On the Number of Observation Nodes in Boolean Networks

TL;DR

The paper tackles the problem of how many observation nodes are required to guarantee observability in Boolean networks, focusing on three BN families: $K$-AND-OR-BNs, $K$-XOR-BNs, and $K$-NC-BNs. It develops information-theoretic lower bounds, including $m \ge \left[(1-K)+\frac{2^{K}-1}{2^{K}}\log_{2}(2^{K}-1)\right]n$ for $K>2$ and a universal bound via counting identical states and fixed points, while also deriving nontrivial best- and worst-case upper bounds through combinatorial constructions. For 2-AND-OR-BNs specifically, a tight lower bound of $m \ge 0.188n$ and a best-case upper bound of $m = \left(\frac{2^{K}-K-1}{2^{K}-1}\right)n$ are established, with a counterexample showing no $m=1$ observability for $n>3$. The results further reveal that $K$-NC-BNs can be easier to observe than some $K$-AND-OR-BNs, as evidenced by an upper bound of $m = \lceil n/K\rceil$ and a lower bound of $(1-\beta_K)n$, highlighting meaningful differences in observability across BN classes. Overall, the paper provides a framework of bounds and constructions to guide observation design in practical BN settings.

Abstract

A Boolean network (BN) is called observable if any initial state can be uniquely determined from the output sequence. In the existing literature on observability of BNs, there is almost no research on the relationship between the number of observation nodes and the observability of BNs, which is an important and practical issue. In this paper, we mainly focus on three types of BNs with $n$ nodes (i.e., $K$-AND-OR-BNs, $K$-XOR-BNs, and $K$-NC-BNs, where $K$ is the number of input nodes for each node and NC means nested canalyzing) and study the upper and lower bounds of the number of observation nodes for these BNs. First, we develop a novel technique using information entropy to derive a general lower bound of the number of observation nodes, and conclude that the number of observation nodes cannot be smaller than $\left[(1-K)+\frac{2^{K}-1}{2^{K}}\log_{2}(2^{K}-1)\right]n$ to ensure that any $K$-AND-OR-BN is observable, and similarly, some lower bound is also obtained for $K$-NC-BNs. Then for any type of BN, we also develop two new techniques to infer the general lower bounds, using counting identical states at time 1 and counting the number of fixed points, respectively. On the other hand, we derive nontrivial upper bounds of the number of observation nodes by combinatorial analysis of several types of BNs. Specifically, we indicate that $\left(\frac{2^{K}-K-1}{2^{K}-1}\right)n,~1$, and $\lceil \frac{n}{K}\rceil$ are the best case upper bounds for $K$-AND-OR-BNs, $K$-XOR-BNs, and $K$-NC-BN, respectively.

Figures (2)

• Figure 1: Comparison between the coefficient $(1-K)+\frac{2^{K}-1}{2^{K}}\log_{2}(2^{K}-1)$ in Theorem \ref{['theorem2']} and the coefficient $\frac{2^{K}-K-1}{2^{K}-1}$ in Theorem \ref{['theorem']}. Here, we find that $\frac{2^{K}-K-1}{2^{K}-1}>(1-K)+\frac{2^{K}-1}{2^{K}}\log_{2}(2^{K}-1)$.
• Figure 2: Comparison between the coefficient $\frac{1}{K}$ in Theorem \ref{['theorem5']} with the coefficient $1-\beta_{K}$ in Proposition \ref{['proposition']}. Here, we find that $\frac{1}{K}>1-\beta_{K}$.

Theorems & Definitions (31)

• Definition 1
• Example 1
• Example 2
• Definition 2
• Definition 3
• Theorem 1
• Proposition 1
• Proposition 2
• Proposition 3
• Remark 1