So we have gamma matrices that satisfy the spacetime algebra relations, $\{\gamma^\mu, \gamma^\nu\} = 2 \eta^{\mu\nu}$. We know that if we set $\sigma^{\mu\nu} = \frac{1}{4}[\gamma^\mu, \gamma^\nu]$ then the $\sigma$ matrices form a basis for a representation of the Lorentz Lie algebra.

Now we want to construct a Lorentz invariant inner product on the spinor space. An inner product on a finite-dimensional complex vector space always takes the form

$$ \tag{1} \langle u, v \rangle = v^\dagger H u $$

where $H$ is a Hermitian matrix. How can we construct $H$ in general without "guessing"?

Clearly such an $H$ must satisfy

$$\tag 2 S^\dagger H S = H$$

where $S$ is an arbitrary Lorentz transformation acting on spinors, and taking the derivative at the identity yields the equivalent condition

$$\tag{3} \sigma^\dagger H + H \sigma = 0$$

where $\sigma$ is an arbitrary linear combination of the $\sigma^{\mu\nu}$'s.

The usual approach I see is to choose a representation of the gamma matrices with $\gamma^0$ Hermitian and $\gamma^i$ anti-Hermitian for $i = 1, 2, 3$. One can easily check in that case that (3) is satisfied when we set $H = \gamma^0$. This gives the Lorentz-invariant inner product

$$\tag{4} \langle u, v \rangle = v^\dagger \gamma^0 u$$

which we can also write as

$$\tag{5} \langle u, v \rangle = \overline{v} u$$

with $\overline{v} = v^\dagger \gamma^0$. This is standard but it seems to require guessing that we should make $\gamma^0$ Hermitian and $\gamma^i$ anti-Hermitian. Moreover, I feel that this presentation is misleading because it appears to single out a privileged direction in space-time, namely the time direction in a fixed frame of reference.


  1. Is it the case that $H$ always has to be a scalar multiple of $\gamma^0$ in order for (3) to hold? If so, why?
  2. If not, how can we construct an appropriate $H$ given an arbitrary representation $\gamma^\mu$ of the spacetime algebra?
  3. Can this be done without relying on the fact that $\eta$ is diagonal? If so, how? If not, why not?

If we choose the signature of the metric $\eta$ to be $(1,-1,-1,-1)$ and choose the gamma matrices $\gamma^{\mu}$ to be unitary (as they can be because they form a representation of a finite group), then it would follow from the commutation relations $\{\gamma^{\mu},\gamma^{\nu}\}=2\eta^{\mu\nu}$ that $\gamma^{0}$ would be Hermitian and $\gamma^{i}\:,i=1,2,3$ would be antihermitian. In such a case we can take $H$ to be $\gamma^0$ or $\gamma^1\gamma^2\gamma^3$. On the other hand, if the signature of the metric $\eta$ is chosen to be $(-1,1,1,1)$ (and again gamma matrices are chosen to be unitary) then $\gamma^{0}$ would be antiheritian and $\gamma^i$ would be hermitian. In this case we can take $H$ to be either $i\gamma^0$ or $i\gamma^1\gamma^2\gamma^3$.

I think the source of nonuniqueness in the choice of $H$ is the fact that the space $\mathbb{C}^4$ is not an irreducible representation of the spin group (though it is an irreducible representation of the Clifford algebra). The two irreducible representation are eigenspaces of the matrix $\gamma_5$. This seems to be the reason that the two choices of $H$ above are related as $H_1=\text{const.}\gamma_5H_2$; however, I am not fully sure about this.

About question 3, I think there is no loss of generality in choosing the matrix $\eta$ to be diagonal as one can always find a basis in which this is true.


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.