Comparing Gravitational Theories

A fair comparison of gravitational theories asks the same questions in the same order. It does not compare the action of one model with the weak-field equations of another and the cosmological fit of a third. To distinguish candidates, compare their fields, symmetries, matter metric, constraints, validity domain, branches, general-relativistic limit, and predictions on specified backgrounds. If consistency has not been established in a regime, that limitation belongs in the comparison.

Required background. Einstein Equations provides the GR baseline and the Bianchi consistency check; Gravitational Actions and Boundary Terms provides the complete variational problem used to compare actions; Variational and Symplectic Geometry provides the reduced phase space, constraints, and charges used to compare physical rather than component counts.

Helpful background. Formulations of General Relativity supplies the gauge, initial-data, and constraint decomposition used to keep background-specific statements comparable.

Questions that distinguish gravitational theories

The theory specification identifies eleven aspects to compare and questions that test their physical meaning.

Aspect	What to specify	Question to check
Variables	Every dynamical field, nondynamical structure, independent connection or frame, and physical matter metric	Does changing which object is varied change the equations or mode count?
Symmetries	Gauge transformations, exact global symmetries, large transformations, and symmetry-breaking parameters	Does the constraint algebra realize the claimed symmetry on the selected branch?
Action and data	Bulk action, boundary completion, initial/boundary data, regularity, asymptotics, and Green-function prescription	Is the variation differentiable and the solution space unambiguous?
Equations and constraints	Differential order, primary and secondary constraints, gauge conditions, and constraint propagation	Is the constraint rank constant on the background being used?
Physical modes	Tensor, vector, and scalar degrees of freedom after reduction	Do Hamiltonian counting and the reduced quadratic action agree?
Local consistency	Kinetic signs, sound speeds, characteristic cones, hyperbolicity, tachyonic timescales, and causal qualifications	Does an adverse momentum direction or background expose a missing instability?
EFT domain	Canonical normalization, operator ordering, strong-coupling scale, cutoff, and first omitted error	Are every curvature, frequency, gradient, and secular parameter controlled?
Branches	Algebraic roots, cosmological branches, center/horizon data, asymptotic charges, and bifurcations	Can the branch be continued without a singularity or rank change?
GR recovery	Scaling path, source map, uniform solution domain, and dimensionless observables	Do traceful and traceless sources approach their GR responses on the same branch?
Predictions	Charges, multipoles, fluxes, polarizations, propagation laws, growth, and lensing potentials predicted by the theory	Are the conventions and validity conditions sufficient to use the prediction in an independent calculation?
Open questions	A sharply stated missing theorem, computation, matching result, or adverse test	What concrete result would resolve the question or change the conclusion?

Each statement must specify its background and approximation. “Three modes” means little unless it says around which solution, with which matter coupling and boundary data. “Stable” must identify the perturbation sectors and wavenumber range. “GR limit” must specify the scaling path and observable. The power-counting page supplies the conditions of validity, while field redefinitions prevent equivalent operator bases from being counted as distinct theories.

Worked comparison: three actions that look deceptively close

Use the common normalization $M_{\mathrm{Pl}}^{-2}=8\pi G$ and let matter couple minimally to $g_{ab}$. Compare GR with metric and Palatini versions of

$$f(\mathcal R)=\mathcal R+\alpha\mathcal R^2.$$

The symbol $\mathcal R$ means $R(g)$ in the metric theory, but $g^{ab}\mathcal R_{ab}(\Gamma)$ in the Palatini theory. That change in the variable varied is dynamical, not typographic.

Theory	Fields	Symmetries	Degrees of freedom	Cutoff or validity	GR limit
GR	$g_{ab}$; matter metric $g_{ab}$	Diffeomorphisms	Two tensor modes on a regular four-dimensional background	Classical GR below the scale where omitted matter or curvature operators matter	Identity baseline
Metric $R+\alpha R^2$	$g_{ab}$, equivalently $g_{ab}$ plus one scalar where $f_{RR}\ne0$	Diffeomorphisms	Two tensor modes plus one scalar on a regular branch	Treat as a fundamental scalar–tensor model or declare the curvature EFT ordering; the interpretations have different high-frequency claims	$\alpha\to0$ with the scalar mass sent above the used scales and boundary data kept on the GR-connected branch
Palatini $\mathcal R+\alpha\mathcal R^2$	$g_{ab}$ and independent torsion-free $\Gamma^a{}_{bc}$; the connection is compatible with a conformal metric	Diffeomorphisms, with projective qualifications for more general metric–affine actions	Two gravitational tensor modes; the auxiliary scalar is algebraically tied to the matter trace in this model	Matter gradients and connection regularity control the effective description; it is not the propagating metric scalar theory	$8\pi G\alpha T\to0$ with the algebraic branch and matter data and derivatives regular

The table establishes only the shared high-level distinctions. The equations show why the two $f$ functions do not define the same dynamics.

Metric variation produces a scalar wave equation

For

$$S_{\mathrm{met}} =\frac{1}{16\pi G}\int d^4x\sqrt{-g}\,f(R)+S_m[g,\psi],$$

metric variation gives

$$f_RR_{ab}-\frac12fg_{ab} -\nabla_a\nabla_bf_R+g_{ab}\Box f_R =8\pi G T_{ab},$$

where $f_R=df/dR$. Taking the trace in four dimensions,

$$3\Box f_R+f_RR-2f=8\pi G T.$$

For $f=R+\alpha R^2$ this becomes

$$6\alpha\Box R-R=8\pi G T.$$

Around Minkowski vacuum, $R$ therefore obeys

$$(\Box-m_s^2)R=0, \qquad m_s^2=\frac{1}{6\alpha}.$$

For $\alpha\gt0$ the extra scalar has positive mass squared; its kinetic sign and the condition $f_R\gt0$ must also be checked on the background. Introducing an auxiliary curvature $\chi$ makes the mode explicit:

$$\varphi=f_{,\chi}=1+2\alpha\chi, \qquad U(\varphi)=\chi\varphi-f(\chi) =\frac{(\varphi-1)^2}{4\alpha}.$$

The action becomes

$$S_{\mathrm{met}} =\frac{1}{16\pi G}\int d^4x\sqrt{-g} \left[\varphi R-U(\varphi)\right]+S_m,$$

the $\omega=0$ Brans–Dicke representation on domains where the Legendre map is invertible. The metric equations, trace dynamics, and scalar representation are derived in Sotiriou and Faraoni 2010, §II.A, pp. 5–7, equations (5)–(18), and §III.A, pp. 12–13, equations (50)–(63).

Palatini variation makes the same symbol auxiliary

Instead vary $g_{ab}$ and a torsion-free connection $\Gamma^a{}_{bc}$ independently:

$$S_{\mathrm{Pal}} =\frac{1}{16\pi G}\int d^4x\sqrt{-g}\,f(\mathcal R) +S_m[g,\psi].$$

When matter does not couple to $\Gamma$, the equations are

$$f_{\mathcal R}\mathcal R_{(ab)} -\frac12f g_{ab}=8\pi G T_{ab}, \qquad \nabla_c^\Gamma\left( \sqrt{-g}\,f_{\mathcal R}g^{ab} \right)=0.$$

Their trace is algebraic:

$$f_{\mathcal R}\mathcal R-2f=8\pi G T.$$

For $f=\mathcal R+\alpha\mathcal R^2$, all quadratic terms cancel and

$$\mathcal R=-8\pi G T.$$

There is no wave equation for $\mathcal R$. The connection is the Levi–Civita connection of $h_{ab}=f_{\mathcal R}g_{ab}$ on a regular branch, so derivatives of the matter trace enter when $\Gamma$ is eliminated. The scalar–tensor representation has the exceptional Brans–Dicke value $\omega=-3/2$ and an algebraic scalar equation. The metric, Palatini, and metric–affine distinctions are set out in Sotiriou and Faraoni 2010, §§III.B–III.C, pp. 13–15, equations (64)–(70), with matter and connection qualifications reviewed in Olmo 2011, §§2–3, pp. 418–435.

The result illustrates why theory labels are insufficient: identical functions $f$ and identical vacuum GR solutions can conceal different initial data, matter response, and propagating content.

From theory to a testable prediction

To connect a theory with observations, first identify quantities it predicts and the conditions under which they can be calculated. Examples are a scalar charge as a function of compactness and equation of state, a polarization-normalized flux, or two cosmological response functions with their gauge and wavenumber domain. Comparing these predictions with data also requires instrument response, source systematics, likelihoods, numerical bounds, and careful interpretation; Tests of Gravity develops these steps. A favorable background solution alone does not establish observational viability, and an empirical constraint on one parameterization does not establish a theorem about an entire theory family.

Common pitfalls

Compare names rather than fields and equations. “$f(R)$ gravity” can mean metric, Palatini, or metric–affine variation, each with different fields and equations.

Treat an untested sector as consistent. Absence of a published instability in one sector or background is not a proof of full nonlinear consistency. State the missing calculation.

Mix backgrounds between rows. A mode count around Minkowski and a stability statement on FLRW do not combine into one background-independent claim.

Exercises

Complete the physical-mode and GR-limit entries for metric and Palatini $R+\alpha R^2$ gravity. Derive the equation that decides whether the scalar is propagating in each case, and explain what must be held fixed as $\alpha\to0$.

Solution

Metric variation gives

$$6\alpha\Box R-R=8\pi G T.$$

The presence of $\Box R$ supplies independent scalar initial data. In vacuum linearized about Minkowski, $R$ is a scalar mode of mass $m_s^2=1/(6\alpha)$, in addition to the two tensor modes. A controlled $\alpha\to0^+$ limit holds source and boundary data on the branch without an independently excited high-frequency scalar, keeps all measured frequencies well below $m_s$, and compares dimensionless observables. Sending the mass to infinity while prescribing finite scalar-wave initial data would not be the same limit.

Palatini variation instead gives

$$f_{\mathcal R}\mathcal R-2f=8\pi G T \quad\Longrightarrow\quad \mathcal R=-8\pi G T.$$

This is algebraic and supplies no independent scalar-wave initial data. The connection equation makes $\Gamma$ compatible with $h_{ab}=f_{\mathcal R}g_{ab}$, so the modification enters through the local matter trace and its derivatives after eliminating $\Gamma$. Its GR limit holds a regular matter configuration fixed while requiring $|\alpha\mathcal R|\ll1$ and staying on a branch with nonzero $f_{\mathcal R}$. Thus both theories can approach GR, but one decouples a propagating massive scalar while the other removes algebraic matter-dependent connection corrections. The limits are not dynamically identical.

References

• Olmo, Gonzalo J. (2011), “Palatini Approach to Modified Gravity: f(R) Theories and Beyond,” International Journal of Modern Physics D 20, 413–462. DOI; Open PDF.
• Sotiriou, Thomas P., and Valerio Faraoni (2010), “f(R) Theories of Gravity,” Reviews of Modern Physics 82, 451–497. DOI; Open PDF.