Doob–Meyer Decomposition

• Doob–Meyer decomposition is a foundational result in stochastic process theory that expresses a submartingale as the sum of a martingale and a predictable, nondecreasing process.
• It employs discrete approximations and hyperfinite sums to connect finite-time partitions with continuous-time behavior, ensuring mechanical predictability.
• Nonstandard analysis techniques, including the Transfer Principle and Standardization Lemma, streamline proofs and extend applications in stochastic calculus and financial modeling.

The Doob–Meyer decomposition is a foundational result in stochastic process theory, characterizing the internal structure of submartingales and, more generally, semimartingales. Given a filtered probability space and a right-continuous submartingale of "class D," the decomposition asserts that the process can be uniquely expressed as the sum of a martingale and a predictable, nondecreasing process. This characterization underlies much of modern stochastic calculus, mathematical finance, and the theory of stochastic integration.

1. Discrete Approximations and Sum Representations

A central feature of the Doob–Meyer decomposition is its connection to discrete-time analogues. For a finite partition $n = \{0 = t_0, t_1, ..., t_m = 1\}$, one defines the "compensator" process $A_t^n$ by summing conditional expectations of increments:

$$A_t^n \equiv \sum_{t_{k+1} \leq t} E(X_{t_{k+1}} - X_{t_k} \mid \mathcal{F}_{t_k}),$$

where $X_t$ is the submartingale, and $\mathcal{F}_{t_k}$ denotes the filtration at time $t_k$. This finite sum generates a predictable, nondecreasing process when $X_t$ is a submartingale, and uniform integrability is verified via standard arguments such as Markov's inequality.

2. Nonstandard Analysis and Hyperfinite Approximation

To transition from discrete to continuous time, the paper exploits nonstandard analysis—specifically, the use of hyperfinite time grids. In the nonstandard universe, one takes a *-finite set $N = \{0 = t_0, t_1, ..., t_M = 1\}$ and constructs

$$A_t^N \equiv {}^* \sum_{t_{k+1} \leq t} {}^* E({}^* X_{t_{k+1}} - {}^* X_{t_k} \mid {}^* \mathcal{F}_{t_k}),$$

where ${}^* E, {}^* X$, and ${}^* \mathcal{F}$ are the nonstandard extensions. The Transfer Principle ensures that properties verified for finite partitions apply to hyperfinite partitions. Crucially, the Concurrence Principle guarantees coverage of all standard real times. Thus, hyperfinite sums serve as discrete-time surrogates for the continuous-time compensator.

3. Standardization, Predictability, and Construction of the Compensator

The standardization (the "Standardize" lemma) is essential for converting nonstandard objects back to standard processes. Given a nonnegative internally measurable function $X$ in the nonstandard universe, tightness conditions allow for the existence of a unique standard integrable function $X_\infty$ such that, for any standard measurable set $A$,

$E(X_\infty; A) \simeq {}^* E(X; {}^* A),$

where $\simeq$ denotes being infinitely close. Applying this lemma to averages derived from hyperfinite sums, one obtains a standard process $A_t$, predictable, nondecreasing, and starting at 0. The martingale component is defined as

$$M_t = E(X_1 - A_1^\infty \mid \mathcal{F}_t), \qquad A_t = X_t - M_t.$$

Iterations of the standardization ensure that $A_t$ possesses the desired properties.

4. Uniqueness and the Doléans–Dade Theorem

Uniqueness follows from the properties of predictable compensators and martingales. Given two decompositions $X_t = A_t + M_t$ and $X_t = A'_t + M'_t$ (with $M_t$ and $M'_t$ martingales), comparison of the nonstandard expectations and application of the Standardize lemma yield

$$E(A_t; F) \simeq {}^* E(A_t^N; {}^* F) = E(A'_t; F),$$

for any standard $F \in \mathcal{F}_t$. Predictability forces equality almost surely. The Doléans–Dade theorem, included as a corollary, states that a standard integrable nondecreasing process is predictable if and only if it is natural—again proven via this machinery.

5. Advantages and Significance of Nonstandard Techniques

Nonstandard analysis provides several benefits over classical proofs:

• Hyperfinite Discretization: Utilization of hyperfinite partitions enables the reduction of complex continuous-time convergence proofs to finite-sum manipulations, directly extending classical discrete-time techniques.
• Transfer Principle: Ensures that finite arguments are valid in the hyperfinite context, removing the necessity for compactness or deep functional analysis tools (e.g., Dunford–Pettis or Komlós's lemma).
• Concurrence Principle: Guarantees that chosen hyperfinite partitions align with all standard times, enabling seamless transition between nonstandard and standard frameworks.
• Standardization Lemma: Extracts standard processes from nonstandard ones, controlling tightness and integrability and preserving key properties such as predictability. The approach circumvents intricate technical arguments found in classical constructions, instead providing an immediate, conceptually clear route from discrete approximations to the continuous decomposition.

6. Impact on Stochastic Process Theory

By encoding continuous-time processes as hyperfinite discrete approximations, the nonstandard approach offers alternative perspectives on the behavior and structure of submartingales. This methodology suggests potential applications for continuous–discrete links in other domains of stochastic analysis. The proof exemplifies how powerful logical principles from nonstandard analysis can replace elaborate analytical machinery, leading to simplification and new insight.

7. Key Formulas and Structural Summary

Below is a summary table of the main formulas:

Step	Formula	Role
Discrete compensator	$$A_t^n = \sum_{t_{k+1} \leq t} E(X_{t_{k+1}}-X_{t_k}\mid\mathcal{F}_{t_k})$$	Finite partition approximation
Hyperfinite sum	$$A_t^N = {}^* \sum_{t_{k+1} \leq t} {}^* E({}^* X_{t_{k+1}}-{}^* X_{t_k}\mid{}^* \mathcal{F}_{t_k})$$	Nonstandard universe construction
Standardization	$E(A_1^\infty; F) \simeq {}^* E(A_1^N; {}^* F)$	Extraction of standard process
Martingale/Compensator	$$M_t = E(X_1 - A_1^\infty \mid \mathcal{F}_t),\, A_t = X_t - M_t$$	Final decomposition

In formal terms, for any standard submartingale $X_t$ of class D, there exists a unique decomposition

$X_t = A_t + M_t, \quad 0 \leq t \leq 1,$

where $A_t$ is a predictable, nondecreasing, standard process starting at 0, and $M_t$ is a standard martingale.

This streamlined nonstandard proof clarifies the structure of submartingales, highlights the essential role of discrete approximations, and demonstrates the conceptual power of nonstandard analysis in semimartingale theory (Matsunaga, 23 Aug 2025).