Numerical Analysis of the Causal Action Principle in Low Dimensions

Felix Finster, Robert H. Jonsson, Niki Kilbertus

Abstract

Causal Fermion Systems (CFS) provide a unified framework for fundamental physics, deriving spacetime and physical laws from the causal action principle. This paper utilizes modern differentiable programming with JAX to numerically solve this variational principle in low dimensions. The study reveals that spacetime points naturally organize into symmetric "discrete Dirac spheres" and introduces geometric visualization tools to bridge abstract operator theory with Lorentzian geometry.

Introduction

The causal action principle is the foundational variational principle of Causal Fermion Systems. It describes how spacetime points, represented as linear operators on a Hilbert space, should be distributed to minimize a specific action. Historically, finding exact minimizers has been analytically challenging, with numerical attempts limited to a small number of points. As the number of points ($m$) and Hilbert space dimension ($f$) increase, the optimization becomes a high-dimensional, non-convex problem. This research scales numerical optimization to over a thousand points, bridging the gap between abstract theory and concrete physical models.

Methodology

The researchers developed a numerical pipeline leveraging differentiable programming and the JAX library. This approach allows for exact gradient computation through complex eigenvalue-based Lagrangian calculations, avoiding the inaccuracies of finite-difference approximations.

To handle the constraints of the problem—where spacetime points must be Hermitian operators with unit trace—the authors implemented a custom parametrization. This maps unconstrained real numbers into the required operator space, ensuring physical constraints are satisfied by design. The optimization strategy employs a two-stage process: L-BFGS for initial descent and standard BFGS for high-precision convergence.

Geometric construction of spacetime points and their causal relations.

Furthermore, the authors introduced "projected spacetime plots." These tools map abstract operator relations into Minkowski-like diagrams, enabling the visualization of emergent causal structures, such as light cones, directly from the mathematical results.

Key Insights: Discrete Dirac Spheres and Causal Geometry

A central finding of the study is the role of discrete Dirac spheres as universal minimizers. When the Hilbert space dimension is minimal ($f=2n$), the optimal configuration consists of points equally distributed on a sphere. In these configurations, every pair of distinct points is "spacelike separated," meaning the interaction Lagrangian between them vanishes.

Visualization of timelike and spacelike separation in the causal fermion system framework.

The geometry of these systems reveals a direct link to Lorentzian physics. For spin dimension $n=1$, causal relations manifest as double cones, providing a bridge to the familiar light-cone structure. The Lagrangian effectively acts as a repulsive potential; to minimize the action, the system pushes points into spacelike regions, "inflating" the Dirac sphere as more points are added.

Results and Scaling

The numerical framework successfully optimized systems with up to $m=1024$ spacetime points, involving over 40,000 parameters. Key results include:

  • Asymptotic Accuracy: Numerical results aligned with analytical lower bounds in the large $f$ limit with high precision.
  • Action Minimization: For $n=1, f=2$, the numerical action approached the theoretical value of approximately 0.138, significantly lower than uniform distribution values.
  • Clustering: Points were observed to cluster on specific hyperbolas in parameter space, suggesting an emergent organization into symmetric configurations even before reaching a perfect Dirac sphere.

Conclusion and Future Directions

While the current approach demonstrates significant scaling, computational intensity remains a factor, with large runs taking several days. Future work will explore linear programming in convex domains to further approximate operator spaces and investigate the asymptotics of the causal action as $m$ grows toward the continuum limit. The study confirms that discrete Dirac spheres are likely the absolute minimizers for $f=2n$, providing a solid foundation for further exploration of Causal Fermion Systems.

Summary updated

BibTeX
@inproceedings{finster2025numerical,
  author = {Felix Finster and Robert H. Jonsson and Niki Kilbertus},
  title = {{Numerical Analysis of the Causal Action Principle in Low Dimensions}},
  booktitle = {Communications},
  year = {2025},
  doi = {10.4310/cms.250930022107},
}
Figure