# Difference between revisions of "Trust-region methods"

Yewenhe0904 (Talk | contribs) (→Introduction) |
Yewenhe0904 (Talk | contribs) |
||

Line 3: | Line 3: | ||

Date Presented: Apr. 10, 2014 | Date Presented: Apr. 10, 2014 | ||

− | ==Introduction== | + | ==Introduction== |

+ | [[File: Trust-Region Method Overview.png|thumb|right|A pictorial view of trust-region method optimization trajectory]] | ||

Trust-region method (TRM) is one of the most important numerical optimization methods in solving nonlinear programming (NLP) problems. It works in a way that first define a region around the current best solution, in which a certain model (usually a quadratic model) can to some extent approximate the original objective function. TRM then take a step forward according to the model depicts within the region. Unlike the line search methods, TRM usually determines the step size before the improving direction (or at the same time). If a notable decrease (our following discussion will based on minimization problems) is gained after the step forward, then the model is believed to be a good representation of the original objective function. If the improvement is too subtle or even a negative improvement is gained, then the model is not to be believed as a good representation of the original objective function within that region. The convergence can be ensured that the size of the “trust region” (usually defined by the radius in Euclidean norm) in each iteration would depend on the improvement previously made. | Trust-region method (TRM) is one of the most important numerical optimization methods in solving nonlinear programming (NLP) problems. It works in a way that first define a region around the current best solution, in which a certain model (usually a quadratic model) can to some extent approximate the original objective function. TRM then take a step forward according to the model depicts within the region. Unlike the line search methods, TRM usually determines the step size before the improving direction (or at the same time). If a notable decrease (our following discussion will based on minimization problems) is gained after the step forward, then the model is believed to be a good representation of the original objective function. If the improvement is too subtle or even a negative improvement is gained, then the model is not to be believed as a good representation of the original objective function within that region. The convergence can be ensured that the size of the “trust region” (usually defined by the radius in Euclidean norm) in each iteration would depend on the improvement previously made. | ||

− | |||

− | |||

==Important Concepts== | ==Important Concepts== | ||

+ | [[File:TRM_stepsize.png|thumb|right|The picture shows both the stepsize and the improving direction is a consequence of a pre-determined trust-region size.]] | ||

'''Trust-region''' | '''Trust-region''' | ||

In most cases, the trust-region is defined as a spherical area of radius <math>\Delta_k</math> in which the trust-region subproblem lies. | In most cases, the trust-region is defined as a spherical area of radius <math>\Delta_k</math> in which the trust-region subproblem lies. | ||

+ | |||

'''Trust-region subproblem''' | '''Trust-region subproblem''' | ||

Line 24: | Line 25: | ||

Where <math>\Delta_k</math> is the trust region radius, <math>g_k</math> is the gradient at current point and <math>B_k</math> is the hessian (or a hessian approximation). It is easy to find the solution to the trust-region subproblem if <math>B_k</math> is positive definite. | Where <math>\Delta_k</math> is the trust region radius, <math>g_k</math> is the gradient at current point and <math>B_k</math> is the hessian (or a hessian approximation). It is easy to find the solution to the trust-region subproblem if <math>B_k</math> is positive definite. | ||

+ | |||

'''Actual reduction and predicted reduction''' | '''Actual reduction and predicted reduction''' | ||

Line 33: | Line 35: | ||

Whether to take a more ambitious step or a more conservative one is depend on the ratio between the actual reduction gained by true reduction in the original objective function and the predicted reduction expected in the model function. Empirical threshold values of the ratio <math>\rho_k</math> will guide us in determining the size of the trust-region. | Whether to take a more ambitious step or a more conservative one is depend on the ratio between the actual reduction gained by true reduction in the original objective function and the predicted reduction expected in the model function. Empirical threshold values of the ratio <math>\rho_k</math> will guide us in determining the size of the trust-region. | ||

− | |||

− | |||

− | |||

==Trust Region Algorithm== | ==Trust Region Algorithm== | ||

Line 80: | Line 79: | ||

==Methods of Solving the Trust-region Subproblem== | ==Methods of Solving the Trust-region Subproblem== | ||

===Cauchy point calculation=== | ===Cauchy point calculation=== | ||

+ | [[File:TRM_CauchyPoint.png|thumb|right|Illustration of Cauchy point calculation.]] | ||

In line search methods, we may find an improving direction from the gradient information, that is, by taking the steepest descent direction with regard to the maximum range we could make. We can solve the trust-region subproblem in an inexpensive way. This method is also denoted as the Cauchy point calculation. We can also express the improving step explicitly by the following closed-form equations | In line search methods, we may find an improving direction from the gradient information, that is, by taking the steepest descent direction with regard to the maximum range we could make. We can solve the trust-region subproblem in an inexpensive way. This method is also denoted as the Cauchy point calculation. We can also express the improving step explicitly by the following closed-form equations | ||

Line 88: | Line 88: | ||

otherwise <math>\tau_k=min~ ({||g_k||}^3/(\Delta_k{g_k}^TB_kg_k),1)</math> | otherwise <math>\tau_k=min~ ({||g_k||}^3/(\Delta_k{g_k}^TB_kg_k),1)</math> | ||

− | |||

− | |||

− | |||

===Limitations and Further Improvements=== | ===Limitations and Further Improvements=== | ||

Though Cauchy point is cheap to implement, like the steepest descent method, it performs poorly in some cases. Varies kinds of improvements are based on including the curvature information from <math>B_k</math>. | Though Cauchy point is cheap to implement, like the steepest descent method, it performs poorly in some cases. Varies kinds of improvements are based on including the curvature information from <math>B_k</math>. | ||

====Dogleg Method==== | ====Dogleg Method==== | ||

− | If <math>B_k</math> is positive definite (we can use quasi-Newton | + | [[File:TRM_Dogleg.png|thumb|right|Illustration of dogleg method trajectory.]] |

+ | If <math>B_k</math> is positive definite (we can use quasi-Newton Hessian approximation &updating to guarantee), then a V-shaped trajectory can be determined by | ||

if <math>0<=\tau<=1,~~p(\tau)=\tau p^U</math> | if <math>0<=\tau<=1,~~p(\tau)=\tau p^U</math> | ||

Line 101: | Line 99: | ||

if <math>1<=\tau<=2,~~p(\tau)=\tau p^U+(\tau-1)(p^B-p^U)</math> | if <math>1<=\tau<=2,~~p(\tau)=\tau p^U+(\tau-1)(p^B-p^U)</math> | ||

− | where <math>p^U=-\frac{g^Tg}{g^TBg}g</math> is the steepest descent direction | + | where <math>p^U=-\frac{g^Tg}{g^TBg}g</math> is the steepest descent direction and <math>p^B</math> is the optimal solution of the quadratic model <math>m_k(p)</math>. Therefore, a further improvement could be achieved compared to using only Cauchy point calculation method in one iteration. (Note that hessian or approximate hessian will be evaluated in dogleg method) |

− | + | ||

− | Note that hessian or approximate hessian will be evaluated | + | |

− | + | ||

− | + | ||

− | |||

====Conjugated Gradient Steihaug’s Method==== | ====Conjugated Gradient Steihaug’s Method==== | ||

− | The most widely used method for solving a trust-region sub-problem is by using the idea of conjugated gradient (CG) method for minimizing a quadratic function since CG guarantees convergence within a finite number of iterations for a quadratic programming. Also, CG Steihaug’s method has the merit of Cauchy point and | + | The most widely used method for solving a trust-region sub-problem is by using the idea of conjugated gradient (CG) method for minimizing a quadratic function since CG guarantees convergence within a finite number of iterations for a quadratic programming. Also, CG Steihaug’s method has the merit of Cauchy point calculation and dogleg method that both in terms of super-linear convergence rate and inexpensiveness to compute.(No expensive Hessian evaluation) |

---- | ---- | ||

Line 166: | Line 159: | ||

<math>\Delta_1=0.050, \Delta_M=5.0, t_1=0.25, t_2=2.0, \eta_1=0.2,\eta_2=0.25,\eta_3=0.75</math> | <math>\Delta_1=0.050, \Delta_M=5.0, t_1=0.25, t_2=2.0, \eta_1=0.2,\eta_2=0.25,\eta_3=0.75</math> | ||

− | |||

− | |||

− | |||

---- | ---- | ||

− | '''Improving Process''' | + | '''Improving Process''' [[File:TRM-Ex1.png|thumb|right|Optimization trajectory of the example function(unconstrained).]] |

<math>Iteration~~~~ 1~~~ Obj.Value=18.3940 ~~~x_1=1.0000~~~ x_2=1.0000~~~ \rho_k=0.9980~~~ \Delta_k=0.0500~~~ norm (p_k)=0.0500~~~ norm( g_k)=18.3940</math> | <math>Iteration~~~~ 1~~~ Obj.Value=18.3940 ~~~x_1=1.0000~~~ x_2=1.0000~~~ \rho_k=0.9980~~~ \Delta_k=0.0500~~~ norm (p_k)=0.0500~~~ norm( g_k)=18.3940</math> |

## Revision as of 19:34, 27 May 2014

Authors: Wenhe (Wayne) Ye (ChE 345 Spring 2014) Steward: Dajun Yue, Fengqi You Date Presented: Apr. 10, 2014

## Contents |

## Introduction

Trust-region method (TRM) is one of the most important numerical optimization methods in solving nonlinear programming (NLP) problems. It works in a way that first define a region around the current best solution, in which a certain model (usually a quadratic model) can to some extent approximate the original objective function. TRM then take a step forward according to the model depicts within the region. Unlike the line search methods, TRM usually determines the step size before the improving direction (or at the same time). If a notable decrease (our following discussion will based on minimization problems) is gained after the step forward, then the model is believed to be a good representation of the original objective function. If the improvement is too subtle or even a negative improvement is gained, then the model is not to be believed as a good representation of the original objective function within that region. The convergence can be ensured that the size of the “trust region” (usually defined by the radius in Euclidean norm) in each iteration would depend on the improvement previously made.

## Important Concepts

**Trust-region**

In most cases, the trust-region is defined as a spherical area of radius in which the trust-region subproblem lies.

**Trust-region subproblem**

If we are using the quadratic model to approximate the original objective function, then our optimization problem is essentially reduced to solving a sequence of trust-region subporblems

Where is the trust region radius, is the gradient at current point and is the hessian (or a hessian approximation). It is easy to find the solution to the trust-region subproblem if is positive definite.

**Actual reduction and predicted reduction**

The most critical issue underlying the trust-region method is to update the size of the trust-region at every iteration. If the current iteration makes a satisfactory reduction, we may exploits our model more in the next iteration by setting a larger . If we only achieved a limited improvement after the current iteration, the radius of the trust-region then should not have any increase, or in the worst cases, we may decrease the size of the trust-region by adjusting the radius to a smaller value to check the model’s validity.

Whether to take a more ambitious step or a more conservative one is depend on the ratio between the actual reduction gained by true reduction in the original objective function and the predicted reduction expected in the model function. Empirical threshold values of the ratio will guide us in determining the size of the trust-region.

## Trust Region Algorithm

Before implementing the trust-region algorithm, we should first determine several parameters. is the upper bound for the size of the trust region. , and ,, are the threshold values for evaluating the goodness of the quadratic model thus for determining the trust-region’s size in the next iteration. A typical set for these values are , and ,,.

**Pseudo-code**

**Set** the starting point at , set the iteration number

**for**

Get the improving step by solving trust-region sub-problem ()

Evaluate from equation()

**if**

**else**

**if** and (full step and model is a good approximation)

**else**

**if**

**else**

(the model is not a good approximation and need to solve another trust-region subproblem within a smaller trust-region)

**end**
>

## Methods of Solving the Trust-region Subproblem

### Cauchy point calculation

In line search methods, we may find an improving direction from the gradient information, that is, by taking the steepest descent direction with regard to the maximum range we could make. We can solve the trust-region subproblem in an inexpensive way. This method is also denoted as the Cauchy point calculation. We can also express the improving step explicitly by the following closed-form equations

if

otherwise

### Limitations and Further Improvements

Though Cauchy point is cheap to implement, like the steepest descent method, it performs poorly in some cases. Varies kinds of improvements are based on including the curvature information from .

#### Dogleg Method

If is positive definite (we can use quasi-Newton Hessian approximation &updating to guarantee), then a V-shaped trajectory can be determined by

if

if

where is the steepest descent direction and is the optimal solution of the quadratic model . Therefore, a further improvement could be achieved compared to using only Cauchy point calculation method in one iteration. (Note that hessian or approximate hessian will be evaluated in dogleg method)

#### Conjugated Gradient Steihaug’s Method

The most widely used method for solving a trust-region sub-problem is by using the idea of conjugated gradient (CG) method for minimizing a quadratic function since CG guarantees convergence within a finite number of iterations for a quadratic programming. Also, CG Steihaug’s method has the merit of Cauchy point calculation and dogleg method that both in terms of super-linear convergence rate and inexpensiveness to compute.(No expensive Hessian evaluation)

**Pseudo-code for CG Steihaug method in solving trust region sub-problem**

*Given tolerance ;*

**Set**

**if**

**return** ;

**for**

**if**

Find such that minimizes

and satisfies ;

**return** ;

**Set** ;

**Set** ;

**if**

Find such that satisfies ;

**return** ;

**Set** ;

**if**

**return** ;

**Set** ;

**Set**

**end **

## Example

Here we use the trust-region method to solve an unconstrained problem as an example. The trust-region subproblems are solved by calculate the Cauchy point.

Starting point

**Improving Process**

## Conclusion

**Trust-Region vs. Line Search**

Line search methods:

- Pick improving direction

- Pick stepsize to minimize

- Update the incumbent solution

Trust-region methods:

- Pick the stepsize (the trust-region subproblem is constrained)

- Solving the subproblem using the approximated model

- If the improvement is acceptable, update the incumbent solution and the size of the trust-region

## References

1. J. Nocedal, S. J. Wright, Numerical optimization. Springer, 1999.

2. Wikipedia page for Trust-Region Methods

3. Yuan, Y. "A review of trust region algorithms for optimization" in ICIAM 99: Proceedings of the Fourth International Congress on Industrial & Applied Mathematics, Edinburgh, 2000 Oxford University Press, USA.