Pith. sign in

REVIEW 4 major objections 4 minor 64 references

Deflation-based certified greedy algorithm and adaptivity for bifurcating nonlinear PDEs

T0 review · 4 major / 4 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read This paper claims that a deflated-greedy reduced-basis algorithm can certify the approximation error on every coexisting branch of a bifurcating nonlinear PDE, without prior knowledge of the bifurcation.

desk verdict A genuinely new certified-ROM idea that deserves refereeing, but the headline branch-certification claim is not actually established by the numerics—the linear estimator is used as a guaranteed bound without proof or verification. read the letter →

arxiv 2501.12361 v3 pith:LKXH7WHK submitted 2025-01-21 math.NA cs.NA

Federico Pichi , Maria Strazzullo This is my paper · ORCID
classification math.NAcs.NA MSC 65N3065N1537M20
keywords reducedbasismethodsgreedyalgorithmdeflationbifurcationaposteriorierrorestimationNavier-StokesequationsCoandaeffectmodelorderreduction
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 reduced-basis greedy framework can be extended to nonlinear parametric PDEs whose solution manifold branches, removing the two obstacles that have kept certification out of reach: the need for a priori knowledge of the coexisting states, and the breakdown of the error estimator near bifurcation points. It introduces the adaptive-greedy algorithm, which starts from a coarse parameter grid and uses the parameter minimizing the reduced inf-sup constant to refine the sampling until the bifurcation point is located. It then introduces the deflated-greedy algorithm, which uses deflation to discover several coexisting snapshots at the same parameter value, builds reduced solutions on every branch, and computes a separate error estimator for each of them. In the Coanda-effect channel-flow test case, where the Navier-Stokes equations admit three coexisting solutions, the paper reports that only the deflated-greedy strategy keeps the reduction error below the tolerance on all branches.

What carries the argument

The machinery has two load-bearing parts. The first is the reduced inf-sup constant $\beta_h^N(\mu)$, the stability factor of the reduced Jacobian evaluated at a reduced solution, which the adaptive-greedy uses as a bifurcation detector and the estimator uses as a denominator; its vanishing at the critical parameter is exactly why standard certification fails near $\mu^*$. The second is the deflation operator $\mathcal{M}(y,u) = (\|y-u\|^r + \sigma)^{-1} I$, which multiplies the residual so that Newton iterations are repelled from already discovered solutions, allowing the algorithm to collect multiple coexisting states at one parameter value. The deflated-greedy wires these together: it computes one estimator value per reduced branch solution, selects the parameter and branch with the largest estimator, adds the corresponding high-fidelity snapshot to the basis, and then deflates the high-fidelity system to find any remaining coexisting snapshots at that same parameter. Continuation gives the reduced Newton solver initial guesses from the previously computed parameter, so the branch identity is carried from one parameter step to the next.

What would settle it

At the $N=25$ deflated-greedy configuration reported in Section 6.3, compute the relative error $E(\mu)$ and the linear estimator $\Delta_{\rm lin}^N(\mu)$ on the full test set for each of the three branches; if the estimator falls below $E$ for any branch in the bifurcating regime, the certification claim is false. A complementary check is to run the same algorithm on a pitchfork bifurcation with an odd number of unequal branches and see whether reduced deflation actually converges to every branch.

Watch

Extended reading notes

Core claim

The central claim is that both known workarounds—POD with a complete, hand-chosen snapshot set, and vanilla-greedy restricted to a single branch—can be replaced by algorithms that need no prior knowledge of the bifurcation. The adaptive-greedy treats the reduced inf-sup constant $\beta_h^N(\mu)$ as a stability indicator: because the high-fidelity inf-sup constant vanishes at the critical parameter $\mu^*$, the parameter that minimizes the reduced constant is taken as an approximation of the bifurcation point, and new training points are inserted around it. The deflated-greedy applies the deflation operator $\mathcal{M}(y,u) = (\|y-u\|^r + \sigma)^{-1} I$ at both levels: at the high-fidelity level it collects multiple coexisting snapshots for the same parameter, and at the reduced level it solves for reduced solutions on different branches. For each reduced branch solution it evaluates the linear a posteriori estimator $\Delta_{\rm lin}^N(\mu) = \|G(B u_N; \mu)\|/\beta_h^N(\mu)$, then enriches the basis with the snapshot belonging to the worst-approximated branch. For the sudden-expansion channel, the paper reports that with $N=25$ basis functions the deflated-greedy keeps the relative reduction error below $\varepsilon=10^{-3}$ for all three branches, whereas POD and vanilla-greedy approximate only the symmetric branch and lose the asymmetric ones.

Load-bearing premise

The certification depends on the standard single-solution error bound still being a true upper bound when applied branch by branch, which requires each reduced branch solution to stay close enough to the corresponding true branch that the single-solution uniqueness argument applies to that branch.

Editorial extensions

If this is right

  • If the claim is right, a reduced model built by deflated-greedy can be used to query any parameter in the trained range and return an accurate solution on any branch the user asks for, not just the stable branch.
  • The adaptive-greedy result means offline sampling can begin from a very coarse parameter grid, with the computational budget concentrated around the detected bifurcation point instead of spread over the uniqueness region.
  • Because the same parameter value can contribute several basis functions, the reduced basis encodes qualitatively different states, which is exactly the information that vanilla-greedy's estimator-driven sampling misses.
  • The deflated estimator's maximum over branches gives a stopping criterion that is meaningful for the whole solution ensemble, so convergence of the reduced model cannot be declared while an entire branch is still unrepresented.

Reading between the lines

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

  • The paper does not plot the estimator against the true error for each branch; verifying the effectivity ratio $\Delta_{\rm lin}^N(\mu)/E(\mu)$ branch by branch would be the natural next check before applying the certification claim to a new problem.
  • The authors note the adaptive strategy could serve as a preprocessing step for deflation, restricting deflated solves to the multi-solution regime; implementing that coupling would directly reduce the offline cost of the deflated-greedy.
  • Because the framework is advertised as agnostic to the type of bifurcation, a natural test is a transcritical or Hopf bifurcation, where branches are not created in symmetric pairs and reduced deflation may or may not converge to every admissible state.
  • If the branch-wise certification transfers to other PDEs, the same one-estimator-per-discovered-branch structure could be applied to time-periodic or stochastic bifurcations, where coexisting attractors play the role of coexisting steady solutions.
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 / 4 minor

Summary. The paper proposes two RB greedy algorithms for bifurcating parametric nonlinear PDEs: an adaptive-greedy that refines the training set near the detected bifurcation point, and a deflated-greedy that discovers multiple coexisting branches via reduced deflation and computes branch-wise error estimators. The methods are tested on the Coanda effect governed by steady Navier-Stokes in a sudden-expansion channel, where three branches coexist below the pitchfork point (mu* ≈ 0.96). The authors report that the deflated-greedy achieves actual reduced errors below epsilon = 1e-3 for all branches and parameters, except for one documented violation at the bifurcation point on the symmetric branch, and that the adaptive-greedy detects mu* starting from a coarse four-point training grid. The paper argues these are the first greedy-based strategies for certified approximation of bifurcating PDEs.

Significance. If the certification claim is substantiated, the work is a meaningful advance: prior POD-based ROMs for bifurcations require a-priori knowledge of the branching structure and do not provide error bounds, while vanilla-greedy misses coexisting branches. The numerical study is honest and fairly thorough: results are compared against independently computed high-fidelity solutions and against the literature value mu* ≈ 0.96, POD and vanilla-greedy baselines are included, and Appendix A explicitly documents that the nonlinear BRR estimator is inapplicable (tau_N > 1) in the bifurcating regime. The main contribution, deflated-greedy with branch-wise estimators, is conceptually novel and clearly presented. However, the central word 'certified' currently rests on an unvalidated linear estimator, and the paper itself concedes an exception to the 'all branches, all parameters' claim, so the significance can only be assessed after these gaps are closed.

major comments (4)
  1. [Section 5.2, Eq. (12)] The branch-wise certification is based entirely on the linear estimator Delta_lin_N(mu) = ||G(Bu_N; mu)|| / beta^h_N(mu). For a nonlinear PDE this is not a proven upper bound: the residual identity yields beta||e|| <= ||G(Bu_N)|| + C||e||^2, which does not imply ||e|| <= ||G||/beta. The rigorous nonlinear estimator (11) requires tau_N(mu) <= 1, and Appendix A reports that this condition is never met in the bifurcating regime. The paper neither proves that each reduced branch solution lies in the local uniqueness ball of the corresponding HF branch nor provides numerical verification that Delta_lin dominates the true error. Since the headline claim in Sections 6.3 and 7 is that the deflated-greedy 'certifies all branches', this missing reliability check is load-bearing and must be addressed.
  2. [Section 6.3, Figure 7g] The text states that 'for all branches and all parameters in Pte, the reduced error is reliably below the tolerance bound', and then immediately concedes: 'The unique point in the parametric space that seems to violate the certification is indeed the one corresponding to the bifurcation point while reconstructing the symmetric branch'. This is an explicit contradiction of the literal 'all branches, all parameters' claim. At minimum, the claim must be restricted to parameters outside a neighborhood of mu*, or the algorithm must be modified so that the symmetric branch at mu* is also certified. As written, the certification statement in Section 7 ('can certify all the branches') is inaccurate.
  3. [Section 4, Eq. (9)] The adaptive-greedy detection of mu* relies on the assumption that 'for large enough N, the reduced inf-sup (9) is a good approximation of the high-fidelity inf-sup (5)', but no convergence statement or quantitative bound is provided. Since 'detecting the bifurcation point starting from scarce information' is a claimed contribution, the paper should either supply a convergence analysis for mu_bif as N increases or explicitly label the detection as a heuristic and validate it with a systematic study (e.g., showing that mu_bif approaches mu* as N grows). The current evidence is a single numerical example, which is suggestive but not sufficient for a general claim.
  4. [Section 6.3, Figures 7 and 9] The certification claim concerns the estimator, yet the paper never plots the estimator against the true error. Figure 7 reports only the actual error E(mu), and Figure 9 shows estimator values at one iteration without overlaying them with E(mu). A direct empirical check would be to plot Delta_lin_N(mu) (or the branch-wise max Delta) together with E(mu) on the test set Pte for the final deflated-greedy basis, for each of the three branches. Such an overlay would either confirm that the estimator dominates the true error or expose the gap that currently undermines the word 'certified'.
minor comments (4)
  1. [Section 6.3, Eq. (24)] The relative error E(mu) is defined as a sum of velocity and pressure relative errors rather than a norm of the combined error; since the two fields have different scales, this additive combination should be justified or replaced by a single norm to avoid masking large errors in one component.
  2. [Remark 2, Section 5.2] The procedure for handling the case where the same parameter maximizes the estimator multiple times is described only in prose; a small pseudocode listing (e.g., a while loop over the sorted estimator values) would make the algorithm unambiguous and easier to reproduce.
  3. [Figure 5] The inset panel in Figure 5 is very small and the axes labels are difficult to read; enlarging the inset and using clearly marked ticks would improve readability.
  4. [Section 2.3, Eq. (12)] The sentence 'for some problems, e.g., the bifurcating ones, such a condition could be difficult to obtain, and the linear estimator is exploited for all greedy iterations' is vague; the paper should specify in advance which problems and under which conditions the linear estimator is used, and state that it is not a proven bound in general.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the reduced-basis claims are validated against independently computed high-fidelity solutions; the main caveats are an unproven linear error bound and a heuristic inf-sup bifurcation indicator, neither of which is a definitional reduction.

full rationale

The derivation chain is not circular. The adaptive-greedy locates the bifurcation point through the reduced inf-sup indicator, and the paper states the underlying assumption explicitly: "for large enough N, the reduced inf-sup (9) is a good approximation of the high-fidelity inf-sup (5)" (Section 4). This is an approximation premise, not a self-referential definition of the detected point. The deflated-greedy selects snapshots by the standard greedy criterion based on the residual estimator (12) and then reports errors (24) against independent HF solutions obtained with Newton and deflation; no parameter is fitted to make the error curves pass, and the detected value mu* ≈ 0.96 agrees with the literature. Self-citations to Pichi et al. prior ROM-bifurcation works supply background and baselines, not the load-bearing proof of the new claims. The main weakness is a correctness gap, not circularity: Appendix A states that the BRR condition tau_N(mu) <= 1 is "never met" in the bifurcating case, so the linear estimator (12) is used even though it is not shown to be a guaranteed upper bound per branch; Section 6.3 also concedes one certification violation at the bifurcation point on the symmetric branch. These concerns undermine the rigor of the "certifies all branches" claim but do not reduce it, by construction, to the algorithm's own inputs.

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

The central claims rest on standard RB machinery plus a few chosen-by-hand parameters. The deflation pair (r, sigma) is the most substantive tuning: the paper itself warns that other PDEs may need different values (Section 6.3), and branch discovery depends on it. Tolerances epsilon, tol, and nref are standard algorithmic choices. The key axioms are the heuristic identification of µ* with the reduced inf-sup minimizer (Section 4) and the per-branch validity of the linear a posteriori bound (Sections 2.3 and 5.2); both are stated, neither is proven. No new physical or mathematical entities are introduced, so the invented-entities ledger is empty.

free parameters (5)
  • deflation exponent r = 2
    Used in the deflation operator M(y,u) = (||y-u||^r + sigma)^(-1) I (Eq. 15). Chosen as a common literature value; Section 6.3 states 'different bifurcating PDEs may need ad-hoc values'. Branch discovery depends on it.
  • deflation shift sigma = 1
    Same deflation operator; chosen by hand from the literature, not by a criterion. Affects whether deflated Newton escapes known roots.
  • adaptive refinement points nref = 4
    Number of equispaced points added per refinement step in Algorithm 5; controls the resolution with which the detected bifurcation point is localized.
  • adaptive inf-sup tolerance tol = 1e-2
    Stopping criterion for the refinement loop comparing mu_bif and mu_prev; controls the accuracy of the detected bifurcation point.
  • greedy tolerance epsilon = 1e-3
    Stopping tolerance on the error estimator for the greedy loops; the certification claim is relative to this tolerance.
assumptions (5)
  • domain assumption The discrete inf-sup constant beta_h(µ) vanishes only at the bifurcation point µ*, and the reduced inf-sup beta_N^h(µ) approximates it well for N large (Section 4, first paragraph).
    Load-bearing for the adaptive-greedy detection: the minimizer of the reduced inf-sup over the training set is assumed to track µ*. Only numerically demonstrated in Figure 6.
  • domain assumption The a posteriori linear estimator (12) remains a valid upper bound for each branch solution found by reduced deflation, i.e., each reduced branch solution lies in the local uniqueness ball of the corresponding HF branch (Sections 2.3 and 5.2).
    Underlies the certification claim of the deflated-greedy; not proven for the multi-branch/deflated setting and never verified by plotting estimator versus true error.
  • standard math G is locally Lipschitz and the inf-sup constant is positive away from µ* (Section 2.3, Eqs. 8-11).
    Standard BRR theory requirements; violated exactly at the bifurcation point, which the paper acknowledges in Figure 7g and Appendix A.
  • domain assumption Deflation with parameters r and sigma and the prescribed solver settings discovers all K = 3 coexisting branches at each sampled parameter (Section 3.2).
    The paper states deflation does not guarantee reaching all solutions in general; the test case relies on it succeeding, and the certification claim covers only the branches actually discovered.
  • domain assumption The finite element discretization resolves the bifurcation (Section 6.1).
    Nh = 24301 Taylor-Hood P2-P1 degrees of freedom are specified, but no mesh size or refinement information is given, so resolution adequacy is assumed.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Deflation-based certified greedy algorithm and adaptivity for bifurcating nonlinear PDEs." pith.science (2026). https://pith.science/paper/LKXH7WHK

@misc{pith2026250112361,
  author       = {Pith},
  title        = {Pith review of: Deflation-based certified greedy algorithm and adaptivity for bifurcating nonlinear PDEs},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/LKXH7WHK}},
  note         = {Machine review of arXiv:2501.12361}
}
read the original abstract

This work deals with tailored reduced order models for bifurcating nonlinear parametric partial differential equations, where multiple coexisting solutions arise for a given parametric instance. Approaches based on proper orthogonal decomposition have been widely investigated in the literature, but they usually rely on some \emph{a-priori} knowledge about the bifurcating model and lack any error estimation. On the other hand, standard certified reduced basis techniques fail to represent correctly the branching behavior, since the error estimator is no longer reliable. The main goal of the contribution is to overcome these limitations by introducing two novel algorithms: (i) the adaptive-greedy, detecting the bifurcation point starting from scarce information over the parametric space, and (ii) the deflated-greedy, certifying multiple coexisting branches simultaneously. The former approach takes advantage of the features of the reduced manifold to detect the bifurcation, while the latter exploits the deflation and continuation methods to discover the bifurcating solutions and enrich the reduced space. We test the two strategies for the Coanda effect held by the Navier-Stokes equations in a sudden-expansion channel. The accuracy of the approach and the error certification are compared with vanilla-greedy and proper orthogonal decomposition.

Figures

Figures reproduced from arXiv: 2501.12361 by the authors.

Figure 1
Figure 1. Schematic representation of the solution ensemble in the case of a pitch￾fork bifurcation. Let us denote by u 1 h = u 1 h (µ) the solution of Equation (1) for a fixed µ ∈ P with initial guess u (0) . We can now build the modified residual as a function of the (renamed) unknown yh ∈ U Nh (14) F(yh; µ) = M(yh, u1 h )G(yh; µ), where, given I the identity over U, r ≥ 1, and σ > 0, the deflation operator M(yh, u1 h ) is … view at source ↗
Figure 2
Figure 2. Comparison of vanilla-greedy methodology and the proposed deflated￾greedy approach. 5.1. Reduced deflation method. To obtain a greedy strategy that is capable of certifying the whole set of (non-unique) solutions U for general nonlinear systems we need to transpose the concept of deflation and continuation at the reduced level. While the latter is straightforward, simply entailing the same mechanism but in a lower d… view at source ↗
Figure 3
Figure 3. , is denoted with Ω. Γin Γout Γw (0, 2.5) (0, 5) (10, 0) (50, 0) (10, 7.5) (50, 7.5) [PITH_FULL_IMAGE:figures/full_fig_p014_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Pitchfork bifurcation for the Coanda effect via point-wise evaluation of the vertical velocity in the channel’s centerline, and the corresponding coexisting solutions for the symmetric and asymmetric branches at Re = 156, i.e. µ = 0.5. where the weak formulation terms …
Figure 5
Figure 5. Figure 5: Evolution of the refinement strategy for the parametric space versus the number of basis/iterations. 6.3. The deflated-greedy performances. In this section, we show the capability of the deflated￾greedy strategy to certify multiple branches while building a reduced ord…
Figure 6
Figure 6. Figure 6: β h N (µ) approximation w.r.t. different iterations for the adaptive-greedy strategy compared with β h (µ) computed on a fine equispaced discretization of P. stopping criterion for the standard and deflated Newton methods is imposed by setting a tolerance tol = 10−10 o…
Figure 7
Figure 7. Figure 7: Relative error and branch certification. Symmetric, upper asymmetric, and lower asymmetric solutions from left to right. POD (N = 25), vanilla-greedy (N = 12), and deflated-greedy (N = 25), are depicted respectively in the top, center, and bottom rows. When applying th…
Figure 8
Figure 8. Figure 8: Average and maximum relative error comparison w.r.t. N between POD (red lines), vanilla-greedy (blue lines) and deflated-greedy (green lines) algorithms for the velocity and pressure asymmetric fields. Finally, we remark that, as it has already been noticed when applyi…
Figure 9
Figure 9. Figure 9: Bound for deflated-greedy and vanilla-greedy algorithms at the fourth iteration. The plot depicts the sampled snapshot(s) and the effect on the detection of different admissible branches. In [PITH_FULL_IMAGE:figures/full_fig_p020_9.png]
Figure 10
Figure 10. Figure 10: Γin Γout Γw (0, 3.75) (0, 7.5) (10, 0) (50, 0) (50, 7.5) [PITH_FULL_IMAGE:figures/full_fig_p025_10.png]
Figure 11
Figure 11. Figure 11: Evolution of τN (µ) for several iterations of the vanilla-greedy for (i) the flow over a backward-facing step and (ii) the sudden-expansion channel, left and right plot, respectively. We perform a vanilla-greedy procedure with Nmax = 35 and tolerance ε = 10−5 with an …

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

64 extracted references · 43 canonical work pages

  1. [1]

    E. L. Allgower and K. Georg. Numerical Continuation Methods, volume 13 of Springer Series in Computational Mathematics. Springer, Berlin, Heidelberg, 1990. doi:10.1007/978-3-642-61257-2

    doi:10.1007/978-3-642-61257-2 1990

  2. [2]

    Ambrosetti and G

    A. Ambrosetti and G. Prodi. A Primer of Nonlinear Analysis . Cambridge Studies in Advanced Mathematics. Cambridge University Press, 1995. URL: https://books.google.it/books?id=dVSTUben3QoC

    work page 1995

  3. [3]

    empirical interpolation

    M. Barrault, N. C. Nguyen, Y. Maday, and A. T. Patera. An “empirical interpolation” method: Application to efficient reduced-basis discretization of partial differential equations.C. R. Acad. Sci. Paris, S´ erie I., 339:667–672, 2004. 22 DEFLATED-GREEDY AND ADAPTIVITY FOR CERTIFIED BIFURCATIONS

    work page 2004

  4. [4]

    Battaglia, S

    F. Battaglia, S. J. Tavener, A. K. Kulkarni, and C. L. Merkle. Bifurcation of Low Reynolds Number Flows in Symmetric Channels. AIAA Journal , 35(1):99–105, 1997. doi:10.2514/2.68

    work page doi:10.2514/2.68 1997

  5. [5]

    D. Bigoni. Nonlinear solid mechanics: bifurcation theory and material instability . Cambridge University Press, 2012

    work page 2012

  6. [6]

    Bifurcation analysis of two-dimensional Rayleigh--B\'enard convection using deflation

    N. Boull´ e, V. Dallas, and P. E. Farrell. Bifurcation analysis of two-dimensional Rayleigh–B \’enard convection using deflation. Physical Review E, 105(5):055106, 2022. arXiv:2102.10576, doi:10.1103/PhysRevE.105.055106

    work page Pith review arXiv 2022

  7. [7]

    J. R. Bravo, G. Stabile, M. Hess, J. A. Hernandez, R. Rossi, and G. Rozza. Geometrically parametrised reduced order models for studying the hysteresis of the Coanda effect in finite element-based incompressible fluid dynamics. Journal of Computational Physics , 509:113058, 2024. doi:10.1016/j.jcp.2024.113058

    work page arXiv 2024

  8. [8]

    Brezzi, J

    F. Brezzi, J. Rappaz, and P. Raviart. Finite Dimensional Approximation of Nonlinear Problems. Part I: Branches of Nonsingular Solutions. Numerische Mathematik , 36(1):1–25, 1980

    work page 1980

Show all 64 references
  1. [9]

    S. L. Brunton, J. L. Proctor, and J. N. Kutz. Discovering governing equations from data by sparse identification of nonlinear dynamical systems. Proceedings of the National Academy of Sciences , 113(15):3932–3937, 2016. doi:10.1073/pnas.1517384113

    2016 doi

  2. [10]

    Buffa, Y

    A. Buffa, Y. Maday, A. T. Patera, C. Prud’homme, and G. Turinici. A priori convergence of the greedy algorithm for the parametrized reduced basis method. ESAIM: Mathematical modelling and numerical analysis , 46(3):595– 603, 2012

    2012

  3. [11]

    Caloz and J

    G. Caloz and J. Rappaz. Numerical analysis for nonlinear and bifurcation problems. In Handbook of Numerical Analysis, volume 5 of Techniques of Scientific Computing (Part 2) , pages 487–637. Elsevier, 1997. doi:10.1016/ S1570-8659(97)80004-X

    1997

  4. [12]

    E. G. Charalampidis, N. Boull´ e, P. E. Farrell, and P. G. Kevrekidis. Bifurcation analysis of stationary solutions of two-dimensional coupled Gross–Pitaevskii equations using deflated continuation. Communications in Nonlinear Science and Numerical Simulation , 87:105255, 2020...

    2020

  5. [13]

    E. G. Charalampidis, P. G. Kevrekidis, and P. E. Farrell. Computing stationary solutions of the two-dimensional Gross–Pitaevskii equation with deflated continuation. Communications in Nonlinear Science and Numerical Simulation, 54:482–499, 2018. doi:10.1016/j.cnsns.2017.05.024

    2018 doi

  6. [14]

    Chaturantabut and D

    S. Chaturantabut and D. C. Sorensen. Nonlinear Model Reduction via Discrete Empirical Interpolation. SIAM Journal on Scientific Computing , 32(5):2737–2764, 2010. doi:10.1137/090766498

    2010 doi

  7. [15]

    P. Ciarlet. Linear and Nonlinear Functional Analysis with Applications . Other Titles in Applied Mathematics. Society for Industrial and Applied Mathematics, 2013

    2013

  8. [16]

    Cort´ es, H

    J. Cort´ es, H. Herrero, and F. Pla. A local ROM for Rayleigh–B´ enard bifurcation problems.Computer Methods in Applied Mechanics and Engineering , 425:116949, 2024. doi:10.1016/j.cma.2024.116949

    2024

  9. [17]

    N. Deng, B. R. Noack, M. Morzy´ nski, and L. R. Pastur. Low-order model for successive bifurcations of the fluidic pinball. Journal of Fluid Mechanics , 884:A37, 2020. doi:10.1017/jfm.2019.959

    2020 doi

  10. [18]

    S. Deparis. Reduced basis error bound computation of parameter-dependent Navier-Stokes equations by the natural norm approach. SIAM Journal on Numerical Analysis , 46(4):2039 – 2067, 2008. doi:10.1137/060674181

    2008 doi

  11. [19]

    H. A. Dijkstra and F. W. Wubs. Bifurcation Analysis of Fluid Flows, 2023. doi:10.1017/9781108863148

    2023 doi

  12. [20]

    Fabiani, F

    G. Fabiani, F. Calabr` o, L. Russo, and C. Siettos. Numerical solution and bifurcation analysis of nonlinear partial differential equations with extreme learning machines. Journal of Scientific Computing , 89(2):44, 2021. doi:10.1007/s10915-021-01650-5

    2021 doi

  13. [21]

    Farrell, A

    P. Farrell, A. Birkisson, and S. Funke. Deflation techniques for finding distinct solutions of nonlinear partial differ- ential equations. SIAM Journal on Scientific Computing , 37(4):A2026 – A2045, 2015. doi:10.1137/140984798

    2015 doi

  14. [22]

    P. E. Farrell, C. H. L. Beentjes, and ´A. Birkisson. The computation of disconnected bifurcation diagrams, 2016. arXiv:1603.00809, doi:10.48550/arXiv.1603.00809

    2016 doi arXiv

  15. [23]

    I. C. Gonnella, M. Khamlich, F. Pichi, and G. Rozza. A stochastic perturbation approach to nonlinear bifurcating problems, 2024. arXiv:2402.16803, doi:10.48550/arXiv.2402.16803

    2024 doi

  16. [24]

    Herrero, Y

    H. Herrero, Y. Maday, and F. Pla. RB (Reduced basis) for RB (Rayleigh–B´ enard).Computer Methods in Applied Mechanics and Engineering , 261–262:132–141, 2013. doi:10.1016/j.cma.2013.02.018

    2013 doi

  17. [25]

    M. Hess, A. Alla, A. Quaini, G. Rozza, and M. Gunzburger. A localized reduced-order modeling approach for PDEs with bifurcating solutions. Computer Methods in Applied Mechanics and Engineering , 351:379–403, 2019. doi:10.1016/j.cma.2019.03.050

    2019 doi

  18. [26]

    J. S. Hesthaven, G. Rozza, and B. Stamm. Certified Reduced Basis Methods for Parametrized Partial Differential Equations. SpringerBriefs in Mathematics. Springer International Publishing AG, Cham, 1st ed. 2016 edition,

    2016

  19. [27]

    C. Jin, Z. Sun, and W. Xu. A novel stochastic bifurcation and its discrimination. Communications in Nonlinear Science and Numerical Simulation , 110:106364, 2022. doi:10.1016/j.cnsns.2022.106364

    2022

  20. [28]

    H. Keller. Lectures on Numerical Methods in Bifurcation Problems. 1988

    1988

  21. [29]

    Khamlich, F

    M. Khamlich, F. Pichi, and G. Rozza. Model order reduction for bifurcating phenomena in fluid-structure in- teraction problems. International Journal for Numerical Methods in Fluids , 94(10):1611 – 1640, 2022. doi: 10.1002/fld.5118

    2022 doi

  22. [30]

    Kielh¨ ofer.Bifurcation Theory: An Introduction with Applications to Partial Differential Equations , volume 156 of Applied Mathematical Sciences

    H. Kielh¨ ofer.Bifurcation Theory: An Introduction with Applications to Partial Differential Equations , volume 156 of Applied Mathematical Sciences. Springer, New York, NY, 2012. doi:10.1007/978-1-4614-0502-3 . DEFLATED-GREEDY AND ADAPTIVITY FOR CERTIFIED BIFURCATIONS 23

    2012 doi

  23. [31]

    Kuehn and K

    C. Kuehn and K. Lux. Uncertainty Quantification of Bifurcations in Random Ordinary Differential Equations. SIAM Journal on Applied Dynamical Systems , 20(4):2295–2334, 2021. doi:10.1137/21M1392073

    2021 doi

  24. [32]

    Kuehn, C

    C. Kuehn, C. Piazzola, and E. Ullmann. Uncertainty quantification analysis of bifurcations of the Allen–Cahn equation with random coefficients, 2024. arXiv:2404.04639

    2024 arXiv

  25. [33]

    Y. A. Kuznetsov. Elements of Applied Bifurcation Theory, volume 112 ofApplied Mathematical Sciences. Springer International Publishing, Cham, 2023. doi:10.1007/978-3-031-22007-4

    2023 doi

  26. [34]

    Laakmann and N

    F. Laakmann and N. Boull´ e. Bifurcation analysis of a two-dimensional magnetic Rayleigh–B´ enard problem. Physica D: Nonlinear Phenomena , 467:134270, 2024. doi:10.1016/j.physd.2024.134270

    2024

  27. [35]

    L´ eger, P

    S. L´ eger, P. Larocque, and D. LeBlanc. Improved Moore-Penrose continuation algorithm for the computation of problems with critical points. Computers & Structures , 281:107009, 2023. doi:10.1016/j.compstruc.2023. 107009

    2023 doi

  28. [36]

    A. Logg, K. Mardal, and G. Wells. Automated Solution of Differential Equations by the Finite Element Method . Springer-Verlag, Berlin, 2012

    2012

  29. [37]

    A. Manzoni. An efficient computational framework for reduced basis approximation and a posteriori error estima- tion of parametrized Navier–Stokes flows. ESAIM: Mathematical Modelling and Numerical Analysis , 48(4):1199– 1226, 2014. doi:10.1051/m2an/2014013

    2014

  30. [38]

    Niroomandi, I

    S. Niroomandi, I. Alfaro, E. Cueto, and F. Chinesta. Model order reduction for hyperelastic materials. Interna- tional Journal for Numerical Methods in Engineering , 81(9):1180–1206, 2010. doi:10.1002/nme.2733

    2010 doi

  31. [39]

    M. A. Olshanskii and L. G. Rebholz. Approximating a branch of solutions to the Navier–Stokes equations by reduced-order modeling. Journal of Computational Physics , 524:113728, 2025. doi:10.1016/j.jcp.2025.113728

    2025

  32. [40]

    F. Pichi. Reduced order models for parametric bifurcation problems in nonlinear PDEs . PhD thesis, Scuola Internazionale Superiore di Studi Avanzati, 2020

    2020

  33. [41]

    Pichi, F

    F. Pichi, F. Ballarin, G. Rozza, and J. S. Hesthaven. An artificial neural network approach to bifurcating phe- nomena in computational fluid dynamics. Computers & Fluids , 254:105813, 2023. doi:10.1016/j.compfluid. 2023.105813

    2023

  34. [42]

    Pichi, B

    F. Pichi, B. Moya, and J. S. Hesthaven. A graph convolutional autoencoder approach to model order reduction for parametrized PDEs. Journal of Computational Physics , 501:112762, 2024. doi:10.1016/j.jcp.2024.112762

    2024

  35. [43]

    Pichi, A

    F. Pichi, A. Quaini, and G. Rozza. A Reduced Order Modeling Technique to Study Bifurcating Phenomena: Application to the Gross–Pitaevskii Equation. SIAM Journal on Scientific Computing , 42(5):B1115–B1135, 2020. doi:10.1137/20M1313106

    2020 doi

  36. [44]

    Pichi and G

    F. Pichi and G. Rozza. Reduced Basis Approaches for Parametrized Bifurcation Problems held by Non-linear Von K´ arm´ an Equations.Journal of Scientific Computing , 81(1):112–135, 2019. doi:10.1007/s10915-019-01003-3

    2019 doi

  37. [45]

    Pichi and G

    F. Pichi and G. Rozza. Reduced Order Models for the Buckling of Hyperelastic Beams. In G. Rozza, G. Sta- bile, M. Gunzburger, and M. D’Elia, editors, Reduction, Approximation, Machine Learning, Surrogates, Em- ulators and Simulators: RAMSES , pages 199–240. Springer Nature Swi...

    2024

  38. [46]

    Pichi, M

    F. Pichi, M. Strazzullo, F. Ballarin, and G. Rozza. Driving bifurcating parametrized nonlinear PDEs by optimal control strategies: Application to Navier–Stokes equations with model order reduction. ESAIM: Mathematical Modelling and Numerical Analysis , 56(4):1361–1400, 2022. d...

    2022

  39. [47]

    Pintore, F

    M. Pintore, F. Pichi, M. Hess, G. Rozza, and C. Canuto. Efficient computation of bifurcation diagrams with a deflated approach to reduced basis spectral element method. Advances in Computational Mathematics , 47(1):1,

  40. [48]

    Pitton, A

    G. Pitton, A. Quaini, and G. Rozza. Computational reduction strategies for the detection of steady bifurcations in incompressible fluid-dynamics: Applications to Coanda effect in cardiology. Journal of Computational Physics , 344:534–557, 2017. doi:10.1016/j.jcp.2017.05.010

    2017 doi

  41. [49]

    Pitton and G

    G. Pitton and G. Rozza. On the Application of Reduced Basis Methods to Bifurcation Problems in Incompressible Fluid Dynamics. Journal of Scientific Computing , 73(1):157–177, 2017. doi:10.1007/s10915-017-0419-6

    2017 doi

  42. [50]

    Quaini, R

    A. Quaini, R. Glowinski, and S. ˇCani´ c. Symmetry breaking and preliminary results about a Hopf bifurcation for incompressible viscous flow in an expansion channel. International Journal of Computational Fluid Dynamics , 30(1):7–19, 2016. doi:10.1080/10618562.2016.1144877

    2016

  43. [51]

    Quarteroni, A

    A. Quarteroni, A. Manzoni, and F. Negri. Reduced Basis Methods for Partial Differential Equations: An Intro- duction. La Matematica per Il 3+2, 92. Springer International Publishing, Cham, 1st ed. 2016. edition, 2016. doi:10.1007/978-3-319-15431-2

    2016 doi

  44. [52]

    L. B. Rall. A Note on the Convergence of Newton’s Method. SIAM Journal on Numerical Analysis , 11(1):34–36,

  45. [53]

    Rheinboldt

    W. Rheinboldt. An adaptive continuation process for solving systems of nonlinear equations. Banach Center Publications, 3(1):129–142, 1978

    1978

  46. [54]

    Rozza, F

    G. Rozza, F. Ballarin, L. Scandurra, and F. Pichi. Real Time Reduced Order Computational Mechanics . SISSA Springer Series. Springer Cham, 2024

    2024

  47. [55]

    Rozza and K

    G. Rozza and K. Veroy. On the stability of the reduced basis method for Stokes equations in parametrized domains. Computer Methods in Applied Mechanics and Engineering , 196(7):1244–1260, 2007

    2007

  48. [56]

    R. Seydel. Practical Bifurcation and Stability Analysis , volume 5 of Interdisciplinary Applied Mathematics . Springer, New York, NY, 2010. doi:10.1007/978-1-4419-1740-9 . 24 DEFLATED-GREEDY AND ADAPTIVITY FOR CERTIFIED BIFURCATIONS

    2010 doi

  49. [57]

    D. J. Silvester. Machine learning for hydrodynamic stability, 2024. arXiv:2407.09572

    2024

  50. [58]

    Tonicello, A

    N. Tonicello, A. Lario, G. Rozza, and G. Mengaldo. Non-intrusive reduced order models for the accurate prediction of bifurcating phenomena in compressible fluid dynamics. Computers & Fluids , 278:106307, 2024. doi:10.1016/ j.compfluid.2024.106307

    2024

  51. [59]

    D. J. Tritton. Physical fluid dynamics . Springer Science & Business Media, 2012

    2012

  52. [60]

    H. Uecker. Numerical Continuation and Bifurcation in Nonlinear PDEs . Other Titles in Applied Mathematics. Society for Industrial and Applied Mathematics, 2021. doi:10.1137/1.9781611976618

    2021 doi

  53. [61]

    Venturi, X

    D. Venturi, X. Wan, and G. E. Karniadakis. Stochastic bifurcation analysis of Rayleigh–B´ enard convection. Journal of Fluid Mechanics , 650:391–413, 2010. doi:10.1017/S0022112009993685. Appendices Appendix A. A numerical comparison between estimators for bifurcating Navier-St...

    2010 doi

  54. [64]

    free-flow

    To obtain an approximation of the Sobolev embedding constant ρ, one can exploit a strategy that combines an eigenvalue problem and a fixed-point algorithm, for further insight see [18, 37]. An important quantity to look at in order to define the nonlinear estimator is τN (µ), ...

  55. [2015]

    doi:10.1007/978-3-319-22470-1

    doi

  56. [2021]

    doi:10.1007/s10444-020-09827-6

    doi

Pith tools

Reviewed August 10, 2026 · model on record in the stance chip above.