If I have a Lorentz invariant equation of motion, like Klein-Gordon equation, is the solution automatically guaranteed to be Lorentz invariant?

I ask this question because of the discussion from Mark Srednicki's Quantum Field Theory section 3 from equations (3.11) to (3.14). If I have a K-G equation, $$ \tag 1 \partial^\mu\partial_\mu\phi -m^2\phi=0,$$ we have a solution of the form $$ \tag 2\exp (i \mathbf{k} \cdot \mathbf{x} \pm i\omega t),$$ which I do not think is Lorentz invariant for solution with $i \mathbf{k} \cdot \mathbf{x} + i\omega t$ as an argument, unless we allow $k^\mu = (-\omega, \mathbf{k})$.

However, he starts constructing a Lorentz invariant solution, and comes up with $$ \tag 3 \phi(\mathbf{x},t) = \int d\tilde{k}[ a(\mathbf{k})e^{ikx} + a^*(\mathbf{k})e^{-ikx}],$$ where $kx = \mathbf{k}\cdot\mathbf{x} - \omega t$. $d\tilde{k}$ is a Lorentz invariant measure and argument of each exponents are Lorentz invariant as well.

However, he says in the beginning that $a(\mathbf{k})$ is an arbitrary function of the wave vector $\mathbf{k}$, which does not sound Lorentz invariant to me. So I am not sure how $\phi(\mathbf{x},t)$ is Lorentz invariant.

  • 3
    $\begingroup$ no. When it does not happen we talk of spontaneous symmetry breaking. Have also a look at this answer of Qmechanic. $\endgroup$
    – glS
    Feb 8, 2015 at 19:19
  • $\begingroup$ so is the solution that Srednicki constructed not a Lorentz invariant one if we allow arbitrary a(k)? $\endgroup$ Feb 8, 2015 at 19:28
  • 1
    $\begingroup$ It would help if you explicitly showed where you think this violates Lorentz covariance/invariance, because I don't see it - $a(k)$ goes to $a(\Lambda^{-1} k)$ under a Lorentz transformation, but Lorentz invariance of the measure means that you can just transform the integration region back to still only have $a(k)$ in there. $\endgroup$
    – ACuriousMind
    Feb 8, 2015 at 19:39
  • 1
    $\begingroup$ Actually, I find transforming integration region back still problematic.. Say I have $\phi(x) \to \phi(\Lambda^{-1} x) = \int d\tilde{k}[ a(\Lambda^{-1}\mathbf{ k})e^{ikx} + a^*(\Lambda^{-1} \mathbf{k})e^{-ikx}] = \int d\tilde{k}[ a(\mathbf{k})e^{i(\Lambda k)x} + a^*(\mathbf{ k})e^{-i(\Lambda k)x}] =? \phi(x)$. $\endgroup$ Feb 9, 2015 at 0:57
  • $\begingroup$ @KyleLee: The transformation of a scalar field is $\phi'(x)=\phi(\Lambda^{-1}x)$. That said, I don't know what you're doing wrong. $\endgroup$
    – Ryan Unger
    Feb 9, 2015 at 11:09

1 Answer 1


In the spirit of the original post, let $k,x$ be 4-vectors and $\mathbf{k}$, $\mathbf{x}$ the spatial components. Then a quantity of the form $$\phi(x) \propto \int dk[ a(k)e^{ikx} + a^*(k)e^{-ikx}]$$ is manifestly Lorentz invariant because it does not explicit contain any free Lorentz indices. What Srednicki does is that he performs the $k^0$ integration, resulting in $$\phi(\mathbf{x},t) = \int \frac{d\mathbf{k}}{f(\mathbf{k})}[ a(\mathbf{k})e^{i\mathbf{k}\cdot\mathbf{x}} + a^*(\mathbf{k})e^{-i\mathbf{k}\cdot\mathbf{x}}],$$ which only includes spatial components. This expression is Lorentz invariant because it is just a different form of the previous one, but it does not manifestly look Lorentz invariant which I assume what causes the confusion. For an explicit form of the function $f$ which will of course be related to the energy as it is the integral over $k^0$, see for example Peskin and Schroeder eqn (2.47).

EDIT: Some more justification:

The Klein-Gordon eqn is $$\partial^\mu\partial_\mu\phi -m^2\phi=0.$$ To solve it, we Fourier transform to momentum space and we get: $$(p^\mu p_\mu -m^2)\tilde\phi=0.$$ The general solution of this eqn is $$\tilde\phi(p)=a(p)\delta(p^\mu p_\mu -m^2),$$ which means that the general solution for the Klein-Gordon is: $$\phi(x)=\frac{1}{(2\pi)^4}\int d^4pe^{ipx}\tilde\phi(p)=\frac{1}{(2\pi)^4}\int d^4pe^{ipx}a(p)\delta(p^\mu p_\mu -m^2)$$ which is manifestly Lorentz invariant. You can then perform the $p^0$ integration as claimed above. I have ignored the complex conjugate term everywhere but should be trivial to restore...

  • $\begingroup$ What about $k^0$ in the argument of exponents ? Why are they still around ? $\endgroup$ Feb 9, 2015 at 11:42
  • $\begingroup$ @KyleLee: They get put on the mass shell: $k^0=\sqrt{\mathbf{k}^2+m^2}$. $\endgroup$
    – Ryan Unger
    Feb 9, 2015 at 11:56
  • $\begingroup$ That's true. Nice ! $\endgroup$ Feb 9, 2015 at 11:58
  • $\begingroup$ Well, one thing that im still confused about is I feel like f(k) should then be appropriately set to make the whole thing Lorentz invariant, not just the measure. I don't see how the coefficients don't play any role in determining f(k) if I integrate the first integral. $\endgroup$ Feb 9, 2015 at 12:00
  • $\begingroup$ @KyleLee: What worries me more is the first equation. What does $a(k)$ mean? Does that imply it has $k^0$ dependence? I can't recall seeing this expansion before. It's normally the $d\mathbf{k}$ one. $\endgroup$
    – Ryan Unger
    Feb 9, 2015 at 12:05

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.