Hamiltonian Constraint Counting

The Dirac constraint algorithm answers a question that the field equations alone can hide: how many independent initial-data functions actually propagate? A singular Legendre transform produces constraints; their Poisson-bracket algebra then separates gauge redundancy from physical restrictions. For a regular constrained system with $N_q$ configuration variables, $N_{\mathrm{FC}}$ independent first-class constraints, and $N_{\mathrm{SC}}$ independent second-class constraints, the local count is $N_{\mathrm{dof}}=N_q-N_{\mathrm{FC}}-N_{\mathrm{SC}}/2$.

This count is necessary before diagnosing kinetic signs, characteristic cones, or strong coupling. It is local in phase space and must be repeated if the constraint rank changes between backgrounds or solution branches.

Required background. Variational and Symplectic Geometry supplies the reduced-phase-space interpretation of gauge directions, while Ordinary Differential Equations supplies the initial-data counting used after reduction.

Helpful background. Einstein Equations provide the gravitational baseline, Gravitational Actions and Boundary Terms explain why the chosen action and boundary data matter, and Formulations of General Relativity supplies the ADM split used in the gravitational check below.

From a singular Lagrangian to the constraint algebra

Let $L(q^A,\dot q^A)$ have $N_q$ configuration variables. The canonical momenta

$$p_A=\frac{\partial L}{\partial \dot q^A}$$

cannot all be inverted for the velocities when the Hessian

$$W_{AB}=\frac{\partial^2L}{\partial\dot q^A\partial\dot q^B}$$

is singular. Relations $\phi_a(q,p)\approx0$ that follow immediately are primary constraints. The weak equality sign means that a relation is imposed only after Poisson brackets have been evaluated. The total Hamiltonian is therefore

$$H_T=H_c+u^a\phi_a,$$

with multipliers $u^a$. Requiring $\dot\phi_a=\{\phi_a,H_T\}\approx0$ may fix multipliers, produce secondary constraints, or reveal an identity. The process continues until no new independent condition appears. This is the finite-dimensional form of the Dirac–Bergmann algorithm developed in Date 2010, Chapter 4.

A constraint is first class if its Poisson bracket with every constraint vanishes on the full constraint surface. First-class chains generate gauge transformations, although a primary and its descendants may combine into one gauge generator. A second-class set has an invertible bracket matrix

$$C_{ab}(x,y)=\{\chi_a(x),\chi_b(y)\}.$$

Second-class constraints are physical restrictions rather than gauge choices. They can be solved explicitly or incorporated with the Dirac bracket. Classification must be done after the complete constraint set is known: a constraint that appears first class at an intermediate stage can pair with a later secondary constraint.

Why the counting formula works

An unconstrained system has a $2N_q$-dimensional phase space. Each independent second-class constraint removes one phase-space direction. Each independent first-class constraint removes two: one direction by restricting the constraint surface and one by quotienting the associated gauge orbit. Hence the physical phase-space dimension is

$$\dim\Gamma_{\mathrm{phys}} =2N_q-2N_{\mathrm{FC}}-N_{\mathrm{SC}}.$$

Dividing by two gives the configuration-space degree-of-freedom count stated in the lead. The result assumes regular, functionally independent constraints and a gauge action of constant rank. Reducible gauge symmetries require subtracting relations among generators; irregular constraints require a separate local analysis. Boundary degrees of freedom also need their own symplectic treatment and are not counted by this bulk formula.

For field theory, $N_q$ is understood per spatial point and each constraint is a local relation. Spatial derivatives inside the constraint algebra do not change the algebraic count, but boundary conditions can convert a would-be gauge transformation into a physical asymptotic symmetry. Thus “two graviton polarizations” is a bulk statement, not a claim that every gravitational boundary charge vanishes.

General relativity in two equivalent bookkeeping conventions

In ADM variables the spatial metric $\gamma_{ij}$ has six components. If lapse $N$ and shift $N^i$ are treated as Lagrange multipliers rather than canonical variables, the Hamiltonian constraint and three momentum constraints are four first-class constraints. Therefore

$$N_{\mathrm{dof}}=6-4=2.$$

The extended phase-space convention gives the same answer. Include $N$ and $N^i$ among the ten configuration variables. Their four vanishing momenta are primary first-class constraints, and preservation produces the four Hamiltonian and momentum constraints. Then $N_q=10$ and $N_{\mathrm{FC}}=8$, again giving two. The canonical decomposition and the role of lapse, shift, and the secondary constraints are explicit in Arnowitt, Deser, and Misner 1962/2008, §§ 3.4 and 4.2–4.6.

This agreement is an independent normalization check on the formula. Mixing conventions—counting lapse and shift in $N_q$ while omitting their primary constraints—would incorrectly produce six modes. The same warning applies in modified gravity when auxiliary fields are introduced: either keep the full extended constraint chain or eliminate auxiliaries before counting, but never combine the two descriptions halfway.

A second-class check: the massive vector

For a Proca field $A_\mu$ in four-dimensional Minkowski space, use $\mathcal L=-F_{\mu\nu}F^{\mu\nu}/4-m^2A_\mu A^\mu/2$ with $F_{\mu\nu}=\partial_\mu A_\nu-\partial_\nu A_\mu$ and $N_q=4$. The spatial momenta are $\pi^i=\dot A_i-\partial_iA_0$. Because $\dot A_0$ is absent, $\pi^0\approx0$ is primary. Preservation yields

$$\mathcal C=\partial_i\pi^i+m^2A_0\approx0.$$

For $m\ne0$, $\{\pi^0(x),\mathcal C(y)\}=-m^2\delta^3(x-y)$, so the pair is second class. The formula gives

$$N_{\mathrm{dof}}=4-\frac12(2)=3,$$

the three polarizations of a massive spin-1 field. In the singular limit $m\to0$, the bracket matrix loses rank and the pair becomes the first-class Maxwell chain. The physical count drops to two. This is not a continuous constraint algebra even though the Lagrangian coefficients vary continuously, and it illustrates why a decoupling or general-relativistic limit must be checked at the level of constraints.

Backgrounds, branches, and the scope of the result

The Hessian rank and constraint matrix are evaluated on a specified region of phase space. A theory may be degenerate on a symmetric background but nondegenerate nearby, or a secondary constraint may disappear on another branch. Such rank changes can signal an accidental gauge symmetry, a strongly coupled mode whose quadratic kinetic term vanished, or an inconsistent truncation. The count therefore belongs to a declared action, matter coupling, background, branch, and boundary-value problem.

Constraint counting does not decide whether the surviving modes have positive energy or a well-posed evolution; those are the subjects of the next pages. Nor does this theory-side calculation import empirical likelihoods or mutable numerical bounds from Tests of Gravity. Stable low-energy quantum general relativity belongs to Quantum Gravity, while generic quantum-field-theory derivations receive only the bounded bridge required for this canonical count.

Common pitfalls

Counting gauge conditions as new physical constraints. A gauge condition pairs with a first-class constraint to select one representative of each orbit. It does not remove an additional physical degree of freedom beyond the two phase-space directions already attributed to that first-class constraint.

Using only the quadratic Hessian. Hessian degeneracy identifies primary candidates, not the completed constraint algebra. Secondary constraints and the rank of their brackets are essential, especially in nonlinear gravitational theories.

Assuming a count is branch independent. A count obtained on Minkowski or FLRW space need not survive on a compact-object branch. Recompute the rank wherever background fields or coupling matrices change.

Exercises

A constrained model has $N_q=7$, three independent first-class constraints, and four independent second-class constraints. Determine its number of physical degrees of freedom and the dimension of its reduced phase space.

Solution

The reduced phase-space dimension is $2(7)-2(3)-4=4$. Dividing by two gives $N_{\mathrm{dof}}=2$. The four second-class constraints must occur as an even-dimensional invertible bracket block; treating them as four gauge generators would incorrectly remove four additional phase-space directions.

Repeat the ADM count after promoting lapse and shift from multipliers to canonical variables.

Solution

The extended configuration space has ten variables. The four momenta conjugate to lapse and shift are primary first-class constraints, and their preservation gives the Hamiltonian and three momentum constraints. Thus there are eight first-class constraints and no second-class constraints, so $10-8=2$ physical degrees of freedom, agreeing with the reduced six-variable convention.

References

• Arnowitt, Richard, Stanley Deser, and Charles W. Misner. 1962/2008. “The Dynamics of General Relativity.” In Gravitation: An Introduction to Current Research, edited by Louis Witten, 227–265. arXiv:gr-qc/0405109.
• Date, Ghanashyam. 2010. “Lectures on Constrained Systems.” arXiv:1010.2062.