-
On the model attractor in a high-dimensional neural network dynamics of reservoir computing
Authors:
Miki U. Kobayashi,
Kengo Nakai,
Yoshitaka Saiki,
Natsuki Tsutsumi
Abstract:
The theory of embedding and generalized synchronization in reservoir computing has recently been developed. Under ideal conditions, reservoir computing exhibits generalized synchronization during the learning process. These insights form a rigorous basis for understanding reservoir computing ability to reconstruct and predict complex dynamics. In this study, we clarified the dynamical system struc…
▽ More
The theory of embedding and generalized synchronization in reservoir computing has recently been developed. Under ideal conditions, reservoir computing exhibits generalized synchronization during the learning process. These insights form a rigorous basis for understanding reservoir computing ability to reconstruct and predict complex dynamics. In this study, we clarified the dynamical system structures of generalized synchronization and embedding by comparing the Lyapunov exponents of a high dimensional neural network within the reservoir computing model with those in actual systems. Furthermore, we numerically calculated the Lyapunov exponents restricted to the tangent space of the inertial manifold in a high dimensional neural network. Our results demonstrate that all Lyapunov exponents of the actual dynamics, including negative ones, are successfully identified.
△ Less
Submitted 16 September, 2025;
originally announced September 2025.
-
A laminar chaotic saddle within a turbulent attractor
Authors:
Hibiki Kato,
Miki U Kobayashi,
Yoshitaka Saiki,
James A. Yorke
Abstract:
Intermittent switchings between weakly chaotic (laminar) and strongly chaotic (bursty) states are often observed in systems with high-dimensional chaotic attractors, such as fluid turbulence. They differ from the intermittency of a low-dimensional system accompanied by the stability change of a fixed point or a periodic orbit in that the intermittency of a high-dimensional system tends to appear i…
▽ More
Intermittent switchings between weakly chaotic (laminar) and strongly chaotic (bursty) states are often observed in systems with high-dimensional chaotic attractors, such as fluid turbulence. They differ from the intermittency of a low-dimensional system accompanied by the stability change of a fixed point or a periodic orbit in that the intermittency of a high-dimensional system tends to appear in a wide range of parameters. This paper considers a case where the skeleton of a laminar state $L$ exists as a proper chaotic subset $S$ of a chaotic attractor $X$, that is, $S\ \subsetneq\ X$. We characterize such a laminar state $L$ by a chaotic saddle $S$, which is densely filled with periodic orbits of different numbers of unstable directions. This study demonstrates the presence of chaotic saddles underlying intermittency in fluid turbulence and phase synchronization. Furthermore, we confirm that chaotic saddles persist for a wide range of parameters. Also, a kind of phase synchronization turns out to occur in the turbulent model.
△ Less
Submitted 13 September, 2024;
originally announced September 2024.
-
The dynamics of the heterochaos baker maps
Authors:
Yoshitaka Saiki,
Hiroki Takahasi,
Kenichiro Yamamoto,
James A. Yorke
Abstract:
The heterochaos baker maps are piecewise affine maps of the unit square or cube introduced in [Nonlinearity 34, 2021, 5744--5761], to provide a hands-on, elementary understanding of complicated phenomena in systems of large degrees of freedom. We review recent progress on a dynamical systems theory of the heterochaos baker maps, and present new results on properties of measures of maximal entropy…
▽ More
The heterochaos baker maps are piecewise affine maps of the unit square or cube introduced in [Nonlinearity 34, 2021, 5744--5761], to provide a hands-on, elementary understanding of complicated phenomena in systems of large degrees of freedom. We review recent progress on a dynamical systems theory of the heterochaos baker maps, and present new results on properties of measures of maximal entropy and the underlying Lebesgue measure. We address several conjectures and questions that may illuminate new aspects of heterochaos and inspire future research.
△ Less
Submitted 13 September, 2024; v1 submitted 1 January, 2024;
originally announced January 2024.
-
Constructing low-dimensional ordinary differential equations from chaotic time series of high/infinite-dimensional systems using radial function-based regression
Authors:
Natsuki Tsutsumi,
Kengo Nakai,
Yoshitaka Saiki
Abstract:
In our previous study (N. Tsutsumi, K. Nakai and Y. Saiki (2022)) we proposed a method of constructing a system of differential equations of chaotic behavior only from observable deterministic time series, which we will call radial function-based regression (RfR) method. The RfR method employs a regression using Gaussian radial basis functions together with polynomial terms to facilitate the robus…
▽ More
In our previous study (N. Tsutsumi, K. Nakai and Y. Saiki (2022)) we proposed a method of constructing a system of differential equations of chaotic behavior only from observable deterministic time series, which we will call radial function-based regression (RfR) method. The RfR method employs a regression using Gaussian radial basis functions together with polynomial terms to facilitate the robust modeling of chaotic behavior. In this paper, we apply the RfR method to several types of relatively high-dimensional deterministic time series generated by a partial differential equation, a delay differential equation, a turbulence model, and intermittent dynamics. The case when the observation includes noise is also tested. We have effectively constructed a system of differential equations for each of these examples, which is assessed from the point of view of time series forecast, reconstruction of invariant sets, and invariant densities. We find that in some of the models, an appropriate trajectory is realized on the chaotic saddle and is identified by the Stagger-and-Step method.
△ Less
Submitted 12 August, 2023;
originally announced August 2023.
-
Machine learning prediction of the MJO extends beyond one month
Authors:
Tamaki Suematsu,
Kengo Nakai,
Tsuyoshi Yoneda,
Daisuke Takasuka,
Takuya Jinno,
Yoshitaka Saiki,
Hiroaki Miura
Abstract:
The prediction of the Madden-Julian Oscillation (MJO), a massive tropical weather event with vast global socio-economic impacts, has been infamously difficult with physics-based weather prediction models. Here we construct a machine learning model using reservoir computing technique that forecasts the real-time multivariate MJO index (RMM), a macroscopic variable that represents the state of the M…
▽ More
The prediction of the Madden-Julian Oscillation (MJO), a massive tropical weather event with vast global socio-economic impacts, has been infamously difficult with physics-based weather prediction models. Here we construct a machine learning model using reservoir computing technique that forecasts the real-time multivariate MJO index (RMM), a macroscopic variable that represents the state of the MJO. The training data was refined by developing a novel filter that extracts the recurrency of MJO signals from the raw atmospheric data and selecting a suitable time-delay coordinate of the RMM. The model demonstrated the skill to forecast the state of MJO events for a month from the pre-developmental stages. Best-performing cases predicted the RMM sequence over two months, which exceeds the expected inherent predictability limit of the MJO.
△ Less
Submitted 29 December, 2022;
originally announced January 2023.
-
Constructing differential equations using only a scalar time-series about continuous time chaotic dynamics
Authors:
Natsuki Tsutsumi,
Kengo Nakai,
Yoshitaka Saiki
Abstract:
We propose a simple method of constructing a system of differential equations of chaotic behavior based on the regression only from a scalar observable time-series data. The estimated system enables us to reconstruct invariant sets and statistical properties as well as to infer short time-series. Our successful modeling relies on the introduction of a set of Gaussian radial basis functions to capt…
▽ More
We propose a simple method of constructing a system of differential equations of chaotic behavior based on the regression only from a scalar observable time-series data. The estimated system enables us to reconstruct invariant sets and statistical properties as well as to infer short time-series. Our successful modeling relies on the introduction of a set of Gaussian radial basis functions to capture local structure. The proposed method is used to construct a system of ordinary differential equations whose orbit reconstructs a time-series of a variable of the well-known Lorenz system as a simple but typical example. A system for a macroscopic fluid variable is also constructed.
△ Less
Submitted 7 August, 2022;
originally announced September 2022.
-
The twisted baker map
Authors:
Yoshitaka Saiki,
Hiroki Takahasi,
James A. Yorke
Abstract:
As a model to provide a hands-on, elementary understanding of "vortex dynamics", we introduce a piecewise linear non-invertible map called a twisted baker map. We show that the set of hyperbolic repelling periodic points with complex conjugate eigenvalues and that without complex conjugate eigenvalues are simultaneously dense in the phase space. We also show that these two sets equidistribute with…
▽ More
As a model to provide a hands-on, elementary understanding of "vortex dynamics", we introduce a piecewise linear non-invertible map called a twisted baker map. We show that the set of hyperbolic repelling periodic points with complex conjugate eigenvalues and that without complex conjugate eigenvalues are simultaneously dense in the phase space. We also show that these two sets equidistribute with respect to the normalized Lebesgue measure, in spite of a non-uniformity in their Lyapunov exponents.
△ Less
Submitted 13 January, 2023; v1 submitted 9 February, 2022;
originally announced February 2022.
-
Dynamical system analysis of a data-driven model constructed by reservoir computing
Authors:
Miki U Kobayashi,
Kengo Nakai,
Yoshitaka Saiki,
Natsuki Tsutsumi
Abstract:
This study evaluates data-driven models from a dynamical system perspective, such as unstable fixed points, periodic orbits, chaotic saddle, Lyapunov exponents, manifold structures, and statistical values. We find that these dynamical characteristics can be reconstructed much more precisely by a data-driven model than by computing directly from training data. With this idea, we predict the laminar…
▽ More
This study evaluates data-driven models from a dynamical system perspective, such as unstable fixed points, periodic orbits, chaotic saddle, Lyapunov exponents, manifold structures, and statistical values. We find that these dynamical characteristics can be reconstructed much more precisely by a data-driven model than by computing directly from training data. With this idea, we predict the laminar lasting time distribution of a particular macroscopic variable of chaotic fluid flow, which cannot be calculated from a direct numerical simulation of the Navier-Stokes equation because of its high computational cost.
△ Less
Submitted 1 October, 2021; v1 submitted 25 February, 2021;
originally announced February 2021.
-
Hausdorff dimension of Cantor intersections and robust heterodimensional cycles for heterochaos horseshoe maps
Authors:
Yoshitaka Saiki,
Hiroki Takahasi,
James A. Yorke
Abstract:
As a model to provide a hands-on, elementary understanding of chaotic dynamics in dimension three, we introduce a $C^2$-open set of diffeomorphisms of $\mathbb R^3$ having two horseshoes with different dimensions of instability. We prove that: the unstable set of one horseshoe and the stable set of the other are of Hausdorff dimension nearly $2$ whose cross sections are Cantor sets; the intersecti…
▽ More
As a model to provide a hands-on, elementary understanding of chaotic dynamics in dimension three, we introduce a $C^2$-open set of diffeomorphisms of $\mathbb R^3$ having two horseshoes with different dimensions of instability. We prove that: the unstable set of one horseshoe and the stable set of the other are of Hausdorff dimension nearly $2$ whose cross sections are Cantor sets; the intersection of the unstable and stable sets contains a fractal set of Hausdorff dimension nearly $1$. As a corollary we detect $C^2$-robust heterodimensional cycles. Our proof employs the theory of normally hyperbolic invariant manifolds and the thicknesses of Cantor sets.
△ Less
Submitted 12 February, 2023; v1 submitted 24 November, 2020;
originally announced November 2020.
-
Quasiperiodic orbits in Siegel disks/balls and the Babylonian problem
Authors:
Yoshitaka Saiki,
James A. Yorke
Abstract:
We investigate numerically complex dynamical systems where a fixed point is surrounded by a disk or ball of quasiperiodic orbits, where there is a change of variables (or conjugacy) that converts the system into a linear map. We compute this "linearization" (or conjugacy) from knowledge of a single quasiperiodic trajectory. In our computations of rotation rates of the almost periodic orbits and Fo…
▽ More
We investigate numerically complex dynamical systems where a fixed point is surrounded by a disk or ball of quasiperiodic orbits, where there is a change of variables (or conjugacy) that converts the system into a linear map. We compute this "linearization" (or conjugacy) from knowledge of a single quasiperiodic trajectory. In our computations of rotation rates of the almost periodic orbits and Fourier coefficients of the conjugacy, we only use knowledge of a trajectory, and we do not assume knowledge of the explicit form of a dynamical system. This problem is called the Babylonian Problem: determining the characteristics of a quasiperiodic set from a trajectory. Our computation of rotation rates and Fourier coefficients depends on the very high speed of our computational method "the weighted Birkhoff average".
△ Less
Submitted 7 November, 2018;
originally announced November 2018.
-
Piecewise-linear maps with heterogeneous chaos
Authors:
Yoshitaka Saiki,
Hiroki Takahasi,
James A. Yorke
Abstract:
Chaotic dynamics can be quite heterogeneous in the sense that in some regions the dynamics are unstable in more directions than in other regions. When trajectories wander between these regions, the dynamics is complicated. We say a chaotic invariant set is heterogeneous when arbitrarily close to each point of the set there are different periodic points with different numbers of unstable dimensions…
▽ More
Chaotic dynamics can be quite heterogeneous in the sense that in some regions the dynamics are unstable in more directions than in other regions. When trajectories wander between these regions, the dynamics is complicated. We say a chaotic invariant set is heterogeneous when arbitrarily close to each point of the set there are different periodic points with different numbers of unstable dimensions. We call such dynamics heterogeneous chaos (or hetero-chaos), While we believe it is common for physical systems to be hetero-chaotic, few explicit examples have been proved to be hetero-chaotic. Here we present two more explicit dynamical systems that are particularly simple and tractable with computer. It will give more intuition as to how complex even simple systems can be. Our maps have one dense set of periodic points whose orbits are 1D unstable and another dense set of periodic points whose orbits are 2D unstable. Moreover, they are ergodic relative to the Lebesgue measure.
△ Less
Submitted 7 October, 2022; v1 submitted 2 November, 2018;
originally announced November 2018.
-
Solving the Babylonian Problem of quasiperiodic rotation rates
Authors:
Suddhasattwa Das,
Yoshitaka Saiki,
Evelyn Sander,
James A Yorke
Abstract:
A trajectory $u_n := F^n(u_0), n = 0,1,2, \dots $ is quasiperiodic if the trajectory lies on and is dense in some $d$-dimensional torus, and there is a choice of coordinates on the torus $\mathbb{T}$ for which $F$ has the form $F(θ) = θ+ ρ\bmod1$ for all $θ\in\mathbb{T}$ and for some $ρ\in\mathbb{T}$. There is an ancient literature on computing three rotation rates $ρ$ for the Moon. %There is a li…
▽ More
A trajectory $u_n := F^n(u_0), n = 0,1,2, \dots $ is quasiperiodic if the trajectory lies on and is dense in some $d$-dimensional torus, and there is a choice of coordinates on the torus $\mathbb{T}$ for which $F$ has the form $F(θ) = θ+ ρ\bmod1$ for all $θ\in\mathbb{T}$ and for some $ρ\in\mathbb{T}$. There is an ancient literature on computing three rotation rates $ρ$ for the Moon. %There is a literature on determining the coordinates of the vector $ρ$, called the rotation rates of $F$. (For $d>1$ we always interpret $\bmod1$ as being applied to each coordinate.) However, even in the case $d=1$ there has been no general method for computing $ρ$ given only the trajectory $u_n$, though there is a literature dealing with special cases. Here we present our Embedding Continuation Method for computing some components of $ρ$ from a trajectory. It is based on the Takens Embedding Theorem and the Birkhoff Ergodic Theorem. Rotation rates are often called "rotation numbers" and both refer to a rate of rotation of a circle. However, the coordinates of $ρ$ depend on the choice of coordinates of $\mathbb{T}$. We explore the various sets of possible rotation rates that $ρ$ can yield. We illustrate our ideas with examples in dimensions $d=1$ and $2$.
△ Less
Submitted 13 July, 2017; v1 submitted 8 June, 2017;
originally announced June 2017.
-
Quantitative Quasiperiodicity
Authors:
Suddhasattwa Das,
Yoshitaka Saiki,
Evelyn Sander,
James A Yorke
Abstract:
The Birkhoff Ergodic Theorem concludes that time averages, i.e., Birkhoff averages, $Σ_{n=0}^{N-1} f(x_n)/N$ of a function $f$ along a length $N$ ergodic trajectory $(x_n)$ of a function $T$ converge to the space average $\int f dμ$, where $μ$ is the unique invariant probability measure. Convergence of the time average to the space average is slow. We introduce a modified average of $f(x_n)$ by gi…
▽ More
The Birkhoff Ergodic Theorem concludes that time averages, i.e., Birkhoff averages, $Σ_{n=0}^{N-1} f(x_n)/N$ of a function $f$ along a length $N$ ergodic trajectory $(x_n)$ of a function $T$ converge to the space average $\int f dμ$, where $μ$ is the unique invariant probability measure. Convergence of the time average to the space average is slow. We introduce a modified average of $f(x_n)$ by giving very small weights to the "end" terms when $n$ is near $0$ or $N-1$. When $(x_n)$ is a trajectory on a quasiperiodic torus and $f$ and $T$ are $C^\infty$, we show that our weighted Birkhoff averages converge 'super" fast to $\int f dμ$ with respect to the number of iterates $N$, i.e. with error decaying faster than $N^{-m}$ for every integer $m$. Our goal is to show that our weighted Birkhoff average is a powerful computational tool, and this paper illustrates its use for several examples where the quasiperiodic set is one or two dimensional. In particular, we compute rotation numbers and conjugacies (i.e. changes of variables) and their Fourier series, often with 30-digit accuracy.
△ Less
Submitted 31 July, 2017; v1 submitted 22 January, 2016;
originally announced January 2016.
-
Quantitative Quasiperiodicity
Authors:
Suddhasattwa Das,
Yoshitaka Saiki,
Evelyn Sander,
James A. Yorke
Abstract:
The Birkhoff Ergodic Theorem concludes that time averages, that is, Birkhoff averages, $Σ_{n=1}^N f(x_n)/N$ of a function $f$ along an ergodic trajectory $(x_n)$ of a function $T$ converges to the space average $\int f dμ$, where $μ$ is the unique invariant probability measure. Convergence of the time average to the space average is slow. We introduce a modified average of $f(x_n)$ by giving very…
▽ More
The Birkhoff Ergodic Theorem concludes that time averages, that is, Birkhoff averages, $Σ_{n=1}^N f(x_n)/N$ of a function $f$ along an ergodic trajectory $(x_n)$ of a function $T$ converges to the space average $\int f dμ$, where $μ$ is the unique invariant probability measure. Convergence of the time average to the space average is slow. We introduce a modified average of $f(x_n)$ by giving very small weights to the "end" terms when $n$ is near $0$ or $N$. When $(x_n)$ is a trajectory on a quasiperiodic torus and $f$ and $T$ are $C^\infty$, we show that our weighted Birkhoff averages converge "super" fast to $\int f dμ$, {\em i.e.} with error smaller than every polynomial of $1/N$. Our goal is to show that our weighted Birkhoff average is a powerful computational tool, and this paper illustrates its use for several examples where the quasiperiodic set is one or two dimensional. In particular, we compute rotation numbers and conjugacies (i.e. changes of variables) and their Fourier series, often with 30-digit precision.
△ Less
Submitted 31 July, 2015;
originally announced August 2015.
