First-hand information for everyone
Researchers prove that the convex lifting of the Mumford–Shah functional — a tool used to relax a tough image‑processing problem — is not exact in two‑dimensional crack‑tip problems. By imposing the crack‑tip Dirichlet data and building an explicit lifted competitor using a radial cut‑off function and a layer‑cake representation, they show the lifted energy is strictly lower than the energy of the crack‑tip subgraph. Because the crack‑tip is already the minimiser of the original functional, this result establishes a strict gap between the two infimum values. Consequently the crack‑tip cannot be calibrated by the approaches of Alberti, Bouchitté and Dal Maso.
The mathematical construction gives a quantitative lower bound for the gap that is positive for all sufficiently large scale parameters L, but the precise magnitude remains unknown. The bound is exponentially small – of the order e^{‑2L} – which explains why numerical experiments based on the convex lifting have never convincingly detected the gap. The study therefore highlights a limitation of current convex relaxation techniques and suggests that careful analytical arguments are needed to uncover such subtle discrepancies.
These findings refine our understanding of how convex relaxations behave in planar fracture problems and indicate that even in simple two‑dimensional settings the relaxation may miss the true minimal energy for special configurations like the crack‑tip.