With larger numbers of DCR subfilters (e.g. 5) the dcrModel solution can sometimes oscillate between iterations of forward modeling. This results in premature termination of the modeling loop, because the convergence may improve by a large amount in one iteration, and by a very small amount in the next, triggering the convergence end condition. One solution is to set a small enough gain on the new model solutions so that modeling converges smoothly, but then more iterations are required to reach the same level of convergence. Instead, the gain should adapt to how well the model is improving compared to predictions.