New least-square algorithms

New algorithms are presented for approximating the minimum of the sum of squares ofM real and differentiable functions over anN-dimensional space. These algorithms update estimates for the location of a minimum after each one of the functions and its first derivatives are evaluated, in contrast with other least-square algorithms which evaluate allM functions and their derivatives at one point before using any of this information to make an update. These new algorithms give estimates which fluctuate about a minimum rather than converging to it. For many least-square problems, they give an adequate approximation for the solution more quickly than do other algorithms.It is a pleasure to thank J. Chesick of Haverford College for suggesting and implementing the application of this algorithm to x-ray crystallography.     

Preconditioned conjugate gradients for solving the transient Boussinesq equations in three-dimensional geometries

In this paper we present a new version of the ‘modified finite element method’ (MFEM) presented by Gresho, Chan, Lee and Upson.¹ The main modification of the original algorithm is the introduction of a cost-effective and memory-saving iterative solver for the discretized Poisson equation for the pressure. The vectorization of the preconditioner has been especially considered. For low Prandtl number problems we also split the advection-diffusion operator of the energy equation into explicit and implicit parts. In that sense the present approach is related to the recent implicitization of the diffusive terms introduced by Gresho and Chan² and by Gresho.³ The algorithm is applied to the study of buoyancy-driven flow oscillations occuring in a horizontal crucible of molten metal under the action of a horizontal temperature gradient.     

On diagonally preconditioning the truncated Newton method for super-scale linearly constrained nonlinear programming

We present an algorithm for super-scale linearly constrained nonlinear programming (LCNP) based on Newton's method. In large-scale programming solving the Newton equation at each iteration can be expensive and may not be justified when far from a local solution. For super-scale problems, the truncated Newton method (where an inaccurate solution is computed by using the conjugate-gradient method) is recommended; a diagonal BFGS preconditioning of the gradient is used, so that the number of iterations to solve the equation is reduced. The procedure for updating that preconditioning is described for LCNP when the set of active constraints or the partition of basic, superbasic and nonbasic (structural) variables have been changed.     

A nuclear magnetic resonance study of toluene-polyisobutylene solutions. 2. Vrentas-Duda theory

The self-diffusion coefficients of toluene in polyisobutylene have been analyzed using the Vrentas-Duda free volume diffusion model. The diffusion coefficients were determined at different temperatures and concentrations, using the pulsed field gradient nuclear magnetic resonance technique. The data were satisfactorily described by the model and the size of the polymer jumping unit was extracted. Comparisons were made with the Fujita free volume theory and the Fujita free volume parameters were extracted from the Vrentas-Duda free volume parameters. From the diffusion data that now available, it can be concluded that for most polymers the jumping unit is about 1.5 times the polymer monomer molecular weight. The activation energy of the toluene diffusion in polyisobutylene is compared with the activation energies of other penetrants in the same polymer. The diffusion data presented in this work show that the energy per mole required to overcome the attractive forces which constrain a diffusing species to its neighbors should be considered to be zero, in order to be able to extract the free volume parameters (from viscosity and diffusion data) with an acceptable uncertainty. ©1995 John Wiley & Sons, Inc.     

Information exchange between independent stochastic systems

We considerN independent, linear, Gaussian stochastic systems, each controlled by a decision maker having independent measurements on his own system. The decision makers agree to cooperate in order to minimize a weighted sum of their own independent quadratic performance indices. For this, they may or may not exchange their measurements on line. If further constraints (such as the restriction to memoryless feedbacks) are imposed, an example shows that the best solution is not always decentralized; that is, exchanging information really improves the overall performance. Moreover, some properties of the best decentralized feedbacks are investigated; they are local minima in a more general class and even absolute minima in particular situations.This research was supported by CNRS, France, Contract No. ATP-2340. The topic of this paper was presented at a Symposium on Management Science, Optimization of Dynamic Problems, held at the European Institute for Advanced Studies in Management, Brussels, Belgium, 1978. The author is grateful to Prof. P. Bernhard and Dr. J. P. Quadrat for helpful discussions.     

On the convergence of policy iteration for controlled diffusions

The convergence of an approximation scheme known as policy iteration has been demonstrated for controlled diffusions by Fleming, Puterman, and Bismut. In this paper, we show that this approximation scheme is equivalent to the Newton-Kantorovich iteration for solving the optimality equation and exploit this equivalence to obtain a new proof of convergence. Estimates of the rate of convergence of this procedure are also obtained.This research was partially supported by NRC Grant No. A-3609.     

非线性最小二乘法的算法

本给出非线性最小二乘的优化条件和几何特征．     

Option pricing and perfect hedging on correlated stocks

We develop a theory for option pricing with perfect hedging in an inefficient market model where the underlying price variations are autocorrelated over a time τ0. This is accomplished by assuming that the underlying noise in the system is derived by an Ornstein-Uhlenbeck, rather than from a Wiener process. With a modified portfolio consisting in calls, secondary calls and bonds we achieve a riskless strategy which results in a closed and exact expression for the European call price which is always lower than Black-Scholes price. We obtain the same price and a modified delta hedging if we start from an effective one-dimensional market model. We compare these strategies and study the sensitivity of the call price to several parameters where the correlation effects are also observed.     

Nonequilibriurn discontinuous phase transitions in a fast ionic conductor model: Coexistence and spinodal lines

Two-dimensional lattice-gas models with attractive interactions and particle-conserving hopping dynamics under the influence of a very large external electric field along a principal axis are studied in the case of off-critical densities. We describe the corresponding nonequilibrium first-order phase transitions, evaluate coexistence and spinodal lines, and make some comparisons with experimental observations on fast ionic conductors.See Ref. 1 (henceforth referred to as II) for references.     

Convergence rates for steady-state derivative estimators

We describe various derivative estimators for the case of steady-state performance measures and obtain the order of their convergence rates. These estimatorsdo not use explicitly the regenerative structure of the system. Estimators based on infinitesimal perturbation analysis, likelihood ratios, and different kinds of finite-differences are examined. The theoretical results are illustrated via numerical examples.