Dirac Quantization and Physical Hilbert Space

Imposing quantum constraints is not merely solving differential equations. A physical state space needs a positive inner product, observables that preserve the constraints, and a controlled treatment of gauge volume. Dirac quantization constructs these objects after quantization; reduced quantization removes gauge directions first. They agree in favorable models, but singular orbit spaces, anomalies, and global gauge choices can make the comparison nontrivial.

Required background. Variational and Symplectic Geometry supplies first-class constraints, gauge orbits, and reduced phase space.

Helpful background. Flat-Spacetime Quantum Field Theory supplies operator quantization; Einstein Equations supplies the gravitational constraint limit; Quantum Information supplies Hilbert-space and positivity concepts; Formulations of General Relativity supplies canonical gravity variables; and Quantum Gravity supplies the programme-wide evidence distinctions.

Kinematical and physical state spaces

Start from a kinematical Hilbert space $\mathcal H_{\rm kin}$ carrying operators $\widehat C_a$ for first-class constraints. Dirac’s prescription seeks

$$\widehat C_a\lvert\Psi\rangle_{\rm phys}=0, \qquad [\widehat O,\widehat C_a]\lvert\Psi\rangle_{\rm phys}=0$$

for physical states and observables. The second equation is stated weakly because two representatives that differ by a constraint act identically on physical states. In gravity, the labels $a$ include spatial points and distinguish momentum from Hamiltonian constraints.

Zero may lie in the continuous spectrum of a constraint. Then its solutions are distributions rather than normalizable vectors in $\mathcal H_{\rm kin}$. Simply restricting the kinematical inner product gives infinity or zero; it does not define $\mathcal H_{\rm phys}$. One instead uses a rigged Hilbert space $\Phi\subset\mathcal H_{\rm kin}\subset\Phi^*$ and a map from test vectors into distributional solutions. The general canonical setting and its observables are organized in Isham 1992, §§3.2–3.4.

Group averaging as a rigging map

Suppose a unimodular gauge group $G$ is represented unitarily by $U(g)$. Formally, group averaging defines

$$\eta(\psi)[\phi]=\int_G dg\,\langle\psi\rvert U(g)\lvert\phi\rangle_{\rm kin}, \qquad \langle\eta(\psi),\eta(\phi)\rangle_{\rm phys}=\eta(\phi)[\psi].$$

For compact $G$, normalized Haar measure makes this the orthogonal projector onto invariant states. For noncompact groups, the integral is generally distributional and its convergence depends on the test space. Non-unimodular groups require modular factors. The method is therefore a prescription with explicit hypotheses, not a universal instruction to “divide by the gauge group.” Giulini and Marolf state the refined algebraic quantization conditions and uniqueness result in Giulini and Marolf 1999, §§2–4.

Positivity is checked after averaging: $\eta(\psi)[\psi]$ must be nonnegative, null vectors must be quotiented out, and the completion gives $\mathcal H_{\rm phys}$. A candidate observable must map the test space to itself and commute appropriately with the group action so that it descends to the quotient. These domain requirements are as important as the formal commutator.

Reduced quantization and its possible mismatch

Reduced quantization solves the classical constraints and quotients gauge orbits first, producing a reduced symplectic space $\Gamma_{\rm red}$. It then quantizes the remaining canonical variables. Locally this can make positivity and evolution transparent. Globally, however, $\Gamma_{\rm red}$ may be singular where stabilizer groups change, and a gauge condition may intersect an orbit more than once or not at all. Quantization need not commute with this singular reduction.

A finite-dimensional system with a free compact group action often gives the same answer by both routes. Gravity is harder: the hypersurface-deformation algebra has structure functions rather than fixed structure constants, and no accepted regulated representation of the full constraint algebra with its physical inner product is known. An anomaly

$$[\widehat C_a,\widehat C_b] =i\widehat f_{ab}{}^c\widehat C_c+\widehat A_{ab}$$

is fatal unless $\widehat A_{ab}$ vanishes on the declared domain or is removed by a justified redefinition. Otherwise applying the commutator to a purported physical state imposes an extra condition unrelated to the classical gauge system.

Independent checks and limitations

Three checks belong together. First, compare the dimension of the averaged state space with classical degree-of-freedom counting. Second, verify positivity and the adjoint relations of physical observables. Third, compute the quantum constraint commutator rather than assuming classical closure survives. Marolf’s review shows how group averaging implements these checks in constrained systems while also documenting convergence and superselection issues Marolf 2000, §§2–4.

The method does not choose a Hamiltonian clock, prove semiclassical general relativity, or solve the regularized gravitational constraints. String and holographic realizations, and semiclassical black-hole thermodynamics, belong to separate treatments. Here the narrow result is a criterion for when a constraint solution space becomes a genuine quantum state space.

Exercises

Let $\mathcal H_{\rm kin}=L^2(S^1,d\theta/2\pi)\otimes\mathcal K$ and impose $\widehat C=-i\partial_\theta$. Evaluate the $U(1)$ group average of $\psi(\theta)=\sum_n e^{in\theta}\psi_n$ and identify the physical inner product.

Solution

The unitary action is $(U(\alpha)\psi)(\theta)=\psi(\theta+\alpha)$. Averaging with normalized Haar measure gives $P\psi=(2\pi)^{-1}\int_0^{2\pi}d\alpha\,U(\alpha)\psi=\psi_0$, because every Fourier mode with $n\ne0$ integrates to zero. Thus the physical space is isomorphic to $\mathcal K$, and $\langle P\psi,P\phi\rangle_{\rm phys}=\langle\psi_0,\phi_0\rangle_{\mathcal K}$. Positivity is immediate, while the discarded Fourier labels are gauge rather than physical excitations. The example works cleanly because the group is compact and the action is represented unitarily; replacing $U(1)$ by a noncompact translation group makes the average distributional.

References

• Giulini, Domenico, and Donald Marolf. “On the Generality of Refined Algebraic Quantization.” Classical and Quantum Gravity 16 (1999): 2479–2488. DOI.
• Isham, Chris J. “Canonical Quantum Gravity and the Problem of Time.” In Integrable Systems, Quantum Groups, and Quantum Field Theories, 1992. arXiv:gr-qc/9210011.
• Marolf, Donald. “Group Averaging and Refined Algebraic Quantization: Where Are We Now?” Conference review, 2000. arXiv:gr-qc/0011112.