Heat Kernels and Seeley–DeWitt Coefficients

The heat kernel converts the spectrum of a quadratic field operator into a local short-distance expansion. In four dimensions, the first three nonzero integrated coefficients organize the vacuum-energy, Einstein–Hilbert, and curvature-squared counterterms. The method is local in the ultraviolet, but its input includes the bundle connection, endomorphism, boundary condition, and a carefully declared Euclidean continuation.

Required background. Gravitational Actions and Boundary Terms provides the variational role of bulk and boundary terms; Curvature supplies the Riemann, Ricci, scalar, and Weyl invariants. Helpful background. Flat-Spacetime Quantum Field Theory provides Gaussian functional integrals and one-loop power counting; Partial Differential Equations and Green Functions supplies elliptic kernels and boundary-value problems; Causality and Global Structure establishes the Lorentzian causal setting recovered after continuation; Quantum Fields in Curved Spacetime provides the state and renormalization context.

Laplace-type operators and the Euclidean convention

Work first on a smooth $d$-dimensional Riemannian manifold with positive-definite metric $g_E$. A second-order operator on a vector bundle is of Laplace type when it can be written uniquely as

$$D=-\left(g_E^{\mu\nu}\nabla_\mu\nabla_\nu+\mathcal E\right),$$

where $\nabla$ contains the Levi–Civita and bundle connections and $\mathcal E$ is a bundle endomorphism. Its connection curvature is

$$\Omega_{\mu\nu}=[\nabla_\mu,\nabla_\nu].$$

This sign convention is the one used in the formulas below. For a Euclidean scalar fluctuation operator

$$D_m=-\nabla_E^2+m^2+\xi R_E=D_0+m^2,$$

we therefore identify $\mathcal E=-\xi R_E$ in $D_0$. Changing the definition of $D$ changes the displayed signs of $\mathcal E$ terms, so a coefficient table cannot be imported without its operator convention. The geometric setup and uniqueness of $\nabla$ and $\mathcal E$ are stated in Vassilevich 2003, §2.1, Eqs. (2.1)–(2.10), PDF.

The Euclidean expansion determines local Lorentzian counterterms by analytic continuation. We return to the site’s $(-,+,+,+)$ metric only after forming covariant invariants and continuing the action. The Lorentzian kernel itself is oscillatory rather than exponentially damped, so the Euclidean small-$t$ localization argument is not a pointwise proof about real-time propagation; this limitation is emphasized in Vassilevich 2003, §2.3, PDF.

Heat kernel and asymptotic coefficients

For a positive elliptic realization of $D$, including its boundary condition when needed, define

$$K(t;x,y;D)=\langle x|e^{-tD}|y\rangle, \qquad K(t;D)=\operatorname{Tr}e^{-tD}.$$

On a compact manifold without boundary,

$$K(t;D)\sim\sum_{k\ge0}t^{(k-d)/2}a_k(D), \qquad t\downarrow0,$$

and the odd coefficients vanish. Our $a_k(D)$ are integrated coefficients: they already include the volume integral and the factor $(4\pi)^{-d/2}$. Locally one may instead write

$$K(t;x,x;D) \sim \frac{1}{(4\pi t)^{d/2}} \sum_{r\ge0}t^r b_{2r}(x;D).$$

Thus $a_{2r}=(4\pi)^{-d/2}\int d^dx\sqrt{g_E}\,\operatorname{tr}b_{2r}$ in this notation. Confusing $a_{2r}$ with $b_{2r}(x)$ loses both an integral and a normalization. Vassilevich gives the spectral definition and this local-versus-integrated relation in Vassilevich 2003, §2.2, Eqs. (2.18)–(2.27), PDF.

The first bulk coefficients

For a closed manifold and no smearing function,

$$a_0(D)=\frac{1}{(4\pi)^{d/2}} \int_M d^dx\sqrt{g_E}\,\operatorname{tr}\mathbf1,$$ $$a_2(D)=\frac{1}{(4\pi)^{d/2}} \int_M d^dx\sqrt{g_E}\, \operatorname{tr}\left(\mathcal E+\frac16R_E\right),$$

and

$$\begin{aligned} a_4(D)=\frac{1}{(4\pi)^{d/2}360} \int_M d^dx\sqrt{g_E}\,\operatorname{tr}\bigl(&60\nabla^2\mathcal E+60R_E\mathcal E+180\mathcal E^2\\ &+12\nabla^2R_E+5R_E^2-2R_{E\mu\nu}R_E^{\mu\nu}\\ &+2R_{E\mu\nu\rho\sigma}R_E^{\mu\nu\rho\sigma} +30\Omega_{\mu\nu}\Omega^{\mu\nu}\bigr). \end{aligned}$$

The functorial derivation fixes the possible invariants by dimension and then determines their universal numerical coefficients through variations, products, and flat-space checks Vassilevich 2003, §4.1, Eqs. (4.13)–(4.28), PDF. For free fields, traces over spin and ghost bundles turn $a_4$ into definite combinations of Weyl-squared, Euler-related, total-derivative, and nonconformal $R^2$ terms Vassilevich 2003, §4.2.2, Eq. (4.35) and Table 1, PDF.

For one real scalar, $\Omega_{\mu\nu}=0$ and $\mathcal E=-\xi R_E$, so direct substitution gives the local coefficient

$$\begin{aligned} b_4={}&\frac1{180} \bigl(R_{E\mu\nu\rho\sigma}R_E^{\mu\nu\rho\sigma} -R_{E\mu\nu}R_E^{\mu\nu}\bigr)\\ &+\frac12\left(\xi-\frac16\right)^2R_E^2 +\frac{1-5\xi}{30}\nabla_E^2R_E. \end{aligned}$$

This step makes the operator-sign check visible: using $\mathcal E=+\xi R_E$ would fail to give the conformal square $(\xi-1/6)^2$.

From coefficients to ultraviolet counterterms

For a real bosonic Gaussian field,

$$\Gamma_E^{(1)}=\frac12\ln\det\frac{D_m}{\mu^2} =-\frac12\int_0^\infty\frac{dt}{t} e^{-tm^2}K(t;D_0),$$

with a regulator understood at $t=0$. In $d=4$, inserting the expansion shows that ultraviolet divergences arise from $a_0$, $a_2$, and $a_4$. Expanding $e^{-tm^2}$ reveals their schematic combination

$$m^4a_0, \qquad m^2a_2, \qquad a_4.$$

They multiply, respectively, local terms of cosmological, Einstein–Hilbert, and curvature-squared type. More precisely, the coefficient of $t^0$ in the massive heat trace is

$$a_4(D_m)=a_4(D_0)-m^2a_2(D_0)+\frac12m^4a_0(D_0).$$

This combination multiplies the logarithmic proper-time integral in four dimensions. A hard proper-time cutoff also displays power divergences from earlier orders; dimensional regularization packages the mass-dependent logarithmic pole in the combination above. Fermions and ghosts change determinant powers and signs; gauge fields also require compatible gauge and ghost boundary conditions. The ultraviolet pole is local, whereas the finite determinant generally contains nonlocal information not reconstructible from finitely many $a_k$.

Boundaries change the coefficient inventory

When $\partial M$ is nonempty, an elliptic operator is incomplete until a boundary operator is specified. Dirichlet, Robin, or mixed data yield additional boundary integrals built from the induced geometry, extrinsic curvature, the Robin endomorphism, and normal derivatives. Odd-index coefficients need no longer vanish; equivalently, half-integer powers appear in the more common local expansion. These are not optional decorations to bulk coefficients: they renormalize boundary couplings required by the chosen variational problem. The boundary structure and the strong-ellipticity qualification are developed in Vassilevich 2003, §§2.1–2.2 and §5, PDF.

An integrated total derivative such as $\int_M\sqrt{g_E}\,\nabla^2R_E$ vanishes on a closed manifold, but becomes a boundary contribution when $\partial M\ne\varnothing$. Therefore one must not discard local derivative terms before the manifold and boundary conditions have been fixed.

Checks on normalization and signs

In flat Euclidean space with $D=-\partial^2$, direct Fourier integration gives

$$K(t;x,y)=\frac{1}{(4\pi t)^{d/2}} \exp\left[-\frac{(x-y)^2}{4t}\right],$$

so $b_0=\mathbf1$ and the normalization of $a_0$ follows. Dimensional analysis gives $[b_{2r}]=2r$, matching every invariant above. For the massless scalar with $\mathcal E=-\xi R_E$,

$$b_2=\left(\frac16-\xi\right)R_E,$$

which vanishes at the four-dimensional conformal value $\xi=1/6$. This is a sign-sensitive check on the identification of $\mathcal E$.

Diagnostic: identify the four-dimensional divergences

Let $D_m=-\nabla_E^2+m^2+\xi R_E$ on a closed four-manifold. Factor the mass as above and determine which local structures can occur in the divergent one-loop action through $a_4$. Explain why the answer changes in the presence of a boundary.

Solution

With $D_0=-(\nabla_E^2+\mathcal E)$ and $\mathcal E=-\xi R_E$, the proper-time product $e^{-tm^2}K(t;D_0)$ gives divergent terms proportional to $m^4a_0$, $m^2a_2$, and $a_4$. Therefore the closed-manifold bulk basis contains $m^4$, $m^2R_E$, and curvature-squared or total-derivative invariants $R_E^2$, $R_{E\mu\nu}R_E^{\mu\nu}$, $R_{E\mu\nu\rho\sigma}R_E^{\mu\nu\rho\sigma}$, and $\nabla^2R_E$, with coefficients depending on $\xi$. This checks dimensions four in every term. A boundary changes the spectral problem and adds half-integer orders and boundary invariants involving extrinsic curvature and the selected boundary operator; moreover, the integrated $\nabla^2R_E$ term becomes a boundary contribution rather than disappearing.

References

• Vassilevich, Dmitri V. “Heat Kernel Expansion: User’s Manual.” Physics Reports 388 (2003): 279–360. DOI. Open PDF, arXiv:hep-th/0306138.