I'm confused about how a contact term vanishes when proving the Ward identity, i.e. the spot immediately following equation 5.52 in Weigand's notes. Spelling out everything concretely, we consider a process with an external photon and some external fermions, giving an $S$-matrix element $$\langle f | i \rangle \sim \xi^\mu \int dx dx_1\ldots \, \partial_x^2 \not\partial_1 \ldots \langle 0 | T A_\mu(x) \psi(x_1)\ldots | 0 \rangle $$ and apply the Schwinger-Dyson equations to turn this into $$\langle f | i \rangle \sim \xi^\mu \int dx dx_1 \ldots \, \not\partial_1 \ldots \langle 0 | T j_\mu(x) \psi(x_1) \ldots | 0 \rangle $$ while throwing away a contact term, as explained here. Now we set the photon polarization $\xi^\mu = k^\mu$ and integrate by parts for $$\langle f | i \rangle \sim \int dx dx_1 \ldots\, \not\partial_1 \partial^\mu \ldots \langle 0 | T j_\mu(x) \psi(x_1) \ldots | 0 \rangle.$$ We now apply the Ward-Takahashi identity, generating more contact terms which are supposed to be zero; however, I don't understand why they are. Directly applying Ward-Takahashi gives a contact term for every external fermion, one of which has the form $$\langle f | i \rangle \sim \int dx dx_1 \ldots \, \not\partial_1 \langle 0 | T e\psi(x_1) \ldots | 0 \rangle \delta(x - x_1) \propto \int dx_1 \ldots \, \not\partial_1 \ldots \langle 0 | T \psi(x_1) \ldots| 0 \rangle.$$ Unlike the previous case, this seems to have precisely the right kind of pole structure; it's exactly what you would get if you applied LSZ.

What am I missing here? Why doesn't this contact term contribute?

  • $\begingroup$ Isn't $\langle f\vert i\rangle$ still supposed to be a connected S-matrix element so the reasoning is still the same? $\endgroup$
    – ACuriousMind
    Dec 9, 2017 at 18:12
  • 1
    $\begingroup$ @ACuriousMind Could you spell out the reasoning in a little more detail? $\endgroup$
    – knzhou
    Dec 9, 2017 at 18:17
  • $\begingroup$ @ACuriousMind In particular I don't see why this is disconnected. If I had to translate this contact term in a diagram, it would look like 'incoming fermion absorbs the incoming photon, and then <rest of diagram>, and the outgoing fermion leaves' which is not necessarily disconnected. $\endgroup$
    – knzhou
    Dec 9, 2017 at 18:31
  • $\begingroup$ The $\not\!\partial\not\partial\langle\bar\psi\psi\rangle$ term does not have the correct pole structure. As per Källén–Lehmann, $\langle\bar\psi\psi\rangle\sim \frac{1}{\not p-m_e}$. OTOH, the expected pole structure is $\frac{1}{\not p-m_e}\frac{1}{\not k-m_e}\frac{1}{k_\gamma^2}$. $\endgroup$ Dec 9, 2017 at 20:19
  • $\begingroup$ @AccidentalFourierTransform Ah, I'm regretting not learning LSZ better. I thought the LSZ reduction formula for two fermions really did look like $\not\partial_1 \not \partial_2 \langle \bar{\psi} \psi \rangle$. What's the key difference between what I wrote and the actual formula? $\endgroup$
    – knzhou
    Dec 9, 2017 at 22:27

1 Answer 1


First off, the following reasons are not correct.

  • The S-matrix contribution from such a contact term can be connected. Diagrammatically, we have contracted an external photon and external fermion together. It is entirely possible for this to be part of a connected diagram.
  • The number of fields is not incorrect. There are exactly as many fermion fields as we started with, and exactly as many $\not\partial$ factors.

The real issue is that the location of the pole has been shifted by the contact term. To see this, we restore the exponential factors for $$\langle f | i \rangle \sim \int dx dx_1 \ldots \, e^{ikx} e^{ik_1 x_1} \ldots \not\partial_1 \langle 0 | T e\psi(x_1) \ldots | 0 \rangle \delta(x - x_1).$$ Evaluating the delta function, we find $$\langle f | i \rangle \sim \int dx_1 \ldots \, e^{i(k+k_1) x_1} \ldots \not\partial_1 \langle 0 | T e\psi(x_1) \ldots | 0 \rangle.$$ This is exactly the LSZ reduction formula for a situation with an external fermion with momentum $k+k_1$. The issue is that $k + k_1$ is never on-shell since we required $k_1$ to be on-shell and $k^2 = 0$, so we don't get a $1/(k+k_1)$ pole from the correlation function. In other words, while the correlation function does have the right number of poles, it doesn't have them in the right places.


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.