Extremal ULDP Mechanisms

• The paper establishes that extremal ULDP mechanisms are the canonical subclass that reduce the complex privacy–utility trade-off optimization to a manageable form.
• It details a staircase mechanism for sensitive symbols combined with direct, invertible outputs for non-sensitive data to balance privacy and utility.
• The analysis culminates in a saddle-point formulation and uBD schemes that achieve the fundamental estimation limits under the ULDP framework.

Searching arXiv for the cited ULDP and related extremal LDP papers. {"query":"all:(\"Fundamental Limit of Discrete Distribution Estimation under Utility-Optimized Local Differential Privacy\" OR \"Extreme Points of the (0,δ)-LDP Polytope with Small Input Size and Arbitrary Output Sizes\" OR \"Factorization by extremal privacy mechanisms: new insights into efficiency\")","max_results":10} {"query":"(Yoon et al., 29 Sep 2025) OR (Rawat et al., 8 Jun 2026) OR (Amorino et al., 29 Jul 2025)","max_results":10} Extremal ULDP mechanisms are the canonical subclass of utility-optimized local differential privacy channels used to analyze discrete distribution estimation when the input alphabet is partitioned into sensitive and non-sensitive symbols. In the ULDP model, sensitive data are subject to ordinary $\epsilon$-LDP on a protected output channel, while non-sensitive data may be revealed through invertible outputs. The extremal subclass combines a staircase mechanism on the sensitive symbols with direct disclosure for the non-sensitive symbols, and it serves as the central reduction device in the exact characterization of the privacy–utility trade-off for discrete distribution estimation (Yoon et al., 29 Sep 2025).

1. ULDP model and estimation problem

The setting considers a categorical random variable

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$

observed by $n$ i.i.d. clients. The alphabet is partitioned into sensitive and non-sensitive symbols,

$$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$

with $1\le v<w$. A privacy mechanism is a conditional distribution

$Q:\mathcal X\to\mathcal P(\mathcal Y),$

and inference is based on privatized outputs $Y_i\sim Q(\cdot\mid X_i)$ (Yoon et al., 29 Sep 2025).

ULDP is defined by a partition of the output alphabet,

$$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$

into protected and invertible outputs. For each protected output $y\in\mathcal Y_{\mathrm P}$, the mechanism satisfies full $\epsilon$-LDP,

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$0

For each invertible output $X\in\mathcal X=[w], \qquad P\in\Delta_w,$1, there exists a unique non-sensitive symbol $X\in\mathcal X=[w], \qquad P\in\Delta_w,$2 such that

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$3

Thus protected outputs conceal all inputs up to the usual multiplicative LDP relation, whereas invertible outputs reveal a non-sensitive input exactly.

The minimax estimation objective is

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$4

with asymptotic counterpart

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$5

The role of extremal ULDP mechanisms is to reduce this optimization over all admissible ULDP channels to a smaller canonical family that remains sufficient for the converse and achievability.

2. Definition and matrix form of extremal ULDP mechanisms

An extremal $X\in\mathcal X=[w], \qquad P\in\Delta_w,$6-ULDP mechanism is defined on the structured output alphabet

$X\in\mathcal X=[w], \qquad P\in\Delta_w,$7

Its protected outputs are indexed by nonempty subsets of the sensitive alphabet, while its invertible outputs are singleton labels for non-sensitive symbols (Yoon et al., 29 Sep 2025).

For every protected output $X\in\mathcal X=[w], \qquad P\in\Delta_w,$8, there exists a nonnegative weight $X\in\mathcal X=[w], \qquad P\in\Delta_w,$9 such that

$n$0

For every invertible output $n$1 with $n$2,

$n$3

Equivalently, the protected component is a staircase-pattern matrix over subsets of $n$4, and the invertible component is a diagonal identity-like structure over the non-sensitive symbols.

Validity is characterized by the normalization constraints

$n$5

These constraints ensure that each sensitive row sums to one.

The matrix representation uses the staircase matrix

$n$6

where $n$7 encodes the binary representation of subset labels. In this form,

• $n$8, so the sensitive block is exactly an extremal LDP staircase mechanism.
• $n$9, so non-sensitive symbols use the same protected-output probabilities as low as allowed by ULDP.
• $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$0 is diagonal/identity-like.
• $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$1.

This construction is deliberately extreme in the sense used by the paper: protected outputs are as informative as possible subject to $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$2-LDP, and non-sensitive symbols are disclosed whenever possible.

3. Degradation theorem and extremal reduction

The paper introduces extremal ULDP mechanisms because the converse proof requires a manageable mechanism class over which Fisher information can be optimized. The key structural theorem states

$$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$3

meaning that every ULDP mechanism is degraded by an extremal ULDP mechanism (Yoon et al., 29 Sep 2025).

Concretely, for any ULDP mechanism $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$4, there exist an extremal mechanism $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$5 and a post-processing channel $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$6 such that

$$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$7

Because post-processing cannot increase information, the statistical difficulty induced by $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$8 is already present in $$\mathcal X_{\mathrm S}=[v], \qquad \mathcal X_{\mathrm N}=\mathcal X\setminus\mathcal X_{\mathrm S}=[v+1:w],$$9. A common misconception is that extremal ULDP mechanisms form only a convenient approximation class; the reduction theorem rules out that interpretation. The extremal family is sufficient for the minimax analysis because every admissible mechanism factors through it by degradation.

This reduction supports a generalized uniform asymptotic Cramér–Rao lower bound. For a compact family $1\le v<w$0 degrading a broader class $1\le v<w$1, and for each interior distribution $1\le v<w$2,

$1\le v<w$3

and after taking the worst case over $1\le v<w$4,

$1\le v<w$5

Applying this theorem with $1\le v<w$6 yields the converse

$1\le v<w$7

The technical ingredients highlighted in the proof are a local perturbation prior $1\le v<w$8, a Bayesian CRLB, the equicontinuity bound

$1\le v<w$9

and compactness together with Dini’s theorem to pass the limit through the infimum over $Q:\mathcal X\to\mathcal P(\mathcal Y),$0.

4. Decomposition of estimation difficulty and the saddle-point formula

The main converse argument is organized around a decomposition of the simplex tangent space into three orthogonal components aligned with the sensitive/non-sensitive split,

$Q:\mathcal X\to\mathcal P(\mathcal Y),$1

The three subspaces are

$Q:\mathcal X\to\mathcal P(\mathcal Y),$2

$Q:\mathcal X\to\mathcal P(\mathcal Y),$3

$Q:\mathcal X\to\mathcal P(\mathcal Y),$4

The paper interprets these as relative proportions among sensitive symbols, relative proportions among non-sensitive symbols, and the total mass $Q:\mathcal X\to\mathcal P(\mathcal Y),$5, respectively (Yoon et al., 29 Sep 2025).

Subspace	Definition	Interpretation
$Q:\mathcal X\to\mathcal P(\mathcal Y),$6	Sensitive coordinates only, zero sum on $Q:\mathcal X\to\mathcal P(\mathcal Y),$7	Relative proportions among sensitive symbols
$Q:\mathcal X\to\mathcal P(\mathcal Y),$8	Non-sensitive coordinates only, zero sum on $Q:\mathcal X\to\mathcal P(\mathcal Y),$9	Relative proportions among non-sensitive symbols
$Y_i\sim Q(\cdot\mid X_i)$0	One-dimensional span of group-contrast vector	Total sensitive mass

The analysis then restricts attention to the one-parameter family

$Y_i\sim Q(\cdot\mid X_i)$1

which is uniform within each group. The lower bound is decomposed into three CRLB-like terms,

$Y_i\sim Q(\cdot\mid X_i)$2

where $Y_i\sim Q(\cdot\mid X_i)$3 is the mixture distribution over block sizes.

The terms are

$Y_i\sim Q(\cdot\mid X_i)$4

$Y_i\sim Q(\cdot\mid X_i)$5

$Y_i\sim Q(\cdot\mid X_i)$6

The objective is concave in $Y_i\sim Q(\cdot\mid X_i)$7 and convex in $Y_i\sim Q(\cdot\mid X_i)$8, so the final characterization is a saddle-point problem. This is the paper’s main structural reduction: the original minimax optimization over all ULDP mechanisms and estimators becomes a finite-dimensional optimization over the sensitive-mass parameter $Y_i\sim Q(\cdot\mid X_i)$9 and the block-size mixture $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$0.

5. uBD schemes as constructive extremal ULDP mechanisms

The achievability side is built around utility-optimized block design schemes. A uBD mechanism uses a mixture of block design mechanisms $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$1, with mixture weights $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$2. For sensitive inputs, it samples $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$3 and applies the corresponding $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$4-uniform block design. For non-sensitive inputs, outputs are chosen so that the overall mechanism becomes an extremal ULDP mechanism (Yoon et al., 29 Sep 2025).

Formally, a uBD mechanism is an extremal ULDP mechanism with

$$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$5

The crucial point is that uBD schemes are not arbitrary approximations: they are specific extremal ULDP mechanisms whose protected-output weights are realized by mixing block designs. This explains why the converse and achievability align so tightly.

The estimator is score-based and follows the same $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$6 decomposition,

$$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$7

with

$$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$8

At the saddle point $$\mathcal Y=\mathcal Y_{\mathrm P}\sqcup \mathcal Y_{\mathrm I},$$9, this estimator saturates the CRLB-like lower bound.

The paper also identifies regimes in which the optimizer takes especially simple forms. When $y\in\mathcal Y_{\mathrm P}$0, the resulting scheme is uRR and is optimal in certain high-privacy/low-sensitivity regimes. In another regime, $y\in\mathcal Y_{\mathrm P}$1 for $y\in\mathcal Y_{\mathrm P}$2, and the resulting scheme coincides with a simple uBD mechanism. In those cases, the converse bound is tight and the uBD scheme exactly attains the fundamental limit.

6. Relation to broader extremal privacy mechanism theory

Extremal ULDP mechanisms sit within a broader line of work in which extremal privacy channels reduce large optimization problems to boundary objects. In classical $y\in\mathcal Y_{\mathrm P}$3-LDP, every privacy mechanism can be factored as

$y\in\mathcal Y_{\mathrm P}$4

in finite alphabets, where $y\in\mathcal Y_{\mathrm P}$5 is extremal, and the same idea extends to continuous spaces via a Choquet representation,

$y\in\mathcal Y_{\mathrm P}$6

In that setting, the extremal channels are staircase mechanisms characterized by

$y\in\mathcal Y_{\mathrm P}$7

and the reduction supports Fisher-information optimization over extremal mechanisms alone (Amorino et al., 29 Jul 2025).

A different extremal geometry appears in the $y\in\mathcal Y_{\mathrm P}$8-LDP polytope. There, the mechanism space

$y\in\mathcal Y_{\mathrm P}$9

is a convex polytope, and the recent characterization of its extreme points for $\epsilon$0 and $\epsilon$1 relies on localization, tight points, perturbation equations, and star configurations. For $\epsilon$2, extremality forces sparse localized support patterns, while for larger alphabets new star-configuration-type extreme mechanisms arise, with extremality governed by subset-sum rigidity and $\epsilon$3 conditions (Rawat et al., 8 Jun 2026).

These adjacent results clarify what is distinctive about the ULDP case. A plausible implication is that extremality under ULDP is best understood as an asymmetric boundary phenomenon: the sensitive component inherits staircase extremality from ordinary LDP, while the non-sensitive component contributes invertible outputs that are absent in the standard LDP model. Another common misconception is that “extremal” necessarily means uniformly maximal disclosure. In ULDP that is not the case. Sensitive data remain constrained by full $\epsilon$4-LDP on the protected channel, whereas direct revelation is permitted only for non-sensitive symbols through the invertible channel.

The resulting conceptual picture is precise. Extremal ULDP mechanisms are the canonical boundary channels for the ULDP estimation problem: they make the converse tractable, they support the exact saddle-point characterization of the optimal privacy–utility trade-off, and they admit constructive realization through uBD schemes that match the lower bound (Yoon et al., 29 Sep 2025).