Scalar Fields on Curved Spacetimes

A real scalar field is the cleanest place to see what curvature changes—and what it does not. Mass and curvature coupling enter the lower-order part of the wave equation, so they change dispersion, tails, and stress–energy without changing the metric light cone that controls propagation. On a globally hyperbolic spacetime, compact Cauchy data still determine one smooth solution in their causal domain.

In the distributional language used below, a test function is smooth with compact support, a distribution is a continuous linear functional on such tests, and the support of a distribution is the smallest closed set outside which it vanishes. A Green operator sends a compact source to a solution with a specified causal support. This is the local bridge needed before using the fuller PDE treatment.

Required background. Partial Differential Equations and Green Functions provides test distributions, Green operators, characteristics, and support conditions; Classical Field Theory supplies action variation, boundary terms, and conserved currents. Helpful background. Flat-Spacetime Quantum Field Theory provides the oscillator and mode interpretation of free fields; Causality and Global Structure establishes Cauchy surfaces, causal futures and pasts, and global hyperbolicity.

The covariant scalar equation

Let $(M,g)$ be a smooth, time-oriented, globally hyperbolic four-dimensional spacetime. With the site signature $(-,+,+,+)$ and curvature convention, take

$$S[\phi] =-\frac12\int_M d^4x\,\sqrt{-g} \left[ \nabla_\mu\phi\nabla^\mu\phi +(m^2+\xi R)\phi^2 \right].$$

Here $m\geq 0$ is the mass and $\xi$ is dimensionless. Varying only the field, integrating the kinetic term by parts, and requiring the boundary variation to vanish gives

$$\delta_\phi S =\int_M d^4x\,\sqrt{-g}\, \delta\phi\, \left(\Box-m^2-\xi R\right)\phi,$$

so

$$P\phi=0, \qquad P\equiv\Box-m^2-\xi R, \qquad \Box=g^{\mu\nu}\nabla_\mu\nabla_\nu.$$

The principal symbol is $\sigma_P(x,k)=g^{\mu\nu}k_\mu k_\nu$ times the identity. Neither $m^2$ nor $\xi R$ appears in it. Consequently the characteristic covectors remain null: curvature coupling can scatter waves inside a light cone, but it does not replace the cone with a mass- or coupling-dependent causal structure. This separation between principal and lower-order terms is the first durable lesson of curved-spacetime propagation.

Cauchy data and causal dependence

Choose a smooth spacelike Cauchy hypersurface $\Sigma$ with future-directed unit normal $n^\mu$. The initial data are

$$\phi\rvert_\Sigma=\phi_0, \qquad n^\mu\nabla_\mu\phi\rvert_\Sigma=\phi_1.$$

For smooth compactly supported $\phi_0$, $\phi_1$, and a compact source $f$, the inhomogeneous problem $P\phi=f$ has a unique smooth solution. If

$$K=\operatorname{supp}\phi_0 \cup\operatorname{supp}\phi_1 \cup\operatorname{supp}f,$$

then $\operatorname{supp}\phi\subset J(K)$. The result also includes continuous dependence on the data in the appropriate test-function topologies. These are precisely the existence, uniqueness, support, and continuity statements proved for normally hyperbolic operators in Bär, Ginoux, and Pfäffle 2007, Chapter 3, §§3.2–3.3, pp. 73–92. Global hyperbolicity matters because every inextendible causal curve crosses $\Sigma$ once; it rules out both missing incoming information and causal information returning without being registered by the data.

This theorem does not say that the field is concentrated only on the null boundary. In generic curved spacetimes, retarded solutions can have tails throughout $J^+(K)$. Finite propagation speed is a support bound, not Huygens’ principle. The algebraic formulation uses the same causal dependence and test-function quotient, as summarized by Hollands and Wald 2015, §§1–2.1.

Conserved current and stress–energy

For any two real solutions $\phi_1$ and $\phi_2$, define the bilinear current

$$j^\mu(\phi_1,\phi_2) =\phi_1\nabla^\mu\phi_2 -\phi_2\nabla^\mu\phi_1.$$

Direct differentiation gives

$$\nabla_\mu j^\mu =\phi_1P\phi_2-\phi_2P\phi_1=0.$$

The mass and curvature potential cancel because they multiply both fields symmetrically. This current will become the symplectic form after integration over a Cauchy surface.

Metric variation instead yields the Hilbert stress tensor

$$\begin{aligned} T_{\mu\nu} ={}&\nabla_\mu\phi\nabla_\nu\phi -\frac12 g_{\mu\nu} \left(\nabla_\alpha\phi\nabla^\alpha\phi+m^2\phi^2\right)\\ &+\xi\left[ G_{\mu\nu}\phi^2 +g_{\mu\nu}\Box(\phi^2) -\nabla_\mu\nabla_\nu(\phi^2) \right]. \end{aligned}$$

The $\xi$ terms are not optional decorations: they arise from varying $R\phi^2$. Using the contracted Bianchi identity and commuting covariant derivatives on the scalar gradient gives

$$\nabla^\mu T_{\mu\nu}=(P\phi)\nabla_\nu\phi,$$

so stress–energy is conserved on shell. This is a prescribed-background statement. Feeding $T_{\mu\nu}$ back into the metric equation belongs to semiclassical or classical backreaction, not to the propagation problem solved here.

Independent checks and limiting cases

In Minkowski spacetime, $R=0$ and a plane wave $e^{-i\omega t+i\mathbf k\cdot\mathbf x}$ obeys

$$\omega^2=\lvert\mathbf k\rvert^2+m^2,$$

which checks the signs of both $\Box$ and the mass term. For minimal coupling, $\xi=0$, the stress tensor reduces to the familiar scalar expression. For $m=0$ and $\xi=1/6$ in four dimensions, the action is classically conformally invariant and the improved tensor has vanishing on-shell trace. These limits separately test the dispersion sign, the metric variation, and the curvature-coupling coefficient.

The conformal trace check can be done directly, without assuming the answer. In four dimensions, contract the displayed stress tensor and use $\Box(\phi^2)=2\nabla_\alpha\phi\nabla^\alpha\phi+2\phi\Box\phi$. On shell the result is

$$T^\mu{}_{\mu} =(6\xi-1)\left[\nabla_\alpha\phi\nabla^\alpha\phi+\xi R\phi^2\right] +(6\xi-2)m^2\phi^2.$$

Setting $m=0$ and $\xi=1/6$ makes each coefficient vanish. This is a classical identity; the later quantum trace anomaly is not a failure of this classical calculation.

The causal check is equally important: changing $m$ or $\xi$ must not change $\sigma_P$. If a calculation predicts a new characteristic cone merely from the potential $m^2+\xi R$, it has confused a lower-order scattering effect with the principal part of the PDE.

Check your understanding

Suppose $R$ is constant on a region and $m^2+\xi R\lt0$ there. Does the scalar equation cease to be normally hyperbolic or propagate outside the metric light cone?

Solution

No. Normal hyperbolicity and the characteristic cone depend only on the principal symbol $g^{\mu\nu}k_\mu k_\nu$, which is unchanged. Because the effective mass multiplies the field without derivatives, it does not change this highest-derivative part. Provided the spacetime and coefficients satisfy the stated smooth global hypotheses, the Cauchy problem still has causal finite propagation. A negative effective squared mass can suggest an instability, but establishing one requires the geometry, spatial spectrum, and boundary conditions. For example, in the flat constant-potential model $P=\Box+\mu^2$, one obtains $\omega^2=\lvert\mathbf k\rvert^2-\mu^2$: modes with $\lvert\mathbf k\rvert\lt\mu$ grow exponentially, while their disturbance remains inside the metric causal cone. Constant scalar curvature alone does not justify this flat-space Fourier decomposition. Causal support and dynamical stability are independent tests.

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 10.4171/037.
• Hollands, Stefan, and Robert M. Wald. “Quantum Fields in Curved Spacetime.” Physics Reports 574 (2015): 1–35. DOI 10.1016/j.physrep.2015.02.001.