\magnification = 1200 \hfuzz=10pt \hsize=4.8in \vsize=7.3in \baselineskip=18pt \hoffset=0.35in \voffset=0.1in \parindent=3pt \def\di{\displaystyle} \def\R{I\!\!R} \def\C{I\!\!\!\!C} \def\N{I\!\!N} \def\T{I\!\!\!T} \def\Q{I\!\!\!\!Q} \def\Z{Z} \def\u{\upsilon} \def\O{\Omega} \def\o{\omega} \def\t{\theta} \def\z{\zeta} \def\fa{{2\pi i\over p}} \def\fp{{2\pi i\over p}} \def\s{\sigma} \def\a{\alpha} \def\b{\beta} \def\k{\kappa} \def\g{\gamma} \def\pp{\pi^{\prime}} \def\eps{\epsilon} \def\A{{\cal A}} \def\U{{\cal U}} \def\F{I\!\!F} \def\QT{{\hat T}} \def\Qf{{\hat f}} \def\l{\ell} \def\lj{\lambda_j} \def\I1{{\cal I}_1} \def\I2{{\cal I}_2} \def\aj{\alpha_j} \def\P{{\cal P}} \def\vj{v_j} \def\st{\sum_{s=0}^{T-1}} \def\sx{\sum_{x\in \Z_p^*}} \def\bk{\beta_k} \def\lk{\lambda_k} \def\L{I\!\!L} \def\t{{\rm T}} \def\G{\Gamma} \def\D{{\cal D}} \def\ti{{1\over \t}} \centerline{\bf DISTRIBUTION OF CLOSED ORBITS} \centerline{\bf FOR LINEAR AUTOMORPHISMS OF TORI} \vglue 0.2cm \vglue 1.0cm \centerline{ Mirko Degli Esposti and Stefano Isola} \vglue 0.5cm \centerline{\it Dipartimento di Matematica, Universit\'a degli Studi di Bologna,} \centerline{\it piazza di Porta S.Donato 5, I-40127 Bologna, Italy} \vskip 0.5cm \centerline{\it e-mail : desposti@dm.unibo.it, isola@dm.unibo.it} \vskip 2cm {\bf Abstract.} In this paper we study the distribution properties of periodic orbits for the linear hyperbolic automorphisms of the $d$-torus. We first obtain an explicit expression of the dynamical zeta function and prove general equidistribution results similar to those obtained for Axiom A flows. We then study in detail some families of periodic orbits living on invariant prime lattices: they have the property that the integral of any character along any single orbit can be reduced to a number theoretic exponential sum over a finite field. This fact enables us to obtain explicit estimates on their asymptotic distributional properties. \vfill \eject {\bf 1. Introduction.} \vskip 0.5cm A number of interesting results on the distribution of periodic orbits for hyperbolic dynamical systems have recently been obtained pursuing strong analogies with general results of number theory and probability theory. Indeed, using the Ruelle zeta function [18] for a hyperbolic flow in the same way as the Riemann zeta function is used to prove the prime number theorem (that is through Wiener-Ikehara Tauberian theorem, see, e.g., [11]), Parry and Pollicott [15] proved a `prime orbit theorem', saying that the number $\pi (x)$ of periodic orbits $\gamma$ with (minimal) period $p(\gamma) \leq x$ is asymptotic to $e^{hx}/ hx$ as $x\to \infty$, where $h$ is the topological entropy of the flow. Moreover, if $\int_{\gamma} \, f$ denotes the (normalized) integral of the continuous function $f$ along $\gamma$, then $(1 /\pi (x)) \, \sum_{p(\gamma)\leq x} \int_{\gamma} \, f \rightarrow \int \, f\, d\mu$ as $x\to \infty$, where $\mu$ is the invariant probability measure of maximal entropy (see [16] and references therein). Other types of theorems yield the equidistribution of periodic orbits for hyperbolic flows in the sense of probability theory [12], [10]. For instance, under certain conditions and for any $\eps >0$, ${\rm Prob}\{ |\int_{\gamma} \, f - \int \, f\, d\mu | > \eps \} \to 0$ as $p(\gamma )\to \infty$, which of course improves the former result. This means that, except for a small (tending to zero) fraction, long periodic orbits are nearly uniformly distributed according to $\mu$. In what follows we shall study the distribution properties of periodic orbits for a particular example: the linear hyperbolic automorphism of the $d$-torus. Besides proving general results as those mentioned above (Section 2), we shall concentrate on some particular families of periodic orbits living on invariant prime lattices (Section 3). These enjoy the nice properties that the integrals of characters over any single orbit can be reduced to (some combinations of) number theoretic exponential sums, like generalized Gauss sums or Kloosterman sums, over a finite field (see [20]). This makes possible to obtain explicit estimates on their asymptotic distributional properties. We point out that the statistical properties of the periodic orbits turn out to be crucial in the investigation of the quantum counterpart of this dynamical system [8], [9]. In particular there is an intriguing connection between the spatial distribution of certain families of closed orbits (see Section 3) and the localization properties of the eigenfunctions of the quantum propagator in the classical limit [4], [5]. \vskip 1cm {\bf 2. The algebraic hyperbolic automorphisms of the $d$-torus.} \vskip 0.5cm We consider the discrete time dynamical system $F\, :\, M\to M$ where $M$ is the $d-$torus $\T^d=\R^d /\Z^d$ (points on $\T^d$ are denoted by $x=(x_1,\dots ,x_d)\in [0,1]^d$) and $F$ is the hyperbolic automorphism of $\T^d$ generated by the $d\times d$ matrix $$ A = \pmatrix{ a_{11} & a_{12} & \ldots & a_{1d}\cr a_{21} & a_{22} & \ldots & a_{2d}\cr \vdots & \vdots & \ddots & \vdots \cr a_{d1} & a_{d2} & \ldots & a_{dd}\cr } \eqno(2.1) $$ where $a_{ij}\in \Z$, $i,j =1\ldots d$, and $A$ has real spectrum $\Lambda=\{\lambda_1, \ldots , \lambda_d\}$ such that $\lambda_1 \lambda_2\dots \lambda_d = 1$. The Lebesgue measure $\mu$ is invariant because $\det A=1$. The above condition on the eigenvalues makes this dynamical system an Anosov one and hence, in particular, ergodic and mixing with respect to $\mu$. Now, an orthonormal basis in $L^2 (\T^d,d\mu)$ is given by the set $$ \{T(n) = e^{2\pi i}|n\in \Z^dJ\}\eqno(2.2) $$ and $A$ acts on points $x=(x_1,\dots ,x_d)$ and on suitably smooth functions $$ f(x)=\sum_{n\in\Z^d}\, f_n\, T(n) $$ respectively as: $$ \cases{ Ax =\left( (a_{11}x_1 +\dots a_{1d}x_d)({\rm mod}\,1), \ldots , (a_{d1}x_1 +\dots a_{dd}x_d)({\rm mod}\,1) \right) \cr f(Ax) = \sum_{n\in\Z^d}\, f_n\, T(A^tn) \cr } \eqno(2.3) $$ where $A^t$ is the transposed matrix of $A$ (notice that we have used the same symbol $A$ to denote the matrix $A$ and the map $F$; this will be repeatedly done in what follows without fear of confusion). Now, if $A$ is a continuous map of a compact metric space $X$ then $h(A)$, the topological entropy, satisfies the restricted variational principle: $$ h(A) = \sup_{\mu \in {\cal M}_A(X)}h_{\mu}(A) \eqno(2.4) $$ where ${\cal M}_A(X)$ the set of the probability measures on the Borel $\sigma$-algebra of $X$ which are $A$-invariant and $h_{\mu}(A)$ is the measure theoretic entropy (see, e.g., [13]). Moreover, $A$ is {\it intrinsically ergodic} if there exist a unique $\mu \in {\cal M}_A(X)$ such that $h(A)=h_{\mu}(A)$. In this case $\mu$ is called the intrinsic measure of $A$, or maximal entropy measure of $A$. Thus, for a linear automorphism of the torus the Haar measure, i.e. the Lebesgue one, is actually an intrinsic measure. We have the \vskip 0.2cm {\bf Theorem 2.1 (Sinai)} {\sl Let $A:\T^d\to \T^d$ be a linear automorphism with eigenvalues $\lambda_1, \ldots \lambda_d$. Let $\mu$ be the Haar measure of $T^d$. Then} $$ h(A)=h_{\mu}(A)=\log \prod_{i\, : \, |\lambda_i| >1}|\lambda_i| \eqno(2.5) $$ \vskip 0.2cm {\it Proof.} See, e.g., [13] p. 265. \vskip 0.2cm Consider again the general situation of a continuous map $A$ of a compact metric space $X$. We say that $x\in X$ is a {\it periodic point} of $A$, of period $n$, if it is a {\sl fixed point} of $A^n$, i.e. $A^nx=x$. We denote by $Fix_n$ the set of such points. It is easy to see that for linear automorphism of the torus the set of periodic points of $A$ is dense in $\T^d$, because it coincides with the subset of $\T^d$ formed by all points having rational coordinates. More generally we have the \vskip 0.2cm {\bf Theorem 2.2 (Bowen-Sinai)} {\it Every topologically mixing hyperbolic homeomorphism $A:X\to X$ is intrinsically ergodic. If $\mu$ denotes its intrinsic measure then for every continuous map $f :X\to \R$ : $$ \int_Xf d\mu = \lim_{n\to \infty}{1\over \# Fix_n} \sum_{x\in Fix_n}f (x)\eqno(2.6) $$ and $A$ has topological entropy} $$ h= \lim_{n\to \infty}{1\over n} \log{\# Fix_n}\eqno(2.7) $$ {\it Proof.} See [13], p.254. \vskip 0.2cm For the linear automorphism of the $d$-torus we have the additional result: \vskip 0.2cm {\bf Proposition 2.1.} $$ \# Fix_n = \left|\, \det(A^n-1)\, \right| = \left|\, \prod_{i=1}^d(\lambda_i^n-1) \, \right| \eqno(2.8) $$ \vskip 0.2cm {\bf Remark.} Notice that from (2.7) and (2.8) one immediately recovers (2.5). \vskip 0.2cm {\it Proof.} The periodic points of period $n$ of the map $A:\T^d\to \T^d$ are given by the solution of the linear system $$ \left( A^n - 1\right)\, x \, = \, k, \qquad x\in \R^d, \; k\in \Z^d\eqno(2.9) $$ It is easy to realize that for $k$ varying in $\Z^d$ the above system determines a regular lattice $L_n$ in the covering space $\R^d$ so that the density of periodic points belonging to $Fix_n$ is given by the inverse of the volume of the fundamental domain $\Delta_n$ of $L_n$. On the other hand, if $\{e_1, \ldots , e_d\}$ denotes the basis of the lattice $L_n$ and each vector $e_i$ is expressed as $e_i = (r_{1i}, \ldots , r_{di})$ then the volume $\u (\Delta_n)$ of $\Delta_n$ is given by $$ \u (\Delta_n) = \left| \det r_{ij} \right| \eqno(2.10) $$ Now, a basis $\{e_1, \ldots , e_d\}$ of the lattice $L_n$ can be easily obtained by solving eq. (2.9) for $d$ distinct and linearly independent vectors $k_1= (k_{11}, \dots , k_{1d}), \ldots , k_d= (k_{d1}, \dots , k_{dd})$, $k_{ij}\in \Z$, such that $$ \det K \equiv \det \left| \, \matrix{ k_{11} & \ldots & k_{1d}\cr \vdots & \ddots & \vdots \cr k_{d1} & \ldots & k_{dd}\cr }\, \right| =1 \eqno(2.11) $$ This yields $$ \u (\Delta_n) = {1\over \Gamma^d} \det \left| \, \matrix{ D_{11} & \ldots & D_{1d}\cr \vdots & \ddots & \vdots \cr D_{d1} & \ldots & D_{dd}\cr }\, \right| \eqno(2.12) $$ where $\Gamma = \left| \, \det \left( A^n-1\right)\, \right|$ and $D_{ij}$ is the determinant of the $d\times d$ matrix obtained by inserting the vector $k_i$ in the place of the $j$-th column of the matrix $A^n-1$ (Cramer's rule). Finally, a tedious but trivial calculation where the basic properties of the determinant are exploited gives $$ \u (\Delta_n) = {\Gamma^{d-1}\det K \over \Gamma^d} \eqno(2.13) $$ Hence, from (2.11) and (2.13) one obtains $$ \# Fix_n = {1\over \u (\Delta_n)} = \left| \, \det \left( A^n-1\right)\, \right| \eqno(2.14) $$ Q.E.D. Let us now restrict ourselves to symplectic matrices, i.e. $d$ is even, say $d=2r$, and $A^tJA=J$, where $J=\left(\matrix{ 0 & E \cr -E & 0 \cr }\right)$, and $E$ denotes the $r\times r$ identity matrix, so that if $\lambda \in \Lambda$ then also $1/\lambda \in \Lambda$. The following result extends an earlier result for $d=2$ [I]. \vskip 0.2cm {\bf Corollary 2.1.} {\it Let $A$ be symplectic and $\Lambda = \{\lambda_1,\ldots , \lambda_r, 1/\lambda_1, \ldots , 1/\lambda_r \}$ where $\lambda_i >1$, $i=1,\dots ,r$. Then,} $$ \# Fix_n = \prod_{i=1}^r\left( \lambda_i^n + \lambda_i^{-n} -2 \right) \eqno(2.15) $$ \vskip 0.2cm {\it Proof.} $$ \left| \prod_{i=1}^d\left( \lambda_i^n -1 \right) \right| = \prod_{i=1}^r\left( \lambda_i^n -1 \right) \, \prod_{i=1}^r\left( 1- \lambda_i^{-n} \right) =\prod_{i=1}^r\left( \lambda_i^n + \lambda_i^{-n} -2 \right) $$ Q.E.D. We now consider the following dynamical zeta function [21]: $$ \zeta (z) = \exp \sum_{n=1}^{\infty}{z^n\over n} \# Fix_n \eqno(2.16) $$ By virtue of Theorem 2.2. the series converges for $|z|< e^{-h}$. Moreover, from Proposition 2.1 it easily follows that for any algebraic hyperbolic automorphisms of the $d$-torus $\zeta(z)$ is a rational function with real zeroes and poles. In the simpler case of an orientation preserving symplectomorphism, from Corollary 2.1 one obtains the following expression: \vskip 0.2cm {\bf Proposition 2.2.} {\it Under the assumptions of Corollary 2.1 we have $$ \zeta (z) = \prod_{\scriptstyle \di i_1\not= \dots \not= i_{r_1} \not= j_1\not= \dots \not=j_{r_2} \atop \scriptstyle \di r_1 + r_2+r_3 = r } {1\over \left(\di 1 - \lambda_{\di i_1}\dots \lambda_{\di i_{r_1}}\lambda_{\di j_1}^{-1}\dots \lambda_{\di j_{r_2}}^{-1} \, \cdot z \right)^{(\di -2)^{\di r_3} } } \eqno(2.17) $$ where the product is over distinct $(r_1+r_2)$-tuples.} \vskip 0.2cm Observe that the above zeta function has $2^r$ simple real poles located at $z = \left(\lambda_{i_1}\dots \lambda_{i_{r_1}}\lambda_{j_1}^{-1}\dots \lambda_{j_{r_2}}^{-1}\right)^{-1}$ where $i_1\not= \dots \not= i_{r_1} \not= j_1\not= \dots \not=j_{r_2}$ and $r_1+r_2=r$. In particular, for $r_2=0$ one gets the first pole $z=\left(\lambda_1\dots \lambda_r\right)^{-1} = e^{-h}$. All the other poles and zeroes have multiplicities larger than one. In particular, for $r_1=r_2=0$ one gets the most degenerate zero (or pole, depending on the parity of $r$) located at $z=1$, with multiplicity $2^r$. \vskip 0.2cm {\bf Remarks.} \vskip 0.2cm 1) For the case $d=2$ ($r=1$) one finds the expression $$ \zeta (z) = { (1-z)^2 \over (1-\lambda z)(1-z/\lambda ) } $$ where $\lambda$ is the largest eigenvalue of $A$ [7]. \vskip 0.2cm 2) If instead of the zeta function (2.16) one considers the following Fredholm determinant: $$ d(z) = \exp - \sum_{n=1}^{\infty}{z^n\over n} \sum_{x\in Fix_n} {1\over |\det (A^n(x)-1)|} $$ then using Proposition 2.1 one gets the expression (valid for any algebraic hyperbolic automorphism $A$ of the $d$-torus): $\, d(z)=1-z$. In particular, this implies the otherwise easily proven result that correlation functions for analytic observables decay faster than any exponential (see [19]). \vskip 0.2cm We now deduce some consequences of the above results. For simplicity' sake we shall give the arguments for the case of the hyperbolic symplectomorphism of the $d$-torus, with $d=2r$, but all the results given in the rest of this Section can be easily extended to the general setting. Denote by $P$ the set of all primitive (i.e., non repeated) periodic orbits of $A$ and by $\pp (n)$ and $\pi (x)$ the number of them whose period is $n$ and less or equal to $x$ respectively, i.e. $$ \pp (n) = \# \{\g \in P \, : \, p(\g ) = n \} \qquad \pi (x) = \# \{\g \in P \, : \, p(\g )\leq x\} \eqno(2.18) $$ \vskip 0.2cm {\bf Proposition 2.3.} {\it $$ \pp (n) \sim { e^{nh} \over n}\eqno(2.19) $$ and $$ \pi (x) \sim { e^{h} \over e^{h} -1}\cdot { e^{xh}\over x}\eqno(2.20) $$ where $f(t)\sim g(t)$ means that $f(t)/g(t)\rightarrow 1$ when $t\to \infty$.} \vskip 0.2cm {\bf Remark.} Proposition 2.2 is the analogue of the {\it prime orbit theorem} proved in the context of Axiom A flows (see [16]). \vskip 0.2cm {\it Proof.} The strategy is to gain insight on the distribution of closed orbits out of the meromorphy domain of $\zeta (z)$. >From Proposition 2.2 we have that $$ {\zeta^{\prime} (z)\over \zeta (z)}={e^h \over 1-e^hz} +g(z) $$ where $g(z)$ is analytic in $\{z\, : \, |z|0$. The rest of the proof makes use of the Wiener-Ikehara Tauberian theorem and proceeds exactly in the same way as in the proof for subshifts of finite type ([16], p.100). Q.E.D. \vskip 0.2cm We now prove that closed orbits exhibit a regularity in a spatial sense. In particular, we show that they are equidistributed on the average with respect to the Lebesgue measure $\mu$. Let $\mu_{\g}$ be the measure defined by $$ \mu_{\g}={1\over p(\g)}\sum_{k=0}^{p(\g)-1}\delta_{A^k(x)}\qquad x\in \g \eqno(2.24) $$ and set $\int_{\gamma}f =\int_{\T^d}f d\mu_{\gamma}$. Then we have the \vskip 0.2cm {\bf Proposition 2.4.} {\it For every continuous map $f :\T^d \to \R$ :} $$ {1\over \pp (n)} \sum_{p(\gamma )=n}\int_{\gamma}f \longrightarrow \int_{\T^d}f d\mu \quad\hbox{as}\quad x\to \infty \eqno(2.25) $$ \vskip 0.2cm {\bf Remark.} This result is an easy consequence of more general results proved in [16]. Nevertheless, for the convenience of the reader, we shall give a simple direct proof based on Theorem 2.2 and Proposition 2.1. \vskip 0.2cm {\it Proof.} We first write the number of fixed points of period $n$ as $$ \# Fix_n = \sum_{l|n} l\cdot \pp (l)\eqno(2.26) $$ where $l|n$ means that $l$ divides $n$. Form Proposition 2.1 we then have $$ \sum_{l|n} l\cdot \pp (l) \sim e^{nh} \eqno(2.27) $$ On the other hand $$ \pp (n) = {1\over n}\left( \sum_{l|n} l\cdot \pp (l) - \sum_{l|n, \; l< n} l\cdot \pp (l) \right) $$ which, together with (2.27), gives $$ \pp (n) \sim {e^{nh}\over n} (1 - C_n) \eqno(2.28) $$ where $C_n$ is of order strictly smaller than one. Analogously, $$ \sum_{x\in Fix_n} f(x) = \sum_{l|n} l\cdot \sum_{p(\gamma )=l}\int_{\gamma}f \eqno(2.29) $$ >From Theorem 2.2 we then have $$ \sum_{l|n} l\cdot \sum_{p(\gamma )=l}\int_{\gamma}f \sim e^{nh}\int_{\T^d}f d\mu \eqno(2.30) $$ On the other hand $$ \sum_{p(\gamma )=n}\int_{\gamma}f = {1\over n}\biggl( \sum_{l|n} l\cdot \sum_{p(\gamma )=l}\int_{\gamma}f - \sum_{l|n, \; l0$ and for any continuous function $f$} $$ \lim_{n\to \infty }\biggl( {\#\{\g\, : \, p(\g )=n,\;\; |\int_{\g}f - \int_{\T^d}f d\mu |> \epsilon \}\over \pp (n)}\biggr) = 0 \eqno(2.32) $$ {\it Proof.} For the sake of simplicity we shall use the notation $ \int_{\T^d}f d\mu = {\bar f}$. For any $k\in \Z_+$ set $$ m_kf(x) = {1\over k}\sum_{l=0}^{k-1}f(A^lx)\eqno(2.33) $$ Now, given $\delta >0$, introduce the sets $$ \eqalign{ S_{\delta ,n} &= \{\gamma \, : \, p(\gamma )=n,\;\; \left|\int_{\gamma}f - {\bar f}\right|\leq \sqrt{\delta} \} \cr R_{\delta,n,k} &= \{\gamma \, : \, p(\gamma )=n,\;\; \int_{\gamma}|m_kf - {\bar f}|\leq \sqrt{\delta} \} \cr } \eqno(2.34) $$ We have $$ \left|\int_{\gamma}f - {\bar f}\right|\leq \left|\int_{\gamma}(f - m_kf)\right|+ \int_{\gamma}|m_kf - {\bar f}| $$ On the other hand it is obvious that $\int_{\gamma}(f - m_kf)=0$, for any $k\in \Z_+$; and thus $$ S_{\delta ,n}\supset R_{\delta ,n,k}\Longrightarrow R_{\delta ,n,k}^c \supset S_{\delta ,n}^c \quad\hbox{for any}\quad \delta >0, \; k\in \Z_+ \eqno(2.35) $$ The ergodicity of $A$ implies that, for $\mu$-almost every $x\in \T^d$, $$ \lim_{k\to \infty} m_kf(x) = {\bar f}\eqno(2.36) $$ Hence, for any continuous $f$, by the Lebesgue dominated convergence theorem we can find a $k_0>0$ such that $$ \int_{\T^d}|m_{k_0}f(x) - {\bar f}|d\mu \leq {\delta \over 2}\eqno(2.37) $$ Set $$ \int_{\T^d} f d\mu_n := {1\over \pp (n)} \sum_{p(\gamma )=n}\int_{\gamma}f $$ Then Proposition 2.4 entails that $d\mu_n$ converges vaguely to the Lebesgue measure. This implies the existence of $n_0(\delta )>0$ such that for any $n>n_0$ $$ \int_{\T^d}|m_{k_0}f(x) - {\bar f}|d\mu_n \leq \delta \eqno(2.38) $$ Hence by the Chebychev inequality we obtain $$ {\# \{\gamma \, : \, p(\gamma )=n,\;\; \int_{\gamma}|m_{k_0}f - {\bar f}|\geq \sqrt{\delta} \}\over \pp (n)} \leq \sqrt{\delta}\eqno(2.39) $$ and therefore, by the second of (2.34): $$ {\# R_{\delta ,n,k_0}^c\over \pp (n)} \leq \sqrt{\delta},\qquad n>n_0\eqno(2.40) $$ If we now put $\delta = 1/k^2$ then from (2.35) we find $$ { \# \{\g\, : \, p(\g )=n,\;\; |\int_{\g}f - \int_{\T^d}f d\mu |> {1\over k} \} \over \pp (n)} \leq {1\over k},\qquad n>n_0 \eqno(2.41) $$ so that, given $\eps >0$, we have for $k > 1/\eps$ and $n > n_0(k)$, $$ { \# \{\g\, : \, p(\g )=n,\;\; |\int_{\g}f - \int_{\T^d}f d\mu |> \eps \} \over \pp (n)} \leq {1\over k}\eqno(2.42) $$ and the assertion follows by taking $k \to \infty$ and consequently $n\to \infty$. Q.E.D. \vskip 1cm {\bf 3. Periodic orbits living on prime lattices} \vskip 0.5cm We now turn to study the asymptotic properties of the probability measures $\mu_{\gamma}$ (cf. (2.24)) supported on {\sl single} periodic orbits $\gamma$'s of $A$ when $p(\gamma) \to\infty$. To this end we first remark the following fact. Consider a sequence $\{ \g_k\}$ of periodic orbits of $A:\T^d \to \T^d$ ordered with non-decreasing periods. Then $\{\mu_{\gamma_k}\}$ converges weakly to $\mu$ as $k\to \infty$ if and only if [1] $$ \int_{\gamma_k} e^{2\pi i} \rightarrow \int \, e^{2\pi i}\, d\mu \eqno(3.1) $$ for any $n\in \Z^d$. Thus, it suffices to consider integrals of characters. \vskip 0.2cm Let $A\in GL(d,\Z)$ with det$\,A=1$, we denote with $f(x)\in\Z[x]$ the corresponding characteristic polynomial. Given an integer $p$, we denote by $\L_p$ the invariant subset of $\T^d$ given by all the rational points with denominator $p$, i.e. $$ \L_p=\{x\in\T^d\,\,\vert p x\in\Z^d\,\,\}\eqno(3.2) $$ $\L_p$ is invariant under the action of $A$, so that any point in $\L_p$ is periodic with period $\leq p^d$. This means that $\L_p$ splits into periodic orbits (which in general may have different periods) of $A$. As it will become more clear in the sequel, to study the structure of periodic orbits living on $\L_p$ amounts to perform with a modulo $p$ arithmetic. In what follows we shall only consider lattices $\L_p$ for primes $p$ such that $f(x)$ completely splits over $\Z_p$, that is $f(x)=\prod_{j=1}^{d} (x-\lj)$, $\lj\in\Z_p$, and call such a prime {\it completely splitting}. As a consequence of Chebotarev's theorem [3], the set of such primes $p$ as positive density among all primes. One can actually give an estimate from below of such a density in terms of the order of the Galois group associated to the polynomial $f(x)$ [3]. Given a completely splitting prime $p$, we denote by $$ \Lambda =\{\lambda_1,\cdots,\lambda_d\}\eqno(3.3) $$ the set of the eigenvalues of $A$, which in general are not all distinct and satisfy $\lambda_1\lambda_2....\lambda_d=1$. In the case $d=2r$ and $A\in SL(r,\Z)$, i.e. $A$ symplectic, $\lj\in\Lambda$ implies $\lj^{-1}\in \Lambda$. For any $j=1,...,d$, we denote by $\t_j$ the order of $\lj$ in the multiplicative group $\Z_p^*$, that is $\lj^{\t_j}=1$ and $\lj^l\neq 1,\,\, 01$) such that $A^{\t(x)}x=x$. Clearly, if $x=\sum_{j\in \D} \aj\vj$ then $$ \t(x)=l.c.m.\{\, \t_j\,: \, j\in \D \}\eqno(3.8) $$ The set $\P$ of all {\it admissible periods} is then given by $$ {\P} =\{\, \t \, = \, l.c.m.\left( \t_{j_1},\t_{j_2},....,\t_{j_k}\,: \, (j_1,....,j_k)\subset (1,.....,d)\right)\, \}\eqno(3.9) $$ $\t_1$ is the smallest period, whereas $\t_{max}={\rm l.c.m.}\{\t_1,\dots,\t_d\} \leq (p-1)^{d}$ is the largest one among the admissible periods. Clearly, the natural ordering of $\P$ induces a filtration of the lattice $L_p$ in subsets of points with common period. \vskip 0.2cm {\bf Remark.} In the two dimensional case, it is known ([14], [4]) that all points in $\L_p\setminus \{0\}$ share the same period: $\t (x) = \t = (p-1)/m$ (for inert primes $p$, $\t (x) = \t = (p+1)/m$, [14]), so that $\P$ is made out of a single point. The number $m={(p-1)/ \t}$ corresponds exactly to the degeneracy of the eigenvalues of the associated quantum propagator [4]. \vskip 0.2cm In order to study the spatial distribution of a given periodic orbit, we are concerned in evaluating the average of an arbitrary (non trivial) character. Given $p$ as above, $x\in \Z_p^d$ and $n\in \Z^d$, we define $$ I_p(x,n)={{1}\over \t(x)}\sum_{s=0}^{\t(x)-1} \,e^{\fp}\eqno(3.10) $$ or, equivalently, $$ I_p(x,n)={{1}\over \t(x)}\sum_{s=0}^{\t(x)-1} \,e^{\fp\sum_{j=1}^{d}\beta_j \lambda_j^s}\eqno(3.11) $$ where $\beta_j=\aj\in\Z_p$. We shall reduce the above expression to (some combinations of) generalized sums over finite fields of the type: $$ \sum_{x\in \Z_p} \chi (f(x)) \psi(g(x)) $$ where $\chi$ is a non trivial multiplicative character of order $d|(p-1)$ of $\Z_p^*$ (i.e. $\chi^d$ is equal to the trivial character $\chi_0$), $\psi$ a non trivial additive character of $\Z_p$ and $f(x), g(x)\in \Z_p [x]$ are given algebraic functions, for instance polynomials, over $\Z_p$. The reference result is then the following: \vskip 0.2cm {\bf Theorem 3.1.} {\it Let $\chi$,$\psi$ be a multiplicative character $\neq {\chi}_0$ of order $d$ with $d\vert (p-1)$, and a non trivial additive character, respectively, of $\Z_p$. Let $f(x)\in \Z_p [x]$ admit $m$ distinct roots, and let $g(x)\in \Z_p [x]$ have degree $n$. Suppose that either $(d,deg(f))=(n,p)=1$, or, more generally, that the polynomials $y^d-f(x)$ and $z^p-z-g(x)$ are absolutely irreducible (i.e. irreducible over any finite algebraic extension of $\Z_p$). Then} $$ \vert \sum_{x\in\Z_p} \chi (f(x)) \psi(g(x))\vert\leq (m+n-1) p^{1/2} $$ {Proof.} See, e.g., [20] page 45. \vskip 0.2cm This result has been obtained by A. Weil as a consequence of the validity of the Riemann hypothesis for curves over finite fields and important extensions to cases where $f,g$ are rational functions over arbitrary finite fields has been achieved by P. Deligne [6]. {\bf Examples.} \item{1)} For $f(x)=g(x)=x$ we have the generalized {\it Gauss sum} $g(\psi ,\chi )$, and, by direct application of Theorem 3.1, $|g(\psi,\chi)|\leq \sqrt{p}$. \vskip 0.2cm \item{2)} Setting $f(x)=x^2-4ab$, $g(x)=x$, $\chi$ the (non trivial) character of order two (Legendre symbol) and $\psi =\exp{\left({2\pi ix\over p}\right)}$, we find the {\it Kloosterman sum} $$ Kl(p,a,b)= \sum_{x\in \Z_p}\left({x^2-4ab \over p}\right)\exp{ \fa x}= \sum_{x\in \Z_p^*}\exp{ {2\pi i \over p} (ax+bx^{-1})} $$ Again, from Theorem 3.1, one has the estimate $\vert Kl(p,a,b)\vert \leq 2\sqrt{p}$. \vskip 0.2cm The next result is useful in reducing to the previous case some sums over cyclic subgroups. \vskip 0.2cm {\bf Lemma 3.1.} {\it $\forall \lambda\in\Z_p^*$, denote $\Lambda_{\lambda}=<\lambda>$ the cyclic subgroup generated by $\lambda$. Let $\#\Lambda_{\lambda}={p-1\over m}=\t$. If $f:\Z_p\longrightarrow\C$ is any complex valued function, then $$ \sum_{s=0}^{\t-1} f({\lambda}^s) ={1\over m}\sum_{j=0}^{m-1} \sum_{x\in{\Z_p}^*} {\chi}_j(x)f(x) $$ where $\{{\chi}_0,\cdots,{\chi}_{m-1}\}$ is a set of multiplicative characters of order $m$.} \vskip 0.2cm {\it Proof.} See the Appendix in [4]. \vskip 0.5cm {\sl 3.1. Ideal lines and ideal orbits.} \vskip 0.2cm As a first step toward the understanding of the distribution of a generic periodic orbit, we now consider those orbits which lay entirely on invariant sublattices of the form $\G_j=\Z_p\vj=\{u\vj\,\,\vert\,u\in\Z_p\} \subset \L_p$ for $j=1,\dots ,d$, where $v_j\in \Z_p^d$ is an eigenvector of $A$. In analogy with the two dimensional case, we call $\G_j$ and $\gamma \subset \G_j$ {\it ideal line} and {\it ideal orbits}, respectively ([2],[4]). Notice that any ideal line is almost completely equidistributed, in the following sense: for any $n\in \Z^d$ such that $\not= 0\, {\rm mod}\, p$, we have $$ \sum_{x\in \G_j} e^{\fp } = \sum_{u\in \Z_p} e^{\fp u}=0\eqno(3.12) $$ Define $$ K_j \equiv K_j(p) = \{n\in \Z^d\setminus \{0\}\, : \, =0\, {\rm mod}\, p\} $$ for $j=1,\dots ,d$, and $$ k_j = \inf_{n\in K_j}|n| $$ where $|n|$ denotes the euclidean length. We then have the following, \vskip 0.2cm {\bf Lemma 3.2.} {\it Let $A$ be irreducible. $$ k(p)=\inf_j k_j \rightarrow \infty \quad\hbox{as}\quad p \to \infty $$ where the limit is taken over an arbitrary increasing sequence of completely splitting primes.} \vskip 0.2cm {\it Proof.} Assume that there exists an infinite sequence of primes $\{p_l\}$ such that $k(p_l)$ remains bounded as $l\to \infty$. By eventually restricting to a subsequence, we can assume that there exist a fixed $n\in\Z^d$ such that for each $l$ there exist an eigenvector $v_{j}\in \Z^d_{p_l}$ such that $=0\,\, {\rm mod}\, p_l$. Given $l\in \Z_+$, let us denote by $M_l\subset \Z_{p_l}^d$ the $\Z_{p_l}$-module generated by $\{B^sn\}_{s\in \Z}$, where $B=A^t$. $M_l$ is a non trivial submodule of dimension $1\leq r_l\leq d-1$. In particular, there exist a submodule $N_l\supset M_l$ with the same dimension $r_l$ and another non trivial submodule $R_l$ such that $Z^d_{p_l}=N_l\bigoplus R_l$. Since this decomposition holds for an infinite number of primes and since $n$ has been chosen independent of $l$, a simple argument shows in fact the existence of a non trivial $B$-invariant $\Z$-module $M\subset \Z^d$, in contradiction with the irreducibility hypothesis. Q.E.D. \vskip 0.2cm {\bf Proposition 3.1.} {\it For any completely splitting prime $p$ and for any $x\in \G_j$ ($x\neq \{0\}$) } $$ \left| I_p(x,n)\right| = \cases{\leq {m_j\over\sqrt{p}}, &if $n\notin K_j$; \cr 1, &otherwise \cr } $$ \vskip 0.2cm {\bf Remark.} \item{1)} If $A$ is irreducible, then Lemma 3.2 and the above result implies that any sequence $\gamma_k\subset \L_{p_k}$ of periodic ideal orbits, where $\{p_k\}$ is any increasing sequence of completely splitting primes such that $m_k/\sqrt{p_k}\to 0$ as $k\to \infty$, becomes equidistributed according to $\mu$ as $k\to \infty$. \item {2)} If $A$ is reducible, e.g. $A=\left( \matrix{ B & 0 \cr 0 & C \cr}\right)$, where $B\in GL(r,\Z)$, $C\in GL(s,\Z)$ with $r+s=d$, then there is an invariant decomposition $\T^d=\T^r\times\T^s$. In this case, one can also find families of periodic orbits which are confined to invariant tori of dimension $< d$. The same technique can then be used to describe the equidistribution of the such periodic orbits with respect the Haar measure supported on these invariant tori. \vskip 0.2cm {\it Proof.} Let $x\in \G_j$, i.e. $x=\alpha v_j$, $\alpha \in\Z_p^*$. If $n\in K_j$ then, clearly, $I_p(x,n)=1$. On the other hand, if $\beta_j=\alpha \neq 0\, {\rm mod}\, p$, using Lemma 3.1 we find $$\eqalign{ &I_p(x,n)={1\over \t_j}\sum_{s=0}^{\t_j-1} e^{{2\pi i\over p}\beta_j\lambda_j^s}= {1\over \t_j m_j}\sum_{x\in\Z_{p}^*}e^{{2\pi i\over p}\beta_j x^{m_j}}\cr &= {1\over \t_j\, m_j}\sum_{k=0}^{m_j-1}\sum_{x\in\Z_{p}^*}\chi_k(x)e^{{2\pi i\over p}\beta_j x}= {1\over \t_j\, m_j}\sum_{k=0}^{m_j-1} g(\psi_{\beta_j},\chi_j)\cr } \eqno(3.16) $$ where we have used the notation $\psi_a(x) = \exp{\left(\di {2\pi iax \over p}\right)}$, $a\in \Z_p$. The first statement then follows from the estimate on the absolute value of an arbitrary Gauss sum. Q.E.D. \vskip 0.5cm {\sl 3.2. Generic orbits.} \vskip 0.2cm Having proved the equidistribution of ideal orbits, we now turn to the generic situation. The simplest cases to treat are those corresponding to a {\it completely non resonant} situation and/or a {\it completely resonant} situation. Let $x\in \L_p$ and write again $x=\sum_{j\in \D} \alpha_j v_j$, where $\D \subseteq \{1,\dots ,d\}$ and $0 \not= \alpha_j\in Z_p$ for any $j\in \D$. \vskip 0.2cm {\bf Definition 3.1.} {\it A point $x\in\L_p$ as above is called {\rm completely non resonant} if $(\t_i,\t_j)=1$ for all $i,j\i \D$, i.e. all the cyclic subgroups generated by $\lambda_j$'s with $j\in \D$ are distinct with trivial intersections.} \vskip 0.2cm {\bf Proposition 3.2.} {\it Let $x\in\L_p$ be completely non resonant and $n\in \Z^d\setminus \cap_{j\in \D}K_j$, then $$ \vert I_p(x,n)\vert\leq {1\over \sqrt{p}}\min_{j\in \D'}m_j $$ where $\D' \subseteq \D$ is such that $\beta_j=\alpha_j \not= 0$ for $j\in \D'$}. \vskip 0.2cm {\it Proof.} Set $k=\# \D$. Let $\{\lambda_1, \dots , \lambda_k\}$ be the eigenvalues corresponding to $v_j$, $j\in \D$, and let $\{\t_1,\dots ,\t_k\}$, $\{m_1, \dots ,m_k\}$ be the corresponding periods and degeneracies, respectively. We have $\t(x)=\t=\t_1\t_2...\t_k$. Set moreover $\t_0=\t_1 \t_2 \dots\t_{k-1}$. Note that $(\t_0,\t_k)=1$ and therefore $\{\lambda_k^{l\t_0}\}_{l=0}^{\t_k-1}=\{\lambda_k^s\}_{s=0}^{\t_k-1}$. We then get $$\eqalign{ I_p(x,n)&=\ti\st e^{\fp[\beta_1\lambda_1^s+\dots\bk\lk^s]}\cr &=\ti\sum_{s=0}^{\t_0-1}\sum_{l=0}^{\t_k-1} e^{\fp[\beta_1\lambda_1^{(s+l\t_0)}+\dots+\bk\lk^{(s+l\t_0)}]}\cr &=\ti\sum_{s=0}^{\t_0-1}e^{\fp(\beta_1\lambda_1^s+\dots+\beta_{k-1}\lambda_{k-1}^s)}\, \sum_{l=0}^{\t_k-1} e^{\fp\bk\lk^{s}\lk^{l\t_0}} \cr &=\ti\sum_{s=0}^{\t_0-1}e^{\fp(\beta_1\lambda_1^s+\dots+\beta_{k-1}\lambda_{k-1}^s)}\, {{1}\over{m_k}}\sum_{l=0}^{m_k-1} g(\psi_{\bk\lk^s},\chi_l) \cr } \eqno(3.17) $$ where, again, $g(\psi_a,\chi)=\sum_{x\in\Z_p^*} \chi(x) e^{\fp ax}$ denotes the Gauss sum. If now $\beta_k \not= 0$ we obtain $$ \vert I_p(x,n)\vert\leq {{\t_0}\over{\t}}\sqrt{p}\leq {{m_k}\over{\sqrt{p}}}\eqno(3.18) $$ which immediately yields the result. Q.E.D. \vskip 0.2cm Let us now consider the opposite case. \vskip 0.2cm {\bf Definition 3.2.} {\it A point $x\in\L_p$ as above is called {\rm completely resonant} if, for all $i,j\i \D$, $<\lambda_i>= <\lambda_j>=<\lambda>$}. \vskip 0.2cm This definition implies that $\t(x)=\t=(p-1)/m$. Moreover, for any $i,j\i \D$ there exist two integers $s_i$, $s_j$ such that $\lambda_i^{s_i}=\lambda_j^{s_j}$. Set again $k=\# \D$. For any $j\in \D$ there exist $\eta_j\in\{-{{\t-1}\over{2}},...,-1,0,1,....,{{\t-1}\over{2}}\}$ such that $\lambda_j=\lambda^{\eta_j}$. We may choose $\lambda$ and the $\eta_j$ in such a way that $\eta =2\, \max_{j\in \D} \vert\eta_j\vert$ is as small as possible. \vskip 0.2cm {\bf Proposition 3.3.} {\it Let $x\in\L_p$ be completely resonant and $n\in \Z^d\setminus \cap_{j\in \D}K_j$,} then $$ \vert I_p(x,n)\vert\leq {m\over \sqrt{p}}\, \eta $$ \vskip 0.2cm {\it Proof.} We have $$\eqalign{ I_p(x,n)&={1\over\t}\sum_{s=0}^{\t-1} e^{\fp[\beta_1\lambda_1^{m\eta_1}+\dots \beta_k\lambda_k^{m\eta_k}]}\cr &={1\over \t m}\sum_{x\in\Z_p^*}\, e^{\fp[\beta_1x^{m\eta_1}+...\beta_kx^{m\eta_k}]}\cr & ={1\over \t \, m}\sum_{j=0}^{m-1}\sum_{x\in\Z_p^*}\,\chi_j(x)\,e^{\fp[\beta_1x^{\eta_1}+\dots+\beta_kx^{ \eta_k}]}\cr } \eqno(3.19) $$ and the result follows by direct application of Theorem 3.1. Q.E.D. \vskip 0.2cm {\bf Remark.} In the case $d=2$ one has $\Lambda =\{\lambda, \lambda^{-1}\}$. Moreover, since any point in $\L_p\setminus \{0\}$ is completely resonant (with $\t(x) = (p-1)/m$) one finds the expression $$\eqalign{ I_p(x,n)&={1\over\t}\sum_{s=0}^{\t-1} e^{\fp[\beta_1\lambda^s+ \beta_2\lambda^{-s}]} \cr &={1\over \t m}\sum_{x\in\Z_p^*}\, e^{\fp[\beta_1x^{m}+\beta_2x^{-m}]}\cr & ={1\over \t \, m}\sum_{j=0}^{m-1}\sum_{x\in\Z_p^*}\,\chi_j(x)\, e^{\fp[\beta_1x+\beta_2x^{-1}]}\cr } \eqno(3.20) $$ and the estimate on the Kloosterman sums can be used to give (see Example 2 above) $$ \left| I_p(x,n)\right| \leq {2 m\over \sqrt{p}} $$ so that any sequence $\gamma_k\subset \L_{p_k}$ of periodic orbits, where $\{p_k\}$ is any increasing sequence of splitting primes such that $m(p_k)/\sqrt{p_k}\to 0$ as $k\to \infty$, becomes equidistributed according to $\mu$ as $k\to \infty$ [4],[5]. \vskip 0.2cm To conclude, we briefly discuss the more general case, corresponding to {\it partially resonant} situations. Given an arbitrary point $x\in \L_p\setminus \{0\}$ let us write it again in the form $x=\sum_{j\in \D} \alpha_j v_j$. For any fixed $\lambda =\lambda_k$, $k\in \D$, we decompose $\D$ as follows $$ \D = \D_1 \cup \D_2 \cup \D_3 $$ where, if $\t=\t_k$ denotes the order of $<\lambda>$: \item{1)} for any $j\in \D_1$: $(\t,\t_j)=1$; \item{2)} for any $j\in \D_2$: $<\lambda_j>=<\lambda>$, ($\t_j=\t$); \item{3)} ({\it partially resonant}) for any $j\in \D_3$: $<\lambda_j>\not=<\lambda>$ and $<\lambda_j>\cap<\lambda>\not= \emptyset$. In particular, if $j\in \D_3$ there exist $s_1=s_1(j)$ and $s_2=s_2(j)$ such that $\lambda^{s_1} = \lambda_j^{s_2}$. \vskip 0.2cm Now, if one can choose $\lambda$ in such a way that the corresponding set $\D_3$ is empty, then the integral $I_p(n,x)$ can be estimated by suitably combining the previous two cases. More precisely, one can first isolate the non resonant eigenvalues as in (3.17) and then proceed as in (3.19). Finally, the case where the presence of partial resonances cannot be avoided seems to be more involved. Nevertheless the same sort of techniques might be used to reduce $I_p(n,x)$ to suitable exponential sums over $\Z_p$ and thus produce, at least in principle, explicit estimates. \vfill \eject {\bf References.} \vskip 1cm \item{ [1]} Billingsley P 1968 {\it Convergence of Probability Measures} (New York: Wiley) \vskip 0.2cm \item{[2]} Bartuccelli F and Vivaldi F 1989 Ideal Orbits of Toral Automorphisms {\it Physica D} {\bf 39} 194 \vskip 0.2cm \item{[3]} Cassels J W S and Fr\"ohlich A eds 1990 {\it Algebraic Number Theory} ({\it Academic Press}) \vskip 0.2cm \item{[4]} Degli Esposti M, Graffi S and Isola S 1994 Classical limit of the quantized hyperbolic toral automorphism {\it Commun. Math. Phys.} to appear \vskip 0.2cm \item{[5]} Degli Esposti M, Graffi S and Isola S 1994 {\it Quantum vs Classical equidistribution results. An overview} ({\it Springer Lectures Notes in Math.}) (Graffi S ed) to be published \vskip 0.2cm \item{[6]} Deligne P 1974 La conjecture de Weil I {\it Publ. Math. I.H.E.S.} {\bf 48} 273-308 \vskip 0.2cm \item{[7]} Isola S 1990 $\zeta-$function and distribution of periodic orbits of toral automorphisms {\it Europhysics Letters} {\bf 11} 517-522 \vskip 0.2cm \item{[8]} Keating J 1991 Asymptotic properties of the periodic orbits of the cat maps {\it Nonlinearity} {\bf 4} 277-307 \vskip 0.2cm \item{[9]} Keating J 1991 The cat maps: quantum mechanics and classical motion {\it Nonlinearity} {\bf 4} 309-341 \vskip 0.2cm \item{[10]} Kifer Y 1994 Large deviations, averaging and periodic orbits of dynamical systems {\it Comm. Math. Phys.} {\bf 162} 33-46 \vskip 0.2cm \item{[11]} Korevaar J 1982 On Newman's quick way to the prime number theorem {\it Mathematical Intelligencer} {\bf 4} 108-115 \vskip 0.2cm \item{[12]} Lalley S P 1987 {\sl Distribution of periodic orbits of symbolic and Axiom A flows} Adv. in Appl. Math. {\bf 8} 154-193 \vskip 0.2cm \item{[13]} Ma\~ne R 1987 {\it Ergodic Theory and Differentiable Dynamics} (New York: Springer) \vskip 0.2cm \item {[14]} Percival I and Vivaldi F 1987 Arithmetical properties of strongly chaotic motions {\it Physica D} {\bf 25} 105-130 \vskip 0.2cm \item{[15]} Parry W and Pollicott M 1983 An analogue of the prime number theorem for closed orbits of Axiom A flows {\it Annals of Math.} {\bf 118} 573-591 \vskip 0.2cm \item{[16]} Parry W and Pollicott M 1990 Zeta functions and the periodic orbit structure of hyperbolic dynamics {\it Ast\'erisque} {\bf 187-188} \vskip 0.2cm \item{[17]} Ram Murthy M 1988 Artin's Conjecture for Primitive Roots {\it Mathematical Intelligencer} {\bf 10} 59-70 \vskip 0.2cm \item{[18]} Ruelle D 1976 Generalized zeta-functions for expanding maps and Anosov flows {\it Inventiones Math.} {\bf 34} 231-242 \vskip 0.2cm \item{[19]} Rugh H H 1994 Generalized Fredholm determinants and Selberg zeta functions for Axiom A dynamical system {\it preprint IHES/P/94/7} \vskip 0.2cm \item{[20]} Schmidt W 1976 {\it Equations over finite fields. An elementary approach.} ({\it Springer Lecture Notes in Math.} {\bf 536}) \vskip 0.2cm \item{[21]} Smale S 1967 Differentiable dynamical systems {\it Bull. Amer. Math. Soc.} {\bf 73} 747-817 \vskip 0.2cm \item{[22]} Walters P 1982 {\it An Introduction to Ergodic Theory} (New York: Springer) \vskip 0.2cm \end