🤖 AI Summary
This paper investigates robust consistent rounding—a high-dimensional continuous-to-discrete mapping from ℝᵈ to k discrete points—under noise resilience constraints: same-color regions must be separated by distance at least t. The central goal is to characterize the precise asymptotic trade-off between k and the tolerable separation t, and to determine the minimal side information required for positive tolerance. Methodologically, it models the problem as a k-color t-separated tiling of ℝᵈ, integrating tools from convex geometry (Brunn–Minkowski inequality, isoperimetric inequalities), algebraic topology (Čech cohomology), and coding theory. Key contributions include: (i) introducing the notion of “consistent rounding” and establishing an information–robustness theoretical framework; (ii) proving that log₂(d+1) bits of side information are necessary and sufficient for positive tolerance; (iii) deriving tight asymptotics—t = (0.561 + o(1))k^{1/3} for d = 3—and asymptotically tight bounds for d = 4, 8, 24 via sphere packings and isoperimetric analysis. This work provides the first information-theoretically optimal characterization of robust consistent rounding in high dimensions.
📝 Abstract
In standard rounding, we want to map each value $X$ in a large continuous space (e.g., $R$) to a nearby point $P$ from a discrete subset (e.g., $Z$). This process seems to be inherently discontinuous in the sense that two consecutive noisy measurements $X_1$ and $X_2$ of the same value may be extremely close to each other and yet they can be rounded to different points $P_1
e P_2$, which is undesirable in many applications. In this paper we show how to make the rounding process perfectly continuous in the sense that it maps any pair of sufficiently close measurements to the same point. We call such a process consistent rounding, and make it possible by allowing a small amount of information about the first measurement $X_1$ to be unidirectionally communicated to and used by the rounding process of $X_2$.
The fault tolerance of a consistent rounding scheme is defined by the maximum distance between pairs of measurements which guarantees that they are always rounded to the same point, and our goal is to study the possible tradeoffs between the amount of information provided and the achievable fault tolerance for various types of spaces. When the measurements $X_i$ are arbitrary vectors in $R^d$, we show that communicating $log_2(d+1)$ bits of information is both sufficient and necessary (in the worst case) in order to achieve consistent rounding for some positive fault tolerance, and when d=3 we obtain a tight upper and lower asymptotic bound of $(0.561+o(1))k^{1/3}$ on the achievable fault tolerance when we reveal $log_2(k)$ bits of information about how $X_1$ was rounded. We analyze the problem by considering the possible colored tilings of the space with $k$ available colors, and obtain our upper and lower bounds with a variety of mathematical techniques including isoperimetric inequalities, the Brunn-Minkowski theorem, sphere packing bounds, and Cech cohomology.
