Feature request: star motions

[Guest]

cos the bright stars are the ones with the briefest lifetimes. a bright star like betelgeuse probably won't even be around for another 1000 years (it's already in its red giant asymptotic phase). it's fun to think about stellar motions for celestia, but the bottom line is that it just won't work, or it'll look ridiculous.

Red giants last quite a bit longer than 1000 years.

[Guest]

When I researched this some years ago, I found that only a few thousand stars have measured radial velocities. That's about as many stars as can be seen with an unaided eye a clear dark night, and the measured stars are not necessarily the brightest ones. The early ones (of spectral class O,B,A etc), that are among the most luminous, often rotate very rapidly which broadens the spectral lines and makes it difficult to measure their doppler shift accurately. If anyone happens to know of a larger database, I'd be happy to learn.

There are certainly more than 30,000 measured RVs. ADS or CDS would have them (ADS under spectral data, I guess). There are certainly many bright stars amongst them. I can produce a plot showing those with full 3D velocity information if you like.

Of course, with only some thousand stars there would be no big problem to show their 3D motion in Celestia, or even to simulate the galactic gravitation over a few thousand years. But I'm not sure it would be that interesting, considering the vast majority of stars (with potentially high 3D velocities) would still be shown static. 5 fps doesn't sound very attractive to me.

Simulating galactic gravitation would be a nightmare. Keplerian orbits wouldn't cut it (or we wouldn't have a flat galactic rotation curve, for a start). Linear approximation would become very inaccurate after a few tens of thousands of years, yes, but it's efficient and reasonable within that time. It is certainly interesting to see how constellations change over time, and where nearby stars are going relative to us. 5fps is not caused by the moving stars, but by the visible 30,000 (this system did not use any sophisticated visibility culling at all), and the processor was doing all the perspective projection without 3D card support, as I said.

I don't believe that this feature is any less scientific than faked stars or "3D" nebulae.

[selden]

Someone wrote

Simulating galactic gravitation would be a nightmare.

Not a nightmare, just very, very slow. Today.

Keplerian orbits wouldn't cut it

So use a simpler approximation for the galactic orbit Keplerian orbits are complicated, if that's the right description, because they describe a patth in a 1/r^2 gravitational field, while the acceleration due to the galactic gravitaional field in the plane of the galaxy is a constant. Of course, the path description does get a little more complicated if one wants to model accurately the path of a star travelling through the plane and out into the halo

(or we wouldn't have a flat galactic rotation curve, for a start).

But the vertical oscillation has a period of ~30 MegaYears. Surely a linear approximation would be good enough over, say 5-10% of that: 2MY or so. And it's not as if adding a sinusoidal term is that much more expensive in CPU time..

Linear approximation would become very inaccurate after a few tens of thousands of years,

What influences are you considering that would have, say, a 10% effect in that short a period? We're already living with positional errors of that order.

[chris]

I think Alexis didn't excellent job of summing up the issues with implementing star motion. Celestia uses an octree to perform fast culling of the star database, and this structure isn't well-suited to moving objects (though it can easily be adapted to handle constrained motion, say, two stars orbiting each other.) There are workarounds . . . If there are only a few thousand moving stars, they can be kept in a separate list. But if most of the stars don't move, I'm not sure if it's worth it to bother with proper motions at all. In short, without better data (i.e. a large database of stellar proper motions that includes radial velocity), I don't feel that motivated to modify Celestia to model stellar motion.

That said, I encourage someone to prove me wrong and point me to a catalog of radial velocities. Extra points if the records contain the HIPPARCOS/Tycho numbers.

--Chris

[Guest]

Not a nightmare, just very, very slow. Today.

I ain't writing no numerical integration code. It would be a nightmare.

So use a simpler approximation for the galactic orbit Keplerian orbits are complicated, if that's the right description, because they describe a patth in a 1/r^2 gravitational field, while the acceleration due to the galactic gravitaional field in the plane of the galaxy is a constant. Of course, the path description does get a little more complicated if one wants to model accurately the path of a star travelling through the plane and out into the halo

I've been arguing for a simpler model, the linear model. Perhaps I'd better state it, to remove any confusion:

The linear model I talk about for the vector position of a star p(t) at time t is just

p(t) = p(0) + v * t,

where p(0) is the observed vector position today and v is the observed 3D vector velocity today. As you can see, it's not exactly computationally taxing.

But the vertical oscillation has a period of ~30 MegaYears. Surely a linear approximation would be good enough over, say 5-10% of that: 2MY or so. And it's not as if adding a sinusoidal term is that much more expensive in CPU time..

I agree, but I don't see why Keplerian orbits need be involved at all.

What influences are you considering that would have, say, a 10% effect in that short a period? We're already living with positional errors of that order.

Errors in the velocity determination would be important to consider, but I really chose "a few tens of thousands of years" as a conservative guess. I don't actually know how long the approximation could be valid.

[Guest]

... I'm not sure if it's worth it to bother with proper motions at all. In short, without better data (i.e. a large database of stellar proper motions that includes radial velocity), I don't feel that motivated to modify Celestia to model stellar motion.

That said, I encourage someone to prove me wrong and point me to a catalog of radial velocities. Extra points if the records contain the HIPPARCOS/Tycho numbers.

--Chris

I think there's a bibliographic catalogue (probably on ADC or CDS) that collected all stellar radial velocities published between 1991 and 1998, and contains ~35,000 stars. If it's too much hassle, it's too much hassle.

[erostosthenes]

Red giants last quite a bit longer than 1000 years.

a super red giant that has been so for all recorded human history isn't likely to is what i was getting at.

[selden]

In addition to the "Hipparcos Input Catalog", it turns out that radial velocities can be deduced from the Hipparcos parallax values in many cases. ( See http://www.astro.lu.se/~dainis/HTML/HIPPARCOS.html, which explains several different techniques.) I'll try to track dowon some URLs for appropriate catalogs later this afternoon.

[selden]

Well, I've found 3 catalogs that look like they might be usable:

The Hipparcos Input Catalog (uses HIC and HD numbers of 118,209 stars, but it isn't obvious how many entries include radial velocities)
http://cdsweb.u-strasbg.fr/cgi-bin/Cat?I/196
ftp: ftp://cdsarc.u-strasbg.fr/cats/I/196

WEB Catalog of Radial Velocities (uses HIC and HD numbers of 20,793 stars)
http://cdsweb.u-strasbg.fr/cgi-bin/Cat?III/190B
ftp: ftp://cdsarc.u-strasbg.fr/cats/III/190B

Astrometric Radial Velocities. III (uses HIP numbers for about 1,000 stars in 11 clusters and associations)
http://cdsweb.u-strasbg.fr/cgi-bin/Cat?J/A%2bA/381/446
ftp: ftp://cdsarc.u-strasbg.fr/cats/J/A+A/381/446

[Guest]

Red giants last quite a bit longer than 1000 years.

a super red giant that has been so for all recorded human history isn't likely to is what i was getting at.

It could last for several more million years yet. I don't think it's really a viable argument against a 4D starmap.

[Guest]

http://adc.gsfc.nasa.gov/adc-cgi/cat.pl?/catalogs/3/3216/

(Bibliographic catalogue of radial velocities appearing in the literature from 1991 to 1998, 33471 records).

[alexis]

Thanks for your input, selden and anonymous! It seems indeed as a large fraction of the Hipparcos database contains radial velocities after all. Now you only have to come up with an efficient "culling scheme" to display them efficiently. Remember, that a radius of interest doesn't help much since you have to predict which stars belong to that region of space, and to do that you need to check every star anyway, to see where it situates at the instant.

The parallax method to determine radial velocities requires very high astrometric precision, and they have really only used the Hipparcos data to check the validity of the method on clusters, where you have good statistics. It will be much more useful in the GAIA sample, where it will be applied to individual stars. I've been to a seminar with Dainis Dravins who researches this, and it's a very interesting method.

Regarding the linear velocity model, the computation of the position in itself will not be a problem of course, the problem is to select which stars to do it for, and which stars get visible. If we choose to check every star every frame, I think simulating a galactic force would be possible by writing

p(t) = p(0) + (v + f(p)*dt)* t

where f(p) is a smooth parametrised vector field representing the galactic force, and dt is the the time difference between 2 frames. This is the simplest Euler integration, and one could of course come up with more accurate schemes (the simplest being dividing up dt if it's large). I say this would have a very minor impact on the performance, in comparison to the projection, flux calculation etc.

A limited time range would also have to be imposed for the validity of the moving star positions (they will diffuse away in the long run). And the question is still what to do with the remaining majority of non-moving stars, throw them out or just let them sit there?

/Alexis

[Guest]

The figure I quoted earlier, of 5 fps for 30,000 stars was from memory (and incorrect). A re-check produced a figure of 20 fps for 30,000 stars running on a 1GHz Athlon w/o 3D card support (I guess projection from 3D to 2D would be handled by 3D cards nowadays - is this so?). Since ~30,000 stars in Hipparcos would appear to have radial velocities, I'm tempted to argue that sophisticated culling is unnecessary. On receiving the request for "time travel", stick only the stars that have velocity information into a list and display that. The stars without velocity information could be faded out in a swish manner, unless explicitly requested. If they are requested, they could be left in the octree data structure and coloured red, or some other method used to identify them as "static". Of course, this method would quickly become inadequate if the number of stars with known radial velocities increased greatly, but that is only likely to happen several years post-GAIA, at which time I will be radically disappointed if commonly available computing power hasn't jumped by another few orders of magnitude and hover cars appear on the market.

I agree that a "sphere of interest" isn't much help in display terms, although plotting it (without using it to calculate visibilities) might produce a useful guide to the user. It might help with indicating when the accuracy situation has become hopeless, after sufficient time in the simulation has passed.

I think I'd be reluctant to use this f(p) model, since it adds an extra set of unknowns (the form of f(p) ) to the situation. Using f(p) = 0 will be inaccurate over long periods, sure, but unless we constrain f(p) very well, so will this model.

[selden]

Alexis,

A quibble: it's not the Hipparcos Catalog (HIP) that has radial velocitis, it's the Hipparcos Input Catalog (HIC), the one they used to plan the mission. Also, it isn't obvious to me that HIC numbers are necessarliy the same as HIP numbers, and a few can't be: HIP includes a few more stars (118,218 vs 118,209 according to the Web pages at CDS).

And, of course, Sol itself moves with respect to the "background" stars, which adds a (very minor) complication.

[chris]

And, of course, Sol itself moves with respect to the "background" stars, which adds a (very minor) complication.

A problem which can be solved by simply defining Celestia's 'universal' coordinate system to be relative to the position of Sol . . . This is more or less the case already.

--Chris

[Aristarkhos]

Count me as someone else who would be interested in seeing proper motions for a subset of stars - whatever is easy to add.

[selden]

And, of course, Sol itself moves with respect to the "background" stars, which adds a (very minor) complication.

A problem which can be solved by simply defining Celestia's 'universal' coordinate system to be relative to the position of Sol . . . This is more or less the case already.

--Chris

Well, that's the complication I meant: if you do that, doesn't it imply that all of the "stationary" background stars are moving in the opposite direction? which may be just a little more computationally intensive than is desirable.

[grin]

It was interesting to read this discussion from the TimeTravel point of 2002.

It's 2006 now, and I wonder whether this highly theoretical discussion had any impact on the code?

And, by the way, which "2d" star-emitting programs do try to correct timetravel star motions? (One of the prettiest ones, Stellarium, seems not to, for example.)

Thanks!

ps: any of you remember of commodore64? 0.98Mhz cpu, 48kb ram. There was a program drawing sky and constellations, providing info on every star displayed (like spectral class, radial velocity ) etc), and was able to project far in the past and the future. Watch the constellations fall apart.... [yes, that's 2D, I know.] Have computer programmers really advanced anything, or it's just the "iron" under our fingers?

[rthorvald]

It was interesting to read this discussion from the TimeTravel point of 2002.

It's be really nice if Celestia could incorporate the proper motions of stars. For example, one would be able to find various stellar associations by watching them move among the others. Not to mention being able to watch the constellations change with time.

I would love - like everybody here, i suppose - to see the stars move... However, as i understand it, this is technically unfeasable.

A question, then:
Would it not be possible, as a crude wrokaround, to have several star catalogs? Say, four or five different ones uptime and downtime from today, with 100000-year intervals re their positions?
Of course one would have to relaunch Celestia when changing star catalog, but still.. It would be an interesting start...

The real question, of course, if the catalogs exist or what amount of work it would be to make them. Of this, i have no idea.

- rthorvald

[selden]

Don't forget that Celestia now can show the motions of stars. You can define a barycenter at the center of the galaxy and have them move around it.

Someone has to translate the measurements of star motions into a form that Celestia can use.

Of course, direct support for motion in the binary star database would be more efficient than textual catalogs.