⚜ PRINCIPIA ORTHOGONA · Vol VI · Roots · WP-93 ← WP-92 · Not a Cusp: Percolation's Own Fold · WP-94 · One Hole, Five Words →
#Scoping Note
Vol VI · Roots · WP-93 · Received 2026-09-02 · Companion to WP-92 · Open

The Triangle, Not the Bubble

Scoping the real obstruction to proving mean-field percolation exponents on $\mathbb{Z}^d$ for $7\le d\le10$ — and, first, fixing a diagrammatic misattribution from an earlier pass at this note, which blamed the wrong two-propagator sum for a threshold that is actually produced by a three-propagator one.
CompanionWP_percolation_cusp_catastrophe.md
WP-92 · geometry/book6/
Methoddirect power-counting + literature check
AI-assisted verification, not AI-generated math
Claim typescoping note
what would close the gap, not a claimed closing
Statusgap open
this page proves nothing new about $7\le d\le10$
Erratum, stated first

A draft pass at scoping this problem attributed the $1/(d-6)$ divergence responsible for percolation's obstruction to $\|G\|_2^2=\int(1-\hat D(k))^{-2}dk$ — a two-propagator (bubble) integral. Direct power counting shows that integral's own threshold is d=4, not 6 (see §1) — it is the self-avoiding-walk/Ising value, not percolation's. The diagram that actually produces dc=6 is the triangle sum, three propagators, and that correction is carried through §§2–3 below.

Nobody doubts the physics: mean-field exponents (β=1, γ=1, δ=2, η=0, ν=1/2 — WP-92, §1–§2) are expected to hold for nearest-neighbor percolation on $\mathbb{Z}^d$ at every $d>6$, including the unproven window $7\le d\le10$. What is scoped here is the separate, narrower question: why the existing proof technique (the lace expansion, closed numerically by NoBLE) stops at $d\ge11$ (Fitzner–van der Hofstad, 2017) rather than reaching all the way to 7, and what closing that gap would actually require.
COMPUTED power-counting worked out directly on this page DEFECT corrects an earlier draft of this scoping note OPEN not resolved by anyone, as far as this note found
§1

Which diagram actually gives $d_c=6$ COMPUTED

The free/mean-field two-point function decays as G(x) ~ |x|^{-(d-2)} (the discrete Newtonian potential), with Fourier transform Ĝ(k) ~ |k|^{-2} near k=0. Two candidate diagrams, two different thresholds:

diagrampropagatorsnear-0 integrandpower countingthreshold
Bubble, ‖G‖₂²2|k|⁻⁴∫r^{d-5}drd = 4
Triangle, ΣG(x)G(x−y)G(y)3|k|⁻⁶∫r^{d-7}drd = 6

The bubble's threshold, d=4, is the self-avoiding-walk and Ising upper critical dimension — a real number, just not percolation's. Percolation's d_c=6 comes from the Aizenman–Newman triangle condition (∇(p_c)=Σ_{x,y}τ(x)τ(x−y)τ(y)<∞), the three-propagator sum, not the two-propagator bubble. Any scoping note that blames a two-propagator quantity for the d=6 threshold has the wrong diagram, regardless of how correct its surrounding architecture is.

§2

The corrected mechanism COMPUTED

Triangle sum: Σ_{x,y} G(x)G(x−y)G(y) = ∫ Ĝ(k)³ dk/(2π)^d Near k=0: Ĝ(k)³ ~ |k|⁻⁶ ⟹ ∫₀ |k|⁻⁶ d^dk ~ ∫ r^{d−7} dr Converges (d>6) — diverges as d → 6⁺, rate ~ 1/(d−6)

NoBLE's bootstrap (WP-92's companion note; Fitzner–van der Hofstad, 2017) closes a loop of three functions f₁,f₂,f₃ bounding ratios of percolation diagrams to their SRW counterparts, via numerical (interval-arithmetic) bounds on convolution integrals built from these propagator sums. As d→6⁺, the triangle-type sums that dominate the higher-order lace-expansion coefficients grow, and the numerical bounds that close the loop at d≥11 stop closing — not because the arithmetic gets less precise, but because the quantity being bounded is genuinely larger and closer to the threshold where it diverges.

What this note is not claiming

It is not claimed that the triangle sum itself is the only quantity NoBLE tracks, nor that fixing this diagram identification amounts to a proof strategy. The correction is narrower: it identifies which diagram's divergence is doing the structural work, so that any future scoping of this gap reasons about the right object.

§3

What closing $7\le d\le10$ would require OPEN

Fitzner and van der Hofstad's own account of the closely related NoBLE extension to lattice trees and lattice animals — same method, same authors, a different but adjacent model — states plainly that pushing their proven range further "would require substantially new ingredients and insights," not sharper interval arithmetic on the existing scheme. This note does not have the equivalent sentence in hand for percolation specifically (the percolation paper's own discussion section was not accessible when checked), so the claim below is an inference from the adjacent case, not a quotation about this one — flagged as such rather than presented as settled.

Candidate directions a genuine attack would need, in roughly increasing order of how much they change the existing method rather than tune it:

What this note does not supply

None of the three directions above is developed here. This is a scoping note: it says what the obstruction is and what shape a resolution would need to have, not a resolution.

§4
[UPDATE 2026-09-02] · two of these three directions have now been tested

Direction 1 — sign-alternating cancellation. Tested directly in Book 4 · The Lace Bootstrap in Intermediate Dimensions. The true zero-lace coefficient was computed at the Brillouin-zone corner for $\mathbb{Z}^7$: $\hat\Pi^{(0)}(k^*) = -84\,p^4 + O(p^6)$ by exact enumeration, and $-0.004782 \pm 0.000430$ by Monte Carlo — negative at $11.1\sigma$. A sign-alternating scheme of the form $\hat\Pi^{(2m)} - \hat\Pi^{(2m+1)} \ge 0$ therefore begins from a negative first term at the corner. The direction is not closed off, but it does not start where it was assumed to start.

Direction 3 — analytic isolation of the pole. Partly mis-stated here. The $1/(d-6)$ is the $d$-dependence of a convergent integral, not a pole in $k$ at fixed $d$. Evaluated exactly, $\Omega(7) = 0.7164$, with $\Omega\cdot(d-6)$ between $0.60$ and $0.72$ across $7 \le d \le 13$ — an $O(1)$ quantity at every integer dimension in range, and no singularity to extract. Isolating the small-$k$ region remains a reasonable numerical tactic; there is simply no pole there. The binding constant is the combinatorial $C$, not $\Omega$: descending from $d=11$ to $d=7$ costs a factor of $5.8$, not a divergence.

What is not claimed

References
  1. Aizenman, M., Newman, C. M. (1984). Tree graph inequalities and critical behavior in percolation models. Journal of Statistical Physics, 36, 107–143.
  2. Aizenman, M., Newman, C. M. (1989/1991). The triangle condition for percolation. Bulletin of the AMS, 21(2); and companion convergence-of-critical-exponents papers.
  3. Hara, T., Slade, G. (1990). Mean-field critical behaviour for percolation in high dimensions. Communications in Mathematical Physics, 128, 333–391.
  4. Fitzner, R., van der Hofstad, R. (2017). Mean-field behavior for nearest-neighbor percolation in d > 10. Electronic Journal of Probability, 22. arXiv:1506.07977.
  5. Fitzner, R., van der Hofstad, R. (2021). NoBLE for lattice trees and lattice animals. Source of the "substantially new ingredients and insights" assessment quoted in §3, for the adjacent (not identical) model.
  6. WP-92 · Not a Cusp: Percolation's Own Fold — the companion note this page scopes an open thread from.