Locally Covariant Wick Polynomials

Higher powers of a quantum field cannot be defined by multiplying distributions at the same point. A useful renormalized Wick power must also mean the same thing on every admissible spacetime. The Hollands–Wald uniqueness theorem shows that locality, covariance, scaling, microlocal regularity, and smooth dependence on the background reduce the apparent freedom to a finite family of local curvature polynomials of the correct engineering dimension.

Required background. Stress–Energy and Conservation provides local tensor observables and metric variation. Curvature supplies the covariant curvature tensors that classify finite counterterms.

Helpful background. Flat-Spacetime Quantum Field Theory defines Fock normal ordering. Partial Differential Equations and Green Functions supplies the distributional field equation. Causality and Global Structure develops admissible local embeddings. Quantum Fields in Curved Spacetime provides the surrounding algebraic and Hadamard framework.

What counts as the same Wick power

Fix a real Klein–Gordon field in a fixed spacetime dimension. A family $\Phi^k_M(f)$ is locally covariant when every admissible embedding $\psi:M\to N$ obeys

$$\mathcal A(\psi)\Phi^k_M(f) =\Phi^k_N(\psi_*f).$$

This naturality condition excludes a prescription that normal-orders against a vacuum chosen independently on each spacetime. The vacuum choices need not be compatible under embeddings, even when each individual normal ordering is finite.

Covariance is only one hypothesis. The Wick powers are Hermitian for real test functions, have the expected commutator with the linear field, satisfy microlocal regularity, depend continuously on smooth metric families and analytically on analytic families in the original theorem, and scale almost homogeneously under $g\mapsto\lambda^{-2}g$ with $m^2\mapsto\lambda^2m^2$. “Almost” allows logarithms generated by renormalization. The analyticity requirement is a hypothesis of the 2001 proof cited here; smooth dependence cannot simply be substituted into that proof without an additional theorem.

The microlocal product criterion is what makes these composite distributions controllable without selecting a vacuum representation; Brunetti, Fredenhagen, and Köhler 1996, pp. 633–652 construct the curved-spacetime Wick algebra and establish the relevant wavefront-set framework.

The finite-renormalization theorem

Let $\Phi^k$ and $\widetilde\Phi^k$ be two Wick-power prescriptions satisfying the axioms for the same field. Then their difference has the triangular form

$$\widetilde\Phi^k =\Phi^k+ \sum_{j=0}^{k-2}\binom{k}{j}C_{k-j}\Phi^j,$$

where $\Phi^0=\mathbf 1$. Each $C_r$ is a locally and covariantly constructed scalar polynomial in the metric, curvature, mass, curvature coupling, and covariant derivatives, with the required scaling dimension $r$. The precise statement, including smoothness and scaling behavior, is Hollands and Wald 2001, §5.1, Theorem 5.1, and §5.2, PDF.

For $k=2$ in four dimensions, dimensional analysis gives

$$\widetilde\Phi^2-\Phi^2 =\alpha m^2+\beta R.$$

For $k=4$, the triangular relation reads $6C_2\Phi^2+4C_3\Phi+C_4$. In this four-dimensional scalar theory with constant parameters $m^2$ and $\xi$, no scalar curvature polynomial of mass dimension three can be constructed, so $C_3=0$. In a more general background theory any field-reflection symmetry must also be imposed consistently. The real coefficients may depend on the dimensionless coupling $\xi$ as allowed by the theorem, but are not functions chosen separately at each point or for each state.

The theorem is a classification, not a uniqueness claim with zero freedom. It says that once two prescriptions satisfy all hypotheses, every difference must have this local form. Numerical values of the finite coefficients require normalization conditions or measurement.

Why the hypotheses force locality

The proof proceeds by induction in $k$. The commutator relation shows that the difference between two $k$th powers has a lower polynomial degree in the basic field. The induction hypothesis then expresses it through lower Wick powers with c-number coefficient fields.

Local covariance then makes each coefficient at $x$ depend only on the background germ near $x$. In the original proof, first restrict to analytic metrics and compare the family $g_{\mu\nu}(sy)$ near $s=0$, together with $m^2\mapsto s^2m^2$. Analytic dependence permits a power series in $s$, while the scaling law selects only the terms of the required dimension. Only finitely many metric derivatives can then appear. The tensorial replacement theorem rewrites them as curvature tensors and covariant derivatives, and continuity extends the result to smooth metrics. This is the actual analytic-and-scaling argument of Hollands and Wald 2001, proof of Theorem 5.1, Eqs. (56)–(66), PDF; dimensional analysis alone would not establish finite local dependence.

Each assumption does visible work. Dropping covariance permits a coordinate-dependent subtraction. Dropping locality permits a coefficient involving an integral of $R$ over all of $M$. Dropping smooth dependence permits discontinuous choices keyed to a global property. Dropping scaling permits infinitely many curvature monomials divided by arbitrary powers of a new scale. These are counterexamples to the conclusion when a hypothesis is removed.

Checks, scope, and limitations

The $k=2$ result provides a quick dimension check: both $m^2$ and $R$ have mass dimension two, matching $\Phi^2$ in four dimensions. A term $R^2$ would have the wrong dimension unless accompanied by an inverse mass scale, which would violate the stated analytic/scaling assumptions at $m=0$. For $k=4$, $C_4$ may contain dimension-four combinations such as $m^4$, $m^2R$, curvature-squared terms, and total derivatives, subject to dimension-specific identities.

The theorem assumes the stated scalar field, local algebra, regularity class, and spacetime category. It does not automatically cover manifolds with boundary, gauge-invariant composite operators with nontrivial cohomology, or nonperturbative interacting fields. Nor does it say that two different finite choices predict identical absolute expectation values; it supplies the complete menu of allowed local redefinitions within its domain.

Exercise

In four dimensions, consider a proposed redefinition $\widetilde\Phi^2=\Phi^2+aR+b m^2+c\,\ell^2R^2$, where $\ell$ is a newly introduced fixed length. Which terms fit the theorem’s scaling classification, and why is the last term excluded under the stated prescription?

Solution

The terms $aR$ and $bm^2$ are local covariant scalars of mass dimension two, exactly matching the engineering dimension of $\Phi^2$. Although $\ell^2R^2$ also has net dimension two, it introduces an independent inverse-mass scale and does not scale almost homogeneously with the original field parameters in the required manner. Allowing it would open an unlimited tower $\ell^{2n-2}R^n$. It is therefore excluded by the scaling and analytic-dependence hypotheses, not by locality alone.

References

• Brunetti, Romeo, Klaus Fredenhagen, and Michael Köhler. “The Microlocal Spectrum Condition and Wick Polynomials of Free Fields on Curved Spacetimes.” Communications in Mathematical Physics 180 (1996): 633–652. DOI. Open PDF.
• 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.