Smoother crash after Bayesian est. of non-stationary model

This model has one shock: technology which is a random walk with drift. Dynare does Bayesian estimation including plots of posteriors for parameters. When trying to produce a plot of the smoothed disturbance the KF/smoother crashes. The stationary version of this model does not crash. What is going on? I am doing something wrong, or is there a problem with Dynare? I can provide mod files for data simulation if desired.
BNTnlc_nonS_PP_steadystate.m (9.54 KB)
NLc88Q109Q3.m (625 Bytes)
BNTnlc_nonS_PP.mod (2.38 KB)

The problem is the result of a variable in the model with a steady state value of 0: the variable is divai. Taking a Taylor series expansion around the log of zero may well explain the problem. At any rate, if I substitute divai out of the model, the problem disappears.

My original steady state routine is very silly. I have a much better one attached to this message here.
BNTnlc_nonS_PP_steadystate.m (2.77 KB)