---
title: 'Causal Fermion Systems: An Operator Framework'
url: https://www.emergentmind.com/topics/causal-fermion-systems
type: topic
---

# Causal Fermion Systems: An Operator Framework

Searching arXiv for recent and foundational papers on causal fermion systems to ground the article in the current literature.
{"query":"all:\"causal fermion systems\"","max_results":10}
{"query":"ti:\"causal fermion systems\" OR abs:\"causal fermion systems\"","max_results":10,"sort_by":"submittedDate","sort_order":"descending"}
Causal fermion systems are an operator-theoretic and measure-theoretic framework in which space-time and its geometric, causal, and spinorial structures are not assumed as primary data, but are encoded in a measure on finite-rank self-adjoint operators on a Hilbert space. In this setting, a physical system is specified by a triple \((\mathcal H,\mathcal F,\rho)\), where \(\mathcal H\) is a separable complex Hilbert space, \(\mathcal F\) is the set of self-adjoint finite-rank operators with bounded signature, and \(\rho\) is the universal measure; space-time is then defined as \(M:=\operatorname{supp}\rho\) [1709.04781]. The framework was developed by Finster and collaborators as a reformulation and extension of the fermionic projector program, and is presented as a setting in which Lorentzian geometry, Dirac theory, gauge interactions, and gravity arise as limiting or effective structures rather than fundamental inputs [1505.05075].

## 1. Foundational definition and conceptual program

A causal fermion system is defined by a separable complex Hilbert space \((\mathcal H,\langle . \mid . \rangle_{\mathcal H})\), a spin dimension \(n\in\mathbb N\), the set \(\mathcal F\subset \mathrm{L}(\mathcal H)\) of all self-adjoint finite-rank operators with at most \(n\) positive and at most \(n\) negative eigenvalues, and a positive measure \(\rho\) on \(\mathcal F\), the universal measure [2411.06450]. In the four-dimensional Dirac setting one has spin dimension \(n=2\), reflecting the signature \((2,2)\) of the pointwise spin scalar product [1709.04781].

The conceptual shift is that space-time is not a manifold on which fields are placed. Rather,
\[
M := \operatorname{supp}\rho \subset \mathcal F
\]
is the space-time of the system [1502.03587]. A space-time point is therefore an operator \(x\in\mathcal F\), more precisely a point in the support of the universal measure. The topology on \(M\) is induced from \(\mathcal F\), viewed as a subset of \(\mathrm L(\mathcal H)\) with the operator norm topology [1709.04781].

This construction unifies geometry and matter at the level of the measure \(\rho\). The framework is called “fermionic” because it starts from a Hilbert space of occupied or physically relevant fermionic states, and “causal” because causal relations are encoded spectrally in operator products and enter directly into the action principle [2411.06450]. Finster’s overviews and the later textbook repeatedly formulate the guiding idea as replacing the manifold-and-fields picture by a measure on operator space from which those structures emerge [1505.05075].

A recurrent misconception is that causal fermion systems simply quantize a Lorentzian manifold. The formalism instead replaces the manifold and metric by operator-theoretic data \((\mathcal H,\mathcal F,\rho)\), from which Lorentzian geometry is recovered only as a limiting case [1709.04781]. Another common misunderstanding is to treat the framework as merely a reformulation of ordinary Dirac theory. The literature instead presents it as a broader class of models intended to accommodate non-smooth, discrete, and “quantum” geometries for which no background manifold is available [2411.06450].

## 2. Construction from Lorentzian spin geometry and local correlation operators

The standard motivating construction begins with a smooth, globally hyperbolic, time-oriented Lorentzian spin manifold \((M,g)\) of dimension four with signature convention \((+,-,-,-)\), its spinor bundle \(SM\), Clifford multiplication
\[
\gamma : T_pM \to \mathrm{L}(S_pM),
\]
and the Dirac operator
\[
\mathcal D = i\gamma^j \nabla_j.
\]
For a mass parameter \(m\in\mathbb R\), the Dirac equation is
\[
(\mathcal D-m)\psi=0
\]
[1709.04781].

Completing smooth spatially compact solutions in the conserved scalar product gives the Hilbert space \((\mathcal H_m,(\cdot\mid\cdot)_m)\). One then chooses a closed subspace \(\mathcal H\subset \mathcal H_m\), interpreted physically as the space of physical wave functions realized in the system, typically including both occupied particle states and Dirac sea states [1709.04781]. Because these wave functions need not be pointwise evaluable, an ultraviolet regularization on a length scale \(\varepsilon\) is introduced,
\[
\mathfrak R_\varepsilon : \mathcal H \to C^0(M,SM).
\]

For each space-time point \(p\in M\), the regularized wave functions define a bounded sesquilinear form
\[
b_p(\psi,\phi) = - \prec (\mathfrak R_\varepsilon\psi)(p)\mid (\mathfrak R_\varepsilon\phi)(p)\succ_p .
\]
By the Fréchet–Riesz theorem this corresponds to a unique bounded linear operator \(F^\varepsilon(p)\) on \(\mathcal H\) such that
\[
(\psi \mid F^\varepsilon(p)\phi)_{\mathcal H}
=
-\prec (\mathfrak R_\varepsilon \psi)(p)\mid (\mathfrak R_\varepsilon \phi)(p)\succ_p .
\tag{1}
\]
Because the spin scalar product at \(p\) has signature \((2,2)\), each \(F^\varepsilon(p)\) is self-adjoint, has rank at most four, and has at most two positive and at most two negative eigenvalues [1709.04781].

This yields a local correlation map
\[
F^\varepsilon : M \to \mathcal F,
\]
and pushing forward the volume measure gives the universal measure
\[
d\rho := (F^\varepsilon)_*\, d\mu_M .
\]
The physical content of the original Lorentzian system is thereby transferred into the operator measure \((\mathcal H,\mathcal F,\rho)\) [1709.04781]. In the Minkowski example reviewed in several papers, the same construction starts from Dirac wave functions in Minkowski space, constructs local correlation operators \(F(x)\), and identifies Minkowski space with \(\operatorname{supp}\rho\) in physically relevant cases where \(F\) is injective with closed image [1502.03587].

The role of regularization is not treated as merely technical. The primer, the textbook, and Oppio’s Minkowski analysis all interpret ultraviolet regularization as the place where microscopic structure enters, potentially modeling Planck-scale deviations from smooth space-time [1709.04781]. Oppio shows in Minkowski space that with a smooth cutoff in momentum space, the local correlation map can be regular, injective, and closed, so that the resulting causal fermion system is regular and its support is a smooth four-dimensional manifold homeomorphic to Minkowski space in the vacuum and in systems with finitely many particles and suitably controlled antiparticles [1909.09229].

## 3. Intrinsic space-time, causality, spin spaces, and geometric structures

Once the triple \((\mathcal H,\mathcal F,\rho)\) is given, the central geometric structures are defined intrinsically. For \(x,y\in M\), the product \(xy\) has rank at most \(2n\). Denoting its non-trivial eigenvalues by
\[
\lambda^{xy}_1,\ldots,\lambda^{xy}_{2n},
\]
the causal relation is defined spectrally: \(x\) and \(y\) are spacelike separated if all \(\lambda^{xy}_j\) have the same absolute value, timelike separated if the \(\lambda^{xy}_j\) are all real and do not all have the same absolute value, and lightlike separated in all other cases [1709.04781]. In the Minkowski vacuum this reproduces the standard Lorentzian distinction between timelike and spacelike separation [1502.03587].

A time direction is encoded by the antisymmetric functional
\[
\mathcal C(x,y)
:=
i\,\operatorname{tr}\big(yx\,\pi_y\pi_x - xy\,\pi_x\pi_y\big),
\]
where \(\pi_x\) is the orthogonal projection onto \(x(\mathcal H)\). One says that \(y\) lies in the future of \(x\) if \(\mathcal C(x,y)>0\), and in the past if \(\mathcal C(x,y)<0\) [1709.04781]. The later textbook notes explicitly that this future relation need not be transitive in general [2411.06450].

For each \(x\in M\), the spin space is
\[
S_xM := x(\mathcal H),
\]
and it carries the spin scalar product
\[
\prec u \mid v \succ_x = - (u \mid x v)_{\mathcal H}.
\]
This is an indefinite inner product of signature \((p,q)\) with \(p,q\le n\) [1709.04781]. A wave function is a map \(\psi:M\to\mathcal H\) with \(\psi(x)\in S_xM\), and every vector \(u\in\mathcal H\) gives a distinguished physical wave function
\[
\psi^u(x)=\pi_x u.
\]
The kernel of the fermionic projector is
\[
P(x,y)=\pi_x\, y \big|_{S_yM}:S_yM\to S_xM,
\]
and satisfies \(P(x,y)^*=P(y,x)\) with respect to the spin scalar products [1709.04781]. The associated closed chain
\[
A_{xy}=P(x,y)P(y,x):S_xM\to S_xM
\]
has eigenvalues equal to the non-trivial eigenvalues of \(xy\), so the fermionic projector carries the full causal information [1502.03587].

The framework also develops analogues of tangent spaces, Clifford structures, spin connection, metric connection, and curvature. A regular point is one for which \(\dim x(\mathcal H)=2n\). In the regular case, the union
\[
SM := \bigcup_{x\in M} S_xM
\]
has the structure of a topological vector bundle over \(M\), with fiber metric of signature \((n,n)\) [1709.04781]. Clifford subspaces are defined as suitable subspaces of symmetric operators on \(S_xM\) satisfying Clifford-type anti-commutation relations, and for \(r=1\) the induced bilinear form gives a Lorentzian metric on the resulting tangent space [1709.04781].

For spin dimension \(n=2\), properly timelike separated points admit a more refined structure based on the closed chain \(A_{xy}\), the Euclidean sign operator \(s_x\), and the directional sign operator \(v_{xy}\). From these one constructs a spin connection
\[
D_{x,y}:S_yM\to S_xM,
\]
a corresponding metric connection \(\nabla_{x,y}:T_y\to T_x\), and curvature as holonomy of these discrete connections [1709.04781]. The 2011 overview already emphasized these constructions as a proposal for a Lorentzian “quantum geometry” [1102.2585].

A recent comparative paper with non-commutative geometry and generalized trace dynamics isolates the two-point character of these constructions as a distinctive feature: instead of encoding geometry primarily by pointwise tensor fields, causal fermion systems encode relations between points via the fermionic projector \(P(x,y)\), treated there as a generalized two-point correlator analogous in role, but not in content, to Synge’s world function [2603.05018]. This suggests that what emerges in the continuum limit is not a bare manifold but a fibered structure carrying spinorial and gauge information.

## 4. Causal action principle, Euler–Lagrange equations, and jet dynamics

The dynamics is governed by the causal action principle. For \(x,y\in\mathcal F\), the spectral weights are
\[
|xy|=\sum_{i=1}^{2n} |\lambda_i^{xy}|,\qquad |(xy)^2|=\sum_{i=1}^{2n} |\lambda_i^{xy}|^2,
\]
and the causal Lagrangian is
\[
\mathcal L(x,y)
=
|(xy)^2|-\frac{1}{2n}|xy|^2
=
\frac{1}{4n}\sum_{i,j=1}^{2n}\big(|\lambda_i^{xy}|-|\lambda_j^{xy}|\big)^2.
\tag{2}
\]
The causal action is
\[
\mathcal S(\rho)=\iint_{\mathcal F\times\mathcal F}\mathcal L(x,y)\,d\rho(x)\,d\rho(y),
\]
to be minimized under the volume, trace, and boundedness constraints
\[
\rho(\mathcal F)=\text{const},\qquad
\int_{\mathcal F}\operatorname{tr}(x)\,d\rho(x)=\text{const},\qquad
\mathcal T:=\iint_{\mathcal F\times\mathcal F}|xy|^2\,d\rho(x)\,d\rho(y)\le C
\]
[1502.03587]. Because \(\mathcal L(x,y)=0\) for spacelike separation, spacelike-separated pairs do not contribute to the action [2411.06450].

Kleiner’s thesis formulates the dynamics in terms of variations of the universal measure of the form
\[
\rho_\tau=(F_\tau)_*(f_\tau\,\rho),
\]
with infinitesimal data encoded by a jet \(\mathfrak v=(b,v)\), where \(b\) is a scalar function and \(v\) a vector field on \(\mathcal F\) [2006.14353]. In the general noncompact causal variational principle, one introduces
\[
\ell(x)=\int_{\mathcal F}\mathcal L(x,y)\,d\rho(y)-\frac{\nu}{2},
\]
and minimizers satisfy the Euler–Lagrange condition
\[
\ell|_M\equiv \inf_{\mathcal F}\ell = 0.
\]
The weak Euler–Lagrange equations are
\[
\nabla_{\mathfrak u}\ell|_M=0\qquad\text{for all }\mathfrak u\in\mathfrak J
\]
in the smooth setting, or corresponding one-sided conditions in the lower-semicontinuous/Lipschitz setting [2006.14353].

Differentiating these equations along families of minimizers yields linearized field equations for jets. A central structural result is that the linearized dynamics carries a conserved symplectic form expressed by surface layer integrals. In the smooth setting, for linearized solutions \(\mathfrak u,\mathfrak v\),
\[
\sigma_\Omega(\mathfrak u,\mathfrak v)
=
\int_\Omega d\rho(x)\int_{M\setminus\Omega}d\rho(y)\,
\Big(
\nabla_{1,\mathfrak u}\nabla_{2,\mathfrak v}\mathcal L(x,y)
-
\nabla_{1,\mathfrak v}\nabla_{2,\mathfrak u}\mathcal L(x,y)
\Big)
\]
vanishes for compact \(\Omega\), yielding a conserved presymplectic form and a Hamiltonian time evolution whenever \(M\) admits a notion of time [2006.14353].

The same thesis derives correction terms forced by the non-differentiable structure of the causal action principle in the lower-semicontinuous setting. Besides the standard linearized equations, there appear a stochastic correction term \(\chi_{w,v}\) and a nonlinear quadratic correction \(\chi^{(2)}_{w,v}\), both interpreted as microscopic effects that can average out macroscopically under the assumption that symmetric derivatives vanish macroscopically [2006.14353]. A plausible implication is that ordinary Hamiltonian dynamics emerges as an effective macroscopic regime rather than a fundamental exact structure.

Noether-type results also appear in this formalism. Symmetries of the Lagrangian or of the universal measure yield conserved surface layer integrals, and in the continuum limit these become the usual conserved Dirac probability integrals and energy-momentum conservation laws [2006.14353].

## 5. Continuum limit, Minkowski vacuum, and effective field equations

A central claim of the theory is that ordinary Lorentzian geometry and effective field equations are recovered in suitable limits. The elementary and overview papers show in the Minkowski vacuum that the fermionic projector kernel can be written as
\[
P(x,y)
=
\int \frac{d^4k}{(2\pi)^4}\,(k_j\gamma^j+m)\,\delta(k^2-m^2)\,\Theta(-k_0)\,e^{-ik(x-y)},
\]
and using Lorentz symmetry one obtains
\[
P(x,y)=\alpha\,\xi_j\gamma^j+\beta\,\mathbb 1,\qquad \xi:=y-x.
\]
The closed chain then has the form
\[
A_{xy}=a\,\xi_j\gamma^j+b\,\mathbb 1,
\]
with eigenvalues
\[
\lambda=b\pm \sqrt{a^2\xi^2},
\]
so the abstract spectral definition reproduces ordinary Minkowski causality [1502.03587].

The Lorentzian-geometric primer makes a stronger statement: if one starts from a globally hyperbolic Lorentzian spin manifold, constructs the corresponding causal fermion system from regularized Dirac solutions, and removes the regularization in a controlled way, then the causal-fermion-system connections converge to the standard metric and spin connections, up to higher-order curvature corrections [1709.04781]. For a future-directed timelike curve subdivided into points \(p_n\), compositions of the discrete spin and metric connections converge to the usual Lorentzian parallel transports as \(N\to\infty\) and \(\varepsilon\searrow 0\) [1709.04781].

The 2015 mini-review and the 2016 monograph present the continuum limit as yielding effective Dirac, Yang–Mills-type, and Einstein equations. In particular, the effective fermionic dynamics is described by
\[
(i\slashed{\partial}+\mathcal B-m)\,u_k(x)=0,
\]
with \(\mathcal B\) interpreted as an effective field describing the collective rearrangement of the fermionic system [1502.03587]. For a lepton model, the continuum limit yields an effective left-handed \(SU(2)\) gauge coupling together with Yang–Mills-type equations, while gravity is governed by
\[
R_{jk}-\frac12 R g_{jk}+\Lambda g_{jk}=\kappa T_{jk}
\]
[1502.03587]. The monograph presents these as the route by which the causal action principle gives rise to interactions of the standard model plus gravity on the level of second-quantized fermionic fields coupled to classical bosonic fields [1605.04742].

Recent work proposes a more systematic extraction of such effective equations from the linearized field equations. The 2025 paper “Construction of Currents in Causal Fermion Systems” defines a CFS current \(J_x(\mathbf v)\) so that the linearized field equations become \(J_x(\mathbf v)=0\), and shows that probing the current with wave functions and Taylor expanding the wave evaluation operator yields a hierarchy of tensorial equations
\[
J_x(\mathbf v)\,\partial_{\mu_1}\cdots\partial_{\mu_n}\Psi(x)^*=0
\]
of increasing rank [2507.09633]. In the \(i\varepsilon\)-regularized Minkowski vacuum, rank one recovers Maxwell’s equations explicitly, while rank two is proposed as the level at which Einstein equations should appear [2507.09633]. The same paper emphasizes that the gravitational interpretation of rank-two equations is not established there, but is a programmatic claim supported by earlier heuristics [2507.09633].

A reasonable caution follows from the literature itself: the broad emergence claims about gauge fields, gravity, and standard-model structures belong to the research program as a whole, whereas individual introductions often only sketch the derivations and refer to longer technical treatments [1502.03587].

## 6. Discrete space-times, relational interpretations, and current debates

The framework is explicitly intended to go beyond smooth Lorentzian manifolds. The 2018 survey on discrete space-times stresses that the causal action principle does not presuppose a background space-time structure, but is a variational principle for space-time itself as well as for all structures in space-time [1812.00238]. In the model case of spin dimension \(n=1\) and \(\mathcal H=\mathbb C^2\), the resulting causal variational principle on the sphere has rigorous results showing that for \(\tau>\sqrt{2}\), the support of any minimizing measure has empty interior, and for \(\tau>\sqrt{6}\), the support is totally disconnected and has Hausdorff dimension at most \(6/7\) [1812.00238]. Numerical results in the same program indicate that if \(\tau>\sqrt{2}\), every minimizing measure is a weighted counting measure [1812.00238]. This is widely interpreted within the literature as evidence for a discreteness or quantization effect at the microscopic level.

Non-smooth local geometry is also addressed by tangent cone constructions. The primer explains that even if no manifold structure is available, one may analyze local structure using tangent cone measures derived from rescaled push-forwards of maps such as
\[
\mathcal A(y)=\pi_x (y-x)x\big|_{S_x},
\]
thereby extracting local directional information without differentiability assumptions [1709.04781]. The same text also introduces a Lorentzian distance based on admissible scales \(\ell_{\min}\) and \(\ell_{\max}\), causal chains, and a supremal length functional, linking causal fermion systems to Lorentzian length spaces and causal-set-like ideas [1709.04781].

A more explicitly relational interpretation is developed in the 2025 paper “Causal Fermion Systems: Spacetime as the web of correlations of a many-body quantum system.” There, the total space-time is described as the support of the universal measure and also as a superposition of one-particle space-times associated with occupied fermionic states, with the fermionic projector written as
\[
P(x,y)=\sum_{i=1}^N |\psi^{u_i}(x)\rangle \langle \psi^{u_i}(y)|
\]
in that many-body interpretation [2504.19272]. For an important vector-scalar subclass that includes the continuum Minkowski vacuum, the paper proves
\[
\mathcal L(x,y)=4\,\mathrm{Var}_\Omega[\tilde A_{xy}],
\]
thereby interpreting minimization of the causal action as minimization of fluctuations in the causal correlation structure of space-time [2504.19272]. This suggests a stronger relational reading of the ontology than is explicit in the earlier overviews.

Several active debates concern interpretation and scope. First, the infinite-dimensional existence theory is not yet fully developed, whereas finite-dimensional existence results are known [2411.06450]. Second, some proposed structures remain conjectural or effective: rank-two tensorial equations are argued to encode Einstein equations but are not proved to do so in the current tensor-hierarchy formalism [2507.09633]. Third, the measurement-theoretic implications are still under development. The 2015 mini-review conjectured that the fundamentally nonlinear Euler–Lagrange equations may give rise to an effective dynamical collapse theory [1502.03587], and the 2024 paper “Causal Fermion Systems as an Effective Collapse Theory” derives in a non-relativistic limit an effective stochastic and nonlinear modification of Schrödinger dynamics together with a deterministic Kossakowski–Lindblad equation for the statistical operator [2405.19254]. This is presented there as an effective consequence of the causal action principle rather than a fundamental postulate [2405.19254].

Another line of development concerns event structures and causation. The 2020 comparison with the ETH approach introduces future algebras in causal fermion systems and reports Minkowski-space results showing that the operator representing a space-time point approximately commutes with the algebra in its causal future, with corrections controlled by the regularization length [2004.11785]. This suggests an analogy between space-time point operators and event-like structures, but the same paper notes that causal fermion system future relations need not be transitive, so the analogy with event algebras is structurally suggestive rather than exact [2004.11785].

A final point of significance is that some geometry–matter relations can be stated directly at the level of the causal action principle, without first deriving Einstein equations. The 2019 paper on two-dimensional area and matter flux introduces intrinsic notions of two-dimensional area, Killing fields, and matter flux in a causal fermion system, and proves that for critical points of the causal action, the area change of two-dimensional surfaces under a Killing flow in null directions satisfies
\[
\frac{d}{d\tau}A(S_\tau)=F(S_\tau),
\]
a Jacobson-type area–flux law within the CFS setting [1910.06161].

Taken together, these developments define causal fermion systems as a research program with a stable operator-measure core and an expanding set of geometric, dynamical, and interpretive layers. Its most persistent thesis is that space-time is not primary but emergent from a universal measure on finite-rank operators; its most persistent open question is how fully the effective structures already identified—Lorentzian geometry, gauge interactions, quantum field theoretic behavior, and possibly collapse dynamics—can be derived in generality from the causal action principle alone.

Source: https://www.emergentmind.com/topics/causal-fermion-systems