Homomorphism Counts: Expressive Power and Query Algorithms
summary
Lovász’s theorem and its database analogue anchor a view where left profile and right profile are vectors of homomorphism counts; Dvořák shows Ck-equivalence matches left profiles restricted to graphs of treewidth < k, and Weisfeiler–Leman/Cai–Fürer–Immerman tools appear throughout. Main claims: many relaxations (Ck, cospectrality) are captureable from the left but provably not from any restricted right profile; chromatic equivalence is captureable from the right but not the left; left Boolean query algorithms equal Boolean combinations of conjunctive queries, and counting over N gives no extra power over Booleans for classes closed under homomorphic equivalence (proof uses CSP dichotomy, Rossman, polynomial interpolation and explicit A′/B′ constructions).
AI-written summary · about
chapters (17)
- 0:00Introduction
- 0:21Homomorphism counts and Poincaré
- 3:34Homomorphism counts concept
- 7:30Two perspectives on homomorphism counts
- 12:04Graph profiles: left and right
- 16:24Lovász’s Theorem and Chaudhuri-Vardi
- 18:57Equivalence from restricted profiles
- 23:48Treewidth and logic result
- 27:05Left vs right profiles for CK
- 31:50Negative results on right profiles
- 38:22Limitations of left profiles
- 39:25Chromatic equivalence and right profiles
- 41:56Query algorithms with homomorphism counts
- 47:40Beyond natural numbers in queries
- 51:59Boolean vs natural number queries
- 58:02Expressive power differences summary
- 1:00:57Q&A
related talks
38mConsistency, Acyclicity, and Positive Semirings
19mAccessing Answers to Unions of Conjunctive Queries with Ideal Time Guarantees: II
26mCounting Subgraphs in Somewhere Dense Graphs
01h 01mDatalog and Equality Saturation
35mLogical Interfaces to Data, Beyond Views
29mDirect Access for Conjunctive Queries with Aggregation
38mThe Intensional-Extensional Problem in Probabilistic Databases
44mIntermediate Relational Algorithm Design
