fmincon is an iterative process. There is a built in limit on the number of steps that it will take. Sometimes that is not enough to find the minimum to within flatness criteria. You can increase the amount of searching.
https://www.mathworks.com/help/optim/ug/current-and-legacy-option-name-tables.html
optimset or optimoptions with appropriate option name and pass that to fmincon
Note: one of the ways that fmincon can run out of steps is if it is following an asymptotic curve. When that happens it is not uncommon that the region is exploring is quite a distance from the actual minimum. fmincon just sees that the value keeps decreasing (asymptotic), and does not know that it could do better by looking in a different area. Therefore you do always want to let it run for thousands of iterations.
