Inhomogeneous Diophantine approximation on

arXiv:0809.3937v1 [math.NT] 23 Sep 2008

curves and Hausdorff dimension

Dzmitry Badziahin∗
York

Abstract

The goal of this paper is to develop a coherent theory for inhomogeneous Diophantine approximation
on curves in Rn akin to the well established homogeneous theory. More specifically, the measure theo-
retic results obtained generalize the fundamental homogeneous theorems of R.C. Baker (1978), Dodson,
Dickinson (2000) and Beresnevich, Bernik, Kleinbock, Margulis (2002). In the case of planar curves, the
complete Hausdorff dimension theory is developed.

1 Introduction
Throughout ψ : R+ → R+ := (0, +∞) denotes a decreasing function and will
be referred to as an approximation function. Let f = (f1 , . . . , fn ) : I → Rn
be a C (n) map defined on an interval I ⊂ R and λ : I → R be a function.
For reasons that will soon be apparent, the function λ will be referred to as
an inhomogeneous function. Let An (ψ, λ) be the set of x ∈ I such that the
inequality
ka · f(x) + λ(x)k < ψ(|a|) (1)
holds for infinitely many a ∈ Zn \ {0}, where k · k denotes the distance
to the nearest integer, |a| := max{|a1 |, . . . , |an |} and a · b stands for the
standard inner product of vectors a and b in Zn . In the special case when
ψ(h) = ψv (h) := h−v for some fixed positive v we will denote An (ψv , λ) by
∗
Supported by EPSRC Grant EP/E061613/1

1
An (v, λ). Furthermore, in the case when the inhomogeneous function λ is
identically zero we write An (ψ) for An (ψ, λ) and An (v) for An (v, λ).
By definition, An (ψ, λ) is the set of x ∈ I such that the corresponding
point f(x) lying on the curve
C := (f1 (x), f2 (x), . . . , fn (x)) : x ∈ I ⊂ Rn

(2)
satisfies the Diophantine condition arising from (1). Within the homogeneous
setup (λ ≡ 0), investigating the measure theoretic properties of An (ψ) dates
back to 1932 and a famous problem of Mahler [24]. The problem states
that when (2) is the Veronese curve given by (x, x2 , . . . , xn ) then An (v) is of
Lebesgue measure zero whenever v > n. Mahler’s problem was eventually
settled by Sprindzuk [28] in 1965 and subsequently Schmidt [27] extended the
result to the case of arbitrary planar curves (i.e. n = 2) with non-vanishing
curvature. These two major results led to what is currently known as the
(homogeneous) theory of Diophantine approximation on manifolds [14].
Diophantine approximation on manifolds has been an extremely active
research area over the past 10 years or so. Rather than describe the activity
in detail, we refer the reader to research articles [3, 4, 8, 10, 15, 23, 29] and the
surveys [6, 22, 25]. Nevertheless, it is worth singling out the pioneering work
of Kleinbock and Margulis [23] in which the fundamental Baker-Sprindzuk
conjecture is established. This has undoubtedly acted as the catalyst to the
works cited above which together constitute a coherent homogeneous theory
for Diophantine approximation on manifolds. The situation for the inhomo-
geneous theory is quite different. Indeed, the inhomogeneous analogue of the
Baker-Sprindzuk conjecture [11, 12] has only just been established in 2008.
The aim of this paper is to develop a coherent theory for inhomogeneous
Diophantine approximation on curves akin to the well established homoge-
neous theory. More precisely, Hausdorff measure theoretic statements for
the sets An (ψ, λ) are obtained. In particular, a complete metric theory is
established in the case of planar curves (n = 2). In short, the results consti-
tute the first precise and general statements in the theory of inhomogeneous
Diophantine approximation on manifolds.

1.1 Main results and corollaries

Before we proceed with the statement of the results, we introduce some useful
notation and recall some standard definitions.

2
The curve C given by (2) is called non-degenerate at x ∈ I if the Wronskian
(j)
w(f1′ , . . . , fn′ )(x) := det(fi (x))16i,j6n
does not vanish. We say that C is non-degenerate if it is non-degenerate
at almost every point x ∈ I. Given a set X ⊂ Rn and a real number
s > 0, Hs (X) will denote the s-dimensional Hausdorff measure of X and
dim X will denote the Hausdorff dimension of X. The latter is defined to
be the infimum over s such that Hs (X) = 0. For the formal definitions and
properties of Hausdorff measure and dimension see [21].

Lower bounds. Our first result enables us to deduce lower bounds for
dim An (v, λ) and represents an inhomogeneous version of the homogeneous
theorem established by Dodson and Dickinson [18]. Furthermore, even within
the homogeneous setup the result is stronger – it deals with Hausdorff mea-
sure rather than just dimension.

Theorem 1 Let f ∈ C (n) (I), ψ be an approximation function and λ ∈

C (2) (I). Assume that w(f1′ , . . . , fn′ )(x) 6= 0 for all x ∈ I. Then for any
0<s61
∞  s
s s
X ψ(q)
H (An (ψ, λ)) = H (I) if · q n = ∞. (3)
q=1
q

Note that whenever the sum in (3) diverges, the theorem implies that
s
H (An (ψ, λ)) > 0. In turn, it follows from the definition of Hausdorff dimen-
sion that dim(An (ψ, λ)) > s. In particular, it is easily verified that the sum
in (3) diverges whenever s < (n + 1)/(1 + τψ ), where
− log ψ(q)
τψ := lim inf
q→∞ log q
is the lower order of 1/ψ at infinity. Thus, Theorem 1 readily gives the
following inhomogeneous version of the Dodson-Dickinson lower bound [18]
for non-degenerate curves.

Corollary 1 Let f, ψ and λ be as in Theorem 1 with τψ > n. Then

n+1
dim An (ψ, λ) > . (4)
τψ + 1

3
 Note that in the case of ψ(q) = q −v we have that τψ = v as one would
expect. In the case of λ being a constant function and ψ(q) = q −v , the above
corollary has previously been obtained by the author in [1]. Bugeaud [16] has
established (4) within the context of approximation by algebraic integers; i.e.
in the case that C is the Veronese curve (xn , . . . , x) and λ : x → xn+1 .
In the case s = 1, the s-dimensional Hausdorff measure Hs is simply
one dimensional Lebesgue measure on the real line R . Thus, Theorem 1
trivially gives rise to a complete inhomogeneous analogue of the theorem of
Beresnevich, Bernik, Kleinbock and Margulis [8] in the case of non-degenerate
curves.

Corollary 2 Let f, ψ and λ be as in Theorem 1. Furthermore, suppose that

the associated curve given by (2) is non-degenerate. Then
∞
X
|An (ψ, λ)| = |I| if hn−1 ψ(h) = ∞.
h=1

Here and elsewhere |X| will stand for the Lebesgue measure of a measurable
subset X of R.

Upper bounds. It is believed that the lower bound for dim An (ψ, λ) given
in Corollary 1 is sharp. Establishing that this is the case, represents a chal-
lenging problem and in general is open even in the homogeneous setup –
it has only been verified in some special cases [2, 7, 13, 19]. In particular,
Baker [2] has settled the problem for planar curves within the homogeneous
setup. To the best of our knowledge, nothing seems to be known in the
inhomogeneous case. The following result, which is an inhomogeneous gen-
eralisation of Baker’s theorem, gives a complete theory for planar curves in
the inhomogeneous case.

Theorem 2 Let ψ : R+ → R+ be an approximation function with τψ > 2.

Let f1 , f2 , λ ∈ C (2) (I) be such that the associated curve C given by (2)n=2 is
non-degenerate everywhere except possibly on a set of Hausdorff dimension
less than τψ3+1 . Then
3
dim A2 (ψ, λ) = .
τψ + 1

4
2 Lower bounds: Proof of Theorem 1
The proof of Theorem 1 will rely on the ubiquitous systems technique as
developed in [9]. Essentially, the notion of ubiquitous system represents a
convenient way of describing the ‘uniform’ distribution of the naturally aris-
ing points (and more generally sets) from a given Diophantine approximation
inequality/problem - see [9, 20].

2.1 Ubiquitous systems in R

For the sake of simplicity, we introduce a restricted notion of ubiquitous
system, which is more than adequate for the applications we have in mind.
Let I0 be an interval in R and P = (Pα )α∈J be a family of resonant points
Pα of I0 indexed by an infinite set J. Next, let β : J → R+ : α 7→ βα be
a positive function on J. Thus the function β attaches a ‘weight’ βα to the
resonant point Pα . Assume that for every t ∈ N the set Jt = {α ∈ J : βα 6
2t } is finite.
Throughout, ρ : R+ → R+ will denote a function such that ρ(t) → 0 as
t → ∞ and it will be referred to as a ubiquitous function. Also, B(x, r) will
denote the ball (or rather the interval) centered at x with radius r.

Definition 1 Suppose that there exists a ubiquitous function ρ and an abso-

lute constant k > 0 such that for any interval I ⊆ I0

[
B Pα , ρ(2t ) ∩ I > k|I|.


lim inf
t→∞
α∈Jt

Then the system (P, β) is called locally ubiquitous in I0 with respect to ρ.

Let (P, β) be a ubiquitous system with respect to ρ and Ψ be an approx-

imation function. Let Λ(P, β, Ψ) be the set of points ξ ∈ R such that the
inequality
|ξ − Pα | < Ψ(βα )
holds for infinitely many α ∈ J. We will be making use of the following
lemma, which is an easy consequence of Corollary 2 (in the case s = 1) and
Corollary 4 (in the case s < 1) from [9].

5
Lemma 1 Let ψ be an approximation function and (P, β) be a locally ubiq-
uitous system with respect to ρ. Suppose there exists a real number λ ∈ (0, 1)
such that ρ(2t+1 ) < λ ρ(2t ) for all n ∈ N. Then,
∞
s s
X Ψ(2t )s
H (Λ(P, β, Ψ)) = H (I0 ) if = ∞.
t=1
ρ(2t )

2.2 Reduction of Theorem 1 to a ubiquity statement

First some notation. Let f = (f1 , . . . , fn ) be as in Theorem 1 and denote by
Fn the set of all functions

a0 + a1 f1 (x) + a2 f2 (x) + . . . + an fn (x)

where a0 , . . . , an are integer coefficients, not all zero. Given a function F ∈

Fn , the height H(F ) of F is defined as

H(F ) := max{|a1 |, . . . , |an |} .

For H > 1, let Fn (H) denote the subclass of Fn given by

Fn (H) = {F ∈ Fn : H(F ) 6 H}.

Given an inhomogeneous function λ, let Rλ = {α ∈ I : ∃ F ∈ Fn , F (α) +

λ(α) = 0}. Then, for α ∈ Rλ the quantity H(α) := min{H(F ) | F ∈
Fn , F (α) + λ(α) = 0} will be referred to as the height of α.
To illustrate the above notions, consider the following concrete example. Let
the functions fi (x) = xi be powers of x. Then Fn is simply the set of all non-
zero integral polynomials of degree at most n. Furthermore, if λ is identically
zero, then Rλ is simply the set of algebraic numbers in I of degree at most
n. On the other hand, if λ(x) = xn+1 then Rλ is simply the set of algebraic
integers in I of degree exactly n + 1.
The key to establishing Theorem 1 is the following ubiquity statement.

Proposition 1 The system (Rλ , H(α)) is locally ubiquitous in I with respect

to ρ(q) = q −n−1 .

We postpone the proof of Proposition 1 to the next section. We now establish

Theorem 1 modulo the proposition. Note that without loss of generality we

6
 (j)
can assume that I is a closed interval. Then, since the functions fi and
λ(k) , 0 6 j 6 n, 1 6 i 6 n, 0 6 k 6 2 are continuous we have that
(j)
∀x∈I |fi (x)| 6 C, |λ(k) (x)| 6 (5)

for some absolute constant C. Therefore we get

|F ′ (x)| 6 n CH(F ) := MH(F ).

Let α ∈ Rλ . Then, by definition there exists a function F ∈ Fn such that

F (α) + λ(α) = 0 and consider the interval
 
J := α − (2M)−1 H(F )−1 ψ(H(F )), α + (2M)−1 H(F )−1 ψ(H(F )) .

For any x ∈ J ∩ I, we have that

|F ′ (x) + λ′ (x)| 6 MH(F ) + C 6 2MH(F ). (6)

The latter inequality holds for all sufficiently large H(F ). Using the Mean
Value Theorem, we obtain

F (x) + λ(x) = F (α) + λ(α) + (F ′ (x2 ) + λ′ (x2 ))(x − α).

By (6), we get that |F (x) + λ(x)| 6 ψ(H(F )) and so it follows that whence

Λ(Rλ , H(α), (2M)−1H(F )−1ψ(H(F ))) ⊂ An (ψ, λ). (7)

In view of the divergent sum condition in Theorem 1, we have that

∞ ∞ s X ∞ s
((2M)−1 2−t ψ(2t ))s X t(n+1) ψ(2t )
 
X
n ψ(h)
−t(n+1)
≍ 2 t
> h = ∞.
t=1
2 t=1
2 h=1
h

Thus, Lemma 1 implies that

Hs (Λ(Rλ , H(α), (2M)−1H(F )−1 ψ(H(F )))) = Hs (I) .

This together with (7) implies that

Hs (An (ψ, λ)) = Hs (I) .

Modulo establishing Proposition 1, this completes the proof of Theorem 1.

7
2.3 Proof of Proposition 1
Without loss of generality we can assume that f1 (x) = x as otherwise we
can use the Inverse Function Theorem to change variables and ensure the
condition. Let Φ(Q, δ) denote the set of x ∈ I such that

|F (x)| < δQ−n (8)

for some F ∈ Fn (Q). We shall make use of the following lemma regarding
the measure of Φ(Q, δ).

Lemma 2 There is an absolute constant δ > 0 with the following property:

for any x0 ∈ I there is a neighborhood I0 ⊂ I of x0 such that for any interval
J ⊂ I0 there exists a sufficiently large Q1 > 0 such that for all Q > Q1 we
have |Φ(Q, δ) ∩ J| < |J|/2.

This lemma is a consequence of Theorem 2.1 in [8]. Since I is compact, it is

easy to see that I0 can be taken to be I.
1
Take Q1 and δ from Lemma 2. Define C1 := δ n+1 and fix some number
Q > C11 Q1 . Let ξ ∈ I\Φ(C1 Q, δ). The goal is to show that we can find
α ∈ Rλ such that

H(α) 6 K1 Q and |ξ − α| 6 K2 Q−n−1 (9)

where the constants K1 and K2 are independent from both Q and J. It

would immediately follow that for Q > C11 Q1 ,

|J| [
6 |J\Φ(C1 Q, δ)| 6 B(α, K2 Q−n−1 ) ∩ J
2
H(α)6K1 Q

and thus
[ |J|
B(α, Q−n−1) ∩ J > . (10)
H(α)6Q
2K2 K1n+1

Taking Q = 2t and setting ρ(H) := H −n−1 , inequality (10) implies that

(Rλ , H(α)) is locally ubiquitous in I with respect to ρ – the statement of
Proposition 1.

8
 We now proceed to establishing (9). Consider the system of inequalities
|an fn (ξ) + . . . + a1 f1 (ξ) + a0 | < Q−n ;
(
(11)
|a1 |, |a2 |, . . . , |an | 6 Q.
It defines a convex body in Rn+1 symmetric about the origin. Consider its
successive minima τ1 , . . . , τn+1 . By definition, τ1 6 τ2 6 . . . 6 τn+1 . Note
that τ1 > C1 . Indeed, otherwise we would have H 6 C1 Q and
|an fn (ξ) + . . . + a0 | 6 C1 Q−n = C1n+1 (C1 Q)−n 6 δH −n ,
a contradiction. By Minkowski’s theorem on successive minima [17],
τ1 · · · τn+1 6 1. Thus, we obtain the bound
τn+1 6 (τ1 · τ2 · · · τn )−1 < C1−n = C2
where C2 is an absolute constant depending only on Q. Finally, by the
definition of τn+1 , we obtain the set of n + 1 linearly independent functions
(j) (j) (j)
Fj (X) = an fn (X) + . . .+ a1 X + a0 , 1 6 j 6 n + 1 with integer coefficients
(j)
ai such that
|Fj (ξ)| 6 C2 Q−n ;
(

(j)
(12)
|ai | 6 C2 Q, i = 1, n.
Now consider the following system of linear equations

 θ1 F1 (ξ) + . . . + θn+1 Fn+1 (ξ) + λ(ξ)


= 0;
 Pn
θ1 F1′ (ξ) + . . . + θn+1 Fn+1
′
(ξ) + λ′ (ξ) = Q+ i=1 |Fi′ (ξ)|; (13)


(1) (n+1)

θ1 aj + . . . + θn+1 aj = 0, 2 6 j 6 n.


We transform this system in the following manner. Take each j-th row
(2 6 j 6 n), multiply it by fj′ (ξ) and subtract the result from the second
(1) (n+1)
row. As a result, the second row will have the form θ1 a1 + . . . + θn+1 a1 .
(1) (n+1)
Similarly we can transform the first row to the form θ1 a0 + . . . + θn+1 a0 .
(j)
Since the matrix (ai ), 0 6 i 6 n, 1 6 j 6 n + 1 is non-degenerate, the
system (13) has a unique solution θ1 , . . . , θn+1 . Choose integers t1 , t2 , . . . , tn
such that |ti − θi | < 1, 1 6 i 6 n + 1. Consider the function
F (X) = t1 F1 (X) + . . . + tn+1 Fn+1 (X) + λ(X)

= xn fn (X) + . . . + x1 X + x0 + λ(X),

9
 (1) (n+1)
where xi = t1 ai + . . . + tn+1 ai . By the first equation in (13), we obtain

|F (ξ)| < (n + 1)C2 Q−n = C3 Q−n .

Further, by the second equation in (13), we obtain |F ′ (ξ)| > Q and

|F ′ (ξ)| < Q + 2 n+1 ′

P
i=1 |Fi (ξ)| 6 Q + 2(n + 1)((n + 1)C2 · CQ)
= (1 + 2(n + 1)2 C2 · C)Q = C4 Q.

Now consider the coefficients xi . They are obviously integers. By the third
equation in (13), we get |xm | 6 (n+1)C2Q = C5 Q for all m > 2. The bounds
for x0 and x1 are given by

|x1 | 6 |F ′(ξ)| + |λ(ξ)| + ni=2 |xi fi′ (ξ)|

P

6 C4 Q + (n − 1)(n + 1)C2 Q · C + C 6 C6 Q

and
Pn
|x0 | 6 |F (ξ)| + |λ(ξ)| + i=1 |xi fi (ξ)|

6 C3 Q−n + (n − 1)(n + 1)|C2 · CQ + C6 CQ + C 6 C7 Q

for sufficiently large Q. Thus, for every ξ ∈ I\Φ(C1 Q, δ) there exists F (x) ∈
Fn such that 

 |F (ξ) + λ(ξ)| 6 C3 Q−n ;


Q 6 |F ′ (ξ) + λ′ (ξ)| 6 C4 Q; (14)



|H(F )| 6 max(C5 , C6 , C7 )Q.


It is easy to check that max{C5 , C6 , C7} = C7 . Hence, |H(F )| 6 C7 Q or

equivalently F ∈ Fn (C7 Q).
The next goal is to show that the function F (x) + λ(x) constructed above
has a root α satisfying conditions (9).

Lemma 3 Let σ(F ) be a set of all x ∈ I satisfying the following system of

inequalities: (
|F (x) + λ(x)| 6 C3 Q−n
Q 6 |F ′(x) + λ′ (x)| 6 C4 Q,

10
where F ∈ Fn (C7 Q). Let Q satisfy the condition
1
(n · C · C7 Q + C) · 2C3 Q−n−1 + C 6 Q.
2
Then for all x0 ∈ σ(F ) ∩ [a + 2C3Q−n−1 , b − 2C3Q−n−1 ] there exists a number
α ∈ (x0 − 2C3 Q−n−1 , x0 + 2C3 Q−n−1 ) such that F (α) + λ(α) = 0.

Proof. By the Mean Value Theorem,

F ′ (x) + λ′ (x) = F ′ (x0 ) + λ′ (x0 ) + (F ′′ (x1 ) + λ′′ (x1 ))(x − x0 ),

where x1 is some point between x and x0 . Taking x ∈ (x0 − 2C3 Q−n−1 , x0 +

2C3 Q−n−1 ) and using (5) we get F ′′ (x1 ) + λ′′ (x1 ) 6 n · C · C7 Q + C. Therefore
1
|(F ′′ (x1 ) + λ′′ (x1 ))(x − x0 )| 6 (n · C · C7 Q + C) · 2C3 Q−n−1 6 Q − C.
2
Finally we get that for all real x such that |x − x0 | 6 2C3 Q−n−1 the following
inequality is satisfied

|F ′ (x) + λ′ (x)| > |F ′(x0 )| − |λ′(x0 )| − |(F ′′(x1 ) + λ′′ (x))(x − x0 )|

> |F ′(x0 )|/2.

In particular it means that the function F ′ (x) + λ′ (x) has the same sign
within the given interval. Again, on using the Mean Value Theorem we get
that F (x) + λ(x) = F (x0 ) + λ(x0 ) + (F ′ (x2 ) + λ′ (x2 ))(x − x0 ), where x2 lies
between x and x0 . Set x = x0 ± 2C3 q −n−1 . Then

|(F ′ (x2 ) + λ′ (x2 ))(x − x0 )| > 2C3 Q−n−1 |F ′(x0 )|/2

> C3 Q−n > |F (x0 ) + λ(x0 )|.

Note that for the two different values of x the expression

(F ′ (x2 ) + λ′ (x2 )) · (x − x0 )

has different signs. Therefore the value of F (x) + λ(x) = F (x0 ) + λ(x0 ) +
(F ′ (x2 ) + λ′ (x2 ))(x − x0 ) has different signs at the two ends of the interval

[x0 − 2C3 Q−n−1 , x0 + 2C3 Q−n−1 ].

11
Thus the function F (x) + λ(x) has a root within this interval and thereby
completes the proof of Lemma 3.
⊠

In view of Lemma 3, we have that for all ξ satisfying system (14) there
exists α with H(α) 6 C7 Q such that

F (α) + λ(α) = 0

and
|ξ − α| < 2C3 Q−n−1 .

Finally, for all ξ ∈ I\Φ(C1 Q, δ) we have constructed a function F (x) ∈ Fn

such that (14) is satisfied. Therefore, by taking K1 = C7 and K2 = 2C3 ,
we find a number α ∈ Rλ satisfying (9). This completes the proof of the
Proposition 1.

3 Upper bounds: Proof of Theorem 2

3.1 Preliminary notes
First of all note that, by Corollary 1, it suffices to establish the lower bound
3
dim A2 (ψ, λ) 6 . (15)
τψ + 1

Note that there is nothing to prove if τψ = 2. Thus, without loss of generality

we can assume that τψ > 2. Further, the definition of τψ readily implies that
for any v < τψ we have that ψ(q) ≪ q −v for all sufficiently large q. It follows
that for any v < τψ we have that A2 (ψ, λ) ⊂ A2 (v, λ). Therefore, (15) will
follow if we consider the special case of ψ(q) = q −v with 2 < v < τψ and
let v → τψ . Therefore, from now on we fix a v > 2 and concentrate on
establishing the bound
3
dim A2 (v, λ) 6 . (16)
v+1

12
3.2 Auxiliary lemmas
As in the proof of Proposition 1, there is no loss of generality in assuming
that f1 (x) = x. Then we simply denote f2 (x) by f (x). With the aim of
establishing Theorem 2 we fix v > 2. By the conditions of Theorem 2, we
3
have that f ′′ (x) 6= 0 for all x except a set of Hausdorff dimension 6 v+1 .
Using the standard arguments – see [5] – we can assume without loss of
generality that
c1 6 |f ′′ (x)| 6 c2 for all x ∈ I, (17)
where c1 , c2 are positive constants.

Lemma 4 (Pyartly [26]) Let δ, ν > 0 and I ⊂ R be some interval. Let

φ(x) ∈ C n (I) be a function such that |φ(n) (x)| > δ for all x ∈ I. Then there
exists a constant c(n) which depends only on n, such that
 ν  n1
|{x ∈ I : |φ(x)| < ν}| 6 c(n) .
δ

Before stating the next lemma recall that F2 is the set of all functions
of the form a0 + a1 x + a2 f (x), where a0 , a1 , a2 are integers not all zero;
H = H(F ) = max{|a1 |, |a2 |}.

Lemma 5 There are constants C1 > 0 and ǫ0 > 0 such that for all F ∈ F2
and any subinterval J ⊂ I of length |J| 6 ǫ0 at least one of the following
inequalities is satisfied for all x ∈ J:

|F ′ (x) + λ′ (x)| > C1 H(F ) or |F ′′ (x) + λ′′ (x)| > C1 H(F ).

Proof. For the case of λ(x) ≡ 0 this is proved in [5, Lemmas 5, 6]. To finish
the proof in inhomogeneous case it is sufficient to note that |λ′ (x)| ≪ 1 and
|λ′′ (x)| ≪ 1.
⊠
In what follows without loss of generality we can assume that |I| 6 ǫ0 –
see [5] for analogues arguments.

Lemma 6 Fix some 0 6 δ 6 1 and a positive number H. Denote by N(δ)

the number of triples (a0 , a1 , a2 ) ∈ Z3 satisfying max{|ai | : i = 0, 1, 2} 6 H

13
such that there exists a solution x ∈ I to the system
(
|F (x) + λ(x)| 6 H −v
(18)
|F ′ (x) + λ′ (x)| 6 H δ .

Then for v > 0, N(δ) ≪ H 1+δ .

Proof. Since |λ′ (x)| ≪ 1, |λ(x)| ≪ 1 and δ > 0, we have that (18) implies
the following system (
|F (x)| ≪ H δ
(19)
|F ′(x)| ≪ H δ .

Subtracting the second inequality of (19) multiplied by x from the first in-
equality of (19) gives
(
|a0 + a2 (f (x) − xf ′ (x))| ≪ H δ
(20)
|a1 + a2 f ′ (x)| ≪ H δ .

If |a1 | = H then we have 2H + 1 possibilities for a2 . By (17), for each

fixed pair (a1 , a2 ) the interval of x satisfying the second inequality of (20) is
of length O(H δ a−1 ′
2 ). Therefore, a2 (f (x) − xf (x)) may vary on an interval
of length O(H δ ) only. Hence, for every fixed pair (a1 , a2 ) we have O(H δ )
possibilities for a0 . Thus, we have O(H δ+1 ) triples (a0 , a1 , a2 ) with |a1 | = H.
Consider the case |a2 | = H. Note that by (17), f ′ (x) is strictly mono-
tonic and finite. Therefore one can change variables by setting t = −f ′ (x);
f (x) − xf ′ (x) = g(t). Note that, by (17), the variable t belongs to some
finite interval J. Furthermore, the function g(t) is bounded, continuously
differentiable on J and |g ′ (t)| ≪ 1. Therefore, the system (20) transforms to

a0
+ g(t) ≪ H δ−1 ;

δ
 |a0 + a2 g(t)| ≪ H ;

 a2
 

|a1 − a2 t| ≪ H δ ; =⇒ a1
− t ≪ H δ−1 ;
  a2
t ∈ J.
 

 t ∈ J.

Note that
 
a1
g = g(t + ∆) = g(t) + ∆g ′ (ξ) = g(t) + O(H δ−1 ) ,
a2

14
 a1
where ∆ = a2
− t. Hence all solutions of the system are also solutions of the
inequality  
a1 a0
g + ≪ H δ−1 .
a2 a2
One can easily check that for |a2 | = H the number of integer solutions of
this inequality is not greater than CH 1+δ for some constant C. Therefore
N(δ) ≪ H 1+δ and the proof is complete. ⊠

Lemma 7 Consider the plane defined by the equation Ax + By + Cz = D

where A, B, C, D are integers with (A, B, C) = 1. Then √ the area S of any
1
triangle on this plane with integer vertices is at least 2 A2 + B 2 + C 2 .

Proof. Denote by x, y and z some points on the considered plane not all
lying on the same line. Take one more integer point v somewhere outside
the plane. We now calculate the volume V of the tetrahedron xyzv.
On one hand the volume of every tetrahedron with integer vertexes is at
least 61 . Therefore V > 16 .
On the other hand, V = 13 Sh, where S is the area of the triangle xyz and
h is the distance between v and the plane. Therefore,
1 1 2
6 V = Sh ⇐⇒ 6 S.
6 3 h
Let v = (α, β, γ). Then
|Aα + Bβ + Cγ − D| 1
h= √ >√ ,
2
A +B +C2 2 A + B2 + C 2
2

√
since α, β and γ are integers and h > 0. Thus, S > 12 A2 + B 2 + C 2 as
required. ⊠

3.3 Proof of Theorem 2

3
Let σ := v+1 be the required bound in (16). The strategy of the proof is to
construct a collection of coverings Di = {dij : j ∈ J} of A2 (v, λ) by intervals
dij such that for any ǫ > 0
X
|dij |σ+ǫ → 0 as i → ∞.
j∈J

15
The bound (16) will then follow from the definition of Hausdorff dimension.
Note that A2 (v, λ) can be represented in one of the following forms
∞ [
\ ∞
A2 (v, λ) = A(a0 , a1 , a2 ) and (21)
n=1 H=n

∞ [
\ ∞
A2 (v, λ) = B(t),
n=1 t=n

where A(a0 , a1 , a2 ) is the set of x ∈ I satisfying

|a0 + a1 x + a2 f (x) + λ(x)| < H −v (22)
for the particular triple (a0 , a1 , a2 );
[
B(t) = A(a0 , a1 , a2 ); H = max{|a1 |, |a2 |}.
2t−1 6H<2t

Therefore for any n ∈ N the collection of sets A(a0 , a1 , a2 ) with H > n is a

covering of A2 (v, λ). Analogously for any n ∈ N the collection of B(t) with
t > n is a covering of A2 (v, λ).
Fix some positive small number ǫ. Divide every set A(a0 , a1 , a2 ) into three
subsets:
A1 (a0 , a1 , a2 ) = {x ∈ A(a0 , a1 , a2 ) : |F ′ (x) + λ′ (x)| > H 1−ǫ }; (23)
n 2−v
o
A2 (a0 , a1 , a2 ) = x ∈ A(a0 , a1 , a2 ) : H 3 < |F ′ (x) + λ′ (x)| 6 H 1−ǫ ; (24)
n 2−v
o
A3 (a0 , a1 , a2 ) = x ∈ A(a0 , a1 , a2 ) : |F ′ (x) + λ′ (x)| 6 H 3 . (25)
(1)
For any of these collections we can construct the associated sets A2 (v, λ),
(2) (3)
A2 (v, λ) and A2 (v, λ) analogously to A2 (v, λ) – see (21). One can easily
check that
(1) (2) (3)
A2 (v, λ) = A2 (v, λ) ∪ A2 (v, λ) ∪ A2 (v, λ).
Therefore it is sufficient to prove (16) for A2 (v, λ) replaced by either of these
subsets.
(1)
The set A2 (v, λ). Since |λ(x)| ≪ 1, we have that
|a1 + a2 f ′ (x) + λ(x)| > H 1−ǫ =⇒ |a1 + a2 f ′ (x)| ≫ H 1−ǫ .

16
Since |f ′′(x)| > d for all x ∈ I, we have that a1 + a2 f ′ (x) is a monotonic
function. Therefore the set of x ∈ I such that |a1 + a2 f ′ (x)| > H 1−ǫ is a
union of at most two intervals. For one interval we have that

a1 + a2 f ′ (x) ≪ −H 1−ǫ

and for the other we have that

a1 + a2 f ′ (x) ≫ H 1−ǫ .

We see that the sign of F ′ (x)+λ′(x) on each of these intervals doesn’t change.
Therefore F (x)+λ(x) is monotonic on them, where F (x) = a0 +a1 x+a2 f (x).
Thus for sufficiently large H the set A1 (a0 , a1 , a2 ) is a union of at most 2
intervals (note that it can be empty, i.e. be a union of empty intervals).
Using Lemma 4 and inequality in (23) we get that the length of each
interval is ≪ H −v−1+ǫ .
(1)
We will use the following cover of A2 (v, λ):
∞
[
Cn = A1 (a0 , a1 , a2 ).
H=n

Note that for a fixed H the number of different pairs (a1 , a2 ) is no greater
than 4H. By (22) there are O(H) possibilities for a0 if (a1 , a2 ) are fixed.
Therefore an appropriate s-volume sum for Cn will be
∞
X ∞
X
2
C≪ H ·H s(ǫ−1−v)
= H 2−s(1+v−ǫ) .
H=n H=n

This sum tends to zero as n → ∞ in case of 2 − s(1 + v − ǫ) < −1, that is

3
s > 1+v−ǫ . Thus,
(1) 3
dim(A2 (v, λ)) 6 . (26)
1+v−ǫ

(2)
The set A2 (v, λ). Here we have the inequality |F ′ (x) + λ′ (x)| 6 H 1−ǫ .
Therefore Lemma 5 implies

∀x ∈ A2 (a0 , a1 , a2 ), |F ′′(x) + λ′′ (x)| ≫ H. (27)

17
In other words |a2 f ′′ (x) + λ′′ (x)| ≫ H. This implies that |a2 | ≫ H. Note
that (27) is also true in the case of x ∈ A3 (a0 , a1 , a2 ).
Let δ be an arbitrary number in (0, 1]. Consider the set
∞ [
\ ∞
Aδ (v, λ) = Aδ (a0 , a1 , a2 ), (28)
n=1 H=n

where Aδ (a0 , a1 , a2 ) is the set of x ∈ A2 (a0 , a1 , a2 ) with the following property

1
H 1− 3 (v+1)δ < |F ′ (x) + λ′ (x)| 6 H 1−δ . (29)

We have that |F ′′ (x) + λ′′ (x)| ≫ H. Therefore, the set Aδ (a0 , a1 , a2 )

consists of at most 4 intervals. Consider the following cover Cn for Aδ (v, λ):
∞
[
Cn = Aδ (a0 , a1 , a2 ).
H=n

By Lemma 6, for a fixed H there exist only O(H 2−δ ) nonempty sets
Aδ (a0 , a1 , a2 ). By Lemma 4 the length of each interval in Aδ (a0 , a1 , a2 ) is
1
bounded by H −v−1+ 3 (v+1)δ . Therefore the corresponding s-volume sum for
Cn is bounded by
∞
H 2−δ · H s(−v−1+ 3 (v+1)δ) .
X 1
(30)
H=n
3
If s > v+1
then the exponent of H is equal to
 
1
2 − δ + s −v − 1 + (v + 1)δ < 2 − δ − 3 + δ = −1.
3
3
Hence for s > v+1 the right hand side of (30) tends to 0 as n → ∞. It follows
3
that dim(Aδ (v, λ)) 6 v+1 for any δ ∈ [0, 1].
For simplicity denote by k the quantity 31 (v + 1). Note that k > 1. For
(2)
l ∈ N consider the set A2 (v, λ) as a union
l
(2)
[
A2 (v, λ) = Aδi (v, λ) ∪ Aδ∗ (v, λ), (31)
i=1

18
where δ1 = ǫ, δi+1 = kδi , δ ∗ = 1. Since k > 1 we have that δi → ∞ as i → ∞.
Therefore there exists a natural number l which depends on ǫ only such that
δl+1 > 1 and δl 6 1. This proves (31).
Since the Hausdorff dimension of each set Aδi (v, λ) and Aδ∗ (v, λ) appear-
3 (2) 3
ing in (31) is not greater than v+1 , we get that dim(A2 (v, λ)) 6 v+1 .

(3)
The set A2 (v, λ). Consider the set
[
B3 (t) := A3 (a0 , a1 , a2 )
2t−1 6H<2t

2−v
and let δ := 3
.
Recall that for all x ∈ A3 (a0 , a1 , a2 ) we have |F ′′ (x)+λ′′ (x)| ≫ H. There-
fore, by Lemma 4, we have that the length of each interval in A3 (a0 , a1 , a2 )
with H ≍ 2t is not greater than

(H −v /H) 2 ≪ 2−t( ).
1 v+1
2

Fix a sufficiently small positive number ǫ1 . Let c = 1 + ǫ1 . For every

t divide the interval I into 2ct equal subintervals of length 2−ct |I| ≪ 2−ct .
These subintervals are divided into two classes:
 3 
• Class I intervals. They include at most O 2t( 2 −c) segments from B3 (t).
• Class II intervals. They include those which are not in class I.
According to this classification consider the sets
∞ [
\ ∞ [
AI (v, λ) = B3 (t) ∩ J;
n=1 t=n class I intervals J

\ ∞
∞ [ [
AII (v, λ) = B3 (t) ∩ J.
n=1 t=n class II intervals J

It follow that
(3)
A2 (v, λ) = AI (v, λ) ∪ AII (v, λ).
The required upper bound for AI (v, λ) will follow on showing the following
lemma.

19
 3
Lemma 8 dim(AI (v, λ)) 6 v+1
(1 + ǫ1 ).
 3 
Proof. Consider a class I interval J. We have at most O 2t( 2 −c) segments
3
from B3 (t) on it. Therefore there are not greater than O(2 2 t ) intervals from
B3 (t) lying inside class I intervals. Consider the following cover of AI (v, λ):
∞
[ [
Cn := B3 (t) ∩ J.
t=n class I intervals J
3
Its v+1
(1 + ǫ1 )-volume is bounded by
∞ ∞ ∞
3(ǫ1 +1)
2 )
−t( v+1
2( 2 − 2 (ǫ1 +1))t =
X 3
X 3 3 X 3
·
2 2
t
·2 v+1 = 2− 2 ǫ 1 t .
t=n t=n t=n

It obviously tends to zero as n → ∞. This finishes the proof of the lemma.

⊠

Let J be a class II interval and F (x) = a0 + a1 x + a2 f (x) ∈ F2 with

2t−1 6 H(F ) < 2t and A3 (a0 , a1 , a2 ) ∩ J 6= ∅. Then
|F (x0 ) + λ(x0 )| ≪ 2−vt and |F ′ (x0 ) + λ′ (x0 )| ≪ 2δt
for some F ∈ F2 and x0 ∈ J. Then for all x ∈ J we have
|F ′(x) + λ′ (x)| = |(F ′ + λ′ )(x0 ) + (x − x0 )(F ′′ + λ′′ )(ξ)| ≪ 2δt + 2(1−c)t .
|F (x)+λ(x)|= |(F +λ)(x0 ) + (x − x0 )(F ′ + λ′ )(x0 ) + (x − x0 )2 (F ′′ + λ′′ )(ξ)|

≪ 2−vt + 2(δ−c)t + 2(1−2c)t .

Choose a sufficiently small ǫ1 > 0 such that
v > 2 + 3ǫ1 . (32)
Then we have
2δt < 2(1−c)t and 2(δ−c)t < 2(1−2c)t .
One can see that 2−vt is always less than the other summands 2(δ−c)t and
2(1−2c)t . Hence in the case of (32) we get the inequalities
|F (x) + λ(x)| ≪ 2(1−2c)t , (33)
|F ′ (x) + λ′ (x)| ≪ 2(1−c)t (34)
for all x ∈ J.

20
Lemma 9 For every fixed J as above all points ~a = (a0 , a1 , a2 ) ∈ Z3 such
that A3 (a0 , a1 , a2 ) ∩ J 6= ∅ lie on a single affine plane.

Proof. Suppose that there exist four integer points ~a, ~b, ~c, d~ not lying on the
same plane such that A3 (~a) ∩ J 6= ∅, A3 (~b) ∩ J 6= ∅, A3 (~c) ∩ J 6= ∅ and
~ ∩ J 6= ∅. It means that the points ~a, ~b, ~c, d~ form a tetrahedron with
A3 (d)
integer vertexes. Therefore its volume is at least 61 .
On the other hand all of these four points must lie inside a paral-
lelepiped R formed by the inequalities (33), (34) and |a2 | < H for a fixed
x ∈ J. The volume of this figure is bounded by
V ≪ 2 · 2t(1−2c) · 2 · 2t(1−c) · 2 · 2t · D −1 ≪ 2t(3−3c) · D −1 ,
where D is the determinant of the matrix
 
1 x f (x)
 0 1 f ′ (x) 
0 0 1
i.e. D = 1. Since c > 1 we have V = o(1) contrary to V > 1/6. The proof is
complete.
⊠
Let the plane from Lemma 9 have the form Ax + By + Cz = D. We
evaluate the intersection area of this plane with parallelepiped R specified
in the proof of the lemma. In order to do this let us consider the body P∆
given by the inequalities

 |F (x) + λ(x)| 6 2t(1−2c) ;
|a | 6 2t ; (35)
 2
|Aa0 + Ba1 + Ca2 − D| 6 ∆,
where ∆ > 0 is a positive parameter. Here a0 , a1 , a2 are viewed as real
variables. The volume of P∆ can be expressed in two different ways. Firstly,
since the determinant of system (35) is B − Ax, we have that
8 · 2t(2−2c) · ∆
V (P∆ ) = . (36)
|B − Ax|
Secondly, V (P∆ ) = S · h where S is the area of the edges defined by the third
inequality of (35) and h is a distance between these edges. That is
2∆
V (P∆ ) = S · √ . (37)
A2 + B 2 + C 2

21
Hence on combining (36) and (37) we obtain that
√
2t(2−2c) · A2 + B 2 + C 2
S≍ . (38)
|B − Ax|
Note that S is the area of the intersection of the plane Aa0 +Ba1 +Ca2 −D = 0
with the figure defined by the first two inequalities of (35). Therefore the
intersection area of this plane with the parallelepiped is not greater than S.
Note that all points a should lie inside this intersection and (38) gives an
estimate for its area.

Case (i): We consider intervals J of type II such that not all points a
associated with J lie on the same line. By Lemma 7, we get that the number
of such points on a fixed interval J is bounded by
√
2t(2−2c) · A2 + B 2 + C 2 √ 2 2t(2−2c)
N≪ / A + B2 + C 2 = . (39)
|B − Ax| |B − Ax|
Since J is not a class I interval we get that for all x ∈ J,
1
|B − Ax| ≪ 2t( 2 −c) . (40)

Similarly to (35) we consider two more systems of inequalities:

  ′
 |F (x) + λ(x)| 6 2t(1−2c) ;  |F (x) + λ′ (x)| 6 2t(1−c) ;
|Aa0 + Ba1 + Ca2 − D| 6 ∆; and |a2 | 6 2t ;
 ′
|F (x) + λ′ (x)| 6 2t(1−c) |Aa0 + Ba1 + Ca2 − D| 6 ∆.


Analogously we get additional bounds for N, namely

2t(2−3c) 2t(2−c)
N≪ and N≪ , (41)
|T | |A|
where
 
1 x f (x)
T = det  A B C  = f ′ (x)(B − Ax) − (C − Af (x)).
0 1 f ′ (x)
Since J is a class II interval then
1
|f ′(x)(B − Ax) − (C − Af (x))| ≪ 2t( 2 −2c) .

22
This result with (40) implies
1
|C − Af (x)| ≪ 2t( 2 −c) . (42)

The second inequality in (41) together with the fact that J is a class II
1
interval implies |A| ≪ 2 2 t .
Fix A. Denote by M(A) the number of possible integer triples (A, B, C)
which can be the coefficients of a plane corresponding to some class II interval.
It follows from (40) and (42) that

M(A) 6 |I| · |A| + 1 ≪ |A|.

B
In fact it is the number of fractions A
in the interval I. Parameter C is
uniquely defined by A and B.
Suppose there exist two class II intervals J1 and J2 with the same coeffi-
cients (A, B, C) of appropriate plane. Applying inequality (40) we get
1 1
∀x ∈ J1 , |B − Ax| ≪ 2t( 2 −c) ; 2t( 2 −c)

t( 12 −c)
1 ⇒ |A(x−y)| ≪ 2 ⇒ |x−y| ≪ .
∀y ∈ J2 , |B − Ay| ≪ 2t( 2 −c) . |A|

Therefore for a fixed (A, B, C) the number x can lie only in the interval of the
1
2t( 2 −c)
length |A|
. Finally we get that the number of class II intervals associated
1
22t
with the triple (A, B, C) is at most |A|
.
We will use the following cover for the AII (v, λ):
∞
[ [
Cn = B3 (t) ∩ J;
t=n J are class II intervals

Using the second inequality in (41) to estimate the number of intervals in

B3 (t) ∩ J we get that the s-volume sum for this cover is bounded by
∞ X 2 21 t 2t(2−c) ∞ X 1
3
−st( v+1 v+1
X X
)
C= · ·2 2 = 2t( 2 −ǫ1 −s( 2 )) 2
,
t=n
|A| |A| t=n
|A|
(A,B,C) (A,B,C)

where (A, B, C) run through possible coefficients of planes corresponding to

23
type II intervals under consideration. Transforming this series we get
t
∞ 22 ∞
X
t( 32 −ǫ1 −s( v+1 ))
X 1 X 3 v+1
2 2 ≪ t · 2t( 2 −ǫ1 −s( 2 ))
|A| t=n
t=n |A|=1 (43)
∞
3 v+1
X
≪ 2t( 2 −s( 2 )) .
t=n

3
If s > v+1
then this series obviously tends to zero as n → ∞.

Case (ii): We consider intervals J of type II such that all points a associated
with J lie on the same line L. Fix such an interval J. Represent this line in
a form:
α + tβ
where α = (α0 , α1 , α2 ) is an integer point on L, β = (β0 , β1 beta2 ) is a vector
between the nearest integer points on L and t is an arbitrary real number.
Then all the vectors (a0 , a1 , a2 ) associated with J are of the form

a0 = α0 + kβ0 , a1 = α1 + kβ1 , a2 = α2 + kβ2 ,

where αi , βi are fixed and k ∈ Z vary. Since J is of class II, there are at least
3
2t( 2 −c) different values k. For each vector (a0 , a1 , a2 ) under consideration we
have that |a0 | ≪ 2t . Hence taking values of |a0 | for two different vectors for
J and subtracting one value from another we get
1
|β0 (k1 − k2 )| ≪ 2t ⇒ |β0 | 6 2t(c− 2 ) . (44)

Similarly we obtain the same inequalities for β1 and β2 .

Now consider inequalities (33) and (34) for the same two vectors. Again
subtracting one inequality from the other we get

|(k1 − k2 )(β0 + β1 x + β2 f (x))| 6 2 · 2t(1−2c) ;

(
(45)
|(k1 − k2 )(β1 + β2 f ′ (x))| 6 2 · 2t(1−c) .

3
Since there are at least 2t( 2 −c) different values k, we can ensure that
3
|k1 − k2 | > 2t( 2 −c) for some k1 , k2 . Dividing these inequalities by |k1 − k2 |

24
and changing variables as in Lemma 6 give the system
( t
|β0 + β2 g(y)| 6 2 · 2− 2 ;
t
|β1 + β2 y| 6 2 · 2− 2
where y = f ′ (x) and g(y) = f (x) − xf ′ (x). Using (44) we get H = max{|β0 |,
1
|β1 |, |β2 |} ≪ 2t(c− 2 ) . Substituting this into the system we get
 1 1
 |β0 + β2 g(y)| ≪ H 1−2c = H − 1+2ǫ1 ;
(46)
1 1
− 1+2ǫ
|β1 + β2 y| ≪ H =H .
 1−2c 1

For a fixed value of β2 the number of possibilities for β1 is O(|β2|). For a

fixed β1 and β2 the number of possibilities for β0 is O(1). Denote by K(β2 )
the number of solutions (β0 , β1 , β2 ) of (46), where β2 is fixed. Then we get
that K(β2 ) ≪ |β2 |.
Suppose that for two different intervals J1 and J2 the parameters β0 , β1
and β2 coincide. Using the second inequality in (46) we get
1
− 1+2ǫ
|β2 (y1 − y2 )| ≪ H 1

where y1 ∈ f ′ (J1 ) and y2 ∈ f ′ (J2 ). Since for all x ∈ I, f ′ (x) > d > 0 the
inequality can be transformed to the form
1 1
− 1+2ǫ
2− 2 t H 1
|x1 − x2 | ≪ ≪ .
|β2 | |β2 |
Therefore the number of class B intervals J with parameters β0 , β1 , β2 is not
greater than
1
2t(c− 2 )
. (47)
|β2 |
t
Further, since |α2 + kβ2 | 6 2t there are at most |β22 | intervals inside J ∩ B3 (t).
Using (47) we have the following upper bound for s-volume sum.
∞ X 2t(c− 21 ) 2t ∞
X v+1
X 3 v+1
X 1
C = · · 2−st( 2 ) = 2t( 2 +ǫ1 −s( 2 ))
t=n
|β2 | |β2 | t=n
|β2 |2
(β0 ,β1 ,β2 ) (β0 ,β1 ,β2 )
∞
X 3 v+1
X K(β2 )
≪ 2t( 2 +ǫ1 −s( 2
))

t=n 1
|β2 |2
|β2 |62t(c− 2 )
∞
3 v+1
X
≪ t · 2t( 2 +ǫ1 −s( 2
))
.
t=n

25
Using the same arguments as in the case when (a0 , a1 , a2 ) lie on a plane we
obtain that ∞
3 v+1
X
C≪ 2t( 2 +2ǫ1 −s( 2 )) .
t=n

Combining this series with (43) we get an estimate

3 + 4ǫ1
dim(AII (v, λ)) 6 .
v+1

Therefore finally for v > 2 + 3ǫ1 we get that

(1) (2)
dim(A2 (v, λ))= dim(A2 (v, λ) ∪ A2 (v, λ) ∪ AI (v, λ) ∪ AII (v, λ))
 
3 3 + 4ǫ1 3
6 max , , (1 + ǫ1 ) .
1+v−ǫ v+1 v+1

Since ǫ and ǫ1 can be made arbitrary small then all values in the max-
3
imum can be made arbitrary close to v+1 , thus implying (15) and thereby
completing the proof of Theorem 2.

Acknowledgements. DB is grateful to Victor Beresnevich, Vasili Bernik and

Sanju Velani for introducing me to the wonderland of metrical Diophantine
approximation and for their numerous helpful discussions.

References
[1] D. A. Badziahin. Inhomogeneous approximations and lower bounds for the Hausdorff
dimension Vestsi Nats. Akad. Navuk Belarusi Ser. Fi z.-Mat. Navuk 126, No. 3, pp.
32–36, 2005. (in Russian)
[2] R.C. Baker. Dirichlet’s Theorem on Diophantine Approximation. Math. Proc. Cam.
Phil. Soc., V. 83, pp. 37 – 59, 1978.
[3] V. Beresnevich. On approximation of real numbers by real algebraic numbers. Acta
Arith. 90 (1999), no.2, pp. 97–112.
[4] V. Beresnevich. A Groshev Type Theorem for Convergence on Manifolds. Acta Math.
Hung., V. 94, pp. 99 – 130, 2002.
[5] V. Beresnevich, V. Bernik. On a metrical theorem of W.M. Schmidt. Acta Arithm.,
V. 75.3, pp. 219 – 233, 1996.

26
 [6] V. Beresnevich, V. I. Bernik, and M. M. Dodson. Regular systems, ubiquity and
Diophantine approximation. A panorama of number theory or the view from Baker’s
garden (Zürich, 1999). CUP. 2002, pp. 260–279.
[7] V. Beresnevich, V. Bernik, M. Dodson. On the Hausdorff dimension of sets of well-
approximable points on nondegenerate curves. Dokl. Nats. Akad. Nauk Belarusi. 46
(2002), no. 6, 18–20.
[8] V. Beresnevich, V. I. Bernik, D. Y. Kleinbock, G. A. Margulis. Metric Diophan-
tine approximation: The Khintchine–Groshev theorem for non-degenerate manifolds..
Moscow Math. J. 2, No. 2, 203–225, 2002.
[9] V. Beresnevich, H. Dickinson, S.L. Velani. Measure Theoretic Laws for Limsup sets
Memoirs of the AMS, V. 179, N 846, 2006.
[10] V. Beresnevich, H. Dickinson, S.L. Velani. Diophantine approximation on planar
curves and the distribution of rational points. Ann. of Math. (2) 166 (2007), no. 3,
pp. 367–426.
[11] V. Beresnevich, S.L. Velani. Simultaneous inhomogeneous diophantine approximation
on manifolds. Preprint, 2007. http://arxiv.org/abs/0710.5685
[12] V. Beresnevich, S.L. Velani. An inhomogeneous transference principle and Diophan-
tine approximation. Preprint, 2008. http://arxiv.org/abs/0802.1837
[13] V. I. Bernik. An application of Hausdorff dimension in the theory of Diophantine
approximation. Acta Arith. 42 (1983), no. 3, 219–253.
[14] V. I. Bernik, M. M. Dodson. Metric Diophantine approximation on manifolds. Cam-
bridge Tracts in Mathematics 137, 1999.
[15] V. I. Bernik, D. Y. Kleinbock, G. A. Margulis. Khintchine-type theorems on mani-
folds: the convergence case for standard and multiplicative versions. Research Notices,
No. 9, pp. 453 – 486, 2001.
[16] Y. Bugeaud. Approximation by Algebraic Integers and Hausdorff Dimension. J.
London Math. Soc., V. 65, pp. 547 – 559, 2002.
[17] J. W. S. Cassels. An introduction to the geometry of numbers. Springer-Verlag, 1959.
[18] H. Dickinson, M. M. Dodson. Extremal Manifolds and Hausdorff Dimension. Duke
Math. J., V. 101, No. 2, pp. 271 – 281, 2000.
[19] M. Dodson, B. Rynne, J. Vickers, Metric Diophantine approximation and Hausdorff
dimension on manifolds. Math. Proc. Cam. Phil. Soc. 105 (1989), 547–558.
[20] M. Dodson, M.V. Melián, D. Pestana, S.L. Velani, Patterson measure and Ubiquity.
Ann. Acad. Sci. Fenn. 20:1 (1995), 37–60.
[21] K. Falconer. Fractal geometry: mathematical foundations and applications. John
Wiley, 1990.
[22] D. Y. Keinbock. Ergodic Theory on Homogeneous Spaces and Metric Number Theory.
Preprint, 2007.

27
[23] D. Y. Kleinbock, G. A. Margulis. Flows on homogeneous spaces and Diophantine
approximations on manifolds. Ann. Math., V. 148, pp. 339-360, 1998.
[24] K. Mahler. Über das Maßder Menge aller S-Zahlen. Ann. Math., 106 (1932), pp.
131–139.
[25] G. A. Margulis. Diophantine approximation, lattices and flows on homogeneous
spaces. A panorama of number theory or the view from Baker’s garden (Zürich,
1999). CUP. 2002, pp. 280–310.
[26] A. Pyartly. Diophantine Approximations on Submanifolds in Euclidean Space. Func.
Anal. and Applications, V. 3(4), pp. 59 – 62, 1969. (in Russian)
[27] W. M. Schmidt. Metrische Sätze über simultane approximation abhängiger Größen.
Monatsch. Math., V. 68, pp. 154-166, 1964.
[28] V.G. Sprindžuk. Mahlers problem in the metric theory of numbers, AMS Providence,
V. 25, RI, 1969.
[29] R. C. Vaughan and S.L. Velani. Diophantine approximation on planar curves: the
convergence theory. Invent. Math., 166 (2006), no. 1, 103–124.

Dzmitry Badziahin
Mathematics Department, University of York, Heslington, York, YO10 5DD, England
Email address: [email protected]

28