M.J. TODD. A study of search directions in primal-dual interior-point methods for semidefinite programming. Optim. Methods Softw., 11&12:1--46, 1999.  Home/Search   Document Details and Download   Summary   Related Articles   Check

.... Gamma1 1 L Gamma1 2 Delta Delta Delta L Gamma1 Gamma1 is a lower triangular matrix with possible nonzero elements specified in F . 3 Primal dual interior point method In this section, we describe a generic framework for primal dual interior point methods applied to SDPs (1) and (2) [10, 13, 16, 18, 21, 24]. Various search directions have been proposed so far for primal dual interior point methods [21] Among others, we restrict ourselves to the HRVW KSH M search direction [10, 13, 16] in this article. In addition, we only consider a simple primal dual path following interior point method which is ....

....elements specified in F . 3 Primal dual interior point method In this section, we describe a generic framework for primal dual interior point methods applied to SDPs (1) and (2) 10, 13, 16, 18, 21, 24] Various search directions have been proposed so far for primal dual interior point methods [21]. Among others, we restrict ourselves to the HRVW KSH M search direction [10, 13, 16] in this article. In addition, we only consider a simple primal dual path following interior point method which is not a Mehrotra type since we implement this simple method in the completion method described in ....

M. J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Optimization Methods and Software 11 & 12 (1999) 1--46.

....X 1 = WD 1 W T (4) where W = L 1 1 L 1 2 L 1 # 1 is a lower triangular matrix with possible nonzero elements specified in F . 3 Primal dual interior point method In this section, we describe a generic framework for primal dual interior point methods applied to SDPs (1) and (2) [10, 13, 16, 18, 21, 24]. Various search directions have been proposed so far for primal dual interior point methods [21] We restrict ourselves to the HRVW KSH M search direction [10, 13, 16] in this article. In addition, we only consider a simple primal dual path following interior point method which is not a Mehrotra ....

....elements specified in F . 3 Primal dual interior point method In this section, we describe a generic framework for primal dual interior point methods applied to SDPs (1) and (2) 10, 13, 16, 18, 21, 24] Various search directions have been proposed so far for primal dual interior point methods [21]. We restrict ourselves to the HRVW KSH M search direction [10, 13, 16] in this article. In addition, we only consider a simple primal dual path following interior point method which is not a Mehrotra type since we implement this simple method in the completion method described in section 5. On ....

M. J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Optimization Methods and Software 11 & 12 (1999) 1--46.

....programming. 1.1.8 Synopsis of Chapter 5 Currently the optimization community disagrees on how to generalize system (1.6) to the setting of semidefinite programming. For a review of twenty search directions for semidefinite programming, including the most common ones, the reader may consult [41]. Our generalization of target maps and their corresponding Newton systems (1.14) define an infinite new family of search directions deriving from 24 generalizations of system (1.6) we call them target directions. These search directions lend themselves to the construction of ....

....as to retrieve a primal dual symmetric square root field for F as a result. For x; s 2 S n Thetan (R) let us define v(x; s) w(x; s) Gamma1=2 xw(x; s) Gamma1=2 . This concept is originally due to Kojima Megiddo Noma Yoshise [24] for LP and to Sturm Zhang [39] for SDP; see also Todd [41]. For a generalization to general instances of selfscaled conic programming, see equation (3.10) The map v : S n Thetan (R) 2 V is primal dual symmetric. In fact, since the dual scaling point satisfies the equation t(x; s) w(x; s) Gamma1 we have v(x; s) w Gamma1=2 xw ....

M. J. Todd. A study of search directions in primal-dual interior-point methods for semidefinite programming. Optimization Methods and Software, 11:1--46, 1999.

....Unlike the LO case, however, there are many possibilities to obtain primal dual search directions. Di erent directions arise when the perturbed optimality conditions are linearized and subsequently symmetrized; a quite comprehensive survey of the search directions obtained this way can be found in [17]. The need for symmetrization arises from the fact that the system of linearized perturbed optimality conditions is overdetermined. A recent idea by Kruk et al. 6] was to avoid symmetrization by solving a least squares problem by the Gauss Newton method. Motivated by [6] we proposed in [10] a ....

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidenite programming, Technical Report 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853, October 1997.

....is not necessarily symmetric. Before Newton s method can be applied to (1.29b) the fundamental operation in primal dual algorithms the domain and range have to match. The different primal dual algorithms differ in the ways that they reconcile the domain and range of these equations. The paper [Todd, 1999] is witness to the intensity of research in SDP interior point methods: It describes twenty techniques for obtaining search directions for SDP. In many of these, the equation (1.29c) is replaced by one whose range lies in SIR n Thetan . That is, it is symmetrized and replaced with an mapping ....

Todd, M. J. (1999). A study of search directions in primaldual interior-point methods for semidefinite programming. Technical report, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY.

.... y i A i = S (40) X S XS = E XS: A crucial observation for SDO is that the above Newton system might have no symmetric solution X. Many researchers have proposed di erent ways of symmetrizing the third equation in the Newton system so that the new system have a unique symmetric solution [20, 21]. In this paper we consider the symmetrization scheme that yields the NT direction [15, 21] Let us de ne the matrix P = X 1 2 (X 1 2 SX 1 2 ) 1 2 X 1 2 = S 1 2 (S 1 2 XS 1 2 ) 1 2 S 1 2 ; 41) and D = P 1 2 . The matrix D can be used to rescale X and S to the same matrix V ....

....we consider the symmetrization scheme that yields the NT direction [15, 21] Let us de ne the matrix P = X 1 2 (X 1 2 SX 1 2 ) 1 2 X 1 2 = S 1 2 (S 1 2 XS 1 2 ) 1 2 S 1 2 ; 41) and D = P 1 2 . The matrix D can be used to rescale X and S to the same matrix V de ned by ( [3, 15, 20, 19]) V : 1 p D 1 XD 1 = 1 p DSD: 42) Obviously the matrices D and V are symmetric, and positive de nite. Also de ning A i : DA i D; i = 1; m; DX : 1 p D 1 XD 1 ; D S : 1 p D SD: 43) Then the NT search direction can be written as the solution of the ....

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Technical Report 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853, October 1997.

.... y i A i = S X S XS = E XS (44) A crucial observation for SDO is that the above Newton system might not have a symmetric solution X. Many researchers have proposed di erent ways of symmetrizing the third equation in the Newton system so that the new system have a unique symmetric solution [26, 27]. In this paper we consider the symmetrization scheme from which the NT direction [18, 27] is derived. Before describing the NT symmetrizing scheme, let us rst introduce the de nition of a matrix function. 19 De nition 4.1 [2] Suppose the matrix X is diagonalizable with X = Q 1 X diag ( 1 ....

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Optimization Methods and Software, 11(1999), pp. 1-46. 30

.... DeltaS DeltaX S = E Gamma XS: A crucial observation for SDO is that the above Newton system might have no symmetric solution DeltaX . Many researchers have proposed different ways of symmetrizing the third equation in the Newton system so that the new system have a unique symmetric solution [20, 21]. In this paper we consider the symmetrization scheme that yields the NT direction [15, 21] Let us define the matrix P = X 1 2 (X 1 2 SX 1 2 ) Gamma 1 2 X 1 2 = S Gamma 1 2 (S 1 2 XS 1 2 ) 1 2 S Gamma 1 2 ; 41) and D = P 1 2 . The matrix D can be used to rescale X ....

....scheme that yields the NT direction [15, 21] Let us define the matrix P = X 1 2 (X 1 2 SX 1 2 ) Gamma 1 2 X 1 2 = S Gamma 1 2 (S 1 2 XS 1 2 ) 1 2 S Gamma 1 2 ; 41) and D = P 1 2 . The matrix D can be used to rescale X and S to the same matrix V defined by ( [3, 15, 20, 19]) V : 1 p D Gamma1 XD Gamma1 = 1 p DSD: 42) Obviously the matrices D and V are symmetric, and positive definite. Also defining A i : DA i D; i = 1; Delta Delta Delta ; m; DX : 1 p D Gamma1 DeltaX D Gamma1 ; D S : 1 p D DeltaSD: 43) Then the NT search ....

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Technical Report 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853, October 1997.

....the product of two symmetric matrices (see (4.24c) is not necessarily symmetric. Before Newton s method can be applied the domain and range have to be reconciled. The various primal dual algorithms differ partly in the manner in which they achieve this reconciliation. The paper of Todd [24] is witness to the intensity of research in SDP interior point methods: It describes twenty techniques for obtaining search directions for SDP, among the most notable being the following: 1) the AHO search direction proposed by Alizadeh, Haeberly and Overton; 2) the KSH HRVW M search direction ....

M. J. Todd. A study of search directions in primal-dual interior-point methods for semidefinite programming. Technical report, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY, February 1999.

.... P 1 = 0 pert. compl. slack. 5.1) Remark 5.5 In fact, most algorithms use the P I = 0 perturbed version of complementary slackness. We speci cally use (5.1) since it allows us to exploit sparsity. However, we pay for this with some loss of accuracy near the optimum. See [112] for a discussion of the many di erent choices for search directions. Two algorithms can be derived. The dual step rst exploits sparsity if many elements are free; while the primal step rst exploits sparsity if many elements are xed. The details are given in [56] We will follow a similar ....

....of the barrier parameter is n = trace ( E P ) 2H (2) P A) Diag (y) 5.15) The Newton direction is dependent on which of the equations (5.13) 5.14) we choose to solve. The equation (5.14) is shown to perform better in many applications. A discussion on various choices is given in [112]. See also [72] However we choose (5.13) below in order to exploit sparsity. The linearization to nd the Newton direction is done below. 70 5.4.3 Primal Dual Feasible Algorithm Dual Step First The algorithm essentially solves for the step h; w and backtracks to ensure both primal and dual ....

M.J. TODD. A study of search directions in primal-dual interiorpoint methods for semidenite programming. Optim. Methods Softw., 11&12:1-46, 1999.

....barrier problem for SDP is: minimize f (X) # j log d j (X) subject to h i (X) 0, i # E . But, as we remarked at the end of Section 3, # j log d j (X) log det X and so this barrier problem matches precisely the one used in the usual interior point methods for SDP (see [13] for a survey of these methods) Hence, whether or not one adds slack variables 16 VANDERBEI AND YURTTAN before introducing the barrier problem is the first and principle difference distinguishing the method presented in this paper from other interior point methods. How does the addition of ....

M.J. Todd. A study of search directions in primal-dual interior-point methods for semidefinite programming. Technical report, Technical Report No. 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853-3801, 1999.

....method can be applied to (1.29a) the fundamental operation in primal dual algorithms the Developments in Interior Point Methods 17 domain and range have to match. The different primal dual algorithms differ in the ways that they reconcile the domain and range of these equations. The paper [Todd, 1999] is witness to the intensity of research in SDP interior point methods: It describes twenty techniques for obtaining search directions for SDP. In many of these, the equation (1.29c) is replaced by one whose range lies in SIR n Thetan . That is, it is symmetrized and replaced with an mapping ....

Todd, M. J. (1999). A study of search directions in primaldual interior-point methods for semidefinite programming. Technical report, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY.

No context found.

M.J. Todd. A study of search directions in primal-dual interior-point methods for semidefinite programming. Interior point methods. Optim. ethods Softw., 11/12:1-46, 1999.

....y, S) # SIR nn IR m SIR nn to a point in IR m SIR nn IR nn (since XS I is in general not symmetric) which is a space of higher dimension. Many authors have suggested di#erent ways of symmetrizing the third equation in (3.1) so that the residual lies in SIR nn . Todd [19] analyzes twenty di#erent search directions for SDP. Next, we introduce some notation that we will use throughout this section. Script letters will denote linear operators on symmetric matrices. In particular, A : SIR nn # IR m is defined by AU : A i . U) m i=1 , 3.2) with adjoint ....

....direction was independently introduced by Helmberg, Rendl, Vanderbei and Wolkowicz [7] Kojima, Shindoh and Hara [10] and Monteiro [11] Finally, Nesterov and Todd [14, 13] introduced the NT direction. The reason for restricting ourselves to the above three directions is twofold. Firstly, Todd [19] introduces the concept of well defined directions , and shows that the H. K. M and NT directions give a unique search direction for every symmetric positive definite X and S and surjective operator A. The AHO direction also enjoys this property if XS SX is symmetric positive semidefinite ....

[Article contains additional citation context not shown here]

M. J. Todd, A study of search directions in primal-dual interior-point methods for semidefinite programming, Technical Report, TR-1205, School of Operations Research and Industrial Engineering, Cornell University, 1997, to appear in Optimization Methods and Software.

No context found.

M.J. TODD. A study of search directions in primal-dual interior-point methods for semidefinite programming. Optim. Methods Softw., 11&12:1--46, 1999.

No context found.

M.J. TODD. A study of search directions in primal-dual interior-point methods for semide nite programming. Optim. Methods Softw., 11&12:1-46, 1999.

No context found.

M.J. TODD. A study of search directions in primal-dual interior-point methods for semidefinite programming. Optim. Methods Softw., 11&12:1--46, 1999.

No context found.

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidenite programming, Technical Report 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853, October 1997. 14

No context found.

M.J. Todd, A study of search directions in primal-dual interior-point methods for semidenite programming, Technical Report 1205, School of Operations Research and Industrial Engineering, Cornell University, Ithaca, NY 14853, October 1997.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC