arXiv subjects
Publications and source records attributed to Mihail N. Kolountzakis.
We give a simple proof of the fact that there exist measures on the real line of discrete support, whose Fourier Transform is also a measure of discrete support, yet this Fourier pair cannot be constructed by repeatedly applying the Poisson Summation Formula finitely many times. More specifically the support of both the measure and its Fourier Tranform are not contained in a finite union of arithmetic progressions.
Suppose $Ω\subseteq\RR^d$ is a bounded and measurable set and $Λ\subseteq \RR^d$ is a lattice. Suppose also that $Ω$ tiles multiply, at level $k$, when translated at the locations $Λ$. This means that the $Λ$-translates of $Ω$ cover almost every point of $\RR^d$ exactly $k$ times. We show here that there is a set of exponentials $\exp(2πi t\cdot x)$, $t\in T$, where $T$ is some countable subset of $\RR^d$, which forms a Riesz basis of $L^2(Ω)$. This result was recently proved by Grepstad and Lev under the extra assumption that $Ω$ has boundary of measure 0, using methods from the theory of quasicrystals. Our approach is rather more elementary and is based almost entirely on linear algebra. The set of frequencies $T$ turns out to be a finite union of shifted copies of the dual lattice $Λ^*$. It can be chosen knowing only $Λ$ and $k$ and is the same for all $Ω$ that tile multiply with $Λ$.
Consider the plane as a union of congruent unit squares in a checkerboard pattern, each square colored black or white in an arbitrary manner. The discrepancy of a curve with respect to a given coloring is the difference of its white length minus its black length, in absolute value. We show that for every radius t>1 there exists a full circle of radius either t or 2t with discrepancy greater than ct^(1/2) for some numerical constant c>0. We also show that for every t>1 there exists a circular arc of radius exactly t with discrepancy greater than ct^(1/2). Finally we investigate the corresponding problem for more general curves and their interiors. These results answer questions posed by Kolountzakis and Iosevich.
A bounded measurable set $Ω$, of Lebesgue measure 1, in the real line is called spectral if there is a set $Λ$ of real numbers ("frequencies") such that the exponential functions $e_λ(x) = \exp(2πi λx)$, $λ\inΛ$, form a complete orthonormal system of $L^2(Ω)$. Such a set $Λ$ is called a {\em spectrum} of $Ω$. In this note we prove that any spectrum $Λ$ of a bounded measurable set $Ω\subseteq\RR$ must be periodic.
Suppose $Λ\subseteq \RR^2$ has the property that any two exponentials with frequency from $Λ$ are orthogonal in the space $L^2(D)$, where $D \subseteq \RR^2$ is the unit disk. Such sets $Λ$ are known to be finite but it is not known if their size is uniformly bounded. We show that if there are two elements of $Λ$ which are distance $t$ apart then the size of $Λ$ is $O(t)$. As a consequence we improve a result of Iosevich and Jaming and show that $Λ$ has at most $O(R^{2/3})$ elements in any disk of radius $R$.
A set $Ω$, of Lebesgue measure 1, in the real line is called spectral if there is a set $Λ$ of real numbers such that the exponential functions $e_λ(x) = \exp(2πi λx)$ form a complete orthonormal system on $L^2(Ω)$. Such a set $Λ$ is called a spectrum of $Ω$. In this note we present a simplified proof of the fact that any spectrum $Λ$ of a set $Ω$ which is finite union of intervals must be periodic. The original proof is due to Bose and Madan.
The number of triangles is a computationally expensive graph statistic which is frequently used in complex network analysis (e.g., transitivity ratio), in various random graph models (e.g., exponential random graph model) and in important real world applications such as spam detection, uncovering of the hidden thematic structure of the Web and link recommendation. Counting triangles in graphs with millions and billions of edges requires algorithms which run fast, use small amount of space, provide accurate estimates of the number of triangles and preferably are parallelizable. In this paper we present an efficient triangle counting algorithm which can be adapted to the semistreaming model. The key idea of our algorithm is to combine the sampling algorithm of Tsourakakis et al. and the partitioning of the set of vertices into a high degree and a low degree subset respectively as in the Alon, Yuster and Zwick work treating each set appropriately. We obtain a running time $O \left(m + \frac{m^{3/2} Δ\log{n}}{t ε^2} \right)$ and an $ε$ approximation (multiplicative error), where $n$ is the number of vertices, $m$ the number of edges and $Δ$ the maximum number of triangles an edge is contained. Furthermore, we show how this algorithm can be adapted to the semistreaming model with space usage $O\left(m^{1/2}\log{n} + \frac{m^{3/2} Δ\log{n}}{t ε^2} \right)$ and a constant number of passes (three) over the graph stream. We apply our methods in various networks with several millions of edges and we obtain excellent results. Finally, we propose a random projection based method for triangle counting and provide a sufficient condition to obtain an estimate with low variance.
This is a survey about tiling by translation only and related questions and methods, especially those that have to do with Fourier Analysis.
Triangle counting is an important problem in graph mining. Clustering coefficients of vertices and the transitivity ratio of the graph are two metrics often used in complex network analysis. Furthermore, triangles have been used successfully in several real-world applications. However, exact triangle counting is an expensive computation. In this paper we present the analysis of a practical sampling algorithm for counting triangles in graphs. Our analysis yields optimal values for the sampling rate, thus resulting in tremendous speedups ranging from \emph{2800}x to \emph{70000}x when applied to real-world networks. At the same time the accuracy of the estimation is excellent. Our contributions include experimentation on graphs with several millions of nodes and edges, where we show how practical our proposed method is. Finally, our algorithm's implementation is a part of the \pegasus library (Code and datasets are available at (http://www.cs.cmu.edu/~ctsourak/).) a Peta-Graph Mining library implemented in Hadoop, the open source version of Mapreduce.
We show that there are polynomials $p_N$ of arbitrarily large degree $N$, with coefficients equal to 0 or 1 (Newman polynomials), such that $$ \liminf_{N \to \infty} N \Linf{p_N^2} \bigl / p_N^2(1) < 1, $$ where $\Linf{q}$ denotes the maximum coefficient of the polynomial $q$ and which, at the same time, are sparse: $p_N(1)/N \to 0$. This disproves a conjecture of Yu \cite{yu}. We build on some previous results of Berenhaut and Saidak \cite{berenhaut-saidak} and Dubickas \cite{dubickas} whose examples lacked the sparsity. This sparsity we create from these examples by randomization.
Consider the plane as a checkerboard, with each unit square colored black or white in an arbitrary manner. In a previous paper we showed that for any such coloring there are straight line segments, of arbitrarily large length, such that the difference of their white length minus their black length, in absolute value, is at least the square root of their length, up to a multiplicative constant. For the corresponding "finite" problem ($N \times N$ checkerboard) we had proved that we can color it in such a way that the above quantity is at most $C \sqrt{N \log N}$, for any placement of the line segment. In this followup we show that it is possible to color the infinite checkerboard with two colors so that for any line segment $I$ the excess of one color over another is bounded above by $C_ε\Abs{I}^{\frac12+ε}$, for any $ε>0$. We also prove lower bounds for the discrepancy of circular arcs. Finally, we make some observations regarding the $L^p$ discrepancies for segments and arcs, $p<2$, for which our $L^2$-based methods fail to give any reasonable estimates.
In this paper we study algorithms for tiling problems. We show that the conditions $(T1)$ and $(T2)$ of Coven and Meyerowitz, conjectured to be necessary and sufficient for a finite set $A$ to tile the integers, can be checked in time polynomial in ${diam}(A)$. We also give heuristic algorithms to find all non-periodic tilings of a cyclic group $Z_N$. In particular we carry out a full classification of all non-periodic tilings of $Z_{144}$.
Consider the plane as a checkerboard, with each unit square colored black or white in an arbitrary manner. We show that for any such coloring there are straight line segments, of arbitrarily large length, such that the difference of their white length minus their black length, in absolute value, is at least the square root of their length, up to a multiplicative constant. For the corresponding ``finite'' problem ($N \times N$ checkerboard) we also prove that we can color it in such a way that the above quantity is at most $C \sqrt{N \log N}$, for any placement of the line segment.
Suppose we put an $ε$-disk around each lattice point in the plane, and then we rotate this object around the origin for a set $Θ$ of angles. When do we cover the whole plane, except for a neighborhood of the origin? This is the problem we study in this paper. It is very easy to see that if $Θ= [0,2π]$ then we do indeed cover. The problem becomes more interesting if we try to achieve covering with a small closed set $Θ$.
Let $G$ be a finite abelian group and $E$ a subset of it. Suppose that we know for all subsets $T$ of $G$ of size up to $k$ for how many $x \in G$ the translate $x+T$ is contained in $E$. This information is collectively called the $k$-deck of $E$. One can naturally extend the domain of definition of the $k$-deck to include functions on $G$. Given the group $G$ when is the $k$-deck of a set in $G$ sufficient to determine the set up to translation? The 2-deck is not sufficient (even when we allow for reflection of the set, which does not change the 2-deck) and the first interesting case is $k=3$. We further restrict $G$ to be cyclic and determine the values of $n$ for which the 3-deck of a subset of $\ZZ_n$ is sufficient to determine the set up to translation. This completes the work begun by Grünbaum and Moore as far as the 3-deck is concerned. We additionally estimate from above the probability that for a random subset of $\ZZ_n$ there exists another subset, not a translate of the first, with the same 3-deck. We give an exponentially small upper bound when the previously known one was $O(1\bigl / \sqrt{n})$.
We give an algorithm for learning symmetric k-juntas (boolean functions of $n$ boolean variables which depend only on an unknown set of $k$ of these variables) in the PAC model under the uniform distribution, which runs in time n^{O(k/\log k)}. Our bound is obtained by proving the following result: Every symmetric boolean function on k variables, except for the parity and the constant functions, has a non-zero Fourier coefficient of order at least 1 and at most O(k/\log k). This improves the previously best known bound of (3/31)k, and provides the first n^{o(k)} time algorithm for learning symmetric juntas.
By analyzing the connection between complex Hadamard matrices and spectral sets we prove the direction ``spectral -> tile'' of the Sectral Set Conjecture for all sets A of size at most 5 in any finite Abelian group. This result is then extended to the infinite grid $\Z^d$ for any dimension d, and finally to Euclidean space. It was pointed out recently by Tao that the corresponding statement fails for |A|=6 in the group $\Z_3^5$, and this observation quickly led to the failure of the Spectral Set Conjecture in $\R^5$ (Tao), and subsequently in $\R^4$ (Matolcsi). In the second part of this note we reduce this dimension further, showing that the direction ``spectral -> tile'' of the Spectral Set Conjecture is false already in dimension 3. In a computational search for counterexamples in lower dimension (one and two) one needs, at the very least, to be able to decide efficiently if a set is a tile (in, say, a cyclic group) and if it is spectral. Such efficient procedures are lacking however and we make a few comments for the computational complexity of some related problems.
We give a new proof of the following interesting fact recently proved by Bower and Michael: if a d-dimensional rectangular box can be tiled using translates of two types of rectangular bricks, then it can also be tiled in the following way. We can cut the box across one of its sides into two boxes, one of which can be tiled with the first brick only and the other one with the second brick. Our proof relies on the Fourier Transform. We also show that no such result is true for three, or more, types of bricks.