So in the last post, we defined the discrete Fourier transform and gave some of its basic properties. At the end, we claimed that it gives us a simple notion of pseudorandomness that allows us to make rigorous the intuition that “pseudorandom subsets of should have many arithmetic progressions.” Today we’re going to justify this notion of pseudorandomness, and work through this — the easy case — of the proof of Roth’s theorem for . At the end, we’ll sketch how to modify the method for the regular, finitary version of the theorem.

## Posts Tagged ‘GILA’

### GILA 4: The pseudorandom case

September 19, 2009### GILA 3: The Fourier transform, and why it works

September 3, 2009At the end of the last post, we outlined an approach to proving Roth’s theorem via a density-increment method. I’ll re-post it here, a little more fleshed out, for convenience.

Sketch.The idea is to deal with “structured sets” (like sets that are themselves long arithmetic progressions) and “pseudo-random sets” in different ways; we saw last time that the density-increment argument looks promising for structured sets, but “random” sets won’t succumb to it.First, we need to make rigorous our intuition that “random enough” sets of positive density will have length-3 APs. The obvious way to do this is by showing the contrapositive. We want to find some suitable “randomness measure” such that random dense sets have high randomness measure. Now let be a positive-density set without 3-APs; we want to show that this condition is enough to guarantee that has low randomness measure.

Now for our modified-probabilistic argument. We want to show that “low-randomness” sets can’t be uniformly distributed (mod d) for every d; this guarantees the existence of an AP on which it has higher density. We pass to this AP; rinse, lather, and repeat, until our density rises above some pre-set bound ( will do nicely.) We also want to introduce a tool that tells us how much a set “looks like” an arithmetic progression with common difference d; this is because we need to obtain quantitative bounds for the density-increment argument, in order to ensure that the densities of our set on the sequence of APs don’t monotonically approach, say, 0.0000001.

So we’re looking for two technical tools; one that quantifies the “randomness” of a set, and one that lets us correlate our set with a candidate arithmetic progression to pass to so that our density-increment argument will actually work quantitatively. As it turns out, though, we can find one tool that will do both jobs for us; the discrete Fourier transform.

### GILA 1: van der Waerden and all that

August 22, 2009I’m beginning my GILA series announced on Thursday with a short exposition of van der Waerden’s theorem, and an attempt to extend its proof to the density version (at least in the k=3 case). Before I start, I should mention something I neglected in the last post: that this series is as much for my benefit as for that of any “interested lay audience” out there. Someone like Terry Tao could undoubtedly do a better series of this type (and, indeed, I’ll be leaning heavily on my borrowed copy of Tao and Van Vu’s *Additive Combinatorics* throughout) but they haven’t, so I’ll try. However, there are certain to be a number of embarrassing mistakes, unjustified statements, lines of thinking left unresolved, etc., throughout (indeed, an anonymous commenter noted two in the introductory post!), and I encourage readers to point them out in the comments.

So, on to van der Waerden’s theorem. van der Waerden’s theorem is one of the early jewels of Ramsey theory (actually predating Ramsey’s article by a couple of years) and can be viewed as a weaker version of Szemeredi’s theorem. It states simply that, for any integers , if you color the positive integers with c colors, then there exists a monochromatic k-term arithmetic progression.

Khinchin, in his wonderful book which unfortunately appears to be out of print (at least in English), gives some history of the problem:

All to whom this question [the theorem in the c = 2 case] was put regarded the problem at first sight as quite simple; its solution in the affirmative appeared to be almost self-evident. The first attempts to solve it, however, led to nought… [T]his problem, provoking in its resistance, soon became the object of general mathematical interest… I made [van der Waerden’s] acquaintance, and learned the solution from him personally. It was elementary, but not simple by any means. The problem turned out to be deep; the appearance of simplicity was deceptive.

This last clause is particularly wonderful, describing as it does not only van der Waerden’s theorem but Ramsey theory and perhaps even combinatorics in general.

The usual proof (given in story-form by Zeilberger here) is not van der Waerden’s, however, but comes from Khinchin, attributed by him to a M.A. Lukomskaya in Minsk. It is a masterpiece, using nothing but a very clever inductive argument and some application of the pigeonhole principle. For the sake of self-containedness, I’ll sketch the argument below the fold.

### GILA 0: Roth’s proof of Roth’s theorem

August 20, 2009I’m returning from my unplanned, unannounced hiatus from blogging with a series on the standard (Fourier-theoretic) proof of Roth’s theorem on arithmetic progressions, otherwise known as the k = 3 case of Szemeredi’s theorem. I’d like to start out by noting that there’s no shortage of good expositions on the topic; a quick Google search will turn up several. This series will differ from these expositions, however, in two key ways.

First, I’ll try to highlight the key ideas of the proof, rather than get bogged down in the calculations. Although, by the end of the series, we should have at least a sketch of a complete proof of Roth’s theorem, I want to make it easy to separate the real insights from the straightfoward calculational machinery. So I’ll likely present a high-level overview of the proof in the main body of these posts, with one or several appendices fleshing out the computations needed to ensure that the overview works.

Second, and more importantly, this is to be a piece of historical fiction. Although I don’t know how Roth came upon his original proof, he was a number theorist and an expert in Diophantine approximation, and was therefore familiar with things like Fourier analysis. I want to approach the problem from the perspective of a combinatorialist roughly contemporaneous with Roth, who’d be familiar with standard combinatorial methods but not with some of the more technical tools, and attempt to re-derive the proof from that viewpoint.

The real meat of the series will start this weekend (which is why this post is numbered 0), but since this is targeted toward a Generally Interested Lay Audience, I’d like to introduce a few topics which a layman might not be familiar with below the cut.