Hyperfinite Robust Representation Theorem

• The theorem converts infinite-dimensional dual representations in risk measures, Dirichlet forms, and G-expectation into hyperfinite formulations recoverable via standard-part operations.
• It leverages non-standard analysis through hyperfinite sets and Loeb measure constructions to transform continuum probabilistic objects into finite combinatorial analogues.
• Its framework underpins coherent risk measurement, plug-in consistency, and bootstrap validity, seamlessly linking population-level theory to finite-sample procedures.

The Hyperfinite Robust Representation Theorem encapsulates a fundamental non-standard analysis paradigm: infinite-dimensional or continuum probabilistic objects and functionals are realized as standard parts (shadows) of hyperfinite, combinatorial formulas. This theorem and its variants systematically convert duality and robust representation formulas from classical analysis to the non-standard, hyperfinite regime, yielding a framework where population-level, asymptotic, or infinite-dimensional results become recoverable from formally finite internal constructions. Hyperfinite robust representation results are central in coherent risk measurement, Dirichlet forms, and sublinear expectations, enabling seamless transfer of structure and limit theorems between infinite and finite cases via standard-part operations.

1. Non-standard Analysis and Hyperfinite Setups

Key results exploit a countably saturated non-standard universe, enabling manipulation of hyperfinite sets (internal sets with a finite, but nonstandardly large, number of atoms indexed by a hypernatural $N$). For instance, in risk theory, given a standard atomless separable probability space $(\Omega,\mathcal{F},P)$, one constructs a Loeb probability space $(I_N,\mathcal{I}_N^L, L(\mu_N))$ from the hyperfinite set $I_N := \{1,\ldots, N\}\subset {}^*\mathbb{N}$ and the internal counting measure $\mu_N(A) = |A|/N$. The Loeb extension produces an isomorphic probability space for $L^\infty$-type functionals, providing a robust framework for measure, integration, and duality in the hyperfinite context (Kania, 31 Jan 2026).

In the analytic context (e.g., Dirichlet forms), compact metric spaces $(X, d)$ are represented by $\delta$-hyperfinite partitions—partitions of ${}^*X$ into hyperfinite families of internal Borel sets of diameter at most $\delta\approx 0$. Probabilistic objects such as measures and transition kernels admit internal, finitely additive counterparts; integration and expectation become hyperfinite sums (Anderson et al., 2020).

2. Hyperfinite Robust Representation: Statement and Mechanism

Consider a convex, monotone, cash-additive, positively homogeneous, Fatou property-satisfying risk measure $\rho:L^\infty \to \mathbb{R}$. The classical robust representation is

$$\rho(X) = \sup_{Z\in \mathcal{Z}} \mathbb{E}[-X Z],$$

for a $\sigma(L^1, L^\infty)$-compact convex set $\mathcal{Z}\subset L^1_+$ with $\mathbb{E}[Z]=1$.

The Hyperfinite Robust Representation Theorem asserts:

• For any $X\in L^\infty$ (identified with a variable in the Loeb space), every $Z\in {}^*\mathcal{Z}$ is S-integrable;
• The internal support functional

$$\phi_X(Z) = \frac{1}{N} \sum_{k=1}^N [-\tilde{X}(k)] Z(k)$$

(with $\tilde{X}$ an internal representative of $X$) achieves its supremum internally over ${}^*\mathcal{Z}$;

• The risk measure is recovered as the standard part:

$$\rho(X) = \operatorname{st}\left( \sup_{Z\in {}^*\mathcal{Z}} \phi_X(Z) \right).$$

Alternatively, by normalization $a(Z)_k := Z(k)/\sum_{j=1}^N Z(j)$, one can rewrite

$$\rho(X) = \operatorname{st} \left( \sup_{a \in \mathcal{A}_N} \sum_{k=1}^N a_k [-\tilde{X}(k)] \right),$$

where $\mathcal{A}_N$ denotes the internal set of weight vectors arising from normalized dual elements (Kania, 31 Jan 2026).

Analogous formulas pertain to hyperfinite Dirichlet forms and hyperfinite $G$-expectations. In these settings, bilinear forms or sublinear expectations on infinite spaces are represented, up to infinitesimal error, as hyperfinite sums over hyperfinite sets of states, weights, or paths, with transfer and standard-part yielding classical theorems (Anderson et al., 2020, Fadina et al., 2018).

3. Finite Shadows and Recovery of Classical Results

Hyperfinite robust representations serve as lifting theorems: population-level objects or infinite-state expressions are shadows—limits under standard-part—of hyperfinite, combinatorial formulas. Concretely, taking $N=n$ finite, the risk estimator

$$\hat{\rho}_n(x) = \sup_{a\in \mathcal{A}_n} \sum_{i=1}^n a_i (-x_i)$$

is the exact finite-sample dual representation for coordinatewise-coherent, law-invariant estimators, corresponding to the finite shadow of the hyperfinite theorem (Kania, 31 Jan 2026). Similarly, the classical robust representation in Dirichlet forms and comparison theorems on Markov chains are obtained as shadows of their hyperfinite analogues (Anderson et al., 2020).

In $G$-expectation, hyperfinite expectations over internal martingale laws converge (in the sense of being infinitesimally close) to the classical expectation over the weak closure of martingale measures, with the hyperfinite supremum attaining the value (Fadina et al., 2018).

4. Key Corollaries and Probability–Statistics Dictionary

Several corollaries emerge directly from the hyperfinite representation principle:

• Discrete Kusuoka representations: Law-invariant coherent risk estimators on $\mathbb{R}^n$ admit dual forms as suprema over mixtures of discrete expected shortfalls at levels $k/n$ (Kania, 31 Jan 2026).
• Spectral plug-in consistency: Canonical plug-in estimators for spectral risk measures converge almost surely and uniformly over Lipschitz spectral classes, with hyperfinite Glivenko–Cantelli theorems and $L^\infty$-discretization error bounds underpinning the analysis.
• Kusuoka plug-in consistency: Plug-in estimators with generalized sup-of-ES structure converge under tightness and uniform ES estimation on $(\delta, 1]$.
• Bootstrap validity and asymptotic normality: Internal bootstrap resampling defined on hyperfinite samples (using Loeb law empirical distributions) delivers, after standard-part, bootstrap consistency (in Kolmogorov distance), while CLT-type results use hyperfinite versions of SLLN and central limit theorems for L-statistics.

The probability–statistics dictionary formalizes correspondences: | Population CRM on $L^\infty$ | Hyperfinite Support Functional | Finite CRE on $\mathbb{R}^n$ | |------------------------------|-----------------------------------------|----------------------------------| | measure $Q\in \mathcal{Q}$ | weight vector $a\in \mathcal{A}_N$ | probability vector $a\in \Delta_n$ | | $\mathbb{E}_Q[-X]$ | $(1/N) \sum_k a_k (-x_k)$ | $\sum_{i=1}^n a_i(-x_i)$ | | $\sup_{Q\in \mathcal{Q}}$ | $\operatorname{internal~sup}_{a\in \mathcal{A}_N}$ | $\sup_{a\in M_n}$ | | $\rho(X)$ | $\operatorname{st}(\sup ~\text{hyperfinite~sum})$ | $\hat\rho_n(x)$ |

This dictionary makes explicit the operational parallel between dual (robust) population representations, their hyperfinite internal realizations, and classical exact finite-sample formulas (Kania, 31 Jan 2026).

5. Transfer Principle, Error Control, and Proof-Theoretic Power

The transfer principle guarantees critical properties—compactness, attainment of suprema, CLT, SLLN—are mirrored internally on hyperfinite structures, bypassing the need for explicit limit or $\delta$–$\epsilon$ arguments. Errors in hyperfinite sums are infinitesimal and removed upon taking standard-part; approximations and discretizations become mathematically exact in the non-standard framework (Anderson et al., 2020).

This methodology enables direct proof translation from finitistic or combinatorial settings to infinite-dimensional or continuum contexts, including fractal state spaces, measure-valued diffusions, and non-smooth kernels, under minimal regularity (e.g., topological, measure-theoretic conditions). It also allows strong bootstrap and convergence results in plug-in risk estimation (Kania, 31 Jan 2026).

6. Applications and Extensions

Hyperfinite robust representation theorems underlie:

• Unified duality and robust formulae for coherent risk measures, coherent risk estimators, and spectral plug-in procedures (Kania, 31 Jan 2026);
• Discrete-to-continuum correspondence for Dirichlet forms and Markov chain comparison theorems on general state spaces (Anderson et al., 2020);
• Foundations of G-expectation as the standard-part of a maximization over hyperfinite martingale measures, justifying discrete approximations and lifting theorems (Fadina et al., 2018).

The framework extends to a variety of infinite-dimensional or pathwise objects whenever robust or dual representations are available in the classical theory. The main prerequisite is the ability to construct appropriate hyperfinite internalizations (support functionals, weights, sample paths, measures) and to ensure regularity conditions (e.g., S-continuity, S-integrability, boundedness) for transfer and standard-part arguments.

7. Significance and Unifying Perspective

The hyperfinite robust representation theorem enables a transparent and systematic dictionary between probabilistic/integral dual representations and their formally finite, combinatorial analogues in non-standard analysis. It bridges population-level theory and finite-sample procedures, allows direct inheritance of limit theorems and bootstrap results, and streamlines proofs by encoding infinite-dimensional phenomena into hyperfinite, internal matrices and vectors—thus offering both conceptual unity and technical tractability across diverse domains of modern probability and risk theory (Kania, 31 Jan 2026, Anderson et al., 2020, Fadina et al., 2018).