I am trying to derive the Klein-Gordon equation for the case of GR using the action:

$$S\left[ {\varphi ,{g_{\mu \nu }}} \right] = \int {\sqrt g {d^4}x\left( { - {1 \over 2}{g^{\mu \nu }}{\nabla _\mu }\varphi {\nabla _\nu }\varphi - {1 \over 2}{m^2}{\varphi ^2}} \right)} \tag{1}$$

So in the Euler - Lagrange equation for the SR case which is: $${\partial _\mu }\left( {{{\partial L} \over {\partial \left( {{\partial _\mu }\varphi } \right)}}} \right) = {{\partial L} \over {\partial \varphi }}\tag{2}$$ I use the correspondence $${\partial \over {\partial \left( {{\partial _\mu }\varphi } \right)}} \to {\nabla _{{\nabla _\mu }\varphi }},{\partial \over {{\partial _\mu }\varphi }} \to {\nabla _\mu } \tag{3}$$ and so, $${\nabla _{{\nabla _\mu }\varphi }}L = - {1 \over 2}\sqrt g {g^{\mu \nu }}{\nabla _\nu }\varphi - {1 \over 2}\sqrt g {g^{\mu \nu }}{\nabla _\mu }\varphi {\nabla _{{\nabla _\mu }\varphi }}\left( {{\nabla _\nu }\varphi } \right) \tag{4}$$ and using this dubious relation that I am unable to prove (extending the SR case of partial derivatives),

$${\nabla _{{\nabla _\mu }\varphi }}\left( {{\nabla _\nu }\varphi } \right) = \delta _\nu ^\mu \tag{5}$$ I finally get the desired KG equation, $${\nabla _\mu }\left( { - \sqrt g {g^{\mu \nu }}{\nabla _\nu }\varphi } \right) = - \sqrt g {m^2}\varphi $$ However, I am very uncomfortable with my assumption 5 that I am unable to prove. Is my analysis correct? I am a 60y old doing this as retirement fun so please don't cut me down too brutally :-)


Note: technically, everywhere we've been using $\sqrt{g}$ it should be $\sqrt{|g|}$, but I'll let the former denote the latter in a slight abuse of notation like the rest of this page.

gj255 has already provided the textbook approach, in which we keep to partial derivatives. You can actually do it a different way; the same from-first-principles argument that derives Eq. (2) can be modified to obtain something that manifestly works with covariant derivatives. Write the Lagrangian density as $L=\sqrt{g}L_0$ with $L_0=\frac{1}{2}\nabla_\mu\varphi\nabla^\mu\varphi-\frac{1}{2}\varphi^2$, so $$0=\delta S=\int d^4x \sqrt{g}\left(\delta\varphi \frac{\partial L_0}{\partial\varphi}+\nabla_\mu\delta\varphi \frac{\partial L_0}{\partial\nabla_\mu\varphi}\right)$$($\nabla_\mu\varphi$ is actually just $\partial_\mu\varphi$, because $\varphi$ is a scalar field), where we have written $\delta\nabla_\mu\varphi=\nabla_\mu\delta\varphi$. But$$\int d^4x\sqrt{g}\nabla_\mu\left(\delta\varphi \frac{\partial L_0}{\partial\nabla_\mu\varphi}\right)=\int d^4x\partial_\mu\left(\sqrt{g}\delta\varphi \frac{\partial L_0}{\partial\nabla_\mu\varphi}\right)$$is a boundary term, so we can rewrite $\delta S=0$ as $$0=\int d^4 x\sqrt{g}\delta\varphi \left(\frac{\partial L_0}{\partial\varphi}-\nabla_\mu\frac{\partial L_0}{\partial\nabla_\mu\varphi}\right),$$giving us an Euler-Lagrange function more in the spirit of what you wanted, viz.$$0=\frac{\partial L_0}{\partial\varphi}-\nabla_\mu\frac{\partial L_0}{\partial\nabla_\mu\varphi}=-m^2\varphi-\nabla_\mu\nabla^\mu\varphi.$$Note: I don't know what the standard symbol is for $L/\sqrt{g}$, but I do know it's sometimes called the scalar Lagrangian density.

  • $\begingroup$ Remark: this approach is probably better suited than mine for more complicated Lagrangians, involving covariant derivatives of vector or tensor fields, for instance. $\endgroup$ – gj255 Jul 19 '18 at 12:09
  • $\begingroup$ @gj255 I discuss it in detail in Sec. 1.3 of my thesis, etheses.whiterose.ac.uk/12277/1/… The one complication is when varying the metric tensor, for which we run into further difficulties. $\endgroup$ – J.G. Jul 19 '18 at 12:30
  • $\begingroup$ Very helpful! Thanks to both @gj255 and J.G for showing how to derive, in effect, Euler-Lagrange equations for Lagrangians having covariant derivatives. I should be able to extend this to case involving variation of a vector such as the vector potential. $\endgroup$ – spinozarabel Jul 19 '18 at 12:33
  • $\begingroup$ @spinozarabel Good luck. If you get stuck, Sec. 2.1 of the above PDF includes a treatment of the vector fields in electromagnetism & Yang-Mills theory. $\endgroup$ – J.G. Jul 19 '18 at 13:06

It's probably more straightforward to express your action in terms of partial derivatives:

$$ S\left[ {\varphi ,{g_{\mu \nu }}} \right] = \int {\sqrt g {d^4}x\left( { - {1 \over 2}{g^{\mu \nu }}{\partial_\mu }\varphi \,{\partial_\nu }\varphi - {1 \over 2}{m^2}{\varphi ^2}} \right)} \,, $$

since $\varphi$ is just a scalar. Then

$$ \frac{\partial L}{\partial (\partial_\mu\varphi)} = -\sqrt{g}\,g^{\mu \nu} \partial_\nu \varphi \quad\text{and}\quad \frac{\partial L}{\partial \varphi} = -\sqrt{g}\,m^2 \varphi^ \,.$$

Finally we need the useful expression for the covariant divergence:

$$ \nabla_\mu V^\mu = \frac{1}{\sqrt{g}} \partial_\mu(\sqrt{g}V^\mu) \,,$$

to reproduce the usual Klein-Gordon equation.

The correspondence you state in $(3)$ is not correct. The normal procedure for turning a theory in flat spacetime into one in curved spacetime involves replacing $\partial_\mu$ by $\nabla_\mu$, yes, but the Euler-Lagrange equations don't change. The E-L equations can be derived for any action, whether in flat spacetime, curved spacetime, or in some other context entirely. One clue that we shouldn't be replacing partial derivatives by covariant derivatives here is the fact that the derivatives are with respect to fields; in the minimal coupling prescription, $\partial_\mu \to \nabla_\mu$ is a statement about how we should take derivatives with respect to position in spacetime.

  • $\begingroup$ The information about scope of changing partial to co-variant derivatives in going from SR to GR is very informative. $\endgroup$ – spinozarabel Jul 19 '18 at 9:52

In order to derive the Klein-Gordon equation you must vary with respect to the scalar field $\phi$. The action reads:

$$S = \int d^{4}x\sqrt{-g}\left(-\cfrac{1}{2}g^{μν}\nabla_{μ}\phi\nabla_{ν}\phi - \cfrac{1}{2}m^{2}\phi ^{2}\right) $$

For the kinetic term you have:

\begin{align} δ(g^{μν}\nabla_{μ}\phi\nabla_{ν}\phi) &= g^{μν}δ(\nabla_{μ}\phi)\nabla_{ν}\phi + g^{μν}δ(\nabla_{ν}\phi)\nabla_{μ}\phi\\ & = δ(\nabla_{μ}\phi)\nabla^{μ}\phi + δ(\nabla_{ν}\phi)\nabla^{ν}\phi \\ & = 2δ(\nabla_{μ}\phi)\nabla^{μ}\phi \\ & = 2\nabla_{μ}(δ\phi)\nabla^{μ}\phi \end{align}

where I've summed the terms with dummy indices. From Leibnitz rule we know that:

$$ \nabla_{μ}(δ\phi\nabla^{μ}\phi) = \nabla_{μ}(δ\phi)\nabla^{μ}\phi + δ\phi\nabla_{μ}\nabla^{μ}\phi $$

The left hand side term in the above expression in spacetime takes the following form:

$$\int d^{4}x\sqrt{-g} \nabla_{μ}(δ\phi\nabla^{μ}\phi).$$

We can see that this term is a surface term so: $$\int d^{4}x\sqrt{-g} \nabla_{μ}(δ\phi\nabla^{μ}\phi) = 0$$ because $δ\phi = 0$ at the surface. So we have:

$$\int d^{4}x \sqrt{-g}\nabla_{μ}(δ\phi)\nabla^{μ}\phi = -\int d^{4}x \sqrt{-g}δ\phi\nabla_{μ}\nabla^{μ}\phi $$

The mass term is:

$$δ(m^{2} \phi ^{2}) = 2\phi δ\phi m^{2}$$

Combining all the above relations one obtains the K-G equation for a massive scalar field:

$$(\Box - m^{2})\phi = 0$$ where $\Box = g^{μν}\nabla_{μ}\nabla_{ν}$ is the D'Alambert operator.

Note - Hint: You can derive K-G equation using the conservation of the energy momentum tensor: $\nabla^{μ}Τ_{μν} = 0$


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.