[Paper Review] Dimer lambda_3 = .453 +/- .001 and Some Other Very Intelligent Guesses

Paul Federbush|ArXiv.org|May 8, 2008

Fuzzy Logic and Control Systems3 citations

TL;DR

This paper proposes a novel algorithm to estimate the dimer constant $\lambda_d$ for dimensions $d=3,4,5$ using an asymptotic series in $1/d$, leveraging successive terms $B_g$ and $B_{g+1}$ from a computed series. The method yields $\lambda_3 = 0.453 \pm 0.001$, $\lambda_4 = 0.5748 \pm 0.0006$, and $\lambda_5 = 0.6785 \pm 0.0001$, with strong empirical support from consistency with the known $\lambda_2$ value and dimensional scaling behavior.

ABSTRACT

Working with a presumed asymptotic series for lambda_d developed in previous work, we make some intelligent guesses for lambda_d with d=3, 4, 5; and estimates for the corresponding errors. We present arguments in favor of these guesses, we earnestly believe they will turn out to be correct. Such approximate values may help stimulate people working on rigorous bounds. In addition to suggesting bounds to prove, there will be the strong motivation to prove me wrong.

Motivation & Objective

• To develop a reliable method for estimating $\lambda_d$ in high dimensions where exact computation is infeasible.
• To provide precise numerical estimates for $\lambda_3$, $\lambda_4$, and $\lambda_5$ based on asymptotic series expansions.
• To validate the proposed algorithm using the known exact value of $\lambda_2$ as a benchmark.
• To stimulate rigorous mathematical bounds by offering well-justified numerical conjectures.

Proposed method

• The method uses successive terms $B_g$ and $B_{g+1}$ from an asymptotic series expansion in $1/d$ to estimate $\lambda_d$.
• The estimate is computed as $\lambda_d = a \pm b$, where $a = \frac{1}{2}(B_g + B_{g+1})$ and $b = |B_g - B_{g+1}|$.
• The optimal $g$ is selected as the index minimizing $|B_g - B_{g+1}|$ for each dimension $d$.
• The algorithm is grounded in the heuristic that the average of two consecutive terms in an asymptotic series gives a stable approximation near the series' best value.
• The method is validated by its accurate recovery of the known $\lambda_2 = 0.296 \pm 0.007$.
• The dimensional trend in $g$ (e.g., $g=2$ for $d=2,3$, $g=4$ for $d=4,5$) supports the algorithm’s consistency with asymptotic theory.

Experimental results

Research questions

• RQ1 Can a simple algorithm based on consecutive terms of an asymptotic series yield accurate estimates for $\lambda_d$ in high dimensions?
• RQ2 Why does the choice of $g$ (the index of the last term used) increase with dimension, and does this reflect underlying asymptotic behavior?
• RQ3 How reliable are the estimates $\lambda_3 = 0.453 \pm 0.001$, $\lambda_4 = 0.5748 \pm 0.0006$, and $\lambda_5 = 0.6785 \pm 0.0001$?
• RQ4 Can this method serve as a foundation for proving rigorous bounds on $\lambda_d$?

Key findings

• The algorithm successfully reproduces the known value $\lambda_2 = 0.296 \pm 0.007$, providing strong support for its validity.
• For $d=3$, the estimate $\lambda_3 = 0.453 \pm 0.001$ is derived from $B_2 = 0.4538$ and $B_3 = 0.4524$, with $|B_2 - B_3| = 0.0014$.
• For $d=4$, the estimate $\lambda_4 = 0.5748 \pm 0.0006$ is based on $B_4 = 0.5751$ and $B_5 = 0.5745$, yielding $|B_4 - B_5| = 0.0006$.
• For $d=5$, the estimate $\lambda_5 = 0.6785 \pm 0.0001$ comes from $B_5 = 0.6785$ and $B_4 = 0.6786$, with $|B_4 - B_5| = 0.0001$.
• The choice of $g=2$ for $d=2,3$ and $g=4$ for $d=4,5$ reflects a dimensional trend consistent with asymptotic theory.
• The authors express confidence that these estimates are correct and suggest they may inspire rigorous mathematical proofs.