# Question about the visible and hidden neurons in neural networks methods

My problem is the following : I found the ground state energy (for the Ising model) with neural networks (more specifically RBM). I reproduced the same result but by increasing every time the ratio $$=\frac{Hidden Neurons}{Visible Neurons}$$ (for example, for a ratio = 1, =3, =5, = 9). I did that because I wanted to show that with a higher ratio, it converged more rapidly to the solution (ground state energy). (I studied the convergence by making the graph of the variance of the energy). But it is not the case at all. It is every time a similar convergence whatever is the ratio. Why ? Of course the runtime increased exponentially (and so, I waited so much time for this deception : not seeing a more rapid convergence whatever the ratio was). I really would like at least to understand why it does not increase the convergence more rapidly when we increase the ratio (ie the number of hidden and visible neurons) ?