LAMPSS – Final Results!

Nevermind Season 8 of Game of Thrones, this is the biggest unveiling of 2019 – the official results of the Lancaster Astrophysics Metal-poor Star Search. So without further ado, LAMPSS present 16 potentially extremely metal-poor ([Fe/H] <-3.0) stars:

Figure 1: Table showing all LAMPSS extremely metal-poor candidates and some of their properties. Distance and age were calculated by us. Note some of the errors on the ages!
Figure 2: LAMPSS_1229, our [Fe/H] = -5.0 candidate, is at the centre of this image from the Hubble Space Telescope.

Once we had applied our conditions for stars, spectral type and metallicity, we had our metal-poor candidates. The nature of candidates was manually checked the source using Hubble Space Telescope imagery ( to ensure the sources were not galaxies. Whilst doing so we found some interesting galaxies:

Figure 3: A jellfyfish galaxy found during our checks of NASA/IPAC archives, characterised by the tail of gas caused by ram pressure stripping. Image on the right is from our own catalogue, viewed on the DS9 software.

We compared our metal-poor candidates with that of WARP. A project running parallel to ours that used INT data, rather than CFHT. We found a definite correlation between two sets of sources:

Figure 4: A plot of WARP’s [Fe/H] value against our [Fe/H] value for a selection of metal-poor sources that appear in both catalogues. There is a definite correlation in the values of metallicity determined by the two groups.

We studied a parameter space that has never been explored before, so, in order to make more meaningful, general comparisons we calculated stellar number densities using the volumes calculated in week 4. By then applying the (g-i) conditions for spectral type, we plotted the number density per spectral type per metallicity for the region of halo surveyed (figure 5). Again, we checked our results against WARPs (figure 6).

Figure 5: Stellar number density per spectral type per metallicity plot for the region of the halo studied. G and K stars are a lot more common than the hotter A and F stars. Also noticeable, is the lack of any [Fe/H] = -4.0 stars.
Figure 6: WARPs results overplotted onto figure 5. They find a lower density for all spectral types and metallicities.

Admittedly, our project has several limitations. All of our predictions are based on the fact that a star lies on the main sequence, however many stars we surveyed will be at different stages in their evolution, thereby making our predictions inaccurate. Additionally, many of the candidates shown in figure 1 are K type stars and, as mentioned in the last blog post, some extrapolation is required to apply our metallicity conditions to K type stars. However, applying Davids old adage – “Within a factor of 10 is fine” – we remain confident that the LAMPSS metal-poor stars are indeed just that.

So that brings our project to a conclusion. We have written a report ( LAMPSS: Discovery of Metal-Poor Stars in the Galactic Halo with the CaHK filter on CFHT MegaCam (Worrell et al. 2019)), which was submitted very comfortably before the deadline. Now all that remains is a presentation at the PLACE conference (4th/5th June). I’d like to thank David Sobral for his considerable help and guidance throughout the project, as well as the rest of LAMPSS for being stellar.

as you were JB x

LAMPSS – Last Stages

I’m sure you’ve all been eagerly anticipating the next iteration of the LAMPSS blog, so sorry for the delay. The final pieces of the puzzle have (finally) fallen into place:

1) Spectral Type Conditions

Figure 1: The full (HK-g)-1.5(g-i) against (g-i) plot we created, including all spectral types.

After considering the inverse temperature against (g-i) plot (figure 5 of the week 4 blog post), we decided the above plot was the best way to separate spectral types of stars by their (g-i) index. The results are shown below:

Figure 2: Table showing the (g-i) conditions used to separate spectral types of stars.

2) Distance Calculations

Karolina completed her distance calculating Python script which employs the distance modulus equation (log10(d/pc) = (m-M+5)/5), to calculate distances using magnitudes in the u, g, r, i and v wavelength bands. The average of the results is used as the final distance to a star. A positional plot of all sources in our catalogue defined as stars by our star-galaxy separation conditions is shown below.

Figure 3: 3D plot showing the position of all stars within the LAMPSS catalogue, spectral type is indicated by colour.

We are searching for stars towards the halo of the Milky Way. By plotting the distance histogram for stars within our sample (figure 4), we conclude the halo stretches from ~25kpc to ~150kpc.

Figure 4: Distance histogram for stars within our catalogue. We judge the halo to cutoff at a distance of ~150kpc.

3) Metallicity Conditions

All that remained before we could start identifying our metal-poor stars towards the halo, was to work out if our stars were metal-poor or not. After considerable effort, Ellie used the Polyfit function on Python to derive metallicity conditions from the colour-colour plot shown in figure 1. Degree 2 polynomials were used to fit lines of different [Fe/H], and provide us with metallicity conditions in terms of g, i and CaHK. The dashed lines represent the mid-point between lines defining integer values of [Fe/H]. Sources were given an integer value of [Fe/H], depending on which pair of dashed lines they lay between.

Figure 5: Finalised metallicity conditions

Due to sources with (g-i)<-1 or (g-i)>1.5 greatly affecting the fitting of the metallicity curves, they were disregarded. This is not a problem for G type stars ( 0.3<(g-i)<0.8), however some K type stars (0.8<(g-i)<2.4) will lie outside the (g-i) range for which these conditions are valid. Therefore, some extrapolation is necessary to classify K type stars by [Fe/H]. Limitations aside, we now have all the tools required to find metal-poor stars. Next time, we present our findings.


LAMPSS – Week 4 – Science!

Hello there, hope you’ve all had a good week and earnt a few quid. This week LAMPSS made progress in several areas.

1) Volume Estimate

We found the minimum apparent magnitude in the HK filter in the COSMOS CaHK catalogue and plotted a histogram (see figure 1) to find the maximum HK apparent magnitude (where it starts to fall off i.e 25.3). The absolute HK magnitude of each type of star (O, A, G etc) was calculated by Karolina’s scripts. The area was calculated using figure 2, giving an area of 1.010 deg^2. With these values, we had everything required for a preliminary volume estimate for each stellar classification. The distance modulus was used to find the maximum possible distance of each type of star for the given apparent and absolute HK magnitudes. The volume we are studying was approximated as a small section of a sphere (approximately 41253 deg^2 in a sphere), with the aforementioned distances being the radius. This resulted in a preliminary volume estimate of 2.86 x 10^9 pc^3 for G-type stars.

Figure 1: Histogram of apparent magnitude in HK filter through 3” aperture. We used the value where the counts begin to drop off (25.3) as the maximum apparent magnitude in HK.
Figure 2: The entirety of the COSMOS CaHK catalogue plotted by position in the sky. This plot was used to estimate the area of the field of view for our study.

2) Numerical Error Script

Chief error analyser Sam completed his numerical error calculator to much excitement. The script can be used in tandem with other scripts, so will be very useful in the weeks to come.

3) Galaxy Removal

We have finalised our criterion for removing galaxies. The cleanest cut was given by the following:

𝐽−𝐾< −0.6143(𝐵−𝑢)−1.6121

𝐵−𝑢< −0.674

This gave 70% completion and 30% contamination, which will have to be taken into consideration later on.

4) Stellar Metallicity Plots

Metallicity affects the HK magnitudes of a star, and as such, a plot of some colour indices including HK against another colour index yields different curves for different metallicities (as well as for different star types). Below is our plot of (HK-g)-1.5(g-i) against g-i (figure 3). Some metal-rich K and M stars appeared off to the right of figure 3, however they were cut out. Comfortingly, it looks rather like the plot from the Pristine Survey included in the week 2 blog post – a comparison is given in figure 4.

Figure 3: Plot of (HK-g)-1.5(g-i) against g-I for stars of different metallicities and types.
Figure 4: Comparison of our own plot with that of the Pristine Survey.

We are also in need of a condition to separate out stars by spectral type. To that end, Karolina started to plot temperature against the g-i colour index (figure 5) to see if there was a relationship between temperature and g-i which cleanly removes other stellar classes. However, there is some overlap between G type stars and the F and K type stars. Further investigation into how to separate G and K type stars from others will be done next week. Both figures 3 and 5 were created using data provided by David from POLLUX, in future we can apply the same methods to obtain results from our actual catalogue.

Figure 5: Plot of inverse temperature against the g-i colour index for F, G and K type stars.

Mainly so we could shoehorn a gif somewhere into this blog, we also plotted normalised flux against wavelength in the region of HK lines for different metallicities, from Solar ([Fe/H] = 0) to hyper metal-poor ([Fe/H] = -5.0). However, this gives an excellent visual representation of the changes in stellar spectra due to metallicity and in particular, the changes in HK magnitude we are looking for in this project. The data for this plot was obtained from POLLUX for a 5000K star (available at

Figure 6: Plot of normalised flux against wavelength for a 5000K star at varying stellar metallicities


LAMPSS – Week 3 – Progress!

This week, on the coding side of things all was well. A python script to correctly determine the magnitude of our Sun was written. In future, we can extend the same script to stars to a distribution of G-type stars generated by us, and see if we get the same magnitudes out from the script as we put in. A script for analytical error was completed. However, since many of the errors we will be dealing with are not symmetrical about the mean, numerical errors will have to be used. Our chief error analyser is in the process of making a numerical error Python script.

There was also progress in the removal of galaxies from our CFHT CaHK catalogue. To do so we used the J & K filters. J and K are two filters with central wavelengths in the near-infrared. The J-K index can be used to distinguish between stars and galaxies because the redshift of an object effects the apparent magnitude of that object in the J and K filters more significantly than in other filters. To find a criterion by which to separate stars and galaxies, we plotted J-K against various combinations of other colour indices on the software TOPCAT. Using the TOPCAT subsets function, the number of galaxies retained and stars lost was found and then used to decide the most effective criterion. An example of this is shown below for a J-K against B-V plot. Stars were defined as having a redshift of zero, and an object in the catalogue was classed as a galaxy if it had z>0.1.

Figure 1: TOPCAT plot of J-K against B-V. Galaxies are the red data points and stars the blue. The equation of the black line was used as the criterion to define stars and galaxies.
Figure 2: TOPCAT subsets table for the J-K against B-V plot.

However, there are a few issues with our method. There are many galaxies below the line in figure 1, meaning our catalogue would still contain galaxies after the galaxy criterion was applied. Furthermore, many entries in the catalogue had a colour magnitude of +/-99.9 as there is no actual data on that colour magnitude. This is not good, as data points were flung out all over the plots. To try and resolve this we considered several alternatives. Firstly, a ratio of the apparent magnitude of an object in a 2” aperture and the apparent magnitude in a 3” aperture should be unity for point-like stars but not for galaxies. Although we soon dismissed this idea, as atmospheric seeing means not all stars are point-like and some galaxies are far enough away to appear point-like leading to lots of contamination. Secondly, we considered using the ‘Stellaricity’ data in the COSMOS catalogue. Stellaricity is a parameter that is equal to 0 for galaxies and 1 for stars. However, many entries in the catalogue have Stellaricity somewhere inbetween 0 and 1, and it is unclear how Stellaricity was quantified by COSMOS, so we cannot use Stellaricity.

Therefore, we must use the J-K method to remove galaxies and find the criterion that gives the highest completion (percentage of stars kept) and lowest contamination (percentage of galaxies kept). We will finalise results from this next week, as well as revise our estimate for the volume of the study region.



To start the week, we began by narrowing our search to G and K type stars. Thanks to 0, B, A and F stars being too short-lived for their own good, we can discard them, as a true low-mass population III must be nearly as old as the universe. Furthermore, by considering the Jeans mass criterion for star formation and using the temperature of the universe at several hundred million years old (when the first stars began forming), it can be shown that first generation M stars are very unlikely.

Figure 1: The Jeans Mass criterion for star formation, if the mass of a gas cloud is greater than MJ then the cloud will collapse (k = Boltzmann constant, T = temperature, µ = mean molecular weight, mH = mass of Hydrogen and Rc = Radius of cloud).

As mentioned last week, we will be using data from the CFHT, with particular interest in the filter including the CaHK lines. CFHT was one of the telescopes used in the Cosmic Evolution Survey (COSMOS), a much larger astronomical survey which covers 2deg2 of the sky and looks to probe the formation and evolution of galaxies. Courtesy of Karolina, the filter profiles of the i, g and CaHK filter of the MegaPrime/MegaCam on the CFHT were plotted. She then proceeded to plot the spectra of a 5000K G type star, opening the doors for the return of everybody’s favourite unit – the erg/cm2/s/Å.

Figure 2: Plot to show the filter profiles of the infrared (i), green (g) and CaHK filter profiles of the CFHT (Information from
Figure 3: Raw spectra of a G type star at 5000K.

We are after the magnitude of the stars in different filter bands in order to make a metallicity scale and start to evaluate which stars in our sample are metal-poor. An example is shown below from the Pristine Survey.

Figure 4: An example of how to separate stars by metallicity by creating a colour-colour plot using i, g and the metallicity-sensitive CaHK magnitude. Taken from the Pristine Survey (Starkenburg et al 2017).

The apparent magnitude (m) is related to flux as follows: m = -2.5log10(F) + ZP. Therefore, we must obtain the flux density per wavelength of stars in different filters and then integrate with respect to wavelength to first get the flux, enabling us to calculate the apparent magnitude in that filter. We first need to interpolate and normalise the filter profiles before the convolution of the filter profiles and stellar spectra can be plotted. The first step next week will be to test our method by obtaining the apparent magnitude of the Sun (A G2v star). As an example, a convolved plot of the i, g and CaHK filter profiles from figure 2 and the G type spectra from figure 3 is shown below.

Figure 5:  The flux density per wavelength of a 5000K G type star through the i, g and CaHK filters.



The project began with a healthy uncertainty about what we were actually doing. As it turned out, we were to be searching for metal-poor stars in the halo of the Milky Way, with the ultimate goal of finding a low-mass Population III star. A task that seems unlikely to prove successful, given that no Population III star has ever been directly observed. The first step was to prepare a brief presentation summarising the project which we delivered to our peers and project supervisor – the esteemed Dr Sobral. We outlined that we would be using data taken by the CFHT using the CaHK filter, a narrow band filter which encompasses the two wavelengths of the spectral lines important to our search. They are the CaHK absorption lines at 3968Å and 3934Å respectively. These two spectral lines of ionized Calcium are very prominent in stellar spectra, as such, are a useful tool in identifying metal-poor stars – which show extremely shallow H and K absorption lines in their spectra. Below is a plot of flux density per wavelength against wavelength for a star from the Sloan Digital Sky Survey (SDSS) spectral database. The H and K absorption lines can be seen to the left of the plot and are very strong in comparison to other spectral lines.

Figure 1: An example of a stars flux density per wavelength against wavelength plot taken from the SDSS spectral database (1Å = 10e-10m and 1erg = 10e-7J).

A lot of background reading was also done by the group during this first week. A fundamental concept is that the metallicity of a star is commonly quantified by the iron content of the star and a logarithmic comparison of that iron content to Solar abundance, as shown by the equation below. The CaHK lines are preferential to the spectral lines of iron because they are more prominent, especially at the low metallicities we hope to be looking at.

Figure 2: Equation that defines stellar metallicity. For example, a star with [Fe/H] = -6.0 would have a millionth of Solar iron abundance.

We took a particular interest in the Pristine survey (Starkenburg et al. 2017), a survey also conducted on the CFHT using the metallicity-sensitive CaHK lines. The survey was successful, and whilst no Population III stars were found, the HK lines proved effective in identifying ultra metal-poor stars ([Fe/H] < -4.0). We also found out that most of these ultra metal-poor stars have an unusual abundance of Carbon ([C/Fe]), which worryingly suggests a ‘metallicity-floor’ for star formation. However, the eloquently named star Pristine 221.8781+9.7844 (Starkenburg et al. 2018) is both ultra metal-poor and has a low [C/Fe] abundance which is a promising sign in our (rather hopeful) search for Population III stars.

Next week will start to work on Python scripts to apply the CaHK filter profile to the spectral data given and thereby work out CaHK magnitudes for the stars. Also, we will start to consider a plethora of other factors, such as, the removal of galaxies from out data set and stellar lifetimes for different spectral types of stars.