I am testing my way to find coordinates with the Gauss elimination. It's too early to say if it will reveal some great secrets, but it looks promising. I'd like to compare it with other methods, but actually I don't know how RedWizzard and Tornsoul and others do this exactly. Any help? I mean, I can read the source code at RedWizzard's GIT repository, but I don't understand it, and I can't use it from PHP.
My implementation of the Gaussian elimination works with 4 reference distances, and if these 4 reference distances are correct, it finds a valid solution always. I then have to check if a neighbour of this solution is a valid solution, too. This is f.ex. for 17 Draconis, (-305.75/272.21875/49.53125) and (-305.75/272.21875/49.5) are possible coordinates for all collected distances. LTT 1593 got two solutions, (81.75/-118.78125/-21.09375) & (81.75/-118.8125/-21.09375). There are some more.
And what I didn't expect: When I have 5 reference distances, and one of them is wrong (at least I suppose it to be wrong), I do calculate the coordinates with every combination using 4 of 5 distances, I may get a valid solution exept the suggested wrong distance, even if it was part of the set of 4 distances used to get the coordinates.
OK, it sounds a bit complicated. I'll try to give an example:
Ross 556 has no coordinates but 5 distances, to George Pantazis, PSPF-LF 2, WISE 0146+4234, LFT 215 and LHS 1375.
For simplicity I call them A, B, C, D and E for now. I think that E, distance from LHS 1375 to Ross 556, is 7.12 Ly, and not 7.13 Ly as TGC states.
When I calculate coordinates using distances A B C and D, I get a solution with A B C D ok, E wrong. As expected.
When I calculate using B C D and E, I get the same solution, where A B C & D are ok, E wrong. Same as before, even when I use the "wrong" distance for calculation.
Another example: col 285 Sector XZ-M B21-6
5 distances, 5 checks. One spills out (4079,-3423,3215)/32 which renders all distances invalid, all other 4 checks give (4080,-3423,3215)/32, and here all distances are good except Halbara.
I think I will make a way to mark these distances "suspicious" in my own database, and have a web frontend where you can enter the system name you're in, and you get a list of these suspicious distances, and enter the correct distances.
Maybe this is a good way to get rid of the multiple distances entered due to typos or having the wrong system name, 328 actually. Yes, I was involved there, too
We have 323 distances between systems with known coordinates, but the distance differs for 0.05 Ly or less.
We have 216 distances where the error is larger. And we have additionally 141 errors between systems with a CR of 5 or higher.
Leads me to another question: How to deal with errors and how to clean up. TGC considers a system to be verified after reaching CR=5. So I think it would be a good thing to offload all distances between systems with CR>=5 to a different place and not use them for "daily work", because we don't need them anymore.
If we have different distances submitted between the same systems, we need someone to visit one of the systems to verify. How many verification do we need to declare a distance as wrong, removing it from the table, or offloading it?
TGC gives a CR of 1 for a set of 5 distances leading to successful trilateration. And if someone enters a distance, it gets a CR of 1, too. I assume TornSoul considered the CRs for distances different than for coordinates, and I think this should be seen the same way.
So in a first step, when collecting TGC and RedWizzard's data, I took the coordinate CR multiplied by 100, and the distance CR multiplied by 20.
When I do my web interface to enter distances, I would like to have registered commanders identified, because I believe someone who cares to register and identify will put more care to distances than someone anonymous. And they will see what they submitted, and be able to change or delete their errors. This is what I am missing most at TGC.
In a second step I will do an API similar to TGC, so RedWizzard's and other applications may submit distances to my database, and every tool using TGC may get more correct data in the same way as from TGC.
Any thoughts on this?
