Causal fermion systems

A causal fermion system of spin dimension ${\displaystyle n\in \mathbb {N} }$ is a triple ${\displaystyle ({\mathcal {H}},{\mathcal {F}},\rho )}$ where

• ${\displaystyle ({\mathcal {H}},\langle .|.\rangle _{\mathcal {H}})}$ is a complex Hilbert space.
• ${\displaystyle {\mathcal {F}}}$ is the set of all self-adjoint linear operators of finite rank on ${\displaystyle {\mathcal {H}}}$ which (counting multiplicities) have at most ${\displaystyle n}$ positive and at most ${\displaystyle n}$ negative eigenvalues.
• ${\displaystyle \rho }$ is a measure on ${\displaystyle {\mathcal {F}}}$.

The measure ${\displaystyle \rho }$ is referred to as the universal measure.

As will be outlined below, this definition is rich enough to encode analogs of the mathematical structures needed to formulate physical theories. In particular, a causal fermion system gives rise to a spacetime together with additional structures that generalize objects like spinors, the metric and curvature. Moreover, it comprises quantum objects like wave functions and a fermionic Fock state.[7]

Inspired by the Langrangian formulation of classical field theory, the dynamics on a causal fermion system is described by a variational principle defined as follows.

Given a Hilbert space ${\displaystyle ({\mathcal {H}},\langle .|.\rangle _{\mathcal {H}})}$ and the spin dimension ${\displaystyle n}$, the set ${\displaystyle {\mathcal {F}}}$ is defined as above. Then for any ${\displaystyle x,y\in {\mathcal {F}}}$, the product ${\displaystyle xy}$ is an operator of rank at most ${\displaystyle 2n}$. It is not necessarily self-adjoint because in general ${\displaystyle (xy)^{*}=yx\neq xy}$. We denote the non-trivial eigenvalues of the operator ${\displaystyle xy}$ (counting algebraic multiplicities) by

${\displaystyle \lambda _{1}^{xy},\ldots ,\lambda _{2n}^{xy}\in {\mathbb {C} }.}$

Moreover, the spectral weight ${\displaystyle |.|}$ is defined by

${\displaystyle |xy|=\sum _{i=1}^{2n}|\lambda _{i}^{xy}|\quad {\text{and}}\quad {\big |}(xy)^{2}{\big |}=\sum _{i=1}^{2n}|\lambda _{i}^{xy}|^{2}{\,}.}$

The Lagrangian is introduced by

${\displaystyle {\mathcal {L}}(x,y)={\big |}(xy)^{2}{\big |}-{\frac {1}{2n}}{\,}|xy|^{2}={\frac {1}{4n}}\sum _{i,j=1}^{2n}{\big (}|\lambda _{i}^{xy}|-|\lambda _{j}^{xy}|{\big )}^{2}\geq 0{\,}.}$

The causal action is defined by

${\displaystyle {\mathcal {S}}=\iint _{{\mathcal {F}}\times {\mathcal {F}}}{\mathcal {L}}(x,y){\,}d\rho (x){\,}d\rho (y){\,}.}$

The causal action principle is to minimize ${\displaystyle {\mathcal {S}}}$ under variations of ${\displaystyle \rho }$ within the class of (positive) Borel measures under the following constraints:

• Boundedness constraint: ${\displaystyle \iint _{{\mathcal {F}}\times {\mathcal {F}}}|xy|^{2}{\,}d\rho (x){\,}d\rho (y)\leq C}$ for some positive constant ${\displaystyle C}$.
• Trace constraint: ${\displaystyle \;\;\;\int _{\mathcal {F}}{\text{tr}}(x){\,}d\rho (x)}$ is kept fixed.
• The total volume ${\displaystyle \rho ({\mathcal {F}})}$ is preserved.

Here on ${\displaystyle {\mathcal {F}}\subset {\mathrm {L} }({\mathcal {H}})}$ one considers the topology induced by the ${\displaystyle \sup }$-norm on the bounded linear operators on ${\displaystyle {\mathcal {H}}}$.

The constraints prevent trivial minimizers and ensure existence, provided that ${\displaystyle {\mathcal {H}}}$ is finite-dimensional.[8] This variational principle also makes sense in the case that the total volume ${\displaystyle \rho ({\mathcal {F}})}$ is infinite if one considers variations ${\displaystyle \delta \rho }$ of bounded variation with ${\displaystyle (\delta \rho )({\mathcal {F}})=0}$.

For ${\displaystyle x,y\in M}$, we denote the non-trivial eigenvalues of the operator ${\displaystyle xy}$ (counting algebraic multiplicities) by ${\displaystyle \lambda _{1}^{xy},\ldots ,\lambda _{2n}^{xy}\in {\mathbb {C} }}$. The points ${\displaystyle x}$ and ${\displaystyle y}$ are defined to be spacelike separated if all the ${\displaystyle \lambda _{j}^{xy}}$ have the same absolute value. They are timelike separated if the ${\displaystyle \lambda _{j}^{xy}}$ do not all have the same absolute value and are all real. In all other cases, the points ${\displaystyle x}$ and ${\displaystyle y}$ are lightlike separated.

This notion of causality fits together with the "causality" of the above causal action in the sense that if two spacetime points ${\displaystyle x,y\in M}$ are space-like separated, then the Lagrangian ${\displaystyle {\mathcal {L}}(x,y)}$ vanishes. This corresponds to the physical notion of causality that spatially separated spacetime points do not interact. This causal structure is the reason for the notion "causal" in causal fermion system and causal action.

Let ${\displaystyle \pi _{x}}$ denote the orthogonal projection on the subspace ${\displaystyle S_{x}:=x({\mathcal {H}})\subset {\mathcal {H}}}$. Then the sign of the functional

${\displaystyle i{\text{Tr}}{\big (}x\,y\,\pi _{x}\,\pi _{y}-y\,x\,\pi _{y}\,\pi _{x})}$

distinguishes the future from the past. In contrast to the structure of a partially ordered set, the relation "lies in the future of" is in general not transitive. But it is transitive on the macroscopic scale in typical examples.[5][6]

For every ${\displaystyle x\in M}$ the spin space is defined by ${\displaystyle S_{x}=x({\mathcal {H}})}$; it is a subspace of ${\displaystyle {\mathcal {H}}}$ of dimension at most ${\displaystyle 2n}$. The spin scalar product ${\displaystyle {\prec }\cdot |\cdot {\succ }_{x}}$ defined by

${\displaystyle {\prec }u|v{\succ }_{x}=-{\langle }u|xv{\rangle }_{\mathcal {H}}\qquad {\text{for all }}u,v\in S_{x}}$

is an indefinite inner product on ${\displaystyle S_{x}}$ of signature ${\displaystyle (p,q)}$ with ${\displaystyle p,q\leq n}$.

A wave function ${\displaystyle \psi }$ is a mapping

${\displaystyle \psi {\,}:{\,}M\rightarrow {\mathcal {H}}\qquad {\text{with}}\qquad \psi (x)\in S_{x}\quad {\text{for all }}x\in M{\,}.}$

On wave functions for which the norm ${\displaystyle {|\!|\!|}\cdot {|\!|\!|}}$ defined by

${\displaystyle {|\!|\!|}\psi {|\!|\!|}^{2}=\int _{M}\left\langle \psi (x){\bigg |}\,|x|\,\psi (x)\right\rangle _{\mathcal {H}}{\,}d\rho (x)}$

is finite (where ${\displaystyle |x|={\sqrt {x^{2}}}}$ is the absolute value of the symmetric operator ${\displaystyle x}$), one can define the inner product

${\displaystyle {\mathopen {<}}\psi |\phi {\mathclose {>}}=\int _{M}{\prec }\psi (x)|\phi (x){\succ }_{x}{\,}d\rho (x){\,}.}$

Together with the topology induced by the norm ${\displaystyle {|\!|\!|}\cdot {|\!|\!|}}$, one obtains a Krein space ${\displaystyle ({\mathcal {K}},{\mathopen {<}}\cdot |\cdot {\mathclose {>}})}$.

To any vector ${\displaystyle u\in {\mathcal {H}}}$ we can associate the wave function

${\displaystyle \psi ^{u}(x):=\pi _{x}u}$

(where ${\displaystyle \pi _{x}:{\mathcal {H}}\rightarrow S_{x}}$ is again the orthogonal projection to the spin space). This gives rise to a distinguished family of wave functions, referred to as the wave functions of the occupied states.

The kernel of the fermionic projector ${\displaystyle P(x,y)}$ is defined by

${\displaystyle P(x,y)=\pi _{x}\,y|_{S_{y}}{\,}:{\,}S_{y}\rightarrow S_{x}}$

(where ${\displaystyle \pi _{x}:{\mathcal {H}}\rightarrow S_{x}}$ is again the orthogonal projection on the spin space, and ${\displaystyle |_{S_{y}}}$ denotes the restriction to ${\displaystyle S_{y}}$). The fermionic projector ${\displaystyle P}$ is the operator

${\displaystyle P{\,}:{\,}{\mathcal {K}}\rightarrow {\mathcal {K}}{\,},\qquad (P\psi )(x)=\int _{M}P(x,y)\,\psi (y)\,d\rho (y){\,},}$

which has the dense domain of definition given by all vectors ${\displaystyle \psi \in {\mathcal {K}}}$ satisfying the conditions

${\displaystyle \phi$ :=\int _{M}x\,\psi (x)\,d\rho (x){\,}\in {\,}{\mathcal {H}}\quad {\text{and}}\quad {|\!|\!|}\phi {|\!|\!|}<\infty {\,}.}

As a consequence of the causal action principle, the kernel of the fermionic projector has additional normalization properties[9] which justify the name projector.

Being an operator from one spin space to another, the kernel of the fermionic projector gives relations between different spacetime points. This fact can be used to introduce a spin connection

${\displaystyle D_{x,y}\,:\,S_{y}\rightarrow S_{x}\quad {\text{unitary}}\,.}$

The basic idea is to take a polar decomposition of ${\displaystyle P(x,y)}$. The construction becomes more involved by the fact that the spin connection should induce a corresponding metric connection

${\displaystyle \nabla _{x,y}\,:\,T_{y}\rightarrow T_{x}\quad {\text{isometric}}\,,}$

where the tangent space ${\displaystyle T_{x}}$ is a specific subspace of the linear operators on ${\displaystyle S_{x}}$ endowed with a Lorentzian metric. The spin curvature is defined as the holonomy of the spin connection,

${\displaystyle {\mathfrak {R}}(x,y,z)=D_{x,y}\,D_{y,z}\,D_{z,x}\,:\,S_{x}\rightarrow S_{x}\,.}$

Similarly, the metric connection gives rise to metric curvature. These geometric structures give rise to a proposal for a quantum geometry.[5]

A minimizer ${\displaystyle \rho }$ of the causal action satisfies corresponding Euler–Lagrange equations.[10] They state that the function ${\displaystyle \ell _{\kappa }}$ defined by

${\displaystyle \ell _{\kappa }(x):=\int _{M}{\big (}{\mathcal {L}}_{\kappa }(x,y)+\kappa \,|xy|^{2}{\big )}\,d\rho (y)\,-\,{\mathfrak {s}}}$

(with two Lagrange parameters ${\displaystyle \kappa }$ and ${\displaystyle {\mathfrak {s}}}$) vanishes and is minimal on the support of ${\displaystyle \rho }$,

${\displaystyle \ell _{\kappa }|_{M}\equiv \inf _{x\in {\mathcal {F}}}\ell _{\kappa }(x)=0\,.}$

For the analysis, it is convenient to introduce jets ${\displaystyle {\mathfrak {u}}:=(a,u)}$ consisting of a real-valued function ${\displaystyle a}$ on ${\displaystyle M}$ and a vector field ${\displaystyle u}$ on ${\displaystyle T{\mathcal {F}}}$ along ${\displaystyle M}$, and to denote the combination of multiplication and directional derivative by ${\displaystyle \nabla _{\mathfrak {u}}g(x):=a(x)\,g(x)+{\big (}D_{u}g{\big )}(x)}$. Then the Euler–Lagrange equations imply that the weak Euler–Lagrange equations

${\displaystyle \nabla _{\mathfrak {u}}\ell |_{M}=0}$

hold for any test jet ${\displaystyle {\mathfrak {u}}}$.

Families of solutions of the Euler–Lagrange equations are generated infinitesimally by a jet ${\displaystyle {\mathfrak {v}}}$ which satisfies the linearized field equations

${\displaystyle \langle {\mathfrak {u}},\Delta {\mathfrak {v}}\rangle |_{M}=0\,,}$

to be satisfied for all test jets ${\displaystyle {\mathfrak {u}}}$, where the Laplacian ${\displaystyle \Delta }$ is defined by  

${\displaystyle \langle {\mathfrak {u}},\Delta {\mathfrak {v}}\rangle (x):=\nabla _{\mathfrak {u}}{\bigg (}\int _{M}{\big (}\nabla _{1,{\mathfrak {v}}}+\nabla _{2,{\mathfrak {v}}}{\big )}{\mathcal {L}}(x,y)\,d\rho (y)-\nabla _{\mathfrak {v}}{\mathfrak {s}}{\bigg )}\,.}$

The Euler–Lagrange equations describe the dynamics of the causal fermion system, whereas small perturbations of the system are described by the linearized field equations.

In the setting of causal fermion systems, spatial integrals are expressed by so-called surface layer integrals.[9][10][11] In general terms, a surface layer integral is a double integral of the form

${\displaystyle \int _{\Omega }{\bigg (}\int _{M\setminus \Omega }\cdots {\mathcal {L}}(x,y)\,d\rho (y){\bigg )}\,d\rho (x)\,,}$

where one variable is integrated over a subset ${\displaystyle \Omega \subset M}$, and the other variable is integrated over the complement of ${\displaystyle \Omega }$. It is possible to express the usual conservation laws for charge, energy, ... in terms of surface layer integrals. The corresponding conservation laws are a consequence of the Euler–Lagrange equations of the causal action principle and the linearized field equations. For the applications, the most important surface layer integrals are the current integral ${\displaystyle \gamma _{\rho }^{\Omega }({\mathfrak {v}})}$, the symplectic form ${\displaystyle \sigma _{\rho }^{\Omega }({\mathfrak {u}},{\mathfrak {v}})}$, the surface layer inner product ${\displaystyle \langle {\mathfrak {u}},{\mathfrak {v}}\rangle _{\rho }^{\Omega }}$ and the nonlinear surface layer integral ${\displaystyle \gamma ^{\Omega }({\tilde {\rho }},\rho )}$.

Based on the conservation laws for the above surface layer integrals, the dynamics of a causal fermion system as described by the Euler–Lagrange equations corresponding to the causal action principle can be rewritten as a linear, norm-preserving dynamics on the bosonic Fock space built up of solutions of the linearized field equations.[4] In the so-called holomorphic approximation, the time evolution respects the complex structure, giving rise to a unitary time evolution on the bosonic Fock space.

If ${\displaystyle {\mathcal {H}}}$ has finite dimension ${\displaystyle f}$, choosing an orthonormal basis ${\displaystyle u_{1},\ldots ,u_{f}}$ of ${\displaystyle {\mathcal {H}}}$ and taking the wedge product of the corresponding wave functions

${\displaystyle {\big (}\psi ^{u_{1}}\wedge \cdots \wedge \psi ^{u_{f}}{\big )}(x_{1},\ldots ,x_{f})}$

gives a state of an ${\displaystyle f}$-particle fermionic Fock space. Due to the total anti-symmetrization, this state depends on the choice of the basis of ${\displaystyle {\mathcal {H}}}$ only by a phase factor.[12] This correspondence explains why the vectors in the particle space are to be interpreted as fermions. It also motivates the name causal fermion system.

Starting on any globally hyperbolic Lorentzian spin manifold ${\displaystyle ({\hat {M}},g)}$ with spinor bundle ${\displaystyle S{\hat {M}}}$, one gets into the framework of causal fermion systems by choosing ${\displaystyle ({\mathcal {H}},{\langle }.|.{\rangle }_{\mathcal {H}})}$ as a subspace of the solution space of the Dirac equation. Defining the so-called local correlation operator ${\displaystyle F(p)}$ for ${\displaystyle p\in {\hat {M}}}$ by

${\displaystyle {\langle }\psi |F(p)\phi {\rangle }_{\mathcal {H}}=-{\prec }\psi |\phi {\succ }_{p}}$

(where ${\displaystyle {\prec }\psi |\phi {\succ }_{p}}$ is the inner product on the fibre ${\displaystyle S_{p}{\hat {M}}}$) and introducing the universal measure as the push-forward of the volume measure on ${\displaystyle {\hat {M}}}$,

${\displaystyle \rho =F_{*}d\mu {\,},}$

one obtains a causal fermion system. For the local correlation operators to be well-defined, ${\displaystyle {\mathcal {H}}}$ must consist of continuous sections, typically making it necessary to introduce a regularization on the microscopic scale ${\displaystyle \varepsilon }$. In the limit ${\displaystyle \varepsilon \searrow 0}$, all the intrinsic structures on the causal fermion system (like the causal structure, connection and curvature) go over to the corresponding structures on the Lorentzian spin manifold.[5] Thus the geometry of spacetime is encoded completely in the corresponding causal fermion systems.

The Euler–Lagrange equations corresponding to the causal action principle have a well-defined limit if the spacetimes ${\displaystyle M:={\text{supp}}\,\rho }$ of the causal fermion systems go over to Minkowski space. More specifically, one considers a sequence of causal fermion systems (for example with ${\displaystyle {\mathcal {H}}}$ finite-dimensional in order to ensure the existence of the fermionic Fock state as well as of minimizers of the causal action), such that the corresponding wave functions go over to a configuration of interacting Dirac seas involving additional particle states or "holes" in the seas. This procedure, referred to as the continuum limit, gives effective equations having the structure of the Dirac equation coupled to classical field equations. For example, for a simplified model involving three elementary fermionic particles in spin dimension two, one obtains an interaction via a classical axial gauge field ${\displaystyle A}$[2] described by the coupled Dirac– and Yang–Mills equations

${\displaystyle {\begin{aligned}(i\partial \!\!\!/\ +\gamma ^{5}A\!\!\!/\ -m)\psi &=0\\C_{0}(\partial _{j}^{k}A^{j}-\Box A^{k})-C_{2}A^{k}&=12\pi ^{2}{\bar {\psi }}\gamma ^{5}\gamma ^{k}\psi \,.\end{aligned}}}$

Taking the non-relativistic limit of the Dirac equation, one obtains the Pauli equation or the Schrödinger equation, giving the correspondence to quantum mechanics. Here ${\displaystyle C_{0}}$ and ${\displaystyle C_{2}}$ depend on the regularization and determine the coupling constant as well as the rest mass.

Likewise, for a system involving neutrinos in spin dimension 4, one gets effectively a massive ${\displaystyle SU(2)}$ gauge field coupled to the left-handed component of the Dirac spinors.[2] The fermion configuration of the standard model can be described in spin dimension 16.[1]

For the just-mentioned system involving neutrinos,[2] the continuum limit also yields the Einstein field equations coupled to the Dirac spinors,

${\displaystyle R_{jk}-{\frac {1}{2}}\,R\,g_{jk}+\Lambda \,g_{jk}=\kappa \,T_{jk}[\Psi ,A]\,,}$

up to corrections of higher order in the curvature tensor. Here the cosmological constant ${\displaystyle \Lambda }$ is undetermined, and ${\displaystyle T_{jk}}$ denotes the energy-momentum tensor of the spinors and the ${\displaystyle SU(2)}$ gauge field. The gravitation constant ${\displaystyle \kappa }$ depends on the regularization length.

Starting from the coupled system of equations obtained in the continuum limit and expanding in powers of the coupling constant, one obtains integrals which correspond to Feynman diagrams on the tree level. Fermionic loop diagrams arise due to the interaction with the sea states, whereas bosonic loop diagrams appear when taking averages over the microscopic (in generally non-smooth) spacetime structure of a causal fermion system (so-called microscopic mixing).[3] The detailed analysis and comparison with standard quantum field theory is work in progress.[4]