Thursday, May 8, 2014

Modeling the Universe

Comparison of a Hubble image (left half) with an image from the Illustris
simulation (right half).  Credit: Illustris Collaboration.  Image source.
Modeling aspects of our world has been a constant effort ever since computers were invented.  Once broken down into mathematical equations, phenomena like basic laws of physics and weather patterns can be simulated by computers.  As computers have become more complex and efficient, they have been able to run more detailed simulations.  Even simple video games and the icons on new desktops behave according to our expectations of natural movement.

The Holy Grail of computing, however, is modelling an entire universe.  In theory this is an unreachable limit, as truly modelling a universe in real time would require a whole universe of computing power, which would just be the Universe modelling itself.  Nevertheless, computers and programmers can march ever closer to this limit as technology advances.

Today, Nature magazine published the results of the latest and (arguably) most sophisticated model of the universe so far (full text here).  The model was produced by Illustris, a collaboration of several organizations and universities, including MIT, Harvard, Cambridge, and Princeton.  The model simulates the interaction of normal matter and a giant web of dark matter in a 350 million lightyear-wide cube over the course of 13 billion years.  It took 8,192 computer cores about three months to run the 12 billion virtual particles through their paces.  As many news sources point out, the results look remarkably like our current universe, and the researchers reported that the distribution and proportions of chemical elements and galaxy shapes also match the known universe.

It may seem like this simulation proves a naturalistic history of the universe, but that is hardly the case.  First of all, the model did not begin with a Big Bang, but with a wide distribution of matter that supposedly resembles the Universe as it was 12 million years after the Big Bang.  Second, as with nearly any model, it is specifically designed to closely match real-world observations at the end.  Generally, scientific models are not cases of setting initial conditions and then letting the system run free to see where it ends up.  Rather, they typically begin with known observations and measurements and work backward from there.

Let's take an example from my wife's work.  She is a groundwater hydrologist at an engineering consulting company in Duluth, and she regularly creates models of aquifer systems for mines and other entities.  Working under the known rules of the physics of aquifers, she starts with field measurements of aquifer depth and other properties, then the computer roughly extrapolates the data to the surrounding area.  She will then tweak certain attributes until the modeled aquifer reflects the behavior of the real aquifer as closely as possible.  It is not intended to be a true-to-life representation of the aquifer, but it is the best guess that can be developed without extensive field research, and it serves as a starting point for future projects that might impact the site.

Similarly, the Illustris model constructs a “best guess” for the development of the Universe in a naturalistic framework starting with several known quantities.  According to the Illustris website, it is “specifically tuned to match two quantities: the present day ratio of the amount of stars to dark matter…, and the total amount of star formation in the universe as a function of time.”  While I don't know the details of how the model was constructed, it is likely that many variables were adjusted until the model gave acceptable results in addition to the predetermined values.

Models only reveal a possible historical scenario; they are not proof of one.  I consider models to be like advanced hypotheses: they are well-developed possible solutions based on extensive data and research, but they have not undergone experimentation to test their accuracy.  Even if the Illustris model is perfectly physically viable, that does not mean it accurately reflects history.  Unfortunately, due to the nature of origins science, I do not believe that such models can be sufficiently tested against actual history so as to serve as proof of past events.  They can only be ruled out.

Of course, the Illustris model is far from perfect.  The authors of the paper admit that there are problems with the simulation, such as stars that are too old.  The simulation is also limited to scales at and above that of small galaxies, so details of stars and planets cannot be compared with reality.  Many of the shortcomings of the model are doubtlessly attributable to technological limitations, but I suspect that many other problems are caused by inaccurate naturalistic assumptions about our Universe.

No comments:

Post a Comment