Trace Anomalies

A classically conformal field can have a conserved stress tensor whose renormalized trace is nonzero on a curved background. The reason is not that the classical equation changed: defining the quantum determinant requires a short-distance prescription and a scale, and no prescription preserves every classical Weyl Ward identity while retaining the other required symmetries. In four dimensions the nontrivial trace separates into Euler and Weyl-squared terms, plus a total derivative whose coefficient can be shifted by a finite local counterterm.

Required background. Gravitational Actions and Boundary Terms provides metric variation and boundary completion; Curvature supplies Weyl, Euler, and total-derivative invariants. Helpful background. Flat-Spacetime Quantum Field Theory provides Ward identities and beta functions; Partial Differential Equations and Green Functions supplies regulated determinants; Causality and Global Structure establishes the Lorentzian setting; Quantum Fields in Curved Spacetime provides renormalized stress–energy and state-independent ultraviolet structure.

Weyl variation as the trace Ward identity

For an infinitesimal Weyl rescaling

$$\delta_\sigma g_{\mu\nu}=2\sigma g_{\mu\nu},$$

the site’s stress-tensor definition gives

$$\delta_\sigma\Gamma[g] =\int d^4x\sqrt{-g}\,\sigma(x) \langle T^\mu{}_{\mu}(x)\rangle.$$

Classically, a massless scalar with $\xi=1/6$, a massless Dirac field, and Maxwell theory in four dimensions have vanishing trace on shell after the appropriate improvement. Quantum mechanically, the regulated effective action depends on the renormalization prescription. The Weyl variation of its logarithmic ultraviolet term remains finite when the regulator is removed and supplies the anomaly.

This statement presumes that diffeomorphism invariance has been maintained, so $\nabla^\mu\langle T_{\mu\nu}\rangle=0$. It is a choice of which Ward identity is preserved, not a failure of general covariance. Duff traces the distinction between removable regularization artifacts and the invariant anomaly in Duff 1994, §§2–3.

Four-dimensional geometric classification

With the global curvature convention and on a manifold without boundary, define

$$C^2=C_{\mu\nu\rho\sigma}C^{\mu\nu\rho\sigma},$$ $$E_4=R_{\mu\nu\rho\sigma}R^{\mu\nu\rho\sigma} -4R_{\mu\nu}R^{\mu\nu}+R^2.$$

The parity-even anomaly of a conformal theory can be written

$$\langle T^\mu{}_{\mu}\rangle =\frac{1}{(4\pi)^2} \left(cC^2-aE_4+b\Box R\right).$$

Here $a$ is the type-A coefficient and $c$ the four-dimensional type-B coefficient. Type A multiplies the Euler density: its integral on a closed manifold is topological, although its local density need not vanish. Type B multiplies a local Weyl invariant and is tied to logarithmic scale dependence. Deser and Schwimmer derive this geometric distinction and identify locally removable total derivatives as the trivial class in Deser and Schwimmer 1993, §§2–3.

For free four-dimensional fields in this normalization,

Field	$a$	$c$
Real conformal scalar	$1/360$	$1/120$
Weyl fermion	$11/720$	$1/40$
Gauge vector, including ghosts	$31/180$	$1/10$

The fermion row gives the parity-even contribution of one two-component Weyl species; a Dirac field contributes twice this row. The vector row is one free Maxwell field. These coefficients follow by changing from the curvature basis in Vassilevich 2003, §4.2.2, Eq. (4.35) and Table 1, PDF, with the fermion determinant sign included. They are independent of the quantum state because they come from the universal ultraviolet singularity. A state can add a conserved state-dependent stress tensor, but it cannot change $a$ or $c$ for the fixed field content.

Derivation from the heat-kernel coefficient

Continue first to a positive-definite Euclidean metric and reduce the quadratic fluctuation operator, including ghosts, to Laplace type. In four dimensions the logarithmic divergence is proportional to the integrated $a_4$. For a classically Weyl-covariant operator, its Weyl variation leaves precisely the dimension-four curvature density. Reorganizing the Riemann-squared, Ricci-squared, and $R^2$ terms into $C^2$ and $E_4$ gives the nontrivial coefficients above, while the Laplacian term produces $\nabla_E^2R_E$.

Only after this Euclidean calculation do we continue to the site’s Lorentzian $(-,+,+,+)$ convention and define the Lorentzian anomaly by $\delta_\sigma\Gamma_L=\int\sqrt{-g}\,\sigma\langle T^\mu{}_{\mu}\rangle$. This avoids an uncontrolled sign transfer from a Euclidean coefficient table. The relation between the heat-kernel coefficient and conformal anomaly, including zero-mode qualifications, is given in Vassilevich 2003, §7.1, PDF.

For the real conformal scalar, the substitution $\mathcal E=-R_E/6$ into the local $b_4$ coefficient makes the calculation explicit:

$$b_4=\frac1{180} \left(R_{Eabcd}R_E^{abcd}-R_{Eab}R_E^{ab}+\nabla_E^2R_E\right).$$

In four dimensions,

$$R_{abcd}R^{abcd}-R_{ab}R^{ab} =\frac32C^2-\frac12E_4.$$

The coefficients are consequently $c=1/120$, $a=1/360$, and $b=1/180$ in this particular local subtraction scheme. This gives the scalar row by algebra, rather than only quoting a table. The $b$ value changes under the finite term below.

On de Sitter spacetime with curvature scale $H$, $C^2=0$, $E_4=24H^4$, and $\Box R=0$. Hence the scalar anomaly is $-H^4/(240\pi^2)$. If the state is de Sitter invariant, symmetry then gives $\langle T_{ab}\rangle=-H^4g_{ab}/(960\pi^2)$, whose orthonormal energy density is positive with the site’s signature. Without that state symmetry the trace alone would not fix the full tensor.

An independent two-dimensional check is especially clean. There is no nonzero local Weyl-squared invariant in two dimensions, so only type A remains and the trace is proportional to $R$. Its coefficient is the central charge, reproducing the fact that one free real scalar contributes one unit of central charge.

Why the total derivative is scheme-dependent

The coefficient $b$ is not on the same footing as $a$ and $c$. Add a finite local term

$$\Delta\Gamma=\alpha\int d^4x\sqrt{-g}\,R^2.$$

For $\delta_\sigma g_{\mu\nu}=2\sigma g_{\mu\nu}$ in four dimensions,

$$\delta_\sigma\sqrt{-g}=4\sigma\sqrt{-g}, \qquad \delta_\sigma R=-2\sigma R-6\Box\sigma.$$

Hence, on a closed spacetime or with boundary terms controlled,

$$\delta_\sigma\Delta\Gamma =-12\alpha\int d^4x\sqrt{-g}\, \sigma\Box R.$$

Therefore $b$ shifts by a freely chosen finite amount: the $\Box R$ contribution is scheme-dependent. No analogous local four-dimensional counterterm shifts $a$ or $c$ while preserving the stated diffeomorphism-invariant setup. The phrase “the trace anomaly coefficient” is thus incomplete unless it says which coefficient.

Local density, integrated anomaly, and boundaries

On a closed four-manifold, $\int\sqrt{g_E}\,E_4$ is proportional to the Euler characteristic and $\int\sqrt{g_E}\,\nabla_E^2R_E=0$. Neither fact allows one to delete the corresponding local density before varying the action. A local stress tensor probes pointwise curvature, and a manifold with boundary converts the total derivative into a genuine boundary integral.

Moreover, the Gauss–Bonnet invariant itself requires a boundary completion when a boundary is present. Heat-kernel coefficients then contain boundary anomaly terms involving intrinsic and extrinsic geometry and the field’s boundary conditions. Bulk values of $a$ and $c$ do not by themselves specify the full integrated anomaly. Duff’s curvature-basis discussion and its boundary qualification appear in Duff 1994, §3, especially Eqs. (12) and (21)–(23).

Common interpretation errors

Topological does not mean locally zero. The integral of $E_4$ is topological on a closed Euclidean four-manifold, but $E_4(x)$ can vary and contributes locally to the trace.

Anomaly does not mean arbitrary scheme dependence. The total-derivative coefficient can move, whereas the type-A and type-B cohomology classes cannot be removed by the same finite local redefinition.

Anomalous trace is not explicit mass breaking. A massive field has classical trace terms in addition to quantum contributions. The formula above is for a classically conformal theory; massive decoupling requires a separate scale-dependent analysis.

Diagnostic: shift the trivial anomaly

Starting from $\delta_\sigma R=-2\sigma R-6\Box\sigma$, compute the Weyl variation of $\alpha\int\sqrt{-g}R^2$ and explain which anomaly coefficient it can change. What extra caution is required with a boundary?

Solution

The measure variation contributes $4\sigma R^2$, while varying the two powers of $R$ contributes $2R(-2\sigma R-6\Box\sigma)$. The algebraic $R^2$ terms cancel, leaving $-12\alpha\int\sqrt{-g}R\Box\sigma$. After integration by parts on a closed manifold this is $-12\alpha\int\sqrt{-g}\sigma\Box R$. Therefore the finite counterterm shifts only the coefficient $b$ of the scheme-dependent total derivative; it does not alter $a$ or $c$. With a boundary, the integration by parts produces a boundary term and the $R^2$ action needs boundary data or a boundary completion, so the local bulk shift cannot be discussed independently of the boundary anomaly.

References

• Deser, Stanley, and Adam Schwimmer. “Geometric Classification of Conformal Anomalies in Arbitrary Dimensions.” Physics Letters B 309 (1993): 279–284. DOI. Open PDF, arXiv:hep-th/9302047.
• Duff, Michael J. “Twenty Years of the Weyl Anomaly.” Classical and Quantum Gravity 11 (1994): 1387–1404. DOI. Open PDF, arXiv:hep-th/9308075.
• Vassilevich, Dmitri V. “Heat Kernel Expansion: User’s Manual.” Physics Reports 388 (2003): 279–360. DOI. Open PDF, arXiv:hep-th/0306138.