Introduction

Topics covered in this chapter:

This part of the manual is intended to provide a general description of the facilities available for modeling with Xpress NonLinear. It is not an exhaustive list of possibilities, and it does not go into very great depth on some of the more advanced topics. All the functions and formats are given in more detail in the second part of this manual and the Xpress-Mosel Reference Manual (Module mmxnlp section).

Xpress Nonlinear consists of:

• the Xpress Optimizer to solve linear, mixed integer linear, and convex quadratic problems,
• Xpress-SLP which uses Successive Linear Programming to solve non-linear models, and
• Artelys Knitro, which is used as a plugin to solve highly nonlinear models.

Note that FICO Xpress Nonlinear is different from FICO Xpress Global, whose documentation can be found in the Global Solver User Guide. FICO Xpress Nonlinear targets finding locally optimal solutions for nonconvex problems. FICO Xpress Global finds globally optimal solutions for nonconvex problems (which might be orders of magnitude harder). For convex problems, both will find globally optimal solutions. It cannot be easily determined a priori which of the two solvers will be superior in performance, since both tackle the same problem with rather different approaches. Note that if both solvers are licensed, Xpress will by default call the global solver, unless user functions are present. For a particular convex problem, it is worth checking both independently.

The functionalities of Xpress Nonlinear extend those of the Xpress Optimizer. Almost any problem that fits into the problem types supported by the Xpress Optimizer are automatically detected and converted into the appropriate format to take advantage of the power of the optimizer's purpose written algorithms.

Xpress-SLP is in essence, a technique which involves making a linear approximation of the original problem at a chosen point, solving the linear approximation and seeing how "far away" the solution point is from the original chosen point. If it is "sufficiently close" then the solution is said to have converged and the process stops. Otherwise, a new point is chosen, based on the solution, and a new linear approximation is made. This process repeats (iterates) until the solution converges. Although this process will find a solution which is the optimum for the linear approximation, there is no guarantee that the solution will be the optimum for the original non-linear problem (that is to say: it may not be the best possible solution to the original problem). Such a solution is called a "local optimum", because it is a better solution than any others in the immediate neighborhood, but may not be better than one a long way away.

The problem of local optima can be thought of as being like trying to find the deepest valley in a range of mountains. You can find a valley relatively easily (just keep going downhill). However, when you reach it, you have no idea whether there is a deeper valley somewhere else, because the mountains block your view. You have found a local optimum, but you do not know whether it is a global optimum. Indeed, in general, finding a global optimum requires a rigorous search.

While Xpress-SLP is most powerful for large or integer nonlinear problems, Knitro which can take advantage of using second order partial derivative information can be more beneficial for highly nonlinear models.

Mathematical programs

There are many specialised forms of models in mathematical programming, and if such a form can be identified, there are usually much more efficient solution techniques available. This section describes some of the major types of problem that Xpress Nonlinear can identify automatically.

Linear programs

Linear programming (LP) involves solving problems of the form

minimize	cT x
subject to	Ax≤b

and in practice this encompasses, via transformations, any problem whose objective and constraints are linear functions.

Such problems were traditionally solved with the simplex method, although recently interior point methods have come to be favoured for larger instances. Linear programs can be solved quickly, and solution techniques scale to enormous sizes of the matrix A. However, few applications are genuinely linear. It was common in the past, however, to approximate general functions by linear counterparts when LPs were the only class of problem with efficient solution techniques.

Convex quadratic programs

Convex quadratic programming (QP) involves solving problems of the form

minimize	cT x + xT Q x
subject to	Ax≤b

for which the matrix Q is symmetric and positive semi-definite (that is, x^T Q x ≥ 0 for all x). This encompasses, via transformations, all problems with a positive semi-definite Q and linear constraints. Such problems can be solved efficiently by interior point methods, and also by quadratic variants of the simplex method.

Convex quadratically constrained quadratic programs

Convex quadratically constrained quadratic programming (QCQP) involves solving problems of the form

minimize	cT x + xT Q x
subject to	Ax≤b
	qjTx + xT Pj x ≤ dj, ∀j

for which the matrix Q and all matrices P_j are positive semi-definite. The most efficient solution techniques are based on interior point methods.

Second order conic problems

Second order conic problems is a special form of a convex quadratically constrained quadratic program, where although the quadratic matrix is not positive semi-definite, the feasible range of the problem is convex, and there are specialized algorithm to solve them.

minimize	cT x + xT Q x
subject to	Ax≤b
	x is in Cj, ∀j

for which the matrix C_j is a convex second order cone and Q is positive semi-definite. The standard form of a second order cone is x^TI x≤y*y where y is non-negative, or (a rotated second order cone) x^TI x≤y*z where y and z are non-negative. Many quadratic problems can be formulated as a second order convex conic problem, including any convex quadratically constrained quadratic programs. Transformation happens automatically for most convertible problems.

General nonlinear optimization problems

Nonlinear programming (NLP) involves solving problems of the form

minimize	f(x)
subject to	gj(x)≤b, ∀j

where f(x) is an arbitrary function, and g(x) are a set of arbitrary functions. This is the most general type of problem, and any constrained model can be realised in this form via simple transformations.

Until recently, few practical techniques existed for tackling such problems, but it is now possible to solve even large instances using Successive Linear Programming solvers (SLP) or second-order methods.

Mixed integer programs

Mixed-integer programming (MIP), in the most general case, involves solving problems of the form

minimize	f(x)
subject to	gj(x)≤b, ∀j
	xk integral

It can be combined with any of the previous problem types, giving Mixed-Integer Linear Programming (MILP), Mixed-Integer Quadratic Programming (MIQP), Mixed-Integer Quadratically Constrained Quadratic Programming (MIQCQP), Mixed-Integer Second Order Conic Problems (MISOCP) and Mixed-Integer Nonlinear Programming (MINLP). Efficient solution techniques now exist for all of these classes of problem.

Technology Overview

In real-world applications, it is vital to match the right optimization technology to your problem. The FICO Xpress libraries provide dedicated, high performance implementations of optimization technologies for the many model classes commonly appearing in practical applications. This includes solvers for linear programming (LP), mixed integer programming (MIP), convex quadratic programming (QP), and convex quadratically constrained programming (QCQP), general nonlinear programming (NLP), and general mixed-integer nonlinear programming (MINLP).

The Simplex Method

The simplex method is one of the most well-developed and highly studied mathematical programming tools. The solvers in the FICO Xpress Optimizer are the product of over 30 years of research, and include high quality, competitive implementations of the primal and dual simplex methods for both linear and quadratic programs. A key advantage of the simplex method is that it can very quickly reoptimize a problem after it has been modified, which is an important step in solving mixed integer programs.

The Logarithmic Barrier Method

The interior point method of the FICO Xpress Optimizer is a state of the art implementation, with leading performance across a variety of large models. It is capable of solving not only the largest and most difficult linear and convex quadratic programs, but also convex quadratically constrained quadratic and second order conic programs. It includes optimized versions of both infeasible logarithmic barrier methods, and also homogeneous self-dual methods.

Outer approximation schemes

A drawback of the barrier methods is that they are not efficiently warm-started. This makes these methods unattractive for solving several related problems, like the ones arising from a branch and bound search. While for linear and convex quadratic problems the simplex methods can be used, there is no immediate such alternative for convex quadratic constrained and second order methods. To bridge the gap, outer approximation cutting schemes are used, which themselves may be warm started by a barrier solution.

Successive Linear Programming

For general nonlinear programs which are very large, highly structured, or contain a significant linear part, the FICO Xpress Sequential Linear Programming solver (XSLP) offers exceptional performance. Successive linear programming is a first order, iterative approach for solving nonlinear models. At each iteration, a linear approximation to the original problem is solved at the current point, and the distance of the result from the selected point is examined. When the two points are sufficiently close, the solution is said to have converged and the result is returned. This technique is thus based upon solving a sequence of linear programming problems and benefits from the advanced algorithmic and presolving techniques available for linear problems. This makes XSLP scalable, as well as efficient for large problems. In addition, the relatively simple core concepts make understanding the solution process and subsequent tuning comparatively straightforward.

Second Order Methods

Also integrated into the Xpress suite is Knitro from Artelys, a second-order method which is particularly suited to large-scale continuous problems containing high levels of nonlinearity. Second order methods approximate a problem by examining quadratic programs fitted to a local region. This can provide information about the curvature of the solution space to the solver, which first-order methods do not have. Advanced implementations of such methods, like Knitro, may as a result be able to produce more resilient solutions. This can be especially noticeable when the initial point is close to a local optimum.

Mixed Integer Solvers

The FICO Xpress MIP solver is a highly scalable parallel branch and bound framework for all classes of mixed integer programs. It is based on a branch and bound search utilizing continuous solvers, advanced cutting planes, in-tree presolving and multiple heuristics, for discovering primal solutions and tightening best bounds. The search is guided by advanced methods for selecting branching variables and estimating sub-tree sizes/efforts.

Mixed integer programming forms the basis of many important applications, and the implementation in the FICO Xpress Suite has proven itself in operation for some of the world's largest organizations. For mixed integer nonlinear problems (MINLPs), we offer FICO Xpress Global to solve convex and non-convex, continuous and mixed-integer, problems to proven global optimality. Furthermore, FICO Xpress Nonlinear offers black-box optimization and some enhanced nonlinear features in a local solver context.

API naming convention

Xpress Nonlinear has been developed as an extension to the XPRS library building on the SLP solver technology, which is reflected in the naming convention. All XPRS API functions are used the same way as normal to build the linear part of the problem, while the API functions prefixed with XSLP are used for all nonlinear aspects, independently of how the problem is solved afterwards (convex quadratic problems by a dedicated solver or Knitro instead of SLP). Some controls have both an XPRS and an XSLP counterpart, for example "XPRS_PRESOLVE" and "XSLP_PRESOLVE". In such cases, "XSLP_PRESOLVE" refers to the nonlinear presolver (even if another solver than SLP is used to solve the problem afterwards) and "XPRS_PRESOLVE" refers to problems that are not deemed as general nonlinear (LP, MIP or convex quadratic); in such cases, if SLP solves one of such problems as part of its iterative process, the XPRS control is respected for such sub-solves.