Singular value decomposition (SVD) provides a simple and straightforwa
rd way of detecting univariant reactions in multicomponent mineral ass
emblages. As originally developed by Fisher (1989), however, the metho
d had the disadvantage that it created model mineral compositions that
were not strictly stoichiometric. Model assemblages of stoichiometric
minerals can be produced by rewriting the mineral composition matrix
in terms of additive and exchange components and then weighting the ad
ditive components more heavily than exchange components during SVD ana
lysis. Prior to SVD analysis, it is generally necessary to use least-s
quares methods to fit mineral analyses to stoichiometric models based
on exchange and additive components; least-squares fitting of this sor
t may be useful for evaluating mineral analyses even when SVD analysis
is not contemplated.