Search

We study derivative-free constrained optimization problems and propose a trust-region method that builds linear or quadratic models around the best feasible and around the best infeasible solutions found so far. These models are optimized within a trust region, and the progressive barrier methodology handles the constraints by progressively pushing the infeasible solutions toward the feasible domain. Computational experiments on 40 smooth constrained problems indicate that the proposed method is competitive with COBYLA, and experiments on two nonsmooth multidisciplinary optimization problems from mechanical engineering show that it can be competitive with the NOMAD software.

The mesh adaptive direct search algorithm (MADS) is an iterative method for constrained blackbox optimization problems. One of the optional MADS features is a versatile search step in which quadratic models are built leading to a series of quadratically constrained quadratic subproblems. This work explores different algorithms that exploit the structure of the quadratic models: the first one applies an l1-exact penalty function, the second uses an augmented Lagrangian and the third one combines the former two, resulting in a new algorithm. It is notable that this latter approach is uniquely suitable for quadratically constrained quadratic problems. These methods are implemented within the NOMAD software package and their impact are assessed through computational experiments on 65 analytical test problems and 4 simulation-based engineering applications.

Suppressing the e ects of liquid loading is a key issue for e cient utilization of mid and late-life wells in shale-gas systems. This state of the wells can be prevented by performing short shut-ins when the gas rate falls below the minimum rate needed to avoid liquid loading. In this paper, we present a Lagrangian relaxation based scheme for shut-in scheduling of distributed shale multiwell systems. The scheme optimizes shut-in times and a reference rate for each multi-well pad, such that the total produced rate tracks a given short-term gas demand for the field. By using simple, frequency-tuned well proxy models, we obtain a compact mixed integer formulation which by Lagrangian relaxation renders a decomposable structure. A set of computational tests demonstrates the merits of the proposed scheme. This study indicates that the method is capable of solving large field-wide scheduling problems by producing good solutions in reasonable computation times.

This paper proposes and explores an algorithm designed to find optimal settings for a process network. Emphasis is put on the system being divisible into components, as this underlying assumption motivates the algorithm in its entirety in that rather simple relations between the system components are modeled as explicit structural constraints, while the significantly more complex relations within each component are approximated based on the underlying simulator data. Although the approach taken in this paper is rather broadly applicable we are, in particular, interested in its application to production optimization problems in the oil and gas industry. We give limited numerical results for one such example that clearly indicates the advantages of our approach. We show the advantages of both decomposing the problem of interest and accounting for the structure from the point of view of exploiting, where ever possible, the explicitly analytic aspects of the problem. The advantage of doing the former is that the considered subproblems are significantly smaller than the overall problem. The advantage of the latter is that one can use derivatives for the analytic parts whereas they are unavailable for the simulators. The underlying approach is a trust-region one with a mixed integer nonlinear program formulation. There are some significant differences in the details of the algorithm from those generally available for such problems.

One important application of our work appears in the robust optimization of simulation-based functions, which may arise due to implementation variables or uncertain parameters. The robust counterpart of an optimization problem without derivatives falls into the category of the bilevel problems under consideration here. We provide numerical illustrations of the application of our algorithmic framework to such robust optimization examples.

We describe a Generalized Semi-Markov Model paired with Monte Carlo simulation that represents the evolution of the systems that constitute an offshore installation. We use the model to assess the performance of a maintenance plan in terms of a production penalty and unplanned shutdowns. In addition to estimating, comparing and improving maintenance plans, our approach enables the determination of the consequences of planning capital expenditures and installation modifications and the value of process improvements. Moreover, the entire framework is designed to be sufficiently flexible to accommodate various ‘what if’ scenarios in addition to other modifications. The simulation model was tested on data from an offshore oil installation, and was well received by the installation's operations personnel. Copyright © 2010 John Wiley & Sons, Ltd.

In this paper we prove global convergence for first- and second-order stationary points of a class of derivative-free trust-region methods for unconstrained optimization. These methods are based on the sequential minimization of quadratic (or linear) models built from evaluating the objective function at sample sets. The derivative-free models are required to satisfy Taylor-type bounds, but, apart from that, the analysis is independent of the sampling techniques. A number of new issues are addressed, including global convergence when acceptance of iterates is based on simple decrease of the objective function, trust-region radius maintenance at the criticality step, and global convergence for second-order critical points.

In recent years there has been a considerable amount of work on the development of numerical methods for derivative-free optimization problems. Some of this work relies on the management of the geometry of sets of sampling points for function evaluation and model building. In this paper we continue the work developed in Conn et al. (2008b, Math. Program., 111, 141-172) for complete or determined interpolation models (when the number of interpolation points equals the number of basis elements), considering now the cases where the number of points is higher (regression models) and lower (underdetermined models) than the number of basis components. We show that the regression and underdetermined models essentially have similar properties to the interpolation model in that the mechanisms and concepts which control the quality of the sample sets, and hence of the approximation error bounds, of the interpolation models can be extended to the over- and underdetermined cases. We also discuss the trade-offs between using a fully determined interpolation model and the over- or underdetermined ones.

Circuit tuning is an important task in the design of custom digital integrated circuits such as high-performance microprocessors. The goal is to improve certain aspects of the circuit, such as speed, area, or power, by optimally choosing the widths of the transistors. This task can be formulated as a large-scale nonlinear, nonconvex optimization problem, where function values and derivatives are obtained by simulation of individual gates. This application offers an excellent example of a nonlinear optimization problem, for which it is very desirable to increase the size of the problems that can be solved in a reasonable amount of time. In this paper we describe the mathematical formulation of this problem and the implementation of a circuit tuning tool. We demonstrate how the integration of a novel state-of-the-art interior point algorithm for nonlinear programming led to considerable improvement in efficiency and robustness. Particularly, as will be demonstrated with numerical results, the new approach has great potential for parallel and distributed computing.

Noise can cause digital circuits to switch incorrectly and thus produce spurious results. Noise can also have adverse power, timing and reliability effects. Dynamic logic is particularly susceptible to charge-sharing and coupling noise. Thus, the design and optimization of a circuit should take noise considerations into account. Such considerations are typically stated as semi-infinite constraints. In addition, the number of signals to be checked and the number of sub-intervals of time during which the checking must be performed can potentially be very large. Thus, the practical incorporation of noise constraints during circuit optimization is a hitherto unsolved problem. This paper describes a novel method for incorporating noise considerations during automatic circuit optimization. Semi-infinite constraints representing noise considerations are first converted to ordinary equality constraints involving time integrals, which are readily computed in the context of circuit optimization based on time-domain simulation. Next, the gradients of these integrals are computed by the adjoint method. By using an augmented Lagrangian optimization merit function, the adjoint method is applied to compute all the necessary gradients required for optimization in a single adjoint analysis, no matter how many noise measurements are considered, and irrespective of the dimensionality of the problem. Numerical results are presented.

A new primal-dual algorithm is proposed for the minimization of non-convex objective functions subject to general inequality and linear equality constraints. The method uses a primal-dual trust-region model to ensure descent on a suitable merit function. Convergence is proved to second-order critical points from arbitrary starting points. Numerical results are presented for general quadratic programs.

The problem of minimizing a sum of Euclidean norms dates from the 17th century and may be the earliest example of duality in the mathematical programming literature. This nonsmooth optimization problem arises in many different kinds of modern scientific applications. We derive a primal-dual interior-point algorithm for the problem, by applying Newton's method directly to a system of nonlinear equations characterizing primal and dual feasibility and a perturbed complementarity condition. The main work at each step consists of solving a system of linear equations (the Schur complement equations). This Schur complement matrix is not symmetric, unlike in linear programming. We incorporate a Mehrotra-type predictor-corrector scheme and present some experimental results comparing several variations of the algorithm, including, as one option, explicit symmetrization of the Schur complement with a skew corrector term. We also present results obtained from a code implemented to solve large sparse problems, using a symmetrized Schur complement. This has been applied to problems arising in plastic collapse analysis, with hundreds of thousands of variables and millions of nonzeros in the constraint matrix. The algorithm typically finds accurate solutions in less than 50 iterations and determines physically meaningful solutions previously unobtainable.

A theoretical framework and a practical algorithm are presented to solve discontinuous piecewise linear optimization problems dealing with functions for which theridges are known. A penalty approach allows one to consider such problems subject to a wide range of constraints involving piecewise linear functions. Although the theory is expounded in detail in the special case of discontinuous piecewiselinear functions, it is straightforwardly extendable, using standard nonlinear programming techniques, tononlinear (discontinuous piecewise differentiable) functions. The descent algorithm which is elaborated uses active-set and projected gradient approaches. It is a generalization of the ideas used by Conn to deal with nonsmoothness in thel1 exact penalty function, and it is based on the notion ofdecomposition of a function into a smooth and a nonsmooth part. The constrained case is reduced to the unconstrained minimization of a (piecewise linear)l1 exact penalty function. We also discuss how the algorithm is modified when it encounters degenerate points. Preliminary numerical results are presented: the algorithm is applied to discontinuous optimization problems from models in industrial engineering. © 1998 The Mathematical Programming Society, Inc. Published by Elsevier Science B.V.

Automating the transistor and wire-sizing process is an important step toward being able to rapidly design high-performance, custom circuits. This paper presents a circuit optimization tool that automates the tuning task by means of state-of-the-art nonlinear optimization. It makes use of a fast circuit simulator and a general-purpose nonlinear optimization package. It includes minimax and power optimization, simultaneous transistor and wire tuning, general choices of objective functions and constraints, and recovery from nonworking circuits. In addition, the tool makes use of designer-friendly interfaces that automate the specification of the optimization task, the running of the optimizer, and the back-annotation of the results of optimization onto the circuit schematic. Particularly for large circuits, gradient computation is usually the bottleneck in the optimization procedure. In addition to traditional adjoint and direct methods, we use a technique called the adjoint Lagrangian method, which computes all the gradients necessary for one iteration of optimization in a single adjoint analysis. This paper describes the algorithms and the environment in which they are used and presents extensive circuit optimization results. A circuit with 6900 transistors, 4128 tunable transistors, and 60 independent parameters was optimized in about 108 min of CPU time on an IBM RISC/System 6000, model 590.

Quasi-Newton algorithms for unconstrained nonlinear minimization generate a sequence of matrices that can be considered as approximations of the objective function second derivatives. This paper gives conditions under which these approximations can be proved to converge globally to the true Hessian matrix, in the case where the Symmetric Rank One update formula is used. The rate of convergence is also examined and proven to be improving with the rate of convergence of the underlying iterates. The theory is confirmed by some numerical experiments that also show the convergence of the Hessian approximations to be substantially slower for other known quasi-Newton formulae. © 1991 The Mathematical Programming Society, Inc.

We consider the minimum cost network flow problem and describe how the non-linear penalty function methods of Conn and Bartels can be specialized to a combinatorial algorithm for this problem. We report on preliminary computational results which show that this method can require fewer pivots than the simplex method while the amount of work required for each pivot is comparable. The algorithm can be proven finite using a modification of Cunningham's strongly feasible basis pivoting rule.

Several algorithms already exist for solving the uncapacitated facility location problem. The most efficient are based upon the solution of the strong linear programming relaxation. The dual of this relaxation has a condensed form which consists of minimizing a certain piecewise linear convex function. This paper presents a new method for solving the uncapacitated facility location problem based upon the exact solution of the condensed dual via orthogonal projections. The amount of work per iteration is of the same order as that of a simplex iteration for a linear program inm variables and constraints, wherem is the number of clients. For comparison, the underlying linear programming dual hasmn + m + n variables andmn +n constraints, wheren is the number of potential locations for the facilities. The method is flexible as it can handle side constraints. In particular, when there is a duality gap, the linear programming formulation can be strengthened by adding cuts. Numerical results for some classical test problems are included.

In this paper we give first- and second-order conditions to characterize a local minimizer of an exact penalty function. The form of this characterization gives support to the claim that the exact penalty function and the nonlinear programming problem are closely related. In addition, we demonstrate that there exist arguments for the penalty function from which there are no descent directions even though these points are not minimizers.

The Barrodale and Roberts algorithm for least absolute value (LAV) regression and the algorithm proposed by Bartels and Conn both have the advantage that they are often able to skip across points at which the conventional simplex-method algorithms for LAV regression would be required to carry out an (expensive) pivot operation. We indicate here that this advantage holds in the Bartels-Conn approach for a wider class of problems: the minimization of piecewise linear functions. We show how LAV regression, restricted LAV regression, general linear programming and least maximum absolute value regression can all be easily expressed as piecewise linear minimization problems.

We describe the results of a series of tests for a class of new methods of trust region type for solving the simple bound constrained minimization problem. The results are encouraging and lead us to believe that the methods will prove useful in solving large-scale problems.

In this paper we motivate and describe an algorithm to solve the nonlinear programming problem. The method is based on an exact penalty function and possesses both global and superlinear convergence properties. We establish the global qualities here (the superlinear nature is proven in [7]). The numerical implementation techniques are briefly discussed and preliminary numerical results are given.

This paper introduces a global approach to the semi-infinite programming problem that is based upon a generalisation of the l1 exact penalty function. The advantages are that the ensuing penalty function is exact and the penalties include all violations. The merit function requires integrals for the penalties, which provides a consistent model for the algorithm. The discretization is a result of the approximate quadrature rather than an a priori aspect of the model.

In this paper we consider the final stage of a ‘global’ method to solve the nonlinear programming problem. We prove 2-step superlinear convergence. In the process of analyzing this asymptotic behavior, we compare our method (theoretically) to the popular successive quadratic programming approach.