Uniqueness and bracket consistency of the stress tensor obtained from a curvature-independent Hamiltonian on arbitrary space-like surfaces (Dirac 1955)

In 1955 Dirac showed that the stress tensor of a classical field theory can be fixed by connecting it with the density of a Hamiltonian formulated on arbitrary space-like surfaces: if the Hamiltonian density $\beta_{\mu}$ is curvature-independent — equivalently, its Poisson brackets with the surface-deformation generators contain the Dirac delta and at most its first derivatives — then $\beta_{\mu}=T_{\mu\nu}l^{\nu}\Gamma$ defines a conserved, symmetric stress tensor; conversely, a Hamiltonian formulation on flat surfaces together with such a $T_{\mu\nu}$ extends to arbitrary surfaces. Four gaps in this program remain without a literature closure: no intrinsic selection principle makes the curvature-independent Hamiltonian, and hence $T_{\mu\nu}$, unique (contact transformations act non-trivially); the Poisson-bracket regularity condition is asserted to be necessary but is not proven necessary and sufficient in general field-theory settings; field variables carrying a free Lorentz index (such as the electromagnetic potential $A_{\mu}$) need special canonical changes and extra terms in the angular-momentum generator that were only sketched; and in gauge theories $\beta_{\mu}$ is ambiguous up to additions proportional to the gauge generator, leaving $T_{\mu\nu}$ determined only up to weakly vanishing terms. The later literature absorbed parts of this program — Kuchař's hyperspace treatment of tensor fields on arbitrary hypersurfaces, the hypersurface-deformation algebra of Hojman-Kuchař-Teitelboim, the general theory of first-class constraints, and covariant gauge-improvement procedures for the energy-momentum tensor — and Kuchař confirmed that no uniqueness of the deformation representation exists for source fields; but no intrinsic selection principle has been found and no complete characterization of the bracket conditions exists. The task is to complete Dirac's correspondence — uniqueness, characterization, generality over field types, and gauge control — within the canonical framework on arbitrary space-like surfaces.

Background

The stress (energy-momentum) tensor $T_{\mu\nu}$ of a classical field theory collects the local densities of energy and momentum: the total four-momentum $P_{\mu}$ is the integral of $T_{\mu\nu}l^{\nu}$ over a space-like hypersurface with unit normal $l^{\nu}$, and local conservation $\partial^{\nu}T_{\mu\nu}=0$ expresses four-momentum balance. The standard requirements on $T_{\mu\nu}$ — correct total charges, local conservation, the existence of a localized angular-momentum density, and symmetry — do not fix it uniquely. One may always add an improvement (superpotential) term, the divergence of an antisymmetric tensor, in four space-time dimensions, of the form $\partial_{\mu}\partial_{\nu}S-g_{\mu\nu}\partial^{2}S$ built from an arbitrary scalar $S$; this changes the local localization of energy and momentum while leaving every total charge and the symmetry and conservation properties intact. Two classical prescriptions remove this ambiguity by extra structure: Belinfante's symmetrization of the canonical Noether tensor derived from the Lagrangian density, and the definition by variation of the coupled action with respect to the metric (documented, with primary references, in the 2016 review of Blaschke, Gieres, Reboud and Schweda). The first involves a non-canonical simplicity choice; the second invokes gravitational coupling, which is not intrinsic to a purely non-gravitational field theory.

Dirac's 1955 paper proposed a third, intrinsic route from Hamiltonian dynamics. In a canonical (Hamiltonian) formulation each field variable is paired with a conjugate momentum and dynamics is generated by a Hamiltonian through Poisson brackets; this is usually set up on flat slices $x^{0}=\text{const}$. Dirac instead let the surface be an arbitrary space-like three-surface, embedded by functions $y_{\mu}(u)$ of surface coordinates $u=(u^{1},u^{2},u^{3})$, promoted the embedding coordinates to canonical variables with conjugate momenta $w_{\mu}$ (so that $[y_{\mu}(u),w_{\nu}(u')]=\delta(u-u')$ where $\delta$ is the Dirac delta on the surface), and wrote the four surface-deformation generator densities $\mathscr{H}_{\mu}$ — which generate deformations of the surface together with the field dynamics — in the split $\mathscr{H}_{\mu}=w_{\mu}+\beta_{\mu}$, where $\beta_{\mu}$ does not depend on the momenta $w_{\mu}$. His key hypothesis is curvature independence: $\beta_{\mu}$ at a surface point is unchanged by infinitesimal deformations of the surface that fix $y_{\mu}$ and its first derivatives at that point, i.e. $\beta_{\mu}$ does not depend on the local extrinsic curvature of the embedding. He showed this is equivalent to a distributional regularity statement about Poisson brackets: the brackets of $\beta_{\mu}$ with the deformation generators may contain $\delta(u-u')$ and its first derivatives $\delta^{r}$, but no second or higher derivatives. Under this condition $\beta_{\mu}$ determines a genuine space-time tensor field through $\beta_{\mu}=T_{\mu\nu}l^{\nu}\Gamma$, with $l^{\nu}$ the unit normal and $\Gamma$ the surface-volume factor; conservation and symmetry of this $T_{\mu\nu}$ then follow from the invariance, under surface deformations, of the total translation and angular-momentum generators $P_{\mu}=\int\beta_{\mu}\,d^{3}u$ and $M_{\mu\nu}=\int(y_{\mu}\beta_{\nu}-y_{\nu}\beta_{\mu})\,d^{3}u$. Conversely, given a Hamiltonian formulation only on flat surfaces together with a suitable stress tensor, the same machinery extends the formulation to arbitrary surfaces. Throughout, Dirac used his weak-equality convention: equations that hold only after the constraints $\mathscr{H}_{\mu}\approx 0$ are imposed are called weak (written $\approx$), while identities are strong.

What the paper did not settle can be stated precisely. First, contact transformations — canonical transformations of the extended phase space (field variables, embedding variables and their momenta) that mix coordinates and momenta — map curvature-independent densities $\beta_{\mu}$ to new curvature-independent densities, so $T_{\mu\nu}$ is fixed only modulo this ambiguity; no classification of, or selection principle for, admissible Hamiltonians was given. Second, the no-higher-derivative regularity condition was asserted to be necessary for the construction and used as a working sufficiency hypothesis, without a characterization of its precise scope. Third, when basic field variables carry a free Lorentz index, as the electromagnetic potential $A_{\mu}$ does, the construction requires a preliminary canonical change to surface-adapted components (the tangential components $A^{r}=A_{\mu}y^{\mu r}$ and the normal component $A_{l}=A_{\mu}l^{\mu}$) and additional terms in the angular-momentum generator $M_{\mu\nu}$; this was indicated but not developed. Fourth, in gauge theories the total surface Hamiltonian contains an arbitrary multiple of a first-class gauge generator $\mathscr{G}$ (a constraint density, $\mathscr{G}\approx 0$, whose Poisson brackets with all constraints vanish weakly and which generates gauge transformations), and the split $\mathscr{H}_{\mu}=w_{\mu}+\beta_{\mu}$ is then ambiguous up to additions proportional to $\mathscr{G}$, so the stress tensor is determined only up to weakly vanishing terms.

For context: the bracket analysis of surface-deformation generators became a cornerstone of canonical general relativity (the algebra of hypersurface deformations), and the improvement ambiguity of $T_{\mu\nu}$ is a standard topic in field theory to this day; the problem below asks for the still-missing completion of Dirac's own program.

Problem Statement

Complete Dirac's 1955 program by answering, within the canonical framework on arbitrary space-like surfaces and without appealing to gravitational coupling or ad hoc simplicity choices:

(1) Uniqueness and selection. Give a classification of curvature-independent Hamiltonian densities $\beta_{\mu}$ modulo admissible contact transformations of the extended phase space (field variables, embedding variables $y_{\mu}$ and their conjugate momenta $w_{\mu}$; admissible means preserving curvature independence), and either a selection principle intrinsic to the Hamiltonian formulation that singles out one $\beta_{\mu}$ and hence a unique stress tensor $T_{\mu\nu}$, or a proof that no such intrinsic principle exists (the induced stress tensors being inequivalent with no canonical criterion to choose among them).

(2) Bracket characterization. Determine precisely for which field theories the regularity condition — the Poisson brackets of the surface densities $\beta_{\mu}$ with the deformation generators $\mathscr{H}_{\mu}$ contain $\delta$ and at most its first derivatives — is both necessary and sufficient (a) for reconstructing a surface-independent tensor field from $\beta_{\mu}$ through $\beta_{\mu}=T_{\mu\nu}l^{\nu}\Gamma$ and (b) for closing the Poisson-bracket algebra of the deformation generators; in particular, establish in which settings higher derivatives of $\delta$ can occur and whether the construction then fails or can be repaired.

(3) Fields with a free Lorentz index. Carry out and verify the construction when the basic field variables carry a free Lorentz index, such as the electromagnetic potential $A_{\mu}$: the canonical change to surface-adapted components, the additional terms in the angular-momentum generator $M_{\mu\nu}$, and the verification of the conservation laws and Poisson-bracket closure in such cases.

(4) Gauge theories. Give a systematic account of the ambiguity $\beta_{\mu}\to\beta_{\mu}+c\,\mathscr{G}$ with $\mathscr{G}$ the first-class gauge generator: how the ambiguity is to be resolved, and state exactly to what extent $T_{\mu\nu}$ is determined up to weakly vanishing terms.

These four items are components of a single objective: to make the correspondence between the surface-deformation Hamiltonian densities and the local stress tensor unique, fully characterized, general over field types, and gauge-controlled, purely within the canonical Hamiltonian framework on arbitrary space-like surfaces. A full resolution requires all four items; a complete answer to (3) or (4) alone settles only the corresponding indexed-field or gauge-theory case and is a partial result toward the overall objective, and within (1) the classification alone leaves the selection question open while within (2) both the reconstruction half (a) and the closure half (b) are required.

References

1. P. A. M. Dirac, "The stress tensor in field dynamics", Il Nuovo Cimento 1(1), 16-36 (1955). DOI: 10.1007/BF02731753, https://doi.org/10.1007/BF02731753.
2. P. A. M. Dirac, "The theory of gravitation in Hamiltonian form", Proceedings of the Royal Society of London A 246, 333-343 (1958). DOI: 10.1098/rspa.1958.0142, https://doi.org/10.1098/rspa.1958.0142.
3. F. J. Belinfante, "On the spin angular momentum of mesons", Physica 6, 887-898 (1939). DOI: 10.1016/S0031-8914(39)90090-X, https://doi.org/10.1016/S0031-8914(39)90090-X.
4. F. J. Belinfante, "On the current and the density of the electric charge, the energy, the linear momentum and the angular momentum of arbitrary fields", Physica 7, 449-474 (1940). DOI: 10.1016/S0031-8914(40)90091-X, https://doi.org/10.1016/S0031-8914(40)90091-X.
5. Curtis G. Callan, Sidney Coleman, Roman Jackiw, "A new improved energy-momentum tensor", Annals of Physics 59, 42-73 (1970). DOI: 10.1016/0003-4916(70)90394-5, https://doi.org/10.1016/0003-4916(70)90394-5.
6. Karel Kuchař, "Geometrodynamics regained: A Lagrangian approach", Journal of Mathematical Physics 15, 708-715 (1974). DOI: 10.1063/1.1666715, https://doi.org/10.1063/1.1666715.
7. Karel Kuchař, "Kinematics of tensor fields in hyperspace. II", Journal of Mathematical Physics 17, 792-800 (1976). DOI: 10.1063/1.522977, https://doi.org/10.1063/1.522977.
8. Karel Kuchař, "Dynamics of tensor fields in hyperspace. III", Journal of Mathematical Physics 17, 801-820 (1976). DOI: 10.1063/1.522978, https://doi.org/10.1063/1.522978.
9. Sergio A. Hojman, Karel Kuchař, Claudio Teitelboim, "Geometrodynamics regained", Annals of Physics 96, 88-135 (1976). DOI: 10.1016/0003-4916(76)90112-3, https://doi.org/10.1016/0003-4916(76)90112-3.
10. D. I. Olive, "Kac-Moody and Virasoro Algebras in Local Quantum Physics", in Fundamental Problems of Gauge Field Theory (NATO ASI Series), 51-92 (1986). DOI: 10.1007/978-1-4757-0363-4_3, https://doi.org/10.1007/978-1-4757-0363-4_3.
11. D. C. Salisbury, "Rosenfeld, Bergmann, Dirac and the Invention of Constrained Hamiltonian Dynamics", arXiv:physics/0701299 (2007). https://arxiv.org/abs/physics/0701299.
12. Marc Henneaux, Claudio Teitelboim, "Quantization of Gauge Systems", Princeton University Press (1992). DOI: 10.1515/9780691213866, https://doi.org/10.1515/9780691213866.
13. D. I. Olive, "Paul Dirac and the Pervasiveness of His Thinking", arXiv:hep-th/0304133 (2003). https://arxiv.org/abs/hep-th/0304133.
14. C. G. Torre, "Covariant Phase Space Formulation of Parametrized Field Theories", arXiv:hep-th/9204055 (1992). https://arxiv.org/abs/hep-th/9204055.
15. Madhavan Varadarajan, "Dirac Quantization of Parametrized Field Theory", arXiv:gr-qc/0607068 (2006). https://arxiv.org/abs/gr-qc/0607068.
16. Daniel N. Blaschke, François Gieres, Méril Reboud, Manfred Schweda, "The energy-momentum tensor(s) in classical gauge theories", Nuclear Physics B 912, 192-223 (2016). DOI: 10.1016/j.nuclphysb.2016.07.001, https://arxiv.org/abs/1605.01121.
17. Nicoleta Voicu, "Energy-momentum tensors in classical field theories - A modern perspective", International Journal of Geometric Methods in Modern Physics 13, 1640001 (2016). DOI: 10.1142/S0219887816400016, https://arxiv.org/abs/1511.08729.
18. Martin Bojowald, Erick I. Duque, Aiden Shah, "Hypersurface deformations", arXiv:2410.18807 (2024). https://arxiv.org/abs/2410.18807.
19. Damianos Iosifidis, Manthos Karydas, Anastasios Petkou, Konstantinos Siampos, "On the geometric origin of the energy-momentum tensor improvement terms", arXiv:2503.21609 (2025). https://arxiv.org/abs/2503.21609.