<div dir="auto">Great summary!<div dir="auto"><br></div><div dir="auto">Update on Hamiltonian MonteCarlo (HMC) that Da will and I were considering ... it is more adapted to sampling constant energy states.</div><div dir="auto"><br></div><div dir="auto">Putting aside the HMC/ergodicity idea for now, we can still use the "electo-potential" idea for finding different kinds of equilibria... depending on which L_p norm is understood in the steepest descent equation</div><div dir="auto"><br></div><div dir="auto">dr/dt=-grad V(r),</div><div dir="auto"><br></div><div dir="auto">Where the "electo-potental" V(r) is defined as </div><div dir="auto"><br></div><div dir="auto">V(r)=Sum(over voter positions R) of the product ||R-r||f(R), where f(R) is the fraction of the voters concentrated at R.</div><div dir="auto"><br></div><div dir="auto">It looks like the L_1 norm is most suitable for an issue space in the form of a Cartesian product of issue axes.</div><div dir="auto"><br></div><div dir="auto">Also argmin{||gradV at k|| : k in K}, where K is the set of candidates, seems to be a previously unknown Smith efficient election method.</div><div dir="auto"><br></div><div dir="auto">It should be highly manipulation resistant because the norm of grad V at k is possibly the best measure of the combined voter pull on candidate k away from its position in "electo-space".</div></div><br><div class="gmail_quote"><div dir="ltr" class="gmail_attr">El sáb., 29 de ene. de 2022 5:35 a. m., Kristofer Munsterhjelm <<a href="mailto:km_elmet@t-online.de">km_elmet@t-online.de</a>> escribió:<br></div><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex">On 28.01.2022 13:53, Toby Pereira wrote:<br>
> I think the problem with these long threads is that there are rarely<br>
> concluding thoughts at the end, so unless you're following very closely,<br>
> you might miss the best stuff. Understandably people post stuff as it<br>
> comes to them, so you get people's "stream of consciousness", but I<br>
> think it would be good if, once a discussion is dying down, the main<br>
> contributors then posted a summary of their overall thoughts and<br>
> findings, and which things they would take forward.<br>
<br>
That's a good idea, though I guess in some cases it could get a bit<br>
difficult to determine when a conversation is over :-)<br>
<br>
But for my threads, the exact spatial model one was discussing how we<br>
could get a better approximation of the probability that a random<br>
election in say a (10 voters, 4 candidates, 3 issues) spatial model will<br>
be a particular given election. This would be useful for constructing<br>
optimal strategically resistant methods in spatial models, and optimal<br>
methods trading off between unmanipulability and VSE.<br>
<br>
But unfortunately doing so seems to require determining a Voronoi map<br>
for each instance of the spatial model (with candidates fixed); in<br>
essence integrating over area of the intersection of some Voronoi<br>
regions, over every possible position of the candidates in issue space.<br>
Which looked pretty daunting to me, so I thought that the best we could<br>
do is sampling the volume of the region with candidates fixed, and then<br>
integrating that.<br>
<br>
Then Daniel and Forest discussed ways to estimate this (I think? Or to<br>
exactly evaluate something close to what I'm looking for), but my diff<br>
eq skills are not nearly up to the task of following them.<br>
<br>
There was also another thread where I was trying to extend my fpA-fpC<br>
method to more than three candidates by making use of a recursive notion<br>
inspired by IRV. I found a more elegant phrasing of the three-candidate<br>
version (with a different tiebreaker when there's a pairwise tie), but<br>
nothing conclusive out to four and beyond; though I do think the<br>
recursive approach itself could be useful later. (Off-list I used the<br>
recursive approach to find out that letting A's score be the minimum of<br>
A's score in any three-candidate sub-election is monotone and preserves<br>
DMT -- though it likely doesn't preserve DMTBR.)<br>
<br>
And this thread (that I'm replying to) is about Copeland//Plurality and<br>
IRV-flavored Condorcet methods. Forest suggested Minmax-elimination,<br>
i.e. Raynaud. The "gross loser" he's referring to is, I think, by<br>
Benham, which restores Plurality compliance to the method.<br>
<br>
-km<br>
</blockquote></div>