Why is "closeloop" on a NARNET using empty cell initial delay states?

Hi,

My question is a bit more theoretical than practical.

I refer to the code in help narnet: https://www.mathworks.com/help/nnet/ref/narnet.html

When closing the network with closeloop, the following code is used and y2 gives the predicted values for the next 20 time samples:

[netc,Xic,Aic] = closeloop(net,Xf,Af);
view(netc);
y2 = netc(cell(0,20),Xic,Aic);

As I write this code in matlab, my final input delays Xf becomes Xic after closing the loop and therefore my understanding is that the values in Xic should be my initial input delays when predicting the next 20 time values with netc. By definition of the time series neural network, it uses the previous inputs to estimate future outputs therefore any predictive task should use 2 initial values to predict the next 20 ones.

However, the help narnet code outputs Xic as an empty cell 0x0

Xic = 

       {}

So, if I look again at the feeding parameters to the following line:

y2 = netc(cell(0,20),Xic,Aic);

Two arguments are empty and contain no information, and the only variable containing the information is Aic which is the initial layers state.

So, should I use the network to predict any future values with any set of data, will the information be stored in Aic? I would believe Aic is intrinsic to the network, and Xic should contain the information? Therefore why is Xic empty and contains no information?

Note I understand the netc view shows the network diagram and it contains no input - so again this get me confused as therefore the network is self-deterministic and would output the same values at all times? Otherwise, what should I input in Aic?

Many thanks for your help, in so I can sleep at night again. And my prior apologies if my question is stupid - I have been reading a fair bit on matlab information and I couldn't find the answer yet.