I am reading Di Francesco's "Conformal Field Theory" and in page 95 he defines a conformal transformation as a mapping $x \mapsto x'$ such that the metric is invariant up to scale:

$$g'_{\mu \nu}(x') = \Lambda(x) g_{\mu \nu} (x).$$

On the other hand we know from GR that under any coordinate transformation the metric changes as

$$ g_{\mu \nu} (x) \mapsto g'_{\mu \nu}(x') = g_{\alpha \beta} (x) \frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} .$$

I feel like there is a notation problem (inconsistency) in these formulas, or maybe I am mixing active and passive coordinate transformations. For instance, if we consider a simple rotation (which is of course a conformal transformation with no rescaling, i.e. $\Lambda(x)=1$) then from the first formula we see that $g'_{\mu \nu}(x') = g_{\mu \nu} (x)$, whereas from the second formula we get something more complicated. Where is the flaw?

In the "String theory" lecture notes by David Tong the same definition of conformal transformation is given. Then he says:

A transformation of the form (4.1) has a diferent interpretation depending on whether we are considering a fixed background metric $g_{\mu \nu}$, or a dynamical background metric. When the metric is dynamical, the transformation is a diffeomorphism; this is a gauge symmetry. When the background is fixed, the transformation should be thought of as an honest, physical symmetry, taking the point $x$ to point $x'$. This is now a global symmetry with the corresponding conserved currents.

I think it has to do with my question, but I don't fully understand it...

  • 2
    $\begingroup$ Related: physics.stackexchange.com/q/38138/2451 , physics.stackexchange.com/q/226464/2451 and links therein. $\endgroup$
    – Qmechanic
    Commented Mar 28, 2019 at 17:48
  • 1
    $\begingroup$ I feel my question is not fully answered there. $\endgroup$
    – MBolin
    Commented Mar 28, 2019 at 17:54
  • 3
    $\begingroup$ The actual definition of a conformal transformation is a mess. By my count, the usual references use at least 5 totally different definitions interchangeably, so you're right to be annoyed. I thought the top answer to the second question Qmechanic linked was good. $\endgroup$
    – knzhou
    Commented Mar 28, 2019 at 17:56

3 Answers 3


OK I think I know what is going on. It's all about primes. Consider an active spacetime transformation:

$$ x^{\mu} \mapsto x'^{\mu}(x) \, ,$$

$$g_{\mu \nu} (x) \mapsto g'_{\mu \nu} (x') = g_{\alpha \beta} (x) \frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} \, .$$

(the transformation of the metric tensor follows from the fact that it is a rank 2 tensor). With this notation both Di Francesco and David Tong are wrong (as far as I understand). The GR book by Zee on the other hand writes it properly. First of all consider an isometry. This is an spacetime transformation as before that leaves the metric invariant, meaning

$$ g'_{\mu \nu} (x') = g_{\alpha \beta} (x) \frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} = g_{\mu \nu} (x') \, .$$

(watch the primes). On the other hand a conformal transformation is a transformation that satisfies a weaker condition: it leaves the metric invariant up to scale, meaning

$$ g'_{\mu \nu} (x') = g_{\alpha \beta} (x) \frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} = \Omega^2(x')g_{\mu \nu} (x') \, .$$

Now there should be no inconsistency. Di Francesco's definition was wrong (according to this convention/notation/understanding) because it compared the metric before and after the transformation at different points, and you have to compare them at the same point.

  • $\begingroup$ Did you check what equation do you get by imposing invariance under a small transformation? I mean, if you use your definition do you get the covariantized Killing equation $\nabla_{(a}v_{b)} = \nabla^cv_c\,g_{ab}$? $\endgroup$
    – MannyC
    Commented Mar 28, 2019 at 23:10
  • $\begingroup$ Yes, I get $\nabla_{\mu} v_{\nu} + \nabla_{\nu} v_{\mu} = \frac{2}{d} (\nabla \cdot v ) g_{\mu \nu}$. $\endgroup$
    – MBolin
    Commented Mar 28, 2019 at 23:32
  • $\begingroup$ Instead if you use the other one there are some pieces missing I imagine? This is very weird... $\endgroup$
    – MannyC
    Commented Mar 28, 2019 at 23:38
  • $\begingroup$ Yes, there are missing pieces and it is logical. Try it. From Di Francesco's definition, for a change of coordinates to be a conformal transformation you should have something like $\frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} = \Omega^2(x) \delta^{\alpha}_{\mu} \delta^{\beta}_{\nu}$, which doesn't seem like a conformal transformation in general. Take a look at Zee. $\endgroup$
    – MBolin
    Commented Mar 28, 2019 at 23:40
  • 1
    $\begingroup$ Damn it, I thought the second block-quoted expression for diffeomorphism in your question to be from Tong which is what I was defending (while thinking that you had quoted it from Tong because I had misread). Yes, now I have no disagreements. Thanks for the responses and for providing a nice addition to my PSE bookmarks ;) $\endgroup$
    – user87745
    Commented Nov 17, 2020 at 13:12

I'm a mathematician, not a physicist, so I learned all of these ideas with different notation, but I think I understand what might be confusing you.

Conformal transformations are indeed a special kind of diffeomorphism, and a rotation (say in the plane with the usual metric) is indeed conformal, so the two formulas you listed had better agree in this case.

But in fact, if your manifold is $\mathbb{R}^2$, your metric is the usual one ($g_{\mu\nu}$ is the identity matrix at every $x$), and your coordinate change is a rotation, the second formula you listed will show you that the metric looks unchanged in the new coordinates. (This is not a coincidence: preserving this metric is exactly the property that makes rotations special in the first place!) That is, there is no conflict between the two formulas here, it's just that seeing it involves a bit of computation.

Working this out is a very good exercise and I don't think you'd gain much from me typing it all out here. A hint that might help you get oriented is that, since rotations are linear in the coordinate system we've chosen, the Jacobian matrix at every point is the same as the matrix for the rotation itself.

  • $\begingroup$ You are right, but I think that is not saying anything. Notice that for a simple metric like Euclidean metric, which doesn't depend on position, the formula for the transformation of the metric $g'_{\mu \nu}(x') = \frac{\partial x^{\alpha}}{\partial x'^{\mu}} \frac{\partial x^{\beta}}{\partial x'^{\nu}} g_{\alpha \beta} (x)$ and the equation for an isometry $g'_{\mu \nu}(x') = g_{\mu \nu}(x') $ look the same. $\endgroup$
    – MBolin
    Commented Mar 28, 2019 at 18:42
  • $\begingroup$ Your answer indeed gives one of the five definitions of the phrase “conformal transformation”. But it isn’t the one that CFT is actually about. Your definition makes conformal invariance just a subset of diffeomorphism invariance. $\endgroup$
    – knzhou
    Commented Mar 28, 2019 at 18:43
  • 1
    $\begingroup$ @NicolasFord A conformal transformation (in the sense of CFT) is a two-step process: first apply a diffeomorphism as you said, then apply an actual rescaling of the metric to cancel the $\Lambda$ factor. (This rescaling is allowed to also rescale the values of other fields.) At the end of the day, $g_{\mu\nu}'(x') = g_{\mu\nu}(x)$. $\endgroup$
    – knzhou
    Commented Mar 28, 2019 at 20:00
  • 1
    $\begingroup$ @NicolasFord You're right that the formula as stated means what you said. However, this isn't the definition that is actually used later in the book. For example, even 5 pages later the book begins to speak of how to decide which classical field theories are conformally invariant. This doesn't make any sense if conformal transformations are a subset of diffeomorphisms, because just about all reasonable theories are diffeomorphism invariant, so knowing a theory is a CFT would be just about useless. $\endgroup$
    – knzhou
    Commented Mar 28, 2019 at 20:02
  • 1
    $\begingroup$ @NicolasFord Yes, I agree 100%, the OP has a different issue. I'm just pointing out this other issue because it's a pet peeve of mine, and because otherwise OP would encounter a nasty surprise in a few pages and get confused all over again. $\endgroup$
    – knzhou
    Commented Mar 28, 2019 at 20:10

I deleted my previous answer because I was very confused when I wrote it. I realized that I just wanted to stress MBolin's answer.

From a differential geometric point of view, an active transformation of space(time) is a diffeomorphism $\phi:M\rightarrow M$. Given a curve $\gamma$ with initial velocity $X=\dot{\gamma}(0)$ at $\gamma(0)$, we can define the new transformed curve $\phi\circ\gamma$ which has velocity $\phi_{*,\gamma(0)}X$ at the transformed initial point $\phi(\gamma(0))$. This is the push-forward of the vector $X$. The conformal condition is then just that the angles between two vectors and their push-forwards coincide. In other words, that there is a function $\Lambda:M\rightarrow\mathbb{R}$ such that for any two vectors $X$ and $Y$ at $p$ we have $$g_{\phi(p)}(\phi_{*,p}X,\phi_{*,p}Y)=\Lambda(p) g_p(X,Y).$$ The left hand side can be thought of as the pull-back of the metric so this condition is at times written as $\phi^*g=\Lambda g$, but this is just a more condensed version of the statement above .

In a coordinate system $x$ defined on a chart containing both $p$ and its transformed point $\phi(p)$, we can write the equation above as $$\Lambda g_{\mu\nu}(x)=\frac{\partial x^\alpha\circ\phi}{\partial x^\mu}\frac{\partial x^\beta\circ\phi}{\partial x^\nu}g_{\alpha\beta}(x)\circ\phi.$$ In here I have used the physicists notation of writing $g_{\mu\nu}(x)$ for the composition $g_{\mu\nu}\circ x$. In here my convention is that $$g_{\mu\nu}(x(p))=g_p\left(\left(\frac{\partial}{\partial x^\mu}\right)_p,\left(\frac{\partial}{\partial x^\nu}\right)_p\right)$$

There is a parallel (although local) discussion on can make from the passive point of view. In here one is instead replacing the coordinates $x$ by a new system $x':=x\circ\phi^{-1}$ (thinking of active vs. passive rotations clarifies why this is the correct coordinates to choose). Then the trick for relating these two is to note that $$\frac{\partial x^\alpha\circ\phi}{\partial x^\mu}=\partial_\mu(x^\alpha\circ\phi\circ x^{-1})\circ x=\partial_\mu(x^\alpha\circ {x'}^{-1})\circ x'\circ\phi=\frac{\partial x^\alpha}{\partial x'^\mu}\circ\phi.$$ Then the conformal condition becomes $$\Lambda g_{\mu\nu}(x)=\left(\frac{\partial x^\alpha}{\partial x'^\mu}\frac{\partial x^\beta}{\partial x'^\nu}g_{\alpha\beta}(x)\right)\circ\phi.$$ We recognize that the term in parenthesis is just the metric in the new coordinates $g'_{\mu\nu}(x')$. Then we have $g'_{\mu\nu}(x')\circ\phi=g'_{\mu\nu}\circ x'\circ\phi=g'_{\mu\nu}\circ x\equiv g'_{\mu\nu}(x)$. We thus recover the statement as in Zee's book that the conformal condition is $$g'_{\mu\nu}(x)=\Lambda g_{\mu\nu}(x).$$


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.