Location via proxy:   [ UP ]  
[Report a bug]   [Manage cookies]                

Hausdorff Dimension of Divergent Geodesics On Product of Hyperbolic Spaces

Download as pdf or txt
Download as pdf or txt
You are on page 1of 31

Hausdorff dimension of divergent geodesics on product of

hyperbolic spaces
Lei Yang∗
arXiv:1303.5993v2 [math.DS] 8 Apr 2013

Abstract

In this article, we consider the diagonal geodesic flows on product of non-compact


hyperbolic surfaces of finite volume,i.e., on the product space T1 (M1 )×· · ·×T1 (Mk ) where
every Mi is some non-compact hyperbolic surface of finite volume and T1 (Mi ) denotes its
unit tangent bundle. And for xi ∈ T1 (Mi ), we denote the geodesic flow starting at xi by
gt (xi ), where t ∈ R. We want to study the points (x1 , . . . , xk ) ∈ T1 (M1 ) × · · · × T1 (Mk ) such
that (gt (x1 ), . . . , gt (xk )) → ∞ as t → ∞. We prove that the set of such points has Hausdorff
dimension 3k − 21 . This work is a generalization of the main results in [2], and the proof
involves some interesting results on the distribution of the orbits of discrete group actions
on Furstenberg boundary ∂H2 .

1 Introduction
Let M be a non-compact Riemann surface with constant curvature −1 and of finite total volume,
and let T1 (M) denote its unit tangent bundle. It is well known that M can be regarded as the quotient
space of the universal hyperbolic space H2 by some discrete subgroup Γ ⊂ Iso(H2 )  SL(2, R), say
M  Γ\H2 . Under the condition when M is of finite total volume, Γ ⊂ SL(2, R) is a lattice. And there is a
natural identification T1 (M)  Γ\SL(2, R). For simplicity, we denote SL(2, 1
 R) by G, then T (M) Γ\G.
 cos θ sin θ

 

Through this article, we use the following notations: let K = SO(2, R) =   : θ ∈ R ,A=


 − cos θ sin θ
 

      

 et  
 1 x

 

at =   : t ∈ R and let N =   : x ∈ R . And each point x ∈ T1 (M), it determines a point
 
−t
  

 e 
 0 1
 

on y ∈ M and also a direction v ∈ T1y (M), then starting from x, we have a unique geodesic flow, which is
denoted by {gt (x)}t∈R . On the Γ \ G level, the geodesic flow corresponds the action of the one-parameter
diagonal subgroup, say {at }t∈R , i.e., for Γx ∈ Γ \ G, the geodesic flow starting from x is just {Γxat }t∈R . In

Department of Mathematics, The Ohio State University, email address: lyang@math.osu.edu

1
this article, we study the diagonal geodesic flows on the product space of hyperbolic surfaces (for some
technical reason we consider the negative geodesic flows, i.e., t is moving in negative direction, and this
is the same as the positive geodesic flows), say T1 (Γ1 \ H2 ) × · · · × T1 (Γk \ H2 )  Γ1 \ G × · · · × Γk \ G, and
we are interested in the points (x1 , . . . , xk ) ∈ Γ1 \ G × · · · × Γk \ G such that (g−t (x1 ), × · · · × g−t (xk )) → ∞
as t → +∞, i.e., for any compact subset K ⊂ Γ1 \ G × · · · × Γk \ G, there exists a constant T (K), such
that for all t > T (K), (g−t (x1 ), . . . , g−t (xk )) < K. If we denote the collection of such points by Dk , we
want to compute the Hausdorff dimension of Dk . For k = 1, i.e., on some single hyperbolic surface, it
is well known that g−t (x) ∈ Γ \ G → ∞ as t → +∞ if and only if g−t (x) is going toward some cusp part
of Γ \ G and it never escapes the cusp. Therefore for every point y ∈ Γ \ H2 , to make the geodesic flow
divergent, the only directions in T1y (Γ \ H2 ) we can choose are the ones whose opposite directions are
pointing to the cusps of Γ \ H2 , and thus the set D1 is of codimension one in Γ \ G. But for k ≥ 2, the set
becomes interesting. Actually, Dk turns out to be a self-similar Cantor-like subset of the whole space.
And in this article, we will prove the following theorem:

Theorem 1.1. For any k ≥ 2, the Hausdorff dimension of Dk ,


1
dimH Dk = 3k −
2
.

The Hausdorff dimension of divergent geodesic trajectories on product of hyperbolic spaces is


closely related to that of divergent diagonal flows on homogeneous spaces G/Γ where G denotes some
semisimple Lie group and Γ ⊂ G is a lattice of G, and it has many applications in number theory. The
reader is referred to [4] and [5] for details in this direction. On the other hand, this topic is also related
to the geodesic flows on translation surfaces and Teichmüller flows on moduli spaces of holomorphic
quadratic differentials since there is a natural SL(2, R)-action on the moduli spaces and the Teichmüller
flows correspond to the orbits of the actions under {at }t∈R . We refer the reader to [3], [9], [10] and
[11] for details about the study in this setting.
This work is inspired by Cheung’s results in [2] in which he deals with the case when the discrete
subgroups are all equal to the arithmetic subgroup SL(2, Z). The main obstacle to generalize from
SL(2, Z) to any lattice Γ is the estimate of the number of cusp points in a given small interval whose
heights are in a given range (here the height of a cusp point is defined later and is an analogue of the
denominator of a rational point qp , say q). To get such an estimate, we need to apply an effective result
of the mixing property of geodesic flows on hyperbolic spaces due to Moore (see [8]).

1.1 SOME BASIC REDUCTION


In this subsection we recall some basic results about group action of SL(2, R) on H2 and get some
trivial reduction of our original problem.
The universal hyperbolic surface is identified with the upper half plane H2 = {z = x + ıy : y > 0}
2 2
with the hyperbolic metric ds2 = dx y+dy
2 , and SL(2, R) acts on H2 by fractional linear transformation:

2
 
g1 g2 
for g =    ∈ G and z ∈ H2 , gz = g1 z+g2 . Then if we consider the map: G → H2 by sending
g3 z+g4
g3 g4
g ∈ G to gı ∈ H2 , it is easy to see that the stabilizer of ı is K, so this map induces  an isomorphism

g1 g2 
2  ∈ G and
G/K  H . On the tangent bundle, the action is represented as follows: for g =  
g3 g4
(z, v) ∈ SL(2, R), then the action g(z, v) = ( gg31 z+g
z+g2
, v
4 (g3 z+g4 )
2 ). This group action is conformal by the

results of fractional linear transformations, i.e., it preserves the hyperbolic metric and also the angles
between any two geodesics. Then we can identify G with T1 (H2 ) by sending g ∈ G to g(ı, ı) ∈ T1 (H2 ).
Under this map we can also identify Γ \ G with T1 (Γ \ H2 ) for any non-compact lattice Γ of G. Then
for x ∈ G  T1 (H2 ), the geodesic flow {gt (x)}t∈R is identified with the orbit of the right action of
{at }t∈R , say {xat }t∈R , and on the level of homogeneous space Γ \ G, the geodesic flow starting from
Γx is equal to {Γxat }t∈R . In this article we insist on studying the negative geodesic flows, i.e., the
negative part of the whole geodesic line, say {Γxa−t}t>0 . Then  for any Γx ∈ Γ \ G,  we can decompose
a 0 
   
x ∈ G as x = u+ (x)p− (x), where p− (x) ∈ P− = 
 
 : a ∈ R \ {0}, b ∈ R and u+ (x) ∈ U + =

−1

b a
 

   
 1 t  
. Then since at p− (x)a−t remains bounded for all large t, therefore at p− (x)a−t →
 
u(t) ∈ : t ∈ R
 
   
0 1

   

ξ(x) ∈ G as t → +∞. Thus on the product space Γ1 \ G × · · · × Γk \ G, (Γ1 x1 a−t , . . . , Γk xk a−t ) → ∞
as t → +∞, then (Γ1 u+ (x1 )a−t at p− (x1 )a−t , . . . , Γk u+ (xk )a−t at p− (xk )a−t ) → ∞, and for at p− (xi )a−t →
ξ(xi ) ∈ G as t → +∞, for i = 1, . . . , k, this implies (Γ1 u+ (x1 )a−t , . . . , Γk u+ (xk )a−t ) also diverges as
t → +∞. Conversely, for the same reason, if (Γ1 u+ (x1 )a−t , . . . , Γk u+ (xk )a−t ) diverges as t → +∞ then so
does (Γ1 x1 a−t , . . . , Γk xk a−t ) as t → +∞. Therefore if we define

Dk = {x = (x1 , . . . , xk ) ∈ Rk : (Γ1 u(x1 )a−t , . . . , Γk u(xk )a−t ) → ∞ as t → +∞} (1)

we have dimH Dk = dimH Dk + dim(P− )k = dimH Dk + 2k. And on the level of H2 , for x ∈ R, the flow
{u(x)a−t }t>0 corresponds to the half geodesic ray {x + e−2t ı : t > 0}, then we have

Dk = {x = (x1 , . . . , xk ) ∈ Rk : Γ1 ×· · ·×Γk \(x1 +ıe−2t , . . . , xk +ıe−2t ) diverges as t → +∞ in Γ1 \H2 ×· · ·×Γk \H2 }


(2)
1 3
And our original problem can be reduced to proving that dim H Dk ≥ k − 2 and dimH D2 = 2.
The paper is organized as follows: in Section 2, we give some basic results concerning the geodesic
flows on a single hyperbolic surface Γ \ H2 and the cusp points of Γ, and deduce the conditions when
a geodesic flow can be close to a cusp of Γ; in Section 3, we prove a technical result concerning the
estimate of the number of cusp points of Γ in a small interval I = [x − d, x + d] with heights within some
range [c1 h, c2 h]; and in Section 4 we give the proof of Theorem 1.1, the first subsection concerns the
lower bound of Hausdorff dimension, the proof depends on the result in Section 3 and the second one
concerns the upper bound, the proof needs a basic estimate about cusps in [?].
Remark: the main ideas and most arguments in this paper also work for studying the divergent
geodesic flows on products of hyperbolic spaces of any dimension n, i.e., {(g−t (x1 ), . . . , g−t (xk ))}t>0 ⊂
T1 (M1 ) × · · · × T1 (Mk ), where Mi  Γi \ Hn is some non-compact hyperbolic space with finite total

3
volume, for i = 1, . . . , k. In a sequent paper, we will prove a similar result in this setting.

2 GEODESIC FLOWS ON Γ \ H2
In this section, we denote G = SL(2, R), and Γ ⊂ G some lattice of G. We may identify G with unit
tangent bundle of the universal hyperbolic space, say T1 (H2 ). Here we use the upper half space model,
the the ideal boundary of H2 , denoted by ∂H2 , is identified with R ∪ ∞. The diagonal flow u(x)a−t
corresponds to a geodesic flow on T1 (H2 ) joining ∞ ∈ ∂H2 and x ∈ R ⊂ ∂H2 . Then to study the quotient
space Γ \ H2 , we may divide H2 into countably many fundamental domains with respect to the Γ-action.
Up to a conjugation, we may assume some fundamental domain has ∞ as one of its cusps, among all the
fundamental domains sharing ∞ as a cusp, we choose one, and denote it by D0 , such that 0 ∈ R ⊂ ∂H2
is not a cusp of D0 . We denote all cusps of D0 by a1 = ∞, a2 , . . . , am , suppose the boundary of D0 has
n sides, then from a basic result of fundamental domains of Γ \ H2 , we may pair all sides of D0 in the
′ ′ ′
following way: {(ei , ei , si )}ni=1 , where ei , ei are sides of D0 and si ∈ Γ, such that for each i, si (ei ) = ei .
′ ′
Note that in this pairing notation, we put both (ei , ei , si ) and (ei , ei , s−1
i ) into our collection. And if we
put S = {si }ni=1 , then a well known result is that S generates Γ (see [1]). Next, we divide D0 into m parts
using only geodesic segments{P j}mj=1 , such that each part P j is connected and has only one cusp a j , then
for any point z ∈ P j ⊂ D0 , we say z is near a j , or close to a j . Then by dividing any fundamental domain
γD0 where γ ∈ Γ via γP j , similarly if z ∈ γP j then we say z is near γai . For a geodesic flow G, if at
some moment it moves into γP j , then we say at this moment G enters the cusp γai .
For any cusp ai , we have the subgroup Γai ⊂ Γ is nontrivial, where  Γai denotes the group of elements
 

  1 kl1
 

in Γ which fix the cusp ai . Then for a1 = ∞, Γ∞ =   : k ∈ Z , for some l1 ∈ R, without loss of
  

0 1
 

  
1 k 

 

generality (take a conjugation of Γ if necessary), we may assume l1 = 1, i.e., Γ∞ =  : k ∈ Z .


0 1  
 
     kli
 
 1 0 1 kli   1 0 1 − ai

 kli  

For general ai , Γai is of form  = : k ∈ Z for some li ∈ R.
 
 1 1 0 1  − 1 1  − kl2i kli 
       
1 + ai
   


ai ai ai
For any geodesic flow G that once enters D0 and near aiat some moment, then the next cusp G enters
1 − kli kli 
will be of form ui (k)s j aq , where ui (k) =  kliai  ∈ Γai and s j ∈ S ∪ {e} where e stands for the
− 2 ai
1 + kli 
ai
identity of G, but s j aq , ai . In fact, G can spend some time near ai , going through several fundamental
domains near ai , say from D0 to ui (k)D0 and then enter another cusp of D0 , or enter another nearby
fundamental domain, which is of form ui (k)s j D0 for some s j ∈ S . When the formal happens then we
put s j = e and thus have the above statement. Then by transforming D0 to γD0 via γ, we have the
following: for any cusp γai , if a geodesic flow G is near γai at some moment, then the next cusp G
enters must be of form γui (k)s j aq .
 
γ1 γ2  √ γ
For a cusp γai , where γ =   ∈ Γ, we define the height of γai , denoted by kγai k = li |γ3 + a4i |,
γ3 γ4

4
√  √ √ −1 
0   li (γ1 + γa2i )

γ1 γ2   li li γ2 
in fact, it is just the lower left entry of    √
li √ −1  =  √ γ4 √ −1 . And also we
γ3 γ4 ai li li (γ3 + ai ) li γ4
 √ −1  √ 
 li 0   li 0 
note that  √ 1 √ −1  Γai  √ 1 √ −1  = Γ∞ . In particular, if ai = ∞, then kγ∞k = |γ3 |. But for
li ai li li ai li
a fixed cusp γai , the choice of γ is flexible, in fact, we can choose any element in the coset γΓai . In this
article we choose γ in the following way: For each cusp ai of D0 , we choose a neighborhood Θi of ai
in ∂H2 , for a1 = ∞, the neighborhood we choose should be Θ1 = (−∞, −M) ∪ (M, ∞) for some large
number M > 0. Then for a fixed cusp γai we choose γ such that γ−1 ∞ < Θi . This always can be done
since if γ−1 ∞ ∈ Θi , then we apply ui (k) to get ui (k)γ−1 ∞, for a suitable k, we may move it out of Θi .
For the case of a1 = ∞, the picture is very clear: if |γ−1 ∞| is larger than M, then we apply u1 (k) to get
u1 (k)γ−1 ∞ = γ−1 ∞ + k, this is pure translation, then obviously   we can make
−1
 |γ ∞| smaller than M for
 1 0  1 0
suitable k. For general case of ai , we translate Θi via  1  to   Θi which is a neighborhood
− ai 1 − a1i 1
 
 1 0
−1  γ−1 ∞ is near ∞, then we may choose suitable k to make
of ∞. Then if γ ∞ ∈ Θi , then  1 
− ai 1
        
1 kli   1 0  1 0  1 0 1 kli   1 0
    γ−1 ∞ =   γ−1 ∞ + kli away from ∞. Thus       γ−1 ∞ will
0 1 − a1i 1 − a1i 1 1
ai 1 0 1 − a1i 1
be out of Θi , i.e., ui (k)γ−1 ∞ < Θi . then we replace γ by γui (−k), then we make γ−1 ∞ < Θi .
′ ′
Near each cusp of D0 , say ai , we can find two horocircles, say Ci and Ci , with radius ri and ri ,
′ ′
respectively, both through ai , such that Ci ⊂ Γai Pi ⊂ Ci . A point x + yı ∈ H2 in inside Ci (and Ci

respectively), if and only if (x − ai )2 + y2 < 2ri y (and (x − ai )2 + y2 < 2ri y respectively). On the other
 √ −1 
 li 0  x+yı
hand, if we apply   √ √  acting on x + yı, we get li = li (aaii x+ai yı
−x)−li yı , and its imaginary part
− aili li − a (x+yı)+li
i

(ai x+ai yı)((ai −x)+yı) a2 y ′


is ℑ li (aaii x+ai yı
−x)−li yı = ℑ li (ai −x)2 +li y2
= li (x−aii)2 +li y2 . Thus if x + yı ∈ Ci (and x + yı ∈ Ci respectively), then
 √ −1 
 li 0  a2 y a2 a2
we have ℑ  √li √  (x + yı) = li (x−aii)2 +li y2 > 2liiri ( > 2l ir′ respectively). From the above calculation,
− ai li i i

we have the following:


a2
Proposition 2.1. For each cusp ai of D0 , there exist two constants, depending only on Γ, ci (= 2liiri ) and
 √ −1 
′ a2i  li 0 
ci (= 2l r′ ), such that, if ℑ   √ √  (x + yı) > ci , then x + yı is near ai , and if x + yı is near ai , then
i i − aili li
 √ −1 
 li 0   ′
ℑ  √li √  (x + yı) > ci . Here conventionally we put ∞1 = 0.
− ai li

Now, for a fixed x ∈ R ⊂ ∂H2 , we denote G x the geodesic flow joining ∞ and x, which corresponds
to the diagonal flow u(x)a−t (let u(x)a−s act on ı), then the element u(x)a−s corresponds to the point
ı ı
x+ e2s
, ∈ γPi ) if and only if γ−1 (x + e2sı ) is near ai , and then we may apply the
it is near γai (i.e., x + e2s  √ −1   
 li 0  −1 ı
γ1 γ2 
above proposition. So we need to calculate ℑ  √li √  γ (x + e2s ). Suppose γ =  , then
− ai li γ3 γ4

5
 √ −1   √ √ 
 li 0  −1  li −1 γ4 − li −1 γ2 
 √li √  γ =  √ √ , so
− ai li li (−γ3 − γa4i ) li (γ1 + γa2i )
 √ −1 
 li 0  −1 ı 1 (γ4 x−γ2 )+γ4 e−2s ı
 √li √  γ (x + e2s
) = li (γ1 + γa2 )−x(γ3 + γa4 )−(γ3 + γa4 )e−2s ı
− ai li i i i
(3)
−2s 2 γ 4 γ
4 −2s γ
1 [(γ4 x−γ2 )+γ4 e ı][(γ1 + ai )−x(γ3 + ai )+(γ3 + ai )e ı]
= li γ2 γ4 2 γ4 2 −4s
[(γ1 + a )−x(γ3 + a )] +(γ3 + a ) e
i i i

and its imaginary part is the following:


γ γ γ γ γ γ
[(γ4 x−γ2 )+γ4 e−2s ı][(γ1 + a2 )−x(γ3 + a4 )+(γ3 + a4 )e−2s ı] 4 −2s −2s 2 4
1 (γ4 x−γ2 )(γ3 + ai )e +γ4 e [(γ1 + ai )−x(γ3 + ai )]
ℑ l1i γ
[(γ1 + a2
i
γ
)−x(γ3 + a4
i
γ
)]2 +(γ3 + a4 )2 e−4s
i
= li γ γ γ
[(γ1 + a2 )−x(γ3 + a4 )]2 +(γ3 + a4 )2 e−4s
i i i i i i
e−2s (γ3 γ4 x+γ42 x/ai −γ2 γ3 −γ2 γ4 /ai +γ1 γ4 +γ2 γ4 /ai −γ3 γ4 x−γ42 x/ai )
= kγai k2 (|x−γai |2 +e−4s )
e−2s (γ1 γ4 −γ2 γ3 )
= kγai k2 (|x−γai |2 +e−4s )
e−2s
= kγai k2 (|x−γai |2 +e−4s )
1
= kγai k2 (e2s |x−γai |2 +e−2s )
(4)
γ2 γ4 2 2 2
√ γ4 γ1 ai +γ2
γ
γ1 + a2
In above, li [(γ1 + ai ) − x(γ3 + ai )] = kγai k |x − γai | since kγai k = li |γ3 + ai |, γai = γ3 ai +γ4 = γ
γ3 + a4
i
,
i
and γ1 γ4 − γ2 γ3 = 1 since γ ∈ SL(2, R).
1 1
When e2s = |x−γa i|
, the imaginary part can obtain its maximum 2kγai k2 |x−γai |
. Therefore we have the
following proposition:

Proposition 2.2. Let G x , ci and ci be defined as above. For each cusp γai , G x is near γai at some
moment if
1
> ci
2kγai k2 |x − γai |
. Conversely, if G x is near γai at some moment, then we have
1 ′
> ci
2kγai k2 |x − γai |
Remark: The above statement can be regarded as an analogue of the condition under which a
rational fraction qp is among the continued fractions of some real number x.
For simplicity, we call G x enters γai deeply if 2kγai k21|x−γai | > ci .

We denote A ≍ B if there exist two positive constants ω and ω (only depending on Γ), such that

ωB ≤ A ≤ ω B; and we denote A ≪ B (A ≫ B respectively) if only the right part (left part respectively)
of the previous inequality holds.
Suppose at some moment G x is near γai , and it leaves γai after a while, then the next cusp it enters
must be of form γui (k)s j aq from the above discussion, where ui (k), s j and aq are defined as above. If

6
   
γ1 γ2  α β j 
we put γ =   and s j =  j , then we have:
γ3 γ4 τj ρj
 
γ1 γ2  1 − kla i
 
kl i  α j β j 
γui (k)s j =    kl i  
kli  

γ3 γ4  − a2i
i
1 + ai τj ρj

 
γ1 [α j (1 − kli ) + τ j kli ] + γ2 [−α j kl2i + τ j (1 + kli
γ1 [β j (1 − kli
+ ρ j kli ] + γ2 [−β j kla2i + ρ j (1 + kli 
ai ai ai )] ai ) ai )]

=  i
kli
γ3 [α j (1 − ai ) + τ j kli ] + γ4 [−α j kla2i + τ j (1 + kli
ai )] γ3 [β j (1 − kli
ai ) + ρ j kli ] + γ4 [−β j kla2i + ρ j (1 + kli 
ai )]
 
i i
(5)
Then we have:
kli
γ1 [α j (1 − ai ) + τ j kli ] + γ2 [−α j kla2i + τ j (1 + kli
ai )] + 1
aq {γ1 [β j (1 − kli
ai ) + ρ j kli ] + γ2 [−β j kla2i + ρ j (1 + kli
ai )]}
i i
γui (k)s j aq = kli
γ3 [α j (1 − ai ) + τ j kli ] + γ4 [−α j kla2i + τ j (1 + kli
ai )] + 1
aq {γ3 [β j (1 − kli
ai ) + ρ j kli ] + γ4 [−β j kla2i + ρ j (1 + kli
ai )]}
i i
(6)
and its height

lq |γ3 [α j (1 − klaii ) + τ j kli ] + γ4 [−α j kla2i + τ j (1 + klaii )] + a1q {γ3 [β j (1 − klaii ) + ρ j kli ] + γ4 [−β j kla2i + ρ j (1 + kli
p
kγui (k)s j aq k = ai )]}|
i i
ρ β β ρ
lq |kli [(τ j + aqj ) − a1i (α j + aqj )](γ3 + γa4i ) + γ3 (α j + aqj ) + γ4 (τ j + aqj )|
p
=
(7)
p √ ρj 1 βj
Since lq li |(τ j + aq ) − ai (α j + aq )| = ks j aq kkai k|ai − s j aq |, especially, if ai = ∞, it will become ks j aq k,
if s j aq = ∞, i.e., s j = e and aq = a1 = ∞, it will become kai k.
Then since |ai − s j aq |ks j aq kkai k is bounded from both above and from below (including the above
two extreme cases), in fact there are only finitely many choices for ai , s j and aq , thus in the above

expression the coefficient of k, which we denote by Ω(k), can be compared with li |γ3 + γa4i | = kγai k,
i.e., Ω(k) ≍ kγai k.
β ρ
In the expression of γui (k)s j aq , the remainder except Ω(k)k is γ3 (α j + aqj ) + γ4 (τ j + aqj ). To estimate
kγui (k)s j aq k, we need to prove the following inequality:
βj ρj γ4
|γ3 (α j + ) + γ4 (τ j + )| ≪ |γ3 + | (8)
aq aq ai
 
γ1 (n) γ2 (n)
Proof of (8). We prove it by contradiction. Suppose for each integer n > 0, there exist γ(n) =   ,
γ3 (n) γ4 (n)
 
α j (n) β j (n)
ai (n), s j (n) =   and aq (n), such that
τ j (n) ρ j (n)
β j (n) ρ j (n) γ4 (n)
|γ3 (n)(α j (n) + ) + γ4 (n)(τ j (n) + )| > n|γ3 (n) + | (9)
aq (n) aq (n) ai (n)
Since there are only finitely many choices for ai (n) s j (n) and aq (n), by passing to a subsequence, we may
assume they remain the same, say equal to ai s j and aq , respectively. And then by choose a subsequence
again we may assume that |γ3 (n)| ≥ |γ4 (n)| always true or |γ3 (n)| ≤ |γ4 (n)| always true. If this is the
formal case. Therefore by dividing both sides by |γ3 (n)| the above inequality becomes:
βj γ4 (n) ρj γ4 (n) 1
|(α j + )+ (τ j + )| > n|1 + | (10)
aq γ3 (n) aq γ3 (n) ai

7
Since | γγ34 (n)
(n)
| ≤ 1, the left hand side remains uniformly bounded for any n, thus the only case to make the
inequality hold is that 1 + γγ43 (n) 1 γ4 (n) γ4 (n) −1
(n) ai → 0 as n → ∞, i.e., − γ3 (n) → ai as n → ∞. Since − γ3 (n) = γ(n) ∞,
we have γ(n)−1 ∞ → ai . This contradicts our choice of γ(n) since previously we choose γ such that
γ−1 ∞ < Θi ∋ ai .
If this is the latter case, then we divide both sides of the inequality by |γ4 (n)| to get

γ3 (n) βj ρj γ3 (n) 1
| (α j + ) + (τ j + )| > n| + | (11)
γ4 (n) aq aq γ4 (n) ai
γ3 (n) 1
Then we get the left hand side remains uniformly bounded, and thus γ4 (n) + ai → 0 as n → ∞. So we
get the same statement as in the formal case and thus the same contradiction.
This completes the proof.

Combining Ω(k) ≍ kγai k and the inequality (8), we have


β ρj
kγui (k)s j aq k = |Ω(k)k + γ3 (α j aqj ) + γ4 (τ j + aq )|
β ρ
≤ |Ω(k)||k| + |γ3 (α j aqj ) + γ4 (τ j + aqj )|
(12)
≪ |k|kγai k + Const.kγai k
≪ |k|kγai k

And for lower bound of kγui (k)s j aq k, we have that there exists some large integer N such that for
|k| ≥ N the following estimate of kγui (k)s j aq k holds:

kγui (k)s j aq k ≫ kkγai k (13)

Combing them, we have for |k| ≥ N, the following estimate holds:

kγui (k)s j aq k ≍ |k|kγai k (14)

For |k| < N, we will prove the same estimate:

kγui (k)s j aq k ≍ |k|kγai k (15)

The upper bound has been proved, so we only need the lower bound, i.e., there exists some constant,
Const., such that
kγui (k)s j aq k ≥ Const.|k|kγai k (16)

where |k| < N. But note that we do not prove this for arbitrary ai , k, s j and aq , instead, we will only
prove this for γai and γui (k)s j aq such that there exists some geodesic G x = {x + e2sı } s∈R , which enters
γai and after a while it leaves γai for γui (k)s j aq . And since we assume |k| is bounded, in the above
inequality the factor |k| is unnecessary. Thus it suffices to prove the following:

ı
Lemma 2.1. There exists some constant Const., such that if some geodesic G x = {x + }
e2s s∈R
enters γai
′ ′ ′
when s = t and leaves it for γ aq when s = t , then kγ aq k ≥ Const.kγai k

8
1
Proof. At first, we claim that there exists some constant Λ, such that e2t < Λ |x−γai|
. If G x enters
1
γai deeply, then the conclusion is already there since at e2s = |x−γa i|
, G x is already near γai , then
e < |x−γai | . If G x does not enter γai deeply, then we have kγai k |x − γai | ≥ 2c1 i , and x + ei2t is near γai
2t 1 2

implies e2t |x − γai |2 + e−2t < ′


1
2ci kγai k2
, which implies e2t |x − γai |2 < ′
1
2ci kγai k2
, and therefore e2t |x − γai | <
1 2ci ci 1

2ci kγai k2 |x−γai |
≤ ′
2ci
. This proves that e2t < ′
ci |x−γai |
. Therefore we prove the claim. Then for the same

reason, e 2t
< Λ |x−γ1′ a | .
q
′ ′ ′
Now from the conditions we know that e2t |x − γai |2 + e−2t < ′
1
2ci kγai k2
and e2t |x − γ aq |2 + e−2t ≥
′ ′ ′ ′
1
2cq kγ′ aq k2
. Since e2t < Λ |x−γ1′ a | , we have e2t |x − γ aq |2 < Λ2 e−2t , and then the latter inequality above
q
′ ′ ′
implies (Λ2 + 1)e−2t > 2c kγ1′ a k2 , then e2t < 2(1 + Λ2 )cq kγ aq k2 . And from the formal inequality
q q
above we have e−2t < 2c′ kγa
1
k2
(here we just drop a positive term e2t |x − γai |2 from the left hand side),
i i
′ ′ ′ ′
i.e., e2t > 2ci kγai k2 . And since t > t, then we have 2(1 + Λ2 )cq kγ aq k2 > 2ci kγai k2 , which leads the
conclusion immediately.
This proves the lemma.

Combing both of them, we have for arbitrary γ ai k and aq , the following holds:

kγui (k)s j aq k ≍ |k|kγai k (17)

The distance between γai and γui (k)s j aq is the following:


kl kli kl kl kli kl
γ
γ1 + a2 γ1 [α j (1− a i )+τ j kli ]+γ2 [−α j +τ j (1+ a i )]+ a1q {γ1 [β j (1− a i )+ρ j kli ]+γ2 [−β j +ρ j (1+ a i )]}
i a2 i i a2 i
|γai − γui (k)s j aq | = γ
γ3 + a4
i
− kli
i
kli kli 1 kli
i
kli kl
i
γ3 [α j (1− a )+τ j kli ]+γ4 [−α j 2 +τ j (1+ a )]+ aq {γ3 [β j (1− a )+ρ j kli ]+γ4 [−β j 2 +ρ j (1+ a i )]}
i a i i a i
i i
ks j aq kkai k|ai −s j aq |
= kγai kkγui (k)s j aq k
(18)
ks j aq k
As we have done above, if ai = ∞, the final line will become if s j aq = ∞, i.e., s j = e and kγai kkγui (k)s j aq k ,
kai k
aq = a1 = ∞, then the final line will become kγai kkγui (k)∞k . Then |ai − s j aq |ks j aq kkai k is bounded from
both above and from below. This deduces the following:
1
|γai − γui (k)s j aq | ≍ (19)
kγai kkγui (k)s j aq k

Combing with the estimate (17), we have the following holds:


1
|γai − γui (k)s j aq | ≍ (20)
|k|kγai k2

Next we study the distance between arbitrary two cusps, γa j and γ ai .

At first, let us look at a simple case, where γ = e.

Lemma 2.2. For any γ ∈ Γ and cusps of D0 , ai and a j , we have

kai kkγa j k|γa j − ai | = ka j kkγ−1 ai k|γ−1 ai − a j | (21)

9
Proof. This follows directly from pure calculation. In fact,
√ p 1
kai kkγa j k|γa j − ai | = li l j | (γ1 + γ2 /a j ) − (γ3 + γ4 /a j )|
√ p a1i
= li l j | a j (γ4 − γ2 /ai ) − (−γ3 + γ1 /ai )|
−γ4 ai +γ2 (22)
= ka j kkγ−1 ai k| −γ3 ai +γ1
− a j|
= |ai |kγ−1 ai k|γ−1 ai − a j |

This proves the lemma.

From the lemma, we have the following proposition:

Proposition 2.3. There exists some constant Const. such that for any γ ∈ Γ, ai and a j , we have

kai kkγa j k|γa j − ai | ≥ Const. (23)

Proof. At first, by replacing γ by ui (k)γ for some suitable k, we can make ui (k)γa j < Θi . And this
replacement does not change γ−1 ai so does not change the value ka j kkγ−1 ai k|γ−1 ai − a j | which equals
kai kkγa j k|ai − γa j |. This shows that this replacement does not change the value kai kkγa j k|γa j − ai |. Then
we can assume that there exists a constant λ such that kai k|γa j − ai | ≥ λ. Then it suffices to prove
kγa j k ≥ Const..
Suppose not, then there exists a sequence γ(n) ∈ Γ and sequence of cusps of D0 , a j (n) such that
kγ(n)a j (n)k > 0 for any n (kγa j k = 0 if and only if a j = ∞ and γ ∈ Γ∞ , and in this case |γa j − ai |kγa j k =
1), and γ(n)a j (n) → 0 as n → ∞. Since the choices of a j (n) are finite, passing to a subsequence we may
assume that a j (n) = a j remains the same. Then we consider the action of Γ ⊂ SL(2, R) on R2 just by
 
linear transformation. Since Γ is a discrete subgroup, and if we denote the stabilizer group of 1, a1j by
Ua j , we have the intersection of Ua j and Γ, Ua j ∩ Γ = Γa j which is cocompact in Ua j , this implies that
 1  γ1 (n) + γ2 (n) 
     
 1 
2 aj 
the orbit Γ  1  is discrete in R . γ(n)  1  = 
     . kγ(n)a j k → 0 tells that the y-coordinate of
aj aj γ3 (n) + γ4a(n)
j
       
 1  1 k  1 k(n)  1 
γ(n)  1  tends to 0 as n → ∞. Since 
   ∈ Γ∞ ⊂ Γ, we can apply u1 (k(n)) =   on γ(n)  
1
aj 0 1 0 1 aj
for some suitable k(n) to remain the y-coordinate and make the x-coordinate γ1 (n)+ γ2a(n)
j
+k(γ3 (n)+ γ4a(n) )
   j

  1 

also bounded, this can always be done since kγ(n)a j k > 0. This shows that  u1 (k(n))γ(n)  1  is
  
 

aj n∈N
 
 1 
bounded, but kγ(n)a j k → 0 implies the previous set is infinite, which contradicts the fact that Γ  1  is
aj
discrete. This proves that kγa j k ≥ Const..
This completes the proof.

10

Now let us consider the general case: for two cusps γa j and γ ai , the distance

γ ′ γ
′ γ1 + a2 γ1 + a2
j
|γa j − γ ai | = γ
γ3 + a4
− γ
i


j γ3 + a4
i
′ ′ ′ ′ ′ ′
′ γ1 γ γ2 γ γ γ ′ γ γ3 γ γ4 γ γ
√ p γ1 γ3 + ai
4 + aj
3 + a2a 4 −γ1 γ3 − 2
ai − 1
aj − a2a 4
i j i j
= li l j kγa j kkγ′ ai k (24)
′ ′ ′ ′ ′ ′ ′ ′
√ p 1
ai
[(γ1 γ4 −γ2 γ3 )+ a1 (γ2 γ4 −γ2 γ4 )]−[(γ1 γ3 −γ1 γ3 )+ a1 (γ1 γ4 −γ2 γ3 )]
j j
= li l j kγa j kkγ′ ai k
′ ′ ′ ′ ′ ′ ′ ′
p [(γ1 γ4 −γ2 γ3 )+ a1 (γ2 γ4 −γ2 γ4 )]−ai [(γ1 γ3 −γ1 γ3 )+ a1 (γ1 γ4 −γ2 γ3 )]
j j
= kai k l j kγa j kkγ′ ai k

On the other hand,  ′ ′


 
′−1
 γ
4 −γ2  γ1 γ2 
γ γ =  ′
 ′ 

  
−γ3 γ1 γ3 γ4
  (25)
γ1 γ4′ − γ2′ γ3 γ2 γ4′ − γ2′ γ4 
=  ′
γ1 γ3 − γ1 γ3′ γ1′ γ4 − γ2 γ3′


Then we have
q ′ ′ 1 ′ ′ ′ ′ 1 ′ ′
l j [(γ1 γ4 − γ2 γ3 ) + (γ2 γ4 − γ2 γ4 )] − ai [(γ1 γ3 − γ1 γ3 ) + (γ1 γ4 − γ2 γ3 )] = kγ′−1 γa j k|γ′−1 γa j −ai |
aj aj
(26)
This shows that
kai kkγ′−1 γa j k|γ′−1 γa j − ai |
|γa j − γ′ ai | = (27)
kγa j kkγ′ ai k
Applying Proposition 2.3 with γ′−1 γ, ai and a j , we have kγ′−1 γa j k|γ′−1 γa j − ai | ≫ 1, and thus get
following proposition for general case:
1
|γa j − γ′ ai | ≫ (28)
kγa j kkγ′ ai k
For a geodesic G x , we have a sequence of cusps {γ(n)a(n)}n∈N such that the geodesic enters them
one after another sequently, and we call it the spectrum of x.
We will prove the following estimate of distance between x and γ(n)a(n):
1
|x − γ(n)a(n)| ≍ (29)
kγ(n)a(n)kkγ(n + 1)a(n + 1)k
1
Proof. Choose a large constant M such that if |x − γai | ≤ Mkγaik
2 then G x enters γai deeply.
1
If |x − γ(n)a(n)| ≥ Mkγ(n)a(n)k2 then the statement is already there since kγ(n + 1)a(n + 1)k ≫
1
kγ(n)a(n)k. Now we assume |x − γ(n)a(n)| < Mkγ(n)a(n)k 2 , then G x enters γ(n)a(n) deeply. Then when G x

leaves γ(n)a(n) for γ(n + 1)a(n + 1), say at s = tn+1 , we have


 √ −1  √ 
 l(n) 0  ı  l(n + 1)−1 0
 γ(n + 1)−1 (x + ı )
−1

ℑ  √l(n) √  γ(n) (x + 2tn+1 ) ≍ ℑ  √
l(n+1) √
− a(n) l(n) e − a(n+1) l(n + 1) e2tn+1

, i.e.,

kγ(n)a(n)k2(e2tn+1 |x − γ(n)a(n)|2 + e−2tn+1 ) ≍ kγ(n + 1)a(n + 1)k2 (e2tn+1 |x − γ(n + 1)a(n + 1)|2 + e−2tn+1 )

11
1
, and for γ(n)a(n), since G x enters it deeply, we have e2tn+1 > |x−γ(n)a(n)| . This implies on the left hand
2 2tn+1 2
side of the previous expression, kγ(n)a(n)k e |x − γ(n)a(n)| is the major term. As we have proved,
1
e2tn+1 ≤ Λ |x−γ(n+1)a(n+1)| then the right hand side is compared with kγ(n + 1)a(n + 1)k2 e−2tn+1 . Therefore

kγ(n)a(n)k2e2tn+1 |x − γ(n)a(n)|2 ≍ kγ(n + 1)a(n + 1)k2 e−2tn+1


p
≍ kγ(n)a(n)k2e2tn+1 |x − γ(n)a(n)|2kγ(n + 1)a(n + 1)k2 e−2tn+1
= |x − γ(n)a(n)|kγ(n)a(n)kkγ(n + 1)a(n + 1)k
(30)
2 −2tn+1
Note that at s = tn+1 , kγ(n + 1)a(n + 1)k e is uniformly bounded from both above and below,
then we have
|x − γ(n)a(n)|kγ(n)a(n)kkγ(n + 1)a(n + 1)k ≍ 1

This proves the statement.

From the above argument, we can also conclude that the time tn when G x enters γ(n)a(n), is asymp-
totically ln kγ(n)a(n)k. And moreover, if G x enters γ(n)a(n) deeply, then at the time T n = 21 ln |x−γ(n)a(n)| 1
,
 √ −1 
 l(n) 0 
ℑ  √l(n) √  γ(n)−1 (x + eı2s ) admits its maximum.
− a(n) l(n)
 √ −1 
 l(n) 0 
From now on, we define the following function for x ∈ R, W x (t) = ℑ   √
l(n) √  γ(n)−1 (x +
− a(n) l(n)
ı
e 2s ) when t ∈ [tn , tn+1 ). Then if G x enters γ(n)a(n) deeply, W x (t) has a local maximum in [tn , tn+1 ) at
s = T n.
Next, we will prove that for two cusps γai and γ′ aq = γui (k)s j aq , if |k| is large enough, then any
point x very close to γ′ aq , say the distance between them is much less than kγ′ a1q k2 has the following
property: G x will enter both γai and γ′ aq , and from leaving γai to entering γ′ aq , it at most spends a
uniformly bounded time, say s0 . Rigorously, the following proposition will be proved:

Proposition 2.4. There exist constants N0 , ǫ and s0 , such that for any two cusps γai and γ′ aq =
γui (k)s j aq , where γ ∈ Γ, ai and aq are two cusps of D0 , ui (k) ∈ Γai and s j ∈ S ∪ {e}, if |k| > N0 ,
then for any x ∈ R, if |x − γ′ aq | < kγ′ aǫ q k2 , then G x enters γai and γ′ aq in the given order, and from leaving
γai to entering γ′ aq , it spends at most s0 .

Proof. Choose N0 large enough so that when |k| > N0 , kγui (k)s j aq k ≍ |k|kγai k ≫ kγai k, then for any
small ǫ > 0, if |x − γ′ aq | < ǫ
kγ′ aq k2
, then the distance between x and γai ,

|x − γai | ≥ |γ′ aq − γai | − |x − γ′ aq |


1 ǫ
≫ kγ′ aq kkγai k − kγ′ aq k2 (31)
1
≫ kγ′ aq kkγai k

Since kγ′ aq k ≫ kγai k.

12
Conversely, we have
|x − γai | ≤ |γ′ aq − γai | + |x − γ′ aq |
1 1
≪ kγ′ aq kkγai k + kγ′ aq k2 (32)
1
≪ kγai kkγ′ aq k

In above, the second inequality is true because γ′ aq = γui (k)s j aq from which it follows |γai − γ′ aq | ≍
1
kγai kkγ′ aq k as we have proved.
1
Therefore we have |x − γai | ≍ kγai kkγ ′ a k , then at first G x will enter γai at some moment, and when
q

it leaves γai , say at s = t, then the next cusp γ”ar G x enters will have height kγ”ar k ≍ kγ′ aq k since
1
|x − γai | ≍ kγai kkγ”ark
.

If γ”ar = γ aq then there is nothing to prove. Suppose not, then we have

kγ”ar k2 (e2t |x − γ”ar |2 + e−2t ) ≍ 1

and moreover we have proved e−2t ≫ e2t |x − γ”ar |2 , then

kγ”ar k2 e−2t ≍ 1

Combing kγ′ aq k ≍ kγ”ar k, we have


kγ′ aq k2 e−2t ≍ 1

which means there exists a constant Const. such that


1
e−2t ≤ Const.
kγ′ aq k2

Then there exists a constant s0 > 0 such that


Const. 1 1 1
e−2t−2s0 ≤ ≤
e2s0 kγ′ aq k2 2cq kγ′ aq k2

And since |γ′ aq − γ”ar | >> kγ′ aq kkγ”a 1


rk
≍ kγ′ a1q k2 , we can choose ǫ > 0 small enough such that if
|x − γ aq | < kγ′ aq k2 , then |x − γ”ar | ≥ Λe |x − γ′ aq |. This fact combing with e−2t ≥ Λ1 |x − γ”ar | implies
′ ǫ 2s0

1
e−2t−2s0 ≥ |x − γ”ar | ≥ |x − γ′ aq |
Λe2s0
Then we have e−2t−2s0 ≥ e2t+2s0 |x − γ′ aq |2 , and thus

kγ′ aq k2 (e2t+2s0 |x − γ′ aq |2 + e−2t−2s0 ) ≤ kγ′ aq k2 2e−2t−2s0


1 (33)
≤ cq

This shows at time s = t + s0 , G x has already entered γ′ aq deeply, and thus proves the proposition.

For σ > 0, we define the subset ∆σ ⊂ R as follows:

Definition 2.1. x ∈ ∆σ if and only if the spectrum of x, {γ(n)a(n)}n∈N has a subsequence {γ′ (n)a′ (n)}n∈N
such that the following properties hold:

13
1. kγ′ (n)a′ (n)k1+σ ≤ kγ′ (n + 1)a′ (n + 1)k ≤ 2kγ′ (n)a′ (n)k1+σ

2. From G x leaving γ′ (n)a′(n) to entering γ′ (n + 1)a′(n + 1), it spends at most s0 , where s0 is defined
as above.

As a close of this section, we will prove the following proposition concerning Hausdorff dimension
of ∆σ :
1
Proposition 2.5. For any σ > 0, the Hausdorff dimension of ∆σ , denoted by dimH (∆σ ) is at least 2+σ .

To prove this proposition, we need the following theorem:

Theorem 2.1. (See [2]) Consider a Cantor set F ⊂ R defined as a nested intersection:
\
F= Fj
j≥0

where F0 is a closed interval and each F j is a disjoint union of finitely many closed intervals. Let (m j ) j≥1
be a sequence of positive integers and (ǫ j ) j≥1 a sequence of real numbers that tend to zero monotonically
and suppose that for each j ≥ 1 there are at least m j intervals of F j contained in each interval of F j−1
and these intervals are separated by gaps of length at least ǫ j . Then the Hausdorff dimension of F
satisfies the lower bound
ln(m1 · · · m j−1 )
dimH F ≥ lim sup
j→∞ − ln(m j ǫ j )

The proof can be found in [6, Example 4.6].

Proof of Proposition 2.5. For some fixed ai , we start with some cusp γ(0)ai with kγ(0)ai k large enough,
ǫ ǫ
and define F0 = [γ(0)ai − kγ(0)aik
2 , γ(0)ai + kγ(0)a k2 ], where ǫ is defined as above. Then assuming F j−1 has
i
ǫ ǫ
been constructed and every interval of it is of form [γai − kγai k2
, γai + kγai k2
], we define F j inductively
as follows:
For each closed interval I = [γai − ǫ
kγai k2
, γai + kγaǫi k2 ] of F j−1 , for any cusp γ′ ai ∈ I, where
γ′ = γui (k), for k > N0 . Then if we take I ′ = [γ′ ai − kγ′ ǫai k2 , γ′ ai + kγ′ ǫai k2 ], then from Proposition 2.4,
we have for any x ∈ I ′ , G x enters γai and then γ′ ai , and from leaving γai to entering γ′ ai , it spends
at most s0 . Note that we have proved that kγui (k)ai k is of form |kγai kΩ(k)k + Θ(γ)| where |Ω(k)| ≍ 1,
i.e., its height is a linear function of k whose coefficient is more or less kγai k, then there are more or
1+σ 1+σ 1+σ
less |[kγai k kγa
,2kγai k ]|
ik
= kγa ik σ
kγai k = kγai k choices of k such that kγui (k)ai k ∈ [kγai k
1+σ
, 2kγai k1+σ ]. Thus
we have at least (up to a constant multiplication) kγai kσ choices of I ′ , and any two different intervals
I1′ = [γ1′ ai − kγ′ ǫai k2 , γ1′ ai + kγ′ ǫai k2 ] and I2′ = [γ2′ ai − kγ′ ǫai k2 , γ2′ ai + kγ′ ǫai k2 ], the gap between them is at least
1 1 2 2
|γ1′ ai − γ2′ ai | − kγ′ ai2ǫ ′ ′ 1 1 1
kkγ′ ai k , since |γ1 ai − γ2 ai | ≫ kγ′ ai kkγ′ ai k , we have the gap is ≫ kγ′ ai kkγ′ ai k ≍ kγai k2+2σ .
1 2 1 2 1 2
Then we define F j by the union of all these I ′ ’s.
ǫ ǫ
From this construction, we have for each interval [γai − kγai k2
, γai + kγai k2
] of F j−1 , the height
(1+σ) j−1 (1+σ) j−1
kγai k ∈ [kγ(0)aik ,2 j−1
kγ(0)ai k ], then the above argument shows that m j ≍ kγai kσ ≥

14
j−1 1 1 j−1
σ σ
kγ(0)ai k(1+σ) and ǫ j ≫ kγai k2+2σ
≥ j−1 . We can put m j = kγ(0)ai k(1+σ) and
2( j−1)(2+2σ) kγ(0)ai k(1+σ) (2+2σ)
1
ǫj = j−1 , and then
2( j−1)(2+2σ) kγ(0)ai k(1+σ) (2+2σ)

P j−1
ln(m1 · · · m j−1 ) = ln mk
Pk=1
j−1 k−1
= k=1 σ(1 + σ) ln kγ(0)ai k
j−1
(1+σ) −1 (34)
= σ (1+σ)−1 ln kγ(0)ai k
= ((1 + σ) j−1 − 1) ln kγ(0)ai k

And j−1
( j−1)(2+2σ)
kγ(0)ai k(1+σ) (2+2σ)
− ln m j ǫ j = ln( m1j ǫ j ) = ln 2 j−1
kγ(0)ai k(1+σ) σ (35)
j−1
= ( j − 1)(2 + 2σ) ln 2 + (1 + σ) (2 + σ) ln kγ(0)ai k
Therefore,
ln(m1 ···m j−1 ) ((1+σ) j−1 −1) ln kγ(0)ai k
lim sup j→∞ − ln m j ǫ j = lim sup j→∞ ( j−1)(2+2σ) ln 2+(1+σ) j−1 (2+σ) ln kγ(0)ai k
1
(36)
= 2+σ
T 1
Then from the above theorem it follows that if we define F = j≥0 F j , then dimH F ≥ 2+σ . On the other
T ǫ ǫ
hand, for any point x ∈ F, suppose x ∈ j≥0 I j , where I j = [γ( j)ai − kγ( j)ai k2 , γ( j)ai + kγ( j)ai k2 ] is a closed
interval of F j , then from our construction, we have kγ( j)ai k ∈ [kγ( j − 1)ai k1+σ , 2kγ( j − 1)ai k1+σ ] and
γ( j) = γ( j − 1)ui (k( j)) for k( j) > N0 (in fact k( j) ≍ kγ( j − 1)ai kσ ), which follows from Proposition 2.4
that G x enters γ( j − 1)ai and then γ( j)ai and moreover from leaving γ( j − 1)ai to entering γ( j)ai it spends
at most s0 . Then for x, {γ( j)ai } j≥0 can be chosen as the subsequence of the spectrum of x satisfying the
conditions in Definition 2.1. This shows that F ⊂ ∆σ .
1
Thus we prove that dimH ∆σ ≥ 2+σ .

Remark: this result is a generalization of Lemma 4.3 in [2].

3 EFFECTIVE COUNTING ON CUSPS


In this section, we consider the following counting problem: given a interval I = [x − d, x + d],
and some large number h, we wonder if there is an effective estimate about the number of the cusps
{γ∞ ∈ I : kγ∞k ∈ [c1 h, c2 h]} for some constants c1 and c2 .
We fix a function f ∈ Cc∞ (G) such that 0 ≤ f ≤ 1 and it is supported on a neighborhood Ω of e,
such that Ω is contained in D0 × Kǫ , where D0 is a fundamental domain of Γ-action on H2 , and Kǫ ⊂ K is
 cos θ sin θ 
a neighborhood of e in K. If we represent every element in K by e(θ) =  , we can choose
− sin θ cos θ
Kǫ = {e(θ) : θ ∈ [−π/10, π/10]}. Then we can identify f with its projection onto Cc∞ (Γ \ G), and let f
satisfy the following conditions: there exists another neighborhood Ω′ ⊂ Ω such that f = 1 on Ω′ , we
denote by S(ψ) the Sobolev norm of a function ψ ∈ H2 (Γ \ G). Let R (and L respectively) denote the
right (and left, respectively) regular action on L2 (G), i.e., R(g)ϕ(x) = ϕ(xg) ( and L(g)ϕ(x) = ϕ(gx)) for

15
g ∈ G and ϕ ∈ L2 (G). It is easy to define similarly the regular right action on L2 (Γ \ G), which is also
denoted by R(g). Then there exists a constant κ such that S(R(t) f)
S( f ) ≤ κ for all t ∈ K, for K is compact.
For γ ∈ Γ, we consider the NAK-decomposition of γ = n(γ)a(γ)k(γ). Then f and L(γ−1 ) f have the
same projection on L2 (Γ\G). We denote R(k(γ))◦L(γ−1 ) f by fγ . It is easy to see that fγ is also supported
on a single fundamental domain T1 (γD0 ), so we can also treat it as a function on Γ \ G. Moreover, fγ is
supported on γΩk(γ)−1 = n(γ)a(γ)(k(γ)Ωk(γ)−1), and on γΩ′ k(γ)−1 = n(γ)a(γ)(k(γ)Ω′k(γ)−1 ), fγ = 1.
If necessary, we may make Ω and Ω′ smaller such that there exists two neighborhoods of identity
Ξ′ ⊂ Ξ ⊂ D0 × Kǫ , satisfying the following: for any k ∈ K, Ξ′ ⊂ kΩ′ k−1 ⊂ kΩk−1 ⊂ Ξ. This can be
done since the subgroup K = SO(2, R) is compact. Then for any γ ∈ Γ, fγ is supported on n(γ)a(γ)Ξ
and on n(γ)a(γ)Ξ′, fγ = 1 (note that NA-component does not act on the K-component). Moreover, if
we treat fγ as a function on Γ \ G, fγ = R(k(γ)) f . Then we consider the evolution of fγ under geodesic
 
et 
flow, say R(at ) fγ , where at is the diagonal element  −t 
. At first we regard R(at ) fγ as a function on
e
G, then R(at ) fγ is supported on n(γ)a(γ)Ξa−t , then if we project it to H2  G/K, it is just the image of
 
ω1 ω2 
this subset acting on ı. Take ξ =    ∈ Ξ, then
ω3 ω4
2t
n(γ)a(γ)ξa−t ı = n(γ)a(γ) ωω31ı+ω
ı+ω2 e
4e
2t (37)
   
1 χ(γ) λ−1 (γ) 
If we put n(γ) =  , and a(γ) =  . Then
0 1 λ(γ)
2t
1 ω1 ı+ω2 e2t
n(γ)a(γ) ωω31ı+ω
ı+ω2 e
4e
2t = λ(γ)2 ω3 ı+ω4 e2t
+ χ(γ)
1 ω1 ω3 +ω2 ω4 e4t 1 e 2t (38)
= λ(γ)2 ω23 +ω24 e4t
+ χ(γ) + ı λ(γ)2
ω2 +ω2 e4t
3 4

 
ω1 ω2  e2t
If necessary, we can make Ω and Ω′ smaller such that for each ξ =   ∈ Ξ, we have
ω23 +ω24 e4t

ω3 ω4
4t
[ 43 e−2t , 34 e−2t ] and ω1 ωω32 +ω2 ω4 e
2 4t ∈ [− 14 , 41 ] for t ≥ 10. Then we have that the projection of n(γ)a(γ)Ξa−t
3 +ω4 e
onto G/K  H2 is contained in
( " #)
1 1 3 1 4 1
x + yı : |x − χ(γ)| ≤ and y ∈ , (39)
4 λ(γ)2 4 λ(γ)2 e2t 3 λ(γ)2 e2t
Now we consider the inner product of R(at ) fγ and L(γ′−1 ) fγ for some γ′ ∈ Γ. Then from our choice
of f , we know that
Z
R(at ) fγ · L(γ′−1 ) fγ dµG (g) ≤ µG (γ′ n(γ)a(γ)Ξ ∩ n(γ)a(γ)Ξa−t ) (40)
G

Since G is semisimple and so unimodular, then µG is both left and right invariant. Therefore,

µG (γ′ n(γ)a(γ)Ξ ∩ n(γ)a(γ)Ξa−t ) = µG (g(γ, γ′ )Ξ ∩ Ξa−t )

where g(γ, γ′ ) = a(γ)−1 n(γ)−1 γ′ n(γ)a(γ).


In [7], the following proposition is proved:

16
Proposition 3.1. See [7, Proposition 3.2] Let G be a real algebraic group, and σ is an involution of G.
Let A = {at } be a one parameter subgroup of G, such that σ(at ) = a−t . Let

H = {g ∈ G : σ(g) = g} U + = {g ∈ G : a−t gat → e as t → ∞}



(41)
U = {g ∈ G : at ga−t → e as t → ∞} Z = ZG (A)

Then there exist constant c and t0 such that for neighborhoods Hr1 ⊂ H, Zr2 ⊂ Z and Ur+3 of e in H, Z
and U + , respectively, small enough, t > t0 and any g ∈ G, we have

µG (gUr+3 Zr2 Hr1 ∩ Ur+3 Zr2 Hr1 a−t ) ≤ ce−2t µU+ (Ur+3 )2 (42)

where µU+ denotes the Haar measure on U + .

To apply the above proposition, we put G = SL(2, R) and σ(g) = (gt )−1 , then H = K, Z = A and
U + = N. Therefore from the above proposition we have that:
Z
R(at ) fγ · L(γ′−1 ) fγ dµG (g) ≤ µG (g(γ, γ′ )Ξ ∩ Ξa−t ) ≤ ce−2t µN (Nr )2
G

where Nr is some small neighborhood of e in N, determined by Ξ. Combining this with the fact that
fγ = 1 on Ξ′ ⊂ Ξ, we have k fγ k22 ≥ µG (Ξ′ ) ≥ c′ µN (Nr )2 . Therefore, the following statement holds:

Proposition 3.2. There exist constants c̃ and t0 such that for t > t0 ,
Z
R(at ) fγ · L(γ′−1 ) fγ dµG (g) ≤ c̃e−2t k fγ k22 (43)
G

where k · k2 denotes the L2 -norm in L2 (G). Note that k fγ k2 = k f k2 is independent of γ.

On the other hand, we can also treat fγ and R(at ) fγ as functions on Γ \ G. We will apply the
following theorem due to Moore:

Theorem 3.1. See [8, Theorem 1.1] Let M be a Riemann surface of constant negative curvature equal
to −1 of finite volume, and let T1 (M) be the unit tangent bundle of M. Let ϕ and ψ be in L2 (T1 (M))
orthogonal to the constants, and satisfying the Hölder condition in the fiber direction for some α > 1/2.
Then if gt is the geodesic flow, we have

|(g∗t ϕ, ψ)| ≤ Le−c|t| (44)

where c depends only on M and L depends on the Hölder norms of ϕ and ψ.

In the above theorem, the Hölder condition of ϕ in the fiber direction for α is defined as follows:

for every x ∈ T1 (M), if we define H(ϕ)(x) = supt∈K |t ϕ(x)−ϕ(x)|
|t|α , where |t| denotes some translate invariant
q
norm on K, then H(ϕ) ∈ L2 (T2 (M)), and the Hölder norm of ϕ is defined to be kϕk22 + kH(ϕ)k22 .
Here T1 (M) is the same as Γ \ G, the geodesic flow acting on L2 (Γ \ G) is the same as R(at ) action,
and from the equivalence between Hölder norm and Sobolev norm, we have the following:
Z Z Z
fγ · R(at ) fγ dµΓ\G − fγ dµΓ\G R(at ) fγ dµΓ\G ≤ LS( fγ )2 e−c|t| (45)
Γ\G Γ\G Γ\G

17
R R
where L and c depend only on Γ. Combining with S( fγ ) ≤ κS( f ) and Γ\G
R(at ) fγ dµΓ\G = f dµΓ\G ,
Γ\G γ
we have Z Z !2
fγ · R(at ) fγ dµΓ\G − fγ dµΓ\G ≤ L′ S( f )2 e−c|t| (46)
Γ\G Γ\G
R R
where L′ and c depend only on Γ. Note that Γ\G
fγ dµΓ\G = Γ\G
f dµΓ\G , we have
Z Z !2
fγ · R(at ) fγ dµΓ\G ≥ f dµΓ\G − L′ S( f )2 e−c|t| (47)
Γ\G Γ\G

Therefore, there exists T 0 > 0 such that for t > T 0 ,


Z Z !2
9
fγ · R(at ) fγ dµΓ\G ≥ f dµΓ\G (48)
Γ\G 10 Γ\G

On the other hand, Z XZ


fγ · R(at ) fγ dµΓ\G = L(γ′ ) fγ · R(at ) fγ dµG (49)
Γ\G γ′ ∈Γ G

Here, on the right hand side, we treat fγ and R(at ) fγ as functions on G.


Define ( Z )
N(t, γ) = γ′ ∈ Γ : L(γ′ ) fγ · R(t) fγ dµG > 0 (50)
G

Then from Proposition 3.2, we have that:


R R
L(γ′ ) fγ · R(at ) fγ dµG
P
f · R(at ) fγ dµΓ\G
Γ\G γ
= γ′ ∈Γ G
R
P ′ (51)
= γ′ ∈N(t,γ) G L(γ ) fγ · R(at ) fγ dµG
−2t
≤ |N(t, γ)|ce k f k22

where |N(t, γ)| denotes the number of elements in N(t, γ). Therefore we have
Z !2
9
|N(t, γ)|ce−2t k f k22 ≥ f dµΓ\G (52)
10 Γ\G

Therefore Z !2
9 1 2t
|N(t, γ)| ≥ f dµΓ\G e (53)
10 Γ\G ck f k22
 
γ1 γ2 
Now, for a fixed cusp γ∞, we may choose a representative γ ∈ γΓ∞ , such that if γ =  , then
γ3 γ4
    
γ1 γ2  1 n γ1 nγ1 + γ2 
|γ4 | < |γ3 |. This can be done since for n ∈ Z,     =  , then we can choose
γ3 γ4 0 1 γ3 nγ3 + γ4
suitable n such that the condition above is satisfied. Then for any γ′ ∈ N(t, γ), we have γ′−1 γΩk(γ)−1 ∩
γΩk(γ)−1 a−t , ∅. Every element on the right hand side corresponds to an element (x + yı, yv2 ) ∈ T1 (H2 ),
where x + yı is containedin the set described
 in (39) and v is very close to i. And on the right hand side if
γ”1 γ”2 
we denote γ” = γ′−1 γ =  , then there exists ω ∈ Ω close to e such that γ”ωk(γ)−1 = (x+yı, yv2 ),
γ”3 γ”4

18
where v is very close to i. Then if we consider the
 NAK-decomposition  of γ” = n(γ”)a(γ”)k(γ”),
 then 
 cos(θ) sin(θ)   cos(θ”) sin(θ”) 
k(γ”) is very close to k(γ). Then we put k(γ) =   , and k(γ”) = 
  ,
− sin(θ) cos(θ) − sin(θ”) cos(θ”)
γ”3
it is easy to check that tan(θ) = − γγ43 and tan(θ”) = − γ” 4
. So from our previous choice of γ we have
γ”1 ı+γ”2 γ”1 γ”3 +γ”2 γ”4 1
|γ”4 | < |γ”3 |. Moreover, γ”ı = γ”3 ı+γ”4 = γ”23 +γ”24
+ ı γ”2 +γ” 2 , if we put γ”4 = kγ”3 , then |k| < 1, and
3 4
thus
1 1 1 1 1 1
= ∈[ , ] (54)
γ”23 + γ”24 γ”23 1 + k2 2 γ”23 γ”23
and
γ”1 γ”1 γ”3 +γ”2 γ”4 |γ”4 |
| γ” 3
− γ”23 +γ”24
| = (γ”23 +γ”24 )|γ”3 |
1 k
= γ”23 1+k2 (55)
1
≤ 2γ”23

1 γ1 1 1 1 1
The same holds for γ also. And since γı = χ(γ) + ı λ(γ)2 , we have |χ(γ) − γ3 | < 2γ32 and λ(γ)2 ∈ [ 2γ32 , γ32 ].
Therefore for any γ′ ∈ N(t, γ), if we set γ” = γ′−1 γ, then |γ”∞ − χ(γ)| ≤ 4λ(γ)1 1
2 + 2kγ”∞k2 and kγ”∞k ∈
2

[ 83 λ(γ)2 e2t , 43 λ(γ)2 e2t ] ⊂ [ 83 kγ∞k2 e2t , 38 kγ∞k2 e2t ]. For t > 10, we have 1
|γ”∞ − χ(γ)| < 2λ(γ)2 , and since
1 1 1
|χ(γ) − γ∞| < 2λ(γ) 2 , then |γ”∞ − γ∞| < λ(γ)2 ≤ kγ∞k2 .
1 1 t
Then if the interval I is of form [γ∞ − kγ∞k 2 , γ∞ + kγ∞k2 ] for some cusp point γ∞, and h = kγ∞ke
q q
and c1 = 38 and c2 = 83 , then from the previous argument, we have that there exists some constant
R 2
9
R = 10 Γ\G
f dµΓ\G ck1f k2 , such that
2

1 2
|{γ”∞ ∈ I : kγ”∞k ∈ [c1 h, c2 h]}| ≥ Re2t = Rh |I| (56)
2
where |I| denotes the length of I.
Remark: this is a generalization of [3, Theorem 2] from Γ = SL(2, Z) to any lattice Γ.
At the end of the section, we prove a basic fact concerning cusp points: for any interval I of length
|I| ≍ X1 , there exists at least one cusp point γai ∈ I with kγai k ≤ X, for all large X > 0.
In fact, for an interval I = [x − cX −1 , x + cX −1 ], where c is a constant to be determined, we
consider the spectrum of x, say {γ(n)a(n)}, and take the last one γ(m)a(m) in it with height kγ(m)a(m)k ≤
1
X, then we claim that γ(m)a(m) ∈ I. By 29, we have |x − γ(m)a(m)| ≍ kγ(m)a(m)kkγ(m+1)a(m+1)k , for
−1
kγ(m + 1)a(m + 1)k > X, we have |x − γ(m)a(m)| < cX for some constant c. Then we choose this
constant in our previous description of I. To sum up, we have the following proposition:

Proposition 3.3. There exists a constant c only depending on the space Γ \ H2 , such that, for any large
number X > 0 and any interval I of length 2c, there exists at least one cusp point γai ∈ I with kγai k ≤ X.

19
4 GEODESIC FLOWS ON PRODUCT SPACE Γ1 \H2 ×· · ·×Γk \H2
4.1 LOWER BOUND
For product of hyperbolic two dimensional spaces, Γ1 \ H2 × · · · × Γk \ H2 , where Γ1 , . . . , Γk ⊂
SL(2, R) are lattices of SL(2, R), i.e., Mi = Γi \ H2 admits a finite SL(2, R) invariant volume. Then the
geodesic flow gt (x1 , . . . , xk ), where (x1 . . . , xk )∈ M1 × · · · , ×Mk corresponds to the point Γ1 × · · · × Γk \
1 x
(ı, . . . , ı)u(x1 ) × · · · × u(xk ), with u(x) =  , corresponds to the projection of the following geodesic
0 1
flow of H2 × · · · × H2 onto M1 × · · · × Mk :
 ı ı 
G(x1 ...xk ) = (x1 + 2s , . . . , xk + 2s ) ∈ H2 × · · · × H2 : s ∈ R (57)
e e
For each xi we define W xi (t) = 1
where γai is the cusp point of Γi which xi + e−2t ı
max(kγai k2 e−2t ,kγai k2 |xi −γai |2 e2t )
 √ −1 
 li 0  −1
is close to. From our previous calculation it is easy to see that W xi (t) ≍ W xi (t) = ℑ  √li √  γ (xi +
− ai li
−2t
e ı). Then for the product space, for x = (x1 , . . . , xk ), we define Wx (t) = max1≤i≤k W xi (t). Then Gx is
divergent, i.e., for any compact subset K ⊂ M1 × · · · × Mk , there exists a constant T depending on K,
such that for any t > T , Gx is always outside K, if and only if Wx (t) → ∞ as t → ∞.
At first, we consider the case when k = 2. as we have done in the previous section, we suppose
that {γ(n)a(n)} and {γ′ (m)a′ (m)} are the spectra of x1 and x2 respectively, and {tn } and {tm′ } are the
sequences of cusp-changing times for x1 and x2 respectively, i.e., for x1 (and for x2 respectively,), when
t ∈ [tn , tn+1 ) (and t ∈ [tm′ , tm+1
′ 1
) respectively), we have W x1 (t) = max(kγ(n)a(n)k2 e−2t ,kγ(n)a(n)k2 |x −γ(n)a(n)|2 e2t ) ,
1
1
(and W x2 (t) = max(kγ′ (m)a′ (m)k2 e−2t ,kγ′ (m)a′ (m)k2 |x −γ′ (m)a′ (m)|2 e2t ) respectively). In the previous section we have
2

proved that e ≍ kγ(n)a(n)k (and e2tm ≍ kγ′ (m)a′ (m)k2 respectively). Then for x = (x1 , x2 ), we can
2tn 2

find a sequence of times {τi } such that at each time t = τi , Wx (t) admits local minimum. Then when
t = τi , either Wx (t) changes from W x1 (t) to W x2 (t) or it changes conversely. Suppose the formal happens,
then suppose τi ∈ [tn , tn+1 ) ∩ [tm′ , tm+1

), since we assume that Wx (t) changes from W x1 (t) to W x2 (t), we
have on the left of τi Wx (t) is decreasing and on the right it is increasing. That means on the left of τi ,
Wx (t) = W x1 (t) = kγ(n)a(n)k2 |x11−γ(n)a(n)|2 e2t and on the right of τi Wx (t) = W x2 (t) = kγ′ (m)a′1(m)k2 e−2t , thus at
t = τi we have the following:

kγ(n)a(n)k2e2τi |x1 − γ(n)a(n)|2 = kγ′ (m)a′ (m)k2 e−2τi

Then this value is equal to their geometric mean, i.e.,


q
kγ(n)a(n)k2e2τi |x1 − γ(n)a(n)|2kγ′ (m)a′ (m)k2 e−2τi = kγ(n)a(n)kkγ′(m)a′ (m)k|x1 − γ(n)a(n)|
1
We have proved that |x − γ(n)a(n)| ≍ kγ(n)a(n)kkγ(n+1)a(n+1)k , therefore the above value is asymptotically
kγ′ (m)a′ (m)k
equal to kγ(n+1)a(n+1)k . So we have

kγ(n + 1)a(n + 1)k


Wx (τi ) ≍
kγ′ (m)a′ (m)k

20
. Similarly, if the latter happens, then
kγ′ (m + 1)a′ (m + 1)k
Wx (τi ) ≍
kγ(n)a(n)k
Then Wx (t) → ∞ if and only if Wx (τi ) → ∞ and this happens if and only if
kγ(n)a(n)k kγ′ (m)a′ (m)k
max( , )→∞
kγ′ (m)a′ (m)k kγ(n)a(n)k
as min(kγ(n)a(n)k, kγ′(m)a′ (m)k) → ∞. In other words, given any large number M, their exists some
number R such that for any spectral cusp γ(n)a(n) of x1 and γ′ (m)a′ (m) of x2 , if min(kγ(n)a(n)k, kγ′(m)a′ (m)k) ≥
R, then
kγ(n)a(n)k kγ′ (m)a′ (m)k
max( , )≥M
kγ′ (m)a′ (m)k kγ(n)a(n)k
In this section, we will prove the following proposition:

Proposition 4.1. For any x1 ∈ ∆σ for Γ1 \ H2 , then the set D(x1 ) = {x2 ∈ R : G(x,y) is divergent } has
Hausdorff dimension 1.

We also quote the following result from [2]:

Proposition 4.2. Let F ⊂ R2 and E its projection onto the x-axis. For each x ∈ E, let Lx be the line
{(x, y) : y ∈ R}. If dimH (F ∩ L x ) ≥ t for all x ∈ E, then

dimH F ≥ t + dimH E

This is Lemma 4.1 in [2] and Corollary 7.12 in [6].


Combining the above two propositions, we have the following:

Proposition 4.3. If we define D2 = {x ∈ R2 : Gx is divergent }, then we have:


3
dimH D2 ≥
2
Then for any product space Γ1 \ H2 × · · · × Γk \ H2 , for any x ∈ Rk , since we just need Wx (t) =
max(W x1 (t), . . . , W xk (t)) → ∞ (as t → ∞), and this happens if we have max(W x1 (t), W x2 (t)) → ∞ (as t →
∞). This shows the Hausdorff dimension of Dk = {x ∈ Rk : Gx is divergent } is at least dimH D2 + (k − 2)
(i.e., we just need G(x1 ,x2 ) to be divergent and let other components be chosen arbitrarily).
Summing up, assuming Proposition 4.1, we have the following theorem:

Theorem 4.1. Let Dk be defined as above, then we have


1
dimH Dk ≥ k − (58)
2
Now the only thing we need to do is to prove Proposition 4.1.

Proof of Proposition 4.1. For a fixed x1 ∈ ∆σ , suppose {γ(n)a(n)} is the refined sequence of cusps of x1 ,
such that

21
1. kγ(n + 1)a(n + 1)k ∈ [kγ(n)a(n)k1+σ, 2kγ(n)a(n)k1+σ]

2. From G x1 leaving γ(n)a(n) to entering γ(n + 1)a(n + 1), it spends at most s0 .


Let qn = kγ(n)a(n)k, hn = lnqqn n and h′n = qn ln qn . We start from some n0 large enough. We choose a cusp
γ′ (n0 )∞ such that kγ′ (n0 )∞k ∈ [c1 hn0 , c2 hn0 ], where c1 , c2 are given in previous section. And we define
Fn0 = In0 to be [γ′ (n0 )∞ − p q12 , γ′ (n0 )∞ + p q12 ], where p ≥ 1 is a constant to be chosen later. Suppose
n0 n0
Fn is already defined and is a union of finitely many disjoint closed intervals each of which is of form
[γ′ (n)∞ − p q12 , γ′ (n)∞ + p q12 ], where kγ′ (n)∞k ∈ [c1 hn , c2 hn ]. Then we construct Fn+1 as follows:
n n
We start with a closed interval of Fn , say In = [γ′ (n)∞ − q12 , γ′ (n)∞ + q12 ], We take γ is of form
n n
   
1 k  α j β j 
γ′ (n)u1(k)s j , where u1 (k) =   ∈ Γ∞ and s j =   ∈ S , then kγ∞k = |kτ j γ′ (n) + α j γ′ (n) +
3 3
0 1 τj ρj
 
γ′ (n) γ2′ (n)
τ j γ4′ (n)|, where γ′ (n) =  1′  and kγ′ (n)∞k = |γ3′ (n)|. We can choose k appropriately such
γ3 (n) γ4′ (n)
that kγ∞k ∈ [h′n − Chn , h′n + Chn ] where C is some constant depending on Γ2 . For n large enough we
have kγ∞k
h′n is very close to 1. And from the result we have proved, there exists a constant C such that

C
|γ′ (n)∞ − γ∞| ≤ kγ′ (n)∞kkγ∞k ≤ Cq2 , where C′ is a constant very close to C. Then we can choose p
n
properly such that I ′ = [γ∞ − kγ∞k1 1
2 , γ∞ + kγ∞k2 ] ⊂ I. Then from the result of previous section we have

the following
|{γ”∞ ∈ I ′ : kγ”∞k ∈ [c1 hn+1 , c2 hn+1 ]}| ≥ 12 R|I ′ |h2n+1
h2
= R (hn+1
′ 2 (59)
n)

= R( qn ln qqnn+1ln qn+1 )2
For each such γ”, we define a closed interval In+1 = [γ”∞ − p q21 , γ”∞ + p q21 ], and define Fn+1 to be
n+1 n+1
q
the union of all such intervals. Using the notation in Theorem 2.1, we can take m j = ( q j−1 ln q jj ln q j−1 )2 , and
(ln q j )2
from what we have proved, the gap between two intervals I j and I ′j of F j is at least Const. h12 ≍ q2j
,
j
(ln q )2
so we can take ǫ j = q2j . Then if we define F = j≥n0 F j , according to Theorem 2.1, we have
T
j

ln mn0 ···m j−1


dimH F ≥ lim sup j→∞ − ln(m j ǫ j ) (60)
q
Since m j = ( q j−1 ln q jj ln q j−1 )2 , then ln m j = 2 ln q j − 2 ln q j−1 − 2 ln ln q j − 2 ln ln q j−1 , in the computation
of the limit we can ignore the terms −2 ln ln q j and −2 ln ln q j−1 for they are much smaller than ln q j and
P j−1
ln q j−1 . Then ln mn0 · · · m j−1 can be estimated by i=n 0
2 ln qi −2 ln qi−1 = 2 ln q j−1 −2 ln qn0 −1 ≍ 2 ln q j−1 ,
(ln q j )2 q2j
for j large. And − ln(m j ǫ j ) = − ln( 2 2
q j q j−1 (ln q j )2 (ln q j−1 )2
) = 2 ln q j−1 + 2 ln ln q j−1 ≍ 2 ln q j−1 . So when j
ln mn0 ···m j−1 2 ln q j−1
is large enough, − ln(m j ǫ j ) ≍ 2 ln q j−1 = 1, and thus,

ln mn0 · · · m j−1
lim sup =1 (61)
j→∞ − ln(m j ǫ j )
This proves that dimH F = 1.
F j , x2 ∈ I j = [γ′ ( j)∞ − p q12 , γ′ ( j)∞ + p q12 ], for each
T
On the other hand, for any x2 ∈ F = j≥n0
j j

j, and kγ′ ( j)∞k ∈ [c1 h j , c2 h j ]. Then from our previous discussion about geodesic flows on Γ \ H2 and

22
distance between a limit point and a cusp, we have that G x2 enters the cusp γ′ ( j)∞ at some moment
q2j
and moreover, the next cusp G x2 enters has height ≍ hj = h′j . Then if we consider the pair (x1 , x2 ), for
x1 ∈ ∆σ , any cusp of x1 has height near kγ(n)a(n)k = qn for some n, and near qn , the largest cusp of x2
smaller than qn is near (up to a constant multiplication) hn = lnqqn n , and the smallest cusp of x2 larger than
qn is near (up to a constant multiplication) h′n = qn ln qn , this implies that for any cusp γ′ (m)a′ (m) of x2
whose height is large enough,

kγ(n)a(n)k kγ′ (m)a′ (m)k


max( , ) ≫ ln qn → ∞ (62)
kγ′ (m)a′ (m)k kγ(n)a(n)k
This shows that F ⊂ D(x1 ). Combining this with dimH F = 1, we complete the proof of Proposition
4.1.

4.2 UPPER BOUND


In this subsection, we will prove the following:

Theorem 4.2. For some small number δ > 0, if we define:

Ek (δ) = {x ∈ Rk : Wx (t) > δ−1 for any t > 0} (63)

Then there exists some small constant δ0 depending on k and the product space Γ1 \ H2 × · · · × Γk \ H2 ,
such that for all 0 < δ < δ0
For k = 2, we have
3
dimH E2 (δ) ≤ + δ1/3 (64)
2
And for k ≥ 3, we have: we have

dimH Ek (δ) ≤ k − 1/2 + δ1/6 (65)

Remark: this theorem is a generalization of the first part of [2, Theorem 6.1], and the main idea
of its proof is the same, but from the case when Γ = SL(2, Z) to general case when Γ is any lattice, we
have some difficulties to overcome. Here we follow the paper of Cheung to deal with the case k = 2 and
k ≥ 3 separately, since they are different in some essential manner.
In general, for any lattice Γ ⊂ SL(2, R), we use notations a, b, c to denote the cusp points of Γ in
R ⊂ ∂H2 . We want to give an upper bound of the number of cusp points lying in some interval I with
heights bounded by X. In particular, we will prove the following statement:

Proposition 4.4. Given an interval I and a positive number X, we have


X 1
≪ X|I| (66)
b∈Q(I,X)
kbk

where Q(I, X) = {b is a cusp point : b ∈ I, kbk ≤ X} and |I| denotes the length of I.

23
Proof. Take all cusps lying in I with height less than or equal to X and order them from small to large,
1
say Q(I, X) = {bi }i=1,...,N with bi < bi+1 for all i. From what we have proved, bi+1 − bi ≫ kbi+1 kkbi k , then
there exists some c > 0 such that bi+1 − bi ≥ kbi+1ckkbi k . Combining kbi+1 k ≤ X, we have
c
bi+1 − bi ≥
Xkbi k
Take the summation for i = 1, . . . , N, we have
c X 1
b N − b1 ≥
X b∈Q(I,X) kbk

For bN − b1 ≤ |I|, we have that


c X 1
|I| ≥
X b∈Q(I,X) kbk
And this leads to our conclusion.

Then for a fixed interval I if we define F I (X) = |Q(I, X)|, then the above statement implies that
Z
1
dF I (t) ≪ |I|X
t≤X t
R
If we define G I (X) = t≤X 1t dF I (t), then dG I (t) = 1t dF I (t) and G I (X) ≪ |I|X.
With this estimate, we are ready to follow Cheung’s proof in [2].
Case k = 2: For the product space Γ1 \ H2 × Γ2 \ H2 , we define Qi ⊂ R ⊂ G/P to be the set of
cusps of Γi \ H2 , for i = 1, 2, and we define Q = Q1 × Q2 . For (a1 , a2 ) ∈ Q, we define B(a1 , a2 ) to be the
1
neighborhood of (a1 , a2 ) consisting of points whose ith component is within the ka1 kka2k
neighborhood
of ai for i = 1, 2. Let J denote the subset of Q consisting of the elements satisfying the following
conditions:
ka1 k < δka2 k or ka2 k < δka1 k (67)
For any (a1 a2 ) ∈ J, we define σ(a1 , a2 ) ⊂ J to be the collection of elements (a′1 , a′2 ) ∈ J such that
Case 1 if ka1 k < δka2 k, then a′1 ⊢ a1 , a′2 |= a2 , and ka′2 k < δka′1 k
Case 2 if ka2 k < δka1 k, then a′2 ⊢ a2 , a′1 |= a1 , and ka′1 k < δka′2 k
Here a′ ⊢ a means there exists some vertical downward geodesic flow G x entering a and leaving
it for a′ and a′ |= a means there exists some vertical downward geodesic flow G x entering a and then
entering a′ after a while (but a′ might not be the very next cusp point G x leaves a for).
We recall the definition of self-similar covering of some subset E ⊂ R2 :

Definition 4.1. See [2, Definition 5.1 and Definition 5.2]


Let B be a countable covering of a subset E ⊂ R2 by bounded subsets of R2 and assume that it is
indexed by some countable set J; let σ be a function from the set J to the set of all nonempty subsets
of J. For any α ∈ J we write B(α) for the element of B indexed by α. We say (B, J, σ) is an indexed
self-similar covering of E if there exists a λ, 0 < λ < 1 such that for every x ∈ E we have a sequence
(α j ) of elements in J satisfying

24
1. ∩B(α j ) = {x}
2. diamB(α j+1) < λdiamB(α j ) for all j, and
3. α j+1 ∈ σ(α j ) for all j.
where diamB denotes the diameter of set B.

Let B denote the collection of all subsets of form B(a1, a2 ) where (a1 , a2 ) ∈ J, We will prove that
(B, J, σ) is a self-similar covering of E2 (δ). In the following reasoning, for simplicity, we treat A ≍ B
the same as A = B, and treat A ≪ B (A ≫ B respectively) the same as A ≤ B (A ≥ B respectively), i.e.,
we ignore the constant multiplications in estimate, this simplicity does not effect the final result.
In fact, for any x = (x1 , x2 ) ∈ E2 (δ), then we have Wx (t) = max(E x1 (t), E x2 (t)) > δ−1 for all t > 0.
And moreover, there is a sequence of times {tn } consisting of all local maxima of Wx (t). Take any tn , and
without loss of generality, we assume that Wx (tn ) = W x1 (tn ). Suppose at t = tn , G x1 is near an and denote
its next cusp point by an+1 . Then Wx (tn+1 ) = W x2 (tn+1 ), and if at this time G x2 is near bn , then from our
previous discussion on the properties of Wx (t), we have that between tn and tn+1 , there is a unique local
minimum τn of Wx (t) and W x1 (τn ) = W x2 (τn ) ≍ kakbn+1
nk
k
. Then since x ∈ E2 (δ), we have kan+1 k
kbn k > δ−1 , i.e.,
kbn k < δkan+1 k. And from the choice of an and bn , we have
1 1
|x1 − an+1 | ≤ <
kan+1 k2 kan+1 kkbn k
Moreover, when t = tn+1 , Wx (tn+1 ) = W x2 (tn+1 ), and at this time, G x2 is near bn , and we denote its next
cusp by bn+1 , and if we consider the next local minima tn+2 , we have Wx (tn+2 ) = W x1 (tn+2 ), and suppose
at this time G x1 is near cusp an+2 , then an+2 |= an+1 , so kan+2 k ≫ kan+1 k, and at the unique minimum
kbn+1 k
τn+1 of Wx (t) between tn+1 and tn+2 we have Wx (τn+1 ) ≍ ka n+2 k
> δ−1 . This means kbn+1 k > δ−1 kan+2 k >
kan+2 k ≫ kan+1 k, therefore,
1 1
|x2 − bn | ≍ ≤
kbn kkbn+1 k kan+1 kkbn k
From the above argument, we have x ∈ B(an+1, bn ), and for the same reason, x ∈ B(an+2, bn+1 ), thus
x ∈ ∩∞
n=1 B(an+1 , bn ). Moreover, we have an+2 |= an+1 and bn+1 ⊢ bn , so (an+2 , bn+1 ) ∈ σ(an+1 , bn ). Finally,
it is easy to see that
1 1
diamB(an+2 , bn+1 ) = <δ = δdiamB(an+1, bn )
kan+2 kkbn+1 k kan+1 kkbn k
Combining all the facts we get above, we conclude that (B, J, σ) is a self-similar covering of E2 (δ).
To get the upper bound of the Hausdorff dimension of E2 (δ), we need the following theorem proved
in [2].

Theorem 4.3. (See [2, Theorem 5.3])


Let (B, J, σ) be an indexed self-similar covering of a subset E ⊂ R2 and suppose there is an s > 0
such that for every α ∈ J X
(diamB(α′)) s ≤ (diamB(α)) s (68)
α′ ∈σ(α)

Then dimH E ≤ s.

25
For a subset B(a1 , a2 ) ∈ B, its diameter diamB(a1, a2 ) ≍ 1
ka1 kka2 k . Then for every (a′1 , a′2 ) ∈ σ(a1 , a2 ),

diamB(a′1, a′2 ) ka1 kka2 k



diamB(a1, a2 ) ka′1 kka′2 k

 s  s
P diamB(a′1 ,a′2 ) P ka1 kka2 k
(a′1 ,a′2 )∈σ(a1 ,a2 ) diamB(a1 ,a2 ) ≍ (a′1 ,a′2 )∈σ(a1 ,a2 ) ka′1 kka′2 k
(69)
ka′ k ka′ k
For symmetry, we assume that ka1 k < δka2 k, if we put ka11 k = a and ka22 k = b, then from the definition of
J, we have a > δ−2 and for fixed a > 0, then b < δa. Then for fixed a > δ−2 , since a′1 ⊢ a1 , there are at
ka′ k
most C̃ choices of a′1 such that ka11 k ∈ [a, a + 1), where C̃ is a constant. Since we ignore all the constant
multiplications and treat them as 1 we may treat C̃ = 1. Note that a′ |= a implies |a′ − a| ≪ kak−2 . We
define
C(a2 , X) = {b is a cusp of Γ2 : |b − a2 | < ka2 k−2 : kbk ≤ Xka2 k}
R R
And define F(X) = |C(a2 , X)|, and G(X) = t≤X 1t dF(t) = ka2 k t≤X tka12 k dF(t), then from Proposition 4.4,
we have
tka2 k
G(t) ≤ ka2 k =t
ka2 k2
Then for any s ∈ ( 23 , 2), we have
 s  s
P diamB(a′1 ,a′2 ) P ka1 kka2 k
(a′1 ,a′2 )∈σ(a1 ,a2 ) diamB(a1 ,a2 ) ≍ (a′1 ,a′2 )∈σ(a1 ,a2 )
ka′1 kka′2 k
R (70)
P 1 1
≤ a>δ−2 a s b≤δa b s−1 dG(b)

Here for the integral we apply integration by parts and the fact G(b) ≤ b:
R 1
R
b≤δa b s−1
dG(b) = (δa)1s−1 G(δa) − b≤δa G(b)db1−s
R
= (δa)1s−1 G(δa) + (s − 1) b≤δa b−sG(b)db
R
≤ (δa)1s−1 (δa) + (s − 1) b≤δa b−s bdb
R (71)
= (δa)2−s + (s − 1) b≤δa b1−s db
s−1
= (δa)2−s + −s+2 (δa)2−s
1 2−s
= 2−s (δa)

We plug it into (70), we have:


 s
P diamB(a′1 ,a′2 ) P 1 1 2−s
(a′1 ,a′2 )∈σ(a1 ,a2 ) diamB(a1 ,a2 ) = a>δ−2 a s 2−s (δa)
δ2−s P 2−2s
= 2−s a>δ−2 a (72)
δ2−s 1 −2 3−2s
= 2−s 2s−3 (δ )
δ3s−4
= (2−s)(2s−3)

1
1 δ3s−4 δ2 1
Then for s = 3
2 + δ 3 , then (2−s)(2s−3) ≍ 1
1 = δ 6 , thus for δ < 2−7 , the above sum is less than 1. This
2 2δ
3

completes the proof of Theorem 4.2

26
Case k > 2: the same as above, let Qi denote the set of cusp points of Γi \ H2 , and we define
Q(k) = Q1 × · · · × Qk × {1, . . . , k} × {1, . . . , k}. For each fixed i ∈ {1, . . . , k} and X > 0, from Proposition
3.3, we may choose a subset Q(X, i) = {γ(n, X, i)a(n, X, i)}n∈Z ⊂ Qi such that

−∞ < · · · < γ(0, X, i)a(0, X, i) < γ(1, X, i)a(1, X, i) < · · · < γ(n, X, i)a(n, X, i) < · · · < +∞

with γ(n + 1, X, i)a(n + 1, X, i) − γ(n, X, i)a(n, X, i) ≍ X −1 for all n ∈ Z, and moreover, for any x ∈ R,
we may choose the cusp γ(n, X, i)a(n, X, i) ∈ Q(X, i) nearest x such that |x − γ(n, X, i)a(n, X, i)| ≪ X −1 .
With this we define the subset J ⊂ Q(k) consisting of elements (a1 , . . . , ak , i, j) ∈ Q(k) satisfying the
following conditions:

ka j k < δkai k and al ∈ Q(kai kka j k, l) for l , i, j (73)

And for each (a1 , . . . , ak , i, j) ∈ J, we define B(a1, . . . , ak , i, j) to be the neighborhood of (a1 , . . . , ak )


1
whose lth component is within the kai kka jk
neighborhood of al . And let A(a1 , . . . , ak , i, j) ⊂ B(a1, . . . , ak , i, j)
be its subset consisting of elements whose ith component is within the ka1i k2 neighborhood of ai . We de-
note B by the collections of all subsets of form B(a1, . . . , ak , i, j) where (a1 , . . . , ak , i, j) ∈ J. Then we
define σ(a1 , . . . , ak , i, j) ⊂ J to be the subset consisting of elements (a′1 , . . . , a′k , j, l) ∈ J satisfying the
following conditions:

1. a′j ⊢ a j

2. kai k < ka′j k and ka j k < ka′l k

3. A(a1 . . . , ak , i, j) ∩ A(a′1 , . . . , a′k , j, l) , ∅

4. kai k2 < ka j kkal k

We may show that (B, J, σ) is a self-similar covering of Ek (δ).


In fact, for any x = (x1 , . . . , xk ) ∈ Ek (δ), we consider the function Wx (t) = max1≤i≤k W xi (t). Then
there is a sequence of times {t p } such that Wx (t) admits local maxima at each t p . Suppose Wx (t p ) =
W xi(p) (t p ), then if ai(p) and bi(p) be the cusps of xi(p) such that G xi(p) enters the formal one right before time
1
t p and enters the latter one right after time t p . From the property of W x (t), we have t p = 21 ln |xi(p)−a i(p) |

1
2 ln ka i(p) kkb i(p) k. Between t p an t p+1 , there is a unique local minimum of Wx (t) at t = τ p , and from our
1 kbi(p) k
previous discussion concerning Wx (t), we have τ p ≍ 2 ln kai(p+1) kkbi(p) k and Wx (τ p ) ≍ kai(p+1) k , then since
x ∈ Ek (δ) we have kai(p+1) k < δkbi(p) k.
We choose a subsequence of {(ai(p) , bi(p) , i(p))} as follows: given (ai(pl ) , bi(pl ) , i(pl )), we define pl+1
to be the smallest p such that kbi(pl ) k2 < kbi(p) kkai(p+1) k, then we have kbi(pl ) k2 < kbi(pl+1 ) kkai(pl+2 ) k, and
moreover we have
kbi(pl ) k2 ≥ kbi(pl+1 −1) kkai(pl+1 ) k ≥ δ−1 kai(pl+1 ) k2

For simplicity, we write (ai(l) , bi(l) , i(l)) instead of (ai(pl ) , bi(pl ) , i(pl )), then we have

kai(l+1) k ≤ δkbi(l) k and kbi(l) k2 < kbi(l+1) kkai(l+2) k (74)

27
and moreover ai(l) and bi(l) are consecutive cusps in the spectrum of xi(l) .
Then for each l, we define (c1 , . . . , ck , i(l), i(l + 1)) ∈ J as follows: ci(l) = bi(l) , ci(l+1) = ai(l+1) ,
and for other subindex l, we choose cl to be the cusp in Q(kci(l) kkci(l+1) k, l) nearest xl . Then we have
(c1 , . . . , ck , i(l), i(l + 1)) ∈ J. And moreover, since ai(l+1) and bi(l+1) are consecutive cusps in the spectrum
of xi(l+1) , we have
1 1 1
|xi(l+1) − ai(l+1) | ≍ < =
kai(l+1) kkbi(l+1) k kai(l+1) kkbi(l) k kci(l+1) kkci(l) k
And also we have
1 1
|xi(l) − bi(l) | ≪ =
kbi(l) k2 kci(l) k2
1
For other subindex l, we have |xl − cl | < kci(l) kkci(l+1) k from our choice of cl . Therefore we have

x ∈ A(c1 , . . . , ck , i(l), i(l + 1)) ⊂ B(c1 , . . . , ck , i(l), i(l + 1))


T
Since diamB(c1, . . . , ck , i(l), i(l + 1)) → 0 as l → ∞, we have l B(c1 , . . . , ck , i(l), i(l + 1)) = {x}
Next, we need to check that (c′1 , . . . , c′k , i(l+1), i(l+2)) ∈ σ(c1 , . . . , ck , i(l), i(l+1)) where (c′1 , . . . , c′k , i(l+
1), i(l + 2)) is the element in J defined as above for l + 1. In fact, at first, we have c′i(l+1) = bi(l+1) ⊢
ai(l+1) = ci(l+1) . Secondly, we have ci(l) = bi(l) , c′i(l+1) = bi(l+1) , ci(l+1) = ai(l+1) and c′i(l+2) = ai(l+2) ,
from our definition of (ai(l) , bi(l) , i(l)), we have kbi(l) k < kbi(l+1) k and kai(l+1) k < kai(l+2) k. Also, x ∈
A(c1 , . . . , ck , i(l), i(l + 1)) ∩ A(c′1 , . . . , c′k , i(l + 1), i(l + 2)) implies it is not empty. Finally, from our con-
struction, we have
kci(l) k2 = kbi(l) k2 < kbi(l+1) kkai(l+2) k = kc′i(l+1) kkc′i(l+2) k
Therefore (c′1 , . . . , c′k , i(l + 1), i(l + 2)) satisfies all conditions of σ(c1 , . . . , ck , i(l), i(l + 1)), this shows that
(c′1 , . . . , c′k , i(l + 1), i(l + 2)) ∈ σ(c1 , . . . , ck , i(l), i(l + 1)).
Finally we have to show that there exists a constant λ such that

diamB(c′1, . . . , c′k , i(l + 1), i(l + 2)) < λdiamB(c1, . . . , ck , i(l), i(l + 1))

In fact, diamB(c′1, . . . , c′k , i(l + 1), i(l + 2)) = kbi(l+1) kka


1
i(l+2) k
and diamB(c1, . . . , ck , i(l), i(l + 1)) = 1
kbi(l) kkai(l+1) k ,
and from our construction of (ai(l) , bi(l) , i(l)), we have
√ √
kai(l+1) kkbi(l) k < δkbi(l) k2 < δkbi(l+1) kkai(l+2) k

Then we have

diamB(c′1, . . . , c′k , i(l + 1), i(l + 2)) < δdiamB(c1 , . . . , ck , i(l), i(l + 1))

Summing up all argument above, we have (B, J, σ) is a self-similar covering of Ek (δ).


Next for some fixed (c1 , . . . , ck , i, j) ∈ J, in order to apply Theorem 4.3, for some s > 0, we need
to estimate the following summation:
!s
X diamB(c′1, . . . , c′k , j, l)
(75)
(c′1 ,...,c′k , j,l)∈σ(c1 ,...,ck ,i, j)
diamB(c1, . . . , ck , i, j)

28
For fixed subindex l, we define

σl (c1 , . . . , ck , i, j) = {(c′1 , . . . , c′k , j, l) ∈ σ(c1 , . . . , ck , i, j)} ⊂ σ(c1 , . . . , ck , i, j)

At first, we study the summation


!s
X diamB(c′1, . . . , c′k , j, l)
(c′1 ,...,c′k , j,l)∈σl (c1 ,...,ck ,i, j)
diamB(c1, . . . , ck , i, j)

for fixed subindex l.


In fact, the above is equal to

 kci kkc j k  s


 
X
 ′ ′ 
(c′1 ,...,c′k , j,l)∈σl (c1 ,...,c ,i, j)
kc j kkcl k
k

kc′j k kc′l k √
We put kc j k = a and kci k = b, then we have a > δ−1 and b < δa. Since c′j ⊢ c j , for fixed a, the
kc′j k
number of cusps of c′j such that kc j k = a is ≪ 1, then we may treat it as 1. And since |c′l − cl | < 1
kci kkc j k , we
define
1
C(l, X) = {c′l : |c′l − cl | < and kc′l k ≤ Xkci k}
kci kkc j k
R 1
And put F(X) = |C(l, X)|, and G(X) = t≤X t
dF(t), from Proposition 4.4, we have
Z
1 Xkci k kci k
G(X) = kci k dF(t) ≤ kci k =X
t≤X tkci k kci kkc j k kc j k

Moreover, for fixed c′j and c′l , and for any other subindex m , i, the cusp c′m must be inside the 1
kci kkc j k and
kc′j kkc′l k
moreover c′m ∈ Q(kc′j kkc′l k, m), there are at most choices
but for the subindex i, we have that
kci kkc j k for c′m ,
kc′ kkc′ k
 ′ ′  
kc kkc k kc k
c′i is within the 1
kci k2
neighborhood of ci , and c′i ∈ Q(kc′j kkc′l k, i), there are at most kcj i k2l = kcijkkc lj k kcijk
choices for c′i . Therefore, suppose s ∈ (k − 21 , k), the above summation is equal to:
 s  s  kc′ kkc′ k k−2 
kci kkc j k kci kkc j k kc j k
P P j l

(c′1 ,...,c′k , j,l)∈σl (c1 ,...,ck ,i, j) kc′j kkc′l k = (c′j ,c′l ) kc′j kkc′l k kci kkc j k kci k
 kc j k  P R (76)
1 √ 1
= kci k a>δ−1 a s−k+2 b≤ δa b s−k+2
dF(b)

Using the same argument as in case k = 2, combining the fact G(b) ≤ b kckcijkk we have:
R 1
R

b≤ δa b s−k+2
dF(b) = √
b≤ √δa
bk−1−s dG(b)
√ R
= G( δa)( δa)k−1−s − b≤ √δa G(b)dbk−1−s
  √ √ k−1−s  
kci k kci k
R
≤ ( δa)( δa) + (s + 1 − k) √ bbk−2−s db (77)
 kc j k  √   √
b≤ δa kc j k
kci k
= kc j k ( δa)
k−s
+ kckcijkk s+1−k
k−s ( δa)
k−s
  √
1 kci k k−s
= k−s kc j k ( δa)

29
Plug the above into the summation (76), we have:
 s   √
kci kkc j k
 kc j k  P kci k
P 1 1 k−s
(c′1 ,...,c′k , j,l)∈σl (c1 ,...,ck ,i, j) kc′ kkc′ k ≤ kci k a>δ−1
a s−k+2 k−s kc j k ( δa)
j l
(k−s)/2
= δ k−s 2k−2s−2
P
a>δ−1 a (78)
δ(k−s)/2 δ2s+1−2k
≤ k−s 2s+1−2k
δ3/2(s−k)+1
= (k−s)(2s+1−2k)

δ3/2(s−k)+1 δ1/4
Then for δ small enough and for s = k − 1/2 + δ1/6 , then we have (k−s)(2s+1−2k) ≍ δ1/6
= δ1/12 ≪ 1. Let l
go over all subindex except j and take the summation of all
!s
X diamB(c′1, . . . , c′k , j, l)
(c′1 ,...,c′k , j,l)∈σl (c1 ,...,ck ,i, j)
diamB(c1, . . . , ck , i, j)

We have the following:


 s  s
P diamB(c′1 ,...,c′k , j,l) P P diamB(c′1 ,...,c′k , j,l)
(c′1 ,...,c′k , j,l)∈σ(c1 ,...,ck ,i, j) diamB(c1 ,...,ck ,i, j) = l, j (c′1 ,...,c′k , j,l)∈σl (c1 ,...,ck ,i, j) diamB(c1 ,...,ck ,i, j)


P
l, j 1
(79)
≍ 1

Thus we claim that there exists some δ0 depending on k and the space Γ1 \ H2 × · · · × Γk \ H2 such that
for all δ < δ0 , we have dimH Ek (δ) ≤ k − 1/2 + δ1/6 .
This complete the proof.
3
Since Dk ⊂ Ek (δ), from dimH E2 (δ) ≤ 2 + δ1/3 and dimH Ek (δ) ≤ k − 1/2 + δ1/6 . By letting δ → 0,
we have dimH Dk ≤ k − 1/2
Combing this with Proposition 4.3 we have the following:

Theorem 4.4. dimH Dk = k − 1/2

Combining all the theorems we get in this section with our discussion in Subsection 1.1, Theorem
1.1 is proved. 

References
[1] A. Beardon, The geometry of discrete groups, Springer-Verlag (1983).

[2] Y. Cheung, Hausdorff dimension of the set of points on divergent trajectories of a homogeneous
flow on a product space, Ergod. Th. Dynam. Sys. 27 (2007), 65-85.

[3] Y. Cheung, Hausdorff dimension of the set of nonergodic directions, Ann. of Math., 158 (2003),
661-178.

[4] Y. Cheung, Hausdorff dimension of the set of Singular Pairs. Annals of Math., 173, (2011), 127-
167.

30
[5] S.G. Dani, Divergent trajectories of flows on homogeneous spaces, J. Reine Angew Math. 359
(1985), 55-89.

[6] K. Falconer, Fractal Geometry. Mathematical Foundations and Applications, John Wiley & Sons
Ltd., Chichester, 1990.

[7] A. Gorodnik, N. Shah, Khinchins theorem for approximation by integral points on quadratic vari-
eties, Math. Annalen 350 (2011), no. 2, 357-380.

[8] C.C. Moore, Exponential decay of correlation coefficients for geodesic flows. Group representa-
tions, ergodic theory, operator algebras and mathematical physics. C.C. Moore, ed., MSRI Publ.
6, Springer-Verlag, New York, 1987.

[9] H. Masur, Hausdorff dimension of the set of nonergodic foliations of a quadratic differential. Duke
Math. Journal, 66 (1992), 387-442.

[10] H. Masur, J. Smillie, Hausdorff dimension of sets of nonergodic measured foliations. Ann. of
Math., 134 (1991), 455-543.

[11] B. Weiss, Divergent trajectories on noncompact parameter spaces. Geom. Funct. Anal. 14 (2004),
no. 1, 94-149.

31

You might also like