Conformal Symmetry

⚡ Key Takeaways

Conformal symmetry is invariance under transformations that preserve angles but not necessarily lengths. The conformal group of $d$ -dimensional spacetime is locally $SO(d+1,1)$ (Euclidean) or $SO(d,2)$ (Lorentzian), with $\tfrac{(d+1)(d+2)}{2}$ generators: translations, rotations, dilatations, and special conformal transformations. It governs critical phenomena, massless field theories, string worldsheets, and the AdS/CFT correspondence. Foundational work spans Liouville (1850), Lie (1880s), Bateman–Cunningham (1909), Polyakov (1970s), and Belavin–Polyakov–Zamolodchikov (1984).

💡 The Core Insight

Demanding that physics look the same after a local rescaling of all lengths—while keeping angles intact—is enormously more restrictive than ordinary Poincaré symmetry. It promotes scale invariance to a full conformal group, fixes the form of two- and three-point functions, organizes operators into representations labeled by scaling dimensions and spins, and in $d=2$ explodes into an infinite tower of conserved currents. Conformal symmetry is the symmetry of criticality and of masslessness.

DefinitionConformal Transformation

Let $(M,g)$ be a (pseudo-)Riemannian manifold of dimension $d$ . A diffeomorphism $\phi: M \to M$ is conformal if

1

\quad (\phi^* g)_{\mu\nu}(x) = \Omega(x)^2\, g_{\mu\nu}(x), \qquad \Omega(x) > 0.

Derivation hint: This is the definition; angles $\cos\theta = \frac{g(u,v)}{\sqrt{g(u,u)g(v,v)}}$ are invariant since $\Omega^2$ cancels between numerator and denominator. An infinitesimal coordinate change $x^\mu \to x^\mu + \epsilon^\mu(x)$ is conformal iff $\epsilon^\mu$ satisfies the conformal Killing equation:

2

\quad \partial_\mu \epsilon_\nu + \partial_\nu \epsilon_\mu = \frac{2}{d}(\partial_\cdot \epsilon)\, \eta_{\mu\nu},

where $\partial_\cdot\epsilon \equiv \partial_\rho \epsilon^\rho$ and on flat space $g=\eta$ . Derivation hint: Vary $\delta g_{\mu\nu} = \partial_\mu\epsilon_\nu + \partial_\nu\epsilon_\mu$ ; demand it be proportional to $\eta_{\mu\nu}$ to preserve conformality, then trace both sides to fix the coefficient as $2/d$ .

TheoremStructure of the Conformal Group (

d\geq 3

)

For flat space $\mathbb{R}^{p,q}$ with $d=p+q\geq 3$ , the solutions of [2] form a finite-dimensional Lie algebra isomorphic to $\mathfrak{so}(p+1,q+1)$ , of dimension $\tfrac{(d+1)(d+2)}{2}$ . The generators are:

Translations $P_\mu$ ( $d$ of them): $\epsilon^\mu = a^\mu$ .
Rotations / Lorentz $M_{\mu\nu}$ ( $\tfrac{d(d-1)}{2}$ ): $\epsilon^\mu = \omega^\mu{}_\nu x^\nu$ , $\omega$ antisymmetric.
Dilatation $D$ (1): $\epsilon^\mu = \lambda x^\mu$ .
Special conformal transformations (SCT) $K_\mu$ ( $d$ ): $\epsilon^\mu = b^\mu x^2 - 2 (b\cdot x) x^\mu$ .

The finite SCT acts as

3

\quad x'^\mu = \frac{x^\mu - b^\mu x^2}{1 - 2\, b\cdot x + b^2 x^2}.

Derivation hint: Insert the polynomial ansatz $\epsilon^\mu = a^\mu + \omega^\mu{}_\nu x^\nu + \lambda x^\mu + (b^\mu x^2 - 2(b\cdot x)x^\mu)$ into [2]. For $d\geq 3$ the third derivative $\partial_\rho\partial_\sigma\partial_\tau\epsilon^\mu=0$ , truncating $\epsilon$ at quadratic order. Exponentiate the SCT generator; equivalently note SCT = inversion $\circ$ translation $\circ$ inversion.

DefinitionInversion

The discrete inversion $I: x^\mu \mapsto x^\mu / x^2$ is conformal with $\Omega(x) = 1/x^2$ . The SCT is the composition

4

\quad K_b = I \circ T_{-b} \circ I,

with $T_{-b}$ a translation by $-b$ . Derivation hint: Compute $I(x) = x/x^2$ , translate by $-b$ , invert again; algebra collapses to [3]. Inversion is not connected to the identity, so it generates the full $O(p+1,q+1)$ beyond the connected component.

DefinitionConformal Algebra Commutators

With $P_\mu, M_{\mu\nu}, D, K_\mu$ , the nonzero brackets are

5

\quad \begin{aligned} &[D, P_\mu] = i\, P_\mu, \qquad [D, K_\mu] = -i\, K_\mu, \\ &[K_\mu, P_\nu] = 2i\,(\eta_{\mu\nu} D - M_{\mu\nu}), \\ &[M_{\mu\nu}, P_\rho] = i(\eta_{\mu\rho}P_\nu - \eta_{\nu\rho}P_\mu),\\ &[M_{\mu\nu}, K_\rho] = i(\eta_{\mu\rho}K_\nu - \eta_{\nu\rho}K_\mu). \end{aligned}

Derivation hint: $P_\mu$ raises dimension by $+1$ , $K_\mu$ lowers by $-1$ as eigen-operators of $\mathrm{ad}_D$ . Rename $J_{\mu\nu}=M_{\mu\nu}$ , $J_{\mu d}=\tfrac12(P_\mu - K_\mu)$ , $J_{\mu,d+1}=\tfrac12(P_\mu+K_\mu)$ , $J_{d,d+1}=D$ to verify the $\mathfrak{so}(p+1,q+1)$ form.

ProofWhy the algebra closes to

\mathfrak{so}(p+1,q+1)

Step 1. Apply $\partial_\rho$ to [2] and permute indices. Taking suitable linear combinations gives

(d-2)\,\partial_\mu\partial_\nu(\partial_\cdot\epsilon) = 0.

For $d\geq 3$ , this forces $\partial_\cdot\epsilon$ to be at most linear in $x$ .

Step 2. Differentiating once more shows $\partial_\mu\partial_\nu\partial_\rho\epsilon_\sigma = 0$ , so $\epsilon^\mu$ is a polynomial of degree $\leq 2$ . Counting coefficients: constant ( $d$ ), linear antisymmetric + trace ( $\tfrac{d(d-1)}{2}+1$ ), quadratic ( $d$ ). Total $\tfrac{(d+1)(d+2)}{2}$ .

Step 3. Identify these with $P,M,D,K$ . Compute commutators of the corresponding vector fields under the Lie bracket of vector fields; one obtains [5].

Step 4. Define the embedding $J_{AB}$ , $A,B \in\{0,\dots,d+1\}$ , as in the hint of [5]. The brackets [5] become the standard $\mathfrak{so}(p+1,q+1)$ relations

[J_{AB}, J_{CD}] = i(\eta_{AC}J_{BD} - \eta_{BC}J_{AD} - \eta_{AD}J_{BC} + \eta_{BD}J_{AC}).

Hence the conformal algebra of $\mathbb{R}^{p,q}$ ( $d\geq 3$ ) is $\mathfrak{so}(p+1,q+1)$ . $\blacksquare$

∎

⚠️ The two-dimensional exception

For $d=2$ the step-1 factor $(d-2)$ vanishes, so $\partial_\cdot\epsilon$ is not forced to be linear. In $z=x^1+ix^2$ the conformal Killing equation reduces to the Cauchy–Riemann equations: every holomorphic map $z\mapsto f(z)$ is conformal. The local algebra is therefore infinite-dimensional—the Witt algebra $\{\ell_n = -z^{n+1}\partial_z\}$ with

6

\quad [\ell_m,\ell_n] = (m-n)\,\ell_{m+n},

which centrally extends to the Virasoro algebra. Only the global $SL(2,\mathbb{C})$ subalgebra ( $\ell_{-1},\ell_0,\ell_1$ ) consists of globally well-defined maps of the Riemann sphere. Derivation hint: With $z=x^1+ix^2$ , [2] in $d=2$ becomes $\partial_{\bar z}\epsilon^z = 0 = \partial_z\epsilon^{\bar z}$ ; expanding $\epsilon^z(z)=\sum_n a_n z^{n+1}$ and computing brackets of $\ell_n$ gives [6].

AP

Alexander Polyakov

b. 1945

Polyakov, with Patashinskii and later independently, advanced the idea (early 1970s) that scale invariance at critical points is enhanced to full conformal invariance, fixing correlation functions by symmetry alone. His 1981 work on the conformal anomaly and quantization of the bosonic string ("Polyakov action") made the central charge and Liouville theory central tools, and his constraints on operator product expansions seeded the conformal bootstrap. The 1984 BPZ paper (Belavin, Polyakov, Zamolodchikov) is the founding document of modern 2D CFT.

🧠 Intuition

Think of conformal maps as transformations that may stretch, but never shear: an infinitesimal circle stays a circle (locally), only its radius changes. In 2D this is exactly complex analyticity—a holomorphic function is a "rotation-and-scaling field," which is why a Mercator-style map distorts areas wildly yet preserves the right angles where meridians cross parallels. Dilatations $D$ are the heart of the story: at a critical point there is no length scale (correlation length $\xi\to\infty$ , lattice spacing $a\to 0$ ), so the theory cannot tell large from small. Adding SCTs $K_\mu$ amounts to "scale invariance that varies smoothly from point to point," consistent only when the trace of the stress tensor vanishes, $T^\mu{}_\mu = 0$ . Operators are then sorted by how they respond to zooming: a primary of dimension $\Delta$ scales as $\phi(\lambda x) = \lambda^{-\Delta}\phi(x)$ , and $\Delta$ directly controls power-law decay of correlators.

Figure

The conformal group generators acting on flat space: translation, rotation, dilatation, and special conformal transformation (inversion-translation-inversion).

```svg

```

ExampleTwo-point function fixed by symmetry

Given: A scalar primary operator $\mathcal{O}(x)$ of scaling dimension $\Delta$ in a $d$ -dimensional CFT. Find its two-point function $\langle \mathcal{O}(x_1)\mathcal{O}(x_2)\rangle$ .

Step 1 — Translation + rotation invariance. The correlator depends only on $r=|x_1-x_2|$ :

\langle \mathcal{O}(x_1)\mathcal{O}(x_2)\rangle = G(r).

Step 2 — Dilatation. Under $x\to\lambda x$ , $\mathcal{O}\to\lambda^{-\Delta}\mathcal{O}$ , so

G(\lambda r) = \lambda^{-2\Delta} G(r) \implies G(r) = \frac{C}{r^{2\Delta}}.

Step 3 — Special conformal check. Under [3], the Jacobian factor is $|x'-y'| = \frac{|x-y|}{\gamma(x)^{1/2}\gamma(y)^{1/2}}$ with $\gamma(x)=(1-2b\cdot x+b^2x^2)$ . The transformation $\mathcal{O}(x)\to \gamma(x)^{\Delta}\mathcal{O}(x')$ leaves $C/r^{2\Delta}$ invariant.

Step 4 — Numerical instance. Take the 3D Ising critical point, magnetization operator $\sigma$ with $\Delta_\sigma \approx 0.5181$ . Then

7

\quad \langle\sigma(x_1)\sigma(x_2)\rangle = \frac{C}{|x_1-x_2|^{2\Delta_\sigma}} \approx \frac{C}{r^{1.036}}.

Derivation hint: Each symmetry kills functional freedom; the only invariant cross-ratio for two points is none, so the form is rigid up to normalization $C$ .

\boxed{\;\langle\mathcal{O}(x_1)\mathcal{O}(x_2)\rangle = \dfrac{C}{|x_{12}|^{2\Delta}}\quad(\text{dimensionless }C)\;}

ExampleCentral charge from the free boson

Given: The 2D free massless scalar $\varphi$ with action $S=\frac{1}{4\pi}\int d^2z\,\partial\varphi\bar\partial\varphi$ . Compute its central charge $c$ .

Step 1 — Stress tensor. $T(z) = -\tfrac12 (\partial\varphi)^2$ (holomorphic component), with the propagator $\langle\partial\varphi(z)\partial\varphi(w)\rangle = -1/(z-w)^2$ .

Step 2 — OPE. Using Wick's theorem,

T(z)T(w) \sim \frac{c/2}{(z-w)^4} + \frac{2T(w)}{(z-w)^2} + \frac{\partial T(w)}{z-w}.

Two contractions of the four $\partial\varphi$ 's give the leading double-contraction term:

\frac{1}{4}\cdot 2 \cdot \frac{1}{(z-w)^4} = \frac{1/2}{(z-w)^4}.

Step 3 — Read off $c$ . Matching to $\frac{c/2}{(z-w)^4}$ :

8

\quad \frac{c/2}{(z-w)^4} = \frac{1/2}{(z-w)^4}\implies c = 1.

Derivation hint: The most singular term in the $TT$ OPE has no operator and is pure c-number; its coefficient is $c/2$ . Count the $2$ ways to fully contract.

\boxed{\;c_{\text{free boson}} = 1\;\text{(dimensionless)}\;}

⚠️ Scale invariance ≠ conformal invariance in general

A common error is to assume any scale-invariant theory is automatically conformally invariant. Scale invariance requires only a conserved dilatation current ( $\partial_\mu(x^\nu T^\mu{}_\nu)=0$ ), i.e. $T^\mu{}_\mu = \partial_\mu V^\mu$ for a "virial current" $V^\mu$ . Full conformal invariance additionally requires this to be improvable to $T^\mu{}_\mu = 0$ (a traceless symmetric stress tensor). In unitary theories this enhancement is proven in $d=2$ (Zamolodchikov–Polchinski) and strongly supported in $d=4$ , but it is not a theorem in general $d$ , and counterexamples (non-unitary, or with unbounded virial current) exist.

CounterexampleElasticity / Maxwell scale vs conformal

Free Maxwell theory $F_{\mu\nu}F^{\mu\nu}$ in $d$ dimensions is scale invariant for all $d$ but conformally invariant only in $d=4$ . For $d\neq 4$ the gauge field has scaling dimension $\Delta_A = \tfrac{d-2}{2}$ that does not match the conformal value required for the stress tensor to be traceless; the trace $T^\mu{}_\mu \propto (d-4)F^2$ vanishes precisely at $d=4$ . Thus "massless" does not guarantee "conformal" without checking tracelessness.

connection

Conformal symmetry is the connective tissue of several major structures:

Renormalization Group: fixed points of the RG flow are scale-invariant and (generically) conformal; critical exponents are scaling dimensions of Primary Operators.
<a class="wiki-link" href="/wiki/virasoro-algebra">Virasoro Algebra</a>: the central extension of the 2D conformal algebra; its representation theory (Kac determinant, minimal models) classifies rational 2D CFTs.
<a class="wiki-link" href="/wiki/ads-cft-correspondence">AdS/CFT Correspondence</a>: a CFT in $d$ dimensions is dual to gravity in $(d+1)$ -dimensional anti–de Sitter space; the conformal group $SO(d,2)$ is exactly the AdS $_{d+1}$ isometry group.
Conformal Bootstrap: uses crossing symmetry of four-point functions plus unitarity to numerically pin down CFT data such as the 3D Ising $\Delta_\sigma$ .
Weyl Transformation: pointwise rescaling of the metric $g\to e^{2\omega}g$ ; the conformal anomaly measures its breaking at the quantum level.

🔬 Where this appears

⏳ History

- 1850 — Liouville's theorem: In

d\geq 3

, all conformal maps are compositions of isometries, dilatations, and inversions—foreshadowing the finite-dimensional group.

- 1870s–1890s — Sophus Lie & Felix Klein: Systematic theory of continuous transformation groups; the conformal group emerges as a Lie group.

- 1909 — Bateman & Cunningham: Discover that Maxwell's equations in vacuum are invariant under the 15-parameter conformal group

SO(4,2)

in 4D.

- 1936 — Dirac: Conformal-space (projective null-cone) formalism, the "Dirac cone," for manifestly conformal field theory.

- 1970–1971 — Polyakov, Migdal, Mack–Salam: Conformal invariance at critical points; conformal Ward identities fixing correlators.

- 1981 — Polyakov: Quantization of the bosonic string, conformal anomaly, and Liouville action; central charge becomes physical.

- 1984 — Belavin, Polyakov, Zamolodchikov (BPZ): Foundational paper of 2D CFT—Virasoro representations, minimal models, the operator algebra.

- 1986 — Zamolodchikov's $c$ -theorem: Monotonic decrease of the central charge under RG flow in 2D.

- 2008–2016 — Numerical conformal bootstrap (Rattazzi, Rychkov, Poland, Simmons-Duffin, et al.): High-precision 3D Ising data, reviving the bootstrap program in

d>2

.

open-problem

Is scale invariance equivalent to conformal invariance in $d=4$ ? While proven in $d=2$ , a fully general, rigorous proof in four dimensions for all unitary, interacting theories remains incomplete; strong arguments (Luty–Polchinski–Rattazzi, Dymarsky–Komargodski–Schwimmer–Theisen) exist but rely on assumptions about the spectrum.

Classifying CFTs in $d\geq 3$ : Unlike the complete classification of unitary minimal models in $d=2$ via the Virasoro algebra, there is no exhaustive classification of higher-dimensional CFTs; the conformal bootstrap carves out the allowed region but does not yet enumerate all theories.

Non-unitary, logarithmic, and irrational CFTs: Logarithmic CFTs (non-diagonalizable $L_0$ ), irrational theories with infinitely many primaries, and CFTs with continuous spectra resist the standard representation-theoretic toolkit.

Conformal anomaly coefficients: The general behavior and possible monotonicity ( $a$ -theorem) of central charges in $d>4$ (e.g. the conjectured 6D $a$ -theorem) is only partially understood.

Formal Statement

Why It Holds / Derivation

Building Intuition

Worked Examples

Common Misconceptions

Connections

Real-World Applications

Historical Development

Subtleties & Limits

See Also