Domains, Signatures, Supports, and Regularity Ledgers

A quantum field is normally a distribution rather than an operator at a point, and a composite field is not obtained by multiplying those distributions formally. Meaning enters only after one specifies the spacetime and signature, test functions, operator domain, support restrictions, regularity or wavefront conditions, and any extension to coincident points. The free Klein–Gordon field and its stress tensor show how these requirements fit together.

Required background. Theorem-First Claim Records supplies the theorem grammar; Test Functions, Distributions, and Support supplies distributional smearing; and Unbounded Operators: Domains, Closure, and Adjoints supplies the operator-domain distinctions. Helpful background. Singular Support and Wavefront Sets gives the microlocal criterion used below; Contact Terms, Domains, and Error Budgets treats coincident contributions; Switching, Smearing, and Regularization explains operational smearing; and Large-N, Loop, and ℏ Hierarchies distinguishes asymptotic expansions from exact operator definitions.

A field expression has several simultaneous domains

For a globally hyperbolic Lorentzian spacetime $M$, an operator-valued distribution is a continuous linear map

$$\phi:C_0^\infty(M)\longrightarrow\operatorname{Op}(\mathcal D), \qquad f\longmapsto\phi(f),$$

where every $\phi(f)$ acts on a declared common dense subspace $\mathcal D$ of a Hilbert space $\mathcal H$. The continuity statement is usually weak: for all $\Psi,\Chi\in\mathcal D$, the scalar map $f\mapsto\langle\Psi,\phi(f)\Chi\rangle$ is a distribution. This is the operator-valued-distribution framework introduced for relativistic fields by Gårding and Wightman Gårding and Wightman 1964, pp. 129–142.

Five domains must not be collapsed into one.

Domain	Typical choice	Question it answers
Background	globally hyperbolic Lorentzian $M$, or a specified Euclidean manifold	Where are the equations and causal relations defined?
Test functions	$C_0^\infty(M)$, Schwartz space $\mathcal S(\mathbb R^d)$, or a section space of a bundle	On what inputs is the distribution continuous?
Operator vectors	a common invariant dense subspace $\mathcal D\subset\mathcal H$	On which vectors are products and commutators defined?
Distributional operation	pullback, product, restriction, or boundary value under a wavefront condition	Is the formal kernel operation defined?
Completion or extension	graph closure, Hilbert completion, distributional extension across a diagonal, or algebraic completion	Which enlarged object is obtained, and is it unique?

Changing one row can invalidate a formula while leaving the others untouched. For example, compact support makes causal propagation and local algebra assignment straightforward, but it does not make $\phi(f)$ bounded. Essential self-adjointness of one smeared field does not place a product $\phi(f)\phi(h)$ on all of $\mathcal H$. Smoothness away from the diagonal does not define a coincidence limit.

Signature is also mathematical data. Lorentzian two-point functions are boundary values with wavefront sets constrained by causal covectors; Euclidean Schwinger functions are distributions in positive metric. Analytic continuation between them is a theorem under spectral, growth, and positivity hypotheses, not the replacement $t=-i\tau$ inside an arbitrary singular expression.

The smeared Klein–Gordon field

Let

$$P=\Box_g+m^2+\xi R$$

be a normally hyperbolic Klein–Gordon operator on a globally hyperbolic spacetime, and let $E=E_{\mathrm{ret}}-E_{\mathrm{adv}}$ be its causal propagator. In a regular representation of the free field, choose the finite-particle domain $\mathcal D_{\mathrm{fin}}$ or another explicitly common invariant domain. For real $f,h\in C_0^\infty(M)$, the field satisfies on that domain

$$\phi(Pf)=0, \qquad \phi(f)^*\supset\phi(f), \qquad [\phi(f),\phi(h)]=iE(f,h)\mathbf 1.$$

The first identity says that the field descends to the quotient $C_0^\infty(M)/P C_0^\infty(M)$. The second is only symmetry on the declared domain; self-adjointness or essential self-adjointness is an additional result. The third is meaningful because both compositions act on the common invariant domain. Global hyperbolicity supplies the advanced and retarded propagators and the causal support property

$$\operatorname{supp}(Ef)\subset J(\operatorname{supp}f).$$

Consequently $E(f,h)=0$ for causally disjoint supports, giving the smeared commutator form of locality.

The distributional support and the operator domain perform different jobs. Shrinking $\operatorname{supp}f$ localizes the observable, but it does not enlarge $\mathcal D_{\mathrm{fin}}$. Conversely, closing the operator does not preserve compact support as an operator-theoretic notion; localization belongs to the net or smearing map.

Products require microlocal compatibility

For scalar distributions $u,v$ on $M$, Hörmander’s product criterion permits $uv$ when there is no point-covector pair for which their singular directions cancel:

$$(x,k)\in\operatorname{WF}(u) \quad\Longrightarrow\quad (x,-k)\notin\operatorname{WF}(v).$$

Under this condition the product is a well-defined distribution and its wavefront set is controlled Hörmander 1990, Theorem 8.2.10. A restriction to the diagonal $\Delta:M\to M\times M$, $x\mapsto(x,x)$, is a pullback. It is defined only if the kernel’s wavefront set avoids the conormal bundle

$$N^*\Delta = \{(x,x;k,-k):k\neq0\}.$$

Two-point functions of a quantum field are singular precisely near the diagonal. Thus the symbolic expression $W(x,x)$ is not licensed merely because $W(x,y)$ is a distribution for $x\neq y$. Nor does a formal product $W(x,y)^2$ exist automatically. The microlocal spectrum condition makes many products away from total diagonals possible, while extension across those diagonals is the renormalization problem developed in curved-spacetime perturbation theory Brunetti and Fredenhagen 2000, §§ 2–5, pp. 627–653.

An extension is not generally unique. If a distribution $t$ is defined on $\mathbb R^n\setminus\{0\}$ with finite scaling degree, it can be extended to the origin, but two extensions may differ by derivatives of the delta distribution up to the degree allowed by the scaling bound. In QFT these local terms become finite renormalization ambiguities. Local covariance, scaling, field equations, and conservation restrict them but do not turn the unextended product into a canonical pointwise value.

A stress-tensor quadratic form on a common domain

The classical scalar stress tensor is a quadratic differential expression in $\phi$. At the quantum level, choose a Hadamard parametrix $H(x,y)$ with the same universal short-distance singularity as a Hadamard two-point function. Let $D_{\mu\nu}(x,y)$ be the bidifferential operator obtained from the classical tensor, including the specified curvature coupling. For a compactly supported smooth test tensor $t^{\mu\nu}$, point splitting defines schematically

$$:T:(t) = \int_M t^{\mu\nu}(x) \left[ D_{\mu\nu}(x,y) \bigl(\phi(x)\phi(y)-H(x,y)\mathbf 1\bigr) \right]_{y=x} d\operatorname{vol}_g(x) +C(t)\mathbf 1.$$

This formula is a prescription, not an ordinary pointwise subtraction. Its meaning is obtained in three stages:

1. pair the bilocal field with a suitable test kernel away from the diagonal;
2. subtract the locally constructed Hadamard singularity so that the required diagonal pullback exists in matrix elements;
3. add only the local curvature terms $C(t)$ allowed by covariance, scaling, conservation, and the chosen renormalization conditions.

Local Wick polynomials satisfying locality, covariance, scaling, continuity, and the field commutation relation are unique only up to a finite family of curvature-dependent parameters Hollands and Wald 2001, Theorems 5.1 and 5.2, pp. 312–320. The result is naturally an operator-valued distribution or quadratic form on a common dense invariant domain, such as $\mathcal D_{\mathrm{fin}}$ in a quasifree Hadamard representation. Symmetry of the smeared form for real $t^{\mu\nu}$ should be checked; self-adjointness, lower bounds, and state-independent closure require separate theorems.

This construction supplies the precise input needed by Null-Smeared Stress Observables. Smearing along a null curve is itself a pullback and may fail without transverse smearing or a suitable wavefront condition. A spacetime-smeared stress tensor therefore does not automatically define every lower-dimensional restriction appearing in a formal energy inequality.

Failure test: plane waves and coincident singularities

The first adversarial replacement is $f(x)=e^{-ip\cdot x}$. A plane wave is neither compactly supported nor Schwartz, so it is outside the test-function domain $C_0^\infty(M)$ and outside $\mathcal S(\mathbb R^d)$. The notation $\phi(p)$ may still be introduced as a Fourier-transformed operator-valued tempered distribution, or as a limit of wave packets, but it is not the ordinary smeared operator $\phi(f)$ on the original domain. One must specify the topology of the limit and whether it yields an operator, a distributional kernel, or only matrix elements.

The second replacement is the raw coincidence product $W(x,y)^2\rvert_{y=x}$. For a Hadamard two-point distribution, singular covectors occur in paired opposite directions along the diagonal, so the product or pullback can meet the forbidden wavefront configuration. A well-defined expression then requires an extension prescription, a subtraction such as Hadamard normal ordering, and finite local counterterms. Euclidean regularity away from coincidence, Lorentzian covariance, or compact support of a later smearing function does not supply that missing extension.

The strongest conclusions that survive are therefore distributional momentum-space matrix elements in the plane-wave case and an off-diagonal bilocal distribution in the coincidence case. Neither adversarial expression is a canonically defined local operator.

Independent checks

Field-equation check. Replace $f$ by $f+Ph$. The smeared field and all observables defined on the quotient should be unchanged. A failure signals that the equation-of-motion ideal was not imposed consistently.

Adjoint check. For $f$ real, verify $\langle\Psi,\phi(f)\Chi\rangle=\langle\phi(f)\Psi,\Chi\rangle$ on the stated common domain. Do not infer equality of closed adjoints from this matrix-element identity.

Causal-support check. If $\operatorname{supp}f$ and $\operatorname{supp}h$ are causally disjoint, use the support of $E$ to obtain $[\phi(f),\phi(h)]=0$. This check is independent of any choice of Hadamard state.

Change-of-parametrix check. If $H'$ is another Hadamard parametrix, $H'-H$ is smooth locally. Substituting $H'$ changes the Wick polynomial by a local smooth curvature-dependent term, not by a new nonlocal singularity. The allowed change must match the finite ambiguity classified by the axioms.

Scaling check. Under a rescaling of lengths, compare the scaling degree of each unextended kernel with the codimension of the diagonal. This predicts whether the extension is unique and the highest derivative of a delta distribution that can occur.

Common pitfalls

Writing $\phi(x)$ as an operator. It is useful notation inside a distributional formula, but the basic object is $f\mapsto\phi(f)$. Point evaluation is generally undefined.

Using one domain for every composite. A common domain for the linear field need not be invariant under arbitrary infinite sums, limits, exponentials, or renormalized composites. Each construction must state and verify its domain.

Treating subtraction as division of infinities. Point splitting subtracts distributions with matched local singular structure before a pullback. It does not assign separate numerical values to $W(x,x)$ and $H(x,x)$.

Exercises

1. Why a plane wave is not a test function. Show that $e^{-ip\cdot x}$ belongs neither to $C_0^\infty(\mathbb R^d)$ nor to $\mathcal S(\mathbb R^d)$. Give a legitimate wave-packet replacement.

Solution

Its support is all of $\mathbb R^d$, so it is not compactly supported. Its absolute value is one, hence $x^\alpha e^{-ip\cdot x}$ is unbounded for every nonzero multi-index $\alpha$; it is not Schwartz. Choose $\chi\in C_0^\infty(\mathbb R^d)$ and set $f_L(x)=L^{-d/2}\chi(x/L)e^{-ip\cdot x}$, or choose a Schwartz envelope. Then $\phi(f_L)$ is defined for each finite $L$. Any $L\to\infty$ statement needs a declared weak, strong, graph, or distributional mode of convergence.

2. Test the square of a delta distribution. In one dimension, $\operatorname{WF}(\delta)=\{(0,k):k\neq0\}$. Apply Hörmander’s criterion to $\delta^2$.

Solution

For every $k\neq0$, both $(0,k)$ and $(0,-k)$ belong to $\operatorname{WF}(\delta)$. The forbidden cancelling pair is therefore present, so the canonical distributional product theorem does not define $\delta^2$. A regularization may produce an extension after subtractions, but its value depends on additional choices and is not the product in the original distribution space.

3. Derive spacelike commutativity. Let $f,h\in C_0^\infty(M)$ have causally disjoint supports. Prove that the free Klein–Gordon fields commute on their common invariant domain.

Solution

The causal propagator has support only on causally related pairs. Hence

$$E(f,h)=\int_{M\times M}f(x)E(x,y)h(y) \,d\operatorname{vol}_g(x)d\operatorname{vol}_g(y)=0.$$

The canonical commutation relation gives $[\phi(f),\phi(h)]=iE(f,h)\mathbf1=0$ on the common invariant domain. This proves the smeared locality statement; it does not assert that unsmeared operators $\phi(x)$ exist.

References

• Brunetti, Romeo, and Klaus Fredenhagen. “Microlocal Analysis and Interacting Quantum Field Theories: Renormalization on Physical Backgrounds.” Communications in Mathematical Physics 208 (2000): 623–661. DOI. Open PDF.
• Gårding, Lars, and Arthur S. Wightman. “Fields as Operator-Valued Distributions in Relativistic Quantum Theory.” Arkiv för Fysik 28 (1964): 129–189. Catalog record.
• Hollands, Stefan, and Robert M. Wald. “Local Wick Polynomials and Time Ordered Products of Quantum Fields in Curved Spacetime.” Communications in Mathematical Physics 223 (2001): 289–326. DOI. Open PDF.
• Hörmander, Lars. The Analysis of Linear Partial Differential Operators I: Distribution Theory and Fourier Analysis. 2nd ed. Berlin: Springer, 1990. DOI.