The laplacian of a tensor $T$ is the tensor defined by $$ \Delta T = g^{ij} \left(\nabla_{\partial_i} \nabla_{\partial_j} T - \nabla_{\nabla_{\partial_i} \partial_j} T\right)=g^{ij} \nabla^2_{ij} T. $$ I would like to calculate the following commutator $$\nabla \Delta T - \Delta \nabla T,$$ for tensors of type $(n,0)$ (i.e. input: $n$ vectors and output: a real number).
With $\Delta \nabla T$ I mean the laplacian of the $(n+1,0)$-tensor $\nabla T$. I have attempted to calculate it many times, but I struggle with taking multiple covariant derivatives and interchanging them. From the liturature I suspect the end result takes the form of a contraction with $T$ and the curvature tensor.

I would greatly appreciate if anyone can show the calculation (in normal coordinates is fine) for this or can give me a reference for this since I can't find any.


1 Answer 1


If you know how to commute covariant derivatives of an arbitrary tensor, then this is not so hard: since the metric commutes with covariant derivatives, we have $$\begin{align} \nabla_k \Delta T - \Delta \nabla_k T &= g^{ij}(\nabla_k \nabla_i \nabla_j T - \nabla_i \nabla_j \nabla_k T) \\ &= g^{ij}([\nabla_k, \nabla_i] \nabla_j T +\nabla_i\nabla_k\nabla_j T- \nabla_i \nabla_j \nabla_k T) \\ &= g^{ij}([\nabla_k, \nabla_i] \nabla_j T + \nabla_i([\nabla_k,\nabla_j] T))\tag{1}.\end{align}$$ Here I'm using the abstract index notation, so e.g. $\nabla_i \nabla_j \nabla_k T = (\nabla^3 T)(\partial_i, \partial_j, \partial_k, \ldots)$ is the covariant third derivative.

Since commutators of covariant derivatives always end up just being some contractions with the curvature, we can immediately see from $(1)$ that the formula will have the form $$[\nabla,\Delta]T = R* \nabla T + \nabla( R * T) = R * \nabla T + \nabla R * T \tag{2},$$ where the notation $S * T$ denotes an arbitrary linear combination of contractions of $S \otimes T$. Thus you should expect not only curvature terms, but also derivatives of the curvature. (That is, unless $n = 0$ - meaning $T$ is just a scalar - in which case the second term vanishes because the Hessian of a scalar is symmetric.) For many purposes $(2)$ is all you need.

To get an explicit expansion of $(1)$ in terms of the curvature tensor, we first need to write down an expression for $[\nabla_k, \nabla_j] T$.$\def\t{\otimes}$ When we take the second covariant derivative of $T = T_{i_1 \ldots i_n}dx^{i_1}\t\cdots\t dx^{i_n}$ and fully expand using the product rule, all the cross terms will vanish thanks to the antisymmetrization; so the derivative commutator also satisfies a Leibniz rule: we find $$[\nabla_k, \nabla_j] T = [\nabla_k,\nabla_j](T_{i_1 \ldots i_n})dx^{i_1}\t\cdots\t dx^{i_n}\\ + T_{i_1 \ldots i_n} ([\nabla_k,\nabla_j]dx^{i_1})\t\cdots \t dx^{i_n} \\ \vdots \\+ T_{i_1 \ldots i_n} dx^{i_1}\t\cdots \t([\nabla_k,\nabla_j]dx^{i_n}).$$ Since the curvature tensor is characterized by $R_{kj}{}^i{}_l dx^l = [\nabla_k,\nabla_j]dx^i$ and the first two derivatives of a scalar commute, this becomes $$[\nabla_k, \nabla_j] T=T_{i_1 \ldots i_n}(R_{kj}{}^{i_1}{}_ldx^l \t \cdots \t dx^{i_n} + \cdots + R_{kj}{}^{i_n}{}_l dx^{i_1} \t \cdots \t dx^l),$$ which (doing some index gymnastics) we can write as

$$[\nabla_k,\nabla_j]T_{i_1\ldots i_n} = R_{kj}{}^l{}_{i_1} T_{l i_2 \ldots i_n}+\cdots + R_{kj}{}^l{}_{i_n}T_{i_1 \ldots i_{n-1} l}.$$

Since you can now commute derivatives of an arbitrary $n$-tensor, you can do it for an $n+1$-tensor too; so if you really want to write out the full mess you can plug this in to $(1)$ for the two commutator terms and then expand everything with the product rule. I don't think there's much insight to be gained by typing it all out - it's just going to be a mess of contractions, some of them of $R$ with $\nabla T$ and some of $\nabla R$ with $T$. Perhaps try it yourself for a very small value of $n$.

  • $\begingroup$ Thank you for your answer! Since I'm not so familiar with calculations with tensors I'm gonna read this step by step and when it's all clear I will certainly accept the answer. Thanks! $\endgroup$
    – abcdef
    Oct 14, 2017 at 15:04
  • $\begingroup$ @AnthonyCarapetis Is your argument valid for any connection or is it valid only for Levi-Civita connection? What may change if the connection has torsion? $\endgroup$
    – user44690
    Jun 25, 2020 at 6:10
  • 2
    $\begingroup$ @user44690: if the connection is not metric-compatible, this argument doesn't work: you'll have extra terms involving $\nabla g.$ For a metric-compatible connection with torsion, I think it goes through unchanged, though I have not checked carefully. $\endgroup$ Jun 25, 2020 at 6:40

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.