egg changed the topic of #principia to: Logs: https://esper.irclog.whitequark.org/principia | <scott_manley> anyone that doubts the wisdom of retrograde bop needs to get the hell out | https://xkcd.com/323/ | <egg> calculating the influence of lamont on Pluto is a bit silly…
_whitelogger has joined #principia
<queqiao-->
<egg> That’s 2/4
_whitelogger has joined #principia
_whitelogger has joined #principia
<queqiao-->
<AGrilledCheeseSandwich> egg, what method did you use to approximate trajectories in principia?
<queqiao-->
<AGrilledCheeseSandwich> Id assume rk4 but i just want to see if there were any other options you considered?
<queqiao-->
<egg> Ah right. And we had added DormandPrince1986RK547FC earlier for the equipotentials.
<queqiao-->
Reply to "AGrilledCheeseSandwich: Id assume rk4 but i just want to see if there were any other options you considered?"
<queqiao-->
<egg> So, Kutta1901Vσ1 (aka RK4) is not used at all in the mod, only in a test.
<queqiao-->
<egg> RK4 is basically never the best choice, it is just a nice thing to show in a lecture because it has a nontrivial but still smallish Butcher tableau.
<queqiao-->
<egg> (No relation to @Butcher.)
<queqiao-->
<egg> I have some graphs lying around comparing various integrators (limited to the symplectic or conjugate symplectic ones, and it only covers what we had in 2016):
<queqiao-->
Reply to "egg: Same with position error:"
<queqiao-->
<sichelgaita> To clarify, in these graphs "left good, right bad". That's why we use QUINLAN_TREMAINE_1990_ORDER_12 when RSS is installed. (The KSP system is not stable enough for this kind of integrator.)
<queqiao-->
<lamont> PVG uses DP5 now, although I don't have pretty graphs and I just used it because of popularity and ode45 in Matlab. Seems the cool kids these days use Tsit5 in Julia.
<queqiao-->
<egg> We don’t have ode45, looks like that’s from Dormand–Prince 1980 rather than 1986
<queqiao-->
<egg> (We haven’t paid much attention to the nonsymplectic integrators really, we just picked one and used it.)
<queqiao-->
<lamont> is that just based on doc comments? i'd expect ode45 was DP1986 along with whatever paper it was that gave the free interpolant
<queqiao-->
<lamont> did i write down what paper that was, 'cuz that was a PITA to find...
<queqiao-->
<egg> Hm. DormandPrince1986RK547FC is a lower_order=4, higher_order=5 method. (The 7 is the number of stages.)
<queqiao-->
<egg> The matlab docs for ode45 do not give the Butcher tableau.
<queqiao-->
<egg> DP80 A family of embedded Runge-Kutta formulae gives a method they call RK5(4)7M.
<queqiao-->
<egg> [DP86] A reconsideration of some embedded Runge-Kutta formulae discusses RK5(4)7FM, RK5(4)7FC, RK5(4)7FS, RK6(5)8M, RK6(5)8C, RK6(5)8S, and RK8(7)13M.
<queqiao-->
Reply to "GoForPDI (less drag=more faster): Get a better computer :magic:"
<queqiao-->
<szymonf.> Thanks bro 🤓
<queqiao-->
<GoForPDI (less drag=more faster)> Being serious, turn down your history and prediction length, and turn up your tolerance
<queqiao-->
Reply to "GoForPDI (less drag=more faster): Being serious, turn down your history and prediction length, and turn up your tolerance"
<queqiao-->
<szymonf.> ❤️
<queqiao-->
<GoForPDI (less drag=more faster)> These will make your trajectory less precise and extend less into the past and future, but hopefully improve your frame rate