On optimality and bounds for internal solutions generated from boundary data-driven Gramians
V. Druskin, S. Moskow, M. Zaslavsky · 2025
Read On optimality and bounds for internal solutions generated from boundary data-driven Gramians by V. Druskin, S. Moskow, M. Zaslavsky (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-2504-15407.txt.
Introduction
Druskin, Moskow, and Zaslavsky study in 2025 how to rebuild inner wave fields from measures taken only at the boundary. The test case is a plasma wave with unknown traits. Their method reshapes known background snapshots using a matrix built from boundary data.
They show the result sits close to the best fit of the true field onto those snapshots. That closeness yields a general error bound for inner fields built from boundary data. In one dimension, they prove settling toward the truth for two forms of start data, at a set rate. Computer tests back the rate and show it cannot be much improved.
Catalog record
| CREATOR | V. Druskin, S. Moskow, M. Zaslavsky |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 47 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-15407
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
On optimality and bounds for internal solutions generated from boundary data-driven Gramians V. Druskin, S. Moskow, M. Zaslavsky Abstract We consider the computation of internal solutions for a time domain plasma wave equation with unknown coefficients from the data obtained by sampling its transfer function at the boundary. The computation is performed by transforming known background snapshots using the Cholesky decomposition of the data-driven Gramian. We show that this approximation is asymptotically close to the projection of the true internal solution onto the subspace of background snapshots. This allows us to derive a generally applicable bound for the error in the approximation of internal fields from boundary data for a time domain plasma wave equation with an unknown potential q q . For general q ∈ L ∞ q\in L^{\infty} , we prove convergence of these data generated internal fields in one dimension for two examples. The first is for piecewise constant initial data and sampling τ \tau equal to the pulse width. The second is piecewise linear initial data and sampling at half the pulse width. We show that in both cases the data generated solutions converge in L 2 L^{2} at order τ \sqrt{\tau} . We present numerical experiments validating the result and the sharpness of this convergence rate. † † runningheads: Bounds for internal solutions from boundary data / V. Druskin, S. Moskow, and M. Zaslavsky 1 Introduction There is a vast literature on the inverse scattering problem of determining an unknown potential q from boundary data for the plasma wave equation u_{tt}-\Delta u+q(x)u=0, (1) see, for example [ 31 , 27 , 37 , 15 , 14 , 33 , 30 , 17 , 36 , 29 , 34 ] . It is well known that knowledge of the interior solution u makes the reconstruction problem for q much easier. Indeed, some methods to compute q from boundary data construct an approximate u as an intermediary step [ 29 ] , and the ability to obtain internal data is a main advantage to coupled physics, or hybrid inversion methods [ 2 ] . One class of inversion methods use reduced order models, or ROMs. Reduced order models (ROMs) in this context are finite dimensional versions of ( 1 ) that can be determined from the boundary data, see for example [ 3 ] , and have led to efficient inversion algorithms [ 5 , 8 , 20 , 19 , 9 , 10 ] . In the work [ 7 ] , it was first found that one could use ROMs to produce accurate internal data from the boundary data only. These approximate fields can then be combined with the Lippmann-Schwinger integral equation for image reconstruction [ 22 , 23 , 11 , 24 , 25 ] , where the image quality depends on the accuracy of the approximation of the internal fields. The accuracy of these fields has mostly been explained in the following intuitive (but non rigorous) way: after sequential orthogonalization of the time snapshots, the perturbed and background orthogonalized snapshots are quite similar due to the cancellation of the reflected wave. One result about the data generated snapshots is in the appendix of [ 11 ] , where the authors construct an exact solution for the case of a piecewise constant medium in one dimension with interfaces exactly coinciding with the time steps. They show that if the initial wave starts with support away from the interfaces, the data-generated solutions are exact. This is because for this special medium, the true snapshots are exactly a linear combination of the background snapshots. In this paper, we consider a general setup, where obtaining the exact solution is not possible. We show that the error in the data-generated solutions can be controlled by the error in the best approximation of the true solutions from the space of background snapshots, and these errors become asymptotically close if the error in the best approximation is small. Then, for well-chosen initial waves, we obtain convergence of the data generated internal solutions for general one-dimensional media. The general error bounds hold for any SISO (single input-single output) problem regardless of dimension, however, in higher dimensions a SISO background subspace will not generally be rich enough for convergence. In higher dimensions a MIMO setup (multiple input -multiple output) will be necessary for convergence, and the approach presented in this manuscript provides a framework for a rigorous higher dimensional study. Convergence is based on the following steps: 1. The data generated internal snapshots can be characterized as the unique (admissible) linear combination of background snapshots that exactly interpolate the true Gramian (mass matrix) M . 2. For a sufficiently rich space of background snapshots, the best approximations will have a mass matrix that is close to the true mass matrix. We show a stability result for the mapping from the mass matrix to the snapshots, and this relates the data-generated solutions to these best approximations. 3. Since the background snapshots are in the approximation space, the best approximation error is based on the regularity of the scattered field, which is much more regular than the singular pulse. The analysis shows that the shape of the pulse and the time sampling interval should be such that the background snapshots form a good approximation space; but not oversampled to cause the mass matrix to be badly conditioned. We prove convergence for two one-dimensional examples. In the first example we use a piecewise constant approximate delta pulse with sampling equal to the pulse width \tau . In the second example, we probe with a piecewise linear approximate delta pulse with the time sampling \tau at half of the pulse width. In both cases we have convergence of the solutions on the order of \sqrt{\tau} for q\in L^{\infty} . The paper is organized as follows. In Section 2, we describe precisely the inverse problem. The method to construct the approximate internal fields from the boundary data is described in Section 3, along with a proof that they interpolate the Gramian exactly. In Section 4 we prove a general error bound. The main results are Proposition 2 and Corollary 1, showing that these data generated internal fields have error that is asymptotically close to the error in the projections of the true solutions onto the background snapshot space. In Section 5 we show that by using a step function initial wave, we obtain convergence as the sampling interval approaches zero. We consider a piecewise linear hat initial wave in Section 6. In Section 7 we discuss how this framework works in higher dimensions, and in Section 8 we show numerical experiments. Section 9 contains a discussion of other possible generalizations and future work. 2 Problem setup We consider first a single input/single output (SISO) problem. That is, for now we assume we have just one source collocated with the receiver, and consider the following wave model problem for a domain \Omega\subset\mathbb{R}^{n} u_{tt}-\Delta u+q(x)u=0\ \ \mbox{in}\ \Omega\times[0,\infty) (2) with initial conditions \displaystyle u(t=0) \displaystyle= \displaystyle g\ \mbox{in}\ \Omega (3) \displaystyle u_{t}(t=0) \displaystyle= \displaystyle 0\ \mbox{in}\ \Omega (4) \displaystyle{\partial u\over{\partial\nu}} \displaystyle= \displaystyle 0\ \mbox{on}\ \partial\Omega (5) where g is initial data representing a localized source near the boundary, and