First the step in ode45 has little meaning since the method is adaptive and will choose the appropriate time step based on a predictor corrector scheme. The 0.01 step that you might specify through the time vector input is merely a set of time values for which you want the value to be computed. Internally the function might use very different time steps.
Second, since the method adapts the time step, the error that you get simply tells you that in order to satisfy the error tolerance it had to use an extremely small time step which seems unreasonable (i.e. a time step of 1e-15 is simply not feasible if you integrate over several units of time). For you information in molecular dynamics such small tie step can be used but the time integration usually only goes for a few nanoseconds and still it might take weeks to compute.
The reason for the small time step is probably that your system is stiff. In mechanical systems, this would correspond to a very stiff spring, in chemical reactions to a very fast reaction. This might also come from a constraint in the system that you are enforcing using a penalization method.
My advice: use another ODE solver (ODE15s, ODE23S) specifically designed to handle stiff problems. You might also play a bit with tolerances and other parameters you can manipulate through "odeset" and explore methods that also use the jacobian of the "ode function". If you still have problems with the stiff integration methods, you might want to check your program for validity. Each time I have had this type of problem and i could not get rid of it, it turned out I had a bug in my program or i wan considering unphysical parameters.
Hope this helps.
A.
