---
title: Wang-Elia Algorithm Overview
url: https://www.emergentmind.com/topics/wang-elia-algorithm
type: topic
---

# Wang-Elia Algorithm Overview

Searching arXiv for the cited papers and related usages of “Wang-Elia Algorithm.”
The term **Wang-Elia algorithm** is used in more than one technical literature, and its meaning depends on context. In residue number systems, it denotes a residue-to-binary conversion method based on Wang’s **New Chinese Remainder Theorem I**, a formulation later analyzed as a restatement of an earlier CRT-based approach rather than a fundamentally new algorithm [0707.3732]. In distributed optimization, it denotes an algorithm proposed in 2010 by J. Wang and N. Elia for distributed consensus optimization, in which each agent maintains a primal estimate and an auxiliary state; a Lyapunov-based analysis established input-to-state stability, linear convergence in the unperturbed case, and explicit links with Gradient Tracking and distributed integral control [2204.10030]. A further usage appears in the literature on Weyl invariant \(E_8\) Jacobi forms, where a “Wang-Elia Algorithm” is identified as a Fourier-expansion and Weyl-orbit method that is explicitly contrasted with a later purely algebraic construction [2201.06895].

## 1. Terminological scope

The name does not designate a single universally fixed procedure across mathematics, signal processing, and control. Instead, the label attaches to distinct algorithmic objects associated with work by authors named Wang and Elia, and the surrounding literature treats these objects independently.

In the residue number system literature, the expression refers to the converter derived from Wang’s **New CRT I** formula for reconstructing an integer \(X\) from residues modulo pairwise coprime moduli \(\{P_1,\ldots,P_n\}\) [0707.3732]. In distributed optimization, the same expression refers to a discrete-time networked algorithm for minimizing a sum of private convex costs over a shared decision variable, with communication over a connected, undirected graph [2204.10030]. In the Jacobi-form literature, the term denotes yet another method, one based on Fourier expansion and analysis of Weyl orbits, and is used mainly as a point of comparison for a newer algebraic alternative [2201.06895].

A central interpretive point is therefore lexical rather than mathematical: the phrase **Wang-Elia algorithm** is domain-specific. Any technical discussion requires immediate disambiguation by field.

## 2. Residue-to-binary conversion and Wang’s New CRT I

In residue number systems, an integer \(X\) is represented by its residues \(x_i\) modulo pairwise coprime moduli \(P_i\), with \(M=\prod_{j=1}^n P_j\). The underlying computational problem is efficient **residue-to-binary conversion**. The classical CRT reconstruction is written as
\[
X = \left(\sum_{i=1}^n s_i \cdot \left(\left(\frac{1}{s_i}\right)_{P_i} x_i \right) \right) \bmod M,
\]
where \(s_i=M/P_i\) and \(\left(\frac{1}{s_i}\right)_{P_i}\) is the inverse of \(s_i\) modulo \(P_i\) [0707.3732].

Wang’s **New CRT I** introduces the alternative expression
\[
X = \left| x_1 + k_1(x_2-x_1)P_1 + k_2(x_3-x_2)P_1P_2 + \ldots + k_{n-1}(x_n-x_{n-1})P_1\ldots P_{n-1} \right|_{M},
\]
with
\[
k_i = \left( \frac{1}{\prod_{j=1}^i P_j} \right)_{\prod_{j=i+1}^n P_j}, \qquad 1 \le i \le n-1.
\]
The so-called **Wang-Elia algorithm** for residue-to-binary conversion is based on this formula [0707.3732].

The relevant historical background includes Huang’s algorithm and the work of Hitz and Kaltofen. Huang proposed computing the CRT formula without modulo-\(M\) operations by representing terms
\[
X_i=s_i\cdot \left(\left(\frac{x_i}{s_i}\right)_{P_i}\right)
\]
in a mixed radix system through precomputed tables, at the cost of substantial storage and precomputation. Hitz and Kaltofen observed that, instead of large tables, one can precompute the mixed-radix representation of the constants \(s_i\) or \(s_i\left(\frac{1}{s_i}\right)_{P_i}\), compute \(x_i\left(\frac{1}{s_i}\right)_{P_i}\) modulo \(P_i\), and then evaluate the sum in mixed radix form using the precomputed digits. They also emphasized the “second form” of CRT,
\[
X = \left( \sum_{i=1}^n s_i \left(\frac{1}{s_i}\right)_{P_i} x_i \right) \bmod M.
\]
This second form is the key comparator for later analysis of Wang’s formula [0707.3732].

## 3. Origin, equivalence, and originality claims in the RNS literature

Beuchat’s analysis establishes that Wang’s **New CRT I** is not a fundamentally new reconstruction principle. The paper shows that the formula is a rewriting of the second CRT form above and that the constants \(k_i\) are merely the mixed radix digits of numbers congruent to \(s_i\left(\frac{1}{s_i}\right)_{P_i}\) modulo \(M\) [0707.3732]. On that basis, the Wang-Elia residue-to-binary method is characterized as a restatement, with minor technical differences, of an algorithmic interpretation already sketched by Hitz and Kaltofen.

The paper also corrects a specific claim by Ananda Mohan. Mohan had argued that Wang’s approach is the same as Huang’s and derivable from the constructive proof of the classical CRT. Beuchat states that this proof is **erroneous**. The correction is precise: Wang’s method does **not** use the mixed-radix conversion of the \(X_i\) terms as Huang’s does; rather, it is a reformulation of the CRT-based approach connected to Hitz and Kaltofen’s second-form viewpoint [0707.3732].

This reassessment directly affects claims of originality. Wang had presented the method as **substantially different** from CRT and MRS approaches, but Beuchat argues that the proper historical placement is different: the formula is a different expression of a known technique rather than a new conversion algorithm. The practical implication given in the paper is that explicit computation of the \(k_i\) constants is unnecessary, because the same result can be obtained through the CRT formulation together with available mixed-radix digit tables. The broader significance stated there is historiographic and engineering-oriented: clarifying the equivalence helps avoid redundant work and refines implementation choices in hardware residue-to-binary converters [0707.3732].

## 4. Distributed consensus optimization algorithm

In distributed optimization, the Wang-Elia algorithm addresses **distributed consensus optimization** with \(N\) agents, each possessing a private convex function \(f_i:\mathbb{R}\to\mathbb{R}\). The collective objective is to minimize the sum of these costs over a shared decision variable \(\theta\), using only local computation and communication over a connected, undirected graph [2204.10030].

Each agent \(i\) maintains two scalar states: \(x_i\in\mathbb{R}\), the current estimate of the optimizer, and \(z_i\in\mathbb{R}\), an auxiliary variable. The perturbed update equations are
\[
\begin{aligned}
x_i^+ &= x_i + \sum_{j \in \mathcal{N}_i} k_{ij} \big( x_j - x_i + z_j - z_i \big) - \gamma \nabla f_i(x_i) + w_i \\
z_i^+ &= z_i - \sum_{j \in \mathcal{N}_i} k_{ij} (x_j - x_i) + s_i ,
\end{aligned}
\]
where \(\mathcal{N}_i\) denotes the neighbor set, \(k_{ij}=k_{ji}>0\) are symmetric weights, \(\gamma>0\) is a stepsize, and \(w_i,s_i\) are additive perturbations modeling errors or uncertainty [2204.10030].

In aggregate form, with \(x=(x_1,\dots,x_N)\), \(z=(z_1,\dots,z_N)\), perturbations \(w,s\), Laplacian \(K\), and
\[
\Phi(x):=(\nabla f_1(x_1),\ldots,\nabla f_N(x_N))^T,
\]
the dynamics become
\[
\begin{aligned}
x^+ &= (I-K)x - Kz - \gamma \Phi(x) + w \\
z^+ &= z + Kx + s .
\end{aligned}
\]
The functional interpretation given in the source is that the \(x\)-update combines consensus forcing with local gradient descent, while the \(z\)-update integrates disagreement among neighboring \(x\)-variables so as to facilitate global consensus. A stated structural property is that **no special initialization** is required for either variable [2204.10030].

## 5. Stability, linear convergence, and relation to Gradient Tracking

The 2022 analysis of the distributed algorithm is Lyapunov-based and assumes that each \(f_i\) is smooth and that the aggregate objective \(\sum f_i\) is strongly convex [2204.10030]. The state is transformed into coordinates separating **average** and **dispersion**, and a composite Lyapunov function \(V\) is constructed to capture both the optimization error and the consensus error.

The main result is an **input-to-state stability** theorem relative to a closed invariant set of optimal equilibria. Specifically, for sufficiently small \(\gamma\), there exist \(\gamma^*,\alpha>0\) such that every solution \((x^t,z^t)\) of the perturbed system satisfies
\[
|(x^t, z^t)|_{\mathcal{O}^*} \leq \alpha \mu_\gamma^t |(x^0, z^0)|_{\mathcal{O}^*}
+ \rho_\gamma \sup_{k < t} |w^k|
+ \tau_\gamma \sup_{k < t} |(w^k, s^k)|,
\]
where \(\mu_\gamma\in(0,1)\) gives the linear decay rate and \(\rho_\gamma,\tau_\gamma\) quantify disturbance amplification [2204.10030]. In the absence of perturbations, the result implies linear convergence of the local estimates and Lyapunov stability of the optimal steady state. Under bounded perturbations, the state remains bounded and close to the optimum, with a disturbance-dependent residual.

The same paper draws explicit comparisons with **Gradient Tracking**. A comparison update is written as
\[
\begin{aligned}
x^+ &= Rx + z - \gamma \Phi(x) \\
z^+ &= Cz - \gamma (C-I)\Phi(x),
\end{aligned}
\]
with stochastic matrices \(R\) and \(C\), and this method requires \(1^T z^0=0\) for convergence [2204.10030]. The contrast emphasized there is structural: Wang-Elia’s \(z\)-dynamics are described as “decoupled” and always filtered out in the \(x\)-update, making the method initialization-free and robust, whereas Gradient Tracking becomes fragile if the auxiliary variables are not carefully initialized. The paper further states that even small constant perturbations can cause divergence in Gradient Tracking, while Wang-Elia remains stable in the ISS sense. Both algorithms are also interpreted as forms of **distributed PI control**, with Wang-Elia implementing integral action distributively and robustly [2204.10030].

## 6. Other uses of the label and related algorithmic distinctions

A separate use of the expression appears in the theory of Weyl invariant \(E_8\) Jacobi forms. In that literature, a later algebraic construction is explicitly said to be distinct from the **Wang-Elia Algorithm**, which is described as relying on Fourier expansion and analysis of Weyl orbits [2201.06895]. Sakai’s construction instead works inside the polynomial algebra
\[
\mathcal{R}=\mathbb{C}[E_4,E_6,a_2,a_3,a_4,b_1,b_2,b_3,b_4,b_5,b_6],
\]
uses a necessary-and-sufficient algebraic criterion for holomorphicity, and does not require Fourier expansion. The paper states that a similar embedding of \(J_{E_8}\) into a polynomial algebra had been considered in the earlier Wang-related work, but that its own theorem is stronger [2201.06895].

This usage reinforces the broader terminological point that **Wang-Elia algorithm** is not a single cross-disciplinary object. In the RNS literature, the term is entangled with debates over CRT reformulation and originality; in distributed optimization, it names a two-state consensus algorithm with ISS and linear convergence guarantees; in the Jacobi-form literature, it denotes a Fourier/Weyl-orbit procedure that serves primarily as a benchmark for a newer algebraic method [0707.3732; 2204.10030; 2201.06895].

A plausible implication is that citations of the term without field qualification are intrinsically ambiguous. For technical precision, the literature is best read as defining three context-bound referents: a residue-to-binary converter associated with Wang’s New CRT I, a distributed consensus optimization scheme due to J. Wang and N. Elia, and a Jacobi-form construction based on Fourier expansion and Weyl orbit analysis.

Source: https://www.emergentmind.com/topics/wang-elia-algorithm