No label defined (Q3976148)

[For the entire collection see Zbl 0728.00003.] It is well known that the vacuum Einstein equations (E) \(G:=\hbox{Ric}- {1\over 2}Rg=0\) for a Lorentzian metric \(g\) (with scalar curvature \(R\)) on a manifold \(M\) are the Euler-Lagrange equations of the Hilbert action principle \(\delta L(g)=0\) with Lagrangian \(L(g)=\int R(g)dv(g)\). But since (E) is of order 2, it should rather follow from a variational principle which contains only first derivatives in \(g\). In fact, such a Lagrangian has been given already by Einstein: The local expression for \(R\) in terms of the Christoffel symbols \(\Gamma\) is \(R=D(\Gamma)+U(\Gamma)\) where \(D\) is a first order differential operator and \(U\) a quadratic polynomial. The second term is first order in \(g\) while the first term gives a divergence whose integral vanishes. Of course, the two terms depend on coordinates (are not ``covariant'') but a more general formula holds for the case where \(\Gamma=\nabla-\nabla_ 0\) for an arbitrary background connection \(\nabla_ 0\) on \(M\) (then there is also curvature term of \(\nabla_ 0\) \(M\) (but this is not varied). The present paper puts this observation in a more general context. For any second order Lagrangian \(L\) it is shown that the Euler equations are of order 2 if \(L\) is linear in the second order derivatives with coefficients depending only on the variables, not their derivatives. In this case, the so-called Poincaré-Cartan form \(\Theta\) which characterizes the solutions of the Euler equation depends only on first derivatives instead of third ones. In the case of the Hilbert Lagrangian, \(\Theta\) minus an exact form is the Poincaré-Cartan form of the first order \(\nabla_ 0\)-dependent Lagrangian introduced above.