I've been reading through a derivation of the LSZ reduction formula (http://www2.ph.ed.ac.uk/~egardi/MQFT_2013/, lecture 2, pages 2-3) and I'm slightly confused about the arguments made about the assumptions:

$$ \begin{aligned} \langle\Omega\vert\phi(x)\vert\Omega\rangle &=0\\ \langle\mathbf{k}\vert\phi(x)\vert\Omega\rangle &=e^{ik\cdot x} \end{aligned} $$

For both assumptions the author first relates $\phi(x)$ to $\phi(0)$ by using the 4-momentum operator $P^{\mu}$, i.e. $$ \phi(x)=e^{iP\cdot x}\phi(0)e^{-iP\cdot x} $$ such that, in the case of the first assumption, one has $$ \langle\Omega\vert\phi(x)\vert\Omega\rangle =\langle\Omega\vert e^{iP\cdot x}\phi(0)e^{-iP\cdot x}\vert\Omega\rangle =\langle\Omega\vert\phi(0)\vert\Omega\rangle $$ where we have used that the vacuum state satisfies $P^{\mu}\lvert\Omega\rangle =0$, such that $e^{-iP\cdot x}\vert\Omega\rangle = \vert\Omega\rangle$.

What I don't understand is, why do we need to relate $\langle\Omega\vert\phi(x)\vert\Omega\rangle$ to $\langle\Omega\vert\phi(0)\vert\Omega\rangle$ in the first place? Both $\langle\Omega\vert\phi(x)\vert\Omega\rangle$ and $\langle\Omega\vert\phi(0)\vert\Omega\rangle$ are Lorentz invariant.

Is it simply because, by showing that for any $x^{\mu}$, $\langle\Omega\vert\phi(x)\vert\Omega\rangle$ is equal to the Lorentz invariant number, $v\equiv\langle\Omega\vert\phi(0)\vert\Omega\rangle$ (in principle $\langle\Omega\vert\phi(x)\vert\Omega\rangle$ could have a different value for each spacetime point $x^{\mu}$), we can then simply shift the field $\phi(x)\rightarrow \phi(x)-v$, such that the condition $\langle\Omega\vert\phi(x)\vert\Omega\rangle=0$ is satisfied? (If this is the case, then I'm guessing the argument is similar for the second condition.)


1 Answer 1


The LSZ formula is based on the following assumptions:

  1. There exists a vector $|\Omega\rangle$ that satisfies $P^\mu|\Omega\rangle=J^{\mu\nu}|\Omega\rangle=0$.

  2. The field transforms according to a certain representation of the Poincaré Group, that is, it satisfies $$ U(a,\Lambda)\phi(x)U(a,\Lambda)^\dagger=D(\Lambda)\phi(\Lambda x+a) $$ where $a\in\mathbb R^4$ and $\Lambda\in SO(1,d)^+$, and $$ U(a,\Lambda)\equiv\mathrm e^{-iP_\mu a^\mu}\mathrm e^{-i\omega_{\mu\nu}J^{\mu\nu}} $$

  3. There exists a certain vector $|\boldsymbol p,\sigma\rangle$ that satisfies $P^\mu|\boldsymbol p,\sigma\rangle=p^\mu|\boldsymbol p,\sigma\rangle$ such that $m^2\equiv p^2$ is an isolated eigenvalue of $P^2$.

  4. The field $\phi(x)$ satisfies $\langle \Omega|\phi(x)|\Omega\rangle=0$.

  5. The field $\phi(x)$ satisfies $\langle \Omega|\phi(x)|\boldsymbol p,\sigma\rangle \neq 0$.

  6. Some other assumptions that are irrelevant for this post (e.g., if the system has a well-defined notion of charge conjugation, then $\phi(x)$ has to commute with $\mathscr C$, and similarly for other internal symmetries).

If $(2)$ is satisfied, and $D(\Lambda)$ is a non-trivial representation of the Lorentz group, then $(4)$ is satisfied automatically; i.e., one need not impose this assumption as a separate condition. Therefore, in this answer we will restrict ourselves to trivial representations of the LG, that is, the scalar representation, where $\phi(x)$ is a scalar field.

In the case of scalar fields, $\langle \Omega|\phi(x)|\Omega\rangle$ is Lorentz invariant regardless of whether it vanishes or not. But we do need to make sure it vanishes, because $(4)$ is a necessary condition for the LSZ formula. Therefore, in order to make sure it vanishes, we note the following: as discussed in the OP, this number satisfies $$ \langle \Omega|\phi(x)|\Omega\rangle=\langle \Omega|\phi(0)|\Omega\rangle $$

Therefore, if for some reason $\langle \Omega|\phi(x)|\Omega\rangle$ is non-zero, we redefine the field $\phi(x)$ through $$ \phi(x)\to\phi(x)-\langle \Omega|\phi(0)|\Omega\rangle $$ which doesn't spoil any of the conditions $1,2,3,5,6$ provided they were already satisfied by the original field, but it ensures that $4$ is satisfied, by construction.

As for the second condition, the argument is as follows: if we use $\langle \Omega|U(a,\Lambda)=\langle \Omega|$ and $U(a,\Lambda)^\dagger|\boldsymbol p\rangle=\mathrm e^{ipa}|\Lambda\boldsymbol p\rangle$, then we can always write $$ \begin{aligned} \langle \Omega|\phi(x)|\boldsymbol p\rangle&=\langle \Omega|\overbrace{U(x,\Lambda)U(x,\Lambda)^\dagger}^1\phi(x)\overbrace{U(x,\Lambda)U(x,\Lambda)^\dagger}^1|\boldsymbol p\rangle\\ &=\overbrace{\langle \Omega|U(x,\Lambda)}^{\langle \Omega|}\overbrace{U^\dagger(x,\Lambda)\phi(x)U(x,\Lambda)}^{\phi(0)}\overbrace{U(x,\Lambda)^\dagger|\boldsymbol p\rangle}^{\mathrm e^{ipx}|\Lambda\boldsymbol p\rangle}\\ &=\langle\Omega|\phi(0)|\Lambda\boldsymbol p\rangle\mathrm e^{ipx} \end{aligned} $$

If we now set $x=0$, we see that this implies that $$ \langle\Omega|\phi(0)|\boldsymbol p\rangle=\langle\Omega|\phi(0)|\Lambda\boldsymbol p\rangle $$ i.e., the matrix element $\langle\Omega|\phi(0)|\boldsymbol p\rangle$ is a scalar; but the only scalar function of $\boldsymbol p$ is $p^2=m^2$, and therefore this matrix element is just a constant, independent of $\boldsymbol p$: $$ \langle \Omega|\phi(x)|\boldsymbol p\rangle=c\, \mathrm e^{ipx} $$

Finally, if, as in $(5)$, we assume that $\langle \Omega|\phi(x)|\boldsymbol p\rangle \neq 0$, then $c\neq 0$ and we can always redefine $\phi(x)$ so that $c=1$; and, as again, this doesn't spoil any of the conditions $1,2,3,4,6$.

  • $\begingroup$ Thanks for the detailed answer. So is the point that one can show that $\langle\Omega\vert\phi(x)\vert\Omega\rangle$, whose value could in principle be different at each spacetime point $x^{\mu}$, is actual equal to the constant $\langle\Omega\vert\phi(0)\vert\Omega\rangle$ $\forall\;x^{\mu}$, and as such we can simply shift the field at each spacetime point to $\phi(x)-\langle\Omega\vert\phi(0)\vert\Omega\rangle$?! $\endgroup$
    – user35305
    Commented Feb 13, 2017 at 14:20
  • $\begingroup$ @user35305 yes, that is correct. Note also that there is nothing special about $x=0$; we could also shift the field to $\phi(x)-\langle\Omega|\phi(\infty)|\Omega\rangle$ or any other point, because $\langle\Omega|\phi(x)|\Omega\rangle$ is independent of $x$. Therefore, we can set $x$ to any other convenient value. $\endgroup$ Commented Feb 13, 2017 at 14:29
  • $\begingroup$ I'm guessing this is the reason why one translates $\langle\Omega\vert\phi(x)\vert\mathbf{p}\rangle$ and $\langle\Omega\vert\phi(0)\vert\mathbf{p}\rangle$ as well?! Like you put in your answer, by carrying out this procedure one can show that $\langle\Omega\vert\phi(x)\vert\mathbf{p}\rangle$ is a Lorentz scalar and so we can again use this information to redefine $\phi(x)$ to ensure that $\langle\Omega\vert\phi(x)\vert\mathbf{p}\rangle =e^{ip\cdot x}$. $\endgroup$
    – user35305
    Commented Feb 13, 2017 at 14:40

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.