Semismooth Newton Algorithm Research Articles

基于ROF模型的修正半光滑牛顿法 The Modified Semismooth Newton Algorithm Based on the ROF Model

本文基于ROF去噪模型的对偶算法提出一个修正的半光滑牛顿法。文中证明了该算法具有Q超线性收敛，同时指出选取适当的参数α可以提高数值计算效率。实验表明，建议的修正算法既能较好的复原图像，又具有较快的收敛速度。 In this paper, based on the dual algorithm of ROF model, we propose a modified semismooth Newton algorithm. Furthermore, we prove that the proposed algorithm converges Q-superlinearly, and also refer that this algorithm can improve the computational efficiency by choosing a suitable parameter α. The simulations show that the new modified algorithm can perfectly restore image and keep the faster conver-gence rate.

New second-order cone linear complementarity formulation and semi-smooth Newton algorithm for finite element analysis of 3D frictional contact problem

We propose a new second-order cone linear complementarity problem (SOCLCP) formulation for the numerical finite element analysis of three-dimensional (3D) frictional contact problems by the parametric variational principle. Specifically, we develop a regularization technique to resolve the multi-valued difficulty involved in the frictional contact law, and use a second-order cone complementarity condition to handle the regularized Coulomb friction law in contact analysis. The governing equations of the 3D frictional contact problem is represented by an SOCLCP via the parametric variational principle and the finite element method, which avoids the polyhedral approximation to the Coulomb friction cone so that the problem to be solved has much smaller size and the solution has better accuracy. In this paper, we reformulate the SOCLCP as a semi-smooth system of equations via a one-parametric class of second-order cone complementarity functions, and then apply the non-smooth Newton method for solving this system. Numerical results confirm the effectiveness and robustness of the SOCLCP approach developed.

Numerical simulation of two-dimensional Bingham fluid flow by semismooth Newton methods

This paper is devoted to the numerical simulation of two-dimensional stationary Bingham fluid flow by semismooth Newton methods. We analyze the modeling variational inequality of the second kind, considering both Dirichlet and stress-free boundary conditions. A family of Tikhonov regularized problems is proposed and the convergence of the regularized solutions to the original one is verified. By using Fenchel’s duality, optimality systems which characterize the original and regularized solutions are obtained. The regularized optimality systems are discretized using a finite element method with (cross-grid P 1 )– Q 0 elements for the velocity and pressure, respectively. A semismooth Newton algorithm is proposed in order to solve the discretized optimality systems. Using an additional relaxation, a descent direction is constructed from each semismooth Newton iteration. Local superlinear convergence of the method is also proved. Finally, we perform numerical experiments in order to investigate the behavior and efficiency of the method.

PDE-Constrained Optimization Subject to Pointwise Constraints on the Control, the State, and Its Derivative

A general Moreau–Yosida-based framework for minimization problems subject to partial differential equations and pointwise constraints on the control, the state, and its derivative is considered. A range space constraint qualification is used to argue existence of Lagrange multipliers and to derive a KKT-type system for characterizing first-order optimality of the unregularized problem. The theoretical framework is then used to develop a semismooth Newton algorithm in function space and to prove its locally superlinear convergence when solving the regularized problems. Further, for maintaining the local superlinear convergence in function space it is demonstrated that in some cases it might be necessary to add a lifting step to the Newton framework in order to bridge an $L^2$-$L^r$-norm gap, with $r>2$. The paper ends by a report on numerical tests.

Semismooth Newton support vector machine

Support vector machines can be posed as quadratic programming problems in a variety of ways. This paper investigates the 2-norm soft margin SVM with an additional quadratic penalty for the bias term that leads to a positive definite quadratic program in feature space only with the nonnegative constraint. An unconstrained programming problem is proposed as the Lagrangian dual of the quadratic programming for the linear classification problem. The resulted problem minimizes a differentiable convex piecewise quadratic function with lower dimensions in input space, and a Semismooth Newton algorithm is introduced to solve it quickly, then a Semismooth Newton Support Vector Machine (SNSVM) is presented. After the kernel matrix is factorized by the Cholesky factorization or the incomplete Cholesky factorization, the nonlinear kernel classification problem can also be solved by SNSVM, and the complexity of the algorithms has no apparent increase. Many numerical experiments demonstrate that our algorithm is comparable with the similar algorithms such as Lagrangian Support Vector Machines (LSVM) and Semismooth Support Vector Machines (SSVM).