[Paper Review] The Z-cubes: a hypercube variant with small diameter

Xuding Zhu|arXiv (Cornell University)|Sep 23, 2015

Interconnection Networks and Systems7 references3 citations

TL;DR

This paper introduces Z-cubes, a novel hypercube variant constructed recursively by connecting two copies of an (n−1)-dimensional Z-cube with a special perfect matching based on a bitwise permutation function. The key contribution is proving that the n-dimensional Z-cube has diameter (1+o(1))n/log₂n, achieving near-optimal diameter for n-regular graphs on 2ⁿ vertices, significantly improving upon prior hypercube variants.

ABSTRACT

This paper introduces a new variant of hypercubes, which we call Z-cubes. The n-dimensional Z-cube $H_n$ is obtained from two copies of the (n-1)-dimensional Z-cube $H_{n-1}$ by adding a special perfect matching between the vertices of these two copies of $H_{n-1}$. We prove that the n-dimensional Z-cubes $H_n$ has diameter $(1+o(1))n/\log_2 n$. This greatly improves on the previous known variants of hypercube of dimension n, whose diameters are all larger than n/3. Moreover, any hypercube variant of dimension $n$ is an n-regular graph on $2^n$ vertices, and hence has diameter greater than $n/\log_2 n$. So the Z-cubes are optimal with respect to diameters, up to an error of order $o(n/\log_2n)$. Another type of Z-cubes $Z_{n,k}$ which have similar structure and properties as $H_n$ are also discussed in the last section.

Motivation & Objective

• To design a new hypercube variant with smaller diameter than existing variants like twisted cubes and spined cubes.
• To address the fundamental limit of hypercube diameter in n-regular graphs on 2ⁿ vertices, which is known to exceed n/log₂n.
• To construct a graph family that achieves diameter close to this theoretical lower bound.
• To explore structural and algorithmic properties of the new Z-cube family, including Hamiltonicity and routing efficiency.

Proposed method

• Define the n-dimensional Z-cube Hₙ recursively from two copies of Hₙ₋₁, using a specific permutation φ that modifies the first κ(n) bits via XOR with the last κ(n) bits.
• Use the permutation φ(x)[1,κ(n)] = x[1,κ(n)] ⊕ x[n−κ(n)+1,n] and φ(x)[κ(n)+1,n] = x[κ(n)+1,n] to determine the matching edges between the two copies.
• Prove that Hₙ is Hamiltonian connected for n ≥ 3, establishing strong connectivity and path existence properties.
• Introduce a simplified variant Zₙ,ₖ using a fixed k, where the permutation φₖ is applied only to the first k bits and last k bits, enabling easier analysis.
• Establish an upper bound on the diameter of Zₙ,ₖ as n/(k+1) + 2ᵏ, which is minimized when k ≈ log₂n − 2log₂log₂n.
• Show that choosing k = ⌈log₂n − 2log₂log₂n⌉ yields a diameter of (1+o(1))n/log₂n, matching the asymptotic bound of Hₙ.

Experimental results

Research questions

• RQ1 What is the diameter of the n-dimensional Z-cube Hₙ, and can it be bounded more tightly than previous hypercube variants?
• RQ2 Is there a polynomial-time algorithm to compute a shortest path between any two vertices in Hₙ?
• RQ3 What is the average distance between two vertices in Hₙ, and how does it compare to standard hypercubes?
• RQ4 What is the minimum number of vertices or edges whose removal makes Hₙ non-Hamiltonian, indicating fault tolerance?
• RQ5 What is the automorphism group of Hₙ, and is Hₙ vertex-transitive?

Key findings

• The n-dimensional Z-cube Hₙ has diameter (1+o(1))n/log₂n, which is asymptotically optimal for n-regular graphs on 2ⁿ vertices.
• The diameter bound is achieved through a recursive construction using a bitwise permutation function that combines the first and last κ(n) bits of each vertex.
• A simplified variant Zₙ,ₖ is introduced with diameter bounded by n/(k+1) + 2ᵏ, which achieves the same asymptotic diameter when k is chosen as ⌈log₂n − 2log₂log₂n⌉.
• The construction of Zₙ,ₖ is simpler than Hₙ, and it allows for a more modular design, though it requires reinitialization when increasing dimension.
• Hₙ is Hamiltonian connected for all n ≥ 3, indicating strong connectivity and path richness.
• The diameter of Hₙ is strictly smaller than that of the standard hypercube Qₙ and all known hypercube variants, including the spined cube with diameter ~n/3.