The star clusters of the Milky Way
Emily L. Hunt | March 4th, 2024 | University of Vienna
1
Nomenclature: the Milky Way's star clusters
Open clusters
Bound, M , young
Globular clusters
Bound, M , old
Associations /
moving groups
Unbound, M , young
≲ 104
​
Sun ≳ 104
​
Sun
≲ 103 ​
Sun
2
Why they're extremely useful
Gaia DR2 Hertzsprung-Russell diagram (Credit:
Gaia Collaboration+18)
Stars in clusters formed at the same time
from the same material
3
Don't just take my word for it...
4
Gaia's impact
Launched in 2013, Gaia is measuring astrometry and photometry
for stars in the Milky Way.
Credit: Lindegren+18
Its accuracy is really incredible!
~10 stars
At least 40 the accuracy
Down to magnitude ~21
9
×
5
How many stars is that?
6
How many stars is that?
7
But there's a catch...
There are many difficulties when trying to work with star clusters:
1. No perfect algorithm to recover
clusters
2. "Invisible" clusters from before Gaia
~50% of clusters are missing!
3. Clusters reported with Gaia
many duplicates + how many are real?
4. The completeness of the census
5. How to even define an open cluster!
Papers reporting new open clusters. Gaia DR2 was released in
2018.
8
Clustering algorithms
Clustering algorithms use user-defined parameters to extract
clusters from data. There are many of them!
A 'toy' 2D dataset After applying DBSCAN
9
HDBSCAN was best!
I tried multiple different algorithms
HDBSCAN was the most sensitive
Sadly, it also reported the most false positives...
10
The false positive problem
However, HDBSCAN is unusable without an extra step to remove
false positives:
Toy 3D dataset After applying HDBSCAN clustering With cluster significance test shading
11
Creating an all-sky catalogue
Recall: unknown how many literature clusters are real & census has
unknown completeness
The solution? An all-sky catalogue!
HDBSCAN most sensitive should get good results!⟹
12
The setup
Performed clustering in three different distance ranges,
totalling almost 13000 different fields
The goal: recover greater than 99% of clusters with signal to
noise ratios over 3σ
13
Add-on: more stars
Many stars in Gaia fainter than G=18 are still usable - I included all
stars with Rybizcki+21 classification over 0.5
Total of 729 million stars (largest ever Gaia clustering analysis)
King 9, without (left) and with (right) these extra stars
14
Add-on: cluster classifications
Cluster colour-magnitude diagrams
(CMDs) are a useful indicator of the
quality of a cluster
I used an approximate Bayesian
neural network to classify cluster
CMDs
CMD classification for a candidate cluster
15
Add-on: cluster photometric parameters
Isochrone fit for to IC 4756
I also made a similar network to infer
photometric parameters (age,
extinction, photometric distance)
16
All in all: we go from this...
17
... to this!
18
18.1
7169 clusters (4105 highly reliable)
2387 new clusters (739 highly reliable)
Plus many extras...
18.2
We accidentally detected tidal tails!
19
We accidentally detected tidal tails!
19.1
On new clusters
There are clearly big advantages to
a single blind search!
There are some very obvious
clusters that were missed
previously
HSC 2384, a new open cluster that was hidden behind IC 2602
20
How many pre-reported clusters do we find?
A big blind search makes it possible to say lots about literature clusters, e.g.:
Recover just 51.6% of clusters in biggest pre-Gaia catalogue, Kharchenko+13.
~1000 missing clusters that we would find if real - hence, probably not
Recover only 18.1% of clusters in Kounkel+20
Unlikely that many of their clusters real - we use same algorithm + better data
Some Gaia-era papers: recover almost all objects; others: not as many...
21
Are all of the clusters we detect bound?
22
Are all of the clusters we detect bound?
22.1
Are all of the clusters we detect bound?
Existing vs. new clusters.
22.2
Distinguishing between bound &
unbound clusters
It's clear I needed to separate open clusters from
unbound moving groups.
But how?
23
Logically: Virial theorem time?
The virial theorem states that an object in
gravitational equilibrium should have
For star clusters, we can express this as:
2T = ∣U∣
Q = ​ =
V
T
​
≈
2GM
ηr ​σ50
2
​ for a bound cluster.
2
1
24
But it didn't work...
Virial ratios for clusters in the catalogue.
They were consistently too large by
a factor of ~10.
The issue: binary stars messing up
velocity dispersion measurements
25
26
Jacobi radii to the rescue!
I spent a while looking for a solution.
A bound cluster will have a radius (the Jacobi radius) at which its
potential is stronger than its host galaxy:
r ​
J
r ​ =J ​(
4Ω − k2 2
M
)
​
3
1
27
Measuring accurate
cluster masses
Problem: cluster masses not
widely measured for Milky Way
clusters
To do this more accurately: I
developed method for selection
effect corrections
The magnitude-dependent selection function of three clusters
28
Measuring accurate
cluster masses II
Stellar masses from
isochrone interpolation
Additional correction for
unresolved binary stars
applied
Kroupa IMF fitted
Corrections are important!
Uncorrected and corrected cluster mass functions
29
Onto Jacobi radii: for three reliable clusters
Jacobi radius determination for three reliable clusters
Intersection =
All three are clear bound
open clusters
r ​
J
30
But what about the 'weird' clusters?
Jacobi radius determination for three suspect clusters
HSC 1131: not bound (disk
stream?)
HSC 2376: not bound
(expanding association?)
HSC 1131: bound! (small,
~60 solar masses)
31
Some limitations
Method is not perfect:
Have to assume spherical clusters
Have to assume circular orbits
Not good for clusters below ~40 MSun
But I think it's still much better than using nothing!
32
How does it change the catalogue's distribution?
The catalogue divided into clusters with (left) and without (right) a valid Jacobi radius.
33
What are the differences between them?
Radius and concentration of clusters vs. mass and age
Moving groups expand with
time; open clusters do not
High-mass open clusters are
very concentrated
Low-mass open clusters
and moving groups less
concentrated
34
The power of cluster masses
The catalogue's completeness depends strongly on mass
Kernel density estimate of cluster distance distribution in mass bins.
Full KDE estimate of cluster mass-distance distribution.
35
Cluster age function
The cluster age function for OCs in the catalogue.
Open cluster catalogues in Gaia
era have fewer old clusters
(likely due to removal of erroneous old
objects)
36
The first ever Gaia cluster mass function
The cluster mass function for OCs in the catalogue
37
Low-mass clusters are destroyed faster
The cluster mass function for OCs in the catalogue The cluster mass function divided into age bins
38
Low-mass clusters are destroyed faster
The cluster mass function divided into age bins
The slope of the cluster mass function, with age
39
Connecting to theory and other galaxies
Young clusters in all galaxies form with initial mass from power law
of slope (Krumholz 2019 + references therein)
New result: can constrain how this flattens with time, due to faster
destruction of low-mass clusters
Our results will be able to constrain rate and intensity of GMC and
spiral arm collisions
≈ −2
40
About individual cluster mass functions
All cluster mass function datapoints for all open clusters within 2 kpc
41
Conclusions
I made the largest ever deduplicated Milky Way star cluster
catalogue
(Hunt & Reffert 2021, 2023)
Jacobi radii and cluster masses can differentiate bound and
unbound clusters effectively
(Hunt & Reffert submitted)
Large catalogue of cluster masses reveals new details on cluster
formation and destruction processes
(also in Hunt & Reffert submitted)
42
I'm currently on the job market!
web:
Largest ever MW cluster catalogue
Jacobi radii to distinguish bound/unbound clusters
Many new results from these mass measurements
emily.space
43