On the gauge-fixed BRST cohomology

A crucial property of the standard antifield-BRST cohomology at non-negative ghost number is that any cohomological class is completely determined by its antifield independent part. In particular, a BRST cocycle that vanishes when the antifields are set equal to zero is necessarily exact. This property, which follows from the standard theorems of homological perturbation theory, holds not only in the algebra of local functions, but also in the space of local functionals. The present paper stresses how important it is that the antifields in question be the usual antifields associated with the gauge invariant description. By means of explicit counterexamples drawn from the free Maxwell-Klein-Gordon system, we show that the property does not hold, in the case of local functionals, if one replaces the antifields of the gauge invariant description by new antifields adapted to the gauge fixation. In terms of these new antifields, it is not true that a local functional weakly annihilated by the gauge-fixed BRST generator determines a BRST cocycle; nor that a BRST cocycle which vanishes when the antifields are set equal to zero is necessarily exact.