## October 31, 2010

### Books to Read While the Algae Grow in Your Fur, October 2010 (Supp'd full with horrors edition)

Algernon Blackwood, In the Realm of Terror: Eight Haunting Tales
Mind candy. At his best when suggesting ways our world is about to open out on to some vast and mysterious and far from benign realm; I see why Lovecraft admired him, and his sentences are better than HPL's, less purple and bloated. But there's something very stuffy and theosophical about Blackwood which I find off-putting.
(And can anyone, in this day and age, take seriously the premise of "A Psychical Invasion", which is that eating a hash brownie sets you up for demonic possession? In fairness, this story does have some fine observation of feline personality.)
A splendidly-written high-fantasy western. (It is by no stretch of the imagination "steampunk".) Gilman takes great themes of what one might call the Matter of America — the encroachment of regimented industrial civilization, the hard-eye anarchic men (and women) of violence, the dream of not just starting the world afresh but of offering the last best hope of earth — and transforms the first two into warring rival pantheons of demons, the third into a noble lost cause. (I think Gilman knows exactly how explosive the last theme is, which is why he manages to handle it without setting it off.) Beneath and behind it all lies the continuing presence of the dispossessed original inhabitants of the continent. A story of great excitement and moment unfolds in this very convincing world, tying together an appealing, if believably flawed, heroine and two finely-rendered anti-heroes, told in prose that is vivid and hypnotic by turns. The story is complete in itself, but leaves open a return to the world, which I really hope will happen soon. The most natural point of comparison is Stephen King's The Dark Tower, especially The Gunslinger, which I love; this is more ambitious in its themes, sounder in its construction, and more satisfying in its execution.
The Half-Made World is the finest rendition I've ever seen of one of our core national myths; go read it.
Manual trackback: Felix Gilman; Crooked Timber
Sequel.
H. P. Lovecraft, Tales (ed. Peter Straub)
Lovecraft — Lovecraft! — gets a Library of America edition: a solid (850 pp.) and handsome volume, with a good selection (including At the Mountains of Madness), and decent end-notes. The stars must indeed be right...
Cherie Priest, Four and Twenty Blackbirds
Mind candy. Ghosts, twisted families, maniacal cultists, and a nice atmosphere of Southern twistedness. A little slow in places but it was a first novel, and I will definitely look for others.
Lauren Willig, Betrayal of the Blood Lily
Mind candy — this time, jalebi.
George G. Roussas, Contiguity of Probability Measures: Some Applications in Statistics
Two sequences of probability measures, say $$P_n$$ and $$Q_n$$ are "contiguous" when, any sequence of measurable sets $$A_n$$ , the probabilities $$P_n(A_n) \rightarrow 0$$ if and only if $$Q_n(A_n) \rightarrow 0$$. This is just slightly weaker than asymptotic mutual absolute continuity --- more exactly, the original sequences of measures can be approximated to arbitrary precision by a new pair of sequences which are mutually absolutely continuous (Theorem 1.5.1). Mutual absolute continuity lets us define Radon-Nikodym derivatives, which is to say, likelihood ratios, and these, as it turns out, can always be put into an exponential form. Assuming that the measures are parametric families of Markov processes, and that contiguity holds, Roussas constructs a clever theory of local approximation by exponential families (whether the original model has that form or not). From these approximation results he then derives locally optimal asymptotic procedures for estimating parameters, testing hypotheses about them, and constructing confidence intervals. Using these methods, results for inference on Markov processes are no harder than the special case of IID data.
A sure grasp of measure-theoretic probability, and of the development of ordinary statistical theory on that basis, are essential pre-requisites. (Someone who could read Schervish, or van der Vaart, or even Cramér, should be fine.) No particular knowledge of Markov processes is required.
Dexter, Season 4
Great as always, and illustrating this point. ROT-13'd spoiler for the very last few minutes: ABBBBBBBBBBB!
Rondo Keele, Ockham Explained
Biographical introduction to the thought of William of Ockham (or Occam, if you prefer the Latin rendition). Does a good job on explaining the background (Aristotle and medieval Catholicism), and clarifying what sort of parsimony principle Ockham advocated. (Cf.) While Keele is not always the sharpest tool in the philosophical shed*, it's readable and short, which is a major accomplishment in itself when dealing with the Scholastics.
*: For instance, he offers an extended critique (pp. 122--128) of Ockham's account of motion, which Keele glosses as follows (p. 118): "Ockham analyzes a sentence like 'X is in motion' as 'X is in a place at time t, and at $$t^{\prime} \neq t$$, X is (continuously, without rest) in another place". (He never quotes a definition of motion from Ockham, that I can see. As stated, this rules out periodic motion, but let that pass.) On this basis, Ockham applies the razor and argues that, while things move, there is no need to posit that they do so by acquiring the "accidental form" of motion. Keele objects that, per Ockham, we cannot truly say that any body is moving at any instant of time, but it's impossible to occupy multiple places at multiple times in a single instant, let alone do so continuously. (Keele even says that Ockham's account conflicts with the differential calculus, and the possibility of calculating an instantaneous velocity by taking the time-derivative of position!) On this basis, Keele suggests that we go back to thinking of motion as an accident inhering in bodies.