I applaud the textbook for attempting to break down the historical wavelength divisions by treating analogous processes in parallel. Still, even this new edition devotes too much space for my taste to what are now pretty outdated techniques in some cases.

Taking into account the Stokes parameters of radiation, the quantities which we can in principle measure are as follows: with some resolution and sampling,

We may broadly classify kinds of observations depending on which of these quantities are of primary interest:

Mixed modes also exist - imaging polarimetry, area spectroscopy, multiobject photometry and spectrophotometry, time-phased spectroscopy. The ultimate detector would give us the location (i.e. direction of arrival), time of arrival, frequency, and polarization direction of each incoming photon. The best we can currently do is x,y,t and with a rather large error n (in some X-ray detectors) and x,y,t for optical photon counters (though there is some promise of improvements for small-field optical and near-IR detectors of this kind).

Major differences among wavelength regimes lie in the technologies needed to perform these operations with the high efficiencies needed for astronomical sources. It is sometimes more natural to detect and measure as photons, other times (especially in the far-infrared and radio) to use wave applications. Mixing these in the same process can be tricky!

Another issue to be stressed is one of philosophy, well enunciated by Feynman: "I am not trying to tell you what to do about cheating on your wife, or fooling your girlfriend, or something like that, when you're not trying to be a scientist, but just trying to be an ordinary human being. We'll leave those problems up to you and your rabbi. I'm talking about a specific, extra type of integrity that is not lying, but bending over backwards to show how you're maybe wrong, that you ought to have when acting as a scientist" (from "Cargo Cult Science" in Surely You're Joking, Mr. Feynman). He goes on to say the ground rules are (1) don't fool yourself, (2) don't fool other scientists, (3) don't fool lay people. Despite the frequency with which they are violated, these are important maxims for any kind of research - it is interesting that much of our accurate knowledge of the world comes from recognizing the role of error!

"But however perfect an instrument may be (and it is the astronomer's business to see that it is perfect), it is the astronomer's further business to look upon it with complete and utter mistrust." (from a letter by David Gill to J.C. Kapteyn, 18 Jan. 1885). I often take this to mean that we should be suspicious of a conclusion particularly to the extent that it accords with our prior notions. Someone phrased it as "seek and ye shall find - but beware if you have to look very hard for what you thought was there all along". If at all possible, one should adopt objective and quantifiable standards of measurement simply to avoid even the hint of personal bias (subconscious though its effects may be). Good examples of this include the Martian canals and van Maanen's announcement of angular rotation of spiral galaxies.

Note that what you need to learn to answer a specific question depends critically on how precisely you pose the question; in general, the greater one's initial ignorance, the more and better data are needed to deal with the problem. By contrast, if we already understand the physical situation, a single well-defined measurement often suffices to tell us something important. For example, to measure a galaxy redshift, if we know nothing about the galaxy spectrum, data of high signal/noise are needed. However, once we know that the vast majority of galaxies have rather similar spectra, we can use cross-correlation against some standard galaxy spectrum (in log frequency space actually) and get a reliable redshift from a spectrum in which few if any individual features are detected. (Gunn's Law: if you can see it, you've got it.) Beware, however, of overrating your physical understanding of a system - we'll come back to this point in discussing Bayesian statistics.

Errors are of two kinds: errors and blunders. Statistics give you errors. You do your own blunders, and no amount of numerical massage can eradicate them. They include identifying the wrong star at the telescope, miscalibrating the instrument, having a dead snake shorting one of your signal lines, seeing a phosphorus spectral feature because you struck a match on the telescope tube, trying to guide the telescope on an asteroid instead of a star, etc. (I didn't make these up - there are documented instances of each).

Last changes: 1/2006	  © 2000-2006