The Time-Slice Axiom

The time-slice axiom says that a region containing a Cauchy surface contains enough observables to generate the algebra of the entire globally hyperbolic spacetime. For a free hyperbolic field this is a theorem: any smearing function can be moved, modulo the field equation, into an arbitrarily thin Cauchy neighborhood. In a general locally covariant theory it is an axiom encoding deterministic dynamics. It is stronger than isotony and different from the claim that a small laboratory can perform every global measurement.

Required background. Flat-Spacetime Quantum Field Theory supplies smeared fields, equations of motion, and the idea that complete Cauchy data determine hyperbolic evolution.

Helpful background. Partial Differential Equations and Green Functions provides the advanced/retarded inverse and exact-sequence facts used in the proof; the needed support step is reconstructed below. Causality and Global Structure defines Cauchy surfaces and domains of dependence. Quantum Fields in Curved Spacetime provides the route from this dynamical axiom to local covariance and relative Cauchy evolution.

The Cauchy-neighborhood statement

Let $M$ be globally hyperbolic and let $N\subset M$ be an admissible open, causally convex subspacetime whose image contains a Cauchy surface of $M$. Write $\alpha_{MN}:\mathcal A(N)\to\mathcal A(M)$ for the algebra morphism induced by inclusion. The time-slice property is

$$\alpha_{MN}\bigl(\mathcal A(N)\bigr)=\mathcal A(M).$$

Because the morphism is already injective, it is an isomorphism. In the functorial language used on the next page, every morphism whose image contains a Cauchy surface is sent to an algebra isomorphism. This is the precise formulation in Brunetti, Fredenhagen, and Verch 2003, §2.2, Definition 2.1(iii), p. 7, PDF.

Three hypotheses do real work. First, the surface is Cauchy for all of $M$, not merely for $N$. Second, the region contains an open neighborhood of that surface, allowing smooth cutoff functions with transition support inside it. Third, global hyperbolicity supplies Green operators and excludes information entering from an uncontrolled timelike boundary. If $N$ is just a diamond that misses part of a Cauchy surface, its algebra is ordinarily a proper subalgebra.

The axiom is sometimes summarized as “all observables live on one time slice.” That wording is misleading. Quantum fields are distributions and are smeared over open spacetime regions; a codimension-one surface alone is not a test-function domain. The assertion concerns any sufficiently thin open Cauchy neighborhood.

Moving a free-field observable into the slice

Consider the real Klein–Gordon operator $P=\Box-m^2-\xi R$ and use

$$E=G_{\mathrm{ret}}-G_{\mathrm{adv}}.$$

The field algebra depends on a test function only through its class in $C_0^\infty(M)/P C_0^\infty(M)$. We will show that every class has a representative supported in $N$.

Choose two smooth spacelike Cauchy surfaces $\Sigma_-$ and $\Sigma_+$ inside $N$, with $\Sigma_+$ to the future of $\Sigma_-$. Let $\chi$ be smooth, $\chi=0$ to the past of $\Sigma_-$ and $\chi=1$ to the future of $\Sigma_+$, with $d\chi$ supported in the slab between them. For $f\in C_0^\infty(M)$ set

$$\phi=Ef, \qquad f_N=P(\chi\phi)=[P,\chi]\phi.$$

The second equality uses $P\phi=0$. Derivatives of $\chi$ vanish outside the chosen slab, while causal propagation keeps the intersection with the causal hull of $\operatorname{supp}f$ spatially compact. Hence $f_N$ is compactly supported inside $N$.

The sign check is important. The section $\chi\phi$ is past-compact, so $G_{\mathrm{ret}}P(\chi\phi)=\chi\phi$. The section $\chi\phi-\phi$ is future-compact, so $G_{\mathrm{adv}}P(\chi\phi)=\chi\phi-\phi$. Therefore

$$Ef_N =G_{\mathrm{ret}}f_N-G_{\mathrm{adv}}f_N =\phi =Ef.$$

Exactness of the Green-hyperbolic sequence then gives $f-f_N=Ph$ for some compactly supported $h$. The field equation implies $\Phi(f)=\Phi(f_N)$, and the latter belongs to the image of $\mathcal A(N)$. The support and exact-sequence ingredients are proved for normally hyperbolic operators in Bär, Ginoux, and Pfäffle 2007, Theorem 3.4.7, pp. 90–91.

From generators to an algebra isomorphism

For the polynomial field algebra, every generator $\Phi(f)$ has a Cauchy-neighborhood representative $\Phi(f_N)$. Products, sums, and star operations are therefore in the image as well, proving surjectivity. In the Weyl algebra the same argument sends $W([f])$ to $W([f_N])$; because $[f]=[f_N]$, their Weyl relations and symplectic phases agree.

Injectivity is not obtained by the cutoff alone. It is part of the accepted algebra morphism for an admissible embedding, or it follows from the faithful construction of the free-field functor. Surjectivity plus injectivity yields the isomorphism. This separation matters in gauge theories, where quotienting constraints or topological sectors can make apparently obvious inclusion maps fail to be injective.

The proof also shows independence of the chosen cutoff. If $\chi_1$ and $\chi_2$ give representatives $f_{N,1}$ and $f_{N,2}$, both have causal propagator $Ef$. Their difference lies in $P C_0^\infty(M)$, so they define the same observable. The construction is dynamical rather than a preferred choice of time coordinate.

What the axiom does and does not determine

An isomorphism of algebras transports states contravariantly. A state on the full algebra restricts to the slice algebra; conversely, composing a slice state with the inverse time-slice map extends it to the full spacetime. This is algebraic evolution without choosing a Schrödinger-picture Hilbert space. It does not select a vacuum or guarantee that two such states are unitarily equivalent.

The time-slice property also enables comparison between two backgrounds that agree outside a compact metric perturbation. Isomorphisms through past and future Cauchy neighborhoods can be composed into the relative Cauchy evolution. Its derivative will measure stress–energy response.

Several stronger claims do not follow:

• The axiom does not say that observables in a bounded diamond generate the whole spacetime algebra; the diamond must contain a complete Cauchy surface.
• It does not imply Einstein causality. Hyperbolic free fields satisfy both, but locality and deterministic propagation are distinct conditions.
• It does not prove interacting existence. In an abstract theory it is an axiom; in perturbative constructions it must be verified order by order.
• It does not survive arbitrary timelike boundaries without boundary data. Incoming information can make the interior slice incomplete.

A dropped-hypothesis test

Take $1+1$ Minkowski spacetime and a bounded double cone $N$ centered at $t=0$. On the complete surface $\Sigma=\{t=0\}$, choose nonzero smooth compact Cauchy data outside $J(N)\cap\Sigma$, and let $u$ be their solution. The causal-propagator theorem supplies a compact $f$ with $Ef=u$. For every $f_N$ supported in $N$, finite propagation gives $\operatorname{supp}Ef_N\subset J(N)$, so $Ef_N$ has zero Cauchy data where we chose the nonzero data of $u$. Therefore $Ef_N\ne Ef$. The source class $[f]$ cannot be represented in $N$, and the free-field inclusion is not surjective. What failed was precisely the complete-Cauchy-surface hypothesis.

This example prevents a common overstatement: “local” in local QFT does not mean that every arbitrarily small region contains all observables. The time-slice property is global in its use of complete Cauchy data, even though the neighborhood can be thin in time.

Exercises

With the retarded-minus-advanced convention, verify the sign in $E P(\chi\phi)=\phi$ for a homogeneous spacelike-compact solution $\phi$. What changes if the causal propagator is defined with the opposite sign?

Solution

Let $h=\chi\phi$. Because $h$ vanishes sufficiently far to the past, the retarded inverse gives $G_{\mathrm{ret}}Ph=h$. Because $h-\phi$ vanishes sufficiently far to the future and satisfies $P(h-\phi)=Ph$, the advanced inverse gives $G_{\mathrm{adv}}Ph=h-\phi$. Hence

$$(G_{\mathrm{ret}}-G_{\mathrm{adv}})Ph =h-(h-\phi)=\phi.$$

If instead $E'=G_{\mathrm{adv}}-G_{\mathrm{ret}}$, the same construction gives $E'Ph=-\phi$. One may replace $f_N$ by $-P(\chi\phi)$, or consistently reverse the quotient-to-solution identification. Reversing only $E$ while leaving the CCR sign unchanged would be inconsistent.

References

• Bär, Christian, Nicolas Ginoux, and Frank Pfäffle. Wave Equations on Lorentzian Manifolds and Quantization. ESI Lectures in Mathematics and Physics. Zürich: European Mathematical Society, 2007. DOI.
• Brunetti, Romeo, Klaus Fredenhagen, and Rainer Verch. “The Generally Covariant Locality Principle—A New Paradigm for Local Quantum Field Theory.” Communications in Mathematical Physics 237 (2003): 31–68. DOI. Open PDF.