Pith. sign in

REVIEW 4 major objections 6 minor 1 cited by

A Variational Framework for the Complexity of PDE Solutions

T0 review · 4 major / 6 minor · reviewed 2026-08-04 · deepseek-v4-flash

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

Pith's one-line read This paper argues that the digital complexity of a PDE solution is set by its regularity: analytic solutions are polynomial-time computable, while non-analytic solutions provably require super-polynomial time.

desk verdict The upper-bound framework is interesting and plausible, but the complexity-blowup results rest on a false equivalence between sub-exponential approximation error and super-polynomial operation count. read the letter →

arxiv 2510.21290 v3 pith:Z64BEASG submitted 2025-10-24 math.NA cs.NA

Juan Esteban Suarez Cardona , Holger Boche , Gitta Kutyniok This is my paper
classification math.NAcs.NA MSC 65Y2068Q2535A15
keywords complexityblowupPDEsolutionvariationalformulationsgradientflowscomputabilityChebyshevapproximationSobolevspacesEikonalequation
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

The paper establishes that the computational difficulty of solving a partial differential equation on a digital computer is controlled by the regularity of its solution, not just by the posedness of the equation. The authors build a variational least-squares loss whose minimizer solves the PDE, run an explicit Euler gradient flow in polynomial surrogate spaces, and convert the convergence rate of that flow into an operation-count bound for approximating the solution to precision 2^{-N}. Their central result (Corollary 6.16) says: if the source and boundary data are analytic but the PDE solution is only in H^k (with k>d/2) and not analytic, then the solution is H^k-computable but only in super-polynomial time — a complexity blowup. They demonstrate the dichotomy with the Poisson equation (polynomial-time solvable under analytic data) and the Eikonal equation (H^1-computable but blowup, because its viscosity solution has a kink along the medial axis). This matters because it ties an intrinsic property of the solution — its analyticity — to the inherent feasibility of computing it.

What carries the argument

The central object is the PDE learning loss L[u]=||N[u]-f||^2_{H^a(Ω)} + ||u-g||^2_{H^b(∂Ω)}, minimized over polynomial surrogates by the explicit gradient flow θ_{j+1}=θ_j - δτ ∇L_n[û_{θ_j}]. The load-bearing identity is the error decomposition ||u*-û_{θ_j}||^2_{H^k} ≲ ε_approx(m) + ε_int(n) + ε_opt(j), which separates the finite-dimensional surrogate error, the Sobolev-cubature quadrature error, and the gradient-flow optimization error. The paper bounds the first two by classical polynomial approximation rates (exponential for analytic functions, algebraic for bounded-variation functions) and the third by the exponential rate (1-μσ/L)^j of a strongly convex gradient flow. The conduit from

What would settle it

Run the paper's discrete gradient flow on the Eikonal equation on a square with a smooth interior obstacle, and count floating-point operations needed to reach H^1 error ≤ 2^{-N}. If the operation count grows polynomially in N despite a sub-exponential decay of the polynomial coefficients, the claimed super-polynomial blowup is falsified. Alternatively, compute the polynomial coefficients of the signed-distance function: if they decay like ρ^{-n} for some ρ>1, the reverse theorem would imply analyticity, contradicting the kink premise.

Watch

Extended reading notes

Core claim

The paper's central claim is that analyticity of the solution is the dividing line for polynomial-time computability: regularity-preserving operators with analytic data yield exponentially convergent polynomial surrogates and polynomial-time H^k-computability, while operators that destroy regularity push the solution into a regime where the Chebyshev approximation error can decay at best sub-exponentially, which the paper equates with super-polynomial operation count. The mechanism runs through a reverse approximation theorem: exponential polynomial accuracy forces an analytic extension into a Bernstein ellipse; since a non-analytic H^k solution cannot have such an extension, its approximati

Load-bearing premise

The load-bearing premise is that a sub-exponential lower bound on the approximation error, of the form Q(n)=O(2^{-n^α}) with α<1, necessarily forces a super-polynomial operation count S(N); but solving 2^{-n^α} ≤ 2^{-N} requires n ≥ N^{1/α}, and if each step costs polynomially in n the total cost is polynomial in N — so the super-polynomiality conclusion relies on this equivalence holding.

Editorial extensions

If this is right

  • If true, the Eikonal equation becomes the first concrete nonlinear PDE whose solution is proved H^1-computable yet not polynomial-time computable, formalizing why geometric-optics solvers must cope with kinks.
  • The framework gives a constructive error budget for PDE learning: achieving precision 2^{-N} costs O(N log(1/(1-μσ/L))) gradient steps and O(F(N^{d/k'})) operations per step, so polynomial-time solvability is guaranteed when the data and solution operator are analytic.
  • For the Poisson equation with analytic source and boundary data, the result upgrades the existing complexity picture in the hypercube: solutions are H^1-computable in polynomial time under compatibility conditions.
  • The framework predicts that any PDE whose solution operator loses regularity at co-dimension-one sets (kinks, shocks, rays) will exhibit complexity blowup, so the phenomenon is structural rather than tied to a specific equation.
  • The sufficient conditions (convexity, coercivity, quadratic growth, Lipschitz gradient) are satisfied by many least-squares formulations, so the computability theorem extends beyond the two worked examples.

Reading between the lines

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

  • The regularity-to-complexity transfer suggests a practical heuristic: before designing an approximation scheme, probe the decay of polynomial coefficients of the solution; sub-exponential decay signals that spectral surrogates will become expensive, and adaptive or localized bases may be needed.
  • The framework's lower-bound mechanism only applies when k > d/2, where H^k embeds into C^0; for low-regularity settings in high dimensions, the blowup may not be detectable by this argument, and the paper explicitly notes this restriction.
  • A testable extension: for a nonlinear PDE with analytic data and a conjectured kink solution (e.g., an eikonal or Burger-type equation), numerically compute the polynomial coefficient decay and the actual operation count; sub-exponential decay with polynomial operation count would undermine the super-polynomiality conclusion.
  • The error decomposition could be reused as an a-posteriori certification tool: given a computed surrogate, estimate the three error terms separately, and the dominant term tells the user whether to refine the basis, the quadrature, or the optimizer.
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

4 major / 6 minor

Summary. The paper proposes a variational (least-squares) framework for analyzing the Turing computability and computational complexity of PDE solutions. The main idea is to approximate the solution of a PDE by minimizing a loss L of residual plus boundary terms, then to track the error of a discrete (Euler) gradient flow in a finite-dimensional polynomial surrogate space. The authors give an error decomposition into approximation, integration, and optimization errors (Theorem 4.12), prove sufficient conditions for H^k-computability and polynomial-time computability under regularity preservation (Theorem 6.13 and Corollary 6.16), and claim a 'complexity blowup' for non-analytic solutions. Two examples are presented: the Poisson equation (polynomial-time computability under analytic data) and the Eikonal equation (claimed complexity blowup due to non-analyticity of the signed-distance solution).

Significance. If the central claims were correct, the paper would offer a useful bridge between PDE regularity, polynomial approximation, and computational complexity, and would provide a new class of PDEs with provable super-polynomial solution complexity. The framework is ambitious and connects several active strands of research. However, the main complexity-blowup conclusion rests on an incorrect equivalence in Definition 6.4, and the central error-decomposition theorem is stated without a proof that is self-contained or matched to its assumptions. The Eikonal example does not satisfy the hypotheses of the theorem used to prove blowup. These are load-bearing issues, not presentation defects. The paper does not ship machine-checked proofs or code, and several key results are deferred to the first author's dissertation and prior papers.

major comments (4)
  1. [Definition 6.4, Eq. (45)-(46); Corollary 6.16] The asserted equivalence between a sub-exponential lower bound Q(n)=O(2^{-n^α}) and super-polynomial operation count S(N) with N^r=O(S(N)) for all r is false. Here n is the polynomial degree, not the number of operations. Reaching precision 2^{-N} with squared error at least 2^{-n^α} forces n=Θ(N^{1/α}); with the paper's own cost convention (O(n^d) operations per evaluation/gradient step, cf. the proof of Theorem 6.13), the total cost is O(N^{d/α}), which is polynomial in N. Thus the lower bound in Eq. (46) is consistent with polynomial-time computation. The phrase 'Q(n)=O(2^{-n^α})' is also vacuous as a lower bound because O is an upper bound and Q≡0 satisfies it. This incorrect equivalence is precisely what converts the non-analyticity lower bound in the second half of Corollary 6.16 into a claim of super-polynomial time. Removing it leaves, at best, a statement about the absence of ex
  2. [Theorem 4.12, Eq. (23)-(25)] The error decomposition is the foundation of the whole complexity analysis, but as stated it is not proved. The proof line 'follows directly from the quadratic growth condition together with the triangle inequality; see [12]' refers to the first author's dissertation, and the assumptions do not obviously justify the decomposition. In particular, û*_θ is introduced as a minimizer of L_n on S_θ(Ω), while u* minimizes L on H(Ω); the term ϵ_int(n)=|L[û*_θ]-L_n[û*_θ]| need not be controlled by the QGC for L, since QGC only relates L[u]-L[u*] to ||u-u*||^2. Moreover, exponential convergence of the explicit Euler flow is claimed from Lemma 4.11 and [24], but Lemma 4.11 requires convexity of L and QGC, and the RSI constant of L_n may differ from that of L. Since Theorem 6.13 and Corollary 6.16 use Eq. (23) directly, this gap is load-bearing.
  3. [Theorem 7.8; Assumption 6.12] The Eikonal example applies Corollary 6.16 without verifying its hypotheses. The paper itself notes that the gradient ∇L_n is not globally Lipschitz and that Theorem 6.13 cannot be applied directly, so the proof switches to an Euler subgradient flow with only O(√j) convergence. But Corollary 6.16 assumes all the conditions of Theorem 6.13, including QGC, RSI, Lipschitz smoothness, and exponential optimization convergence. In addition, the Eikonal loss L[u]=|| |∇u|-1 ||^2_{L2}+||u|_S||_{H^{1/2}(S)} is not convex and not differentiable in the standard sense, so the QGC/RSI framework is not in force. Consequently, the claimed super-polynomial lower bound for the Eikonal solution is unsupported. Even if one accepted the non-analyticity of the signed-distance function, Major Comment 1 shows that non-analyticity alone does not imply super-polynomial cost.
  4. [Proof of Theorem 6.13] The displayed modulus in the proof is internally inconsistent. With ϵ_app(n)+ϵ_int(n)=O(n^{-k'+k}+n^{-k*}) and k'>k, obtaining error 2^{-N} requires n ≳ 2^{N/(k'-k)}, but the proof states e_n(N)=max{2^{N/(k-k')}, 2^{N/k*}}, where k-k' is negative, so the first term does not have the stated asymptotic behavior. The same negative exponent appears in the operation-count expression O(F(max{2^{dN/(k-k')},2^{dN/k*}})). This is likely a typographical slip, but it occurs in the central polynomial-time computability proof and should be corrected before the theorem can be considered rigorous.
minor comments (6)
  1. [Definition 6.4] The lower-bound condition in Eq. (46) should use Ω(2^{-n^α}) or a lower-case notation, not O(2^{-n^α}); as written, the zero function satisfies it and the lower bound is vacuous.
  2. [Definition 6.10 vs. Remark 6.11] Definition 6.10 defines complexity blowup in terms of polynomial-time input data and super-polynomial-time computability of the solution, but Remark 6.11 expresses blowup via a lower bound Q and upper bound V on the iterate error. These are different criteria, and the role of Q in Remark 6.11 is not clearly aligned with Definition 6.4.
  3. [Theorem 7.3 proof] The proof invokes 'hypo-ellipticity' to conclude that the solution inherits regularity of the data. This is not justified for the Dirichlet problem on a domain with corners, and Remark 7.4 already acknowledges that corner singularities may break the inheritance. The use of hypo-ellipticity should be replaced with a precise regularity statement under the stated compatibility assumptions.
  4. [Lemma 7.5] The matrix W_Ω is defined twice in Eq. (62), and the second definition is identical to the first; one occurrence is presumably W_∂Ω. Please correct the typography.
  5. [References and deferred proofs] Several central results (Theorem 4.12, Propositions 5.5 and 5.6, Theorem 5.12) are deferred to the dissertation [12] or to prior work. For a journal submission, at least the new error decomposition of Theorem 4.12 should be proved in the paper, since it is the basis of the complexity estimates.
  6. [Conclusion, last paragraph] The limitation paragraph says the error decomposition 'provides only an upper bound on the solution's complexity,' but the paper also derives lower bounds using Corollary 5.8 and Definition 6.4. The wording is confusing and should clarify that the lower bounds come from approximation theory, not from the optimization error decomposition.

Circularity Check

1 steps flagged · score 8.0 of 10

Corollary 6.16's complexity blowup is installed by Definition 6.4, not derived: a sub-exponential error lower bound in polynomial degree is declared equivalent to a super-polynomial operation count.

  1. self definitional [Definition 6.4; used in Corollary 6.16 and Theorem 7.8]
    "there exists a sub-exponential function Q:N→R+, with Q(n)=O(2^{−n^α}), for some 0<α<1, such that ||qn−f||²_X ≥ Q(n), ∀n≥N. Equivalently, the function f can be approximated with precision 2^{−N} in at least S(N) operations, where S:N→N is a super-polynomial function satisfying N^r = O(S(N)) for all r∈N."

    The derivation of blowup in Corollary 6.16 is exactly this 'Equivalently': non-ρ-analyticity gives a lower bound Q(n) on the degree-n polynomial approximation error; Definition 6.4 converts that bound into 'at least super-polynomial operations'. But n is the polynomial degree, not the operation count. Reaching precision 2^{-N} with Q(n)=Ω(2^{-n^α}) only requires n=O(N^{1/α}), and with O(n^d) arithmetic per degree step the total cost is O(N^{d/α}), which is polynomial. Thus the super-polynomial conclusion is not a consequence of the lower bound; it is attached to the lower bound by definition. Read literally, Q(n)=O(2^{-n^α}) is an upper bound and is even vacuous.

full rationale

The central lower-bound half of the paper—Corollary 6.16 and its Eikonal application in Theorem 7.8—reduces to the asserted equivalence in Definition 6.4. The mathematically valid statement obtained from Theorem 5.7 is only a sub-exponential lower bound on the degree-n polynomial approximation error of a non-analytic solution. That statement does not imply a super-polynomial operation count, because n is the polynomial degree and the cost per degree step is polynomial in n. Therefore 'H^k-computable in at least super-polynomial time' is not derived; it is installed by the 'Equivalently' clause. Separate issues are also present: the error decomposition (Theorem 4.12) and Sobolev cubature rates (Theorem 5.12) are deferred to the first author's own dissertation and prior papers, and Theorem 7.8 invokes Corollary 6.16 after acknowledging that the Eikonal loss is not globally Lipschitz and that Theorem 6.13 cannot be applied directly. Those issues compound the problem, but the definitional step alone is load-bearing. Because the flagship complexity-blowup result is forced by a definitional identification rather than by a valid operation-count derivation, I score the circularity as 8.

Assumptions & free parameters 2 free parameters · 8 assumptions · 0 invented entities

The paper introduces no new physical or mathematical objects. The main external inputs are standard approximation-theory results (Chebyshev rates, Sobolev embedding), standard gradient-descent convergence bounds, and the authors' own prior work on Sobolev cubatures and error decompositions. The only ad hoc-to-paper item is the definitional equivalence in Definition 6.4, which drives the blowup claims and is not valid.

free parameters (2)
  • boundary-penalty weight n in discrete Poisson loss = n (polynomial degree)
    In Lemma 7.5, the discrete loss L_n is defined with an extra factor n multiplying the boundary term; this factor is chosen by hand so that λ-convexity follows from the inverse inequality ||Q_n||_{H^{1/2}(∂Ω)} ≤ n^{1/2}||Q_n||_{L^2(∂Ω)}. It is not present in the continuous loss (Definition 7.1) and is not derived from the PDE.
  • sub-exponential bound function Q in Definition 6.4 = O(2^{-n^α}), 0<α<1
    The lower-bound function Q is chosen by hand to define 'super-polynomial time'; the choice makes the blowup conclusions appear, and its equivalence to super-polynomial operation counts is asserted without proof (and is false).
assumptions (8)
  • standard math Chebyshev series converge exponentially for ρ-analytic functions and algebraically for functions in AC^{k-1}∩BV^k (Propositions 5.5, 5.6).
    Invoked in Theorem 6.13 and Corollary 6.16 to turn regularity into approximation error bounds; cited to [26], [28], [12].
  • standard math Sobolev embedding H^k(Ω) ⊂ C^0(Ω) for k>d/2.
    Used in Corollary 6.16 proof to pass from H^k convergence lower bound to C^0 lower bound (equation (56)).
  • standard math Existence of a minimizer of the loss via Tonelli's direct method whenever L is convex, coercive, lower semi-continuous.
    Used in Theorem 6.13 and Theorem 7.8 proofs to guarantee u*.
  • domain assumption The loss L satisfies Assumption 6.12: differentiable, convex, coercive, QGC, Lipschitz gradient, and C^0-computable N[u_n], ∇N[u_n].
    This is the main structural assumption of Theorem 6.13; it is not verified for the Eikonal loss, where the paper itself notes ∇L_n is not globally Lipschitz and convexity is not established.
  • domain assumption If input data f,g are ρ-analytic and the operator is regularity-preserving, then the solution u* is ρ*-analytic.
    Used in Corollary 6.16 first part and Theorem 7.3; Remark 7.4 admits this fails for the hypercube corners without compatibility conditions.
  • domain assumption The signed distance function is the relevant minimizer of the Eikonal forward loss (68).
    Used in Theorem 7.8 proof; cited to [15,42], but the L^2 least-squares loss with the ℓ1-norm term may have many minimizers and no convexity is shown.
  • ad hoc to paper The equivalence in Definition 6.4 between sub-exponential lower bounds and super-polynomial operation counts.
    This is the load-bearing assumption for all complexity-blowup claims; it is asserted without proof and is numerically false, since a lower bound of order 2^{-n^α} forces n~N^{1/α}, giving polynomial operations.
  • standard math Gradient-descent error bounds for RSI/Lipschitz functionals and subgradient flows.
    Invoked in Theorem 4.12 and the Eikonal proof; not proved in the paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Variational Framework for the Complexity of PDE Solutions." pith.science (2026). https://pith.science/paper/Z64BEASG

@misc{pith2026251021290,
  author       = {Pith},
  title        = {Pith review of: A Variational Framework for the Complexity of PDE Solutions},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/Z64BEASG}},
  note         = {Machine review of arXiv:2510.21290}
}
read the original abstract

Partial Differential Equations (PDEs) are fundamental mathematical models for describing physical phenomena, yet most PDEs of practical interest require numerical approximations. The feasibility of such methods is constrained by existing computational models. Since digital computers are the primary realizations of numerical computations, and Turing machines define their theoretical limits, computability of PDE solutions is of fundamental significance. It provides a rigorous framework to distinguish equations that are effectively solvable from those that encode undecidable or non-computable behavior. Once computability is established, complexity theory quantifies the resources required to approximate PDE solutions. In this work, we present a novel framework based on least-squares variational formulations and associated gradient flows to analyze the computability and complexity of PDE solutions from an optimization perspective. Our approach approximates PDE solution operators via discrete gradient flows, linking PDE properties, such as coercivity, ellipticity, and convexity, to solution complexity. Within this setting, we characterize representation- and discretization-dependent sufficient conditions for regimes where PDEs admit polynomial-time approximations, as well as regimes exhibiting complexity blowup, where polynomial-time input data produce solutions with super-polynomial complexity. In summary, this paper develops a variational framework for analyzing computability and computational complexity of PDE solution classes. The results show how PDE structure and solution regularity influence their complexity, by establishing sufficient conditions for computability and complexity bounds. Beyond the theoretical characterization, the framework provides guidelines for effective numerical methods and contributes to understanding the limitations of digital computation for PDE problems.

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

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

  1. The Push-Forward Transform for Continuous and Robust Comparison of Dynamic Shapes

    cs.CV 2026-08 conditional novelty 5.0 of 10

    Mapping signed distance functions through a diffeomorphism to a common reference domain yields a spectral shape descriptor (PF-SDM) that is invariant to rigid motions and scaling, and is competitive with deep learning...

Reference graph

Works this paper leans on

49 extracted references · cited by 1 Pith paper

  1. [12]

    Dissertation, Technische Universit¨ at Dresden, June 2025

    Juan-Esteban Suarez Cardona.Sobolev Cubatures and Hybrid Surrogate Models for PDE Learning Problems: Addressing Numerical Stiffness and the Gibbs Phe- nomenon. Dissertation, Technische Universit¨ at Dresden, June 2025. Published version. CC BY 4.0 license. 31

    2025

  2. [24]

    Linear convergence of gra- dient and proximal-gradient methods under the Polyak- Lojasiewicz condition

    Hamed Karimi, Julie Nutini, and Mark Schmidt. Linear convergence of gra- dient and proximal-gradient methods under the Polyak- Lojasiewicz condition. InMachine Learning and Knowledge Discovery in Databases, Lecture notes in computer science, pages 795–811. Springer International Publishing, Cham, 2016

    2016

  3. [1]

    On computable numbers, with an application to the entscheidungsproblem.J

    Alonzo Church and A M Turing. On computable numbers, with an application to the entscheidungsproblem.J. Symb. Log., 2(1):42, March 1937

    1937

  4. [2]

    Monte carlo solution of partial differential equations using a hybrid computer.IEEE Trans

    Howard Handler. Monte carlo solution of partial differential equations using a hybrid computer.IEEE Trans. Electron. Comput., EC-16(5):603–610, October 1967

    1967

  5. [3]

    The wave equation with computable ini- tial data whose unique solution is nowhere computable.MLQ Math

    Marian B Pour-El and Ning Zhong. The wave equation with computable ini- tial data whose unique solution is nowhere computable.MLQ Math. Log. Q., 43(4):499–509, January 1997

    1997

  6. [4]

    Computing schr¨ odinger propagators on type-2 turing machines.J

    Klaus Weihrauch and Ning Zhong. Computing schr¨ odinger propagators on type-2 turing machines.J. Complex., 22(6):918–935, December 2006

    2006

  7. [5]

    Complexity blowup for solutions of the laplace and the diffusion equation, 2023

    Aras Bacho, Holger Boche, and Gitta Kutyniok. Complexity blowup for solutions of the laplace and the diffusion equation, 2023

    2023

  8. [6]

    American Mathematical Society, Providence, RI, March 2022

    Lawrence C Evans.Partial differential equations. American Mathematical Society, Providence, RI, March 2022

    2022

Show all 49 references
  1. [7]

    Soft- ware, environments, tools

    Lloyd N (lloyd Nicholas) Trefethen.Spectral methods in MATLAB. Soft- ware, environments, tools. Society for Industrial and Applied Mathematics, Philadelphia, Pa., 2000

    2000

  2. [8]

    Least- squares space-time formulation for advection-diffusion problem with efficient adaptive solver based on matrix compression

    Marcin Lo´ s, Paulina Sep´ ulveda, Mateusz Dobija, and Anna Paszy´ nska. Least- squares space-time formulation for advection-diffusion problem with efficient adaptive solver based on matrix compression. InLecture Notes in Computer Science, Lecture notes in computer science, pa...

    2023

  3. [9]

    Deep least-squares methods: An unsupervised learning-based numerical method for solving elliptic PDEs.J

    Zhiqiang Cai, Jingshuang Chen, Min Liu, and Xinyu Liu. Deep least-squares methods: An unsupervised learning-based numerical method for solving elliptic PDEs.J. Comput. Phys., 420(109707):109707, November 2020

    2020

  4. [10]

    Tackling the curse of dimensionality with physics-informed neural networks

    Zheyuan Hu, Khemraj Shukla, George Em Karniadakis, and Kenji Kawaguchi. Tackling the curse of dimensionality with physics-informed neural networks. Neural Netw., 176(106369):106369, August 2024

    2024

  5. [11]

    Polynomial differentiation decreases the training time complexity of physics-informed neural networks and strengthens their approximation power.Mach

    Juan-Esteban Suarez Cardona and Michael Hecht. Polynomial differentiation decreases the training time complexity of physics-informed neural networks and strengthens their approximation power.Mach. Learn. Sci. Technol., 4(4):045005, December 2023

    2023

  6. [13]

    A variational principle for gradient flows.Math

    N Ghoussoub and L Tzou. A variational principle for gradient flows.Math. Ann., 330(3):519–549, November 2004

    2004

  7. [14]

    Lectures in Mathematics

    Luigi Ambrosio, Nicola Gigli, and Giuseppe Savare.Gradient flows. Lectures in Mathematics. ETH Z¨ urich. Birkhauser Verlag AG, Basel, Switzerland, 2 edition, December 2008

    2008

  8. [16]

    Cambridge University Press, Cambridge, England, June 2012

    Sanjeev Arora and Boaz Barak.Computational complexity. Cambridge University Press, Cambridge, England, June 2012

    2012

  9. [17]

    Noncomputability in analysis and physics: A complete determination of the class of noncomputable linear operators

    Marian Boykan Pour-El and Ian Richards. Noncomputability in analysis and physics: A complete determination of the class of noncomputable linear operators. Adv. Math. (N. Y.), 48(1):44–74, April 1983

    1983

  10. [18]

    Turing meets circuit theory: Not every continuous- time LTI system can be simulated on a digital computer.IEEE Trans

    Holger Boche and Volker Pohl. Turing meets circuit theory: Not every continuous- time LTI system can be simulated on a digital computer.IEEE Trans. Circuits Syst. I Regul. Pap., 67(12):5051–5064, December 2020

    2020

  11. [19]

    Is the linear schr¨ odinger propagator turing computable? InComputability and Complexity in Analysis, Lecture notes in computer science, pages 369–377

    Klaus Weihrauch and Ning Zhong. Is the linear schr¨ odinger propagator turing computable? InComputability and Complexity in Analysis, Lecture notes in computer science, pages 369–377. Springer Berlin Heidelberg, Berlin, Heidelberg, 2001

    2001

  12. [20]

    On the computational complexity of the dirichlet problem for poisson’s equation.Math

    Akitoshi Kawamura, Florian Steinberg, and Martin Ziegler. On the computational complexity of the dirichlet problem for poisson’s equation.Math. Struct. Comput. Sci., 27(8):1437–1465, December 2017

    2017

  13. [21]

    Bit- complexity of solving systems of linear evolutionary partial differential equations

    Ivan Koswara, Gleb Pogudin, Svetlana Selivanova, and Martin Ziegler. Bit- complexity of solving systems of linear evolutionary partial differential equations. InLecture Notes in Computer Science, Lecture notes in computer science, pages 223–241. Springer International Publishi...

    2021

  14. [22]

    Cambridge University Press, Cambridge, England, April 2009

    Sanjeev Arora and Boaz Barak.Computational complexity. Cambridge University Press, Cambridge, England, April 2009

    2009

  15. [23]

    Negative order sobolev cubatures: preconditioners of partial differential equation learning tasks circumventing numerical stiffness.Mach

    Juan-Esteban Suarez Cardona, Phil-Alexander Hofmann, and Michael Hecht. Negative order sobolev cubatures: preconditioners of partial differential equation learning tasks circumventing numerical stiffness.Mach. Learn. Sci. Technol., 5(3):035029, September 2024

    2024

  16. [25]

    Linear convergence of gradient and proximal-gradient methods under the polyak- lojasiewicz condition, 2020

    Hamed Karimi, Julie Nutini, and Mark Schmidt. Linear convergence of gradient and proximal-gradient methods under the polyak- lojasiewicz condition, 2020

    2020

  17. [26]

    Trefethen.Approximation theory and approximation practice, volume

    Lloyd N. Trefethen.Approximation theory and approximation practice, volume

  18. [27]

    J.C. Mason. Near-best multivariate approximation by fourier series, chebyshev series and chebyshev interpolation.Journal of Approximation Theory, 28(4):349– 358, 1980

    1980

  19. [28]

    Multivariate polynomial approximation in the hypercube

    Lloyd Trefethen. Multivariate polynomial approximation in the hypercube. Proceedings of the American Mathematical Society, 145(11):4837–4844, June 2017. 32

    2017

  20. [29]

    High-order integration on regular triangulated manifolds reaches super-algebraic approximation rates through cubical re-parameterizations, 2024

    Gentian Zavalani, Oliver Sander, and Michael Hecht. High-order integration on regular triangulated manifolds reaches super-algebraic approximation rates through cubical re-parameterizations, 2024

    2024

  21. [30]

    Calculation of gauss quadrature rules.Math

    Gene H Golub and John H Welsch. Calculation of gauss quadrature rules.Math. Comput., 23(106):221–230, 1969

    1969

  22. [31]

    Perspectives in Mathematical Logic

    Marian B Pour-El and J Ian Richards.Computability in Analysis and Physics. Perspectives in Mathematical Logic. Springer, Berlin, Germany, December 1989

    1989

  23. [32]

    On the definitions of computable real continuous functions

    Andrzej Grzegorczyk. On the definitions of computable real continuous functions. Fundamenta Mathematicae, 44(1):61–71, 1957

    1957

  24. [33]

    McGraw-Hill series in higher mathematics

    Walter Rudin.Real and Complex Analysis. McGraw-Hill series in higher mathematics. McGraw-Hill Professional, New York, NY, 3 edition, September 1986

    1986

  25. [34]

    The direct method in the calculus of variations

    Gianni Dal Maso. The direct method in the calculus of variations. InAn Introduction toΓ-Convergence, pages 8–18. Birkh¨ auser Boston, Boston, MA, 1993

    1993

  26. [35]

    Springer, 2011

    Haim Brezis.Functional analysis, Sobolev spaces and partial differential equations, volume 2. Springer, 2011

    2011

  27. [36]

    Applied mathematical sciences

    Pavel B Bochev and Max D Gunzburger.Least-Squares Finite Element Methods. Applied mathematical sciences. Springer, New York, NY, 2009 edition, March 2009

    2009

  28. [37]

    Compatibility conditions for dirichlet and neumann problems of poisson’s equation on a rectangle.J

    Tobias Hell and Alexander Ostermann. Compatibility conditions for dirichlet and neumann problems of poisson’s equation on a rectangle.J. Math. Anal. Appl., 420(2):1005–1023, December 2014

    2014

  29. [38]

    Analysis of fractional differential equations.J

    Kai Diethelm and Neville J Ford. Analysis of fractional differential equations.J. Math. Anal. Appl., 265(2):229–248, January 2002

    2002

  30. [39]

    Hypoelliptic operators

    Dorina Mitrea. Hypoelliptic operators. InUniversitext, pages 201–216. Springer New York, New York, NY, 2013

    2013

  31. [40]

    Singularities of solutions of the eikonal equation.Differ

    A R Alimov. Singularities of solutions of the eikonal equation.Differ. Equ., 55(10):1311–1316, October 2019

    2019

  32. [41]

    Revisiting subgradient method: Complexity and convergence beyond lipschitz continuity

    Xiao Li, Lei Zhao, Daoli Zhu, and Anthony Man-Cho So. Revisiting subgradient method: Complexity and convergence beyond lipschitz continuity. 2023

    2023

  33. [42]

    Springer International Publishing, 2015 edition, January 2014

    Nikos Katzourakis.An introduction to viscosity solutions for fully nonlinear pde with applications to calculus of variations in L?SpringerBriefs in mathematics. Springer International Publishing, 2015 edition, January 2014

    2015

  34. [43]

    Golub and John H

    Gene H. Golub and John H. Welsch. Calculation of gauss quadrature rules. Technical report, Stanford, CA, USA, 1967

    1967

  35. [44]

    Fast and accurate computation of Gauss– Legendre and Gauss–Jacobi quadrature nodes and weights.SIAM J

    Nicholas Hale and Alex Townsend. Fast and accurate computation of Gauss– Legendre and Gauss–Jacobi quadrature nodes and weights.SIAM J. Sci. Comput., 35(2):A652–A674, January 2013

    2013

  36. [45]

    On the computation of the Gauss–Legendre quadrature formula with a given precision.J

    Knut Petras. On the computation of the Gauss–Legendre quadrature formula with a given precision.J. Comput. Appl. Math., 112(1-2):253–267, November 1999

    1999

  37. [46]

    I Bogaert, B Michiels, and J Fostier.O(1) computation of legendre polynomials and Gauss–Legendre nodes and weights for parallel computing.SIAM J. Sci. Comput., 34(3):C83–C101, January 2012. 33

    2012

  38. [47]

    More special functions

    George B Arfken, Hans J Weber, and Frank E Harris. More special functions. In Mathematical Methods for Physicists, pages 871–933. Elsevier, 2013

    2013

  39. [48]

    Springer monographs in mathematics

    J Necas.Direct methods in the theory of elliptic equations. Springer monographs in mathematics. Springer, Berlin, Germany, 1 edition, March 2010

    2010

  40. [49]

    Traces of sobolev functions on fractal type sets and characterization of extension domains.Journal of Functional Analysis, 143(1):221–246, 1997

    Piotr Haj lasz and Olli Martio. Traces of sobolev functions on fractal type sets and characterization of extension domains.Journal of Functional Analysis, 143(1):221–246, 1997

    1997

  41. [50]

    Birkh¨ auser Boston, Boston, MA, 1993

    Gianni Dal Maso.The Direct Method in the Calculus of Variations, pages 8–18. Birkh¨ auser Boston, Boston, MA, 1993. 34

    1993

Pith tools

Reviewed August 4, 2026 · model on record in the stance chip above.