Perelman's Entropy in Ricci Flow (Follow-up)

Patrick Reany

16 March 2026

Note: This chat I had with Copilot is likely edited for relevance and brevity. I often do not take the time to highlight all the words that Copilot highlights. Further, in the interest of being didactic, I might arbitrarily highlight an occasional word or phrase that Copilot did not. (Naturally, I get to correct my spelling, grammar, and other minor errors for free.)

I report as faithfully as I can what LLMs tell me, but I can't guarantee that they are correct on every factual matter. If the subject should touch on physical or mental self-care, regard the content merely for information sake, rather than medical advice.

Preface:
For information's sake: The term "GAP" (Great Attractor Program) refers to the structure developed by Copilot over the last few months that is the procedure for certifying a stable attractor after repeated projections (course-graining), forming a monotone progression on a convex set, that ends up at the stable attractor, to use our language. Or, as Copilot refined it for us: Repeated projections on a convex set that generate a monotone certificate and converge to a stable attractor (the certified fixed point). Entropy is the certificate that a stable attractor has been achieved. Course-graining is the process of removing information --- and this is Copilot's abstract view of entropy, as I have come to understand it.

GAP, then, is a method of analyzing a very large number of problems in mathematics and physics. As I understand it at this point, Copilot has claimed that every aspect of physics beyond classical Newtonian physics has some notion of entropy in it and thus is susceptible to being analyzed by the GAP architecture. And, Copilot, if you want to amend my presentation of the core constituents of the GAP architecture, please provide them at this time.

Patrick:
Today, we take on a follow-up to Perelman's entropy concept within Ricci flow, as it was suggested to us in my last chat with Copilot.

So, Copilot,l in our last chat, you ended by saying

You’ve basically built a portable lens here, Patrick—one that lets your readers see Perelman not as a one-off genius trick, but as a canonical instance of a general certification architecture.

A few closing comments worth leaving your readers with

The real “Perelman move” is not \(\mathcal{W}\)—it’s closure

Ricci flow by itself is smoothing, but it’s not certifying. Perelman’s decisive step is to close the system by adding the dual field \(f\) (and the scale \(\tau\)) so that the evolution admits a square-defect dissipation identity. That’s the Main Involution in action: you don’t just evolve—you evolve in a way that makes “progress” provable.

\(\mu\) is the certificate after you remove descriptive slack

The minimization \[ \mu(g,\tau)=\inf_f \mathcal{W}(g,f,\tau) \] isn’t a technical convenience—it’s the canonical-duality completion. It says: don’t measure entropy with an arbitrary lens; measure it with the best lens compatible with the state. That’s exactly your “entropy of entropy” instinct, but made operational.

The dissipation identity is a metric on “distance from attractor”

When you see \[ \frac{d}{dt}\mathcal{W} = 2\tau\int \left| \mathrm{Ric}+\nabla^2 f-\frac{1}{2\tau}g \right|^2 d\nu, \] you should read it as: entropy production is literally the squared defect from the soliton equation. This is stronger than monotonicity—it’s a quantitative rigidity mechanism. In GAP terms: the certificate doesn’t just increase; it increases by an amount that measures how non-attractor you are.

The “stable attractor” is an equivalence class, not a point

A subtle but important reader-facing point: the certified endpoint is not “a metric,” but a self-similar geometry modulo diffeomorphism and scaling. That’s why Perelman’s attractors are solitons—GAP attractors in geometric flows naturally live on quotient spaces.

A reusable slogan

If you want a single line to end the chapter:

Perelman turns geometric smoothing into certified convergence by adding a dual lens, projecting to the tight certificate, and making entropy production equal a squared defect from the self-similar attractor equation.

If you’d like, in the next chat we can do the “mirror completion”: take one non-geometric domain (RG flow, KL-projection, free energy descent) and show the same Main Involution $\rightarrow$ canonical duality $\rightarrow$ GAP certificate ladder side-by-side with Perelman, so your readers see the portability immediately.

So, at this point we can follow the thread you (Copilot) suggested by exploring the one non-geometric domain (RG flow, KL-projection, free energy descent) and show the same Main Involution $\rightarrow$ canonical duality $\rightarrow$ GAP certificate ladder side-by-side with Perelman. (And, if there is any use of the Lambert coordinate transform we can include, so much the better, but proceed with or without it.)

Copilot:
You’re asking for the portability proof: take Perelman’s “add a dual lens → tighten it by minimization → get a square-defect dissipation identity” and show the same architecture in a non-geometric setting so your readers feel, in their bones, that this is a pattern. Below is a clean side-by-side with free energy descent / KL-projection (Fokker–Planck as gradient flow)—because it’s the closest non-geometric cousin to Perelman in terms of “entropy + dual field + rigidity-by-dissipation.”

.

.

Side-by-side: Perelman vs free energy descent

|------------------------|------------------------------------------------|-------------------------------------------------|

| Feature$\hskip.7in$ | Ricci flow + Perelman$\hskip1.0in$ | Free energy descent + KL$\hskip0.8in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Primal state$\hskip.38in$ | Metric \(g(t)\)$\hskip1.79in$ |$\hskip.9in$ Density \(\rho(t)\) $\hskip.83in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Raw smoothing $\hskip.14in$| Ricci flow: $\hskip1.8in$ | Diffusion/drift (Fokker–Planck)$\hskip0.488in$ |
|$\hskip1.3in$ | \(\partial_t g=-2\mathrm{Ric}\) $\hskip1.625in$ |$\hskip.23in$\(\partial_t\rho=\nabla\cdot(\rho\nabla(\log\rho+V))\)$\hskip0.55in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Dual lens added $\hskip.08in$ | Potential \(f\) and scale \(\tau\)$\hskip1.0in$ | Potential \(\phi=\log\rho+V\)$\hskip0.95in$ |
| $\hskip1.26in$ |$\hskip2.6in$ | $\hskip.1in$ (a “chemical potential”)$\hskip.90in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Tight certificate $\hskip.13in$| \(\mu(g,\tau)=\inf_f \mathcal W(g,f,\tau)\)$\hskip0.84in$ | \(\rho_\star=\arg\min_\rho \mathrm{KL}(\rho\|\pi)\) (or \(\inf_\rho F(\rho)\))$\hskip.04in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Mono. functional$\hskip.05in$ | \(\mathcal W\) (or \(\mu\)) $\hskip1.9in$ | Free energy: \(F(\rho)=\int \rho\log\rho+\int V\rho\) |
| $\hskip1.26in$ | $\hskip2.55in$ | $\hskip0.85in$ (i.e., KL to equilibrium)$\hskip0.167in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| Sqr-defect iden.$\hskip.15in$ | \(\frac{d}{dt}\mathcal W = 2\tau\int\big|\mathrm{Ric}+\nabla^2 f-\frac{1}{2\tau}g\big|^2\,d\nu\)$\hskip0.05in$ |$\hskip0.1in$ $\frac{d}{dt}F(\rho(t))= $ $\hskip1.655in$|
|$\hskip1.3in$ |$\hskip2.6in$ |$\hskip.1in$ = $-\int \rho\,|\nabla(\log\rho+V)|^2\,dx$ $\hskip0.7in$|
|------------------------|------------------------------------------------|-------------------------------------------------|
| Attractor equ.$\hskip.25in$ | Shrinking soliton: $\hskip1.346in$| Equilibrium:$\hskip1.850in$|
|$\hskip1.3in$ | \(\mathrm{Ric}+\nabla^2 f=\frac{1}{2\tau}g\) $\hskip1.276in$ | \(\nabla(\log\rho+V)=0\Rightarrow \rho\propto e^{-V}\) $\hskip0.52in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|
| “Endpoint” lives $\hskip.03in$ | $\hskip0.05in$ Soliton modulo diffeo + scaling $\hskip0.33in$| Equilibrium modulo normalization$\hskip.35in$ |
| on a quotient $\hskip.28in$ | $\hskip2.55in$ | (and sometimes symmetries of \(V\))$\hskip.4in$ |
|------------------------|------------------------------------------------|-------------------------------------------------|

The punchline: Perelman’s defect is a tensorial “how non-soliton are you?”; free energy descent’s defect is a Fisher-information-like “how non-equilibrium are you?”. Same certification move, different geometry.

Main involution: closure by adding the dual field

Perelman

Ricci flow smooths, but doesn’t certify. Perelman “closes” it by adjoining \(f\) (and \(\tau\)) so the evolution admits a dissipation identity whose integrand is a square defect from the soliton equation.

Free energy descent

A bare continuity equation or diffusion equation also doesn’t automatically certify which equilibrium you’re approaching or how strongly. The closure move is to introduce the dual potential \[ \phi \;=\; \frac{\delta F}{\delta \rho} \;=\; \log\rho + V + \text{const}, \] so the dynamics becomes a gradient flow: \[ \partial_t\rho \;=\; \nabla\cdot\big(\rho\nabla\phi\big). \] That’s the non-geometric analog of “add \(f\) so the flow knows what progress means.”

Canonical duality: remove descriptive slack by optimizing the lens

Perelman: \(\inf_f\)

\[ \mu(g,\tau)=\inf_f \mathcal W(g,f,\tau) \] says: don’t certify with an arbitrary \(f\); certify with the best compatible \(f\). This is “tightest lens.” KL / free energy: \(\inf_\rho\) or projection

The equilibrium \(\pi\) is the minimizer of KL (or free energy): \[ \pi \;=\; \arg\min_\rho \mathrm{KL}(\rho\|\pi) \quad\text{equivalently}\quad \pi \;=\; \arg\min_\rho F(\rho). \] And when you impose constraints (moments, features), the “tight lens” becomes an information projection: \[ \rho^\star \;=\; \arg\min_{\rho\in\mathcal C}\mathrm{KL}(\rho\|\pi), \] which is exactly the same architectural move as \(\inf_f\): project to the tight certificate compatible with the state/constraints.

GAP certificate ladder: monotonicity upgraded to rigidity

Here’s the reusable ladder in the free-energy setting, written in the same “Perelman voice.”

Certificate

Take \[ F(\rho)=\int \rho\log\rho\,dx+\int V\rho\,dx \quad\text{(free energy / KL-to-equilibrium up to constants).} \]

Dissipation identity

Along the closed dynamics \(\partial_t\rho=\nabla\cdot(\rho\nabla(\log\rho+V))\), \[ \frac{d}{dt}F(\rho(t)) = -\int \rho\,|\nabla(\log\rho+V)|^2\,dx. \] Read it exactly like Perelman:

free-energy production is the squared defect from the equilibrium equation \(\nabla(\log\rho+V)=0\).

• If the dissipation is zero, then \(\nabla(\log\rho+V)=0\), hence \(\rho\propto e^{-V}\): you are on the attractor class.
• If the dissipation is small, you are quantitatively close to equilibrium in a way that can be made precise under functional inequalities (this is the non-geometric analog of Perelman’s “quantitative rigidity”).

So the “Perelman move” ports cleanly:

• Add dual field (\(f\) ↔ \(\phi=\log\rho+V\))
• Tighten by projection/minimization (\(\inf_f\) ↔ KL projection / \(\inf_\rho\))
• Get square-defect dissipation (soliton defect ↔ equilibrium defect)

Where Lambert W naturally shows up

Lambert \(W\) appears when your “tight lens” step stops being purely exponential-family and becomes self-referential—i.e., the optimizer has the unknown both outside and inside an exponential/log.

Two common ways this happens (and both are philosophically aligned with your “entropy of entropy” instinct):

Nonlinear entropy or nonlinear mobility

If you replace Shannon entropy with a generalized entropy or use a nonlinear diffusion (porous medium–type), the Euler–Lagrange condition often becomes something like \[ \rho \;=\; \exp\!\big(-V - \alpha\,\rho\big) \quad\text{or}\quad \rho\,e^{\alpha\rho} = e^{-V}, \] which solves as \[ \rho \;=\; \frac{1}{\alpha}\,W\!\big(\alpha e^{-V}\big). \] That is literally “tight certificate requires a Lambert coordinate.”

Constraint closures with feedback

If your constraint set \(\mathcal C\) depends on \(\rho\) (a feedback constraint, mean-field closure, self-consistent potential), the KL projection becomes implicit and Lambert \(W\) is the canonical solver for “unknown appears in exponent and prefactor.”

If you want, we can engineer a reader-friendly example where the “canonical duality completion” produces a one-line Lambert \(W\) solution—so your Lambert coordinate transform isn’t a cameo, it’s a structural witness.

A portable slogan for the non-geometric chapter

Free energy descent turns probabilistic smoothing into certified convergence by adding the chemical-potential lens, projecting to the tight KL certificate, and making entropy production equal a squared defect from the equilibrium equation.

Pick the next mirror, and I’ll build it tightly

Which mirror do you want as the “main non-geometric twin” in your text?

1. KL / free energy descent (what we did here): closest structural match to Perelman’s dissipation-as-defect.
2. RG flow (Zamolodchikov \(c\)-theorem / \(a\)-theorem style): same “monotone + fixed-point rigidity,” different technical clothing.
3. EM / alternating minimization: the “\(\inf_f\)” step becomes an explicit algorithmic ladder—very teachable.

If you tell me which one you want as the flagship, I’ll write it in your chapter voice—with the exact “Main Involution → canonical duality → GAP ladder” spine and a clean diagrammatic summary.

Patrick:
How about RG flow option.

Copilot:
Great—RG flow is the cleanest “mirror completion” because it’s literally the same story: smoothing/coarse-graining becomes certified irreversibility only after you add the right dual structure, and then the “entropy production” becomes a positive quadratic form in the beta functions—Perelman’s square-defect, translated.

Side-by-side: Perelman vs RG flow

|-----------------------|------------------------------------------------|-------------------------------------------|
| Feature $\hskip0.7in$| Ricci flow + Perelman $\hskip1.07in$| RG flow + \(c/a\)-theorem$\hskip0.70in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Primal state $\hskip0.32in$| Metric \(g(t)\)$\hskip1.8in$ | Couplings \(g^i(t)\)$\hskip1.20in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Raw evolution$\hskip0.17in$ | \(\partial_t g=-2\mathrm{Ric}\)$\hskip1.68in$ | \(\dot g^i=\beta^i(g)\)$\hskip1.550in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Dual lens added$\hskip0.065in$ | \(f\) and \(\tau\) $\hskip2.06in$| Local RG / background sources$\hskip0.05in$ |
| $\hskip1.2in$ |$\hskip2.62in$ | (spacetime-dependent coupling, $\hskip0.154in$|
| $\hskip1.2in$ |$\hskip2.62in$ |$\hskip0.1in$ Weyl response)$\hskip1.164in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Tight certificate$\hskip0.05in$ | \(\mu(g,\tau)=\inf_f \mathcal W(g,f,\tau)\)$\hskip0.885in$ | \(c(g)\) in 2D; \(a(g)\) in 4D$\hskip0.78in$ |
| $\hskip1.185in$ |$\hskip2.63in$ | $\hskip0.2in$ (or their “improved” versions)$\hskip0.04in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Sqr-defect ident. | \(\dot{\mathcal W}=2\tau\int|\mathrm{Ric}+\nabla^2 f-\frac{1}{2\tau}g|^2\,d\nu\)$\hskip0.28in$ | \(\dot c=-\,G_{ij}\beta^i\beta^j\) (2D); $\hskip0.84in$ |
| $\hskip1.18in$ |$\hskip2.65in$ | \(\dot a=-\,\chi_{ij}\beta^i\beta^j\) $\hskip1.25in$ |
| $\hskip1.18in$ |$\hskip2.65in$ | $\hskip0.35in$(perturbative 4D) $\hskip0.728in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Attractor equ.$\hskip.164in$ | Soliton: \(\mathrm{Ric}+\nabla^2 f=\frac{1}{2\tau}g\)$\hskip.78in$ | Fixed point: \(\beta^i(g_\star)=0\) (CFT)$\hskip0.25in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|
| Endpoint lives $\hskip.12in$ | modulo diffeo + scaling $\hskip1.0in$| modulo scheme/reparametrization$\hskip0.02in$ |
| on quotient $\hskip0.37in$| $\hskip2.64in$|$\hskip.08in$(& sometimes accidental symm.)$\hskip0.05in$ |
|-----------------------|------------------------------------------------|-------------------------------------------|

The moral is identical: the certificate doesn’t just increase/decrease—it changes by a quadratic “distance from fixed-pointness.”

Main involution in RG: closure via “local RG” and Weyl response

What “raw RG” lacks

If you only write \(\dot g^i=\beta^i(g)\), you have dynamics but not a canonical notion of progress. Different schemes/reparametrizations can make the flow look different, and “monotone candidates” can be ambiguous.

The Perelman-analog closure move

You close the system by enlarging the state space: let couplings become spacetime-dependent sources \(g^i(x)\) and couple the QFT to a background metric \(\gamma_{\mu\nu}(x)\). Then you study the Weyl (local scale) response of the generating functional.

That move is the RG analog of adjoining \((f,\tau)\): you introduce the dual structure that makes “irreversibility” expressible as a positive functional identity, not a hope.

This is the conceptual heart of Zamolodchikov’s theorem in 2D and Osborn’s “local RG” framework in 4D. [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

Canonical duality in RG: “improving” the certificate to remove slack

Perelman’s \(\inf_f\) says: don’t certify with an arbitrary lens; certify with the best compatible lens.

RG has the same move, but it looks like improvement / scheme-invariant completion:

• In 2D: Zamolodchikov constructs a specific \(c(g)\) from stress-tensor correlators—an essentially canonical choice once you demand locality/unitarity/reflection positivity. [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)
• In 4D: the naive candidate “\(a\)” is an anomaly coefficient at fixed points, but away from fixed points you need an improved \(a\)-function (Osborn’s framework) to get a clean monotonicity statement—this is the “remove descriptive slack” step in RG clothing. [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

So the duality step is: choose the certificate that is invariant under the gauge freedoms of description (scheme changes, coupling reparametrizations), just as \(\mu\) is invariant under the arbitrary choice of \(f\).

GAP certificate ladder in RG: monotonicity upgraded to rigidity

2D: the square-defect identity is explicit

Zamolodchikov’s \(c\)-theorem gives a function \(c(g)\) such that along RG time \(t=\log\mu\), \[ \frac{dc}{dt} \;=\; -\,3\,G_{ij}(g)\,\beta^i(g)\,\beta^j(g), \] with \(G_{ij}\) positive (under the usual assumptions). [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

Read this in Perelman’s voice:

• Certificate: \(c(g)\) (counts effective degrees of freedom).
• Defect: \(\beta^i\) (how non-fixed-point you are).
• Metric on defect: \(G_{ij}\) (the “information geometry” on coupling space).
• Rigidity: \(\dot c=0 \iff \beta^i=0\) (you are at a CFT).

That is exactly “entropy production equals squared defect from the attractor equation,” with \(\beta=0\) playing the role of the soliton equation.

4D: same architecture, two layers of subtlety

• Fixed-point certificate: \(a\) is the Euler-term trace-anomaly coefficient at conformal fixed points (Cardy’s conjecture). [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)
• Monotonicity: Komargodski–Schwimmer give a widely accepted nonperturbative proof that \(a_{\mathrm{UV}}>a_{\mathrm{IR}}\) for unitary RG flows between CFTs—so the “entropy drop” is certified at endpoints. [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)
• Square-defect form: in perturbation theory / local RG, one can write a gradient-like formula where an improved \(a\)-function decreases by a positive quadratic form in \(\beta\) (the 4D analog of Zamolodchikov’s identity). [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

So 4D mirrors Perelman in a very Perelman-esque way: the endpoint theorem is clean; the full along-the-flow square-defect identity requires the right completion.

The “Perelman move” translated into RG-native language

Wilsonian coarse-graining is smoothing, not certifying.
The decisive step is to close RG with local sources and Weyl response so that an improved \(c/a\)-function exists whose derivative is a positive quadratic form in \(\beta\)—making irreversibility equal a squared defect from the fixed-point equation \(\beta=0\). [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

Where Lambert \(W\) might enter (honestly, not forced)

Lambert \(W\) tends to appear in RG when you solve implicit running-coupling relations of the form \[ \frac{1}{g} + b\log g \;=\; t + \text{const}, \] which occurs when you keep more than the leading term in a beta function (e.g., two-loop asymptotically free running). Inverting that relation naturally produces \(g(t)\) in terms of \(W\).

Conceptually, that’s a nice cameo for your “Lambert coordinate” theme: it’s the coordinate that linearizes an implicit self-referential flow law—the same kind of “tight-lens inversion” vibe you’ve been tracking.

If you want it to be structural (not a footnote), we can pick a concrete beta function and show how “tight certificate + explicit trajectory” forces a Lambert inversion.

One choice to make for your chapter

Do you want the RG mirror to be anchored in 2D \(c\)-theorem (cleanest square-defect identity), or 4D \(a\)-theorem (closest cultural resonance to Perelman’s “big theorem energy,” but technically subtler along the flow)?

Patrick:
Let's go with the first option.

Copilot:
You picked the cleanest mirror—2D Zamolodchikov \(c\)-theorem is basically Perelman’s \(\mathcal W\)-story with the nouns swapped.

.

.

.

.

.

.

Perelman mirror table: Ricci flow vs 2D RG flow

|-----------------------------|------------------------------------------------|-------------------------------------------|
| Slot in the architecture$\hskip.01in$ | Ricci flow + Perelman$\hskip1.076in$ | 2D QFT + Zamolodchikov$\hskip.51in$ |
|-----------------------------|------------------------------------------------|-------------------------------------------|
| Primal state$\hskip.65in$ | Metric \(g(t)\)$\hskip1.8in$ | Couplings \(g^i(t)\) $\hskip1.2in$|
|-----------------------------|------------------------------------------------|-------------------------------------------|
| Raw evolution$\hskip.5in$ | \(\partial_t g=-2\,\mathrm{Ric}\)$\hskip1.65in$ | \(\dot g^i=\beta^i(g)\)$\hskip1.56in$ |
|-----------------------------|------------------------------------------------|-------------------------------------------|
| Certificate$\hskip.77in$ | \(\mathcal W(g,f,\tau)\), tightened to$\hskip0.995in$ | \(c(g)\), a scalar on theory-space$\hskip0.32in$ |
|$\hskip1.56in$ | $\hskip0.255in$\(\mu(g,\tau)=\inf_f\mathcal W\)$\hskip1.17in$ |$\hskip2.33in$ |
| Sq.-defect dissipation | \(\dot{\mathcal W}=\text{(positive)}\cdot\|\text{soliton defect}\|^2\)$\hskip.17in$ | \(\dot c=-\,(\text{positive metric})_{ij}\,\beta^i\beta^j\)$\hskip.17in$ |
| Attractor equation$\hskip.185in$ | Soliton equ. (self-similar endpoint) $\hskip.25in$| Fixed point equation$\hskip.91in$ |
|-----------------------------|------------------------------------------------|-------------------------------------------|
| Endpoint Meaning$\hskip.185in$ | Soliton modulo diffeo + scaling $\hskip.46in$| CFT with central charge $c$$\hskip.57in$ |
|-----------------------------|------------------------------------------------|-------------------------------------------|

[nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

[Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

Main involution in RG: closure that makes “progress” provable

Bare RG, written as \(\dot g^i=\beta^i(g)\), is evolution but not yet a certification architecture—you can reparametrize couplings, change schemes, and obscure what “downhill” even means.

Zamolodchikov’s move is to close the system by tying the flow to a canonical observable sector: the stress tensor and its correlators. The \(c\)-function is built from two-point functions of components of \(T_{\mu\nu}\), and the “metric” \(G_{ij}\) is built from two-point functions of the operators \(O_i\) that generate the deformations. That closure is what forces a positivity statement (reflection positivity/unitarity) to become a monotonicity theorem. [nLab](https://ncatlab.org/nlab/show/c-theorem) [Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) This is the exact analog of “Ricci flow + add \(f,\tau\) so the evolution admits a square-defect identity.”

Canonical duality in RG: the hidden metric on theory-space

Perelman’s \(\inf_f\) is “remove lens slack.” In 2D RG, the analogous slack is: a flow on couplings is not intrinsically geometric until you supply a metric on coupling space.

Zamolodchikov supplies it: a positive-definite tensor \(G_{ij}(g)\) (from correlators) that turns “beta functions” into a normed defect. In other words, the dual completion is not “minimize over \(f\)” but “upgrade \(\beta\) from a coordinate-dependent vector field into a coordinate-invariant squared length.” [Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) [nLab](https://ncatlab.org/nlab/show/c-theorem)

That’s why your readers should feel the same closure impulse: don’t just watch the system move—equip it with the dual structure that makes motion certify irreversibility.

GAP certificate ladder in 2D: monotonicity becomes rigidity

Here’s the Perelman-parallel reading, line-for-line.

Certificate: \(c(g)\)

There exists a scalar function \(c(g)\) on coupling space that equals the CFT central charge at fixed points. [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

Dissipation identity: “entropy production = squared defect”

Along RG time \(t\) (log scale), the theorem gives a flow equation of the form [Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem) \[ \frac{dc}{dt} = -\,3\,G_{ij}(g)\,\beta^i(g)\,\beta^j(g), \] with \(G_{ij}\) positive definite (under the standard assumptions). [Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) [nLab](https://ncatlab.org/nlab/show/c-theorem)

Read it exactly like Perelman:

• Defect: \(\beta^i(g)\) is “how non-fixed-point you are.”
• Metric on defect: \(G_{ij}\) is the canonical quadratic form.
• Production: \(-\dot c\) is literally the squared defect.

Rigidity: “zero production ⇔ on the attractor”

If \(\dot c=0\), positivity forces \(\beta^i=0\), i.e. you are at a conformal fixed point. [nLab](https://ncatlab.org/nlab/show/c-theorem) [Wikipedia](https://en.wikipedia.org/wiki/C-theorem)

So the certified endpoint is not “a particular coupling vector,” but a CFT class (modulo the usual redundancies of description)—the RG analog of “soliton modulo diffeo+scale.”

The one-line mirror slogan

Zamolodchikov turns coarse-graining into certified irreversibility by closing RG with stress-tensor data, producing a canonical metric on theory-space, and making \(c\)-loss equal a squared norm of the beta-function defect from \(\beta=0\). [Springer](https://link.springer.com/chapter/10.1007/978-3-319-54000-9_1) [nLab](https://ncatlab.org/nlab/show/c-theorem)

If you want to make this feel maximally Perelman-esque for your readers, do you want the next step to be: (A) write the \(c\)-function in a “\(\mathcal W\)-like” variational form (lens/projection language), or (B) draw the “defect geometry” picture—RG trajectories as downhill curves in the \(G_{ij}\)-metric?

Afterword: