6
$\begingroup$

Consider a smooth manifold $M$ together with an affine connection (or covariant derivative) $\nabla$ on the tangent bundle $TM$.

Is it possible to have an affine connection, possibly with non-zero torsion, such that $\nabla_{X,Y}^2 = \nabla_{Y,X}^2$? I.e. the second covariant derivative is symmetric/commutative, while the covariant derivative itself is not (since it might have torsion)?

How are such connections or manifolds called? If anyone can refer me to an article or reference, that would be great.

PS. Note that the connection is not necessarily curvature-free in the case that $\nabla_{X,Y}^2 = \nabla_{Y,X}^2$, but instead all the curvature is due to torsion. That is, $R(X,Y)Z = \nabla_{T(X,Y)}Z$.

$\endgroup$
3
  • $\begingroup$ Flat connection as opposed to flat torsion free connection. $\endgroup$
    – Deane Yang
    Commented Dec 24, 2023 at 21:12
  • $\begingroup$ Since the connection possibly has torsion, one still has the torsion induced curvature component. That is, one has $R(X,Y)Z = \nabla_{T(X,Y)}Z$. And so the connection is not necessarily flat, but instead all the curvature is due to torsion. $\endgroup$
    – Khaled T.
    Commented Dec 24, 2023 at 21:14
  • $\begingroup$ I don’t know much but maybe this is relevant: math.stackexchange.com/a/2174322/10584 $\endgroup$
    – Deane Yang
    Commented Dec 24, 2023 at 21:45

1 Answer 1

6
$\begingroup$

If the second covariant derivative of every vector field $Z$ is symmetric in the sense that $\nabla(\nabla Z)$ (which is a section of $TM\otimes T^*M\otimes T^*M$) is a section of the sub bundle $TM\otimes S^2(T^*M)$, then the connection $\nabla$ is torsion-free and flat.

This follows directly from Cartan's structure equations (cf. Kobayashi and Nomizu). Here is the calculation. Let $\omega = (\omega^i)$ be a coframing on an open set in $M$ and let $\theta = (\theta^i_j)$ be the connection forms of $\nabla$ relative to the coframing $\omega$. The Cartan structure equation are $$ \mathrm{d}\omega^i = -\theta^i_j\wedge\omega^j +\tfrac12 T^i_{jk}\,\omega^j\wedge\omega^k \quad\text{and}\quad \mathrm{d}\theta^i_j = -\theta^i_k\wedge\theta^k_j + \tfrac12 R^i_{jkl}\,\omega^k\wedge\omega^l, $$ where $T^i_{jk} = - T^i_{kj}$ and $R^i_{jkl} = -R^i_{jlk}$ are, respectively, the components of the torsion and curvature of $\nabla$ in the coframing $\omega$. (Note the use of the Einstein summation convention both here and below.)

If $Z$ is an arbitrary vector field in the domain of $\omega$, with $\omega$-components $z^i = \omega^i(Z)$, then there are functions $z^i_j$ and $z^i_{jk}$ that satisfy the equations $$ \mathrm{d}z^i = -\theta^i_j\,z^j + z^i_j\,\omega^j \quad\text{and}\quad \mathrm{d}z^i_j = -\theta^i_k\,z^k_j + \theta^k_j\, z^i_k + z^i_{jk}\,\omega^k. $$ Here, $z^i_j$ and $z^i_{jk}$ are the components of the first and second covariant derivatives of $Z$ relative to the coframing $\omega$.

Suppose that $z^i_{jk}=z^i_{kj}$ for every vector field $Z$. (This is the component version of the condition that $\nabla(\nabla Z)$ be a section of $TM\otimes S^2(T^*M)$.) Use the above structure equations and the symmetry assumption to compute $$ 0 = \mathrm{d}(\mathrm{d}z^i) = \tfrac12\bigl(-R^i_{jkl}z^j + T^j_{kl}z^i_j\bigr)\,\omega^k\wedge\omega^l. $$ Thus, $-R^i_{jkl}z^j + T^j_{kl}z^i_j = 0$ for every vector field $Z$. (I.e., the relation $R(X,Y)Z = \nabla_{T(X,Y)}Z$ holds for every triple of vector fields $X$, $Y$, and $Z$.) Since the value of $Z$ and its first covariant derivative can be specified arbitrarily at any point of $M$, this relation can only hold for all vector fields $Z$ if and only if $T^j_{kl}$ and $R^i_{jkl}$ vanish identically, i.e., the connection $\nabla$ is both torsion-free and flat.

$\endgroup$
2
  • $\begingroup$ Thank you for your elaborate answer. I have two more question, however. Excuse me if it might be silly, I am not so familiar with Cartan’s approach. 1) I have the idea that you have assumed metricity of the connection here. Is that correct? If so, would the situation be different for a non-metric connection? 2) Similarly, have you assumed here that the coframe defines a holonomic frame? What if the frame is not holonomic? $\endgroup$
    – Khaled T.
    Commented Dec 26, 2023 at 14:29
  • $\begingroup$ @KhaledT.: I didn't assume anything about a metric. This argument works for any affine connection. The choice of local coframing $\omega$ is for convenience (to avoid having to define and work on the full coframe bundle); it doesn't matter which coframing $\omega$ (holonomic or not holonomic) that one chooses. $\endgroup$ Commented Dec 26, 2023 at 18:12

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .