Background
meOnly() folds the forcing into an augmented matrix column
(src/expm.cpp:272-296), exactly as it does a constant infusion rate. So f is
treated as constant across a substep: exact for a genuinely constant forcing,
first-order in hmax otherwise. That is why hmax still governs accuracy even
once the iteration has converged.
Change
Evaluate f at both substep ends and integrate the linear interpolant exactly
via a second augmented column (a phi_2 term), making the substep second-order in
hmax.
Status
Deferred by decision. The iteration lands first with constant-per-substep
forcing plus a test pinning the observed order of convergence in hmax; this
issue is what that test would then show improving. Not scheduled.
Blocked by #1185.
Background
meOnly()folds the forcing into an augmented matrix column(
src/expm.cpp:272-296), exactly as it does a constant infusion rate. Sofistreated as constant across a substep: exact for a genuinely constant forcing,
first-order in
hmaxotherwise. That is whyhmaxstill governs accuracy evenonce the iteration has converged.
Change
Evaluate
fat both substep ends and integrate the linear interpolant exactlyvia a second augmented column (a phi_2 term), making the substep second-order in
hmax.Status
Deferred by decision. The iteration lands first with constant-per-substep
forcing plus a test pinning the observed order of convergence in
hmax; thisissue is what that test would then show improving. Not scheduled.
Blocked by #1185.