Group Actions and Quotient Geometry

A symmetry reduces Einstein’s equations only where its orbits form a sufficiently regular geometric family. If a Lie group acts freely and properly by isometries, the orbit space is a smooth manifold and the spacetime fields can be reorganized into fields on that lower-dimensional space. Fixed points, null orbits, and changes of isotropy do not invalidate symmetry, but they do prevent the entire quotient from being treated as one ordinary smooth bundle.

Required background. Use Killing fields and isometry groups for the infinitesimal symmetry condition and the Einstein equation for the parent field system.

Helpful background. Hypersurface normals and tangential projectors supplies the projection idea reused for nondegenerate symmetry orbits.

Orbits, isotropy, and a smooth quotient

Let a Lie group $G$ act smoothly on a spacetime $(M,g)$ by a map

$$\Phi:G\times M\longrightarrow M, \qquad \Phi_a^*g=g.$$

The orbit through $p$ is $G\!\cdot p$, and the isotropy subgroup

$$G_p=\{a\in G:\Phi_a(p)=p\}$$

measures how much of the group fixes that point. Its dimension controls the orbit dimension:

$$\dim(G\!\cdot p)=\dim G-\dim G_p.$$

The cleanest reduction occurs when the action is free, so every $G_p$ is trivial, and proper, so sequences of group transformations cannot produce non-Hausdorff orbit identifications. Then the projection $\pi:M\to B=M/G$ makes $M$ a principal $G$-bundle over a smooth manifold $B$ of dimension $\dim M-\dim G$. Independent Killing generators provide the vertical tangent directions. This is the regular setting underlying the invariant mixed-basis decomposition in Pons and Talavera 2004, §§2.1–2.2, pp. 5–8, PDF.

Freeness is sufficient, not necessary, for obtaining useful geometry. If all isotropy groups have the same conjugacy class, one can work on a smooth orbit-type stratum. When the isotropy type changes, $M/G$ is naturally stratified: each regular stratum is smooth, while exceptional orbits sit on lower-dimensional pieces. Silently extending regular-bundle formulas across those pieces is a common source of false boundary conditions and division by a vanishing Killing norm.

Which fields descend

A scalar descends to $B$ precisely when it is constant along every orbit. For a generator $X$, this is $\mathcal L_X f=0$. A differential form must be both invariant and horizontal—its contraction with every vertical generator must vanish—to be the pullback of a form on $B$. Invariant tensors need not themselves be horizontal, so invariance alone does not turn every spacetime tensor into a tensor of the same type on the quotient.

Suppose the orbit metric is nondegenerate. At each regular point the tangent space then splits into vertical and orthogonal horizontal parts,

$$T_pM=V_p\oplus H_p, \qquad V_p=T_p(G\!\cdot p), \qquad H_p=V_p^\perp.$$

For horizontal lifts $X^H,Y^H$ of vectors on $B$, the quotient metric is

$$h(\pi_*X^H,\pi_*Y^H)=g(X^H,Y^H).$$

If the removed orbit is timelike, $h$ is Riemannian; if it is spacelike, $h$ retains one timelike direction. A null orbit has $V_p\cap V_p^\perp\ne0$, so this orthogonal construction is degenerate and requires a different null reduction.

The spacetime metric contains more information than $h$. In an adapted local trivialization it also carries an internal orbit metric and connection-like cross terms. Those fields become scalars and gauge potentials on $B$. Thus a vacuum Einstein metric in the parent dimension normally reduces to gravity coupled to effective matter fields, not to vacuum gravity in fewer dimensions. Geroch’s one-Killing formalism makes this norm–twist content explicit Geroch 1971, pp. 918–924.

Fixed axes as an independent check

Minkowski spacetime in cylindrical coordinates provides a decisive test:

$$ds^2=-dt^2+dr^2+r^2d\phi^2+dz^2, \qquad X=\partial_\phi, \qquad g(X,X)=r^2.$$

For $r>0$, the $SO(2)$ action has one-dimensional circular orbits and the quotient metric is

$$h=-dt^2+dr^2+dz^2.$$

At $r=0$, the whole rotation group fixes the point and the orbit dimension falls to zero. Including that stratum gives a half-space whose edge records the fixed axis; the regular quotient itself has $r\gt0$ and excludes this edge. It is not one principal $SO(2)$-bundle across the axis. The parent spacetime is perfectly smooth there. The apparent singularity is in the regular-orbit variables, not in curvature.

This example also separates three failure modes. Nonfree actions produce exceptional strata, nonproper actions can spoil Hausdorffness, and null orbits make the orthogonal quotient metric degenerate. None is diagnosed merely by checking that a Killing equation holds.

Exercise

For the cylindrical Minkowski metric above, show that an invariant scalar $F(t,r,z)$ has four-dimensional wave operator

$$\Box F=-\partial_t^2F+\partial_r^2F+\frac1r\partial_rF+\partial_z^2F.$$

Why is this not simply the wave operator of the quotient metric $h$?

Solution

The determinant is $\sqrt{-g}=r$, so

$$\Box F=\frac1r\partial_\mu\!\left(rg^{\mu\nu}\partial_\nu F\right) =-\partial_t^2F+\frac1r\partial_r(r\partial_rF)+\partial_z^2F.$$

Expanding the radial term gives the stated formula. The quotient metric is flat with $\sqrt{-h}=1$, so its intrinsic wave operator lacks $r^{-1}\partial_rF$. That extra term records the changing volume $2\pi r$ of the symmetry orbit. A correct reduction therefore retains the orbit-size scalar rather than discarding it with the coordinate $\phi$.

References

• Geroch, Robert. “A Method for Generating Solutions of Einstein’s Equations.” Journal of Mathematical Physics 12, no. 6 (1971): 918–924. DOI.
• Pons, Josep M., and Pere Talavera. “Consistent and Inconsistent Truncations: Some Results and the Issue of the Correct Uplifting of Solutions.” Nuclear Physics B 678 (2004): 427–454. DOI. Open PDF.