On the Tail Transition of First Arrival Position Channels: From Cauchy to Exponential Decay
Extracted source text · arXiv 2511.19074
\documentclass[11pt]{article}
%\documentclass[11pt,twocolumn]{article}
%\usepackage[margin=0.75in]{geometry}
% --- Page Setup (arXiv-safe) ---
\usepackage[margin=1in]{geometry}
\usepackage{setspace}
\onehalfspacing
% --- Packages ---
\usepackage{cite}
\usepackage{amsmath,amssymb,amsfonts}
\usepackage{algorithmic}
\usepackage{graphicx}
\usepackage{textcomp}
\usepackage{xcolor}
\usepackage{url}
\usepackage{bm}
% --- TikZ ---
\usepackage{tikz}
\usetikzlibrary{arrows.meta, decorations.pathmorphing, patterns}
% --- Custom Commands ---
\newcommand{\norm}[1]{\left\lVert#1\right\rVert}
\newcommand{\E}{\mathbb{E}}
\newcommand{\R}{\mathbb{R}}
\newcommand{\todo}[1]{\textcolor{red}{[TODO: #1]}}
%%----
% --- Title and Author (arXiv format) ---
\title{\textbf{On the Tail Transition of First Arrival Position Channels: From Cauchy to Exponential Decay}}
\author{
\textbf{Yen-Chi Lee} \\
Department of Mathematics, National Central University, Taiwan \\
Email: \texttt{yclee@math.ncu.edu.tw}
}
\date{} % arXiv usually leaves date empty
%%----
\begin{document}
\maketitle
% ==============================================================================
% ABSTRACT
% ==============================================================================
\begin{abstract} %v1c
While the zero-drift First Arrival Position (FAP) channel is rigorously known to be Cauchy-distributed, practical molecular communication systems typically operate with non-zero drift. This letter characterizes the transition from heavy-tailed Cauchy behavior to light-tailed exponential decay. Through asymptotic analysis, we identify a critical spatial scale $n_c=\sigma^2/v$ separating diffusion- and drift-dominated regimes, revealing that the channel effectively behaves as a ``Truncated Cauchy'' model. Numerical results show that Gaussian approximations severely underestimate capacity at low drift, while the zero-drift case provides the appropriate performance lower bound for systems where drift assists particle transport.
\end{abstract}
%%----
\iffalse
\begin{abstract} %v1b
While the zero-drift First Arrival Position (FAP) channel is rigorously known to be Cauchy-distributed, practical molecular communication systems typically operate with non-zero drift. This letter characterizes the transition from heavy-tailed Cauchy behavior to light-tailed exponential decay. Through asymptotic analysis, we identify a critical spatial scale $n_c=\sigma^2/v$ separating diffusion- and drift-dominated regimes, revealing that the channel effectively behaves as a ``Truncated Cauchy'' model. Numerical results further show that Gaussian approximations severely underestimate capacity at low drift, while the zero-drift limit provides the physical worst-case capacity baseline that all non-zero-drift systems must exceed.
\end{abstract}
\fi
%%----
\iffalse
\begin{IEEEkeywords}
Molecular communication, first arrival position (FAP), heavy-tailed noise, Cauchy distribution, channel capacity, phase transition.
\end{IEEEkeywords}
\fi
% ==============================================================================
% I. INTRODUCTION
% ==============================================================================
\section{Introduction}
Molecular Communication (MC) via diffusion relies on the stochastic motion of information particles to convey messages \cite{Akyildiz2008}. While the majority of MC literature focuses on First Arrival Time (FAT), the spatial dimension---specifically, the First Arrival Position (FAP)---offers a compelling alternative.
In FAT systems, information is strictly coupled to time; thus, the long temporal tail inherent to diffusion causes significant errors and interference with subsequent symbols.
FAP effectively \textbf{decouples} information from this stochastic timing. Since the information is encoded in the spatial impact point, a delayed arrival does not invalidate the signal; the random arrival time becomes irrelevant noise rather than a source of error. This mechanism inherently mitigates \textbf{temporal} Inter-Symbol Interference (ISI) \cite{Lee_TCOM24, Pandey_WCL18}.
Beyond increasing degrees of freedom for single-link communication, FAP serves as the fundamental spatial impulse response for \textbf{Molecular MIMO} \cite{Koo2016} and \textbf{Index Modulation (IM)} \cite{Gursoy2019} systems.
As argued in \cite{Lee_TCOM24} and realized in microfluidic setups \cite{Kuscu2016}, practical biosensors in lab-on-a-chip devices are inherently fabricated as planar arrays rather than the spherical approximations often used in early theoretical studies.
Consequently, the statistical behavior of the FAP on this receiving plane becomes critical in governing the \textit{spatial interference} (cross-talk) between adjacent sensing elements.
Specifically, a heavy-tailed noise distribution implies long-range interference that decays slowly with distance, severely limiting the feasible density of MIMO arrays.
Characterizing the FAP noise statistics is therefore a prerequisite for robust system design. In our recent work \cite{Lee_ICC24}, we bridged the gap between diffusion physics and stable information theory by proving that the FAP channel mathematically converges to a multidimensional additive Cauchy noise channel in the zero-drift limit. This established the zero-drift FAP channel as a concrete physical realization of the abstract stable noise model studied by Fahs et al. \cite{Fahs_ISIT12} and Verd\'{u} \cite{Verdu_Entropy23}, necessitating constraints based on \textit{logarithmic moments} (or ``strength'' \cite{Verdu_Entropy23}) rather than conventional variance.
However, this ``physical Cauchy limit'' represents a singular state where the Péclet number is zero. In most microfluidic or biological flows, a non-zero drift velocity $\mathbf{v}$ introduces a restoring force that suppresses the heavy tails. This induces a \textbf{physical phase transition} in the noise statistics: from algebraic decay ($\sim n^{-2}$) to exponential decay ($\sim e^{-c|n|}$), theoretically recovering finite moments. Understanding this transition is vital to avoid design pitfalls. Assuming a Gaussian model (light tail) when the drift is weak can lead to a disastrous underestimation of outlier interference; conversely, adhering to the Cauchy model (heavy tail) when the drift is strong may result in overly conservative designs.
In this letter, we rigorously analyze the tail transition of the FAP channel. We derive a \textit{critical cutoff length} based on the interplay between the microscopic diffusion parameter $\sigma^2$ and the macroscopic drift velocity magnitude $v = \|\mathbf{v}\|$. We show that the FAP channel operates in a ``Truncated Cauchy'' regime. Furthermore, numerical capacity analysis reveals that in low-drift environments, standard Gaussian approximations fail significantly, whereas the zero-drift Cauchy model provides a robust physical lower bound for reliable communication.
% ==============================================================================
% II. SYSTEM MODEL AND EXACT DISTRIBUTION
% ==============================================================================
\section{System Model and Exact Distribution}
% --- FIGURE 1: SYSTEM MODEL ---
\begin{figure}[!h]
\centering
\resizebox{0.55\textwidth}{!}{%
\begin{tikzpicture}[
>=Latex,
node distance=2cm,
font=\sffamily\small
]
\def\TxX{5}
\def\RxX{0}
\def\Ymax{2.5}
\def\Ymin{-2.5}
\draw[blue!80!black, very thick] (\RxX, \Ymin) -- (\RxX, \Ymax) node[above] {Rx};
\fill[pattern=north east lines, pattern color=blue!30] (\RxX-0.3, \Ymin) rectangle (\RxX, \Ymax);
\draw[red!80!black, very thick, dashed] (\TxX, \Ymin) -- (\TxX, \Ymax) node[above] {Tx};
\node[draw, fill=yellow!90!orange, minimum width=2cm, minimum height=1cm, align=center] (source) at (\TxX+2.5, 0) {Information\\Source};
\draw[->, thick] (source) -- (\TxX, 0);
\node[draw, fill=yellow!90!orange, minimum width=2cm, minimum height=1cm, align=center] (sink) at (\RxX-2.5, 0) {Information\\Sink};
\draw[<-, thick] (sink) -- (\RxX, 0);
\fill[red] (\TxX, 0.5) circle (3pt);
\fill[blue] (\RxX, 1.2) circle (3pt);
\draw[red, thick, decorate, decoration={random steps,segment length=3pt,amplitude=4pt}]
(\TxX, 0.5) -- (3.5, 0.8) -- (2.0, 0.2) -- (1.0, 1.5) -- (\RxX, 1.2);
\node[red!80!black, font=\footnotesize] at (2.5, 1.8) {Information Particles};
\draw[->, thick, color=red!60!black] (3.5, -0.5) to[bend left=20] node[below, font=\footnotesize] {Drift Velocity $\mathbf{v}$} (1.5, -0.5);
\draw[<->, thick] (\RxX, \Ymin-0.3) -- (\TxX, \Ymin-0.3) node[midway, below] {$\lambda$};
\draw[gray, thin, dashed] (\RxX, \Ymin) -- (\RxX, \Ymin-0.5);
\draw[gray, thin, dashed] (\TxX, \Ymin) -- (\TxX, \Ymin-0.5);
\end{tikzpicture}%
}
\caption{
Abstract FAP channel model featuring hyperplane-shaped transmitters (Tx) and receivers (Rx). The particle undergoes Brownian motion with diffusion coefficient $D$ and drift velocity $\mathbf{v}$ over a transmission distance $\lambda$. The Rx registers the arrival point on the transverse plane.
}
\label{fig:channel_model}
\end{figure}
% ---------------------------------
We consider a diffusive MC system as illustrated in Fig.~\ref{fig:channel_model}, where a transmitter releases a particle at $\mathbf{x}_{\text{in}}$ on a hyperplane. The particle undergoes Brownian motion with diffusion coefficient $D$ (where $\sigma^2 = 2D$) and a constant drift velocity $\mathbf{v}$ perpendicular to the receiving plane located at distance $\lambda$. Let $v = \|\mathbf{v}\|$ denote the drift speed. The received position on the transverse line is modeled as $Y = X + N$, where $N \in \mathbb{R}$ represents the random lateral displacement.
\subsection{The Physical Cauchy Baseline (Zero-Drift)}
The noise distribution is governed by the drift velocity $\mathbf{v}$. As derived in \cite{Lee_ICC24}, in the limit of vanishing drift ($v \to 0$), the complex FAP distribution degenerates into a multivariate Cauchy distribution:
\begin{equation} \label{eq:cauchy_limit}
\lim_{v \to 0} f_N(n|v) = \frac{\lambda}{\pi (n^2 + \lambda^2)}.
\end{equation}
This convergence provides the fundamental baseline for our analysis: the channel is inherently heavy-tailed in the absence of external flow.
\subsection{The General Case (Non-Zero Drift)}
For a general drift $v \neq 0$, the exact PDF of the noise $N$ on a 1D line (2D system) is given by \cite{Lee_TCOM24}:
\begin{equation} \label{eq:exact_pdf}
f_N(n|v) = \frac{v\lambda}{\pi \sigma^2 \sqrt{n^2 + \lambda^2}} e^{\frac{v\lambda}{\sigma^2}} K_1\left( \frac{v}{\sigma^2}\sqrt{n^2+\lambda^2} \right) e^{-\frac{v\sqrt{n^2+\lambda^2}}{\sigma^2}},
\end{equation}
where $K_1(\cdot)$ is the modified Bessel function of the second kind. While \eqref{eq:exact_pdf} is exact, its complex form obscures the physical intuition regarding the channel's tail behavior. We now perform an asymptotic analysis to reveal the underlying structure.
% ==============================================================================
% III. THE ANATOMY OF THE TAIL
% ==============================================================================
\section{The Anatomy of the Tail: Asymptotic Analysis}
The behavior of the noise distribution is governed by the argument of the Bessel function. Let us define the dimensionless parameter $z(n) = \frac{v}{\sigma^2}\sqrt{n^2+\lambda^2}$.
\subsection{Regime 1: The Cauchy Core (Diffusion Dominated)}
Consider the case where the drift is small or the impact position is close to the center, such that $z(n) \ll 1$. Using the asymptotic expansion $K_1(z) \approx 1/z$ \cite{Abramowitz1972}, the PDF becomes:
\begin{equation}
\begin{split}
f_N(n|v) &\approx \frac{v\lambda}{\pi \sigma^2 \sqrt{n^2 + \lambda^2}} \left(\frac{\sigma^2}{v\sqrt{n^2+\lambda^2}}\right) \\
&\quad \times \underbrace{\exp\left(\frac{v(\lambda - \sqrt{n^2+\lambda^2})}{\sigma^2}\right)}_{\to 1} \\
&\propto \frac{1}{n^2+\lambda^2}.
\end{split}
\end{equation}
This recovers the Cauchy distribution \cite{Lee_ICC24}, characterized by an algebraic heavy tail $O(n^{-2})$. In this spatial region, the diffusion force dominates the drift.
\subsection{Regime 2: The Exponential Tail (Drift Dominated)}
Consider the tail behavior where $|n|$ is large, such that $z(n) \gg 1$. Using the expansion $K_1(z) \approx \sqrt{\frac{\pi}{2z}}e^{-z}$ \cite{Abramowitz1972}, the PDF behaves as:
\begin{equation}
f_N(n|v) \sim \frac{e^{-z}}{\sqrt{z} \sqrt{n^2+\lambda^2}} \cdot e^{-z} \approx C \cdot n^{-3/2} \exp\left( - \frac{2v}{\sigma^2} |n| \right).
\end{equation}
Crucially, the tail now exhibits exponential decay. This implies that for any $v \neq 0$, all moments of the distribution (including variance) are finite.
\subsection{The Critical Length Scale}
The transition between the Cauchy regime and the Exponential regime occurs when $z(n) \approx 1$. This defines a critical spatial scale $n_c$:
\begin{equation} \label{eq:critical_scale}
\frac{v}{\sigma^2} n_c \approx 1 \implies n_c = \frac{\sigma^2}{v}.
\end{equation}
The physical interpretation of $n_c$ is the \textit{cutoff length}. For $|n| < n_c$, the noise is heavy-tailed (Cauchy-like); for $|n| > n_c$, it is light-tailed. The non-zero drift FAP channel thus acts as a ``Truncated Cauchy'' channel.
\subsection{Extension to 3D Systems (2D Receiving Plane)}
While the analytical derivations above focus on a 2D system (1D receiving line) for clarity, the identified phase transition is universal and applies to 3D systems (2D receiving plane) relevant to planar MIMO arrays.
In a 3D system, the noise becomes a vector $\mathbf{N} \in \mathbb{R}^2$. The exact PDF involves an exponential term $\exp(-v\|\mathbf{n}'\|/\sigma^2)$, where $\|\mathbf{n}'\| = \sqrt{\|\mathbf{n}\|^2 + \lambda^2}$ denotes the Euclidean propagation distance \cite{Lee_TCOM24}.
Similar asymptotic analysis reveals:
\begin{itemize}
\item \textbf{Core ($\|\mathbf{n}\| \ll n_c$):} The distribution converges to a \textit{Bivariate Cauchy} distribution ($\propto \|\mathbf{n}\|^{-3}$) \cite{Lee_ICC24}, offering 2 spatial degrees of freedom.
\item \textbf{Tail ($\|\mathbf{n}\| \gg n_c$):} The distribution decays exponentially, recovering finite moments.
\end{itemize}
Crucially, the critical spatial scale $n_c = \sigma^2/v$ remains the fundamental physical boundary governing the transition from heavy-tailed interference to light-tailed behavior, regardless of the dimensionality.
% ==============================================================================
% IV. CAPACITY ANALYSIS AND NUMERICAL RESULTS
% ==============================================================================
\section{Capacity Analysis and Numerical Results}
To understand the impact of this tail transition on information transmission, we evaluate the channel capacity under a peak amplitude constraint $A$.
\subsection{Numerical Setup}
We perform numerical simulations with parameters: transmission distance $\lambda=10$, receiver constraint $A=200$, and diffusion coefficient $D=100$ (units normalized). We compute the exact mutual information $I(X; Y)$ assuming a uniform input distribution $X \sim \mathcal{U}[-A, A]$.
% --- ADDED REPRODUCIBILITY DETAILS HERE ---
For comparison, the Gaussian approximation (dashed line) represents the capacity of an AWGN channel $C_{\text{Gauss}} = \frac{1}{2}\ln(1 + P_X/\sigma_N^2(v))$. Here, $\sigma_N^2(v)$ is the noise variance computed numerically from the exact PDF in \eqref{eq:exact_pdf}, and $P_X = A^2/3$ is the average signal power, chosen to match the variance of the uniform input.
% ------------------------------------------
We focus on the 1D simulation to demonstrate the transition effect clearly; the qualitative behavior extends to 2D receiving planes.
\subsection{Results and Interpretation}
Fig.~\ref{fig:capacity} illustrates the capacity as a function of drift velocity. The figure features three distinct curves: the exact capacity (red solid), the Gaussian approximation (blue dashed), and a theoretical baseline (black dotted).
% --- FIGURE 2: CAPACITY RESULTS ---
\begin{figure}[ht]
\centering
% UPDATED FILENAME TO MATCH MATLAB OUTPUT
\includegraphics[width=0.7\linewidth]{Figures/Fig2_Transition.eps}
\caption{Capacity transition of the FAP channel from the Cauchy regime to the Gaussian regime. The significant gap at low drift velocities highlights the failure of standard variance-based Gaussian approximations.}
\label{fig:capacity}
\end{figure}
% ----------------------------------
The \textbf{black dotted line}, labeled as the \textit{Zero-Drift Cauchy Limit} \cite{Lee_ICC24}, serves as a critical benchmark.
\begin{itemize}
\item \textbf{Mathematical Definition:} This line represents the theoretical capacity derived in our previous work \cite{Lee_ICC24}: $C_{\text{Cauchy}} = \ln(A/\lambda)$. Given our simulation parameters ($A=200, \lambda=10$), this yields a constant value of $\ln(20) \approx 3.0$ nats. This aligns with the general information-theoretic capacity of additive Cauchy channels under output strength constraints recently derived by Verd\'{u} \cite[Thm. 13]{Verdu_Entropy23}.
\item \textbf{Physical Meaning (Geometric Limit):} Being horizontal, this limit is independent of drift velocity. It signifies the \textit{geometric limit} of the channel. As $v \to 0$, the noise becomes purely Cauchy, and the capacity is determined solely by the ratio of the receiver aperture $A$ to the diffusion scale $\lambda$.
\end{itemize}
A crucial observation is the behavior of the \textbf{red solid line (Exact Capacity)} relative to the baselines:
\begin{enumerate}
\item \textbf{Robustness in Low Drift:} As $v \to 0$, the exact capacity does not collapse; instead, it stabilizes and parallels the black Cauchy limit.
\item \textbf{Failure of Gaussian Approximation:} Conversely, the Gaussian approximation (blue dashed line) crashes to zero in the low-drift regime. This is because the noise variance explodes as the distribution approaches the Cauchy limit. The Gaussian model incorrectly interprets this high variance as a complete loss of signal fidelity.
\end{enumerate}
These results confirm that in the diffusion-dominated regime ($n_c > A$), relying on variance-based Gaussian models leads to a disastrous underestimation of capacity. The Zero-Drift Cauchy model \cite{Lee_ICC24} provides the correct physical baseline for robust system design.
%%----
\subsection{Impact on Spatial Interference}
To quantify the cross-talk potential in Molecular MIMO systems, we analyze the \textit{interference probability} $P_{\text{int}}(r) = \mathbb{P}(|N| > r)$, which represents the likelihood that a particle intended for a receiver at the origin is erroneously detected by a neighboring receiver located at a separation distance $r$.
Fig.~\ref{fig:interference} presents the simulation results generated using two distinct methods.
First, for the \textbf{Zero-Drift Baseline (Black Dotted)}, we utilize the closed-form Cumulative Distribution Function (CDF) corresponding to the Cauchy PDF in \eqref{eq:cauchy_limit} to calculate $P_{\text{int}}(r) = 1 - \frac{2}{\pi}\arctan\left(\frac{r}{\lambda}\right)$.
Using the asymptotic expansion $\arctan(x) \approx \frac{\pi}{2} - \frac{1}{x}$ for large $x$, we observe that $P_{\text{int}}(r) \approx \frac{2\lambda}{\pi r} \propto r^{-1}$, confirming the algebraic heavy-tail decay depicted in Fig.~\ref{fig:interference}.
Second, for \textbf{Non-Zero Drift (Red and Blue)} velocities $v=0.1$ (low) and $v=5.0$ (high), we perform numerical integration of the exact PDF kernel given in \eqref{eq:exact_pdf}. The probability is computed as $P_{\text{int}}(r) = 2 \int_{r}^{\infty} f_N(n|v) \, dn$, where the factor of 2 accounts for the symmetry of the distribution.
\begin{figure}[ht]
\centering
\includegraphics[width=0.7\linewidth]{Figures/Fig3_Interference_r.eps}
\vspace{3mm} % 加一小點空間讓 caption 更自然
\caption{Spatial interference probability versus separation distance $r$. The zero-drift baseline is computed analytically via the Cauchy CDF, while non-zero drift curves are obtained by numerically integrating the exact Bessel-based PDF. For the high-drift case (blue dashed), the exponential truncation begins visibly after the critical scale $n_c \approx 40\,\mu\text{m}$.}
\label{fig:interference}
\end{figure}
The results illustrate the physical phase transition vividly:
\begin{itemize}
\item \textbf{High Drift ($v=5.0$):} We explicitly mark the critical scale $n_c = \sigma^2/v = 40\,\mu\text{m}$ in Fig.~\ref{fig:interference}. For separation distances $r > n_c$, the interference drops precipitously (exponentially). This confirms that $n_c$ serves as a safe ``cutoff distance'' for dense antenna packing.
\item \textbf{Low Drift ($v=0.1$):} The theoretical critical scale is $n_c = 2000\,\mu\text{m}$, which lies beyond the plotted range ($r \le 1000\,\mu\text{m}$). Consequently, the red curve remains tightly coupled to the zero-drift Cauchy baseline ($\sim r^{-1}$) throughout the figure. This signifies that in diffusion-dominated regimes, interference persists over extremely long distances, preventing dense integration.
\end{itemize}
% ==============================================================================
% V. DISCUSSION: OPTIMAL INPUTS AND SHAPING LOSS
% ==============================================================================
\section{Discussion: Optimal Inputs and Shaping Loss}
The capacity results presented in Fig.~\ref{fig:capacity} reveal a persistent gap between the numerical capacity (red solid line) and the theoretical zero-drift limit (black dotted line). This gap provides insight into the choice of input distribution, a critical factor in modulation design.
\subsection{The Shaping Loss in Zero-Drift Regime}
In the zero-drift limit ($v \to 0$), the channel becomes an additive Cauchy noise channel. The theoretical capacity limit (black dotted line in Fig.~\ref{fig:capacity}) is achieved by an optimal input distribution $X^\star$ which is Cauchy-distributed \cite{Lee_ICC24, Verdu_Entropy23}:
\begin{equation}
C_{\text{opt}} = \max_{P_X: \, \mathrm{strength}(Y) \le A} I(X; Y) = \ln\left(\frac{A}{\lambda}\right).
\end{equation}
However, practical MC transmitters are typically constrained by a finite emission aperture, necessitating a peak amplitude constraint $|X| \le A$. In our simulations, we employed a \textbf{Uniform distribution} $X \sim \mathcal{U}[-A, A]$.
At high signal strength ($A \gg \lambda$), the entropy of the output $Y$ is dominated by the input geometry. The mutual information for the uniform input can be approximated as:
\begin{equation}
I_{\text{unif}} \approx h(X_{\text{unif}}) - h(N) = \ln(2A) - \ln(4\pi\lambda).
\end{equation}
Consequently, the \textit{shaping loss} $\Delta$, defined as the capacity gap between the optimal Cauchy input and the practical uniform input, is asymptotically given by:
\begin{equation}
\Delta = C_{\text{opt}} - I_{\text{unif}} \approx \ln\left(\frac{A}{\lambda}\right) - \ln\left(\frac{A}{2\pi\lambda}\right) = \ln(2\pi) \approx 1.83 \text{ nats}.
\end{equation}
This theoretical gap aligns well with the numerical results in Fig.~\ref{fig:capacity} (showing a gap of $\approx 1.5$ nats). The loss arises because the Uniform distribution lacks the heavy tails of the optimal Cauchy input, leading to a mismatch with the channel's native noise statistics. Nevertheless, the fact that the capacity remains high even with a sub-optimal Uniform input highlights the inherent robustness of FAP modulation.
\subsection{The Transition to Discrete Inputs}
As the drift velocity increases ($v \to \infty$), the noise statistics transition toward a Gaussian distribution with variance $\sigma_v^2$ decreasing with $v$. In this regime, the channel can be approximated as an additive Gaussian noise channel under a peak amplitude constraint $|X| \le A$:
\begin{equation}
Y \approx X + Z, \quad Z \sim \mathcal{N}(0, \sigma_v^2), \quad |X| \le A.
\end{equation}
It is a classical result by Smith that the capacity-achieving distribution for this setting is discrete with a finite number of mass points \cite{Smith1971}:
\begin{equation}
P_{X_{\text{high}}}^\star(x) = \sum_{k=1}^{M} w_k \delta(x - x_k),
\end{equation}
where $x_k \in [-A, A]$ and $M$ depends on the signal-to-noise ratio $A^2/\sigma_v^2$.
This creates a structural trade-off:
\begin{itemize}
\item \textbf{Low Drift:} The optimal input $f_{X_{\text{low}}}^\star(x)$ is a continuous, heavy-tailed Cauchy density to match the channel's infinite variance.
\item \textbf{High Drift:} The optimal input $P_{X_{\text{high}}}^\star(x)$ collapses into a discrete constellation to exploit the light-tailed Gaussian noise.
\end{itemize}
Our results demonstrate that the simple Uniform input, $f_{X}(x) = \frac{1}{2A}$ for $|x|\le A$, serves as a robust compromise. It possesses the continuous support required for the low-drift regime while maximizing the differential entropy $h(X)$ under the peak constraint, thereby approximating the high-SNR performance in the drift-dominated regime.
% ==============================================================================
% VI. CONCLUSION
% ==============================================================================
\section{Conclusion}
This letter analyzed the transition of FAP noise from a heavy-tailed Cauchy distribution to a light-tailed exponential distribution. We identified a critical length scale $n_c = \sigma^2/v$ separating the diffusion core from the drift tail. Our results reinforce the importance of the physical Cauchy limit \cite{Lee_ICC24} as a fundamental baseline for low-drift molecular channels, where traditional Gaussian assumptions fail. Furthermore, characterizing this tail behavior is shown to be critical for determining the feasible density of Molecular MIMO arrays under spatial interference constraints.
% ==============================================================================
% REFERENCES
% ==============================================================================
\bibliographystyle{IEEEtran}
\bibliography{main}
\end{document}