I have been reading a little bit about structural zeros and chi-square goodness of fit tests.

I've not been able to find a clear answer as to whether 'structural zeros' are only relevant in the observed data... In other words, is it a violation of the assumptions of the chi-square to have 0 for an expected value?

Example:

x: 1929, 3283, 293, 2015

p: 0.2, 0.6, 0, 0.2

Where x are the observed counts and p the expected proportions.

Is that a legitimate test, or would I have to eliminate the 0, 293 value, ignoring the observed count for the unexpected variable?

Example:

x: 1929, 3283, 2015

p: 0.2, 0.6, 0.2

Thank you