Pith. sign in

REVIEW 3 major objections 5 minor 46 references

The Most Dangerous Seed: Nonlinear Optimal Perturbations in Rayleigh-Taylor Instability

T0 review · 3 major / 5 minor · reviewed 2026-08-08 · deepseek-v4-flash

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

Pith's one-line read The paper claims that the initial perturbation maximizing kinetic-energy growth in two-dimensional compressible Rayleigh-Taylor instability is a coherent, interface-localized wave packet, and that it outperforms random perturbations of…

desk verdict Solid first application of CNOP to an astrophysical instability, but the 'most dangerous seed' claim is undercut by the objective mismatch: the paper maximizes squared kinetic-energy anomaly, not total kinetic energy. read the letter →

arxiv 2608.04085 v1 pith:SDQWWP6A submitted 2026-08-04 astro-ph.GA astro-ph.HEastro-ph.SRphysics.flu-dyn

Suoqing Ji , Bin Shi This is my paper · ORCID
classification astro-ph.GAastro-ph.HEastro-ph.SRphysics.flu-dyn
keywords Rayleigh-Taylorinstabilityconditionalnonlinearoptimalperturbationfinite-timegrowthnon-normalmodesturbulentmixingdifferentiablesimulationspectralanalysis
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 tries to establish that the most dangerous initial perturbation in a two-dimensional compressible Rayleigh-Taylor instability is not a random jumble but a coherent wave packet concentrated near the density interface and dominated by a narrow band of Fourier modes. Using the conditional nonlinear optimal perturbation (CNOP) method, the authors maximize the squared kinetic-energy anomaly at a chosen target time under a fixed initial-energy budget, and they report that the resulting seed produces kinetic-energy enhancements of factors 20.7, 32.5, and 8.6 over random perturbations of the same energy in their three configurations. The seed's structure depends on the optimization horizon: short horizons select a compact spectral band close to the linear regime, while longer horizons spread energy over more modes and larger scales. If correct, the result would mean that the spatial organization of a seed, not merely its amplitude, controls the efficiency of Rayleigh-Taylor-driven mixing in astrophysical fluids.

What carries the argument

The carrying mechanism is the conditional nonlinear optimal perturbation (CNOP) formulation: an optimization problem that maximizes the objective $J(\delta u_0)=\int_\Omega \left[e_k^*(r;\delta u_0)-e_k^*(r;0)\right]^2 dV$, the squared departure of the kinetic-energy density from the unperturbed baseline at the target time, subject to the initial-energy constraint $C(\delta u_0)=\int_\Omega \frac{1}{2}\rho_0(r)\,\delta u_0^2\, dV \le \varepsilon$. The solution is found by a projected quasi-Newton method that uses exact gradients obtained by reverse-mode automatic differentiation through the discretized compressible Euler equations, and the resulting seed is characterized by Fourier spectral energy analysis. The CNOP solution is what connects the chosen objective to the physical claim about the most dangerous seed.

What would settle it

Take one of the reported CNOP runs and launch the same projected quasi-Newton optimization from many hundreds of random seeds, plus structured seeds such as pure single modes, two-mode superpositions, and phase-randomized copies of the CNOP spectrum. If any of these reaches a strictly larger objective value than the reported optimum at the same energy budget, the claim that this is the most dangerous seed is false; if none does, the global-optimality objection is weakened but not eliminated.

Watch

Extended reading notes

Core claim

The central discovery is that, for a compressible ideal-gas two-layer atmosphere with Atwood number 1/3 and a finite-thickness interface, the initial vertical-velocity perturbation that maximizes the finite-time kinetic-energy anomaly is a coherent, interface-localized wave packet. At target time $t^*=2$ (about 4.1 e-folding times of the reference mode), the packet's horizontal Fourier energy is concentrated in modes $n=9$ to $15$ ($N=128$) or $n=9$ to $16$ ($N=256$), a band that is not the linear fastest-growing mode, because linear theory gives no finite preferred wavelength for an inviscid sharp interface. The preferred scales instead emerge from the interplay of finite interface thickness, finite optimization time, and nonlinear dynamics. At the longer horizon $t^*=3$, the spectrum broadens toward lower wavenumbers, and growth is increasingly carried by interactions among many modes, so the five-mode or 90%-power reconstructions that work at $t^*=2$ begin to fail. The paper's controlled comparisons show that the optimized seed initiates vigorous bubble-and-spike overturning and a substantially wider mixing layer, while random perturbations of equal energy remain comparatively quiescent.

Load-bearing premise

The reported seed is the most dangerous one only if the numerical optimizer converged to the global maximum of the nonconvex objective; the paper itself states that convergence from multiple random initializations does not constitute a mathematical proof of a global optimum.

Editorial extensions

If this is right

  • If an astrophysical system shows Rayleigh-Taylor-driven mixing stronger than expected for weak random seeding, the cause may be a coherent seed structure rather than a larger seed energy budget.
  • Because the CNOP seed's structure depends on the optimization time, linear stability theory alone cannot predict the initial perturbations that dominate finite-time nonlinear growth; finite-time nonlinear optimization is needed.
  • The five-dominant-mode and 90%-power reconstructions reproduce early nonlinear growth, so a small set of modes suffices when the evolution is short, but longer evolution requires the full distributed spectrum.
  • The method's gradient-based optimization can be extended to magnetized, radiative, or multiphase flows, where objective functions based on magnetic energy, emissivity, or mixed mass could replace kinetic-energy anomaly.
  • Short-horizon optima stay near the linear regime and select a compact spectral band, while long-horizon optima exploit nonlinear bubble-and-spike dynamics and shift to larger scales; this connects the seed to the emergent morphology.

Reading between the lines

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

  • The paper restricts perturbations to the vertical velocity component; extending the search to full velocity fields (including horizontal/shear perturbations) might yield a seed that is even more dangerous, since shear-induced Kelvin-Helmholtz roll-up could be seeded from the start.
  • A direct test of the phase-coherence hypothesis would be to randomize the Fourier phases of the CNOP seed while keeping its power spectrum; if the randomized version performs nearly as well, coherence is not the key factor, but if not, phase alignment is essential.
  • The two-dimensional setting likely changes the mode-selection and nonlinear cascade; in three dimensions, vortex stretching and the extra degree of freedom may move the optimal band and weaken the reported kinetic-energy ratios.
  • The chosen objective (squared kinetic-energy anomaly) defines 'danger' in one particular way; using mixed-mass or mixing-width as the objective might select a different optimal seed, so the title's 'most dangerous' is relative to the optimization target.
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

3 major / 5 minor

Summary. The paper applies the conditional nonlinear optimal perturbation (CNOP) method to the two-dimensional compressible Rayleigh-Taylor instability in an astrophysical setting. The authors maximize an objective defined by the squared kinetic-energy density anomaly at a target time, subject to a fixed initial kinetic-energy budget, using a differentiable JAX-based finite-volume solver and a projected L-BFGS optimizer. They report that the resulting optimal seeds are coherent, interface-localized wave packets with a narrow dominant horizontal mode band at short optimization horizons and a broader spectrum at longer horizons. Forward simulations compare each CNOP seed with a single localized random perturbation and with a five-mode reconstruction, finding larger integrated kinetic energy and mixing-layer growth for the CNOP cases. The paper concludes that the spatial organization of the seed, not just its amplitude, controls the efficiency of RT-driven mixing.

Significance. If the central claims hold, the paper would be a useful first demonstration of CNOP for an astrophysical instability, showing that finite-time nonlinear optimal perturbations are structurally coherent and that seed organization matters for RT-driven mixing. The use of differentiable simulation to compute exact discrete gradients is a methodological strength, and the paper includes convergence diagnostics and controlled forward runs at two resolutions and two horizons. The central structural finding, that a coherent interface-localized seed produces much stronger mixing than an incoherent random seed of equal energy, is interesting and worth publishing. However, the quantitative 'most dangerous seed' claim is undermined by a mismatch between the objective that is actually maximized and the physical quantity used to report the gain, and by the use of a single random realization as the comparison baseline.

major comments (3)
  1. [§3.1, Eq. (8); §3.3; Conclusion 3] The objective actually maximized is J = ∫Ω [e_k^*(r) − e_k^0(r)]^2 dV, the squared L2 norm of the kinetic-energy density anomaly, not the integrated kinetic energy ∫Ω e_k^* dV. Maximizing J does not, in general, maximize kinetic-energy growth; because the integrand is squared, J rewards spatial concentration of the final kinetic-energy field independently of its total amplitude. The factors 20.7, 32.5, and 8.6 reported in Conclusion 3 and shown in Fig. 3 are ratios of integrated kinetic energy, but the CNOP seed is not the maximizer of that quantity. The coherent, interface-localized wave-packet structure may therefore be partly an artifact of the L2 objective rather than the intrinsic structure of the physically 'most dangerous' perturbation. Please either reformulate the objective as total kinetic energy (or its anomaly) and rerun the optimizations, or reframe the abstract, title, and conclusions to state explicitly that the seed maximizes the squared kinetic-energy field anomaly.
  2. [§3.6, Table 3; Conclusion 3] Each comparison uses a single Random realization per case, with no ensemble averaging or error bars. A single draw from a Gaussian random field can be atypical, so the reported kinetic-energy factors of 20.7, 32.5, and 8.6 and the corresponding mixing-width differences have no quantified uncertainty. The claim that CNOP seeds outperform 'random perturbations' would be substantially strengthened by rerunning the forward comparisons with an ensemble of independent random seeds and reporting the median and spread, or at least by showing that the single shown realization is representative.
  3. [§5, last paragraph; title and abstract] The manuscript correctly acknowledges in §5 that convergence from multiple random initializations 'does not constitute a mathematical proof of a global optimum.' Despite this, the title 'The Most Dangerous Seed' and the abstract's statement that the perturbation 'maximizes the kinetic energy growth' assert global optimality. Even after the objective-mismatch issue is resolved, the global-optimality caveat should be reflected in the title and conclusions, for example by saying 'a particularly dangerous seed' or by explicitly stating that the optimum is local unless stronger evidence is provided.
minor comments (5)
  1. [§4.2, Eq. (12); §3.6, 5modes] The construction of the 5modes initial perturbation is under-specified: the Fourier coefficients are defined along the single row y_peak, but it is not explained how the selected horizontal modes are extended in the vertical direction. Please describe the full 2D initial field used in the 5modes runs.
  2. [Appendix A, Fig. A1] The axis labels in Fig. A1 read 'J (per cell)' and '‖∇J‖ (per component)', whereas Eq. (8) defines J as an integral over the domain. Please clarify whether the plotted quantity is J/Ω or another normalization, and ensure the convergence criterion is consistent with the definition used in the optimization.
  3. [§5, limitations paragraph] There is a typo in the sentence '... but not a constitute mathematical proof of a global optimum'; it should read 'but does not constitute a mathematical proof.'
  4. [§4.1] In the sentence 'The close argeement between T2_N128 and T2_N256...', 'argeement' should be 'agreement.'
  5. [Abstract and §3.2] The abstract says the CNOP perturbation 'maximizes the kinetic energy growth,' but Eq. (8) maximizes a squared-field anomaly. Please harmonize the terminology throughout, including the phrase 'most energetic response' in §3.2, so that the reader is not misled about the objective.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the optimized seed is the argmax of the stated objective, and all reported structural, kinetic-energy, and mixing results are forward-simulated outputs rather than inputs to the optimization.

full rationale

The paper's derivation chain is self-contained at the circularity level. The CNOP seed is explicitly defined as the argmax of the objective J in Eq. (8) subject to the energy constraint in Eq. (9), so the statement that the seed maximizes J is true by construction. However, every substantive reported finding—the coherent interface-localized morphology, the spectral band n~9–16, the kinetic-energy enhancement factors 20.7, 32.5, and 8.6, and the mixing-width growth—is obtained by forward Euler simulations initialized with that seed, not assumed in the objective. The comparison against random perturbations is a genuine computation, and the optimizer could in principle have returned any field satisfying the energy budget. No fitted parameter is renamed as a prediction: the perturbation energy epsilon is fixed, and the optimized objective J is not the same quantity as the reported total kinetic energy. The self-citations (Shi & Sun 2023; Shi & Ma 2024) are methodological and not load-bearing, and no uniqueness theorem from the authors' prior work is invoked. The reviewer's concern that J is the squared kinetic-energy-density anomaly rather than total kinetic energy is a legitimate construct-validity caveat for the 'most dangerous' wording, and the paper itself discloses this limitation in Section 5; but because the kinetic-energy and mixing results are measured forward quantities that do not reduce to J by construction, this is a correctness and interpretation risk rather than circularity.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new physical entities. Its central claim rests on three chosen numerical parameters (h_s, epsilon, h_loc) and on the modeling assumption that 2D inviscid Euler is a sufficient testbed. The most fragile input is h_s, which sets the short-wavelength cutoff and therefore participates in determining the preferred mode band. The optimization's convergence to a global optimum is an assumption, not a proved fact.

free parameters (3)
  • Interface smoothing length h_s = 0.02 L
    Chosen initial condition parameter; sets the finite transition width that suppresses short wavelengths and participates in selecting the CNOP preferred band (k h_s ~ 1.3 for n ~ 10). No h_s sensitivity study is reported.
  • Perturbation energy budget epsilon = 1e-6
    Chosen constraint amplitude; the optimal perturbation structure is defined relative to this energy budget, so the 'most dangerous seed' is budget-dependent. No variation in epsilon is explored.
  • Initial guess localization width h_loc = 0.1 L
    Width of the random Gaussian field used as the optimizer initial guess; the paper claims seed insensitivity but does not show a systematic study of h_loc.
assumptions (3)
  • domain assumption The 2D inviscid compressible Euler equations with ideal-gas EOS and uniform gravity adequately represent the astrophysical RT instability of interest.
    Invoked by the model setup in Section 2.1; excludes cooling, magnetic fields, 3D effects, and explicit dissipation, as acknowledged in the limitations.
  • standard math The linear RT growth rate sqrt(A g k) with finite-thickness corrections is a valid reference for interpreting the CNOP regime (t*/tau_RT).
    Used in Sections 2.2 and 5 to argue the preferred band arises from finite h_s and finite time; relies on classical linear stability theory for the smoothed profile.
  • domain assumption The projected L-BFGS optimizer converged to a stationary point, and the reported CNOP is treated as the global optimum.
    Appendix A reports decay of the stationarity measure, but global optimality is unproven; the paper explicitly acknowledges this in Section 5.

how reviews work

0 comments
Cite this review

Pith. "Pith review of The Most Dangerous Seed: Nonlinear Optimal Perturbations in Rayleigh-Taylor Instability." pith.science (2026). https://pith.science/paper/SDQWWP6A

@misc{pith2026260804085,
  author       = {Pith},
  title        = {Pith review of: The Most Dangerous Seed: Nonlinear Optimal Perturbations in Rayleigh-Taylor Instability},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/SDQWWP6A}},
  note         = {Machine review of arXiv:2608.04085}
}
read the original abstract

Long-term instabilities in astrophysical fluids are inherently nonlinear, where even small-amplitude perturbations can trigger dramatic instability. However, owing to the complex interactions among non-normal modes, the perturbation structures responsible for the greatest growth remain poorly understood. In this paper, we employ the nonlinear optimization method known as the conditional nonlinear optimal perturbation (CNOP) to identify the most dangerous initial velocity perturbation, i.e., the perturbation that maximizes the kinetic energy growth in the two-dimensional compressible Rayleigh-Taylor instability in astrophysical hydrodynamics. Compared with random perturbations, the optimal perturbation forms a coherent wave-packet structure localized around the density interface. We investigate its dependence on spatial resolution and optimization time horizon through two sets of numerical experiments. For a fixed optimization time, increasing the spatial resolution produces a more sharply localized wave packet, whereas for a fixed spatial resolution, increasing the optimization time causes the wave packet to become progressively more dispersed. Furthermore, we analyze the optimal perturbations in Fourier space using the fast Fourier transform (FFT), which provides a clearer characterization of the spectral distribution. Higher-resolution simulations concentrate most of the perturbation energy into only a few dominant modes, while longer optimization times distribute the energy over a broader range of modes. These results indicate that short optimization time horizons involve relatively weak modal interactions and remain closer to the linear regime, whereas longer optimization times enhance nonlinear modal interactions, broaden the spectral distribution, and reduce the predictive capability of linear stability theory.

Figures

Figures reproduced from arXiv: 2608.04085 by the authors.

Figure 1
Figure 1. Initial vertical-velocity perturbations. The left panel shows the localized random field used as the initial guess. The remaining panels show the CNOPs obtained by solving the optimization problem(10) for the cases: T2_N128, T2_N256, and T3_N128. All perturbations satisfy the same kinetic-energy constraint, 𝐶(𝛿𝑢0) = 10−6 . 0 20 40 60 Fourier mode n 0.00 0.05 0.10 0.15 Normalized spectral energy En N = 128;t ¤ = 2 0 … view at source ↗
Figure 2
Figure 2. Normalized spectral energy 𝐸𝑛 of CNOPs across the horizontal Fourier modes 𝑛. The orange bars indicate the five dominant modes used to construct the 5modes comparison runs, whereas the blue bars represent the remaining Fourier modes. 4.3. Total Kinetic-Energy Growth In the following two subsections, we investigate the nonlin￾ear growth induced by the CNOP seeds [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗
Figure 3
Figure 3. The comparison of total kinetic energy growth generated by CNOPs. For the baseline case, T2_N128, the nonlinear evolution initi￾ated by the dominant 5modes approximation closely follows that of the full CNOP, whereas the localized random pertur￾bation produces a markedly different evolution. As the grid resolution increases from 𝑁 = 128 to 𝑁 = 256, the discrep￾ancy between the dominant 5modes approximation and the f… view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: The comparison of mixing-layer growth generated by CNOPs with the mixing width given by Eq. (13). sequently evolve into coherent large-scale bubbles and spikes during the nonlinear development. Although no analytical theory currently predicts the preferred spectral ban…
Figure 5
Figure 5. Figure 5: Nonlinear evolution of the density field for the three initial perturbations: T2_N128, T2_N256, and T3_N128. at the grid size 𝑁 = 128 and 𝑛 = 9–16 at the grid size 𝑁 = 256. Their near resolution independence in￾dicates that the preferred scales are physical rather than…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

46 extracted references · 21 canonical work pages

  1. [1]

    A., & Hawley, J

    Balbus, S. A., & Hawley, J. F. 1991, Astrophysical Journal, Part 1 (ISSN 0004-637X), vol. 376, July 20, 1991, p. 214-233., 376, 214

    work page 1991

  2. [2]

    2018, Software

    Bradbury, J., Frostig, R., Hawkins, P., et al. 2018, Software. http://github.com/google/jax

    work page 2018

  3. [3]

    Burrows, A., Hayes, J., & Fryxell, B. A. 1995, The Astrophysical Journal, 450, 830, doi:10.1086/176188

    doi:10.1086/176188 1995

  4. [4]

    M., & Farrell, B

    Butler, K. M., & Farrell, B. F. 1992, Physics of Fluids A: Fluid Dynamics, 4, 1637, doi:10.1063/1.858386

    doi:10.1063/1.858386 1992

  5. [5]

    1961, Hydrodynamic and Hydromagnetic Stability (Oxford University Press)

    Chandrasekhar, S. 1961, Hydrodynamic and Hydromagnetic Stability (Oxford University Press)

    work page 1961

  6. [6]

    Chevalier, R. A. 1976, The Astrophysical Journal, 207, 872, doi: 10.1086/154549

    work page doi:10.1086/154549 1976

  7. [7]

    L., Dimits, A., et al

    Dimonte, G., Youngs, D. L., Dimits, A., et al. 2004, Physics of Fluids, 16, 1668, doi:10.1063/1.1688328

    doi:10.1063/1.1688328 2004

  8. [8]

    2009, Journal of Geophysical Research: Oceans, 114, C04022, doi:10.1029/2008JC004925 Faucher-Giguère, C.-A., & Oh, S

    Duan, W., Liu, X., Zhu, K., & Mu, M. 2009, Journal of Geophysical Research: Oceans, 114, C04022, doi:10.1029/2008JC004925 Faucher-Giguère, C.-A., & Oh, S. P. 2023, Annual Review of Astronomy and Astrophysics, 61, 131

    work page doi:10.1029/2008jc004925 2009

Show all 46 references
  1. [9]

    Field, G. B. 1965, The Astrophysical Journal, 142, 531, doi: 10.1086/148317

    1965 doi

  2. [10]

    2000, ACM Transactions on Mathematical Software, 26, 19, doi:10.1145/347837.347846

    Griewank, A., & Walther, A. 2000, ACM Transactions on Mathematical Software, 26, 19, doi:10.1145/347837.347846

    2000

  3. [11]

    1986, SIAM Journal on Numerical Analysis, 23, 707, doi:10.1137/0723046

    Grippo, L., Lampariello, F., & Lucidi, S. 1986, SIAM Journal on Numerical Analysis, 23, 707, doi:10.1137/0723046

    1986 doi

  4. [12]

    Gronke, M., & Oh, S. P. 2018, Monthly Notices of the Royal Astronomical Society: Letters, 480, L111, doi: 10.1093/mnrasl/sly131

    2018 doi

  5. [13]

    R., Millman, K

    Harris, C. R., Millman, K. J., Van Der Walt, S. J., et al. 2020, nature, 585, 357

    2020

  6. [14]

    Hunter, J. D. 2007, Computing in science & engineering, 9, 90 14 0.4 0.2 0.0 0.2 0.4 0.4 0.2 0.0 0.2 0.4 y T2_N128-90%Power-2D t = 0.4 0.4 0.2 0.0 0.2 0.4 t = 0.8 0.4 0.2 0.0 0.2 0.4 t = 1.2 0.4 0.2 0.0 0.2 0.4 t = 1.6 0.4 0.2 0.0 0.2 0.4 t = 2.0 0.4 0.2 0.0 0.2 0.4 0.4 0.2 0....

    2007

  7. [15]

    P., & Masterson, P

    Ji, S., Oh, S. P., & Masterson, P. 2019, Monthly Notices of the Royal Astronomical Society, 487, 737, doi: 10.1093/mnras/stz1248

    2019 doi

  8. [16]

    2018, Nature, 561, 498, doi:10.1038/s41586-018-0525-0

    Jiang, Y.-F., Cantiello, M., Bildsten, L., et al. 2018, Nature, 561, 498, doi:10.1038/s41586-018-0525-0

    2018 doi

  9. [17]

    L., & Stone, J

    Jun, B.-I., Norman, M. L., & Stone, J. M. 1995, The Astrophysical Journal, 453, 332, doi:10.1086/176393

    1995 doi

  10. [18]

    Juniper, M. P. 2011, Journal of Fluid Mechanics, 667, 272

    2011

  11. [19]

    Kerswell, R. R. 2018, Annual Review of Fluid Mechanics, 50, 319, doi: 10.1146/annurev-fluid-122316-045042

    2018 doi

  12. [20]

    2003, Astronomy & Astrophysics, 408, 621, doi: 10.1051/0004-6361:20030863

    Kifonidis, K., Plewa, T., Janka, H.-T., & Müller, E. 2003, Astronomy & Astrophysics, 408, 621, doi: 10.1051/0004-6361:20030863

    2003 doi

  13. [21]

    2026, The Astrophysical Journal, 1003, arXiv:2605.03540

    Luo, T.-X., Zhou, P., Ng, C.-Y., et al. 2026, The Astrophysical Journal, 1003, arXiv:2605.03540

    2026 arXiv

  14. [22]

    A., & Arnett, D

    Meakin, C. A., & Arnett, D. 2007, The Astrophysical Journal, 667, 448, doi:10.1086/520318

    2007 doi

  15. [23]

    Henningson, D. S. 2011, Physical Review Letters, 106, 134502, doi: 10.1103/PhysRevLett.106.134502

    2011 doi

  16. [24]

    2003, Chinese Science Bulletin, 48, 1045, doi: 10.1007/BF03184224

    Mu, M., & Duan, W. 2003, Chinese Science Bulletin, 48, 1045, doi: 10.1007/BF03184224

    2003 doi

  17. [25]

    2003, Nonlinear Processes in Geophysics, 10, 493, doi:10.5194/npg-10-493-2003

    Mu, M., Duan, W., & Wang, B. 2003, Nonlinear Processes in Geophysics, 10, 493, doi:10.5194/npg-10-493-2003

    2003 doi

  18. [26]

    2006, Journal of the Atmospheric Sciences, 63, 1587, doi:10.1175/JAS3703.1

    Mu, M., & Zhang, Z. 2006, Journal of the Atmospheric Sciences, 63, 1587, doi:10.1175/JAS3703.1

    2006 doi

  19. [27]

    Nocedal, J., & Wright, S. J. 2006, Numerical Optimization, 2nd edn. (Springer)

    2006

  20. [28]

    Parker, E. N. 1958, Physical Review, 109, 1874

    1958

  21. [29]

    Pringle, C. C. T., & Kerswell, R. R. 2010, Physical Review Letters, 105, 154502, doi:10.1103/PhysRevLett.105.154502

    2010 doi

  22. [30]

    Pringle, C. C. T., Willis, A. P., & Kerswell, R. R. 2012, Journal of Fluid Mechanics, 702, 415, doi:10.1017/jfm.2012.192

    2012 doi

  23. [31]

    1883, Proceedings of the London Mathematical Society, 14, 170, doi:10.1112/plms/s1-14.1.170

    Rayleigh, L. 1883, Proceedings of the London Mathematical Society, 14, 170, doi:10.1112/plms/s1-14.1.170

    doi

  24. [32]

    Richtmyer, R. D. 1960, Communications on pure and applied mathematics, 13, 297

    1960

  25. [33]

    Schmid, P. J. 2007, Annual Review of Fluid Mechanics, 39, 129, doi: 10.1146/annurev.fluid.38.050304.092139

    2007

  26. [34]

    2024, Nonlinear Processes in Geophysics, 31, 165

    Shi, B., & Ma, J. 2024, Nonlinear Processes in Geophysics, 31, 165

    2024

  27. [35]

    2023, Nonlinear Processes in Geophysics, 30, 263

    Shi, B., & Sun, G. 2023, Nonlinear Processes in Geophysics, 30, 263

    2023

  28. [36]

    2025, astronomix - differentiable MHD in JAX, Zenodo, doi:10.5281/ZENODO.17782162 15

    Storcks, L. 2025, astronomix - differentiable MHD in JAX, Zenodo, doi:10.5281/ZENODO.17782162 15

    2025 doi

  29. [37]

    P., & Gronke, M

    Tan, B., Oh, S. P., & Gronke, M. 2021, Monthly Notices of the Royal Astronomical Society, 502, 3179, doi: 10.1093/mnras/stab053

    2021 doi

  30. [38]

    Taylor, G. I. 1950, Proceedings of the Royal Society of London, Series A, 201, 192, doi:10.1098/rspa.1950.0052

    1950

  31. [39]

    F., Spruce, M., & Speares, W

    Toro, E. F., Spruce, M., & Speares, W. 1994, Shock Waves, 4, 25, doi: 10.1007/BF01414629

    1994 doi

  32. [40]

    N., Trefethen, A

    Trefethen, L. N., Trefethen, A. E., Reddy, S. C., & Driscoll, T. A. 1993, Science, 261, 578, doi:10.1126/science.261.5121.578

    1993 doi

  33. [41]

    J., ZuHone, J

    Turk, M., Goldbaum, N. J., ZuHone, J. A., et al. 2025, Introducing yt 4.0: Analysis and Visualization of Volumetric Data, Tech. rep., Manubot

    2025

  34. [42]

    J., Smith, B

    Turk, M. J., Smith, B. D., Oishi, J. S., et al. 2010, The Astrophysical Journal Supplement Series, 192, 9, doi: 10.1088/0067-0049/192/1/9

    2010 doi

  35. [43]

    E., et al

    Virtanen, P., Gommers, R., Oliphant, T. E., et al. 2020, Nature methods, 17, 261

    2020

  36. [44]

    Wang, C.-Y., & Chevalier, R. A. 2001, The Astrophysical Journal, 549, 1119, doi:10.1086/319439

    2001 doi

  37. [45]

    2018, The Astrophysical Journal, 857, 121

    Yuan, F., Yoon, D., Li, Y.-P., et al. 2018, The Astrophysical Journal, 857, 121

    2018

  38. [46]

    2025, The Astrophysical Journal, 985, 178

    Zhang, H., Xia, H., Ji, S., et al. 2025, The Astrophysical Journal, 985, 178

    2025

Pith tools

Reviewed August 8, 2026 · model on record in the stance chip above.