Pith. sign in

REVIEW 5 major objections 5 minor 1 cited by

Goal-oriented optimal sensor placement for PDE-constrained inverse problems in crisis management

T0 review · 5 major / 5 minor · reviewed 2026-08-06 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Five goal-oriented sensors match the reconstruction accuracy of a 96-sensor grid, and a steered mobile sensor beats a stationary one within 7 seconds.

desk verdict A competent, clearly written integration of established Bayesian OED tools for contaminant tracking; the realistic geometry is a plus, but the method is not new and the validation is self-consistent, with a genuine sign inconsistency in Eq. (9) that must be fixed before the headline claims are fully supported. read the letter →

arxiv 2507.02500 v2 pith:IUJ3BJL7 submitted 2025-07-03 math.NA cs.NA

Marco Mattuschka , Noah An der Lan , Max von Danwitz , Daniel Wolff , Alexander Popp This is my paper · ORCID
classification math.NAcs.NA MSC 65N2162K0535R30
keywords goal-orientedoptimalexperimentaldesignC-optimalBayesianinverseproblemssensorplacementdynamicsteeringairbornecontaminanttransportlow-rankHessianapproximationadvection-diffusion
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper is trying to show that you do not need many sensors to locate and track an airborne contaminant source if you choose where to measure based on the specific decision you care about. It formulates source identification as a Bayesian inverse problem governed by an advection-diffusion PDE and then selects sensor locations by a C-optimal design criterion that minimizes posterior variance of a chosen quantity of interest, using a low-rank Hessian approximation to keep the computation fast. In the paper's experiments, five selected stationary sensors match the reconstruction quality of a 96-sensor grid in a critical region, eight sensors are enough to monitor a region over a time window, and one mobile sensor steered by the same criterion reaches good reconstruction in about 7 seconds. The authors claim the framework is transferable to other crisis-management sensing tasks because the criterion only depends on the model through the chosen quantity-of-interest vector.

What carries the argument

The load-bearing object is the C-optimal design criterion applied to the posterior covariance of the quantity of interest: instead of minimizing the trace of the full posterior covariance (A-optimality), one minimizes the variance of a single linear functional $\rho = P(m)$ of the parameter. The goal-oriented operator $P$ is encoded as a vector $c$ in the finite-element space — for a space-time region of interest, $c$ is the solution of an adjoint advection-diffusion problem $c = K^*(1_{[T_0,T_{\mathrm{final}}]\times P})$ — and the design objective becomes $c^T H(w)^{-1} c$. This is evaluated cheaply through a low-rank approximation of the preconditioned Hessian (randomized SVD of the preconditioned forward operator plus Sherman-Morrison-Woodbury), and sparse layouts are obtained by adding an $\ell^1$ penalty on sensor weights and solving with L-BFGS-B. The gradient of the objective with respect to each sensor weight reduces to a squared forward response $(F^i(q))^2$, which makes the optimization tractable.

What would settle it

Generate synthetic observations from a finer or otherwise different discretization of Equation 1, or perturb the wind field, and rerun the C-optimal five-sensor design; if reconstruction error in the region of interest no longer matches the 96-sensor configuration, the result hinges on the test setup. A controlled tracer-release experiment with independent ground truth would also settle it.

Watch

Extended reading notes

Core claim

The paper's central claim is that goal-oriented C-optimal experimental design, accelerated by low-rank approximation of the preconditioned Hessian, produces sparse stationary sensor layouts and effective dynamic steering trajectories for PDE-constrained Bayesian inverse problems. In the airborne contaminant test case, five optimally placed stationary sensors reconstruct the initial condition in a critical subregion with accuracy comparable to the full 96-sensor configuration, eight sensors suffice for monitoring a region over a time window, and a mobile sensor steered by the same criterion substantially improves the reconstruction within about 7 seconds where a stationary sensor alone fails to give a reliable estimate within 12 seconds. The authors present this as evidence that decision-relevant uncertainty, rather than global field uncertainty, is the right target for sensor placement in crisis management.

Load-bearing premise

The load-bearing premise is that the numerical validation is meaningful despite being self-consistent: the same discretized advection-diffusion model generates the 'true' measurements, defines the likelihood, and supplies the forward map, so the claimed five-sensor accuracy may not survive model error or unknown conditions.

Editorial extensions

If this is right

  • In the static source-identification case, five sensors selected by the goal-oriented C-optimal criterion reconstruct the initial condition inside the region of interest with accuracy comparable to the full 96-sensor grid.
  • When the goal is monitoring a region over a time window, eight sensors suffice to reconstruct the concentration evolution and concentrate the uncertainty reduction in that region.
  • Steering one mobile sensor with the same design criterion, alongside a poorly placed stationary sensor, yields substantially better reconstruction after about 7 seconds than the stationary sensor alone does after 12 seconds.
  • Because the design objective depends on the quantity of interest through a single vector $c$, the same algorithm transfers to other physical models and other user-defined quantities of interest.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the same comparison were repeated with data from an independent forward model — different discretization, unknown or unsteady wind, or sensor faults — the reported sensor counts and times could shift; the paper does not establish robustness to such model error.
  • The five-sensor advantage is demonstrated on synthetic data in which the ground truth and the inversion share a discretization; a laboratory tracer release with measured wind would separate algorithmic merit from self-consistency.
  • The dynamic steering loop recentres the quantity of interest on the current maximum of the reconstruction, which is a heuristic; one could extend it to account for posterior uncertainty in the location of the maximum, or to multiple mobile sensors.
  • Because the goal-oriented criterion reduces to a fixed vector, the method's computational cost is dominated by the low-rank Hessian approximation, suggesting it could scale to three-dimensional crisis geometries if the spectrum of the preconditioned Hessian decays similarly.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 5 minor

Summary. The paper proposes a Bayesian goal-oriented optimal experimental design framework for stationary and mobile sensor placement in PDE-constrained inverse problems, applied to airborne contaminant source identification. The authors use a Laplacian-like trace-class prior, precondition the Hessian, approximate the preconditioned misfit Hessian by a rank-r eigendecomposition, and combine a C-optimality criterion with an ℓ1 penalty to select sparse sensor configurations. For dynamic steering, the QoI is centered at the maximum of the current maximum-a-posteriori reconstruction and the mobile sensor moves to the highest-weight candidate location. Three numerical experiments on a campus geometry obtained from OpenStreetMap are presented: reconstructing the initial condition in a region with five stationary sensors, monitoring a space-time QoI with eight sensors, and steering a mobile sensor alongside one fixed sensor. The paper claims accuracy comparable to a dense 96-sensor configuration and faster reconstruction with the mobile sensor.

Significance. If the results hold, the framework would be a useful practical extension of earlier dynamic-steering work to goal-oriented designs and complex geometries, with potential real-time capability due to the low-rank offline/online decomposition. The paper gives an explicit adjoint-based formulation for space-time quantities of interest and reports a substantial speedup for reduced-order forward evaluations. Strengths include the clear finite-element setting, the use of a trace-class prior, and the three demonstration cases. However, the validation is entirely synthetic and largely qualitative, no code or data is provided, and the printed low-rank and derivative formulas contain sign inconsistencies. The contribution is therefore promising but not yet fully supported as written.

major comments (5)
  1. [§3.2, Eq. (9)] Equation (9) is inconsistent with Equation (8) and with the formula used in Section 4.1. Since the preconditioned Hessian is Hmisfit + I (Eq. (8)), the Sherman-Morrison-Woodbury identity gives (I + Hmisfit)^{-1} ≈ I - V_r D_r V_r^T with D_r = diag(λ/(1+λ)). Equation (9) instead states (Hmisfit_h(w) - I)^{-1} ≈ I + V_r D_r V_r^T, changing both the sign inside the inverse and the sign of the low-rank correction. Section 4.1 uses c_h^T A_h^{-1}(I - V_r D_r V_r^T)A_h^{-1}c_h, so the printed derivation and the implemented objective are mutually contradictory. Because no code is provided, the reader cannot determine which form was actually used; if Eq. (9) were used, the minimized quantity would not be the posterior variance and the five- and eight-sensor layouts in Section 5 would not be C-optimal in the stated sense. Please correct the equation and provide a reproducible implementation or a numerical check of the objective.
  2. [§4.1, Eq. (12)] The derivative formula has the wrong sign. For Γpost(w) = (F^T W Γ_noise^{-1} W F + Γ_pr^{-1})^{-1} and the trace objective c^T H^{-1}c, the derivative with respect to w_i is -(F_i H^{-1}c)^2/σ^2, where F_i is the i-th row of the observation operator. Equation (12) states a positive value. Combined with the ℓ1 penalty of Eq. (10), a positive gradient would drive all weights to zero during minimization, so either the sign is a typographical error or the optimization used a different objective; in either case the reported sparse configurations are not supported by the printed formulas.
  3. [§5.1 and §5.2] The central claim that five (resp. eight) sensors achieve accuracy comparable to the full 96-sensor configuration is supported only by visual comparison of reconstructed fields (Figures 9 and 10) and qualitative variance plots. No error norms, signal-to-noise metrics, or posterior-variance reductions are reported. Please provide quantitative errors, e.g., relative L² errors in the quantity of interest and in the reconstructed initial condition, for both the sparse and the full configurations.
  4. [§5] The numerical validation is self-consistent rather than independent: the ground truth is generated by the same discretized advection-diffusion model used in the likelihood, the wind field is assumed known and divergence-free, and the noise variance and prior hyperparameters are hand-chosen. The experiments therefore do not test the method under model error, unknown wind, or sensor faults. The conclusion in Section 6 that the method proves real-world capability is too strong; please add robustness experiments (e.g., perturbed wind, different noise levels, or model discrepancy) or soften the claims accordingly.
  5. [§3.3 and §5] The rank r=200 truncation and the binarization threshold are not justified. Figure 8 shows singular-value decay for F and F∘A, but not for the preconditioned Hessian used in Eq. (9), and no convergence study of the C-optimal objective with respect to r is given. Since the design depends on the low-rank approximation, please report the spectrum of the preconditioned misfit Hessian and the sensitivity of the selected layouts to r, α, and the threshold.
minor comments (5)
  1. [§2.3] In the formula for mmap(w), Γ_pr mpr should be Γ_pr^{-1} mpr; as printed, the equation is dimensionally inconsistent and omits the inverse of the prior covariance.
  2. [§4.1 and §5] The symbol P is overloaded as both the QoI operator and the spatial subset (e.g., P1 and P2 in Section 5); please use distinct notation for the operator and the subset.
  3. [Figure 8] The right panel claims that the reduced-order model approximates the forward operator fairly well, but no error metric or maximum relative error is given; please add a quantitative error plot.
  4. [§5.3] The mobile sensor speed of approximately 40 m/s greatly exceeds typical UAV speeds; if this is a deliberate abstraction, state this and discuss the implications for the steering claim.
  5. [Throughout] There are minor typographical errors, including 'Navier-Stockes' for 'Navier-Stokes', 'rigth' for 'right' in the Figure 9 caption, and inconsistent capitalization of section references.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the OED derivation is self-contained; synthetic validation is self-consistent but not circular.

full rationale

After walking the derivation chain, I find no step in which a claimed prediction or first-principles result reduces to its own inputs by construction. The C-optimal objective in Section 4.1 is obtained by standard adjoint calculus: the operator P is defined independently, its adjoint P* is computed, and the posterior-variance expression is ⟨c,H^{-1}c⟩; the low-rank approximation is cited from external literature ([2,15,18]) and is used as an approximation rather than as the definition of the quantity being evaluated. The internal sign inconsistency between Eq. (9) and the formula used in Section 4.1 is a reproducibility/correctness concern, not a circularity: it does not equate a prediction to an input by construction. Sensor sparsity is controlled by a user-chosen regularization parameter α (Eq. (10)), and the resulting layouts are evaluated by applying the same Bayesian inverse problem; no fitted parameter is renamed as a prediction. The numerical experiments generate synthetic observations from the same discretized forward model ('we use the forward simulation of Equation 1 ... as the ground truth'), so the validation is self-consistent rather than independent—a limitation for external validity, but a standard in-silico test that does not make the derivation circular. Self-citations ([16], [20]) concern grid generation and an earlier digital-twin simulation; they are not load-bearing for the OED results.

Assumptions & free parameters 6 free parameters · 7 assumptions · 0 invented entities

The framework is grounded in standard Bayesian and optimal experimental design theory, but the numerical results depend on several hand-chosen hyperparameters (prior covariance, noise, sparsity penalty, low-rank rank, steering window) and on the assumption that the same model generates data and performs inference. The steering heuristic and rank truncation are adopted without independent justification.

free parameters (6)
  • Prior covariance hyperparameters eta, gamma = eta = 8, gamma = 800
    Chosen in Section 3.2 and Section 5 to define A = 8I - 800Delta. The posterior covariance and the optimal experimental design objective depend directly on this prior.
  • Noise variance sigma^2 = sigma = 0.005
    Assumed sensor noise in all experiments (Section 5). It scales the likelihood and the C-optimal design criterion.
  • Sparsity regularization parameter alpha = 0.1 (OED 1), 1.0 (OED 2), unspecified (OED 3)
    Balances information gain against the number of sensors in Equation (10). It controls sparsity and the final sensor count in the reported layouts.
  • Low-rank truncation rank r = r = 200 spectral values
    The Hessian inverse is approximated using 200 eigenvalues (Section 5). The accuracy of the design and steering depends on the assumed eigenvalue decay.
  • Binarization threshold for sensor weights = not stated
    Weights are thresholded to a binary sensor configuration in Section 3.3, but the threshold is never specified, and it determines the final sensor count.
  • Steering QoI size and time window = 40 m square, [Tstep, Tstep + 2 s]
    Defined in OED 3 (Section 5.3). The steering behavior depends on the chosen region size and time horizon, which are set by hand without sensitivity analysis.
assumptions (7)
  • domain assumption Wind field v is sufficiently smooth, bounded, and divergence-free, and is known exactly.
    Used to define the advection-diffusion forward model and adjoint (Section 2.1, Equation 1). The inverse problem does not infer or marginalize over wind uncertainty.
  • domain assumption The prior for the initial condition is Gaussian with covariance Gamma_pr = A^-2, where A = eta I - gamma Delta with Robin boundary condition.
    Makes the inverse problem well-posed and defines the metric used in the optimal experimental design objective (Section 3.2, Equations 7-8), following [2,17].
  • domain assumption The initial condition lies in H^1_{Gamma_-}(Omega) and is zero on the inflow boundary.
    Parameter space restriction stated in Section 2.2. The adjoint and prior covariance are set up on this space.
  • ad hoc to paper The Hessian spectrum decays fast enough for a rank-200 approximation to be accurate for design and steering.
    Observed experimentally for the preconditioned operator in Figure 8. It is not guaranteed for other geometries or transport parameters, and the design uses this approximation.
  • ad hoc to paper Synthetic observations are generated by the same discretized forward model used for inversion and design.
    Section 5 uses "the forward simulation of Equation 1 ... as the ground truth." This makes the numerical validation self-consistent and avoids addressing model error.
  • ad hoc to paper For steering, the QoI region is set as a 40 m square centered at the maximum of the current MAP reconstruction.
    Heuristic in Section 4.2 and Figure 7. No convergence, robustness, or sensitivity analysis is given for this choice.
  • standard math Linear Gaussian Bayesian updates and the Sherman-Morrison-Woodbury formula.
    Used in Sections 2.2-2.3 and Equation (9) to derive the posterior covariance and the low-rank inverse approximation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Goal-oriented optimal sensor placement for PDE-constrained inverse problems in crisis management." pith.science (2026). https://pith.science/paper/IUJ3BJL7

@misc{pith2026250702500,
  author       = {Pith},
  title        = {Pith review of: Goal-oriented optimal sensor placement for PDE-constrained inverse problems in crisis management},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/IUJ3BJL7}},
  note         = {Machine review of arXiv:2507.02500}
}
read the original abstract

This paper presents a novel framework for goal-oriented optimal static sensor placement and dynamic sensor steering in PDE-constrained inverse problems, utilizing a Bayesian approach accelerated by low-rank approximations. The framework is applied to airborne contaminant tracking, extending recent dynamic sensor steering methods to complex geometries for computational efficiency. A C-optimal design criterion is employed to strategically place sensors, minimizing uncertainty in predictions. Numerical experiments validate the approach's effectiveness for source identification and monitoring, highlighting its potential for real-time decision-making in crisis management scenarios.

Figures

Figures reproduced from arXiv: 2507.02500 by the authors.

Figure 1
Figure 1. Unmanned systems operated as autonomous sensor platforms at German Aerospace Center (DLR). Images: [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Forward simulation of airborne contaminant transport on a campus geometry. Estimated wind vector field [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Inverse Problem. Measurements at 96 equidistantly spaced sensor positions (left) and reconstructed initial [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (10 more)
Figure 4
Figure 4. Figure 4: Region of interest P defining the QoI (left) and point-wise variance σ 2 (x) of mmap obtained with five selected sensors (green spheres, right) [PITH_FULL_IMAGE:figures/full_fig_p007_4.png]
Figure 5
Figure 5. Figure 5: Illustration of the QoI 1[T QoI 0 ,T QoI final]×P (left) and solution c = K∗ (1[T QoI 0 ,T QoI final]×P ) of the transport problem induced by the adjoint operator K∗ (right). can extract the diagonal of the inverse Hessian matrix in the finite element basis, assigning …
Figure 6
Figure 6. Figure 6: Dynamic sensor steering test case with stationary sensor (green) and mobile sensor (purple trajectory). (a) [PITH_FULL_IMAGE:figures/full_fig_p009_6.png]
Figure 7
Figure 7. Figure 7: Algorithm for dynamic sensor steering based on goal-oriented sensor placement [19, 1]. [PITH_FULL_IMAGE:figures/full_fig_p009_7.png]
Figure 8
Figure 8. Figure 8: Reduced-order modeling. Decay of singular values of [PITH_FULL_IMAGE:figures/full_fig_p010_8.png]
Figure 9
Figure 9. Figure 9: OED 1. Weights of optimal sensor configuration to monitor [PITH_FULL_IMAGE:figures/full_fig_p011_9.png]
Figure 10
Figure 10. Figure 10: OED 2. Weights of optimal configuration to monitor concentration evolution in [PITH_FULL_IMAGE:figures/full_fig_p011_10.png]
Figure 11
Figure 11. Figure 11: OED 2. Point-wise variance σ 2 as measure of the uncertainty of the inferred parameter for the optimal configuration for P2 (left) compared to the full sensor grid (right) predicted correctly in region P2 during the time from 5 s to 12 s. This formulation results in t…
Figure 12
Figure 12. Figure 12: OED 3. Data fusion of stationary sensor (marked in green) and mobile sensor (marked in purple). Mobile [PITH_FULL_IMAGE:figures/full_fig_p013_12.png]
Figure 13
Figure 13. Figure 13: OED 3. Comparison of predictions based on data fusion of mobile sensor and stationary sensor (left column) [PITH_FULL_IMAGE:figures/full_fig_p014_13.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Sparse Source Identification in Transient Advection-Diffusion Problems with a Primal-Dual-Active-Point Strategy

    math.NA 2025-11 conditional novelty 6.0 of 10

    A primal-dual-active-point algorithm with Radon-norm regularization identifies sparse contaminant sources in advection-diffusion problems from scarce sensor data, beating L2-regularized baselines in synthetic 2D/3D be...

Reference graph

Works this paper leans on

25 extracted references · 8 canonical work pages · cited by 1 Pith paper

  1. [1]

    From Data to Decisions: A Real-Time Measurement–Inversion–Prediction–Steering Framework for Hazardous Events and Health Monitoring, pages 195–227

    Sonja Wogrin, Arjun Singh, Douglas Allaire, Omar Ghattas, and Karen Willcox. From Data to Decisions: A Real-Time Measurement–Inversion–Prediction–Steering Framework for Hazardous Events and Health Monitoring, pages 195–227. Springer International Publishing, Cham, 2023. ISBN 978-3-031-27986-7. doi:10.1007/978-3- 031-27986-7_8

    doi:10.1007/978-3- 2023

  2. [2]

    hIPPYlib: An extensible software framework for large-scale inverse problems governed by PDEs: Part I: Deterministic inversion and linearized Bayesian inference

    Umberto Villa, Noemi Petra, and Omar Ghattas. hIPPYlib: An extensible software framework for large-scale inverse problems governed by PDEs: Part I: Deterministic inversion and linearized Bayesian inference. ACM Transactions on Mathematical Software , 47(2):1–34, 2021. ISSN 0098-3500. doi:10.1145/3428447. URL https://doi.org/10.1145/3428447

    doi:10.1145/3428447 2021

  3. [3]

    Model variational inverse problems governed by partial differential equations

    Noemi Petra and Georg Stadler. Model variational inverse problems governed by partial differential equations. ICES REPORT 11-05, The Institute for Computational Engineering and Sciences, The University of Texas at Austin, 2011. URL https://math.nyu.edu/~stadler/papers/PetraStadler11.pdf

    work page 2011

  4. [4]

    Nathan Halko, Per-Gunnar Martinsson, and Joel A. Tropp. Finding structure with randomness: Probabilistic algorithms for constructing approximate matrix decompositions. SIAM Review, 53(2):217–288, 2011. ISSN 0036-1445. doi:10.1137/090771806

    doi:10.1137/090771806 2011

  5. [5]

    Randomized algorithms for the low-rank approximation of matrices

    Edo Liberty, Franco Woolfe, Per-Gunnar Martinsson, Vladimir Rokhlin, and Mark Tygert. Randomized algorithms for the low-rank approximation of matrices. Proceedings of the National Academy of Sciences, 104(51):20167– 20172, 2007. doi:10.1073/pnas.0709640104

    doi:10.1073/pnas.0709640104 2007

  6. [6]

    Local and superlinear convergence for truncated iterated projections methods

    Trond Steihaug. Local and superlinear convergence for truncated iterated projections methods. Mathematical Programming, 27(2):176–190, 1983. ISSN 0025-5610. doi:10.1007/BF02591944

    doi:10.1007/bf02591944 1983

  7. [7]

    Alen Alexanderian and Arvind K. Saibaba. Efficient D-optimal design of experiments for infinite-dimensional Bayesian linear inverse problems. SIAM Journal on Scientific Computing, 40(5):A2956–A2985, 2018. ISSN 1064-8275. doi:10.1137/17M115712X

    doi:10.1137/17m115712x 2018

  8. [8]

    Goal-oriented optimal approximations of Bayesian linear inverse problems

    Alessio Spantini, Tiangang Cui, Karen Willcox, Luis Tenorio, and Youssef Marzouk. Goal-oriented optimal approximations of Bayesian linear inverse problems. SIAM Journal on Scientific Computing, 39(5):S167–S196,

    work page

Show all 25 references
  1. [9]

    Ahmed Attia, Alen Alexanderian, and Arvind K. Saibaba. Goal-oriented optimal design of experiments for large- scale Bayesian linear inverse problems. Inverse Problems, 34(9):095009, 2018. doi:10.1088/1361-6420/aad210

    2018 doi

  2. [10]

    An offline-online decomposition method for efficient linear Bayesian goal-oriented optimal experimental design: Application to optimal sensor placement

    Keyi Wu, Peng Chen, and Omar Ghattas. An offline-online decomposition method for efficient linear Bayesian goal-oriented optimal experimental design: Application to optimal sensor placement. SIAM Journal on Scientific Computing, 45(1):B57–B77, 2023. ISSN 1064-8275. doi:10.1137...

    2023 doi

  3. [11]

    Optimal experimental design for infinite-dimensional Bayesian inverse problems governed by PDEs: a review

    Alen Alexanderian. Optimal experimental design for infinite-dimensional Bayesian inverse problems governed by PDEs: a review. Inverse Problems, 37(4):043001, 2021. ISSN 0266-5611. doi:10.1088/1361-6420/abe10c

    2021 doi

  4. [12]

    Elman and Tengfei Su

    Howard C. Elman and Tengfei Su. A low-rank solver for the stochastic unsteady Navier–Stokes prob- lem. Computer Methods in Applied Mechanics and Engineering , 364:112948, 2020. ISSN 0045-7825. doi:https://doi.org/10.1016/j.cma.2020.112948

    2020

  5. [13]

    Choosing observation operators to mitigate model error in Bayesian inverse problems

    Nada Cvetkovi´c, Han Cheng Lie, Harshit Bansal, and Karen Veroy. Choosing observation operators to mitigate model error in Bayesian inverse problems. SIAM/ASA Journal on Uncertainty Quantification, 12(3):723–758,

  6. [14]

    Lineare Funktionalanalysis

    Hans Wilhelm Alt. Lineare Funktionalanalysis. Springer Berlin Heidelberg, Berlin, Heidelberg, 2012. ISBN 978-3-642-22260-3. doi:10.1007/978-3-642-22261-0

    2012 doi

  7. [15]

    A-optimal design of experiments for infinite-dimensional Bayesian linear inverse problems with regularized ℓ0-sparsification

    Alen Alexanderian, Noemi Petra, Georg Stadler, and Omar Ghattas. A-optimal design of experiments for infinite-dimensional Bayesian linear inverse problems with regularized ℓ0-sparsification. SIAM Journal on Scientific Computing, 36(5):A2122–A2148, 2014. ISSN 1064-8275. doi:10....

    2014 doi

  8. [16]

    Contaminant dispersion simulation in a digital twin framework for critical infrastructure protection

    Max von Danwitz, Jacopo Bonari, Philip Franz, Lisa Kühn, Marco Mattuschka, and Alexander Popp. Contaminant dispersion simulation in a digital twin framework for critical infrastructure protection. In 9th European Congress on Computational Methods in Applied Sciences and Engine...

    2024 doi

  9. [17]

    Mitigating the influence of the boundary on PDE-based covariance operators

    Yair Daon and Georg Stadler. Mitigating the influence of the boundary on PDE-based covariance operators. Inverse Problems and Imaging, 12(5):1083–1102, 2018. ISSN 1930-8337. doi:10.3934/ipi.2018045

    2018 doi

  10. [18]

    Tobin Isaac, Noemi Petra, Georg Stadler, and Omar Ghattas. Scalable and efficient algorithms for the prop- agation of uncertainty from data through inference to prediction for large-scale problems, with application to flow of the Antarctic ice sheet. Journal of Computational P...

    2015 doi

  11. [19]

    Model Reduction for Dynamic Sensor Steering: A Bayesian Approach to Inverse Problems

    Sonja Wogrin. Model Reduction for Dynamic Sensor Steering: A Bayesian Approach to Inverse Problems . Master’s thesis, Massachusetts Institute of Technology. Computation for Design and Optimization Program., 2008. URL http://hdl.handle.net/1721.1/43739

    2008

  12. [20]

    Towards real-time urban physics simulations with digital twins

    Jacopo Bonari, Lisa Kühn, Max von Danwitz, and Alexander Popp. Towards real-time urban physics simulations with digital twins. In 2024 28th International Symposium on Distributed Simulation and Real Time Applications (DS-RT), pages 18–25, 2024. doi:10.1109/DS-RT62209.2024.00013

    2024

  13. [21]

    Baratta, Joseph P

    Igor A. Baratta, Joseph P. Dean, Jørgen S. Dokken, Michal Habera, Jack S. Hale, Chris N. Richardson, Marie E. Rognes, Matthew W. Scroggs, Nathan Sime, and Garth N. Wells. DOLFINx: The next generation FEniCS problem solving environment, 2023

    2023

  14. [22]

    Linear convergence of accelerated conditional gradient algorithms in spaces of measures

    Konstantin Pieper and Daniel Walter. Linear convergence of accelerated conditional gradient algorithms in spaces of measures. ESAIM: Control, Optimisation and Calculus of Variations, 27:38, 2021. ISSN 1292-8119. doi:10.1051/cocv/2021042

    2021

  15. [23]

    Investigation of reinforcement learning for shape optimization of 2D profile extrusion die geometries

    Clemens Fricke, Daniel Wolff, Marco Kemmerling, and Stefanie Elgeti. Investigation of reinforcement learning for shape optimization of 2D profile extrusion die geometries. Advances in Computational Science and Engineering, 1(1):1–35, 2023. doi:10.3934/acse.2023001. 16

    2023 doi

  16. [2017]

    doi:10.1137/16M1082123

    ISSN 1064-8275. doi:10.1137/16M1082123

    doi

  17. [2024]

    doi:10.1137/23M1602140

    doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.