SearcharxivSearch

arXiv subjects

Ilya Schurov

Publications and source records attributed to Ilya Schurov.

6 recordsLinked to original sources

Learning complexity of many-body quantum sign structures through the lens of Boolean Fourier analysis

We study sign structures of the ground states of spin-$1/2$ magnetic systems using the methods of Boolean Fourier analysis. Previously it was shown that the sign structures of frustrated systems are of complex nature: specifically, neural networks of popular architectures lack the generalization ability necessary to effectively reconstruct sign structures in supervised learning settings. This is believed to be an obstacle for applications of neural quantum states to frustrated systems. In the present work, we develop an alternative language for the analysis of sign structures based on representing them as polynomial functions defined on the Boolean hypercube - an approach called Boolean Fourier analysis. We discuss the relations between the properties of the Boolean Fourier series and the learning complexity of sign structures, and demonstrate that such polynomials can potentially serve as variational ansätze for the complex sign structures that dramatically outperform neural networks in terms of generalization ability. While ansätze of this type cannot yet be directly used in the context of variational optimization, they indicate that the complexity of sign structures is not an insurmountable curse, and can potentially be learned with better designed NQS architectures. Finally, we show how augmenting data with Boolean functions can aid sign prediction by neural networks.

cond-mat.dis-nn

Bachet's game with lottery moves

Bachet's game is a variant of the game of Nim. There are $n$ objects in one pile. Two players take turns to remove any positive number of objects not exceeding some fixed number $m$. The player who takes the last object loses. We consider a variant of Bachet's game in which each move is a lottery over set $\{1,2,\ldots, m\}$. The outcome of a lottery is the number of objects that player takes from the pile. We show that under some nondegenericity assumptions on the set of available lotteries the probability that the first player wins in subgame perfect Nash equilibrium converges to $1/2$ as $n$ tends to infinity.

math.OC

Global bifurcations in the two-sphere: a new perspective

We construct an open set of structurally unstable three parameter families whose weak and so called moderate topological classification defined below has a numerical invariant that may take an arbitrary positive value. Here and below "families" are "families of vector fields in the two-sphere". This result disproves an Arnold's conjecture of 1985. Then we construct an open set of six parameter families whose moderate topological classification has a functional invariant. This invariant is an arbitrary germ of a smooth map $(\mathbb R_+, a)\to(\mathbb R_+, b)$. More generally, for any positive integers $d$ and $d'$, we construct an open set of families whose topological classification has a germ of a smooth map $\left(\mathbb R_+^d, a\right)\to\left(\mathbb R_+^{d'}, b\right)$ as an invariant. Any smooth germ of this kind may be realized as such an invariant. These results open a new perspective of the global bifurcation theory in the two sphere. This perspective is discussed at the end of the paper.

math.DS

Duck factory on the two-torus: multiple canard cycles without geometric constraints

Slow-fast systems on the two-torus are studied. As it was shown before, canard cycles are generic in such systems, which is in drastic contrast with the planar case. It is known that if the rotation number of the Poincare map is integer and the slow curve is connected, the number of canard limit cycles is bounded from above by the number of fold points of the slow curve. In the present paper it is proved that there are no such geometric constraints for non-integer rotation numbers: it is possible to construct generic system with as simple as possible slow curve and arbitrary many limit cycles.

math.DS

On the adjacency quantization in the equation modelling the Josephson effect

We investigate two-parametric family of non-autonomous ordinary differential equations on the two-torus $$\dot x=\frac{dx}{dt}=ν\sin x + a + s \sin t, \ a,ν,s\in\rr; \ ν\neq0 \text {is fixed},$$ that model the Josephson effect from superconductivity. We study its rotation number as a function of parameters $(a,s)$ and its {\it Arnold tongues}: the level sets of the rotation number that have non-empty interior. Its Arnold tongues have many non-typical properties: they exist only for integer rotation numbers (V.M.Buchstaber, O.V.Karpov, S.I.Tertychnyi (2010); Yu.S.Ilyashenko, D.A.Ryzhov, D.A.Filimonov (2011)); their boundaries are given by pairs of analytic curves (V.M.Buchstaber, O.V.Karpov, S.I.Tertychnyi (2004, 2012)). Numerical experiments and theoretical investigations (V.M.Buchstaber, O.V.Karpov, S.I.Tertychnyi (2006); A.V.Klimenko and O.L.Romaskevich (2012)) show that each Arnold tongue forms an infinite chain of adjacent domains separated by adjacency points and going to infinity in asymptotically vertical direction. Recent numerical experiments had also shown that the adjacencies of each Arnold tongue have one and the same integer abscissa $a$ equal to the corresponding rotation number. We prove this fact for every fixed $ν$ with $|ν|\leq1$. In the general case we prove a weaker statement: the abscissa of each adjacency point is integer; it has the same sign, as the rotation number; its modulus is no greater than that of the rotation number. The proof is based on the representation of the differential equations under consideration as projectivizations of complex linear differential equations on the Riemann sphere (V.M.Buchstaber, O.V.Karpov, S.I.Tertychnyi (2004); R.L.Foote (1998); Yu.S.Ilyashenko, D.A.Ryzhov, D.A.Filimonov (2011)), and the classical theory of complex linear equations.

math.DS

Ducks on the torus: existence and uniqueness

We show that there exist generic slow-fast systems with only one (time-scaling) parameter on the two-torus, which have canard cycles for arbitrary small values of this parameter. This is in drastic contrast with the planar case, where canards usually occur in two-parametric families. Here we treat systems with a convex slow curve. In this case there is a set of parameter values accumulating to zero for which the system has exactly one attracting and one repelling canard cycle. The basin of the attracting cycle is almost the whole torus.

math.DS