A Decomposition Approach for the Gain Function in the Feedback Particle Filter
Pith reviewed 2026-05-22 22:48 UTC · model grok-4.3
The pith
Decomposing the Poisson equation into two solvable parts yields exact gain functions for the feedback particle filter when the observation is polynomial.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
The central claim is that the Poisson equation with probability-weighted Laplacian can be decomposed, via a tunable scalar parameter, into a pair of exactly solvable linear equations whenever the observation function is a polynomial; the solutions are then combined to recover the gain function used to steer particles in the feedback particle filter.
What carries the argument
Decomposition of the Poisson equation with a free parameter into two exactly solvable equations.
If this is right
- Computational cost grows linearly in the number of particles and in the degree of the observation polynomial.
- Estimation accuracy exceeds that of the bootstrap particle filter and the constant-gain FPF on the tested examples.
- CPU time is the lowest among all compared methods that achieve comparable accuracy.
- The method remains resampling-free and control-oriented like the original FPF.
Where Pith is reading between the lines
- The same decomposition idea might extend to other observation classes if a suitable free parameter can still be chosen to cancel higher-order terms.
- Because the two sub-equations are solved independently, the approach could be parallelized across particles or polynomial terms.
- If the observation is only approximately polynomial, the method supplies a natural truncation error bound linear in the neglected coefficients.
Load-bearing premise
The observation function must be a polynomial.
What would settle it
Run the decomposition on a quadratic observation function and check whether the recovered gain exactly satisfies the original Poisson equation for a known stationary density.
read the original abstract
The feedback particle filter (FPF) is an innovative, control-oriented and resampling-free adaptation of the traditional particle filter (PF). In the FPF, individual particles are regulated via a feedback gain, and the corresponding gain function serves as the solution to the Poisson's equation equipped with a probability-weighted Laplacian. Owing to the fact that closed-form expressions can only be computed under specific circumstances, approximate solutions are typically indispensable. This paper is centered around the development of a novel algorithm for approximating the gain function in the FPF. The fundamental concept lies in decomposing the Poisson's equation into two equations that can be precisely solved, provided that the observation function is a polynomial. A free parameter is astutely incorporated to guarantee exact solvability. The computational complexity of the proposed decomposition method shows a linear correlation with the number of particles and the polynomial degree of the observation function. We perform comprehensive numerical comparisons between our method, the PF, and the FPF using the constant-gain approximation and the kernel-based approach. Our decomposition method outperforms the PF and the FPF with constant-gain approximation in terms of accuracy. Additionally, it has the shortest CPU time among all the compared methods with comparable performance.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The paper proposes a decomposition approach for approximating the gain function in the feedback particle filter (FPF) by splitting the weighted Poisson equation into two exactly solvable sub-equations when the observation function is a polynomial. A free parameter is introduced to ensure solvability of the decomposed system. The resulting algorithm has computational complexity linear in the number of particles and the polynomial degree. Numerical comparisons are presented against the standard particle filter (PF) and the constant-gain FPF (and a kernel-based variant), claiming superior accuracy and the shortest CPU time among methods with comparable performance.
Significance. If the decomposition yields an exact solution to the original Poisson equation for polynomial observations and the reported numerical advantages hold under reproducible conditions, the method provides a computationally attractive exact-construction alternative within this restricted class. The linear scaling and absence of resampling are strengths relative to standard PF. However, the polynomial restriction on the observation function is a fundamental scope limitation that prevents direct transfer to general nonlinear filtering problems where closed-form gains are unavailable.
major comments (2)
- [decomposition section / Poisson equation formulation] The decomposition (described in the main algorithmic section) introduces a free parameter to guarantee exact solvability of the two sub-equations, but it is not shown that the resulting gain satisfies the original weighted Poisson equation (the one equipped with the probability-weighted Laplacian) for arbitrary values of this parameter; the paper must clarify whether the solution is independent of the free parameter or requires a specific choice that recovers the true gain.
- [numerical experiments section] The outperformance claim in accuracy and CPU time is demonstrated only on polynomial observation examples; because the exact decomposition construction does not extend outside this class, the numerical superiority does not support broader claims about FPF gain approximation in general settings.
minor comments (3)
- [abstract] The abstract and introduction should explicitly state the polynomial restriction on the observation function at the outset rather than only in the method description.
- [numerical experiments section] Numerical results would benefit from reporting error bars, number of Monte Carlo runs, and exact experimental conditions (state dimension, noise levels, polynomial degrees tested) to allow assessment of statistical significance.
- [method section] Notation for the free parameter and the decomposed operators should be introduced with a clear equation reference to avoid ambiguity when comparing to the original Poisson equation.
Simulated Author's Rebuttal
We thank the referee for the careful reading and constructive feedback. We address the two major comments point-by-point below. Minor revisions have been prepared to improve clarity on the scope and the role of the free parameter.
read point-by-point responses
-
Referee: [decomposition section / Poisson equation formulation] The decomposition (described in the main algorithmic section) introduces a free parameter to guarantee exact solvability of the two sub-equations, but it is not shown that the resulting gain satisfies the original weighted Poisson equation (the one equipped with the probability-weighted Laplacian) for arbitrary values of this parameter; the paper must clarify whether the solution is independent of the free parameter or requires a specific choice that recovers the true gain.
Authors: We agree that the manuscript should have made this explicit. The free parameter is not arbitrary: it is chosen so that the sum of the two sub-solutions exactly recovers the solution of the original weighted Poisson equation for polynomial observations. In the revised manuscript we add a short lemma (new Appendix A) proving that, for this specific choice, the decomposed gain satisfies the original equation identically. For other values of the parameter the sum generally does not solve the original equation, so the construction is not parameter-independent. revision: yes
-
Referee: [numerical experiments section] The outperformance claim in accuracy and CPU time is demonstrated only on polynomial observation examples; because the exact decomposition construction does not extend outside this class, the numerical superiority does not support broader claims about FPF gain approximation in general settings.
Authors: The referee is correct that the exact decomposition applies only to polynomial observations. All numerical examples in the paper use polynomial observation functions precisely because that is the setting in which the method is exact. We do not claim, and have not claimed, superiority for arbitrary nonlinear observations. In the revision we have added an explicit sentence in the abstract and in the concluding paragraph stating that the reported accuracy and timing advantages are demonstrated for the polynomial-observation class. revision: yes
Circularity Check
No circularity: derivation is a direct algorithmic construction
full rationale
The paper presents a decomposition method that splits the weighted Poisson equation into two exactly solvable sub-equations when the observation function is polynomial, using an introduced free parameter to enforce solvability. This is an explicit algorithmic construction whose output (the gain function) is obtained by solving the sub-equations, not by fitting parameters to data or redefining the target quantity. Numerical comparisons to PF and constant-gain FPF are external benchmarks; no self-citation chains, ansatz smuggling, or renaming of known results appear as load-bearing steps. The central claim therefore remains independent of its own inputs.
Axiom & Free-Parameter Ledger
free parameters (1)
- free parameter
axioms (1)
- domain assumption The gain function is the solution to Poisson's equation with probability-weighted Laplacian.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.