Decomposing $\delta A_\mu$#

The perturbation of the 1-form decomposes as:

$\delta A_\mu = \left(\delta A_0,\; \partial_i \mathcal{A} + \delta A_i^T\right),$

where $\delta A_0$ and $\mathcal{A}$ are scalar perturbations and $\delta A_i^T$ is a transverse vector perturbation ($\partial^i \delta A_i^T = 0$). The count: $1 + 1 + 2 = 4$ components, matching the 4 components of $A_\mu$.

Notice the structure: the spatial part $\delta A_i$ is split into a longitudinal piece $\partial_i \mathcal{A}$ (a gradient — it points along the direction of propagation for a plane wave) and a transverse piece $\delta A_i^T$ (perpendicular to the direction of propagation). This is the Helmholtz decomposition of a vector field into curl-free and divergence-free parts.

Which Perturbations Propagate?#

From the background analysis in Post 2, we know the $\beta = 0$ component of the equation of motion $\nabla_\alpha F^{\alpha\beta} = 2V_X A^\beta$ is a constraint — it contains no second-order time derivatives. At the perturbation level, this constraint determines $\delta A_0$ in terms of the other variables. It is not a propagating degree of freedom.

The remaining perturbations — the longitudinal scalar $\mathcal{A}$ and the transverse vector $\delta A_i^T$ — are the dynamical ones: they carry their own time derivatives in the action.

The count:

$\underbrace{1}_{\delta A_0\text{ (constraint)}} + \underbrace{1}_{\mathcal{A}\text{ (dynamical)}} + \underbrace{2}_{\delta A_i^T\text{ (dynamical)}} = 4 \text{ components}, \quad 3 \text{ DOFs}.$

This matches the DOF count of 3 from Post 2, as it must. In a generic background these would be the three polarizations of a massive vector: two transverse (from $\delta A_i^T$) and one longitudinal (from $\mathcal{A}$).

The twist on this background. Our background sits at a critical point, $V_X = 0$, and that changes how the modes behave. Expand the potential to second order in $\delta A_\mu$:

$V(X_0 + \delta X) \simeq V_0 + \underbrace{V_X}_{=\,0}\,(2A^\mu\delta A_\mu + \delta A_\mu\delta A^\mu) + \tfrac{1}{2}V_{XX}\,(2A^0\delta A_0)^2 .$

Only $\delta A_0$ gets a “mass” term. The spatial perturbations enter only through the gauge-invariant $\delta F_{\mu\nu}$. So the transverse modes $\delta A_i^T$ propagate like the two polarizations of a massless photon. The longitudinal mode $\mathcal{A}$ enters only as $\dot{\mathcal{A}} - \delta A_0$, so the action is invariant under $\mathcal{A} \to \mathcal{A} + \lambda(\vec x)$. After $\delta A_0$ is eliminated, $\mathcal{A}$ has a kinetic term but no gradient energy ($\omega^2 = 0$): it does not propagate as a wave. The perturbations around this background therefore do not look like a Proca field. This is the structure familiar from “bumblebee” models of spontaneous Lorentz violation, where a vector with a potential minimized at $A_\mu A^\mu \neq 0$ has photon-like Nambu-Goldstone modes.

The Metric Sector#

The metric perturbations decompose as usual. On the isotropic background where $F_{\mu\nu} = 0$ and $V_X = 0$, the vector field’s perturbations couple to the metric perturbations through the linearized Einstein equations. The key results:

• Scalar sector: The metric scalars $\Phi$, $B$, $\Psi$, $E$ are non-propagating (determined by constraints and gauge choices), as in standard GR. The new dynamical scalar is $\mathcal{A}$ from the vector field.
• Vector sector: The metric vectors $B_i^T$, $E_i^T$ are non-propagating. The new dynamical vectors are $\delta A_i^T$.
• Tensor sector: The gravitational wave modes $h_{ij}^{TT}$ propagate as usual — the vector field has no tensor perturbation to contribute.

The massive vector adds 3 DOFs to the 2 gravitational wave DOFs, giving 5 in total: $\mathcal{A}$, $\delta A_i^T$ (2 components), and $h_{ij}^{TT}$ (2 components). (With the caveat above about how $\mathcal{A}$ behaves on the $V_X = 0$ background.)

Decomposing $\delta B_{\mu\nu}$#

The 2-form perturbation has six independent components, which split naturally into temporal-spatial and purely spatial parts:

$\delta B_{0i} \equiv \delta\mathcal{B}_i, \qquad \delta B_{ij} = \epsilon_{ijk}\,\delta\mathcal{C}^k.$

The second relation exploits the fact that an antisymmetric $3 \times 3$ matrix $\delta B_{ij}$ has 3 independent components — the same number as a 3-vector. The Levi-Civita symbol $\epsilon_{ijk}$ provides the map: just as the magnetic field $B^k = \frac{1}{2}\epsilon^{ijk}F_{ij}$ encodes the spatial part of the electromagnetic field strength, the vector $\delta\mathcal{C}^k$ encodes the spatial part of $\delta B_{ij}$.

Each of these 3-vectors gets its own SVT decomposition:

$\delta\vec{\mathcal{B}} = \delta\vec{\mathcal{B}}^T + \nabla(\delta\mathcal{B}), \qquad \delta\vec{\mathcal{C}} = \delta\vec{\mathcal{C}}^T + \nabla(\delta\mathcal{C}),$

where $\delta\mathcal{B}$ and $\delta\mathcal{C}$ are scalars, $\delta\vec{\mathcal{B}}^T$ and $\delta\vec{\mathcal{C}}^T$ are transverse vectors, and:

$\nabla \cdot \delta\vec{\mathcal{B}}^T = 0, \qquad \nabla \cdot \delta\vec{\mathcal{C}}^T = 0.$

The total count: $\delta\mathcal{B}$ (1) + $\delta\vec{\mathcal{B}}^T$ (2) + $\delta\mathcal{C}$ (1) + $\delta\vec{\mathcal{C}}^T$ (2) = 6 components. ✓

No Tensor Perturbations#

Notice something important: the 2-form perturbation has no tensor (transverse-traceless) sector. The SVT decomposition produces only scalars and vectors — no rank-2 transverse-traceless tensors. This is a direct consequence of the antisymmetry of $B_{\mu\nu}$: an antisymmetric tensor simply doesn’t have enough structure to produce a TT piece.

The physical consequence: the 2-form field cannot source gravitational waves at linear order. In a universe where the only matter content is the 2-form, the tensor perturbations $h_{ij}^{TT}$ of the metric satisfy the vacuum wave equation (on the given background). They propagate freely, but the dark energy field does not generate them. This is not unique to the 2-form, though. A quintessence scalar, or the massive vector above, has no linear tensor perturbation either, and SVT decoupling does the rest. In every one of these models, gravitational waves are sourced by the dark energy only at second order in perturbations.

Decomposing the Gauge Parameter#

The gauge transformation $B_{\mu\nu} \to B_{\mu\nu} + \partial_\mu\theta_\nu - \partial_\nu\theta_\mu$ acts on the perturbations through the gauge parameter $\theta_\mu$, which we decompose as:

$\theta_\mu = \left(\theta_0,\; \vec{\theta}^T + \nabla\theta\right),$

where $\theta_0$ and $\theta$ are scalar functions, and $\vec{\theta}^T$ is a transverse vector ($\nabla \cdot \vec{\theta}^T = 0$). This gives $1 + 1 + 2 = 4$ gauge parameters, but the residual symmetry $\theta_\mu \to \theta_\mu + \partial_\mu\varphi$ removes one scalar freedom, leaving 3 effective gauge parameters — matching the count from Post 2.

How Each Perturbation Transforms#

Working out the effect of the gauge transformation on each SVT component:

Temporal-spatial scalars and vectors:

$\delta\vec{\mathcal{B}}^T \to \delta\vec{\mathcal{B}}^T + \dot{\vec{\theta}}^T, \qquad \delta\mathcal{B} \to \delta\mathcal{B} + \dot{\theta} - \theta_0.$

The transverse vector shifts by the time derivative of the gauge parameter’s transverse part. The scalar shifts by the combination $\dot{\theta} - \theta_0$.

Purely spatial scalars and vectors:

$\delta\vec{\mathcal{C}}^T \to \delta\vec{\mathcal{C}}^T + \nabla \times \vec{\theta}^T, \qquad \delta\mathcal{C} \to \delta\mathcal{C}.$

The transverse vector shifts by the curl of $\vec{\theta}^T$. And here is the key result: $\delta\mathcal{C}$ is gauge-invariant. It doesn’t transform at all under gauge transformations.

The reason $\delta\mathcal{C}$ is gauge-invariant is structural. It sits inside $\delta B_{ij} = \epsilon_{ijk}\partial^k(\delta\mathcal{C})$ — the longitudinal part of the spatial perturbation. The gauge transformation adds $\partial_i\theta_j - \partial_j\theta_i$ to $B_{ij}$. Decomposing this into the Levi-Civita form, it contributes only a curl (transverse) piece, which goes into $\delta\vec{\mathcal{C}}^T$, not into $\delta\mathcal{C}$. The longitudinal spatial scalar is untouched.

Gauge-Invariant Combinations#

Beyond the manifestly gauge-invariant $\delta\mathcal{C}$, there is one more gauge-invariant combination:

$\delta\dot{\vec{\mathcal{C}}}^T - \nabla \times \delta\vec{\mathcal{B}}^T.$

This can be verified directly: under the gauge transformation, $\delta\dot{\vec{\mathcal{C}}}^T$ shifts by $\nabla \times \dot{\vec{\theta}}^T$, while $\nabla \times \delta\vec{\mathcal{B}}^T$ shifts by $\nabla \times \dot{\vec{\theta}}^T$ — the shifts cancel.

Since the field strength $H_{\mu\nu\rho}$ is gauge-invariant, the action can only depend on these gauge-invariant quantities. The physical content of the perturbation theory is entirely captured by $\delta\mathcal{C}$ and the combination $\delta\dot{\vec{\mathcal{C}}}^T - \nabla \times \delta\vec{\mathcal{B}}^T$.

Identifying the Propagating Degree of Freedom#

We can now use the gauge freedom to simplify the perturbation content:

1. Use $\vec{\theta}^T$ (2 components) to set $\delta\dot{\vec{\mathcal{C}}}^T = 0$. This is a valid gauge choice. The gauge-invariant combination then reduces to $-\nabla \times \delta\vec{\mathcal{B}}^T$, which no longer carries an independent time derivative.

2. Use $\theta_0$ and $\theta$ (effectively 1 scalar freedom, after accounting for the residual symmetry) to set $\delta\mathcal{B} = 0$.

After gauge fixing, the perturbation variables $\delta\vec{\mathcal{B}}^T$ and $\delta\mathcal{B}$ are fixed by gauge choices, while $\delta\vec{\mathcal{C}}^T$ is constrained (it appears without second-order time derivatives in the equations of motion).

What’s left? The gauge-invariant scalar $\delta\mathcal{C}$. It appears in the action with a time derivative $\delta\dot{\mathcal{C}}$ — specifically, $X = -\frac{1}{12}H_{\mu\nu\rho}H^{\mu\nu\rho}$ expanded to second order in perturbations contains $(\delta\dot{\mathcal{C}})^2$ — making it a genuine dynamical variable with a second-order evolution equation.

Only $\delta\mathcal{C}$ propagates. One degree of freedom, as predicted by the background counting and by Hodge duality. Five of the six perturbation components are either gauge artifacts or constrained variables.

The Physical Identity of $\delta\mathcal{C}$#

What is $\delta\mathcal{C}$, physically?

Recall from Post 1 that a massless 2-form in 4D is Hodge-dual to a scalar field. The duality maps $H_{\mu\nu\rho}$ to $\partial_\mu\phi$ via $H_{\mu\nu\rho} = \epsilon_{\mu\nu\rho\sigma}\partial^\sigma\phi$. On the background, the constant $H_{xyz} = h_0$ corresponds to a homogeneously rolling scalar. Because the Levi-Civita tensor carries a factor $\sqrt{-g} = a^3$, the dual scalar has $\dot\phi \propto h_0/a^3$. This is exactly the kinetic-dominated (stiff) behavior of a free massless scalar, matching $w = +1$ for the free 2-form.

At the perturbation level, $\delta\mathcal{C}$ is the perturbation of this dual scalar. It’s the fluctuation in $\phi$ around its rolling background. All the machinery of antisymmetric tensors, 3-form field strengths, gauge symmetries, and SVT decompositions ultimately distills down to a single scalar fluctuation — precisely what you’d get if you’d worked with the dual scalar from the start.

This is Hodge duality earning its keep: it provides a consistency check on the entire perturbation analysis. If we’d gotten any number other than 1 for the propagating DOFs, or if the surviving perturbation had been a vector rather than a scalar, something would be wrong.

The 2-Form#

To connect perturbations to observations, we need the perturbed stress-energy tensor $\delta T^\mu{}_\nu$. This enters the linearized Einstein equations and determines how the dark energy perturbations affect the gravitational potentials, and hence the CMB and large-scale structure.

Working in the scalar sector (which dominates at late times, since vector perturbations decay), there is a useful shortcut. $T^\mu{}_\nu$ depends on $B_{\mu\nu}$ only through the gauge-invariant $H_{\mu\nu\rho}$. The temporal-spatial scalar $\delta\mathcal{B}$ is pure gauge ($\delta B_{0i} = \partial_i\delta\mathcal{B}$ contributes $\partial_j\partial_k\delta\mathcal{B} - \partial_k\partial_j\delta\mathcal{B} = 0$ to $H$), so it cannot appear. The 2-form enters only through $\delta\mathcal{C}$, via $\delta H_{xyz} = \nabla^2\delta\mathcal{C}$ and $\delta H_{0jk} = \epsilon_{jkl}\,\partial_l\delta\dot{\mathcal{C}}$. The perturbation of $X$ is:

$\delta X = 2X\left(-3\Psi + \frac{\nabla^2\delta\mathcal{C}}{h_0}\right),$

where $\Psi$ is the scalar metric perturbation, and $\nabla^2$ and $\partial_i$ here are comoving coordinate derivatives.

The components of the perturbed stress-energy tensor are:

$\delta T^0{}_0 = f_X\,\delta X,$

$\delta T^i{}_j = -(f_X + 2Xf_{XX})\,\delta X\;\delta^i{}_j,$

$\delta T^0{}_i = 2Xf_X\left(a\,\partial_i B_{\text{metric}} + \frac{a^2}{h_0}\,\partial_i\delta\dot{\mathcal{C}}\right),$

$\delta T^i{}_0 = -\frac{2Xf_X}{h_0}\,\delta^{ij}\partial_j\delta\dot{\mathcal{C}},$

where $B_{\text{metric}}$ is the scalar metric perturbation from $\delta g_{0i}$ (not to be confused with the 2-form field $B_{\mu\nu}$), and $f_{XX} = d^2f/dX^2$. These are exactly the components of a perturbed perfect fluid with $\rho + p = 2Xf_X$ and velocity $u^i = -\partial_i\delta\dot{\mathcal{C}}/h_0$.

Several features are worth noting:

The energy density perturbation $\delta\rho = \delta T^0{}_0$ depends on $\delta X$, which involves both the metric perturbation $\Psi$ and the propagating 2-form perturbation $\delta\mathcal{C}$. The dark energy density fluctuates in response to both the gravitational potential and the field’s own dynamics.

The anisotropic stress vanishes at linear order: $\delta T^i{}_j$ is proportional to $\delta^i{}_j$. So the 2-form dark energy, at linear order, does not produce a difference between the two Newtonian potentials $\Phi$ and $\Psi$, just like a perfect fluid (or a quintessence field).

The pressure perturbation is tied to the density perturbation: $\delta p/\delta\rho = 1 + 2Xf_{XX}/f_X$. This ratio is the sound speed of the 2-form fluid (see below). As a check, the free theory $f = X$ gives $c_s^2 = 1$, as it should for the dual free scalar.

The momentum flux $\delta T^0{}_i$ is set by $\delta\dot{\mathcal{C}}$. The gauge-dependent $\delta\mathcal{B}$ drops out of every component, as it must.

The Massive Vector#

The perturbed stress-energy tensor of the massive vector has a richer structure, reflecting its 3 propagating DOFs. On the background where $F_{\mu\nu} = 0$ and $V_X = 0$, the perturbation of $T_{\mu\nu}$ involves:

• Scalar sector: $\delta A_0$ (constrained) and $\mathcal{A}$ (propagating) contribute to $\delta\rho$, $\delta p$, and the momentum flux.
• Vector sector: $\delta A_i^T$ (propagating) contributes transverse momentum flux — a feature absent in scalar dark energy models.

The vector perturbations are particularly interesting: they represent rotational modes of the dark energy field. In standard $\Lambda$CDM, there are no propagating vector perturbations at all (cosmological vector modes decay). A massive vector dark energy model generically excites these modes, providing a potential observational signature — though detecting them would require extraordinary precision, as they are expected to be small.

Gravitational Waves#

The tensor sector is, somewhat disappointingly, not where these models differ at linear order. The 2-form cannot source gravitational waves at linear order, because its perturbations are purely scalar and vector. The same is true of quintessence and of the massive vector on an isotropic background. All three are minimally coupled to gravity, so gravitational waves propagate exactly as in GR on the corresponding background $a(t)$: no modified friction term, and a propagation speed equal to $c$. They affect gravitational waves only through the expansion history, and through second-order sourcing.

That’s a feature, not a bug, given the observational bound $|c_T - c| \lesssim 10^{-15}$ from GW170817. Modified gravity theories with non-minimal couplings often change the GW friction term or speed. Minimally coupled $p$-form dark energy does not, so an observed modification of GW propagation would point away from these models.

Sound Speed and Clustering#

The sound speed of dark energy perturbations determines whether the dark energy clusters (forms inhomogeneities) or remains smooth. For the 2-form, the single propagating perturbation $\delta\mathcal{C}$ has sound speed

$c_s^2 = \frac{dp/dX}{d\rho/dX} = 1 + \frac{2Xf_{XX}}{f_X},$

which you can read off either from the ratio $\delta p/\delta\rho$ above or from the quadratic action for $\delta\mathcal{C}$. Gradient stability requires $c_s^2 \geq 0$. If $c_s^2 \neq 1$, the dark energy clusters differently from a cosmological constant (which has no perturbations at all) or from quintessence (which generically has $c_s^2 = 1$).

Equation of State Evolution#

Perhaps the most accessible signature is the time dependence of $w$. A cosmological constant has $w = -1$ exactly. Quintessence has $w > -1$ (but can be very close to $-1$). Mathematically, the 2-form can give either $w > -1$ or $w < -1$ (phantom), depending on the sign of $f_X$:

$w = -1 + \frac{2Xf_X}{f}.$

The phantom case ($w < -1$, when $f_X < 0$) doesn’t escape the usual no-go, though. The kinetic term of $\delta\mathcal{C}$ is proportional to $f_X$, so $f_X < 0$ makes the one propagating degree of freedom a ghost, exactly as for a phantom scalar. Equivalently, $\rho + p = 2Xf_X < 0$ violates the null energy condition, and for a single propagating field that always comes with a ghost or a gradient instability. There is no loophole here: by Hodge duality, the 2-form is a scalar in disguise. A healthy 2-form ($f_X > 0$, $c_s^2 \geq 0$) behaves like a (generally non-canonical) quintessence field, with $w > -1$ that evolves in time.

The Massive Vector’s Distinct Signature#

The massive vector, despite being a mere cosmological constant at the background level, has genuinely distinct perturbation physics. Its 3 DOFs — compared to the 2-form’s 1 — mean richer perturbation spectra. The transverse modes $\delta A_i^T$ produce vector-type perturbations that, if detected, would be a smoking gun for spin-1 dark energy. The longitudinal mode $\mathcal{A}$ is the delicate one. On the $V_X = 0$ background it has no gradient energy, which is a warning sign of strong coupling, so its phenomenology has to be treated with care (for example by going beyond linear order, or by adding derivative self-interactions as in generalized Proca theories).

The Bigger Picture#

These two theories are representatives of a much larger landscape. The massive vector is the simplest case of generalized Proca theory, which allows derivative self-interactions while keeping the equations of motion second-order — the spin-1 analogue of Horndeski (scalar-tensor) theory. The massless 2-form with a general $f(X)$ is the analogue of $k$-essence for $p$-forms.

The systematic exploration of this landscape — what theories are healthy (ghost-free, stable), what background solutions they admit, what perturbation spectra they predict, and whether current or future observations can distinguish them from $\Lambda$CDM — is an active area of research. The tools we’ve developed in this series (gauge analysis, DOF counting, SVT decomposition, background and perturbation equations) are the basic toolkit for this program.

The cosmological constant problem remains unsolved. But the search for its resolution has led us through a beautiful corner of theoretical physics — where differential geometry, gauge theory, and cosmology intersect — and the journey is far from over.

Conventions used throughout this series: metric signature $(+,-,-,-)$, natural units $c = \hbar = 1$. Friedmann equations: $H^2 = 8\pi G\rho/3$, $\ddot{a}/a = -4\pi G(\rho + 3p)/3$.