-
Extremal Eigenvalues of Random Kernel Matrices with Polynomial Scaling
Authors:
David Kogan,
Sagnik Nandy,
Jiaoyang Huang
Abstract:
We study the spectral norm of random kernel matrices with polynomial scaling, where the number of samples scales polynomially with the data dimension. In this regime, Lu and Yau (2022) proved that the empirical spectral distribution converges to the additive free convolution of a semicircle law and a Marcenko-Pastur law. We demonstrate that the random kernel matrix can be decomposed into a "bulk"…
▽ More
We study the spectral norm of random kernel matrices with polynomial scaling, where the number of samples scales polynomially with the data dimension. In this regime, Lu and Yau (2022) proved that the empirical spectral distribution converges to the additive free convolution of a semicircle law and a Marcenko-Pastur law. We demonstrate that the random kernel matrix can be decomposed into a "bulk" part and a low-rank part. The spectral norm of the "bulk" part almost surely converges to the edge of the limiting spectrum.
In the special case where the random kernel matrices correspond to the inner products of random tensors, the empirical spectral distribution converges to the Marcenko-Pastur law. We prove that the largest and smallest eigenvalues converge to the corresponding spectral edges of the Marcenko-Pastur law.
△ Less
Submitted 22 October, 2024;
originally announced October 2024.
-
Correlation of coalescence times in a diploid Wright-Fisher model with recombination and selfing
Authors:
David Kogan,
Dimitrios Diamantidis,
John Wakeley,
Wai-Tong Louis Fan
Abstract:
The correlation among the gene genealogies at different loci is crucial in biology, yet challenging to understand because such correlation depends on many factors including genetic linkage, recombination, natural selection and population structure. Based on a diploid Wright-Fisher model with a single mating type and partial selfing for a constant large population with size $N$, we quantify the com…
▽ More
The correlation among the gene genealogies at different loci is crucial in biology, yet challenging to understand because such correlation depends on many factors including genetic linkage, recombination, natural selection and population structure. Based on a diploid Wright-Fisher model with a single mating type and partial selfing for a constant large population with size $N$, we quantify the combined effect of genetic drift and two competing factors, recombination and selfing, on the correlation of coalescence times at two linked loci for samples of size two. Recombination decouples the genealogies at different loci and decreases the correlation while selfing increases the correlation. We obtain explicit asymptotic formulas for the correlation for four scaling scenarios that depend on whether the selfing probability and the recombination probability are of order $O(1/N)$ or $O(1)$ as $N$ tends to infinity. Our analytical results confirm that the asymptotic lower bound in [King, Wakeley, Carmi (Theor. Popul. Biol. 2018)] is sharp when the loci are unlinked and when there is no selfing, and provide a number of new formulas for other scaling scenarios that have not been considered before. We present asymptotic results for the variance of Tajima's estimator of the population mutation rate for infinitely many loci as $N$ tends to infinity. When the selfing probability is of order $O(1)$ and is equal to a positive constant $s$ for all $N$ and if the samples at both loci are in the same individual, then the variance of the Tajima's estimator tends to $s/2$ (hence remains positive) even when the recombination rate, the number of loci and the population size all tend to infinity.
△ Less
Submitted 18 October, 2023;
originally announced October 2023.
-
Cyclic and well-rounded lattices
Authors:
Lenny Fukshansky,
David Kogan
Abstract:
We focus on two important classes of lattices, the well-rounded and the cyclic. We show that every well-rounded lattice in the plane is similar to a cyclic lattice, and use this cyclic parameterization to count planar well-rounded similarity classes defined over a fixed number field with respect to height. We then investigate cyclic properties of the irreducible root lattices in arbitrary dimensio…
▽ More
We focus on two important classes of lattices, the well-rounded and the cyclic. We show that every well-rounded lattice in the plane is similar to a cyclic lattice, and use this cyclic parameterization to count planar well-rounded similarity classes defined over a fixed number field with respect to height. We then investigate cyclic properties of the irreducible root lattices in arbitrary dimensions, in particular classifying those that are simple cyclic, i.e. generated by rotation shifts of a single vector. Finally, we classify cyclic, simple cyclic and well-rounded cyclic lattices coming from rings of integers of Galois algebraic number fields.
△ Less
Submitted 16 February, 2022; v1 submitted 10 October, 2021;
originally announced October 2021.
-
On average coherence of cyclotomic lattices
Authors:
Lenny Fukshansky,
David Kogan
Abstract:
We introduce maximal and average coherence on lattices by analogy with these notions on frames in Euclidean spaces. Lattices with low coherence can be of interest in signal processing, whereas lattices with high orthogonality defect are of interest in sphere packing problems. As such, coherence and orthogonality defect are different measures of the extent to which a lattice fails to be orthogonal,…
▽ More
We introduce maximal and average coherence on lattices by analogy with these notions on frames in Euclidean spaces. Lattices with low coherence can be of interest in signal processing, whereas lattices with high orthogonality defect are of interest in sphere packing problems. As such, coherence and orthogonality defect are different measures of the extent to which a lattice fails to be orthogonal, and maximizing their quotient (normalized for the number of minimal vectors with respect to dimension) gives lattices with particularly good optimization properties. While orthogonality defect is a fairly classical and well-studied notion on various families of lattices, coherence is not. We investigate coherence properties of a nice family of algebraic lattices coming from rings of integers in cyclotomic number fields, proving a simple formula for their average coherence. We look at some examples of such lattices and compare their coherence properties to those of the standard root lattices.
△ Less
Submitted 4 November, 2022; v1 submitted 14 December, 2020;
originally announced December 2020.
-
On the geometry of nearly orthogonal lattices
Authors:
Lenny Fukshansky,
David Kogan
Abstract:
Nearly orthogonal lattices were formally defined in [4], where their applications to image compression were also discussed. The idea of ``near orthogonality" in $2$-dimensions goes back to the work of Gauss. In this paper, we focus on well-rounded nearly orthogonal lattices in~$\mathbb R^n$ and investigate their geometric and optimization properties. Specifically, we prove that the sphere packing…
▽ More
Nearly orthogonal lattices were formally defined in [4], where their applications to image compression were also discussed. The idea of ``near orthogonality" in $2$-dimensions goes back to the work of Gauss. In this paper, we focus on well-rounded nearly orthogonal lattices in~$\mathbb R^n$ and investigate their geometric and optimization properties. Specifically, we prove that the sphere packing density function on the space of well-rounded lattices in dimension $n\geq 3$ does not have any local maxima on the nearly orthogonal set and has only one local minimum there: at the integer lattice~$\mathbb Z^n$. Further, we show that the nearly orthogonal set cannot contain any perfect lattices for~$n \geq 3$, although it contains multiple eutactic (and even strongly eutactic) lattices in every dimension. This implies that eutactic lattices, while always critical points of the packing density function, are not necessarily local maxima or minima even among the well-rounded lattices. We also prove that a (weakly) nearly orthogonal lattice in~$\mathbb R^n$ contains no more than~$4n-2$ minimal vectors (with any smaller even number possible) and establish some bounds on coherence of these lattices.
△ Less
Submitted 19 July, 2021; v1 submitted 8 March, 2020;
originally announced March 2020.
-
On The 2D Phase Retrieval Problem
Authors:
Dani Kogan,
Yonina C. Eldar,
Dan Oron
Abstract:
The recovery of a signal from the magnitude of its Fourier transform, also known as phase retrieval, is of fundamental importance in many scientific fields. It is well known that due to the loss of Fourier phase the problem in 1D is ill-posed. Without further constraints, there is no unique solution to the problem. In contrast, uniqueness up to trivial ambiguities very often exists in higher dimen…
▽ More
The recovery of a signal from the magnitude of its Fourier transform, also known as phase retrieval, is of fundamental importance in many scientific fields. It is well known that due to the loss of Fourier phase the problem in 1D is ill-posed. Without further constraints, there is no unique solution to the problem. In contrast, uniqueness up to trivial ambiguities very often exists in higher dimensions, with mild constraints on the input. In this paper we focus on the 2D phase retrieval problem and provide insight into this uniqueness property by exploring the connection between the 2D and 1D formulations. In particular, we show that 2D phase retrieval can be cast as a 1D problem with additional constraints, which limit the solution space. We then prove that only one additional constraint is sufficient to reduce the many feasible solutions in the 1D setting to a unique solution for almost all signals. These results allow to obtain an analytical approach (with combinatorial complexity) to solve the 2D phase retrieval problem when it is unique.
△ Less
Submitted 26 May, 2016;
originally announced May 2016.
-
An isoperimetric inequality for conjugation-invariant sets in the symmetric group
Authors:
Neta Atzmon,
David Ellis,
Dmitry Kogan
Abstract:
We prove an isoperimetric inequality for conjugation-invariant sets of size $k$ in $S_n$, showing that these necessarily have edge-boundary considerably larger than some other sets of size $k$ (provided $k$ is small). Specifically, let $T_n$ denote the Cayley graph on $S_n$ generated by the set of all transpositions. We show that if $A \subset S_n$ is a conjugation-invariant set with…
▽ More
We prove an isoperimetric inequality for conjugation-invariant sets of size $k$ in $S_n$, showing that these necessarily have edge-boundary considerably larger than some other sets of size $k$ (provided $k$ is small). Specifically, let $T_n$ denote the Cayley graph on $S_n$ generated by the set of all transpositions. We show that if $A \subset S_n$ is a conjugation-invariant set with $|A| = pn! \leq n!/2$, then the edge-boundary of $A$ in $T_n$ has size at least $$c \cdot \frac {\log_2 (\tfrac 1{p})}{\log_2 \log_2 (\tfrac 2{p})}\cdot n \cdot |A|,$$ where $c$ is an absolute constant. (This is sharp up to an absolute constant factor, when $p = Θ(1/s!)$ for any $s \in \{1,2,...,n\}$.) It follows that if $p = n^{-Θ(1)}$, then the edge-boundary of a conjugation-invariant set of measure $p$ is necessarily a factor of $Ω(\log n / \log \log n)$ larger than the minimum edge-boundary over all sets of measure $p$.
△ Less
Submitted 28 October, 2014; v1 submitted 16 September, 2014;
originally announced September 2014.
