Time sensitivity analysis - results variation
 

I have encountered an oddity when performing time sensitivity analysis and wanted to figure out the problem.

So, as usual with my new set up, I performed grid independence analysis and performed time sensitivity analysis. I'm currently performing constant time step transient simulation. So, I was comparing the results as time steps are reduced by two fold for each simulation.

Usually, as time step becomes smaller, results become similar to previously tested time step. (I.e. 1s simulation results vary by 10% to 0.5s simulation, 0.5s simulation results vary by 2% to 0.2s simulation results, etc.) Untill it is deemed that the time step independence has been reached when the difference is negligible.

However, I encountered a problem where 1s results vary negligibly to 0.2s results but 0.5s result is significantly different to the other two results. (Middle result is different to smaller+bigger time steps). 0.5s results exhibit different time characteristics compared to the other two. What could be the potential problem that could be causing this phenomenon?

simulate with 0.1 sec, if its same comparing to 0.2s, go on with 0.2sec