Global Convergence

We present a line-search algorithm for large-scale continuous optimization. The algorithm is matrix-free in that it does not require the factorization of derivative matrices. Instead, it uses iterative linear system solvers. Inexact step computations are supported in order to save computational expense during each iteration. The algorithm is an interior-point approach derived from an inexact Newton method for equality constrained optimization proposed by Curtis, Nocedal, and Wächter [SIAM J. Optim., 20 (2009), pp. 1224-1249, with additional functionality for handling inequality constraints. The algorithm is shown to be globally convergent under loose assumptions. Numerical results are presented for nonlinear optimization test set collections and a pair of PDEconstrained model problems.

Conjugate gradient methods are widely used for unconstrained optimization, especially large scale problems. The strong Wolfe conditions are usually used in the analyses and implementations of conjugate gradient methods. This paper presents a new version of the conjugate gradient method, which converges globally, provided the line search satisfies the standard Wolfe conditions. The conditions on the objective function are also weak, being similar to those required by the Zoutendijk condition.

This paper explores the convergence of nonlinear conjugate gradient methods without restarts, and with practical line searches. The analysis covers two classes of methods that are globally convergent on smooth, nonconvex functions. Some properties of the Fletcher-Reeves method play an important role in the first family, whereas the second family shares an important property with the Polak-Ribire method. Numerical experiments are presented.

Al~ract-In this paper we give a tutonal account of several of the most recent adapuve control results for ngld robot mampulators Our intent ~s to lend some perspecUve to the growing hst of adapuve control results for mampulators by providing a unified framework for comparison of those adapUve control algorithms which have been shown to be globally convergent In most cases we are able to stmphfy the denvatmns and proofs of these results as well *

We consider the problem of minimizing the sum of a smooth function and a separable convex function. This problem includes as special cases bound-constrained optimization and smooth optimization with ℓ1-regularization. We propose a (block) coordinate gradient descent method for solving this class of nonsmooth separable problems. We establish global convergence and, under a local Lipschitzian error bound assumption, linear convergence for this method. The local Lipschitzian error bound holds under assumptions analogous to those for constrained smooth optimization, e.g., the convex function is polyhedral and the smooth function is (nonconvex) quadratic or is the composition of a strongly convex function with a linear mapping. We report numerical experience with solving the ℓ1-regularization of unconstrained optimization problems from Moré et al. in ACM Trans. Math. Softw. 7, 17–41, 1981 and from the CUTEr set (Gould and Orban in ACM Trans. Math. Softw. 29, 373–394, 2003). Comparison with L-BFGS-B and MINOS, applied to a reformulation of the ℓ1-regularized problem as a bound-constrained optimization problem, is also reported.

The ordered subsets EM (OSEM) algorithm has enjoyed considerable interest for emission image reconstruction due to its acceleration of the original EM algorithm and ease of programming. The transmission EM reconstruction algorithm converges very slowly and is not used in practice. In this paper, we introduce a simultaneous update algorithm called separable paraboloidal surrogates (SPS) that converges much faster than the transmission EM algorithm. Furthermore, unlike the "convex algorithm" for transmission tomography, the proposed algorithm is monotonic even with nonzero background counts. We demonstrate that the ordered subsets principle can also be applied to the new SPS algorithm for transmission tomography to accelerate "convergence", albeit with similar sacrifice of global convergence properties as OSEM. We implemented and evaluated this ordered subsets transmission (OSTR) algorithm. The results indicate that the OSTR algorithm speeds up the increase in the objective function by roughly the number of subsets in the early iterates when compared to the ordinary SPS algorithm. We compute mean square errors and segmentation errors for different methods and show that OSTR is superior to OSEM applied to the logarithm of the transmission data. However, penalized-likelihood reconstructions yield the best quality images among all other methods tested.

This paper considers jointly optimal design of crosslayer congestion control, routing and scheduling for ad hoc wireless networks. We first formulate the rate constraint and scheduling constraint using multicommodity flow variables, and formulate resource allocation in networks with fixed wireless channels (or single-rate wireless devices that can mask channel variations) as a utility maximization problem with these constraints. By dual decomposition, the resource allocation problem naturally decomposes into three subproblems: congestion control, routing and scheduling that interact through congestion price. The global convergence property of this algorithm is proved. We next extend the dual algorithm to handle networks with timevarying channels and adaptive multi-rate devices. The stability of the resulting system is established, and its performance is characterized with respect to an ideal reference system which has the best feasible rate region at link layer.

The global and local convergence properties of a class of augmented Lagrangian methods for solving nonlinear programming problems are considered. In such methods, simple bound constraints are treated separately from more general constraints and the stopping rules for the inner minimization algorithm have this in mind. Global convergence is proved, and it is established that a potentially troublesome penalty parameter is bounded away from zero.

Variable selection is fundamental to high-dimensional statistical modeling. Many variable selection techniques may be implemented by maximum penalized likelihood using various penalty functions. Optimizing the penalized likelihood function is often challenging because it may be nondifferentiable and/or nonconcave. This article proposes a new class of algorithms for finding a maximizer of the penalized likelihood for a broad class of penalty functions. These algorithms operate by perturbing the penalty function slightly to render it differentiable, then optimizing this differentiable function using a minorize-maximize (MM) algorithm. MM algorithms are useful extensions of the well-known class of EM algorithms, a fact that allows us to analyze the local and global convergence of the proposed algorithm using some of the techniques employed for EM algorithms. In particular, we prove that when our MM algorithms converge, they must converge to a desirable point; we also discuss conditions under which this convergence may be guaranteed. We exploit the Newton-Raphson-like aspect of these algorithms to propose a sandwich estimator for the standard errors of the estimators. Our method performs well in numerical tests.

BABDCR is a package of Fortran 90 subroutines for the solution of linear systems with bordered almost block diagonal coefficient matrices. It is designed to handle matrices with blocks of the same size, that is, having a block upper bidiagonal structure with an additional block in the right upper corner. The algorithm implemented in the package performs cyclic reduction of the coefficient matrix in order to reduce the fill-in due to the corner block.

In this paper, we study both the local and global convergence of various iterative methods for solving the variational inequality and the nonlinear complementarity problems. Included among such methods are the Newton and several successive overrelaxation algorithms. For the most part, the study is concerned with the family of linear approximation methods. These are iterative methods in which a sequence of vectors is generated by solving certain linearized subproblems. Convergence to a solution of the given variational or complementarity problem is established by using three different yet related approaches. The paper also studies a special class of variational inequality problems arising from such applications as computing traffic and economic spatial equilibria. Finally, several convergence results are obtained for some nonlinear approximation methods.

In this two-part paper, we propose a decentralized strategy, based on a game-theoretic formulation, to find out the optimal precoding/multiplexing matrices for a multipoint-to-multipoint communication system composed of a set of wideband links sharing the same physical resources, i.e., time and bandwidth. We assume, as optimality criterion, the achievement of a Nash equilibrium and consider two alternative optimization problems: 1) the competitive maximization of mutual information on each link, given constraints on the transmit power and on the spectral mask imposed by the radio spectrum regulatory bodies; and 2) the competitive maximization of the transmission rate, using finite order constellations, under the same constraints as above, plus a constraint on the average error probability. In this first part of the paper, we start by showing that the solution set of both noncooperative games is always nonempty and contains only pure strategies. Then, we prove that the optimal precoding/multiplexing scheme for both games leads to a channel diagonalizing structure, so that both matrix-valued problems can be recast in a simpler unified vector power control game, with no performance penalty. Thus, we study this simpler game and derive sufficient conditions ensuring the uniqueness of the Nash equilibrium. Interestingly, although derived under stronger constraints, incorporating for example spectral mask constraints, our uniqueness conditions have broader validity than previously known conditions. Finally, we assess the goodness of the proposed decentralized strategy by comparing its performance with the performance of a Pareto-optimal centralized scheme. To reach the Nash equilibria of the game, in Part II, we propose alternative distributed algorithms, along with their convergence conditions. Index Terms-Ad hoc networks, cognitive radio, game theory, linear precoding, mutual information, Nash equilibrium.

Distributed and efficient resource allocation is critical for fully realizing the benefits of cooperative communications in large scale communication networks. This paper proposes two auction mechanisms, the SNR auction and the power auction, that determine relay selection and relay power allocation in a distributed fashion. A single-relay network is considered first, and the existence and uniqueness of the Nash Equilibrium (i.e., the auction's outcome) are proved. It is shown that the power auction achieves the efficient allocation by maximizing the total rate increase, and the SNR auction is flexible in trading off fairness and efficiency. For both auctions, the distributed best response bid updates globally converge to the unique Nash Equilibrium in a completely asynchronous manner. The analysis is then generalized to networks with multiple relays, and the existence of the Nash Equilibrium is shown under appropriate conditions. Simulation results verify the effectiveness and robustness of the proposed algorithms.

We consider the global and local convergence properties of a class of Lagrangian barrier methods for solving nonlinear programming problems. In such methods, simple bound constraints may be treated separately from more general constraints. The objective and general constraint functions are combined in a Lagrangian barrier function. A sequence of such functions are approximately minimized within the domain defined by the simple bounds. Global convergence of the sequence of generated iterates to a first-order stationary point for the original problem is established. Furthermore, possible numerical difficulties associated with barrier function methods are avoided as it is shown that a potentially troublesome penalty parameter is bounded away from zero. This paper is a companion to previous work of ours on augmented Lagrangian methods.

A navigation functions' based methodology, established in our previous work for centralized multiple robot navigation, is extended to address the problem of decentralized navigation. In contrast to the centralized case, each agent plans its actions without knowing the destinations of the other agents. Asymptotic stability is guaranteed by the existence of a global Lyapunov function for the whole system, which is actually the sum of the separate navigation functions. The collision avoidance and global convergence properties are verified through simulations.

Centroidal Voronoi tessellations (CVTs) are Voronoi tessellations of a bounded geometric domain such that the generating points of the tessellations are also the centroids (mass centers) of the corresponding Voronoi regions with respect to a given density function. Centroidal Voronoi tessellations may also be defined in more abstract and more general settings. Due to the natural optimization properties enjoyed by CVTs, they have many applications in diverse fields. The Lloyd algorithm is one of the most popular iterative schemes for computing the CVTs but its theoretical analysis is far from complete. In this paper, some new analytical results on the local and global convergence of the Lloyd algorithm are presented. These results are derived through careful utilization of the optimization properties shared by CVTs. Numerical experiments are also provided to substantiate the theoretical analysis.

In the last few years, the support vector machine (SVM) method has motivated new interest in kernel regression techniques. Although the SVM has been shown to exhibit excellent generalization properties in many experiments, it suffers from several drawbacks, both of a theoretical and a technical nature: the absence of probabilistic outputs, the restriction to Mercer kernels, and the steep growth of the number of support vectors with increasing size of the training set. In this paper, we present a different class of kernel regressors that effectively overcome the above problems. We call this approach generalized LASSO regression. It has a clear probabilistic interpretation, can handle learning sets that are corrupted by outliers, produces extremely sparse solutions, and is capable of dealing with large-scale problems. For regression functionals which can be modeled as iteratively reweighted least-squares (IRLS) problems, we present a highly efficient algorithm with guaranteed global convergence. This defies a unique framework for sparse regression models in the very rich class of IRLS models, including various types of robust regression models and logistic regression. Performance studies for many standard benchmark datasets effectively demonstrate the advantages of this model over related approaches.

We consider adaptive stabilization for a class of nonlinear second-order systems. Interpreting the system states as position and velocity, the system is assumed to have unknown, nonparametric position-dependent damping and stiffness coefficients. Lyapunov methods are used to prove global convergence of the adaptive controller. Furthermore, the controller is shown to be able reject constant disturbances and to asymptotically track constant commands. For illustration, the controller is used to stabilize the van der Pol limit cycle, the Duffing oscillator with multiple equilibria, and several other example systems.

This paper considers the noncooperative maximization of mutual information in the Gaussian interference channel in a fully distributed fashion via game theory. This problem has been studied in a number of papers during the past decade for the case of frequency-selective channels. A variety of conditions guaranteeing the uniqueness of the Nash Equilibrium (NE) and convergence of many different distributed algorithms have been derived. In this paper we provide a unified view of the state-of- the-art results, showing that most of the techniques proposed in the literature to study the game, even though apparently different, can be unified using our recent interpretation of the waterfilling operator as a projection onto a proper polyhedral set. Based on this interpretation, we then provide a mathematical framework, useful to derive a unified set of sufficient conditions guaranteeing the uniqueness of the NE and the global convergence of waterfilling based asynchronous distributed algorithms. The proposed mathematical framework is also instrumental to study the extension of the game to the more general MIMO case, for which only few results are available in the current literature. The resulting algorithm is, similarly to the frequency-selective case, an iterative asynchronous MIMO waterfilling algorithm. The proof of convergence hinges again on the interpretation of the MIMO waterfilling as a matrix projection, which is the natural generalization of our results obtained for the waterfilling mapping in the frequency-selective case.

The problem of robust H∞ filtering for continuous-time uncertain linear systems with multiple time-varying delays in the state variables is investigated. The uncertain parameters are supposed to belong to a given convex bounded polyhedral domain. The aim is to design a stable linear filter assuring asymptotic stability and a prescribed H∞ performance level for the filtering error system, irrespective of the uncertainties and the time delays. Sufficient conditions for the existence of such a filter are established in terms of linear matrix inequalities, which can be efficiently solved by means of powerful convex programming tools with global convergence assured. An example illustrates the proposed methodology

We present a simple and unified technique to establish convergence of various minimization methods. These contain the (conceptual) proximal point method, as well as implementable forms such as bundle algorithms, including the classical subgradient relaxation algorithm with divergent series.

We investigate the global convergence of estimation of distribution algorithms (EDAs). In EDAs, the distribution is estimated from a set of selected elements, i.e., the parent set, and then the estimated distribution model is used to generate new elements. In this paper, we prove that: 1) if the distribution of the new elements matches that of the parent set exactly, the algorithms will converge to the global optimum under three widely used selection schemes and 2) a factorized distribution algorithm converges globally under proportional selection.

In this paper we propose a new line search algorithm that ensures global convergence of the Polak-Ribi ere conjugate gradient method for the unconstrained minimization of nonconvex di erentiable functions. In particular, we show that every limit point produced by the Polak-Ribi ere iteration is a stationary point of the objective function. Moreover, we prove that, asymptotically, the rst stationary point along the search direction can be accepted and that, under strong convexity assumptions, the known global convergence results can be reobtained as a special case. From a computational point of view, we may expect that an algorithm incorporating the stepsize acceptance rules proposed here will retain the same good features of the Polak-Ribi ere method, while avoiding pathological situations.

Recently, we propose a nonlinear conjugate gradient method, which produces a descent search direction at every iteration and converges globally provided that the line search satisfies the weak Wolfe conditions. In this paper, we will study methods related to the new nonlinear conjugate gradient method. Specifically, if the size of the scalar yS^ with respect to the one in the new method belongs to some interval, then the coiTesponding methods are proved to be globally convergent; otherwise, we are able to construct a convex quadratic example showing that the methods need not converge. Numerical experiments are made for two combinations of the new method and the Hestenes-Stiefel conjugate gradient method. The initial results show that, one of the hybrid methods is especially efficient for the given test problems.

In this manuscript we investigate the global convergence of the implicit residual-based a posteriori error estimates of Adjerid et al. (2002) [3]. The authors used the discontinuous Galerkin method to solve one-dimensional transient hyperbolic problems and showed that the local error on each element is proportional to a Radau polynomial. The discontinuous Galerkin error estimates under investigation are computed by solving a local steady problem on each element. Here we prove that, for smooth solutions, these a posteriori error estimates at a fixed time t converge to the true spatial error in the L2L2 norm under mesh refinement.

In this paper, the filter technique of Fletcher and Leyffer (1997) is used to globalize the primal-dual interior-point algorithm for nonlinear programming, avoiding the use of merit functions and the updating of penalty parameters. The new algorithm decomposes the primal-dual step obtained from the perturbed first-order necessary conditions into a normal and a tangential step, whose sizes are controlled by a trust-region type parameter. Each entry in the filter is a pair of coordinates: one resulting from feasibility and centrality, and associated with the normal step; the other resulting from optimality (complementarity and duality), and related with the tangential step. Global convergence to first-order critical points is proved for the new primal-dual interior-point filter algorithm.

The Godard or constant modulus algorithm (CMA) equalizer is perhaps the best known and the most popular scheme for blind adaptive channel equalization. Most published works on blind equalization convergence analysis are confined to T-spaced equalizers with real-valued inputs. The common belief is that analysis of fractionally spaced equalizers (FSE's) with complex inputs is a straightforward extension with similar results. This belief is, in fact, untrue. In this paper, we present a convergence analysis of GodardCMA FSE's that proves the important advantages provided by the FSE structure. We show that an FSE allows the exploitation of the channel diversity that supports two important conclusions of great practical significance: 1) A finite-length channel satisfying a length-and-zero condition allows Godard/CMA FSE to be globally convergent, and 2) the linear FSE filter length need not be longer than the channel delay spread. Computer simulation demonstrates the performance improvement provided by the adaptive Godard FSE.

A new active-set method for smooth box-constrained minimization is introduced. The algorithm combines an unconstrained method, including a new line-search which aims to add many constraints to the working set at a single iteration, with a recently introduced technique (spectral projected gradient) for dropping constraints from the working set. Global convergence is proved. A computer implementation is fully described and a numerical comparison assesses the reliability of the new algorithm.

The sequential Iterative Water-Filling Algorithm (IWFA) proposed by Yu et al. is by now a popular lowcomplexity algorithm to compute the Nash equilibrium point of the power allocation game in a Gaussian frequency-selective multiuser interference channel. The algorithm is based on a distributed sequential updating where, at each iteration, the users choose their power allocation, one after the other. However, this sequential updating strategy may slow down its convergence time excessively when the number of users is high. In this paper, we propose an alternative distributed algorithm, called Simultaneous Iterative Water-Filling Algorithm (SIWFA), where at each iteration, all the users update their power allocations simultaneously, rather than sequentially. This reduces the convergence time considerably, specially when the number of users is large. Our main contribution is to provide a unified set of sufficient conditions for the convergence of both IWFA and SIWFA, that are less stringent than those known in the literature for IWFA. These conditions guarantee the convergence of both algorithms also in the presence of spectral mask constraints imposed on the power allocations of the users. T

Figure 1: The emergence and spread of lowest-low fertility in Europe during 1990-2002 future population trends, with current patterns ranging from countries that stabilized at moderately belowreplacement fertility levels to lowest-low fertility countries with TFR declines below 1.3 (Figure 2). For instance, several European countries that were among the first to experience sustained below-replacement fertility in the late 1960s and early 1970s, including Denmark, France, the Netherlands and the United Kingdom, exhibit relatively high fertility in 2002. Moreover, the Dutch, Danish and French TFRs have increased during the last decade to levels of 1.72 (The Netherlands), 1.77 (Denmark), and 1.89 (France) (Council of Europe 2003), and several other European countries exhibit even higher TFRs. These trends

Magneto-and electroencephalography (M/EEG) measure the electromagnetic fields produced by the neural electrical currents. Given a conductor model for the head, and the distribution of source currents in the brain, Maxwell's equations allow one to compute the ensuing M/EEG signals. Given the actual M/EEG measurements and the solution of this forward problem, one can localize, in space and in time, the brain regions than have produced the recorded data. However, due to the physics of the problem, the limited number of sensors compared to the number of possible source locations, and measurement noise, this inverse problem is ill-posed. Consequently, additional constraints are needed. Classical inverse solvers, often called Minimum Norm Estimates (MNE), promote source estimates with a small ℓ 2 norm. Here, we consider a more general class of priors based on mixed-norms. Such norms have the ability to structure the prior in order to incorporate some additional assumptions about the sources. We refer to such solvers as Mixed-Norm Estimates (MxNE). In the context of M/EEG, MxNE can promote spatially focal sources with smooth temporal estimates with a twolevel ℓ 1 /ℓ 2 mixed-norm, while a three-level mixed-norm can be used to promote spatially non-overlapping sources between different experimental conditions. In order to efficiently solve the optimization problems of MxNE, we introduce fast first-order iterative schemes that for the ℓ 1 /ℓ 2 norm give solutions in a few seconds making such a prior as convenient as the simple MNE. Furhermore, thanks to the convexity of the optimization problem, we can provide optimality conditions that guarantee global convergence. The utility of the methods is demonstrated both with simulations and experimental MEG data.

This paper introduces a general implicit iterative method for finding zeros of a maximal monotone operator in a Hilbert space which unifies three previously studied strategies: relaxation, inertial type extrapolation and projection step. The first two strategies are intended to speed up the convergence of the standard proximal point algorithm, while the third permits one to perform inexact proximal iterations with fixed relative error tolerance. The paper establishes the global convergence of the method for the weak topology under appropriate assumptions on the algorithm parameters.

New degrees of freedom can be optimized in mask shapes when the source is also adjustable, because required image symmetries can be provided by the source rather than the collected wave front. The optimized mask will often consist of novel sets of shapes that ...

A recurrent neural network for the optimal control of a group of interconnected dynamic systems is presented in this paper. On the basis of decomposition and coordination strategy for interconnected dynamic systems, the proposed neural network has a two-level hierarchical structure: several local optimization subnetworks at the lower level and one coordination subnetwork at the upper level. A goal-coordination method is used to coordinate the interactions between the subsystems. By nesting the dynamic equations of the subsystems into their corresponding local optimization subnetworks, the number of dimensions of the neural network can be reduced significantly. Furthermore, the subnetworks at both the lower and upper levels can work concurrently. Therefore, the computation efficiency, in comparison with the consecutive executions of numerical algorithms on digital computers, is increased dramatically. The proposed method is extended to the case where the control inputs of the subsystems are bounded. The stability analysis shows that the proposed neural network is asymptotically stable. Finally, an example is presented which demonstrates the satisfactory performance of the neural network.

This letter exploits the cyclic prefix to create a blind adaptive globally convergent channel-shortening algorithm, with a complexity like least mean squares. The cost function is related to that of the shortening signal-to-noise solution of Melsa et al., and simulations are provided to demonstrate the performance of the algorithm.

Projection Technique is used to suggest a unified and general iterative algorithm for computing the approximate solution of a new class of quasi variational inequalities. The convergence properties of this algorithm are also considered. Several special cases which can be obtained from the general results are also discussed.

What is the capacity of the uplink of a radio network of receivers? We consider a spread spectrum model in which each user is decoded by all the receivers in the network (macrodiversity). We use a carrier-to-interference performance criterion that we derive from Shannon theory; each user must find the right transmitter power level to satisfy its carrier-tointerference constraint. Satisfying this requirement for all users is equivalent to solving a fixed point problem. We use this power control problem to derive the network capacity region and find that the feasibility of a configuration of users is independent of their positions in the network; each user can be assigned a bandwidth that is independent of the user's position in the network. Our capacity region is an upper bound over all schemes that treat the interference of other users as pure noise. To show that the capacity can be realized in practice, we propose a decentralized power adaptation algorithm and prove global convergence to the fixed point via a monotonicity argument.