AN APPLICATION OF METRIC DIOPHANTINE APPROXIMATION IN HYPERBOLIC SPAC E TO QUADRATIC FORMS

For any real T, a lim sup set WGty (T) of T-(well)-approximabl e points is defined for discrete groups ~ acting on the Poincar é model of hyperbolic space . Here y is a ` distinguished point ' on the sphere at infinity whose orbit under G corresponds to the rationals (which can be regarded as the orbit of the point at infinity under the modular group) in the classical theory of diophantin e approximation . In this paper the Hausdorff dimension of the set WG , y (T} is determined for geometrically finite groups of the first kind . Consequently, by considering the hyperboloid model of hyperbolic space , this result is shown to have a natural but non trivial interpretatio n in terms of quadratic forms .


. Introductio n
The Euclidean norm of a vector x ~(x 1 , . . ., xk , xk + 1 ) in R k+1 wil l be denoted by 114, i .e .iixii = ixx + . . .Denote by M(B" -}-1 ) the group of orientation preserving Moebius transformations preserving the unit ball B" +1 .The group M(8" +1 ) preserves the aboye metric p .Let G be an arbitrary discrete subgroup of M(B k+1 ) and denote by L(G) the limit set of G, the set of cluster points in S' of any orbit of G in B" +1 .The limit set is either empt y or contains one, two or an infinite number of points .The group G is said to be elementary if the number of points in L(G) is at most two , and nonelementary otherwise .Also the group G is said to be of th e first kind if L(G) = S k and of the second kind otherwise .So all elementary groups are of the second kind .In this paper, G will be restricte d to nonelementary geometrically finite discrete subgroups of M(B 1 +1 ) .
By definition, there exists some convex fundamental polyhedron wit h finitely many faces which implies that G is finitely presentad (see [7] an d [13] for further details) .The reader is referred to [11, [2] and [7] for a general introduction to the theory of discrete groups and hyperboli c geometry .
The following definition of T-approximable is the natural extension of Defiñition 1 given in [14] .
Definition 1 .For each real number T, a paint x in the set L(G) is said to be T-approximable with respect to a non-empty finite subset A of L(G) if for some y in A the inequality ilx -g(y) II 5 L(0, g(0)) -T is satisfied for infinitely many g in G .Note that for groups of the first kind 5(G) = 1e ; so Theorem 2 just implies that for these groups equality holds in Theorem 1 .Sinc e dim(UyEA WG ;y (T)} = maxyEA {dim WG ;y (T ) } for any finite set A, th e following result is a direct consequence of Theorem 2 .
Corollary 1 .Let A be a non empty finite set of parabolic cusps if there are any, and a finite set of hyperbolic fixed points otherwise .Fo r T > 1, k where WG ;A (T) is the set of points x in Sk which are T-approxirnabl e with respect to A .
Remark .For groups with parabolic elements, Theorem 2 also appears in the work of M .V .Melián and D .Pestana [6] .They also give k '7 -the equivalent geometric result in terms of the `rate ' of excursions by geodesics into a cuspidal end of the associated hyperbolic manifold .
In Section 2, we provide the reader with the results required to prov e the aboye theorems .In view of these results, the proofs of the theorem s follow on using essentially the same arguments as in the Ftiichsian cas e (k = 1) [14] ; and will therefore be merely sketched .On t he whole, th e results of Section 2 are generalizations of Pattersons results for Fuchsian groups stated in Section 2 of [14], to geometrically finite groups with the restriction that all parabolic cusps (if there any) are of maximial rank .
The main purpose of this article is to give an interpretation of Theorem 2 in terms of quadratic forms .For this, suppose that c E R and is a bijection for positive c and p : passing through the origin is mapped by p onto a single point o n S k , different rays are mapped onto different points .However the map p : II n Q(0) ---~S k , where II is a k + 1 -dimensional hyperplane such that II n Q(0) is a subset of Q + (0), is a bij ect ion and moreover is easily seen to be bi-Lipschitz .
denote the quadratic form on R k+1 x R .The function q(z) of the single variable z determines the symmetric bilinear form q (z 1 , Z2) by th e formula ¡ 1 where b 1 , b 2 , . . ., bk + 2 is a basis for the lattice ( i .e .a set of k+ 2 linearly independent elements of R k+2), and u 1 , u2 , . . . ,uk + 2 run through Z .Let A be a lattice in R k+l x R on which the quadratic form q takes integral values, i .e .q : A ----~Z .
The group O (k + 1, 1) acts on each Q(c) and leaves the quadratic for m q(z) and the bilinear form q(z1, z2) invariant .The action is transitive if c 0 and is also transitive on Q(0)\{O} .Let lk+ 1 denote the k + 1dimensional identity matrix .Then in matrix notation A E O(k + 1,1) i f A T JA = J where 1 0 . .0 0 -4+ 1 0 whence (det A) 2 = 1 .For further details see [1, page 371 and [2, Sec- tion 3 .71 .Far c positive, Patterson [10] has shown that M( B k +1 ) , the group of orientation preserving Moebius transformations preserving the unit ball 8 k+1 , can be identified with the subgroup G + (k Let F be the maximal subgroup of 0+(k + 1,1) which preserves the lattice A introduced aboye .By definition r preserves each Q(c) .It i s known that I' acts discontinuously on Q(1) and that the quotient Q(1)/ F has finite volume .Also, if q takes on the value 0 on A \{0}, so that (Q(0)\{o})nA O, the quotient Q(1)/r is not compact .Let G = pI'p -1 , then G is a discrete subgroup of M(Bk+ 1 ) with finite covolume .Since G is of finite covolume it is certainly of the first kind, and by a theorem o f w.P. Thurston ([11, Prop .8 .4.3D is also geometrically finite .Hence G is a geometrically finite group of the first kind .Furthermore, if Q(1)/ F is not compact then the quotient B ' +1 / G is not compact and so G contains parabolic elements [131 .In this case, to each parabolic cusp o f G there corresponds a ray on Q(0) such that its intersection with 11\0 } is nonempty ; and any point of (Q(0)\{0}) n A( O) corresponds to some parabolic cusp of G .
In Section 3, we prove the following interpretation of Theorem 2 i n terms of the quadratic form q and the lattice A introduced aboye .
Theorem 3 .Let q, A be as above and let a be a positive real number .In arder to simplify notation, write L g for L (o , g(0)) .It is a well known fact that for any positive real number N

Suppose that
where the implied positive multiplicative constant is dependent only o n the group G ([7], [9]) .A direct consequence of inequality (1) is that fo r t < 6(G ) (2) The reader is referred to [14] for the proof of (2) .In fact inequalitie s (1) and (2) are valid for arbitrary discrete subgroups of M(B'+1 ) .Given these estimates, the proof of Theorem 1 now follows on using the same arguments as those in Section 4 of [14] .
The proof of Theorem 2 relies on the concept of a `ubiquitous system ' ([4, Section 5] in [14] } , and the following results are essential in setting up such a system .Theorem 4 .Suppose that G has no parabolic elements .Let rl, n' be the set of fixed points of a hyperbolic subgroup of G .Then there is a positive constant c with the following property : for each x in L(G) , N > 1, there exist y in {i, ri' }, g in G so tha t <N and ii x-9(Y)II < Ñ .
Theorem 5 .Suppose G has parabolic elements only of maximal rank k and let P be a complete set of representatives of the cusps of G .Then there is a positive constant c with the following property: for each x in L(G), N > 1, there exist y in P, g in G so tha t Lg < N and ll x -g(y) II < c .\ /L 9 N These theorems are a generalization of Dirichlet's theorem in the theory of global diophantine approximation and were proved by Patterso n in [8] for Fuchsian groups .It should be noted 'that for Fuchsian groups (1) the concepts of geometrically finite and finitely presentad coincide an d that all parabolic cusps are of maximal rank .The proofs of the aboye theorems make use of the fact that G has a convex fundamental polyhedron with finitely many faces and that in the case of Theorem 5, ther e exists an orbit point of the origin within a bounded distance (dependent only on G and P) of the summit of a horoball belonging to a complete sety of horoballs [12] for G .This latter fact is guaranteed by considering only groups with cusps of maximal rank .However, it should be possible to remove this restriction .In any case, for the purpose of this pape r Theorem 5 is required in the proof of Theorem 2 in which G is of the first kind, and so all cusps are automatically of maximal rank .
For a subgroup H of G, let G i i H be a set of representatives of the cosets {gH : g E G} so chosen that if g E G i IH, h E H then Lg Ç L gh .So in this notation the minimum of L gh (h E H) occurs when gh E Gil H .
The foliowing geometrical results are extensions of Propositions 8 .1 an d 8 .2 in [81, and can be proved using the same arguments as in the Fuchsian case [8] .
Proposition 1 .Suppose G has parabolic elements only of maxima l rank k and let P be a complete set of representatives of the paraboli c cusps of G .Then there are constants c1 , c2 , c3 , depending only on G and P with the following property : if p, q E P, g E Gil Gq there is a n element h in Gil G p so tha t I -9( q )II < cl and c2 Ly L h c 3 L y .9 Proposition 2 .Suppose that has G no parabolic elements .Let rl, rl ' be the set of fixed points of a hyperbolic subgroup of G .Then there are constants c4, c5 , c6, depending only on G and {r], rl'} with the following property: if g E Gil G,in , there is an element h in Gil G,m, so tha t

Ilgeri') -h ( 97)iI < L and cS Ly < L h < c6 Ly . 9
Providing the group G has parabolic elements, it should be note d that the hypotheses of Theorem 5 and Proposition 1 are satisfied for all non-elementary geometrically finite groups of the first kind and fo r all non-elementary finitely generated Fuchsian groups .The two propositions play an important role in `decoupling' the parabolic cusps or th e hyperbolic fixed points in order to obtain results for a single paraboli c cusp or a single hyperbolic fixed point .
Let G be a group of the first kind and y be an arbitrary paraboli c cusp of G if there are any, and a hyperbolic fixed point otherwise .By S .L .vELAN I Theorem 1, dim WG ;y (T) Ç k / T for T > 1 ; so in order to prove Theorem 2 it is sufficient to show that dim WG ;y ( T ) > k/T(T > 1) .This lower boun d inequality is obtained by using the concept of ubiquity and the reader is referred to [4] and/or [14] for the definition of ubiquity .
Let SZ be an open subset of Sk and let R. = {g(y) E : g E G} be a set of points in 12 .Far a positive constant c 7 , let : G : )3 c 7 be a positive function on G and le t : Hl + -> IR + :x(c 7 x) -T , be a positive decreasing function with ~~x} -> 0 as x --+ oo .On combin- ing the results stated in this section with the arguments used to prove Theorem 2 in [141 ; it can be verified that the system (R., ~3} is ubiquitous relative to the function where c 8 is a positive constant .Basically, this implies that for each positive integer N there exists a Lebesgue measurable subset A(N) o f such that (i) lim n ~~19\A(n) ~ = O, where 1X 1 denotes the k--dimensiona l Lebesgue measure of a measurable set X ; and (u) for any x in A(N) , there exists a g in G with L g Ç c7 N such that the inequality II x-g (y ii < A(N) is satisfied .The function a associated with the system (7?., ~) above, arises naturally from Theorem 5 and Proposition 1 in the cas e when G has parabolic elements and from Theorem 4 and Proposition 2 otherwise .The set A(N) is an approximating set for SZ in the measur e theoretical sense and is obtained by `thickening' each point g(y) in 11 with L g Ç c7 N by a A(N)-neighbourhood .
It follows from Theorem 6 in [14]

. Proof of Theorem 3
Let 1' be the maximal subgroup of G+ (k + 1, 1) preserving the lattic e A and consider the action of F on Q(1) .Then G = pI'p -1 is a geometrically finite group of the first kind, and since Q + (o) n A 0 the grou p G contains parabolic elements .Let A be a complete set of inequivalent parabolic cusps .Since G is geometrically finite, the set A is finite .Consider the set WG ;A (r) = UyEA WG ;y (r) wher e WG;y (r) = {x E S k : I I x-()II ç L9 T far infinitely many g in G} .
For any 'y in F let g = p-yp -1 be the corresponding element in the group G .Patterson in [10] shows that there exist positive constants cg, cm, so that if z E Q + (o), then for any -y in 1' there exists fy * in -y I' z such that II-y(z) 11 < c 9 L g and lly*(z)ll ~ c i oL 9* 11 z 11 .
Here I' z = {¡y E F : -y(z) = z} is the stabilizer of z, and so -y(z) = 'y* (z) ; and g* E g Gp(Z) where Gp(Z) = {g E G : g(p(z)) = p(z)} is the stabilizer of the point p(z) in S k .This shows that Lg can be taken to be comparable with ll'y(z)M for the optimal choice of g, Le .for g in GI I Gp(Z ) in terms of the notation introduced in Section 2 .
In the set WG ;y (r) replace the g(y) by p(z) for some z in Q+ (o) nA and take L g to be comparable with lizM .On writing g(y) = p(z), x = p(( ) where ( E Q (o), it is easily verified that Remark .In the case where Q(0) n A = {0}, elements of Q(0) ar e approximated by a pair of hyperbolic fixed points .An analogue to Theorem 10 in [10] should be possible for this case . 1 hope to pursue this problem in the near future .
I would like to thank Maurice Dodson for introducing me to the theory of metric diophantine approximation, and for his continual help an d encouragement .I am also most grateful to the Mathematisches Institut, Góttingen and S .J .
+ x~+ xl +1 } 1 /2 .The k + 1dimensional unit ball B k+l = {x E : llxll < 1 } is a model of k + 1 -dimensional hyperbolic space and supports a metri c p derived from the differentia l dP= dx ~1 -lixII 2 S .L .VELAN I The unit ball model is usually referred to as the Poincaré model .Geodesics far the metric p are ares of circles orthogonal to the uni t sphere S I and straight lines through the origin .
Let , WG ;y (T) = {x E L(G) : N II Ç L(0, g(0)) --T for infinitely many g in G } denote the set of T-approximable points with respect to y in L(G), an d denote by WG; A (T ) the set of points x in L(G) which are T -approximabl e with respect to A .Thus WG ; A ( T ) = U WG ;Y( T ) • yE A With reference to the aboye definition let y be an arbitrary parabolic cusp of G is there any, and a hyperbolic fixed point otherwise .Let dim F denote the Hausdorff dimension of a subset F of H1 k +1 .For the definition of Hausdorff dimension and for further details the reader is referred to [5] .The following results are a generalization of the metric results give n in [14] .Theorem 1 .For T > 1 , dimWG;y (T) < S(G) Theorem 2 .If the group G is of the first kind, then for T > 1 dim WG ;y (T ) Hence for 7 > 1, the correct lower bound inequality follows .
( E IInQ(0) : 4((, z )i Ç 11(11 ifor i .m .z in Q + (0)f1A} , where H is a k + 1-dimensional hyperplane of the form described earlier .In view of (3) and the discussion aboye, the map p = Wn ~ WG ;A (r) is bi-Lipschitz .Since Hausdorff dimension is invariant under bi-Lipschit z transformations, the set Wn also has Hausdorff dimension k/r for 'r > 1 .Consider a ray on Q(0) passing through the origin and containing a point of the set Wn .Excluding the origin, each point on the ray belongs to the set W( 2 r -1), since q(r(, z) = rq((,z) for r E R\{0} ; and each point in W(2 7--1) Ties on some ray passing through the origin an d containing a point in Wn .Henc e W(2 7 -1) _ {r( : ( E Wn, r E R\{0}} , from which it follows that for T > 1 dimW(2T-1)=dimWn+l= ~+1 .T Putting cx = 27 -1 the assertion of Theorem 3 follows immediately .