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…
<queqiao->
⟨lamont⟩ tried to look at what you're doing but i got very confused with "gateux derivatives" and other such stuff.
<queqiao->
⟨lpg4999⟩ it's just a misspelling for "gato". this is the multilingual cat channel, not the multilingual cake chanel
<queqiao->
⟨Al₂Me₆⟩ I want to say that’s what they call the directional derivative? I believe they’re doing gradient descent based on the user’s current flight plan.
<queqiao->
⟨lamont⟩ but if you've got L-BFGS you should roughly have matlab level fminunc capability. and from skimming it looks like you're trying to minimize "J = ∆v" (or "J = ∆v^2"?) subject to the celestial distance constraint? (and not sure if you're throwing that into J with a penalty function or if you're doing lagrange multipliers?)
<queqiao->
⟨lamont⟩ (yeah the sensitivities of the flight plan to variations in the plan is definitely a derivative that you'd want to compute, i assume that is more or less what is being computed...)
<queqiao->
⟨lamont⟩ but more to the point is that for that kind of a problem you'll get better results doing both forward and backward integration with a "meet in the middle approach"
<queqiao->
⟨lamont⟩ that'll increase the dimensionality of the problem though
<queqiao->
⟨lamont⟩ but the idea being that "∆v (x,y,z)" are optimizer variables at the start, you then forward propagate those for "0.5*∆t", then you have optimizer variables at the end (you could just pick the 6-vector "(vf,rf)" as optimizer variables) and then backward propagate by "0.5*∆t". then you have constraints at point "1" in the middle where "v1- = v1+" and "r1- = r1+"
<queqiao->
⟨lamont⟩ but you can also be more cleverer and constrain rf to be on a surface with fixed radius with theta, phi being free (working in spherical coordinates) and then you're halfway to having your constraint automatically applied. and you can also probably eliminate a degree of freedom in vf by only considering the tangent space to rf and enforce the periapsis constraint that way.
<queqiao->
⟨lamont⟩ derivatives get a bit messier though and the problem becomes larger, but you guarantee that you always "hit" your terminal conditions (because you start there).
<queqiao->
⟨lamont⟩ then the optimizer just crawls around forward propagating on one side and backward propagatingo on the other side, trying to find a match point at the midpoint of the trajectory.
<queqiao->
⟨lamont⟩ (which is dramaticaly more stable than trying to only do forward shooting and hit a celestial)
<queqiao->
⟨ashnoom⟩ Here is an interesting one. But I am not sure if it is Principia's or Kerbalism's problem.
<queqiao->
⟨ashnoom⟩ I have a Orbital Perturbation Experminent running around Venus. I have a 65deg inclination. But when it is _not_ the active vessel, then the inclination in the Kerbalism window chnges to <50
<queqiao->
⟨ashnoom⟩ I brought my inclination up by 2.2points, now I at least get my science :p
<queqiao->
⟨quattrorumbass⟩ could I use this with my current 80s rp1 game or should I try it with a different save?
<queqiao->
⟨quattrorumbass⟩ * save
<queqiao->
⟨sichelgaita⟩ This should be compatible with any recent save (less than 3-4 months). Of course, save your save before using this version.
<queqiao->
⟨quattrorumbass⟩ ty!
<queqiao->
⟨egg⟩ To an extent this effort stems from having the BFGS hammer because of the equipotential work and seeing something that looked like a nail.