Thank you Davide. Further probing shows it might be related to optimization : if I turn the vector generating loop optimisation away from symbolic to either numeric or none, the exceptions stop. Although the optimised and non-optimised variants do look to be very similar, so that might be spurious. This all originated in a larger document, with two same sized arrays generated by two similar function calls : one generated an exception when typing the vector name, the other didn't. N was originally 32, there are lots of values of N that don't cause an exception.