Differential Obstructions to Curvature-Dependent Conformal Transformations

Abstract: Curvature-dependent conformal rules are local forward assignments on known metrics, but they are not generically local changes of metric variables. We study the nondegenerate class $\widetilde g_{μν}=F(R[g])g_{μν}$, with $F&gt;0$ and $F_R\neq0$. Introducing an independent auxiliary scalar localizes the forward map, while recovering the original metric requires a differential constraint. The complete inverse metric tangent map contains the nonpolynomial projector $ξ<sup>μξ<sup>ν/ξ<sup>2$; hence no differentiable finite-jet inverse, i.e., a formula involving only finitely many derivatives at the same point, exists on an open set of unrestricted metrics. Branchwise functional inverses may nevertheless exist after boundary or Cauchy data are specified. Metric $f(R)$ gravity gives an explicit realization: its local Einstein-frame scalar--tensor representation is a parent theory, whereas a metric-only Einstein-side description requires a differential section governed by a normal operator. We derive the pulled-back classical Hessian and its off-shell embedding term, and in the quadratic model show how constrained Gaussian elimination produces the scalaron nonlocal kernel, the corresponding normal determinant for the displayed measure, and the zero-mode compatibility condition. Exact parent and metric solutions remain equivalent; the obstruction concerns locality and the off-shell variational and fluctuation domains. These results provide a precise framework for assessing curvature-dependent frame transformations in modified gravity and clarify their implications for effective actions, semiclassical analyses, and quantum frame equivalence.

• The paper proves that nondegenerate transformations of the form \(\widetilde g_{\mu\nu}=F(R)g_{\mu\nu}\) cannot generally have differentiable finite-jet metric-only inverses because their linear response contains the nonpolynomial factor \(\xi^\mu\xi^\nu/\xi^2\).
• It shows that metric \(f(R)\) gravity remains classically equivalent to its scalar–tensor parent on shell, while off-shell reconstruction requires a differential constraint, boundary or Cauchy data, and a Green-function prescription.
• The paper derives physical and quantum consequences—including scalaron poles, Yukawa corrections, cosmological response, nonlocal quadratic kernels, and possible zero-mode failures—while distinguishing locality obstructions from absolute noninvertibility or instability.

The paper establishes a precise locality obstruction for conformal transformations whose conformal factor depends on the Ricci scalar of the metric being transformed. While such maps are local in the forward direction—given a known metric $g_{\mu\nu}$, one computes $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$ directly—their inversion requires reconstructing the curvature of an unknown preimage metric, a self-consistent differential fixed-point problem rather than an algebraic substitution. The central result is that, for the nondegenerate class $F>0$, $F_R\neq0$, the complete inverse metric response contains the nonpolynomial momentum factor $\xi^\mu\xi^\nu/\xi^2$, so that no differentiable finite-jet (finitely many derivatives at a point) metric-only inverse exists on an open set of unrestricted configurations.

The finite-jet obstruction

Writing the preimage as $g_{\mu\nu}=F(R_J)^{-1}\widetilde g_{\mu\nu}$, self-consistency demands $R_J=R[F(R_J)^{-1}\widetilde g]$, which expands into a differential fixed-point equation whose formal iteration generates arbitrarily high derivatives. To make this precise, the authors introduce an independent auxiliary scalar $X$ via $\widetilde g_{\mu\nu}=F(X)g_{\mu\nu}$ and impose the projection constraint $$\widehat{\mathcal P}_F[\widetilde g,X]\equiv R[F(X)^{-1}\widetilde g]-X=0$$. Linearizing about a projected background and computing principal symbols yields

$\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$0

A finite-order differential operator has polynomial symbol; the factor $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$1 is nonpolynomial, so any hypothetical $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$2 finite-jet inverse would contradict this expression. The argument is careful to note that derivative dependence alone does not imply noninvertibility—certain derivative-dependent transformations do admit finite-order inverses—and that the obstruction is one of locality, not of absolute invertibility: branchwise functional inverses exist once a functional domain, boundary or Cauchy data, and a Green prescription (e.g., retarded) are specified. Nonlocality here does not entail acausality.

A concrete illustration of off-shell degeneracy is given by fibers of the forward map: for $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$3 and $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$4, homogeneous positive-branch preimages obey a nonlinear oscillator equation, so the same transformed Minkowski metric admits infinitely many regular off-shell preimages. A global metric-only chart therefore requires a section prescription.

Metric $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$5 gravity: parent theory versus metric-only description

Metric $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$6 gravity provides the physical realization. On a regular Legendre branch ($\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$7, $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$8), the theory is equivalent to a scalar–tensor parent action in the Jordan frame, and the Einstein-frame transformation $\widetilde g_{\mu\nu}=F(R[g])g_{\mu\nu}$9 is genuinely local because $F>0$0 is an independent coordinate. The original metric theory is recovered only by imposing the differential constraint

$F>0$1

which identifies the parent scalar with the curvature of the reconstructed metric. Crucially, exact regular solutions satisfy this constraint automatically—a parent-shell identity shows $F>0$2 follows from the full parent equations—so classical on-shell equivalence is untouched. The distinction arises off shell, where arbitrary parent configurations need not lie in the projected set $F>0$3.

The normal operator and its observable consequences

Linearizing the projection splits fluctuations into a metric part $F>0$4 and a scalar part governed by the "normal operator"

$F>0$5

On constant projected backgrounds this reduces to a Klein–Gordon operator with mass parameter $F>0$6, related to its Jordan counterpart by $F>0$7. An intrinsic cross-check from the Jordan trace equation confirms the same pole. The sourced tangent relation $F>0$8 makes explicit that the scalaron, while intrinsic to the metric theory, cannot generically be written as a local finite-order functional of the Einstein metric alone.

This structure has direct observational content. In the weak-field limit of $F>0$9, the static section gives the Yukawa response $F_R\neq0$0 with charge $F_R\neq0$1, reproducing the standard $F_R\neq0$2 corrections to the Newtonian potentials. In cosmology, prescribing an Einstein-frame expansion history does not determine $F_R\neq0$3 algebraically: the projection becomes a nonlinear second-order ODE requiring branch and Cauchy data, with the linearized form being a driven damped oscillator sourced by $F_R\neq0$4. In the quasistatic subhorizon regime, the same denominator organizes the effective Newton coupling $F_R\neq0$5 and slip ratio, up to background-curvature terms neglected in that approximation.

Projected Hessian and constrained Gaussian theory

A main formal result is the pulled-back Hessian of the metric-only Einstein-side action $F_R\neq0$6:

$F_R\neq0$7

where the second term—the embedding correction—is proportional to the parent scalar Euler derivative and vanishes on the common shell. Geometrically it is the field-theoretic analogue of the second fundamental form of the constraint surface. Consequently three distinct quadratic objects must not be conflated: the projected metric Hessian, the tangent-restricted parent Hessian (which omits the embedding term and is neither of the others), and the unrestricted parent Hessian. This inequivalence appears before any determinant is evaluated and explains the known on-shell agreement but off-shell mismatch between metric and scalar–tensor one-loop calculations (Ruf et al., 2017, Ruf et al., 2017, Ohta, 2017).

In the quadratic model, constrained Gaussian elimination over the delta-functional-enforced constraint produces simultaneously the nonlocal scalaron kernel

$F_R\neq0$8

the normal determinant associated with the displayed factorized measure, and, when zero modes are present, a cokernel compatibility condition on the metric source. The paper emphasizes that the determinant's power and phase depend on the fundamental measure and that the pushforward Jordan measure may modify it; a complete one-loop comparison still requires gauge fixing, ghosts, regularization, and zero-mode prescriptions. Below the scalaron scale, the kernel admits a controlled derivative expansion $F_R\neq0$9; promoting a finite truncation to an exact kinetic operator would introduce spurious roots outside EFT validity.

Breakdown of the graph section

Beyond nonlocality, the metric-only graph can fail entirely when $\xi^\mu\xi^\nu/\xi^2$0 is not invertible on its chosen domain. For $\xi^\mu\xi^\nu/\xi^2$1, the condition $\xi^\mu\xi^\nu/\xi^2$2 gives candidate degeneracies at $\xi^\mu\xi^\nu/\xi^2$3—regular Legendre points with a massless constant vertical mode—but these are diagnostics only, becoming actual obstructions solely when the mode belongs to the chosen domain and satisfies its boundary conditions. Similarly, on compact Euclidean backgrounds a zero mode occurs when $\xi^\mu\xi^\nu/\xi^2$4 (e.g., $\xi^\mu\xi^\nu/\xi^2$5 on $\xi^\mu\xi^\nu/\xi^2$6). The paper carefully separates these chart failures from Legendre degeneracy, tachyonic instability (which does not by itself spoil the Cauchy problem), and genuine flat directions of the action Hessian: a normal zero mode alone does not imply a collective coordinate.

Limitations and open questions

The theorem applies strictly to the nondegenerate Ricci-scalar class; extensions require case-by-case audit. In Palatini $\xi^\mu\xi^\nu/\xi^2$7 the scalar constraint is algebraic before connection elimination, so the obstruction is absent there, while hybrid metric–Palatini and general metric-affine theories demand coupled scalar–connection normal-operator matrices whose analysis is left open. The construction also stops short of a numerical one-loop effective action: the induced measure, constraint Jacobian, projected gauge and ghost sectors, and regularization remain unspecified. Finally, the global nonlinear projection problem—uniqueness of branchwise inverses, coexistence of multiple off-shell preimages, genuine folds or bifurcations beyond linearized diagnostics—is unresolved, as is the possible existence of exceptional symmetry-reduced sectors admitting local inverse responses.

Conclusion

The paper demonstrates that curvature-dependent conformal rules are local forward assignments on known metrics but not generically local changes of metric variables: their inverses contain inverse differential operators, require functional domain and boundary or Cauchy data, and can fail at spectral obstructions. In metric $\xi^\mu\xi^\nu/\xi^2$8 gravity, the scalar–tensor Einstein frame is a local parent theory, while a metric-only Einstein-side description exists only through a differential section whose Green inverse governs weak-field charges, cosmological response, and the nonlocal scalaron kernel. The practical prescription is an audit procedure—identify the carrier of the conformal factor, derive the elimination constraints, and test the complete composite response $\xi^\mu\xi^\nu/\xi^2$9—before interpreting any curvature-dependent transformation as a genuine change of off-shell variables.