The Conformal-Factor Problem

The Euclidean Einstein action is not bounded below on the space of real positive-definite metrics. Rapid conformal variations can drive it to minus infinity, so the naive weight $e^{-S_E}$ grows rather than decays. This is the conformal-factor problem. It does not prove that every gravitational path integral is meaningless, but it forces any Euclidean or complex saddle prescription to state its integration cycle, gauge treatment, and relation to an original Lorentzian amplitude.

Required background. Gravitational Actions and Boundary Terms supplies the Euclidean action and fixed boundary data; Flat-Spacetime Quantum Field Theory supplies Gaussian contours, determinants, and Wick rotation.

Helpful background. Einstein Equations supplies the saddle equations; Quantum Information supplies state and norm distinctions; and Quantum Gravity supplies the comparison between formal prescriptions and controlled evidence.

The negative conformal direction

Use the Euclidean convention

$$S_E[g]=-\frac{1}{16\pi G}\int d^4x\sqrt g\,(R-2\Lambda)$$

with boundary terms fixed separately. Write $g_{\mu\nu}=e^{2\phi}\bar g_{\mu\nu}$. In four dimensions,

$$R=e^{-2\phi}\left(\bar R-6\bar\Box\phi-6(\bar\nabla\phi)^2\right), \qquad \sqrt g=e^{4\phi}\sqrt{\bar g}.$$

Take $\phi$ smooth with compact support away from the boundary, so the boundary geometry and its action are unchanged. Integration by parts uses

$$\int\sqrt{\bar g}\,e^{2\phi}\bar\Box\phi =-2\int\sqrt{\bar g}\,e^{2\phi}(\bar\nabla\phi)^2.$$

Thus

$$S_E=-\frac{1}{16\pi G}\int d^4x\sqrt{\bar g} \left[e^{2\phi}\left(\bar R+6(\bar\nabla\phi)^2\right)-2\Lambda e^{4\phi}\right].$$

The gradient term has a negative overall sign. Taking a smooth conformal mode with fixed small amplitude but increasing wavenumber makes this contribution arbitrarily negative. Gibbons, Hawking, and Perry give the decomposition and its contour interpretation in Gibbons, Hawking, and Perry 1978, pp. 143–148.

This direction is related to the indefinite DeWitt supermetric, but it is not removed merely by saying that the conformal mode is gauge. Local Weyl transformations are not a gauge symmetry of Einstein gravity. Diffeomorphism gauge fixing alone does not remove the off-shell conformal instability. Equally, the instability does not mean there is a propagating scalar ghost among the two on-shell Lorentzian graviton polarizations: an off-shell Euclidean integration direction and a physical propagating mode are different objects.

Candidate contour prescriptions

The original conformal rotation sets a suitably normalized fluctuation $\phi=iy$ near a saddle, turning a growing Gaussian into a decaying one. This defines a local steepest-descent contour, including a Jacobian and orientation. Extending it globally is harder because nonlinear terms, gauge zero modes, and other negative modes can change the descent directions.

A Lorentzian-first prescription begins with $e^{iS_L}$ and complexifies lapse or metric variables only to deform the original oscillatory cycle. Picard–Lefschetz flow then decides which saddle thimbles have nonzero intersection number with that cycle. This avoids declaring the real Euclidean slice fundamental, but singular endpoints and infinite-dimensional flow remain nontrivial.

A third position treats the conformal mode together with the measure, ghosts, and renormalization group. Some cancellations or effective actions can change the naive quadratic picture in a restricted model. Such a result must show the complete determinant and contour; a formal cancellation of one negative eigenvalue is not a nonperturbative definition.

Negative modes are not all the same

The high-frequency conformal instability is distinct from a finite number of physical negative modes around a Euclidean saddle. A physical negative mode can signal thermodynamic instability or an imaginary part associated with decay. A gauge zero mode instead calls for a collective coordinate. The Hessian spectrum, boundary conditions, and contour determine which interpretation applies.

As an adverse check, suppose one rotates every negative mode in the same direction without tracing the original cycle. The local Gaussian converges, but its phase can be wrong, and the chosen thimble may have zero intersection with the Lorentzian contour. Feldbrugge, Lehners, and Turok show in minisuperspace that the lapse contour and thimble intersections can select a background saddle weighting different from a guessed Euclidean saddle prescription Feldbrugge, Lehners, and Turok 2017, §§II–IV.

Results and open questions as of 2026-08-20

The unbounded real Euclidean action is established; the globally correct contour for the full gravitational path integral is not settled. The competing positions differ in what they take as primary data.

Prescription or position	Evidence or controlled result	Scope, adverse result, or limitation
Local conformal rotation	Converts the quadratic conformal Gaussian near selected Euclidean saddles into a convergent integral	Global nonlinear cycle and phase are prescribed; additional negative modes require separate treatment
Lorentzian Picard–Lefschetz definition	Original oscillatory cycle fixes contributing thimbles in finite-dimensional and minisuperspace models	Extension to the full metric configuration space and singular boundaries remains unresolved
Measure-and-ghost cancellation	Complete one-loop calculations can remove gauge artifacts and identify genuine modes	Gauge cancellation does not remove every off-shell conformal direction or define topology sums
Canonical or causal formulations	Avoid a fundamental real Euclidean metric integral and retain causal structure	Must still construct dynamics, continuum limit, and correspondence with known Lorentzian amplitudes

The conservative conclusion is that Euclidean gravity is a saddle-calculation tool with a declared contour, not an ordinary positive statistical measure. A stronger claim that one contour supplies the unique nonperturbative theory remains conjectural. The problem is separate from string and holographic constructions and from semiclassical thermodynamic interpretation, which are treated elsewhere.

Any proposed cure must also preserve gluing: amplitudes computed on adjacent regions should compose after integration over their shared boundary data. Convergence achieved at the price of a contour that violates this composition law would not define the intended quantum dynamics.

The conclusions could change after a mathematically controlled Lorentzian metric cycle is shown to induce a regulator-independent conformal contour, or after independent calculations establish that a proposed measure and ghost treatment removes the instability while preserving boundary composition and physical amplitudes beyond a fixed truncation.

Exercises

Let $a\gt0$ and consider the divergent formal integral $I=\int_{-\infty}^{\infty}d\phi\,e^{a\phi^2}$. Rotate to $\phi=iy$ and determine the convergent value and its phase. Why is this not yet a solution of the gravitational problem?

Solution

With the contour oriented from $-i\infty$ to $+i\infty$, $d\phi=i\,dy$ and $a\phi^2=-ay^2$. Thus the rotated integral is $I_{\mathcal C}=i\int_{-\infty}^{\infty}dy\,e^{-ay^2}=i\sqrt{\pi/a}$. The opposite orientation changes the sign. This local calculation fixes convergence only after a contour and orientation are chosen. Gravity has infinitely many coupled modes, gauge and stabilizer factors, nonlinear singularities, boundary data, and an original Lorentzian cycle whose intersection numbers decide whether this rotated direction actually contributes. The toy Gaussian tests the Jacobian and phase; it cannot select the full gravitational contour.

References

• Feldbrugge, Job, Jean-Luc Lehners, and Neil Turok. “Lorentzian Quantum Cosmology.” Physical Review D 95 (2017): 103508. DOI.
• Gibbons, G. W., S. W. Hawking, and M. J. Perry. “Path Integrals and the Indefiniteness of the Gravitational Action.” Nuclear Physics B 138 (1978): 141–150. DOI.