Summary
RC_TPI is a single scalar compared against the resource-constraint error at every period with np.any, so there is no way to exempt the terminal period without simultaneously waiving the check across the whole path.
https://github.com/PSLmodels/OG-Core/blob/master/ogcore/TPI.py#L1839
if (np.any(np.absolute(RC_error) >= p.RC_TPI)) and ENFORCE_SOLUTION_CHECKS:
raise RuntimeError("Transition path equlibrium not found " + "(RC_error)")Why this matters
The last period of a truncated transition path is not a true steady state, so RC_error[T-1] is routinely larger than the interior periods by orders of magnitude. Country packages that hit this have no expressive option other than raising the global tolerance.
A concrete example downstream: OG-UK sets p.RC_TPI = 0.2 with an in-code comment stating that only t = T-1 fails and "all other periods are well within 1e-4" (oguk/api.py:979-983). The effect is that a mid-path violation of up to 0.2 — the same order as the terminal error the setting was written to route around — now passes silently on every transition run.
Suggested change
Check the interior and terminal periods separately, e.g.
- compare
RC_error[:-1] against a tight tolerance (RC_TPI, default unchanged), and
- handle
RC_error[-1] under a separate, looser RC_TPI_terminal parameter (or exclude it and note the truncation in the docs).
This is backwards compatible for anyone currently passing the check, and it restores the diagnostic for the interior path.
Happy to open a PR if the approach looks right.
Summary
RC_TPIis a single scalar compared against the resource-constraint error at every period withnp.any, so there is no way to exempt the terminal period without simultaneously waiving the check across the whole path.https://github.com/PSLmodels/OG-Core/blob/master/ogcore/TPI.py#L1839
Why this matters
The last period of a truncated transition path is not a true steady state, so
RC_error[T-1]is routinely larger than the interior periods by orders of magnitude. Country packages that hit this have no expressive option other than raising the global tolerance.A concrete example downstream: OG-UK sets
p.RC_TPI = 0.2with an in-code comment stating that onlyt = T-1fails and "all other periods are well within 1e-4" (oguk/api.py:979-983). The effect is that a mid-path violation of up to 0.2 — the same order as the terminal error the setting was written to route around — now passes silently on every transition run.Suggested change
Check the interior and terminal periods separately, e.g.
RC_error[:-1]against a tight tolerance (RC_TPI, default unchanged), andRC_error[-1]under a separate, looserRC_TPI_terminalparameter (or exclude it and note the truncation in the docs).This is backwards compatible for anyone currently passing the check, and it restores the diagnostic for the interior path.
Happy to open a PR if the approach looks right.