4 damped (modified) Newton methods

Size: px
Start display at page:

Download "4 damped (modified) Newton methods"

Transcription

1 4 damped (modified) Newton methods 4.1 damped Newton method Exercise 4.1 Determine with the damped Newton method the unique real zero x of the real valued function of one variable f(x) = x 3 +x 2 using the step size rule t k = 1+ x k 1 and the starting point x 0 = Show, that the sequence of iterations x k for some C 2 > C 1 > 0 and ε > 0 and all x k with x k 1 the estimation (Q-convergence of order 3/2, faster than linear, slower than quadratic) C 1 x k 1 3/2 x k+1 1 C 2 x k 1 3/2 satisfies. 2. How the step size is to choose in order to achieve a Q-convergence of order r +1/n? Exercise 4.2 At the strictly quadratic convex function of n variables f(x) = x T Qx+q T x a damped Newton method is applied with scaled Armijo rule step size and secant parameter α (0,1/2) (because of the uniform convexity of f the negative gradient direction need not to be used). The starting scaled test step size is given through t = max {1, f(xk ) T d k } (d k ) T d k where d k satisfies the Newton equation Qd k = f(x k ). 1. Show, that the above test step size the armijo condition in x k w.r.t. the direction d k satisfies, whenever 1 t < 2(1 α) is valid. 2. Show, that the above damped Newton method is only Q-linearly convergent in case of 1 < t < 2(1 α) with the convergence rate C = t Determine for the function no. 4 of two variables f(x) = 1 4 (x 1 5) 2 +(x 2 6) 2 all starting points for which the damped Newton method converges in one step, i.e. for which the above parameter t = 1. Exercise 4.3 Prove: Suppose, that the function f C 2 (D,R) with open set D and W := N f (f(x 0 )) D satisfies on this level set W the (m,m) condition m d 2 d T H f (x)d M d 2, then for each sequence of points x k W the assigned sequence of Newton directions d k := H f (x k ) f(x k ) is strictly gradient like (sgl). Experiment 4.1 Start the file modnewton01. It is a long time experiment for the damped Newton method (method no. 3) using the linesearches from 1.0 till 8.1. For 20 stochastic points it needs about 5 minutes. You find the result file at C:\temp\<user>\results\modNewton01.txt 20

2 1. What kind of linesearches is bad (see also exercise 4.2 for the reasons)? 2. What kind of linesearches should be preferred (see also the convergence theorem and the subsequent statement)? 3. When the damped Newton method is only linearly convergent? Experiment 4.2 Start modnewton0201.m { linsmode,1.0, tol,1e-8, x0,[7,11]}); It gives a answer to question 1 in experiment 4.1 for the strictly convex quadratic function no. 4. Which theoretical length t has the starting step size in this case? See item 1. in exercise 4.2. Is the numerical value for t k in the iterational table the same? We used α = 10 4 (default value). Repeat the experiment (second command) with a value of α = 0.2 by using a second coordinate for linsmode { linsmode,[1.0,0.2], tol,1e-8, x0,[7,11]}); The condition in item 1 exercise 4.2 is not satisfied. In spite of this the method needs more than one step, why? Experiment 4.3 Start modnewton0202.m linsmode=4.1 (Armijo backtracking with t=1 preference) uses as first test t = 1. However, if d k is too large, the linesearch algorithm reduce its length. Since by the theory d k 0 this can happen only finitely many. If firstly the length is not reduced, the minimum is achieved in this full Newton step. { linsmode,4.1, tol,1e-8, x0,[6,11]}); The same holds for linsmode=5.1 (Powell/Wolfe backtracking with t=1 preference) has the same reduction and therefore the same effect. { linsmode,5.1, tol,1e-8, x0,[6,11]}); Experiment 4.4 Start modnewton0203.m. { linsmode,6.0, tol,1e-8, x0,[6,11]}); Golden section (GS) can be used to approximate the perfect step size with a given relative error. We use as default relative error The first interval [0,λ] for the application of GS is found by the Armijo rule (gives λ) and some subsequent extension to a λ such that φ(0) > φ( λ) φ(λ) is valid (otherwise φ no local minimum on positive real axis) (see goldsection.m). In spite of the step length is (very few!!) different from 1 we need more 21

3 than one step. Second, since the step length does not converge to one we have only Q-linear convergence. We repeat the experiment with fminbnd (combination of GS and quadratic interpolation, linsmode=[6.0,1e-2,1], 3rd coordinate is set) instead of pure GS. First QI yields for quadratic function the perfect step length hence the damped Newton method needs only one step. { linsmode,[6.0,1e-2,1], tol,1e-8, x0,[6,11]}); If we choose the relative error smaller than the tolerance tol of the damped Newton method, then again the minimum is achieved as expected in one step. But it takes a lot of function value calculations. Attention do not use some 3rd coordinate of linsmode!! { linsmode,[6.0,1e-10], tol,1e-8, x0,[6,11]}); Experiment 4.5 Start modnewton0204.m. The backtracking linesearch (linsmode=4.1) can attain the lower backtracking safeguard 0.1 over a lot of steps, whenever t = 1 does not satisfy AR, the first qudratic interpolation is smaller than 0.1 but 0.1 is valid for AR. Hence the damped Newton method can be over a long time an algorithm with constant step length smaller than one. Here is an example with the 2-dimensional banana function. his=psolve(1,3,{ picinfo,12, holdon,1},{ linsmode,4.1, tol,1e-8}); The backtracking Powell/Wolfe linesearch (linsmode=5.1) has only seldom this effect because of the additional lower bound by the twisted tangent condition. Hence it is faster. his=psolve(1,3,{ picinfo,12, holdon,1},{ linsmode,5.1, tol,1e-8}); Experiment 4.6 Start modnewton0205.m. The best result (for this example) for the damped Newton method generates a linesearch with final interpolation and use of the step with smallest function value. This is done by linsmode = 3.0(scaled) or linsmode = 3.1(starting value 1). Similar but more complicated linesearches of this kind can be found in Spellucci Sometimes also LS 3.1 can cause because of interpolation longer steps at the starting phase of the damped Newton method, which yields acceleration. his=psolve(1,3,{ picinfo,12, holdon,1},{ linsmode,3.0, tol,1e-8}); his=psolve(1,3,{ picinfo,12, holdon,1},{ linsmode,3.1, tol,1e-8}); Experiment 4.7 Start modnewton0206.m. In case of non regular hessian at the final point the linesearch linsmode=3.0 has advantages and the damped Newton method can be faster than the (local) Newton method. Because of longer step length than 1 the damped Newton method avoids the eigenvector direction for the eigenvalue zero, which is given for the Murphy function no. 9. We compare both methods. his=psolve(9,[1,3],{ picinfo,12, holdon,1},{ linsmode,3.0, tol,1e-8}); 22

4 Experiment 4.8 The damped Newton method can run as steepest descent over a large number of steps, when ever a point with negative eigenvalue of the hessian is reached (please compute the eigenvalues of the hessian at this point). This may happen for all implemented linesearches, less often for linesearches with final interpolation. We study the banana function in higher dimensions. Start modnewton03.m. We use commands of the kind (only short info, no picture from 2.-9.) his=psolve(50,[1,3],{ picinfo,0, txtinfo,1},... { linsmode,linsmode, tol,1e-8, maxit,maxit},{dim}); with dim as below and maxit large enough. 1. Banana dim =20, linsmode = steepest descent steps from 250 Iterations 2. Banana dim =20, linsmode = steepest descent steps from 39 Iterations (by chance faster (?) than local Newton method, but study the costs and CPU-time!) 3. Banana dim =10, linsmode = steepest descent steps from 803 Iterations The (local) Newton method needs for the same starting points in case of dim = 20 and dim = 10 only 46 resp. 34 iterations The replacement of the negative gradient direction by a stochastic cone direction (nearly in the half space of descent directions by setting of the option tau0, 5 attempts) yields no significantly better results, here for linsmode = 1.1 and dim x = 10. We use the command his=psolve(50,[1,3],{ picinfo,0, txtinfo,1},... { linsmode,1.1, tol,1e-8, maxit,1000, tau0,[pi/2.1]},{10}); 4.2 damped modified Newton method Exercise 4.4 Prove: Let Q R n n be a symmetric indefinite matrix and D R n n be a positively definite diagonal matrix, then there is µ 0 > 0 such that the matrix Q + µ 0 D is positively semi definite (one eigenvalue is zero) and Q+µ 0 D SPD n for all µ > µ 0. Exercise 4.5 Prove: Suppose, that the function f C 2 (D,R) with open set D, then for each bounded sequence of points x k D with assigned sequence of regularized Hessians H k the arising sequence of modified Newton directions d k := H k f(x k ) is strictly gradient like (sgl). Experiment 4.9 start modnewton04, be aware that m=10 (cf. row 10). We shorten a longtime experiment (originally 250 stochastically chosen starting points) and compare iterations and function value computations for the damped modified Newton method applied to the banana function of dimensions 10,15,20. Its behavior is much more comfortable than the former considered damped version (compare with the number of iterations for the 2 (!!!) dimensional banana function in experiment 4.1). The calls are executed according to the command 23

5 hist=psolve(50,4,{ picinfo,0, txtinfo,0},... { linsmode,ls(i), x0,xstart(j,:), maxit,maxit, tol,tol},{dim}); where LS=[1.0,1.1,2.0,2.1,3.0,3.1,4.1,5.1,6.0,7.0,7.1,8.0,8.1], dim=[10,15,20], j=1:m and maxit = 1e+4 being far away of really used number of iterations. The result file can be studied in C:\temp\<user>\results\modNewton04a.txt Experiment 4.10 Start modnewton06.m (for Matlab under 32 bit processor) or modnewton06a.m (for Matlab under 64 bit processor) which bases on a mat-file generated by a former run of modnewton05 (determination of points in which the Newton method is divergent). We consider the 20th starting point with divergence of the Newton method and study the damped modified Newton method with the call load modnewton0501 x0div % load the x0div variable of a mat-file his=psolve(50,4,{ picinfo,2, txtinfo,2, holdon,1},... { linsmode,3.1, x0,x0div{3}(20,:), tol,tol, maxit,500},{20}); We consider it for tol = 1e-3,1e-5,1e-8. Till tol = 1e-6 the AR condition can be correctly interpreted. The Armijo condition can still be well interpreted and yields for tol = 1e-3 yet a smooth function also in floating point arithmetic. The method stops regular with gradient condition. For tol = 1e-8 the methods stops with to small step size, caused by Armijo rule in LS 3.1 since the Armijo condition cannot be correctly interpreted by the numerics. The following steps show, that step length 1 is possible and yields a final point with quite better gradient norm 1e-13 (see command window). Figure 2 illustrates the Armijo condition (left and right hand side) of a smooth function in floating point arithmetics under digit extinction. Conclusion (see lecture). Finally we run the local Newton method to show that this is not a substantial alternative. 4.3 nonmonotonic damped (modified) Newton method Experiment 4.11 Start modnewton07: We use the nonmonotonic Armijo linesearch for the damped (modified) Newton method (method no. 3 (4)). The main command for this LS 9 is LS = 9.0xxs, where m = xx {1 : 99} determines the maximal number of going back in the function value history and s 0 : 9 is the number of starting pure Armijo steps. We start for the 50-dim Banana function with m = 1 : 20 (i. e. 20 times) twice method 4 (modified damped Newton with nonmonoton AR and mixed linesearch) and ones method 3 (damped Newton with nonmonoton AR), with the call his=psolve(50,method,{ picinfo,0, txtinfo,0},... { linsmode,ls, tol,1e-8, maxit,500},{50}); For comparison we start method 3 and 4 with LS 3.1 his=psolve(50,method,{ picinfo,0, txtinfo,0},... { linsmode,3.1, tol,1e-8, maxit,500},{50}); and the local Newton method (1) his=psolve(50,1,{ picinfo,0, txtinfo,0},... { tol,1e-8, maxit,500},{50}); 24

6 1. damped modified Newton method (= 4), xx=01:20, LS = 9.0xx1 result in: c:\temp\<user>\results\modnewtond07_50_50_20_1nmd.txt 2. damped modified Newton method (=4), xx=01:20, LS = [9.0xx1,0,0,1] mixed with LS 3.1 during modifying result in: c:\temp\<user>\results\modnewtond07_50_50_20_1_lsmixednmd.txt 3. damped Newton method (=3), xx=01:20, LS = 9.0xx1 result in: c:\temp\<user>\results\modnewtond07_50_50_20_1nd.txt The best results (till 40% less iterations) are achieved for (1.) damped modified Newton method with LS The mixed LS in case of (2.) damped modified Newton (till 10% less iterations) is not so successful in this example, best for LS The damped Newton method with negative gradient as alternative (3.) is similar to (2.) but 30 steepest descent steps destroy the performance. Please repeat the experiment for the Wood function no 15., Banana 2 dim. no 1, no. 2, no. 14 For function no. 2 the nonmonotonic strategy is unexpectedly contraproductive. Experiment 4.12 Start modnewton09: We investigate with problem no. 50 in higher dimensions(dim=50)howmuchstepswithpurearmijosearchatthebeginningcanberecommended. The literature says 5. Is this true for our implementation? The main command is his=psolve(50,4,{ picinfo,0, txtinfo,0},... { linsmode,9+m/1000+j/10000, tol,1e-8, maxit,250},{dim}); We use the information directly from the output his to construct the table and set m =15 (maximal number updatings) and j =1:10 designs the number of pure Armijo steps at the beginning. Experiment 4.13 Start modnewton10: We demonstrate that the nonmonotonic strategy can yield to stagnation for damped regularized Newton method as already known from damped Newton method. In modnewton10a we investigate the reason for this behavior. The main call is his=psolve(50,method,{ picinfo,12, txtinfo,-12},... { linsmode,ls, tol,1e-8, maxit,250},{4}); with method = 3,4 and LS = , LS = 3.1. The IOparameter-option txtinfo,-12 causes that all iterations are printed only in the file C:/temp/<use>/results/NEWTONMD-LS P050-0.txt but not on the screen. Here in this example the mixed linesearch LS = [9.0101,0,0,1] (available only for method 4) overcome this deteriorated situation which is not ever the case. Try to find an example. The pure linesearch LS = 3.1 for method = 4 runs to another stationary point. We check its kind. Experiment 4.14 Start modnewton10a: The regularization takes place at an area where the gradient is very flat and the hessian is indefinite. Eigenvalue 1 is about -1.5 and the other three ones are about 40. The used regularization routine modelhess (Dennis/Schnabel) overestimates the regularization parameter to µ = Hence all eigenvalues of the modified hessian are larger than 40. Since the gradient coordinates with respect to the eigenvector basis are about of 10 2 the direction d calculated with the modified hessian is with very small (ever step length 25

7 λ = 1). Since we use the same regularization program the input data of this program changes very slow. Hence, the area with this bad properties cannot be leaved fast enough and we have such small steps over a lot of iterations. Remark: The local Newton methods (also step length λ = 1!!) makes in the direction of the first eigenvector 30 times longer steps, but with ascent! 4.4 Newton method, hessian with forward differences Experiment 4.15 Start modnewton08a: The modification of the hessian by use of 2nd forward differences is contra productive and destroys the superlinearity in examples with bad condition of the hessians. The stopping condition should not be too small in such a case as done in this example. Thereafter, study the main call with tol = 1e-8 and second vorward differences: his=psolve(50,1,{ picinfo,2, txtinfo,2, holdon,1, diffmode,3},... { tol,tol, maxit,50, x0,0.9*ones(1,20)},{20}); What happens? modnewton08b:what is the numerical reason, that the method cannot catch the stopping condition? We study the x k and the d k at iteration k=8,9,10,11,12 (in EdOptLab use his.x(k+1,:))? Attention, the directions d k have to be externally computed. We use the call his=psolve(50,1,{ picinfo,2, txtinfo,2, holdon,1, diffmode,diffmode},... { tol,1e-13, maxit,15, x0,0.9*ones(1,20)},{20}); for diffmode=3,2,1 and study the last 4 essential iterations. Then we compare the behavior with the other both diffmodes = 2, damped Gauss-Newton method Continuation of introduction: Because of (J(x k ) T r(x k )) T d k = (J(x k ) T r(x k )) T (J(x k ) T J(x k )) 1 J(x k ) T r(x k ) < 0 in case of rankj(x k ) = n and F(x k ) = J(x k ) T r(x k ) 0 outside of a stationary point, the Gauss-Newton direction d k := (J(x k ) T J(x k )) 1 J(x k ) T r(x k ) is ever a descent direction. The handicap of the Newton-method in case of nonconvex problem functions cannot occur. In the following experiments we repeat the former experiments with the damped Gauss-Newton method (method=2) and compare it for respective experiments with the damped (modified) Newton method (method = 3 ( 4) ). Use suitable linesearches according to the possibilities (e.g. linsmode =1.0, 1.1,2.0,2.1,3.0, 3.1, 4, 5, 7.0, 7.1, 8.0, 8.1). Experiments: Experiment 4.16 Start gsn1d.m: Consider problem , where the 3 points (t,y) = (1,2),(2,4),(3,y) should be approximated by the ansatz function h(x,t) = e xt. We have 26

8 y = 8, 1, 8 for problem 213, 214 resp Observe that the problem has only one dimension. Use the command (damped Gauss-Newton method) his = gaussnewton(213,2,{ picinfo,12, txtinfo,3},... { linsmode,3.1, tol,1e-8, maxit,200}). under EdOptLab and study the convergence properties for several line searches. Which line search is the best? For comparison calculate the solution also with a simple Newton method for the necessary optimality condition. Experiment 4.17 Start gsn1d.m: Repeat Experiment 4.16, except the elementary calculation, for problem where the points should be approximated by the ansatz function h(x, t) = x 1 e x 2 t.thisproblemistwodimensional. Problem216usesthepointset (t,y) {1,2,4,5,8} {3,4,6,11,20}. Problem 217 has the additional point (4.1,46) and problem 218 has w.r.t. 216 slightly disturbed y-values. Which line search is the best? Experiment 4.18 Start gsn3d.m: Compare the performance of the damped Gauss-Newton method his = gaussnewton(201,2,{ picinfo,2, txtinfo,3},... { linsmode,3.1, tol,1e-12}) for problem 201 and the damped (modified) Newton method his = PSOLVE(1,[3,4],{ picinfo,12, txtinfo,3},... { linsmode,3.1, tol,1e-12}) for problem 1 using the standard starting point. Discuss the results for several linesearches. Also here we have problems with the expensive time of AD. It seems, that for a lot of dimensions in the function vector the AD needs linearly more CPU-time. Therefore make comparisons with symbolic differentiation (diffmode = 4 for Newton, diffmode = 5 for Gauss-Newton). However here we have long initilization time for the constructing of the m-files. Experiment 4.19 Start gsn4d.m: Compare the performance of the damped Gauss-Newton method his = gaussnewton(250,2,{ picinfo,2, txtinfo,3},... { linsmode,3.1, tol,1e-12},{dim}) for problem 250 and the damped (modified) Newton method his = PSOLVE(50,[3,4],{ picinfo,12, txtinfo,3},... { linsmode,3.1, tol,1e-12},{dim}) for problem 50 using the standard starting point. Discuss the results for several dimension, say dim = 5, 10, 20, 50, 100 and for several linesearches. Here the time problem for AD is more serious with increasing dimension. 27

Unconstrained optimization

Unconstrained optimization Chapter 4 Unconstrained optimization An unconstrained optimization problem takes the form min x Rnf(x) (4.1) for a target functional (also called objective function) f : R n R. In this chapter and throughout

More information

Nonlinear Optimization: What s important?

Nonlinear Optimization: What s important? Nonlinear Optimization: What s important? Julian Hall 10th May 2012 Convexity: convex problems A local minimizer is a global minimizer A solution of f (x) = 0 (stationary point) is a minimizer A global

More information

ECE580 Exam 1 October 4, Please do not write on the back of the exam pages. Extra paper is available from the instructor.

ECE580 Exam 1 October 4, Please do not write on the back of the exam pages. Extra paper is available from the instructor. ECE580 Exam 1 October 4, 2012 1 Name: Solution Score: /100 You must show ALL of your work for full credit. This exam is closed-book. Calculators may NOT be used. Please leave fractions as fractions, etc.

More information

LECTURE 22: SWARM INTELLIGENCE 3 / CLASSICAL OPTIMIZATION

LECTURE 22: SWARM INTELLIGENCE 3 / CLASSICAL OPTIMIZATION 15-382 COLLECTIVE INTELLIGENCE - S19 LECTURE 22: SWARM INTELLIGENCE 3 / CLASSICAL OPTIMIZATION TEACHER: GIANNI A. DI CARO WHAT IF WE HAVE ONE SINGLE AGENT PSO leverages the presence of a swarm: the outcome

More information

ECE580 Partial Solution to Problem Set 3

ECE580 Partial Solution to Problem Set 3 ECE580 Fall 2015 Solution to Problem Set 3 October 23, 2015 1 ECE580 Partial Solution to Problem Set 3 These problems are from the textbook by Chong and Zak, 4th edition, which is the textbook for the

More information

Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem

Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem min{f (x) : x R n }. The iterative algorithms that we will consider are of the form x k+1 = x k + t k d k, k = 0, 1,...

More information

FALL 2018 MATH 4211/6211 Optimization Homework 4

FALL 2018 MATH 4211/6211 Optimization Homework 4 FALL 2018 MATH 4211/6211 Optimization Homework 4 This homework assignment is open to textbook, reference books, slides, and online resources, excluding any direct solution to the problem (such as solution

More information

Unconstrained minimization of smooth functions

Unconstrained minimization of smooth functions Unconstrained minimization of smooth functions We want to solve min x R N f(x), where f is convex. In this section, we will assume that f is differentiable (so its gradient exists at every point), and

More information

Lecture 5: September 12

Lecture 5: September 12 10-725/36-725: Convex Optimization Fall 2015 Lecture 5: September 12 Lecturer: Lecturer: Ryan Tibshirani Scribes: Scribes: Barun Patra and Tyler Vuong Note: LaTeX template courtesy of UC Berkeley EECS

More information

Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem

Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem Lecture 4 - The Gradient Method Objective: find an optimal solution of the problem min{f (x) : x R n }. The iterative algorithms that we will consider are of the form x k+1 = x k + t k d k, k = 0, 1,...

More information

1 Newton s Method. Suppose we want to solve: x R. At x = x, f (x) can be approximated by:

1 Newton s Method. Suppose we want to solve: x R. At x = x, f (x) can be approximated by: Newton s Method Suppose we want to solve: (P:) min f (x) At x = x, f (x) can be approximated by: n x R. f (x) h(x) := f ( x)+ f ( x) T (x x)+ (x x) t H ( x)(x x), 2 which is the quadratic Taylor expansion

More information

E5295/5B5749 Convex optimization with engineering applications. Lecture 8. Smooth convex unconstrained and equality-constrained minimization

E5295/5B5749 Convex optimization with engineering applications. Lecture 8. Smooth convex unconstrained and equality-constrained minimization E5295/5B5749 Convex optimization with engineering applications Lecture 8 Smooth convex unconstrained and equality-constrained minimization A. Forsgren, KTH 1 Lecture 8 Convex optimization 2006/2007 Unconstrained

More information

Numerical solutions of nonlinear systems of equations

Numerical solutions of nonlinear systems of equations Numerical solutions of nonlinear systems of equations Tsung-Ming Huang Department of Mathematics National Taiwan Normal University, Taiwan E-mail: min@math.ntnu.edu.tw August 28, 2011 Outline 1 Fixed points

More information

Unconstrained Optimization

Unconstrained Optimization 1 / 36 Unconstrained Optimization ME598/494 Lecture Max Yi Ren Department of Mechanical Engineering, Arizona State University February 2, 2015 2 / 36 3 / 36 4 / 36 5 / 36 1. preliminaries 1.1 local approximation

More information

x k+1 = x k + α k p k (13.1)

x k+1 = x k + α k p k (13.1) 13 Gradient Descent Methods Lab Objective: Iterative optimization methods choose a search direction and a step size at each iteration One simple choice for the search direction is the negative gradient,

More information

, b = 0. (2) 1 2 The eigenvectors of A corresponding to the eigenvalues λ 1 = 1, λ 2 = 3 are

, b = 0. (2) 1 2 The eigenvectors of A corresponding to the eigenvalues λ 1 = 1, λ 2 = 3 are Quadratic forms We consider the quadratic function f : R 2 R defined by f(x) = 2 xt Ax b T x with x = (x, x 2 ) T, () where A R 2 2 is symmetric and b R 2. We will see that, depending on the eigenvalues

More information

Methods for Unconstrained Optimization Numerical Optimization Lectures 1-2

Methods for Unconstrained Optimization Numerical Optimization Lectures 1-2 Methods for Unconstrained Optimization Numerical Optimization Lectures 1-2 Coralia Cartis, University of Oxford INFOMM CDT: Modelling, Analysis and Computation of Continuous Real-World Problems Methods

More information

The Steepest Descent Algorithm for Unconstrained Optimization

The Steepest Descent Algorithm for Unconstrained Optimization The Steepest Descent Algorithm for Unconstrained Optimization Robert M. Freund February, 2014 c 2014 Massachusetts Institute of Technology. All rights reserved. 1 1 Steepest Descent Algorithm The problem

More information

Motivation: We have already seen an example of a system of nonlinear equations when we studied Gaussian integration (p.8 of integration notes)

Motivation: We have already seen an example of a system of nonlinear equations when we studied Gaussian integration (p.8 of integration notes) AMSC/CMSC 460 Computational Methods, Fall 2007 UNIT 5: Nonlinear Equations Dianne P. O Leary c 2001, 2002, 2007 Solving Nonlinear Equations and Optimization Problems Read Chapter 8. Skip Section 8.1.1.

More information

Nonlinear Optimization for Optimal Control

Nonlinear Optimization for Optimal Control Nonlinear Optimization for Optimal Control Pieter Abbeel UC Berkeley EECS Many slides and figures adapted from Stephen Boyd [optional] Boyd and Vandenberghe, Convex Optimization, Chapters 9 11 [optional]

More information

Line Search Methods for Unconstrained Optimisation

Line Search Methods for Unconstrained Optimisation Line Search Methods for Unconstrained Optimisation Lecture 8, Numerical Linear Algebra and Optimisation Oxford University Computing Laboratory, MT 2007 Dr Raphael Hauser (hauser@comlab.ox.ac.uk) The Generic

More information

Numerical Optimization Prof. Shirish K. Shevade Department of Computer Science and Automation Indian Institute of Science, Bangalore

Numerical Optimization Prof. Shirish K. Shevade Department of Computer Science and Automation Indian Institute of Science, Bangalore Numerical Optimization Prof. Shirish K. Shevade Department of Computer Science and Automation Indian Institute of Science, Bangalore Lecture - 13 Steepest Descent Method Hello, welcome back to this series

More information

5 Quasi-Newton Methods

5 Quasi-Newton Methods Unconstrained Convex Optimization 26 5 Quasi-Newton Methods If the Hessian is unavailable... Notation: H = Hessian matrix. B is the approximation of H. C is the approximation of H 1. Problem: Solve min

More information

IPAM Summer School Optimization methods for machine learning. Jorge Nocedal

IPAM Summer School Optimization methods for machine learning. Jorge Nocedal IPAM Summer School 2012 Tutorial on Optimization methods for machine learning Jorge Nocedal Northwestern University Overview 1. We discuss some characteristics of optimization problems arising in deep

More information

Numerical Optimization

Numerical Optimization Numerical Optimization Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Spring 2010 Emo Todorov (UW) AMATH/CSE 579, Spring 2010 Lecture 9 1 / 8 Gradient descent

More information

Lecture Notes to Accompany. Scientific Computing An Introductory Survey. by Michael T. Heath. Chapter 5. Nonlinear Equations

Lecture Notes to Accompany. Scientific Computing An Introductory Survey. by Michael T. Heath. Chapter 5. Nonlinear Equations Lecture Notes to Accompany Scientific Computing An Introductory Survey Second Edition by Michael T Heath Chapter 5 Nonlinear Equations Copyright c 2001 Reproduction permitted only for noncommercial, educational

More information

CPSC 540: Machine Learning

CPSC 540: Machine Learning CPSC 540: Machine Learning First-Order Methods, L1-Regularization, Coordinate Descent Winter 2016 Some images from this lecture are taken from Google Image Search. Admin Room: We ll count final numbers

More information

Numerical Methods I Solving Nonlinear Equations

Numerical Methods I Solving Nonlinear Equations Numerical Methods I Solving Nonlinear Equations Aleksandar Donev Courant Institute, NYU 1 donev@courant.nyu.edu 1 MATH-GA 2011.003 / CSCI-GA 2945.003, Fall 2014 October 16th, 2014 A. Donev (Courant Institute)

More information

Methods that avoid calculating the Hessian. Nonlinear Optimization; Steepest Descent, Quasi-Newton. Steepest Descent

Methods that avoid calculating the Hessian. Nonlinear Optimization; Steepest Descent, Quasi-Newton. Steepest Descent Nonlinear Optimization Steepest Descent and Niclas Börlin Department of Computing Science Umeå University niclas.borlin@cs.umu.se A disadvantage with the Newton method is that the Hessian has to be derived

More information

CS 542G: Robustifying Newton, Constraints, Nonlinear Least Squares

CS 542G: Robustifying Newton, Constraints, Nonlinear Least Squares CS 542G: Robustifying Newton, Constraints, Nonlinear Least Squares Robert Bridson October 29, 2008 1 Hessian Problems in Newton Last time we fixed one of plain Newton s problems by introducing line search

More information

17 Solution of Nonlinear Systems

17 Solution of Nonlinear Systems 17 Solution of Nonlinear Systems We now discuss the solution of systems of nonlinear equations. An important ingredient will be the multivariate Taylor theorem. Theorem 17.1 Let D = {x 1, x 2,..., x m

More information

8 Numerical methods for unconstrained problems

8 Numerical methods for unconstrained problems 8 Numerical methods for unconstrained problems Optimization is one of the important fields in numerical computation, beside solving differential equations and linear systems. We can see that these fields

More information

min f(x). (2.1) Objectives consisting of a smooth convex term plus a nonconvex regularization term;

min f(x). (2.1) Objectives consisting of a smooth convex term plus a nonconvex regularization term; Chapter 2 Gradient Methods The gradient method forms the foundation of all of the schemes studied in this book. We will provide several complementary perspectives on this algorithm that highlight the many

More information

TMA4180 Solutions to recommended exercises in Chapter 3 of N&W

TMA4180 Solutions to recommended exercises in Chapter 3 of N&W TMA480 Solutions to recommended exercises in Chapter 3 of N&W Exercise 3. The steepest descent and Newtons method with the bactracing algorithm is implemented in rosenbroc_newton.m. With initial point

More information

Examination paper for TMA4180 Optimization I

Examination paper for TMA4180 Optimization I Department of Mathematical Sciences Examination paper for TMA4180 Optimization I Academic contact during examination: Phone: Examination date: 26th May 2016 Examination time (from to): 09:00 13:00 Permitted

More information

Shiqian Ma, MAT-258A: Numerical Optimization 1. Chapter 3. Gradient Method

Shiqian Ma, MAT-258A: Numerical Optimization 1. Chapter 3. Gradient Method Shiqian Ma, MAT-258A: Numerical Optimization 1 Chapter 3 Gradient Method Shiqian Ma, MAT-258A: Numerical Optimization 2 3.1. Gradient method Classical gradient method: to minimize a differentiable convex

More information

Math 411 Preliminaries

Math 411 Preliminaries Math 411 Preliminaries Provide a list of preliminary vocabulary and concepts Preliminary Basic Netwon s method, Taylor series expansion (for single and multiple variables), Eigenvalue, Eigenvector, Vector

More information

Programming, numerics and optimization

Programming, numerics and optimization Programming, numerics and optimization Lecture C-3: Unconstrained optimization II Łukasz Jankowski ljank@ippt.pan.pl Institute of Fundamental Technological Research Room 4.32, Phone +22.8261281 ext. 428

More information

Lecture 7: Minimization or maximization of functions (Recipes Chapter 10)

Lecture 7: Minimization or maximization of functions (Recipes Chapter 10) Lecture 7: Minimization or maximization of functions (Recipes Chapter 10) Actively studied subject for several reasons: Commonly encountered problem: e.g. Hamilton s and Lagrange s principles, economics

More information

Optimization. Totally not complete this is...don't use it yet...

Optimization. Totally not complete this is...don't use it yet... Optimization Totally not complete this is...don't use it yet... Bisection? Doing a root method is akin to doing a optimization method, but bi-section would not be an effective method - can detect sign

More information

Math 164: Optimization Barzilai-Borwein Method

Math 164: Optimization Barzilai-Borwein Method Math 164: Optimization Barzilai-Borwein Method Instructor: Wotao Yin Department of Mathematics, UCLA Spring 2015 online discussions on piazza.com Main features of the Barzilai-Borwein (BB) method The BB

More information

Math 273a: Optimization Netwon s methods

Math 273a: Optimization Netwon s methods Math 273a: Optimization Netwon s methods Instructor: Wotao Yin Department of Mathematics, UCLA Fall 2015 some material taken from Chong-Zak, 4th Ed. Main features of Newton s method Uses both first derivatives

More information

Conditional Gradient (Frank-Wolfe) Method

Conditional Gradient (Frank-Wolfe) Method Conditional Gradient (Frank-Wolfe) Method Lecturer: Aarti Singh Co-instructor: Pradeep Ravikumar Convex Optimization 10-725/36-725 1 Outline Today: Conditional gradient method Convergence analysis Properties

More information

Outline. Math Numerical Analysis. Intermediate Value Theorem. Lecture Notes Zeros and Roots. Joseph M. Mahaffy,

Outline. Math Numerical Analysis. Intermediate Value Theorem. Lecture Notes Zeros and Roots. Joseph M. Mahaffy, Outline Math 541 - Numerical Analysis Lecture Notes Zeros and Roots Joseph M. Mahaffy, jmahaffy@mail.sdsu.edu Department of Mathematics and Statistics Dynamical Systems Group Computational Sciences Research

More information

Math Numerical Analysis

Math Numerical Analysis Math 541 - Numerical Analysis Lecture Notes Zeros and Roots Joseph M. Mahaffy, jmahaffy@mail.sdsu.edu Department of Mathematics and Statistics Dynamical Systems Group Computational Sciences Research Center

More information

Unconstrained optimization I Gradient-type methods

Unconstrained optimization I Gradient-type methods Unconstrained optimization I Gradient-type methods Antonio Frangioni Department of Computer Science University of Pisa www.di.unipi.it/~frangio frangio@di.unipi.it Computational Mathematics for Learning

More information

Introduction to unconstrained optimization - direct search methods

Introduction to unconstrained optimization - direct search methods Introduction to unconstrained optimization - direct search methods Jussi Hakanen Post-doctoral researcher jussi.hakanen@jyu.fi Structure of optimization methods Typically Constraint handling converts the

More information

Numerical optimization

Numerical optimization THE UNIVERSITY OF WESTERN ONTARIO LONDON ONTARIO Paul Klein Office: SSC 408 Phone: 661-111 ext. 857 Email: paul.klein@uwo.ca URL: www.ssc.uwo.ca/economics/faculty/klein/ Numerical optimization In these

More information

5 Overview of algorithms for unconstrained optimization

5 Overview of algorithms for unconstrained optimization IOE 59: NLP, Winter 22 c Marina A. Epelman 9 5 Overview of algorithms for unconstrained optimization 5. General optimization algorithm Recall: we are attempting to solve the problem (P) min f(x) s.t. x

More information

Roots of equations, minimization, numerical integration

Roots of equations, minimization, numerical integration Roots of equations, minimization, numerical integration Alexander Khanov PHYS6260: Experimental Methods is HEP Oklahoma State University November 1, 2017 Roots of equations Find the roots solve equation

More information

Trajectory-based optimization

Trajectory-based optimization Trajectory-based optimization Emo Todorov Applied Mathematics and Computer Science & Engineering University of Washington Winter 2012 Emo Todorov (UW) AMATH/CSE 579, Winter 2012 Lecture 6 1 / 13 Using

More information

10. Unconstrained minimization

10. Unconstrained minimization Convex Optimization Boyd & Vandenberghe 10. Unconstrained minimization terminology and assumptions gradient descent method steepest descent method Newton s method self-concordant functions implementation

More information

Math 409/509 (Spring 2011)

Math 409/509 (Spring 2011) Math 409/509 (Spring 2011) Instructor: Emre Mengi Study Guide for Homework 2 This homework concerns the root-finding problem and line-search algorithms for unconstrained optimization. Please don t hesitate

More information

CS 450 Numerical Analysis. Chapter 5: Nonlinear Equations

CS 450 Numerical Analysis. Chapter 5: Nonlinear Equations Lecture slides based on the textbook Scientific Computing: An Introductory Survey by Michael T. Heath, copyright c 2018 by the Society for Industrial and Applied Mathematics. http://www.siam.org/books/cl80

More information

Second Order Optimization Algorithms I

Second Order Optimization Algorithms I Second Order Optimization Algorithms I Yinyu Ye Department of Management Science and Engineering Stanford University Stanford, CA 94305, U.S.A. http://www.stanford.edu/ yyye Chapters 7, 8, 9 and 10 1 The

More information

Comparison of Modern Stochastic Optimization Algorithms

Comparison of Modern Stochastic Optimization Algorithms Comparison of Modern Stochastic Optimization Algorithms George Papamakarios December 214 Abstract Gradient-based optimization methods are popular in machine learning applications. In large-scale problems,

More information

4 Newton Method. Unconstrained Convex Optimization 21. H(x)p = f(x). Newton direction. Why? Recall second-order staylor series expansion:

4 Newton Method. Unconstrained Convex Optimization 21. H(x)p = f(x). Newton direction. Why? Recall second-order staylor series expansion: Unconstrained Convex Optimization 21 4 Newton Method H(x)p = f(x). Newton direction. Why? Recall second-order staylor series expansion: f(x + p) f(x)+p T f(x)+ 1 2 pt H(x)p ˆf(p) In general, ˆf(p) won

More information

Lecture 3: Linesearch methods (continued). Steepest descent methods

Lecture 3: Linesearch methods (continued). Steepest descent methods Lecture 3: Linesearch methods (continued). Steepest descent methods Coralia Cartis, Mathematical Institute, University of Oxford C6.2/B2: Continuous Optimization Lecture 3: Linesearch methods (continued).

More information

EAD 115. Numerical Solution of Engineering and Scientific Problems. David M. Rocke Department of Applied Science

EAD 115. Numerical Solution of Engineering and Scientific Problems. David M. Rocke Department of Applied Science EAD 115 Numerical Solution of Engineering and Scientific Problems David M. Rocke Department of Applied Science Multidimensional Unconstrained Optimization Suppose we have a function f() of more than one

More information

Numerical Optimization of Partial Differential Equations

Numerical Optimization of Partial Differential Equations Numerical Optimization of Partial Differential Equations Part I: basic optimization concepts in R n Bartosz Protas Department of Mathematics & Statistics McMaster University, Hamilton, Ontario, Canada

More information

Line Search Algorithms

Line Search Algorithms Lab 1 Line Search Algorithms Investigate various Line-Search algorithms for numerical opti- Lab Objective: mization. Overview of Line Search Algorithms Imagine you are out hiking on a mountain, and you

More information

Outline. Scientific Computing: An Introductory Survey. Nonlinear Equations. Nonlinear Equations. Examples: Nonlinear Equations

Outline. Scientific Computing: An Introductory Survey. Nonlinear Equations. Nonlinear Equations. Examples: Nonlinear Equations Methods for Systems of Methods for Systems of Outline Scientific Computing: An Introductory Survey Chapter 5 1 Prof. Michael T. Heath Department of Computer Science University of Illinois at Urbana-Champaign

More information

Optimization 2. CS5240 Theoretical Foundations in Multimedia. Leow Wee Kheng

Optimization 2. CS5240 Theoretical Foundations in Multimedia. Leow Wee Kheng Optimization 2 CS5240 Theoretical Foundations in Multimedia Leow Wee Kheng Department of Computer Science School of Computing National University of Singapore Leow Wee Kheng (NUS) Optimization 2 1 / 38

More information

EAD 115. Numerical Solution of Engineering and Scientific Problems. David M. Rocke Department of Applied Science

EAD 115. Numerical Solution of Engineering and Scientific Problems. David M. Rocke Department of Applied Science EAD 115 Numerical Solution of Engineering and Scientific Problems David M. Rocke Department of Applied Science Taylor s Theorem Can often approximate a function by a polynomial The error in the approximation

More information

Notes for CS542G (Iterative Solvers for Linear Systems)

Notes for CS542G (Iterative Solvers for Linear Systems) Notes for CS542G (Iterative Solvers for Linear Systems) Robert Bridson November 20, 2007 1 The Basics We re now looking at efficient ways to solve the linear system of equations Ax = b where in this course,

More information

Lecture 8. Root finding II

Lecture 8. Root finding II 1 Introduction Lecture 8 Root finding II In the previous lecture we considered the bisection root-bracketing algorithm. It requires only that the function be continuous and that we have a root bracketed

More information

Unconstrained minimization

Unconstrained minimization CSCI5254: Convex Optimization & Its Applications Unconstrained minimization terminology and assumptions gradient descent method steepest descent method Newton s method self-concordant functions 1 Unconstrained

More information

Optimization with Scipy (2)

Optimization with Scipy (2) Optimization with Scipy (2) Unconstrained Optimization Cont d & 1D optimization Harry Lee February 5, 2018 CEE 696 Table of contents 1. Unconstrained Optimization 2. 1D Optimization 3. Multi-dimensional

More information

Some definitions. Math 1080: Numerical Linear Algebra Chapter 5, Solving Ax = b by Optimization. A-inner product. Important facts

Some definitions. Math 1080: Numerical Linear Algebra Chapter 5, Solving Ax = b by Optimization. A-inner product. Important facts Some definitions Math 1080: Numerical Linear Algebra Chapter 5, Solving Ax = b by Optimization M. M. Sussman sussmanm@math.pitt.edu Office Hours: MW 1:45PM-2:45PM, Thack 622 A matrix A is SPD (Symmetric

More information

Proximal Newton Method. Ryan Tibshirani Convex Optimization /36-725

Proximal Newton Method. Ryan Tibshirani Convex Optimization /36-725 Proximal Newton Method Ryan Tibshirani Convex Optimization 10-725/36-725 1 Last time: primal-dual interior-point method Given the problem min x subject to f(x) h i (x) 0, i = 1,... m Ax = b where f, h

More information

Conjugate Gradient (CG) Method

Conjugate Gradient (CG) Method Conjugate Gradient (CG) Method by K. Ozawa 1 Introduction In the series of this lecture, I will introduce the conjugate gradient method, which solves efficiently large scale sparse linear simultaneous

More information

7.2 Steepest Descent and Preconditioning

7.2 Steepest Descent and Preconditioning 7.2 Steepest Descent and Preconditioning Descent methods are a broad class of iterative methods for finding solutions of the linear system Ax = b for symmetric positive definite matrix A R n n. Consider

More information

Gradient Descent. Dr. Xiaowei Huang

Gradient Descent. Dr. Xiaowei Huang Gradient Descent Dr. Xiaowei Huang https://cgi.csc.liv.ac.uk/~xiaowei/ Up to now, Three machine learning algorithms: decision tree learning k-nn linear regression only optimization objectives are discussed,

More information

Convex Optimization. Problem set 2. Due Monday April 26th

Convex Optimization. Problem set 2. Due Monday April 26th Convex Optimization Problem set 2 Due Monday April 26th 1 Gradient Decent without Line-search In this problem we will consider gradient descent with predetermined step sizes. That is, instead of determining

More information

Convex Optimization. Newton s method. ENSAE: Optimisation 1/44

Convex Optimization. Newton s method. ENSAE: Optimisation 1/44 Convex Optimization Newton s method ENSAE: Optimisation 1/44 Unconstrained minimization minimize f(x) f convex, twice continuously differentiable (hence dom f open) we assume optimal value p = inf x f(x)

More information

Numerical optimization

Numerical optimization Numerical optimization Lecture 4 Alexander & Michael Bronstein tosca.cs.technion.ac.il/book Numerical geometry of non-rigid shapes Stanford University, Winter 2009 2 Longest Slowest Shortest Minimal Maximal

More information

Proximal Newton Method. Zico Kolter (notes by Ryan Tibshirani) Convex Optimization

Proximal Newton Method. Zico Kolter (notes by Ryan Tibshirani) Convex Optimization Proximal Newton Method Zico Kolter (notes by Ryan Tibshirani) Convex Optimization 10-725 Consider the problem Last time: quasi-newton methods min x f(x) with f convex, twice differentiable, dom(f) = R

More information

Written Examination

Written Examination Division of Scientific Computing Department of Information Technology Uppsala University Optimization Written Examination 202-2-20 Time: 4:00-9:00 Allowed Tools: Pocket Calculator, one A4 paper with notes

More information

IP-PCG An interior point algorithm for nonlinear constrained optimization

IP-PCG An interior point algorithm for nonlinear constrained optimization IP-PCG An interior point algorithm for nonlinear constrained optimization Silvia Bonettini (bntslv@unife.it), Valeria Ruggiero (rgv@unife.it) Dipartimento di Matematica, Università di Ferrara December

More information

How do we recognize a solution?

How do we recognize a solution? AMSC 607 / CMSC 764 Advanced Numerical Optimization Fall 2010 UNIT 2: Unconstrained Optimization, Part 1 Dianne P. O Leary c 2008,2010 The plan: Unconstrained Optimization: Fundamentals How do we recognize

More information

Maria Cameron. f(x) = 1 n

Maria Cameron. f(x) = 1 n Maria Cameron 1. Local algorithms for solving nonlinear equations Here we discuss local methods for nonlinear equations r(x) =. These methods are Newton, inexact Newton and quasi-newton. We will show that

More information

Computational Finance

Computational Finance Department of Mathematics at University of California, San Diego Computational Finance Optimization Techniques [Lecture 2] Michael Holst January 9, 2017 Contents 1 Optimization Techniques 3 1.1 Examples

More information

Numerical Optimization

Numerical Optimization Numerical Optimization Unit 2: Multivariable optimization problems Che-Rung Lee Scribe: February 28, 2011 (UNIT 2) Numerical Optimization February 28, 2011 1 / 17 Partial derivative of a two variable function

More information

ECS550NFB Introduction to Numerical Methods using Matlab Day 2

ECS550NFB Introduction to Numerical Methods using Matlab Day 2 ECS550NFB Introduction to Numerical Methods using Matlab Day 2 Lukas Laffers lukas.laffers@umb.sk Department of Mathematics, University of Matej Bel June 9, 2015 Today Root-finding: find x that solves

More information

Optimization and Calculus

Optimization and Calculus Optimization and Calculus To begin, there is a close relationship between finding the roots to a function and optimizing a function. In the former case, we solve for x. In the latter, we solve: g(x) =

More information

Part 3: Trust-region methods for unconstrained optimization. Nick Gould (RAL)

Part 3: Trust-region methods for unconstrained optimization. Nick Gould (RAL) Part 3: Trust-region methods for unconstrained optimization Nick Gould (RAL) minimize x IR n f(x) MSc course on nonlinear optimization UNCONSTRAINED MINIMIZATION minimize x IR n f(x) where the objective

More information

Iterative Methods for Smooth Objective Functions

Iterative Methods for Smooth Objective Functions Optimization Iterative Methods for Smooth Objective Functions Quadratic Objective Functions Stationary Iterative Methods (first/second order) Steepest Descent Method Landweber/Projected Landweber Methods

More information

Numerical Optimization

Numerical Optimization Unconstrained Optimization Computer Science and Automation Indian Institute of Science Bangalore 560 01, India. NPTEL Course on Unconstrained Minimization Let f : R n R. Consider the optimization problem:

More information

14. Nonlinear equations

14. Nonlinear equations L. Vandenberghe ECE133A (Winter 2018) 14. Nonlinear equations Newton method for nonlinear equations damped Newton method for unconstrained minimization Newton method for nonlinear least squares 14-1 Set

More information

Deep Learning. Authors: I. Goodfellow, Y. Bengio, A. Courville. Chapter 4: Numerical Computation. Lecture slides edited by C. Yim. C.

Deep Learning. Authors: I. Goodfellow, Y. Bengio, A. Courville. Chapter 4: Numerical Computation. Lecture slides edited by C. Yim. C. Chapter 4: Numerical Computation Deep Learning Authors: I. Goodfellow, Y. Bengio, A. Courville Lecture slides edited by 1 Chapter 4: Numerical Computation 4.1 Overflow and Underflow 4.2 Poor Conditioning

More information

Optimality Conditions

Optimality Conditions Chapter 2 Optimality Conditions 2.1 Global and Local Minima for Unconstrained Problems When a minimization problem does not have any constraints, the problem is to find the minimum of the objective function.

More information

Scientific Computing: An Introductory Survey

Scientific Computing: An Introductory Survey Scientific Computing: An Introductory Survey Chapter 5 Nonlinear Equations Prof. Michael T. Heath Department of Computer Science University of Illinois at Urbana-Champaign Copyright c 2002. Reproduction

More information

STA141C: Big Data & High Performance Statistical Computing

STA141C: Big Data & High Performance Statistical Computing STA141C: Big Data & High Performance Statistical Computing Lecture 8: Optimization Cho-Jui Hsieh UC Davis May 9, 2017 Optimization Numerical Optimization Numerical Optimization: min X f (X ) Can be applied

More information

Numerisches Rechnen. (für Informatiker) M. Grepl P. Esser & G. Welper & L. Zhang. Institut für Geometrie und Praktische Mathematik RWTH Aachen

Numerisches Rechnen. (für Informatiker) M. Grepl P. Esser & G. Welper & L. Zhang. Institut für Geometrie und Praktische Mathematik RWTH Aachen Numerisches Rechnen (für Informatiker) M. Grepl P. Esser & G. Welper & L. Zhang Institut für Geometrie und Praktische Mathematik RWTH Aachen Wintersemester 2011/12 IGPM, RWTH Aachen Numerisches Rechnen

More information

1. Method 1: bisection. The bisection methods starts from two points a 0 and b 0 such that

1. Method 1: bisection. The bisection methods starts from two points a 0 and b 0 such that Chapter 4 Nonlinear equations 4.1 Root finding Consider the problem of solving any nonlinear relation g(x) = h(x) in the real variable x. We rephrase this problem as one of finding the zero (root) of a

More information

Lecture 44. Better and successive approximations x2, x3,, xn to the root are obtained from

Lecture 44. Better and successive approximations x2, x3,, xn to the root are obtained from Lecture 44 Solution of Non-Linear Equations Regula-Falsi Method Method of iteration Newton - Raphson Method Muller s Method Graeffe s Root Squaring Method Newton -Raphson Method An approximation to the

More information

SOLUTIONS to Exercises from Optimization

SOLUTIONS to Exercises from Optimization SOLUTIONS to Exercises from Optimization. Use the bisection method to find the root correct to 6 decimal places: 3x 3 + x 2 = x + 5 SOLUTION: For the root finding algorithm, we need to rewrite the equation

More information

Coordinate Descent and Ascent Methods

Coordinate Descent and Ascent Methods Coordinate Descent and Ascent Methods Julie Nutini Machine Learning Reading Group November 3 rd, 2015 1 / 22 Projected-Gradient Methods Motivation Rewrite non-smooth problem as smooth constrained problem:

More information

MATH 3795 Lecture 13. Numerical Solution of Nonlinear Equations in R N.

MATH 3795 Lecture 13. Numerical Solution of Nonlinear Equations in R N. MATH 3795 Lecture 13. Numerical Solution of Nonlinear Equations in R N. Dmitriy Leykekhman Fall 2008 Goals Learn about different methods for the solution of F (x) = 0, their advantages and disadvantages.

More information

Outline. Additional Nonlinear Systems. Abstract. Finding Equilibrium Points Numerically. Newton s Method

Outline. Additional Nonlinear Systems. Abstract. Finding Equilibrium Points Numerically. Newton s Method Outline Finding Equilibrium Points Numerically Additional Nonlinear Systems James K. Peterson Department of Biological Sciences and Department of Mathematical Sciences Clemson University June 13, 2017

More information