Quadratic Programming Algorithms
Quadratic Programming Definition
Quadratic programming is the problem of finding a vector x thatminimizes a quadratic function, possibly subject to linear constraints:
(1) |
such that A·x≤b, Aeq·x=beq, l≤x≤u.
interior-point-convex
quadprog
Algorithm
The interior-point-convex
algorithm performs thefollowing steps:
Presolve/Postsolve
Generate Initial Point
Predictor-Corrector
Stopping Conditions
Infeasibility Detection
Note
The algorithm has two code paths. It takes one when the Hessianmatrix H is an ordinary (full) matrix of doubles,and it takes the other when H is a sparse matrix.For details of the sparse data type, see Sparse Matrices.Generally, the algorithm is faster for large problems that have relativelyfew nonzero terms when you specify H as sparse. Similarly, the algorithm is fasterfor small or relatively dense problems when you specify H as full.
Presolve/Postsolve
The algorithm first tries to simplify the problem by removing redundancies and simplifying constraints. The tasks performed during the presolve step can include the following:
Check if any variables have equal upper and lowerbounds. If so, check for feasibility, and then fix and remove thevariables.
Check if any linear inequality constraint involves only one variable. If so, check for feasibility, and then change the linear constraint to a bound.
Check if any linear equality constraint involves only one variable. If so, check for feasibility, and then fix and remove the variable.
Check if any linear constraint matrix has zero rows. If so, check for feasibility, and then delete the rows.
Determine if the bounds and linear constraints are consistent.
Check if any variables appear only as linear terms in the objective function and do not appear in any linear constraint. If so, check for feasibility and boundedness, and then fix the variables at their appropriate bounds.
Change any linear inequality constraints to linearequality constraints by adding slack variables.
If the algorithm detects an infeasible or unbounded problem, it halts and issues an appropriate exit message.
The algorithm might arrive at a single feasible point, whichrepresents the solution.
If the algorithm does not detect an infeasible or unbounded problem in the presolve step, and if the presolve has not produced the solution, the algorithm continues to its next steps. After reaching a stopping criterion, the algorithm reconstructs the original problem, undoing any presolve transformations. This final step is the postsolve step.
For details, see Gould and Toint [63].
Generate Initial Point
The initial point x0
for the algorithm is:
Initialize
x0
toones(n,1)
, wheren
is the number of rows in H.For components that have both an upper bound
ub
anda lower boundlb
, if a component ofx0
isnot strictly inside the bounds, the component is set to(ub+lb)/2
.For components that have only one bound, modify thecomponent if necessary to lie strictly inside the bound.
Take a predictor step (see Predictor-Corrector), with minor corrections for feasibility,not a full predictor-corrector step. This places the initial pointcloser to the central path without entailingthe overhead of a full predictor-corrector step. For details of thecentral path, see Nocedal and Wright [8], page 397.
Predictor-Corrector
The sparse and full interior-point-convex algorithms differmainly in the predictor-corrector phase. The algorithms are similar,but differ in some details. For the basic algorithm description, seeMehrotra [47].
The algorithms begin by turning the linear inequalities Ax <= b into inequalities of the form Ax >= b by multiplying A and b by -1. This has no bearing on the solution, but makes the problem of the same form found in some literature.
Sparse Predictor-Corrector
Full Predictor-Corrector
Sparse Predictor-Corrector.Similar to the fmincon
interior-point algorithm,the sparse interior-point-convex
algorithm tries to finda point where the Karush-Kuhn-Tucker (KKT) conditionshold. For the quadratic programming problem described in Quadratic Programming Definition, these conditions are:
Here
is the extendedlinear inequality matrix that includes bounds written as linear inequalities. is the corresponding linearinequality vector, including bounds.
s is the vector of slacks thatconvert inequality constraints to equalities. s haslength m, the number of linear inequalities andbounds.
z is the vector of Lagrange multiplierscorresponding to s.
y is the vector of Lagrange multipliersassociated with the equality constraints.
The algorithm first predicts a step from the Newton-Raphsonformula, then computes a corrector step. The corrector attempts tobetter enforce the nonlinear constraint sizi=0.
Definitions for the predictor step:
rd, thedual residual:
req, theprimal equality constraint residual:
rineq, theprimal inequality constraint residual, which includes bounds and slacks:
rsz, thecomplementarity residual:
rsz = Sz.
S is the diagonal matrix of slack terms, z isthe column matrix of Lagrange multipliers.
rc, theaverage complementarity:
In a Newton step, the changes in x, s, y,and z, are given by:
(2) |
However, a full Newton step might be infeasible, because ofthe positivity constraints on s and z.Therefore, quadprog
shortens the step, if necessary,to maintain positivity.
Additionally, to maintain a “centered” positionin the interior, instead of trying to solve sizi=0, the algorithmtakes a positive parameter σ, and tries tosolve
sizi = σrc.
quadprog
replaces rsz inthe Newton step equation with rsz+ΔsΔz–σrc1, where 1 isthe vector of ones. Also, quadprog
reorders theNewton equations to obtain a symmetric, more numerically stable systemfor the predictor step calculation.
After calculating the corrected Newton step, the algorithm performsmore calculations to get both a longer current step, and to preparefor better subsequent steps. These multiple correction calculationscan improve both performance and robustness. For details, see Gondzio [5].
Full Predictor-Corrector.The full predictor-corrector algorithm does not combine boundsinto linear constraints, so it has another set of slack variablescorresponding to the bounds. The algorithm shifts lower bounds tozero. And, if there is only one bound on a variable, the algorithmturns it into a lower bound of zero, by negating the inequality ofan upper bound.
is the extended linear matrixthat includes both linear inequalities and linear equalities. is the corresponding linearequality vector. also includesterms for extending the vector x with slack variables s thatturn inequality constraints to equality constraints:
where x0 means theoriginal x vector.
The KKT conditions are
(3) |
To find the solution x, slack variables anddual variables to Equation3,the algorithm basically considers a Newton-Raphson step:
(4) |
where X, V, W,and T are diagonal matrices corresponding to thevectors x, v, w,and t respectively. The residual vectors on thefar right side of the equation are:
rd, thedual residual
rp, theprimal residual
rub, theupper bound residual
rvx, thelower bound complementarity residual
rwt, theupper bound complementarity residual
The algorithm solves Equation4 by first converting it to the symmetricmatrix form
(5) |
where
All the matrix inverses in the definitions of D and R aresimple to compute because the matrices are diagonal.
To derive Equation5 from Equation4, notice that thesecond row of Equation5 isthe same as the second matrix row of Equation4. The first row of Equation5 comes from solvingthe last two rows of Equation4 for Δv andΔw, and then solving for Δt.
To solve Equation5,the algorithm follows the essential elements of Altman and Gondzio [1]. The algorithm solves the symmetricsystem by an LDL decomposition. As pointed out by authors such asVanderbei and Carpenter [2], this decomposition is numericallystable without any pivoting, so can be fast.
After calculating the corrected Newton step, the algorithm performsmore calculations to get both a longer current step, and to preparefor better subsequent steps. These multiple correction calculationscan improve both performance and robustness. For details, see Gondzio [5].
The full quadprog
predictor-corrector algorithmis largely the same as that in the linprog
'interior-point'
algorithm,but includes quadratic terms as well. See Predictor-Corrector.
References
[1] Altman, Anna and J. Gondzio. Regularized symmetric indefinite systems in interior point methods for linear and quadratic optimization. Optimization Methods and Software, 1999. Available for download here.
[2] Vanderbei, R. J. and T. J. Carpenter. Symmetric indefinite systems for interior point methods. Mathematical Programming 58, 1993. pp. 1–32. Available for download here.
Stopping Conditions
The predictor-corrector algorithm iterates until it reachesa point that is feasible (satisfies the constraints to within tolerances)and where the relative step sizes are small. Specifically, define
The algorithm stops when all of these conditions are satisfied:
where
rc essentially measuresthe size of the complementarity residuals xv and tw,which are each vectors of zeros at a solution.
Infeasibility Detection
quadprog
calculates a merit function φ atevery iteration. The merit function is a measure of feasibility. quadprog
stopsif the merit function grows too large. In this case, quadprog
declaresthe problem to be infeasible.
The merit function is related to the KKT conditions for theproblem—see Predictor-Corrector.Use the following definitions:
The notation and means the linear inequalitycoefficients, augmented with terms to represent bounds for the sparsealgorithm. The notation similarly representsLagrange multipliers for the linear inequality constraints, includingbound constraints. This was called z in Predictor-Corrector, and was called y.
The merit function φ is
If this merit function becomes too large, quadprog
declaresthe problem to be infeasible and halts with exit flag -2
.
trust-region-reflective
quadprog
Algorithm
Many of the methods used in Optimization Toolbox™ solversare based on trust regions, a simple yet powerfulconcept in optimization.
To understand the trust-region approach to optimization, considerthe unconstrained minimization problem, minimize f(x),where the function takes vector arguments and returns scalars. Supposeyou are at a point x in n-spaceand you want to improve, i.e., move to a point with a lower functionvalue. The basic idea is to approximate f witha simpler function q, which reasonably reflectsthe behavior of function f in a neighborhood N aroundthe point x. This neighborhood is the trust region.A trial step s is computed by minimizing (or approximatelyminimizing) over N. This is the trust-region subproblem,
(6) |
The current point is updated to be x+s if f(x+s)<f(x);otherwise, the current point remains unchanged and N,the region of trust, is shrunk and the trial step computation is repeated.
The key questions in defining a specific trust-region approachto minimizing f(x) are how tochoose and compute the approximation q (definedat the current point x), how to choose and modifythe trust region N, and how accurately to solvethe trust-region subproblem. This section focuses on the unconstrainedproblem. Later sections discuss additional complications due to thepresence of constraints on the variables.
In the standard trust-region method ([48]), the quadratic approximation q isdefined by the first two terms of the Taylor approximation to F at x;the neighborhood N is usually spherical or ellipsoidalin shape. Mathematically the trust-region subproblem is typicallystated
(7) |
where g is the gradient of f at the current point x, H is the Hessian matrix (the symmetric matrix of second derivatives), D is a diagonal scaling matrix, Δ is a positive scalar, and ‖ . ‖ is the 2-norm. Good algorithms exist for solving Equation7 (see [48]); such algorithms typically involve the computation of all eigenvalues of H and a Newton process applied to the secular equation
Such algorithms provide an accurate solution to Equation7. However, theyrequire time proportional to several factorizations of H.Therefore, for large-scale problems a different approach is needed.Several approximation and heuristic strategies, based on Equation7, have been proposedin the literature ([42] and [50]). The approximation approach followedin Optimization Toolbox solvers is to restrict the trust-regionsubproblem to a two-dimensional subspace S ([39] and [42]).Once the subspace S has been computed, the workto solve Equation7 istrivial even if full eigenvalue/eigenvector information is needed(since in the subspace, the problem is only two-dimensional). Thedominant work has now shifted to the determination of the subspace.
The two-dimensional subspace S isdetermined with the aid of a preconditionedconjugate gradient process described below. The solver defines S asthe linear space spanned by s1 and s2,where s1 is in the directionof the gradient g, and s2 iseither an approximate Newton direction, i.e.,a solution to
(8) |
or a direction of negative curvature,
(9) |
The philosophy behind this choice of S isto force global convergence (via the steepest descent direction ornegative curvature direction) and achieve fast local convergence (viathe Newton step, when it exists).
A sketch of unconstrained minimization using trust-region ideasis now easy to give:
Formulate the two-dimensional trust-region subproblem.
Solve Equation7 to determine the trial step s.
If f(x + s)< f(x),then x = x + s.
Adjust Δ.
These four steps are repeated until convergence. The trust-regiondimension Δ is adjusted according to standard rules. In particular,it is decreased if the trial step is not accepted, i.e., f(x + s)≥ f(x).See [46] and [49] for a discussion of this aspect.
Optimization Toolbox solvers treat a few important specialcases of f with specialized functions: nonlinearleast-squares, quadratic functions, and linear least-squares. However,the underlying algorithmic ideas are the same as for the general case.These special cases are discussed in later sections.
The subspace trust-region method is used to determine a searchdirection. However, instead of restricting the step to (possibly)one reflection step, as in the nonlinear minimization case, a piecewise reflective linesearch is conducted at each iteration. See [45] for details of the line search.
Preconditioned Conjugate Gradient Method
A popular way to solve large, symmetric, positive definite systems of linear equations Hp=–g is the method of Preconditioned Conjugate Gradients (PCG). This iterative approach requires the ability to calculate matrix-vector products of the form H·v where v is an arbitrary vector. The symmetric positive definite matrix M is a preconditioner for H. That is, M=C2, where C–1HC–1 is a well-conditioned matrix or a matrix with clustered eigenvalues.
In a minimization context, you can assume that the Hessian matrix H is symmetric. However, H is guaranteed to be positive definite only in the neighborhood of a strong minimizer. Algorithm PCG exits when it encounters a direction of negative (or zero) curvature, that is, dTHd≤0. The PCG output direction p is either a direction of negative curvature or an approximate solution to the Newton system Hp=–g. In either case, p helps to define the two-dimensional subspace used in the trust-region approach discussed in Trust-Region Methods for Nonlinear Minimization.
Linear Equality Constraints
Linear constraints complicate the situation described for unconstrainedminimization. However, the underlying ideas described previously canbe carried through in a clean and efficient way. The trust-regionmethods in Optimization Toolbox solvers generate strictly feasibleiterates.
The general linear equality constrained minimization problemcan be written
(10) |
where A is an m-by-n matrix(m≤n). Some Optimization Toolbox solverspreprocess A to remove strict linear dependenciesusing a technique based on the LU factorization of AT [46]. Here A is assumedto be of rank m.
The method used to solve Equation10 differs from the unconstrained approachin two significant ways. First, an initial feasible point x0 iscomputed, using a sparse least-squares step, so that Ax0=b.Second, Algorithm PCG is replaced with Reduced Preconditioned ConjugateGradients (RPCG), see [46], in orderto compute an approximate reduced Newton step (or a direction of negativecurvature in the null space of A). The key linearalgebra step involves solving systems of the form
(11) |
where approximates A (smallnonzeros of A are set to zero provided rank isnot lost) and C is a sparse symmetric positive-definiteapproximation to H, i.e., C=H.See [46] for more details.
Box Constraints
The box constrained problem is of the form
(12) |
where l is a vector of lower bounds, and u isa vector of upper bounds. Some (or all) of the components of l canbe equal to –∞ and some (or all) of the components of u canbe equal to ∞. The method generates a sequence of strictlyfeasible points. Two techniques are used to maintain feasibility whileachieving robust convergence behavior. First, a scaled modified Newtonstep replaces the unconstrained Newton step (to define the two-dimensionalsubspace S). Second, reflectionsare used to increase the step size.
The scaled modified Newton step arises from examining the Kuhn-Tuckernecessary conditions for Equation12,
(13) |
where
and the vector v(x) isdefined below, for each 1≤i≤n:
If gi<0 and ui<∞ then vi=xi–ui
If gi≥0 and li>–∞ then vi=xi–li
If gi<0 and ui=∞ then vi=–1
If gi≥0 and li=–∞ then vi=1
The nonlinear system Equation13 is not differentiable everywhere.Nondifferentiability occurs when vi=0. You can avoidsuch points by maintaining strict feasibility, i.e., restricting l<x<u.
The scaled modified Newton step sk forthe nonlinear system of equations given by Equation13 is defined as the solution to thelinear system
(14) |
at the kth iteration, where
(15) |
and
(16) |
Here Jv playsthe role of the Jacobian of |v|. Each diagonalcomponent of the diagonal matrix Jv equals0, –1, or 1. If all the components of l and u arefinite, Jv=diag(sign(g)).At a point where gi=0, vi mightnot be differentiable. is defined atsuch a point. Nondifferentiability of this type is not a cause forconcern because, for such a component, it is not significant whichvalue vi takes. Further,|vi| will still be discontinuousat this point, but the function |vi|·gi iscontinuous.
Second, reflections are used to increasethe step size. A (single) reflection step is defined as follows. Givena step p that intersects a bound constraint, considerthe first bound constraint crossed by p; assumeit is the ith bound constraint (either the ithupper or ith lower bound). Then the reflectionstep pR=p exceptin the ith component, where pRi=–pi.
active-set
quadprog
Algorithm
After completing a presolve step, the active-set
algorithm proceeds in two phases.
Phase 1 — Obtain a feasible point with respect to all constraints.
Phase 2 — Iteratively lower the objective function while maintaining a list of the active constraints and maintaining feasibility in each iteration.
The active-set
strategy (also known as a projection method) is similar to the strategy of Gill et al., described in [18] and [17].
Presolve Step
The algorithm first tries to simplify the problem by removing redundancies and simplifying constraints. The tasks performed during the presolve step can include the following:
Check if any variables have equal upper and lowerbounds. If so, check for feasibility, and then fix and remove thevariables.
Check if any linear inequality constraint involves only one variable. If so, check for feasibility, and then change the linear constraint to a bound.
Check if any linear equality constraint involves only one variable. If so, check for feasibility, and then fix and remove the variable.
Check if any linear constraint matrix has zero rows. If so, check for feasibility, and then delete the rows.
Determine if the bounds and linear constraints are consistent.
Check if any variables appear only as linear terms in the objective function and do not appear in any linear constraint. If so, check for feasibility and boundedness, and then fix the variables at their appropriate bounds.
Change any linear inequality constraints to linearequality constraints by adding slack variables.
If the algorithm detects an infeasible or unbounded problem, it halts and issues an appropriate exit message.
The algorithm might arrive at a single feasible point, whichrepresents the solution.
If the algorithm does not detect an infeasible or unbounded problem in the presolve step, and if the presolve has not produced the solution, the algorithm continues to its next steps. After reaching a stopping criterion, the algorithm reconstructs the original problem, undoing any presolve transformations. This final step is the postsolve step.
For details, see Gould and Toint [63].
Phase 1 Algorithm
In Phase 1, the algorithm attempts to find a point x
that satisfies all constraints, with no consideration of the objective function. quadprog
runs the Phase 1 algorithm only if the supplied initial point x0
is infeasible.
To begin, the algorithm tries to find a point that is feasible with respect to all equality constraints, such as X = Aeq\beq
. If there is no solution x
to the equations Aeq*x = beq
, then the algorithm halts. If there is a solution X
, the next step is to satisfy the bounds and linear inequalities. In the case of no equality constraints set X = x0
, the initial point.
Starting from X
, the algorithm calculates M = max(A*X – b, X - ub, lb – X)
. If M <= options.ConstraintTolerance
, then the point X
is feasible and the Phase 1 algorithm halts.
If M > options.ConstraintTolerance
, the algorithm introduces a nonnegative slack variable γ for the auxiliary linear programming problem
such that
Here, ρ is the ConstraintTolerance
option multiplied by the absolute value of the largest element in A
and Aeq
. If the algorithm finds γ = 0 and a point x that satisfies the equations and inequalities, then x is a feasible Phase 1 point. If there is no solution to the auxiliary linear programming problem x with γ = 0, then the Phase 1 problem is infeasible.
To solve the auxiliary linear programming problem, the algorithm sets γ0 = M + 1, sets x0 = X
, and initializes the active set as the fixed variables (if any) and all the equality constraints. The algorithm reformulates the linear programming variables p to be the offset of x from the current point x0, namely x = x0 + p. The algorithm solves the linear programming problem by the same iterations as it takes in Phase 2 to solve the quadratic programming problem, with an appropriately modified Hessian.
Phase 2 Algorithm
In terms of a variable d, the problem is
(17) |
Here, Ai refers to the i
th row of the m-by-n matrix A.
During Phase 2, an active set , which is an estimate of the active constraints (those on the constraint boundaries) at the solution point.
The algorithm updates at each iteration k, forming the basis for a search direction dk. Equality constraints always remain in the active set . The search direction dk is calculated and minimizes the objective function while remaining on any active constraint boundaries. The algorithm forms the feasible subspace for dk from a basis Zk whose columns are orthogonal to the estimate of the active set (that is, ). Therefore, a search direction, which is formed from a linear summation of any combination of the columns of Zk, is guaranteed to remain on the boundaries of the active constraints.
The algorithm forms the matrix Zk from the last n–l columns of the QR decomposition of the matrix , where l is the number of active constraints and l < n. That is, Zk is given by
(18) |
where
After finding Zk, the algorithm looks for a new search direction dk that minimizes q(d), where dk is in the null space of the active constraints. That is, dk is a linear combination of the columns of Zk: for some vector p.
Viewing the quadratic as a function of p by substituting for dk, gives
(19) |
Differentiating this equation with respect to p yields
(20) |
∇q(p) is referred to as the projected gradient of the quadratic function because it is the gradient projected in the subspace defined by Zk. The term is called the projected Hessian. Assuming the projected Hessian matrix is positive semidefinite, the minimum of the function q(p) in the subspace defined by Zk occurs when ∇q(p)=0, which is the solution of the system of linear equations
(21) |
The algorithm then takes a step of the form
where
Due to the quadratic nature of the objective function, only two choices of step length α exist at each iteration. A step of unity along dk is the exact step to the minimum of the function restricted to the null space of . If the algorithm can take such a step without violating the constraints, then this step is the solution to the quadratic program (Equation18). Otherwise, the step along dk to the nearest constraint is less than unity, and the algorithm includes a new constraint in the active set at the next iteration. The distance to the constraint boundaries in any direction dk is given by
which is defined for constraints not in the active set, and where the direction dk is towards the constraint boundary, that is, .
When the active set includes n independent constraints, without location of the minimum, the algorithm calculates the Lagrange multipliers λk, which satisfy the nonsingular set of linear equations
(22) |
If all elements of λk are nonnegative, xk is the optimal solution of the quadratic programming problem Equation1. However, if any component of λk is negative, and the component does not correspond to an equality constraint, then the minimization is not complete. The algorithm deletes the element corresponding to the most negative multiplier and starts a new iteration.
Sometimes, when the solver finishes with all nonnegative Lagrange multipliers, the first-order optimality measure is above the tolerance, or the constraint tolerance is not met. In these cases, the solver attempts to reach a better solution by following the restart procedure described in [1]. In this procedure, the solver discards the current set of active constraints, relaxes the constraints a bit, and constructs a new set of active constraints while attempting to solve the problem in a manner that avoids cycling (repeatedly returning to the same state). If necessary, the solver can perform the restart procedure several times.
Note
Do not try to stop the algorithm early by setting large values of the ConstraintTolerance
and OptimalityTolerance
options. Generally, the solver iterates without checking these values until it reaches a potential stopping point, and only then checks to see whether the tolerances are satisfied.
Occasionally, the active-set
algorithm can have difficulty detecting when a problem is unbounded. This issue can occur if the direction of unboundedness v is a direction where the quadratic term v'Hv = 0. For numerical stability and to enable a Cholesky factorization, the active-set
algorithm adds a small, strictly convex term to the quadratic objective. This small term causes the objective function to be bounded away from –Inf
. In this case, the active-set
algorithm reaches an iteration limit instead of reporting that the solution is unbounded. In other words, the algorithm halts with exit flag 0
instead of –3
.
References
[1] Gill, P. E., W. Murray, M. A. Saunders, and M. H. Wright. A practical anti-cycling procedure for linearly constrained optimization. Math. Programming 45 (1), August 1989, pp. 437–474.
Warm Start
When you run the quadprog
or lsqlin
'active-set'
algorithm with a warm start object as the start point, the solver attempts to skip many of the Phase 1 and Phase 2 steps. The warm start object contains the active set of constraints, and this set can be correct or close to correct for the new problem. Therefore, the solver can avoid iterations to add constraints to the active set. Also, the initial point might be close to the solution for the new problem. For more information, see optimwarmstart.
MATLAB Command
You clicked a link that corresponds to this MATLAB command:
Run the command by entering it in the MATLAB Command Window. Web browsers do not support MATLAB commands.
Select a Web Site
Choose a web site to get translated content where available and see local events and offers. Based on your location, we recommend that you select: .
You can also select a web site from the following list:
Americas
- América Latina (Español)
- Canada (English)
- United States (English)
Europe
- Belgium (English)
- Denmark (English)
- Deutschland (Deutsch)
- España (Español)
- Finland (English)
- France (Français)
- Ireland (English)
- Italia (Italiano)
- Luxembourg (English)
- Netherlands (English)
- Norway (English)
- Österreich (Deutsch)
- Portugal (English)
- Sweden (English)
- Switzerland
- Deutsch
- English
- Français
- United Kingdom (English)
Asia Pacific
- Australia (English)
- India (English)
- New Zealand (English)
- 中国
- 日本 (日本語)
- 한국 (한국어)
Contact your local office