Weak Formulations of Einstein's Equations

A weak Einstein equation is not specified by writing $G_{\mu\nu}=8\pi G T_{\mu\nu}$ beneath a low-regularity metric. One must say how curvature is defined, which test objects are allowed, which nonlinear products exist, how gauge is handled, and what convergence preserves the equation. Distributional, reduced-hyperbolic, variational, and measure-valued formulations answer those questions differently.

Required background. Analysis for Mathematical Relativity supplies distributions, weak convergence, and compactness; The Einstein Equation supplies the classical system; PDEs and Green Functions for Gravitation supplies test-function formulations.

Helpful background. Causality and Global Structure helps separate a weak field equation from causal or inextendibility conclusions.

Distributional geometry

If $g$ is a Geroch–Traschen regular metric—locally bounded with locally bounded inverse and weak first derivatives in $L^2_{\mathrm{loc}}$—then its Riemann and Einstein tensors are tensor distributions. The equation

$$G(g)+\Lambda g=8\pi G T$$

means equality after pairing both sides with every smooth compactly supported test tensor density of the complementary type. This formulation is coordinate invariant under smooth coordinate changes because the curvature distribution was constructed tensorially.

Its advantage is direct geometric meaning. Its cost is a restrictive metric class. It admits codimension-one shell curvature but does not make arbitrary distribution-valued metrics or codimension-two concentrations legal. Geroch and Traschen establish the definition and stability properties in Geroch–Traschen 1987, §§II–VI.

Even within this class, a distributional Einstein equation alone does not automatically provide a unique evolution. Existence, constraint propagation, and causal interpretation need separate theorems.

Weak reduced equations

After harmonic gauge fixing, the principal part is a quasilinear wave operator. For a representative component,

$$g^{\alpha\beta}\partial_\alpha\partial_\beta u =Q(g,\partial g),$$

multiply by a compactly supported smooth test function $\varphi$ and integrate once by parts:

$$-\int \partial_\beta u\, \partial_\alpha\!\left(g^{\alpha\beta}\varphi\right)\,d^4x =\int Q\varphi\,d^4x.$$

This identity makes sense when the coefficient, weak derivative, and quadratic terms lie in compatible Lebesgue spaces. It is useful for energy methods and weak limits, but it is gauge dependent. One must additionally impose the gauge constraint weakly and prove that its subsidiary equation forces it to vanish. Otherwise the solution is only a weak solution of the reduced system.

For instance, locally bounded $g^{-1}$, $\partial g^{-1}\in L^2$, and $\partial u\in L^2$ make the left-hand side finite: expanding its derivative gives one bounded-coefficient term tested against $\partial\varphi$ and one $L^1$ product tested against $\varphi$. This verifies that the identity is defined; it does not yet prove solvability or uniqueness.

Changing to rough coordinates can also change the apparent admissible class. Geometric uniqueness therefore needs a regularity class for the comparison diffeomorphism, not only for the metric components.

Variational formulations

A variational solution makes the first variation of an action vanish against compactly supported metric variations. Because the Einstein–Hilbert action contains second derivatives before boundary terms are separated, low-regularity use requires care: choose a first-order form or a controlled boundary term, specify the metric space, and verify that the action and its variation are finite.

Variational structure can expose conserved currents and constraints, but stationarity need not imply the full distributional Einstein equation unless integration by parts and the fundamental lemma are valid in the selected class. Likewise, the weak contracted Bianchi identity requires enough regularity to justify diffeomorphism variations. It is not a formal license to differentiate every distributional product.

Compensated compactness and measure-valued limits

Suppose smooth solutions $g_n$ have uniform weak bounds and converge to $g$. Linear terms pass to the limit, but quadratic expressions in $\partial g_n$ may not. Special div–curl or null structures can force cancellation; this is compensated compactness. When oscillations survive, their effect is recorded by a defect measure or a Young measure rather than discarded.

In high-frequency gravitational limits, the defect can behave like an effective stress tensor. Under polarized $U(1)$ symmetry, Huneau and Luk prove rigorous instances in which weak vacuum limits satisfy an Einstein equation coupled to effective massless matter; their review separates proved symmetry classes from the general Burnett conjecture in Huneau–Luk 2024, §§1.1–1.3 and §§3–6.

A measure-valued formulation records the probability distribution of unresolved gradients. Its barycenter need not be a classical or distributional vacuum solution, because averaging and nonlinear evaluation do not commute.

Strong convergence illustrates what removes the simplest defect. If $\Gamma_n\to\Gamma$ strongly in $L^2$ on every compact coordinate set, then componentwise

$$\|\Gamma_n\Gamma_n-\Gamma\Gamma\|_{L^1} \le(\|\Gamma_n\|_{L^2}+\|\Gamma\|_{L^2}) \|\Gamma_n-\Gamma\|_{L^2}\longrightarrow0.$$

Also $\partial\Gamma_n\to\partial\Gamma$ distributionally by integration against derivatives of test functions. The mixed-index curvature therefore converges distributionally. Any additional metric contractions must still be justified in the chosen metric-convergence class. Weak $L^2$ convergence alone does not supply this estimate.

Four formulations compared

Formulation	Basic unknown and test	Main strength	Main limitation
Distributional geometry	Regular metric; test tensor densities	Coordinate-invariant curvature and thin shells	Restrictive products and codimension support
Reduced hyperbolic	Gauge-fixed components; scalar or tensor tests	Compatible with PDE energy methods	Gauge and subsidiary constraints must be recovered
Variational	Metric in a finite-action class; compact variations	Encodes symmetries and weak Euler–Lagrange equations	Boundary terms and equivalence to distributional equations require proof
Measure-valued or defect	Limit metric plus oscillation measure	Retains nonlinear high-frequency energy	Usually weaker uniqueness and geometric interpretation

No row universally contains the others. A shell spacetime may fit distributional geometry cleanly; a high-frequency sequence may require a defect measure; a local existence proof may be most natural in reduced wave coordinates.

Solved diagnostic: weak limits do not preserve squares

For the sequence below, compute the weak limits of its first derivative and of the derivative squared. What defect remains?

Solution

On the circle, let

$$u_n(x)=\frac{1}{n}\sin(nx).$$

Then $u_n\to0$ uniformly and $u_n'=\cos(nx)\rightharpoonup0$ weakly in $L^2$. Nevertheless,

$$(u_n')^2=\cos^2(nx)\rightharpoonup\frac12,$$

not zero. Thus the weak limit of the quadratic term is not the quadratic term of the weak limit. The missing $1/2$ is the simplest model of a defect measure. Einstein’s quadratic connection terms have additional tensorial cancellations, but those cancellations must be proved rather than assumed.

The impulsive-wave and thin-source examples show how the distributional formulation works when the singular support is a hypersurface.

References

• Geroch, Robert, and Jennie Traschen. “Strings and Other Distributional Sources in General Relativity.” Physical Review D 36 (1987): 1017–1031. DOI.
• Huneau, Cécile, and Jonathan Luk. “High-Frequency Solutions to the Einstein Equations.” arXiv:2404.07659, 2024.

Further reading

• LeFloch, Philippe G., and Cristinel Mardare. “Definition and Stability of Lorentzian Manifolds with Distributional Curvature.” Portugaliae Mathematica 64 (2007): 535–573. EuDML record.