---
title: Discrete Kusuoka Representation
url: https://www.emergentmind.com/topics/discrete-kusuoka-representation
type: topic
---

# Discrete Kusuoka Representation

The discrete Kusuoka representation provides a finite or combinatorial analogue of the classical Kusuoka representation in several domains, most notably in coherent risk measurement, self-similar analysis on fractals (such as Sierpiński gaskets), and stochastic analysis (cubature). In all cases, the representation characterizes fundamental functionals—risk, measure, or expectation—as explicit supremums or sums over weighted elementary building blocks, providing insight into their structure and enabling tractable computations in discrete or finite settings.

## 1. Discrete Kusuoka Representations in Coherent Risk Theory

In the context of risk measures, the discrete Kusuoka representation describes any law-invariant coherent risk estimator (CRE) $\rho_n:\mathbb{R}^n\to\mathbb{R}$ on a finite sample as the supremum over mixtures of discrete expected shortfalls (dES). Here, for a vector $x=(x_1,\ldots,x_n)$ of losses or profits and $k=1,\ldots,n$,
\[
\dES_{k/n}(x) = -\frac{1}{k} \sum_{i=1}^k x_{i:n},
\]
where $x_{1:n}\leq \cdots \leq x_{n:n}$ are the order statistics. The main result asserts the existence of a nonempty convex set $\mathcal{M}_n\subset \Delta_n$ of weight-vectors such that
\[
\rho_n(x) = \sup_{\mu\in\mathcal{M}_n}\sum_{k=1}^n \mu_k\,\dES_{k/n}(x),
\]
where $\Delta_n$ is the simplex in $\mathbb{R}^n$ and $\mu_k$ encodes how much tail risk is emphasized at each quantile level. This translates the abstract risk estimator into a worst-case over elementary tail losses, with explicit combinatorial structure suitable for optimization and statistical analysis [2602.00784].

## 2. Discrete Kusuoka on Sierpiński Gaskets and Fractal Energy Measures

For self-similar fractals such as the level-$k$ Sierpiński gasket $SG_k$, the discrete Kusuoka representation specifies the construction of energy measures and Laplacians via finite graph approximations. At each resolution level $m$, the vertices $V_m$ and edges $E_m$ of the approximating graph provide a setting in which energy forms $E_m(u,u)$ and energy measures $v_m$ are defined. Given boundary-normalized harmonic functions $(h_0,h_1,h_2)$, the discrete energy measures $v_i$ assign "energy-mass" to each vertex, and the total Kusuoka measure is $v = v_0 + v_1 + v_2$.

The essential recursion is: for every cell labeled by a word $w$ of length $m$, the energy mass vector $\mu(C)$ transforms under explicit matrices $M_i$:
\[
\mu(F_i C) = M_i\,\mu(C),
\]
with the total measure further obeying a variable weight self-similarity:
\[
v = \sum_{i=1}^d Q_i\,v\circ F_i^{-1},
\]
where $Q_i$ are scalar weights derived from $M_i$. This recursion yields, for each $x\in V_m$,
\[
v_m(\{x\}) = \sum_{|w|=m} Q_w\,\delta_{F_w(q_i)}(x),
\]
with $Q_w$ the word product of weights along the iterated function system (IFS). In the limit $m\to\infty$, the continuous Kusuoka measure emerges, encoding fine-scale energy dissipation and the energy Laplacian [1411.2371].

## 3. Discrete Kusuoka in Financial Super-Replication

Within financial mathematics, the discrete Kusuoka representation characterizes super-replication prices under transaction costs in multivariate, discrete-time models. Given a reference discrete model for $d$ assets, under proportional transaction costs of order $\epsilon_n=1/\sqrt{n}$, the value $V_n(F)$ of super-replicating a European payoff is shown to admit the dual representation:
\[
V_n(F) = \sup_{\mathbb{Q} \in Q_n} \mathbb{E}_{\mathbb{Q}}[F(W_n(S^n))],
\]
where $Q_n$ encodes constraints on consistent (martingale) price system dynamics under admissible volatility bands. The scaling limit of these discrete representations as $n\to\infty$ recovers a $G$-expectation on a set of continuous martingale laws where the volatility matrix lies in a convex set $\Gamma$ parameterized by transaction cost coefficients and simplex structure. This discrete-to-continuum duality is essential for robust pricing and hedging under realistic frictions [1510.04537].

## 4. Atomic (Finite) Space Kusuoka and Minimal Representations

On a genuinely discrete, finite probability space $(\Omega, \mathcal{F}, \mathbb{P})$ with $n$ atoms, any law-invariant coherent risk measure admits a Kusuoka representation via suprema over mixtures of discrete Average-Value-at-Risk (AV@R). Specifically,
\[
\rho(X) = \sup_{\mu\in\mathcal{M}_{\min}}\int_0^1 \mathrm{AV@R}_\alpha(X)\,d\mu(\alpha),
\]
where $\mathcal{M}_{\min}$ is the unique minimal set of measures for which the supremum formula holds, distinguished by extremality under first-order stochastic dominance. The weights for $X = (x_{(1)},...,x_{(n)})$ (ordered values) are explicit functionals of $\mu$, providing a fully combinatorial, non-atomic analogue of the continuous Kusuoka formula. Procedures for pruning $\mathcal{M}$ down to $\mathcal{M}_{\min}$ via stochastic order ensure uniqueness and extremality [1210.7257].

## 5. Discrete Kusuoka Cubature in Stochastic Analysis

Kusuoka's discrete cubature on Wiener space constructs high-order approximations to expectations of SDE functionals via weighted sums over deterministic ODE flows. A cubature formula of order $m$ comprises bounded-variation paths $W^i$ and weights $w_i$ such that expectations involving iterated integrals up to order $m$ are matched. For example:
\[
\mathbb{E}[f(X(t,x))] \approx \sum_{i=1}^N w_i\,f(\Phi^i(t,x)),
\]
where $\Phi^i$ is the ODE flow along $W^i$. The Ninomiya–Victoir (NV) scheme provides a practical, level-5 realization, using sequences of random variables and ODE flows that are semi-explicit in certain stochastic volatility models. These representations allow one to replace stochastic simulations with deterministic, structure-preserving summations, providing both qualitative insight and computational efficiency [1009.4818].

## 6. Matrix-Valued Gibbs Formulation and Fractal Self-Similarity

Kusuoka's measure on self-similar sets admits a fully discrete, matrix-valued Gibbs formulation. For a post-critically finite iterated function system (IFS), the measure is constructed as follows:
- Matrix-valued transfer operator $L_G$ acts on symmetric matrix-valued functions via $D v_i(x)$, the Jacobians of the IFS maps.
- The unique (up to scaling) positive-definite eigenfunction/eigenmeasure pair $(Q,T)$ yields the scalar Kusuoka measure by the trace pairing
  \[
  \mu_K(E) = \int_E \mathrm{tr}[Q(x)\,T(dx)].
  \]
- On each cell of level $n$, matrix products $D_{i_0}\cdots D_{i_{n-1}}$ provide explicit combinatorial weights, leading to dual characterizations as both a Gibbs state and energy measure for Dirichlet-form analysis.

This framework unifies the discrete measure, energy scaling, and spectral properties in a single, matrix-algebraic setting, facilitating explicit computations even for non-affine or non-scalar cases [2005.12028].

## 7. Connections Between Discrete and Continuous Kusuoka Representations

Discrete Kusuoka representations are not merely analogues, but strong finite approximations which, in the limit (e.g., as $n \to \infty$ in risk, or $m\to\infty$ in graphs), recover the classical continuous Kusuoka framework. In all domains—risk, fractals, financial super-replication—the discrete structure makes computation effective, reveals underlying extremal or duality patterns, and provides an entry point for probabilistic, combinatorial, or numerical methods. This suggests discrete Kusuoka theory serves simultaneously as a structural tool for understanding the geometry of coherent functionals and as a constructive method for practical computation in non-continuous settings.

Source: https://www.emergentmind.com/topics/discrete-kusuoka-representation