Optimization Options Reference

Optimization Options

下表介绍了优化选项。 使用optimoptions函数或fminbnd，fminsearch，fzero或lsqnonneg的optimset创建选项。


有关可用选项值和默认值的信息，请参阅各个函数参考页。


选项的默认值根据您使用选项作为输入参数调用的优化函数而有所不同。 您可以通过输入optimoptions（@solvername）或等效的optimoptions（'solvername'）来确定任何优化函数的默认选项值。 例如，

optimoptions(@fmincon)

返回默认“interior-point”fmincon算法的选项列表和默认值。 要找到另一个fmincon算法的默认值，请设置Algorithm选项。 例如，

opts = optimoptions(@fmincon,'Algorithm','sqp')

optimoptions“隐藏”一些选项，这意味着它不显示其值。 这些选项不会出现在此表中。 而是显示在 Hidden Options中。


Optimization Options

Option Name	Description	Used by Functions	Restrictions
AbsoluteGapTolerance	Nonnegative real. intlinprog stops if the difference between the internally calculated upper (U) and lower (L) bounds on the objective function is less than or equal to AbsoluteGapTolerance: U – L <= AbsoluteGapTolerance.	intlinprog	optimoptions only
AbsoluteMaxObjectiveCount	Number of F(x) to minimize the worst case absolute values.	fminimax
Algorithm	Chooses the algorithm used by the solver.	fmincon, fminunc, fsolve, linprog, lsqcurvefit, lsqlin, lsqnonlin, quadprog
BranchRule	Rule for choosing the component for branching: 'maxpscost' — The fractional component with maximum pseudocost. See Branch and Bound. 'mostfractional' — The component whose fractional part is closest to 1/2. 'maxfun' — The fractional component with maximal corresponding component in the absolute value of objective vector f.	intlinprog	optimoptions only
CheckGradients	Compare user-supplied analytic derivatives (gradients or Jacobian, depending on the selected solver) to finite differencing derivatives.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin	optimoptions only. For optimset, use DerivativeCheck
ConstraintTolerance	Tolerance on the constraint violation.	fgoalattain, fmincon, fminimax, fseminf, intlinprog, linprog, lsqlin, quadprog	optimoptions only. For optimset, use TolCon
CutGeneration	Level of cut generation (see Cut Generation): 'none' — No cuts. Makes CutMaxIterations irrelevant. 'basic' — Normal cut generation. 'intermediate' — Use more cut types. 'advanced' — Use most cut types.	intlinprog	optimoptions only
CutMaxIterations	Number of passes through all cut generation methods before entering the branch-and-bound phase, an integer from 1 through 50. Disable cut generation by setting the CutGenerationoption to 'none'.	intlinprog	optimoptions only
Display	Level of display. 'off' displays no output. 'iter' displays output at each iteration, and gives the default exit message. 'iter-detailed' displays output at each iteration, and gives the technical exit message. 'notify' displays output only if the function does not converge, and gives the default exit message. 'notify-detailed' displays output only if the function does not converge, and gives the technical exit message. 'final' displays just the final output, and gives the default exit message. 'final-detailed' displays just the final output, and gives the technical exit message.	All. See the individual function reference pages for the values that apply.
EqualityGoalCount	Specify the number of objectives required for the objective fun to equal the set goal. Reorder your objectives, if necessary, to have fgoalattain achieve the first EqualityGoalCount objectives exactly.	fgoalattain	optimoptions only. For optimset, use GoalsExactAchieve
FiniteDifferenceStepSize	Scalar or vector step size factor for finite differences. When you set FiniteDifferenceStepSize to a vector v, forward finite differences steps delta are delta = v.*sign′(x).*max(abs(x),TypicalX); where sign′(x) = sign(x) except sign′(0) = 1. Central finite differences are delta = v.*max(abs(x),TypicalX); Scalar FiniteDifferenceStepSizeexpands to a vector. The default is sqrt(eps) for forward finite differences, and eps^(1/3) for central finite differences.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin	optimoptions only. For optimset, use FinDiffRelStep
FiniteDifferenceType	Finite differences, used to estimate gradients, are either 'forward' (the default), or 'central' (centered), which takes twice as many function evaluations but should be more accurate. 'central'differences might violate bounds during their evaluation in fmincon interior-point evaluations if the HonorBounds option is set to false.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin	optimoptions only. For optimset, use FinDiffType
FunctionTolerance	Termination tolerance on the function value.	fgoalattain, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, lsqcurvefit, lsqlin, lsqnonlin, quadprog	optimoptions only. For optimset, use TolFun
HessianApproximation	Method of Hessian approximation: 'bfgs', 'lbfgs', {'lbfgs',Positive Integer}, or 'finite-difference'. Ignored when HessianFcn or HessianMultiplyFcn is nonempty.	fmincon	optimoptions only. For optimset, use Hessian
HessianFcn	Function handle to a user-supplied Hessian (see Including Hessians).	fmincon, fminunc	optimoptions only. For optimset, use HessFcn
HessianMultiplyFcn	Handle to a user-supplied Hessian multiply function. Ignored when HessianFcn is nonempty.	fmincon, fminunc, quadprog	optimoptions only. For optimset, use HessMult
Heuristics	Algorithm for searching for feasible points (see Heuristics for Finding Feasible Solutions): 'none' 'rss' 'round' 'rins'	intlinprog	optimoptions only
HeuristicsMaxNodes	Strictly positive integer that bounds the number of nodes intlinprog can explore in its branch-and-bound search for feasible points. See Heuristics for Finding Feasible Solutions.	intlinprog	optimoptions only
HonorBounds	The default true ensures that bound constraints are satisfied at every iteration. Turn off by setting to false.	fmincon	optimoptions only. For optimset, use AlwaysHonorConstraints
IntegerPreprocess	Types of integer preprocessing (see Mixed-Integer Program Preprocessing): 'none' — Use very few integer preprocessing steps. 'basic' — Use a moderate number of integer preprocessing steps. 'advanced' — Use all available integer preprocessing steps.	intlinprog	optimoptions only
IntegerTolerance	Real from 1e-6 through 1e-3, where the maximum deviation from integer that a component of the solution x can have and still be considered an integer. IntegerTolerance is not a stopping criterion.	intlinprog	optimoptions only
JacobianMultiplyFcn	User-defined Jacobian multiply function. Ignored unless SpecifyObjectiveGradient is true for fsolve, lsqcurvefit, and lsqnonlin.	fsolve, lsqcurvefit, lsqlin, lsqnonlin
LPMaxIterations	Strictly positive integer, the maximum number of simplex algorithm iterations per node during the branch-and-bound process.	intlinprog	optimoptions only
LPOptimalityTolerance	Nonnegative real where reduced costs must exceed LPOptimalityTolerance for a variable to be taken into the basis.	intlinprog	optimoptions only
MaxFunctionEvaluations	Maximum number of function evaluations allowed.	fgoalattain, fminbnd, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin	optimoptions only. For optimset, use MaxFunEvals
MaxIterations	Maximum number of iterations allowed.	All but fzero and lsqnonneg	optimoptions only. For optimset, use MaxIter
MaxFeasiblePoints	Strictly positive integer. intlinprog stops if it finds MaxFeasiblePoints integer feasible points.	intlinprog	optimoptions only
MaxNodes	Strictly positive integer that is the maximum number of nodes the solver explores in its branch-and-bound process.	intlinprog
MaxTime	Maximum amount of time in seconds allowed for the algorithm.	intlinprog, linprog
NodeSelection	Choose the node to explore next. 'simplebestproj' — Best projection. See Branch and Bound. 'minobj' — Explore the node with the minimum objective function. 'mininfeas' — Explore the node with the minimal sum of integer infeasibilities. See Branch and Bound.	intlinprog	optimoptions only
ObjectiveCutOff	Real greater than -Inf. The default is Inf.	intlinprog	optimoptions only
ObjectiveImprovementThreshold	Nonnegative real. intlinprog changes the current feasible solution only when it locates another with an objective function value that is at least ObjectiveImprovementThreshold lower: (fold – fnew)/(1 + fold) > ObjectiveImprovementThreshold.	intlinprog	optimoptions only
ObjectiveLimit	If the objective function value goes below ObjectiveLimit and the iterate is feasible, then the iterations halt.	fmincon, fminunc, quadprog
OptimalityTolerance	Termination tolerance on the first-order optimality.	fgoalattain, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, linprog (interior-pointonly), lsqcurvefit, lsqlin, lsqnonlin, quadprog	optimoptions only. For optimset, use TolFun
OutputFcn	Specify one or more user-defined functions that the optimization function calls at each iteration. See Output Function or intlinprog Output Functions and Plot Functions.	fgoalattain, fminbnd, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, fzero, intlinprog, lsqcurvefit, lsqnonlin
PlotFcn	Plots various measures of progress while the algorithm executes. Select from predefined plots or write your own. @optimplotx plots the current point @optimplotfunccount plots the function count @optimplotfval plots the function value @optimplotconstrviolation plots the maximum constraint violation @optimplotresnorm plots the norm of the residuals @optimplotfirstorderopt plots the first-order of optimality @optimplotstepsize plots the step size @optimplotmilp plots the gap for mixed-integer linear programs See Plot Functions or intlinprog Output Functions and Plot Functions.	fgoalattain, fminbnd, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, fzero, intlinprog, lsqcurvefit, lsqnonlin. See the individual function reference pages for the values that apply.	optimoptions only. For optimset, use PlotFcns
RelativeGapTolerance	Real from 0 through 1. intlinprog stops if the relative difference between the internally calculated upper (U) and lower (L) bounds on the objective function is less than or equal toRelativeGapTolerance: (U – L) / (abs(U) + 1) <= RelativeGapTolerance. intlinprog automatically modifies the tolerance for large L magnitudes: tolerance = min(1/(1+|L|), RelativeGapTolerance)	intlinprog	optimoptions only
RootLPAlgorithm	Algorithm for solving linear programs: 'dual-simplex' — Dual simplex algorithm 'primal-simplex' — Primal simplex algorithm	intlinprog	optimoptions only
RootLPMaxIterations	Nonnegative integer that is the maximum number of simplex algorithm iterations to solve the initial linear programming problem.	intlinprog	optimoptions only
ScaleProblem	For fmincon interior-point and sqpalgorithms, true causes the algorithm to normalize all constraints and the objective function by their initial values. Disable by setting to the default false.	fmincon
Simplex Use Algorithm instead	If 'on', function uses the simplex algorithm.	linprog	optimset only
SpecifyConstraintGradient	User-defined gradients for the nonlinear constraints.	fgoalattain, fmincon, fminimax	optimoptions only. For optimset, use GradConstr
SpecifyObjectiveGradient	User-defined gradients or Jacobians for the objective functions.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin	optimoptions only. For optimset, use GradObj or Jacobian
StepTolerance	Termination tolerance on x.	All functions except linprogand lsqlin, and the quadprog'active-set' algorithm	optimoptions only. For optimset, use TolX
SubproblemAlgorithm	Determines how the iteration step is calculated.	fmincon
TypicalX	Array that specifies typical magnitude of array of parameters x. The size of the array is equal to the size of x0, the starting point. Primarily for scaling finite differences for gradient estimation.	fgoalattain, fmincon, fminimax, fminunc, fsolve, lsqcurvefit, lsqlin, lsqnonlin, quadprog
UseParallel	When true, applicable solvers estimate gradients in parallel. Disable by setting to false.	fgoalattain, fmincon, fminimax, fminunc, fsolve, lsqcurvefit, lsqnonlin.

Hidden Options

optimoptions “hides” some options, meaning it does not display their values. To learn how to view these options, and why they are hidden, see View Options.

Function reference pages list these options in italics.

• Hidden Optimization Toolbox Options

• Hidden Global Optimization Toolbox Options

Hidden Optimization Toolbox Options

This table lists the hidden Optimization Toolbox™ options.

Options that optimoptions Hides

Option Name	Description	Used by Functions	Restrictions
Diagnostics	Display diagnostic information about the function to be minimized or solved.	All but fminbnd, fminsearch, fzero, and lsqnonneg
DiffMaxChange	Maximum change in variables for finite differencing.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin
DiffMinChange	Minimum change in variables for finite differencing.	fgoalattain, fmincon, fminimax, fminunc, fseminf, fsolve, lsqcurvefit, lsqnonlin
FunValCheck	Check whether objective function and constraints values are valid. 'on' displays an error when the objective function or constraints return a value that is complex, NaN, or Inf. Note FunValCheck does not return an error for Infwhen used with fminbnd, fminsearch, or fzero, which handle Inf appropriately. 'off' displays no error.	fgoalattain, fminbnd, fmincon, fminimax, fminsearch, fminunc, fseminf, fsolve, fzero, lsqcurvefit, lsqnonlin
HessPattern	Sparsity pattern of the Hessian for finite differencing. The size of the matrix is n-by-n, where n is the number of elements in x0, the starting point.	fmincon, fminunc
HessUpdate	Quasi-Newton updating scheme.	fminunc
InitBarrierParam	Initial barrier value.	fmincon
InitDamping	Initial Levenberg-Marquardt parameter.	fsolve, lsqcurvefit, lsqnonlin	optimoptions only
InitTrustRegionRadius	Initial radius of the trust region.	fmincon
JacobPattern	Sparsity pattern of the Jacobian for finite differencing. The size of the matrix is m-by-n, where mis the number of values in the first argument returned by the user-specified function fun, and n is the number of elements in x0, the starting point.	fsolve, lsqcurvefit, lsqnonlin
LPPreprocess	Type of preprocessing for the solution to the relaxed linear program (see Linear Program Preprocessing): 'none' — No preprocessing. 'basic' — Use preprocessing.	intlinprog	optimoptions only
MaxPCGIter	Maximum number of iterations of preconditioned conjugate gradients method allowed.	fmincon, fminunc, fsolve, lsqcurvefit, lsqlin, lsqnonlin, quadprog
MaxProjCGIter	A tolerance for the number of projected conjugate gradient iterations; this is an inner iteration, not the number of iterations of the algorithm.	fmincon
MaxSQPIter	Maximum number of iterations of sequential quadratic programming method allowed.	fgoalattain, fmincon, fminimax
MeritFunction	Use goal attainment/minimax merit function (multiobjective) vs. fmincon (single objective).	fgoalattain, fminimax
PrecondBandWidth	Upper bandwidth of preconditioner for PCG. Setting to 'Inf' uses a direct factorization instead of CG.	fmincon, fminunc, fsolve, lsqcurvefit, lsqlin, lsqnonlin, quadprog
Preprocess	Level of LP preprocessing prior to simplex or dual simplex algorithm iterations.	linprog	optimoptions only
RelLineSrchBnd	Relative bound on line search step length.	fgoalattain, fmincon, fminimax, fseminf
RelLineSrchBndDuration	Number of iterations for which the bound specified in RelLineSrchBnd should be active.	fgoalattain, fmincon, fminimax, fseminf
ScaleProblem	When using the Algorithmoption 'levenberg-marquardt', setting the ScaleProblem option to 'jacobian' sometimes helps the solver on badly-scaled problems.	fsolve, lsqcurvefit, lsqnonlin
TolConSQP	Constraint violation tolerance for the inner SQP iteration.	fgoalattain, fmincon, fminimax, fseminf
TolPCG	Termination tolerance on the PCG iteration.	fmincon, fminunc, fsolve, lsqcurvefit, lsqlin, lsqnonlin, quadprog
TolProjCG	A relative tolerance for projected conjugate gradient algorithm; this is for an inner iteration, not the algorithm iteration.	fmincon
TolProjCGAbs	Absolute tolerance for projected conjugate gradient algorithm; this is for an inner iteration, not the algorithm iteration.	fmincon

Hidden Global Optimization Toolbox Options

For the reasons these options are hidden, see Options that optimoptions Hides (Global Optimization Toolbox).

Options that optimoptions Hides

Option Name	Used by Functions
Cache	patternsearch
CacheSize	patternsearch
CacheTol	patternsearch
DisplayInterval	particleswarm, simulannealbnd
FunValCheck	particleswarm
HybridInterval	simulannealbnd
InitialPenalty	ga, patternsearch
MaxMeshSize	patternsearch
MeshRotate	patternsearch
MigrationDirection	ga
MigrationFraction	ga
MigrationInterval	ga
PenaltyFactor	ga, patternsearch
PlotInterval	ga, patternsearch, simulannealbnd
StallTest	ga
TolBind	patternsearch