Pith. sign in

REVIEW 2 major objections 4 minor 42 references

A differentiable digital twin of a closed-Brayton gas-cooled reactor makes gradient-based parameter inversion competitive with established filters on three of four operating regimes.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 03:26 UTC pith:TLAJ4K22

load-bearing objection A careful, honest benchmark of AD-based inversion on a nuclear DAE twin — worth reviewing, but the stated bias floor and the reported mean errors need reconciliation before the headline numbers can be trusted. the 2 major comments →

arxiv 2607.23176 v1 pith:TLAJ4K22 submitted 2026-07-25 eess.SY cs.SY

An Adjoint-Based Differentiable Physics Framework for Online Parameter Inversion in Closed-Brayton Gas-Cooled Reactor Digital Twins

Chengyuan Li , Shanfang Huang , Jian Deng This is my paper
classification eess.SY cs.SY
keywords digital twindifferentiable physicsadjoint methodautomatic differentiationparameter inversion4D-Varclosed Brayton cyclegas-cooled reactor
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This paper tries to establish that a first-principles reactor digital twin can be made differentiable end to end, and that the resulting exact gradients let a variational estimator recover a physical reactivity coefficient from noisy, partial sensor streams as accurately as — and often more accurately than — established ensemble and unscented filters. The payoff is practical: gradient-based inversion, long considered impractical for nonlinear plant models, becomes a viable online-monitoring tool across a reactor's operating range rather than only at steady state. A sympathetic reader would care because the method replaces hand-derived adjoint code with automatic differentiation through the implicit solver, making the whole estimator family cheap to deploy, and because the paper gives a clear scoping rule for when to use which estimator.

Core claim

The central claim is that reverse-mode automatic differentiation through an implicit BDF differential-algebraic solver of a closed-Brayton He-Xe reactor exposes exact parameter sensitivities, and that an AD-Hessian incremental 4D-Var estimator built on those sensitivities attains the lowest mean error on the reflector reactivity coefficient on three of the four benchmark corners: 0.43% under transient full observation (versus 5.11% for the unscented filter), 0.43% on the partial-observation steady-state noise scan, and 2.23% pooled on the combined transient-partial corner (versus 2.69% for the ensemble filter). The unscented filter keeps its best-linear-unbiased advantage at the controlled s

What carries the argument

The load-bearing object is the end-to-end differentiable digital twin itself: an implicit differential-algebraic equation system with a singular mass matrix, integrated by a BDF scheme that admits reverse-mode automatic differentiation, so one backward solve yields the full parameter gradient. On top of it sit a quasi-random low-discrepancy warm-start stage and a trust-region inner loop whose gradient and Hessian both come from reverse-mode AD, forming the AD-Hessian incremental 4D-Var estimator; a calibrated-Huber loss and per-direction Tikhonov priors handle outliers and rank-deficient directions; and the differentiability simplifications (smooth-tanh event ramps, fast quasi-equilibrium re

Load-bearing premise

The paper assumes that the roughly 0.5% deterministic bias introduced by its three differentiability simplifications is small compared to measurement noise, while its best reported mean errors are at or below that floor, and it validates the twin only against a sibling model, not against independent experimental truth.

What would settle it

Run the same four benchmark corners with synthetic truth generated by an independent high-fidelity simulation or by a physical closed-Brayton test loop, and measure the realized bias of the AD-Hessian estimator at the lowest noise level; if the bias exceeds the paper's stated ~0.5% floor or the estimator no longer beats the ensemble filter on the transient-partial corner, the central claim of competitive gradient-based inversion would be falsified.

Watch this falsifier. Get emailed when new claim-graph text bears on it.

If this is right

  • If the central claim holds, gradient-based inversion can be deployed for reactor digital twins without hand-written tangent-linear and adjoint codes, since the backward pass is obtained automatically through the implicit solver.
  • The estimator's advantage on transient and partial-observation corners means online parameter monitoring is feasible during load-following and off-normal transients, not only at steady state.
  • The bias-decomposition result implies that further accuracy gains will come from reducing the differentiability simplifications' deterministic bias rather than from more statistical efficiency in the estimators.
  • The Randomize-Then-Optimize posterior sampler, if validated beyond the pilot, offers calibrated credible intervals for safety-relevant monitoring on the same differentiable plant.
  • The deployment rule — gradient methods for transient or partial-observation monitoring at low-to-moderate noise, sequential filters for the controlled steady-state corner and extreme noise — is actionable for plant operators.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • Editorial inference: Because the synthetic truth comes from a sibling model sharing the same transport-property and point-kinetics physics, the claimed 0.43% errors are consistency measures between two implementations, not fidelity to a real plant; a truly independent validation could shift all absolute errors without necessarily changing the ranking.
  • Editorial inference: The stated ~0.5% deterministic bias floor sits at or above several headline mean errors, suggesting that at the lowest noise levels the realized bias must be smaller than the paper's own bound; if an external test shows the bias is actually that large, the low-noise advantages would be partly an artifact of the shared-model setup.
  • Editorial inference: The same AD-through-implicit-DAE recipe should transfer to other nonlinear plant models (thermal-hydraulic loops, chemical process systems, power cycles) whenever their governing equations are smooth enough for implicit integration, making the estimator family a general tool rather than a reactor-specific one.
  • Editorial inference: The bimodal posterior on the rank-deficient recuperator coefficient indicates that multi-parameter inversion needs either additional sensors, stronger priors, or explicit multi-modality handling if trustworthy joint estimates are required.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 4 minor

Summary. The paper develops an end-to-end differentiable digital twin of a closed-Brayton He–Xe reactor: a JAX/diffrax implicit BDF DAE formulation with reverse-mode automatic differentiation through the implicit solver. On this twin it builds an AD-Hessian incremental 4D-Var estimator with Sobol pre-screening and trust-region inner solves, and benchmarks it against EnKF, ETKF, UKF, and finite-difference 4D-Var on a 2×2 scenario matrix crossing steady/transient excitation with full/partial observation. The headline results report the lowest mean relative error on the reflector coefficient α_refl on three of the four corners (S2: 0.43%, S3: 0.43%, S4 pooled: 2.23%), with the UKF winning the controlled steady-state corner. The authors argue that residual errors are dominated by a deterministic bias floor from three differentiability-driven simplifications rather than by statistical inefficiency, since all estimator variances lie within roughly 3× the Cramér–Rao bound. A Randomize-Then-Optimize posterior sampler is presented in pilot form.

Significance. If the reported numbers hold, this is a meaningful contribution: it demonstrates that gradient-based inversion through an implicit DAE plant model can be competitive with established sequential filters, without a surrogate, and it packages several textbook devices (incremental 4D-Var, trust regions, Sobol warm starts, full AD Hessians) into a working pipeline for a reactor class that has received little attention in the digital-twin literature. The closed-form CRLB derivation, the paired Wilcoxon/bootstrap testing, the ablation structure, and the promised code release are genuine strengths. The main limitations are that the synthetic truth is generated by the authors' own NuHeXSys FMU, so the benchmarks establish sibling-model consistency rather than plant fidelity, and that an internal inconsistency between the stated deterministic bias floor and the sub-0.5% mean errors must be resolved before the central superiority claim can be accepted.

major comments (2)
  1. [§2.6, §5.4, §5.6, Table 2] Section 2.6 states that the three differentiability simplifications 'impose a deterministic offset on α̂_refl of order 5×10⁻³ in relative units'; Section 5.6 says the residual RMSE is 'dominated by a bias floor' from these simplifications. Table 2 reports Adjoint Incremental mean relative errors of 0.43% on S2 and S3, and §5.4 reports '0.40–0.46%' across the S3 noise levels. For a deterministic bias b, E|e| ≥ |E[e]| = |b|; a population bias of 5×10⁻³ therefore cannot be consistent with sample means below 4.3×10⁻³ unless the realized bias is materially smaller than the stated offset. This is not a finite-sample artefact at the lowest S3 noise, where the reported CRLB-scaled variances are small. The paper must reconcile the §2.6 bound with the §5.6 bias-floor claim and the Table 2 numbers, for example by reporting the measured bias from the decomposition in §5.6 and revising one of the thr
  2. [§2.1, §5.1, §6] The synthetic truth for every benchmark scenario is generated by the OpenModelica FMU of the authors' own NuHeXSys model, which shares the twin's transport-property and point-kinetics physics. Table 1 therefore validates consistency between sibling models built from the same physical closures, not fidelity to a real plant. The Section 6 acknowledgement that the benchmark is twin-to-twin is appropriate, but the abstract's conclusion that gradient-based inversion is competitive 'across a reactor's operating range' overstates the external validity of the evidence. Please scope the headline claims to the twin-to-twin numerical comparison, or add an independent validation benchmark (e.g., the Sandia SBL-30 data mentioned in Section 6).
minor comments (4)
  1. [§5.6 / Abstract] The claim that 'every estimator's variance sits within roughly 3× the CRLB' is demonstrated only for the S3 partial-observation noise scan; the abstract generalizes it to all scenarios. Clarify that this statement is scenario-specific, or provide the corresponding analysis for S1, S2, and S4.
  2. [§3.7, Table 2] Several baseline cells are based on incomplete seed sets: EnKF 4/8 (S1), 10/15 (S2), 29/40 (S3); ETKF 5/8 (S1), 4/10 (S2). The authors argue that the cap truncation acts in the filters' favour, but this is an assertion. Report the missing-seed fraction per cell and a sensitivity check (e.g., worst-case imputation) to confirm that the rankings in Table 2 are not affected.
  3. [§3.2, §3.5] The phrase 'exact parameter sensitivities' should be qualified: the reverse-mode AD through the implicit BDF solver yields the discrete adjoint of the numerical integrator, not exact continuous sensitivities of the DAE. The distinction matters when solver tolerances are finite.
  4. [Supplementary Material] The manuscript repeatedly refers to Supplementary Material Sections S1–S8 for seed lists, ablation results, CRLB tabulations, calibration audits, and sequential-tracking tests, but no supplement is included with the manuscript. These materials are load-bearing for several detailed claims; please ensure they are available at review time.

Circularity Check

0 steps flagged

No derivation-equals-input circularity: the estimator pipeline and CRLB are self-contained, but the benchmark truth is the authors' own NuHeXSys FMU, a disclosed self-referential validation caveat rather than a circular derivation.

full rationale

The claimed derivation chain is not circular. Reverse-mode AD through the implicit BDF DAE (Sec. 3.2) produces sensitivities at the cost of one backward solve; the proposed estimator (Sec. 3.5) and the closed-form CRLB (Eqs. 12-13) are derived from the model's own sensitivities and checked numerically. All estimators share identical seed-aligned observation streams, and no fitted parameter is relabeled as a prediction. The main self-referential element is benchmark provenance, not derivation: Section 2.1 states that 'A full-fidelity OpenModelica FMU of the plant generates the synthetic data for every benchmark scenario' and that the loop/core 'follow the authors' NuHeXSys code [24], reformulated here for inversion'; Section 2.2 takes the transport-property map from the same reference. The paper itself concedes in Section 6 that 'the present benchmark is twin-to-twin'. This shared ancestry weakens external fidelity but does not make the estimator-vs-filter comparison circular, because the inverted α̂_refl is not inserted into the observation model. The stated 5×10⁻³ differentiability bias floor (Sec. 2.6) versus the 0.43% S2/S3 mean errors is an internal consistency/correctness issue to resolve, not a reduction of outputs to inputs. No uniqueness-importation, ansatz-smuggling, or renaming pattern is present. Score 2 reflects the minor self-citation in benchmark provenance only.

Axiom & Free-Parameter Ledger

6 free parameters · 8 axioms · 0 invented entities

The benchmark rests on six hand-chosen hyperparameters and eight assumptions the reader did not pay for upstream. Heaviest items: the asserted-but-unmeasured ~0.5% differentiability bias floor on α̂_refl (Section 2.6), which is at the same magnitude as the headline errors and is never checked against an independent reference; and the closed provenance in which both the truth-generating FMU and the differentiable twin descend from the authors' own NuHeXSys code [24], leaving the Sandia design point [23] as the only external anchor. Hyperparameter risk is partially mitigated by fixing settings across scenarios and by paired statistical tests on shared seeds. No new physical entities are postulated.

free parameters (6)
  • Tikhonov weight λ = selected per scenario by L-curve heuristic
    Section 3.3: λ is chosen at the curvature corner of the L-curve — a fit to the noise/prior trade-off that affects every regularised inversion result.
  • Per-direction prior weight λ_hA = ≈35.4 = (29748/5000)²
    Section 3.1 (text below Eq. 9) and Section 5.3: explicitly chosen to 'mirror the EnKF prior covariance on h_Arcp' — the A/B-test configuration is tuned to the baseline it is compared against.
  • Sobol pre-screening grid size / warm-starts / outer steps = N_Sobol=30, K=3, K_outer=15
    Section 3.5: hand-chosen hyperparameters; the h_Arcp recovery on S2 is admitted to depend on Sobol-grid coverage (Table 2 note ⋄).
  • Smooth-tanh ramp time τ_ramp = 10 s
    Section 2.6: one of the three differentiability simplifications; its contribution to the deterministic bias floor is estimated only by order of magnitude.
  • Moderately informative prior width σ_θ = 0.5·|θ_0|
    Section 4.5: fixed across scenarios by hand; contributes to the Bayes-Tikh and Huber+prior variants.
  • Baseline hyperparameters = EnKF N=80, inflation 1.05; ETKF same; UKF α=1e-3, β=2, κ=0; FD step not stated
    Section 3.7: baseline tuning affects the head-to-head gaps; the finite-difference 4D-Var gradient step size is not reported in the text.
axioms (8)
  • domain assumption Six-group point kinetics with three reactivity-feedback channels (Eq. 2), parameters from the Prometheus pin-block reference [23, 24]
    Section 2.3: standard reactor kinetics; the inversion target α_refl is one of the feedback coefficients, so the estimator's accuracy is bounded by this model's fidelity.
  • domain assumption Lumped two-mass core thermal model is 'lossless for the inversion': radial peaking leaves no signature on the sensed bulk channels (Section 2.3)
    Section 2.3: central reduction that keeps the adjoint affordable; if radial or axial peaking couples into the bulk sensors, the recovered bulk coefficients would be biased.
  • domain assumption The OpenModelica FMU truth generator is a faithful ground truth for the plant; benchmark is twin-to-twin (Sections 2.1, 6)
    The FMU is the authors' own NuHeXSys code [24], sharing transport-property and kinetics physics with the twin; fidelity is externally anchored only at the Sandia design point [23], with no experimental rig data.
  • ad hoc to paper The three differentiability simplifications (smooth-tanh events, fast-T1 reduction, explicit-P1) impose a bias small relative to the inversion target (Section 2.6)
    Asserted as 'order 5×10⁻³ in relative units' and 'comparable to the recovery target on second-scale safety events'; the paper's own headline errors (0.43%) sit at or below this floor, and the bias is never measured against an independent source.
  • domain assumption Parallel-observer assumption: the truth state at t_inject is known as initial condition (Sections 4.2, 2.6)
    Standard for twin benchmarks but unattainable on a real plant; affects the S2 and S4 transient results.
  • ad hoc to paper BDF propagator stalls are integrator pathology rather than filter algorithmic property; cap-truncation does not favor the proposed estimator (Section 3.7)
    The paper asserts truncation 'is unlikely to be missing-at-random' and favours the filters; this missingness claim is untestable from the text and is what the paired Wilcoxon tests rely on.
  • standard math Adjoint DAE well-posedness and reverse-mode AD correctness through the implicit solver (Cao et al. [18]; diffrax/JAX)
    Section 3.2: relies on continuous-adjoint DAE theory and the correctness of the discrete reverse pass; no formal verification is shipped.
  • domain assumption CRLB computed from the twin's sensitivity ∂y_sim/∂θ is the relevant information bound (Eq. 12–14)
    The bound measures information relative to the twin's model, not the FMU's or the plant's; if the twin's sensitivities are biased, the bound and the efficiency claims shift.

pith-pipeline@v1.3.0-alltime-deepseek · 18837 in / 36961 out tokens · 307370 ms · 2026-08-01T03:26:08.381449+00:00 · methodology

0 comments
read the original abstract

Digital twins for advanced reactors must invert physical parameters online from noisy, partial sensor streams while the plant is rarely at steady state. Gradient-based inversion is the natural tool for this task, but the forward models in routine use are seldom differentiable end to end, so practitioners fall back on derivative-free filters. We present an end-to-end-differentiable digital twin of a closed-Brayton gas-cooled reactor that propagates reverse-mode automatic differentiation through an implicit differential-algebraic plant model, exposing exact parameter sensitivities. The twin drives an AD-Hessian incremental 4D-Var estimator, benchmarked against ensemble, unscented, and finite-difference variational baselines over a two-by-two matrix that crosses steady with transient excitation and full with partial observation. No single estimator wins everywhere: the unscented filter keeps its best-linear-unbiased advantage on the controlled steady-state corner, while the proposed estimator attains the lowest mean error on the reflector coefficient on the other three corners --- reaching \SI{0.43}{\percent} under transient full observation, roughly an order of magnitude below the unscented filter, and matching the ensemble filter at low-to-moderate noise on the combined transient-partial corner. Every estimator's variance sits within a small factor of the Cram\'er--Rao bound, so the residual error reflects a deterministic bias floor from the twin's differentiability simplifications rather than statistical inefficiency, and the advantage is one of robustness to the resulting multi-modal loss landscape, which a component ablation localises by regime. Differentiating a first-principles plant model thus makes gradient-based inversion competitive with established filters across a reactor's operating range.

Figures

Figures reproduced from arXiv: 2607.23176 by Chengyuan Li, Jian Deng, Shanfang Huang.

Figure 1
Figure 1. Figure 1: Pipeline of the proposed AD-Hessian Incremental 4D-Var estimator (Stage A Sobol pre-screening, Stage B trust [PITH_FULL_IMAGE:figures/full_fig_p009_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Benchmark-scenario design space along observability (horizontal) and dynamic regime (vertical). S1 is the controlled [PITH_FULL_IMAGE:figures/full_fig_p011_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Twin validation against the OpenModelica reference for the three channels reaching steady state on the 5000 s horizon: [PITH_FULL_IMAGE:figures/full_fig_p013_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Scenario S1 head-to-head: bars = eight-seed means of the six core estimators and the ETKF supplement (EnKF on [PITH_FULL_IMAGE:figures/full_fig_p014_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Scenario S2 joint αrefl+hArcp inversion. Top row: Prx and TBeO/T1 traces. Bottom row: αˆrefl (c) and hAˆ rcp (d) relative errors. Bars are ten-seed means; whiskers are min/max. EnKF aggregated over 10/15 completed seeds (†); ETKF over 4/10 (‡). Panel (d) hArcp is only weakly identifiable: the across-seed mean is 28 % (median 21 %, per-seed range ∼3–60 %), and the 0.12 % mark is a single coverage-lucky Sobo… view at source ↗
Figure 6
Figure 6. Figure 6: Scenario S2 sensitivity-matrix SVD: effective rank one ( [PITH_FULL_IMAGE:figures/full_fig_p016_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Scenario S3 partial-observation steady-state inversion: top, four-channel sensor traces at [PITH_FULL_IMAGE:figures/full_fig_p017_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Scenario S4 head-to-head on the combined transient-and-partial-observation budget. Left: [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Scenario S3 total error and bias decomposition: (a) relative RMSE [PITH_FULL_IMAGE:figures/full_fig_p018_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Cost–accuracy Pareto front on Scenarios S2–S4. The dashed staircase splits the sequential-filter segment [PITH_FULL_IMAGE:figures/full_fig_p019_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Joint (αrefl, hArcp) RTO posterior on Scenario S2 (medium pilot, Nseeds = 3, NRTO = 5). The strongly identifiable αrefl is tightly recovered near the truth (star); the rank-deficient hArcp is bimodal, with mass split between the truth basin (24000W/K, dotted line) and a displaced mode near 30–38 kW/K. 6. Conclusions We have presented an end-to-end-differentiable digital twin of a closed-Brayton gas-cooled… view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

42 extracted references · 5 canonical work pages

  1. [1]

    H. Gong, T. Zhu, Z. Chen, Y. Wan, Q. Li, Parameter identification and state estimation for nuclear reactor operation digital twin, Annals of Nuclear Energy 180 (2023) 109497.doi:10.1016/j.anucene. 2022.109497

    arXiv 2023

  2. [2]

    Y. Do, M. Zebrowitz, J. Stahl, F. Zhang, iFANnpp: Nuclear power plant digital twin for robots and autonomous intelligence, Annals of Nuclear Energy 227 (2026) 111993.doi:10.1016/j.anucene.2025. 111993

    work page doi:10.1016/j.anucene.2025 2026

  3. [3]

    L. Gong, C. Peng, Q. Huang, Toward a digital twin for nuclear reactors: Data assimilation in thermal- hydraulics using EnKF, Progress in Nuclear Energy 195 (2026) 106305.doi:10.1016/j.pnucene. 2026.106305

    arXiv 2026

  4. [4]

    L. Hong, H. Gong, H. Ji, J. Lu, H. Li, Q. Li, Optimizing near-carbon-free nuclear energy systems: Advances in reactor operation digital twin through hybrid machine learning algorithms for parameter identification and state estimation, Nuclear Science and Techniques 35 (8) (2024) 135.doi:10.1007/ s41365-024-01494-2

    2024

  5. [5]

    Carrassi, M

    A. Carrassi, M. Bocquet, L. Bertino, G. Evensen, Data assimilation in the geosciences: An overview of methods, issues, and perspectives, Wiley Interdisciplinary Reviews: Climate Change 9 (5) (2018) e535. doi:10.1002/wcc.535

    doi:10.1002/wcc.535 2018

  6. [6]

    Evensen, The Ensemble Kalman Filter: Theoretical formulation and practical implementation, Ocean Dynamics 53 (4) (2003) 343–367.doi:10.1007/s10236-003-0036-9

    G. Evensen, The Ensemble Kalman Filter: Theoretical formulation and practical implementation, Ocean Dynamics 53 (4) (2003) 343–367.doi:10.1007/s10236-003-0036-9

    doi:10.1007/s10236-003-0036-9 2003

  7. [7]

    P.L.Houtekamer, H.L.Mitchell, AsequentialensembleKalmanfilterforatmosphericdataassimilation, Monthly Weather Review 129 (1) (2001) 123–137.doi:10.1175/1520-0493(2001)129<0123:ASEKFF> 2.0.CO;2

    doi:10.1175/1520-0493(2001)129 2001

  8. [8]

    J. L. Anderson, An adaptive covariance inflation error correction algorithm for ensemble filters, Tellus A 59 (2) (2007) 210–224.doi:10.1111/j.1600-0870.2006.00216.x

    arXiv 2007

  9. [9]

    M. Asch, M. Bocquet, M. Nodet, Data Assimilation: Methods, Algorithms, and Applications, Funda- mentals of Algorithms, SIAM, 2016.doi:10.1137/1.9781611974546

    doi:10.1137/1.9781611974546 2016

  10. [10]

    S. J. Julier, J. K. Uhlmann, Unscented filtering and nonlinear estimation, Proceedings of the IEEE 92 (3) (2004) 401–422.doi:10.1109/jproc.2003.823141

    arXiv 2004

  11. [11]

    S. M. Kay, Fundamentals of Statistical Signal Processing: Estimation Theory, Prentice Hall, 1993

    1993

  12. [12]

    Le Dimet, O

    F.-X. Le Dimet, O. Talagrand, Variational algorithms for analysis and assimilation of meteorological observations: Theoretical aspects, Tellus A 38 (2) (1986) 97–110.doi:10.3402/tellusa.v38i2.11706

    doi:10.3402/tellusa.v38i2.11706 1986

  13. [13]

    R. H. Reichle, Data assimilation methods in the Earth sciences, Advances in Water Resources 31 (11) (2008) 1411–1418.doi:10.1016/j.advwatres.2008.01.001. 21

    work page doi:10.1016/j.advwatres.2008.01.001 2008

  14. [14]

    Bonavita, E

    M. Bonavita, E. Hólm, L. Isaksen, M. Fisher, The evolution of the ECMWF hybrid data assimilation system, Quarterly Journal of the Royal Meteorological Society 142 (694) (2016) 287–303.doi:10. 1002/qj.2652

    2016

  15. [15]

    Doucet, A

    A. Doucet, A. M. Johansen, A tutorial on particle filtering and smoothing: Fifteen years later, in: D. Crisan, B. Rozovskii (Eds.), The Oxford Handbook of Nonlinear Filtering, Oxford University Press, 2011, pp. 656–704

    2011

  16. [16]

    Tarantola, Inverse Problem Theory and Methods for Model Parameter Estimation, SIAM, 2005

    A. Tarantola, Inverse Problem Theory and Methods for Model Parameter Estimation, SIAM, 2005. doi:10.1137/1.9780898717921

    doi:10.1137/1.9780898717921 2005

  17. [17]

    P. C. Hansen, Discrete Inverse Problems: Insight and Algorithms, SIAM, 2010.doi:10.1137/1. 9780898718836

    doi:10.1137/1 2010

  18. [18]

    Y. Cao, S. Li, L. Petzold, R. Serban, Adjoint sensitivity analysis for differential-algebraic equations: The adjoint DAE system and its numerical solution, SIAM Journal on Scientific Computing 24 (3) (2003) 1076–1089.doi:10.1137/S1064827501380630

    work page doi:10.1137/s1064827501380630 2003

  19. [19]

    Innes, Don’t unroll adjoint: Differentiating SSA-form programs, arXiv preprint arXiv:1810.07951 (2018)

    M. Innes, Don’t unroll adjoint: Differentiating SSA-form programs, arXiv preprint arXiv:1810.07951 (2018)

    Pith/arXiv arXiv 2018

  20. [20]

    Rackauckas, Y

    C. Rackauckas, Y. Ma, J. Martensen, C. Warner, K. Zubov, R. Supekar, D. Skinner, A. Ramad- han, A. Edelman, Universal differential equations for scientific machine learning, arXiv preprint arXiv:2001.04385 (2020)

    Pith/arXiv arXiv 2001

  21. [21]

    Kidger, On neural differential equations, arXiv preprint arXiv:2202.02435 (2021)

    P. Kidger, On neural differential equations, arXiv preprint arXiv:2202.02435 (2021)

    Pith/arXiv arXiv 2021

  22. [22]

    Sapienza, J

    F. Sapienza, J. Bolibar, F. Schäfer, B. Groenke, et al., Differentiable programming for differential equations: A review, arXiv preprint arXiv:2406.09699 (2024)

    arXiv 2024

  23. [23]

    S. A. Wright, R. J. Lipinski, M. E. Vernon, T. Sanchez, Closed brayton cycle power conversion systems for nuclear reactors: Modeling, Operations, and Validation, Technical Report SAND2006-2518, Sandia National Laboratories (2006)

    2006

  24. [24]

    C. Li, L. Guo, S. Huang, J. Deng, Optimization and simulation of startup control for space nuclear power systems with closed Brayton cycle based on NuHeXSys, Progress in Nuclear Energyhttps:// www.sciencedirect.com/science/article/abs/pii/S0149197025000083. NuHeXSys analysis code with non-ideal He–Xe properties, multi-channel core thermal-hydraulics, and ...

    2025

  25. [25]

    Courtier, J.-N

    P. Courtier, J.-N. Thépaut, A. Hollingsworth, A strategy for operational implementation of 4D-Var, using an incremental approach, Quarterly Journal of the Royal Meteorological Society 120 (519) (1994) 1367–1387.doi:10.1002/qj.49712051912

    doi:10.1002/qj.49712051912 1994

  26. [26]

    Trémolet, Model-error estimation in 4d-var, Quarterly Journal of the Royal Meteorological Society 133 (626) (2007) 1267–1280.doi:10.1002/qj.94

    Y. Trémolet, Model-error estimation in 4d-var, Quarterly Journal of the Royal Meteorological Society 133 (626) (2007) 1267–1280.doi:10.1002/qj.94

    doi:10.1002/qj.94 2007

  27. [27]

    I. M. Sobol’, On the distribution of points in a cube and the approximate evaluation of integrals, USSR Computational Mathematics and Mathematical Physics 7 (4) (1967) 86–112.doi:10.1016/ 0041-5553(67)90144-9

    1967

  28. [28]

    A. R. Conn, N. I. M. Gould, P. L. Toint, Trust-Region Methods, MPS-SIAM Series on Optimization, SIAM, 2000.doi:10.1137/1.9780898719857

    doi:10.1137/1.9780898719857 2000

  29. [29]

    Nocedal, S

    J. Nocedal, S. J. Wright, Numerical Optimization, Springer Series in Operations Research and Financial Engineering, Springer, 1999.doi:10.1007/978-0-387-40065-5. 22

    doi:10.1007/978-0-387-40065-5 1999

  30. [30]

    Bradbury, R

    J. Bradbury, R. Frostig, P. Hawkins, M. J. Johnson, C. Leary, D. Maclaurin, G. Necula, A. Paszke, J. VanderPlas, S. Wanderman-Milne, Q. Zhang, JAX: Composable transformations of Python+NumPy programs, software,http://github.com/google/jax(2018)

    2018

  31. [31]

    Kidger, diffrax: Differentiable solvers for odes, sdes, cdes in JAX, Software,https://github.com/ patrick-kidger/diffrax(2022)

    P. Kidger, diffrax: Differentiable solvers for odes, sdes, cdes in JAX, Software,https://github.com/ patrick-kidger/diffrax(2022)

    2022

  32. [32]

    R. T. Q. Chen, Y. Rubanova, J. Bettencourt, D. K. Duvenaud, Neural ordinary differential equations, Advances in Neural Information Processing Systems 31 (2018)

    2018

  33. [33]

    A. C. Hindmarsh, P. N. Brown, K. E. Grant, S. L. Lee, R. Serban, D. E. Shumaker, C. S. Wood- ward, SUNDIALS: Suite of nonlinear and differential/algebraic equation solvers, ACM Transactions on Mathematical Software 31 (3) (2005) 363–396.doi:10.1145/1089014.1089020

    arXiv 2005

  34. [34]

    U. M. Ascher, L. R. Petzold, Computer Methods for Ordinary Differential Equations and Differential- Algebraic Equations, SIAM, 1998.doi:10.1137/1.9781611971392

    doi:10.1137/1.9781611971392 1998

  35. [35]

    D. C. Liu, J. Nocedal, On the limited memory BFGS method for large scale optimization, Mathematical Programming 45 (1–3) (1989) 503–528.doi:10.1007/BF01589116

    doi:10.1007/bf01589116 1989

  36. [36]

    P. C. Hansen, Analysis of discrete ill-posed problems by means of the L-curve, SIAM Review 34 (4) (1992) 561–580.doi:10.1137/1034115

    doi:10.1137/1034115 1992

  37. [37]

    J. M. Bardsley, Randomize-then-optimize: A method for sampling from posterior distributions in nonlinear inverse problems, SIAM Journal on Scientific Computing 36 (4) (2014) A1895–A1910. doi:10.1137/140964023

    work page doi:10.1137/140964023 2014

  38. [38]

    Z. Wang, J. M. Bardsley, Randomize-then-optimize algorithms for variational inference, SIAM Journal on Numerical Analysis 56 (6) (2018) 3450–3477.doi:10.1137/18M1166691

    work page doi:10.1137/18m1166691 2018

  39. [39]

    Doucet, N

    A. Doucet, N. de Freitas, N. Gordon (Eds.), Sequential Monte Carlo Methods in Practice, Information Science and Statistics, Springer, 2001.doi:10.1007/978-1-4757-3437-9

    doi:10.1007/978-1-4757-3437-9 2001

  40. [40]

    G. Kitagawa, Monte carlo filter and smoother for non-gaussian nonlinear state space models, Journal of Computational and Graphical Statistics 5 (1) (1996) 1–25.doi:10.1080/10618600.1996.10474692

    arXiv 1996

  41. [41]

    C. H. Bishop, B. J. Etherton, S. J. Majumdar, Adaptive sampling with the ensemble transform Kalman filter. Part I: Theoretical aspects, Monthly Weather Review 129 (3) (2001) 420–436.doi:10.1175/ 1520-0493(2001)129<0420:ASWTET>2.0.CO;2

    2001

  42. [42]

    Fritzson, A

    P. Fritzson, A. Pop, K. Abdelhak, A. Ashgar, B. Bachmann, et al., The OpenModelica integrated environment for modeling, simulation, and model-based development, Modeling, Identification and Control 41 (4) (2020) 241–295.doi:10.4173/mic.2020.4.1. 23

    doi:10.4173/mic.2020.4.1 2020