diff --git a/main.tex b/main.tex index 2918abe4ad74c8567448ff59d05e82a6ce20feb9..5e4b908952d3d8cc4ad523627b2edb50c565ca20 100644 --- a/main.tex +++ b/main.tex @@ -693,7 +693,38 @@ Goals: \item noticeably slow (show benchmark) \end{itemize} -\section{Refinement} +The main downside of this algorithm is that many $r$,$c$ pairs which are tested +end up not yielding any solutions for the problem. +In fact, solutions $u$ to our problem with high rank must have $\mu(u)$ close to +$\beta_{-}$: +\begin{align*} + 0 &\leq \chern_1^{\beta_{-}}(u) \leq \chern_1^{\beta_{-}}(u) \\ + 0 &\leq \mu(u) - \beta_{-} \leq \frac{\chern_1^{\beta_{-}}(v)}{r} +\end{align*} +In particular, it's the $\chern_1^{\beta_{-}}(v-u) \geq 0$ conditions which +fails for $r$,$c$ pairs with large $r$ and $\frac{c}{r}$ too far from $\beta_{-}$. +This condition is only checked within the internal loop. +This, along with a conservative estimate for a bound on the $r$ values (as +illustrated in example \ref{exmpl:recurring-first}) occasionally leads to slow +computations. + +Here are some benchmarks to illustrate the performance benefits of the +alternative algorithm which will later be described in this article [ref]. + +\begin{center} +\begin{tabular}{ |r|l|l| } + \hline + Choice of $v$ on $\mathbb{P}^2$ + & $(3, 2\ell, -2)$ + & $(3, 2\ell, -\frac{15}{2})$ \\ + \hline + Computation time for earlier [ref] program & \sim 20s & >1hr \\ + Computation time for [ref] program & \sim 50ms & \sim 50ms \\ + \hline +\end{tabular} +\end{center} + +\section{Tighter Bounds} \label{sec:refinement} To get tighter bounds on the rank of destabilizers $E$ of some $F$ with some