A survey of some arithmetic applications of ergodic theory in negative curvature

This paper is a survey of some arithmetic applications of techniques in the geometry and ergodic theory of negatively curved Riemannian manifolds, focusing on the joint works of the authors. We describe Diophantine approximation results of real numbers by quadratic irrational ones, and we discuss various results on the equidistribution in $\mathbb R$, $\mathbb C$ and in the Heisenberg groups of arithmetically defined points. We explain how these results are consequences of equidistribution and counting properties of common perpendiculars between locally convex subsets in negatively curved orbifolds, proven using dynamical and ergodic properties of their geodesic flows. This exposition is based on lectures at the conference"Chaire Jean Morlet: G\'eom\'etrie et syst\`emes dynamiques", at the CIRM, Luminy, 2014. We thank B. Hasselblatt for his strong encouragements to write this survey.

In Subsection 2.2 of this survey, we introduce a general framework of Diophantine approximation in measured metric spaces, in which most of our arithmetic corollaries are inserted (see the end of Subsection 2.2 for references concerning this framework).In order to motivate it, we first recall in Subsection 2.1 some very basic and classical results in Diophantine approximation (see for instance [Bug1,Bug2]).A selection (extracted from [PaP1,PaP4,PaP7]) of our arithmetic results are then stated in Subsections 2.3, 2.4 and 2.5, where we indicate how they fit into this framework: Diophantine approximation results (à la Khintchine, Hurwitz, Cusick-Flahive and Farey) of real numbers by quadratic irrational ones, equidistribution of rational points in R (for various height functions), in C and in the Heisenberg group, ... We will explain in Subsection 4.2 the starting point of their proofs, using the geometric and ergodic tools and results previously described in Subsection 4.1, where we give an exposition of our work in [PaP6]: an asymptotic formula as t → +∞ for the number of common perpendiculars of length at most t between closed locally convex subsets D − and D + in a negatively curved Riemannian orbifold, and an equidistribution result of the initial and terminal tangent vectors v − α and v + α of the common perpendiculars α in the outer and inner unit normal bundles of D − and D + , respectively.Common perpendiculars have been studied, in various particular cases, sometimes not explicitly, by Basmajian, Bridgeman, Bridgeman-Kahn, Eskin-McMullen, Herrmann, Huber, Kontorovich-Oh, Margulis, Martin-McKee-Wambach, Meyerhoff, Mirzakhani, Oh-Shah, Pollicott, Roblin, Shah, the authors and many others (see the comments after Theorem 15 below, and the survey [PaP5] for references).
Section 3 presents the background notions on the geometry in negative curvature, describes various useful measures, and recalls the basic results about them, due to works of Patterson, Sullivan, Bowen, Margulis, Babillot, Roblin, Otal-Peigné, Kleinbock-Margulis, Clozel, Oh-Shah, Mohammadi-Oh and the authors (see for instance [Rob2,Bab,OP,MO,PaP3]).See [PaPS, BrPP] for extensions to manifolds with potentials and to trees with potentials.
Let us denote by c A the complementary subset of a subset A of any given set, by µ the total mass of a measure µ, by Leb R and Leb C the Lebesgue measures on R and C, by ∆ x the unit Dirac mass at any point x in any topological space, and by * the weak-star convergence of measures on any locally compact space.

Basic and classic Diophantine approximation
When denoting a rational number p q ∈ Q, we will assume that p and q are coprime and that q > 0. For every irrational real number x ∈ R − Q, let us define the approximation exponent ω(x) of x as ω(x) = lim sup p q ∈Q, q→+∞ − ln x − p q ln q .
The Dirichlet theorem implies that which motivates the definition of the the approximation constant c(x) of x ∈ R − Q as c(x) = lim inf p q ∈Q, q→+∞ q 2 x − p q .
The convention varies, some other references consider c(x) −1 or (2c(x)) −1 as the approximation constant.The Lagrange spectrum for the approximation of real numbers by rational ones is Sp Lag = {c(x) : x ∈ R − Q} .
Given ψ : N → ]0, +∞[, the set of ψ-well approximable real numbers by rational ones is Again the convention varies, some other references consider q → ψ(q)/q or similar instead of ψ.
The Khintchine theorem gives a necessary and sufficient criteria for the ψ-well approximable real number to have full or zero Lebesgue measure: Finally, the equidistribution of Farey fractions (which is closely related with the Mertens formula) is an equidistribution theorem of the rational numbers in R when their denominator tends to +∞: π 2 12 s p q ∈Q, |q|≤s Leb R .

An approximation framework
The general framework announced in the introduction concerns the quantitative answers to how dense a given dense subset of a given topological set is.Let (Y, d, µ) be a metric measured space (see for instance [Gro,Chap. 3 1 2 ] and [Hei] for generalities), with Y a subspace of a topological space X, let Z be a countable (to simplify the setting in this survey) subset of X whose closure contains Y (for instance a dense orbit of a countable group of homeomorphisms of X), and let H : Z → ]0, +∞[ be a map called a height function which is proper (for every r > 0, the set H −1 (]0, r]) is finite).The classical Diophantine approximation problems of subsection 2.1 fit into this framework with X = R, Y = R − Q, Z = Q and H : p q → q or H : p q → q 2 , considered modulo translation by integers.The height functions in question are invariant under translation by Z, and the properness condition is satisfied in the quotient space Z\R or, equivalently, by restriction to the unit interval.
We endow Z with the Fréchet filter of the complementary subsets of its finite subsets: z ∈ Z tends to infinity if and only if z leaves every finite subset of Z. Generalising the definitions of Subsection 2.1, for every y ∈ Y , we may define the approximation exponent The approximation problems may be subdivided into several classes, as follows, possibly by taking the appropriate height function, for appropriate maps ψ : ]0, +∞[ → ]0, +∞[.• The approximation exponent problem: study the map y → ω(y).
• The Lagrange problem: study the Lagrange spectrum for the approximation of the elements of Y by the elements of Z with height function H. • The Jarnik-Besicovich problem: compute the Hausdorff dimension of the set of y ∈ Y with c(y) ≥ ψ(H(y)).
• The counting problem: study the asymptotics as s tends to +∞ of Card{y ∈ Y : H(y) ≤ s} .
• The equidistribution problem: study the set of weak-star accumulation points as s tends to +∞ of the probability measures The equidistribution problem, closely linked to the counting problem, is the one we will concentrate on in Subsections 2.3, 2.4 and 2.5.This framework is not new (see for instance the works of Kleinbock), and many results have developped some aspect of it.
(1) For instance, X could be the boundary at infinity of a Gromov hyperbolic metric space, Y a (subset of) the limit set of a discrete group Γ of isometries of this hyperbolic space, Z could be the orbit under Γ of some point x ∈ X, and H : Z → ]0, +∞[ could be γx → 1 + d X (A 0 , γB 0 ) where A 0 , B 0 are subsets of X, with B 0 invariant under the stabilizer of x in Γ. Numerous aspects of this particular case have been developped, by Patterson, Sullivan, Dani, Hill, Stratmann, Velani, Bishop-Jones, Hersonsky-Paulin, Parkkonen-Paulin, and the most complete and general version is due to Fishman-Simmons-Urbański [FSU], to which we refer in particular for their thorough long list of references.
(2) The case when X = R N , Y is a submanifold (or a more general subset) of X, Z = Q N has been widely studied, under the name of Diophantine approximation on curves, submanifolds and fractals subsets, by many authors, including Kleinbock-Margulis, Bernik, Dodson, Beresnevich, Velani, Kleinbock-Weiss and others, see for instance [BerD, BaBV] and their references.
(3) If X = X(R) is for instance the set of real points of an algebraic manifold defined over Q, if Z = X(Q) is the set of rational points, if Y is the (Hausdorff) closure of Z in X or this closure minus Z, there are many results, in particular when X is homogeneous, on the above Diophantine approximation problems.Similarly, if X = G/H is a homogeneous space of a semisimple connected Lie group G, if Z is the orbit in X of a lattice in G and if Y is the closure of Z in X, most of the above problems have been stated and studied for instance in [GGN3,GGN2,GGN1].We refer to the works of Benoist, Browning, Colliot-Thélène, Duke, Einsiedler, Eskin, Gorodnik, Heath-Brown, Lindenstrauss, Margulis, Mozes, Oh, Ratner, Quint, Rudnick, Salberger, Sarnak, Shah, Tomanov, Ullmo, Venkatesh, Weiss and many others for equidistribution results in homogeneous spaces, see for instance [Bre,BF,Har,EiW,Serr,GreT,Kim,BenQ1,BenQ2].
In the next three subsections, we give a sample of the results obtained by the authors on the above problems.

Diophantine approximation in R by quadratic irrationals
Our first results concern the Diophantine approximation of real numbers, where we replace the approximating rationals by quadratic irrational numbers.We refer for instance to [Bug1] and its references for very different Diophantine approximation results by algebraic numbers.
We denote by α σ the Galois conjugate of a real quadratic irrational α, and by tr α = α + α σ its trace.We approximate the real points by the elements of the orbit of a fixed quadratic irrational α 0 by homographies under PSL 2 (Z) (and their Galois conjugates).We denote by γ • x the action by homography of γ ∈ PSL 2 (R) on x ∈ P 1 (R) = R ∪ {∞}.
Let X = R (with the standard Euclidean distance and Lebesgue measure), let H is an appropriate height function on each PSL 2 (Z)-orbit under homographies (working modulo translation by Z) of a given quadratic irrational.We refer to [PaP1,PaP2] for a proof of this, and for more algebraic expressions of this height function and its important differences with classical ones.
We consider in the first statement below the particular case when α 0 is the Golden Ratio φ = 1+ √ 5 2 , and we refer to [PaP1] for the general version.One way this restriction simplifies the statement is that the Golden Ratio is in the same orbit under PSL 2 (Z) as its Galois conjugate, which is not the case of every quadratic irrational (see for instance [Sar]).The next result is proven in [PaP1,Theo. 1.3,Prop. 1.4], with a mistake in the Hurwitz constant corrected in the erratum of loc.cit., thanks to Bugeaud, who gave another way to compute it in [Bug3], using only continued fractions techniques.
Theorem 1 (Parkkonen-Paulin) With the notation X, Y, Z, H as above, the Lagrange spectrum Sp Lag is closed, bounded, with Hurwitz constant 3 √ 5 − 1.For every ψ : ]0, +∞[ → ]0, +∞[ such that t → ln ψ(e t ) is Lipschitz, we have The exact value of the Hurwitz constant for the Diophantine approximation of real numbers by elements of an orbit under PSL 2 (Z) (or a congruence subgroup) of a general quadratic irrational (and its Galois conjugate) is an interesting open problem.
Let α 0 , β 0 be fixed integral quadratic irrationals, and let R α 0 , R β 0 be the regulators of the lattices Z + Zα 0 , Z + Zβ 0 respectively.The integrality assumption is only present here in order to simplify the statements below in this survey, see [PaP4] for the general version.
The following result is an equidistribution result of the traces of the quadratic irrationals in a given orbit (by homographies) under PSL 2 (Z) of a quadratic irrational, using the above height function H : α → 2 |α−α σ | .We refer to [PaP4,Theo. 4.1] for a version with additional congruence assumptions, and to [PaP4,Theo. 4.2] and [PaP4,Theo. 4.4] for extensions to quadratic irrationals over an imaginary quadratic number field (using relative traces) or over a rational quaternion algebra.
Theorem 2 (Parkkonen-Paulin) As s → +∞, we have We introduced in [PaP4] another height function for the Diophantine approximation of real numbers by the elements of the orbit (by homographies) under PSL 2 (Z) of β 0 , which measures their relative complexity with respect to α 0 .Let be the standard crossratio of a quadruple (a, b, c, d) of pairwise distinct points in R. For every β ∈ PSL 2 (Z) • β 0 − {α 0 , α σ 0 }, let We prove in [PaP4] that the map H α 0 is an appropriate height function modulo the action (by homographies) of the fixator PSL 2 (Z) α 0 of α 0 in PSL 2 (Z).The following theorem is a counting result of quadratic irrationals relative to a given one.
Theorem 3 (Parkkonen-Paulin) There exists κ > 0 such that, as s → +∞, We refer to [PaP4,Theo. 4.9] for a more general version, including additional congruence assumptions, and to [PaP4,Theo. 4.10] for an extension to quadratic irrationals over an imaginary quadratic extension of Q. Theorem 3 fits into the framework of Subsection 2.2 with

Equidistribution of rational points in R and C
In this section, we will consider equidistribution results in R and C of arithmetically defined points.
First, we would like to consider again the approximation of points in R by points in Q, but to change the height function H, using an indefinite rational binary quadratic form Q which is not the product of two rational linear forms, by taking It is easy to see that this is (locally) an appropriate height function outside the roots α, α σ of t → Q(t, 1): for every s ≥ 0, the number of p q ∈ Q such that H p q ≤ s is locally finite in R − {α, α σ }.This fits in the framework of Subsection 2.2 as follows.Let which acts by homographies on P 1 (R), fixing α and α σ , acting properly discontinuously on P 1 (R) − {α, α σ }.We therefore study the Diophantine approximation problems with We only consider in this survey the particular case when related to the Golden Ratio, as the knowledgeable reader has already seen!).We refer to [PaP4,Theo. 5.10] for the general result, with error term estimates and additional congruence assumptions, and to [GorP] for an extension to n-ary norm forms for general n > 2.
Theorem 4 (Parkkonen-Paulin) As s → +∞, we have Note that the measure to which the rational points equidistribute when counted with this multiplicity is no longer the Lebesgue measure, but is the natural smooth measure invariant under the real group of automorphs of Q (unique up to multiplication by a locally constant positive function).Now, let us turn to the Diophantine approximation of complex numbers by Gaussian rational ones.Every element of the imaginary quadratic field K = Q(i) of Gaussian rational numbers may and will be written p q with p, q ∈ Z[i] relatively prime Gaussian integers, and this writing is unique up to the multiplication of p and q by the same invertible Gaussian integer 1, −1, i or −i.In particular, the map is well defined, and is clearly an appropriate height function on We hence consider, with the notation of Subsection 2.2, the spaces The following result (due to [Cos,Coro. 6.1] albeit in a less explicit form) is an equidistribution result of the Gaussian rational points in the complex field, analogous to the Mertens theorem on the equidistribution of Farey fractions in the real field.We denote, here and in Subsection 2.5, by and by   ζ its Dedekind zeta function.The pictures below show the fractions where p, q ∈ Z[i] are relatively prime Gaussian integers with |q| ≤ 5 and |q| ≤ 10.The fact that there is a large white region around the fractions p q ∈ Q[i] with |q| small will be explained in Subsection 4.2.
Theorem 5 (Cosentino, Parkkonen-Paulin) As s → +∞, We refer to [PaP4,Theo. 1.1] for a version of this theorem valid for any imaginary quadratic number field (see below a plot in the square 3 ] are relatively prime Eisenstein integers with |q| ≤ 5 and |q| ≤ 10), with additional congruence assumptions on u and v, to [PaP4,Theo. 1.2] for analogous results in Hamilton's quaternion division algebra, and to [Cos,PaP4] for error term estimates.

Equidistribution and counting in the Heisenberg group
In this section, we will consider Diophantine approximation results of elements of the Heisenberg group by arithmetically defined points.
Recall that the 3-dimensional Heisenberg group is the nilpotent real Lie group with underlying manifold and law (w 0 , w)(w 0 , w ) = (w 0 + w 0 + w w, w + w ).A standard model in control theory of Heis 3 , with underlying manifold R 3 , may be obtained by the change of variables (x, y, t) ∈ R 3 with w = x + iy and t = 2 Im w 0 .We endow Heis 3 with its Haar measure (that is 1 2 dx dy dt in the above coordinates (x, y, t) ) and with the (almost) distance d Cyg defined below.
The Cygan distance d Cyg on Heis 3 (see [Gol,page 160], sometimes called the Korányi distance in sub-Riemannian geometry, though Korányi [Kor] attributes it to Cygan [Cyg]) is the unique left-invariant distance on Heis 3 such that the Cygan distance from (w 0 , w) ∈ Heis 3 to (0, 0) is the Cygan norm |(w 0 , w)| Cyg = 2|w 0 |.Note that with the aforementioned change of variables (x = Re w, y = Im w, t = 2 Im w 0 ), we do recover the standard formulation of the Cygan norm on R 3 (which is, by the way, equivalent to the Guivarc'h norm introduced much earlier in [Gui]): We will denote by B Cyg (x, r) the ball for the Cygan distance of center x and radius r.
The modified Cygan distance d Cyg on Heis 3 is a minor variation of the Cygan distance, introduced in [PaP1, §4.4]: it is the unique left-invariant map d Cyg on Heis 3 × Heis 3 such that Though d Cyg might not be a distance, it is hence close to the Cygan distance, and it will allow error terms estimates in the following results.
We consider again the imaginary quadratic number field K = Q(i).The Heisenberg group Heis 3 is the set of real points of an algebraic group defined over Q, with group of rational points equal to Heis 3 (Q) = Heis 3 ∩(K × K).It has also a natural Z-structure, with the group of integral points equal to Heis The following beautiful result (see [GaNT,Theo. 1.1] for a more general version), whose tools are in harmonic analysis, solves the analog in the Heisenberg group of the Gauss circle problem.It would be interesting to know if it is still valid for general imaginary quadratic number field K.Note that 1 Haar Heis 3 (Heis (see Equation ( 24) in [PaP7], the volume Vol Heis 3 used in this reference being Vol Heis 3 = 8 Haar Heis 3 ), and that Haar Heis 3 (B Cyg (0, R)) = Haar Heis 3 (B Cyg (0, 1)) R 4 .
Theorem 6 (Garg-Nevo-Taylor) As R → +∞, we have Any element in Heis 3 (Q) may and will be written ( a c , b c ), where a, b, c ∈ O K are relatively prime (that is, the ideal of O K generated by a, b, c is equal to O K ), and satisfy c = 0 and 2 Re(ac) = |b| 2 .This writing is unique up to the multiplication of a, b, c by the same element of O × K .In particular, the map H : Heis is well defined, and is clearly an appropriate height function on Heis 3 (Q) modulo the action of Heis 3 (Z).
The following result is an equidistribution theorem of the set of rational points in Heis 3 , analogous to the equidistribution of Farey fractions in the real field.We denote by ζ Riemann's zeta function.Note that the exponent 4 that appears below is the same as in the above theorem of Garg-Nevo-Taylor, it is the Hausdorff dimension of the Cygan distance.
Theorem 7 (Parkkonen-Paulin) As s → +∞, we have We refer to [PaP7,Theo. 13] for a version of this theorem valid for any imaginary quadratic number field, with additional congruence assumptions, and with error term.The next result, analogous to the Mertens theorem in the real field, follows from the version with error term of Theorem 7, by Equation (1).
Corollary 8 (Parkkonen-Paulin) There exists κ > 0 such that, as s → +∞, We now turn to equidistribution and counting results in the Heisenberg group of arithmetically defined topological circles, relating them to Diophantine approximation problems.
Let us consider the Hermitian form of signature (1, 2) on C3 defined by Using homogeneous coordinates in the complex projective plane P 2 (C), Poincaré's hypersphere2 is the projective isotropic locus of h which is a real-analytic submanifold of P 2 (C) diffeomorphic to the 3-sphere S 3 .The projective action on P 2 (C) of the projective special unitary group PSU h of h preserves H S .The Alexandrov compactification Heis 3 ∪{∞} of the Heisenberg group Heis 3 identifies with Poincaré's hypersphere by mapping (w 0 , w) to [w 0 : w : 1] and ∞ to [1 : 0 : 0].We identify Heis 3 with its image in P 2 (C), called Segre's hyperconic, that we will think of as the projective model of the Heisenberg group.
As defined by von Staudt, 3 a chain in Poincaré's hypersphere H S is an intersection, nonempty and not reduced to a point, with H S of a complex projective line in P 2 (C).It is called finite if it does not contain ∞ = [1 : 0 : 0].A chain C separates the complex projective line containing it into two real discs D ± (C), which we endow with their unique Poincaré metric (of constant curvature −1) invariant under the stabiliser of C in PSU h .
If π : Heis 3 → C is the canonical Lie group morphism (w 0 , w) → w, then the chains are the images, under the elements of PSU h , of the vertical chains, that are the union with {∞} of the fibers of π.In particular, the finite chains are ellipses in (the aforementioned coordinates (x, y, t) of) Heis 3 whose images under π are Euclidean circles in C. We refer for instance to [Gol,§4.3] for these informations and more on the chains.
A chain C will be called arithmetic (over where n 0 is the order of the pointwise stabiliser of C 0 in PSU h (O K ).
We refer to [PaP7,Theo. 19] for a version of this theorem valid for any imaginary quadratic number field (the pictures below represent two views of a part of an orbit of an arithmetic chain when K = Q(i √ 2)) and with additional congruence assumptions.
Given a complex projective line L in P 2 (C), there is a unique order 2 complex projective map with fixed point set L, called the reflexion in L. Given a finite chain C, contained in the complex projective line L(C), the center of C (see for instance [Gol,4.3.3]),denoted by cen(C) ∈ H S − {∞} = Heis 3 , is the image of ∞ = [1 : 0 : 0] under the reflexion in L(C).We also prove in [PaP7,Theo. 20] (a version valid for any imaginary quadratic number field and allowing additional congruence assumptions of) the following result saying that the centers of the finite arithmetic chains in a given PSU h (O K )-orbit of a given arithmetic chain C 0 with d Cyg -diameter at least equidistribute in the Heisenberg group towards the normalised Haar measure 2 |D K | Haar Heis 3 .
The above result fits into the framework of Subsection 2.2 with This equidistribution phenomenon can be understood by looking at the following picture, representing a different view of the same orbit of arithmetic chains as the one above Theorem 9.
3 Measures in negative curvature

A classical link between basic Diophantine approximation and hyperbolic geometry
Let us briefly explain a well-known link between hyperbolic geometry and Diophantine approximation problems, which goes back to Gauss and Ford (see also [Seri]).This will start to explain why the proofs of Theorems 1, 2, 3, 4, 5, 7, 9 and 10 all use real or complex hyperbolic geometry.
For n ≥ 2, let H n R be the upper halfspace model of the real hyperbolic n-space, with underlying manifold {x = (x 1 , . . ., be the closed Euclidean disc of center p q + 1 2 ψ( p q )q 2 i and radius 1 2 ψ( p q )q 2 , with its tangency point to the horizontal axis removed.Then (H x ) x∈Q∪{∞} is a family of horodiscs in H 2 R , with pairwise disjoint interiors, centred at the parabolic fixed points of Γ.If ψ is constant, then this family is Γ-equivariant, and if ψ = 1, then it is maximal.
A link between Diophantine approximation of real numbers by rational ones and hyperbolic geometry is the following one: for every ξ ∈ R, we have if and only if the geodesic line . Many Diophantine approximation properties of ξ may be explained by the behaviour of the image of the geodesic L ξ in the modular curve Γ\H 2 R .For instance, the coefficients of the continued fraction expansion of ξ ∈ R − Q are bounded if and only if the positive subray of L ξ has a bounded image in Γ\H 2 R (see for instance [Seri]).Hence, it is useful for arithmetic applications to study the dynamical and ergodic properties of the geodesic flows in negative curvature, and we develop these topics in the following two sections.

Negative curvature background
We refer for instance to [BriH,Rob2,PaPS] for definitions, proofs and complements concerning this subsection.
Let M (for instance H 2 R ) be a complete simply connected (smooth) Riemannian manifold with (dimension at least 2 and) pinched negative sectional curvature −b 2 ≤ K ≤ −1, and let x 0 ∈ M be a fixed basepoint.Let Γ (for instance PSL 2 (Z)) be a nonelementary (not virtually nilpotent) discrete group of isometries of M , and let M be the quotient Riemannian orbifold Γ\ M .See [Rob2,BrPP] to relax the pinching and manifold assumptions on M .
We denote by ∂ ∞ M the boundary at infinity of M , that is the quotient space of the space of geodesic rays ρ : [0, +∞[ → M in M , two of them being equivalent if the Hausdorff distance between their images is finite.The class of a geodesic ray is called its point at infinity.
The Busemann cocycle of M is the map where ρ is any geodesic ray with point at infinity ξ.The above limit exists and is independent of ρ.The horosphere with center ξ ∈ ∂ ∞ M through x ∈ M is {y ∈ M : β ξ (x, y) = 0}, and {y ∈ M : β ξ (x, y) ≤ 0} is the horoball centered at ξ bounded by this horosphere.For instance, in H n R , the horoballs are either the subspaces {(x 1 , . . ., x n ) ∈ H n R : x n ≥ a} for a > 0 or the closed Euclidean balls contained in the closure of H n R , tangent to the horizontal hyperplane, minus the tangency point.
Let M ∪ ∂ ∞ M be the geometric compactification of M , and let ΛΓ = Γx 0 − Γx 0 be the limit set of Γ.Let π : T 1 M → M be the unit tangent bundle of M .

For every
respectively, be the endpoints at −∞ and +∞ of the geodesic line defined by v. Let ∂ 2 ∞ M be the subset of ∂ ∞ M × ∂ ∞ M which consists of pairs of distinct points at infinity of M .Hopf 's parametrisation of T 1 M is the homeomorphism which identifies T 1 M with ∂ 2 ∞ M ×R, by the map v → (v − , v + , s), where s is the signed distance of the closest point to x 0 on the geodesic line defined by v to π(v).
The geodesic flow is the smooth flow (φ t ) t∈R on T 1 M defined, in Hopf's coordinates, by φ and its strong unstable manifold is ) and π(W − (v)) in M of the strong stable and strong unstable manifolds of v are the horospheres through π(v) centered at v + and v − , respectively (see the picture below on the left hand side).These horospheres bound horoballs denoted by HB ± (v).The strong stable manifolds and strong unstable manifolds are the (smooth) leaves of topological foliations in T 1 M that are invariant under the geodesic flow and the group of isometries of M , denoted by W + and W − respectively.
For every v ∈ T 1 M , let d W − (v) and d W + (v) be Hamenstädt's distances on the strong unstable and strong stable leaf of v, defined as follows (see [HP3,Appendix] and compare with [Ham]): for all w, z ∈ W ± (v), let The above limit exists, and Hamenstädt's distance induces the original topology on W ± (v), though it has fractal properties: the Hausdorff dimension of d W + (v) is in general bigger than the topological dimension of W ± (v).For all w, z ∈ W ± (v) and t ∈ R, and for every isometry γ of M , we have These dilation/contraction properties of Hamenstädt's distances under the geodesic flow are a strengthening of the Anosov property of the geodesic flow (see the above picture on the right hand side).

The various measures
We refer for instance to [Rob2,PaP3,PaPS,BrPP] for definitions, proofs and complements concerning this subsection.We introduce here the various measures that will be useful for our ergodic study in Section 4 A Patterson-Sullivan measure is a family (µ x ) x∈ M of finite nonzero measures on ∂ ∞ M , whose support is the limit set ΛΓ of Γ, such that, for all γ ∈ Γ, x, y ∈ M and ξ ∈ ∂ ∞ M , Such a family exists, and if P Γ (δ Γ ) = +∞, then, for all x ∈ M , for the weak-star convergence of measures (see for instance [Rob2]).
Let C be a nonempty proper closed convex subset of M , with stabiliser Γ C in Γ, such that the family (γC) γ∈Γ/Γ C of subsets of M is locally finite.
The inner (respectively outer) unit normal bundle is orthogonal to a contact hyperplane to C and points towards (respectively away from) C (see [PaP3] for more precisions, and note that the boundary ∂C of C is not necessarily C 1 , hence may have more than one contact hyperplane at some point, and that it is not necessarily true that exp(tv) belongs (respectively does not belong) to C for t > 0 small enough).000000 000000 000000 111111 111111 111111 Using this homeomorphism, we defined in [PaP3] the outer (respectively inner) skinning measure on T 1 M as the measure σ This measure is independent of x 0 and satisfies γ * σ ± C = σ ± γC for every γ ∈ Γ. Hence the measure γ∈Γ/Γ C σ ± γC (w) is a Γ-invariant locally finite measure on T 1 M , therefore defining (see for instance [PaP6,§2.4]for details) a (locally finite) measure on T 1 M = Γ\T 1 M , called the outer (respectively inner) skinning measure of C on T 1 M , and denoted by σ + C (respectively σ − C ).Note that the measure σ + HB − (w) (respectively σ − HB + (w) ) coincides with the Margulis measure (see for instance [Mar3,Rob2]) on the strong unstable leaf W − (w) (respectively strong stable leaf W + (w) ), for every w ∈ T M .
When M has constant curvature and Γ is geometrically finite, when C is an ball, horoball or totally geodesic submanifold, the (outer) skinning measure of C has been introduced by Oh and Shah [OhS3,OhS2], who coined the term, with beautiful applications to circle packings, see also [HP2,Lemma 4.3] for a closely related measure.The terminology comes from McMullen's proof of the contraction of the skinning map (capturing boundary information for surface subgroups of 3-manifold groups) introduced by Thurston to prove his hyperbolisation theorem.
The Bowen-Margulis measure on T 1 M (associated to a given Patterson-Sullivan measure) is the measure m BM on T 1 M given by the density in Hopf's parametrisation.The Bowen-Margulis measure m BM is independent of x 0 , and it is invariant under the actions of the group Γ and of the geodesic flow.Thus (see for instance [PaP6,§2.4]),it defines a measure m BM on T 1 M which is invariant under the quotient geodesic flow, called the Bowen-Margulis measure on T 1 M .We refer for instance to [Led, PaPS] for the extensions to the case with potential of the Patterson-Sullivan and Bowen-Margulis measures (the latter becomes the Gibbs measure), and to [PaP6] for the extensions to the case with potential of the skinning measures.
Let C be a nonempty proper closed convex subset of M .Let Note that the measure dµ W 0± (w) depends on W ± (w), not only on W 0± (w).
The following result (see [PaP3,Prop. 8]) says that the Bowen-Margulis measure disintegrates over the skinning measure of C. When M has constant curvature and Γ is torsionfree and for special convex sets C, this result is implicit in [OhS2].
Proposition 11 (Parkkonen-Paulin) The restriction to U ± C of the Bowen-Margulis measure m BM disintegrates by the fibration f ± C : We summarize in the following statement the ergodic properties of the Bowen-Margulis measure that we will use in Section 4.
Theorem 12 Assume that m BM is finite.
(1) (Patterson, Sullivan, Roblin) We have P Γ (δ Γ ) = +∞ and the Patterson-Sullivan measure is unique up to a multiplicative constant; hence the Bowen-Margulis measure m BM is uniquely defined, up to a multiplicative constant.
(2) (Bowen, Margulis, Otal-Peigné) When normalised to be a probability measure, the Bowen-Margulis measure on T 1 M is the unique measure of maximal entropy of the geodesic flow.(3) (Babillot) If the set of lengths of closed geodecics in M generates a dense subgroup of R, then m BM is mixing under the geodesic flow.
(4) (Kleinbock-Margulis, Clozel) If M is a symmetric space and Γ an arithmetic lattice, then there exist c, κ > 0 and ∈ N such that for all φ, ψ ∈ C c (T 1 M ) and all t ∈ R, we have Here are a few comments on these results.The Bowen-Margulis measure m BM is finite for instance when M is compact, or when Γ is geometrically finite and its critical exponent is strictly bigger than the critical exponents of its parabolic subgroups (as it is the case when M is locally symmetric), by [DOP].
For the second assertion, we refer to [Mar2] and [Bow] when M is compact, and to [OP] under the weaker assumption that m BM is finite.We refer to [Rob2] for a proof of the first assertion, to [Bab,Thm. 1] for the third one, and to [PaPS] for the extensions of the first three assertions to the case with potential.The assumption of Assertion (3), called non-arithmeticity of the length spectrum holds for instance, by [Dal1,Dal2], when M is locally symmetric or 2-dimensional, or when Γ contains a parabolic element, or when ΛΓ is not totally disconnected.
In Assertion (4), called the exponential decay of correlation property of the Bowen-Margulis measure, we denote by • the Sobolev norm of regularity .We refer to [KM1] for a proof of this last assertion, with the help of [Clo,Theorem 3.1] to check its spectral gap property and of [KM2,Lemma 3.1] to deal with finite cover problems.Note that the spectral gap property has been checked by [MO] if M = H n R and Γ is only assumed to be geometrically finite with δ Γ ≥ n − 2 if n ≥ 3 and δ Γ ≥ 1 2 if n = 2, thus providing the first infinite volume examples for which Assertion (4) holds.
When M is locally symmetric with finite volume, the Bowen-Margulis measure m BM coincides, up to a multiplicative constant, with the Liouville measure, that is the Riemannian measure Vol T 1 M of Sasaki's metric on T 1 M .See for instance [PaP5,§7] when M is real hyperbolic.In particular, the measure is finite in this case.More precisely (see [PaP5,§7] and [PaP6,§6]), if M = H n R and if M has finite volume, normalizing the Patterson-Sullivan measure so that its total mass is the volume of the (n−1)-sphere with its standard spherical metric (so that µ x 0 = Vol(S n−1 )), we have, denoting by Vol N the Riemannian volume of any Riemannian manifold N ,

Geometric equidistribution and counting
In this section, we will link the Diophantine approximation problems of Subsections 2.3, 2.4 and 2.5 to general geometric equidistribution and counting problems on the common perpendiculars between two locally convex subsets in a negatively curved Riemannian orbifold.Assume for simplicity in the introduction of this Section 4 (thus avoiding problems of regularity, multiplicities and finiteness), that N is a compact negatively curved Riemannian manifold, and that D − and D + are proper nonempty disjoint closed locally convex subsets of M with smooth boundaries.A common perpendicular from D − to D + is a locally geodesic path in N starting perpendicularly from D − and arriving perpendicularly to D + .
There is exactly one such common perpendicular in every homotopy class of paths starting from D − and ending in D + , where during the homotopy the origin of the path remains in D − and its terminal point remains in D + .In particular, there are at most countably many such common perpendiculars, and at most finitely many when their length is bounded.
Even when N is a closed hyperbolic surface and D − , D + are simple closed geodesic (see the picture below), the result (see Equation ( 4)) was not known before appearing in [PaP6].
We give in Subsection 4.1 (refering to [PaP6] for complete statements and proofs) an asymptotic formula as t → +∞ for the number of common perpendiculars of length at most t from D − to D + , and an equidistribution result as t → +∞ of the initial and terminal tangent vectors v − α and v + α of these common perpendiculars α in the outer and inner unit normal bundles of D − and D + , respectively.Although we do use Margulis's mixing ideas, major new techniques needed to be developed to treat the problem in the generality considered in [PaP6], some of them we will indicate in Subsection 4.1.
Here is a striking corollary of Theorem 15 in a very different context, that apparently does not involve negative curvature dynamics or geometry.Let Γ be a geometrically finite discrete subgroup of PSL 2 (C) (acting by homographies on P 1 (C) = C ∪ {∞}).Assume that Γ does not contain a quasifuchsian subgroup with index at most 2, and that its limit set ΛΓ is bounded and not totally disconnected in C.These assumptions are only here to ensure that the domain of discontinuity ΩΓ = (C ∪ {∞}) − ΛΓ of Γ has infinitely many connected components (only one of them unbounded).The following result gives a precise asymptotic as tends to 0 on the counting function of the (finite) number of these connected components whose diameter are at least .
The multiplicative constant has an explicit value, that requires some more notation, and does involve hyperbolic geometry.We denote by (Ω i ) i∈I a family of representatives, modulo the action of Γ, of the connected components of ΩΓ whose stabilisers have infinite index in Γ.For every i ∈ I, let C Ω i be the convex hull of Ω i in the upper-half space model of the 3-dimensional real hyperbolic space H 3 R , and let σ − C Ω i be the (inner) skinning measure of C Ω i for Γ.We also denote by HB ∞ the horoball in H 3 R consisting of points with vertical coordinates at least 1, and by σ + HB∞ its (outer) skinning measure for Γ.
Corollary 13 (Parkkonen-Paulin) Let Γ be a geometrically finite discrete group of PSL 2 (C), with bounded and not totally disconnected limit set in C, which does not contain a quasifuchsian subgroup with index at most 2. Assume that the Hausdorff dimension δ of the limit set of Γ is at least 1 2 .Then there exists κ > 0 such that the number of connected components of the domain of discontinuity ΩΓ of Γ with diameter at least is equal, as → 0, to We refer to [PaP6,Coro. 25] for a proof of this result, with the error term coming from [PaP6,Theo. 28] and [MO], as explained in the discussion of Assertion (4) of Theorem 12.This corollary largely extends the result of  when all the connected components of the domain of discontinuity are assumed to be round discs.Note that the fractal geometry of the boundary of general convex hulls is an important feature that needs to be adressed by non-homogeneous dynamics arguments.

Equidistribution and counting of common perpendicular
Let ( M , Γ, M ) be as in the beginning of Subsection 3.2.A common perpendicular from a closed convex subset A − in M to a closed convex subset A + in M is a geodesic arc α in M whose initial tangent vector v − α belongs to ∂ 1 + A − and terminal tangent vector v + α belongs to ∂ 1 − A + .Note that there exists such a common perpendicular if and only if A − and A + are nonempty, proper, with disjoint closures in M ∪ ∂ ∞ M .It is then unique, and its length is positive.
Let C ± be nonempty proper closed convex subsets of M , with stabiliser Γ C ± in Γ, such that the family (γC ± ) γ∈Γ/Γ C ± of subsets of M is locally finite.We denote by Perp(C − , C + ) the set of images in M = Γ\ M of the common perpendiculars from γ − C − to γ + C + as γ ± ranges over Γ, and, for every t > 0, by Perp(C − , C + , t) the subset of the ones with length at most t.For every α ∈ Perp(C − , C + ), we denote by v − α and v + α its initial tangent vector and terminal tangent vector, which belong to the image in Since Γ might have torsion, and since Γ C ± \C ± does not necessarily embed in M = Γ\ M , each element α of Perp(C − , C + , t) comes with a natural multiplicity m(α).Denote by α any common perpendicular in M between translates of C − and C + with image α in M and by Γ α its pointwise stabiliser in Γ.Then Note that the numerator and the denominator are finite by the local finiteness of the families (γC ± ) γ∈Γ/Γ C ± and the discreteness of Γ, and they depend only on the orbit of α under Γ.This multiplicity is indeed natural.Concerning the denominator, in any counting problem of objects possibly having symmetries, the appropriate counting function consists in taking as the multiplicity of an object the inverse of the cardinality of its symmetry group.The numerator is here in order to take into account the fact that the elements γ ± in Γ such that α is a common perpendicular from γ − C − to γ + C + are not necessarily unique, even modulo Γ C ± .The natural counting function of the common perpendiculars between the images of C ± is then the map m(α) .
The reader can assume for simplicity that Γ is torsionfree and that Γ C ± \C ± embeds in M = Γ\ M by the map induced by the inclusion of C ± in M , in which case all multiplicities are 1.
Below, we state our equidistribution result, in the outer and inner unit normal bundles of the images in M of C − and C + , of the initial and terminal tangent vectors of the common perpendiculars between the images of C − and C + , as well as our asymptotic formula as t → +∞ for the number of common perpendiculars of length at most t between the images of C − and C + .We refer respectively to [PaP6,Theo. 14,28] and [PaP6,Coro. 20,28] for more general versions, involving more general locally finite families of convex subsets, versions with weights coming from potentials (the Bowen-Margulis measure being then replaced by the Gibbs measure of [PaPS]), and for version with error terms under an additional assumption of exponential decay of correlations (see Theorem 12 (4)).
Theorem 14 (Parkkonen-Paulin) Assume that m BM is finite and mixing under the geodesic flow, and that the skinning measures σ ± C ∓ are finite.For the weak-star convergence of measures on T 1 M × T 1 M , we have Theorem 15 (Parkkonen-Paulin) Assume that m BM is finite and mixing under the geodesic flow, and that the skinning measures σ ± C ∓ are finite and nonzero.Then, as t → +∞, The counting function N C − , C + (t) has been studied in various special cases since the 1950's and in a number of recent works, sometimes in a different guise, see the survey [PaP5] for more details.A number of special cases were known before our result: • C − and C + are reduced to points, by for instance [Hub], [Mar1] and [Rob2], • C − and C + are horoballs, by [BeHP], [HP1], [Cos] and [Rob2] without an explicit form of the constant in the asymptotic expression, • C − is a point and C + is a totally geodesic submanifold, by [Her], [EsM] and [OhS1] in constant curvature, • C − is a point and C + is a horoball, by [Kon] and [KonO] in constant curvature, and [Kim] in rank one symmetric spaces, • C − is a horoball and C + is a totally geodesic submanifold, by [OhS3] and [PaP2] in constant curvature, and • C − and C + are (properly immersed) locally geodesic lines in constant curvature and dimension 3, by [Pol].
As a new particular case, if M has constant curvature −1, if the images in M of C − and C + are closed geodesics of lengths − and + , respectively, then the number of common perpendiculars (counted with multiplicity) from the image of C − to the image of C + of length at most s satisfies, as s → +∞, When M is a closed hyperbolic surface and C − = C + , this formula (4) has been obtained by Martin-McKee-Wambach [MMW] by trace formula methods, though obtaining the case C − = C + seems difficult by these methods.
The family ( (α)) α∈Perp(C − , C + ) (with multiplicities) will be called the marked ortholength spectrum from C − to C + .The set of lengths (with multiplicities) of elements of Perp(C − , C + ) will be called the ortholength spectrum of C − , C + .This second set has been introduced by Basmajian [Bas] (under the name "full orthogonal spectrum") when M has constant curvature, and the images in M of C − and C + are disjoint or equal embedded totally geodesic hypersurfaces or embedded horospherical cusp neighbourhoods or embedded balls (see also [BriK] when M is a compact hyperbolic manifold with totally geodesic boundary and the images in M of C − and C + are exactly ∂M ).The two results above are hence major contributions to the asymptotics of marked ortholength spectra.
Let us give a brief sketch of the proof of Theorem 15, refering to [PaP6, §4.1] for a full proof.
Step 1.In this technical step, we start by constructing dynamical neighbourhoods and test functions around the outer/inner unit normal bundles of our convex sets, that will be appropriately pushed forward/backwards by the geodesic flow (using the nice contraction/dilation properties of Hamenstädt's distances compared to the ones of the induced Riemannian metric in variable curvature).We fix R > 0 big enough, and we will let η > 0, a priori small enough, tend to 0.
For all w ∈ T 1 M , let V + w be the open ball of center w and radius R for Hamenstädt's distance on the strong stable leaf W + (w) of w.For every η > 0, let 000 000 000 000 000 111 111 111 111 111 be the measurable and m BM -almost everywhere finite map (since its denominator is positive if w − ∈ ΛΓ) defined by .
Let us denote by 1 A the characteristic function of a subset A. Consider the map ( This map is Γ-invariant and measurable, hence it defines a measurable map ψ − η : By the disintegration result of Proposition 11, the functions ψ ∓ η are integrable and Step 2. In this step, we use the mixing property of the geodesic flow, as first introduced by Margulis in his thesis (see for instance [Mar3]).Due to the symmetry of the problem, a one-sided pushing of the geodesic flow, as in all the previous works using Margulis's ideas, is not sufficient, and we need to push simultaneously the outer and inner unit normal vectors to the convex sets in opposite directions.For all t ≥ 0, let Then the mixing hypothesis of the geodesic flow and Equation ( 6) imply that Step 3.
In this step, we give another estimate of a η (t), relating it to the counting of common perpendiculars.One of the main new ideas in the proof (see [PaP6,§2.3]for a complete version) is an effective study of the geometry and the dynamics of the accidents that occur around midway of the pushing by the geodesic flow, yielding an effective statement of creation of common perpendiculars.
In order to give an idea of this phenomenon, assume that v ∈ T 1 M belongs to the support of the function ψ . This is equivalent to assuming that φ ±t/2 v belongs to the support of ψ ± η .If v is a lift of v to T 1 M , by the definition of the maps ψ ∓ η , this is equivalent to asking that there exist γ ± ∈ Γ, w ± ∈ ∂ 1 ∓ C ± and s ± ∈ ]−η, η[ such that φ ±(t/2+s ± ) v belongs to Hamenstädt's balls V ∓ w ± .If R is fixed, when η > 0 is small enough and t is big enough, negative curvature estimates say that v is very close to the tangent vector at the midpoint of a common perpendicular from γ − C − to γ + C + whose length is close to t (see [PaP6,§2.3]).
To obtain a precise estimate on a η (t), given a fundamental domain F for the action of Γ on T 1 M , we apply Fubini's theorem, as in Sarnak's "unfolding technique": , as well as a fine analysis (especially refined for the error term estimates) of the intrinsic geometry in variable curvature (almost everywhere defined second fundamental form, ...) of the outer/inner unit normal bundle pushed a long time by the geodesic flow.We then conclude by a Cesaro-type of argument in order to consider all common perpendiculars with length at most T , as T tends to +∞, and by letting η tend to 0.

Towards the arithmetic applications
As we already hinted to, Theorems 2, 3, 4, 5, 7, 9 and 10 all follow from Theorem 14 or Theorem 15, though many more tools and ideas are needed, in particular volume computations of arithmetic orbifolds.
We only indicate the very beginning of the proof of these theorems, giving a bit more details on Theorems 1 to 4.
To prove Theorem 5, we apply Theorem 14 with M = H 3 R , Γ the Bianchi group PSL 2 (O K ), and C − = C + any horoball centered at ∞ in the upper halfspace model of H 3 R .Note that the cusps of the noncompact finite volume hyperbolic orbifold PSL 2 (O K )\H 3 R correspond to the ideal classes of O K (in particular if K = Q(i), there is only one cusp), see for instance [ElGM].Keeping the same C − and taking C + centered at a parabolic fixed point defining the cusp allows version of Theorem 5 when p and q are varying in a given fractional ideal of O K (when the class number of the imaginary quadratic number field K is larger than 1).
To prove Theorem 7, we consider the Hermitian form h : (z 0 , z 1 , z 2 ) → −z 0 z 2 − z 2 z 0 + |z 1 | 2 on C 3 whose signature is (1, 2).We apply Theorem 14 with M the projective model {[z 0 : z 1 : z 2 ] ∈ P 2 (C) : h(z 0 , z 1 , z 2 ) < 0} of the complex hyperbolic plane H 2 C , with Γ the Picard group PSU h (O K ) = PSU h ∩ PGL 3 (O K ), and with C − = C + any horoball centered at ∞ = [1 : 0 : 0].See [PaP7,Theo. 12,13] for the other ingredients of the proof.The reason why large white regions appear around the points p q ∈ K with |q| small in the figures before and after Theorem 5 is that the horoballs in the Γ-orbit of C − centered at these points have large Euclidean radius, hence it is (quadratically) difficult to fit disjoint horoballs in this orbit below them.
Replacing in the above data C + by the convex hull in H 2 C of an arithmetic chain C 0 , applying Theorem 15 and 14 is the very first step for proving Theorem 9 and 10, respectively.See [PaP7,Theo. 19,20] for the other ingredients of the proof.
To prove Theorems 2, 3 and 4, we apply Theorem 14 or Theorem 15 with M the upper halfplane model of H 2 R and Γ = PSL 2 (Z) (or appropriate finite index subgroups when we want versions with additional congruence assumptions).Note that the modular curve PSL 2 (Z)\H 2 R , being arithmetic hyperbolic, has its Bowen-Margulis measure proportional to its Liouville measure, hence finite, and its geodesic flow is mixing, with exponential decay of correlation.We then take The key input, also crucial to prove Theorem 1, is the well-known hyperbolic geometry understanding of quadratic irrationals.A real number α is a quadratic irrational if and only if it is fixed by (the action by homography of) an element γ ∈ PSL 2 (Z) with | tr γ| > 2. Then α σ is the other fixed point of γ, and the geodesic line L α = ]α, α σ [ maps to a closed geodesic in the hyperbolic orbifold PSL 2 (Z)\H 2 R .In particular, the image of ∂ 1 ± L α in PSL 2 (Z)\T 1 H 2 R is compact, and the skinning measures σ ± Lα are positive and finite.The first hint that there is a connection between quadratic irrationals and common perpendiculars is the following one.Let HB ∞ be the horoball centered at ∞, consisting of the points of H 2 R with Euclidean height at least 1.Note that its stabiliser in PSL 2 (Z) acts cocompactly on ∂ 1 ± HB ∞ , hence the skinning measures σ ± HB∞ are positive and finite.Then, by an easy computation in hyperbolic geometry, the common perpendicular between HB ∞ and the geodesic line ]α, α σ [ (assuming that they are disjoint) has length ln H(α), where Another important observation to prove Theorem 4 (taking α = 1+ √ 5 2 the Golden Ratio, C − = L α and C + = HB ∞ ) is that since the modular curve has finite volume, the skinning measure on ∂ 1 + C − is homogeneous.Hence, on each of the two connected components of ∂ 1 + C − which are naturally parametrised by R, it is proportional to the Lebesgue measure, and this Lebesgue measure projects by the negative endpoint map v → v − to a measure on R − {α, α σ } proportional to d Leb R (t)  |t 2 −t−1| , explaining the limit in Theorem 4. See [PaP4,Theo. 6] for a complete proof.
H (y) of y by the elements of Z with height function H as ω(y) = lim sup z∈Z − ln d(y, z) ln H(z) , and the approximation constant c(y) = c X, Y, Z, H (y) of y by the elements of Z with height function H as c(y) = lim inf z∈Z H(z) d(y, z) .The Lagrange spectrum Sp Lag = Sp X, Y, Z, H for the approximation of the elements of Y by the elements of Z with height function H is Sp Lag = {c(y) : y ∈ Y } .Its least upper bound will be called the Hurwitz constant for the approximation of the elements of Y by the elements of Z with height function H.Given ψ : ]0, +∞[ → ]0, +∞[, the set WA ψ = WA ψ, X, Y, Z, H of ψ-well approximable elements of Y by the elements of Z with height function H is of the chains.The picture below shows part of an orbit of arithmetic chains under the arithmetic lattice PSU h (O K ).Theorem 9 (Parkkonen-Paulin) Let C 0 be an arithmetic chain over K in the hypersphere H S .Then there exists a constant κ > 0 such that, as > 0 tends to 0, the number of chains modulo PSU h (O K ) ∞ in the PSU h (O K )-orbit of C 0 with d Cyg -diameter at least is equal to 512 ζ(3) Covol(C 0 ) • C − a horoball centered at ∞ and C + the geodesic line ]α 0 , α σ 0 [ in H 2 R with points at infinity α 0 , α σ 0 for Theorem 2, • C − the geodesic line ]α 0 , α σ 0 [ and C + the geodesic line ]β 0 , β σ 0 [ for Theorem 3, • C − the geodesic line ]α 0 , α σ 0 [ and C + a horoball centered at ∞ for Theorem 4.