Pith. sign in

REVIEW 2 major objections 6 minor 25 references

A Unified Approach for Deriving Optimal Finite Differences

T0 review · 2 major / 6 minor · reviewed 2026-08-14 · deepseek-v4-flash

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

Pith's one-line read A unified optimization framework derives finite-difference schemes that meet order, spectral, and stability requirements together.

desk verdict A solid, clearly presented unification of order, spectral, and stability constraints for finite-difference design; the LMI stability constraint and symmetry proofs are new and the central claims hold up. read the letter →

arxiv 1908.02890 v1 pith:AL4OPSU7 submitted 2019-08-08 physics.comp-ph physics.flu-dyn

Komal Kumari , Raktim Bhattacharya , Diego A. Donzis This is my paper · ORCID
classification physics.comp-phphysics.flu-dyn MSC 65M0665M1290C2265D25
keywords finitedifferencesspectralaccuracyoptimizationstabilitymodifiedwavenumberlinearmatrixinequalitydispersionrelationcomputationalfluiddynamics
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 claims that the three requirements that usually shape a finite-difference scheme—formal order of accuracy, spectral (wavenumber-by-wavenumber) resolution, and stability of the fully discretized equation—can be folded into a single optimization problem. Minimizing a weighted $L^2$ spectral error under order-of-accuracy equality constraints and a stability inequality yields optimal coefficients analytically when stability is ignored, and numerically when it is included. The framework shows that spectrally optimal stencils are symmetric for even derivatives and antisymmetric for odd derivatives, and that imposing stability biases only the odd-derivative stencils. If the approach is right, finite-difference design becomes a systematic tradeoff: one can trade spectral accuracy for very large stable time steps, or decouple resolution from formal order to build low-order schemes that outperform high-order standard ones on multiscale problems.

What carries the argument

The load-bearing object is the weighted spectral-error functional $\|e(\eta)\|_{L^2}^2 = \int_0^\pi \gamma(\eta) |\sum_m a_m e^{jm\eta} - (j\eta)^d|^2 d\eta$, with $\gamma(\eta)$ a user-chosen weight over wavenumbers, together with the order-of-accuracy constraints $a_d^T X_d = y_d$ and, for fully discrete stability, the linear matrix inequality $\|I_N + \sum_d (\Delta t/\Delta x^d) \beta_d A_d^\Phi\|_2 \le 1$. The KKT system for the stability-free subproblem yields the optimal coefficients analytically, and the stability inequality turns the combined problem into a convex (semidefinite) optimization, Eq. (52), that can be solved with standard tools. The weighting function is what decouples spectral resolution from formal order: extra stencil points can be spent on resolving chosen wavenumber bands rather than raising truncation order.

What would settle it

Take the optimized M=4 second-order schemes from Fig. 9, choose a pair $(r_c, r_d)$ inside the claimed stable region, and compute the eigenvalues of $I + r_c A_1^\Phi + r_d A_2^\Phi$ directly; if any eigenvalue has modulus greater than 1, or if a fully discrete run of the advection-diffusion equation with those parameters grows in time, then the 2-norm stability guarantee is not capturing the true stability boundary and the large-time-step claim would need to be restated as a sufficient condition only.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that the optimal finite-difference coefficients for a derivative of order $d$ and a given stencil are the solution of the quadratic program in Eq. (52): minimize the weighted $L^2$ norm of the spectral error $e(\eta)=\sum_m a_m e^{jm\eta}-(j\eta)^d$ subject to order-accuracy constraints and, when full discretization is considered, a linear-matrix-inequality stability constraint obtained by bounding the 2-norm of the evolution operator. For the accuracy-plus-spectral problem, the KKT equations give the coefficients in closed form, and the paper proves two structural results: for even $d$ the optimal error is purely real, so the coefficients are symmetric, and for odd $d$ it is purely imaginary, so the coefficients are antisymmetric. When stability is added, the framework yields schemes that bias only the odd-derivative stencils—upwinding the convective term—while leaving even-derivative stencils symmetric, and it can produce explicit schemes whose guaranteed stable time steps are about an order of magnitude larger than those of standard schemes of the same order. The paper also shows numerically that optimized second-order schemes can reproduce the physics of diffusion, advection-diffusion, Burgers, and wave problems with accuracy comparable to or better than standard eighth-order schemes, especially at high wavenumbers.

Load-bearing premise

The load-bearing premise is that bounding the 2-norm of the evolution operator by one is a faithful enough proxy for true spectral-radius stability; if that bound is much looser than the actual stability limit for non-normal operators, then the very-large-time-steps result is a guaranteed lower bound rather than an exact characterization.

Editorial extensions

If this is right

  • Spectral resolution and formal order of accuracy cease to be tied: with a fixed stencil and low formal order, optimized schemes can resolve high wavenumbers far better than standard high-order schemes of the same stencil.
  • Stability-aware optimization produces upwinded first-derivative stencils and symmetric second-derivative stencils, giving a principled justification for the common practice of stabilizing convection while leaving diffusion central.
  • Explicit schemes can be designed to remain stable at time steps an order of magnitude larger than standard explicit schemes by accepting larger spectral error, which is directly useful for massively parallel simulations where implicit time stepping is costly.
  • Existing optimized schemes, such as dispersion-relation-preserving and low-dissipation low-dispersion explicit schemes, are recovered by the framework as special cases, so the framework unifies them rather than competing with them.
  • Semi-discrete stability of the optimized schemes depends only on the even-derivative terms of the PDE, so stability can be assessed and enforced term by term.

Reading between the lines

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

  • Beyond the paper: the same optimization with a non-constant $\gamma(\eta)$ could be used to design schemes that preserve a target dispersion relation over a prescribed band, effectively turning Eq. (52) into a dispersion-relation-preserving design tool with built-in stability.
  • Beyond the paper: since the 2-norm stability condition is sufficient but not necessary, replacing it with the exact spectral-radius condition or a pseudospectral bound would likely enlarge the guaranteed stable region in Figs. 8 and 9; this is a testable numerical extension using eigenvalue computations.
  • Beyond the paper: the framework's symmetry and antisymmetry proofs rely only on the form of the spectral error and the positivity of $\gamma$, so they should carry over to boundary stencils and to compact (implicit) finite differences, provided the stencil is expressed in the same linear form.
  • Beyond the paper: a natural next step, which the authors flag as ongoing, is to optimize space and time discretizations jointly; the current framework fixes the temporal scheme and would need a non-convex treatment to couple both.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

2 major / 6 minor

Summary. The paper proposes a unified optimization framework for deriving finite-difference schemes that simultaneously accounts for formal order of accuracy, spectral resolution, and stability. Order conditions are written as linear equality constraints, spectral error is minimized as a weighted L2 norm in wavenumber space, and stability of the fully discrete scheme is imposed through a matrix 2-norm bound expressed as a linear matrix inequality. For the accuracy-plus-spectral subproblem, the authors derive an analytical KKT solution and prove that the optimal stencil is symmetric for even derivatives and antisymmetric for odd derivatives, yielding purely dissipative or purely dispersive spectral errors, respectively. Stability is then treated either by maximizing the time step for a fixed optimal stencil or by optimizing stencil coefficients for a fixed time step under the stability constraint. The framework is applied to diffusion, linear advection-diffusion, Burgers, and wave equations, with comparisons against standard finite differences and analytical solutions. The central claim is that the most general formulation, Eq. (52), is a well-posed convex problem whose solutions expose exploitable tradeoffs between accuracy and stability, including explicit schemes stable at very large time steps.

Significance. If the claims hold, this is a valuable contribution to computational physics: it provides a mathematically clean, problem-specific way to design finite-difference operators, and it unifies strands of the optimized-scheme literature (e.g., dispersion-relation-preserving and bandwidth-optimized schemes) under one convex formulation. The analytical KKT solution and the symmetry/antisymmetry lemmas are genuine structural results, and the numerical tests against analytical solutions of the diffusion, advection-diffusion, and Burgers equations provide credible support for the performance claims. The paper also correctly identifies and demonstrates an interesting tradeoff: for fixed formal order, additional stencil points can be spent on stability rather than on spectral accuracy, with the odd-derivative stencil absorbing the bias. The main reservations are a gap in the justification that the stability-constrained problem reduces to O(Sd) degrees of freedom, and a mathematical misstatement in the stability condition as written.

major comments (2)
  1. [Sec. 2.3.4, Eq. (52); Lemma 3] The assertion that 'as a consequence of Lemma 3 for periodic domains, the optimization problem in Eq. (52) has a total of n = Sd degrees of freedom' is not justified by Lemma 3 as stated and proved. Appendix C proves uniformity of the optimal stencil only for the equality-constrained quadratic program (46)-(48), whose KKT system does not include the stability LMI (45). Since the objective and the stability constraint in Eq. (52) are invariant under cyclic permutation of the grid-point rows, one can in fact recover a uniform-row minimizer by averaging over cyclic shifts, but this argument is not given. Please either supply this missing argument or reformulate Eq. (52) directly over a single stencil and state that the periodic stability analysis is performed on the resulting circulant operator. Without this, the claimed O(Sd) computational cost and the interpretation of Figs. 9-11 are not fully supported.
  2. [Sec. 2.3.2, Eq. (43)] The stability condition is written as lambda_max(I_N + sum_d (Delta t/(Delta x)^d) beta_d A_d^Phi) <= 1. For the non-symmetric evolution operator in Eq. (42), in particular for the advective (skew-symmetric) part, eigenvalues need not be real, so 'lambda_max' is not defined and this inequality is neither necessary nor sufficient for the spectral-radius condition needed for bounded powers of the evolution operator. The subsequent replacement by the 2-norm bound in Eq. (44) is correct and is what the numerical results actually use, but Eq. (43) should be corrected to rho(...) <= 1, with the usual caveat about semisimple unit-modulus eigenvalues.
minor comments (6)
  1. [Appendices A and B, Sec. 4] The proofs of Lemmas 1 and 2 conclude that the antisymmetric (or symmetric) part vanishes from integral conditions involving S^T a or C^T a. This conclusion requires gamma to be positive, or at least supported on a set of positive measure where the finite trigonometric sums are linearly independent. The examples with gamma = 0 outside [0, 2.5] satisfy this, but the statement in Section 4 that symmetry holds 'regardless of the functional form of gamma(eta)' is too broad as written; the required hypothesis on gamma should be stated in the lemmas.
  2. [Sec. 2.3.2, Eq. (44)] The paper should state explicitly that for the periodic uniform grids used in Figs. 8-10 the evolution operator is circulant and hence normal, so ||G||_2 = rho(G) and the 2-norm LMI is not conservative for those cases. This would clarify why the stability regions are tight, and it would underline the caveat, already given for multistep schemes in Sec. 2.3.5, that the 2-norm bound can be conservative for the non-periodic extension sketched in Sec. 4.
  3. [Sec. 3.2] The text near Fig. 14 refers to 'SDF2' and 'SDF8'; these should be 'SFD2' and 'SFD8' for consistency with the rest of the paper.
  4. [Sec. 3.2] The statement that the framework reproduces the schemes of [3,8] is made only qualitatively. A short table or error comparison would give readers a concrete way to verify that claim.
  5. [Throughout] There are numerous typographical errors and inconsistencies (e.g., 'exmaple', 'dicrete', 'dissiption', 'meangingful', 'stems for' instead of 'stems from'). A careful proofreading pass is needed before publication.
  6. [Sec. 3] No code or data repository is provided. Since the paper's central tool is a convex optimization formulation solved with cvx, releasing the scripts and full coefficient tables would materially improve reproducibility and would allow readers to test the stability-region claims independently.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found: the optimization framework is self-contained and its derived properties follow from the stated minimization and constraints.

full rationale

The paper's central claim is that Eq. (52) minimizes a weighted L2 spectral error, defined independently in Eq. (19) as the difference between the modified and exact wavenumber responses, subject to the order-accuracy equalities Eq. (30) and the stability LMI Eq. (45). The coefficients are not fitted to any target output; the KKT solution Eq. (25) follows from positive definiteness of Qd and full row rank of Xd, and Lemmas 1 and 2 are proven in Appendices A and B by decomposing stencils into symmetric and antisymmetric parts. The stability regions in Figs. 8-10 are generated from the LMI condition, not from matching a desired stability boundary, and the numerical tests in Sec. 3 compare against analytical solutions (Eqs. 62, 64, 67) rather than against the optimization objective. The only backward-looking validation, the statement in Sec. 3.2 that optimized fourth-order schemes reproduce the coefficients of Refs. [3,8], is an external check, not an input to the derivation. No self-citation is load-bearing, and no fitted parameter is renamed as a prediction. The acknowledged conservativeness of the 2-norm bound in Sec. 2.3.5 is a limitation, not a circularity.

Assumptions & free parameters 0 free parameters · 5 assumptions · 0 invented entities

The paper introduces no new physical entities and no fitted constants. Its user-selected weighting function gamma(eta), the target wavenumber interval, and the order and stencil choices are design inputs, not parameters fitted to reproduce the results. The analytical claims rest on standard Fourier analysis, KKT theory, and an LMI stability reformulation, plus periodic-domain assumptions.

assumptions (5)
  • standard math Fourier/modal analysis defines spectral error via modified wavenumber (Section 2.2).
    The framework treats spatial derivatives through their action on Fourier modes, which is the standard von Neumann or modified-wavenumber framework cited in the paper.
  • standard math Karush-Kuhn-Tucker conditions solve the equality-constrained quadratic program (Eq. 24).
    The analytical optimal coefficients rely on KKT conditions for a convex QP with full-row-rank constraint matrix X_d.
  • standard math Schur complement equivalence for the LMI in Eq. (45).
    The 2-norm stability condition is rewritten as a block matrix inequality using a standard Schur complement argument.
  • domain assumption Uniform grid and periodic domain for all analytical results (Section 2.3.4 and Appendix D).
    The shift-operator construction and Lemma 3's grid-point invariance use periodic boundary conditions; non-periodic boundaries are deferred to future work in Section 4.
  • domain assumption The matrix 2-norm bound is a sufficient proxy for spectral-radius stability (Eq. 44).
    Stability is guaranteed by bounding the 2-norm rather than the exact spectral radius, making the reported stability regions conservative.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Unified Approach for Deriving Optimal Finite Differences." pith.science (2026). https://pith.science/paper/AL4OPSU7

@misc{pith2026190802890,
  author       = {Pith},
  title        = {Pith review of: A Unified Approach for Deriving Optimal Finite Differences},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/AL4OPSU7}},
  note         = {Machine review of arXiv:1908.02890}
}
read the original abstract

A unified approach to derive optimal finite differences is presented which combines three critical elements for numerical performance especially for multi-scale physical problems, namely, order of accuracy, spectral resolution and stability. The resulting mathematical framework reduces to a minimization problem subjected to equality and inequality constraints. We show that the framework can provide analytical results for optimal schemes and their numerical performance including, for example, the type of errors that appear for spectrally optimal schemes. By coupling the problem in this unified framework, one can effectively decouple the requirements for order of accuracy and spectral resolution, for example. Alternatively, we show how the framework exposes the tradeoffs between e.g. accuracy and stability and how this can be used to construct explicit schemes that remain stable with very large time steps. We also show how spectrally optimal schemes only bias odd-order derivatives to remain stable, at the expense of accuracy, while leaving even-order derivatives with symmetric coefficients. Schemes constructed within this framework are tested for diverse model problems with an emphasis on reproducing the physics accurately.

Figures

Figures reproduced from arXiv: 1908.02890 by the authors.

Figure 1
Figure 1. Optimal coefficients for various stencil sizes [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Spectral accuracy for the schemes presented in Fig. 1 and Table 1. Top, middle and bottom rows show the modified [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗
Figure 3
Figure 3. Optimal spectral error for various stencil size [PITH_FULL_IMAGE:figures/full_fig_p009_3.png] view at source ↗
Figures from the paper (16 more)
Figure 5
Figure 5. Figure 5: Similar conclusions to the second derivative example shown above are observed. As [PITH_FULL_IMAGE:figures/full_fig_p009_5.png]
Figure 4
Figure 4. Figure 4: Optimal coefficients for various stencil size [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Spectral accuracy for the schemes presented in Fig. 4 and Table 2. Top, middle and bottom rows show the modified [PITH_FULL_IMAGE:figures/full_fig_p011_5.png]
Figure 6
Figure 6. Figure 6: Optimal spectral error for various stencil size [PITH_FULL_IMAGE:figures/full_fig_p011_6.png]
Figure 7
Figure 7. Figure 7: Spectral error for various γ(η) for second derivative. Standard eighth order scheme (black), γ(η) = 1 (magenta), γ(η) = exp(−η/0.1) (blue) and γ(η) = exp(−η/0.06) (red) for η ∈ [0, 2.5] and 0 otherwise. Different methods to assess stability of a given discretized equat…
Figure 8
Figure 8. Figure 8: Stability region for the advection-diffusion equation with optimal coefficients [PITH_FULL_IMAGE:figures/full_fig_p016_8.png]
Figure 9
Figure 9. Figure 9: Tradeoff between stability and spectral accuracy with respect to ( [PITH_FULL_IMAGE:figures/full_fig_p018_9.png]
Figure 10
Figure 10. Figure 10: Contours of measure A for M = 2 for (a) first derivative (b) second derivative. (c) Optimal coefficients for different values of rc and rd as marked, with the same symbols, in (a). 0 1 2 3 -4 -3 -2 -1 0 1 0 1 2 3 0 1 2 3 4 η η a) (b)( <[e(η)] =[e(η)] [PITH_FULL_IMAGE…
Figure 11
Figure 11. Figure 11: Dispersion error (a) and dissipation error (b) for the first derivative with M = 2 at different values of rc and rd as marked with the same color in [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Normalized error in spectral energy for diffusion equation. Solid lines correspond to semi-discrete integration [PITH_FULL_IMAGE:figures/full_fig_p022_12.png]
Figure 13
Figure 13. Figure 13: Normalized error in spectral energy for advection-diffusion equation. Different lines correspond to different numerical [PITH_FULL_IMAGE:figures/full_fig_p023_13.png]
Figure 14
Figure 14. Figure 14: (a) Ratio of numerical to actual speed. (b) Normalized phase error for advection-diffusion equation. Different lines [PITH_FULL_IMAGE:figures/full_fig_p024_14.png]
Figure 15
Figure 15. Figure 15: Evolution of space-averaged energy normalized by the initial space-averaged energy with normalized time [PITH_FULL_IMAGE:figures/full_fig_p025_15.png]
Figure 16
Figure 16. Figure 16: Energy spectrum for the non-linear advection-diffusion equation at [PITH_FULL_IMAGE:figures/full_fig_p026_16.png]
Figure 17
Figure 17. Figure 17: (a) Evolution of space-averaged dissipation (b) Evolution of space-averaged fourth order moment of velocity gradient [PITH_FULL_IMAGE:figures/full_fig_p026_17.png]
Figure 18
Figure 18. Figure 18: Solution of the wave-equation for t = 2 with a Gaussian initial condition. Different lines correspond to different numerical schemes: SFD8 with RK4 in time (black), OFD2 with RK2 in time (magenta), and SFD2 with RK2 in time (red). The blue dashed-star line is the anal…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

25 extracted references · 25 canonical work pages

  1. [1]

    Hirsch, Numerical computation of internal and external flows, volume 1, Wiley, New York, 1994

    C. Hirsch, Numerical computation of internal and external flows, volume 1, Wiley, New York, 1994

    work page 1994

  2. [2]

    S. K. Lele, Compact finite-difference schemes with spectral-like resolution, J. Comp. Phys. 103 (1992) 16–42

    work page 1992

  3. [3]

    C. K. Tam, J. C. Webb, Dispersion-relation-preserving finite difference schemes for computational acoustics, J. Comp. Phys. 107 (1993) 262 – 281

    work page 1993

  4. [4]

    J. W. Kim, D. J. Lee, Optimized compact finite difference schemes with maximum resolution, AIAA Journal 34 (1996) 887–893

    work page 1996

  5. [5]

    Z. Wang, R. Chen, Optimized weighted essentially nonoscillatory schemes for linear waves with discon- tinuity, J. Comp. Phys. 174 (2001) 381 – 404

    work page 2001

  6. [6]

    Zhuang, R

    M. Zhuang, R. F. Chen, Applications of high-order optimized upwind schemes for computational aeroacoustics, AIAA Journal 40 (2002) 443–449

    work page 2002

  7. [7]

    Ponziani, S

    D. Ponziani, S. Pirozzoli, F. Grasso, Development of optimized weighted-ENO schemes for multiscale compressible flows, Int. J. Numer. Meth. Fluids 42 (2003) 953–977

    work page 2003

  8. [8]

    Bogey, C

    C. Bogey, C. Bailly, A family of low dispersive and low dissipative explicit schemes for flow and noise computations, J. Comp. Phys. 194 (2004) 194 – 214

    work page 2004

Show all 25 references
  1. [9]

    M. P. Martin, E. M. Taylor, M. Wu, V. G. Weirs, A bandwidth-optimized WENO scheme for the effective direct numerical simulation of compressible turbulence, J. Comp. Phys. 220 (2006) 270–289

    2006

  2. [10]

    T. K. Sengupta, S. K. Sircar, A. Dipankar, High accuracy schemes for DNS and acoustics, Journal of Scientific Computing 26 (2006) 151–193

    2006

  3. [11]

    J. Fang, Z. Li, L. Lu, An optimized low-dissipation monotonicity-preserving scheme for numerical simulations of high-speed turbulent flows, J. Sci. Comput. 56 (2013) 67–95

    2013

  4. [12]

    Zhang, Z.-X

    J.-H. Zhang, Z.-X. Yao, Optimized explicit finite-difference schemes for spatial derivatives using maxi- mum norm, J. Comp. Phys. 250 (2013) 511 – 526

    2013

  5. [13]

    Pirozzoli, Performance analysis and optimization of finite-difference schemes for wave propagation problems, J

    S. Pirozzoli, Performance analysis and optimization of finite-difference schemes for wave propagation problems, J. Comp. Phy. 222 (2007) 809–831

    2007

  6. [14]

    Ashcroft, X

    G. Ashcroft, X. Zhang, Optimized prefactored compact schemes, J. Comp. Phys. 190 (2003) 459 – 477

    2003

  7. [15]

    Haras, S

    Z. Haras, S. Ta’asan, Finite difference schemes for long-time integration, J. Comp. Phy. 114 (1994) 265 – 279

    1994

  8. [16]

    Sengupta, G

    T. Sengupta, G. Ganeriwal, S. De, Analysis of central and upwind compact schemes, J. Comp. Phy. 192 (2003) 677 – 694

    2003

  9. [17]

    T. K. Sengupta, A. Bhole, Error dynamics of diffusion equation: Effects of numerical diffusion and dispersive diffusion, Journal of Computational Physics 266 (2014) 240 – 251

    2014

  10. [18]

    D. P. Bertsekas, Nonlinear programming, Athena scientific Belmont, 1999

    1999

  11. [19]

    Grant, S

    M. Grant, S. Boyd, Y. Ye, CVX: Matlab software for disciplined convex programming, 2008

    2008

  12. [20]

    Ben-Tal, A

    A. Ben-Tal, A. Nemirovski, Lectures on modern convex optimization: analysis, algorithms, and engi- neering applications, SIAM, 2001. 34

    2001

  13. [21]

    Stoer, R

    J. Stoer, R. Bulirsch, Introduction to numerical analysis, volume 12, Springer Science & Business Media, 2013

    2013

  14. [22]

    R. Y. Zhang, J. Lavaei, Efficient algorithm for large-and-sparse LMI feasibility problems, IEEE CDC (2018) 6868–6875

    2018

  15. [23]

    Eberhard, The partial differential equation ut +uux =µuxx, Communications on Pure and Applied Mathematics 3 (1950) 201–230

    H. Eberhard, The partial differential equation ut +uux =µuxx, Communications on Pure and Applied Mathematics 3 (1950) 201–230

    1950

  16. [24]

    J. D. Cole, On a quasi-linear parabolic equation occurring in aerodynamics, Quarterly of Applied Mathematics 9 (1951) 225–236

    1951

  17. [25]

    Shokin, K

    Y. Shokin, K. Roesner, The Method of Differential Approximation, Scientific Computation, Springer Berlin Heidelberg, 2011. 35

    2011

Pith tools

Reviewed August 14, 2026 · model on record in the stance chip above.