We hope you are all staying safe. With massive vaccination programs across the globe we hope you and your loved ones are getting back to what used to be normal. With that out of the way, let us circle back to Property Testing. This month was less sleepy as compared to the two preceding months and we saw six papers in total (two of them explore problems in quantum property testing). Without further ado, let us take a deeper dive.

GSF-locality is not sufficient for proximity-oblivious testing, by Isolde Adler, Noleen Kohler, Pan Peng (arXiv) The notion of proximity oblivious testers was made explicit in the seminal work of Goldreich and Ron in 2009 [GR09]. A proximity oblivious tester for a graph property is a constant query tester that rejects a graph with probability that monotonically increases with distance to the property. (Edit: Correction) A property is called proximity oblivious testable (or PO testable) if it has a one sided proximity oblivious tester. [GR09] gave a characterization of which properties \(\Pi\) are PO testable in the bounded degree model if and only if it is a “local” property of some kind which satisfies a certain non propagation condition. [GR09] conjectured that all such “local” properties satisfy this non propagation condition. This paper refutes the above conjecture from [GR09].

Coming up next. More action on triangle freeness.

Testing Triangle Freeness in the General Model in Graphs with Arboricity \(O(\sqrt n)\), by Reut Levi (arXiv) PTReview readers are likely to be aware that triangle freeness has been a rich source of problems for developing new sublinear time algorithms. This paper considers the classic problem of testing triangle freeness in general graphs. In the dense case, algorithms with running time depending only on \(\varepsilon\) are known thanks to the work of Alon, Fischer, Krivelevich and Szegedy. In the bounded degree case, Goldreich and Ron gave testers with query complexity \(O(1/\varepsilon)\). This paper explores the problem in general graph case and proves an upper bound of \(O(\Gamma/d_{avg} + \Gamma)\) where \(\Gamma\) is the arboricity of the graph. The author also shows that this upperbound is tight for graphs with arboricity at most \(O(\sqrt n)\). Curiously enough, the algorithm does not take arboricity of the graph as an input and yet \(\Gamma\) (the arboricity) shows up in the upper and lower bounds.

Testing Dynamic Environments: Back to Basics, by Yonatan Nakar and Dana Ron (arXiv) Goldreich and Ron introduced the problem of testing “dynamic environments” in 2014. Here is the setup for this problem. You are given an environment that evolves according to a local rule. Your goal is to query some of the states in the system at some point of time and determine if the system is evolving according to some fixed rule or is far from it. In this paper, the authors consider environments defined by elementary cellular automata which evolve according to threshold rules as one of the first steps towards understanding what makes a dynamic environment tested efficiently. The main result proves the following: if your local rules satisfy some conditions, you can use a meta algorithm with query complexity \(poly(1/\varepsilon)\) which is non adaptive and has one sided error. And all the threshold rules indeed satisfy these conditions which means they can be tested efficiently.

Identity testing under label mismatch, by Clement Canonne and Karl Wimmer (arXiv) This paper considers a classic problem distribution testing with the following twist. Let \(q\) denote a distribution supported on \([n]\). You are given access to samples from another distribution \(p\) where \(p = q \circ \pi\) where \(\pi\) is some unknown permutation. Thus, I relabel the data and I give you access to samples from the relabeled dataset. Under this promise, note that identity testing becomes a trivial problem if \(q\) is known to be uniform over \([n]\). The authors develop algorithms for testing and tolerant testing of distributions under this additional promise of \(p\) being a permutation of some known distribution \(q\). The main result shows as exponential gap between the sample complexity of testing and tolerant testing under this promise. In particular, identity testing under the promise of permutation has sample complexity \(\Theta(\log^2 n)\) whereas tolerant identity testing under this promise has sample complexity \(\Theta(n^{1-o(1)})\).

Testing symmetry on quantum computers, by Margarite L. LaBorde and Mark M. Wilde (arXiv) This paper develops algorithms which test symmetries of a quantum states and changes generated by quantum circuits. These tests additionally also quantify how symmetric these states (or channels) are. For testing what are called “Bose states” the paper presents efficient algorithms. The tests for other kinds of symmetry presented in the paper rely on some aid from a quantum prover.

Quantum proofs of proximity, by Marcel Dall’Agnol, Tom Gur, Subhayan Roy Moulik, Justin Thaler (ECCC) The sublinear time (quantum) computation model has been gathering momentum steadily over the past several years. This paper seeks to understand the power of \({\sf QMA}\) proofs of proximity for property testing (recall \({\sf QMA}\) is the quantum analogue of \({\sf NP}\)). On the algorithmic front, the paper develops sufficient conditions for properties to admit efficient \({\sf QMA}\) proofs of proximity. On the complexity front, the paper demonstrates a property which admits an efficient \({\sf QMA}\) proof but does not admit a \({\sf MA}\) or an interactive proof of proximity.