Hidden Simplicity of the Gravity Action

Clifford Cheung and Grant N. Remmen

[7mm] Walter Burke Institute for Theoretical Physics

[-1mm] California Institute of Technology, Pasadena, CA 91125 e-mail:,


We derive new representations of the Einstein-Hilbert action in which graviton perturbation theory is immensely simplified. To accomplish this, we recast the Einstein-Hilbert action as a theory of purely cubic interactions among gravitons and a single auxiliary field. The corresponding equations of motion are the Einstein field equations rewritten as two coupled first-order differential equations. Since all Feynman diagrams are cubic, we are able to derive new off-shell recursion relations for tree-level graviton scattering amplitudes. With a judicious choice of gauge fixing, we then construct an especially compact form for the Einstein-Hilbert action in which all graviton interactions are simply proportional to the graviton kinetic term. Our results apply to graviton perturbations about an arbitrary curved background spacetime.

1 Introduction

The modern scattering amplitudes program has revealed a striking simplicity in gravity that suggests an underlying structure not yet fully understood. At the same time, groundbreaking progress on the experimental front of gravitational wave astronomy [1, 2, 3] has created new opportunities for utilizing insights from formal theory. It is therefore critical that we fully appraise to what extent theoretical advances in gravity have anything to offer by way of real-world applications.

In this paper, we present alternative representations of the Einstein-Hilbert (EH) action that exhibit an immense reduction in the complexity of graviton perturbation theory. Our results involve a general procedure for integrating in auxiliary fields to recast the EH action into a form in which all interactions truncate at finite order. In the minimal construction presented in this paper, we expand the EH action about a flat background in terms of graviton perturbations and a single auxiliary field interacting via purely cubic vertices,


where . Since the corresponding Feynman diagrams are cubic, the mechanics of graviton perturbation theory are drastically simplified. From the cubic Feynman rules, it is then straightforward to derive new off-shell recursion relations for graviton amplitudes, in analogy with the Berends-Giele recursion relations for Yang-Mills theory [4, 5]. In this cubic representation, the Einstein field equations take the form of two coupled first-order differential equations that are at most quadratic in the fields. Note that this construction is a field redefinition away from the first-order Palatini formalism [6], whose cubic structure was emphasized long ago in LABEL:Deser:1969wk, though not in the context of graviton perturbation theory.

Subsequently, we show how a judicious choice of graviton field basis and gauge fixing yields an especially simple form of the EH action. In the phenomenologically relevant case of dimensions, we obtain the gauge-fixed action111Our notational conventions are , , and . Throughout, denotes differentiation to the right, while denotes differentiation to the left.


where is the inverse of expressed as a geometric series in the graviton. Here we have defined the kinetic tensor


whose trace corresponds to the graviton kinetic term. Remarkably, all graviton interaction vertices are given trivially by the kinetic tensor multiplied by powers of the graviton. This simplicity stands in stark contrast with graviton perturbation theory in the conventional approach, where interaction vertices grow intractably lengthy and complex for increasing powers of the graviton.

Let us put our results in context with some other recent approaches related to finding simpler ways of calculating quantities in classical and perturbative quantum gravity, as well as applying the techniques of scattering amplitudes to problems in classical and semiclassical gravity [8, 9]. Indeed, finding ways of simplifying calculations in general relativity is a particularly relevant and pressing problem in light of LIGO’s recent detections of gravitational waves [1, 2, 3]. In particular, the celebrated BCJ double copy [10] relating amplitudes in gauge theory and gravity has been explored in classical contexts [11, 12, 13]. In LABEL:twofold, a field redefinition and gauge fixing of the EH action was found that allowed the Lagrangian to exhibit the twofold Lorentz invariance whose existence was suggested at the level of amplitudes by the double copy; further, the perturbation theory for the action in LABEL:twofold is simpler than the canonical perturbation theory of the EH action. In this paper, we will make simplicity of the action the goal, independent of consideration of the double copy or twofold Lorentz invariance (though this property will make an appearance in Sec. 2.4).

The remainder of this paper is organized as follows. In Sec. 2 we construct a cubic representation of the EH action by integrating in a single auxiliary field. We then derive Feynman rules and off-shell recursion relations for graviton scattering amplitudes. Afterward, in Sec. 3 we derive a further simplified representation of the EH action by exploiting the freedom of gauge fixing. We then discuss the generalization of our results to curved spacetime in Sec. 4 and conclude in Sec. 5.

2 Cubic Formulation

In this section we reformulate the EH action as a theory of purely cubic interactions. To do this, we devise a convenient field basis for the graviton in which the action arises from integrating out a single auxiliary field. We derive the corresponding Feynman rules and off-shell recursion relations for graviton scattering amplitudes. As we will see, the resulting cubic formulation is compact and enjoys an enhanced twofold Lorentz symmetry.

2.1 Lagrangian

2.1.1 Field Basis

All of our results are derived directly from the EH action in dimensions,


working in mostly-plus signature. As shown in LABEL:twofold, the corresponding Lagrangian can be rewritten in the form


where the ellipses denote total derivative contributions that we hereafter neglect. From the second equality in Eq. (5), it is clear that the Lagrangian is naturally a function of the variables222These are sometimes referred to in the literature as the “gothic” variables . We will use the notation for clarity and consistency with LABEL:twofold.


where by definition the “lowered” fields and “raised” fields are inverses of each other, so


In terms of the and fields, the EH action becomes


with the associated Lagrangian


where for later convenience we have defined the vector


which characterizes variations in the volume element.

While the EH action is not conformally invariant, the notion of conformal weight will be a handy bookkeeping tool for terms in the action. Under a conformal transformation, the metric transforms as


which acts on the natural variables in Eq. (6) according to


In particular, the conformal weights are and for the metric and volume measure, respectively, and for the Lagrangian, which is consistent with the mass dimension of the gravitational constant, . In order to abide by the conformal weight counting, the EH Lagrangian must take the schematic form


In other words, every term must carry one more factor of than . For instance, without any additional manipulation, the EH action in Eq. (9) is of the form . As we will see, the EH action can be rewritten in various forms that are homogeneous in powers of and , i.e., for which for a single power . The EH action has many elegant properties when recast into a homogeneous form.

2.1.2 Auxiliary Fields

The conventional approach to graviton perturbation theory entails interaction vertices of arbitrarily high order. That is, the nonlinearities of the action are present for arbitrarily high . However, we will now see how this tower of interactions can be resummed by introducing as few as one auxiliary field. The crux of our construction is to treat as the fundamental field and generate all factors of by integrating out an auxiliary field (or vice versa with and swapped).

To be concrete, let us now describe how to recast the EH action in Eq. (9) into the homogeneous form . We substitute in Eq. (7) to transform the term into a term of the form , yielding


Since Eq. (14) is a quadratic form in , it is natural to treat this field as fundamental and integrate in an auxiliary field that generates the remaining factors. By inverting the term in parentheses in Eq. (14), we obtain the equivalent action


where is an auxiliary field. Note that Eq. (15) is fully equivalent to the EH action, albeit with interactions that truncate at quintic order.

This procedure generalizes in the obvious way. By inserting the Kronecker delta function in Eq. (7) into Eq. (14) in various ways, we can rearrange the Lagrangian into a form with interaction vertices that truncate at any arbitrary but finite order. For example, from Eq. (15) one can derive an alternative quintic action in terms of rather than , plus an auxiliary field with all lowered indices. As we will soon see, the minimal construction of this type results in a cubic Lagrangian.

Returning to Eq. (15), we derive the equation of motion for ,


where is the Christoffel symbol written as an implicit function of the metric in terms of and through Eq. (6). From the above relation, it is natural to define a new auxiliary field with the same index structure as the Christoffel symbol,


so the action in Eq. (15) takes an even simpler form,


In this basis, the natural field variable is and integrating out generates all factors of . The Lagrangian in Eq. (18) is a primary result of this paper: a cubic representation of the EH action in terms of the graviton and a single auxiliary field.

Since Eq. (18) is equivalent to the EH action, the associated equations of motion are equivalent to the Einstein field equations. The equation of motion for the graviton field is


Note that the left-hand side is equal to the Ricci tensor, , which follows from the Jacobian relating and , as derived in LABEL:twofold. Meanwhile, the equation of motion for the auxiliary field is


The two coupled first-order differential equations in Eqs. (19) and (20) are equivalent to the Einstein field equations.

Let us now comment on one final cubic representation of the action. After some rearrangement, Eq. (20) can be written as


Motivated by the link between the auxiliary field and the Christoffel symbol, we go to a field basis in which the auxiliary field is literally equal to the Levi-Civita connection on shell, so


and the action becomes


By construction, the equations of motion set on shell. Partially integrating Eq. (23), we obtain yet another alternative form of the EH action,


plugging in from the definition in Eq. (6). Substituting for in the expression in parentheses yields the Ricci tensor written in terms of Christoffel symbols. The action in Eq. (23) is closely related to the Palatini formalism [6] in which one takes the EH action and treats the connection as a priori independent of the metric; see also Refs. [7, 15].

2.2 Perturbation Theory

The Lagrangians in Eqs. (15), (18), and (23) treat either or as the fundamental fields. However, since we have made no assumptions about the size of the field values, these actions apply for arbitrarily large deviations away from flat space. This is the case even though we have chosen to write these Lagrangians purely in terms of partial rather than covariant derivatives.

On the other hand, it is still of practical interest to study gravity perturbatively in powers of graviton fluctuations about a flat background in Cartesian coordinates, . For Lagrangians in which the fundamental field is , we define , in which case is a geometric series in the graviton. Meanwhile, for those Lagrangians in which the fundamental field is , we use a different but physically equivalent field basis , for which is a geometric series. Note that these two uses of are inequivalent, but are related to each other by a field redefinition (and similarly are related to the graviton field in canonical perturbation theory via a different field redefinition).

Though elegant, the action in Eq. (18) is not yet in a form appropriate for perturbation theory, since there is explicit mixing between the graviton and the auxiliary field. In this section, we will show how to unmix these states and derive the propagators and Feynman vertices for the corresponding graviton perturbation theory. To eliminate the mixing between the gravition and the auxiliary field , we apply the field shift


where indices on and are raised and lowered using the background metric .

After diagonalizing the quadratic term in Eq. (18), we add the gauge-fixing term


so that the graviton propagator is well defined. This gauge choice coincides with harmonic (de Donder) gauge for , i.e., the requirement for the trace-reversed field . Upon gauge fixing, the Lagrangian333For notational convenience, we suppress the normalization of the action in our discussion of the Feynman rules. To convert to the canonically normalized scattering amplitudes, simply multiply the amplitude computed using our Feynman rules by a factor of , together with a factor of for each external graviton. becomes


where the quadratic terms are


and the cubic terms are


where . In this gauge, the graviton propagator takes the simple -independent form


corresponding to propagation from to . The auxiliary field propagator takes the form


corresponding to propagation from to . Meanwhile, the interaction vertices are


The above Feynman rules are summarized in Fig. 1.

Figure 1: Summary of Feynman propagators and vertices for the cubic gravity action in Eq. (27).

2.3 Recursion Relations

Since the Lagrangian in Eq. (18) is comprised of purely cubic interactions, we can derive explicit off-shell recursion relations for tree-level graviton scattering amplitudes in analogy with the Berends-Giele recursion relations for Yang-Mills theory. In fact, these gravity recursion relations are in a sense simpler than for Yang-Mills theory, as the action in Eq. (18) has no quartic interactions.

To begin, let us define the off-shell graviton current , corresponding to an insertion of a graviton field branching off into a set of on-shell gravitons, and the off-shell auxiliary field current , corresponding to an insertion of the auxiliary field branching off into a set of on-shell gravitons. Here the dependence on momentum flowing through the current, , is shown explicitly. The currents are also implicit functions of the momenta and polarization tensors of the remaining on-shell external states. We adopt a convention in which the on-shell gravitons are incoming and the off-shell leg is outgoing, while all gravitons are incoming for the vertices. The currents are equal to


where and are semi-on-shell amplitudes with all legs on-shell except for one leg with momentum corresponding to an off-shell graviton or auxiliary field, respectively.

The graviton current satisfies the recursion relations


while the auxiliary field current satisfies


where each sum runs over all partitions of the set of on-shell graviton labels into distinct subsets and .

As with the Berends-Giele recursion relations, the above equations are to be solved iteratively, order by order in the number of external on-shell gravitons. The initialization step of the recursion relations involves just a single on-shell graviton, where and . The latter vanishes because we are interested only in currents with gravitons as on-shell external states and because we have used the transformation in Eq. (25) to obtain the action in Eq. (27) in which the graviton and auxiliary field do not mix. Using the recursion relations in Eqs. (34) and (35), we have calculated the off-shell graviton current up to fourth order in on-shell gravitons, obtaining the correct three-particle, four-particle, and five-particle amplitudes.

2.4 Enhanced Symmetries

The study of graviton scattering amplitudes has revealed a number of noteworthy surprises, including enhanced cancellations in supergravity theories [16, 17, 18] and the so-called “bonus relations” that arise in the BCFW recursion relations [19, 20]. Another miraculous result is the celebrated “double copy” construction relating graviton scattering amplitudes to the squares of gluon amplitudes, e.g., via the KLT [21] and BCJ [10] relations. In the former representation, graviton scattering amplitudes are expressed as products of Lorentz invariant gluon amplitudes, suggesting a hidden twofold Lorentz invariance within gravity. In LABEL:twofold it was shown that with a careful choice of field basis and gauge fixing one obtains a form of the EH action that exhibits this symmetry explicitly. At the level of the action, twofold Lorentz invariance is manifested as a consistent labeling of all indices as one of two types (e.g., barred and unbarred indices as in ), which are separately contracted [22, 23].

Though not by design, the cubic action in Eq. (18) automatically exhibits a twofold Lorentz symmetry in the spirit of the construction in LABEL:twofold. In particular, one can write Eq. (18) as


where as before we have expanded in before promoting the graviton to a general tensor and the auxiliary field to a pair of fields and . Derivatives can carry either unbarred or barred indices, or , while the metric enters either as or . The action in Eq. (36) is explicitly invariant under an twofold Lorentz symmetry that acts separately on barred and unbarred indices. As discussed in LABEL:twofold, this symmetry can be made manifest in the Lagrangian by introducing an auxiliary extra set of spacetime dimensions.

That there exists a simple cubic formulation of the EH action with twofold Lorentz invariance is particularly enticing given the BCJ prescription in which graviton amplitudes are obtained by squaring the numerators of gluon amplitudes that are expressed in a particular cubic form. A BCJ-compliant action for gravity is conceivable, since BCJ duality has already been demonstrated as a manifest symmetry of a particular cubic representation of the nonlinear sigma model [24]. Unfortunately, when the graviton and auxiliary fields are unmixed, the twofold Lorentz invariance is no longer manifest.

3 Simplified Formulation

Thus far we have only exploited the freedom of choosing a field basis to simplify the Lagrangian. However, gauge fixing also offers enormous leeway in reformulating the action.444Though we introduced a simple gauge-fixing term in Sec. 2.2, we did not use this freedom to make the perturbation theory of the pure gravity action as simple as possible. As we will see, with an appropriate nonlinear gauge fixing it is possible to eliminate the term in Eq. (9) in order to write the action in the homogeneous form . The resulting action is strikingly simple, allowing for a closed-form expression for graviton interaction vertices at arbitrarily high order.

3.1 Lagrangian

3.1.1 Gauge Fixing

We define the full gauge-fixed action to be


where the gauge-fixing term is chosen to be


for some vector . With the benefit of hindsight, we make the special choice


for some constant . The gauge-fixing term in Eq. (38) corresponds to the gauge condition . In terms of coordinates , which are treated as real scalar functions on the spacetime manifold, the gauge condition is equivalent to a condition on the coordinates,


that is, , where is the covariant derivative defined with respect to the full metric . Our gauge condition for general is thus a hybrid of the harmonic and unimodular gauge conditions. To derive this coordinate condition, we used the geometric identities and .

Reshuffling terms and dropping total derivatives, we obtain the full gauge-fixed action


where the Lagrangian is given by


As advertised, every term in Eq. (42) has two fields and one field. To turn this property to our advantage, we again use a field basis in which the graviton perturbations enter as


where as before is simply a geometric series in the graviton field. Rearranging terms via integration by parts, we write Eq. (42) as


where the kinetic tensor is a two-derivative quadratic form in the graviton,


The graviton kinetic term is given by , while all higher-order interactions are simply related to this term by trivial powers of . That is, the graviton interaction is of fixed length and complexity to arbitrarily high order in the graviton. This contrasts sharply with the conventional picture of graviton perturbation theory, where tremendous effort is required to compute the interaction vertex at any given order. Restricting to and setting , we obtain another primary result of this paper, given by the action defined in Eqs. (2) and (3).

3.1.2 Auxiliary Fields

Just as in Sec. 2, the EH action in Eq. (44) can be reformulated as a simple cubic theory of the graviton plus auxiliary fields. That this is possible should be unsurprising since the theory, in terms of graviton perturbations, has the structure of a simple geometric series. Specifically, we find that Eq. (44) is generated by the cubic action


where and are general two-index fields. On shell, one has , which after plugging back into Eq. (46) yields Eq. (44).

We emphasize here that all of the nontrivial derivative structure of gravity is encoded in the kinetic tensor . In particular, the kinetic tensor shoulders triple duty, forming the basis of the graviton kinetic term, the interactions, and the interactions. The remaining terms—the quadratic term and the and interactions—all have trivial index structure.

3.2 Perturbation Theory

In this section we derive the Feynman rules for the action in Eq. (44). As we will see, the interaction vertices are extremely simple. Let us first compute the propagator in our chosen field basis and gauge fixing. Following LABEL:Brandt, a general graviton propagator can be expanded as


where the basis tensors are


Inverting the kinetic term associated with Eq. (45), we obtain the coefficients for the tensor structures in the graviton propagator:


Considering and choosing for the gauge fixing, we find that the propagator takes a particularly simple form,


The geometric series form of our gravity action in Eq. (44) means that the graviton interactions have precisely the same structure as the kinetic term. As a result, we can write down an analytic formula for all Feynman vertices at any order. We first define , where


The term in the action (44) is , where . Thus, the corresponding -point Feynman vertex is


where is the momentum-space version of Eq. (51) obtained by sending and to and , respectively, and where we have raised all indices via . The sum in Eq. (52) runs over each element of the symmetric group of permutations on the set of the legs. For the special case of the three-particle vertex, we obtain