Can you please elaborate on the suggested fix? it would be great to provide a code snippet that exhibits the problem.

Let's take the code of Composition Function 3 as an example, which has F13, F14, F16, F18 with different dimensions. This is the original code:

global initial_flag
persistent func_num func o sigma lambda bias M

[ps,D] = size(x);
func_num = 6;
lb = -5; ub = 5;
if initial_flag==0 % then load optima.mat and generate the composition function
...
end

We just want to only load optima.mat for the first time and directly use the persistent variables func_num, func, o, sigma ,lambda, bias, M in the next calls. This works fine during running F13 (2D). But when changing to F14 (3D), the dimension of input x changed. So there will be a mismatch of the dimension of the persistent variable o .

To address this issue, I add two persistent variables initial_flag and dim and delete the global variable initial_flag. In this way, at every call we not only judge if the initial_flag==0, but also judge if the dimension of the input x has changed. Therefore, when F13 (2D) changes to F14 (3D), we will reload optima.mat and generate the new compositional function.