A growth estimate for the planar Mumford–Shah minimizers at a tip point: An alternative proof of David–Léger
Yi Ru-Ya Zhang · 2025
Read A growth estimate for the planar Mumford–Shah minimizers at a tip point: An alternative proof of David–Léger by Yi Ru-Ya Zhang (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-15605.txt.
Introduction
Yi Ru-Ya Zhang published a research paper in 2025 giving a new proof of a growth bound for the Mumford-Shah problem, a math model of image splitting and crack growth. The bound says how fast a best-fit solution can change near the tip of a crack. An older paper proved a global version of the same bound.
Zhang's route rests on a two-case split and on the shape of the region left once the crack set is removed. It differs from both earlier proofs of the estimate. The new argument also yields the bound in a purely local form. Such estimates anchor the wider theory of free-discontinuity problems, where cracks and edges are part of the unknown.
Catalog record
| CREATOR | Yi Ru-Ya Zhang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-15605
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A growth estimate for the planar Mumford–Shah minimizers at a tip point: An alternative proof of David–Léger Thanks: The author is funded by National Key R&D Program of China (Grant No. 2021YFA1003100), NSFC grant No. 12288201 and the Chinese Academy of Sciences. Yi Ru-Ya Zhang Abstract . Let Ω ⊂ ℝ 2 \Omega\subset\mathbb{R}^{2} be a bounded domain and u ∈ S B V ( Ω ) u\in SBV(\Omega) be a local minimizer of the Mumford–Shah problem in the plane, with 0 ∈ S ¯ u 0\in\overline{S}_{u} being a tip point and B 1 ⊂ Ω B_{1}\subset\Omega . Then there exist absolute constants C > 0 C>0 and 0 < r 0 < 1 0<r_{0}<1 such that | u ( x ) − u ( 0 ) | ≤ C r 1 2 for any x ∈ B r and 0 < r < r 0 . |u(x)-u(0)|\leq Cr^{\frac{1}{2}}\quad\text{ for any }\ x\in B_{r}\ \text{ and }\ 0<r<r_{0}. This estimate is a local version of the original one in [ 11 , Proposition 10.17] . Our result is based on a dichotomy and the John structure of Ω ∖ S ¯ u \Omega\setminus\overline{S}_{u} , different from the one by David–Léger [ 11 ] or Bonnet–David [ 8 , Lemma 21.3] . 2020 Mathematics Subject Classification 49Q20, 49N60 1. Introduction The Mumford–Shah functional, introduced by Mumford and Shah in [ 21 ] , is a well–known model in image processing. In their seminal paper [ 14 ] , De Giorgi, Carriero, and Leaci established the existence of minimizers for a weaker formulation of the Mumford–Shah problem through direct methods, drawing on a lower semicontinuity result by De Giorgi and Ambrosio [ 13 ] . To be more specific, for any bounded domain \Omega\subset\mathbb{R}^{n} , it was introduced in [ 13 ] a subspace of BV(\Omega) , denoted by SBV(\Omega) , in which the functions only has jump discontinuities (see Section 2 for more details). Then for a function u\in SBV(\Omega) , its \lambda -Mumford–Shah energy on an open set \Omega\subseteq\mathbb{R}^{n} is defined by MS_{\lambda}(u,\,\Omega):=\int_{\Omega}{\left|Du\right|}^{2}\,dx+\lambda\mathcal{H}^{n-1}\left(S_{u}\cap\Omega\right), where \lambda>0 and S_{u}\subset\Omega is the set of discontinuity points of u . A function u\in SBV_{\mathop{\mathrm{}}{\,loc\,}}(\Omega) is a local \lambda -minimizer if for any x\in\Omega with B_{r}(x)\subset\Omega,\,r>0 and every open set U\subset\subset\Omega\cap B_{r}(x) , we have MS_{\lambda}(u,\,U)<\infty together with MS_{\lambda}(u,\,U)\leq MS_{\lambda}(v,\,U), whenever \{u\neq v\}\subset\subset U. The Euler–Lagrange equation [ 4 , Theorem 7.35] for a local minimizer is that, for any \eta\in C_{0}^{1}(\Omega;\,\mathbb{R}^{n}), \int_{\Omega\setminus\overline{S}_{u}}|Du|^{2}{\rm div}\eta-2\langle\nabla u,\nabla u\cdot\nabla\eta\rangle\,dx+\lambda\int_{S_{u}}{\rm div}_{\tau}\eta\,d\mathcal{H}^{n-1}=0, (1.1) where {\rm div}_{\tau} denotes the tangential divergence. This in particular implies that u is harmonic in \Omega\setminus\overline{S}_{u} and satisfies the zero Neumann boundary value condition on both sides of \overline{S}_{u} ; see [ 4 , (7.42)] . Moreover the (weak) mean curvature of \overline{S}_{u} equals to the jump of the gradient [|Du|^{2}]^{\pm} , according to [ 4 , Theorem 7.38] . Ambrosio, Fusco and Pallara proved in [ 1 , 2 ] that, when \Omega\subset\mathbb{R}^{n} and u\in SBV(\Omega) is a local Mumford–Shah minimizer, there exists a subset \Sigma\subset\overline{S}_{u} , which is relatively closed and \mathcal{H}^{n-1}(\Sigma)=0 , such that the set \overline{S}_{u}\setminus\Sigma is the union of C^{1,\,\frac{1}{4}} -hypersurfaces. Moreover, both u and Du have a Hölder continuous extension to \overline{S}_{u}\setminus\Sigma . For more details, see also [ 4 , Theorem 8.1] , along with the survey [ 18 ] , and [ 3 , 17 , 20 ] for more recent results. Nevertheless, much more is known about the planar case. In particular, in an earlier result [ 7 ] , Bonnet proved that an isolated connected component of \overline{S}_{u} is a finite union of C^{1,\,1} -arcs. His result is based on the so–called Bonnet’s monotonicity formula, which is applicable when the discontinuity set in the plane is connected. Later, David [ 10 ] demonstrated a version of \epsilon -regularity for the minimizers, and many additional results regarding Mumford–Shah local minimizers in the plane were established by Bonnet and David in [ 8 ] . In particular, in the monograph [ 12 , Theorem 69.29] , David proved that if \overline{S}_{u} at a tip point x\in\overline{S}_{u}\setminus{S}_{u} with B(x,\,2)=:B_{2}(x)\subset\Omega , is sufficiently close to a single radius in the Hausdorff distance within (which implies \overline{S}_{u} is connected in B_{1}(x) [ 12 , Lemma 69.8] ), then \overline{S}_{u} is locally C^{1,\,1}_{\mathop{\mathrm{}}{\,loc\,}} in B_{1}(x) . Recently, this result was independently improved by [ 5 ] and [ 16 ] , showing that one actually obtains C^{1,\,1} (and even C^{2,\,\alpha} )-regularity up to the end point of \overline{S}_{u} , again under the assumption that \overline{S}_{u}\cap B_{2}(x) is sufficiently close to a single radius in the Hausdorff distance. The planar results mentioned above generally rely on the a priori assumption that \overline{S}_{u} is connected. In this paper, we examine the growth of u near a tip point without assuming the connectedness of the discontinuity set. This estimate serves as a localized version of the original estimate for global minimizers presented in [ 11 , Proposition 10.17] , via a completely different argument; readers may also refer to [ 15 , Proposition 4.6.1] . Theorem 1.1 . Let \Omega\subset\mathbb{R}^{2} be a bounded domain, u\in SBV(\Omega) be a local minimizer for the Mumford–Shah problem, and 0\in\Omega be a tip point, i.e. 0\in\overline{S}_{u}\setminus S_{u} . Suppose that B_{1}\subset\Omega . Then there exist absolute constants C>0 and 0<r_{0}<1 so that, for any 0<r<r_{0} , one has |u(x)-u(0)|\leq Cr^{\frac{1}{2}}\quad\text{ for any }\ x\in B_{r}. Our estimate is not straightforward because u\in SBV(\Omega) might not satisfy a Poincaré-type inequality, even if we can estimate the growth of the L^{2} -norm of Du within every disk (see Lemma 2.1 ). The discontinuity of u poses a significant challenge. It is worth noting that the harmonic conjugate v of u can be defined with Hölder continuity exponent \frac{1}{2} . This modulus of continuity follows directly from Lemma 2.1 as v\in W^{1,\,2}(\Omega); see e.g. [ 15 , Proposition 4.5.1] . Remark 1.2 . Some readers might compare our proof with the proof of Lemma 21.3 in [ 8 ] . However, it is important to note that they are not the same. Firstly, our estimate is directly based on an Ahlfors regularity lemma (see Lemma 2.1 below), which provides an L^{2} -estimate of |Du| . In contrast, the proof of Lemma 21.3 in [ 8 ] relies on an L^{\infty} -estimate, as seen in equation [ 8 , (21.1)] . Secondly, the proof of Bonnet–David assumes that x and 0 are in the same component of B\setminus\overline{S}_{u} (see [ 8 , (21.2)] ). However, this assumption is addressed in our work through Corollary 1.4 below. More precisely, our Lemma 3.1 and Lemma 3.2 may share some similarities with the proof of [ 8 , Lemma 21.3] . To some extent, Lemma 3.1 can be seen as an L^{2} -version of [ 8 , (21.14), Lemma 21.3] . Similarly, the finiteness of N in Lemma 3.2 is related to the finiteness of the set E in the proof of [ 8 , Lemma 21.3] . To handle the L^{2} -integral directly, one needs to use carrot John subdomains (or Boman chains). Therefore, we must demonstrate that one can cover B_{r}\setminus\overline{S}_{u} with a uniformly finite number