Regularity and positivity of solutions of the Consensus-Based Optimization equation: unconditional global convergence

Lukang Sun; Massimo Fornasier

REVIEW 3 major objections 5 minor 2 cited by

Regularity and positivity of solutions of the Consensus-Based Optimization equation: unconditional global convergence

T0 review · 3 major / 5 minor · reviewed 2026-08-09 · deepseek-v4-flash

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

Pith's one-line read This paper proves that, for d>1, any smooth initial density for the CBO equation becomes strictly positive everywhere except at the moving consensus point, eliminating the initial-support hypothesis from global-convergence proofs.

desk verdict A credible and valuable positivity result for CBO, but the abstract over-sells and the application of the main theorem to the CBO equation needs a small repair. read the letter →

arxiv 2502.01434 v2 pith:JMN72B75 submitted 2025-02-03 math.AP

Massimo Fornasier , Lukang Sun This is my paper · ORCID

classification math.AP MSC 35B6535Q8435Q9135Q9337N4060H10

keywords consensus-basedoptimizationglobalnonlinearFokker-Planckequationdegeneratediffusionpositivityofsolutionsregularitymean-fieldlimitunconditionalconvergence

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 targets a missing piece in the theory of Consensus-Based Optimization (CBO), a particle method for global minimization whose mean-field limit is a nonlinear Fokker–Planck equation with degenerate diffusion. Earlier convergence proofs required the initial particle distribution to already charge a neighborhood of the global minimizer. The paper shows that for dimension d>1, any smooth initial density produces a solution that is strictly positive everywhere except at the moving consensus point, so that initial-support condition is unnecessary and CBO converges globally without it. To reach this conclusion, it proves existence, uniqueness, and classical regularity for a broad class of degenerate drift-diffusion equations by a Galerkin approximation. The positivity then follows from a parabolic Harnack inequality applied after shifting the equation into the consensus-point frame.

What carries the argument

The central object is the CBO mean-field PDE ∂tρ = div((v - v_alpha(ρ,t))ρ) + Δ(‖v - v_alpha(ρ,t)‖²ρ), where v_alpha is the Gibbs-weighted consensus point. The proof rewrites this as a linear degenerate drift-diffusion equation with G(v,t) = ‖v - v_alpha(t)‖² and J(v,t) = v - v_alpha(t), then shifts coordinate by v_alpha(t) to obtain ∂tρ̃ = ‖v‖² Δρ̃ + ⟨5v + P(t), ∇ρ̃⟩ + 3d ρ̃, where P(t) = d v_alpha(t)/dt is bounded by Lemma 2.24 under the polynomial-growth assumptions (160). On any annulus B_R(0) \ B_ε(0) the shifted equation is uniformly parabolic, so the parabolic Harnack inequality turns a single pocket of mass into strict positivity everywhere outside the consensus point. The Galerkin machinery—truncating coefficients on expanding tori, proving uniform Sobolev estimates for the approximating solutions, and passing to the limit in the cutoff radius—supplies the classical regularity needed to apply Harnack and to identify the PDE solution with the law of the underlying SDE.

What would settle it

Compute or simulate d v_alpha(t)/dt for a CBO PDE solution whose objective f satisfies (144) but has ∇f and Δf growing faster than polynomially, such as f(v) = $e^{{‖v‖²}}$; if dv_alpha/dt is unbounded on a finite time interval, Lemma 2.24 fails and the claimed unconditional positivity is not established by this route. Alternatively, a direct numerical solution of the d=2 PDE starting from a smooth density supported in a half-space should show positive density everywhere except at the consensus point after any positive time; a persistent zero region would refute Theorem 2.25.

Watch

Extended reading notes

Core claim

The paper's central claim is Theorem 2.25: if the objective f satisfies the growth conditions (144), the CBO PDE (145) has a unique measure solution in a weighted-Sobolev class Γ; if additionally ∇f and Δf grow at most polynomially, as in (160), then for d>1 the density satisfies rho(v,t)>0 for all t>0 and all v different from the consensus point v_alpha(rho,t). Because the earlier convergence theorems [24, Theorems 3.7 and 3.8] only require initial mass in a ball around the global minimizer, this positivity makes assumption (7) superfluous and yields unconditional global convergence. The supporting regularity result, Theorem 2.5, establishes unique classical solutions in class Γ′ for a general family of degenerate drift-diffusion equations (11)-(12), with additional weighted integrability in class Γ under Assumption 2.17.

Load-bearing premise

The load-bearing premise is that the consensus point moves at bounded speed, which the paper proves only under polynomial growth of ∇f and Δf; if that bound fails, the shifted equation is not shown to have bounded coefficients and the positivity argument does not go through.

Editorial extensions

If this is right

The initial-support condition (7) is removed, so convergence guarantees no longer require sampling particles near a global minimizer.
The CBO PDE is well-posed: it has a unique measure solution, and that solution is in fact a classical solution in class Γ, so further quantitative analysis can build on it.
The regularity theorem extends to the divergence-form equation (12), giving classical solutions for the whole family of degenerate drift-diffusion equations satisfying Assumptions 2.6, 2.7, and 2.17.
For d>1, positivity holds globally in time and space except at the consensus point; the d=1 case genuinely fails, as the paper shows by an explicit invariant-region argument.
The same results carry over to CBO on smooth compact manifolds without boundary, so constrained CBO on hypersurfaces inherits positivity and unconditional convergence.

Reading between the lines

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

The paper leaves implicit that the Harnack-shift mechanism could be made quantitative: tracking how fast positivity spreads away from the consensus point would yield explicit lower bounds on ρ, which could sharpen the convergence rates inherited from the earlier Lyapunov estimates.
A natural testable extension is to relax the polynomial-growth assumption (160): since Lemma 2.24 is the only place that assumption enters, any objective for which dv_alpha/dt remains bounded would satisfy the same positivity conclusion through an unchanged proof.
Because the regularity theorem is stated for general equations (11)-(12), the positivity argument likely adapts to other consensus-based variants—constrained, multi-objective, or momentum-enhanced—whose mean-field equations share the same degenerate-diffusion structure.

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.

Desk Editor's Note

read the letter

Short version: this paper has a real result—positivity of CBO densities for d>1, which removes the initial-support assumption from the convergence theory—and a substantial regularity theorem for degenerate drift-diffusion equations that supports it. The proof is long but built from standard tools, and the Galerkin-plus-cutoff strategy is plausible. It deserves a serious referee. But the abstract overstates what is proved, and the application of the main theorem to the CBO equation has a repairable but real mismatch that must be fixed.

What is genuinely new: the well-posedness and regularity result for equations (11)-(12) under Assumptions 2.6/2.7/2.17, and the full-support theorem for the CBO density away from vα. The d=1 obstruction in Remark 2.23 is a nice touch. The paper is honest enough to state the extra growth conditions (160) before the positivity claim.

Where it is soft: the stress-test note claims a contradiction between the drift coefficient 5 in the expansion of (145) and the coefficient 3 in (146). That specific arithmetic does not hold up—div(‖w‖²∇ρ) contributes 2⟨w,∇ρ⟩, so the total drift really is 5. The real problem is that Theorem 2.5 is stated for equation (11) with reaction +ρ and drift ⟨J,∇ρ⟩, while (146) has 3dρ and 3⟨w,∇ρ⟩. You can absorb both: take J=3w and g=(3d−1)ρ, or make an exponential time change to convert 3dρ into ρ. Either works under the same assumptions, but the paper doesn't say so. As written, the bridge to the CBO equation is not literally established.

The other soft spots are more minor: the positivity argument leans on [24, Proposition 23] without stating its hypotheses, and Lemma 2.24 is sketched by reference to [14, Lemma 3.3]. Both are likely true under (144)+(160), but the paper should say what it imports. And the abstract's 'unconditional' should be qualified by d>1 and (160).

Bottom line: the central claim is credible and the fix is straightforward. Send it to referees, but ask them to verify the linearization step and the imported propositions.

Referee Report

3 major / 5 minor

Summary. The paper develops a regularity theory for degenerate drift-diffusion equations of the form ∂tρ = div(G∇ρ) + ⟨J,∇ρ⟩ + ρ + g and the conservative variant (12), under growth and structural conditions on G,J,g (Assumptions 2.6, 2.7, 2.17). The main theorem 2.5 establishes existence, uniqueness, and regularity (classes Γ, Γ′, Γ″) via Galerkin approximation on expanding tori. The authors apply this to the CBO mean-field PDE (145), showing that the unique measure solution lies in Γ. Under d > 1 and the additional growth condition (160), a change of variables v ↦ v + vα(t) puts the equation in uniformly parabolic form on annuli, and a parabolic Harnack argument, seeded by a mass-persistence result [24, Proposition 23], yields positivity of ρ(v,t) for v ≠ vα(t). Combined with [24, Theorems 3.7 and 3.8], this is claimed to remove the initial-support assumption (7) and give unconditional global convergence of CBO. The paper also contains a d = 1 counterexample showing that the positivity statement can fail in one dimension.

Significance. If the proof is completed, the paper would be a significant contribution: it supplies the first well-posedness and regularity analysis for the degenerate CBO Fokker–Planck equation and removes the mass-near-the-minimizer assumption that has been a standing technical limitation. The Galerkin argument is detailed and largely self-contained; the use of [24] for the final convergence step is appropriate and is an independent, peer-reviewed result; and the d = 1 counterexample in Remark 2.23 is a valuable delimitation. However, the advertised 'unconditional' convergence is qualified by d > 1 and the extra growth condition (160), and two technical bridges in the application to CBO need repair before the main claim is fully justified.

major comments (3)

[§2.4, Eqs. (146)–(148), Theorem 2.5] Theorem 2.5 is stated for equations (11)/(12), whose zeroth-order term is +ρ + g. In contrast, the linearized CBO equation (146) has zeroth-order coefficient 3d, namely ∂tρ = div(‖v − vα(t)‖²∇ρ) + 3⟨v − vα(t),∇ρ⟩ + 3dρ. Therefore the assertion 'By Theorem 2.5 ... equation (146) has a unique weak solution and ρ ∈ Γ' is not justified as written. The same gap affects equation (148) in §2.4.1. This is load-bearing because ρ ∈ Γ is exactly what allows the shifted equation (154) to be treated classically and the Harnack step to be applied. The gap is repairable, for example by rescaling τ = 3dt and G,J by 1/(3d), or by stating Theorem 2.5 for ∂tρ = div(G∇ρ) + ⟨J,∇ρ⟩ + cρ + g with c > 0, but this reduction must be supplied explicitly in the revision.
[§2.4.2, Eqs. (153)–(155)] The parabolic Harnack inequality is invoked from [22, Section 7.1, Theorem 10] for the non-divergence equation (154), which contains the drift 5v + P(t) and the zeroth-order term 3dρ̃. The cited theorem is a Harnack inequality for divergence-form equations without lower-order terms. A Harnack inequality does hold for uniformly parabolic operators with bounded lower-order coefficients (e.g., Krylov–Safonov, or after the substitution u = e^{−3dt}ρ̃ and an appropriate transformation to divergence form), but the manuscript must either quote a theorem covering this class or perform the reduction explicitly. As written, the positivity conclusion is not fully supported by the cited result.
[Abstract and Theorem 2.25] Theorem 2.25 proves positivity only for d > 1 and under the additional growth condition (160), and Remark 2.23 explicitly shows that the d = 1 statement is false. The abstract and introduction nevertheless state that CBO solutions 'remain positive and maintain full support' and that convergence is 'unconditional', omitting both conditions. This overstates the proven result and should be corrected in the abstract, the introduction, and the title if it is retained. The body of the paper is more careful than the abstract, but the advertised central claim is the unconditional statement.

minor comments (5)

[§2.4.1, Eq. (148)] The diffusion coefficient in (148) is written as ‖v − vα(ρ,t)‖ρ̇; it should be ‖v − vα(ρ,t)‖²ρ̇ to match (145) and (146).
[Assumption 2.17, Eq. (118), and Remark 2.18] The symbol f is used in the integrability condition (118); from the proof in §2.3.3, and specifically (128), the condition is on the inhomogeneous term g, not on the objective function f. Please replace f by g (or introduce it clearly).
[§2.4.2, after Eq. (155)] The phrase 'ǫ < 0 small enough' should be '0 < ε < R'; the current inequality is nonsensical as written.
[§2.3.3, Eqs. (139)–(141)] The definition of h in the weak-uniqueness argument is garbled: the expression '(η_{R,δ}(z_ϵ))_ϵ' is not a clear definition. Please rewrite the truncation and mollification step in display form.
[§2.4.2, use of [24, Proposition 23]] The precise hypotheses of [24, Proposition 23] are not stated before it is used to seed the Harnack argument; since this is the step that produces the initial positive mass, name the assumptions of the proposition and verify them for equation (151).

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the regularity and positivity derivation is self-contained, and the imported self-cited results are independent published results not equivalent to the paper's inputs.

full rationale

The main theorem (Theorem 2.5) is proved inside the paper by Galerkin approximation from Assumptions 2.6, 2.7, and 2.17; it is not derived by fitting a parameter or by defining the solution class in terms of the CBO conclusion. The CBO application fixes v_alpha(t) from an independently established measure solution in [14] and then applies Theorem 2.5 to the linearized equation; the later identification rho_bar = rho is a standard uniqueness argument rather than a reformulation of the target. The positivity result is derived within the paper via the shifted equation (153), the polynomial-growth bound in Lemma 2.24, and the parabolic Harnack inequality, with the initial positive-mass seed supplied by [24, Proposition 23]. That cited proposition is an independent, peer-reviewed, parameter-free result, and the final convergence statement imports [24, Theorems 3.7 and 3.8] as external results; although [24] shares an author, its role is to supply the quantitative convergence once positivity removes assumption (7), not to assume the positivity conclusion. No equation in the paper is defined in terms of the result it is used to prove, and no fitted quantity is renamed as a prediction. The apparent coefficient mismatch in the expansion of the CBO equation (drift coefficient 5 in (153) versus 3 in (146)) is a correctness or consistency concern about whether Theorem 2.5 applies exactly as written, not a circularity, so it is not scored here.

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

The central claim rests on prior results from [14,24], on the technical growth assumptions for G (verified for the CBO equation), and on standard parabolic regularity tools. No free parameters are fitted to data.

assumptions (5)

domain assumption Existence, uniqueness, and quantitative convergence of the CBO mean-field dynamics under the initial support condition (7) [24, Theorem 3.7, 3.8].
Imported in the conclusion to turn positivity into unconditional global convergence; not re-derived in this paper.
domain assumption Persistent positive mass: if ρ0(B_r(v))>0 then ρt(B_r(v))>0 for t>0 [24, Proposition 23].
Used in Section 2.4.2 to seed the Harnack inequality; its hypotheses are not restated.
domain assumption Unique strong solution of the nonlinear CBO SDE [14, Theorem 3.2].
Used in Section 2.4.1 to conclude uniqueness of the measure solution.
domain assumption Objective f satisfies (144) and, for the positivity step, the polynomial growth bounds (160) on ∇f and Δf; initial density smooth in H^m for all m; d>1.
These are the paper's standing hypotheses for the main application.
standard math Standard PDE tools: Galerkin method, Sobolev embeddings, Gronwall lemma, Faà di Bruno formula, parabolic Harnack inequality, trace theorem.
Used throughout Sections 2.3 and 2.4.2 without derivation.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Regularity and positivity of solutions of the Consensus-Based Optimization equation: unconditional global convergence." pith.science (2026). https://pith.science/paper/JMN72B75

@misc{pith2026250201434,
  author       = {Pith},
  title        = {Pith review of: Regularity and positivity of solutions of the Consensus-Based Optimization equation: unconditional global convergence},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/JMN72B75}},
  note         = {Machine review of arXiv:2502.01434}
}

read the original abstract

Introduced in 2017 \cite{B1-pinnau2017consensus}, Consensus-Based Optimization (CBO) has rapidly emerged as a significant breakthrough in global optimization. This straightforward yet powerful multi-particle, zero-order optimization method draws inspiration from Simulated Annealing and Particle Swarm Optimization. Using a quantitative mean-field approximation, CBO dynamics can be described by a nonlinear Fokker-Planck equation with degenerate diffusion, which does not follow a gradient flow structure. In this paper, we demonstrate that solutions to the CBO equation remain positive and maintain full support. Building on this foundation, we establish the {\it unconditional} global convergence of CBO methods to global minimizers. Our results are derived through an analysis of solution regularity and the proof of existence for smooth, classical solutions to a broader class of drift-diffusion equations, despite the challenges posed by degenerate diffusion.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

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

Stochastic Zeroth-Order Method for Computing Generalized Rayleigh Quotients
math.OC 2025-12 conditional novelty 6.0 of 10

A random one-dimensional line-search algorithm computes the largest generalized Rayleigh quotient using only matrix-vector products with A and B, without adjoints or B^{-1}, with almost-sure convergence to the global ...
Consensus-based optimization for closed-box adversarial attacks and a connection to evolution strategies
math.OC 2025-06 conditional novelty 5.0 of 10

Consensus-based optimization matches or beats natural evolution strategies as a closed-box adversarial attack method in easier attack scenarios, and consensus hopping is shown to be a gradient-descent-like limit of CBO.

Reference graph

Works this paper leans on

48 extracted references · 39 canonical work pages · cited by 2 Pith papers

[24]

Fornasier, T

M. Fornasier, T. Klock, and K. Riedl. Consensus-based o ptimization methods converge globally. SIAM Journal on Optimization , 34(3):2973–3004, 2024

2024
[1]

Bae, S.-Y

H.-O. Bae, S.-Y. Ha, M. Kang, H. Lim, C. Min, and J. Yoo. A co nstrained consensus based optimization algorithm and its application to ﬁnance. Appl. Math. Comput. , 416:Paper No. 126726, 10, 2022. 30

work page 2022
[2]

CBX: Python and Julia packages for consensus-based interacting particle methods

R. Bailo, A. Barbaro, S. N. Gomes, K. Riedl, T. Roith, C. To tzeck, and U. Vaes. CBX: Python and Julia packages for consensus-based interacting particle methods. arXiv preprint arXiv:2403.14470, 2024

work page Pith review arXiv 2024
[3]

Beddrich, E

J. Beddrich, E. Chenchene, M. Fornasier, H. Huang, and B. Wohlmuth. Constrained consensus-based optimization and numerical heuristics fo r the few particle regime. arXiv:2410.10361, 2024

arXiv 2024
[4]

Bellavia and G

S. Bellavia and G. Malaspina. A discrete consensus-base d global optimization method with noisy objective function. Journal of Optimization Theory and Applications , 206(1):20, 2025

work page 2025
[5]

Benfenati, G

A. Benfenati, G. Borghi, and L. Pareschi. Binary interac tion methods for high dimensional global optimization and machine learning. Appl. Math. Optim. , 86(1):9, 2022

work page 2022
[6]

Blum and A

C. Blum and A. Roli. Metaheuristics in combinatorial opt imization: Overview and con- ceptual comparison. ACM Comput. Surv. , 35(3):268–308, 2003

work page 2003
[7]

Bonandin and M

S. Bonandin and M. Herty. Consensus-based algorithms fo r stochastic optimization prob- lems. arXiv:2404.10372, 2024

arXiv 2024

Show all 48 references

[8]

Borghi, M

G. Borghi, M. Herty, and L. Pareschi. A consensus-based a lgorithm for multi-objective optimization and its mean-ﬁeld description. In 2022 IEEE 61st Conference on Decision and Control (CDC) , pages 4131–4136. IEEE, 2022

2022
[9]

Borghi, M

G. Borghi, M. Herty, and L. Pareschi. An adaptive consens us based method for multi- objective optimization with uniform Pareto front approxim ation. Applied Mathematics & Optimization, 88(2):1–43, 2023

2023
[10]

Borghi, M

G. Borghi, M. Herty, and L. Pareschi. Constrained conse nsus-based optimization. SIAM J. Optim. , 33(1):211–236, 2023

2023
[11]

Bungert, F

L. Bungert, F. Hoﬀmann, D. Y. Kim, and T. Roith. Mirrorcbo : A consensus-based opti- mization method in the spirit of mirror descent. arXiv:2501.12189, 2025

2025 arXiv
[12]

Bungert, P

L. Bungert, P. Wacker, and T. Roith. Polarized consensu s-based dynamics for optimization and sampling. Mathematical Programming, pages 1–31, 2024

2024
[13]

Byeon, S

J. Byeon, S. Ha, G. Hwang, D. Ko, and J. Yoon. Consensus, e rror estimates and applications of ﬁrst- and second-order consensus-based optimization al gorithms. Mathematical Models and Methods in Applied Sciences , 35(2):345–401, Feb. 2025. Publisher Copyright: © 2025 World Sci...

2025
[14]

J. A. Carrillo, Y.-P. Choi, C. Totzeck, and O. Tse. An ana lytical framework for consensus- based global optimization method. Mathematical Models and Methods in Applied Sciences , 28(06):1037–1066, 2018

2018
[15]

J. A. Carrillo, F. Hoﬀmann, A. M. Stuart, and U. Vaes. Cons ensus-based sampling. Stud. Appl. Math. , 148(3):1069–1140, 2022

2022
[16]

J. A. Carrillo, S. Jin, L. Li, and Y. Zhu. A consensus-bas ed global optimization method for high dimensional machine learning problems. ESAIM Control Optim. Calc. Var. , 27:Paper No. S5, 22, 2021. 31

2021
[17]

J. A. Carrillo, C. Totzeck, and U. Vaes. Consensus-base d optimization and ensemble kalman inversion for global optimization problems with constrain ts. In Modeling and Simulation for Collective Dynamics , pages 195–230. World Scientiﬁc, 2023

2023
[18]

J. A. Carrillo, N. G. Trillos, S. Li, and Y. Zhu. FedCBO: R eaching group consensus in clustered federated learning through consensus-based opt imization. J. Mach. Learn. Res. , 25(214):1–51, 2024

2024
[19]

Chenchene, H

E. Chenchene, H. Huang, and J. Qiu. A consensus-based al gorithm for non-convex multi- player games. arXiv:2311.08270, 2024

2024 arXiv
[20]

Cipriani, H

C. Cipriani, H. Huang, and J. Qiu. Zero-inertia limit: f rom particle swarm optimization to consensus-based optimization. SIAM J. Math. Anal. , 54(3):3091–3121, 2022

2022
[21]

Constantine and T

G. Constantine and T. Savits. A multivariate Faa di Brun o formula with applications. Transactions of the American Mathematical Society , 348(2):503–520, 1996

1996
[22]

L. C. Evans. Partial Diﬀerential Equations , volume 19. American Mathematical Society, 2022

2022
[23]

Fornasier, H

M. Fornasier, H. Huang, L. Pareschi, and P. S¨ unnen. Con sensus-based optimization on hypersurfaces: Well-posedness and mean-ﬁeld limit. Math. Models Methods Appl. Sci. , 30(14):2725–2751, 2020

2020
[25]

Fornasier, L

M. Fornasier, L. Pareschi, H. Huang, and P. S¨ unnen. Con sensus-based optimization on the sphere: Convergence to global minimizers and machine learn ing. J. Mach. Learn. Res. , 22:237:1–237:55, 2021

2021
[26]

Fornasier, P

M. Fornasier, P. Richt´ arik, K. Riedl, and L. Sun. Conse nsus-based optimization with truncated noise. Eur. J. Appl. Math. , pages 1–24, 2024

2024
[27]

Fornasier and L

M. Fornasier and L. Sun. A PDE framework of consensus-ba sed optimization for objectives with multiple global minimizers. Communications in Partial Diﬀerential Equations , to appear
[28]

Gendreau and J.-Y

M. Gendreau and J.-Y. Potvin, editors. Handbook of Metaheuristics . Springer, New York, NY, USA, 2 edition, 2013

2013
[29]

N. J. Gerber, F. Hoﬀmann, and U. Vaes. Mean-ﬁeld limits fo r consensus-based optimization and sampling. arXiv:2312.07373, 2023

2023
[30]

Grassi, H

S. Grassi, H. Huang, L. Pareschi, and J. Qiu. Mean-ﬁeld p article swarm optimization. In Modeling and Simulation for Collective Dynamics , pages 127–193. World Scientiﬁc, 2023

2023
[31]

Grassi and L

S. Grassi and L. Pareschi. From particle swarm optimiza tion to consensus based opti- mization: stochastic modeling and mean-ﬁeld limit. Mathematical Models and Methods in Applied Sciences, 31(08):1625–1657, 2021

2021
[32]

S.-Y. Ha, M. Kang, and D. Kim. Emergent behaviors of high -dimensional Kuramoto models on Stiefel manifolds. Automatica, 136:Paper No. 110072, 2022. 32

2022
[33]

Hansen, A

N. Hansen, A. Auger, R. Ros, O. Mersmann, T. Tuˇ sar, and D . Brockhoﬀ. COCO: A platform for comparing continuous optimizers in a black-bo x setting. Optimization Methods and Software, 36(1):114–144, 2021

2021
[34]

Herty, Y

M. Herty, Y. Huang, D. Kalise, and H. Kouhkouh. A multisc ale consensus-based algorithm for multi-level optimization. Mathematical Models and Methods in Applied Sciences , to appear
[35]

Huang, J

H. Huang, J. Qiu, and K. Riedl. On the global convergence of particle swarm optimization methods. Applied Mathematics & Optimization , 88(2):30, 2023

2023
[36]

Huang, J

H. Huang, J. Qiu, and K. Riedl. Consensus-based optimiz ation for saddle point problems. SIAM J. Control Optim. , 62(2):1093–1121, 2024

2024
[37]

Huang, M

Y. Huang, M. Herty, D. Kalise, and N. Kantas. Fast and rob ust consensus-based optimiza- tion via optimal feedback control. arXiv:2411.03051, 2024

2024 arXiv
[38]

Jingrun, J

C. Jingrun, J. Shi, and L. Lyu. A consensus-based global optimization method with adap- tive momentum estimation. Communications in Computational Physics , 31(4):1296–1316, 2022

2022
[39]

Kalise, A

D. Kalise, A. Sharma, and M. V. Tretyakov. Consensus-ba sed optimization via jump- diﬀusion stochastic diﬀerential equations. Mathematical Models and Methods in Applied Sciences, 33(02):289–339, 2023

2023
[40]

Klamroth, M

K. Klamroth, M. Stiglmayr, and C. Totzeck. Consensus-b ased optimization for multi- objective problems: a multi-swarm approach. Journal of Global Optimization , 89(3):745– 776, 2024

2024
[41]

Ko, S.-Y

D. Ko, S.-Y. Ha, S. Jin, and D. Kim. Convergence analysis of the discrete consensus- based optimization algorithm with random batch interactio ns and heterogeneous noises. Mathematical Models and Methods in Applied Sciences , 32(06):1071–1107, 2022

2022
[42]

Pinnau, C

R. Pinnau, C. Totzeck, O. Tse, and S. Martin. A consensus -based model for global opti- mization and its mean-ﬁeld limit. Math. Models Methods Appl. Sci. , 27(1):183–204, 2017

2017
[43]

E. Platen. An introduction to numerical methods for sto chastic diﬀerential equations. In Acta numerica, 1999 , volume 8 of Acta Numer. , pages 197–246. Cambridge Univ. Press, Cambridge, 1999

1999
[44]

K. Riedl. Leveraging memory eﬀects and gradient informa tion in consensus-based optimisa- tion: On global convergence in mean-ﬁeld law. European Journal of Applied Mathematics , 35(4):483–514, 2024

2024
[45]

C. Totzeck. Trends in consensus-based optimization. I n Active Particles, Volume 3: Ad- vances in Theory, Models, and Applications , pages 201–226. Springer, 2021

2021
[46]

Totzeck, R

C. Totzeck, R. Pinnau, S. Blauth, and S. Schotth¨ ofer. Anumerical comparison of consensus- based global optimization to other particle-based global o ptimization schemes. PAMM, 18(1):e201800291, 2018

2018
[47]

Totzeck and M.-T

C. Totzeck and M.-T. Wolfram. Consensus-based global o ptimization with personal best. Math. Biosci. Eng. , 17(5):6026–6044, 2020

2020
[48]

N. G. Trillos, S. Li, K. Riedl, and Y. Zhu. CB 2O: Consensus-based bi-level optimization. arXiv:2411.13394, 2024. 33 3 Appendix The proof directly follow the proofs of Lemmas 2.9, 2.11 and 2 .12. Let us ﬁx any t ∈ [0,T ], and for simplicity, we denote G(·) := G(·,t ). Then we ...

2024 arXiv

Pith tools

Reviewed August 9, 2026 · model on record in the stance chip above.

[1] [24]

Fornasier, T

M. Fornasier, T. Klock, and K. Riedl. Consensus-based o ptimization methods converge globally. SIAM Journal on Optimization , 34(3):2973–3004, 2024

2024

[2] [1]

Bae, S.-Y

H.-O. Bae, S.-Y. Ha, M. Kang, H. Lim, C. Min, and J. Yoo. A co nstrained consensus based optimization algorithm and its application to ﬁnance. Appl. Math. Comput. , 416:Paper No. 126726, 10, 2022. 30

work page 2022

[3] [2]

CBX: Python and Julia packages for consensus-based interacting particle methods

R. Bailo, A. Barbaro, S. N. Gomes, K. Riedl, T. Roith, C. To tzeck, and U. Vaes. CBX: Python and Julia packages for consensus-based interacting particle methods. arXiv preprint arXiv:2403.14470, 2024

work page Pith review arXiv 2024

[4] [3]

Beddrich, E

J. Beddrich, E. Chenchene, M. Fornasier, H. Huang, and B. Wohlmuth. Constrained consensus-based optimization and numerical heuristics fo r the few particle regime. arXiv:2410.10361, 2024

arXiv 2024

[5] [4]

Bellavia and G

S. Bellavia and G. Malaspina. A discrete consensus-base d global optimization method with noisy objective function. Journal of Optimization Theory and Applications , 206(1):20, 2025

work page 2025

[6] [5]

Benfenati, G

A. Benfenati, G. Borghi, and L. Pareschi. Binary interac tion methods for high dimensional global optimization and machine learning. Appl. Math. Optim. , 86(1):9, 2022

work page 2022

[7] [6]

Blum and A

C. Blum and A. Roli. Metaheuristics in combinatorial opt imization: Overview and con- ceptual comparison. ACM Comput. Surv. , 35(3):268–308, 2003

work page 2003

[8] [7]

Bonandin and M

S. Bonandin and M. Herty. Consensus-based algorithms fo r stochastic optimization prob- lems. arXiv:2404.10372, 2024

arXiv 2024