Many ordinary differential equation (ODE) and differential algebraic e
quation (DAE) codes terminate the nonlinear iteration for the correcto
r equation when the difference between successive iterates (the step)
is sufficiently small. This termination criterion avoids the expense o
f evaluating the nonlinear residual at the final iterate. Similarly, J
acobian information is not usually computed at every time step but onl
y when certain tests indicate that the cost of a new Jacobian is justi
fied by the improved performance in the nonlinear iteration. In this p
aper, we show how an out-of-date Jacobian coupled with moderate ill co
nditioning can lead to premature termination of the corrector iteratio
n and suggest ways in which this situation can be detected and remedie
d. As an example, we consider the method of lines (MOL) solution of Ri
chards' equation (RE), which models ow through variably saturated poro
us media. When the solution to this problem has a sharp moving front,
and the Jacobian is even slightly ill conditioned, the corrector itera
tion used in many integrators can terminate prematurely, leading to in
correct results. While this problem can be solved by tightening the to
lerances for the solvers used in the temporal integration, it is more
efficient to modify the termination criteria of the nonlinear solver a
nd/or recompute the Jacobian more frequently. Of these two, recomputat
ion of the Jacobian is the more important. We propose a criterion base
d on an estimate of the norm of the time derivative of the Jacobian fo
r recomputation of the Jacobian and a second criterion based on a cond
ition estimate for tightening of the termination criteria of the nonli
near solver.