arxiv: 2604.15761 · v1 · submitted 2026-04-17 · 💻 cs.NE · math.OC

Recognition: unknown

Frenetic Cat-inspired Particle Optimization: a Markov state-switching hybrid swarm optimizer with application to cardiac digital twinning

Jorge S\'anchez , Guadalupe Garc\'ia-Isla , Sandra Perez-Herrero , Beatriz Trenor , Javier Saiz

Authors on Pith no claims yet

Pith reviewed 2026-05-10 07:55 UTC · model grok-4.3

classification 💻 cs.NE math.OC

keywords swarm optimizationMarkov switchingparticle swarmblack-box optimizationcardiac digital twinhybrid optimizerevolutionary computationinverse problem

0 comments

The pith

FCPO combines particle swarm dynamics with a Markov state-switching controller to cut runtime while reaching target accuracy on benchmarks and cardiac model calibration.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops Frenetic Cat-inspired Particle Optimization as a hybrid swarm method to handle expensive black-box problems where evaluation budgets are limited. It pairs PSO-like particle motion with an explicit Markov controller that switches the swarm between exploration and refinement states on the fly. Four supporting mechanisms are added: bounded state-conditioned steps, an elite-difference jump to escape plateaus, covariance-guided local search around top solutions, and gradual population shrinkage. Across ten CEC 2022 test cases the method records the lowest average runtime while matching or beating classical swarms on multimodal functions; in a ventricular activation twin it meets the 0.1 mV RMSE target in roughly 40 iterations with stable results from varied starts.

Core claim

FCPO integrates PSO-like dynamics with an explicit-state Markov switching controller to schedule exploration and refinement operators online. The controller triggers state-conditioned bounded motion, an elite-difference global jump operator, eigen-space guided local refinement from elite covariance, and linear population size reduction. On five CEC 2022 functions at dimensions 10 and 20 it records the lowest mean runtime (0.183 s) across all cases, 2.3 times faster than CMA-ES and 2.6 times faster than L-SHADE in the authors' Python implementation. On the composition function F10 at D=20 it also returns the best mean objective value while remaining faster than CMA-ES. In the ventricular-ECG-

What carries the argument

The explicit-state Markov switching controller that dynamically selects among bounded motion, elite-difference jumps, eigen-space refinement, and population reduction according to the swarm's current state.

If this is right

On the tested CEC functions FCPO supplies a better accuracy-runtime trade-off than classical particle swarms and remains faster than CMA-ES on multimodal cases.
The method reaches the target ECG fidelity (RMSE below 0.1 mV) in the ventricular activation twin within about 40 iterations with consistent convergence across restarts.
Linear population reduction limits late-stage cost while the jump and eigen-space operators help escape stagnation on structured and hybrid landscapes.
The approach is positioned as practical for other inverse problems where each objective evaluation is computationally heavy.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The Markov controller could be grafted onto other swarm or evolutionary methods to automate their exploration-refinement balance without manual tuning.
If the speed advantage persists on additional expensive tasks, FCPO-style switching may become a default choice for calibration loops in simulation-based medical modeling.
The eigen-space refinement step might be combined with surrogate models to further reduce evaluations in even higher-dimensional cardiac or physiological fitting problems.

Load-bearing premise

The particular Markov transition rules, elite-difference jump, eigen-space refinement, and linear population schedule together produce a general accuracy-runtime advantage that holds outside the five chosen CEC functions and the single cardiac calibration example.

What would settle it

Applying FCPO to a wider set of CEC or real-world expensive problems and measuring whether its mean runtime remains lower than CMA-ES while objective values stay competitive or better.

Figures

Figures reproduced from arXiv: 2604.15761 by Beatriz Trenor, Guadalupe Garc\'ia-Isla, Javier Saiz, Jorge S\'anchez, Sandra Perez-Herrero.

**Figure 1.** Figure 1: The algorithm iteratively combines population reduction, state-dependent particle updates, fitness evaluation, late-stage refinement, and Markov-state transitions until termination. In the figure, X and V denote particle positions and velocities, S the vector of particle states, A the Markov transition matrix, P𝑏𝑒𝑠𝑡 the personalbest positions, g𝑏𝑒𝑠𝑡 the global-best solution, 𝑃 the current population size,… view at source ↗

**Figure 2.** Figure 2: Pareto-style runtime–accuracy comparison for the CEC 2022 benchmark subset at A) 𝐷 = 10 and B) 𝐷 = 20. Markers show each optimizer’s mean runtime and relative median error over 30 runs. Relative median error is normalized by the best median final objective value obtained in each case, so the best method is located at 1 (horizontal dashed line) and larger values indicate worse performance. Lower values on b… view at source ↗

**Figure 3.** Figure 3: Convergence profiles on representative CEC benchmark functions at 𝐷 = 20. The best-so-far error is reported as a function of the number of function evaluations (NFE), using logarithmic scales on both axes, for FCPO and six baseline optimizers. Panels correspond to: A) F1: Rotated Zakharov, B) F2: Rotated Rosenbrock, C) F6: Hybrid Function 1, and D) F10: Composition Function 2. 6.2. Overall trends across al… view at source ↗

**Figure 4.** Figure 4: Ablation analysis of FCPO on representative benchmark functions. The convergence behavior of the full FCPO algorithm is compared with three reduced variants, namely FCPO_NoZoom, FCPO_NoEigen, and FCPO_NoLPSR, using NFE-matched best-so-far error curves. Panels correspond to: A) F6: Hybrid Function 1 (𝐷 = 10), B) F6: Hybrid Function 1 (𝐷 = 20), C) F10: Composition Function 2 (𝐷 = 10), and D) F10: Composition… view at source ↗

**Figure 5.** Figure 5: Ventricular activation digital-twin calibration using FCPO. A) Simulated biventricular activation time map (posterior and anterior views) obtained from the optimized Purkinje–myocyte junctions (PMJs) distribution; colors denote local activation time (ms) relative to the onset, with the range shown in the color bar. B) Twelve-lead ECG comparison between the clinical target (black dashed) and the ECG simulat… view at source ↗

**Figure 6.** Figure 6: Convergence and activation-time standard deviation of FCPO in ventricular digital-twin calibration. A) Best loss versus iteration, shown as the mean over 10 independent runs (black line) with ±1 standard deviation (gray band). B) Spatial endocardial activation variability over 10 independent runs, shown as the nodewise standard deviation of activation time, 𝜎𝑡𝑎 (x), computed across all final best solutions… view at source ↗

read the original abstract

Designing optimizers that remain effective under tight evaluation budgets is critical in expensive black-box settings such as cardiac digital twinning. We propose Frenetic Cat-inspired Particle Optimization (FCPO), a hybrid swarm method that couples particle swarm optimization-like dynamics with an explicit-state Markov switching controller to schedule exploration and refinement operators online. FCPO integrates (i) state-conditioned bounded motion, (ii) an elite-difference global jump operator to escape stagnation, (iii) eigen-space guided local refinement from elite covariance, and (iv) linear population size reduction to control late-stage computational cost. We benchmark FCPO on five representative functions from the Congress on Evolutionary Computation (CEC) 2022 suite (F1, F2, F3, F6 and F10) at dimensions D$\in${10,20} over 30 independent runs, comparing against PSO, CSO, CLPSO, SHADE, L-SHADE and CMA-ES. FCPO achieves the lowest mean runtime across the ten benchmark cases (average 0.183 s), about 2.3x faster than CMA-ES and 2.6x faster than L-SHADE in our Python implementation. On the multimodal composition function F10 at D=20, FCPO attains the best mean objective (9.625x 10^2 $\pm$ 1.275x 10^3) and remains faster than CMA-ES (0.602 s vs. 1.126 s mean runtime). On structured landscapes (F1--F3) and on the hybrid function (F6), CMA-ES remains the most accurate method, while FCPO substantially improves over classical swarms and maintains a favorable accuracy--runtime trade-off. Finally, in a ventricular activation digital twin calibration task, FCPO reaches the target electrocardiogram (ECG) fidelity (RMSE<0.1 mV) within ~ 40 iterations and produces physiologically plausible activation maps with robust convergence across repeated initializations, supporting its use as a practical optimizer for expensive inverse problems.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

FCPO packages familiar swarm components with a Markov switch and shows runtime wins on limited tests, but lacks the stats and controls needed for strong claims.

read the letter

FCPO is a new hybrid swarm optimizer using Markov state switching to manage exploration and refinement, and it claims faster runtimes than standard methods on a few benchmarks while working for cardiac model fitting. The combination of bounded PSO motion, elite jumps, covariance refinement, and shrinking population under Markov control is the fresh packaging. The pieces come from PSO and CMA-ES lines, but this specific controller setup is new. It does a solid job reporting consistent Python runtimes across 30 runs and applying the method to calibrate a ventricular activation model to ECG data, hitting low error in reasonable iterations. The weak points are the small set of only five CEC functions, missing statistical significance tests despite the runs, no ablations of the operators, and no head-to-head in the cardiac case. The high variance on F10 means the 'best mean' might not be reliable, and without fixed budgets it's unclear if the speed is from the algorithm or the implementation. This is aimed at evolutionary computation folks who need practical tools for costly black-box optimization, or biomedical engineers building digital twins. Someone in those areas could find the runtime comparisons and the application useful. It deserves a serious referee because the results are presented with enough detail to evaluate and the application adds relevance. I'd send it for peer review, but flag the need for more rigorous testing like p-values and component ablations in the revisions.

Referee Report

4 major / 2 minor

Summary. The paper proposes Frenetic Cat-inspired Particle Optimization (FCPO), a hybrid swarm optimizer that uses an explicit-state Markov controller to switch between particle-swarm-like bounded motion, an elite-difference global jump operator, eigen-space local refinement from elite covariance, and linear population-size reduction. It reports results on five CEC 2022 functions (F1, F2, F3, F6, F10) at D=10 and D=20 over 30 runs, claiming the lowest mean runtime (0.183 s average) across the ten cases—2.3× faster than CMA-ES and 2.6× faster than L-SHADE in the authors’ Python implementation—while attaining the best mean objective on F10 at D=20. The method is further applied to ventricular activation digital-twin calibration, where it reaches ECG RMSE < 0.1 mV in roughly 40 iterations with robust convergence.

Significance. If the accuracy–runtime trade-off is confirmed under controlled evaluation budgets and statistical testing, FCPO would supply a practical adaptive swarm method for expensive black-box problems, particularly inverse modeling tasks such as cardiac digital twinning. The explicit Markov scheduling of exploration and refinement operators is a structured hybridization that could be reusable; the cardiac example demonstrates direct applicability to physiologically constrained optimization.

major comments (4)

[Benchmark results] Benchmark results (implicitly Section 4 or 5): the claim that FCPO attains the lowest mean runtime across all ten cases is not supported by statistical significance tests (Wilcoxon or Friedman with post-hoc p-values). The reported 2.3× and 2.6× speed-ups versus CMA-ES and L-SHADE therefore remain descriptive rather than inferential, especially given the large standard deviation on F10 (962.5 ± 1275).
[Experimental design] Experimental design: no ablation or component-removal experiments are presented for the four core operators (Markov state transitions, elite-difference jump, eigen-space refinement, linear population reduction). Without these, it is impossible to determine whether the reported advantage stems from the proposed combination or from implementation details, function selection, or the single Python runtime environment.
[Cardiac application] Cardiac digital-twin calibration section: convergence to RMSE < 0.1 mV within ~40 iterations is shown, yet no head-to-head comparison against CMA-ES, L-SHADE or the other baselines is provided under identical evaluation budgets, initializations, or stopping criteria. This omission prevents assessment of whether FCPO’s practical utility exceeds that of established methods on the target application.
[Method and parameter settings] Parameter reporting: the Markov transition probabilities, elite-difference jump scale, and population-reduction rate are listed as free parameters but are neither tabulated with exact values nor subjected to sensitivity analysis, undermining reproducibility and claims of robustness.

minor comments (2)

[Introduction] The abstract and introduction refer to “Frenetic Cat-inspired” behavior without a concise paragraph linking feline movement heuristics to the specific Markov states or operators; a short motivation subsection would improve clarity.
[Experimental protocol] Runtime figures are wall-clock times from a single Python implementation; adding a fixed function-evaluation budget column would allow fairer algorithmic comparison independent of language overhead.

Simulated Author's Rebuttal

4 responses · 0 unresolved

We thank the referee for the constructive and detailed comments, which have helped us strengthen the manuscript. We address each major comment below and indicate the revisions made.

read point-by-point responses

Referee: Benchmark results (implicitly Section 4 or 5): the claim that FCPO attains the lowest mean runtime across all ten cases is not supported by statistical significance tests (Wilcoxon or Friedman with post-hoc p-values). The reported 2.3× and 2.6× speed-ups versus CMA-ES and L-SHADE therefore remain descriptive rather than inferential, especially given the large standard deviation on F10 (962.5 ± 1275).

Authors: We agree that statistical significance testing is required to support the runtime claims rigorously. In the revised manuscript we have added Wilcoxon signed-rank tests comparing runtimes of FCPO against each baseline across all ten benchmark cases. The tests show statistically significant advantages (p < 0.05) for FCPO in eight of the ten cases. For F10 at D=20 we retain the reported mean and standard deviation but explicitly discuss the high variance and note that the mean runtime advantage remains consistent with the other cases. These additions appear in the updated Section 4 and a new supplementary table. revision: yes
Referee: Experimental design: no ablation or component-removal experiments are presented for the four core operators (Markov state transitions, elite-difference jump, eigen-space refinement, linear population reduction). Without these, it is impossible to determine whether the reported advantage stems from the proposed combination or from implementation details, function selection, or the single Python runtime environment.

Authors: The referee correctly notes the absence of ablation experiments. Because the Markov controller dynamically schedules the operators, fully independent removals are not straightforward. We have nevertheless added a partial ablation study in the revised manuscript: each operator is disabled individually on a subset of functions while keeping the remaining components and the state machine intact. The results, presented in a new subsection of Section 5, indicate that every operator contributes measurably to the observed runtime-accuracy trade-off, with the full combination performing best. We acknowledge that a complete factorial design would be desirable and have flagged this as future work. revision: partial
Referee: Cardiac digital-twin calibration section: convergence to RMSE < 0.1 mV within ~40 iterations is shown, yet no head-to-head comparison against CMA-ES, L-SHADE or the other baselines is provided under identical evaluation budgets, initializations, or stopping criteria. This omission prevents assessment of whether FCPO’s practical utility exceeds that of established methods on the target application.

Authors: We concur that direct comparisons on the cardiac task are essential. In the revised manuscript we have added head-to-head results for FCPO versus CMA-ES and L-SHADE on the ventricular activation calibration problem. All methods were run from identical initial populations, with the same evaluation budget (maximum 100 iterations) and identical stopping criterion (RMSE < 0.1 mV or budget exhausted). The new experiments, reported in Section 6, show that FCPO reaches the target fidelity in fewer iterations on average while producing comparable activation maps. Corresponding figures and statistical summaries have been included. revision: yes
Referee: Parameter reporting: the Markov transition probabilities, elite-difference jump scale, and population-reduction rate are listed as free parameters but are neither tabulated with exact values nor subjected to sensitivity analysis, undermining reproducibility and claims of robustness.

Authors: We thank the referee for highlighting this reproducibility issue. We have inserted a new table (Table 2) that lists every hyper-parameter with its exact numerical value used in all reported experiments, including the Markov transition matrix entries, the elite-difference jump scale factor, and the linear population-reduction schedule. In addition, we performed a sensitivity analysis by perturbing each key parameter by ±20 % and report the resulting changes in runtime and final objective value in Appendix B. The analysis confirms that performance remains stable within the tested ranges, supporting the robustness claim. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical benchmarks on independent CEC functions and separate cardiac task

full rationale

The paper proposes FCPO as an algorithmic construction (Markov state-switching + elite jump + eigen refinement + linear reduction) and reports wall-clock runtimes plus objective values on five external CEC 2022 functions plus one independent ventricular activation calibration whose ECG target is external to the optimizer design. No equations, fitted parameters, or self-citations are shown that reduce the reported performance numbers to quantities defined by the same inputs; the evaluation remains statistically independent of the method's internal definitions.

Axiom & Free-Parameter Ledger

3 free parameters · 1 axioms · 0 invented entities

The central claim rests on several design choices whose values are not derived from first principles or external benchmarks but selected to produce the reported performance.

free parameters (3)

Markov state transition probabilities
Switching rates between exploration and refinement states are introduced without derivation and must be set to achieve the claimed scheduling behavior.
Elite-difference jump scale
Magnitude of the global jump operator based on elite differences is a tunable design parameter.
Population reduction rate
Linear schedule for shrinking swarm size is chosen rather than derived.

axioms (1)

domain assumption Markov chain state transitions can be chosen to improve the exploration-refinement trade-off over fixed schedules
The paper assumes the explicit-state controller provides online adaptation benefits without proving optimality of the transition matrix.

pith-pipeline@v0.9.0 · 5701 in / 1483 out tokens · 48633 ms · 2026-05-10T07:55:06.972165+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

3 extracted references · 3 canonical work pages

[1]

Maurice Clerc and James Kennedy

doi: 10.1007/978-3-540-36668-3_94. Maurice Clerc and James Kennedy. The particle swarm—explosion, stability, and convergence in a multidimensional complex space.IEEE Transactions on Evolutionary Computation, 6(1):58–73, 2002. doi: 10.1109/4235.985692. Joaquín Derrac, Salvador García, Daniel Molina, and Francisco Herrera. A practical tutorial on the use of...

work page doi:10.1007/978-3-540-36668-3_94 2002
[2]

Karli Gillette, Matthias A

doi: 10.1007/s10472-010-9213-y. Karli Gillette, Matthias A. F. Gsell, Anton J. Prassl, Elias Karabelas, Ursula Reiter, Gert Reiter, Thomas Grandits, Christian Payer, Darko Štern, Martin Urschler, Jason D. Bayer, Christoph M. Augustin, Aurel Neic, Thomas Pock, Edward J. Vigmond, and Gernot Plank. A framework for the generation of digital twins of cardiac e...

work page doi:10.1007/s10472-010-9213-y
[3]

Thomas Grandits, Karli Gillette, Aurel Neic, Jason Bayer, Edward Vigmond, Thomas Pock, and Gernot Plank

doi: 10.1016/j.media.2021.102080. Thomas Grandits, Karli Gillette, Aurel Neic, Jason Bayer, Edward Vigmond, Thomas Pock, and Gernot Plank. An inverse eikonal method for identifying ventricular activation sequences from epicardial activationmaps.JournalofComputationalPhysics,419:109700,2020. doi: 10.1016/j.jcp.2020.109700. Thomas Grandits, Alexander Efflan...

work page doi:10.1016/j.media.2021.102080 2021