After running some test cases, it looks pretty clear that if your model includes DMIG entries for hardware, that hardware is not included in either the Grid Point weight generator summary OR the modal solution when attempting a differential stiffening (STATSUB routine) solution. Run a simple cantilevered beam model, then replace the second half of it with a DMIG representation of the mass and stiffness. Each model will give the exact same answers for a normal modes solution. But now if you apply a distributed load to only the first physical element and set that load to 0.0 lb/in you should once again get exactly the same results. For the model with only CBEAM elements you will. However, for the model where half is a DMIG, you will get the exact results for only a cantilver beam of the half-length (the length still represented by CBEAMs). Somewhere in the STATSUB routine NX is losing the DMIG entries. Anyone else seen this?