Search

In this paper, we study zeroth-order algorithms for nonconvex-concave minimax problems, which have attracted much attention in machine learning, signal processing, and many other fields in recent years. We propose a zeroth-order alternating randomized gradient projection (ZO-AGP) algorithm for smooth nonconvex-concave minimax problems; its iteration complexity to obtain an varepsilon -stationary point is bounded by crO (\varepsilon 4), and the number of function value estimates is bounded by scrO (dx + dy) per iteration. Moreover, we propose a zeroth-order block alternating randomized proximal gradient algorithm (ZO-BAPG) for solving blockwise nonsmooth nonconvexconcave minimax optimization problems; its iteration complexity to obtain an varepsilon -stationary point is bounded by scrO (varepsilon 4), and the number of function value estimates per iteration is bounded by O (Kdx +dy). To the best of our knowledge, this is the first time zeroth-order algorithms with iteration complexity guarantee are developed for solving both general smooth and blockwise nonsmooth nonconvex-concave minimax problems. Numerical results on the data poisoning attack problem and the distributed nonconvex sparse principal component analysis problem validate the efficiency of the proposed algorithms. [ABSTRACT FROM AUTHOR]

Recent years have seen significant advances in quantum/quantum-inspired technologies capable of approximately searching for the ground state of Ising spin Hamiltonians. The promise of leveraging such technologies to accelerate the solution of difficult optimization problems has spurred an increased interest in exploring methods to integrate Ising problems as part of their solution process, with existing approaches ranging from direct transcription to hybrid quantum- classical approaches rooted in existing optimization algorithms. While it is widely acknowledged that quantum computers should augment classical computers, rather than replace them entirely, comparatively little attention has been directed toward deriving analytical characterizations of their interactions. In this paper, we present a formal analysis of hybrid algorithms in the context of solving mixed-binary quadratic programs (MBQP) via Ising solvers. By leveraging an existing completely positive reformulation of MBQPs, as well as a new strong-duality result, we show the exactness of the dual problem over the cone of copositive matrices, thus allowing the resulting reformulation to inherit the straightforward analysis of convex optimization. We propose to solve this reformulation with a hybrid quantum-classical cutting-plane algorithm. Using existing complexity results for convex cutting-plane algorithms, we deduce that the classical portion of this hybrid framework is guaranteed to be polynomial time. This suggests that when applied to NP-hard problems, the complexity of the solution is shifted onto the subroutine handled by the Ising solver. [ABSTRACT FROM AUTHOR]

In this paper we consider finding a second-order stationary point (SOSP) of nonconvex equality constrained optimization when a nearly feasible point is known. In particular, we first propose a new Newton-conjugate gradient (Newton-CG) method for finding an approximate SOSP of unconstrained optimization and show that it enjoys a substantially better complexity than the Newton-CG method in [C. W. Royer, M. O'Neill, and S. J. Wright, Math. Program., 180 (2020), pp. 451-488]. We then propose a Newton-CG based augmented Lagrangian (AL) method for finding an approximate SOSP of nonconvex equality constrained optimization, in which the proposed Newton-CG method is used as a subproblem solver. We show that under a generalized linear independence constraint qualification (GLICQ), our AL method enjoys a total inner iteration complexity of O(ε7/peration complexity of O(ε7/2 min\{ n, ε3/4\}) for finding an (ε, ε)-SOSP of nonconvex equality constrained optimization with high probability, which are significantly better than the ones achieved by the proximal AL method in [Y. Xie and S. J. Wright, J. Sci. Comput., 86 (2021), pp. 1-30]. In addition, we show that it has a total inner iteration complexity of O(ε11/2) and an operation complexity of O(ε11/2 min\{ n, ε5/4\}) when the GLICQ does not hold. To the best of our knowledge, all the complexity results obtained in this paper are new for finding an approximate SOSP of nonconvex equality constrained optimization with high probability. Preliminary numerical results also demonstrate the superiority of our proposed methods over the other competing algorithms. [ABSTRACT FROM AUTHOR]

Proximal mappings are essential in splitting algorithms for both convex and noncon-vex optimization. In this paper, we show that proximal mappings of every prox-bounded function are nonexpansive if and only if they are firmly nonexpansive if and only if they are averaged if and only if the function is convex. Lipschitz proximal mappings of prox-bounded functions are also characterized via hypoconvex or strongly convex functions. Our results generalize a recent result due to Rockafellar [ABSTRACT FROM AUTHOR]

Given closed convex sets Ci, i=1, ... l, and some nonzero linear maps Ai, i=1, ... l, of suitable dimensions, the multiset split feasibility problem aims at finding a point in ... based on computing projections onto Ci and multiplications by Ai and AiT. In this paper, we consider the associated best approximation problem, i.e., the problem of computing projections onto ...; we refer to this problem as the best approximation problem in multiset split feasibility settings (BA-MSF). We adapt the Dykstra's projection algorithm, which is classical for solving the BA-MSF in the special case when all Ai - I, to solve the general BA-MSF. Our Dykstra-type projection algorithm is derived by applying (proximal) coordinate gradient descent to the Lagrange dual problem, and it only requires computing projections onto Ci and multiplications by Ai and AiT in each iteration. Under a standard relative interior condition and a genericity assumption on the point we need to project, we show that the dual objective satisfies the Kurdyka-Łojasiewicz property with an explicitly computable exponent on a neighborhood of the (typically unbounded) dual solution set when each Ci is C1,α.-cone reducible for some α∈ (0,1]: this class of sets covers the class of C²-cone reducible sets, which include all polyhedrons, second-order cone, and the cone of positive semidefinite matrices as special cases. Using this, explicit convergence rate (linear or sublinear) of the sequence generated by the Dykstra-type projection algorithm is derived. Concrete examples are constructed to illustrate the necessity of some of our assumptions. [ABSTRACT FROM AUTHOR]

This paper proposes a path-based approach for the minimization of a continuously differentiable function over sparse symmetric sets, which is a hard problem that exhibits a restrictiveness-hierarchy of necessary optimality conditions. To achieve the more restrictive conditions in the hierarchy, state-of-the-art algorithms require a support optimization oracle that must exactly solve the problem in smaller dimensions. The path-based approach developed in this study produces a path-based optimality condition, which is placed well in the restrictiveness-hierarchy, and a method to achieve it that does not require a support optimization oracle and, moreover, is projection-free. In the development process, new results are derived for the regularized linear minimization problem over sparse symmetric sets, which give additional means to identify optimal solutions for convex and concave objective functions. We complement our results with numerical examples. [ABSTRACT FROM AUTHOR]

In this paper, we have studied a decomposition method for solving a class of non-convex two-stage stochastic programs, where both the objective and constraints of the second-stage problem are nonlinearly parameterized by the first-stage variables. Due to the failure of the Clarke regularity of the resulting nonconvex recourse function, classical decomposition approaches such as Benders decomposition and (augmented) Lagrangian-based algorithms cannot be directly generalized to solve such models. By exploring an implicitly convex-concave structure of the recourse function, we introduce a novel decomposition framework based on the so-called partial Moreau envelope. The algorithm successively generates strongly convex quadratic approximations of the recourse function based on the solutions of the second-stage convex subproblems and adds them to the first-stage mas-ter problem. Convergence has been established for both a fixed number of scenarios and a sequential internal sampling strategy. Numerical experiments are conducted to demonstrate the effectiveness of the proposed algorithm. [ABSTRACT FROM AUTHOR]

In this paper, we consider a class of structured nonsmooth difference-of-convex (DC) minimization in which the first convex component is the sum of a smooth and a nonsmooth function, while the second convex component is the supremum of finitely many convex smooth functions. The existing methods for this problem usually have weak convergence guarantees or need to solve lots of subproblems per iteration. Due to this, we propose hybrid algorithms for solving this problem in which we first compute approximate critical points and then check whether these points are approximate D-stationary points. Under suitable conditions, we prove that there exists a subsequence of iterates of which every accumulation point is a D-stationary point. Some preliminary numerical experiments are conducted to demonstrate the efficiency of the proposed algorithms. [ABSTRACT FROM AUTHOR]

Symmetries of a set are linear transformations that map the set to itself. A fundamental domain of a symmetric set is a subset that contains at least one representative from each of the symmetric equivalence classes (orbits) in the set. This paper contributes a novel polynomial algorithm for constructing minimal polytopic fundamental domains of polytopic sets. Our algorithm is applicable for generic linear symmetries of the set and has linear complexity in the number of facets and dimension of the symmetric polytope, e.g., the feasible region of an optimization problem. In addition, this paper contributes a novel polynomial algorithm for mapping an element of the polytope to its representative in the fundamental domain. The algorithms are demonstrated in four examples--two illustrative and two practical. In the first practical example, we show that a minimal fundamental domain of the hypercube under the symmetric group is the set of points with sorted elements. In the second practical example, we show how the construction algorithm can be applied to the max-cut problem. [ABSTRACT FROM AUTHOR]

The paper deals with stochastic difference-of-convex-functions (DC) programs, that is, optimization problems whose cost function is a sum of a lower semicontinuous DC function and the expectation of a stochastic DC function with respect to a probability distribution. This class of nonsmooth and nonconvex stochastic optimization problems plays a central role in many practical applications. Although there are many contributions in the context of convex and/or smooth stochastic optimization, algorithms dealing with nonconvex and nonsmooth programs remain rare. In deterministic optimization literature, the DC algorithm (DCA) is recognized to be one of the few algorithms able to effectively solve nonconvex and nonsmooth optimization problems. The main purpose of this paper is to present some new stochastic DCAs for solving stochastic DC programs. The convergence analysis of the proposed algorithms is carefully studied, and numerical experiments are conducted to justify the algorithms' behaviors. [ABSTRACT FROM AUTHOR]

In this paper we consider large-scale composite optimization problems having the objective function formed as a sum of two terms (possibly nonconvex); one has a (block) coordinatewise Lipschitz continuous gradient and the other is differentiable but nonseparable. Under these general settings we derive and analyze two new coordinate descent methods. The first algorithm, referred to as the coordinate proximal gradient method, considers the composite form of the objective function, while the other algorithm disregards the composite form of the objective and uses the partial gradient of the full objective, yielding a coordinate gradient descent scheme with novel adaptive stepsize rules. We prove that these new stepsize rules make the coordinate gradient scheme a descent method, provided that additional assumptions hold for the second term in the objective function. We present a complete worst-case complexity analysis for these two new methods in both convex and nonconvex settings, provided that the (block) coordinates are chosen random or cyclic. Preliminary numerical results also confirm the efficiency of our two algorithms for practical problems. [ABSTRACT FROM AUTHOR]

An adaptive regularization algorithm for unconstrained nonconvex optimization is presented in which the objective function is never evaluated but only derivatives are used. This algorithm belongs to the class of adaptive regularization methods, for which optimal worst-case complexity results are known for the standard framework where the objective function is evaluated. It is shown in this paper that these excellent complexity bounds are also valid for the new algorithm despite the fact that significantly less information is used. In particular, it is shown that if derivatives of degree one to p are used, the algorithm will find an ε 1-approximate first-order minimizer in at most O(ε (p+1)/p 1 ) iterations and an (ε 1, ε 2)-approximate second-order minimizer in at most O(max[ε¹ (p+1)/p 1, ε (p+1)/(p 1) 2 ]) iterations. As a special case, the new algorithm using first and second derivatives, when applied to functions with Lipschitz continuous Hessian, will find an iterate xk at which the gradient's norm is less than ε 1 in at most O(ε1 3/2) iterations. [ABSTRACT FROM AUTHOR]

Packing and covering semidefinite programs (SDPs) appear in natural relaxations of many combinatorial optimization problems as well as a number of other applications. Recently, several techniques were proposed, which utilize the particular structure of this class of problems, to obtain more efficient algorithms than those offered by general SDP solvers. For certain applications, such as those described in this paper, it may be desirable to obtain sparse dual solutions, i.e., those with support size (almost) independent of the number of primal constraints. In this paper, we give an algorithm that finds such solutions, which is an extension of a logarithmic-potential based algorithm of Grigoriadis et al. [SIAM J. Optim. 11 (2001), pp. 1081-1091] for packing/covering linear programs. [ABSTRACT FROM AUTHOR]

Recent papers indicate that some algorithms for constrained optimization may exhibit worst-case complexity bounds that are very similar to those of unconstrained optimization algorithms. A natural question is whether well-established practical algorithms, perhaps with small variations, may enjoy analogous complexity results. In the present paper we show that the answer is positive with respect to inexact restoration algorithms in which first-order approximations are employed for defining the subproblems. [ABSTRACT FROM AUTHOR]

We study the random reshuffling (RR) method for smooth nonconvex optimization problems with a finite-sum structure. Though this method is widely utilized in practice, e.g., in the training of neural networks, its convergence behavior is only understood in several limited settings. In this paper, under the well-known Kurdyka-Łojasiewicz (KL) inequality, we establish strong limit-point convergence results for RR with appropriate diminishing step sizes; namely, the whole sequence of iterates generated by RR is convergent and converges to a single stationary point in an almost sure sense. In addition, we derive the corresponding rate of convergence, depending on the KL exponent and suitably selected diminishing step sizes. When the KL exponent lies in [0,1/2], the convergence is at a rate of O(t-1) with t counting the number of iterations. When the KL exponent belongs to (1/2,1), our derived convergence rate is of the form O(t-q) with q∈(0,1) depending on the KL exponent. The standard KL inequality-based convergence analysis framework only applies to algorithms with a certain descent property. We conduct a novel convergence analysis for the nondescent RR method with diminishing step sizes based on the KL inequality, which generalizes the standard KL framework. We summarize our main steps and core ideas in an informal analysis framework, which is of independent interest. As a direct application of this framework, we establish similar strong limit-point convergence results for the reshuffled proximal point method. [ABSTRACT FROM AUTHOR]

Finding a zero of a sum of maximally monotone operators is a fundamental problem in modern optimization and nonsmooth analysis. Assuming that the resolvents of the operators are available, this problem can be tackled with the Douglas-Rachford algorithm. However, when dealing with three or more operators, one must work in a product space with as many factors as there are operators. In groundbreaking recent work by Ryu and by Malitsky and Tam, it was shown that the number of factors can be reduced by one. A similar reduction was achieved recently by Campoy through a clever reformulation originally proposed by Kruger. All three splitting methods guarantee weak convergence to some solution of the underlying sum problem; strong convergence holds in the presence of uniform monotonicity. In this paper, we provide a case study when the operators involved are normal cone operators of subspaces and the solution set is thus the intersection of the subspaces. Even though these operators lack strict convexity, we show that striking conclusions are available in this case: strong (instead of weak) convergence and the solution obtained is (not arbitrary but) the projection onto the intersection. To illustrate our results, we also perform numerical experiments. [ABSTRACT FROM AUTHOR]

This paper presents smoothing schemes for obtaining approximate stationary points of unconstrained or linearly constrained composite nonconvex-concave min-max (and hence nonsmooth) problems by applying well-known algorithms to composite smooth approximations of the original problems. More specifically, in the unconstrained (resp., constrained) case, approximate stationary points of the original problem are obtained by applying, to its composite smooth approximation, an accelerated inexact proximal point (resp., quadratic penalty) method presented in a previous paper by the authors. Iteration complexity bounds for both smoothing schemes are also established. Finally, numerical results are given to demonstrate the efficiency of the unconstrained smoothing scheme. [ABSTRACT FROM AUTHOR]

In this paper, we introduce a stochastic projected subgradient method for weakly convex (i.e., uniformly prox-regular) nonsmooth, nonconvex functions---a wide class of functions which includes the additive and convex composite classes. At a high level, the method is an inexact proximal-point iteration in which the strongly convex proximal subproblems are quickly solved with a specialized stochastic projected subgradient method. The primary contribution of this paper is a simple proof that the proposed algorithm converges at the same rate as the stochastic gradient method for smooth nonconvex problems. This result appears to be the first convergence rate analysis of a stochastic (or even deterministic) subgradient method for the class of weakly convex functions. In addition, a two-phase variant is proposed that significantly reduces the variance of the solutions returned by the algorithm. Finally, preliminary numerical experiments are also provided. [ABSTRACT FROM AUTHOR]

Recently, in a series of papers [Y. Rahimi, C. Wang, H. Dong, and Y. Lou, SIAM J. Sci. Comput., 41 (2019), pp. A3649-A3672; C. Wang, M. Tao, J. Nagy, and Y. Lou, SIAM J. Imaging Sci., 14 (2021), pp. 749-777; C. Wang, M. Yan, and Y. Lou, IEEE Trans. Signal Process., 68 (2020), pp. 2660-2669; P. Yin, E. Esser, and J. Xin, Commun. Inf. Syst., 14 (2014), pp. 87-109], the ratio of ℓ1 and ℓ2 norms was proposed as a sparsity inducing function for noiseless compressed sensing. In this paper, we further study properties of such model in the noiseless setting, and propose an algorithm for minimizing ℓ1/ℓ2 subject to noise in the measurements. Specifically, we show that the extended objective function (the sum of the objective and the indicator function of the constraint set) of the model in [Y. Rahimi, C. Wang, H. Dong, and Y. Lou, SIAM J. Sci. Comput., 41 (2019), pp. A3649-A3672] satisfies the Kurdyka--Lojasiewicz (KL) property with exponent 1/2; this allows us to establish linear convergence of the algorithm proposed in [C. Wang, M. Yan, and Y. Lou, IEEE Trans. Signal Process., 68 (2020), pp. 2660-2669] (see equation 11) under mild assumptions. We next extend the ℓ1/ℓ2 model to handle compressed sensing problems with noise. We establish the solution existence for some of these models under the spherical section property [S. A. Vavasis, Derivation of Compressive Sensing Theorems from the Spherical Section Property, University of Waterloo, 2009; Y. Zhang, J. Oper. Res. Soc. China, 1 (2013), pp. 79-105] and extend the algorithm in [C. Wang, M. Yan, and Y. Lou, IEEE Trans. Signal Process., 68 (2020), pp. 2660-2669] (see equation 11) by incorporating moving-balls-approximation techniques [A. Auslender, R. Shefi, and M. Teboulle, SIAM J. Optim., 20 (2010), pp. 3232-3259] for solving these problems. We prove the subsequential convergence of our algorithm under mild conditions and establish global convergence of the whole sequence generated by our algorithm by imposing additional KL and differentiability assumptions on a specially constructed potential function. Finally, we perform numerical experiments on robust compressed sensing and basis pursuit denoising with residual error measured by \ell 2 norm or Lorentzian norm via solving the corresponding ℓ1/ℓ2 models by our algorithm. Our numerical simulations show that our algorithm is able to recover the original sparse vectors with reasonable accuracy. [ABSTRACT FROM AUTHOR]

This paper develops a novel limited-memory method to solve dynamic optimization problems. The memory requirements for such problems often present a major obstacle, particularly for problems with PDE constraints such as optimal flow control, full waveform inversion, and optical tomography. In these problems, PDE constraints uniquely determine the state of a physical system for a given control; the goal is to find the value of the control that minimizes an objective. While the control is often low dimensional, the state is typically more expensive to store. This paper suggests using randomized matrix approximation to compress the state as it is generated and shows how to use the compressed state to reliably solve the original dynamic optimization problem. Concretely, the compressed state is used to compute approximate gradients and to apply the Hessian to vectors. The approximation error in these quantities is controlled by the target rank of the sketch. This approximate first- and second-order information can readily be used in any optimization algorithm. As an example, we develop a sketched trust-region method that adaptively chooses the target rank using a posteriori error information and provably converges to a stationary point of the original problem. Numerical experiments with the sketched trust-region method show promising performance on challenging problems such as the optimal control of an advection-reaction-diffusion equation and the optimal control of fluid flow past a cylinder. [ABSTRACT FROM AUTHOR]

In this paper, we propose two new solution schemes to solve the stochastic strongly monotone variational inequality (VI) problems: the stochastic extra-point solution scheme and the stochastic extra-momentum solution scheme. The first one is a general scheme based on updating the iterative sequence and an auxiliary extra-point sequence. In the case of a deterministic VI model, this approach includes several state-of-the-art first-order methods as its special cases. The second scheme combines two momentum-based directions: the so-called heavy-ball direction and the optimism direction, where only one projection per iteration is required in its updating process. We show that if the variance of the stochastic oracle is appropriately controlled, then both schemes can be made to achieve optimal iteration complexity of O(κ ln (1/∈)) to reach an ∈ -solution for a strongly monotone VI problem with condition number κ. As a specific application to stochastic VI, we demonstrate how to incorporate a zeroth-order approach for solving stochastic minimax saddle-point problems in our schemes, where only noisy and biased samples of the objective can be obtained, with a total sample complexity of O (κ/∈). [ABSTRACT FROM AUTHOR]

In this paper, we develop new primal-dual algorithms to solve a class of nonsmooth and nonlinear convex-concave minimax problems, which covers many existing and brand-new models as special cases. Our approach relies on a combination of a generalized augmented Lagrangian function, Nesterov's accelerated scheme, and adaptive parameter updating strategies. Our algorithmic framework is single-loop and unifies two important settings: general convex-concave and convex-linear cases. Under mild assumptions, our algorithms achieve O (1/k) convergence rates through three different criteria: primal-dual gap, primal objective residual, and dual objective residual, where k is the iteration counter. Our rates are both ergodic (i.e., on a weighted averaging sequence) and nonergodic (i.e., on the last-iterate sequence). These convergence rates can be boosted up to O (1/k²) if only one objective term is strongly convex (or, equivalently, its conjugate is L-smooth). To the best of our knowledge, this is the first algorithm achieving optimal rates on the primal last-iterate sequence for convex-linear minimax problems. As a byproduct, we specify our algorithms to solve a general convex cone constrained program with both ergodic and nonergodic rate guarantees. We test our algorithms and compare them with two recent methods on two numerical examples. [ABSTRACT FROM AUTHOR]

Conjugate gradient methods are important first-order optimization algorithms both in Euclidean spaces and on Riemannian manifolds. However, while various types of conjugate gradient methods have been studied in Euclidean spaces, there are relatively fewer studies for those on Riemannian manifolds (i.e., Riemannian conjugate gradient methods). This paper proposes a novel general framework that unifies existing Riemannian conjugate gradient methods such as the ones that utilize a vector transport or inverse retraction. The proposed framework also develops other methods that have not been covered in previous studies. Furthermore, conditions for the convergence of a class of algorithms in the proposed framework are clarified. Moreover, the global convergence properties of several specific types of algorithms are extensively analyzed. The analysis provides the theoretical results for some algorithms in a more general setting than the existing studies and new developments for other algorithms. Numerical experiments are performed to confirm the validity of the theoretical results. The experimental results are used to compare the performances of several specific algorithms in the proposed framework. [ABSTRACT FROM AUTHOR]

This paper generalizes the optimized gradient method (OGM) [Y. Drori and M. Teboulle, Math. Program., 145 (2014), pp. 451-482], [D. Kim and J. A. Fessler, Math. Program., 159 (2016), pp. 81-107], [D. Kim and J. A. Fessler, J. Optim. Theory Appl., 172 (2017), pp. 187-205] that achieves the optimal worst-case cost function bound of first-order methods for smooth convex minimization [Y. Drori, J. Complexity, 39 (2017), pp. 1-16]. Specifically, this paper studies a generalized formulation of OGM and analyzes its worst-case rates in terms of both the function value and the norm of the function gradient. This paper also develops a new algorithm called OGM-OG that is in the generalized family of OGM and that has the best known analytical worst-case bound with rate O(1/N1.5) on the decrease of the gradient norm among fixed-step first-order methods. This paper also proves that Nesterov's fast gradient method [Y. Nesterov, Dokl. Akad. Nauk. USSR, 269 (1983), pp. 543-547], [Y. Nesterov, Math. Program., 103 (2005), pp. 127-152] has an O(1/N1.5) worst-case gradient norm rate but with constant larger than OGM-OG. The proof is based on the worst-case analysis called the Performance Estimation Problem in [Y. Drori and M. Teboulle, Math. Program., 145 (2014), pp. 451-482]. [ABSTRACT FROM AUTHOR]

In this paper we describe a systematic procedure to analyze the convergence of degenerate preconditioned proximal point algorithms. We establish weak convergence results under mild assumptions that can be easily employed in the context of splitting methods for monotone inclusion and convex minimization problems. Moreover, we show that the degeneracy of the preconditioner allows for a reduction of the variables involved in the iteration updates. We show the strength of the proposed framework in the context of splitting algorithms, providing new simplified proofs of convergence and highlighting the link between existing schemes, such as Chambolle-Pock, forward Douglas-Rachford, and Peaceman-Rachford, that we study from a preconditioned proximal point perspective. The proposed framework allows us to devise new flexible schemes and provides new ways to generalize existing splitting schemes to the case of the sum of many terms. As an example, we present a new sequential generalization of forward Douglas-Rachford along with numerical experiments that demonstrates its interest in the context of nonsmooth convex optimization. [ABSTRACT FROM AUTHOR]

The golden ratio primal-dual algorithm (GRPDA) is a new variant of the classical Arrow-Hurwicz method for solving structured convex optimization problems, in which the objective function consists of the sum of two closed proper convex functions, one of which involves a composition with a linear transform. The same as the Arrow-Hurwicz method and the popular primal-dual algorithm (PDA) of Chambolle and Pock, GRPDA is full-splitting in the sense that it does not rely on solving any subproblems or linear system of equations iteratively. Compared with PDA, an important feature of GRPDA is that it permits larger primal and dual stepsizes. However, the stepsize condition of GRPDA requires that the spectral norm of the linear transform is known, which can be difficult to obtain in some applications. Furthermore, constant stepsizes are usually overconservative in practice. In this paper, we propose a linesearch strategy for GRPDA, which not only does not require the spectral norm of the linear transform but also allows adaptive and potentially much larger stepsizes. Within each linesearch step, only the dual variable needs to be updated, and it is thus quite cheap and does not require any extra matrix-vector multiplications for many special yet important applications such as a regularized least-squares problem. Global iterate convergence and 0(1/N) ergodic convergence rate results, measured by the function value gap and constraint violations of an equivalent optimization problem, are established, where N denotes the iteration counter. When one of the component functions is strongly convex, faster O(l/N²) ergodic convergence rate results, quantified by the same measures, are established by adaptively choosing some algorithmic parameters. Moreover, when the subdifferential operators of the component functions are strongly metric subregular, a condition that is much weaker than strong convexity, we show that the iterates converge R-linearly to the unique solution. Numerical experiments on matrix game and LASSO problems illustrate the effectiveness of the proposed linesearch strategy. [ABSTRACT FROM AUTHOR]

We introduce a new feasible corrector-predictor (CP) interior-point algorithm (IPA), which is suitable for solving linear complementarity problem (LCP) with P* (K)-matrices. We use the method of algebraically equivalent transformation (AET) of the nonlinear equation of the system which defines the central path. The AET is based on the function φ(ί) = t -- t and plays a crucial role in the calculation of the new search direction. We prove that the algorithm has O iteration complexity, where κ is an upper bound of the handicap of the input matrix. To the best of our knowledge, this is the first CP IPA for P*(κ)-LCPs which is based on this search direction. We implement the proposed CP IPA in the C++ programming language with specific parameters and demonstrate its performance on three families of LCPs. The first family consists of LCPs with P* (K)-matrices. The second family of LCPs has the P-matrix defined by Csizmadia. Eisenberg-Nagy and de Klerk [Math. Program., 129 (2011), pp. 383--402] showed that the handicap of this matrix should be at least 22n--8 -- 1/4. Namely, from the known complexity results for P*(κ)-LCPs it might follow that the computational performance of IPAs on LCPs with the matrix defined by Csizmadia could be very poor. Our preliminary computational study shows that an implemented variant of the theoretical version of the CP IPA (Algorithm 4.1) presented in this paper, finds a e-approximate solution for LCPs with the Csizmadia matrix in a very small number of iterations. The third family of problems consists of the LCPs related to the copositivity test of 88 matrices from [C. Bras, G. Eichfelder, and J. Judice, Comput. Optim. Appl., 63 (2016), pp. 461--493]. For each of these matrices we create a special LCP and try to solve it using our IPA. If the LCP does not have a solution, then the related matrix is strictly copositive, otherwise it is on the boundary or outside the copositive cone. For these LCPs we do not know whether the underlying matrix is P*(k) or not, but we could reveal the real copositivity status of the input matrices in 83 out of 88 cases (accuracy ≤ 94%). The numerical test shows that our CP IPA performs well on the sets of test problems used in the paper. [ABSTRACT FROM AUTHOR]

In this paper, we study smooth stochastic multilevel composition optimization problems, where the objective function is a nested composition of T functions. We assume access to noisy evaluations of the functions and their gradients, through a stochastic first-order oracle. For solving this class of problems, we propose two algorithms using moving-average stochastic estimates, and analyze their convergence to an e-stationary point of the problem. We show that the first algorithm, which is a generalization of [S. Ghadimi, A. Ruszczynski, and M. Wang, SIAM J. Optim., 30 (2020), pp. 960-979] to the T level case, can achieve a sample complexity of OT(1/ε6) by using minibatches of samples in each iteration, where OT hides constants that depend on T. By modifying this algorithm using linearized stochastic estimates of the function values, we improve the sample complexity to OT(1/ε4). This modification not only removes the requirement of having a minibatch of samples in each iteration, but also makes the algorithm parameter-free and easy to implement. To the best of our knowledge, this is the first time that such an online algorithm designed for the (un)constrained multilevel setting obtains the same sample complexity of the smooth single-level setting, under standard assumptions (unbiasedness and boundedness of the second moments) on the stochastic first-order oracle. [ABSTRACT FROM AUTHOR]

In this paper, we consider a class of single-ratio fractional minimization problems, in which the numerator of the objective is the sum of a nonsmooth nonconvex function and a smooth nonconvex function while the denominator is a nonsmooth convex function. In this work, we first derive its first-order necessary optimality condition, by using the first-order operators of the three functions involved. Then we develop first-order algorithms, namely, the proximity-gradientsubgradient algorithm (PGSA), PGSA with monotone line search (PGSA ML), and PGSA with nonmonotone line search (PGSA NL). It is shown that any accumulation point of the sequence generated by them is a critical point of the problem under mild assumptions. Moreover, we establish global convergence of the sequence generated by PGSA or PGSA ML and analyze its convergence rate, by further assuming the local Lipschitz continuity of the nonsmooth function in the numerator, the smoothness of the denominator, and the Kurdyka--Łojasiewicz (KL) property of the objective. The proposed algorithms are applied to the sparse generalized eigenvalue problem associated with a pair of symmetric positive semidefinite matrices, and the corresponding convergence results are obtained according to their general convergence theorems. We perform some preliminary numerical experiments to demonstrate the efficiency of the proposed algorithms. [ABSTRACT FROM AUTHOR]

In this paper we propose an efficient method for solving the spherically constrained homogeneous polynomial optimization problem. The new approach has the following three main ingredients. First, we establish a block coordinate descent type search method for nonlinear optimization, with the novelty being that we accept only a block update that achieves the maximum improvement, hence the name of our new search method: maximum block improvement (MBI). Convergence of the sequence produced by the MBI method to a stationary point is proved. Second, we establish that maximizing a homogeneous polynomial over a sphere is equivalent to its tensor relaxation problem; thus we can maximize a homogeneous polynomial function over a sphere by its tensor relaxation via the MBI approach. Third, we propose a scheme to reach a KKT point of the polynomial optimization, provided that a stationary solution for the relaxed tensor problem is available. Numerical experiments have shown that our new method works very efficiently: for a majority of the test instances that we have experimented with, the method finds the global optimal solution at a low computational cost. [ABSTRACT FROM AUTHOR]

In this paper, we focus on the problem of minimizing a nonconvex function over the unit simplex. We analyze two well-known and widely used variants of the Frank--Wolfe algorithm and first prove global convergence of the iterates to stationary points, both when using exact and Armijo line search. Then we show that the algorithms identify the support in a finite number of iterations (the identification result does not hold for the classic Frank--Wolfe algorithm). This, to the best of our knowledge, is the first time a manifold identification property has been shown for such a class of methods. [ABSTRACT FROM AUTHOR]

This paper studies bilevel polynomial optimization. We propose a method to solve it globally by using polynomial optimization relaxations. Each relaxation is obtained from the Karush-Kuhn-Tucker (KKT) conditions for the lower level optimization and the exchange technique for semi-infinite programming. For KKT conditions, Lagrange multipliers are represented as polynomial or rational functions. The Moment-sum-of-squares relaxations are used to solve the polynomial optimization relaxations. Under some general assumptions, we prove the convergence of the algorithm for solving bilevel polynomial optimization problems. Numerical experiments are presented to show the efficiency of the method. [ABSTRACT FROM AUTHOR]

Multiple-input multiple-output (MIMO) detection is a fundamental problem in wireless communications and it is strongly NP-hard in general. Massive MIMO has been recognized as a key technology in fifth generation (5G) and beyond communication networks, which on one hand can significantly improve the communication performance and on the other hand poses new challenges of solving the corresponding optimization problems due to the large problem size. While various efficient algorithms such as semidefinite relaxation (SDR) based approaches have been proposed for solving the small-scale MIMO detection problem, they are not suitable to solve the large-scale MIMO detection problem due to their high computational complexities. In this paper, we propose an efficient quadratic programming (QP) relaxation based algorithm for solving the large-scale MIMO detection problem. In particular, we first reformulate the MIMO detection problem as a sparse QP problem. By dropping the sparse constraint, the resulting relaxation problem shares the same global minimizer with the sparse QP problem. In sharp contrast to the SDRs for the MIMO detection problem, our relaxation does not contain any (positive semidefinite) matrix variable and the numbers of variables and constraints in our relaxation are significantly less than those in the SDRs, which makes it particularly suitable for the large-scale problem. Then we propose a projected Newton based quadratic penalty method to solve the relaxation problem, which is guaranteed to converge to the vector of transmitted signals under reasonable conditions. By extensive numerical experiments, when applied to solve small-scale problems, the proposed algorithm is demonstrated to be competitive with the state-of-the-art approaches in terms of detection accuracy and solution efficiency; when applied to solve large-scale problems, the proposed algorithm achieves better detection performance than a recently proposed generalized power method. [ABSTRACT FROM AUTHOR]

This paper aims at developing two versions of the generalized Newton method to compute local minimizers for nonsmooth problems of unconstrained and constrained optimization that satisfy an important stability property known as tilt stability. We start with unconstrained minimization of continuously differentiable cost functions having Lipschitzian gradients and suggest two second-order algorithms of Newton type: one involving coderivatives of Lipschitzian gradient mappings, and the other based on graphical derivatives of the latter. Then we proceed with the propagation of these algorithms to minimization of extended-real-valued prox-regular functions, while covering in this way problems of constrained optimization, by using Moreau envelopes. Employing advanced techniques of second-order variational analysis and characterizations of tilt stability allows us to establish the solvability of subproblems in both algorithms and to prove the Q-superlinear convergence of their iterations. [ABSTRACT FROM AUTHOR]

In this paper, we propose a primal-dual algorithm with a novel momentum term using the partial gradients of the coupling function that can be viewed as a generalization of the method proposed by Chambolle and Pock in 2016 to solve saddle point problems defined by a convex-concave function L(x,y)=f(x)+Φ(x,y)--h(y) with a general coupling term Φ(x,y) that is not assumed to be bilinear. Assuming ∇xΦ(*, y) is Lipschitz for any fixed y, and ∇yΦ(*, *) is Lipschitz, we show that the iterate sequence converges to a saddle point; and for any (x,y), we derive error bounds in terms of ... for the ergodic sequence ... In particular, we show O(1/k) rate when the problem is merely convex in x. Furthermore, assuming Φ(x,*) is linear for each fixed x and f is strongly convex, we obtain the ergodic convergence rate of O(1/k²) -- we are not aware of another single-loop method in the related literature achieving the same rate when Φ is not bilinear. Finally, we propose a backtracking technique which does not require the knowledge of Lipschitz constants while ensuring the same convergence results. We also consider convex optimization problems with nonlinear functional constraints and we show that using the backtracking scheme, the optimal convergence rate can be achieved even when the dual domain is unbounded. We tested our method against other state-of-the-art first-order algorithms and interior-point methods for solving quadratically constrained quadratic problems with synthetic data, the kernel matrix learning, and regression with fairness constraints arising in machine learning. [ABSTRACT FROM AUTHOR]

We consider a linear problem over a finite set of integer vectors and assume that there is a verification oracle, which is an algorithm being able to verify whether a given vector optimizes a given linear function over the feasible set. Given an initial solution, the algorithm proposed in this paper finds an optimal solution of the problem together with a path, in the 1-skeleton of the convex hull of the feasible set, from the initial solution to the optimal solution found. The length of this path is bounded by the sum of distinct values which can be taken by the components of feasible solutions, minus the dimension of the problem. In particular, in the case when the feasible set is a set of binary vectors, the length of the constructed path is bounded by the number of variables, independently of the objective function. [ABSTRACT FROM AUTHOR]

In this paper, we propose two novel non-stationary first-order primal-dual algorithms to solve non-smooth composite convex optimization problems. Unlike existing primal-dual schemes where the parameters are often fixed, our methods use predefined and dynamic sequences for parameters. We prove that our first algorithm can achieve an O(1/k) convergence rate on the primal-dual gap, and primal and dual objective residuals, where k is the iteration counter. Our rate is on the non-ergodic (i.e., the last iterate) sequence of the primal problem and on the ergodic (i.e., the averaging) sequence of the dual problem, which we call the semi-ergodic rate. By modifying the step-size update rule, this rate can be boosted even faster on the primal objective residual. When the problem is strongly convex, we develop a second primal-dual algorithm that exhibits an O(1/k²) convergence rate on the same three types of guarantees. Again by modifying the step-size update rule, this rate becomes faster on the primal objective residual. Our primal-dual algorithms are the first ones to achieve such fast convergence rate guarantees under mild assumptions compared to existing works, to the best of our knowledge. As byproducts, we apply our algorithms to solve constrained convex optimization problems and prove the same convergence rates on both the objective residuals and the feasibility violation. We still obtain at least O(1/k²) rates even when the problem is "semi-strongly" convex. We verify our theoretical results via two well-known numerical examples. [ABSTRACT FROM AUTHOR]

A quadratically constrained quadratic program (QCQP) is an optimization problem in which the objective function is a quadratic function and the feasible region is defined by quadratic constraints. Solving nonconvex QCQP to global optimality is a well-known NP-hard problem and a traditional approach is to use convex relaxations and branch-and-bound algorithms. This paper makes a contribution in this direction by showing that the exact convex hull of a general quadratic equation intersected with any bounded polyhedron is second-order cone representable. We present a simple constructive proof and some preliminary applications of this result. [ABSTRACT FROM AUTHOR]

In this paper we consider the unconstrained minimization problem of a smooth function in ℝn in a setting where only function evaluations are possible. We design a novel randomized derivative-free algorithm--the stochastic three points (STP) method-and analyze its iteration complexity. At each iteration, STP generates a random search direction according to a certain fixed probability law. Our assumptions on this law are very mild: roughly speaking, all laws which do not concentrate all measures on any halfspace passing through the origin will work. For instance, we allow for the uniform distribution on the sphere and also distributions that concentrate all measures on a positive spanning set. Although our approach is designed to not explicitly use derivatives, it covers some first order methods. For instance, if the probability law is chosen to be the Dirac distribution concentrated on the sign of the gradient, then STP recovers the signed gradient descent method. If the probability law is the uniform distribution on the coordinates of the gradient, then STP recovers the randomized coordinate descent method. The complexity of STP depends on the probability law via a simple characteristic closely related to the cosine measure which is used in the analysis of deterministic direct search (DDS) methods. Unlike in DDS, where O(n) (n is the dimension of x) function evaluations must be performed in each iteration in the worst case, our method only requires two new function evaluations per iteration. Consequently, while the complexity of DDS depends quadratically on n, our method depends linearly on n. [ABSTRACT FROM AUTHOR]

We study the iteration complexity of the optimistic gradient descent-ascent (OGDA) method and the extragradient (EG) method for finding a saddle point of a convex-concave unconstrained min-max problem. To do so, we first show that both OGDA and EG can be interpreted as approximate variants of the proximal point method. This is similar to the approach taken in (A. Nemirovski (2004), SIAM J. Optim., 15, pp. 229-251) which analyzes EG as an approximation of the "conceptual mirror prox." In this paper, we highlight how gradients used in OGDA and EG try to approximate the gradient of the proximal point method. We then exploit this interpretation to show that both algorithms produce iterates that remain within a bounded set. We further show that the primal-dual gap of the averaged iterates generated by both of these algorithms converge with a rate of O(1/k). Our theoretical analysis is of interest as it provides the first convergence rate estimate for OGDA in the general convex-concave setting. Moreover, it provides a simple convergence analysis for the EG algorithm in terms of function value without using a compactness assumption. [ABSTRACT FROM AUTHOR]

Stochastic dominance, a pairwise comparison between random variables, is an effective tool for expressing risk aversion in stochastic optimization. In this paper, we develop a family of primal-dual algorithms for optimization problems with stochastic dominance-constraints. First, we develop an offline primal-dual algorithm and bound its optimality gap as a function of the number of iterations. Then, we extend this algorithm to the online setting where only one random sample is given in each decision epoch. We give probabilistic bounds o n the optimality gap in this setting. This technique also yields an online algorithm for the stochastic dominance-constrained multiarmed bandit with partial feedback. The paper concludes by discussing a dual approach for a batch learning problem with robust stochastic dominance constraints. [ABSTRACT FROM AUTHOR]

This paper introduces the notion of a weighted complementarity problem (wCP), which consists of finding a pair of vectors (x, s) belonging to the intersection of a manifold with a cone such that their product in a certain algebra, x o s, equals a given weight vector w. When w is the zero vector, then wCP reduces to a complementarity problem. The motivation for introducing the more general notion of a wCP lies in the fact that several equilibrium problems in economics can be formulated in a natural way as wCPs. Moreover, those formulations lend themselves to the development of highly efficient algorithms for solving the corresponding equilibrium problems. For example, Fisher's competitive market equilibrium model can be formulated as a wCP that can be efficiently solved by interior-point methods. Moreover, it is shown that the quadratic programming and weighted centering problem, which generalizes the notion of a linear programming and weighted centering problem recently proposed by Anstreicher, can be formulated as a special linear monotone wCP. The main contribution of the paper is to introduce and analyze two interior-point methods for solving general monotone linear wCPs. [ABSTRACT FROM AUTHOR]

In this paper, we discuss a variational inequality problem with the gradient of a nonconvex objective function in which the constraint set composed of the absolute set and the subsidiary sets is not feasible. We formulate a compromise solution of the problem by using a solution to a variational inequality for the gradient over a subset of the absolute set with the elements closest to the subsidiary sets in terms of the norm. The objective of this paper is to enable us to discuss a network bandwidth allocation problem with a nonconcave utility function in which the compoundable constraints about the preferable transmission rate fall in the infeasible region. For such a bandwidth allocation, the ideal bandwidth must be quickly shared among traffic sources so as to satisfy the capacity constraints (absolute constraints) and all sources' demands regarding the transmission rate (subsidiary constraints) as much as possible. Iterative algorithms for the bandwidth allocation should thus not involve auxiliary optimization problems or complicated computations. Here, we devise an algorithm based on an iterative technique for triple-hierarchical constrained nonconvex optimization to achieve the optimal bandwidth allocation. Numerical examples for the bandwidth allocation demonstrate the effectiveness and convergence of the proposed algorithm. [ABSTRACT FROM AUTHOR]

This paper develops an optimization-based theory for the existence and uniqueness of equilibria of a noncooperative game wherein the selfish players' optimization problems are nonconvex and there are side constraints and an associated price clearance to be satisfied by the equilibria. A new concept of equilibrium for such a noncouvex game, which we term a "quasi-Nash equilibrium" (QNE), is introduced as a solution of the variation inequality (VI) obtained by aggregating the first order optimality conditions of the players' problems while retaining the convex constraints (if any) ill the defining set of the VI. Under a second-order sufficiency condition from nonlinear programming, a QNE becomes a local Nash equilibrium of the game. Uniqueness of a QNE is established using a degree-theoretic proof. Under a key bounded ness property of the Karush-Kuhn-Tucker multipliers of the nonconvex constraints and the positive definiteness of the Hessians of the players' Lagrangian functions, we establish the single-valuedness of the players' best-response maps, from which the existence of a Nash equilibrium (NE) of the nonconvex game follows. We also present a distributed algorithm for computing an NE of such a game and provide a matrix-theoretic condition for the convergence of the algorithm. An application is presented that pertains to a special multi-leader follower game wherein the nonconvexity is due to the followers' equilibrium conditions in the leaders' optimization problems. Another application to a cognitive radio paradigm in a signal processing game is described in detail in [G. Scutari and J.S. Pang, IEEE Trans. Inform. Theory, submitted; J.S. Pang and G. Scutari, Joint IEEE Trans. Signal Process, submitted]. [ABSTRACT FROM AUTHOR]

This paper introduces a novel algorithm to approximate the matrix with minimum nuclear norm among all matrices obeying a set of convex constraints. This problem may be understood as the convex relaxation of a rank minimization problem and arises in many important applications as in the task of recovering a large matrix from a small subset of its entries (the famous Netflix problem). Off-the-shelf algorithms such as interior point methods are not directly amenable to large problems of this kind with over a million unknown entries. This paper develops a simple first-order and easy-to-implement algorithm that is extremely efficient at addressing problems in which the optimal solution has low rank. The algorithm is iterative, produces a sequence of matrices {Xk,Yk}, and at each step mainly performs a soft-thresholding operation on the singular values of the matrix Yk. There are two remarkable features making this attractive for low-rank matrix completion problems. The first is that the soft-thresholding operation is applied to a sparse matrix; the second is that the rank of the iterates {Xk} is empirically nondecreasing. Both these facts allow the algorithm to make use of very minimal storage space and keep the computational cost of each iteration low. On the theoretical side, we provide a convergence analysis showing that the sequence of iterates converges. On the practical side, we provide numerical examples in which 1, 000 x 1, 000 matrices are recovered in less than a minute on a modest desktop computer. We also demonstrate that our approach is amenable to very large scale problems by recovering matrices of rank about 10 with nearly a billion unknowns from just about 0.4% of their sampled entries. Our methods are connected with the recent literature on linearized Bregman iterations for ℓ1 minimization, and we develop a framework in which one can understand these algorithms in terms of well-known Lagrange multiplier algorithms. [ABSTRACT FROM AUTHOR]

In [SIAM J. Optim., 16 (2006), pp. 1110-1136], Roos proved that the devised fullstep infeasible algorithm has O(n) worst-case iteration complexity. This complexity bound depends linearly on a parameter Κ̄(&zata;, which is proved to be less than √2n. Based on extensive computational evidence (hundreds of thousands of randomly generated problems), Roos conjectured that Κ̄(ζ) = 1 (Conjecture 5.1 in the above-mentioned paper), which would yield an O(√n) iteration full-Newton step infeasible interior-point algorithm. In this paper we present an example showing that Κ̄(ζ) is in the order of √n, the same order as that proved in Roos's original paper. In other words, the conjecture is false. [ABSTRACT FROM AUTHOR]

In this paper, we study active set identification results for the away-step Frank--Wolfe algorithm in different settings. We first prove a local identification property that we apply, in combination with a convergence hypothesis, to get an active set identification result. We then prove, for nonconvex objectives, a novel O() convergence rate result and active set identification for different step sizes (under suitable assumptions on the set of stationary points). By exploiting those results, we also give explicit active set complexity bounds for both strongly convex and nonconvex objectives. While we initially consider the probability simplex as feasible set, in an appendix we show how to adapt some of our results to generic polytopes. [ABSTRACT FROM AUTHOR]

This paper studies the class of two-stage stochastic programs with a linearly biparameterized recourse function defined by a convex quadratic program. A distinguishing feature of this new class of nonconvex stochastic programs is that the objective function in the second stage is linearly parameterized by the first-stage decision variable, in addition to the standard linear parameterization in the constraints. While a recent result has established that the resulting recourse function is of the difference-of-convex (dc) kind, the associated dc decomposition of the recourse function does not provide an easy way to compute a directional stationary solution of the two-stage stochastic program. Based on an implicit convex-concave property of the bi-parameterized recourse function, we introduce the concept of a generalized critical point of such a recourse function and provide a sufficient condition for such a point to be a directional stationary point of the stochastic program. We describe an iterative algorithm that combines regularization, convexification, and sampling and establish the subsequential convergence of the algorithm to a generalized critical point, with probability 1. [ABSTRACT FROM AUTHOR]

Undirected graphical models have been especially popular for learning the conditional independence structure among a large number of variables where the observations are drawn independently and identically from the same distribution. However, many modern statistical problems would involve categorical data or time-varying data, which might follow different but related underlying distributions. In order to learn a collection of related graphical models simultaneously, various joint graphical models inducing sparsity in graphs and similarity across graphs have been proposed. In this paper, we aim to propose an implementable proximal point dual Newton algorithm (PPDNA) for solving the group graphical Lasso model, which encourages a shared pattern of sparsity across graphs. Though the group graphical Lasso regularizer is nonpolyhedral, the asymptotic superlinear convergence of our proposed method PPDNA can be obtained by leveraging on the local Lipschitz continuity of the Karush--Kuhn--Tucker solution mapping associated with the group graphical Lasso model. A variety of numerical experiments on real data sets illustrates that the PPDNA for solving the group graphical Lasso model can be highly efficient and robust. [ABSTRACT FROM AUTHOR]

In this paper, we provide the first provable linear-time (in terms of the number of nonzero entries of the input) algorithm for approximately solving the generalized trust region subproblem (GTRS) of minimizing a quadratic function over a quadratic constraint under some regularity condition. Our algorithm is motivated by and extends a recent linear-time algorithm for the trust region subproblem by Hazan and Koren [Math. Program., 158 (2016), pp. 363–381]. However, due to the nonconvexity and noncompactness of the feasible region, such an extension is nontrivial. Our main contribution is to demonstrate that under some regularity condition, the optimal solution is in a compact and convex set and lower and upper bounds of the optimal value can be computed in linear time. Using these properties, we develop a linear-time algorithm for the GTRS. [ABSTRACT FROM AUTHOR]