I am following the notes by Freed about the Dirac operator in section 2. I am trying to understand the Weitzenbock equality that for a curved-space spinor's Dirac operator $D\!\!\!/\ $ and the associated 'Laplacian' $\Delta = \nabla^*\nabla $ satisfy

$$D\!\!\!/\ D\!\!\!/\ - \Delta = \frac{R}{4}$$

I'm trying to reproduce this formula using different notation, but I'm having difficulty defining the Laplacian $\Delta$. I know how to define the curved space Dirac operator in terms of an orthonormal frame $e^\mu_a$ and a basis of gamma matrices $\gamma^a$ as:

$$D\!\!\!/\ = \gamma^a e^\mu_a(\partial_\mu + \omega_\mu^I T^I) =: \gamma^a e^\mu_a \nabla_\mu$$

where I'm writing the $SO(n)$ spin connection as $\omega_\mu^I T^I$ for the Lie algebra generators $T^I$ and coefficents $\omega$. However, I'm confused on how the connection Laplacian should be defined. Using the naive definition of

$$\tilde{\Delta} = (\partial_\mu + \omega_\mu^I T^I)(\partial^\mu + \omega^{\mu I} T^I) = e^a_\mu(\partial_a + \omega_a^I T^I)e^{b\mu}(\partial_b + \omega_b^I T^I) $$

does not reproduce the right identity, and actually gives a differential operator rather than a scalar for $D\!\!\!/\ D\!\!\!/\ - \Delta$.

I have a feeling that the above expression for the Laplacian doesn't make sense for spinor fields, since implicitly I'm raising the spinor index using $g^{\mu\nu}$, whereas everything should in principle be doable via tetrads. What is the correct covariant way to write the operators $\nabla, \nabla^*$ and $\Delta$?


I figured it out, and typed it out below.


1 Answer 1


First, my expression for the Laplacian above was off, and the definition of $\nabla^*$ and $\nabla^*\nabla$ is a bit more subtle than I thought. $\nabla^*$ is the adjoint of $\nabla$. In particular, some vector $X$, $\nabla^*_X$ is the adjoint of $\nabla_X = X^\mu \nabla_\mu$. This can be computed from trying to integrate $\langle w, \nabla_X v \rangle = \int_M d(vol) w_\nu X^\mu \nabla_\mu v^\nu$ by parts to get

$$\langle w, \nabla_X v \rangle = \int_M d(vol) w_\nu X^\mu \nabla_\mu v^\nu = -\int_M d(vol) v_\nu \nabla_\mu (X^\mu w^\nu) =: \langle \nabla^*_X w, v \rangle$$

so that $\nabla_X^* w^\nu = -\nabla_\mu(X^\mu w^\nu)$ in local coordinates.

Note that all of these definitions were covariant derivatives with respect to a fixed orthonormal frame field. The operator $\nabla^*\nabla$ is actually referring to $tr(\nabla^*\nabla) =: \Delta$, where the trace means given some local orthonormal basis $e^\mu_a$, we define the laplacian as

$$(\Delta w)^\mu = \sum_a(\nabla^*_{e_a}\nabla_{e_a}) = \sum_a -\nabla_\nu(e_a^\nu e_a^\rho \nabla_\rho w^\mu)$$

which we can rewrite as (using Freed's convention that $\{\gamma^a,\gamma^b\}= -2 \delta^{ab}$)

$$(\Delta w)^\mu = \frac{1}{2}\nabla_\nu(e_a^\nu \{\gamma^a,\gamma^b\} e_b^\rho \nabla_\rho w^\mu)$$

Now, we want to compare $D\!\!\!/\ D\!\!\!/\ $ with $\Delta$. If we work in Riemann normal coordinates centered at $x=0$, we note that it's possible to choose $e^\mu_a = \delta^\mu_a + O(x^2)$, so that $\omega_\mu = 0 + O(x)$. This means we can freely rewrite the above expression as

$$(\Delta w)^\mu = \frac{1}{2} e_a^\nu \{\gamma^a,\gamma^b\} \nabla_\nu(e_b^\rho \nabla_\rho w^\mu) = \frac{1}{2} \{\gamma^\nu,\gamma^\rho\} \nabla_\nu \nabla_\rho w^\mu $$

where we abuse notation and write $\mu = 0,...,d$ to represent the flat space coordinates locally. And we also rewrite $D\!\!\!/\ D\!\!\!/\ $ as

$$(D\!\!\!/\ D\!\!\!/\ w)^\mu = \gamma^\nu \gamma^\rho \nabla_\nu \nabla_\rho w^\mu$$

so that

$$D\!\!\!/\ D\!\!\!/\ - \Delta = -\frac{1}{2}[\gamma^\nu,\gamma^\rho] \nabla_\nu\nabla_\rho = -\frac{1}{2}\gamma^\nu \gamma^\rho [\nabla_{\nu},\nabla_{\rho}]$$

From here, it becomes straightforward, since $[\nabla_{\nu},\nabla_{\rho}]$ gives the curvature tensor. One tricky point is the eventual need for the identity $\gamma^i\gamma^j\gamma^k\gamma^l R_{ijkl} = 2R$ (which is explained in Freed's notes clearly). This can be shown by noting that the Riemann tensors symmetries require $i \neq j, k \neq l$. The case where $j \neq k \neq l$ are all distinct vanishes by the Bianchi identity $R_{ijkl} + R_{iklj} + R_{iljk} = 0$. The case where $j=k$ gives $\sum_j \gamma^i \gamma^j \gamma^j \gamma^l R_{ijjl} = - \gamma^i \gamma^l R_{il} = R$. the case $j=l$ similarly gives $R$, so the total sum is $2R$. Putting this all together will give the desired result after dealing appropriately with the factors of 2 from the spin connection.

  • $\begingroup$ This was a great derivation! Quick question, above you kept using $\nabla$, for different steps, my question is: is this $\nabla$ the covariant derivative or the Laplacian? Thank you $\endgroup$ Apr 10, 2020 at 16:46
  • $\begingroup$ $\nabla$ was the covariant derivative and $\Delta$ is the Laplacian, which in this case is defined as the trace of $\nabla^*\nabla$, where $\nabla^*$ is the adjoint of $\nabla$ $\endgroup$
    – Joe
    Apr 10, 2020 at 21:45

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

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