Table of Contents
Fetching ...

Mathematical comparison of classical and quantum mechanisms in optimization under local differential privacy

Yuuya Yoshida

TL;DR

The paper investigates the boundary between classical and quantum mechanisms in optimization under local differential privacy. It formalizes classical $ε$-DP and classical-quantum $ε$-DP, defines the essentially classical subset $\mathrm{EC}_n(ε)$, and proves that $\mathrm{EC}_n(ε)\neq\mathrm{CQ}_n(ε)$ for all $n\ge3$, establishing a quantum advantage in optimization tasks under CQ-$ε$-DP. It provides a concrete objective based on the RLD Fisher information that yields equal classical and essentially classical optima, but strict superiority for CQ optima when $n\ge3$, together with explicit bounds via auxiliary functions. The work also constructs explicit CQ-$ε$-DP $n$-tuples not lying in $\mathrm{EC}_n(ε)$, demonstrating the practical gap between these sets, and discusses implications for dimension handling and future refinements of the bounds. Overall, the results quantify the separation between classical- and quantum-driven DP mechanisms and illuminate when quantum resources provide a genuine advantage in private data optimization.

Abstract

Let $\varepsilon>0$. An $n$-tuple $(p_i)_{i=1}^n$ of probability vectors is called $\varepsilon$-differentially private ($\varepsilon$-DP) if $e^\varepsilon p_j-p_i$ has no negative entries for all $i,j=1,\ldots,n$. An $n$-tuple $(ρ_i)_{i=1}^n$ of density matrices is called classical-quantum $\varepsilon$-differentially private (CQ $\varepsilon$-DP) if $e^\varepsilonρ_j-ρ_i$ is positive semi-definite for all $i,j=1,\ldots,n$. Denote by $\mathrm{C}_n(\varepsilon)$ the set of all $\varepsilon$-DP $n$-tuples, and by $\mathrm{CQ}_n(\varepsilon)$ the set of all CQ $\varepsilon$-DP $n$-tuples. By considering optimization problems under local differential privacy, we define the subset $\mathrm{EC}_n(\varepsilon)$ of $\mathrm{CQ}_n(\varepsilon)$ that is essentially classical. Roughly speaking, an element in $\mathrm{EC}_n(\varepsilon)$ is the image of $(p_i)_{i=1}^n\in\mathrm{C}_n(\varepsilon)$ by a completely positive and trace-preserving linear map (CPTP map). In a preceding study, it is known that $\mathrm{EC}_2(\varepsilon)=\mathrm{CQ}_2(\varepsilon)$. In this paper, we show that $\mathrm{EC}_n(\varepsilon)\not=\mathrm{CQ}_n(\varepsilon)$ for every $n\ge3$, and estimate the difference between $\mathrm{EC}_n(\varepsilon)$ and $\mathrm{CQ}_n(\varepsilon)$ in a certain manner.

Mathematical comparison of classical and quantum mechanisms in optimization under local differential privacy

TL;DR

The paper investigates the boundary between classical and quantum mechanisms in optimization under local differential privacy. It formalizes classical -DP and classical-quantum -DP, defines the essentially classical subset , and proves that for all , establishing a quantum advantage in optimization tasks under CQ--DP. It provides a concrete objective based on the RLD Fisher information that yields equal classical and essentially classical optima, but strict superiority for CQ optima when , together with explicit bounds via auxiliary functions. The work also constructs explicit CQ--DP -tuples not lying in , demonstrating the practical gap between these sets, and discusses implications for dimension handling and future refinements of the bounds. Overall, the results quantify the separation between classical- and quantum-driven DP mechanisms and illuminate when quantum resources provide a genuine advantage in private data optimization.

Abstract

Let . An -tuple of probability vectors is called -differentially private (-DP) if has no negative entries for all . An -tuple of density matrices is called classical-quantum -differentially private (CQ -DP) if is positive semi-definite for all . Denote by the set of all -DP -tuples, and by the set of all CQ -DP -tuples. By considering optimization problems under local differential privacy, we define the subset of that is essentially classical. Roughly speaking, an element in is the image of by a completely positive and trace-preserving linear map (CPTP map). In a preceding study, it is known that . In this paper, we show that for every , and estimate the difference between and in a certain manner.

Paper Structure

This paper contains 7 sections, 21 theorems, 95 equations.

Key Result

Proposition 1.4

For all $\varepsilon>0$, $\mathrm{EC}_2(\varepsilon)=\mathrm{CQ}_2(\varepsilon)$.

Theorems & Definitions (45)

  • Definition 1.1: Classical $\varepsilon$-DP Duchi and classical-quantum $\varepsilon$-DP ISIT2020
  • Definition 1.2: Monotonicity for CPTP maps
  • Definition 1.3: Essentially classical element
  • Proposition 1.4
  • Definition 1.5
  • Theorem 1.6
  • Theorem 1.7
  • Corollary 1.8
  • Proposition 1.9
  • Definition 2.1: RLD Fisher information book1
  • ...and 35 more