---
title: Jump-Time Averaged State in Stochastic Processes
url: https://www.emergentmind.com/topics/jump-time-averaged-state
type: topic
---

# Jump-Time Averaged State in Stochastic Processes

“Jump-time averaged state” is not a standard term with a single accepted definition across stochastic-process theory. In the averaging-principle literature, the closest rigorous object is usually the **averaged slow dynamics** obtained by freezing the slow variable and averaging the slow drift against the invariant law, or more generally the occupation law, of a fast jump-driven subsystem. In jump-location and empirical-process literature, the nearest object is instead the distribution of the process **evaluated at jump epochs**. In hazard-based and filtering formulations, the same phrase points to hazard-weighted jump-event distributions or to posterior beliefs after jump innovations [1701.07983], [2212.05218], [1807.08068], [1701.01920].

## 1. Terminological scope and principal meanings

In the terminology of "Weak order in averaging principle for stochastic differential equations with jumps" [1701.07983], the object corresponding to a jump-time averaged state is the **averaged dynamics** or **averaged slow motion** of a slow–fast jump-diffusion. There, the averaging is performed by freezing the slow variable, constructing the invariant measure of the fast jump-diffusion, and replacing the slow drift by its invariant-measure average.

Several later formulations make the same conceptual point in different settings. For two time-scale regime-switching processes on a countably infinite state space, the effective state is obtained by averaging the slow drift against the invariant distribution \(\pi^x\) of the fast jump chain conditional on frozen slow state \(x\), and the paper explicitly emphasizes that this is **not** an average over explicit jump times [2212.05218]. For non-autonomous reaction–diffusion equations with Poisson random measures, the averaged object is defined through a **time-dependent evolution family of measures** \(\mu_t^x\) and a long-time Bohr mean, again not through averaging over jump epochs [1807.08068].

A distinct usage appears when the process is sampled **at jump times themselves**. "Jump Locations of Jump-Diffusion Processes with State-Dependent Rates" [1701.01920] makes the jump-epoch viewpoint primary: the central object is the sequence of jump locations \(X_i=X_{t_i}\) and its stationary distribution \(p_\star\). Large-deviation formulations sharpen this distinction by separating the ordinary time-averaged density \(p(x)\) from the empirical distributions \(Q^{-}(x)/n\) and \(Q^{+}(x)/n\) of pre-jump and post-jump states per jump event [2104.10392].

The resulting terminology is therefore context-dependent. A “jump-time averaged state” may mean an invariant-measure average over a fast jump mechanism, an event-sampled jump-location distribution, a hazard-weighted jump-event state law, or a posterior belief updated by jump observations.

## 2. Averaged slow state in slow–fast jump-diffusions

The canonical averaging-principle formulation is the two-time-scale jump-diffusion system
\[
\begin{aligned}
dX_t^\epsilon &= a(X_t^\epsilon,Y_t^\epsilon)\,dt+b(X_t^\epsilon)\,dB_t+c(X_{t-}^\epsilon)\,dP_t,\qquad X_0^\epsilon=x, \\
dY_t^\epsilon &= \frac1\epsilon f(X_t^\epsilon,Y_t^\epsilon)\,dt+\frac1{\sqrt\epsilon}g(X_t^\epsilon,Y_t^\epsilon)\,dW_t+h(X_{t-}^\epsilon,Y_{t-}^\epsilon)\,dN_t^\epsilon,\qquad Y_0^\epsilon=y,
\end{aligned}
\]
where \(P_t\) has intensity \(\lambda_1\) and \(N_t^\epsilon\) has intensity \(\lambda_2/\epsilon\) [1701.07983]. The slow variable \(X^\epsilon\) carries drift, diffusion, and jumps, while the fast variable \(Y^\epsilon\) is accelerated in drift, diffusion, and jump intensity.

The standing assumptions are \(C^2\)-regularity with bounded first and second derivatives, boundedness of \(a,b,c\), nondegeneracy of the fast diffusion, and a dissipativity condition that includes the jump coefficient \(h\). Under these hypotheses, the frozen fast motion
\[
dY_t^x(y)=f(x,Y_t^x(y))\,dt+g(x,Y_t^x(y))\,dW_t+h(x,Y_{t-}^x(y))\,dN_t,\qquad Y_0^x=y,
\]
has a unique invariant measure \(\mu^x(dy)\), exponential contraction,
\[
\mathbb{E}\|Y_t^x(y)-Y_t^x(y')\|^2 \le \|y-y'\|^2 e^{-\beta t},
\]
and ergodicity [1701.07983].

The averaged drift is then
\[
\bar a(x):=\int_{\mathbb{R}^m} a(x,y)\,\mu^x(dy)
       =\lim_{t\to+\infty}\mathbb{E}\,a(x,Y_t^x(y)).
\]
This is the precise mathematical content of the jump-time averaged state in this setting: the slow drift is averaged against the equilibrium law of the fast jump-diffusion. The corresponding effective slow equation is
\[
d\bar X_t=\bar a(\bar X_t)\,dt+b(\bar X_t)\,dB_t+c(\bar X_{t-})\,dP_t,\qquad \bar X_0=x.
\]

Two structural points are essential. First, only the **slow drift** is averaged, because in this model \(b\) and \(c\) do not depend on the fast variable. Second, the two jump mechanisms play different roles: the fast jump term \(h\,dN_t^\epsilon\) changes the invariant measure \(\mu^x\) and therefore changes \(\bar a(x)\), whereas the slow jump term \(c\,dP_t\) survives explicitly in the limit equation [1701.07983].

## 3. Generator formulation, Poisson equation, and weak order \(1\)

The weak approximation theory is formulated through infinitesimal generators. For a slow test function \(\Phi\in C_b^2(\mathbb{R}^n,\mathbb{R})\),
\[
\mathcal{L}_1\Phi(x)
= \bigl(a(x,y),D_x\Phi(x)\bigr)_{\mathbb{R}^n}
+\frac12 \operatorname{Tr}\!\left[D_{xx}^2\Phi(x)\cdot b(x)b^T(x)\right]
+\lambda_1\bigl(\Phi(x+c(x))-\Phi(x)\bigr),
\]
while the frozen fast generator is
\[
\mathcal{L}_2\Psi(y)
= \bigl(f(x,y),D_y\Psi(y)\bigr)_{\mathbb{R}^m}
+\frac12 \operatorname{Tr}\!\left[D_{yy}^2\Psi(y)\cdot g(x,y)g^T(x,y)\right]
+\lambda_2\bigl(\Psi(y+h(x,y))-\Psi(y)\bigr).
\]
Hence the full generator is \(\mathcal{L}^\epsilon=\mathcal{L}_1+\epsilon^{-1}\mathcal{L}_2\), and the averaged generator is
\[
\bar{\mathcal L}\Phi(x)
= \bigl(\bar a(x),D_x\Phi(x)\bigr)_{\mathbb{R}^n}
+\frac12 \operatorname{Tr}\!\left[D_{xx}^2\Phi(x)\cdot b(x)b^T(x)\right]
+\lambda_1\bigl(\Phi(x+c(x))-\Phi(x)\bigr).
\]

For \(u^\epsilon(t,x,y)=\mathbb{E}\phi(X_t^\epsilon(x,y))\) with \(\phi\in C_b^3(\mathbb{R}^n,\mathbb{R})\), the asymptotic expansion
\[
u^\epsilon(t,x,y)=u_0(t,x,y)+\epsilon u_1(t,x,y)+r^\epsilon(t,x,y)
\]
leads to
\[
\mathcal L_2 u_0=0,\qquad
\partial_t u_0=\mathcal L_1 u_0+\mathcal L_2 u_1.
\]
Since \(\mathcal L_2u_0=0\), the leading term does not depend on \(y\), and after integrating against \(\mu^x\) one obtains
\[
\partial_t u_0(t,x)=\bar{\mathcal L}u_0(t,x).
\]
Thus \(u_0=\bar u\), where \(\bar u(t,x)=\mathbb E\,\phi(\bar X_t(x))\).

The first correction solves the Poisson equation
\[
\mathcal L_2 u_1(t,x,y)
=\bigl(\bar a(x)-a(x,y),D_x\bar u(t,x)\bigr)_{\mathbb R^n},
\]
with representation
\[
u_1(t,x,y)=\int_0^{+\infty}\mathbb E\,\rho(t,x,Y_s^x(y))\,ds.
\]
This is the first-order weak correction generated by the fast jump-diffusion semigroup.

The main theorem is that for any \(T>0\) and \(\phi\in C_b^3(\mathbb{R}^n,\mathbb{R})\),
\[
\left|\mathbb{E}\phi(X_T^\epsilon(x,y))-\mathbb{E}\phi(\bar X_T(x))\right|
\le C_{T,\phi,x,y}\,\epsilon.
\]
The weak order in the averaging principle is therefore \(1\), while the paper explicitly compares this with strong averaging order \(1/2\), so the weak rate is essentially twice the strong one [1701.07983].

## 4. Generalizations: fast pure-jump switching and non-autonomous jump-driven SPDEs

For fully coupled two-time-scale regime-switching systems, the slow component is a diffusion in \(\mathbb R^d\) and the fast component is a pure jump process on a countably infinite state space \(S=\{1,2,\dots\}\). With frozen slow state \(x\), the fast chain has generator \(Q(x)\) and invariant law \(\pi^x=(\pi_i^x)_{i\in S}\), and the averaged drift is
\[
\bar b(x)=\sum_{i\in S} b(x,i)\,\pi_i^x.
\]
The limit system is the deterministic ODE
\[
d\bar X_t=b(\bar X_t)\,dt,\qquad \bar X_0=x_0,
\]
because the diffusion coefficient is multiplied by \(\sqrt{\varepsilon}\) and vanishes in the limit [2212.05218].

In that framework, the decisive issue is the regularity of \(x\mapsto \pi^x\). Under uniform strong ergodicity,
\[
|\pi^x-\pi^y|_{\mathrm{var}} \le C_\pi |x-y|,
\]
so the averaged drift is Lipschitz and the slow process converges in \(L^1\) to the unique averaged limit. Under weaker ergodicity,
\[
|\pi^x-\pi^y|_{\mathrm{var}} \le K_4 |x-y|^{1/2},
\]
the averaged drift is only Hölder continuous; the limit ODE still admits a solution, but uniqueness may fail, and only weak convergence is obtained in general [2212.05218]. This makes explicit that the averaged state may exist while the averaged trajectory is not uniquely determined.

For non-autonomous stochastic reaction–diffusion equations with Poisson random measures, the autonomous invariant-measure picture is replaced by a **time-dependent evolution family of measures** \((\mu_t^x)_{t\in\mathbb R}\) associated with the frozen fast equation. The basic averaged coefficient is
\[
\Phi(t,x):=\int_{\mathbb H} B_1(t,x,y)\,\mu_t^x(dy),
\]
and the effective drift is the Bohr mean
\[
\bar B_1(x):= \lim_{T\to\infty}\frac1T\int_0^T \int_{\mathbb H} B_1(t,x,y)\,\mu_t^x(dy)\,dt.
\]
The averaged equation keeps the slow Wiener and Poisson terms explicitly:
\[
du(t)=\big[A_1(t)u(t)+\bar B_1(u(t))\big]dt
+F_1(t,u(t))\,dw^{Q_1}(t)
+\int_{\mathbb Z}G_1(t,u(t),z)\,\tilde N_1(dt,dz).
\]
Only the coupling through the fast variable is averaged out [1807.08068].

## 5. Event-sampled states at jump epochs

A different and literal meaning arises when the state is sampled **at jump times**. In the framework of jump-diffusions with state-dependent jump intensity \(\lambda(x)\), jump times are \(t_i\) and jump locations are
\[
X_i:=X_{t_i}.
\]
Their densities \(p_i\) evolve through an absorbing-process construction, and the stationary jump-location distribution \(p_\star\) is characterized by
\[
0 = L u_\star - \lambda u_\star + J \lambda u_\star,\qquad p_\star=\lambda u_\star,
\]
with normalization
\[
\int_{-\infty}^{\infty} u_\star(x)\lambda(x)\,dx=1.
\]
If \(\hat p_s\) is the stationary density of the full reinjected process, then
\[
p_\star(x)=\frac{\lambda(x)\hat p_s(x)}{\int \lambda(y)\hat p_s(y)\,dy}.
\]
Hence the stationary distribution of jump locations is the same as the stationary distribution of the full process **if and only if** \(\lambda\) is constant [1701.01920]. This is the basic event-bias relation for jump-time sampling.

Large-deviation theory makes the same distinction at the empirical level. The ordinary time-averaged density is
\[
p(x)=\frac{1}{T}\int_0^T dt\, \delta(x(t)-x),
\]
whereas the empirical jump-flow is
\[
Q(x,y)=\frac{1}{T}\sum_{t:\,x(t^+)\neq x(t^-)}
\delta(x(t^+)-x)\,\delta(x(t^-)-y).
\]
Its marginals
\[
Q^{-}(y)=\int dx\,Q(x,y),\qquad Q^{+}(x)=\int dy\,Q(x,y)
\]
are the empirical densities of pre-jump and post-jump states per unit physical time, and
\[
n=\int dx\,dy\,Q(x,y)
\]
is the jump density. The normalized jump-epoch distributions are therefore \(Q^{-}(x)/n\) and \(Q^{+}(x)/n\) [2104.10392]. In this sense, a jump-time averaged state is not \(p(x)\) but the appropriate marginal of \(Q\).

The same structure appears in linear-response theory for Markov jump processes. The paper on martingale-based linear response treats jump-additive functionals
\[
\sum_{s\in(0,t]}\alpha(s,X_{s-},X_s),
\]
and the jump count \(N_t\) is obtained by taking \(\alpha\equiv 1\). This provides the numerator and denominator needed for jump-count-normalized empirical state averages, even though the normalized ratio itself is not the primary object of that work [2201.02982].

## 6. Hazard-weighted jump-event states and filtered post-jump beliefs

In state-dependent hazard models, the natural jump-time state law is neither an invariant-measure average nor a stationary jump-flow marginal. For the Ornstein–Uhlenbeck internal-state model
\[
dX_t=-\epsilon X_t\,dt+\sqrt2\,dB_t,
\]
with jump time
\[
T=\inf\left\{t:\int_0^t \Lambda(X_s)\,ds>\Gamma\right\},
\qquad \Gamma\sim\exp(1),
\]
the density \(f(x,t)\) of still-alive particles satisfies
\[
\frac{\partial f}{\partial t}
-\epsilon\frac{\partial}{\partial x}(xf)
-\frac{\partial^2 f}{\partial x^2}
=-\Lambda(x)f.
\]
The waiting-time density is
\[
n(t)=\int_{-\infty}^{\infty}\Lambda(x)f(x,t)\,dx,
\]
so the joint jump-event density in state–time space is
\[
j(x,t)=\Lambda(x)f(x,t).
\]
Consequently,
\[
\mathbb{P}(X_T\in dx\mid T=t)
=\frac{\Lambda(x)f(x,t)}{n(t)}\,dx,
\qquad
\mathbb E[X_T\mid T=t]
=\frac{\int x\,\Lambda(x)f(x,t)\,dx}
{\int \Lambda(x)f(x,t)\,dx}.
\]
This is the natural jump-time averaged state in a hazard-driven setting [2303.06910].

In partially observed jump-diffusion systems, the nearest analogue is not a state value but a **belief distribution**. The filtered conditional law is
\[
\pi_t(d\theta)=\mathbb P(\Theta_t\in d\theta\mid\mathcal F_t^X),
\]
and the low-dimensional state summary exposed to the forecast model is
\[
\beta_t=\int \varphi(\theta)\,\pi_t(d\theta).
\]
The encoder performs a Strang-splitting update
\[
q_{k+1}\approx
\mathcal C_h\circ\mathcal B_h\circ\mathcal A_{\Delta t}
\circ\mathcal B_h\circ\mathcal C_h(q_k),
\]
where \(\mathcal C_h\) is the **jump innovation**. In that framework, the closest rigorous object to a jump-time averaged state is the posterior belief, or the belief feature \(\beta_t\), after the jump-innovation update [2605.24548].

A persistent misconception is therefore that jump-time averaging always means averaging over discrete jump epochs. The literature shows four non-equivalent constructions: invariant-measure averaging of a fast jump mechanism, occupation-measure averaging for fast switching, empirical state distributions at jump events, and hazard- or observation-weighted post-jump laws. What they share is not a common formula, but a common structural role: they replace the full jump-resolved dynamics by an effective state description adapted either to fast jump clocks or to event-conditioned sampling.

Source: https://www.emergentmind.com/topics/jump-time-averaged-state