# American Institute of Mathematical Sciences

## Journals

JIMO
In this paper, we analyze the convergence properties of a nonlinear Lagrangian method based on Log-Sigmoid function for nonconvex semidefinite programming (NCSDP) problems. It is different from other convergence analysis, because the subproblem in our algorithm is inexactly solved. Under the constraint nondegeneracy condition, the strict complementarity condition and the second order sufficient conditions, it is obtained that the nonlinear Lagrangian algorithm proposed is locally convergent by choosing a proper stopping criterion and the error bound of solution is proportional to the penalty parameter when the penalty parameter is less than a threshold.
keywords: nonconvex semidefinite programming Löwner operator nonlinear Lagrangian method Log-Sigmoid function. Inexact solution
JIMO
We present a nonlinear Lagrangian method for nonconvex semidefinite programming. This nonlinear Lagrangian is generated by a Löwner operator associated with Log-Sigmoid function. Under a set of assumptions, we prove a convergence theorem, which shows that the nonlinear Lagrangian algorithm is locally convergent when the penalty parameter is less than a threshold and the error bound of the solution is proportional to the penalty parameter.
keywords: Löwner operator. nonlinear Lagrangian nonconvex semidefinite programming
JIMO
In this paper, we discuss a system of differential equations based on the projection operator for solving the box constrained variational inequality problems. The equilibrium solutions to the differential equation system are proved to be the solutions of the box constrained variational inequality problems. Two differential inclusion problems associated with the system of differential equations are introduced. It is proved that the equilibrium solution to the differential equation system is locally asymptotically stable by verifying the locally asymptotical stability of the equilibrium positions of the differential inclusion problems. An Euler discrete scheme with Armijo line search rule is introduced and its global convergence is demonstrated. The numerical experiments are reported to show that the Euler method is effective.
keywords: global convergence. differential inclusion asymptotical stability differential equations Box constrained variational inequality problem
DCDS-B
In this paper we propose a new type of $\theta$-scheme with four parameters ($\{\theta_i\}_{i=1}^4$) for solving the backward stochastic differential equation $-dy_t=f(t,y_t,z_t) dt - z_t dW_t$. We rigorously prove some error estimates for the proposed scheme, and in particular, we show that accuracy of the scheme can be high by choosing proper parameters. Various numerical examples are also presented to verify the theoretical results.
keywords: error estimate second order numerical tests. Backward stochastic differential equations $\theta$-scheme
DCDS-S
A global attractor in $L^2$ is shown for weakly dissipative $p$-Laplace equations on the entire Euclid space, where the weak dissipativeness means that the order of the source is lesser than $p-1$. Half-time decomposition and induction techniques are utilized to present the tail estimate outside a ball. It is also proved that the equations in both strongly and weakly dissipative cases possess an $(L^2,L^r)$-attractor for $r$ belonging to a special interval, which contains the critical exponent $p$. The obtained attractor is proved to be approximated by the corresponding attractor inside a ball in the sense of upper strictly and lower semicontinuity.
keywords: weak dissipativeness lower semicontinuity of attractors regularity of attractors Global attractors p-Laplace equations unbounded domains.
DCDS-B
A bi-spatial pullback attractor is obtained for non-autonomous and stochastic FitzHugh-Nagumo equations when the initial space is $L^2(\mathbb{R}^n)^2$ and the terminate space is $H^1(\mathbb{R}^n)\times L^2(\mathbb{R}^n)$. Some new techniques of positive and negative truncations are used to investigate the regularity of attractors for coupling equations and to correct the essential mistake in [T. Q. Bao, Discrete Cont. Dyn. Syst. 35(2015), 441-466]. A counterexample is given for an important lemma for $H^1$-attractor in several literatures included above.
keywords: stochastic FitzHugh-Nagumo equations bi-spatial attractors pullback attractors truncation method. Random dynamical system
DCDS-B

Backward compact dynamics is deduced for a non-autonomous Benjamin-Bona-Mahony (BBM) equation on an unbounded 3D-channel. A backward compact attractor is defined by a time-dependent family of backward compact, invariant and pullback attracting sets. The theoretical existence result for such an attractor is derived from the backward flattening property, and this property is proved to be equivalent to the backward asymptotic compactness in a uniformly convex Banach space. Finally, it is shown that the BBM equation has a backward compact attractor in a Sobolev space under some suitable assumptions, such as, backward translation boundedness and backward small-tail. Both spectrum decomposition and cut-off technique are used to give all required backward uniform estimates.

keywords: Benjamin-Bona-Mahony equations non-autonomous systems pullback attractors backward compact attractors backward flattening property increasing absorption spectrum decomposition unbounded channel
JIMO
In this paper, to design a piecewise linear contractual function, we consider to solve the single-level nonconvex programming with integral operator which is equivalent to the principal-agent bilevel programming model with continuous distribution. A modified constraint shifting homotopy method for solving the Karush-Kuhn-Tucker system of the discrete nonconvex programming is proposed and the global convergence from any initial point in shifted feasible set is proven under some mild conditions. A simple homotopy path tracing algorithm is given and is implemented in Matlab. For some typical risk averse utility functions and the typical distribution functions which simultaneously satisfy monotone likelihood ratio condition and convexity of the distribution function condition, some numerical tests to design the piecewise linear contract are done by our homotopy method as well as by using fmincon in Matlab, LOQO and MINOS and, as a comparison, the piecewise constant contracts are also designed by solving the single-level nonconvex programming which is equivalent to the principal-agent bilevel programming model with corresponding discrete distributions. Numerical tests show that: to design a piecewise linear contract, which is much better than a piecewise constant contract, it needs only to solve a much lower dimensional optimization problem and hence needs much less computing time. Numerical experiences also show that the modified constraint shifting homotopy method is feasible and robust.
keywords: nonconvex programming. homotopy method bilevel programming Principal-agent model piecewise linear contractual function
JIMO

Semi-supervised learning is an attractive method in classification problems when insufficient training information is available. In this investigation, a new semi-supervised classifier is proposed based on the concept of maximum vector-angular margin, (called S$^3$MAMC), the main goal of which is to find an optimal vector $c$ as close as possible to the center of the dataset consisting of both labeled samples and unlabeled samples. This makes S$^3$MAMC better generalization with smaller VC (Vapnik-Chervonenkis) dimension. However, S$^3$MAMC formulation is a non-convex model and therefore it is difficult to solve. Following that we present two optimization algorithms, mixed integer quadratic program (MIQP) and DC (difference of convex functions) program algorithms, to solve the S$^3$MAMC. Compared with the supervised learning methods, numerical experiments on real and synthetic databases demonstrate that the S$^3$MAMC can improve generalization when the labelled samples are relatively few. In addition, the S$^3$MAMC has competitive experiment results in generalization compared to the traditional semi-supervised classification methods.

keywords: Vector-angular margin semi-supervised learning support vector machine DC programming mixed integer programming
MBE

The circadian clock is a self-sustaining oscillator that has a period of about 24 hours at the molecular level. The oscillator is a transcription-translation feedback loop system composed of several genes. In this paper, a scalar nonlinear differential equation with two delays, modeling the transcriptional co-regulation in mammalian circadian clock, is proposed and analyzed. Sufficient conditions are established for the asymptotic stability of the unique nontrivial positive equilibrium point of the model by studying an exponential polynomial characteristic equation with delay-dependent coefficients. The existence of the Hopf bifurcations can be also obtained. Numerical simulations of the model with proper parameter values coincide with the theoretical result.

keywords: Circadian clock delay differential equations stability Hopf bifurcation