Optional Strong Supermartingales

• Optional strong supermartingales are processes that generalize classical supermartingales by accommodating irregularities like discontinuities and non-càdlàg paths.
• The Mertens decomposition extends the Doob-Meyer framework, decomposing these processes into martingale and nondecreasing components critical for stochastic integration and BSDEs.
• Komlós-type limits and Fatou convergence techniques ensure stability and structural analysis, underpinning applications in reflected BSDEs and nonlinear expectation theory.

An optional strong supermartingale is an $\mathbb{R}$-valued process that generalizes the classical notion of supermartingales to the setting of optional processes, particularly accommodating possible discontinuities and non-càdlàg paths. The development of this concept underpins a large class of stochastic processes with irregular sample paths, plays a central role in the theory of stochastic integration, and forms a cornerstone in the analysis of reflected backward stochastic differential equations (RBSDEs), nonlinear expectations, and the characterization of random times such as honest times. Key foundational works include Mertens' extension of the Doob-Meyer decomposition, the generalization to $g$-supermartingale systems, and deep connections with reflected BSDEs and optimal stopping problems.

1. Definitions and Core Properties

An optional strong supermartingale is an $\mathcal{O}$-measurable process $X = (X_t)_{t\ge0}$ such that for any pair of stopping times $\sigma \leq \tau$, the integrability $X_\tau \in L^1$ holds and the strong supermartingale property

$$E[X_\tau \mid \mathcal{F}_\sigma] \le X_\sigma \quad \text{a.s.}$$

is satisfied. The process is said to be of class (D) if $\{ X_\tau : \tau < \infty \}$ is uniformly integrable; of class (DL) if the same holds on each finite time-horizon $[0,T]$ (Czichowsky et al., 2013, Li, 2018, Grigorova et al., 2015).

Optional strong (super)martingales are optional processes not requiring right-continuity, and can possess both left and right jumps. The class (D) condition is essential for uniqueness and the applicability of general decomposition theorems (Grigorova et al., 2015).

2. Mertens Decomposition and Generalizations

The Mertens decomposition is a refinement of the Doob-Meyer decomposition for (optional strong) supermartingales of class (D) that need not be càdlàg (Grigorova et al., 2015): $X_t = M_t - A_t - C_{t-}$ where $g$0 is a càdlàg uniformly integrable martingale, $g$1 is a predictable, right-continuous, nondecreasing process with $g$2, and $g$3 is an adapted, right-continuous, pure-jump, nondecreasing process with $g$4. At each predictable time $g$5, $g$6 and $g$7, enforcing left-continuity at predictable times and right-upper-semicontinuity, respectively (Grigorova et al., 2015).

The decomposition generalizes further: for $g$8-supermartingale systems, the Doob-Meyer-Mertens decomposition provides aggregation and decomposition for $g$9-supermartingales (via BSDEs) even without right-continuity, yielding predictable $\mathcal{O}$0, predictable $\mathcal{O}$1, and an orthogonal martingale $\mathcal{O}$2 in the representation (Bouchard et al., 2015).

3. Limits, Stability, and Fatou Convergence

Given a sequence $\mathcal{O}$3 of nonnegative martingales, there exists a sequence of convex combinations $\mathcal{O}$4 and a limiting process $\mathcal{O}$5 such that $\mathcal{O}$6 in probability for every stopping time $\mathcal{O}$7, with $\mathcal{O}$8 an optional strong supermartingale (Czichowsky et al., 2013). This construction coincides, except on a countable set, with the Fatou limit: $\mathcal{O}$9 with corrections at certain stopping times. No version of almost-sure simultaneous convergence at all stopping times exists (Czichowsky et al., 2013).

Additionally, similar Komlós-type results hold for limits of optional strong supermartingales, their left limits, and their stochastic integrals, with explicit limit descriptions relating $X = (X_t)_{t\ge0}$0 for optional and predictable versions (Czichowsky et al., 2013).

4. Applications to BSDEs and Nonlinear Expectations

In the study of reflected BSDEs with irregular obstacles, optional strong supermartingales provide the natural class for the solution process when the obstacle is only right-upper-semicontinuous. The solution to such an RBSDE can be decomposed as an optional strong supermartingale, and the Mertens decomposition yields the additive structure for the process increments, as well as minimality (Skorokhod) conditions for the increasing parts (Grigorova et al., 2015).

Beyond the classical linear case, the general Doob-Meyer-Mertens decomposition for $X = (X_t)_{t\ge0}$1-supermartingale systems provides a robust tool for BSDEs with constraints or nonlinear drivers, crucial for risk measures and nonlinear expectation theory (Bouchard et al., 2015).

5. Characterization via Drawdown and Honest Times

The technology of optional strong supermartingales underpins the drawdown and multiplicative decompositions central to the theory of Azéma supermartingales and the characterization of honest times (Li, 2018). Any finite honest time $X = (X_t)_{t\ge0}$2 can be associated bijectively with a class of nonnegative local optional supermartingales $X = (X_t)_{t\ge0}$3 with continuous running supremum and $X = (X_t)_{t\ge0}$4: $X = (X_t)_{t\ge0}$5 where $X = (X_t)_{t\ge0}$6 is the Azéma supermartingale (Li, 2018).

This representation admits both additive (drawdown) and multiplicative (relative drawdown) forms and extends to semimartingales of class $X = (X_t)_{t\ge0}$7, accommodating jumps in the finite variation part and allowing for broad generalizations of pricing formulas (e.g., the Madan-Roynette-Yor formula).

6. Structural Features, Semimartingale Classes, and Optional Decompositions

An optional semimartingale $X = (X_t)_{t\ge0}$8 is of class $X = (X_t)_{t\ge0}$9 if $\sigma \leq \tau$0, $\sigma \leq \tau$1 has no right jumps (left-continuous and continuous parts), and the Skorokhod reflection condition holds: $\sigma \leq \tau$2 with $\sigma \leq \tau$3. Such structure enables one to recover broad generalizations in optimal stopping and option pricing, encapsulated in level-set representations and various dualities (Li, 2018).

Additionally, the general Doob-Meyer-Mertens theory under (nonlinear) $\sigma \leq \tau$4-expectations enables optional decompositions even in the presence of portfolio constraints, imposing $\sigma \leq \tau$5 almost everywhere for a process $\sigma \leq \tau$6 integrand with respect to a continuous martingale $\sigma \leq \tau$7 (Bouchard et al., 2015).

7. Summary Table of Key Structural Results

Result/Property	Decomposition	Key Features
Mertens Decomposition	$\sigma \leq \tau$8	$\sigma \leq \tau$9: càdlàg martingale; $X_\tau \in L^1$0: predictable nondecreasing; $X_\tau \in L^1$1: pure-jump, right-continuous, nondecreasing (Grigorova et al., 2015)
$X_\tau \in L^1$2-Doob-Meyer-Mertens	$X_\tau \in L^1$3	$X_\tau \in L^1$4-driver, predictable $X_\tau \in L^1$5, with minimality under $X_\tau \in L^1$6-expectation (Bouchard et al., 2015)
Drawdown/Multiplicative Form	$X_\tau \in L^1$7	$X_\tau \in L^1$8 $X_\tau \in L^1$9; continuous running supremum, drawdown links (Li, 2018)
Fatou Limit Construction	$$E[X_\tau \mid \mathcal{F}_\sigma] \le X_\sigma \quad \text{a.s.}$$0	Coincides with optional strong supermartingale limit (Czichowsky et al., 2013)

The corpus of results on optional strong supermartingales forms a unified theory connecting general process decompositions, stability under limits, nonlinear expectations, reflected BSDEs, and the fine structure of random times in stochastic analysis (Czichowsky et al., 2013, Grigorova et al., 2015, Bouchard et al., 2015, Li, 2018).