Abstract:We consider generalized solutions of the Perona-Malik equation in dimension one, defined as all possible limits of solutions to the semi-discrete approximation in which derivatives with respect to the space variable are replaced by difference quotients.
Our first result is a pathological example in which the initial data converge strictly as bounded variation functions, but strict convergence is not preserved for all positive times, and in particular many basic quantities, such as the supremum or the total variation, do not pass to the limit. Nevertheless, in our second result we show that all our generalized solutions satisfy some of the properties of classical smooth solutions, namely the maximum principle and the monotonicity of the total variation.
The verification of the counterexample relies on a comparison result with suitable sub/supersolutions. The monotonicity results are proved for a more general class of evolution curves, that we call $uv$-evolutions.
What problem does this paper attempt to address?
The problems that this paper attempts to solve mainly focus on the properties of the generalized solutions of the Perona - Malik equation, especially whether some basic properties of these solutions are preserved during the time - evolution process. Specifically, the paper focuses on the following aspects:
1. **Lack of strict convergence**: The paper provides a counter - example to show that even if the initial data strictly converges in the space of functions of bounded variation, this strict convergence is not necessarily preserved at positive times. This means that for positive time \(t > 0\), the maximum value and the total variation may not strictly converge to the maximum value and the total variation of the limit solution. This indicates that some important properties may be lost during the process from semi - discrete approximation to continuous solution.
2. **Maximum principle and monotonicity of total variation**: Despite the above problems, the paper proves that all generalized solutions obtained through semi - discrete approximation still satisfy the maximum principle and the monotonicity of total variation. That is:
- For each \(t\geq0\), the function \(t\mapsto M^{+}(t)\) is non - increasing, and \(t\mapsto M^{-}(t)\) is non - decreasing, where \(M^{+}(t)\) and \(M^{-}(t)\) represent the essential supremum and infimum of \(u(t, x)\) respectively.
- For each \(t\geq0\), the functions \(t\mapsto TV^{+}(t)\) and \(t\mapsto TV^{-}(t)\) are both non - increasing, where \(TV^{+}(t)\) and \(TV^{-}(t)\) represent the positive total variation and negative total variation of \(u(t, x)\) respectively.
To prove these results, the paper introduces a class of solutions called **uv - evolution** and proves that all generalized solutions obtained through semi - discrete approximation are uv - evolution. The definition of uv - evolution is as follows:
### Definition 2.8 (One - dimensional uv - evolution with homogeneous Neumann boundary conditions)
In the interval \((a, b)\subseteq\mathbb{R}\), a **uv - evolution** is a pair of measurable functions \(u:(0,+\infty)\times(a, b)\to\mathbb{R}\) and \(v:(0,+\infty)\times(a, b)\to\mathbb{R}\) satisfying the following properties:
1. **Time regularity**: The function \(u\) has a weak derivative \(u_{t}\) with respect to time, and
\[
u_{t}\in L^{1}((0, T)\times(a, b))\quad\forall T > 0.
\]
2. **Spatial regularity**: For almost all \(t > 0\),
- The function \(x\mapsto u(t, x)\) belongs to \(BV((a, b))\),
- The function \(x\mapsto v(t, x)\) belongs to \(W^{1,1}((a, b))\).
3. **Evolution equation**: The functions \(u\) and \(v\) satisfy
\[
u_{t}(t, x)=v_{x}(t, x)\quad\text{in}\ (0,+\infty)\times(a, b).
\]
4. **Neumann boundary conditions**: For almost all \(t > 0\),
\[
v(t, a)=v(t, b)=0.
\]
5. **Sign condition**: For almost all \(t > 0\),
\[
v(t, x)\cdot Du(t, x)\geq0\quad\text{as a measure in}\ (a, b).
\]
Through these definitions, the paper proves that all uv -