CONFORMAL TRANSFORMATIONS THAT CLICKBONUS ⑥ (full length) / The other half of the series title

For fifteen episodes we have said "conformal transformation" while meaning only one of the two things that go by that name.

The other conformal transformation The Weyl weight had a proper name all along — the scaling dimension \(\Delta\).
And a world in which nothing but \(\Delta\) exists as an observable is real. It is the critical point of water.

What you need: division, inequalities, and some comfort with having nothing but pure numbers \(\mathcal{O}\to\Omega^{-\Delta}\mathcal{O}\)

This series has consistently used "conformal transformation" for the Weyl transformation — the operation that swaps your ruler point by point. But physics has a second thing with that name: the conformal group of flat spacetime. Set the two side by side and the "weight" we have been computing all along acquires its proper name — and we walk into a world where no dimensionful quantity means anything at all. There the observables are \(\Delta\), a pure number, and nothing else. And that world is not speculation: water, carbon dioxide and uniaxial magnets all fall into it at their critical points.

01There are two conformal transformations

This is a standing source of confusion, so let us separate them first.

Weyl transformationConformal group
What it doesrewrites the metric, \(g\to\Omega^2g\)moves the spacetime (a coordinate change)
Freedomone function \(\Omega(x)\) = infinite-dimensionalfinite-dimensional (for \(d\ne2\))
Charactera gauge-like rewriting (bookkeeping)a genuine symmetry
In this seriesthe star of episodes 1 through bonus ⑤this episode

The conformal group is the set of coordinate changes of flat spacetime that alter the metric by no more than an overall factor. Count them, and the answer is a function of the dimension alone.

Counting

The breakdown

$$\underbrace{\tfrac{d(d-1)}{2}}_{\text{rotations}}+\underbrace{d}_{\text{translations}}+\underbrace{1}_{\text{dilatation}}+\underbrace{d}_{\text{special conformal}}=\frac{(d+1)(d+2)}{2}$$

In four dimensions

$$6+4+1+4=15\qquad(\text{five more than Poincaré's }10)$$

Only five wider than Poincaré symmetry. The additions are the dilatation and the special conformal transformations.

dimension \(d\)number of conformal symmetriesnote
13conformal quantum mechanics
26the formula gives only the global part; locally it is infinite-dimensional
310the arena of critical phenomena
415our spacetime
628

Two dimensions is the exception. There the local conformal transformations are infinite in number (one for every holomorphic function), which is why the theory can be solved exactly. That is why the worldsheet of a string is two-dimensional.

02\(\Delta\) is what this series has been computing all along

The central object in a conformal field theory is the scaling dimension \(\Delta\). Its definition is:

The definition of \(\Delta\)
$$\mathcal{O}(x)\ \longrightarrow\ \Omega^{-\Delta}\,\mathcal{O}(x)$$

That is exactly the Weyl weight.

So episode 3's table of weights, episode 7's \(\Omega^{D-4}\), bonus ④'s \(\tilde m=\Omega^{-1}m\) — all of them were computing \(\Delta\), without knowing its name. "The other conformal transformation" was in the room the whole time.

And \(\Delta\) has two parts.

Classical and quantum $$\Delta=\underbrace{\Delta_{\text{classical}}}_{\text{engineering dimension}}+\underbrace{\gamma}_{\text{anomalous dimension}}$$

\(\gamma\ne0\) means "the weight you computed classically is wrong" — episode 8's anomaly, operator by operator.

Episode 8 measured the anomaly with the \(\beta\) function (\(1/137\to1/128\)). At the level of individual operators, that same thing is \(\gamma\). And \(\gamma\) can be measured in the laboratory — section 05.

03A world containing nothing but \(\Delta\)

In a theory where conformal symmetry is exact (a conformal field theory, CFT), a dimensionful quantity cannot exist even in principle: there is no standard of length. So what is left? A list of \(\Delta\)s, and the coefficients for multiplying operators together. Both dimensionless.

And \(\Delta\) is not free to be anything. Unitarity — probabilities not going negative — imposes a floor.

operatorconstraint on \(\Delta\)in \(d=4\)meaning
scalar\(\Delta\ge\dfrac{d-2}{2}\)\(\ge1\)equality means a free field
conserved current \(J_\mu\)\(\Delta=d-1\)\(=3\)exact (conservation fixes it)
stress tensor \(T_{\mu\nu}\)\(\Delta=d\)\(=4\)exact
Episode 8 appears here "\(\Delta(T_{\mu\nu})=d\) exactly" is a restatement of \(T^\mu_\mu=0\): if the trace vanishes, the dimension of the stress tensor cannot move. The conformal anomaly of episode 8 is precisely the failure of that. One and the same fact, said once as "a trace survives" and once as "a dimension shifts".
◇ ◇ ◇

04The bootstrap — feeding in nothing but pure numbers

Here is the heart of it. There is a way to determine \(\Delta\) in a CFT without ever writing down a Lagrangian. It is called the conformal bootstrap.

input 1
crossing symmetryexpand a four-point function in two different orders and demand the same answer
input 2
unitarityprobabilities are not negative — every expansion coefficient is positive
input 3
a symmetry (e.g. \(\mathbb{Z}_2\))nothing more than "flipping every spin changes nothing"

That is the entire input. No coupling constant, no cutoff, no lattice spacing — not one dimensionful quantity. What comes out is a forbidden region of \(\Delta\) values, and its boundary has a kink.

And the three-dimensional Ising model sits on that kink. Nature is parked on a corner of a purely mathematical exclusion plot.

The 3d Ising CFT (bootstrap, 2016)
$$\Delta_\sigma=0.5181489(10),\qquad \Delta_\varepsilon=1.412625(10)$$

Everything about critical phenomena follows from these two numbers.

First read off the anomalous dimension. A free scalar in three dimensions has \(\Delta=(d-2)/2=0.5\) exactly, so

$$\gamma_\sigma=\Delta_\sigma-0.5=0.0181489$$

which is to say that "how wrong the Weyl weight is" is known to seven digits. And it has a familiar name in statistical physics: \(\eta=2\gamma_\sigma=0.0362978\).

05Six critical exponents out of two numbers

Every critical exponent in the statistical-physics textbooks is a translation of \(\Delta_\sigma\) and \(\Delta_\varepsilon\).

The dictionary (\(d=3\)) $$\eta=2\Delta_\sigma-d+2,\qquad \nu=\frac{1}{d-\Delta_\varepsilon}$$ $$\alpha=2-d\nu,\quad \beta=\frac{\nu(d-2+\eta)}{2},\quad \gamma=\nu(2-\eta),\quad \delta=\frac{d+2-\eta}{d-2+\eta}$$

Only the first two are independent; the other four are derived.

exponentbootstrapexperiment / Monte Carlowhat it measures
\(\eta\)0.0362980.0363(2)falloff of correlations
\(\nu\)0.6299710.6300(17)divergence of the correlation length
\(\alpha\)0.1100870.110(5)divergence of the specific heat
\(\beta\)0.3264190.3265(15)shape of the coexistence curve
\(\gamma\)1.2370751.2372(5)divergence of the susceptibility
\(\delta\)4.7898414.789(2)the critical isotherm

And one combination holds for any \((\Delta_\sigma,\Delta_\varepsilon)\) you care to plug in.

The thing that does not move
$$\alpha+2\beta+\gamma=2.000000\qquad(\text{identically, independent of }\Delta_\sigma,\ \Delta_\varepsilon)$$

All six exponents swing wildly, and this one combination never leaves 2. Exactly the structure of bonus ④'s \(Q=2/3\).

Here is the point of the episode. Water, carbon dioxide, uniaxial magnets and binary alloys all obey these same numbers at their critical points. The shape of the molecules, the strength of the bonds — none of it enters. The material is stage scenery and only the pure numbers are physics: what this series has been asserting all along, existing as an experimental fact rather than a metaphor.

Figure: on the left, the \((\Delta_\sigma,\Delta_\varepsilon)\) plane (grey is forbidden by unitarity, ★ is the Ising point). On the right, bars for the six critical exponents, the thin vertical line being the experimental value. Move the knobs and all six bars move — and the only place where all six land on their marks is the Ising point. Even so, \(\alpha+2\beta+\gamma\) never leaves 2.
Δ_σ = 0.518149
current value experimental mark forbidden by unitarity

06Why the critical point is a point, not a surface

\(\Delta\) has one more decisive job: deciding whether an operator can be ignored.

The criterion is nothing but a comparison with \(d\)
$$\Delta<d\ \Rightarrow\ \text{relevant (it grows; you must turn a knob to cancel it)}$$ $$\Delta>d\ \Rightarrow\ \text{irrelevant (it dies on its own; ignore it)}$$

Put numbers in for 3d Ising.

operator\(\Delta\)versus \(d=3\)corresponding knob
\(\sigma\) (\(\mathbb{Z}_2\) odd)0.518149relevantmagnetic field (forbidden if you keep the symmetry)
\(\varepsilon\) (\(\mathbb{Z}_2\) even)1.412625relevanttemperature
\(\varepsilon'\) (\(\mathbb{Z}_2\) even)3.829510irrelevantnone needed

As long as the symmetry is kept, there is exactly one relevant operator, \(\varepsilon\). So there is exactly one knob — temperature.

Conclusion of section 06

How many knobs you must turn to reach criticality is fixed by comparing pure numbers.

\(\Delta_\varepsilon=1.41\ <\ 3\ <\ \Delta_{\varepsilon'}=3.83\) — that single inequality is why the critical point is a point and not a surface. And the fact that \(\varepsilon'\) exceeds \(d\) is universality itself: every microscopic detail is irrelevant and flows away.

07The sign of a mass means nothing

Finally, the most extreme illustration of this series' conclusion: the holographic (AdS/CFT) dictionary.

Bookkeeping and physics as two readings of one thing
$$\Delta(\Delta-d)=m^2R^2\qquad\Longleftrightarrow\qquad \Delta=\frac{d}{2}+\sqrt{\frac{d^2}{4}+m^2R^2}$$

The left side is the boundary's dimensionless \(\Delta\), the right the bulk's dimensionful mass. One object, seen from inside or from outside.

And an almost unbelievable consequence follows. For \(\Delta\) to be real one needs only \(m^2R^2\ge-d^2/4\) — so \(m^2\) is allowed to be negative.

\(d\)BF bound \(m^2R^2\ge\)\(\Delta\) thereat \(m^2R^2=0\)
2\(-1\)1.0\(\Delta=2\)
3\(-2.25\)1.5\(\Delta=3\)
4\(-4\)2.0\(\Delta=4\)

In flat spacetime \(m^2<0\) is the definition of a tachyon, of instability. In AdS it is not.

The one line of this episode

What decides stability is not the sign of \(m^2\) but whether \(\Delta\) is real.

The sign of a dimensionful quantity means nothing on its own; only dimensionless ones can decide. Episode 10's decision procedure, promoted to a theorem about stability.

08An open door — helium's 6σ

Having told a very clean story, here is something that does not fit. Next door to Ising, in the 3d XY universality class — the λ point of liquid helium-4.

Not matching

Theory (bootstrap, with Monte Carlo agreeing)

$$\Delta_s=1.51136(22)\ \Rightarrow\ \nu=0.671754,\qquad \alpha=2-3\nu=-0.01526(30)$$

Experiment (Lipa et al. 2003, measured on the Space Shuttle to remove gravity)

$$\alpha=-0.0127(3)$$

A difference of 0.00256, roughly . Both sides are precise, and they do not agree.

The specific-heat exponent is so close to zero that logarithmic corrections are delicate, and both the experimental analysis and the theoretical extrapolation have been argued over. It is still unresolved.

The honest line

Section 01's separation of the Weyl transformation from the conformal group is a pedagogical split. The two are closely related — the conformal group is the combination of diffeomorphisms and Weyl transformations that preserves flat space. Also, "Weyl invariance implies conformal invariance" is not true in general (the converse is), and holds only under conditions.

The bootstrap of section 04 is a numerical method. "Ising sits on the kink" means the kink's location agrees with Monte Carlo, not that the kink has been proven to be Ising (modern "island" methods do isolate it to high precision). The quoted errors on \(\Delta\) are numerical and include systematic estimates.

The "experiment / Monte Carlo" column in section 05 gives representative values across several measurements and computations, not a single source. Real fluids and magnets are not exact CFTs; the exponents appear only sufficiently close to the critical point, with corrections to scaling.

Section 07's \(\Delta(\Delta-d)=m^2R^2\) is the relation for a scalar field; with spin the form changes. AdS/CFT itself is a conjecture and is not proven in general. The discrepancy in section 08 is a genuine unresolved problem, and it is not settled whether the fault lies with experiment or with theory.

Exercises
  1. Four-dimensional conformal symmetry has 15 generators, five more than Poincaré's 10. What are the five?
    Show the answer
    One dilatation and four special conformal transformations. The first stretches everything uniformly; the second is "invert, translate, invert again". In general \(d\) you gain \(1+d\), for a total of \(d(d-1)/2+d+1+d=(d+1)(d+2)/2\).
  2. State the relation between \(\eta\) and the anomalous dimension.
    Show the answer
    \(\eta=2\gamma_\sigma\). A free scalar in three dimensions has \(\Delta=(d-2)/2=0.5\), so \(\gamma_\sigma=\Delta_\sigma-0.5=0.0181489\) and \(\eta=0.0362978\). \(\eta\) is twice "how wrong the classical Weyl weight is" — episode 8's anomaly, measured operator by operator.
  3. Why does reaching the critical point need only one knob, the temperature?
    Show the answer
    Because among the \(\mathbb{Z}_2\)-even operators only one, \(\varepsilon\), is relevant (\(\Delta<d=3\)); the next one, \(\varepsilon'\), has \(\Delta=3.83>3\) and is irrelevant. The number of knobs equals the number of relevant operators, and that is fixed by comparing pure numbers with \(d\). Incidentally, \(\varepsilon'\) being irrelevant is also the reason for universality.
  4. Show that \(\alpha+2\beta+\gamma=2\) identically.
    Show the answer
    Just substitute: \(\alpha+2\beta+\gamma=(2-d\nu)+\nu(d-2+\eta)+\nu(2-\eta)=2-d\nu+d\nu-2\nu+\nu\eta+2\nu-\nu\eta=2\). \(\nu\), \(\eta\) and \(d\) all cancel. That is why the figure's knobs never move it — this is not physics but a consequence of six exponents being functions of two numbers.
  5. (Harder) Why is \(m^2<0\) allowed in AdS?
    Show the answer
    Because the stability condition is not "\(m^2>0\)" but "\(\Delta\) is real". Since \(\Delta=d/2+\sqrt{d^2/4+m^2R^2}\), \(\Delta\) stays real all the way down to \(m^2R^2=-d^2/4\), and fluctuations are not amplified. The negative curvature of AdS acts as a box that supports a negative \(m^2\) up to that point. The sign of a dimensionful quantity cannot decide; the reality of a dimensionless one is the true criterion — the Breitenlohner–Freedman bound.

SummaryThe weight had a name

There are two "conformal transformations": the Weyl transformation this series has used for fifteen episodes (infinite-dimensional, a gauge-like rewriting) and the conformal group of flat spacetime (finite-dimensional, a genuine symmetry, 15 generators in four dimensions). Only in \(d=2\) is the conformal group also infinite-dimensional — which is why a string's worldsheet is two-dimensional.

The two were connected all along. The definition of a CFT's central object is \(\mathcal{O}\to\Omega^{-\Delta}\mathcal{O}\), so \(\Delta\) is the Weyl weight. Episode 3's weight table and episode 7's \(\Omega^{D-4}\) were computing \(\Delta\) without naming it. The gap from the classical weight, \(\gamma\), is the anomalous dimension — episode 8's anomaly measured operator by operator.

The conformal bootstrap uses no dimensionful input whatsoever — crossing symmetry, unitarity, and a symmetry. From that come the 3d Ising values \(\Delta_\sigma=0.5181489\) and \(\Delta_\varepsilon=1.412625\), from which all six critical exponents follow, and water, carbon dioxide and uniaxial magnets obey the same numbers. \(\gamma_\sigma=0.0181489\): "the error in the Weyl weight" is known to seven digits. And \(\alpha+2\beta+\gamma\) never leaves 2, whatever \(\Delta\) you feed it.

\(\Delta\) even fixes how many knobs it takes to reach criticality. \(\Delta_\varepsilon=1.41<3<\Delta_{\varepsilon'}=3.83\) — that one inequality is why the critical point is a point, and it is universality itself. Finally the holographic dictionary \(\Delta(\Delta-d)=m^2R^2\) ties bookkeeping (the bulk mass) to physics (the boundary \(\Delta\)) as a single object, and declares that stability is decided not by the sign of \(m^2\) but by whether \(\Delta\) is real. Episode 10's decision procedure turned out to be a theorem about stability.

This document is bonus episode ⑥ of the series "Conformal Transformations That Click", written for high-school and university students who enjoy physics. That the \(d\)-dimensional conformal algebra has dimension \((d+1)(d+2)/2\), and that only for \(d=2\) is the local conformal algebra infinite-dimensional (Virasoro), are standard. That the scaling dimension \(\Delta\) is the weight under a Weyl transformation (\(\mathcal{O}\to\Omega^{-\Delta}\mathcal{O}\)) and that \(\Delta=\Delta_{\text{classical}}+\gamma\) with \(\gamma\) the anomalous dimension are also standard. The scalar unitarity bound \(\Delta\ge(d-2)/2\), \(\Delta=d-1\) for a conserved current and \(\Delta=d\) for the stress tensor are standard results. The 3d Ising CFT values \(\Delta_\sigma=0.5181489(10)\), \(\Delta_\varepsilon=1.412625(10)\) and \(\Delta_{\varepsilon'}=3.82951(61)\) come from the conformal bootstrap (El-Showk, Paulos, Poland, Rychkov, Simmons-Duffin, Vichi and others, 2012–2016). The conversions to critical exponents (\(\eta=2\Delta_\sigma-d+2\), \(\nu=1/(d-\Delta_\varepsilon)\), and so on) are standard; the exponent values in the table and the identity \(\alpha+2\beta+\gamma=2\) were computed for this article. The "experiment / Monte Carlo" column gives approximate representative values across several measurements and computations, not values from a single source. The 3d XY value \(\Delta_s=1.51136(22)\) is from the bootstrap work of Chester et al., and \(\alpha=-0.0127(3)\) is from the space-based experiment of Lipa et al. (2003); the disagreement between them is an unresolved problem. The AdS/CFT relation \(\Delta(\Delta-d)=m^2R^2\) for a scalar and the Breitenlohner–Freedman bound \(m^2R^2\ge-d^2/4\) are standard, but the AdS/CFT correspondence itself is an unproven conjecture. The bootstrap is a numerical method; "the Ising model sits on the kink" means agreement with Monte Carlo, not a proof. Real materials are not exact CFTs and show these exponents only near the critical point. ── To print, use your browser's Print → Save as PDF (sliders freeze and answers are hidden in the print version).

Print / save as PDF: ⌘+P (Ctrl+P on Windows). On screen, the two sliders move \(\Delta\); try to find the point where all six bars land on their marks. "Show the answer" opens each solution.