Gaussian characteristic-function martingale from covariance loss (source code)

= Gaussian characteristic-function martingale from covariance loss
{c}
{title2=$\exp(iM_t-V_t/2)$}

Let $M$ be a continuous local <martingale> on $[0,T]$, with deterministic $M_0=m$, and let $V$ be a continuous nonnegative adapted <finite-variation process> with deterministic $V_0=v<\infty$. Suppose $d[M]_t=-dV_t$ and $V_T=0$ <almost surely>. For each real $u$, the <Itô formula> shows
$$
Z_t(u)=\exp(iuM_t-u^2V_t/2)
$$
is a complex local <martingale>: its two finite-variation terms cancel. Its absolute value is at most one, so it is a true bounded <martingale>. Consequently
$$
\mathbb E e^{iuM_T}=e^{ium-u^2v/2},
$$
which identifies the entire <normal distribution> $N(m,v)$ by its <characteristic function>.

For example, $M_t=B_t$ and $V_t=T-t$ give $B_T\sim N(0,T)$. The residual-variance condition $V_T=0$ is essential for this conclusion. With $M_t=0$ and $V_t=1$ throughout, the displayed exponential is still a bounded <martingale>, but its terminal <mean> variable is the constant zero, not $N(0,1)$. In a field-revelation construction, $M_t$ represents a <conditional mean> and $V_t$ the remaining conditional <variance>; only after the latter has vanished does the <martingale> expectation identify the terminal <statistical sample>[sample] law.