The Hamiltonian was defined only up to a variation
In the covariant phase space formalism, the Hamiltonian H[ξ] associated with a diffeomorphism generated by a vector field ξ is not given directly. It is defined implicitly by an equation involving its variation, and earlier work argued that this equation could not generically be solved to obtain H[ξ] itself. Harlow and Wu solved this for pure gravity by using the boundary terms needed for a well-posed variational problem. This paper extends that result to gravity coupled to matter, where all matter fields except true scalars carry gauge freedoms.
The authors' key move is to require more of the action than classical physics demands. Classically, a Lagrangian need only be invariant under gauge transformations up to a total derivative. Quantum mechanically, the action used in a path integral must be exactly gauge invariant. The paper shows that imposing this stronger requirement, together with boundary conditions that make the variational problem well posed, is what allows the Hamiltonian's defining equation to be integrated. The result is a closed expression for an arbitrary theory of gravity coupled to matter, without any integrability condition.
Boundary terms are not bookkeeping
The variational problem is made well posed by adding a (d−1)-form boundary term ℓ to the Lagrangian. The standard condition would be that Θ(φ,δφ) + δℓ vanish on the boundary, where Θ is the presymplectic potential. Following Harlow and Wu, the paper relaxes this to allow a total derivative: on the boundary, Θ(φ,δφ) + δℓ equals dD(φ,δφ). This weaker condition is the consistency condition that later enters the proof.
The paper also uses a dressed symplectic current, built by adding d- and δ-exact terms to the presymplectic potential. Its defining property is that it vanishes on-shell when restricted to the boundary, by virtue of the consistency condition. This is why the boundary data are not incidental: they determine which symplectic structure makes the Hamiltonian well defined. The authors stress that the additional boundary terms are relevant both for the phase space and the Hamiltonian and for evaluating the Euclidean action.
Diffeomorphisms act together with a compensating gauge transformation
For fields with gauge freedom, the action of an infinitesimal diffeomorphism is not simply minus the Lie derivative. It must be supplemented by an induced gauge transformation whose parameter Λξ depends on both the vector field ξ and the field configuration: δξ = −Lξ + δΛξ. For a gauge field A, the parameter takes the form Λξ = ıξA − Pξ, where Pξ is the momentum map.
This structure is what makes the gauge-invariant charges possible. When ξ is a Killing vector k of a solution, the combination of the Lie derivative and the compensating transformation annihilates the field, and the associated on-shell conserved charge is the generalized Komar charge, built from the Noether–Wald charge QNW[k] plus a term wk. The paper's Hamiltonian construction generalizes this machinery so that it applies, for instance, to theories in which the gravitational field is described by a vielbein.
The prescription is applied to explicit theories
The authors compute the Hamiltonian for several theories using their prescription, including general relativity, Einstein–Maxwell theory, and minimal five-dimensional supergravity, which contains Chern–Simons terms. The Chern–Simons case is the one that requires the most care: the Lagrangian is modified by subtracting a background Einstein–Maxwell term and adding a transgression term γQ(A, Ā), so that the new Lagrangian leads to the same equations of motion, is gauge invariant by construction, and vanishes when evaluated on the background fields ēa and Ā.
The paper also examines how ambiguities affect the result. Because several objects are defined only up to total derivatives, the authors track how Θ, Ξ, ℓ, D, R, and QNW shift under the addition of total derivatives. They show that the Hamiltonian charge H[ξ] changes only by a total derivative, so the Hamiltonian obtained by integrating over a (d−2)-dimensional boundary is exactly invariant. This is a statement about the construction's robustness, not about a physical prediction.
The Euclidean action becomes a boundary integral
The second main result connects the Hamiltonian to Gibbons and Hawking's Euclidean approach to quantum gravity. The authors show that, when the action has the required properties and is rotated to Euclidean signature, the on-shell Euclidean action can be expressed in terms of the covariant phase space Hamiltonian, as an integral over a (d−2)-dimensional boundary. This extends earlier results for pure gravity to generic diffeomorphism-invariant field theories.
The same ingredients that made the Hamiltonian integrable are essential here: the boundary terms that guarantee a well-posed variational principle and exact gauge invariance are exactly what allows the Euclidean action to be written as a boundary Hamiltonian. The paper presents this as a structural link between the δ-integrability of the Hamiltonian and the suitability of the action for a Euclidean path integral.
Black-hole backgrounds recover the quantum statistical relation
In black-hole backgrounds, the Hamiltonian gives directly the quantum statistical relation, including terms related to non-gravitational charges and chemical potentials, as expected. For minimal five-dimensional supergravity, the paper evaluates the Hamiltonian on a boundary at infinity and on the horizon. At infinity, after removing vanishing ϑ terms, the integral reduces to β(M − ΩJ + Φ∞Q). At the horizon, where k = 0 and Pk = Φ is constant, the integral reduces to S + ΦHQ.
Combining the two gives the Euclidean action IE = β(M − ΩJ + Φ∞Q) − (S + βΦHQ) = β(M − ΩJ − μQ) − S, with μ = ΦH − Φ∞. The authors emphasize that this recovery is gauge invariant: the horizon statement does not require working in a special regular gauge for the Maxwell field. The result is a consistency check of the construction in a setting where the answer is known, not a new prediction of black-hole thermodynamics.
What the closed expression does not yet cover
The paper's own discussion lists the extensions it has not treated. Theories with scalar fields and scalar charges are left for future work, as is the addition of a cosmological constant and, more generally, terms containing a dimensionful constant, which play the role of thermodynamical variables in Smarr formulas and the first law. The authors also note that deriving the Hamiltonian with the general prescription can be technically difficult: theories with higher-order curvature terms require complicated boundary terms.
A further limitation is built into the setup: the explicit examples use Dirichlet boundary conditions, and the authors state that one would like to derive the Hamiltonian with other boundary conditions related to other statistical ensembles. The quantum statistical relation recovered here is therefore tied to that choice of boundary conditions and to the Einstein–Maxwell class of theories, including Chern–Simons terms, that the paper actually computes.
The most useful way to read the result is as a priority ordering for follow-up work. The construction shows that exact gauge invariance of the action, not merely invariance up to a total derivative, is the ingredient that makes the Hamiltonian integrable and the Euclidean action a boundary integral. The open cases the authors name, scalars, a cosmological constant, higher-curvature terms, and other boundary conditions, are precisely where that ingredient has yet to be supplied.
Sources used in this answer
On the quantum statistical relation and the covariant phase space Hamiltonian
Boundary terms that make the variational principle well posed and the action exactly gauge invariant also make the covariant phase space Hamiltonian integrable, yielding a closed expression that reproduces the quantum statistical relation in black-hole backgrounds, including non-gravitational charges and chemical potentials.
