Conditions for violation of Jacobi identity for covariant derivative

130 Views Asked by At

I wish to know if there is a situation when the Jacobi identity for the covariant derivative $\nabla_{\mu}$ $$\chi_{\mu\nu\rho}\equiv [\nabla_{\mu}, [\nabla_{\nu}, \nabla_{\rho}]]+[\nabla_{\rho}, [\nabla_{\mu}, \nabla_{\nu}]]+[\nabla_{\nu}, [\nabla_{\rho}, \nabla_{\mu}]] = 0$$ can be violated. Can anyone suggest an example where this may happen?

Edit: Just for the sake of adding some context. Consider a connection with torsion. Then, one will find that

$$[\nabla_{\mu}, \nabla_{\nu}]\varphi = 2T^{\lambda}_{~~~\mu\nu}\partial_{\lambda}\varphi$$ where $T^{\lambda}_{~~\mu\nu} = \Gamma^{\lambda}_{[\mu\nu]}$. Now, we may look at $$[\nabla_{\rho},[\nabla_{\mu}, \nabla_{\nu}]] = 2\nabla_{\rho}T^{\lambda}_{~~\mu\nu}\nabla_{\lambda}+2T^{\lambda}_{~~\mu\nu}T^{\tau}_{~~\rho\lambda}\nabla_{\tau}$$ which implies that $$\chi_{\mu\nu\rho} = 2\nabla_{\rho}T^{\lambda}_{~~\mu\nu}\nabla_{\lambda}+2T^{\lambda}_{~~\mu\nu}T^{\tau}_{~~\rho\lambda}\nabla_{\tau} + \text{cyclic}(\mu, \nu, \rho) $$ which will only be satisfied if torsion is zero. It seems the Jacobi identity is violated unless one works with a specific kind of torsion. Am I correct in my thinking?

2

There are 2 best solutions below

0
On BEST ANSWER

Ok, so first of all, the Jacobi identity is always true for the covariant derivative. If $\psi$ is a section of any vector bundle on which the covariant derivative acts (indices suppressed), we have $$ [\nabla_i,[\nabla_j,\nabla_k]]\psi=\nabla_i[\nabla_j,\nabla_k]\psi-[\nabla_j,\nabla_k]\nabla_i\psi \\ = \nabla_i\nabla_j\nabla_k\psi-\nabla_i\nabla_k\nabla_j\psi-\nabla_j\nabla_k\nabla_i\psi+\nabla_k\nabla_j\nabla_i\psi. $$ If we take the cyclic sum, we get $$[\nabla_i,[\nabla_j,\nabla_k]]\psi+\text{cyclic} =\color{red}{\nabla_i\nabla_j\nabla_k\psi}+\color{green}{\nabla_j\nabla_k\nabla_i\psi}+\color{purple}{\nabla_k\nabla_i\nabla_j\psi} \\ -\color{blue}{\nabla_i\nabla_k\nabla_j\psi}-\color{brown}{\nabla_j\nabla_i\nabla_k\psi}-\nabla_k\nabla_j\nabla_i\psi \\ -\color{green}{\nabla_j\nabla_k\nabla_i\psi}-\color{purple}{\nabla_k\nabla_i\nabla_j\psi}-\color{red}{\nabla_i\nabla_j\nabla_k\psi} \\ +\nabla_k\nabla_j\nabla_i\psi+\color{blue}{\nabla_i\nabla_k\nabla_j\psi}+\color{brown}{\nabla_j\nabla_i\nabla_k\psi}. $$ The identically colored pairs cancel pairwise, hence the Jacobi identity is true for the covariant derivative.


What (apparantly) stumps OP is that if the covariant derivatives are expanded explicitly, we get - combined with the Jacobi identity - a nontrivial identity. Which is btw faulty in the OP.

For a scalar field $\phi$, the Ricci identity is $$ [\nabla_i,\nabla_j]\phi=-T^k_{\ ij}\nabla_k\phi, $$ whereas for a covector field $\omega_i$, the Ricci identity is $$ [\nabla_i,\nabla_j]\omega_k=-R^l_{\ kij}\omega_l-T^l_{\ ij}\nabla_l\omega_k. $$

We thus get $$ [\nabla_i,[\nabla_j,\nabla_k]]\phi=\nabla_i([\nabla_j,\nabla_k]\phi)-[\nabla_j,\nabla_k]\nabla_i\phi=\nabla_i(-T^l_{\ jk}\nabla_l\phi)+R^l_{\ ijk}\nabla_l\phi+T^l_{\ jk}\nabla_l\nabla_i\phi \\ =(R^l_{\ ijk}-\nabla_i T^l_{\ jk}+T^p_{\ jk}T_{\ ip}^l)\nabla_l\phi, $$ hence we get the relation (the cyclic permutation is up to a constant factor same as the antisymmetrization) $$ R^l_{\ [ijk]}=\nabla_{[i}T^l_{\ jk]}-T^l_{\ [i|p|}T^p_{\ jk]}. $$

This is nothing else than the "algebraic" Bianchi identity (which is only truly algebraic if the torsion vanishes), which has been obtained as a consequence of the Jacobi identity.

4
On
  1. For what it's worth, the Jacobi identity (JI) follows immediately from the associativity of composition of operators, so any violation of the JI must somehow violate this premise. E.g. the unbounded operators could be defined on a domain where composition is not well-defined. (Another idea is to act on a non-associative Colombeau algebra, although I have not tried to work out any examples.)

  2. In case of torsion, let us mention that the second Bianchi identity (whose proof uses the JI repeatedly) acquires extra torsion terms.