- Reimplementing the destructor using BFS avoids the problem of possible stack overflow in the destructor.
- Added an experimental version (
lambda_exp) designed to avoid double-counting the same part of the equation during a run (there may be unknown problems).