Abstract:Complex valued measures of finite total variation are a powerful signal model in many applications. Restricting to the d -dimensional torus, finitely supported measures can be exactly recovered from their trigonometric moments up to some order if this order is large enough. Here, we consider the approximation of general measures, e.g., supported on a curve, by trigonometric polynomials of fixed degree with respect to the 1-Wasserstein distance. We prove sharp lower bounds for their best approximation and (almost) matching upper bounds for effectively computable approximations when the trigonometric moments of the measure are known. A second class of sum of squares polynomials is shown to interpolate the indicator function on the support of the measure and to converge to zero outside.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is how to approximate a measure by trigonometric polynomials of a fixed degree given the finite trigonometric moments of a measure. Specifically, the authors focus on the best approximation problem of general measures (such as measures supported on curves) under the 1 - Wasserstein distance, and the effective computational methods for such approximation. In addition, they also study another type of polynomial - sum of squares polynomials, which can interpolate on the support set of the measure and approach zero outside the support set.
### Background and Problem Description of the Paper
- **Background**: Complex - valued measures with finite total variation are a powerful signal model in many applications. When restricted to the d - dimensional torus, a measure with finite support can be exactly recovered from its trigonometric moments if the order is high enough.
- **Problem**: For general measures (for example, measures supported on curves), how to use trigonometric polynomials of a fixed degree to approximate these measures, and give strict lower bounds and almost - matching upper bounds for the best approximation under the 1 - Wasserstein distance? At the same time, explore whether a class of sum of squares polynomials can interpolate the indicator function on the support set of the measure and approach zero outside the support set.
### Main Contributions
1. **Strict Lower and Upper Bounds of Approximation Error**:
- Provide strict lower and upper bounds of point - wise approximation error.
- Give an upper bound of the approximation error under the 1 - Wasserstein distance, which decays inversely linearly (up to a logarithmic factor) as the polynomial degree increases.
2. **Relationship between Approximation Theory and Lipschitz Functions**:
- Establish a simple connection between approximation under the 1 - Wasserstein distance and the known approximation theory of Lipschitz functions.
- Relate the best polynomial approximation problem of measures to the best approximation results in \(L^1(T^d)\) and \(C(T^d)\).
3. **Interpolation Behavior of Sum of Squares Polynomials**:
- Analyze the point - wise convergence properties of a class of sum of squares polynomials \(p_{1,n}\), which can interpolate the indicator function on the support set of the measure and approach zero outside the support set.
- Prove that the maximum points of \(p_{1,n}\) on the support set of a discrete measure or a measure supported on an algebraic curve can be used to estimate the position of the support set.
### Methods and Techniques
- **Convolution Method**: Construct trigonometric polynomial approximation by convolving the measure with a specific kernel (such as the Fejér kernel).
- **Sum of Squares Polynomials**: Construct a class of sum of squares polynomials \(p_{1,n}\), and theoretically analyze their interpolation and support set estimation properties.
- **Wasserstein Distance**: Use the 1 - Wasserstein distance to quantify the quality of approximation and give strict lower and upper bounds.
### Application Prospects
- **Signal Processing**: In signal processing, especially in the super - resolution problem of sparse signals, these results can provide more effective signal recovery methods.
- **Image Processing**: In image processing, especially in the contour reconstruction of natural images and the phase recovery problem in X - ray crystallography, these theoretical results can provide new perspectives and methods.
Through these contributions, this paper not only promotes the research on measure approximation theoretically, but also provides powerful tools and methods for practical applications.