PlanetFinder

Using Spotfire in the Search for Extrasolar Planets

Motivation:

The search for extrasolar planets is a recent area of interest in astronomy research. In the last few years, refinements in instruments have lead to successful detection of planets in over 70 solar systems besides our own. These detections have been accomplished using a method known as ‘Doppler spectroscopy’, which is capable of detecting planets with a certain range of characteristics. Other methods exist with the potential of discovering planets outside this range of characteristics, but they are not yet as refined and require more resources. It is therefore important to make careful choices about which star to take measurements from. Since it may be more likely to find additional planets in solar systems already containing planets, it seems logical to turn these new instruments first to those stars that already have confirmed planets. Below are a chart of planet detection methods and a description of their limitations. On the chart, the light green rectangle represents the range of planets that may have human habitable temperatures and gravities.

Doppler spectroscopy uses the Doppler shift induced in a star’s spectrum to detect motion caused by a planet. It is limited to finding planets with large mass or a close orbit and only works with F or cooler class star spectra.

Astrometry (SIM) uses the change in a star’s position induced by a planet in orbit. It is capable of detection planets of smaller masses only at larger orbits (dark green line on chart). It is limited to stars within 10 parsecs (about 33 light years) of us and includes only a small portion of the human habitable planet space.

Transit photometry (Kepler) uses the change in light intensity from a star as a planet eclipses our line of sight. It is capable of finding planets in most of the human habitable planet space. However, only about 0.5% of planets in this range will be oriented such that they are detectable and it also requires more than 4 years of observation.

Data:

The California and Carnegie Planet Search (http://exoplanets.org/) has a collection of data on currently known planets, including planet mass, orbital period, orbital distance, and orbit eccentricity. From the SIMBAD online astronomical database (http://simbad.u-strasbg.fr/Simbad) I collected data on the stars that these planets orbit, including stellar spectra classification, visual magnitude, distance and galactic coordinates. From this information I also calculated XYZ coordinates of the star relative the sun. This data was imported from Excel (used for calculations) into Spotfire (http://www.spotfire.com/) with great ease. When all was said and done, I ended up with these entries for each planet:

Mass (Mj, Jupiter masses)

Period (days)

AU (orbital distance, earth = 1 AU)

E (eccentricity, how elliptical an orbit is, 0 = circular)

# in system (total number of planets detected in this solar system)

Star mass (in solar masses)

Star spectral type (sun is a G, F is hotter, K and M are cooler)

Spectral subtype (0 to 9, indicates how close spectra is to next cooler type, sun is G5)

Luminosity class (indicates stellar size and evolution, sun is G5V)

Distance (in light years)

Galactic azimuth (degrees from angle to galactic core)

Galactic elevation (degrees from angle of galactic plane)

X Y Z (in light years, where +X is towards galactic core and +Z is above galactic plane)

Using Spotfire to demonstrate an obvious correlation:

Orbital mechanics dictates that a planet’s orbital period is related to its orbital distance. This was easily and clearly demonstrated using Spotfire by producing a scatter plot of AU vs. Period. The relationship is obvious.

Of course, Spotfire lets you do other cool things to convey more information. So I made the shape circular (after all, they are planets), and I made the size correspond to the mass of the planet. All this took a few clicks of the button, and I could even set the drawing order based on planet mass so the larger circles did not completely occlude the smaller. I also wanted to convey some information with color. Some stars have more than one planet, so I wanted to use color to distinguish planets that were in the same system.

Now, here I discovered a small annoyance. Spotfire very easily lets you assign color to a data dimension, so I set color to the number of planets parameter. This worked ok, but not perfect. The single planet systems all showed up one color (yellow). Upsilon Andi, the only star that has 3 planets, had its planets show up another color (blue). But the several systems with two planets all were the same color so you could not tell which shared stars. Ultimately, to get the display seen above, I had to set every entry to the color yellow, then set colors to each star entry that had more than one planet. So Spotfire easily allows mapping a display parameter to an entry, but it would have been even more powerful to map a display parameter to a subset of entries as well.

But once I finished, I had the scatter plot above, which shows that orbital period is a function only of AU. It can be seen that planet mass is not a factor and that multiple planets in the same system also does not effect period (which I suspect astronomers already knew).

Using Trellis Plots to compare data in different classes:

A trellis plot can be used to examine similar relationships in several classes of data. For an example, lets compare the visual magnitude of stars with their distance. There should be some general relationship since more distance stars should be less bright (higher magnitude value). Actually, this relationship should be distance squared. But now we will use a trellis plot to create this scatter plot for each stellar spectra class. This way we can easily see if every spectral class behaves in the same manner and note discrepancies.

We can see that they are generally grouped along an arc. But to look for more detailed relations, lets use color to represent star subclass. Blue will be hotter stars and red will cooler stars (matching the actual star colorings). Since stars also have a luminosity class, we can use size for the different classes. Luminosity class V is a normal generation star, where lower class numbers are giant class stars, and hence are brighter. So I wanted size to be larger for lower classes. I brought up a pop-up menu on the query device for Luminosity class (a check box) and under the Set Properties menu item I found reverse sorting. So this accomplished what I desired and above you see a plot that shows star magnitude along with the parameters that should affect it (distance, type, subtype and luminosity class). It is interesting to note that there are only cooler (red) stars in the F spectral class plot. This is because the Doppler spectroscopy method does not work well with the spectra of hotter stars. F class is the hottest class of the 4 above and data is only available for the cooler of these.

Adding columns from data:

One technique of astrometry can only detect planets if their mass – AU product is above a certain ratio. So to add the ability to query for planets detectable by astrometry, we use a Spotfire function that will add data columns based on functions of existing columns and we add this product. We can now generate queries for SIM 10 and SIM 500 astrometry methods, which have different thresholds for this product.

Creating a detailed query and outputting to HTML:

So, lets say we are searching for likely candidate systems to use Transit Photometry on. We want to search systems with existing planets based on a guess that if a system has one planet, it may have more. We think that eccentricity might play a role. We are looking for habitable planets, so we would like the system to have an open band (no planet) near that range. So we set up a scatter plot to show eccentricity vs. AU. We set size based on multiple planets in the system (looking for systems like ours with many planets). Mapping star subtype to color will let us see how closely the star matches our sun (without putting filter limits on it). So now we set sliders and check boxes (too easy) to set our query as follows:

Stellar Spectra class G only (sun like)

Luminosity class V only (sun like)

Star mass 0.9 to 1.1 (within 10% of our sun)

Since Spotfire cannot do OR queries, we cannot set an exclusion band for AU around 1, so we will just have to do this by inspection. We end up with a decent selection and looking at ones outside the exclusion area, you’ll find two planets exist in star 47 U Ma and both have stable orbits well outside 2 AU. I will mark these two records. Now we can output the resulting plot to an HTML file, complete with the query and selected records.

planets.sfs

Thursday, March 07, 2002 12:55:22 AM

Scatter Plot

Scatter Plot

SQL Query

select
*
from
W:\viz\planets.xls
where
Luminosity Class in (V) and
((0.9 <= Star Mass) and
(Star Mass <= 1.1)) and
Star Type in (G)

Table of Marked Record(s)

# in system	AU	Dist (ly)	E	GC(AZ)	GC(EL)	Luminosity Class	Mass (Mj)	Period (days)	Star	Star Mass	Star Subtype	Star Type	V Magnitude	X (ly)	Y (ly)	Z (ly)	"Mass AU ratio"	"Mass AU product"
2	2.09	45.965999999999994	0.06	175.78	63.37	V	2.56	1090.5	47UMa	1.03	1	G	5.1	39.03073955225916	-5.866530405908071	23.55967629234893	1.2248803827751198	5.3504
2	3.78	45.965999999999994	0	175.78	63.37	V	0.76	2640	47UMa	1.03	1	G	5.1	39.03073955225916	-5.866530405908071	23.55967629234893	0.20105820105820107	2.8728

From this we can see that 47 U Ma may be an excellent star for studying with Transit Photometry. It has one planet somewhat larger than Jupiter at an orbit outside that of Mars and another planet slightly smaller than Jupiter at an orbit inside that of Jupiter. It is a G1V star with mass almost equal our sun, but slightly hotter, a younger version. And it is a mere 47 light years away.

3D plots:

3D plots are hard to do well, but for some things they are maybe useful, and at least cool looking. I did a 3D plot of star location in XYZ coordinate space for a 200 light year cube with earth at the center. I assigned color to star type (matching their real colors)

To include another demonstration of using new column data to add visualization, I added a binned column based on star magnitude. I set bins to be < 6, 6 – 8 and > 8. I set display shape to these bins. Solid objects are magnitude < 6, these are stars that can be seen with the naked eye. Hollow shells are magnitude 6-8, these are stars that can be seen with binoculars. Magnitude > 8 stars can only be seen with a telescope and these are crosshairs on the display. Not massively useful, but way cool. Now this is what I call a starfield display!

Conclusions:

For quick looks and basic queries of complex data, Spotfire can’t be beat. Its quick, gives immediate feedback (incremental, reversible) and easy to use. Digging a little deeper, you find a broad range of functionality, though sometimes it took me a bit of looking to get what I wanted, but its clear even a moderate amount of practice and one would know the full abilities of the software. It provides a variety of graphs for display, including 3D, though the usefulness of that is questionable.

To keep things simple and manageable, Spotfire uses only conjunctive queries. On at least one display I wished to filter out a band of data, a disjunctive query. Although Spotfire gave no straightforward method of doing this, it still could have been achieved by creating a binned data column based on the band. Then one could filter out the bin corresponding to the band of data.

One thing I did wish for. Twice I added data columns of data. One of these, the binned data, added functionality to the display but setting display shape to reflect it. But for the mass – AU product, no easy way existed to show this as a constraint. It would be useful to be able to configure Spotfire to show filter bounds within the display.