Quantum Ergodicity for Point Scatterers on Arithmetic Tori
aa r X i v : . [ m a t h . A P ] M a r QUANTUM ERGODICITY FOR POINT SCATTERERS ONARITHMETIC TORI
P ¨AR KURLBERG AND HENRIK UEBERSCH ¨AR
Abstract.
We prove an analogue of Shnirelman, Zelditch and Colin de Verdi`e-re’s Quantum Ergodicity Theorems in a case where there is no underlyingclassical ergodicity. The system we consider is the Laplacian with a deltapotential on the square torus. There are two types of wave functions: oldeigenfunctions of the Laplacian, which are not affected by the scatterer, andnew eigenfunctions which have a logarithmic singularity at the position of thescatterer. We prove that a full density subsequence of the new eigenfunctions equidistribute in phase space . Our estimates are uniform with respect to thecoupling parameter, in particular the equidistribution holds for both the weakand strong coupling quantizations of the point scatterer. Introduction
The point scatterer, namely the Laplacian with a delta potential, on a two-dimensional flat manifold is a popular model in the study of the transition betweenchaos and integrability in quantum systems. In 1990 Seba [14] considered this op-erator on a rectangle with irrational aspect ratio and Dirichlet boundary conditionsand argued that the spectrum and eigenfunctions of the point scatterer display fea-tures such as level repulsion and a Gaussian value distribution, both of which arepresent in quantum systems with chaotic classical dynamics (cf. [4] and [2]), suchas the quantization of the geodesic flow on hyperbolic manifolds or the flow in theSinai billiard. In fact the point scatterer can be understood as a limit of the Sinaibilliard where the radius shrinks to zero faster than the semiclassical wavelength.The subject of this paper is a point scatterer on a flat torus. It has two typesof eigenfunctions: first there are old eigenfunctions of the Laplacian, namely thosewhich vanish at the position of the scatterer; the nonzero eigenvalues remain thesame, though with multiplicities reduced by 1. Secondly, there are new eigenfunc-tions which diverge logarithmically near the position of the scatterer; the corre-sponding eigenvalues have multiplicity 1 and interlace with the old Laplace eigen-values.We shall only be concerned with the set of new eigenfunctions, i.e., the ones whichare affected by the scatterer. In [11] it was proved that a full density subsequence of the new eigenfunctions equidistribute in position space in the special case of asquare torus. We extend the results of [11] and prove that a full density subsequenceof these eigenfunctions in fact equidistribute in phase space — we thus establishan analogue of Shnirelman, Zelditch and Colin de Verdi`ere’s Quantum Ergodicity Date : September 10, 2018.P.K. was partially supported by grants from the G¨oran Gustafsson Foundation and the SwedishResearch Council. See Section 3 for a precise definition of a “full density subsequence”.
Theorem in a case where there is no underlying chaotic dynamics and no classicalergodicity.An analogue of this result for a cubic 3D torus was recently obtained by N.Yesha [18]. The situation for a square torus is very different from irrational tori,where the eigenfunctions are expected to localise in phase space on a finite numberof momentum vectors [7, 1]. In the case where the aspect ratio is diophantine thiscan be proven rigorously for a full density subsequence of new eigenfunctions [8].1.1.
Spectrum of the point scatterer.
The formal operator − ∆ + αδ x , α ∈ R is realized using von Neumann’s theory of self-adjoint extensions. We simply statethe most important facts in this section in order to formulate the results of thispaper. For a more detailed discussion of the self-adjoint realization of the pointscatterer we refer the reader to the introduction and appendix of the paper [11].Let T = R / π Z . We consider the restriction of the positive Laplacian − ∆ tothe domain D = C ∞ c ( T \ { x } )of functions which vanish near the position of the scatterer: H = − ∆ | D The operator H is symmetric, but fails to be self-adjoint, in fact H has deficiencyindices (1 , H ϕ , ϕ ∈ ( − π, π ], which are restrictions of theadjoint H ∗ to the domain of functions f ∈ Dom( H ∗ ) which satisfy the asymptotic f ( x ) = C (cid:18) cos (cid:16) ϕ (cid:17) log | x − x | π + sin (cid:16) ϕ (cid:17)(cid:19) + o (1) , x → x for some constant C ∈ C . The case ϕ = π corresponds to α = 0. In this paper wewill study the operators H ϕ , ϕ ∈ ( − π, π ).The spectrum of the operator H ϕ consists of two parts: “old” and“new” eigen-values. Since H ϕ is a self-adjoint realization of a rank one perturbation of theLaplacian, the effect is that each nonzero old Laplace eigenvalue appears, withmultiplicity reduced by 1, in the spectrum of H ϕ . Further, each old Laplace eigen-value gives rise to a new eigenvalue with multiplicity 1. In fact, these new eigen-values interlace with the multiplicity one sequence associated with the old Laplaceeigenvalues.There are two types of eigenfunctions of H ϕ associated with the two parts of thespectrum:(A) “Old” eigenfunctions which vanish at x and therefore are not affected bythe scatterer. These are simply eigenfunctions of the unperturbed Lapla-cian.(B) “New” eigenfunctions which feature a logarithmic singularity at x ; in factthey are given by Green’s functions G λ = (∆ + λ ) − δ x .We will study how eigenfunctions of type (B) are distributed in phase space asthe eigenvalue tends to infinity. Denote by S the set of distinct eigenvalues of theLaplacian on T , namely integers which can be represented as a sum of two squares: S := { n ∈ Z : n = x + y | x, y ∈ Z } UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 3
28 30 32 34 36 38 40 - - - Figure 1.
The picture shows a plot of the l. h. s. of equation(1.1) as a function of λ . The zeroes are the new eigenvalues corre-sponding to the self-adjoint extension with parameter ϕ = 0.For given n ∈ S denote its multiplicity by r ( n ) := X n = | ξ | ξ ∈ Z , i.e. the number of ways n can be written as a sum of two squares.The eigenvalues of type (B) are solutions to the equation(1.1) X n ∈ S r ( n ) (cid:18) n − λ − nn + 1 (cid:19) = c tan (cid:16) ϕ (cid:17) (see Figure 1 for a plot of the l. h. s.) where(1.2) c = X n ∈ S r ( n ) n + 1 . As mentioned earlier, they interlace with the distinct Laplace eigenvalues S = { < < < < < < · · · } as follows(1.3) λ ,ϕ < < λ ,ϕ < < λ ,ϕ < < λ ,ϕ < < λ ,ϕ < < λ ,ϕ < < · · · where the new eigenvalue associated with n ∈ S is denoted by λ n,ϕ ; note that λ n,ϕ < n .1.2. Strong coupling.
In the physics literature equation (1.1) is referred to as a“weak coupling” quantization. In fact, (cf. [12]) the new eigenvalues λ m,ϕ “clump”with the Laplace eigenvalues m ∈ S in the sense that for a full density subsequenceof S, 0 < m − λ m,ϕ ≪ m ) − o (1) . P ¨AR KURLBERG AND HENRIK UEBERSCH¨AR
In particular the eigenvalue spacing distribution of the point scatterer coincideswith that of the Laplacian, and the effect of the scatterer on the spectrum is quiteweak in this quantization. (In a sense it corresponds to letting α → λ → ∞ .)Shigehara [15] and later Bogomolny, Gerland and Schmit [3], with the intent offinding a model exhibiting level repulsion, considered another quantization some-times referred to as a “strong coupling” quantization. There are various ways toarrive at this quantization condition from equation (1.1). For example, one maytruncate the summation outside an energy window of size O ( λ δ ) where δ > X n ∈ S | n − n + ( λ ) |≤ n + ( λ ) δ r ( n ) (cid:18) n − λ − nn + 1 (cid:19) = c tan (cid:16) ϕ (cid:17) , where n + ( λ ) denotes the smallest element of S which is larger than λ . (With λ denoting such a solution, the corresponding “new” eigenfunction is defined as acertain Green’s function G λ , cf. Section 1.3.) A summation by parts argument (seefor instance Lemma 3.1 in [16]) shows that X n ∈ S | n − n + ( λ ) | >n + ( λ ) δ r ( n ) (cid:18) n − λ − nn + 1 (cid:19) = − π log λ + O δ (1)and hence the truncation given by (1.4) is equivalent to a logarithmic renormalisa-tion of the r. h. s. of (1.1), namely, as λ → ∞ ,(1.5) X n ∈ S r ( n ) (cid:18) n − λ − nn + 1 (cid:19) = − π (1 + o δ (1)) log λ = c tan (cid:16) ϕ λ (cid:17) , if we allow ϕ λ to depend on λ appropriately, and where the o δ (1) error term dependson the exponent δ . Since the error term depends on δ we note that there is no uniquechoice of strong coupling quantization; the key point is matching the leading orderlogarithmic term. The renormalization in (1.5) can be viewed as letting a boundarycondition vary with the energy. Consequently D ϕ λ , the domain of the operator H ϕ λ is varying; this setting is reminiscent of problems in semiclassical analysis whereboundary conditions are allowed to depend on the semiclassical parameter ~ . Remark 1.
In the weak coupling quantization the lowest new eigenvalue is alwaysnegative, but for the strong coupling quantization the lowest new eigenvalue maybe either positive or negative. In the case of a positive lowest new eigenvalue,this eigenvalue would be denoted λ to keep our notation consistent, in particularensuring that λ n < n for n ∈ S always holds. We remark that in the statement of our main result, Theorem 1.1, the sequenceΛ = { λ n } will denote any increasing sequence of numbers which interlace with S . In particular, it applies to the eigenvalues of the weak, as well as the strong,coupling quantizations.1.3. Semiclassical Measures.
Let a ∈ C ∞ ( S ∗ T ). Denote by Op( a ) a zero-orderpseudo-differential operator associated with a (see subsection 2.1 for more details.)Let g λ = G λ / k G λ k , λ / ∈ S , where we recall that S denotes the set of Laplaceeigenvalues and that G λ = (∆ + λ ) − δ x . We are interested in weak limits of UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 5 measures dµ λ defined by the identity(1.6) h Op( a ) g λ , g λ i = Z S ∗ T adµ λ . Main Result.
The following theorem holds generally for the L -normalizedGreen’s functions g λ . It states that the measures dµ λ defined by (1.6) convergeweakly to Liouville measure as λ → ∞ along a full density subsequence of any increasing sequence Λ which interlaces with S . (Recall that S denotes set of unper-turbed Laplace eigenvalues, namely the set of integers which can be represented asa sum of two squares.) Theorem 1.1.
Let Λ be an increasing sequence which interlaces with S . For m ∈ S ,denote by λ m the largest element of Λ which is smaller than m ∈ S . There existsa full density subsequence S ′ ⊂ S , that does not depend on Λ , such that for all a ∈ C ∞ ( S ∗ T ) , (1.7) lim m →∞ m ∈ S ′ h Op( a ) g λ m , g λ m i = Z S ∗ T a ( x, ϕ ) dx dϕ vol( S ∗ T ) . As already noted, the theorem holds in particular for the new eigenvalues of theweak and strong coupling quantizations of a point scatterer. Hence we have thefollowing corollary of Theorem 1.1.
Corollary 1.2.
Quantum Ergodicity holds for the new eigenfunctions of weakly,as well as strongly, coupled point scatterers on T . Remark 2.
Recall that the new eigenvalues in the strong coupling limit are givenby the set of solutions { λ m } m to (1.5) (or alternatively, solutions to (1.4) ), withcorresponding new eigenfunctions given by the Green’s functions G λ m . Althoughthese Green’s functions are eigenfunctions of different operators { H ϕ λm } m (in fact,the domains of the operators change), it is natural to say that quantum ergodicityholds in the strong coupling limit if a full density subset of the collection of neweigenfunctions equidistribute. We further note that the counting function, or Weyl’s law, for the set of neweigenvalues (cf. Theorem 3.3) satisfies |{ n : λ n ≤ x }| ≪ x √ log x = o ( x ) , while the counting function for the full set of eigenvalues (new and old, with mul-tiplicity) is the same as for the unperturbed Laplacian, hence ≫ x . Consequently,the sequence of new eigenvalues is of density zero within the full spectrum, andthe approach of proving Quantum Ergodicity for the set of new eigenfunctions bycomputing first or second moments of matrix coefficients (e.g., see [19]) with respectto the full set of eigenfunctions seems unlikely to succeed.1.5. Acknowledgements.
We would like to thank Zeev Rudnick and StephaneNonnenmacher for valuable discussions about this problem and for many helpfulremarks which have led to the improvement of this paper.The authors are also very grateful to the referee for a careful reading of the paperand for many comments and suggestions that improved the exposition.
P ¨AR KURLBERG AND HENRIK UEBERSCH¨AR The matrix elements
Quantization of phase space observables.
Consider a classical symbol a ∈ C ∞ ( S ∗ T ), where S ∗ T ≃ T × S denotes the unit cotangent bundle of T .We may expand a in the Fourier series(note that the Fourier coefficients decay rapidly since a is smooth)(2.1) a ( x, φ ) = X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) e i h ζ,x i +i kφ . We choose a complex realization of the unit cotangent bundle S ∗ T and parametrisethe unit circle S at position x ∈ T by the complex exponential map ϕ e i ϕ .We now want to associate with a a pseudodifferential operator Op( a ) : C ∞ ( T ) → C ∞ ( T ). We choose the following symbol (we associate with ξ = ( ξ , ξ ) the com-plex number ˜ ξ := ξ + i ξ and note that e i k arg ˜ ξ = ( ˜ ξ/ | ˜ ξ | ) k )(2.2) σ a ( x, ξ ) = P ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) (cid:16) ˜ ξ | ˜ ξ | (cid:17) k e i h ζ,x i , ξ = 0 P ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) e i h ζ,x i , ξ = 0 . Claim: The symbol σ a , as defined above, belongs to the class of toroidal symbols S , ( T × Z ) as defined in [13], Part II, Section 4.1.2, Defn. 4.1.7, p. 344. To seethis, define the difference operators∆ ξ j f ( ξ ) = f ( ξ + e j ) − f ( ξ ) , where e = (1 , e = (0 , f : R → R a smooth function,∆ β ξ ∆ β ξ f ( ξ ) = ∂ β ξ ∂ β ξ f ( ξ ) (cid:12)(cid:12)(cid:12) ξ = ξ ′ , for some ξ ′ = ξ + ( β ′ , β ′ ) with ( β ′ , β ′ ) ∈ [0 , β ] × [0 , β ]. With f k ( ξ ) denoting thereal, or imaginary, part of ( ˜ ξ/ | ˜ ξ | ) k , a quick calculation then gives that for integers α , α , β , β ≥ (cid:12)(cid:12)(cid:12) ∂ α x ∂ α x ∆ β ξ ∆ β ξ f k ( ξ ) e i h ζ,x i (cid:12)(cid:12)(cid:12) ≪ α ,α ,β ,β k β + β | ζ | α | ζ | α (1 + | ξ | ) − β − β This bound, together with the rapid decay of Fourier coefficients of a , implies that | ∂ α x ∂ α x ∆ β ξ ∆ β ξ σ a ( x, ξ ) | ≤ C a,α ,α ,β ,β (1 + | ξ | ) − β − β , thus confirming the claim.The action of the pseudodifferential operator Op( a ) is then defined by multi-plication on the Fourier side, analogously to Defn. 4.1.9 in [13]. Therefore, wehave (Op( a ) f )( x ) = X ξ ∈ Z σ a ( x, ξ ) ˆ f ( ξ ) e i h ξ,x i = X ξ ∈ Z \{ } X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) ˜ ξ | ˜ ξ | ! k ˆ f ( ξ ) e i h ξ + ζ,x i + X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) ˆ f (0) e i h ζ,x i (2.3) UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 7
We can now read off the action of Op( a ) on the Fourier coefficients:(2.4) \ (Op( a ) f )( ξ ) = X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) ˜ ξ − ˜ ζ | ˜ ξ − ˜ ζ | ! k ˆ f ( ξ − ζ ) , ξ = ζ (recall that ˜ ξ := ξ + i ξ and that the Fourier coefficients ˆ a ( ζ, k ) decay rapidly) and(2.5) \ (Op( a ) f )( ζ ) = X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) ˆ f (0) . In terms of the Fourier coefficients the matrix elements of Op( a ) can be written as(2.6) h Op( a ) f, f i = X ξ ∈ Z \ (Op( a ) f )( ξ ) ˆ f ( ξ )In particular, for the observable e ζ,k ( x, φ ) = e i h ζ,x i +i kφ , we have(2.7) h Op( e ζ,k ) f, f i = X ξ ∈ Z \{ ζ } ˜ ξ − ˜ ζ | ˜ ξ − ˜ ζ | ! k ˆ f ( ξ ) ˆ f ( ξ − ζ ) + ˆ f ( ζ ) ˆ f (0) . Mixed modes. If ζ = 0, we have the bound(2.8) | h Op( e ζ,k ) f, f i | ≤ X ξ ∈ Z | ˆ f ( ξ ) || ˆ f ( ξ − ζ ) | . In the case f = g λ = G λ / k G λ k we have the L -expansion G λ ( x, x ) = 14 π X ξ ∈ Z c ( ξ ) e i h x,ξ i where c ( ξ ) = | ξ | − λ . We obtain(2.9) | h Op( e ζ,k ) g λ , g λ i | ≤ P ξ ∈ Z | c ( ξ ) || c ( ξ − ζ ) | P ξ ∈ Z | c ( ξ ) | . In [11] it was proved that there exists a full density subsequence S ′ ⊂ S such thatfor any nonzero lattice vector ζ ∈ Z the matrix elements of Op( e ζ,k ) vanish as n → ∞ along S ′ . The following result was obtained. Theorem 2.1. (Rudnick-U., 2012)
Let Λ be an increasing sequence which inter-laces with S . Denote by λ m the largest element of Λ which is smaller than m ∈ S .There exists a subsequence S ′ ⊂ S of full density such that for any ζ ∈ Z , ζ = 0 , k ∈ Z (2.10) lim n →∞ n ∈ S ′ h Op( e ζ,k ) g λ n , g λ n i = 0 . Remark 3.
The above result is only stated for the weak coupling quantization in [11] . However, the proof in fact works for any interlacing sequence, in particularfor the strong coupling quantization.To see this, we briefly recall the key steps of the proof in [11] . The first step isto show that the Green’s functions G λ can be approximated by truncated Green’sfunctions G λ,L , where L = λ δ for a specific choice of δ > and the truncationdrops all lattice vectors ξ outside an annulus A ( λ, L ) = { ξ ∈ Z | || ξ | − λ | ≤ L } . The subsequence S ′ ⊂ S is chosen in such a way to ensure that the latticepoints inside the annulus A ( λ, L ) are sufficiently well-spaced; this then implies that P ¨AR KURLBERG AND HENRIK UEBERSCH¨AR c ( ξ − ζ ) ≪ /L for ζ ∈ Z fixed and ξ ∈ A ( λ, L ) . A second condition requires that theneighboring Laplace eigenvalues are not too far apart in order for the lower bound k G λ k ≫ /λ o (1) to hold. These two key properties only depend on the arithmeticproperties of the neighboring Laplace eigenvalues, and not on the location of thenew eigenvalue itself. Pure momentum modes.
Let us consider the case ζ = 0. We rewrite thematrix elements as (cf. eq. (2.4)) h Op( e ,k ) g λ , g λ i = P ξ ∈ Z \{ } ( ˜ ξ/ | ˜ ξ | ) k | c ( ξ ) | + | c (0) | P ξ ∈ Z | c ( ξ ) | = λ + P n ∈ S \{ } w k ( n )( n − λ ) λ + P n ∈ S \{ } r ( n )( n − λ ) (2.11)where w k ( n ), for n ∈ S , is a certain exponential sum defined as follows: withΛ n := { z = x + iy ∈ Z [ i ] : | z | = n } , denoting the set Gaussian integers of norm n (we can interpret these as latticepoints lying on a circle of radius √ n ), we define(2.12) w k ( n ) := X z ∈ Λ n (cid:18) z | z | (cid:19) k . Pure momentum observables on the square torus
We begin by introducing some convenient notation. Given a set S ⊂ Z , let S ( x ) := S ∩ [1 , x ]. We say that a subset S ⊂ S is of full density if | S ( x ) | =(1 + o (1)) · | S ( x ) | as x → ∞ . In what follows, S will always denote the set ofintegers that can be represented as sums of two integer squares.For k = 0, we can now construct a full density subsequence S ′ k ⊂ S such that h Op( e ,k ) g λ n , g λ n i → λ n → ∞ along n ∈ S ′ k . (Recall that λ n denotes theperturbed eigenvalue associated with the Laplace eigenvalue n ∈ S .) Proposition 3.1.
For a given integer k = 0 , there exists a subsequence S ′ k ⊂ S ,of full density, such that for n ∈ S ′ k (3.1) | h Op( e ,k ) g λ n , g λ n i | ≪ (log λ n ) / − log 2 / o (1) . We note that 1 / − log 2 / − . · · · < Preliminary Results.
Before we can give the proof of Proposition 3.1 westate a number of necessary results whose proof can be found in the number theoryliterature, or in Section 5. We first recall Rieger’s bound on pair correlation typesums for integers that are sums of two squares.
Theorem 3.2 ([10], Satz 2) . Let f ( n ) denote the characteristic function of S , theset integers representable as sums of two integer squares. If < | h | ≪ x then X n ≤ x f ( n ) f ( n + h ) ≪ c ( h ) x log x where c ( h ) := Q p | hp ≡ (1 + 1 /p ) . UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 9
Remark 4.
Rieger’s result is stated for h > and summing over n ≤ x + h ,but since f ( n ) = 0 for n < and we assume | h | ≪ x , the above formulationfollows immediately (albeit possibly with a worse absolute constant.) Moreover,since c ( h ) ≤ P d | h /d , it easily follows that (3.2) X < | h |≤ T c ( h ) ≪ T, T → ∞ . We shall also need to recall a fundamental fact about the size of S ( x ). Theorem 3.3 (Landau, see [9], § . There exists c > such that (3.3) | S ( x ) | = c · x √ log x (1 + O (1 / log x )) as x → ∞ . Given n ∈ S , let ω ( n ) denote the number of prime divisors of n that are con-gruent to one modulo four, i.e., ω ( n ) := X p | n,p ≡ S . Proposition 3.4.
We have (3.4) 1 | S ( x ) | X n ∈ S ( x ) ω ( n ) = 12 log log x + O (log log log x ) and (3.5) 1 | S ( x ) | X n ∈ S ( x ) ω ( n ) = 14 (log log x ) + O ((log log x ) · log log log x )This, together with Chebychev’s inequality, immediately gives the following nor-mal order result on ω ( n ). Corollary 3.5.
Fix ǫ > . Then, as x → ∞ , (3.6) |{ n ∈ S ( x ) : | ω ( n ) −
12 log log n | < (log log n ) / ǫ }| = | S ( x ) | · (1 + o ǫ (1)) . From the corollary, we deduce (see Section 5.2 for details) the following weakanalog of a normal order result for r ( n ). Corollary 3.6. As x → ∞ , (3.7) |{ n ∈ S ( x ) : r ( n ) = (log n ) (log 2) / ± o (1) }| = | S ( x ) | · (1 + o (1)) . We shall also need the following L -bound on the exponential sums w k ( n ), Proposition 3.7. If k = 0 , then (3.8) X n ∈ S ( x ) | w k ( n ) | ≪ k x In particular, by Chebychev’s inequality, the number of n ∈ S ( x ) for which | w k ( n ) | >T is at most x/T , and we find that | w k ( n ) | ≤ (log n ) / ǫ holds for almost all n ∈ S ( x ) . The result readily follows from a Halberstam-Richert type inequality, see section5.1 for more details.
Proof of Proposition 3.1.
We begin by noting that since the set Λ n isinvariant under multiplication by i , w k ( n ) = 0 unless 4 | k . Hence the case k | k , and k = 0.We next introduce some further notation. Given m ∈ S , let m + , m − ∈ S denote the nearest neighbor (in S ) to the right, respectively left, and similarly, let m ++ , m −− denote the second nearest neighbors to the right, respectively left.Define S ⊂ S by successively removing a zero density subset of elements forwhich the following properties do not hold. Namely, let S consist of those m ∈ S for which the following properties, as m → ∞ , all hold:(1) Multiplicities are near their (logarithmic) normal order in the followingsense: r ( m ) = (log m ) (log 2) / ± o (1) , r ( m − ) = (log m ) (log 2) / ± o (1) . (2) There is nearly square root cancellation in exponential sums: | w k ( m ) | ≤ (log m ) / o (1) , | w k ( m − ) | ≤ (log m ) / o (1) . (3) There are no near neighbors: m + − m ≥ (log m ) / − o (1) , and m − m − ≥ (log m ) / − o (1) .(4) There are no near second neighbors: m ++ − m + ≥ (log m ) / − o (1) , and m − − m −− ≥ (log m ) / − o (1) .(5) Neighbors are not too far away: m + − m ≤ (log m ) / o (1) , m − m − ≤ (log m ) / o (1) , and m − − m −− ≤ (log m ) / o (1) .(6) There are not too many “close” neighbors in the following sense: for T ≪ m , |{ n ∈ S : | n − m | ≤ T }| ≪ T (log T ) (log m ) / − o (1) (7) For W ∈ [(log m ) / · (log log m ) , (log m ) ] there are ≫ W / ((log m ) / (log log m )(log W ) )elements in S that lie between m and n if | w k ( n ) | ≥ W .(8) For W ≥ (log m ) there are ≫ W / / log W elements in S that lie between m and n if | w k ( n ) | ≥ W .(9) For ǫ > G ∈ [2 , m − ǫ ], H G ( m ) := X n ∈ S,n = m | m − n |≥ G | m − n | ≪ ǫ (log G ) G (log m ) / − o (1) . Remark 5.
We tacitly assume that o (1) is chosen so that (log m ) o (1) → ∞ as m ∈ S tends to infinity; we will also use the convention that the sign of o (1) isimportant, in particular (log m ) − o (1) → . We defer the proof that S has full density inside S to Section 3.3. Remark 6.
Here, and what follows we will without comment make use of the factthat log m ∼ log m − ∼ log m + ∼ log m ++ etc. To see this, any crude bound on | m + − m | , | m ++ − m | etc suffices, e.g. the trivial bound | m + − m | ≪ m / whichfollows from bounding the distance to the nearest square. Moreover, we also use thefact that for almost all m ∈ S ( x ) , e.g. m ∈ [ x/ log x, x ] , we have log m ∼ log x . UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 11
The following will be used to show that the numerator in (2.11) is essentiallygiven by two terms.
Lemma 3.8. If m ∈ S , then (3.9) X n ∈ S,n = m,m − | w k ( n ) || m − n | ≪ m ) / − o (1) Proof.
Fix m ∈ S ( x ). To simplify the notation, let L = log m . To bound the sum X n ∈ S,n = m,m − | w k ( n ) || m − n | we split it into parts according to the size of | w k ( n ) | . Small | w k ( n ) | : | w k ( n ) | ≤ L / (log L ) . Since m ∈ S , its nearest neighbors, byproperty 3 are of distance at least L / − o (1) away from m . Thus, the contributionfrom n for which | w k ( n ) | ≤ L / (log L ) is, by property 9, ≪ X n ∈ S | n − m |≥ L / − o (1) L / (log L ) | m − n | = L / · (log L ) · H L / − o (1) ( m ) ≪ L / (log L ) L / − o (1) L / − o (1) = 1 L / − o (1) . Medium | w k ( n ) | : | w k ( n ) | ∈ [ L / (log L ) , L ]. For terms in the sum for which n ≥ m , we use the crude bound | w k ( n ) | ≤ r ( n ) ≪ √ n and find that the totalcontribution is ≪ P n ≥ m n − / ≪ m − / , and hence it is enough to consider termsfor which n < m .Let W i = 2 i L / (log L ) for integer i ≥ i L / (log L ) ≤ L andconsider n such that | w k ( n ) | ∈ [ W i , W i +1 ]. By property 7, the number of elementsin S between n and m is ≫ W i / ( L / o (1) (log W i ) ). Thus, using the bound onthe number of close neighbors, i.e., take T = | n − m | in property 6 (note that T ≪ m when n < m ), we must have T (log T ) (log m ) / − o (1) ≫ W i / ( L / o (1) (log W i ) )),which implies that | n − m | ≫ W − o (1) i . Thus, by property 9 (take G = W − o (1) i ), X n ∈ S | w k ( n ) |∈ [ W i ,W i +1 ] | w k ( n ) || n − m | ≪ X n ∈ S : | m − n |≫ W − o (1) i W i | n − m | ≪ W i (log W i ) L / − o (1) W − o (1) i = 1 L / − o (1) · W − o (1) i = 1 L / − o (1) · (2 i L / (log L ) ) − o (1) ≪ L / − o (1) (3 / i . Summing over relevant i ≥
0, we find that the total contribution is ≪ L / − o (1) . Large | w k ( n ) | : | w k ( n ) | ≥ L . Let W i = 2 i L and consider n such that | w k ( n ) | ∈ [ W i , W i ]. By property 8, we must then have | n − m | ≫ W / i / log W i , and hencethe contribution is (using the bound P k ≥ A /k ≪ /A ) ≪ X n ∈ S | n − m |≫ W / i / log W i W i | n − m | ≪ W i · log W i W / i ≪ W / − o (1) i = 1(2 i L ) / − o (1) . Summing over i ≥
0, the total contribution is ≪ L − o (1) X i ≥ − (1 / − o (1)) i ≪ L − o (1) . (cid:3) Proof of Proposition 3.1 using Lemma 3.8.
Recalling that m − < λ m < m , we notethat that | λ m − n | ≥ | m − n | if n > m . Moreover, for n ≤ m −− the minimum of | λ m − n | / | m − n | = 1 − m − λ m m − n (as n ≤ m −− ranges over elements in S ) is attainedfor n = m −− , and consequently | λ m − n | / | m − n | ≥ | λ m − m −− | / | m − m −− | ≥ | m − − m −− | / | m − m −− | which, by properties 4, and 5, is ≫ (log m ) / − o (1) / (log m ) / o (1) = 1 / (log m ) o (1) .Hence | λ m − n | ≫ (log m ) − o (1) | m − n | holds for n = m, m − , and thus X n ∈ S,n = m,m − | w k ( n ) || λ m − n | ≤ (log m ) o (1) · X n ∈ S,n = m,m − | w k ( n ) || m − n | Let M = min( | λ m − m | , | λ m − m − | ). Trivially λ m ≫ m / , and by property 1,Lemma 3.8 implies that1 /λ m + P n ∈ S | w k ( n ) || λ m − n | /λ m + P n ∈ S r ( n ) | λ m − n | ≪ O (1 /m ) + ( | w k ( m ) | + | w k ( m − ) | ) /M + m ) / − o (1) O (1 /m )+(log m ) (log 2) / − o (1) M which, by property 2, is(3.10) ≪ (log m ) / o (1) + M (log m ) / − o (1) (log m ) (log 2) / − o (1) . Recalling that λ m ∈ [ m − , m ], property 5 implies that M ≪ (log m ) o (1) and wethus find that (3.10) is ≪ (log m ) / o (1) + (log m ) o (1) (log m ) / − o (1) (log m ) (log 2) / − o (1) = 1(log m ) (log 2) / − / − o (1) = o (1)as m → ∞ , since (log 2) / − / . · · · . Recalling that log m ≫ log λ m (cf.Remark 6) the proof is concluded. (cid:3) Proof that S has full density. Property (1).
That r ( m ) = (log m ) (log 2) / o (1) holds for almost all m ∈ S follows from corollary 3.6. To ensure that r ( m − ) = (log m ) (log 2) / o (1) also holds,we remove the right neighbor of those m for which r ( m ) = (log m ) (log 2) / o (1) isnot true; this removes another zero density set. (By Remark 6, log m + = (1 + o (1)) log m .) UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 13
Property (2).
By Proposition 3.7 X n ∈ S ( x ) | w k ( n ) | ≪ x and Chebychev’s inequality, together with | S ( x ) | ∼ cx/ √ log x , then gives that | w k ( m ) | ≤ (log m ) / o (1) holds for almost all m ∈ S ( x ). Removing right neighbors,as in the proof of Property (1), the same holds for | w k ( m − ) | .3.3.3. Property (3).
Let f denote the characteristic function of S . By Theorem 3.2, X m ≤ x X h :0 < | h |≤ (log m ) / − o (1) f ( m ) f ( m + h ) ≪ x log x X h :0 < | h |≤ (log x ) / − o (1) c ( h ) ≪ x log x · (log x ) / − o (1) . Thus, by Chebychev’s inequality, X h :0 < | h |≤ (log m ) / − o (1) f ( m ) f ( m + h ) < m in S ( x ). Consequently, almost all m ∈ S have no nearbyneighbors.3.3.4. Property (4).
We use the same proof as the one used for showing that Prop-erty (3) holds.3.3.5.
Property (5).
Let n < n . . . < n I ≤ x denote ordered representativesof the elements in S ( x ), and let s i = n i +1 − n i . Since P i
The argument is similar to the one used to prove property3: again let f denote the characteristic function of S . Then, as T ≪ m ≤ x ,Theorem 3.2 gives that X m ≤ x X h :0 < | h |≤ T f ( m ) f ( m + h ) ≪ x log x X h :0 < | h |≤ T c ( h ) ≪ xT log x and Chebychev’s inequality implies that X h :0 < | h |≤ T f ( m ) f ( m + h ) ≥ T (log T ) (log x ) / − o (1) holds for at most x (log x ) / o (1) (log T ) exceptional elements m ∈ S ( x ). Taking T i =2 i , removing the exceptional elements, and summing over i we find that we haveremoved ≪ x (log x ) / o (1) X i ≥ /i = o ( | S ( x ) | )elements. Thus, property 6 holds for T being a power of two. To see that it holds for all T ≪ m , take i to be the smallest integer such that T i = 2 i ≥ T and note that T ∈ [ T i / , T i ].3.3.7. Property (7).
Given n ∈ S such that w = | w k ( n ) | ∈ [(log n ) / (log log n ) , (log n ) ] , remove 2 · w / ((log n ) / (log log n )(log w ) ) neighbors to the left of n , and 2 · w / ((log n ) / (log log n )(log w ) ) neighbors to the right, and let R n denote the setof such removed elements.Fix x and consider the number of removed elements in [1 , x ]. We claim that if l ∈ S ( x ) has been removed, then l ∈ R n for some n ≤ x . To show this, we notethat given an integer t , we can always find l , l ∈ S such that l < t < l , and l − l ≪ √ t (just take nearby squares), and since | w k ( n ) | ≤ r ( n ) ≤ n o (1) , any R n will be contained in an interval of length ≪ n / o (1) .Hence it suffices to bound the union of R n for n ≤ x . The removed contributionfrom n for which n ≤ x/ (log x ) is at most x · (log x ) (log x ) = o ( | S ( x ) | ) (here we use theassumption w ≤ (log n ) ).On the other hand, for n ∈ [ x/ (log x ) , x ], we have log n = (1 + o (1)) log x . Let W i = 2 i (log x ) / (log log x ), and consider n ∈ S ( x ) such that | w k ( n ) | ∈ [ W i , W i ].By Proposition 3.7 and Chebychev’s inequality, the number of such n is ≪ xW i ,and the total number of removed elements is thus ≪ xW i · W i (log x ) / (log log x )(log W i ) ≪ x (log x ) / (log log x ) · i Summing over i ≥ ≪ x (log x ) / (log log x ) · i = o ( | S ( x ) | ) . Property (8).
Arguing as before, if | w k ( n ) | ≥ (log n ) let w = | w k ( n ) | andremove the nearest 2 w / / log w neighbors to the right and left of n ; let R n denotethe set of removed neighbors.Fix x and consider the number of removed elements in [1 , x ]. We first note that | w k ( n ) | ≤ r ( n ) ≪ n / holds for all n ∈ S . Consequently R n , if non-empty,contains at most n / neighbors of n which (since | S (2 y ) | − | S ( y ) | ≫ y/ √ log y forall y by Landau) implies that if l ∈ S ( x ) and l belongs to some R n , then n ≤ x .Consider first the removed contribution coming from R n for which n ≤ √ x .Since | w k ( n ) | ≤ r ( n ) ≪ n / , the total contribution is ≪ √ x · ( x / ) / = o ( | S ( x ) | ) . If n ∈ [ √ x, x ] and | w k ( n ) | ≥ (log n ) , we have | w k ( n ) | ≥ (log x ) / . Define W i = 2 i · (log x ) /
100 and consider the removed contribution from R n forwhich | w k ( n ) | ∈ [ W i , W i ]. By Proposition 3.7 and Chebychev’s inequality, thenumber of such n ∈ S (2 x ) is ≪ xW i and the associated removed contribution is ≪ x · ( W / i / log W i ) W i ≪ xW / i log W i ≪ x (2 i (log x ) ) / = x i/ log x . UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 15
Summing over i ≥ ≪ X i ≥ x i/ log x = o ( | S ( x ) | ) . Property (9).
The final property is an immediate consequence of the followingLemma.
Lemma 3.9. If ǫ > then for almost all m ∈ S ( x ) , we have, for any T ∈ [2 , x − ǫ ] , X n ∈ S | n − m |≥ T m − n ) ≪ (log T ) T (log x ) / − o (1) Proof.
We first bound the sum over n ∈ S \ S (2 x ), i.e., those n for which n ≥ x : X n ∈ S | n − m |≥ Tn ≥ x m − n ) ≤ X k ≥ x /k ≪ /x = o (cid:18) (log T ) T (log x ) / − o (1) (cid:19) . (Recall that m ≤ x since m ∈ S ( x ), and that T ≤ x − ǫ .)Next we note that X m,n ∈ S (2 x ) | n − m |≥ T m − n ) = X k ≥ T |{ m, n ∈ S (2 x ) : | m − n | = k }| k By Theorem 3.2, |{ m, n ∈ S (2 x ) : | m − n | = h }| ≪ x · c ( h )log x and, by partial summation and using that c ( h ) is bounded on average (cf. (3.2), X m,n ∈ S (2 x ) | n − m |≥ T m − n ) ≪ x log x X h ≥ T c ( h ) h ≪ xT log x . By Chebychev’s inequality, the number of m ∈ S (2 x ) for which P n ∈ S (2 x ) | n − m |≥ T m − n ) ≥ (log T ) T (log x ) / − o (1) holds is thus ≪ xT log x (cid:30) (log T ) T (log x ) / − o (1) = x (log x ) / o (1) · (log T ) Taking T i = 2 i , summing over i ≪ log x , and recalling that | S ( x ) | ∼ x/ √ log x wefind that the property holds in the special case of T being a power of two. Theresult for general T follows by taking the largest i such that T i = 2 i ≤ T and notingthat T i ∈ [ T / , T ]. (cid:3) Proof of Theorem 1.1
Let a ∈ C ∞ ( S ∗ T ) be a smooth observable with rapidly decaying Fourier expan-sion a ( x, φ ) = X ζ ∈ Z ,k ∈ Z ˆ a ( ζ, k ) e i h ζ,x i +i kφ . Since Op( e ζ,k ) is unitary (cf. (2.4)) for all ζ, k , we have | h Op( e ζ,k ) g λ , g λ i | = | h g λ , g λ i | , and the rapid decay of Fourier coefficients then shows that given ǫ > J such that | h (Op( a ) − Op( P J )) g λ , g λ i | ≤ X | ζ | , | k | >J | ˆ a ( ζ, k ) || h Op( e ζ,k ) g λ , g λ i | ≤ ǫ (4.1)holds uniformly in λ , where P J ( x, φ ) is the trigonometric polynomial(4.2) P J ( x, φ ) = X ζ ∈ Z ,k ∈ Z | ζ | , | k |≤ J ˆ a ( ζ, k ) e i h ζ,x i +i kφ obtained by truncating the Fourier expansion of a . Hence it is enough to show thatfor any fixed J ≥ h Op( P J ) g λ , g λ i → S ∗ T ) Z S ∗ T a dµ = ˆ a (0 , λ → ∞ along a full density subsequence of S .Now, given J ≥
1, let ˜ S J := \ | k |≤ J ( S ′ k ∩ S ′ )where S ′ ⊂ S denotes the full density sequence of Theorem 2.1. (Since S ′ and S ′ k have full densities for all k = 0, so does ˜ S J for all J .) It follows from the previoustwo sections that(4.4) h Op( P J ) g λ , g λ i → S ∗ T ) Z S ∗ T P J dµ = ˆ a (0 , λ ∈ ˜ S J → ∞ .In order to construct the full density sequence of Theorem 1.1 we use a standarddiagonalisation argument (see for instance [4]) to extract such a sequence from thelist of sequences { ˜ S J } J . By construction ˜ S J +1 ⊂ ˜ S J . Choose M J such that for all X > M J (4.5) { λ ∈ ˜ S J | λ ≤ X } { λ ∈ S | λ ≤ X } ≥ − J and let S ′∞ be such that S ′∞ ∩ [ M J , M J +1 ] = ˜ S J ∩ [ M J , M J +1 ] for all J . Then S ′∞ ∩ [0 , M J +1 ] contains ˜ S J ∩ [0 , M J +1 ] and therefore S ′∞ is of full density in S .Moreover, for any J ≥
1, we have(4.6) h Op( P J ) g λ , g λ i = Z S ∗ T P J dµ λ → S ∗ T ) Z S ∗ T P J dµ = ˆ a (0 , λ → ∞ along S ′∞ since S ′∞ ∩ ( M J +1 , ∞ ) ⊂ ˜ S J ∩ ( M J +1 , ∞ ).5. Number theoretic background
Bounding mean values of multiplicative functions.
We recall that r ( n ) / multiplicative function , i.e., r ( mn ) / r ( m ) / · r ( n ) / m, n ) = 1, andsimilarly w k ( n ) / UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 17
In particular, both functions are determined by their values at prime powers, andwe have r ( p e )4 = e + 1 if p ≡ p ≡ e is even, or if p = 2,0 if p ≡ e is odd.For p ≡ θ p ∈ [0 , π/
4) by cos θ p = x/ p x + y , where x + y = p for x, y ∈ Z and 0 ≤ y ≤ x . We then have (if 4 | k ) w k ( p e )4 = P el =0 e i · θ p · ( e − l ) if p ≡ p ≡ e is even,0 if p ≡ e is odd, ± p = 2.In particular (again for 4 | k ), w k (2) / − k/ , and for odd primes we have(5.1) w k ( p )4 = ( kθ p ) for p ≡ p ≡ f be a non-negative multiplicative function such that for all primepowers f ( p k ) ≪ γ k holds for some γ <
2, and X p ≤ x f ( p ) = x log x · ( τ + o (1)) , as x → ∞ , for some constant τ . Satz 1 of Wirsing [17] then implies that X n ≤ x f ( n ) ≪ τ x log x · Y p ≤ x (cid:18) f ( p ) p + f ( p ) p + . . . (cid:19) . Proof of Proposition 3.7.
For k fixed, define a multiplicative function f ( n ) :=( | w k ( n ) | / (recall that | w k ( n ) | / f ( p ) = p = 2 , (2 cos( kθ p )) for p ≡ , p ≡ X p ≤ x f ( p ) = 1 + X p ≤ x,p ≡ f ( p ) = 1 + X p ≤ x,p ≡ (2 cos(2 πkθ p )) . Thus Hecke’s result on angular equidistribution of split Gaussian primes (see [6])gives that(5.2) X p ≤ x f ( p ) = x log x · (cid:18) · Z (2 cos(2 πkθ )) dθ + o (1) (cid:19) = x log x · (1 + o (1))Hence Wirsing’s Satz 1 applies (also note that f ( p k ) ≪ k for all p, k ), thus X n ≤ x | w k ( n ) | ≪ X n ≤ x f ( n ) ≪ x log x · Y p ≤ x (cid:18) f ( p ) p + f ( p ) p + . . . (cid:19) . Now, since P ∞ k =2 f ( p k ) /p k ≤ P ∞ k =2 ( k + 1) /p k ≪ /p , we find that Y p ≤ x (cid:18) f ( p ) p + f ( p ) p + . . . (cid:19) ≪ exp X p ≤ x f ( p ) p = exp (log log x + O (1)) , where the final equality follows from (5.2) and partial summation. Hence X n ≤ x | w k ( n ) | ≪ x log x · exp(log log x + O (1)) ≪ x. Erd¨os-Kac Theory.
Let ω ( n ) denote the number of distinct prime factorsof an integer n . The celebrated Erd¨os-Kac theorem assert that the distributionof n ω ( n ) − log log n √ log log n o n ≤ x is given by the standard normal distribution as x → ∞ ; inparticular, a typical integer of size x has about log log x prime factors. We shallneed some analogous, but weaker, results for elements in S .Recall that given n ∈ S , ω ( n ) denotes the number of prime factors, congruentto one modulo four, of n ; i.e., with P ′ p denoting the sum over p ≡ ω ( n ) := ′ X p | n . Proof of Proposition 3.4.
Using that at most four primes p ≥ x / can dividean integer n ≤ x , together with P ′ p ≤ x / |{ n ∈ S ( x ) : p | n }| = P ′ p ≤ x / | S ( x/p ) | , wefind that X n ∈ S ( x ) ω ( n ) = X n ∈ S ( x ) ′ X p | n X n ∈ S ( x ) ′ X p | n,p ≤ x / O (1) = ′ X p ≤ x / | S ( x/p ) | + O ( | S ( x ) | )By Landau, | S ( x/p ) | = cxp √ log( x/p ) · (1 + O (1 / log( x/p )), and thus, what will bethe main term, is given by ′ X p ≤ x / | S ( x/p ) | = ′ X p ≤ x / cxp p log( x/p ) · (1 + O (1 / log( x/p ))If p ∈ [ x / log log x , x / ] then log( x/p ) ≫ log x and thus the contribution from suchprimes p is ≪ x √ log x ′ X p ∈ [ x / log log x ,x / ] /p ≪ x √ log x · log (cid:18) log x / log x / log log x (cid:19) ≪ x log log log x √ log x which is of the same order as the claimed error term in the first assertion of theProposition.Now, if p ≤ x / log log x then log p ≤ log x/ log log x and thus p log( x/p ) = p log x − log p = p log x (cid:18) − O (cid:18) x (cid:19)(cid:19) UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 19
Hence, by the analogue of Mertens’ theorem for primes in progressions , togetherwith 1 / log( x/p ) ≪ / log x for p ≤ x / log log x , we find that ′ X p ≤ x / log log x cxp p log( x/p ) · (1 + O (1 / log( x/p ))= cx √ log x · (1 + O (1 / log x )) · (1 + O (1 / log log x ))) ′ X p ≤ x / log log x /p = cx √ log x · (1 + O (1 / log log x ))) · (cid:18)
12 log (cid:18) log x log log x (cid:19) + O (1) (cid:19) = cx √ log x · (1 + O (1 / log log x ))) · ( 12 log log x + O (log log log x )) = cx √ log x ( 12 log log x + O (log log log x ))Dividing by | S ( x ) | and again using Landau’s Theorem, the proof of the first asser-tion is concluded.The variance estimate is similar: since n ≤ x can have at most 4 prime divisors p ≥ x / , we have X n ∈ S ( x ) ω ( n ) = X n ∈ S ( x ) ′ X p ≤ x,p | n = X n ∈ S ( x ) ′ X p ≤ x / ,p | n O (1) = X n ∈ S ( x ) ′ X p ≤ x / ,p | n + 2 ′ X p ≤ x / ,p | n O (1) The total contribution from the last two terms in the inner sum is, by our firstassertion (regarding the mean value of ω ( n )), ≪ x log log x √ log x + | S ( x ) | ≪ x log log x √ log x . With [ a, b ] = ab/ ( a, b ) denoting the least common multiple of integers a, b , wehave X n ∈ S ( x ) ′ X p ≤ x / ,p | n = ′ X p ,p ≤ x / | S ( x/ [ p , p ]) | = ′ X p ,p ≤ x / | S ( x/p p ) | + ′ X p ≤ x / | S ( x/p ) | − ′ X p ≤ x / | S ( x/p ) | The latter two terms are of lower order than the claimed main term — the argumentused to estimate the mean of ω ( n ) implies that ′ X p ≤ x / | S ( x/p ) | ≪ x log log x √ log x We shall only need that P ′ p ≤ x /p = 1 / · log log x + O (1), a simple consequence of the primenumber theorem for arithmetic progressions. and, again using Landau, we find that ′ X p ≤ x / | S ( x/p ) | ≪ x √ log x ′ X p ≤ x / /p ≪ x √ log x . As for the double sum over small primes, again by Landau,(5.3) ′ X p ,p ≤ x / | S ( x/p p ) | = ′ X p ,p ≤ x / cxp p p log( x/ ( p p )) (1 + O (1 / log( x/ ( p p ))))= ′ X p ,p 12 log log x − O (log log log x ) (cid:19) · (1 + O (1 / log log x ))Dividing by | S ( x ) | and using Landau again, the main term is thus14 (log log x ) + O ((log log x ) · log log log x ) . UANTUM ERGODICITY FOR TORAL POINT SCATTERERS 21 Proof of Corollary 3.6. Define a multiplicative function f ( n ) := r ( n )4 · ω ( n ) If p ≡ f ( p e ) = ( e + 1) / 2; if p ≡ f ( p e +1 ) = 0 whereasfor even exponents f ( p e ) = 1. Using Wirsing’s Satz 1 again, we find (recall that P ′ p ≤ x denotes the over primes p ≡ X n ∈ S ( x ) f ( n ) = X n ≤ x f ( n ) ≪ x log x exp X p ≤ x f ( p ) /p = x log x exp ′ X p ≤ x /p + O (1) ≪ x log x exp (cid:18) 12 log log x + O (1) (cid:19) ≪ x (log x ) / ≪ | S ( x ) | (here we again have used Mertens’ Theorem for arithmetic progressions.) Cheby-shev’s inequality then implies that the number of n ∈ S ( x ) for which f ( n ) ≥ log log log n is o ( | S ( x ) | ). In particular, we find that2 ω ( n ) ≤ r ( n ) / ≤ ω ( n ) · log log log n holds for almost all n ∈ S ( x ). Now, since Corollary 3.5 implies that ω ( n ) =(1 / o (1)) log log n for almost all n ∈ S ( x ), we find that r ( n ) = 2 (1 / o (1)) log log n = (log n ) (log 2) / o (1) holds for almost all n ∈ S ( x ). References [1] G. Berkolaiko, J. P. Keating and B. Winn, No quantum ergodicity for star graphs , Comm.Math. Phys. Vol. 250, 259–285, 2004.[2] O. Bohigas, M. J. Giannoni and C. Schmit, Characterization of Chaotic Quantum Spectraand Universality of Level Fluctuation Laws , Phys. Rev. Lett. 52, 1–4, 1984.[3] E. Bogomolny, U. Gerland and C. Schmit, Singular Statistics , Phys. Rev. E, Vol. 63, No. 3,2001.[4] Y. Colin de Verdi`ere, Ergodicit´e et fonctions propres du laplacien. Comm. Math. Phys. Vol.102, No. 3, 497–502, 1985.[5] L. Fainsilber, P. Kurlberg, and B. Wennberg. Lattice points on circles and discrete velocitymodels for the Boltzmann equation , SIAM J. Math. Anal. 37, No. 6, 1903–1922 (electronic),2006.[6] E. Hecke. Eine neue Art von Zetafunktionen und ihre Beziehungen zur Verteilung derPrimzahlen. Math. Z. 6, No. 1-2, 11–51, 1920.[7] J. P. Keating, J. Marklof and B. Winn, Localised eigenfunctions in ˘Seba billiards. J. Math.Phys. 51, no. 062101.[8] P. Kurlberg and H. Uebersch¨ar, Localised semiclassical measures for point scatterers ondiophantine tori. in preparation[9] E. Landau, Handbuch der Lehre von der Verteilung der Primzahlen. 2 B¨ande. Chelsea Pub-lishing Co., New York, 1953. 2nd ed., With an appendix by Paul T. Bateman.[10] G. J. Rieger. Aufeinanderfolgende Zahlen als Summen von zwei Quadraten. Nederl. Akad.Wetensch. Proc. Ser. A 68 = Indag. Math. , 27:208–220, 1965.[11] Z. Rudnick, H. Uebersch¨ar, Statistics of wave functions for a point scatterer on the torus .Comm. Math. Phys., Vol. 316, No. 3, 763–782, 2012.[12] Z. Rudnick, H. Uebersch¨ar, On the eigenvalue spacing distribution for a point scatterer on aflat torus . Annales Henri Poincar´e, to appear[13] M. Ruzhansky, V. Turunen, Pseudo-Differential Operators and Symmetries: BackgroundAnalysis and Advanced Topips , Birkh¨auser (2010).[14] P. ˘Seba, Wave chaos in a singular quantum billiard . Phys. Rev. Lett. 64, 1855–1858, 1990. [15] T. Shigehara, Conditions for the appearance of wave chaos in quantum singular systems witha pointlike scatterer , Phys. Rev. E, Vol. 50, No. 6, 1994.[16] H. Uebersch¨ar, Quantum Chaos for point scatterers on flat tori , Phil. Trans. R. Soc. A 372(2014), 20120509.[17] E. Wirsing. Das Asymptotische Verhalten von Summen ¨uber multiplikative Funktionen. Math. Ann. 143, 75–102, 1961.[18] N. Yesha, Uniform distribution for the perturbed eigenfunctions of a point scatterer on thethree-dimensional torus , in preparation[19] S. Zelditch, Uniform distribution of eigenfunctions on compact hyperbolic surfaces , DukeMath. J. 55, 919–941, 1987. Department of Mathematics, KTH Royal Institute of Technology, SE-10044, Stock-holm, Sweden E-mail address : [email protected] Institut de Physique Th´eorique, CEA Saclay, 91191 Gif-sur-Yvette Cedex, France. E-mail address ::