Computational combinatorics
Say, you have written a paper. You want to submit it to a journal. But in what field? More often than not, the precise field/area designation for this paper is easy to determine, or at least easy to place it into some large category. Even if the paper is in between fields, this is often well regarded and understood situation, nothing wrong with that. Say, the paper is resolving a problem in field X with tools from field Y. Submit to Xjournal unless the application is routine and the crux of the innovation is in refining the tools. Then submit to Yjournal.
However, when it comes to CS, things are often less clear. This is in part because of the novelty of the subject, and in part due to the situation in CS theory, which is in constant flux and search for direction (a short Wikipedia article is as rather vague and unhelpful, even more so than these generic WP articles tend to be).
The point of this post is to introduce/describe the area of “Computational Combinatorics“. Although Google returns 20K hits for this term (including experts, courses, textbooks), the meaning is either obscure or misleading. We want to clarify what we mean, critique everyone else, and make a stake for the term!
1) What I want computational combinatorics to mean is “theoretical CS aspects of combinatorics” (and to a lesser extend “practical..”), which is essentially part of combinatorics but the tools and statements use compute science terminology (for a concise description of complexity aspects, see dated but excellent survey by David Shmoys and Eva Tardos). I will give a recent example below, but basically if you want to prove a negative result in combinatorics (as in “one should not expect a nice formula for the number of 3colorings or perfect matchings of a general graph”), then CS language (and basic tools) is a way to go. When people use “computational combinatorics” to mean “basic results in combinatorics that are useful for further studies of computer science”, they are being misleading. A proper name for such course is “Introduction to Combinatorics” or “Combinatorics for Computer Scientists”, etc.
2) In two recent papers, Jed Yang and I proved several complexity results on tilings. To explain them, let me start with the following beautiful result by Éric Rémila, built on earlier papers by Thurston, Conway & Lagarias, and Kenyon & Kenyon:
Tileability of a simply connected region in the plane with two types of rectangles can be decided in polynomial time.
First, we show that when the number of rectangles is sufficiently large (originally about 10^{6}, later somewhat decreased), one should not expect such a result. Formally, we prove that tileability is NPhard in this case. We then show that in 3dim the topology of the region gives no advantage. Among other results, we prove that tileability of contractible regions with 2x2x1 slabs is
NPcomplete, and counting 2x1x1 domino tilings of contractible regions is #Pcomplete.
Now, the CS Theory point of view on these types of results changed drastically over time. Roughly, 30 years ago they were mainstream. About 20 years ago they were still of interest, but no longer important. Nowadays they are marginal at best – the field has moved on. My point is that the result are of interest in Combinatorics and Combinatorics only. Indeed, it has long been observed that applying combinatorial group theory to tilings (as done by Thurston, Rémila, etc.) is more of an art than a science. Although we believe that already for three general rectangles in the plane the problem is intractable, proving such a result is exceedingly difficult. Our various results solve weak versions of this problem.
3) The ontology (classification) in mathematics has always been a mess (this is not unusual). For example, combinatorial enumeration is the same as enumerative combinatorics. On the other hand, as far as I can tell, analytic geometry has nothing to do with geometric analysis. There is also no “monotonicity” to speak about: even though group theory is a part of algebra, the geometric group theory is neither a part of geometric algebra, nor of algebraic geometry, although traditionally contains combinatorial group theory. Distressingly, there are two completely different (competing) notions of “algebraic combinatorics” (see here and there), and algebraic graph theory which is remarkably connected to both of these. The list goes on.
4) So, why name a field at all, given the mess we have? That’s mostly because we really want to incorporate the CS aspects of combinatorics as a legitimate branch of mathematics. Theory CS is already over the top combinatorial (check out the number of people who believe that P=?NP will be resolved with combinatorics), but when a problem arises in combinatorics from within, this part of combinatorics needs a name to call home. I propose using the term computational combinatorics, in line with computational group theory, computational geometry, computational topology, etc., as a part of the loosely defined computational mathematics. I feel that the adjective “computational” is broad and flexible enough to incorporate both theoretical/complexity aspects as well as some experimental work, and combinatorial software development (as in WZ theory), compared to other adjectives, such as “algorithmic”, “computable”, “effective”, “computersciency”, etc. So, please, AMS, next time you revise your MSC, consider adding “Computational Combinatorics” as 05Fxx.
P.S. A well known petition asks for graph theory to have its own MSC code (specifically, 07), due to the heavy imbalance in the number of graph theory vs. the rest of combinatorics papers. Without venturing an opinion, let me mention that perhaps, adding a top level “computational combinatorics” subfield of combinatorics will remedy this as well – surely some papers will migrate there from graph theory. Just a thought…

March 14, 2018 at 12:26 amICM Paper  Igor Pak's blog