Doob martingale

In the mathematical theory of probability, a Doob martingale (named after Joseph L. Doob, ^[1] also known as a Levy martingale) is a stochastic process that approximates a given random variable and has the martingale property with respect to the given filtration. It may be thought of as the evolving sequence of best approximations to the random variable based on information accumulated up to a certain time.

When analyzing sums, random walks, or other additive functions of independent random variables, one can often apply the central limit theorem, law of large numbers, Chernoff's inequality, Chebyshev's inequality or similar tools. When analyzing similar objects where the differences are not independent, the main tools are martingales and Azuma's inequality.^{[ clarification needed ]}

Definition

Let ${\displaystyle Y}$ be any random variable with ${\displaystyle \mathbb {E} [|Y|]<\infty }$. Suppose ${\displaystyle \left\{{\mathcal {F}}_{0},{\mathcal {F}}_{1},\dots \right\}}$ is a filtration, i.e. ${\displaystyle {\mathcal {F}}_{s}\subset {\mathcal {F}}_{t}}$ when ${\displaystyle s<t}$. Define

${\displaystyle Z_{t}=\mathbb {E} [Y\mid {\mathcal {F}}_{t}],}$

then ${\displaystyle \left\{Z_{0},Z_{1},\dots \right\}}$ is a martingale, ^[2] namely Doob martingale, with respect to filtration ${\displaystyle \left\{{\mathcal {F}}_{0},{\mathcal {F}}_{1},\dots \right\}}$.

To see this, note that

• ${\displaystyle \mathbb {E} [|Z_{t}|]=\mathbb {E} [|\mathbb {E} [Y\mid {\mathcal {F}}_{t}]|]\leq \mathbb {E} [\mathbb {E} [|Y|\mid {\mathcal {F}}_{t}]]=\mathbb {E} [|Y|]<\infty }$;
• ${\displaystyle \mathbb {E} [Z_{t}\mid {\mathcal {F}}_{t-1}]=\mathbb {E} [\mathbb {E} [Y\mid {\mathcal {F}}_{t}]\mid {\mathcal {F}}_{t-1}]=\mathbb {E} [Y\mid {\mathcal {F}}_{t-1}]=Z_{t-1}}$ as ${\displaystyle {\mathcal {F}}_{t-1}\subset {\mathcal {F}}_{t}}$.

In particular, for any sequence of random variables ${\displaystyle \left\{X_{1},X_{2},\dots ,X_{n}\right\}}$ on probability space ${\displaystyle (\Omega ,{\mathcal {F}},{\text{P}})}$ and function ${\displaystyle f}$ such that ${\displaystyle \mathbb {E} [|f(X_{1},X_{2},\dots ,X_{n})|]<\infty }$, one could choose

${\displaystyle Y:=f(X_{1},X_{2},\dots ,X_{n})}$

and filtration ${\displaystyle \left\{{\mathcal {F}}_{0},{\mathcal {F}}_{1},\dots \right\}}$ such that

${\displaystyle {\begin{aligned}{\mathcal {F}}_{0}&:=\left\{\phi ,\Omega \right\},\\{\mathcal {F}}_{t}&:=\sigma (X_{1},X_{2},\dots ,X_{t}),\forall t\geq 1,\end{aligned}}}$

i.e. ${\displaystyle \sigma }$-algebra generated by ${\displaystyle X_{1},X_{2},\dots ,X_{t}}$. Then, by definition of Doob martingale, process ${\displaystyle \left\{Z_{0},Z_{1},\dots \right\}}$ where

${\displaystyle {\begin{aligned}Z_{0}&:=\mathbb {E} [f(X_{1},X_{2},\dots ,X_{n})\mid {\mathcal {F}}_{0}]=\mathbb {E} [f(X_{1},X_{2},\dots ,X_{n})],\\Z_{t}&:=\mathbb {E} [f(X_{1},X_{2},\dots ,X_{n})\mid {\mathcal {F}}_{t}]=\mathbb {E} [f(X_{1},X_{2},\dots ,X_{n})\mid X_{1},X_{2},\dots ,X_{t}],\forall t\geq 1\end{aligned}}}$

forms a Doob martingale. Note that ${\displaystyle Z_{n}=f(X_{1},X_{2},\dots ,X_{n})}$. This martingale can be used to prove McDiarmid's inequality.

McDiarmid's inequality

Main article: McDiarmid's inequality

The Doob martingale was introduced by Joseph L. Doob in 1940 to establish concentration inequalities such as McDiarmid's inequality, which applies to functions that satisfy a bounded differences property (defined below) when they are evaluated on random independent function arguments.

A function ${\displaystyle f:{\mathcal {X}}_{1}\times {\mathcal {X}}_{2}\times \cdots \times {\mathcal {X}}_{n}\rightarrow \mathbb {R} }$ satisfies the bounded differences property if substituting the value of the ${\displaystyle i}$th coordinate ${\displaystyle x_{i}}$ changes the value of ${\displaystyle f}$ by at most ${\displaystyle c_{i}}$. More formally, if there are constants ${\displaystyle c_{1},c_{2},\dots ,c_{n}}$ such that for all ${\displaystyle i\in [n]}$, and all ${\displaystyle x_{1}\in {\mathcal {X}}_{1},\,x_{2}\in {\mathcal {X}}_{2},\,\ldots ,\,x_{n}\in {\mathcal {X}}_{n}}$,

${\displaystyle \sup _{x_{i}'\in {\mathcal {X}}_{i}}\left|f(x_{1},\dots ,x_{i-1},x_{i},x_{i+1},\ldots ,x_{n})-f(x_{1},\dots ,x_{i-1},x_{i}',x_{i+1},\ldots ,x_{n})\right|\leq c_{i}.}$

McDiarmid's Inequality ^[1]  — Let ${\displaystyle f:{\mathcal {X}}_{1}\times {\mathcal {X}}_{2}\times \cdots \times {\mathcal {X}}_{n}\rightarrow \mathbb {R} }$ satisfy the bounded differences property with bounds ${\displaystyle c_{1},c_{2},\dots ,c_{n}}$.

Consider independent random variables ${\displaystyle X_{1},X_{2},\dots ,X_{n}}$ where ${\displaystyle X_{i}\in {\mathcal {X}}_{i}}$ for all ${\displaystyle i}$. Then, for any ${\displaystyle \varepsilon >0}$,

${\displaystyle {\text{P}}\left(f(X_{1},X_{2},\ldots ,X_{n})-\mathbb {E} [f(X_{1},X_{2},\ldots ,X_{n})]\geq \varepsilon \right)\leq \exp \left(-{\frac {2\varepsilon ^{2}}{\sum _{i=1}^{n}c_{i}^{2}}}\right),}$

${\displaystyle {\text{P}}(f(X_{1},X_{2},\ldots ,X_{n})-\mathbb {E} [f(X_{1},X_{2},\ldots ,X_{n})]\leq -\varepsilon )\leq \exp \left(-{\frac {2\varepsilon ^{2}}{\sum _{i=1}^{n}c_{i}^{2}}}\right),}$

and as an immediate consequence,

${\displaystyle {\text{P}}(|f(X_{1},X_{2},\ldots ,X_{n})-\mathbb {E} [f(X_{1},X_{2},\ldots ,X_{n})]|\geq \varepsilon )\leq 2\exp \left(-{\frac {2\varepsilon ^{2}}{\sum _{i=1}^{n}c_{i}^{2}}}\right).}$

See also

• Catalog of articles in probability theory
• Concentration inequality – a summary of McDiarmid's and several similar inequalities.
• Expected value and Variance
• Fuzzy logic and Fuzzy measure theory
• Glossary of probability and statistics
• List of probability topics
• List of publications in statistics
• List of statistical topics
• Notation in probability
• Probability distribution

References

1. Doob, J. L. (1940). "Regularity properties of certain families of chance variables" (PDF). Transactions of the American Mathematical Society. 47 (3): 455–486. doi: 10.2307/1989964 . JSTOR   1989964.
2. Doob, J. L. (1953). Stochastic processes. Vol. 101. New York: Wiley. p. 293.