There are a lot of Gaia DR3 variables with the remark “Period adopted is from vari_short_timescale table as there is no period in the Special Object Study table.”, for example Gaia DR3 1921804591431954176.
Is it true that this information is an estimate and that own measurements using CCD photometry would be helpful because they provide a better value based on the light curve?
I would also like to ask this more generally, for other cases than Gaia DR3 1921804591431954176: Which sources of period information are so inaccurate that it is better to use your own measurements using CCD photometry?
most of these periods are spurious, the number of Gaia observations is usually not large enough to get a reliable period, and we have seen such short periods in red dwarfs, where they are highly unlikely.
So new observations are useful.
Regarding getting new observations to improve periods, if you asked me, I would reply that the cases when new photometry is more valuable is when the amplitudes are small and thus survey data like ZTF or ASAS-SN are not really useful to reliable detect variations (e.g. 0.01-0.05 mag.) or when stars are too bright for those surveys (e.g. brihter than 10 for ASAS-SN or 12.5 for ZTF).
The above is specially true when there are no TESS observations.
But all data are welcomed.
Gaia results in general are not good because of the observing cadence I mentioned above, and also classifications are not reliable because they only provided rough or multiple types.
Before submitting data on any star, especially low amplitude objects, it is important to check that the data being collected is precise enough to show consistent variations, that is, getting to know your setup and its performance.
I recently did a quick check for some Gaia DR3 variables of unknown type (VAR) with very short periods (<~0.025 d). I would say that the majority of such periods are wrong and need to be redefined (for relatively bright stars, it is possible in many cases with TESS data). I agree with @Sebastian_Otero regarding the cadence: this is the main source of the period uncertainty.
By the way, misclassifications are quite common for Gaia DR3 variable stars. Gaia frequently classifies other types of variable stars as RS Canum Venaticorum types, while for many YSOs, they are classified as LPVs.
I ran some scripts recently to try and identify these misclassifications.
The results show that, as of today (2025.6.3), there are at least 185 misclassified YSOs in the VSX that were classified by Gaia as L-type variables.
However, putting that aside, Gaia’s classification of ECL appears comparatively more accurate, although the periods might be incorrect. Is it possible to use Gaia’s ECL classification to improve the classification of unclassified variable stars in the VSX in bulk? @Sebastian_Otero
For example, stars from the GDS or the 3P1-PS1 catalog – variables in the former are often assigned the generic ‘VAR’ type, while the latter contains many misclassified eclipsing binaries.
Additionally, there are QSO/AGN catalogs from SDSS or LAMOST.
These could be used to improve the classification of ZTF entries in the VSX. Since they have spectroscopic confirmation, this improvement would be more accurate.
For instance, ZTF J122935.25+145103.3 is actually an AGN, but the VSX classifies it as an SR variable (I plan to submit a revision for this object soon).
It seems that many (if not most) short-period variables in the VSX, imported from the Gaia DR3 catalogue and classified as VAR (or other poorly defined types), have incorrect periods. It would probably be safer to mark all their periods as “uncertain.”
Here is a sample of 23 Gaia DR3 variables in Lynx with the VAR type, maximum magnitude between 9 and 13, and amplitude > 0.09. A quick check using TESS data shows that 9 of them are EAs with periods > 1 day (with one exception), 4 can be classified as ELL (the period is either incorrect or must be doubled), a couple are ROT variables, and the others do not show any periodic variability.
Name AUID Coords Const Type Period Mag MagMax Delimiter MagMin Band Range Real Type Real Period
Gaia DR3 1000179513940148992 -- 06 48 15.68 +56 02 57.3 Lyn VAR 0.0370734 13.17 - 13.34 G 13.17 - 13.34 G 0.17 ?? can't find
Gaia DR3 1001056447478944640 -- 06 41 02.33 +57 08 32.7 Lyn VAR 0.0192549 13.37 - 13.47 G 13.37 - 13.47 G 0.1 EA 2.0004
Gaia DR3 1002308000948537472 -- 07 03 56.20 +59 15 25.7 Lyn VAR 0.1489882 12.98 - 13.09 G 12.98 - 13.09 G 0.11 EA 45.258
Gaia DR3 1003420977297924608 -- 06 46 11.64 +60 38 57.6 Lyn VAR 0.0356101 10.68 - 10.80 G 10.68 - 10.80 G 0.12 EA 4.1148
Gaia DR3 1028460194252643840 -- 08 22 16.03 +52 01 06.2 Lyn VAR 0.0145877 12.55 - 12.66 G 12.55 - 12.66 G 0.11 gdor? >1
Gaia DR3 1034801073514684288 -- 08 18 20.96 +56 45 50.2 Lyn VAR 0.214227 11.51 - 11.67 G 11.51 - 11.67 G 0.16 ELL, period*2 0.428454
Gaia DR3 717277512872954880 -- 08 42 26.01 +34 48 41.9 Lyn VAR 0.0145782 12.24 - 12.48 G 12.24 - 12.48 G 0.24 ?? can't find
Gaia DR3 813285390740288512 -- 09 29 35.62 +40 30 21.0 Lyn VAR 0.0777802 10.88 - 11.05 G 10.88 - 11.05 G 0.17 EA, eccentric 8.4157
Gaia DR3 816515309225903616 -- 09 06 58.06 +41 52 42.4 Lyn VAR 0.0365103 13.49 - 13.65 G 13.49 - 13.65 G 0.16 ?? can't find
Gaia DR3 902895279166421888 -- 08 16 44.76 +35 07 34.8 Lyn VAR 0.480367 13.13 - 13.28 G 13.13 - 13.28 G 0.15 ELL, wrong period 0.4285
Gaia DR3 903633979181255168 -- 08 30 40.98 +35 05 33.8 Lyn VAR 0.0248096 13.44 - 13.60 G 13.44 - 13.60 G 0.16 EA, 1.363
Gaia DR3 906980583338597632 -- 08 05 13.76 +36 38 09.2 Lyn VAR 0.0343689 12.31 - 12.60 G 12.31 - 12.60 G 0.29 ?? can't find
Gaia DR3 908044150383986816 -- 08 20 00.34 +37 22 39.4 Lyn VAR 0.0145597 13.40 - 13.63 G 13.40 - 13.63 G 0.23 ?? can't find
Gaia DR3 909161224133968256 -- 08 08 43.94 +40 02 02.0 Lyn VAR 0.201003 12.83 - 12.95 G 12.83 - 12.95 G 0.12 ELL, period*2 0.402006
Gaia DR3 919537693322550784 -- 08 00 11.51 +38 52 29.1 Lyn VAR 0.037912 11.94 - 12.32 G 11.94 - 12.32 G 0.38 EA, eccentric 5.048
Gaia DR3 920041888123732096 -- 07 50 38.00 +38 30 06.0 Lyn VAR 0.0367668 13.32 - 13.52 G 13.32 - 13.52 G 0.2 EA 0.9155
Gaia DR3 920173004885657216 -- 07 43 41.60 +38 29 09.4 Lyn VAR 0.0185788 13.37 - 13.46 G 13.37 - 13.46 G 0.09 ?? can't find
Gaia DR3 921478812382219264 -- 08 02 20.73 +40 39 06.6 Lyn VAR 0.0191968 13.40 - 14.02 G 13.40 - 14.02 G 0.62 EA, eccentric? 5.92
Gaia DR3 922984383101571584 -- 07 59 12.43 +43 07 46.2 Lyn VAR 0.0184381 13.16 - 13.26 G 13.16 - 13.26 G 0.1 EA 6.3357
Gaia DR3 980939301701684096 -- 07 03 38.96 +51 54 49.5 Lyn VAR 0.0356601 12.29 - 12.42 G 12.29 - 12.42 G 0.13 ?? ??
Gaia DR3 981334541771383296 -- 07 06 27.87 +52 17 18.5 Lyn VAR 0.0144251 12.16 - 12.28 G 12.16 - 12.28 G 0.12 ROT 0.742
Gaia DR3 983304523010713216 -- 07 25 53.01 +51 29 39.5 Lyn VAR 0.318567 12.55 - 12.67 G 12.55 - 12.67 G 0.12 ELL, period*2 0.637134
Gaia DR3 985316251332889984 -- 07 43 21.52 +54 02 16.2 Lyn VAR 0.1244546 12.96 - 13.09 G 12.96 - 13.09 G 0.13 ROT|ELL 0.5815
I have visually inspected more than 600 VSX stars in TESS data those last few month.
What I can say is that : apart from the classical bright var stars, that shows perfect textbook and easy LC, most of the Gaia stars included in the VSX the last couple of years down to mag 16 (and fainter) don’t even show any significant/visible variation in any TESS sectors, what ever could be the cadence, 30 min to 2 min, or have false type of variation/ false periode. A very bad work.
Any amateur looking at those LC would never had classified the incrimined Gaia stars as variable.
That is also true for a large number of var ELL stars, contained in the 15000ELL EB table 1 (false period), for example.
The ZTF variables and ATOID are generally relatively bright stars with continuous and large amplitude variations, and are of good quality, except maybe the faintest of them - I have inspected dozen of them.
Sorry for beeing so direct, but unfortunately, that’s a fact that anybody can check, by looking at a randomly choosen number of Gaia ID in the VSX.
Hello everyone,
as the two people above said. In fact, I have some doubts about certain classifications in Gaia DR3. “VAR” is not an accurate variable-star type, but rather a variable star lacking classification and other information.
Compared with categories such as GDS, however, I feel that Gaia DR3 has a noticeably larger number of problematic entries labeled VAR.
Actually, I am not sure whether VSX really needs to import all or most of the entries classified as VAR in Gaia DR3. After all, judging from the current situation, the error rate in this broad category is indeed rather high. Up to now, there are 166,029 sources in VSX with Gaia DR3 identifiers whose classification is “VAR” (perhaps we can round it to 166,000), and very many of them are likely problematic.
But obviously, at present the VSX team does not have a large amount of time and energy to check these entries. At this stage, there are also some papers that make batch corrections to variable-star classifications or periods in VSX, but up to now, it is clear that there is already a serious backlog of relevant papers and journals that still need to be corrected or incorporated.
I do come across Gaia VARS. I ask for AUIDs for the brighter ones, SNR> 60-100, in hopes of finding a curve for them. Sometimes I get the AUID sometimes not. Sometimes there is rejection because the amplitude is not stated and VSX rejects it as less than 0.02 mags. I feel that the stars typed as VAR are an invitation to discovery. If I am taking data on a half dozen other stars in the FOV, why throw away the VAR?
IMHO, there is nothing wrong with a noisy, dim light curve as long as enough measurements can be made over a few years so that a polynomial fit can be applied to discover epoch and minima. That is especially good for unobserved stars.
The survey may mislabel them, but amateurs can discover the actual parameters.
The trouble is that the Gaia teams classified as VAR a very large amount of stars for which none can see any small variation. You may be right, TRE, when you says that the variations could be very low, just above the noise, and so are needing statistical computations to be seen. To this level of variation, you would probably classify as variable maybe 5 to 10 % of the stars, if not more.
But, if so, I wonder what could be the real effectiveness of an algorithm that classify as variable such stars, and in the same time can’t see true variables, with large amplitudes, and very clear patterns such EA/EB, EW, ROT and other type of variables.
Not sure Machine learning is very very good at this game.
One should take a look at the star :
ATOID J088.3238+26.8133
Which is classified as IRR in the ATLAS catalog of variable stars ( J/AJ/156/241/table4 - A first catalog of variable stars measured by ATLAS (Heinze+, 2018))
(IRR/LPV = The acronyms stand for “long-period” and “irregular” variables. These classes serve as “catch-all” bins for objects that do not seem to fit into any of our more specific categories. The LPV class contains objects whose variations appear to be dominated by low frequencies, corresponding to P\ga5 days, while the IRR class contains objects whose dominant frequencies are higher. Most of the stars classified as LPV or IRR (especially the latter) do not show coherent variations that can be folded cleanly with a single period. Hence, both classes are in some sense “irregular,” though the characteristic timescales are different. Among the objects that cannot be cleanly phased to a single period, the LPV class surely includes many semiregular red giant variables, while the IRR class has a large number of cataclysmic binaries)
Indeed, there is a lot of revisions in front of us.
I wonder why such bad jobs are accepted from the pros in publications. Sob sob.
yes, those short periods are likely bogus, as mentioned in the original posts in this thread.
Fixing them is a good data-mining project for the community.
For all of you that are taking the time to analyze these stars and share your results, they would have more value of they are submitted to VSX.
When we have so many objects analyzed, it is nor practical to submit them individually, so we recommend the batch upload approach. That is preparing a spreadsheet with all the relevant columns and submit as many of them as possible. Light curves can be uploaded to a dedicated website for our review.
This is what several observers like Gabriel Murawski and Jiashuo Zhang have been doing. Check the links to see how they did it.
Even the ones that do not show variability are worth revising (as “CST”) to prevent people from observing them (and request AUIDs for that).
most of the Gaia stars included in the VSX the last couple of years down to mag 16 (and fainter) don’t even show any significant/visible variation in any TESS sectors, what ever could be the cadence, 30 min to 2 min, or have false type of variation/ false periode. A very bad work.
Any amateur looking at those LC would never had classified the incrimined Gaia stars as variable.
Importing the Gaia variables to VSX took a lot of work and in the process, we were able to see some of the problems the catalog (actually the several catalogs incluced in the release) had.
When working with such large datasets you have to make assumptions because you can’t check one star at a time.
We have not added to VSX a rather large subset of stars that were obviously bogus or have a large chance to be so, e.g. stars fainter than mag. 21 which is the survey’s limiting magnitude, stars fainter than mag. 20 with no actual type published, stars brighter than mag. 8 that may be saturated, a subset of CV candidates that are actually a result of contamination from very close companions, and other objects that were identified as artifacts. Nearly a million objects have not been imported. But we decided to upload the rest because we saw that there are bona-fide variables there too.
Having wrong types or periods in lists like these is not unusual. We need to get used to that in the sky survey era, especially for datasets without high cadence like Gaia.
Getting help from the community in revising these objects is always welcomed. Gaia gave us a huge dataset to work with.
Submitting batch uploads as explained in my reply to Max is a useful way to help improve VSX and show the true nature of thosee Gaia variables and detect which ones are not variable at all.
Right now we are doing a revision of 60,000 of these objects using information in SIMBAD, in order to add spectral types, identify extragalactic objects wrongly classified, and inconsistencies between the published spectral types and the current classification.
This kind of work might become interesting data-mining projects for people wanting to help properly characterizing these stars.
You mentioned other surveys like ZTF and ATLAS.
ZTF spans much longer and has a much higher cadence so their results are much better.
ATLAS is not as good and the main problem it has is that it is not VSX friendly. They haven’t used any standard classification scheme and they have published different periods found with different algorithms, sometimes one of these periods is the correct one, sometimes it is the other one. They have also published millions of stars actually flagged as dubious. Some ZTF catalogs of suspects and Gaia catalogs of variable stars are also presenting variable stars without useful information and are not being considered for VSX inclusion.
Determining what to include and what not is not a straightforward thing, but Gaia has an advantage: their coordinate precision that allows unambiguous identifications. Something that will help us correct lots of wrong identifications and improve cross-matching when importing future variable star catalogs (like the ATLAS one that hasn’t been imported yet because of the issues mentioned above).
So observers like you are welcomed to submit revisions when you find something is incorrect.
Nothing but irregular there!
But that is an example of potential contribution from observers or data-miners.
And about acceptance or not of this kind of work. They are more focused on quantity than in quality. But I wouldn’t blame them, we are talking about millions of objects. I agree with you that machine learning hasn’t yet reached a state when automated classifications are trustable. It is rapidly evolving though. Even then, light curve inspection to properly classify objects (when we have so many specific subtypes) is still needed.
I still think that having all these objects catalogued brings observers an opportunity to contribute, like Ray said.
yes it could be an eclipsing binary with some irregularities on a the smaller star. ROT + absidal motion ? I’m not sure, of course.
The Y are flux with sky background removed. X are JD.
I do not remember exactly the mag of this star, but it could be something like 12.3 mag, in ATLAS system.
Thanx for all those informations about the VSX, Gaia and other surveys.
I agree : this is not an easy task.
Of course it is impossible to check individually all the stars imported in the VSX. And I do think it is not the responsibility of the VSX team to do so. Possibibly, the professionnals could do a better job when assigning a var flag to a star.
About submitting revisions to the VSX : I have now dozen of possible revisions to submit to the VSX. Unfortunatly, the process is particularly not straightforward, and something like discouraging.
I now have visually checked more than 45000 LC in less than 3 months, and have learned a lot, and I could share some stories as well. In this process, I also individually checked 624 VSX stars. And found more than 80 uncatalogued variable stars. That is more than 10% of the present VSX stars, in the inspected sky areas.
A friend is writing a code that greatly help to do this job. We are planning to expand this work with other amateurs.
I wrote you about all this stuff in december or january, asking you quite a lot of questions. You never replied, alas. I know you don’t have enought time to answer, but some details only known by you could help us.
We are now in a state where we can’t no longer submit individual stars to the VSX. There are so much stars to submit or to revise!
We are now seeking professionnals here in France to publish this job, with some hopes : some are interested, and we will meet them in june.
However, some help from you would accelerate the process, if you like.
There are some catalogs we’re not sure.
Is there any process/ any chance to submit some stars by batches ?
Hi guys, that is not the light curve shape of an eclipsing binary.
And you can’t just see if there is apsidal motion by looking at a few cycles in a JD light curve.
It is an SXARI variable. They have such light curve shape with sharp minima (it might even be an SXARI/E).
It is classified as B9 in 2024AJ…168…25G and that is consistent with the above classification.
Spectral types are important when you try to classify an object. The more information you have, the easier would be to make some conclusion.
You are welcomed to submit it to VSX with an epoch of minimum given.
As I mentioned in this thread, we can try the batch upload option in case when there are too many variables, because individual submission is a bad use of both the observer and the moderator’s time.
Right now we are at the limit when it comes to handling the VSX submission/revision rate.
Something in the order of 20-40 d delay in the queue, and receiving several submissions a day.
You said that the process is discouraging. In the context of this discussion, having mentioned all the issues that the batch imports from surveys have, the good thing is that individual submissions do not hace any of that. They have to have compelling evidence of variability and follow the standards mentioned in the VSX supporting documentation.
We are working on a new version of VSX and we will apply some filters to try to speed up the whole process. We receive too many submissions that do not follow any of our guidelines, and it takes too much time to process them. Much of them shouldn’t have gone through.
Recently, a lot of people have started submitting AI-based submissions, underestimating the knowledge on variable star classification and survey data that it takes to end up with a good final analysis.
I agree with you when you say that it is better to publish these findings in some journal
. Then we would be able to import those results from the resulting publication (we are behind with those updates too, it takes a lot of time and resources).
When mentioning the batch upload submission mode for VSX, I have to mention that for that to work, the observer should have experience classifying variable stars, otherwise, we wouldn’t be able to import their files as they are, so it needs some training.
I checked and found your Jan. 20 email. It was so long that I couldn’t reply it at the time and then it got lost. Sorry for that. I usually gave priority to the discussions in the moderation queue and lately, it has been overwhelming. I will take the chance to reply to you later today since we are discussing these issues.
Yes, what I have been mentioning in this thread. But submitting individually at the beginning might work as training in order to get experienced enough to be elegible for that approach.
Well… it is an eclipser of sorts, esp. if a /E. Single star, magnetically confined circum-stellar material that “occults” the star during spin rotation of the star. Sounds like an eclipser to me! There are too many variable types now…
p.s. Note: I in the original question did not use the word “binary”. I did just use the term “eclipser”. So I didn’t mean to imply that it was an “eclipsing binary” or “eclipsing binary star”. Oh, bother… best to use the specific proposed type… one must keep up with all the VSX variable types!
