq — T n ¨ µP ψ2 , T ´α .
` ˘ ` ˘

As a consequence, one infers from Proposition 5.3 that

´ ¯
Voln S:P pK, T , αq ! N
r P pψ1 , T , αq ď N: pK, T , αq
P
´ ¯
ď NP 2 , T , αq ! Voln SP pK, T , αq .
r pψ :

Here, the left-most inequality relies on the fact that the exponent δ in (5.18) is, by
assumption, strictly positive. This establishes the conclusion of Theorem 5.1 in
Case (1’) when 0 ă α ă 1.

As for Case (2’), it relies on the arguments developed in the proof of Theorem 2.8
(in particular in those present in the proofs of Lemmata 2.12 and 2.13). To see this, fix
k and ψk be smooth maps in Cc pR q pointwise monotonic as
an integer k ě 1. Let ψ´ ` 8 n

k tends to infinity such that

χKp1´1{kq ď ψ´ `
k ď χK ď ψk ď χKp1`1{kq (5.19)

(recall here that, given λ ą 0, one sets Kpλq “ λ ¨ K and that the inequalities between the
characteristic functions are indeed valid under the assumptions that K is star-shaped
with respect to the origin). Upon choosing k larger than some integer k0 , assume
furthermore that the maps ψ˘ k also meet the conditions stated ` in˘ (5.9).
˘ From [43, The-
orem 1.4.1, p.25], they can be chosen so that the constant CN ψk , K defined in (5.15)
is bounded above as
K ! kN
` ˘
CN ψ˘ k , (5.20)
for some implicit constant depending only on N ě 1. Given k ě k0 , set then

N
r ´ pT , αq “ N
` ˘
r P ψ´ , T , α when EpT ,αq “ ApT {p2∆P pKqq, T ´α q
P,k k

and
N
r ` pT , αq “ N
` ˘
r P ψ` , T , α when EpT ,αq “ BpT {p2∆P pKqq, T ´α q
P,k k
 smooth counting 87

so that
r ´ pT , αq ď N: pT , αq ď N
N P,k P
r ` pT , αq .
P,k (5.21)
To emphasize the analogy with the proof of Theorem 2.8, given c P p0, 1q, let further-
more ż
νP pK, cq “ χt|Ppxq|ďcu ¨ dx (5.22)
K
and ż
` ˘
ν˘
P pk, cq “ µP ψ˘
k ,c “ χt|Ppx|ďcu ¨ ψ˘
k pxq ¨ dx.
Rn
Then, for any k ě 1,
ˇ ´ ¯ˇ ˇ ˘ˇ
ˇNP,k pT , αq ´ Voln SP pK, T , αq ˇ “ ˇN
ˇr˘ : ˇr˘ n
` ´α ˇ
, αq T ν K, T
ˇ
P,k pT ´ ¨ P ˇ
ˇ ˘ˇ
ď ˇNP,k pT , αq ´ T ¨ νP k, T
n
ˇr˘ ˘
` ´α ˇ
ˇ
(5.19)
ˇ `
` T n ¨ ˇν` ´α
˘ ´
` ˘ˇ
´α ˇ
P k, T ´ νP k, T .
Since χK ď ψ` k0 , repeating the calculations done in the proof of Lemma 2.12, one
obtains that for any c ą 0 and any k ě k0 ,
`
ˇν pk, cq ´ ν´ pk, cqˇ ! νP pk0 , cq
ˇ ` ˇ
P P k
with an implicit constant independent of k and c. From Proposition 5.3, this implies
that
` ´α q ν` ´α q ˙
ˆ
M νP pk0 , T P pk0 , T
ˇ ´ ¯ˇ
ˇNP,k pT , αq ´ Voln SP pK, T , αq ˇ ! T ¨ CN pψk , Kq ¨
ˇr˘ : ˇ n ˘
`
Tδ k
(5.23)
for some δ ą 0 and some integers M, N ě 1. Here, the inclusions
Supp ψ`
k0 Ă Kp1 ` 1{k0 q Ă Kp2q yield that
´ ¯
` ˘ νP pK, 2´q ¨ T ´α q Voln S:P pK, 2q{α T , αq
ν`
P k0 , T
´α
ď “ ˘n
2n
`
(5.22) 2n ¨ 2q{α T
´ ¯
Voln S:P pK, T , αq ` ˘
! “ νP K, T ´α ,
Tn
where the second last relation is easily deduced from the explicit volume estimate
obtained in Case (3) of Theorem 1.1. It is then a consequence of the inequalities (5.20)
and (5.23) that
ˇ ´ ¯ˇ ´ ¯ ˆ kMN 1 ˙
ˇNP,k pT , αq ´ Voln SP pK, T , αq ˇ ! Voln SP pK, T , αq ¨
ˇr˘ : :
.
ˇ
`
Tδ k
In view of the inequalities (5.21), specialising k to be the integer part of T δ{p1`MNq
provides the sought power saving. This completes the deduction of Theorem 5.1 from
Proposition 5.3 when α P p0, 1q.
 non–stationary phase analysis and estimate of the error term 88

The proof of Proposition 5.3 relies on the Poisson summation formula, which im-
plies that the counting function (5.11) can be expanded as
ÿ pP,ψq
Nr P pψ, T , αq “ G
p
T ,α pkq . (5.24)
kPZn

pP,ψq pP,ψq
Here, G p
T ,α denotes the Fourier transform of the map GT ,α defined in (5.12). Explici-
tly, an elementary change of variables shows that, given ξ P Rn ,

p pP,ψq pξq “ T n ¨ H
G p pP,ψq pξq , (5.25)
T ,α T ,α

where
ż
p pP,ψq pξq “
H e p´T ξ ¨ xq ¨ EpT ,αq pP pxqq ¨ ψ pxq ¨ dx. (5.26)
T ,α
Rn

The term G p pP,ψq p0q obtained when k “ 0 is referred to as the leading term in the
T ,α
sum (5.24), and the remaining sum as the error term. These two quantities are analysed
separately, starting with the latter.

5.2 non–stationary phase analysis and estimate of

the error term
The following claim essentially establishes that the error term has fast decay in the
variable T .

Lemma 5.4 (Asymptotic behavior of the error term). Let α P p0, 1q and η P p0, 1q. Fix an
integer j ě 1 such that jη ą n. Assume that ψ is a smooth map supported in the compact set
K introduced in (5.9) and that T is a parameter larger than 21{p1´ηq . Then,
ˇ ˇ
ˇ ˇ
ˇÿ ˇ
pP,ψq
GT ,α pkqˇ ! Cj pψ, Kq ¨ T n´jη ¨ logpT q,
ˇ p ˇ
ˇ
ˇ ˇ
ˇ k‰0n ˇ
kPZ

where the constant Cj pψ, Kq is defined in (5.15) and where the implicit constant does not
depend on the smooth map ψ.

The proof of this statement relies on the following effective non-stationary phase
estimate which can be found in [43, Theorem 7.7.1].

Proposition 5.5 (Effective non–stationary phase estimate). Let j ě 1 be an integer and

let ψ be in Cjc pRn q with support contained in a compact set K. Assume that the real–valued
 non–stationary phase analysis and estimate of the error term 89

function f is j ` 1 times continuously differentiable in an open neighbourhood V of K . Then,

for any real parameter λ ą 0, it holds that
Cj pψ, Kq
ˇż ˇ
ˇ ˇ
ˇ n e pλfpxqq ¨ ψpxq ¨ dxˇ ! λj ¨ δ pfqj ¨ min ␣1, δ pfqj (
ˇ ˇ
R ψ ψ

assuming that the quantity

δψ pfq “ inf }∇f pxq}
xPK
should be strictly positive. In the above inequality, the implicit constant is independent from ψ
and λ; furthermore, it remains bounded as long as the function f remains in a bounded subset
of the set of j ` 1 times continuously differentiable functions on V.
Proof of Lemma 5.4. Fix k ‰ 0 in Zn . Then, by Fourier inversion,
ż T {p2∆P pKqq ˆż ˙
pP,ψq
H
p
T ,α pkq “ E
ppT ,αq ptq ¨ e ptPpxq ´ T k ¨ xq ¨ ψpxq ¨ dx ¨ dt, (5.27)
´T {p2∆P pKqq Rn

where the real ∆P pKq is defined in (5.10) and Hp pP,ψq pkq in (5.26). Let η P p0, 1q and
T ,α
T ě 21{p1´ηq be as in the assumption. Fix t P r´T {p2∆P pKqq, T {p2∆P pKqqs. Then, given
x P Supp ψ Ă K,
}t ¨ ∇Ppxq ´ T k} ě T ¨ }k} ´ |t| ¨ }∇Ppxq} ě T ¨ }k} ´ |t| ¨ ∆P pKq. (5.28)
(5.10)

As a consequence, the inequality }t ¨ ∇Ppxq ´ T k} ě pT ¨ }k}qη is met as soon as

}k}η 1
}k} ě ` ¨
T 1´η 2
In turn, under the assumption that T ě 21{p1´ηq , this condition is fulfilled for all k P
Zn z t0u. Proposition 5.5 (applied with λ “ 1) then implies that for any integer j ě 1,

ˇ ! Cj pψ, Kq ¨
ˇż ˇ
ˇ ˇ
ˇ e ptPpxq ´ T k ¨ xq ¨ ψpxq ¨ dx (5.29)
R pT ¨ }k}qjη
ˇ n ˇ

Under the assumption jη ą n, one thus obtains that

ˇ ˇ
ˇ ˇ
ˇÿ ˇ ÿ ˇˇ pP,ψq ˇ
pP,ψq n
GT ,α pkqˇ ď T ¨ ˇHT ,α pkqˇ
ˇ p ˇ p ˇ
ˇ
ˇ ˇ (5.25)
ˇ k‰0n ˇ k‰0
kPZ kPZn
¨ ˛
˚ÿ
Cj pψ, Kq ¨ T n´jη ¨ ˝ }k}´jη ‚
‹
!
(5.27) & (5.29)
k‰0
kPZn
˜ż ¸
ˇ ˇ
ˇEpT ,αq ptqˇ ¨ dt
ˇp ˇ
ˆ
|t|ďT {p2∆P pKqq
˜ż ¸
ˇ ˇ
! Cj pψ, Kq ¨ T n´jη ¨ ˇEpT ,αq ptqˇ ¨ dt . (5.30)
ˇp ˇ
|t|ďT {p2∆P pKqq
 analysis of the leading term 90

To evaluate this last factor, let t be any real lying in the interval
r´T {p2∆P pKqq, T {p2∆P pKqqs. From the elementary estimate
ˇ ˇ " *
ˇ ˇ ˇ sin p2πT ´α tq ˇ ´α 1
ˇχpIpT ´α q ptqˇ “ ˇˇ ˇ ď min T , ,
πt ˇ |πt|

one deduces that

ˇ ˇ ˇ ˇ ˇˇ´ ¯ ˇ
ˇEpT ,αq ptqˇ ď χ pIpT ´α q ptq ` ˇ EpT ,αq ´ χ
ˇp ˇ ˇ
ˇ ˇ p pIpT ´α q ptqˇ
ż
ˇ ˇ ˇ ˇ
ď χ ˇ pIpT ´α q ptq `
ˇ ˇEpT ,αq pxq ´ χIpT ´α q pxqˇ ¨ dx
" R * ż ˆ ˙
1 T ˇ ˇ
ď min 2, ` θ ¨ ˇ|x| ´ T ´α ˇ ¨ dx
(5.7) & (5.8) |t| R 2∆P pKq
" *
1 1
! min 2, `
|t| T

with an implicit constant depending on the integral of θ over its domain of definition.
Inserting this estimate in (5.30) then concludes the proof.

5.3 analysis of the leading term

p pP,ψq p0q in the sum (5.24) reduces to the following
The estimation of the leading term G T ,α
claim :

Lemma 5.6 (Asymptotic behavior of the leading term). Keep the assumptions and the
notations of Proposition 5.3, assuming in particular that α P p0, 1q. Then, there exist integers
M, N ě 1 such that for any δ P p0, 1 ´ αq, it holds that
´α q
M µP pψ, T
ˇ ˘ˇˇ
ˇ p pP,ψq
ˇHT ,α p0q ´ µP ψ, T ´α ˇ ! pCN pψ, Kqq ¨
`
¨
Tδ
p pP,ψq p0q is defined in (5.26), the map µP pψ, ¨ q in (5.13) and
In this inequality, the quantity H T ,α
the constant CN pψ, Kq in (5.15). Also, the implicit constant is independent of the smooth map
ψ.

Proof. It follows from Theorem 5.2 that

ˇ ˘ˇ
ˇ p pP,ψq ` ´α ˇ
ˇHT ,α p0q ´ µP ψ, T ˇ
ż ˇ ˇ
ˇBpT {p2∆P pKqq, T ´α q pPpxqq ´ ApT {p2∆P pKqq, T ´α q pPpxqqˇ ¨ ψ pxq ¨ dx
ˇ ˇ
ď
żR ˆ
n
˙
T ˇ ˇ
´α ˇ
ď θ ¨ |Ppxq| ´ T
ˇ ¨ ψ pxq ¨ dx, (5.31)
Rn 2∆P pKq
 analysis of the leading term 91

where the constant ∆P pKq is defined in (5.10).

Let δ ą 0 stand for some small exponent, the range of which is to be determined.
Define the set
" *
ˇ 2 ¨ ∆P pKq
SP pT , αq “ x P R : ˇ|Ppxq| ´ T ˇ ď
n
ˇ ´α
T 1´δ

and decompose the integral in (5.31) as

ż ˆ ˙
T ˇ ´α ˇ
ˇ
θ ¨ |Ppxq| ´ T
ˇ ¨ ψ pxq ¨ dx
Rn 2∆P pKq
ż ˆ ˙
T ˇ ´α
ˇ
“ χSP pT ,αq pxq ¨ θ ¨ ˇ|Ppxq| ´ T ˇ ¨ ψ pxq ¨ dx
Rn 2∆P pKq
ż ˆ ˙
` ˘ T ˇ ´α
ˇ
` 1 ´ χSP pT ,αq pxq ¨ θ ¨ ˇ|Ppxq| ´ T ˇ ¨ ψ pxq ¨ dx. (5.32)
Rn 2∆P pKq

The first integral on the right–hand side of this equation can be bounded as follows :
ż ˆ ˙
T ˇ ˇ
´α ˇ
χSP pT ,αq pxq ¨ θ ¨ |Ppxq| ´ T
ˇ ¨ ψ pxq ¨ dx
Rn 2∆P pKq
ˆ ˙ ż
ď max θpxq ¨ χSP pT ,αq pxq ¨ ψ pxq ¨ dx. (5.33)
xě0 Rn

Introducing the map µP pψ, ¨ q defined in (5.13), this last integral becomes
ż
χSP pT ,αq pxq ¨ ψ pxq ¨ dx
Rn ˆ ˙ ˆ ˙
1 2∆P pKq 1 2∆P pKq
“ µP ψ, α ` ´ µP ψ, α ´
T T 1´δ T T 1´δ
1 plog T qm
! pCN pψ, KqqM ¨ 1´δ ¨ αpρ´1q (5.34)
(5.14) & (5.16) T T

with an implicit constant depending on the parameters α, ρ, m and on the polynomial

Ppxq. Here, the inequality (5.34) holds as long as

0 ă δ ă 1´α (5.35)

since this condition guarantees that T ´1`δ “ o pT ´α q as T tends to infinity. Under this
assumption, the upper bound (5.33) implies that
ż ˆ ˙
T ˇ ˇ
´α ˇ
χSP pT ,αq pxq ¨ θ ¨ |Ppxq| ´ T
ˇ ¨ ψ pxq ¨ dx
Rn 2∆P pKq
plog T qm
! pCN pψ, KqqM ¨ 1´δ`αpρ´1q ¨ (5.36)
T
 analysis of the leading term 92

As for the second term on the right–hand side of the equation (5.32), one infers from
the definition of the set SP pT , αq and from the fast decay of the function θ that for any
large β ą 0,
ż ˆ ˙
` ˘ T ˇ ˇ
1 ´ χSP pT ,αq pxq ¨ θ ¨ ˇ|Ppxq| ´ T ˇ ¨ ψ pxq ¨ dx ! T ´βδ
´α
(5.37)
R n 2∆ P pKq

with an implicit constant depending only on the choice of β and on the volume of the
compact set K containing the support of the map ψ (recall that this map is assumed
to meet the conditions stated in (5.9)).

Under the restriction (5.35), one can choose β ą 0 large enough so that inequali-
ties (5.36) and (5.37) imply that
ˇ
ˇ p pP,ψq
ˇ
α ˇ pCN pψ, KqqM
ˇHT ,α p0q ´ µP pψ, T qˇ ! 1´δ`αpρ´1q´η ¨
T
Since the restriction (5.35) also guarantees that 1 ` αpρ ´ 1q ´ δ ą αρ when α ă 1, in
view of the asymptotic expansion (5.14), this provides the required power saving in
the error term upon choosing η ą 0 small enough. This completes the proof of the
lemma.

Completion of the proof of Proposition 5.3. In view of the Poisson summation for-
mula (5.24), the sought estimate (5.18) is an immediate consequence of Lemmata 5.4
and 5.6 when α P p0, 1q.
6 DETERMINISTIC COUNTING IN THIN
DOMAINS

Throughout this final chapter, let α ą 1 be a real and let Fpxq “ pF1 pxq, . . . , Fp pxqq be a
p-tuple of homogeneous forms in n ě 2 variables, each of degree d ě 2. Assume that
the set K Ă Rn meets the conditions pH1 q, pH2 q and pH3 q stated in the Introduction
(Chapter 1) and that it is furthermore semialgebraic (recall that this means that it is
given as a finite union of sets that can be defined as finitely many equalities and
inequalities involving polynomial maps). Given a large parameter T ě 1, recall also the
definitions of the set S:F pK, T , αq and of the corresponding counting function NF: pK, T , αq
given in Chapter 1, namely
! )
S:F pK, T , αq “ x P T ¨ K : }Fpxq} ď T d´α (6.1)

and ´ ¯
NF: pK, T , αq “ # SF pK, T , αq X Z .
: n
(6.2)
The main goal of this final chapter is to establish Theorem 6.1 below, which comprises
Theorem 1.6 stated in the Introduction as a particular case when α ą 1 (the case α ď 1
being settled in the previous chapter).

To this end, first are recalled further notations and definitions introduced in Chap-
ter 1. When v P Sn´1 and σ P R, the set vK pσq denotes the affine hyperplane

vK pσq :“ tx P Rn : v ¨ x “ σu . (6.3)

Fixing a compact set K containing K in its interior and contained in the open set U
defined as part of the support restriction condition (1.10), let for any given ε ą 0

µF,K pv, σ, εq “ Voln´1 x P K X vK pσq : }Fpxq} ď ε

`␣ (˘

and
MF pK, εq “ sup sup µF,K pv, σ, εq . (6.4)
vPSn´1 σPR
The "biggest level of flatness" that can be achieved when intersecting the sublevel set
tx P K : |Fpxq| ď εu with affine hyperplanes is quantified by the value of the real
ˆ ˙
log MF pK, εq
qF pKq “ lim inf . (6.5)
εÑ0` log ε

93
 chapter 6. deterministic counting in thin domains 94

This is a well–defined real number under the assumption pH2 q.

The dimension dim pZK pFqq of the algebraic variety ZR pFq over the set K is denoted
by τF pKq and its codimension by τpF pKq, namely by τpF pKq “ n ´ dim pZK pFqq. Recall
that in the case that the variety ZK pFq has smooth complete intersection, it holds that
τpF pKq “ p.

The main result in this chapter can now be stated as follows :

Theorem 6.1. Let α ą 1 be a real. Assume that K is a semialgebraic set meeting the assump-
tions pH1 q, pH2 q and pH3 q stated in the Introduction. Let it be contained in the interior of a
compact set K Ă U, where the open set U is defined in the support restriction condition (1.10).
Also, let rF pKq be the nonnegative real introduced in the assumption pH3 q. Then, with the
above notations, the following claims are verified :

(Z1) The upper bound

´ ¯
NF: pK, T , αq ! Voln S:F pK, T , αq ` T τF pKq´δF pK,Kq (6.6)

holds for some exponent δF pK, Kq ą 0 depending on the polynomial Ppxq and on the
compact sets K Ă K if either of the following two mutually exclusive assumptions are
satisfied :

(i) in the generic case when

qF pKq ą rF pKq ´ 1, (6.7)
one requires that the measure of flatness qF pKq should be bounded below in this
sense :
τ pKq
qF pKq ą n ´ 1 ´ rF pKq ¨ F ¨ (6.8)
τpF pKq
One can then choose any exponent δF pK, Kq such that

n ¨ pn ´ 1 ´ qF pKqq
0 ă δF pK, Kq ă τF pKq ´ ¨ (6.9)
n ´ 1 ´ qF pKq ` rF pKq

(ii) in the degenerate case when

qF pKq ď rF pKq ´ 1, (6.10)

one requires that the measure of flatness qF pKq should be bounded below in this
sense :
qF pKq ą τpF pKq ´ 1. (6.11)
One can then choose any exponent δF pK, Kq such that

0 ă δF pK, Kq ă qF pKq ´ τpF pKq ` 1. (6.12)

 chapter 6. deterministic counting in thin domains 95

In the inequality (6.6), the volume term on the right–hand side is explicitly determined
by Case (2) in Theorem 1.1.

(Z2) In the case that that the algebraic variety ZR pFq is not contained in a linear subspace of
dimension n ´ p and that, furthermore, the set of homogeneous forms Fpxq has smooth
complete intersection over K (recall that this is understood in the sense that the map (1.6)
does not vanish over K), then the conditions (6.7) and (6.8) are both met. As a conse-
quence, an estimate of the form (6.6) indeed holds.

Claim (Z2) is established in Section 6.3.2 below. Most of the work in proving
the above statement deals with establishing Claim (Z1). To this end, it is enough to
consider the case of p “ 1 homogeneous form, say Ppxq, of degree denoted by q ě 1.
The general case stated in (Z1) then follows upon applying the results obtained in this
particular situation to the polynomial PF pxq “ }Fpxq}2 and upon squaring all relations
involving this polynomial. Explicitly, this is saying that it suffices to replace in what
follows the polynomial Ppxq with PF pxq, the real ε ą 0 with ε2 , the integer q with 2d
and the real α ą 1 with 2α. The details of this elementary verification are left to the
reader.

In view of this reduction, it is natural to establish (Z1) employing the notations

corresponding to the analogues in the case p “ 1 of the set S:F pK, T , αq and of the
counting function NF: pK, T , αq. These are respectively the set S:P pK, T , αq and the
counting function NP: pK, T , αq introduced in Chapter 5 — see the definitions (5.1)
and (5.2).

In order to establish the claim (Z1), let j ě 1 be an integer and let φ be in Cjc pRq and
ψ in Cjc pRn q such that

χr´1,1s ď φ ď χr´2,2s and χK ď ψ ď χK . (6.13)

As the proof relies on semialgebraic geometry, the maps φ and ψ are also assumed
to be semialgebraic (meaning that their graphs are semialgebraic sets). The existence
of such maps is guaranteed, for instance, by [52, Proposition 4.8].

Fix
ε ě T ´α . (6.14)
 non–stationary phase analysis and reduction to a finite sum 96

It then follows from the Poisson Summation formula and from the homogeneity of the
polynomial Ppxq that
ˆ ˙ ˆ ˙
ÿ k Ppkq
NP pK, T , αq ď
:
ψ ¨φ
n T ε ¨ Tq
kPZ
ÿ ż ´ ¯
n
“ T ¨ ψpxq ¨ φ ε´1 ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx
kPZn R
n
ż ´ ¯
n
“ T ¨ ψpxq ¨ φ ε´1 ¨ Ppxq ¨ dx
Rn
ÿ ż ´ ¯
n ´1
`T ¨ ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx. (6.15)
kPZn zt0u R
n

Upon specialising the real ε ą 0 and the maps φ and ψ, the integral in the former
term in this last expression is related in Section 6.3 to the volume of the set of points
x P K such that |Ppxq| ă T ´α . The analysis of the oscillatory integrals and of the
resulting sum in the latter term in (6.15) constitutes the main substance of the proof
and will be carried out along these lines : by a non–stationary phase analysis developed
in Section 6.1 below, the sum is first reduced to a finite one up to an error term with
rapid decay in the quantity }T k}, provided that the integer j ě 1 defining the regularity
of the maps φ and ψ is large enough. To deal with this finite sum, upon decomposing
in polar coordinates the nonzero vector T k as
´T k “ λv, where λą0 and v P Sn´1 , (6.16)
the goal is to obtain, under suitable assumptions, a uniform decay estimate for the
Fourier coefficient appearing in (6.15); namely an estimate of the form
εν
ż ´ ¯
´1
ψpxq ¨ φ ε ¨ Ppxq ¨ e pλv ¨ xq ¨ dx ! γ (6.17)
Rn λ
for some positive exponents γ and ν. The crucial requirement here, which prevents
the reduction of the problem to a (more or less standard) local harmonic analysis esti-
mate of oscillatory integrals, is that the implicit constant in the inequality above must
be independant of the choice of the unit vector v P Sn´1 . This issue is overcome in
Section 6.2 with the help of o–minimal theory. The sum appearing in (6.15), when
taken over finitely many nonzero integer vectors k, can then be estimated upon spe-
cialising λ to the value }T k} in (6.17) and upon optimising the value of ε under the
constraint (6.14). This will be seen to imply Theorem 6.1.

6.1 non–stationary phase analysis and reduction

to a finite sum
This section deals with the problem of reducing the infinite sum appearing in (6.15) to
a finite one through a non–stationary phase analysis.
 non–stationary phase analysis and reduction to a finite sum 97

Lemma 6.2. Let β ą 0 be a large parameter and let κ P p0, 1q be a small one. Fix T ě 2 and
ε P p0, 1{T q and assume that φ ě 0 is in Cjc pRq and ψ ě 0 in Cjc pRn q, where
n ` β ¨ p1 ´ κq
j ě 1` ¨ (6.18)
κ
Then, there exists a constant aP pψ, κq ą 0 depending on the parameter κ ą 0 and on the sup
norm of the polynomial Ppxq over the support of ψ such that
ÿ ż ´ ¯
´1
ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx
kPZn zt0u R
n
ˇż ´ ¯ ˇ
β
ÿ ˇ ˇ
´1
! pεT q ` ˇ ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dxˇˇ .
Rn
ˇ
1ď}k}ďaP pψ,κq¨pT εq´1{p1´κq

Here, the implicit constant depends on the parameters j, β, κ and on the weight functions ψ and
φ.
Proof. Let k P Zn zt0u be an integer vector and let L ě 1 be a real. By Fourier inversion,
ż ´ ¯
´1
ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx
Rn ˆż ˙
ż ´ ¯
´1
“ φ̂ptq ¨ ψpxq ¨ e tε ¨ Ppxq ´ T k ¨ x ¨ dx ¨ dt
|t|ďL Rn
loooooooooooooooooooooooooooooooooooooomoooooooooooooooooooooooooooooooooooooon
“I1 pLq
ż ˆż ´ ¯ ˙
´1
` φ̂ptq ¨ ψpxq ¨ e tε ¨ Ppxq ´ T k ¨ x ¨ dx ¨ dt .
|t|ąL R n
loooooooooooooooooooooooooooooooooooooomoooooooooooooooooooooooooooooooooooooon
“I2 pLq

Since φ is in Cjc pRq, `repeated

˘ integration by parts show that its Fourier transform
t ÞÑ φ̂ptq decays as O t ´j when t Ñ 8. As a consequence,
ˆż ˙ ˆż ˙
|I2 pLq| ď ψpxq ¨ dx ¨ |φ̂ptq| ¨ dt ! L´j`1 . (6.19)
Rn |t|ąL

As for the integral I1 pLq, it can be estimated with the help of the effective
non–stationary phase result stated in Proposition 5.5 (upon letting λ “ 1 and
fpxq “ tε´1 ¨ Ppxq ´ T k ¨ x therein). To this end, given t P r´L, Ls and x P Supp ψ, note
first that, upon setting ∆P pKq “ maxxPK }∇Ppxq}, it holds that
› ›
› ´1
›tε ¨ ∇Ppxq ´ T k› ě T }k} ´ Lε´1 ¨ ∆P pKq. (6.20)
›
(6.13)

Under the assumption that T ě 2, this quantity is larger than pT }k}qκ whenever
L ¨ ∆P pKq
}k} ě ` ˘¨ (6.21)
T ε ¨ 1 ´ 2´p1´κq
 geometric tomography on semi–algebraic sets 98

Proposition 5.5 then yields that

|I1 pLq| ! pT }k}q´jκ (6.22)
with an implicit constant depending on the map ψ and on the integer j. Specialise the
inequalities (6.19) and (6.20) to the case that L “ }k}κ and note that the condition (6.21)
then amounts to
ˆ ˙1{p1´κq
´1{p1´κq ∆P pKq
}k} ě aP pψ, κq ¨ pT εq , where aP pψ, κq “ . (6.23)
1 ´ 2´p1´κq
One then obtains that
ÿ ż ´ ¯
ψpxq ¨ φ ε´1 ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx
kPZn zt0u R
n

ÿ ˇż ´ ¯ ˇ
ˇ ´1
ˇ
ď ˇ ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dxˇˇ
Rn
ˇ
1ď}k}ďaP pψ,κq¨pT εq´1{p1´κq
ÿ
` pI1 p}k}κ q ` I2 p}k}κ qq .
}k}ąaP pψ,κq¨pT εq´1{p1´κq

Here, given any integer j satisfying the bound (6.18),

ÿ ÿ
pI1 p}k}κ q ` I2 p}k}κ qq ! }k}´κpj´1q
(6.19)&(6.22)
}k}ąaP pψ,κq¨pT εq´1{p1´κq }k}ąaP pψ,κq¨pT εq´1{p1´κq

! pT εqpκ¨pj´1q´nq{p1´κq
! pT εqβ
(6.18)

since T ε ă 1 by assumption. This concludes the proof of the lemma.

6.2 geometric tomography on semi–algebraic sets

In view of Lemma 6.2 and of the discussion held in the introduction to this chapter,
the problem is reduced to obtaining uniform decay estimates for the oscillatory integral
appearing in the relation (6.15) when the integer vector k varies in a bounded domain.
To this end, the integral is first expressed as the Fourier transform of the so–called
Gel’fand–Leray function in Section 6.2.1 below. The properties of this function relevant
to the proof of Theorem 6.1 are then established in the following Section 6.2.2.

6.2.1 Uniform Fourier Decay of the Gel’fand–Leray Function

Define the Gel’fand–Leray function gP depending on the variables pv, ε, σq P Sn´1 ˆ R2

and on the weight functions ψ and φ as the map
ż ´ ¯
pv, ε, σq P Sn´1
ˆ R ÞÑ xgP pv, ε, σq , pφ, ψqy “
2
ψ pxq ¨ φ ε´1 ¨ Ppxq ¨ dx, (6.24)
vK pσq
 geometric tomography on semi–algebraic sets 99

where the affine subspace vK pσq is defined in (6.3) (this slightly unusual notation for
the function gP takes into account the fact that it depends both on φ and ψ). It satifies
the property that, given ε ą 0, λ ě 1 and v P Sn´1 , the oscillatory integral in (6.15) can
be decomposed after an elementary change of variables as the Fourier transform of the
Gel’fand–Leray function as follows :
ż ´ ¯ ż
´1
ψ pxq ¨ φ ε ¨ Ppxq ¨ e pλv ¨ xq ¨ dx “ xgP pv, ε, σq , pφ, ψqy ¨ e pλσq ¨ dσ. (6.25)
Rn R

This transformation can be seen as a formalisation of the principle of geometric tomo-

graphy according to which the properties of a sufficiently nice set can be understood
from the knowledge of the volume of their slices with lower dimensional families of
sets.

The main property of interest satisfied by the Gel’fand–Leray function is contained

in the following lemma, which is proved in the next section.
Lemma 6.3. Recall that the weight functions ψ and φ introduced in (6.13) are assumed to
be semialgebraic. Then, there exists an integer bP pφ, ψq ě 1 depending on these weights
and on the polynomial Ppxq only such that for any given real number ε ą 0 and any vector
v P Sn´1 , the real line can be partitioned into at most bP pφ, ψq intervals on the interior of
which, whenever nonempty, the map

σ P R ÞÑ xgP pv, ε, σq , pφ, ψqy (6.26)

is continuously differentiable and monotonic.

This immediately implies the main ingredient in the proof of the counting estimate
in Theorem 6.1 :
Corollary 6.4. Under the assumptions of the previous lemma, given ε ą 0, set

MP pψ, φ, εq “ sup sup xgP pv, ε, σq , pφ, ψqy .

vPSn´1 σPR

Then, it holds that

ż ´ ¯ MP pψ, φ, εq
ψ pxq ¨ φ ε´1 ¨ Ppxq ¨ e pλv ¨ xq ¨ dx ! (6.27)
Rn λ
for some implicit constant independent of the choice of v P Sn´1 and of λ ą 0.
Proof. Express the oscillatory integral on the left–hand side of the inequality (6.27) as
the Fourier transform of the Gel’fand–Leray function as in the equation (6.25). Then,
decompose this Fourier transform as
˜ż ¸
ż c
ÿ
xgP pv, ε, σq , pφ, ψqy ¨ e pλσq ¨ dσ “ xgP pv, ε, σq , pφ, ψqy ¨ e pλσq ¨ dσ .
R j“1 Ij
(6.28)
 geometric tomography on semi–algebraic sets 100

Here, c “ cP pφ, ψ, v, εq is at most equal to the integer bP pφ, ψq present in the statement
pPq
of Lemma 6.3 and, for each 1 ď j ď c “ cP pφ, ψ, v, εq, the set Ij “ Ij pφ, ψ, v, εq is an
interval with nonempty interior where the restriction of the Gel’fand–Leray function
seen as a map in the variable σ is continuously differentiable and monotonic.

pPq pPq
Fixing the integer j and denoting by dj “ dj pφ, ψ, v, εq and by ej “ ej pφ, ψ, v, εq
the endpoints of the interval Ij , one obtains by integration by parts that
ˇż ˇ
ˇ ˇ
ˇ xgP pv, ε, σq , pφ, ψqy ¨ e pλσq ¨ dσˇ
ˇ ˇ
ˇ Ij ˇ
ˇ ˇ
ˇ rxg pv, ε, σq , pφ, ψqy ¨ e pλσqsσ“ej ż ej ˇ
ˇ P σ“dj 1 d ˇ
“ˇ ˇ ´ ¨ pxgP pv, ε, σq , pφ, ψqyq ¨ e pλσq ¨ dσˇˇ
ˇ 2iπλ 2iπλ dj dσ ˇ
supσPIj |xgP pv, ε, σq , pφ, ψqy|
ď ,
λ
where the last inequality follows from the Triangle Inequality and from the assumption
of the monotonicity of the map σ ÞÑ xgP pv, ε, σq , pφ, ψqy restricted to the interval Ij . The
definition of the quantity MP pψ, φ, εq and the decomposition (6.28) then yield that
ˇż ˇ
ˇ xgP pv, ε, σq , pφ, ψqy ¨ e pλσq ¨ dσˇ ď cP pφ, ψ, v, εq ¨ MP pψ, φ, εq
ˇ ˇ
ˇ
R
ˇ λ
bP pφ, ψq ¨ MP pψ, φ, εq
ď ,
λ
which completes the proof of the statement.

6.2.2 Counting Intervals of Monotonicity through O–Minimality

The goal in this section is to establish Lemma 6.3. To this end, recall that an o–minimal
structure is a collection of families of sets O “ tOk ukě0 satisfying the following proper-
ties :

(i) for each k ě 0, the family Ok is made of subsets of Rk stable under complemen-
tation and union and containing the empty set ;

(ii) for any integer k ě 0, if A P Ok , then A ˆ R P Ok`1 and R ˆ A P Ok`1 ;

(iii) ␣
for any integer k ě 0( and any indices 1 ď i ă j ď k, the set
px1 , . . . , xk q P Rk : xi “ xj lies in Ok ;

(iv) given an integer k ě 0 and a set A P Ok`1 , denoting by π : Rk`1 Ñ Rk the

canonical projection map, it holds that πpAq P Ok ;

(v) Any x P R is such that txu P O1 ; furthermore, px, yq P R2 : x ă y P O2 ;

␣ (
 geometric tomography on semi–algebraic sets 101

(vi) the family O1 consists precisely of the finite unions of points and open intervals
in the real line.

A set is definable in the structure O if it belongs to a family Ok for some k ě 0, and a

map is definable if so is its graph.

The following generic process to generate o–minimal structures is described by

Scanlon [71]( : given k ě 0, fix a class of so–called distinguished functions Fk “
f : Rk Ñ R . An atomic set is a set of the form
␣

! ) ! )
x P Rk : apxq ă bpxq or x P Rk : apxq “ bpxq ,

where a, b : Rk Ñ R are any well–defined functions taken as the compositions of the

coordinate maps, constant maps and distinguished maps. Letting
ď
F “ Fk ,
kě0

denote by OF the smallest collection of subsets of Rk (as k ě 0 varies) containing the

atomic sets and stable under the images of the canonical projections π : Rk`1 Ñ Rk
and by taking complements and finite unions. Then, OF defines an o–minimal
structure.

The following examples are of particular importance in the proof of Lemma 6.3 :

(a) when F :“ Falg is the set of all polynomials defined over Rk for some (varying)
k ě 0, the collection OF is the set of all semialgebraic sets. That the above condi-
tion (iv) should be satisfied is the content of the Tarski–Seidenberg Theorem [30,
§2.10], the other properties being easily verified.

(b) when F :“ Fan is defined as the union of Falg and of all restricted analytic func-
tions (these are maps f : Rk Ñ R vanishing outside r´1, 1sk which are the
restriction to r´1, 1sk of a function which is real analytic in a neighbourhood of
r´1, 1sk ), one obtains the o–minimal structure of restricted analytic functions. The
proof in this case is due to Denef and van den Dries [29]. A detailed description
of the sets in Fan is, furthermore, provided in [31] : these are precisely the glob-
ally subanalytic ones; that is, sets V Ă Rk such that for each x P Rk , there exists
a relatively compact semianalytic set X contained in Rk`l for some integer l ě 1
and a neighbourhood U of x such that the intersection V X U is the projection of
X onto the first k coordinates (recall here that a set is semianalytic if it is defined
locally around any of its points as a finite sequence of equations or inequalities
involving analytic maps, or a finite union of such sets).

(c) when F :“ Fan,exp is defined as the union of Fan and of the singleton consisting
of the exponential function exp : R Ñ R, one obtains the o–minimal structure
 geometric tomography on semi–algebraic sets 102

of the expansian of restricted analytic functions by the exponential map. The proof in
this case is due to van den Dries and Miller [32]. It should be noted that the
logarithm function log : Rą0 Ñ R is (definable in Fan,exp since its graph is the set
px, yq P R2 : px ą 0q ^ px “ exppyqq .
␣

The proof of Lemma 6.3 requires to work with the parametric integral of a semial-
gebraic map (see the definition of the Gel’fand–Leray function in (6.24) and the equa-
tion (6.26)). It turns out that such a parametric integral map need not remain in the
structure of semialgebraic sets. To see this, it is enough to consider the semialgebraic
function #
1{y if px ą 0q ^ p1 ă y ă xq;
gpx, yq “
0 otherwise.
Integrating with respect to the variable y then defines a parametric integral expressed
as the logarithm function, which is not semialgebraic. This is the reason explaining
the introduction of the larger structure Fan,exp (which, in particular, comprises the
logarithm function).

Proof of Lemma 6.3. Under the assumption that the weights ψ and φ are semialge-
braic, the Gel’fand–Leray map pv, ε, σq P Sn´1 ˆ R2 ÞÑ xgP pv, ε, σq , pφ, ψqy defined
in (6.24) is a parametric integral of a globally subanalytic function (since semialgebraic
functions are globally subanalytic). The seminal work [59] by Lion and Rolin proves
that a parametric integral of a globally subanalytic function, and therefore the
Gel’fand–Leray map itself, is definable in the o–minimal structure Fan,exp . (Kaiser
exhibits in [49] a smaller o-minimal structure containing all parametric integrals of
semialgebraic functions. The so-called constructible functions form a class of functions
containing all such integrals which is optimal in a sense detailed in Section 6.4 below.)

The proof relies on the existence of a C1 –cell decomposition of the domain of

definition of the Gel’fand–Leray map. For the purpose of the present considerations,
the actual (technical) definition of a cell is irrelevant so that the reader is referred
to [30, Chap. 3] for further details. It is enough to mention here that these are, in some
suitable sense, the "building blocks" of an o–minimal structure (here Fan,exp ). They
are definable and stable under the same operations as this structure (in particular
under canonical projections such as in the above property (iv)) and can be used to
partition any definable set into finitely many pieces. Furthermore, and this is the
above mentioned existence claim, it follows from [30, Theorem 7.3.2] that the domain
of definition of any definable function can be partitioned into finitely many cells where
the function is C1 : in the case under consideration, this means that there exists a
finite collection D of cells partitioning the set Sn´1 ˆ R2 such that the restriction of the
Gel’fand–Leray map to each cell D in D is continuously differentiable. In other words,
for each D P D, there exists a definable open set UD and a C1 map ΓD : UD Ñ R such
that the restrictions to D of ΓD and of the Gel’fand–Leray map xgP , pφ, ψqy coincide.
 completion of the proof of the main theorem 103

Theorem 7.3.2 in [30] establishes that any definable set can be partitioned into
finitely many cells which are C1 (this is imposing the regularity of the boundary of
the cells, see loc. cit. for further details). Consider then a partition A of the domain
Sn´1 ˆ R2 into C1 –cells inducing for each D P D partitions of the sets
" * " *
` BΓD ´ BΓD
AD “ pv, ε, σq P D : ě0 and AD “ pv, ε, σq P D : ă0 .
Bσ Bσ

These sets are definable as the partial derivatives of a definable function are definable
on their domain of definition (see, e.g., [80, Point (3), §2.3] for a justification of this
claim). Since the union over D P D of the sets A´D and AD covers the domain S
` n´1 ˆ R2 ,

each cell in the family A lies in some A˘D.

Fix pv0 , ε0 q P Sn´1 ˆ R and denote by π : pv, ε, σq P Sn´1 ˆ R2 ÞÑ pv, εq P Sn´1 ˆ R

the canonical projection. Consider the various sets

Apv0 ,ε0 q “ tσ P R : pv0 , ε0 , σq P Au (6.29)

obtained as the projection over the real line of the preimages under the projection map
π of all those cells A P A such that pv0 , ε0 q P πpAq. From [30, Proposition 3.3.5], they
form a cell decomposition of the real line denoted by Apv0 ,ε0 q .

Consider a set of the form (6.29) not reduced to a finite union of points and choose
D P D such that A Ă A˘ D . Then, over each of the finitely many open intervals making its
interior, the map σ P R ÞÑ xgP pv0 , ε0 , σq , pφ, ψqy is C1 and monotonic as it coincides
therein with the continuously differentiable map σ ÞÑ ΓD pv0 , ε0 , σq whose derivative
keeps a constant sign. Lemma 6.3 is then an immediate consequences of the following
three observations :

‚ the collection Apv0 ,ε0 q is contained in A;

‚ the collection A has a finite cardinality;

‚ given a set in A, the number of connected components of its projection over R

(with respect to the third coordinate σ P R ) is finite from the above property (vi).

6.3 completion of the proof of the main theorem

The first part (§6.3.1) of this antepenultimate section is devoted to the completion of
the proof of the claim (Z1) in Theorem 6.1 and the second one (§6.3.2) to the proof of
(Z2).
 completion of the proof of the main theorem 104

6.3.1 The Counting Bound

The goal is here to establish the counting bound (6.6), which constitutes the substance
of Claim (Z1) in Theorem 6.1. Recall that the discussion following the statement of
the theorem makes it clear that it suffices to establish this claim in the case of a single
polynomial Ppxq.
Proof of Claim (Z1) in Theorem 6.1. Let

T ´α ď ε ă T ´1 , (6.30)

where, by assumption, α ą 1. Let β ą 0 be a large parameter and let κ P p0, 1q

be a small parameter. Fix for the time being two nonnegative compactly supported
semialgebraic maps φ and ψ which satisfy the inequalities (6.13). These maps are
assumed to be j ě 1 times continuously differentiable, where j is any integer large
enough (depending on n, β and κ) so that the assumption (6.18) of the Non–Stationary
Phase Lemma 6.2 holds. Recall that the existence of such maps is guaranteed, e.g.,
by [52, Proposition 4.8].

From the Poisson summation estimate (6.15) and from Lemma 6.2, one has that
ˆż ´ ¯ ˙
NP pK, T , αq ! T ¨
: n
ψpxq ¨ φ ε ¨ Ppxq ¨ dx ` T n ¨ pεT qβ
´1

¨ R
n
˛
ˇż ´ ¯ ˇ
ÿ ˇ ˇ
` Tn ¨ ˝ ´1
ˇ n ψpxq ¨ φ ε ¨ Ppxq ¨ e p´T k ¨ xq ¨ dxˇ
ˇ ˇ‚
R
1ď}k}ďaP pψ,κq¨pT εq´1{p1´κq

for some implicit constant depending on j, β, κ and on the weight functions ψ and φ.
In this inequality, Corollary 6.4 applied to the polar decomposition (6.16) implies that
for any given k P Zn zt0u,
ż ´ ¯ MP pψ, φ, εq
ψpxq ¨ φ ε´1 ¨ Ppxq ¨ e p´T k ¨ xq ¨ dx !
Rn T ¨ }k}
with an implicit constant depending only on the polynomial Ppxq and on the weights
ψ and φ. As a consequence,
ˆż ´ ¯ ˙
NP pK, T , αq ! T ¨
: n
ψpxq ¨ φ ε ¨ Ppxq ¨ dx ` T n ¨ pεT qβ
´1
Rn
` T n´1 ¨ pT εq´pn´1q{p1´κq ¨ MP pψ, φ, εq

in such a way that

ˆż ´ ¯ ˙
NP: n
pK, T , αq ! T ¨ ψpxq ¨ φ ε ¨ Ppxq ¨ dx ` T n ¨ pεT qβ
´1
Rn
κ¨pn´1q n´1
` T´ 1´κ ¨ ε´ 1´κ ¨ MP pψ, φ, εq. (6.31)
 completion of the proof of the main theorem 105

Recall that K and K are two compact sets such that the former is contained in the
interior of the latter and such that the latter is contained in the open set U defined as
part of the support restriction condition (1.10). Given any real c P p0, 1q and any smooth
´ ¯
and compactly supported map ψ, recall also the definition of the quantity µP ψ, c
r r
in (2.23). Choose then ψ
r such that
r ď χK .
χK ď ψ ď ψ (6.32)
Under the assumption pH3 q applied
´ ´ ¯ to the ´ polynomial
¯¯ Ppxq, and with the notation
of Lemma 2.5, it holds that rP ψ , mP ψ
r r “ prP pKq, mP pKqq. Here, under the
assumption pH2 q, the point pX1 q in Proposition 2.16 implies that rP pKq ą 0. Then, from
Lemma 2.5,
ż ´ ¯ ż ´ ¯
´1
ψpxq ¨ φ ε ¨ Ppxq ¨ dx ď ψpxq
r ¨ χt|Ppxq|ď2εu ¨ dx “ µP ψ, r 2ε
Rn (6.13) Rn (2.23)

! rP pKq
ε ¨ |log ε|mP pKq´1 .
(2.24)

As a consequence,
κ¨pn´1q n´1
NP: pK, T , αq ! T n ¨ εrP pKq ¨ |log ε|mP pKq´1 ` T n ¨ pεT qβ ` T ´ 1´κ ¨ ε´ 1´κ ¨ MP pψ, φ, εq.
(6.31)

Let η ą 0. Provided that ε ą 0 is small enough, it follows from the definition of the
measure of flatness qP pKq in (6.5) that
n´1
NP: pK, T , αq ! T n ¨ εrP pKq ¨ |log ε|mP pKq´1 ` T n ¨ pεT qβ ` ε´ 1´κ `qP pKq´η . (6.33)

The goal is first to prove the counting bound (6.6) in the generic case (i) of the state-
ment of the theorem. To this end, note that the condition (6.7) implies that, provided
that η and κ are small enough,
n
ρP pK, K, κ, ηq :“ n´1 ą 1. (6.34)
1´κ ´ q P pKq ` rP pKq ` η
The counting bound is then established by a distinction of cases :
‚ assume that
α ď ρP pK, K, κ, ηq (6.35)
and set ε “ T ´α . Then, the inequality (6.33) yields
NP: pK, T , αq ! T n´rP pKq¨α ¨ |log T |mP pKq´1 ` T n`βp1´αq . (6.36)
Under the assumption that α ą 1, it is enough to choose β ą α ¨ rP pKq{pα ´ 1q
so that this sum should be bounded by constant multiple of the first term which,
from Case (2) in Theorem 1.1, is the volume of the set S:P pK, T , αq. In other words,
´ ¯
NP: pK, T , αq ! Voln S:P pK, T , αq . (6.37)
 completion of the proof of the main theorem 106

‚ Assume now that the converse to the inequality (6.35) holds, namely that

α ą ρP pK, K, κ, ηq. (6.38)

In this case, choose

ε “ T ´ρP pK,K,κ,ηq (6.39)
in such a way that the inequality (6.33) yields

NP: pK, T , αq ! T ρpP pK,K,κ,ηq`γ ` T n`βp1´αq

with ˜ ¸
rP pKq
pP pK, K, κ, ηq :“ n ¨ 1 ´ n´1
ρ
1´κ ´ qP pKq ` rP pKq ` η
for any γ ą 0 (this extra factor is to absorb the possible logarithmic contribution
in (6.33)). The condition (6.8) ensures that one can choose δP pK, Kq ą 0 lying in
the interval determined by the inequalities (6.9). Provided that the parameters
γ, η and κ are chosen small enough, it then holds that

pP pK, K, κ, ηq ` γ ă τP pKq ´ δP pK, Kq.

ρ

Under the assumption α ą 1 and upon choosing the parameter β large enough,
this is saying that

NP: pK, T , αq ! T τP pKq´δP pK,Kq . (6.40)

Thus, under the assumptions (6.7) and (6.8), the inequalities (6.37) and (6.40) establish
the counting bound (6.6) for any α ą 1 and for any exponent δP pK, Kq ą 0 in the range
determined by (6.9), namely,
´ ¯
NP pK, T , αq ! Voln SP pK, T , αq ` T τP pKq´δP pK,Kq .
: :
(6.41)

It remains to prove this bound in the degenerate case (ii) of Claim (Z1) in Theo-
rem 6.1 when it is assumed that the condition (6.10) holds. The reason why the above
argument is not applicable anymore is that the relation (6.34) need not hold any longer
when the parameters κ and η are small enough. As a consequence, the condition (6.35)
is neither met under the assumption α ą 1 nor can one necessarily set ε as in (6.39)
when the complementary assumption (6.38) is satisfied because of the constraints
imposed in (6.30).

To deal with this situation, set ε “ T ´p1`ηq for the same η ą 0 as the one fixed above.
Then, the inequality (6.33) yields the upper bound
n´1
NP: pK, T , αq ! T n´rP pKq¨p1´η{2q ` T n´η¨β ` T p1`ηq¨p 1´κ ´qP pKq`ηq .
 completion of the proof of the main theorem 107

If γ ą 0 is given and if η and κ are chosen suitably small, the assumption of degenera-
cy (6.10) implies that

NP: pK, T , αq ! T n´1´qP pKq`γ .

Under the assumption (6.11), one can fix a real δP pK, Kq ą 0 in the range determined
by the inequalities (6.12). It is then enough to choose γ ą 0 small enough so that

n ´ 1 ´ qP pKq ` γ ď τP pKq ´ δP pK, Kq.

This establishes the bound (6.41) in this degenerate case also and thus completes the
proof of the claim (Z1) in Theorem 6.1.

6.3.2 The Case of Smooth Varieties

The goal is now to establish Claim (Z2) of Theorem 6.1. To this end, fix p ě 1 homo-
geneous forms F1 pxq, . . . , Fp pxq P Rrxs such that the smoothness condition given by the
nonvanishing of the map (1.6) over K holds. In this case, the dimension of the algebraic
variety ZR pFq over K equals τF pKq “ n ´ τpF pKq “ n ´ p. Furthermore, from Point (2) in
Theorem 1.1, one then has that rP pKq “ p (recall here that K is a compact set contained
in the interior of K Ă U and that it contains the set C in its interior, where C and U are
defined in the support restriction condition (1.10)). Under these assumptions, the in-
equalities (6.7) and (6.8) are the same : showing that they hold amounts to establishing
that
qF pKq ą τpF pKq ´ 1 “ p ´ 1. (6.42)
Proof that the inequality (6.42) holds under the assumptions of the claim (Z2) in Theorem 6.1.
Given ρ ą 0, denote by NF pK, ρq the ρ–tubular neighbourhood of the variety ZK pFq :
it is defined as the set of all those points in Rn lying at a distance less than ρ from
ZK pFq.

Given open sets O1 and O2 contained in Rn , the set O1 satisfies the unique nearest
point property with respect to ZO2 pFq :“ ZR pFq X O2 if for every y P O1 , there exists
a unique point πF pyq P ZO2 pFq minimising the distance from y to ZO2 pFq. The reach
ρF pyq of a point y P ZO2 pFq is then defined as the supremum of all those real numbers
ρ ą 0 such that the open ball with radius ρ centered at y has the unique nearest point
property with respect to ZO2 pFq.

Under the assumption that the map (1.6) does not vanish over K, and therefore
over a small enough neighbourhood O of it, a classical result by Federer (see [34, Theo-
rem 4.12]) implies that ρF pyq ą 0 at any point y P ZO pFq. Since, from [34, Remark 4.2],
the reach map y P ZK pFq ÞÑ ρF pyq is continuous and since the set K is assumed to be
compact, one infers the existence of a real ρF pKq ą 0 such that the projection map

πF : y P NF pK, ρF pKqq ÞÑ πF pyq P ZR pKq

 completion of the proof of the main theorem 108

is well–defined. From there on, establishing the inequality (6.42) requires an elemen-
tary lemma :
Lemma 6.5 (Tubular neigbourhoods are comparable to the set of small solutions to
inequalities under the assumption of smoothness). Let ρF pKq ą 0 be as above. Define the
constants

αF pKq “ min αi pKq´1 and βF pKq “ max t1, βi pKqu ,

1ďiďp 1ďiďp

where for i “ 1, . . . , p,
! )
αi pKq “ max }∇Fi pxq} : x P NF pK, ρF pKqq

and ! )
2
βi pKq “ max ~∇ Fi pxq~ : x P NF pK, ρF pKqq .

Here, ∇2 Fi pxq denotes the n ˆ n Hessian matrix associated to the homogeneous form Fi at the
point x and ~ ¨ ~ denotes the matrix norm induced by the Euclidean norm. Set furthermore
" ´ ! )¯1{2 *
´1
θF pKq “ max 1, max ~GrF pxq ~ : x P NF pK, ρF pKqq ,

where GrF pxq is the Gram p ˆ p matrix whose pi, jqth entry is the scalar product
∇Fi pπF pxqq ¨ ∇Fj pπF pxqq, and let
" *
˚ 1
ρF pKq “ min ρF pKq, ? ¨ (6.43)
2 n ¨ θF pKq ¨ βF pKq

Then, all these constants are well–defined under the smoothness assumption that the
map (1.6) does not vanish. Moreover, whenever 0 ă ε ă ρ˚F pKq, it holds that

NF pK, αF pKq ¨ εq Ă MF pK, εq Ă NF pK, 2 ¨ θF pKq ¨ εq , (6.44)

where
MF pK, εq “ tx P NF pK, ρ˚F pKqq : |Fi pxq| ď ε for all 1 ď i ď pu .

Proof. Let x P NF K, ρ˚F pKq . Given 1 ď i ď p, Taylor–Lagrange’s Theorem provides

` ˘

the existence of a point yx lying in the line segment rπF pxq, xs such that

|Fi pxq| “ |∇Fi pyx q ¨ px ´ πF pxqq| ď }∇Fi pyx q} ¨ }x ´ πF pxq} ,

where the last relation follows from the Cauchy–Schwarz inequality. This is easily
seen to imply the first inclusion in (6.44).
 completion of the proof of the main theorem 109

As for the second one, fix again a point x P NF K, ρ˚F pKq and an index 1 ď i ď p.
` ˘

The Taylor–Lagrange Theorem at order two guarantees the existence of a point yx P

rπF pxq, xs such that
ˇ ˇ
ˇ 1 T 2
ˇ
|Fi pxq| “ ˇˇ∇Fi pπF pxqq ¨ px ´ πF pxq ` ¨ px ´ πF pxqq ¨ ∇ Fi pyx q ¨ px ´ πF pxqqˇˇ
2
ě |∇Fi pπF pxqq ¨ px ´ πF pxqq| ´ βi pKq ¨ }x ´ πF pxq}2 , (6.45)

where px ´ πF pxqqT denotes the transpose of the (column) vector x ´ πF pxq. This vector
lies in the subspace normal to the manifold ZK pFq at the point πF pxq. Under the
smoothness assumption that the map (1.6) does not vanish, a basis for this normal
space is the set of linearly independent vectors p∇Fi pπF pxqqq1ďiďp . The smoothness
assumption also guarantees that the p ˆ p Gram matrix GrF pxq formed from these
vectors is invertible.

The norm of the vector x ´ πF pxq lying in the normal subspace under consideration
then satisfies the relations

}x ´ πF pxq}2 “ vF pxqT ¨ GrF pxq´1 ¨ vF pxq ď θF pKq2 ¨ }vF pxq}2 ,

where vF pxq is the p–dimensional (column) vector whose coordinates are the scalars
∇Fi pπF pxqq ¨ px ´ πF pxqq for 1 ď i ď p. If one assumes that |Fi pxq| ă ε for some ε ą 0
and for all 1 ď i ď p, the inequality (6.45) then yields that

}x ´ πF pxq} ? }x ´ πF pxq}
ε ą ´ n ¨ βF pKq ¨ }x ´ πF pxq}2 ě ¨
θF pKq (6.43) 2 ¨ θF pKq

This establishes the second inclusion in (6.44) and thus completes the proof.

To resume the proof of the inequality (6.42), note that the Łojasiewicz Inequality
(as stated, e.g., in [63, Theorem 4.1]) guarantees the existence of constants c ą 0 and
λ ą 0 depending on the compact set K and on the set of homogeneous forms Fpxq “
pF1 pxq, . . . , Fp pxqq such that

}Fpxq} ě c ¨ dist px, ZR pFqqλ .

Here, the notation on the right–hand side stands for the Euclidean distance from a
point x P K to the set ZR pFq. This implies that for any ρ ą 0, there exists εF pK, ρq ą 0
such that for all 0 ă ε ă εF pK, ρq, the inequality }Fpxq} ă ε forces the point x P K to lie
in the ρ–neighbourhood of the variety ZR pFq. Combined with Lemma 6.5, one obtains
the existence of real numbers rεF pKq ą 0 and cF pKq ą 0 such that for any 0 ă ε ă r εF pKq
and any v P S n´1 and σ P R,

x P K X vK pσq : }Fpxq} ă ε Ă vK pσq X MF pK, εq Ă vK pσq X NF pK, cF pKq ¨ εq .

␣ (
 completion of the proof of the main theorem 110

As a consequence, for ε ą 0 small enough,

Voln´1 x P K X vK pσq : }Fpxq} ď ε

`␣ (˘ ` ˘
ď Voln´1 vK pσq X NF pK, cF pKq ¨ εq . (6.46)

To bound from above the volume on the right–hand side of this inequality, denote
by NMF pKq the normal bundle of the variety ZK pFq : it is the union over x P ZK pFq of
the normal spaces NMF pK, xq where px, yq P NMF pK, xq if x P ZK pFq and if y lies in the
subspace orthogonal to the tangent space to ZK pFq at the point x. The Tubular Neigh-
bourhood Theorem states that there exists a continous mapping x P ZK pFq ÞÑ δpxq ą 0
such that the map px, yq ÞÑ x ` y realises a diffeomorphism between the open subset
of NMF pKq defined as tpx, yq P NMF pKq : }y} ă δpxqu and its image. The proof of
this statement in [57, Theorem 6.24] makes it also clear that the mapping δ is uniformly
bounded below by a strictly positive constant under the assumption that the reach of
the manifold ZK pFq is strictly positive.

One thus obtains for some ρ rF pKq ą 0 a diffeomorphism

r“ρ

px, yq P NMF pKq rr

ρs ÞÑ t “ x ` y P NF pK, ρ
rq

with inverse

t P NF pK, ρ ρs .
rq ÞÑ px, yq “ pπF ptq, t ´ πF ptqq P NMF pKq rr (6.47)

Here, one has set NMF pKq rr ρs “ tpx, yq P NMF pKq : x P K and }y} ă ρ ru, where
the vector y lies in a p–dimensional subspace (namely, the subspace normal to the
pn ´ pq–dimensional tangent subspace to the variety ZK pFq at a given point).

Given ε ą 0 small enough, up to a multiplicative constant absorbing the sup–norm

of the Jacobian of the change of variables (6.47), the volume on the right–hand side of
the inequality (6.46) is thus at most the product of two nonnegative quantities : on the
one hand, the (surface) measure of the set of x P ZK pFq lying at a distance less than
cF pKq ¨ ε from a point in the affine space vK pτq and, on the other, the maximal volume
of the intersection of a p–dimension ball of radius cF pKq ¨ ε with vK pσq. In other words,

Voln´1 vK pσq X NF pK, cF pKq ¨ εq ! τF ptx P K : |v ¨ x ´ σ| ď cF pKq ¨ εuq ¨ εp´1 ,

` ˘

(6.48)
where τF denotes the surface measure of the restriction of the variety ZR pFq to a small
enough neighbourhood of K where the gradient map (1.6) does not vanish, and where
the implicit constant in (6.48) is independent of v and σ. To complete the proof, it
suffices to show that the surface measure of the set under consideration decays as a
positive power in the variable ε.

From the compacity of K, the problem can be reduced to finitely many similar
local estimates which can be dealt with the theory of pC, αq–good functions along the
lines of [2, Corollary 3]. As this analytic approach requires rather tedious calculations
 determining the measure of flatness 111

and a distinction of cases to obtain estimates uniform in the parameters v and σ, the
claim is established hereafter with a more global approach.

To this end, let ψ be in C8

c pR q such that it is constant equal to 1 over K and such
n

that its support is contained in the interior of the neighbourhood of K used to define
τF . Define then the volume element

dµF “ ψ ¨ dτF . (6.49)

In particular, the induced measure µF is absolutely continuous with respect to the

measure τF .

Under the assumption that the variety ZK pFq is not contained in a linear subspace
of dimension n ´ p, it follows from [72, p.351, Point (iii) & Theorem 2] that there exists
κ ą 0 such that the Fourier transform of the measure µF satisfies the decay property

µF pξq| ! }ξ}´κ
|p (6.50)

for all ξ P Rn z t0u and for some implicit constant depending on the choice of ψ. From
the Esséen Concentration Inequality [75, §2.2.11], it then holds that for any v P Sn´1
and σ P R,
˜ż ¸
µF ptx P K : |v ¨ x ´ σ| ď cF pKq ¨ εuq ! pcF pKq ¨ εq ¨ µF ptvq| ¨ dt
|p
|t|ďpcF pKq¨εq´1

for some implicit constant independent of the choice of v, σ and ε. Upon combining
this relation with (6.49) and (6.50), one obtains from the inequality (6.48) that for some
κ ą 0,
Voln´1 vK pσq X NF pK, cF pKq ¨ εq ! εp´1`κ .
` ˘

As a consequence, the definition of the quantity MF pK, εq in (6.4) implies that

MF pK, εq ! εp´1`κ

in such a way that

ˆ ˙
log MF pK, εq
qF pKq :“ lim inf ě p ´ 1 ` κ ą p ´ 1.
εÑ0` log ε
This completes the proof of the inequality (6.42).

6.4 determining the measure of flatness

The practical use of Theorem 6.1 relies on the determination of the measure of flatness
introduced in (6.5). This final section is devoted to the development of the theory ena-
bling one to determine effectivity the value of this quantity. To this end, following the
 determining the measure of flatness 112

discussion held after the statement of Theorem 6.1, consider, without loss of generality,
the case of a single real homogeneous polynomial Ppxq of degree q ě 2. One is thus
interested in finding effectively the value of the real
ˆ ˙
log MP pK, εq
qP pKq “ lim inf , (6.51)
εÑ0` log ε
where
MP pK, εq “ sup sup µP,K pv, σ, εq
vPSn´1 σPR
and
x P K X vK pσq : |Ppxq| ď ε .
`␣ (˘
µP,K pv, σ, εq “ Voln´1
Here, K is assumed to be any compact semialgebraic set with nonempty interior
intersecting non trivially the algebraic variety ZR pPq. Under these assumptions, the
real qP pKq is well-defined.

Fix a real number ε ą 0, a vector v P Sn´1 and a scalar σ P R. The first step is
to make explicit the above-defined volume µP,K pv, σ, εq of the slice of the sublevel set
tx P K : |Ppxq| ď εu with the affine space vK pσq. To this end, denote by Rv the rotation
matrix mapping the last element en P Sn´1 of the canonical basis onto the unit vector
v P Sn´1 and leaving the orthogonal of the plane spanned by en and v unchanged (in
the degenerate case when v “ en , the matrix Rv is to be taken as the identity). Then,
one can decompose the vector x P K X vK pσq as x “ Rv py, σq, where y P Rn´1 lies in
the projection onto the first n ´ 1 coordinates of the preimage of the semialgebraic set
K by R´1
v . One is thus reduced to computing the n ´ 1 volume of the set of vectors y
such that
|pP ˝ Rv q py, σq| ď ε with py, σq P R´1
v pKq . (6.52)
This is achieved in the following statement.
Proposition 6.6. Keep the above notations and recall that the set K Ă Rn is assumed to be
bounded and semialgebraic. Then, there exists a semialgebraic set
Sn pKq Ă Rn´2 ˆ R ˆ Sn´1 ˆ p0, 1q
depending on K which can be partitioned into N ě 1 semialgebraic subsets Dj (where 1 ď
j ď N) such that the following property holds : over each of these subsets are defined lpjq ě 0
analytic semialgebraic maps ξ1 ă ξ2 ă ¨ ¨ ¨ ă ξlpjq : Dj Ñ R such that, defining for any
given pσ, v, εq P R ˆ Sn´1 ˆ p0, 1q the projected sets
! )
Dj pv, σ, εq “ t P R n´2
: pt, v, σ, εq P Dj ,

the volume µP,K pv, σ, εq is the sum of finitely many integrals of the form
ż
pξi`1 pt, v, σ, εq ´ ξi pt, v, σ, εqq ¨ dt.
Dj pv,σ,εq

Here, 1 ď j ď N and 1 ď i ď lpjq whenever lpjq ě 1.

 determining the measure of flatness 113

The definition of the semialgebraic set Sn pKq is explicitly given in the proof of the
proposition, which makes it also clear that it can be constructed effectively. This proof
can be inferred from the following result essentially due to Coste [28] :
Lemma 6.7. Let Q pxq P R rxs be a polynomial in n ě 1 variables. Assume that B Ă Rn´1 is
a connected semialgebraic set and that k and d are integers such that k ď d and such that for
every point px P B, the polynomial Q px1 , p xq has degree d in the variable x1 and exactly k distinct
roots in C. Then, there exist l ď k analytic semialgebraic maps ζ1 ă ¨ ¨ ¨ ă ζl : B Ñ R which
can be effectively determined such that, for every p x P B, the set of real roots of the polynomial
xq is exactly tζ1 pp
Q px1 , p xqu. Furthermore, for any i “ 1, . . . , l, the multiplicity of
xq , . . . , ζl pp
the root ζi pp
xq is constant for p x P B.
Proof. This is [28, Proposition 2.6] with the exception that it is shown there that the
maps ξi are just continuous rather than analytic. This additional feature is obtained
as a consequence of the continuity of the roots of a polynomial as functions of the
coefficients. A stronger statement established in [12] is true : the distinct roots of a
polynomial are analytic functions of the coefficients in the open set where the roots
retain their multiplicities. This yields the above statement.
Deduction of Proposition 6.6 from Lemma 6.7. Consider the real polynomial

Qpy, σ, v, εq “ ppP ˝ Rv q py, σqq2 ´ ε2 .

Even if it means reindexing the coordinates of the vector y “ py1 , . . . , yn´1 q P Rn´1 ,
assume without loss of generality that the variable y1 appears therein explicitly. Then,
denote by Sn pKq the image of the set
! )
py, σ, v, εq P R n´1
ˆRˆS n´1 ´1
ˆ p0, 1q : py, σq P Rv pKq (6.53)

under the projection map

py, σ, v, εq P Rn´1 ˆ R ˆ Sn´1 ˆ p0, 1q ÞÑ py

p , σ, v, εq P Rn´2 ˆ R ˆ Sn´1 ˆ p0, 1q,

where yp “ py2 , . . . , yn´1 q P Rn´2 when y “ py1 , y2 , . . . , yn´1 q P Rn´1 . From the
Tarski-Seidenberg Theorem (see Point (a) in §6.2.2), the set Sn pKq is also semialgebraic.

Partition Sn pKq into finitely many semialgebraic subsets such that over each of
them, the polynomial Q seen as a function of the variable y1 alone with coefficients
in Sn pKq keeps a constant degree with its complex roots retaining their multiplicity.
This is indeed possible since the degree of Q in the variable y1 is determined by
mutually exclusive semialgebraic conditions (namely, the vanishing of a suitable set
of coefficients) and so is the multiplicity of its set of roots (namely, the vanishing of
discriminants polynomials whose variables are the set of coefficients).

From [28, Theorem 2.23], each of the elements of the semialgebraic partition
thus obtained has finitely many semialgebraic connected components. Lemma 6.7
 determining the measure of flatness 114

then implies that over each such connected component, there exists a finite number
(possibly zero) of semialgebraic analytic maps describing the roots of the polynomial
Q (in the variable y1 ) as functions of the coefficients py p , σ, v, εq. Denote this finite
number of maps by ζ2 , . . . , ζl´1 and define corresponding maps ξ2 , . . . , ξl´1 in the
following
` way :˘ when 2 ď i ď l ´ 1, ξi is the restriction of ζi to the preimage
ζ´1
i π1 vpR ´1 pKqq , where π pR´1 pKqq is the semialgebraic image of R´1 pKq under
1 v v
the projection map py, σq ÞÑ y1 . Since from [28, Corollary 2.9], the preimage of a
semialgebraic set under a semialgebraic map is also semialgebraic, the maps ξi , where
1 ď i ď l ´ 1, remain semialgebraic. Then, by definition, for all 2 ď i ď l ´ 1, the point
ppξi pyp , σ, v, εq , y
p q , σ, v, εq lies in the set (6.53). Even if it means further considering
separately the restrictions of the maps ξi (2 ď i ď l ´ 1) to the (finite number of)
connected components of their common domain of definition, they may be assumed
to be defined over connected semialgebraic sets.

Let then ξ1 and ξl be the infimum and the supremum of the admissible values of
y1 , respectively, when py p , σ, v, εq belongs to the connected semialgebraic set, say D,
defining the domain of definition of the maps ξi , where 2 ď i ď l ´ 1 (these infimum
and supremum are finite under the assumption that the set K is bounded). The Mean
Value Theorem implies that if Qpy1 , y p , v, σ, εq ď 0 for some ppy1 , y p q , σ, v, εq lying in the
set (6.53), there exists an index 1 ď i ď l such that ξi py p , σ, v, εq ď y1 ď ξi`1 py p , σ, v, εq.

Fixing the components v, σ and ε, it follows␣ from the Tarski–Seidenberg ( Theo-

rem [30, §2.10] that the projected set D pv, σ, εq :“ y pPR n´2 p , σ, v, εq P D is semi-
: py
algebraic. Assuming that its pn ´ 2q–dimensional Lebesgue measure is positive, the
definition of the polynomial Q yields that the measure of the set of points y P Rn´1
such that |pP ˝ Rv q py, σq| ď ε with py, σq P R´1 v pKq is a finite sum of integrals of the
form ż
p , v, σ, εq ´ ξi py
pξi`1 py p , v, σ, εqq ¨ dyp.
Dpv,σ,εq

The proof is complete upon indexing the (finitely many) semialgebraic sets D pv, σ, εq
which have positive Lebesgue measure.

The next step is to show that, given ε ą 0, the suprema defining the real MP pK, εq
in (6.4) are attained at some point pv, σq “ pvpεq, σpεqq P Sn´1 ˆ R. Combined with
Proposition 6.6, this implies that the value of MP pK, εq can be determined in a finite
number of steps.

Proposition 6.8. Recall that the set K Ă Rn is assumed to be compact. Fix ε ą 0. Then, there
exists pvpεq, σpεqq P Sn´1 ˆ R such that

MP pK, εq “ µP,K pvpεq, σpεq, εq .

 determining the measure of flatness 115

To prove the statement, it is convenient to introduce, given a value of ε ą 0, the

map
ż
µP,K p ¨ , ¨ , εq : pv, σq ÞÑ µP,K pv, σ, εq “ χt|pP˝Rv qpy,σq| ď εu ¨ χR´1 pKq py, σq ¨ dy.
v
Rn´1
(6.54)

Proof. Since the set K is bounded, the support of the map µP,K p ¨ , ¨ , εq is compact. To
establish the claim, it is therefore enough to show that this map is upper semicontinu-
ous (so that it then assumes a maximal value over its support).

Let then pvk , σk qkě1 be a sequence in the support of µP,K p ¨ , ¨ , εq converging to a

point pv0 , σ0 q in this support. The goal is to prove that

lim sup µP,K pvk , σk , εq ď µP,K pv0 , σ0 , εq . (6.55)

kÑ8

To this end, one infers first from Fatou’s Lemma (which is applicable from the assump-
tion that the set K is bounded) that
ż ´ ¯
lim sup µP,K pvk , σk , εq ď lim sup χt|pP˝Rv qpy,σk q| ď εu ¨ χR´1 pKq py, σk q ¨ dy.
vk
kÑ8 Rn´1 kÑ8 k

Fix y P Rn´1 and let vkl , σkl lě1 be a subsequence (depending on y P Rn´1 ) realising
` ˘

the upper limit of the sequence of maps inside the integral on the right–hand side.
By the continuity of the polynomial map defined by Ppxq and from the closedness of
´ set K,¯ `upon taking
ˇthe ˘ˇˇ
limits as the integer l ě 1 tends to infinity in the relations
ˇ P ˝ Rvkl y, σkl ˇ ď ε and Rvkl y, σkl P K, it holds that |pP ˝ Rv0 q py, σ0 q| ď ε and
ˇ ` ˘

Rv0 py, σ0 q P K. As a consequence,

´ ¯
lim sup χt|pP˝Rv qpy,σk q| ď εu ¨ χR´1 pKq py, σk q ď χt|pP˝Rv qpy,σk q| ď εu ¨ χR´1 pKq py, σ0 q
k vk 0 v0
kÑ8

in such a way that (6.55) is verified. This completes the proof.

The conclusive step in the theory developed in this final section is to show that a
conjecture in o–minimality (communicated to the authors by Raf Cluckers) enables
one to make the calculation of the measure of flatness qP pKq completely effective in
the following sense : the lower limit of the ratio plog MP pK, εqq{plog εq defining it
in (6.51) is an actual limit which differs from this ratio up to an explicit error term as
ε Ñ 0` .

To this end, one needs to consider a relevant class of functions to analyse the pro-
perties of the map µP,K p ¨ , ¨ , εq introduced in (6.54) (given a fixed value of ε ą 0) and
then of the map ε ą 0 ÞÑ MP pK, εq. This is the class of constructible functions over a
given globally subanalytic set X Ă Rk defined in [22, §2] as the ring of real–valued
 determining the measure of flatness 116

functions over X generated by functions which either are globally subanalytic or the
logarithms of positive globally subanalytic functions. In other words, a function f is
constructible over X if it can be expressed as
a
ÿ b
ź
fpxq “ fj pxq ¨ log fij pxq
i“1 j“1

for some integers a, b ě 0, where the fi ’s and the fij ’s are subanalytic with fij ą 0 for
all i and all j. More generally, a map is said to be constructible if it is constructible
over some subanalytic set.

The main property satisfied by constructible functions, which is of fundamental

importance in the present considerations, is the stability under parametric integration
established in [25, Theorem 1.3]. This is saying that for any globally subanalytic sub-
sets A Ă Rn and B Ă Rm şand for any map f : A ˆ B Ñ R which is constructible (over
A ˆ B), the map y P A ÞÑ B fpy, xq ¨ dx is also constructible (over A). As a consequence,
the map µP,K p ¨ , ¨ , εq is constuctible as a function defined over a semialgebraic set as
the parametric integral over a semialgebraic domain of a semialgebraic function.

The function

MP pK, ¨ q : ε ą 0 ÞÑ MP pK, εq “ max max µP,K pv, σ, εq

vPSn´1 σPR

thus turns out to be the parametric maximum of a constructible function. The class
of constructible functions is nevertheless not stable under taking parametric maxima :
an explicit counterexemple to this claim is worked out in [48]. A weaker form of this
stability property is nevertheless expected to hold :

Conjecture 6.9 (Raf Cluckers). Given a constructible, non–negative function F : pε, xq P

X ÞÑ Fpε, xq defined over some subanalytic set X Ă p0, 8q ˆ Rk , there exist a constant δ ą 0
and a constructible function G : p0, 8q Ñ R of the form

Gpεq “ c ¨ |log ε|l ¨ εa (6.56)

for some real c ě 0, some rational number a and some integer l ě 0 such that

δ ¨ Gpεq ď sup Fpε, xq ď Gpεq for all ε ă δ.

xPRk

A more general form of this conjecture would assert that, under suitable as-
sumptions, the parametric supremum of a constructible function defined over any
subanalytic set is "sandwiched" between constant multiples of a constructible function
depending on the remaining variables. This is inspired by analogy to the p-adic case
where the corresponding result holds (even uniformly in p) — see [23, Theorem B]
 determining the measure of flatness 117

and [24, Theorem 2.1.3].

If Conjecture 6.9 holds, keeping the notation therein, the function MP pK, ¨ q satisfies
the inequalities
δ ¨ c ¨ |log ε|l ¨ εa ď MP pK, εq ď c ¨ |log ε|l ¨ εa
when 0 ă ε ă δ. As a consequence, for such values of ε,
ˇ ˇ
log pc ¨ lq log |log ε| ˇ log MP pK, εq ˇ log pc ¨ l ¨ δq log |log ε|
` ď ˇˇ ´ aˇˇ ď ` ,
log ε log ε log ε log ε log ε

implying in particular that

ˆ ˙
log MP pK, εq log |log ε|
“ qP pKq ` O , where qP pKq “ a.
log ε log ε

With the help of Propositions 6.6 and 6.8, this relation shows that the measure of
flatness qP pKq can be determined effectively in a finite number of steps assuming the
validity of Conjecture 6.9.
determining the measure of flatness 118
 REFERENCES

[1] Adiceam, F. Vertical shift and simultaneous Diophantine approximation on poly-

nomial curves. Proc. Edinb. Math. Soc. (2) (2) 58 (2015), no. 1, 1–26.

[2] Adiceam, F.; Beresnevich, V.; Levesley, J.; Velani, S. and Zorin, E. Diophantine
approximation and applications in interference alignment. Adv. Math. 302 (2016),
231–279.

[3] Athreya, J.S. and Margulis, G. Logarithm laws for unipotent flows. I. J. Mod. Dyn. 3
(2009), no. 3, 359–378.

[4] Athreya, J.S. and Margulis, G. Values of random polynomials at integer points.
J. Mod. Dyn. 12 (2018), 9–16.

[5] Atiyah, M.F. Resolution of singularities and division of distributions. Comm. Pure
Appl. Math. 23 (1970), 145–150.

[6] Arnold, V. I.; Gusein-Zade, S. M. and Varchenko, A. N. Singularities of differen-

tiable maps. Volume 2.Monodromy and asymptotics of integrals. Translated from
the Russian by Hugh Porteous and revised by the authors and James Montaldi.
Reprint of the 1988 translation. Modern Birkhäuser Classics. Birkhäuser/Springer,
New York, 2012.

[7] Bandi, P.; Ghosh, A. and Han, J. A generic effective Oppenheim theorem for sys-
tems of forms. J. Number Theory218 (2021), 311–333.

[8] Barlet, D. Contribution effective de la monodromie aux développements asympto-

tiques. Ann. Sci. École Norm. Sup. (4) 17 (1984), no. 2, 293–315.

[9] Basu, S.; Pollack, R. and Roy, M.-F. Algorithms in real algebraic geometry. Second
edition. Algorithms and Computation in Mathematics, 10. Springer–Verlag, Berlin,
2006.

[10] Benoist, Y. and Oh, H. Effective equidistribution of S–integral points on symmetric

varieties. Ann. Inst. Fourier (Grenoble) 62 (2012), no. 5, 1889–1942.

[11] Beresnevich, V.; Vaughan, R.; Velani, S. and Zorin, E. Diophantine approximation
on manifolds and the distribution of rational points: contributions to the conver-
gence theory. Int. Math. Res. Not. IMRN 2017, no. 10, 2885–2908.

[12] Brillinger, D. The analyticity of the roots of a polynomial as functions of the coef-
ficients. Math. Mag. 39 (1966), 145–147.

119
 REFERENCES 120

[13] Buterus, P.; Götze, F. and Hille, T. On small values of indefinite diagonal quadratic
forms at integer points in at least five variables. Trans. Amer. Math. Soc. Ser. B 9
(2022), 1–34.

[14] Buterus, P.; Götze, F.; Hille, T. and Margulis, G. Distribution of values of quadratic
forms at integral points. Invent. Math. 227 (2022), no. 3, 857–961.

[15] Bernšteı̆n, I and Gel’fand, S. Meromorphy of the function Pλ . (Russian) Funkcional.

Anal. i Priložen. 3, 1969, no. 1, 84–85.

[16] Berkesch, C. and Leykin, A. Algorithms for Bernstein-Sato polynomials and mul-
tiplier ideals. ISSAC 2010—Proceedings of the 2010 International Symposium on Sym-
bolic and Algebraic Computation, 99–106, ACM, New York, 2010.

[17] Bourgain, J. A quantitative Oppenheim theorem for generic diagonal quadratic

forms. Israel J. Math. 215 (2016), no. 1, 503–512.

[18] Briançon, J.; Granger, M. and Maisonobe, P. Sur le polynôme de Bernstein des
singularités semi-quasi-homogènes. Prépublication de l’Université de Nice, no.138,
Nov. 1986.

[19] Briançon, J.; Granger, M.; Maisonobe, P. and Miniconi, M. Algorithme de calcul
du polynôme de Bernstein : cas non dégénéré. Ann. Inst. Fourier39 (1989), no.3,
553–610.

[20] Briançon, J. and Maisonobe, P. Algorithms for Bernstein-Sato polynomials and

multiplier ideals. Caractérisation de l’existence du polynôme de Bernstein relatif.
Progress in Mathematics, vol. 134, 1996. 215–236. Acte du colloque de La Rabida.

[21] Chambert–Loir, A. and Tschinkel, Y. Igusa integrals and volume asymptotics in

analytic and adelic geometry. Confluentes Math., 2 (2010), no. 3, 351–429.

[22] Cluckers, R.; Comte, G.; Miller, D.; Rolin, J-P. and Servi, T. Integration of oscilla-
tory and subanalytic functions. Duke Math. J. 167 (2018), no. 7, 1239-1309.

[23] Cluckers, R.; Glazer, I. and Hendel, Y. A number theoretic characterization of

E-smooth and (FRS) morphisms: estimates on the number of Z{pk Z-points. To
appear in Algebra and Number Theory, available at arXiv:2103.00282.

[24] Cluckers, R.; Gordon, J. and Halupczok, I. Uniform analysis on local fields and
applications to orbital integrals. Trans. Amer. Math. Soc. Ser. B 5 (2018), 125-166.

[25] Cluckers, R. and Miller, D. Stability under integration of sums of products of real
globally subanalytic functions and their logarithms. Duke Math. J. 156 (2011), no.
2, 311–348.

[26] Colin de Verdière, Y. Nombre de points entiers dans une famille homothétique de
domains de R. nn. Sci. École Norm. Sup. (4) 10 (1977), no. 4, 559–575.
 REFERENCES 121

[27] Colzani, L.; Gigante, G. and Travaglini, G. Trigonometric approximation and a

general form of the Erdős Turán inequality. Trans. Amer. Math. Soc. 363 (2011), no.
2, 1101–1123.

[28] Coste, M. An introduction to semialgebraic geometry, 2002. Available at https://

gcomte.perso.math.cnrs.fr/M2/CosteIntroToSemialGeo.pdf.

[29] Denef, J. and van den Dries, L. p–adic and real subanalytic sets. Ann. of Math. (2)
128 (1988), no. 1, 79–138.

[30] van den Dries, L. Tame topology and o-minimal structures. London Mathematical
Society Lecture Note Series, 248. Cambridge University Press, Cambridge, 1998.

[31] van den Dries, L. and Miller, C. Geometric categories and o–minimal structures.
Duke Math. J. 84 (1996), no. 2, 497–540.

[32] van den Dries, L. and Miller, C. On the real exponential field with restricted ana-
lytic functions. Israel J. Math. 85 (1994), no. 1–3, 19–56.

[33] Eskin, A.; Margulis, G. and Mozes, S. Upper bounds and asymptotics in a quan-
titative version of the Oppenheim conjecture. Ann. of Math. (2) 147 (1998), no. 1,
93–141.

[34] Federer, H. Curvature measures. Trans. Amer. Math. Soc. 93 (1959), 418–491.

[35] Fouvry, E. and Waldschmidt, M. Number of integers represented by families of

binary forms. To appear in Acta Arith., available at arXiv:2206.03733.

[36] Gel’fand, I. and Shilov, G. Generalized functions. Vol. I: Properties and operations.
Translated by Eugene Saletan Academic Press, New York-London, 1964.

[37] Ghosh, A.; Gorodnik, A. and Nevo, A. Optimal density for values of generic
polynomial maps. Amer. J. Math. 142 (2020), no. 6, 1945–1979.

[38] Ghosh, A.; Kelmer, D. and Yu, S. Effective density for inhomogeneous quadratic
forms I:Generic forms and fixed shifts. Int. Math. Res. Not. IMRN 2022, no. 6,
4682–4719.

[39] Ghosh, A.; Kelmer, D. and Yu, S. Effective density for inhomogeneous quadratic
forms II: fixed forms and generic shifts. Preprint available at arXiv:2001.10990v3.

[40] Greenblatt, M. Fourier transforms of indicator functions, lattice point discrepancy,

and the stability of integrals. Math. Ann. 380 (2021), no. 3-4, 1959–1990.

[41] Gyoja, A. Theory of prehomogeneous vector spaces without regularity condition.

Publ. Res. Inst. Math. Sci. 27 (1991), no. 6, 861–922.
 REFERENCES 122

[42] Ho, W. How many rational points does a random curve have? Bull. Amer. Math.
Soc. (N.S.) 51 (2014), no. 1, 27–52.

[43] Hörmander, L. The analysis of linear partial differential operators. I. Distribution the-
ory and Fourier analysis. Second edition. Grundlehren der mathematischen Wis-
senschaften, 256. Springer–Verlag, Berlin, 1990.

[44] Huang, J. Rational points near planar curves and Diophantine approximation. Adv.
Math. 274 (2015), 490–515.

[45] Huang, J. The density of rational points near hypersurfaces. Duke Math. J. 169
(2020), no. 11, 2045–2077.

[46] Igusa, J-I. An introduction to the theory of local zeta functions. AMS/IP Studies in
Advanced Mathematics, 14. American Mathematical Society, Providence, RI. In-
ternational Press, Cambridge, MA, 2000.

[47] Jeanquartier, P. Développement asymptotique de la distribution de Dirac attachée

à une fonction analytique. C. R. Acad. Sci. Paris Sér. A–B 201 (1970), A1159–A1161.

[48] Kaiser, T. Growth of Log-Analytic Functions. Available at : arXiv:2301.02187

[math.LO].

[49] Kaiser, T. Integration of semialgebraic functions and integrated Nash functions.

Math. Z. 275 (2013), no. 1-2, 349–366.

[50] Kashiwara, M. D–modules and microlocal calculus. Translated from the 2000 Japanese
original by Mutsumi Saito. Translations of Mathematical Monographs, 217.
Iwanami Series in Modern Mathematics. American Mathematical Society, Prov-
idence, RI, 2003.

[51] Kashiwara, M. B-functions and holonomic systems. Rationality of roots of B-

functions. Invent. Math. 38 (1976/77), no. 1, 33–53.

[52] Kawakami, T. Equivariant differential topology in an o–minimal expansion of the

field of real numbers. Topology Appl. 123 (2002), no. 2, 323–349.

[53] Kelmer, D. and Yu, S. Values of random polynomials in shrinking targets.

Trans. Amer. Math. Soc. 373 (2020), no. 12, 8677–8695.

[54] Kleinbock, D. and Skenderi, M. Khintchine-type theorems for values of subhomo-

geneous functions at integer points. Monatsh. Math. 194 (2021), no. 3, 523–554.

[55] Kollár, J. Singularities of pairs. Algebraic geometry – Santa Cruz 1995, 221–287,
Proc. Sympos. Pure Math., 62, Part 1, Amer. Math. Soc., Providence, RI, 1997.
 REFERENCES 123

[56] Lavrik, A. The principal term of the divisor problem and the power series of the
Riemann zeta–function in a neighborhood of a pole (Russian). Number theory,
mathematical analysis and their applications. Trudy Mat. Inst. Steklov. 142 (1976),
165–173, 269.

[57] Lee, J. Introduction to smooth manifolds. Second edition. Graduate Texts in Mathe-
matics, 218. Springer, New York, 2013.

[58] Lichtin, B. Poles of |fpz, wq|2s and roots of the b-function. Ark. Mat. 27 (1989), no.
2, 283–304.

[59] Lion, J.-M. and Rolin, J.-P. Intégration des fonctions sous-analytiques et volumes
des sous-ensembles sous-analytiques. Ann. Inst. Fourier (Grenoble) 48 (1998), no. 3,
755–767.

[60] Loeser, F. Quelques conséquences locales de la théorie de Hodge. Ann. Inst. Fourier
(Grenoble) 35 (1985), no. 1, 75–92.

[61] Loeser, F. Une estimation asymptotique du nombre de solutions approchées d’une

équation p-adique. Invent. Math. 85 (1986), no. 1, 31–38.

[62] Loeser, F. Une estimation asymptotique du nombre de solutions approchées d’une

équation p-adique. Invent. Math. 85 (1986), no. 1, 31–38.

[63] Malgrange, B. Ideals of differentiable functions. Tata Institute of Fundamental Re-

search Studies in Mathematics, 3 Tata Institute of Fundamental Research, Bombay;
Oxford University Press, London, 1967.

[64] Malgrange, B. Intégrales asymptotiques et monodromie. Ann. Sci. École

Norm. Sup. (4) 7 (1974), 405–430 (1975).

[65] Malgrange, B. Polynômes de Bernstein-Sato et cohomologie évanescente. Analysis

and topology on singular spaces, II, III (Luminy, 1981), 243–267, Astérisque, 101–102,
Soc. Math. France, Paris, 1983.

[66] Margulis, G. Discrete subgroups and ergodic theory. Number theory, trace formulas
and discrete groups (Oslo, 1987), 377–398, Academic Press, Boston, MA, 1989.

[67] Randol, B. A lattice-point problem. Trans. Amer. Math. Soc. 121 (1966), 257–268.

[68] Saito, M. On microlocal b–function. Bull. Soc. Math. France 122 (1994), no. 2, 163–
184.

[69] Saito, M. On Real Log Canonical Thresholds. Available at arXiv:0707.2308.

[70] Sarnak, P. Values at integers of binary quadratic forms. Harmonic analysis and num-
ber theory (Montreal, PQ, 1996), 181–203, CMS Conf. Proc., 21, Amer. Math. Soc.,
Providence, RI, 1997.
 REFERENCES 124

[71] Scanlon, T. A proof of the André–Oort conjecture using mathematical logic [after
Pila, Wilkie and Zannier]. Séminaire Bourbaki: Vol. 2010/2011. Exposés 1027–1042.
Astérisque No. 348 (2012), Exp. No. 1037, ix, 299–315.

[72] Stein, E. Harmonic analysis: real-variable methods, orthogonality, and oscillatory inte-
grals. With the assistance of Timothy S. Murphy. Princeton Mathematical Series,
43. Monographs in Harmonic Analysis, III. Princeton University Press, Princeton,
NJ, 1993.

[73] Serre, J–P. Lectures on the Mordell–Weil theorem. Translated from the French and
edited by Martin Brown from notes by Michel Waldschmidt. Aspects of Mathe-
matics, E15. Friedr. Vieweg & Sohn, Braunschweig, 1989.

[74] Stewart, C. and Xiao, S. On the representation of integers by binary forms. Math.
Ann. 375 (2019), no. 1-2, 133–163.

[75] Tao, T. Topics in random matrix theory. Graduate Studies in Mathematics, 132. Amer-
ican Mathematical Society, Providence, RI, 2012.

[76] Tenenbaum, G. Introduction to analytic and probabilistic number theory. Third edition.
Translated from the 2008 French edition by Patrick D. F. Ion. Graduate Studies in
Mathematics, 163. American Mathematical Society, Providence, RI, 2015.

[77] Vanderkam, J. Values at integers of homogeneous polynomials. Duke Math. J. 97

(1999), no. 2, 379–412.

[78] Vaughan, R.C. and Velani, S. Diophantine approximation on planar curves: the
convergence theory. Invent. Math. 166 (2006), no. 1, 103-124.

[79] Watanabe, S. Algebraic geometry and statistical learning theory. Cambridge Mono-
graphs on Applied and Computational Mathematics, 25. Cambridge University
Press, Cambridge, 2009.

[80] Wilkie, A. O-minimal structures. Séminaire Bourbaki. Vol. 2007/2008. Astérisque

No. 326 (2009), Exp. No. 985, vii, 131–142 (2010).