Dirac and Lagrangian reductions in the canonical approach to the first-order form of the Einstein-Hilbert action (Q2368900)

The Einstein equation in vacuo, \(\text{Ric}_{ij} = 0\) (where Ric denotes the Ricci curvature tensor \(\text{Ric}_{ij} = R^{k}_{ijk}\) of a spacetime metric \(g\)), is the Euler-Lagrange equation of the Einstein-Hilbert variation pinciple \(\delta\int R(g)\,dv_g = 0\) where \(R(g) = \text{trace } \text{Ric}(g) = g^{ij} \text{Ric}_{ij}\) (scalar curvature) and \(dv_g = |\det g|^{1/2}dx\) (volume element of \(g\)). It is astonishing and has been observed in different contexts that both the Einstein-Hilbert Lagrangian \(L = R(g)|\det g|^{1/2}\) and its Euler-Lagrange equation are of the same order in \(g\) (order 2) though the Euler-Lagrange equation involves derivatives of the Lagrangian. The explanation is that the second order term of \(L\) is in divergence form and hence its integral vanishes if the variation is zero outside a compact subset. Omitting this so called ``surface term'' one obtains a first order Lagrangian with the same Euler-Lagrange equation. One of the ideas to quantize gravity is to pass from the Lagrangian to a Hamiltonian formalism and to use canonical quantization (replacing Poisson brackets by commutators of linear operators on a Hilbert space). In the present paper, only this classical transition from Lagrange to Hamilton is discussed. One needs, however, a Lagrangian containing only first order derivatives of the dynamical variables. There are two ways making the original Einstein-Hilbert Lagrangian \(L(g)\) first order: (1) cutting off the second order divergence term from \(L\) as explained above; the remaining Lagrangian is first order in \(g = (g_{ij})\) but no longer invariant under arbitrary coordinate changes, (2) treating the metric \(g\) and the affine connection (Christoffel symbols) \(\Gamma = (\Gamma_{ij}^{k})\) as independent variables, considering \(L\) as a function in these variables and their first derivatives and using the relation between \(g\) and \(\Gamma\) at the end. This method preserves the full coordinate invariance but blows up the number of variables. Passing from Lagrange to Hamilton, we first distinguish time from space derivatives in the Lagrangian. Now \(L = L(q,v)\) where \(q\) are the dynamical variables (\(g\) or \(g,\Gamma\)) and their space derivatives, and \(v = \dot q\). Next put \(p = L_v(q,v)\). If this equation can be solved for \(v\) yielding \(v = v(p,q)\), we may pass to the Hamiltonian \(H(p,q) = pv - L(q,v)\) where the variable \(v\) is replaced by the function \(v(p,q)\). This is possible iff the partial Hessian \(L_{vv}\) is non-degenerate. But in the cases (1) and (2) at hand it fails, and it is particularly bad for the Lagrangian of (2): part of the \(v\)'s are the time derivatives of \(\Gamma\) which occur only linearly! There is a more general method investigated first by Dirac which applies here: One considers \(H(p,q,v) = pv - L(q,v)\) and uses the equation \(\Phi := p-L_q = 0\) as a constraint defining a submanifold of the \((p,q,v)\)-space on which \(H\) must be restricted. Thus the variational principle \(0 =\delta\int L = \delta\int(pv-H)\) holds under the constraint \(\Phi = 0\). Using Lagrange multipliers, \(H\) has to be replaced with \(\hat H = H + \lambda\Phi\). The Lagrange multipliers \(\lambda\) are treated as extra velocities \(v\) and the constraints \(\Phi\) as their conjugate momenta. According to Dirac, there are various kinds of constraints which have to be treated differently in several steps. The programme is carried out in two ways for a simpler model, a 2-dimensional modification of the Lagrangian in (2): In Section 3, the Dirac analysis is done with the full set of \(g\) and \(\Gamma\) variables while in Section 4 first the number of variables is reduced by inserting those relations between \(g\) and \(\Gamma\) which do not contain time derivatives (``Lagrange reduction''). Surprisingly and unfortunately, the two results are different. The possible reasons are discussed in the ``Conclusion'' section.