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->
⟨.auer⟩ the thing I didn´t like about stock node markers is that ~mostly of the time~ in some cases it becomes cluttered by overlapping other nodes displaying info, so I always end using the UI, well honestly i hate it...
<queqiao->
⟨kerbmario⟩ ⟪lpg4999⟫ has no one made a mod that embeds a web […] ⮪ you know embedded webview would be brilliant for pulling up htmls or txts in game like mod docs
<queqiao->
⟨egg⟩ ⟪fiora⟫ honestly that alone would be amazing. […] ⮪ It turns out that it is somewhat tricky to do 4-dimensional optimisation given the shape of our types. Do you think optimizing just the ∆v might be a useful first step?
<queqiao->
⟨test_account9540⟩ unfortunately, no (
<queqiao->
⟨egg⟩ +components
<queqiao->
⟨test_account9540⟩ when the maneuver is planned manually I always have to tune both of them
<queqiao->
⟨test_account9540⟩ But why id the optimization 4-dimensional? Are you optimizing all three dV components?
<queqiao->
⟨test_account9540⟩ * is
<queqiao->
⟨test_account9540⟩ when the maneuver is planned manually I always have to tune both dV and time
<queqiao->
⟨egg⟩ Yes, plus time.
<queqiao->
⟨test_account9540⟩ But it should be enough to optimize prograde dV and time
<queqiao->
⟨egg⟩ That would likely be equally messy.
<queqiao->
⟨test_account9540⟩ If the vessel is in the correct parking orbit given by TWP, there should always be an impact trajectory with a prograde only burn
<queqiao->
⟨test_account9540⟩ If only one component is optimized then it probably should be time, rather then dV I think
<queqiao->
⟨test_account9540⟩ * than
<queqiao->
⟨test_account9540⟩ dV is reasonably well known from TWP but time is not known at all
<queqiao->
⟨egg⟩ We may be talking about different things. This is about automating the fine tuning, not the initial guess. Single dimensional optimization is useless, you can do that well and quickly enough by hand.
<queqiao->
⟨test_account9540⟩ But then it has to be two dimensional, time and prograde dV
<queqiao->
⟨egg⟩ right, but the moment it is inhomogeneous it might as well be four-dimensional, it is the same mess either way.
<queqiao->
⟨test_account9540⟩ One dimensional might work I think. How long would one optimization attempt take? If it's quick enough, it might be possible to optimize time for dV given by TWP, if no impact is found, increase by 1m/s and try again, repeat 300 times 🤔
<queqiao->
⟨test_account9540⟩ +a prograde maneuver with
<queqiao->
⟨test_account9540⟩ +dV
<queqiao->
⟨test_account9540⟩ * dV by
<queqiao->
⟨egg⟩ ... no.
<queqiao->
⟨lamont⟩ you really need an SQP solver
<queqiao->
⟨lamont⟩ you have a nonlinear problem without a feasible initial guess
<queqiao->
⟨lamont⟩ and it would likely be best to use sims-flanaghan transcription of the problem using patched conics first, in order to generate a near-solution, and then feed that as an initial guess into the full SQP problem with n-body physics.
<queqiao->
⟨lamont⟩ although for an impact on the moon maybe that is overly complex and just a lambert solver could get it close enough.
<queqiao->
⟨test_account9540⟩ this is about interplanetary transfers
<queqiao->
⟨lamont⟩ oh yeah, you need to solve the patched conic problem first with sims-flanaghan
<queqiao->
⟨lamont⟩ (and maybe use sims-flanaghan with the n-body problem as well)
<queqiao->
⟨lamont⟩ i've been fighting with the MJ transfer planner for years and you just can't really do it the naive way
<queqiao->
⟨egg⟩ The problem isn't so much the methods as it is the plumbing.
<queqiao->
⟨lamont⟩ well sims-flanaghan is pretty necessary because you do forward/backward integration from the SOI boundaries and meet in the middle of the SOI or helicoentric trajectories. that eliminates the problem of the variation in the impulse being magnified so much by the time you integrate out to the target celestial SOI that you miss the SOI completely.
<queqiao->
⟨egg⟩ We are looking at a more local problem for now. You have a flyby and want to lower it. That is easy in principle. In practice our gradient descent wants something that fits in our 3d geometry type system.