Abstract:We introduce a notion of tractability for ill-posed operator equations in Hilbert space. For such operator equations the asymptotics of the best possible rate of reconstruction in terms of the underlying noise level is known in many cases. However, the relevant question is, which level of discretization, again driven by the noise level, is required in order to achieve this best possible accuracy. The proposed concept adapts the one from Information-based Complexity.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the tractability of linear ill - posed problems in Hilbert spaces. Specifically, the author introduces a new concept to measure the degree of discretization driven by the noise level in order to achieve the best possible reconstruction accuracy. This concept draws on Information - based Complexity theory and shows the relevance of this concept in dealing with the curse of dimensionality through several examples.
### Problem Background
1. **Linear Ill - posed Problems**
- Formally represented as the operator equation \(Ax = y\), where \(A\) is a bounded linear operator from the real infinite - dimensional Hilbert space \(X\) to \(Y\).
- The noise model is \(y^{\delta}=Ax + \delta\xi\), where the norm of the unknown noise element \(\xi\) is restricted to be within 1, that is, \(\|Ax - y^{\delta}\|_Y\leq\delta\).
- The goal is to approximately reconstruct the unknown element \(x\) from the noisy data \(y^{\delta}\).
2. **Known Results**
- In many cases, it is known that under appropriate smoothness assumptions, the optimal reconstruction rate is related to the noise level \(\delta\).
- The core of the problem lies in determining the discretization level required to achieve this optimal reconstruction accuracy.
### Main Contributions
1. **Introduction of the Tractability Concept**
- The author proposes a new concept for measuring the tractability of linear ill - posed problems, which takes into account the effects of the noise level \(\delta\) and the spatial dimension \(d\).
- This concept is proposed to more accurately understand the difficulty of solving numerical problems at a given accuracy.
2. **Theoretical Analysis**
- By studying the singular value decomposition (SVD) of the operator equation, the author derives the expression for the minimum discretization level \(k^{*}(\delta, d)\).
- By analyzing the asymptotic behavior of \(k^{*}(\delta, d)\), the author draws conclusions about the tractability of the problem.
3. **Example Analysis**
- The paper shows the behavior of \(k^{*}(\delta, d)\) under different conditions through multiple examples, especially when the dimension \(d\) increases.
- For example, in some cases, even if the noise level \(\delta\) is small, but when the dimension \(d\) is large, \(k^{*}(\delta, d)\) will grow exponentially, making the problem intractable.
### Conclusion
- By introducing the tractability concept, the author provides a new method to evaluate and understand the difficulty of solving linear ill - posed problems in high - dimensional cases.
- This concept not only takes into account the influence of the noise level but also the influence of the spatial dimension, thus providing more valuable guidance for solving practical problems.
### Formula Summary
- **Noise Model**
\[
y^{\delta}=Ax + \delta\xi,\quad\|\xi\|_Y\leq1
\]
- **Minimum Discretization Level**
\[
k^{*}(\delta)=\begin{cases}
1, & \text{if }\Theta(s_1^2)\leq\delta,\\
\max\{k\in\mathbb{N}:\Theta(s_k^2)>\delta\}, & \text{otherwise}
\end{cases}
\]
- **Asymptotically Optimal Error**
\[
e(H_{\phi},\delta)\leq\phi(\Theta^{-1}(\delta))
\]
- **Multidimensional Case**