Heinz Horner

The sphericalp-spin interaction spin-glass model

The relaxational dynamics for local spin autocorrelations of the sphericalp-spin interaction spin-glass model is studied in the mean field limit. In the high temperature and high external field regime, the dynamics is ergodic and similar to the behaviour in known liquid-glass transition models. In the static limit, we recover the replica symmetric solution for the long time correlation. This phase becomes unstable on a critical line in the (T, h) plane, where critical slowing down is observed with a cross-over to power law decay of the correlation function ∝t−ν, with an exponent ν varying along the critical line. For low temperatures and low fields, ergodicity in phase space is broken. For small fields the transition is discontinuous, and approaching this transition from above, two long time scales are seen to emerge. This dynamical transition lies at a somewhat higher temperature than the one obtained within replica theory. For larger fields the transition becomes continuous at some tricritical point. The low temperature phase with broken ergodicity is studied within a modified equilibrium theory and alternatively for adiabatic cooling across the transition line. This latter scheme yields rather detailed insight into the formation and structure of the ergodic components.

Dynamics of learning for the binary perceptron problem

A polynomial learning algorithm for a perceptron with binary bonds and random patterns is investigated within dynamic mean field theory. A discontinuous freezing transition is found at a temperature where the entropy is still positive. Critical slowing down is observed approaching this temperature from above. The fraction of errors resulting from this learning procedure is finite in the thermodynamic limit for all temperatures and all finite values of the number of patterns per bond. Monte-Carlo simulations on larger samples (N≧127) are in quantitative agreement. Simulations on smaller samples indicate a finite bound for the existence of perfect solutions in agreement with the replica theory and the zero entropy criterion. This suggests that perfect solutions exist also in larger samples but cannot be found with a polynomial procedure as expected for a combinatorial hard problem.

Lattice dynamics of quantum crystals

A theory of lattice dynamics for quantum crystals is developed. This is done by summing an infinite class of diagrams of the usual anharmonic expansion and by avoiding the harmonic approximation as starting point. The zero order of the expansion given in this paper corresponds to the harmonic approximation with an effective potential. Higher orders correspond to higher anharmonic corrections with the same potential. Since the new potential varies more slowly the expansion seems to converge more rapidly than the usual anharmonic expansion.Numerical calculations on bcc He3 show that the ground state energy is lowered by about 3–4 cal/mol by taking into account long range correlations due to phonons. The elastic constants and the Debye temperature are calculated in zero and second order. The lowering of the bulk modulus due to the second order is about 10%. Experiments agree quite well with the second order results.

Neural networks with low levels of activity: Ising vs. McCulloch-Pitts neurons

The performance of neural networks used as associative memory for uncorrelated patterns with prescribed mean activity is analyzed within the replica symmetric mean field theory. The optimal representation of the possible states of the neutrons, active or inactive, is found to depend on the mean activity. For activity equal one half Ising neurons and for low activities McCulloch-Pitts neurons are optimal. In this optimal representation the noise due to noncondensed patterns is reduced.

Lattice distortions due to deuterium in niobium

Abstract The quasielastic diffuse scattering of neutrons from deuterium interstitials in niobium has been measured and compared to data calculated from a Kanzaki force model with central forces acting on nearest and next-nearest neighbours. It is found that the cubic symmetry of the long range distortion field of H in Nb is already established in the immediate neighbourhood of the defect.

Transients and basins of attraction in neutral network models

Approximate dynamic mean field equations for a generalized Hopfield model are derived, which allow to calculate transient properties of this model. These equations are exact for short times and yield the replica symmetric solution as a stationary solution. They allow reliable computation of retrieval trajectories and basins of attraction of retrieval states, as demonstrated by comparison with simulations. The equations are derived for networks with arbitrary mean activity and results are given for the standard model and for low activities.

Goldstone mode singularities and equation of state of an isotropic magnet

Static properties of an isotropic magnet are calculated in the whole critical region including the magnetization curve. The method proposed is a resummation of renormalized perturbation theory without use of recursion relations. This is possible because only special diagrams or subdiagrams show infrared divergencies at the magnetization curve due to Goldstone modes. The arguments given are heavily based on Ward identities. The resulting perturbation theory is well behaved in the total critical region and exhibits explicitely the form of the Goldstone mode singularities at the magnetization curve. The equation of state is calculated including two-loop contribution. Resulting effective exponents are then correct in orderε in the whole critical region. In various limits agreement with known results is found. A one-loop calculation of the correlation functions is also given.

Long time tail correlations in discrete chaotic dynamics

A scaling relation is derived connecting the exponent of the algebraically decaying correlation and response functions with the degree of intermittency and the order of the maximum. It is universal, i.e. within a large class independent of the correlated variables. This implies universal 1/f-like spectra. The corrections to scaling are investigated, too.

Non-equilibrium dynamics of simple spherical spin models

Abstract:We investigate the non-equilibrium dynamics of spherical spin models with two-spin interactions. For the exactly solvable models of the d-dimensional spherical ferromagnet and the spherical Sherrington-Kirkpatrick (SK) model the asymptotic dynamics has for large times and large waiting times the same formal structure. In the limit of large waiting times we find in both models an intermediate time scale, scaling as a power of the waiting time with an exponent smaller than one, and thus separating the time-translation-invariant short-time dynamics from the aging regime. It is this time scale on which the fluctuation-dissipation theorem is violated. Aging in these models is similar to that observed in spin glasses at the level of correlation functions, but different at the level of response functions, and thus different at the level of experimentally accessible quantities like thermoremanent magnetization.

Drift, creep and pinning of a particle in a correlated random potential

The motion of a particle in a correlated random potential under the influence of a driving force is investigated in mean field theory. The correlations of the disorder are characterized by a short distance cutoff and a power law decay with exponent γ at large distances. Depending on temperature and γ, drift with finite mobility, creep or pinning is found. This is in qualitative agreement with results in one dimension. This model is of interest not only in view of the motion of particles or manifolds in random media, it also improves the understanding of glassy non-equilibrium dynamics in mean field models. The results, obtained by numerical integration and analytic investigations of the various scaling regimes in this problem, are compared with previous proposals regarding the long time properties of such systems and with replica calculations.