c·t = CONST, THAT CLICKS EPISODE 44 ／ Part V — where the tool breaks
What was demanded was never absence
The scale only has
to become irrelevant
A lattice spacing is a smallest length, yet conformal field theories are computed on lattices.
Because conformal invariance belongs to the fixed point, not to the lattice.
Last time we found that a theory with a smallest length cannot be conformally invariant. And yet, as a matter of fact — conformal field theories are computed on lattices every day, and a lattice spacing is a smallest length. This time we count out why there is no contradiction. And there, at last, is the reason the tool that has been breaking all through Part V survives anyway.
01The apparent contradiction
02The core — the scale need not be absent, only irrelevant
observables can depend only on the dimensionless ratio \(\xi/a\) (Episode 3) — and at criticality \(\xi/a\to\infty\)
The main point of this episode
Conformal invariance is not a property of the lattice but of the fixed point the lattice theory flows to.
The scale is not absent; it becomes irrelevant.
── This is Episode 3's procedure exactly: the physics is in the ratio, and when the ratio diverges the answer stops depending on \(a\).
03How well do they agree? — the 3D Ising model
| Exponent | Conformal bootstrap (continuum) | Lattice Monte Carlo | Relative difference | \(\sigma\) |
|---|---|---|---|---|
| \(\nu\) | \(0.6299709(4)\) | \(0.63002(10)\) | \(7.8\times10^{-5}\) | \(0.5\) |
| \(\eta\) | \(0.0362978(20)\) | \(0.03627(10)\) | \(7.7\times10^{-4}\) | \(0.3\) |
Kos et al. (2016) for the bootstrap, Hasenbusch (2010) for the lattice. The lattice answer and the continuum answer agree to four digits in \(\nu\) and three in \(\eta\) (both within \(1\sigma\)) — the "smallest length" of the lattice spacing is nowhere in the answer.
04Universality — different insides, the same numbers
The details of the lattice and the details of the molecules have both vanished from the answer — that is what "becoming irrelevant" means experimentally.
05How fast does it vanish? — the correction exponent \(\omega\)
| \(\xi/a\) | residual error \((a/\xi)^\omega\) | bits |
|---|---|---|
| \(10\) | \(1.5\times10^{-1}\) | \(2.8\) |
| \(10^2\) | \(2.2\times10^{-2}\) | \(5.5\) |
| \(10^3\) | \(3.2\times10^{-3}\) | \(8.3\) |
| \(10^4\) | \(4.8\times10^{-4}\) | \(11.0\) |
| \(10^6\) | \(1.0\times10^{-5}\) | \(16.5\) |
Reaching six digits (\(10^{-6}\)) would need \(\xi/a>1.7\times10^7\) — no lattice that large fits. In practice one uses improved actions (which cancel the leading correction) and extrapolation. The agreement in §03 is not the achievement of a naive lattice computation; it is the achievement of the work done to cancel the corrections.
Figure: how fast the lattice's traces vanish. Refining the lattice reduces the error as \((a/\xi)^{0.83}\), but six digits would need \(\xi/a>10^7\), so improved actions cancel the leading correction instead. Move the slider.
06The link to Episode 14
07Where it did not work — the \(\lambda\) transition of helium
difference \(0.00255\), combined error \(0.00042\) → \(6.0\sigma\), i.e. 29.0 bits on Episode 19's scale
The measurement was made aboard the Space Shuttle (Lipa et al. 2003), to avoid gravitational pressure gradients. This discrepancy is unresolved. Whether universality fails, or a systematic error lies on the experimental or the theoretical side, is not settled — universality works very well, but it is not universal magic.
08In the end this too was (A)/(B)
| Reading | The limit | Conformal invariance | Consequence |
|---|---|---|---|
| (A) \(a\) is a regulator to be removed | take the continuum limit | exact at the fixed point | the lattice is a tool |
| (B) \(a\) is physical (spacetime really is discrete) | do not remove it | approximate | violations could be observed |
Episode 36's (A)/(B) applies to the lattice spacing as well. And the GRB constraints of Episode 43 §05 were a measurement testing (B).
Part V's answer
Episode 43: a smallest length is incompatible with conformal invariance (excluded by hypothesis).
Episode 44: and the tool survives anyway — provided the scale becomes irrelevant.
── What the tool demanded was never "no scale", but "no scale left in the answer".
If it is irrelevant in the renormalisation-group sense, conformal invariance returns at long distances to any accuracy you like.
(1) §03's agreement is a consequence of the lattice side having taken the continuum limit. These are not raw numbers from a finite lattice but values extrapolated from several lattice sizes with improved actions cancelling the leading correction — per §05, a naive computation on a \(10^3\) lattice still carries a 0.3 per cent error. "The lattice gives the same answer" properly means "carefully taking the continuum limit gives the same answer".
(2) §02's "irrelevant" is renormalisation-group jargon. It does not mean "beside the point" but an operator whose coefficient shrinks under the RG transformation — and which operators are irrelevant differs from theory to theory and is not obvious. For gravity, that is exactly the open question of Episode 35's asymptotic safety.
(3) §07's helium discrepancy is still under discussion. Systematic errors on the experimental side (finite-size effects, temperature control) and the theoretical error budget have both been suggested — this document claims nothing beyond "it is unresolved". The \(6.0\sigma\) simply combines the quoted errors and would move substantially with a different systematic estimate.
(4) §04's universality examples are a textbook summary. In real experiments the number of reliable digits depends on how the critical point is approached and how corrections are handled — not all of these are confirmed to the same precision, and the liquid–gas exponents are not known as precisely as the magnetic ones.
(5) §08's (B) is not a claim that spacetime is discrete. It is a statement that the reading is logically available; current observations only exclude the linear effect (Episode 43 §05) — whether spacetime is discrete is unresolved, and this document endorses neither answer.
Exercises
- The lattice has a smallest length, so why can conformal field theories be computed on it?
Show the answer
Because conformal invariance is a property of the fixed point the lattice theory flows to, not of the lattice. Observables can depend only on \(\xi/a\), which diverges at criticality, so \(a\) drops out of the answer — the scale is not absent but irrelevant. - How many bits does halving the lattice spacing buy?
Show the answer
0.83 bits (the 3D Ising correction exponent \(\omega=0.8303\)). Even at \(\xi/a=10^3\) a 0.3 per cent error remains (8.3 bits), and six digits would need \(\xi/a>1.7\times10^7\), beyond any lattice — hence improved actions and extrapolation. - What is the experimental meaning of universality?
Show the answer
That completely different microscopic contents give identical exponents — lattice spins, the liquid–gas critical point of water, binary fluids and uniaxial ferromagnets all give \(\nu=0.6300\). The details of the lattice and of the molecules have vanished from the answer: that is what "irrelevant" looks like in the laboratory. - Is there a case where universality is in question?
Show the answer
The \(\lambda\) transition of \(^4\)He. The 3D XY class gives \(\alpha=2-3\nu=-0.01525(30)\), while the Space Shuttle experiment gives \(-0.0127(3)\) — a \(6.0\sigma\), 29-bit discrepancy. It is unresolved; whether universality fails or a systematic error is responsible is not settled. - (Harder) Taking Episodes 43 and 44 together, what does the tool actually demand?
Show the answer
Not "no scale" but "no scale left in the answer". As Episode 43 showed, a smallest length excludes exact conformal invariance; but if that scale is irrelevant in the RG sense, conformal invariance returns at long distances to any accuracy you like — that is Part V's answer.
Summary: what the tool demanded was never absence
Episode 43 showed that a theory with a smallest length cannot be conformally invariant. Yet conformal field theories are computed on lattices. The reason there is no contradiction: conformal invariance is a property of the fixed point the lattice theory flows to, not of the lattice. Observables can depend only on \(\xi/a\), which diverges at criticality, so \(a\) drops out of the answer. The scale is not absent; it becomes irrelevant.
In practice, the 3D Ising exponents from the continuum bootstrap and from lattice Monte Carlo agree to four digits in \(\nu\) (\(0.5\sigma\)). And there is universality — lattice spins, the liquid–gas critical point of water, binary fluids, uniaxial ferromagnets, all with the same \(\nu=0.6300\). The details of the lattice and of the molecules have vanished from the answer.
But the vanishing has a speed. Lattice traces die as \((a/\xi)^{0.83}\) — 0.83 bits per halving of the spacing. Six digits would need \(\xi/a>1.7\times10^7\), which no lattice can hold. The agreement in §03 is not the achievement of a naive computation but of improved actions and extrapolation — the work of cancelling the corrections.
And there is a case where it did not work — the \(\lambda\) transition of \(^4\)He. Against the 3D XY prediction \(\alpha=-0.01525(30)\), the Space Shuttle experiment gives \(-0.0127(3)\): a \(6.0\sigma\), 29-bit discrepancy, still unresolved. Universality works very well, but it is not universal magic.
In the end this too was (A)/(B) — is \(a\) a regulator to be removed, or a physical discreteness? Episode 43's GRB constraints were a measurement of (B). Part V's answer: what the tool demanded was never "no scale", but "no scale left in the answer".
This document is Episode 44 of "c·t = const, That Clicks" (the eighth of Part V), written for physics-minded high-school and university readers. The renormalisation group, universality, the continuum limit of lattice theories and the conformal bootstrap are all established standard material and nothing here is a new claim — the numbers are computed in kenshou/calc48.py. The values quoted are the 3D Ising bootstrap of Kos, Poland, Simmons-Duffin & Vichi (2016) and the lattice Monte Carlo of Hasenbusch (2010), the 3D XY bootstrap of Chester et al. (2019), and the \(^4\)He experiment of Lipa et al. (2003). §03's agreement follows from the lattice side having taken the continuum limit — these are not raw finite-lattice numbers but values extrapolated from several sizes with improved actions cancelling the leading correction, and a naive computation on a \(10^3\) lattice still carries a 0.3 per cent error. §02's "irrelevant" is renormalisation-group jargon, and which operators are irrelevant differs between theories and is not obvious — for gravity that is exactly the open question of Episode 35. §07's helium discrepancy is still under discussion, with systematic errors suggested on both the experimental and the theoretical side; this document claims nothing beyond "it is unresolved", and the \(6.0\sigma\) simply combines the quoted errors. §04's universality examples are a textbook summary and not all are confirmed to the same precision. §08's (B) is not a claim that spacetime is discrete but a statement that the reading is logically available — whether spacetime is discrete is unresolved, and this document endorses neither answer. The "improved action" curve in the figure is schematic and does not represent the performance of any particular method. ── To make a PDF, use your browser's Print dialogue (sliders freeze and answers are hidden in the print version).