Some remarks on global asymptotic stability of neural networks with constant time delay (Q2471158)

The paper deals with the following system being referred to as delay neural network (DNN) with constant delay: \[ {dx(t)\over dt}=-A x(t)+W_0 f(x(t))+W_1 f(x(t-\tau)), \] where \(x=(x_1,\dots,x_n)^T\) is the state vector, \(f(x)=(f_1(x_1),\dots,f_n(x_n))\) and \(f_j(x_j)=g_j(x_j+u^*_j)-g_j(x_j)\) and \(f_j(0)=0\) for all \(j\). The function \(f_j(\cdot)\) is said to satisfy condition (H), when \[ 0\leq {f_j(\xi_1)-f_j(\xi_2)\over \xi_1-\xi_2}\leq\sigma_j\,,\quad j=1,\dots,n. \] The following result is proved by \textit{S. Arik} [Phys. Lett. A 311, No. 6, 504--511 (2003; Zbl 1098.92501)]: Theorem. The origin of the considered DNN with constant delay is the unique equilibrium point and it is globally asymptotically stable if the condition (H) is satisfied and there exist a positive definite matrix \(P=P^T\) and a positive diagonal matrix \(D=\operatorname{diag}(d_j)\) such that \[ 2DA\Sigma^{-1}-DW_0-W_0^TD-W_1^TPW_1-DP^{-1}D>0, \] where \(\Sigma=\operatorname{diag}(\sigma_i)\) is a positive diagonal matrix. The present paper provides an alternative proof of this result.