Eclipses for BC crashes

[jimbabwean]

I am unable to find eclipses in the year 2477 BC. I attach script, traceback log and ephemeris printout.

test.py.txt
traceback.txt
print_eph.txt

[brandon-rhodes]

Which version of Skyfield are you using? Support for a two-part ephemeris like DE441 was only added recently. Try upgrading to the latest version with something like:

pip install -U skyfield

Hopefully that should get you working. If not, then simply comment here again, specifying which Skyfield version you are using, and which exact ephemeris file you are using — de441_sub.bsp isn't a filename I recognize, so I'm guessing it might be one you built yourself?

[jimbabwean]

I ran the update, but still have the error. The ephemeris that I am using comes from splicing together the two parts that NASA provide for de441, and then excerpting years 5000 BC to 4000 AD with the NASA tool, as that meets my needs without having to store a 3 GB file on disk. The ephemeris works if I do a simple right ascension, declination request for the position of the moon in 2477 BC, but crashes if I ask for the lunar eclipses for that year.

In other words:

t = ts.utc(-2476, 4, 4)
moon = eph['moon']
earth = eph['earth']
astro = earth.at(t).observe(moon)
ra, dec, distance = astro.radec()
print(ra, dec)

works, and gives this output:

18h 38m 43.10s -28deg 40' 11.4"

but

t0 = ts.utc(-2476, 1, 1)
t1 = ts.utc(-2475, 1, 1)
t, y, details = eclipselib.lunar_eclipses(t0, t1, eph)

causes it to crash with this (among a lot of other stuff):

File "/Library/Frameworks/Python.framework/Versions/3.11/lib/python3.11/site-packages/jplephem/spk.py", line 224, in generate
raise OutOfRangeError(
jplephem.exceptions.OutOfRangeError: segment only covers dates 1969-07-30 through 4000-01-04

If I print the ephemeris with print(eph) I get this in the output (among a lot of other stuff):

JD -105155.50 - JD 2440432.50 (-5001-12-29 through 1969-07-29)
3 -> 399 EARTH BARYCENTER -> EARTH
3 -> 301 EARTH BARYCENTER -> MOON

So, it seems that the ephemeris has enough data to answer the question about the lunar eclipses in -2476?

Thanks for your help.

[brandon-rhodes]

All right — I'm back from a trip to deliver a talk at a Python conference, my first conference in quite a while, and so it's time for me to start catching up on open Skyfield issues.

1. Could you print the entire ephemeris and share the output here? Seeing only two of its segments isn't enough for me to guess what's going on, as looking for eclipses will probably also involve planets like Jupiter and Saturn.
2. Could you share the whole traceback, including all the "lot of other stuff"? Given just the line you have shared, I can't tell whether the error occurred when trying to produce the initial astrometric position, or whether it occurred as the library tried to produce an apparent position, which involves the positions of Jupiter and Saturn.

[jimbabwean]

Those three files posted at the beginning of the thread provide the full traceback, the script, and a print of the ephemeris. The skyfield version being used is 1.53. The de441_sub.bsp file was made by merging the NASA provided two part ephemeris files for de441 and using the config file: LEAPSECONDS_KERNEL = naif0012.tls
SPK_KERNEL = de441.bsp
SOURCE_SPK_KERNEL = de441_part-1.bsp
SOURCE_SPK_KERNEL = de441_part-2.bsp
BEGIN_TIME = 5001 B.C. DEC 31 00:00:00.000
END_TIME = 4000 JAN 02 00:00:00.000

[brandon-rhodes]

Thanks for drawing my attention back to the files linked up at the top!

Comparing the exception traceback to the code inside of Skyfield, it looks like when I wrote the lunar eclipse routine back in November 2020, I made it faster by doing low-level ephemeris operations instead of using the main .at(t) method. Which means that the lunar eclipse routine didn't benefit from the recent work I did to support multi-segment ephemerides like DE441.

So I'll need to choose between (a) abandoning efficiency by taking away the low-level calls and replacing them with more expensive ones, and (b) somehow reworking things so that the eclipse routine can share the support for multi-segment ephemerides, but without getting too much slower.

It's getting late in the day, so I'll probably commence in earnest tomorrow.

[jimbabwean]

Thank you Brandon. You're amazing!

[brandon-rhodes]

So, I just realized that I didn't offer any workaround to go ahead and let you keep working on your project. Try doing this right after loading the ephemeris:

eph.segments = [
    s for s in eph.segments
    if s.spk_segment.start_jd == -3100015.5
]

This will toss out of RAM all of the segments that aren't the old BC ones, leaving you only the ones you want. So Skyfield won't grab the more recent segments about be confused about your ancient date. Try it out and see if it makes your script work!

[jimbabwean]

The reason why I was examining eclipse data in the ancient past, was trying to see whether the much debated Babylonian clay tablet VAT 4956 had other years that satisfied the events mentioned. One such event is an eclipse on a certain day of the year. I found the eclipse that I was looking for by using Stellarium with the NASA ephemeris. I thought I'd let you know about the bug for the sake of helping you get your amazing library to work with the split ephemeris tables. Your library makes a lot of task very easy. Thanks!