The stepsize before the solver exists usually becomes small. So if you choose "tspan" in the call to the ode integrator as a 2-element vector, many solutions are collected in this region, and you can examine the curves with the plot command.
Another alternative is to choose many output times in advance so that you can be sure to capture a point in the critical region.
Or you can repeat the run up to a time slightly smaller than the time when the solver got into trouble.
