In Figure 4, two groups of results at the steady state according to different initial u are shown. With exactly the same parameter, as we choose 0 as the initial guess of u, the small size jump (dip) is removed, but it is still there when we use f as the initial u. In fact, this phenomenon is intrinsic to our model. As the initial u is chosen to be f, to remove the small size jump, u will have to increase in the interval. Since u is updated according to (29), if u increased in the interval, it would have to increase gradually. However, this gradual increase cannot happen. As suggested by Theorem 1, the gradual increase will not change the value of the first term in our model (6) too much, but will make the fitting term larger. Therefore, the small size jump will be preserved. The above statements also provide an explanation for the case when the initial u is set to be 0. Based on the observation, we always choose 0 as the initial guess of u in our numerical experiments.
The above example also suggests that our proposed functional is not a convex functional, which is different from the ROF model.