Using the bare perturbation theory, for a $\lambda\phi^4-$theory in $d$-dimensions, the regularized self-energy turns out to be $$\Sigma=-\frac{\lambda_0 m_0^2}{16\pi^2\epsilon}+\text{finite}\tag{1}$$ where $\epsilon=4-d$. This contribution modifies the pole of the propagator from $$m_0^2\to m^2= m_0^2+\Sigma=m_0^2\Big(1-\frac{\lambda_0}{16\pi^2\epsilon}\Big)+\text{finite}\tag{2}$$ where $m^2$ is the physical mass. Why is it that the finite term often neglected and $m^2$ is just defined as $$m^2=m_0^2\Big(1-\frac{\lambda_0}{16\pi^2\epsilon}\Big)?$$

  • $\begingroup$ I recognize that this doesn't answer your question, but life is so much better if you use renormalized perturbation theory. In that case the infinities cancel in such a more natural way, and one can also think about the finite pieces of the counter terms much more easily, too; see, e.g., Sterman p 285-287. $\endgroup$ Commented May 28, 2017 at 18:46
  • $\begingroup$ Is it not for the same reason you neglect the higher order terms in the naive Kallen-Lehmann expansion -- because they're all branch cuts not poles? $\endgroup$
    – gautampk
    Commented May 28, 2017 at 19:12

2 Answers 2


Why is it that the finite term often neglected

The finite term is NOT neglected. Rather, it could be absorbed by a counter term. If your text book just says "neglecting finite term", you should throw the book away immediately and ask for a refund.

To cancel or not to cancel the finite part by the counter term makes the all the difference between modified minimal subtraction ($\bar{MS}$) and minimal subtraction ($MS$) schemes. It's just a human convention that has no physics impact.

When it comes to a counter term, below is the rule of thumb,

  • It has to be local, which means no additional momentum dependence other than that prescribed by the original Lagrangian term.
  • It has to take care of all the divergences, which means the divergent portion is fixed and the finite/non-divergent portion could be set to any value you prefer.

This is the philosophy behind renormalization: one has a Lagrangian $\mathcal{L}(g_i)$ where $g_i$, $i=1,\ldots,N$ are some couplings (let me think of the mass as just another coupling). And then a machinery that takes as an input $\mathcal{L}(g_i)$ and outputs some observables $\Gamma^{(n)}(g_i;p_j)$, which will be typically correlation functions or scattering amplitudes ($p_j$ being the external momenta). This machinery consists in the computation of Feynman diagrams.

The experiments are able to fix the $\Gamma$'s for some configuration of the external momenta, for example one could say $$ \Gamma^{(n)}(g_i;p_j)|_{p_j\to \mu} = \tilde{g}_n\,,\quad n=1,\ldots,N \,.$$ Where $\tilde{g}_n$ is just a number given by experiments. So what we want to do is to tweak the $g_i$'s so that we get the desired result. As we know, after we regularize the theory, the answer is something of the form $$ \tilde{g}_n = f_n^{(0)}(g_i) + \frac{f_n^{(1)}(g_i)}{\varepsilon} + \frac{f_n^{(2)}(g_i)}{\varepsilon^2} + \cdots\,, $$ where the $f_n^{(l)}$ are some finite functions of the couplings. For any $\varepsilon>0$ this fixes the $N$ couplings $g_i$ in terms of the $N$ observables $\tilde{g}_n$ and we could just leave it at that. The $g_i$'s are functions $g_i(\varepsilon,\tilde{g}_n)$ and they can be fed to the Lagrangian, which in turns produces other observables and determines them in terms of the previous $N$ experiments as follows $$ \Gamma^{(m)}(g_i(\varepsilon,\tilde{g}_n);p_j)|_{p_j\to \mu} = \tilde{g}_m(\tilde{g}_n)\,,\quad m\neq1,\ldots,N\,. $$ It is crucial that the $\varepsilon$ dependence disappears in the end, but this is ensured by the theorem of power counting in renormalization theory.

I said that we could leave it at that because this is all we need from a theory, we want it to predict infinitely many experiments starting from $N$ of them as an input. However, for all practical purposes, it's better to compute once and for all the divergent part in $g_i(\varepsilon,\tilde{g}_n)$ and express it as $$ g_i(\varepsilon,\tilde{g}_n) = h_i^{(0)}(\tilde{g}_n) + \frac{h_i^{(1)}(g_i)}{\varepsilon} + \frac{h_i^{(2)}(g_i)}{\varepsilon^2} + \cdots\,. $$ We know how the divergent part works, so let's just focus on the finite piece. $h_i^{(0)}(\tilde{g}_n)$ is what we call the renormalized coupling. But this is totally arbitrary, we could also call $h_i^{(0)}(\tilde{g}_n) - 2\pi$ the renormalized coupling, as long as we specify that the divergent part is $2\pi + \mathrm{poles}$.

I'm sorry if I decided not to address your specific questions but rather give you a more general answer.


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.