学术翻译：固定采样伪梯度数量导致算法线性收敛至最优解邻域

The academic translation is as follows:

If the number of samples for pseudo-gradient is fixed, it will no longer guarantee the progressive reduction of uncertainty, and thus cannot provide algorithmic convergence in the expected sense. The following inference shows that when algorithm 1 uses a constant sampling capacity, it can only achieve linear convergence to the neighborhood of the optimal solution, namely 'F Nash equilibrium'. It can be seen that the boundary D depends on network structure, batch size, step size, and problem parameters η, L, ν.