In this seminar, I’m going to be discussing joint work with Kyle Hambrook. This work is still in progress, so be aware that it may still contain errors at this point. A set E Ă Rn is said to have Fourier dimension at least s if the following condition holds: there exists a probability measure µ supported on E with the property that, for every ǫ ą 0, there exists a constant Cǫ such that µ ppξq ď Cǫ |ξ|´s{2`ǫ . (If this holds “on average”, then E has Hausdorff dimension at least s. So the Fourier dimension of E is no more than the Hausdorff dimension of E). Note that the Fourier dimension of E depends on the ambient space. A set E is called a Salem set if the Fourier dimension of E is equal to the Hausdorff dimension of E. Example: The sphere in Rn is a Salem set. Nonexamples: Hyperplanes in Rn have Fourier dimension 0. The Cantor middle-thirds set has Fourier dimension 0. The first examples of Salem sets of fractional dimension are due to Salem [5] : Salem’s constructions were random fractals. The well-approximable numbers were shown to be a Salem set by Kaufman, [3]. This was later explained in detail by Bluhm, [2]. In the local-field setting, Papadimitropoulos modified Salem’s construction to work over local fields. The goal of my joint project with Kyle is to show that the well-approximable numbers in the discrete valuation ring Zp are a Salem set. Definition 0.1. The p-adic integers Zp are the set ÿ cj p j j where cj is a number from 0 to p ´ 1. Addition and multiplication of finite sums is just like base-p arithmetic in Z. The topology is generated by the p-adic absolute value, which is |x|p “ pj , where j is the first nonzero index in the sum. With respect to this absolute value, Zp is a compact abelian group, and therefore comes equipped with a finite Haar measure: the Haar measure of a ball is equal to its radius. Definition 0.2. The set of well-approximable numbers of degree τ in Zp is the set tx : there are infinitely many pairs r, q P Z such that |qx´r|p ď maxp|r|, |q|q´τ u. It is easily seen from the Dirichlet principle that every x is approximable of degree 2. A result of Morotskaya, together with a later result of Bernik and Morotskaya, imply that the numbers approximable of degree τ has Hausdorff dimension τ2 for any τ ą 2 [4] [1]. So the goal is to show that the Fourier dimension of this set is also τ2 . 1 I will now discuss what the additive characters on the p-adic numbers Qp look like. There is a locally constant additive character χ such that χp1q “ ep p1 q and χ is constant on cosets of pZp - note that it |x|p ă 1, then χpxq “ 1. Letting χ be this character, each character of Qp is given by χpyxq for some y P Qp . If |y|p ă 1, this character is trivial on Zp . So the family of characters can be viewed as eptyxp´1 up q, where t¨up is the p-adic fractional part, and y is thought of as an element of Qp {Zp . Define Qpαq to be the set of p-adic integers of the form r{q where |r|, |q| ď pα and pq, pq “ 1. Define Qpα,τ q to be the set of p-adic integers in the p´τ α thickening of this set. Define the function Fα to be the measure on Qpα,τ q weighted by the number of pairs pr, qq corresponding to a given rational number. ś I will pick a quickly growing sequence tαj u and define µk “ kj“1 Fαj . I will show that the µ pk ’s are convergent in the supremum norm, so the µk ’s have a nonzero weak limit µ, which will be our desired probability measure. I will start by estimating the Fpα ’s. Observe that we can obtain µ pk has a k-fold convolution of the Fα ’s for appropriate α. Note that Fα is constant on balls of radius p´τ α . This implies that Fpα psq “ 0 for any s with |s|p ą pτ α . Now, it’s time to compute the Fourier coefficient Fpα psq for s such that |s|p ď τα p . If s is nonzero, we can write s “ s1 p´γ´1 , where s1 is a p-adic integer with absolute value 1. Fpα psq “ ż Zp “C Fα psqeptsxuq dx ÿ ż r{qPQα |x´r{q|p epts1 x{pγ uq dx. ăp´τ α Here, C is an appropriate constant so that Fpα p0q “ 1. Now, the expression in the exponent does not depend on anything other than the first γ digits of s1 x (and therefore, not on any component of x other than the first γ digits). Therefore, the integrand does not depend on x: because γ ă τ α, The integrand is constant and equal to epts1 rq ˚ {pγ up q, where q ˚ denotes the inverse of q mod pγ . So we are left with ÿ “ Cpτ α eps1 rq ˚ {pγ q. r{qPQα We therefore only need to check the region where α ă γ ď τ α. We split this sum up over different values of q: ÿ ÿ Cpτ α eps1 rq ˚ {pγ q qăpα rďpα pq,pq“1 If we split this sum up over q, it’s trivial to see that this sum cancels if γ ď α. ` ˘´1 Now, we can use the Evidently, the appropriate constant C is p´τ α #Qpαq 2 geometric series bound and the trivial bound on the inner sum: the inner sum is obviously less than or equal to pα in absolute value, but we also have that it’s less than or equal to pγ {ps1 q ˚ q by the geometric series bound. So the goal is to show that s1 q ˚ is usually large. The trick here is simple: suppose s1 q ˚ ” y mod pγ , where y ă N and q ă N :“ pα Then we have that s1 ” yq mod pγ . In other words, we have that s1 ` zpγ “ yq for some z. But the right hand side is bounded above by M N , so the left side is too, and there are only MN pγ possible values of z. By the divisor bound, each of these contributes no more than pM N qop1q possible solutions, so we have a total of no more than pMN q1`op1q pγ γ such pairs q, y. This shows that s1 q ˚ α 1`op1q times. can only be smaller than N mod p at most pp Npqγ Note that the statement that |s1 q ˚ | is no more than N mod pγ is equivalent to saying that the angle ps1 q ˚ {pγ q is no larger than pNγ . So if we start with N “ pγ´α , that will tell us the number of angles that are smaller than p´α . This is the smallest angle we’re interested in because the trivial bound is better for smaller angles. So we apply the bound for each N from pγ´α to pγ . We then perform a dyadic summation: The number of q such that s1 q ˚ is between 2j pγ´α and 2j`1 pγ´α is no more than 2j - so each family of angles contributes approximately the same amount to the sum- a total of pα`opαq . Because there are popαq summands, we get the bound of pα`opαq on the sum. Summarizing, we have the following bounds: $ “ 1 if s “ 0 ’ ’ ’ &“ 0 if 0 ă |s| ď pα p Fα psq α α ’ ď p if p ă |s|p ď pτ α ’ ’ % “ 0 if |s|p ą pτ α Let αj “ 2j . Then for large enough j, we have that for αj ă |s|p ď τ αj , µ pk psq “ pFα1 ¨ ¨ ¨ Fαk q^ psq “ ˚kℓ“1 Fppαℓ q psq ÿ Fpα1 ps1 q ¨ ¨ ¨ Fpαk psk q. “ s1 `...`sk “s Note that if sℓ is nonzero and in the support of Fpαpℓq for some ℓ ą j, the growth rate of the αk ’s guarantees that we do not have s1 ` . . . ` sk “ s. Therefore, this convolution operation stabilizes after j convolutions. Furthermore, it’s immediately clear that we can’t have s1 ` . . . ` sj “ s if the sj term is zerothis would violate the ultrametric inequality. Therefore, we get that µ pk psq “ 0 for k ă j and µ pk psq “ µ pj psq for k ě j. We now need to find a bound on the number of decompositions for which each piece s1 ` . . . ` sj “ s, where each |sj | ď pαj . Clearly, the digits of sj after the p´αj´1 digit are uniquely determined. There are at most pαj´1 choices 3 for the remaining digits. After making this choice, there are at most pαj´2 choices for sj´1 . Continuing in this manner, we get that there are a total of no more than pαj´1 `¨¨¨`α1 terms in the sum, each of which has size no more than p´αj `opαj q . Therefore, by choosing the αj to grow sufficiently rapidly, we can guarantee that α1 ` . . . ` αj´1 À αj , so we get that the µ pj psq À p´αj `opαj q . In particular, this proves that the µ pj ’s converge uniformly, because sups‰0 µ pj psq Ñ 0 as j Ñ 8. Let µ p be the limit of the µj ’s. We then know that the µj ’s converge weakly to a measure µ with µ p as its Fourier transform. Because of the bounds ´ 1 `op1q p on Fpαj q psq for each j, we have the desired bound: µ ppsq ď |s|p τ . References [1] V. I. Bernik and I. L. Morotskaya. Diophantine approximations in Qp and Hausdorff dimension. Vestsı̄ Akad. Navuk BSSR Ser. Fı̄z.-Mat. Navuk, (3):3– 9, 123, 1986. [2] Christian Bluhm. On a theorem of Kaufman: Cantor-type construction of linear fractal Salem sets. Ark. Mat., 36(2):307–316, 1998. [3] R. Kaufman. On the theorem of Jarnı́k and Besicovitch. Acta Arith., 39(3):265–267, 1981. [4] I. L. Morotskaya. Hausdorff dimension and Diophantine approximations in Qp . Dokl. Akad. Nauk BSSR, 31(7):597–600, 668, 1987. [5] R. Salem. On singular monotonic functions whose spectrum has a given Hausdorff dimension. Ark. Mat., 1:353–365, 1951. 4