Is the Disformal Cure for Curvature-Coupled Vector Fields Only Leading-Order?

A massive vector field (a "Proca" field) that couples directly to spacetime curvature is a popular ingredient in dark-energy and modified-gravity models. It is also known to misbehave. Around flat space its longitudinal polarization and ordinary gravitational waves both become strongly interacting at a low energy scale. In curved space it develops runaway modes that grow exponentially, which is the "tachyonic" behaviour of a negative effective mass-squared. On expanding cosmological backgrounds it carries an unexpected sixth degree of freedom whose propagation speed is exactly zero.

A proposed fix redefines the metric by subtracting a term proportional to $A_\mu A_\nu$, a so-called disformal transformation, and this appears to remove the gravitational-wave strong coupling and the runaway modes. This note asks whether that fix is real or an artifact of working to leading order. The test is a full, non-linear count of the system's independent degrees of freedom on generic backgrounds.

The argument rests on one structural observation. The disformal map in question contains no derivatives of the field, and for small $\beta A^2$ it can be inverted. A change of variables of that kind cannot change the number of degrees of freedom of a theory. So if the original curvature-coupled theory carries a sixth, extra mode on cosmological backgrounds, the transformed theory must carry it too. That leaves two possibilities. The cure could hold only to first order in the coupling, with the pathology returning at second order. Or the cure could amount to a quietly different theory, in which other matter couples to the new metric. We propose to settle which, and to show that the strong coupling, the curved-space runaway and the sixth mode with zero speed are three symptoms of a single non-linear effect. That effect is the loss of a constraint, of the kind that drives the Ostrogradsky instability, and it is invisible around the trivial background $A_\mu = 0$.

Core Claim

A healthy massive vector has a time component $A_0$ with no time derivative in the action. That absence is a constraint, and it removes one would-be degree of freedom, leaving five once gravity is included. The term $\beta R_{\mu\nu}A^\mu A^\nu$ contains second time derivatives of the metric multiplied by $A^0A^0$. Integrating by parts turns this into a mixing between the metric's rate of change and $\dot A_0$. The quantity that decides whether $A_0$ becomes dynamical is the determinant of the kinetic matrix, written $\mathcal{H}$. This is the matrix of second derivatives of the Lagrangian $L$ with respect to all velocities $\dot q_I$, where the coordinates $q_I$ include $A_0$:

$$\mathcal{H} \equiv \det\left(\frac{\partial^2 L}{\partial \dot q_I\,\partial \dot q_J}\right).$$

Our hypothesis is that $\mathcal{H}$ vanishes identically when $A^0 = 0$ but is generically of order $\beta (A^0)^2$ once the field has a background value. A non-zero value means the constraint is lost and an extra, possibly ghostly mode propagates. A ghost here is a mode with negative kinetic energy, the field-theory version of the Ostrogradsky instability, in which energy is unbounded from below.

The hypothesis explains three published results at once if it holds. Around Minkowski with $A=0$, the extra mode has zero kinetic coefficient, so it is infinitely strongly coupled. That fits the scale $L_{\beta\,\text{str}} \sim (\beta/M_{\text{pl}}^2 m^2)^{1/3}$ found in Unveiling the Inconsistency of the Proca Theory with Nonminimal Coupling to Gravity. On cosmological backgrounds where $A_0 \neq 0$, the mode switches on, which fits the six modes and $c_s=0$ reported in On the cosmological degrees of freedom of Proca field with non-minimal coupling to gravity. Its effective mass-squared then picks up curvature-dependent pieces of order $\beta R$ that can be negative, which fits the curved-space runaway.

The disformal map is

$$g_{\mu\nu} \to \tilde g_{\mu\nu} = g_{\mu\nu} - \frac{\beta}{M_{\text{pl}}^2}A_\mu A_\nu .$$

To first order in $\beta$, it generates a term $-\beta G_{\mu\nu}A^\mu A^\nu$ from the Einstein–Hilbert action, which cancels the Ricci-tensor coupling up to a harmless Ricci-scalar piece. At order $\beta^2$, the curvature of $\tilde g$ contains $(\nabla A)^2$ and $A\nabla\nabla A$ structures. Whether those terms are degenerate, meaning $\mathcal{H}=0$, is exactly the open question. Our prediction is that they are not degenerate to all orders.

Relation to Prior Work

The work that motivates this proposal comes from one line of degree-of-freedom analyses. Unveiling the Inconsistency of the Proca Theory with Nonminimal Coupling to Gravity finds two things: the Ricci-tensor coupling makes the longitudinal vector mode and the gravitational waves strongly coupled at the same scale, while the Ricci-scalar coupling stays benign. It then proposes the disformal map. That analysis is linearized and works to first order in $\beta$. We keep the full non-linear structure and the higher orders in $\beta$.

On the cosmological degrees of freedom of Proca field with non-minimal coupling to gravity finds the sixth mode on cosmological and anisotropic (Bianchi I) backgrounds. Its authors say explicitly that $c_s=0$ makes the linear result untrustworthy. We supply the non-linear criterion they call for.

On the Kalb-Ramond field with non-minimal coupling to gravity shows the same pattern for a rank-2 antisymmetric field. Ricci-tensor couplings are removable by a disformal map, while Riemann couplings need non-local maps. That result constrains our hypothesis. It predicts that the Hessian obstruction for Riemann couplings cannot be cancelled by any algebraic redefinition, and that becomes our second test case.

Two papers show that leading-order analyses hide ghosts and supply methods we reuse. Pathologies of dimension-zero scalar fields rewrites a fourth-order scalar theory with auxiliary fields and exposes a ghost that a gauge-symmetry argument had claimed to remove. We borrow the auxiliary-field technique to make the $\dot A_0$ mixing explicit. Galileons coupled to massive gravity: general analysis and cosmological solutions finds kinetic terms that vanish identically on self-accelerating solutions. That is the same symptom of a zero Hessian on special backgrounds, and we use it as an external comparison rather than a target.

The brane constructions in New Class of Effective Field Theories from Embedded Branes obtain ghost-free theories by inheriting degeneracy from Lovelock geometry, so the absence of extra modes is guaranteed by construction there. We check whether degeneracy survives a field redefinition that has no such geometric protection.

Research Program

We first write the non-minimally coupled Proca theory, with couplings $\alpha R A^2$ and $\beta R_{\mu\nu}A^\mu A^\nu$, in a 3+1 split of space and time. We then run a full Dirac constraint analysis, which classifies the Hamiltonian constraints to decide how many degrees of freedom propagate. This step does not linearize around any background. It yields the Hessian $\mathcal{H}$ as an explicit function of $A_0$, the spatial components $A_i$, the lapse and the extrinsic curvature. It also identifies the field-space locus where the constraint survives.

Next we apply the exact disformal map, with no truncation in $\beta$, and expand the transformed action order by order up to $\beta^3$. At each order we recompute the Hessian. This separates "healthy to first order in $\beta$" from "healthy exactly." It also tells us how the original theory and the minimally coupled Proca theory with matter on $\tilde g$ actually differ physically.

Third, we evaluate the surviving extra mode, if there is one, on spatially flat expanding backgrounds and on Bianchi I backgrounds. On each we extract the sign of its kinetic term, which decides ghost versus healthy. We also extract its effective mass-squared and compare its instability rate with the Hubble rate $H$, the criterion Goon et al. used for tachyons in Galileons on cosmological backgrounds.

Finally, we repeat the Hessian analysis for the massive Kalb-Ramond field with Ricci and Riemann couplings, as a test of whether the criterion generalizes.

All derivations will be done in a computer algebra system and released as worked problems. That makes them reusable as harder test items alongside the nine problems of LLMs with in-context learning for Algorithmic Theoretical Physics. That benchmark's reported failure modes include incorrect reduction of higher derivatives and mistaking flat for cosmological backgrounds, and our problems exercise exactly those.

Baselines

The naive reference point is the linearized analysis around Minkowski with $A_\mu=0$. This is the standard way to count degrees of freedom. It reports five modes and a strong-coupling scale, but no extra mode and no ghost. We will rerun it inside our pipeline, using the same symbolic code and gauge choices. Beating it means finding, on a background it cannot see, a mode whose existence it misses.

The strongest existing methods are two published results. The first is the first-order-in-$\beta$ disformal resolution of Unveiling the Inconsistency of the Proca Theory with Nonminimal Coupling to Gravity, which reports that the map removes tensor strong coupling and the curved-space runaway. The second is the second-order cosmological perturbation analysis of On the cosmological degrees of freedom of Proca field with non-minimal coupling to gravity, which reports six propagating modes (two scalar, two vector, two tensor) and $c_s=0$ for the extra scalar. These two results are the strongest competitors because each is the most complete published analysis of its own question, and they appear to point in opposite directions. Their reported outputs are structural rather than numerical, and there is no reported quantitative accuracy metric for either on Lacuna.

For fairness, our pipeline must first reproduce both published results exactly, on the same backgrounds and with the same coupling ranges, before any new claim is made. Beating them means one of two outcomes. Either we show that the disformal frame, kept to all orders, still carries the sixth mode on cosmological backgrounds, which contradicts the first-order cure. Or we prove degeneracy to all orders, which would show that the sixth mode found in the cosmological analysis is removable and resolve that paper's open question.

The ablation that isolates our key ingredient is the same Hamiltonian analysis with the disformal map truncated at first order in $\beta$. The full non-linear Hessian and the first-order version must disagree for the proposal's central claim to hold. The Ricci-scalar-only theory ($\beta=0$) serves as a negative control: there the Hessian should vanish identically.

Why not just…?

Why not just stay with linear perturbations around Minkowski and de Sitter? On those backgrounds $A_0=0$, which is precisely where our predicted Hessian vanishes. The linear analysis is therefore built to miss the mode. The cosmological result of On the cosmological degrees of freedom of Proca field with non-minimal coupling to gravity, six modes with $c_s=0$, already shows that the mode count jumps once $A_0\neq 0$. The degree-of-freedom discontinuities in Revisiting stability in new general relativity show the same failure in a different theory.

Why not just treat it as an effective field theory and declare the extra mode beyond the cutoff? This is a genuine threat. If the extra mode's kinetic coefficient of order $\beta (A^0)^2$ only becomes important above $L_{\beta\,\text{str}}$ on realistic cosmological backgrounds, the pathology is out of reach and the first-order cure is adequate in practice. Our third step computes the ratio of the extra mode's characteristic scale to the strong-coupling scale, using the background $A_0$ needed for cosmic acceleration. If that ratio places the mode above the cutoff, we will say so: it would mean the first-order cure is enough at practical scales even though it is not exact. The Pathologies of dimension-zero scalar fields case shows that ghosts in such theories are not always removable by an appeal to a cutoff, but it does not settle this specific theory.

Why not just set $\beta=0$ and keep only the benign Ricci-scalar coupling? That is a sound phenomenological choice, and we include it as the control. It does not answer whether the disformal cure is valid. The cure is also used as a template elsewhere, notably for Kalb-Ramond fields in On the Kalb-Ramond field with non-minimal coupling to gravity. Knowing whether it is exact or only first-order affects every theory that relies on it.

Why not just evolve the non-linear equations numerically and watch for blow-up? A numerical blow-up cannot tell a ghost from a tachyon from a strong-coupling artifact. It also cannot count degrees of freedom. We will use short numerical evolutions only as a consistency check on the sign of the instability rate that the Hamiltonian analysis predicts.

Decisive Checks

The pipeline is validated by reproducing three published results. It must recover five modes around Minkowski with the cube-root scaling of $L_{\beta\,\text{str}}$. It must recover six modes and $c_s=0$ on expanding backgrounds. And it must recover the Kalb-Ramond split, in which Ricci-tensor couplings are removable and Riemann couplings are not.

The main test is the sign and order in $\beta$ of the Hessian in the disformal frame. A non-zero result at order $\beta^2$ on expanding backgrounds would confirm that the cure holds only to leading order. Exact vanishing would refute our hypothesis and instead show that the sixth mode is a removable artifact of the frame.

A secondary test is the Riemann-coupled Kalb-Ramond theory. Our account predicts a Hessian obstruction there that no algebraic map can cancel. If such a map does cancel it, the unifying account fails.

If the hypothesis survives, a single computable quantity sorts curvature-coupled vector and tensor theories into three classes: genuinely healthy, healthy only to leading order, and fatally ghostly.