Your software calculates the unit operations sequentially. You have a recycle in your flowsheet so the unit ops are calculated more than once. The 'tear' is a stream that is monitored each calculation cycle for differences. When the tear streams have results that are 'close enough' to the previous cycle, then the software determines that the solution 'converged' successfully. The tear tolerance sets how close is 'close enough'. You can find the Tear Convergence Sheet in the Convergence group of your Simulation parameters.
The artificial intelligence in Aspen has made the suggestion that your tear tolerance is too loose. This is probably NOT the real cause of your simulation problem. Based on your question, I suggest you look closely at how you specified your purge stream.
For troubleshooting your simulation, you may want to observe the calculations more closely. Put in calculation breaks or step through the calcs and observe the intermediate results. Are the results moving in the right direction? Are they reasonable? If not, figure out how to re-specify the model to do what you want.