Post has attachment
After two days, my screen is finally filled with orange stringy sausage thingies. Hurrah for science !

I've been re-examining an old data set, searching for hydrogen streams that may have gone unnoticed. Some known streams are very long indeed - these can't have been missed, because they'd be extremely obvious in the data. But shorter features could be hiding. Bright galaxies are a lot like bright light sources in ordinary photographs - they appear much larger than they actually are. One way to limit this is to plot contours, which show the structure much more clearly than intensity maps. As long as the galaxies aren't too bright, non-circular features in the contours generally show up much more easily than having to carefully try and adjust the contrast of a flux map.

The reason the galaxies look like these long cigar-like blobs is because the third dimension is velocity. We don't have great spatial resolution so generally the galaxies only appear as a circularish blob at any given velocity, occasionally just about resolved into something more interesting. But we have great velocity resolution. Because the galaxies are rotating, this means we detect them at many different velocity channels.

This method seems to be working pretty well : there are 2-3 nice examples of streams that are almost certainly real and maybe a dozen other weaker candidates. None of these would have shown up very clearly in standard maps. And I couldn't have made this figure back when the data first came in - I could have plotted the contours for all 93 galaxies here but it would probably have taken closer to two weeks rather than days.
Photo

Post has attachment
A little evening's diversion. A few weeks ago there was this ALMA press release (which I came across again today) about observations of the gas around the star LL Pegasi (http://www.almaobservatory.org/en/press-room/press-releases/1122-alma-adds-a-new-dimension-to-a-hubble-space-telescope-result). It was already fairly famous from Hubble observations thanks to its remarkably neat spiral pattern. The ALMA observations add velocity information and I wanted to see what this would look like in 3D. Actually I've been wondering about this for a while since there was a similar-ish press release about another sort-of similar object some time ago.

For those who aren't familiar with these types of observations, have a look at the gif in the press release first. There you see the data in a slightly more usual format, as a series of images. Each one shows the gas at some particular velocity along our line of sight. What I've done here is use each image as the slice of a 3D cube - it's fun to look at (maybe even useful) but it doesn't show you the true 3D structure of the object.

The last time something similar like this was doing the rounds I couldn't find the original FITS data I needed to display it. This time I didn't bother. I took the gif, converted it into a sequence of png images, then wrote a Python script to convert the image sequence into a FITS cube and then ran it through FRELLED (what else ? http://www.rhysy.net/frelled-1.html). Oh, and I interpolated extra velocity channels because there weren't very many in the gif (there might be more in the original data, I don't know). So there's a fair amount of extra processing for this one, but probably nothing that would result in any serious differences from the raw data.

What does it all mean ? Haven't got a clue, I just thought it would be nice to look at.
Animated Photo

Post has attachment
More experiments with all-sky HI data using frequency to set the radial distance. The realtime display in Blender gives a rather different appearance to the rendered views (see recent posts in this collection) because faces are one-sided in the realtime view - so the opposite half of the spheres aren't visible. Gives a more interesting and subtle effect.

Post has attachment
Geometrical Jiggery-Pokery

More fun with the all-sky HI data... sort of.

Recently I posted what the data looks like if you assume velocity is the same as distance. Of course it isn't really, but transforming it into true distance isn't so easy.

The data from the telescope consists of maps of the sky each one at a slightly different line-of-sight velocity. Knowing the position and velocity of each point in the map, it's possible to convert this into distance with some fairly ugly trigonometry (http://ircamera.as.arizona.edu/astr_250/Lectures/Lec_22sml.htm).

There are some intrinsic limitations to this that can't be avoided. For instance, if the gas is closer to the centre of the Galaxy than the Sun, the equations give two equally valid solutions and there's no easy way to decide which is correct. So that data has to be chucked out. Another is that if you've looking directly towards or away from the Galactic centre, you don't get meaningful velocity information - we can only measure velocity along our line of sight, but at those angles gas is moving entirely across the sky except for some small random motions. So data at angles close to the centre and anti-centre needs to be thrown out too.

Then there's the problem of how to display the data. Previously (http://astrorhysy.blogspot.cz/2013/11/damn-thats-nice-piece-of-gas.html) I tried to convert the raw velocity cube to a distance cube by applying the equations to each pixel in the original data. So knowing position, velocity and intensity of any point gives a corresponding position, distance and intensity in the new data cube.

This creates an extra problem. Although the original data is fully sampled (that is, each map of the sky at each velocity channel is complete - every pixel has measured values), that isn't automatically the case for the output distance cube. To simplify, imagine that position x,y corresponds to i,j in the new data set. The problem is essentially that position x+1,y+1 corresponds to something like i+2,j+2 - there are gaps. You could try interpolating the missing values, but it's not so easy - and you still lose data, because in some regions in turns out that multiple points in the original coordinates correspond to the same point in the new coordinates. You need the resolution to be adaptive.

Which is where this funky geometry comes in. What we have here are a series of planes in Blender, each one corresponding to a different velocity channel. By transforming the vertices to the corresponding distance, the faces between them automatically interpolate the missing regions. The animation just overlays each successive velocity channel converted into distance.

.... or at least that's the idea. I'm really not sure if this is working correctly. The funky shape might be a consequences of the non-trivial equations, or I might have set something wrong. It's very hard to visualise what the equations look like in 3D, which is what this is supposed to help with. This requires further thought, but it's quite nice to watch (the flickering dark shadows are Blender rendering artifacts).
Animated Photo

Post has attachment
You've probably seen the press release going around about the "most detailed map of the Milky Way" (http://phys.org/news/2016-10-scientists-milky.html). Strictly speaking it's the most sensitive, highest-resolution all-sky HI survey of the Milky Way, but I think we can forgive the simplified headline in this instance.

Here's a little comparison between the previous Leiden/Argentine/Bonn survey, the new HI4PI survey, and the GALFA-HI survey. It's a bit crude - HI4PI is definitely better than LAB, but I probably haven't done it justice. More details would probably show up if I chose better settings. Still, GALFA-HI is clearly still far superior in terms of resolution.

I was hoping to use the new data to remake the Hydrogen Sky project (http://www.rhysy.net/the-hydrogen-sky.html) - last time I relied heavily on GALFA-HI, because the low resolution of LAB doesn't give such nice results. The problem with GALFA is that it's an Arecibo survey, so it has fantastic resolution and sensitivity but limited coverage. Which makes it rather tricky to find photographs of the correct orientation to overlay the data with any kind of accuracy.

Although HI4PI is better than LAB, I'm not sure it's worth the effort. The LAB data is only 250 MB, HI4PI is 32 GB (high spatial and velocity resolution and 32 bit data files instead of 16). I'm not even sure my computer (even in work) could load a 32 GB file into memory and the improvement wouldn't be that dramatic. Not that the survey isn't much better for science, just not so much for visualisation, unfortunately.
Animated Photo

Post has attachment

Post has attachment
More fun with all-sky HI data. The radial distance here is velocity, not real distance, hence the weird-looking structures. Colour is scaled based on the flux range in each velocity channel, so you see a lot more structure than in the earlier versions.

Might try another attempt at converting this to true distance, but my spare time is pretty close to zero for the next couple of weeks.
Animated Photo

Post has attachment
Quick animation test. Needs more frames really. There's nothing new here (detailed explanation : http://astrorhysy.blogspot.cz/2013/11/damn-thats-nice-piece-of-gas.html). But last time it involved a laborious manual procedure. Now I've got a pipeline to process data of arbitrary coordinate systems (Cartesian is for wimps !), and it seems to be working. Sort of semi-functional in the realtime view but that needs a lot more work. Turns out you can get away with quite low-poly spheres without any noticeable distortion, which is good.
Animated Photo

Post has attachment
That's what it's supposed to look like,
Photo

Post has attachment
Coordinate mix-up turned my spherical HI data into some sort of minimalist exploding peach...
Photo
Wait while more posts are being loaded