I suppose I should explain how I got these. I compared each solution to a 1200x1200 accelerated solution. Assuming that we can expect 6.5-7 bits of precision to be gained with each doubling, there should be more than enough precision in the 1200x1200 solution for these graphs to be very precise. On the other hand, if I had gone all the way up to a 1024x1024 solution, there would only be about 1.5 bits difference, which would most likely cause a significant deviation from the linear relationship seen in the log-log comparision graph.
Later this week I hope to repeat this analysis with various natural solutions, which I should be able to take up 1024x1024 (perhaps even 1448x1448, depending on how much precision I can get out of the matrix solver). I expect to see an additional 2.5-4.0 bits for each doubling, based on initial observations I made a week or two ago.
Edit: changed "2.5-3.0 bits" to "2.5-4.0 bits", to be a bit more conservative on my estimate for the natural solution.
Later this week I hope to repeat this analysis with various natural solutions, which I should be able to take up 1024x1024 (perhaps even 1448x1448, depending on how much precision I can get out of the matrix solver). I expect to see an additional 2.5-4.0 bits for each doubling, based on initial observations I made a week or two ago.
Edit: changed "2.5-3.0 bits" to "2.5-4.0 bits", to be a bit more conservative on my estimate for the natural solution.
~ Jay Daniel Fox