-
Optimal regularizations for data generation with probabilistic graphical models
Authors:
Arnaud Fanthomme,
F Rizzato,
S Cocco,
R Monasson
Abstract:
Understanding the role of regularization is a central question in Statistical Inference. Empirically, well-chosen regularization schemes often dramatically improve the quality of the inferred models by avoiding overfitting of the training data. We consider here the particular case of L 2 and L 1 regularizations in the Maximum A Posteriori (MAP) inference of generative pairwise graphical models. Ba…
▽ More
Understanding the role of regularization is a central question in Statistical Inference. Empirically, well-chosen regularization schemes often dramatically improve the quality of the inferred models by avoiding overfitting of the training data. We consider here the particular case of L 2 and L 1 regularizations in the Maximum A Posteriori (MAP) inference of generative pairwise graphical models. Based on analytical calculations on Gaussian multivariate distributions and numerical experiments on Gaussian and Potts models we study the likelihoods of the training, test, and 'generated data' (with the inferred models) sets as functions of the regularization strengths. We show in particular that, at its maximum, the test likelihood and the 'generated' likelihood, which quantifies the quality of the generated samples, have remarkably close values. The optimal value for the regularization strength is found to be approximately equal to the inverse sum of the squared couplings incoming on sites on the underlying network of interactions. Our results seem largely independent of the structure of the true underlying interactions that generated the data, of the regularization scheme considered, and are valid when small fluctuations of the posterior distribution around the MAP estimator are taken into account. Connections with empirical works on protein models learned from homologous sequences are discussed.
△ Less
Submitted 2 December, 2021;
originally announced December 2021.
-
Inference of compressed Potts graphical models
Authors:
Francesca Rizzato,
Alice Coucke,
Eleonora de Leonardis,
J. P. Barton,
Jérôme Tubiana,
Remi Monasson,
Simona Cocco
Abstract:
We consider the problem of inferring a graphical Potts model on a population of variables, with a non-uniform number of Potts colors (symbols) across variables. This inverse Potts problem generally involves the inference of a large number of parameters, often larger than the number of available data, and, hence, requires the introduction of regularization. We study here a double regularization sch…
▽ More
We consider the problem of inferring a graphical Potts model on a population of variables, with a non-uniform number of Potts colors (symbols) across variables. This inverse Potts problem generally involves the inference of a large number of parameters, often larger than the number of available data, and, hence, requires the introduction of regularization. We study here a double regularization scheme, in which the number of colors available to each variable is reduced, and interaction networks are made sparse. To achieve this color compression scheme, only Potts states with large empirical frequency (exceeding some threshold) are explicitly modeled on each site, while the others are grouped into a single state. We benchmark the performances of this mixed regularization approach, with two inference algorithms, the Adaptive Cluster Expansion (ACE) and the PseudoLikelihood Maximization (PLM) on synthetic data obtained by sampling disordered Potts models on an Erdos-Renyi random graphs. We show in particular that color compression does not affect the quality of reconstruction of the parameters corresponding to high-frequency symbols, while drastically reducing the number of the other parameters and thus the computational time. Our procedure is also applied to multi-sequence alignments of protein families, with similar results.
△ Less
Submitted 3 January, 2020; v1 submitted 30 July, 2019;
originally announced July 2019.
-
Nonequilibrium Statistical Mechanics of Systems with Long-Range Interactions: Ubiquity of Core-Halo Distributions
Authors:
Yan Levin,
Renato Pakter,
Felipe B. Rizzato,
Tarcísio N. Teles,
Fernanda P. da C. Benetti
Abstract:
Systems with long-range (LR) forces, for which the interaction potential decays with the interparticle distance with an exponent smaller than the dimensionality of the embedding space, remain an outstanding challenge to statistical physics. The internal energy of such systems lacks extensivity and additivity. Although the extensivity can be restored by scaling the interaction potential with the nu…
▽ More
Systems with long-range (LR) forces, for which the interaction potential decays with the interparticle distance with an exponent smaller than the dimensionality of the embedding space, remain an outstanding challenge to statistical physics. The internal energy of such systems lacks extensivity and additivity. Although the extensivity can be restored by scaling the interaction potential with the number of particles, the non-additivity still remains. Lack of additivity leads to inequivalence of statistical ensembles. Before relaxing to thermodynamic equilibrium, isolated systems with LR forces become trapped in out-of-equilibrium quasi-stationary state (qSS), the lifetime of which diverges with the number of particles. Therefore, in thermodynamic limit LR systems will not relax to equilibrium. The qSSs are attained through the process of collisionless relaxation. Density oscillations lead to particle-wave interactions and excitation of parametric resonances. The resonant particles escape from the main cluster to form a tenuous halo. Simultaneously, this cools down the core of the distribution and dampens out the oscillations. When all the oscillations die out the ergodicity is broken and a qSS is born. In this report, we will review a theory which allows us to quantitatively predict the particle distribution in the qSS. The theory is applied to various LR interacting systems, ranging from plasmas to self-gravitating clusters and kinetic spin models.
△ Less
Submitted 3 October, 2013;
originally announced October 2013.
-
Alternate islands of multiple isochronous chains in wave-particle interactions
Authors:
M. C. de Sousa,
I. L. Caldas,
A. M. Ozorio de Almeida,
F. B. Rizzato,
R. Pakter
Abstract:
We analyze the dynamics of a relativistic particle moving in a uniform magnetic field and perturbed by a standing electrostatic wave. We show that a pulsed wave produces an infinite number of perturbative terms with the same winding number, which may generate islands in the same region of phase space. As a consequence, the number of isochronous island chains varies as a function of the wave parame…
▽ More
We analyze the dynamics of a relativistic particle moving in a uniform magnetic field and perturbed by a standing electrostatic wave. We show that a pulsed wave produces an infinite number of perturbative terms with the same winding number, which may generate islands in the same region of phase space. As a consequence, the number of isochronous island chains varies as a function of the wave parameters. We observe that in all the resonances, the number of chains is related to the amplitude of the various resonant terms. We determine analytically the position of the periodic points and the number of island chains as a function of the wave number and wave period. Such information is very important when one is concerned with regular particle acceleration, since it is necessary to adjust the initial conditions of the particle to obtain the maximum acceleration.
△ Less
Submitted 14 November, 2013; v1 submitted 11 June, 2013;
originally announced June 2013.
-
Controlling chaos in wave-particle interactions
Authors:
M. C. de Sousa,
I. L. Caldas,
F. B. Rizzato,
R. Pakter,
F. M. Steffens
Abstract:
We analyze the behavior of a relativistic particle moving under the influence of a uniform magnetic field and a stationary electrostatic wave. We work with a set of pulsed waves that allows us to obtain an exact map for the system. We also use a method of control for near-integrable Hamiltonians that consists in the addition of a small and simple control term to the system. This control term creat…
▽ More
We analyze the behavior of a relativistic particle moving under the influence of a uniform magnetic field and a stationary electrostatic wave. We work with a set of pulsed waves that allows us to obtain an exact map for the system. We also use a method of control for near-integrable Hamiltonians that consists in the addition of a small and simple control term to the system. This control term creates invariant tori in phase space that prevent chaos from spreading to large regions and make the controlled dynamics more regular. We show numerically that the control term just slightly modifies the system but is able to drastically reduce chaos with a low additional cost of energy. Moreover, we discuss how the control of chaos and the consequent recovery of regular trajectories in phase space are useful to improve regular particle acceleration.
△ Less
Submitted 31 July, 2012; v1 submitted 9 October, 2011;
originally announced October 2011.
-
Standard map in magnetized relativistic systems: fixed points and regular acceleration
Authors:
M. C. de Sousa,
F. M. Steffens,
R. Pakter,
F. B. Rizzato
Abstract:
We investigate the concept of a standard map for the interaction of relativistic particles and electrostatic waves of arbitrary amplitudes, under the action of external magnetic fields. The map is adequate for physical settings where waves and particles interact impulsively, and allows for a series of analytical result to be exactly obtained. Unlike the traditional form of the standard map, the pr…
▽ More
We investigate the concept of a standard map for the interaction of relativistic particles and electrostatic waves of arbitrary amplitudes, under the action of external magnetic fields. The map is adequate for physical settings where waves and particles interact impulsively, and allows for a series of analytical result to be exactly obtained. Unlike the traditional form of the standard map, the present map is nonlinear in the wave amplitude and displays a series of peculiar properties. Among these properties we discuss the relation involving fixed points of the maps and accelerator regimes.
△ Less
Submitted 29 July, 2010;
originally announced July 2010.
-
Statistical Mechanics of Unbound Two Dimensional Self-Gravitating Systems
Authors:
Tarcísio N. Teles,
Yan Levin,
Renato Pakter,
Felipe B. Rizzato
Abstract:
We study, using both theory and molecular dynamics simulations, the relaxation dynamics of a microcanonical two dimensional self-gravitating system. After a sufficiently large time, a gravitational cluster of N particles relaxes to the Maxwell-Boltzmann distribution. The time to reach the thermodynamic equilibrium, however, scales with the number of particles. In the thermodynamic limit,…
▽ More
We study, using both theory and molecular dynamics simulations, the relaxation dynamics of a microcanonical two dimensional self-gravitating system. After a sufficiently large time, a gravitational cluster of N particles relaxes to the Maxwell-Boltzmann distribution. The time to reach the thermodynamic equilibrium, however, scales with the number of particles. In the thermodynamic limit, $N\to\infty$ at fixed total mass, equilibrium state is never reached and the system becomes trapped in a non-ergodic stationary state. An analytical theory is presented which allows us to quantitatively described this final stationary state, without any adjustable parameters.
△ Less
Submitted 19 April, 2010; v1 submitted 1 April, 2010;
originally announced April 2010.
-
Thermodynamics of driven collisionless systems
Authors:
Felipe B. Rizzato,
Renato Pakter,
Yan Levin
Abstract:
A statistical theory is presented which allows to calculate the stationary state achieved by a driven system after a process of collisionless relaxation. The theory is applied to study an electron beam driven by an external electric field. The Vlasov equation with appropriate boundary conditions is solved analytically and compared with the molecular dynamics simulation. A perfect agreement is fo…
▽ More
A statistical theory is presented which allows to calculate the stationary state achieved by a driven system after a process of collisionless relaxation. The theory is applied to study an electron beam driven by an external electric field. The Vlasov equation with appropriate boundary conditions is solved analytically and compared with the molecular dynamics simulation. A perfect agreement is found between the theory and the simulations. The full current-voltage phase diagram is constructed.
△ Less
Submitted 1 April, 2009;
originally announced April 2009.
-
Collisionless relaxation in gravitational systems: From violent relaxation to gravothermal collapse
Authors:
Yan Levin,
Renato Pakter,
Felipe B. Rizzato
Abstract:
Theory and simulations are used to study collisionless relaxation of a gravitational $N$-body system. It is shown that when the initial one particle distribution function satisfies the virial condition -- potential energy is minus twice the kinetic energy -- the system quickly relaxes to a metastable state described {\it quantitatively} by the Lynden-Bell distribution with a cutoff. If the initi…
▽ More
Theory and simulations are used to study collisionless relaxation of a gravitational $N$-body system. It is shown that when the initial one particle distribution function satisfies the virial condition -- potential energy is minus twice the kinetic energy -- the system quickly relaxes to a metastable state described {\it quantitatively} by the Lynden-Bell distribution with a cutoff. If the initial distribution function does not meet the virial requirement, the system undergoes violent oscillations, resulting in a partial evaporation of mass. The leftover particles phase separate into a core-halo structure. The theory presented allows us to quantitatively predict the amount and the distribution of mass left in the central core, without any adjustable parameters. On a longer time scale $τ_G \sim N$ collisionless relaxation leads to a gravothermal collapse.
△ Less
Submitted 22 August, 2008;
originally announced August 2008.
-
Coherence and incoherence in extended broad band triplet interaction
Authors:
G. I. de Oliveira,
F. B. Rizzato
Abstract:
In the present analysis we study the transition from coherent to incoherent dynamics in a nonlinear triplet of broad band combs of waves. Expanding the analysis of previous works, this paper investigates what happens when the band of available modes is much larger than that of the initial narrower combs within which the nonlinear interaction is not subjected to selection rules involving wave mom…
▽ More
In the present analysis we study the transition from coherent to incoherent dynamics in a nonlinear triplet of broad band combs of waves. Expanding the analysis of previous works, this paper investigates what happens when the band of available modes is much larger than that of the initial narrower combs within which the nonlinear interaction is not subjected to selection rules involving wave momenta. Here selection rules are present and active, and we examine how and when coherence can be defined.
△ Less
Submitted 31 January, 2008;
originally announced January 2008.
-
Image Effects on the Transport of Intense Nonaxisymmetric Charged Beams
Authors:
Renato Pakter,
Yan Levin,
Felipe B. Rizzato
Abstract:
The effect of conducting pipes on the equilibrium of intense nonaxisymmetric continuous beams of charged particles is investigated. For a cylindrical pipe and an elliptical beam, we obtain an exact closed form analytical expression for the electrostatic potential. Using a variational principle, we then explore the distortions that equilibrium beams suffer due to the conducting channel. Finally,…
▽ More
The effect of conducting pipes on the equilibrium of intense nonaxisymmetric continuous beams of charged particles is investigated. For a cylindrical pipe and an elliptical beam, we obtain an exact closed form analytical expression for the electrostatic potential. Using a variational principle, we then explore the distortions that equilibrium beams suffer due to the conducting channel. Finally, we present an exact proof that despite the nonlinear forces acting on beams of arbitrary cross section inside conducting pipes of arbitrary shape, the density of these beams remains homogeneous and their cross sectional area remains the sa me as the one in free-space.
△ Less
Submitted 14 December, 2007;
originally announced December 2007.
-
Wave breaking and particle jets in intense inhomogeneous charged beams
Authors:
Felipe B. Rizzato,
Renato Pakter,
Yan Levin
Abstract:
This work analyzes the dynamics of inhomogeneous, magnetically focused high-intensity beams of charged particles. While for homogeneous beams the whole system oscillates with a single frequency, any inhomogeneity leads to propagating transverse density waves which eventually result in a singular density build up, causing wave breaking and jet formation. The theory presented in this paper allows…
▽ More
This work analyzes the dynamics of inhomogeneous, magnetically focused high-intensity beams of charged particles. While for homogeneous beams the whole system oscillates with a single frequency, any inhomogeneity leads to propagating transverse density waves which eventually result in a singular density build up, causing wave breaking and jet formation. The theory presented in this paper allows to analytically calculate the time at which the wave breaking takes place. It also gives a good estimate of the time necessary for the beam to relax into the final stationary state consisting of a cold core surrounded by a halo of highly energetic particles.
△ Less
Submitted 19 October, 2007;
originally announced October 2007.
-
Superconducting pipes and levitating magnets
Authors:
Yan Levin,
Felipe B. Rizzato
Abstract:
Motivated by a beautiful demonstration of the Faraday's and Lenz's law in which a small neodymium magnet falls slowly through a conducting non-ferromagnetic tube, we consider the dynamics of a magnet falling through a superconducting pipe. Unlike the case of normal conducting pipes, in which the magnet quickly reaches the terminal velocity, inside a superconducting tube the magnet falls freely.…
▽ More
Motivated by a beautiful demonstration of the Faraday's and Lenz's law in which a small neodymium magnet falls slowly through a conducting non-ferromagnetic tube, we consider the dynamics of a magnet falling through a superconducting pipe. Unlike the case of normal conducting pipes, in which the magnet quickly reaches the terminal velocity, inside a superconducting tube the magnet falls freely. On the other hand, to enter the pipe the magnet must overcome a large electromagnetic energy barrier. For sufficiently strong magnets, the barrier is so large that the magnet will not be able to penetrate it and will be suspended over the front edge. We calculate the work that must done to force the magnet to enter a superconducting tube. The calculations show that superconducting pipes are very efficient at screening magnetic fields. For example, the magnetic field of a dipole at the center of a short pipe of radius $a$ and length $L \approx a$ decays, in the axial direction, with a characteristic length $ξ\approx 0.26 a$. The efficient screening of the magnetic field might be useful for shielding highly sensitive superconducting quantum interference devices, SQUIDs. Finally, the motion of the magnet through a superconducting pipe is compared and contrasted to the flow of ions through a trans-membrane channel.
△ Less
Submitted 15 September, 2006;
originally announced September 2006.
-
Electromagnetic braking: a simple quantitative model
Authors:
Yan Levin,
Fernando L. da Silveira,
Felipe B. Rizzato
Abstract:
A calculation is presented which quantitatively accounts for the terminal velocity of a cylindrical magnet falling through a long copper or aluminum pipe. The experiment and the theory are a dramatic illustration of the Faraday's and Lenz's laws and are bound to capture student's attention in any electricity and magnetism course.
A calculation is presented which quantitatively accounts for the terminal velocity of a cylindrical magnet falling through a long copper or aluminum pipe. The experiment and the theory are a dramatic illustration of the Faraday's and Lenz's laws and are bound to capture student's attention in any electricity and magnetism course.
△ Less
Submitted 6 September, 2006; v1 submitted 31 March, 2006;
originally announced March 2006.
-
Stability of off-axis motion for intense particle beams in periodically focusing channels
Authors:
Jorge S. Moraes,
Renato Pakter,
Felipe B. Rizzato
Abstract:
A general equation for the centroid motion of free, continuous, intense beams propagating off-axis in solenoidal periodic focusing fields is derived. The centroid equation is found to be independent of the specific beam distribution and may exhibit unstable solutions. A new Vlasov equilibrium for off-axis beam propagation is also obtained. Properties of the equilibrium and the relevance of centr…
▽ More
A general equation for the centroid motion of free, continuous, intense beams propagating off-axis in solenoidal periodic focusing fields is derived. The centroid equation is found to be independent of the specific beam distribution and may exhibit unstable solutions. A new Vlasov equilibrium for off-axis beam propagation is also obtained. Properties of the equilibrium and the relevance of centroid motion to beam confinement are discussed.
△ Less
Submitted 19 October, 2004;
originally announced October 2004.
-
Nonlinear dynamics of electromagnetic pulses in cold relativistic plasmas
Authors:
Alexandre Bonatto,
Renato Pakter,
Felipe Rizzato
Abstract:
In the present analysis we study the self consistent propagation of nonlinear electromagnetic pulses in a one dimensional relativistic electron-ion plasma, from the perspective of nonlinear dynamics. We show how a series of Hamiltonian bifurcations give rise to the electric fields which are of relevance in the subject of particle acceleration. Connections between these bifurcated solutions and r…
▽ More
In the present analysis we study the self consistent propagation of nonlinear electromagnetic pulses in a one dimensional relativistic electron-ion plasma, from the perspective of nonlinear dynamics. We show how a series of Hamiltonian bifurcations give rise to the electric fields which are of relevance in the subject of particle acceleration. Connections between these bifurcated solutions and results of earlier analysis are made.
△ Less
Submitted 19 October, 2004;
originally announced October 2004.
-
Breakdown of Modulational Approximations in Nonlinear Wave Interaction
Authors:
G. J. L. Gerhardt,
M. Frichembruder,
F. B. Rizzato,
S. R. Lopes
Abstract:
In this work we investigate the validity limits of the modulational approximation as a method to describe the nonlinear interaction of conservative wave fields. We focus on a nonlinear Klein-Gordon equation and suggest that the breakdown of the approximation is accompanied by a transition to regimes of spatiotemporal chaos.
In this work we investigate the validity limits of the modulational approximation as a method to describe the nonlinear interaction of conservative wave fields. We focus on a nonlinear Klein-Gordon equation and suggest that the breakdown of the approximation is accompanied by a transition to regimes of spatiotemporal chaos.
△ Less
Submitted 26 May, 1999;
originally announced May 1999.
-
Chaotic Interaction of Langmuir Solitons and Long Wavelength Radiation
Authors:
R. Erichsen,
G. I. de Oliveira,
F. B. Rizzato
Abstract:
In this work we analyze the interaction of isolated solitary structures and ion-acoustic radiation. If the radiation amplitude is small solitary structures persists, but when the amplitude grows energy transfer towards small spatial scales occurs. We show that transfer is particularly fast when a fixed point of a low dimensional model is destroyed.
In this work we analyze the interaction of isolated solitary structures and ion-acoustic radiation. If the radiation amplitude is small solitary structures persists, but when the amplitude grows energy transfer towards small spatial scales occurs. We show that transfer is particularly fast when a fixed point of a low dimensional model is destroyed.
△ Less
Submitted 26 May, 1998;
originally announced May 1998.
-
Chaos and Energy Redistribution the Nonlinear Interaction of Two Spatio-Temporal Wave Triplets
Authors:
S. R. Lopes,
F. B. Rizzato
Abstract:
In this paper we examine the spatio-temporal dynamics of two nonlinearly coupled wave triplets sharing two common modes. Our basic findings are the following. When spatial dependence is absent, the homogeneous manifold so obtained can be chaotic or regular. If chaotic, it drives energy diffusion from long to small wavelengths as soon as inhomogeneous perturbations are added to the system. If reg…
▽ More
In this paper we examine the spatio-temporal dynamics of two nonlinearly coupled wave triplets sharing two common modes. Our basic findings are the following. When spatial dependence is absent, the homogeneous manifold so obtained can be chaotic or regular. If chaotic, it drives energy diffusion from long to small wavelengths as soon as inhomogeneous perturbations are added to the system. If regular, one may yet have two distinct cases: (i) energy diffusion is again present if the inhomogeneous modes are linearly unstable and triplets are effectively coupled; (ii) energy diffusion is absent if the inhomogeneous modes are linearly stable or the triplets are uncoupled.
△ Less
Submitted 8 January, 1998;
originally announced January 1998.
-
Separatrix Reconnections in Chaotic Regimes
Authors:
G. Corso,
F. B. Rizzato
Abstract:
In this paper we extend the concept of separatrix reconnection into chaotic regimes. We show that even under chaotic conditions one can still understand abrupt jumps of diffusive-like processes in the relevant phase-space in terms of relatively smooth realignments of stable and unstable manifolds of unstable fixed points.
In this paper we extend the concept of separatrix reconnection into chaotic regimes. We show that even under chaotic conditions one can still understand abrupt jumps of diffusive-like processes in the relevant phase-space in terms of relatively smooth realignments of stable and unstable manifolds of unstable fixed points.
△ Less
Submitted 15 January, 1998; v1 submitted 8 January, 1998;
originally announced January 1998.
-
Nonintegrable Interaction of Ion-Acoustic and Electromagnetic Waves in a Plasma
Authors:
Felipe B. Rizzato,
Sergio R. Lopes,
Abraham C. -L. Chian
Abstract:
In this paper we re-examine the one-dimensional interaction of electromagnetic and ion acoustic waves in a plasma. Our model is similar to one solved by Rao et al. (Phys. Fluids, vol. 26, 2488 (1983)) under a number of analytical approximations. Here we perform a numerical investigation to examine the stability of the model. We find that for slightly over dense plasmas, the propagation of stable…
▽ More
In this paper we re-examine the one-dimensional interaction of electromagnetic and ion acoustic waves in a plasma. Our model is similar to one solved by Rao et al. (Phys. Fluids, vol. 26, 2488 (1983)) under a number of analytical approximations. Here we perform a numerical investigation to examine the stability of the model. We find that for slightly over dense plasmas, the propagation of stable solitary modes can occur in an adiabatic regime where the ion acoustic electric field potential is enslaved to the electromagnetic field of a laser. But if the laser intensity or plasma density increases or the laser frequency decreases, the adiabatic regime loses stability via a transition to chaos. New asymptotic states are attained when the adiabatic regime no longer exists. In these new states, the plasma becomes rarefied, and the laser field tends to behave like a vacuum field.
△ Less
Submitted 9 December, 1996; v1 submitted 5 December, 1996;
originally announced December 1996.