For millennia, mankind gazed out into the mesmerizing night sky, without knowing that what they were seeing were stars just like our Sun and billions of other galaxies making rest of our Universe. The ``seeing by eyes" later turned into ``observing through telescopes" and today the detection and discovery of objects in the heavens depends far more than ever on statistical analysis tools, theoretical models and algorithms that are built by the scientists down herefor optimally extracting out interesting objects from the astrophysical data sets which exist out there.
The first ever significant and noteworthy detections of objects in the cosmos made by a member of homosapiens from the planet Earth, in my view, can be traced back to the time of Tycho Brahe's observations. On November 11, 1572, on his way back home, Brahe was contemplating stars in the clear sky when a new and unusual star almost as bright as Jupiter sitting in the Cassiopeia constellation grabbed his attention (as mentioned in his 1573 book, ``De Nova Stella" (On the new star) ) . Tycho reports (from Burnham's Celestial Handbook):
I was not ashamed to doubt the trustworthyness of my own eyes. I had no further doubts. A miracle indeed, one that has never been previously seen before our time, in any age since the beginning of the world.
This remarkable discovery was important as it refuted the previously believed Aristotelian belief in an unchanging celestial realm.
The instruments Brahe used for observing the night sky were mural instruments or sextants - instruments which can only mark positions of the stars on the celestial sky. The discovery of refracting telescope by German-Dutch spectacle maker Hans Lippershey in 1608 created a revolution in the field of astronomy. The invention soon proliferated across the Europe and the scientists began making their own telescopes which led to a plethora of astrophysical discoveries. One of them was smart enough to make a telescope of magnifying capability that could see celestial bodies more distinctly than was ever possible before. Galileo Galilei built a telescope in 1609 and pointed it towards the sky and made stupendous discoveries, hitting the ball right out of the park. For the first time he observed hitherto-unknown objects in the heavens - spots on the Sun, craters on the Moon, the stars of the Milky Way, and in 1610 he observed the first four satellites of Jupiter- Io, Europa, Ganymede, and Callisto. Proving that not everything orbits Earth, he promoted the Copernican view of a Sun-centered universe (for which he obtained unpleasant reward from the Roman Catholic Inquisition- the infamous story we all know).
From there on, every scientist tried to construct his own version of the telescope by improving the design of the previous one. Best of these instruments were constructed by the Christian Huygens of the Netherlands. Huygens discovered Saturn's ring and its largest satellite Titan in 1655. In 1781 amateur astronomer William Herschel discovered the planet Uranus. With the help of his sister, Herschel built over 400 telescopes during his lifetime which he used for observing the night sky. Henrietta Leavitt discovered more than 2000 variable stars called Cepheids between 1907 and 1921 while studying the photographic plates at Harvard to understand fundamental properties of the stars. The period-luminosity relationship of these peculiar set of stars which had their luminosity flux fluctuating periodically (hence variable stars) play crucial role in measuring distances to the other galaxies. Her work helped American astronomer Edwin Hubble to measure galaxy distances in the 1920s, which led to his realization that the universe is infact expanding. Maarten Schmidt in 1963 discovered the first quasar providing a probe of the universe back to substantial redshifts (or time) while studying the spectrum of radio source 3C 273. He fathom the unfamiliar and perplexing emission lines in the spectrum of the source 3C 273 were the Hydrogen lines red shifted because the object was located billion of light years away.
Such detections fall under the category of what is called direct imaging. The telescope is turned towards the sky and the object of interest becomes apparent in the images of the telescope. The object (star, structure or some signal) overwhelmingly stands out from the background (stars or noise) because of some kind of extraordinary physical property it possessesand its high signal strength and easily gets an attention. The only major effort lies is in building the instrument (telescope or spectrometer). The accidental discovery of CMB by Penzias and Wilson (1965) was an unintended detection of steady microwave signal which they initially took to be a very strong noise signal being detected by their receiver. However, this noise signal was 100 times stronger than they were otherwise expecting, which made them thinking that something mysterious was going on. Soon after talking with Prof. Bernard F. Burke about Jim Peebles paper on Universe’s Blackbody radiation, Penzias and Wilson realised the significance of their discovery.
Earlier, scientists used simple telescopes to explore the heavens. As the technology emerged and grew, experimentalists began pushing down the luminosity detection limit of the telescopes in order to observe more and more objects in the sky. Today we are well on our way to unravelling many of Universe's mystery, living in what maybe the most remarkable age of astronomical discoveries. Today, 400 years after Galileo, astronomers use giant mirrors on top of the remote mountains to survey the cosmos. Scientists have even launched telescopes into space, high above the disturbing effects of our atmosphere where the Universe is transparent and the view is breath taking. On 25 April 1990, a vision became a reality that was set forth by joint collaboration of European Space Agency (ESA) and National Aeronautics and Space Administration (NASA) together in 1970s of working together to design and build Hubble Space Telescope. The unprecedented deep images of the Abell cluster (cluster of galaxies) delivered by Hubble is stunning. However, even after employing high tech new generation telescopes there always remain a large number of objects that go unnoticed in the images simply because of their low contrast or/and heavy background contamination or/and observational errors in the data making them difficult to be easily detected. For example, direct imaging of exoplanets is extremely difficult or in most cases impossible simply because being small and dim, planets are easily lost in the bright glare of giant stars they orbit. But that does not mean the planet is not there. In such cases, we just need to be clever enough to be able to detect such low contrast objects. Detections in such cases then require some sort ofdetection algorithm that can filter out the object of interest from the data.
These detection algorithmsare mathematical models which are a convolution of the physical theory and some signal processing mechanism. The physical theory part helps in identifying the data points that behave in accordance to the phenomena we are interested in finding, while the signal processing part filters out the data points that follow the physical theory, hence revealing the structure. Clever innovations lead to advancement of these techniques that then make it feasible to apply them in different and complicated scenarios to detect different kinds of astrophysical structures, objects, events or waveforms. These mathematical tools then require 3 conditions for a positive detection:
· The signal of interest is infact there in the data which is being analysed.
· The signal follows the predicted physical model/pattern.
· The signal of interest is above the detection limit. Detection limit depends on the signal strength, detection technique in work, precision of the data measurement and the level of contamination (unwanted background noise).
There are different algorithms that exist in the literature today used for signal processing and structure detection. However Match Filters(MF) are widely used in the field of astrophysics - going from detection of clusters of galaxies to characterisation of light curves of stars with eclipsing exoplanets, extracting broad spectrum to detect supernovas,etc. Match filtering (originally developed by Wiener in 1949 for signal processing) is a template fitting technique used to extract the signal (object of interest) efficiently from the contamination (background or unwanted stars). It detects the signal of known shape by maximizing the signal to noise ratio (SNR) in the presence of additive stochastic noise. It does this by weighing the data points that follow the predicted model/pattern heavily which in turn increases the power of these desirable data points hence uncovering the hidden signal. For better detection of gravitational waves, the LIGO (Laser Interferometer Gravitational-wave Observatory) group used a form of Match Filter technique which was prepared based on the general relativistic models of the binary blackhole (BBH) merger waveforms using which they identified two signals GW150914 and GW151226.PyCBC and GstLAL were 2 different Match Filters made that used common waveform templatesbut differed in implementation. The MF was built to get a better quantified signal from the data. The GW signals are extremely weak signals immersed in an intense noise background. Pulling out a wave signal of such a low intensity from a dense background is only possible by using Match Filtering or some other signal processing technique.
Scientists often label new findings or new detections in terms of sigma. When Higgs boson was discovered in July 2012, scientists tossed around the phrase ``Its a 5-sigma detection" to describe the strength of the discovery. The probability of a value being measured while dealing with normal distribution depends on the mean (μ) and the standard deviation (σ). The further a measurement is from μ (i.e. towards either end of the bell curve, see Figure), the less and less likely it is of being measured at random, or due to a fluctuation in the background or just by-chance. A 3σ detection event has a 0.3% probability of occurring by chance, and a 5σ event has just a 0.00006% probability of occurring by chance. Physicists traditionally call a 3σ detection ``evidence" (that they take as being the minimum to be believed), while a 5σ detection is considered a ``discovery" (which is essentially 0% probability of the result being false). If the model is good and data is of good quality, the s value is high too. Rene Descartes said - When you do not have the power to follow the truth, follow what is most probable. “What happens if something that’s really just part of the background looks like a plausible signal?” That’s exactly what the sigma tells you! The higher the sigma value, the less likely it is that a signal is just a random fluctuation in the background model.
Another concept that is needed to be verified for scientific detections and discoveries is reproducibility. Just saying ``Aha! we have a detection" is not sufficient. In some sense it should be reproducible in order to convince people that infact the results are not any kind of artifact or a fake and the detected signal is not any kind of deceitfulness by the nature. The past few years have seen a slew of announcements of major discoveries in astrophysics such as detection of gravitational waves, detection of dark matter particles producing γ-rays, X-rays scattering off nuclei underground, and even evidence in the cosmic microwave background for gravitational waves caused by the rapid inflation of the early Universe. Some of these turned out to be false alarms and evidences for others are also inconspicuous (Not so for the Gravitational Wave detection atleast. The first detection in September 2016 was gold and the third one in January 2017 was platinum). There is nothing more deceptive than an obvious fact. Reproducibility means that the original data can be analyzed (by an independent investigator) using some technique to obtain the same results of the original study. Since the procedure followed to reproduce the scientific results is based on some full proof scientific theory, reproducibility also ensures that we understand the underlying science behind the phenomenon. In astrophysics, these scientific procedures are usually computer simulations which are used to revive the present day cosmological scenario and the structures within to check the validity of physical parameters in play. Many people seem to conflate the ideas of reproducible and correctness, but they are not the same thing. A study could be reproducible and still be wrong (the conclusion or claim can be wrong). If my claim has some value, then the correctness of the claim will be determined by whether others come to similar conclusions or not after they have attempted to reproduce my results. Reproducibility is important because it ensures that the results are correct, ensures transparency giving us confidence in understanding exactly what was done and infact it is the only thing that an investigator can guarantee about a study.
Karl Popper once said that “Non-reproducible single occurrences are of no significance to science”. Science today is progressing on the principles of sample collection and statistical analysis. We need more and more useful data in order to help Science advance in the right direction. These unbiased computational detection techniques are useful as not only they help in detecting useful low strength signals, which otherwise would be impossible to detect, but also prohibit us from confusing an otherwise noise fluctuation, some kind of observational bias or any other form of artifact in the data with a positive detection. Therefore, a detection of an astrophysical object is done at 2 levels. At first level, data is collected. This is what we call sky observation. At second stage, detection algorithms and tools are used to identify objects and phenomenathat might exist in this observed data.
Today, where on one hand we have a scenario where big machines such as LHC are being upgraded in order to increase sensitivity of the detector, data from space missions such as Gaia/ESA (launched in November 2013) is being collected and improved and NASA JWST space telescope is to be launched in October 2018 to explore cosmos with unprecedented precision, on the other hand scientists are already applying detection algorithms on the existing data and are cooking new techniques and numerical recipes to work with the new data sets in order to make the most of it. The curiosity to understand the overall physiology of the Universe is what drives and demands the discovery of more interesting features in galactic and extragalactic sky, which seems impossible without the employment of the detection algorithms. The more of unknown in the Universe can only become known when both the machines and the algorithms are wisely planned, built and executed.