We have shown how, working in a field-theoretic framework, one can der
ive expressions for the even moments of the valence parts of g(1,2)(x)
. These expressions cannot be written as matrix elements of local oper
ators and do not coincide with the analytic continuation to n = even i
nteger of the OPE results. Just as for the OPE one can in some cases a
rgue that the hadronic matrix elements should be small, leading to app
roximate sum rules for the moments of the valence parts of g(1,2)(x).
But, most importantly, for the case n=2 we have proved rigorously that
the hadronic matrix element vanishes, yielding an exact sum rule. We
have argued that the convergence properties of this sum rule are good
and are a further test of QCD.