Foundations of Quasirandomness Joshua N. Cooper UCSD / South Carolina Quasirandomness Regularity Chung, Graham, Wilson ‘89 Graphs Szemerédi ’76ish Simonovits, Sós ‘91 Graphs Chung, Graham ’90-92 Gowers, Tao – ‘05 Tournaments, Subsets of Zn,… Nagle/Rödl/Skokan/Schacht – ’05 Frankl/Rödl – ‘92, ‘01 Chung ‘91 Chung, Graham ’90-92 Hypergraphs Chung ‘91 Hypergraphs Kohayakawa, Rödl, Skokan ’02 Hypergraphs (p≠.5) JC ’03 Permutations JC ’05 JC ‘05 Permutations The rough idea of quasirandomness: A universe: the class of combinatorial objects OBJ A property: P(o), true a.s. for large objects o OBJ A sequence: o1, o2, o3, … Define {oi} to be quasirandom if P(oi) “asymptotically”. A (weak) example: OBJ is the class of graphs, P(G) is the property 1 o(1) | V (G) | | E (G) | 2 2 where o(1) 0 as | V (G ) | . For each random-like property P, one can define P-quasirandomness. Some types of quasirandomness imply other ones: P1 Q1 P3 Q2 P2 P4 Q3 By transitivity, the property cliques form a poset: P1 Q1 P3 Q2 P2 P4 Q3 The quasirandom property cliques studied historically have been surprisingly large, i.e., include a large number of very different random-like properties. Furthermore, many of the cliques look similar, even in different universes OBJ. So what exactly is quasirandomness? An information theoretic idea: Suppose that we have a space X, and a subset of k points of X… ? … then, X is quasirandom if learning whether or not the points are “related” tells us almost nothing about “where” the points are. An information theoretic idea: Suppose that we have a space X, and a subset of k points of X… … then, X is quasirandom if learning whether or not the points are “related” tells us almost nothing about “where” the points are. “Related”: A relation “Where”: A family R⊂Xk L of subsets L⊂Xk “local sets” Let x be a uniformly random choice of an element from Xk, and write 1R for the indicator of the event that x in R. Then R is quasirandom with respect to L whenever, for all L L, PL H1R | L H1R o1 PL H1R | L H1R o1 Intuition: The statement only has force when Intuition: Learning that x P(L) is “not too small”, i.e., (1). L (“where x is”) tells you almost nothing new about the event R(x). Theorem 1. Suppose that min(P(R),1-P(R)) random with respect to = (1). L iff PR L P( R)P( L) o1 for all L L. Then R is quasi- Corollary 2. Write |X|=n. Suppose that min(|R|,nk-|R|) = (nk). Then R is quasirandom with respect to L iff R L for all L L. RL n k o nk … which is why we recover quasirandomness in all its guises when we set: Object Type Local Sets Relations Graphs / Tournaments S×T, for subsets S,T⊂V(G) binary symmetric / antisymmetric Subsets of Zn arithmetic progressions (or intervals for “weak” quasirandomness) unary Permutations k-uniform hypergraphs Sets π(I) ∩ J for intervals I, J “closed” k-uniform hypergraphs binary (inversions) totally symmetric k-ary Definition. A k-uniform hypergraph is called “closed” when it is equal to its image under the closure operator u ° d, where d(H ) = the set of all (k-1)-edges contained in edges of H u(H ) = the set of all k-edges spanned by a Kk(k-1) in H H d(H ) u ° d(H ) We wish to reproduce and generalize the theorems appearing in different versions of quasirandomness. For example: Definition. For a set Y ⊂ Xk, we write π(Y) for the projection of Y onto the coordinates {2,…,k}. L of local sets is called robust if, whenever Y ⊂ X : Y → L is any mapping, L includes the set Definition. The family and y y . yY X X X y1, (y1) y2, (y2) y3, (y3) y y 3 j 1 j j L and L is robust, X, π(R∩({x}×Xk)) is quasirandom with respect to π(L). Theorem 3. Let k > 1. If R is quasirandom with respect to then, for almost all x All of the local set systems with k > 1 previously mentioned are robust. (And so is the set of all Cartesion products.) Translation into two sample contexts: Corollary 4. If a tournament T is quasirandom, then almost all out-degrees are n/2 + o(n). Corollary 5. If a hypergraph quasirandom. H is quasirandom, then almost all vertex links are Current questions (some of which are partially solved): (1) What are the conditions on R sufficient to prove the converse of the theorem on the previous slide? (2) What about substructure counts, i.e., “patterns”? (3) What role does a group structure on X play? (4) Is there a spectral aspect of quasirandomness that goes beyond what is already known? Is it possible to make sense of this question for k > 2? (5) Describe the structure of the poset of property cliques induced by the possible families of local sets. Thank you!