With the default pl_linearize=1, AMPL supplies both an SOS2 as well as its linearization. MP accepts SOS2 and either passes it to the solver, or provides its own linearization. Thus, AMPL's linearization should ideally be deleted (although it's simple and might be more numrically robust). However, MP's linearization assumes sum(lambda)=1 which is not the case in AMPL's encoding of PL expressions.
With the default pl_linearize=1, AMPL supplies both an SOS2 as well as its linearization. MP accepts SOS2 and either passes it to the solver, or provides its own linearization. Thus, AMPL's linearization should ideally be deleted (although it's simple and might be more numrically robust). However, MP's linearization assumes sum(lambda)=1 which is not the case in AMPL's encoding of PL expressions.
See also ampl/escrow/issues/93.