Idius Land!

10 million particles

Sat, 19 May 2012 05:00:38 +0000

Here's a video of the current version of GSnap manipulating a freshly created GADGET-2 snapshot containing slightly more than 10 million particles. With my current computing resources, it appears that I won't be able to run full merger simulations with this many particles, but it's nice to see that GSnap can easily handle snapshots of this size. This video also demonstrates some of the GUI improvements mentioned in the previous post.

View with 720p HD quality for best results.

GSnap is Usable

Mon, 16 Apr 2012 01:04:09 +0000

The GSnap GUI workflow is now much more streamlined. Clicking on any point in the preview window now shifts the selected point to the center of the frame and rotates the snapshot so that the current viewing direction becomes the new z-axis. A context menu and keyboard shortcuts have been added to the preview window as well. A particularly useful shortcut allows the user to compute

$\sigma_*$ for stars appearing in the slit by simply pressing "s" while the preview window is active. There is little need to switch back and forth between the main window and the preview window.

GSnap now has a simple logo. It will eventually have its own web page and a built-in help browser / user manual. At this point, I need to work on other aspects of my research project. Before I begin the other work though, I thought it would be good to summarize GSnap's current capabilities.

GSnap can:

Read and write GADGET type 1 (Springel) snapshot files.
Compute the mass-weighted velocity dispersion ( $\sigma_*$ ) of any particle type using a diffraction slit (mask) of arbitrary dimensions from an arbitrary direction.
Compute the mean, maximum, minimum, and standard deviation of $\sigma_*$ for a set of random viewing directions (1000 directions, by default).
Arbitrarily shift particle positions and velocities (i.e., shift the origin and drift velocity of the coordinate system independently for each type of particle).
Rotate the particle positions and velocities (i.e., rotate the coordinate system independently for each type of particle).
Compute the center of mass of any particle type or any combination of particle types.
Measure $\sigma_*$ for different stellar age groups.
Create simple visualizations of a snapshot. Images can be saved to disk as PNG or JPEG images. The system can also be viewed interactively.
Compute the stellar velocity dispersion by focusing measurements on the center of mass of the system (the default) or focusing the measurements on the progenitor nuclei.
Create fairly high quality visualizations of the gas component using a simple ray casting algorithm and SPH smoothing kernel. The simulation smoothing lengths are not used in this process; the code computes smoothing lengths itself. So technically, any particle type could be visualized using the volume rendering technique (dark matter halos, for instance).
Interpolate particle positions between snapshot files using a 3rd-order interpolation scheme. This is primarily for the purpose of creating movies of the merger simulations. An arbitrary number of intermediate snapshots can be generated.
Create plots of stellar velocity dispersion vs. time (if provided with a directory full of snapshots).
Create plots of black hole accretion rate vs. time (if provided with a directory full of snapshots).

More features are planned.

GSnap is written in C++. Its only dependency is the Qt Framework. The automation scripts are written in Python. Since the Qt Framework and Python are available for GNU/Linux, MacOS, and Windows, GSnap can be used on all three major platforms without modification. When I finish my Ph.D. research, GSnap will be made publicly available (GPL licenced).

Adding new features is fairly straightforward. In particular, it would be easy to add the ability to read more file formats—including file formats from other astrophysical simulation programs. Most of the current features deal with measuring

$\sigma_*$ because my dissertation work primarily deals with the evolution of

$\sigma_*$ during galaxy mergers. GSnap is not intended to compete with other codes such as Matt Turk's yt, Klaus Dolag's Splotch, Patrik Jonsson's Sunrise, Nick Gnedin's IFrIT, Daniel Price's Splash, or any other analysis + visualization code; it is simply designed to help me with my particular research project.

GSnap GUI Improvements

Fri, 13 Apr 2012 22:24:35 +0000

Here's a video demonstrating new features of the GSnap GUI. It's clear that the work flow could be streamlined quite a bit.

Set the quality to HD for best viewing results.

Volume Rendering 2.0

Sat, 07 Apr 2012 03:30:08 +0000

I made some improvements to the volume rendering scheme today. Here are the snapshots from the previous post, rendered using the improved code:

And here's a video animation of the test simulation:

The video plays very fast because I didn't create interpolated frames.

Volume Rendering

Tue, 03 Apr 2012 03:48:20 +0000

While on vacation, I have been playing with adding a simple volume rendering scheme to GSnap. The scheme uses my own smoothing kernel to distribute each particle's mass among the surrounding voxels and a volume hashing scheme to accelerate the nearest neighbor search. I'll eventually use the volume rendering scheme to estimate dust attenuation when making more realistic images of GADGET snapshots.

It needs a lot of work before it will be ready to use. Here are some early results:

Nucelus Centered vs. CM-Frame Velocity Dispersion

Sun, 25 Mar 2012 07:29:16 +0000

GSnap now allows the user to specify nucleus-centered velocity dispersion measurements. Previously, when computing velocity dispersion statistics for 1000 random directions, the diffraction slit was placed on the center of mass of the merging system. This meant that early in the merger process, when the progenitors were separated significantly, some measurements of

$\sigma_*$ were based on a very small number of stars in the outskirts of the galaxies. With the nucleus-centered technique,

$\sigma_*$ measurements always focus on the nuclei of the progenitor galaxies. The plots below illustrate the differences between the two techniques.

Nucleus-centered:
CM centered:

Using the nucleus-centered technique, the left half of the plot is much flatter and the values of

$\sigma_*$ are larger. The right halves of these time series are essentially the same because the two nuclei move closer and eventually coincide at the center of mass of the system. The lower halves of the plots show the accretion rate---just as in the previous post.

Accretion Rate vs. Time

Sun, 18 Mar 2012 09:01:02 +0000

My advisor suggested that I plot black hole accretion rate versus time (as a proxy for AGN activity vs. time) and compare that with the velocity dispersion vs time plot as well as the morphologies of the systems. Here are my first (very tentative) results on my test simulation.

The first half of the plot will be somewhat more meaningful once I have implemented the progenitor-centered velocity dispersion calculation. The relatively flat section in the first half of the plot corresponds to times when the two progenitor nuclei were separated significantly; the measurement of the mean

$\sigma_*$ was performed on some lines of sight that did not pass through either progenitor. Increasing the spatial resolution (by increasing the number of particles and decreasing the softening length) will also provide more meaningful results in general. At the moment, it appears that there is likely some mild AGN activity immediately following the first encounter. A second era of accretion begins during the second pass, then quickly fades. The accretion rate is highest during the transition from the oscillatory stage to the phase mixing stage as the system rapidly approaches its quiescent velocity dispersion.

Below, I present some simple GSnap-generated images at various stages during the merger. Each snapshot is shown from several different viewing directions. I will eventually produce analogous Sunrise images and spectra from higher resolution simulations.

At 0.5 Gyr:

At 1.5 Gyr:

At 2.25 Gry:

At 2.35 Gyr:

At 2.5 Gyr:

At 2.66 Gyr:

At 3.0 Gyr:

At 4.0 Gyr:

GSnap Previewer - Optimized

Sat, 17 Mar 2012 23:37:37 +0000

After thorough optimization and OpenMP parallelization, here's the newest incarnation of the GSnap previewer. It's much faster than the proptotype! Note: The video quality is significantly better if you play it at 480p quality by clicking the quality button at the bottom of the player.

GSnap Previewer

Wed, 14 Mar 2012 04:36:01 +0000

I've added an interactive previewer to my GSnap code so that I don't have to rely on IFrIT for viewing GADGET snapshots. The code still only depends on the Qt framework. Here's a video of the previewer in action:

GSnap Progress

Mon, 12 Mar 2012 04:14:21 +0000

Significant progress has been made recently on my analysis code, GSnap and its associated Python scripts. Most notably, I have completed the same sort of analysis that was performed in Stickley & Canalizo (2012) on a dissipative GADGET-2 merger simulation and I have implemented a convenient snapshot viewing routine so that the snapshot can be viewed with the same program that analyzes and manipulates the systems, rather than using two separate programs.

It appears that the evolution of stellar velocity dispersion in dissipative mergers is qualitatively similar to the process that I described for dissipationless mergers. Here's a preview:

I'll soon add another method of measuring velocity dispersion that centers on the nucei of the progenitors rather than staying fixed on the center of mass of the system. I'll also take a look at the velocity dispersions of "old" star particles versus younger star particles.

Here are some examples of the first few snapshots that I visualized with GSnap:

I could eventually add color and use the SPH information and stellar ages to add a bit of realism, but this is all I really need. In order to compress the dynamic range of the brightness into something that the image format could handle, I had to normalize the raw image data so that the brightest pixels had a value of 1.0. Then, I re-scaled the value of each pixel,

$f_{ij}$ using a power law transformation,

$f_{ij}\,\rightarrow\,f^{\,\gamma}_{ij}.$

The exponent,

$\gamma$ is an adjustable parameter; I've achieved the best results with

$\gamma=0.35 \pm 0.05$ .

I have been notified that I will soon be registered on the Triton Resource at the San Diego Supercomputing Center. UCR has given me enough computing time to run one or two simulations.