Abstract:We revisit the relation between two fundamental property testing models for bounded-degree directed graphs: the bidirectional model in which the algorithms are allowed to query both the outgoing edges and incoming edges of a vertex, and the unidirectional model in which only queries to the outgoing edges are allowed. Czumaj, Peng and Sohler [STOC 2016] showed that for directed graphs with both maximum indegree and maximum outdegree upper bounded by $d$, any property that can be tested with query complexity $O_{\varepsilon,d}(1)$ in the bidirectional model can be tested with $n^{1-\Omega_{\varepsilon,d}(1)}$ queries in the unidirectional model. In particular, if the proximity parameter $\varepsilon$ approaches $0$, then the query complexity of the transformed tester in the unidirectional model approaches $n$. It was left open if this transformation can be further improved or there exists any property that exhibits such an extreme separation.
We prove that testing subgraph-freeness in which the subgraph contains $k$ source components, requires $\Omega(n^{1-\frac{1}{k}})$ queries in the unidirectional model. This directly gives the first explicit properties that exhibit an $O_{\varepsilon,d}(1)$ vs $\Omega(n^{1-f(\varepsilon,d)})$ separation of the query complexities between the bidirectional model and unidirectional model, where $f(\varepsilon,d)$ is a function that approaches $0$ as $\varepsilon$ approaches $0$. Furthermore, our lower bound also resolves a conjecture by Hellweg and Sohler [ESA 2012] on the query complexity of testing $k$-star-freeness.
What problem does this paper attempt to address?
The problem that this paper attempts to solve is about the relationship between two basic models in directed graph property testing. Specifically, it is about the relationship between the query complexity of the two - way model and the one - way model in bounded - degree directed graphs. The main contribution of the paper is to prove that for the testing of subgraph - freeness containing \(k\) source components, \(\Omega(n^{1 - 1/k})\) queries are required in the one - way model. This directly gives the property that first explicitly shows the separation of query complexity between the two - way model and the one - way model. That is, it can be tested with \(O_{\varepsilon,d}(1)\) query complexity in the two - way model, while \(\Omega(n^{1 - f(\varepsilon,d)})\) query complexity is required in the one - way model, where \(f(\varepsilon,d)\) is a function that approaches 0 when \(\varepsilon\) approaches 0. In addition, their lower - bound results also solve the conjecture of Hellweg and Sohler [ESA 2012] about the query complexity of testing \(k\)-star - freeness.
### Brief Introduction to the Paper Background
1. **Directed Graph Property Testing**: Property testing is a framework for studying extremely fast (randomized) algorithms, which is used to solve the relaxed version of the classic decision problems of graphs. Given a graph property \(P\), the goal is to design a property tester that can accept any graph satisfying \(P\) with high probability and reject graphs that are "far" from satisfying \(P\), that is, a significant part of the graph representation (such as the adjacency matrix or adjacency list) needs to be modified to make it satisfy \(P\).
2. **Two - way Model and One - way Model**: Bender and Ron introduced two basic models to study directed graph property testing:
- **Two - way Model**: The algorithm is allowed to query the out - edges and in - edges of vertices.
- **One - way Model**: The algorithm is only allowed to query the out - edges of vertices, and not allowed to query the in - edges.
3. **Existing Work**: Czumaj, Peng and Sohler [STOC 2016] showed that for directed graphs whose maximum in - degree and maximum out - degree are both upper - bounded by \(d\), any property that can be tested with \(O_{\varepsilon,d}(1)\) query complexity in the two - way model requires \(n^{1-\Omega_{\varepsilon,d}(1)}\) queries in the one - way model. In particular, if the proximity parameter \(\varepsilon\) approaches 0, the query complexity in the one - way model approaches \(n\).
### Main Contributions of the Paper
1. **Query Complexity of Subgraph - freeness**: The paper proves that for the testing of subgraph - freeness containing \(k\) source components, \(\Omega(n^{1 - 1/k})\) queries are required in the one - way model.
2. **Separation of Property Testing**: This result directly gives the property that first explicitly shows the separation of query complexity between the two - way model and the one - way model. That is, it can be tested with \(O_{\varepsilon,d}(1)\) query complexity in the two - way model, while \(\Omega(n^{1 - f(\varepsilon,d)})\) query complexity is required in the one - way model, where \(f(\varepsilon,d)\) is a function that approaches 0 when \(\varepsilon\) approaches 0.
3. **Solving the Conjecture**: In addition, their lower - bound results also solve the conjecture of Hellweg and Sohler [ESA 2012] about the query complexity of testing \(k\)-star - freeness.
### Technical Details
1. **Frequency Variables**: A sequence of frequency variables \(X_A\) is defined. A number is randomly and uniformly selected from the set of different elements appearing in the sequence, and then \(X_A\) represents its frequency.
2. **Proportional Moments**: Two random variables \(X_1\) and \(X_2\) are said to have \(k - 1\) proportional moments if \(\frac{E[X_2]}{E[X_1]}=\frac{(E[X_2])^{k - 1}}{(E[X_1])^{k - 1}}\)