Equivalent growth of N_DFL and 1/δ_TRL
AX_triangle_removal_vs_dfl_growth
Source: arXiv:2104.11626 — Removal lemmas and approximate homomorphisms
Informal Conjecture on similar growth of $N_{DFL}(\varepsilon)$ and $1/\delta_{TRL}(\varepsilon)$ · arXiv:2104.11626
Diagnostic tag ill-posed high confidence
The phrase “similar growth” specifies no asymptotic equivalence, and while the same-parameter Theta interpretation is false, plausible logarithmic or polynomial-equivalence versions remain open.
Caveat. The argument uses the standard normalization in which N_DFL bounds the number of triangles; edge-count and endpoint conventions only change constants or parameter rescalings.
Attack verdicts are unrefereed model self-reports. An ill-posed label is a diagnostic of the catalog statement supplied to the model, not a literature-review status or a peer-reviewed result.
Status open high confidence
The conjecture that $N_{DFL}(\varepsilon)$ and $1/\delta_{TRL}(\varepsilon)$ have similar growth remains open as of May 2026. No follow-up paper resolving or substantially advancing this informal conjecture was found in the literature. The gap between the known lower bound $(1/\varepsilon)^{c\log(1/\varepsilon)}$ (from the Behrend construction) and the upper bound tower$(O(\log(1/\varepsilon)))$ (Fox) remains unresolved. A 2025 paper (arXiv:2507.05231) improves the constant in the lower-bound construction for the triangle removal lemma but does not address the specific comparison between $N_{DFL}$ and $1/\delta_{TRL}$.
Reviewer notes. No follow-up found that directly addresses the conjecture. The paper arXiv:2507.05231 ('An improved construction for the triangle removal lemma', 2025) is tangentially related—it improves the constant in the lower-bound construction showing $1/\delta_{TRL}(\varepsilon) \geq (1/\varepsilon)^{C_{\rm new}\log_2(1/\varepsilon)}$—but does not address the comparison with $N_{DFL}(\varepsilon)$. The conjecture is phrased informally ('We suspect') and is likely a long-horizon open problem given the difficulty of the existing bounds.
Context
The authors note that the best known lower bound on $N_{DFL}(\varepsilon)$ has the form $(1/\varepsilon)^{c\log(1/\varepsilon)}$ arising from the Behrend construction, while $1/\delta_{TRL}(\varepsilon)$ satisfies $1/\delta_{TRL}(\varepsilon) \leq \text{tower}(O(\log(1/\varepsilon)))$ (Fox [8]). Theorem 1.12 provides supporting evidence: if $N_{DFL}(\varepsilon)$ grows subexponentially in $\varepsilon^{-1}$, then so does $1/\delta_{TRL}(\varepsilon)$. If the suspected similarity holds, Theorem 1.10 and inequality (1.1) would together give comparable lower and upper bounds on $M_{TFL}(\varepsilon)$.
Notes. PDF source — math may be garbled; stated as 'We suspect' in running prose without a labelled environment. The PDF extraction appears truncated after the start of Section 2, so further conjectures or open problems in later sections may be missing.
Source paper
Removal lemmas and approximate homomorphisms
Jacob Fox, Yufei Zhao · 2022-01-13
https://arxiv.org/abs/2104.11626
PDF source