Expected visits¤

For a square transition, expected_visits() computes the expected number of visits to each segregating state, starting from one mutant copy. It includes generation zero and stops at the first endpoint hit.

import jax
import drifft

jax.config.update("jax_enable_x64", True)
transition = drifft.Transition(drifft.FFT(64), drifft.Underdominance(0.01))
result = transition.expected_visits()
if not bool(result.converged):
    raise RuntimeError(f"{result.status}: residual={result.residual_norm}")
visits = result.visits

Solver settings and limitations¤

Choose solver tolerances for the residual you need, then check converged before interpreting the visit vector. Long absorption times can amplify errors in the underlying transition even when the solve converges.

Choose solver="gmres" (default) or "bicgstab"; adjust rtol, atol, and iteration limits as needed. The transition API lists defaults and available settings.

result.converged requires a successful Lineax status, finite output, and a fresh infinity-norm residual no larger than atol+rtol. Always check it. result.stats preserves the installed Lineax solver's iteration counters; GMRES num_steps is not a count of individual matvecs. Failures retain the candidate iterate for diagnosis, but it must not be treated as a valid occupation vector.

Compare visits and total absorption time under tighter backend settings, using the accuracy guide. The model chapter derives the occupation equation and its transience requirement.