You are here

Kevin Jardine's blog

Can we detect the galactic bar using Gaia DR2?

Submitted by Kevin Jardine on 14 February, 2019 - 16:07

The galactic bar is located in the inner galaxy at a large distance from the Sun. Its closest approach lies in the heavily dust obscured first quadrant. Nevertheless, can we detect the bar using the Gaia DR2 data set? I think that the answer is yes if we are willing to map it using quantitative star counts and do not expect a great deal of detail.

Here's the region in the first quadrant that we want to map:

What helps is that the bar extends far above and below the galactic plane and ends in a point. That creates a geometry that can be detected despite all the data limitations.

If we just try to map the bar in the galactic plane, we do not get an encouraging result because of all the dust extinction:

This is a density map of stars between 0 and 50 degrees in galactic longitude and between 3 and 8 kpc. (The parallax errors are so high at this distance that I did not impose any error constraints on the stars except RUWE < 1.4)

However, if we look 400 parsecs below the galactic plane, we can avoid the extinction but now the parallax errors are so high that much of the detail is blurred out:

What I think helps is a more quantitative approach using star counts. Here's a graph that analyzes the star counts and distribution, also for a slice taken 400 parsecs below the galactic plane:

The x axis shows the galactic longitude. The blue data and left y axis shows the total Gaia star count between 3 and 8 kpc for that degree longitude at 400 parsecs below the graphics plane. The red dots and right y axis shows how far you need to go in distance beyond 3 kpc to find two-thirds of the stars for that degree longitude.

What you find is that we start with about 4500 stars close to zero degrees longitude and that as we approach 20 degrees the star count drops and so does the distance required to find two-thirds of the stars. What this says to me is that from 0 to 20 degrees the stars are in a structure that starts wide and gradually narrows to a point - just what you would expect from a bar shaped structure.

I've tried numerous slices and I see the same pattern with the point (end of the bar) occurring around 20 degrees from 400 parsecs below the galactic plane and beyond, as well as about 25-30 degrees from 400 parsecs above the galactic plane and beyond. Larger than about 20-25 degrees we see a pattern that is not so clear but arguably could be two spiral arms passing at different distances from the end of the bar below the galactic plane and one spiral arm above the galactic plane (the arm that passes closest to the end of the bar runs entirely below the galactic plane).

So in short, in the region close to the bar, we seem to see a spiral arm (the Centaurus arm) intersect the bar close to and beneath the tip.

Within 400 parsecs of the galactic plane, the pattern becomes more chaotic, probably because of dust extinction.

Searching the galaxy map

Submitted by Kevin Jardine on 10 July, 2018 - 00:16
Galaxy map search interface
The latest version of the galaxy map supports searching for stars and displaying their locations.

The new version of the galaxy map, ( released today, supports searching for stars. You can enter any Gaia DR2 source identifier or Simbad identifier in the search box at the upper right and if the object has a parallax, its position will be shown on the map. The most probable location will be shown with a green cross hair and other possible locations are shown along a light blue line. If Simbad has more information on the object, some of it will be shown in a pop up box, including common names, coordinates, z height, spectrum and absolute magnitude.

Because Simbad includes all the Hipparcos parallaxes, it is possible to search for Hipparcos stars even if they are too bright to appear in Gaia DR2. Simbad identifiers include common names so, for example, you can search for "Betelgeuse".

By default, the Bailer-Jones distance catalog is used to determine Gaia DR2 distances. You can turn off Bailer-Jones distances to use the simple reciprocal calculation instead (only available when err/plx < 0.2) by unticking the Bailer-Jones distances box..

Currently the Bailer-Jones catalog is only available through the Heidelberg ARI TAP service which does not include the CORS headers that allow queries to be accessed via JavaScript applications. To solve this I am using a sneaky trick involving a NodeJS proxy running on my own server that accesses the Heidelberg service. I hope that remains stable until the Bailer-Jones catalog is available via Vizier (or the TAP service adds the appropriate headers to support client applications).

In addition to search, I've added a display of longitude and galactic plane distance at the bottom left corner that tracks your mouse pointer (not available for mobile devices).

Making a galaxy map, Part 6: Resources

Submitted by Kevin Jardine on 26 June, 2018 - 09:41
Galaxy map meshes in Gaia Sky
Gaia Sky comes with most of the mesh sets preinstalled.

Here are some resources for the galaxy map.


Python scripts for creating and selecting the meshes:

Gaia Sky key frame animator:


All map meshes set up in Blender with lighting and materials:

Gaia Sky meshes:

All but the two largest mesh sets (Non ionizing ranges ge25 lt50 and Non ionizing peaks ge25 lt50) are part of the normal Gaia Sky install.

Individual map meshes:

These are for use outside of Gaia Sky and are larger more detailed versions. As the hot star, dust and HII region meshes were created using different bin sizes and orientations, it probably makes sense to convert them to a standard coordinate system representing parsecs where the galactic centre (0 degrees) is to the right and the sun is at (0,0,0).

Here are the transformations:

hot star meshes: multiply mesh coordinates by 3, translate by (-3027,-2997,-597), rotate by -90 degrees
dust meshes: multiply mesh coordinates by 5, translate by (-2045,-1995,-300), rotate by -90 degrees
HII regions: no transformation should be necessary.

Mesh used for hot star 3D print:

Mesh used for all star 3D print:


SVG source and background images:

Star lists

Ionizing stars

Making a galaxy map, Part 5: Overlays

Submitted by Kevin Jardine on 22 June, 2018 - 00:22
Galaxy map with
Galaxy map with three mesh sets: very ionizing, some ionizing, and no ionizing with 50 or more stars.

Over the last few blog posts I've outlined the process I used to create a hot star map of the galaxy within 3000 pc. Along the way I mentioned many different choices I made for the map, from star filtering, to density calculations, to mesh selections.

How can we tell that these choices were appropriate and actually created the map I desired?

Answering this question is made more difficult by the revolutionary nature of the Gaia DR2 data set. There has never been a data set large enough and accurate enough to make a map of the galaxy beyond about 300 parsecs before. Gaia DR2 allows a map with a radius that is an order of magnitude larger. There are no other maps for comparison.

There is, fortunately, a very relevant and highly studied set of stars that we can use as a check on the map: the list of known ionizing stars, which is included as an overlay on the map. These extremely hot stars, either Wolf-Rayet stars or stars of class O-B3, emit intense ultraviolet radiation, ionizing any nearby hydrogen clouds by ripping electrons apart from protons, and producing red glowing HII regions visible for thousand of parsecs.

These ionizing stars, and the HII regions they create, are an obvious spiral arm marker in other galaxies, and creating lists of these stars in the Milky Way has been an important project for astronomers interested in mapping our home galaxy.

Roberta Humphreys first published a large list of highly luminous stars, including many ionizing stars, in her highly cited 1978 paper, Studies of luminous stars in nearby galaxies. I. Supergiants and O stars in the Milky Way [ADS link: 1978ApJS...38..309H] It was subsequently expanded by Douglas McElroy and Cindy Blaha and released in 1984. The catalog and a SIMBAD cross match is available from my blog post here.

More recently, researchers have created extensive catalogs of O stars and Wolf-Rayet stars.

Astronomers have also published lists of ionizing stars for individual HII regions. I have compiled these results into a list of more than 700 stars as described in this blog post.

Wright et. al., "The massive star population of Cygnus OB2", Mon. Not. R. Astron. Soc. 449, 741 (2015) have a list of ionizing stars in Vizier.

Finally, almost all of the 340 thousand stars I used to create the map have 2MASS cross matches. About half of these 2MASS identifiers have entries in the SIMBAD astronomical database. I downloaded the basic data for these stars and selected the stars with ionizing spectral types.

After combining all these lists and cross matching with the Gaia DR2 catalog using the 2MASS, Tycho-2 and UCAC4 cross match lists, I ended up with a list of more than 6000 ionizing stars. (I'll include this list in my resource post next week.)

I used the distance estimates from C.A.L. Bailer-Jones et. al. in "Estimating distances from parallaxes IV: Distances to 1.33 billion stars in Gaia Data Release 2" to place the stars on the map. About 5000 of the stars were within 3000 parsecs.

Here are the stars below:

and here is the map with just the very ionizing mesh set:

You can do a more detailed comparison by toggling on and off the ionizing stars on the interactive map site.

I conclude from this comparison that the map includes almost all the regions with known ionizing stars (the main exception in the above image being the highly reddened Cyg OB2 region, which shows up only in the cooler meshes). However the "very ionizing meshes" (which are selected using Gaia DR2 colours) also include a few regions that have few ionizing stars in my list. In a future blog post, I'll look more carefully at these "extra" regions. It will be interesting to see what happens to the map with Gaia DR3, currently expected in late 2020, which should have more accurate photometry.

It is also interesting to compare my map with the recently released preprint on star clusters in Gaia DR2:
Cantat-Gaudin, T. et. al., "A Gaia DR2 view of the Open Cluster population in the Milky Way". Unfortunately the electronic data mentioned in that paper does not seem to have been released yet (or at least I cannot find it) so I'll have to resort to some ugly image scraping from the paper. I'll replace this with a proper map when I can get the data:

The blue and purple clusters are the youngest clusters. The yellow clusters are cooler clusters more than a billion years old so are not relevant for my hot star map.

All of these maps are consistent with three basic patterns for the ionizing stars within 3000 parsecs:

  • The ionizing stars in the Perseus arm are concentrated in the Cassiopeia region between 115-140°. The ionizing stars in the third quadrant (180-270°) are not very concentrated except perhaps in the Vela molecular ridge.
  • The Sagittarius arm is visible in the fourth quadrant (270-360°), especially in the Carina region and to a weaker extent in the first quadrant (0-90°)
  • The Orion spur is visible as an extended diagonal slash between the Perseus and Sagittarius arms.

The full face-on map here also includes two additional features derived from subsets of the ionizing star list: HII regions are positioned on the map as described in this blog post and OB association labels are determined from the median distance to the star members as given in the 1984 Humphreys catalog, with the exception of the Cyg OB7 label, which I moved closer to the position of the Sh 2-117 HII region. The Cyg OB7 stars as listed in the Humphreys catalog were very widely distributed in Gaia DR2, leaving no clear position, but Cyg OB7 is often associated with this HII region, which includes the Pelican and North America nebulae.

There were also a small number of OB association labels from the Humphreys list that were positioned near no obvious star concentration or isosurface region (such as R 103 and R 105) and I removed those from the map. You can see these in the pdf linked from this blog post.

Finally, the full map includes dust concentrations within 2000 parsecs using the data cube provided at the download link on this page:

which is described in this preprint:

"3D maps of interstellar dust in the Local Arm: using Gaia, 2MASS and APOGEE-DR14" by Lallement et al.

Next week, in the last blog post in this series, I'll provide links to star lists, meshes, source code and other resources for the galaxy map.

Making a galaxy map, Part 4: Enclosure

Submitted by Kevin Jardine on 21 June, 2018 - 08:08
Foggy mountain range
Finding structure in Gaia DR2 is like mapping foggy terrain. We may not know exactly how high the peaks go but we can find the mountain ranges nonetheless.

As I mentioned in my last blog post, because of strong dust extinction and reddening in some regions within 3000 parsecs as well as data selection effects and perhaps instrument magnitude limits, two regions that in reality may have similar hot star density may have very different density values in Gaia DR2. In this blog post I describe a technique using structural enclosure analysis for mapping star concentrations using density isosurfaces in the absence of precise density values.

To explain the enclosure technique I'll first step back a dimension and look at how we can analyze structure in contour lines on a topographic elevation map.

Consider the topographic map on the left side of this image:

If we are only interested in the structure of the region, we can remove the contour lines within individual peaks to get the simplified map on the right side.

We can represent the structure of this map using a tree diagram:

which just tells us that there are four peaks, one of which has two subpeaks. The tree diagram removes all the shape, size and location data but reveals the basic map structure.

It turns out that trees can represent any kind of enclosure, regardless of dimension. So they can be used to represent the enclosure structure of hot star density isosurfaces as well.

Another way to represent a tree is through enclosed circles. This circle diagram represents exactly the same tree. This diagram also has no location, shape or size data, but can be a clearer way to represent a complex tree:

The wonderful D3.js JavaScript library has a system to display any tree as a set of enclosed circles, so on this site I've put a circle enclosure diagram for the entire set of connected hot star isosurface regions with a density of 5% or more and with each region containing 5 or more stars (click to view the entire diagram in a separate window):

The background colour represents the star density and the circle size represents (very roughly) the number of stars it contains. Hovering your mouse over a region (when displayed on its separate page) gives the name of the region and a star count. The name of the region consists of the two numbers separated by a dash. The first number gives the density mesh the region is found in and the second number gives a sequential number for the connected regions found in the mesh.

This enclosure diagram is a detailed guide to the structures contained inside the 3D print I showed at the top of my previous blog post.

It turns out that the isosurface regions have some unusual enclosure properties that we can exploit to create our map.

Before I describe those, I think it would be helpful to view some annotated details of the above image:

On the left side:

On the right side:

We can immediately see density problems in these images. For example, the left image implies that there are more hot stars in the nearby southern stars surrounding the Sco OB2 and Vel OB2 associations in the solar neighbourhood than in the Carina region, one of the largest star formation regions known in the Milky Way. Clearly this is not true - it is an artifact of the star selection process, most likely caused by dust reddening, the restriction to very low error parallaxes, or perhaps instrument magnitude limits.

For this reason, a star map based purely on Gaia DR2 star density values would be highly misleading. Fortunately, a careful examination of the enclosure diagram suggests an alternative approach.

We can see that the density distribution is found in three distinct kinds of regions. First, except for a dense concentration in the solar neighbourhood, all the isosurface meshes contain one large region (which I call a plateau) and numerous smaller denser regions (which I call ranges after mountain ranges). The remaining regions occur as density peaks within the ranges.

The plateaus, being large enclosing structures, do not provide the kind of detailed site specific information that the ranges and peaks do, so I excluded them from the map. Looking at the list of ranges greatly simplified the mapping process - although there are about 33 thousand connected regions with a density of 5% or more, there are only about 6 thousand ranges. So this was more than an 80% reduction in the regions that needed to be considered for the map. Most of the connected regions turned out to be peaks contained within these 6 thousand ranges.

6 thousand ranges would still create too much clutter for a map. Without being able to use density, I needed an alternative criteria to select the most important ranges.

I decided to select ranges based on the number of ionizing stars they contained. Well, not exactly - what I actually did was to count the number of stars with a Gaia DR2 colour index less than -0.2. According to Eric Mamajek's very useful stellar colour table, the O-B3 ionizing stars have colour indices lower than this limit in the standard B-V colour system.

It actually seems unlikely to me that the Gaia DR2 colours are precise enough for a specific limit to select ionizing stars, but as you will see in the next blog post, I used an independent list of known ionizing stars as a sanity check for the map.

After some experiments, I split the ranges into three groups: the very ionizing ranges that contained 10 or more stars with ionizing colours, the some ionizing ranges that contained 4 to 9 stars with ionizing colours, and the rest with less than 4 stars with ionizing colours.

I then split the ranges with few or no stars with ionizing colours into three groups based on the total number of hot stars they contained: those with 50 or more, those with 25 to 49 stars, and those with less than 25 stars.

As the last group of ranges with less than 25 stars consisted of many small star clumps that tended to clutter up the map, I did not use those, much as a map of a continent does not attempt to show every town or village.

I also left the set of non-ionizing ranges with 25 to 49 hot stars off the face-on map poster, but it is available as a separate download for Gaia Sky.

For each set of ranges, I also created a set of peaks. After some experimenting, I selected the peaks inside each range with 50 or fewer hot stars. This made it possible to render translucent ranges containing several smaller denser peaks.

I will provide links to the four mesh sets (each containing one range file and one peak file) next week as part of the last blog post in this series.


Subscribe to RSS - Kevin Jardine's blog