## Robust Convex Quadratically Constrained Programs (2002)

Venue: | Mathematical Programming |

Citations: | 22 - 2 self |

### BibTeX

@ARTICLE{Goldfarb02robustconvex,

author = {D. Goldfarb and G. Iyengar},

title = {Robust Convex Quadratically Constrained Programs},

journal = {Mathematical Programming},

year = {2002},

volume = {97},

pages = {495--515}

}

### OpenURL

### Abstract

In this paper we study robust convex quadratically constrained programs, a subset of the class of robust convex programs introduced by Ben-Tal and Nemirovski [4]. Unlike [4], our focus in this paper is to identify uncertainty structures that allow the corresponding robust quadratically constrained programs to be reformulated as second-order cone programs. We propose three classes of uncertainty sets that satisfy this criterion and present examples where these classes of uncertainty sets are natural. 1 Problem formulation A generic quadratically constrained program (QCP) is defined as follows.

### Citations

9740 | Statistical learning theory
- Vapnik
- 1998
(Show Context)
Citation Context ...Pattern classification using hyperplanes is called linear discrimination. In a typical application of linear discrimination, the hyperplane (w, b) is chosen by solving the following quadratic program =-=[8, 19, 28]-=-. minimize 1 2 �w�2 + C �� i=1 ξi � , subject to wT xi + b ≥ 1 − ξi, if yi = +1, w T xi + b ≥ 1 + ξi, if yi = −1, ξi ≥ 0, i = 1, . . . , l. Instead of solving (40), one typically solves its dual given... |

2465 | A tutorial on support vector machines for pattern recognition
- Burges
- 1998
(Show Context)
Citation Context ...Pattern classification using hyperplanes is called linear discrimination. In a typical application of linear discrimination, the hyperplane (w, b) is chosen by solving the following quadratic program =-=[8, 19, 28]-=-. minimize 1 2 �w�2 + C �� i=1 ξi � , subject to wT xi + b ≥ 1 − ξi, if yi = +1, w T xi + b ≥ 1 + ξi, if yi = −1, ξi ≥ 0, i = 1, . . . , l. Instead of solving (40), one typically solves its dual given... |

2070 |
Matrix Computations
- Golub, Loan
- 1996
(Show Context)
Citation Context ...∈Rn } �Ax − b�2 , (45) where A = [a1, . . . , am] T ∈ R m×n and b ∈ R n . If m ≥ n and the matrix A has full column rank, the solution of this optimization problem is given by x ∗ = � A T A) −1 A T b =-=[13]-=-. Even when additional linear and convex quadratic constraints are imposed on the solution x, such as �x� 2 ≤ M, the linear least squares problem (45) is still a convex QCP. In many applications of le... |

2043 |
Robust Statistics
- Huber
- 1981
(Show Context)
Citation Context ...− µ)(�µ − µ) T ] = (I − KC) T Σ(I − KC) + K T DK. (67) The non-robust version of this measurement model (i.e. Σ = Σ0 and D = D0 for fixed Σ0 and D0) is the well-known Gaussian linear stochastic model =-=[14]-=-. The robust measurement model developed here is a variant of the model proposed by Calafiore and El Ghaoui [9] where the a priori covariance Σ was known exactly and the noise covariance D ∈ � D : D −... |

1576 |
Portfolio Selection
- Markowitz
(Show Context)
Citation Context ...folio φ is given by rφ = r T φ. The objective is to choose a portfolio that maximizes some measure of “return” on the investment subject to appropriate constraints on the associated “risk”. Markowitz =-=[20, 21]-=- proposed a model for portfolio selection in which the “return” is the expected value E[rφ] of the portfolio return, the “risk” is the variance Var � � rφ of the return, and the optimal portfolio φ ∗ ... |

832 | Semidefinite programming
- Vandenberghe, Boy
- 1996
(Show Context)
Citation Context ...nvestigated a version of (5) in which the uncertainty structures Si are generalized ellipsoids and showed that the resulting robust optimization problem can be reduced to a semidefinite program (SDP) =-=[1, 22, 26]-=-. In this paper we explore uncertainty structures for which the corresponding robust problems can be reformulated as SOCPs. Our interest in this class of structures stems from the fact that both the w... |

499 | Interior point methods in semidefinite programming with applications to combinatorial optimization
- Alizadeh
- 1995
(Show Context)
Citation Context ...nvestigated a version of (5) in which the uncertainty structures Si are generalized ellipsoids and showed that the resulting robust optimization problem can be reduced to a semidefinite program (SDP) =-=[1, 22, 26]-=-. In this paper we explore uncertainty structures for which the corresponding robust problems can be reformulated as SOCPs. Our interest in this class of structures stems from the fact that both the w... |

369 |
Interior-point polynomial algorithms in convex programming
- Nesterov, Nemirovskii
- 1994
(Show Context)
Citation Context ...that the matrix Q is positive semidefinite. Suppose Q � 0. Then Q = V T V for some V ∈ R m×n and the quadratic constraint x T Qx ≤ −(2qT x + γ) is equivalent to the second-order cone (SOC) constraint =-=[2, 18, 22]-=- �� � � 2Vx � � (1 + γ + 2qT �� ���� ≤ 1 − γ − 2q x) T x. (2) ∗ Submitted to Math Programming, Series B. Please do not circulate. † IEOR Department, Columbia University, Email: gold@ieor.columbia.edu.... |

292 | Robust convex optimization
- Ben-Tal, Nemirovski
- 1998
(Show Context)
Citation Context ...rb † G. Iyengar ‡ July 1st, 2002 Abstract In this paper we study robust convex quadratically constrained programs, a subset of the class of robust convex programs introduced by Ben-Tal and Nemirovski =-=[4]-=-. Unlike [4], our focus in this paper is to identify uncertainty structures that allow the corresponding robust quadratically constrained programs to be reformulated as second-order cone programs. We ... |

249 | Adjustable robust solutions of uncertain linear programs
- Ben-Tal, Goryashko, et al.
- 2004
(Show Context)
Citation Context ... j� , (47) where without any loss of generality �v� ≤ 1, �u� ≤ 1, and ξ j ∼ N (0, Ω j ), j = 1, . . . , l. Without the stochastic term, the uncertainty set (47) has the affine structure considered in =-=[4, 5, 6]-=-. The term � l j=1 uj ξ j models the imperfect knowledge of the stochastic perturbations in a – the decision maker knows the total variance and modes Ω j but does not know the variance of each of the ... |

204 |
Linear Matrix Inequalities in System and Control Theory
- Boyd, Ghaoui, et al.
- 1994
(Show Context)
Citation Context ... � c0 bT 0 � p� − � ci bT i � � 0. b0 A0 i=1 τi bi Ai Moreover, if p = 1 then the converse holds if there exists x0 such that F1(x0) > 0. For a discussion of the S-procedure and its applications, see =-=[7]-=-. Since v = 0 is strictly feasible for 1 − v T Gv ≥ 0, the S-procedure implies that (28) holds for all 1 − vT Gv ≥ 0 if and only if there exists a τ ≥ 0 such that � M = ν − τ − yT 0 (F0 + ηN)y0 −ryT 0... |

162 | Applications of second-order cone programming
- Lobo, Vandenberghe, et al.
- 1998
(Show Context)
Citation Context ...that the matrix Q is positive semidefinite. Suppose Q � 0. Then Q = V T V for some V ∈ R m×n and the quadratic constraint x T Qx ≤ −(2qT x + γ) is equivalent to the second-order cone (SOC) constraint =-=[2, 18, 22]-=- �� � � 2Vx � � (1 + γ + 2qT �� ���� ≤ 1 − γ − 2q x) T x. (2) ∗ Submitted to Math Programming, Series B. Please do not circulate. † IEOR Department, Columbia University, Email: gold@ieor.columbia.edu.... |

158 | Second-order cone programming
- Alizadeh, Goldfarb
- 2003
(Show Context)
Citation Context ...that the matrix Q is positive semidefinite. Suppose Q � 0. Then Q = V T V for some V ∈ R m×n and the quadratic constraint x T Qx ≤ −(2qT x + γ) is equivalent to the second-order cone (SOC) constraint =-=[2, 18, 22]-=- �� � � 2Vx � � (1 + γ + 2qT �� ���� ≤ 1 − γ − 2q x) T x. (2) ∗ Submitted to Math Programming, Series B. Please do not circulate. † IEOR Department, Columbia University, Email: gold@ieor.columbia.edu.... |

152 | Robust solutions to least-squares problems with uncertain data
- Ghaoui, Lebret
- 1997
(Show Context)
Citation Context ... themselves tractable optimization problems. Robustness as applied to uncertain least squares problems and uncertain semidefinite programs was independently studied by El Ghaoui and his collaborators =-=[11, 12]-=-. In keeping with the formulation proposed by Ben-Tal and Nemirovski, a generic robust convex QCP is given by minimize c T x subject to x T Qix + 2q T i x + γi ≤ 0, for all (Qi, qi, γi) ∈ Si, i = 1, .... |

108 | Robust portfolio selection problems - Goldfarb, Iyengar - 2003 |

97 | Lectures on Modern Convex Optimization
- Nemirovski
- 2005
(Show Context)
Citation Context ... (x, ξ) ∈ K ⊂ R m , ∀ξ ∈ U, where ξ are the uncertain parameters, U is the uncertainty set, x ∈ R n is the decision vector, K is a convex cone and, for fixed ξ ∈ U, the function F (x, ξ) is K-concave =-=[4, 6]-=-. The optimization problem (4) is called a robust optimization problem. Ben-Tal and Nemirovski established that for certain classes of uncertainty sets U, robust counterparts of linear programs, quadr... |

84 | Robust solutions to uncertain semidefinite programs
- Ghaoui, Oustry, et al.
- 1999
(Show Context)
Citation Context ... themselves tractable optimization problems. Robustness as applied to uncertain least squares problems and uncertain semidefinite programs was independently studied by El Ghaoui and his collaborators =-=[11, 12]-=-. In keeping with the formulation proposed by Ben-Tal and Nemirovski, a generic robust convex QCP is given by minimize c T x subject to x T Qix + 2q T i x + γi ≤ 0, for all (Qi, qi, γi) ∈ Si, i = 1, .... |

56 | Robust truss topology design via semidefinite programming
- Ben-Tal, Nemirovski
- 1997
(Show Context)
Citation Context ...oblems are typically sensitive to parameter fluctuations, the errors in the input parameters tend to get amplified in the decision vector. This, in turn, leads to a sharp deterioration in performance =-=[3, 10]-=-. The problem of choosing a decision vector in the presence of parameter perturbation was formalized by Ben-Tal and Nemirovski [4, 5] as follows: minimize c T x subject to F (x, ξ) ∈ K ⊂ R m , ∀ξ ∈ U,... |

50 | Portfolio Optimization with Linear and Fixed Transaction Costs - Lobo, Fazel, et al. - 2006 |

28 | Robust linear and support vector regression
- Mangasarian, Musicant
- 2000
(Show Context)
Citation Context ...Pattern classification using hyperplanes is called linear discrimination. In a typical application of linear discrimination, the hyperplane (w, b) is chosen by solving the following quadratic program =-=[8, 19, 28]-=-. minimize 1 2 �w�2 + C �� i=1 ξi � , subject to wT xi + b ≥ 1 − ξi, if yi = +1, w T xi + b ≥ 1 + ξi, if yi = −1, ξi ≥ 0, i = 1, . . . , l. Instead of solving (40), one typically solves its dual given... |

11 |
A New Method of Channel Shortening With Applications to Discrete Multi Tone (DMT) Systems. Part II: Training,” submitted to
- Pal, Iyengar, et al.
(Show Context)
Citation Context ... .. haM−1 . .. . . . . .. . .. . .. . .. . .. . .. . .. . .. 0 . . 0 hj0 hj1 hj2 hj3 . ⎤ ⎥ ⎥, ⎥ ⎦ ⎡ ⎢ gj = ⎢ ⎣ � 0 . . . . . . �� (n+m−1)×n 0 hj,m−1 � gj0 gj1 . gj,n−1 Under fairly general conditions =-=[24, 23]-=- the system of equations (58) has a solution provided p ≥ 2, i.e. on can find finite impulse response filters Gj(z), j = 1, . . . , p that can shorten the channel to any given target D(z). This equali... |

9 | Robust linear programming and optimal control
- Vandenberghe, Boyd, et al.
- 2002
(Show Context)
Citation Context ...ainty sets that meet this criterion. Adding robustness reduces the sensitivity of the optimal decision to fluctuations in the parameters and can often result in significant improvement in performance =-=[3, 16, 27]-=-. Typically, the complexity of the deterministic reformulation of the robust problem is higher than the non-robust version of the problem. However, since the worst case complexity of convex QCPs is co... |

7 |
The effect of errors in means, variances and covariances on optimal portfolio choice
- Chopra, Ziemba
- 1993
(Show Context)
Citation Context ...oblems are typically sensitive to parameter fluctuations, the errors in the input parameters tend to get amplified in the decision vector. This, in turn, leads to a sharp deterioration in performance =-=[3, 10]-=-. The problem of choosing a decision vector in the presence of parameter perturbation was formalized by Ben-Tal and Nemirovski [4, 5] as follows: minimize c T x subject to F (x, ξ) ∈ K ⊂ R m , ∀ξ ∈ U,... |

1 | Ghaoui. Minimum variance estimation with uncertain statistical model
- Calafiore, El
- 2001
(Show Context)
Citation Context ...= Σ0 and D = D0 for fixed Σ0 and D0) is the well-known Gaussian linear stochastic model [14]. The robust measurement model developed here is a variant of the model proposed by Calafiore and El Ghaoui =-=[9]-=- where the a priori covariance Σ was known exactly and the noise covariance D ∈ � D : D −1 = D −1 0 + L∆R + RT ∆ T L T � 0, �∆� ≤ 1 � . They show that the problem of choosing the gain matrix K to mini... |