Can a continuous-in-probability process jump?

Yes. Random jump times are compatible with the definition, as the uniform-jump example shows.

Does the property require a continuous-time index?

It concerns limits in the index set's topology; time-indexed real intervals are the common setting discussed here.

Does it prove that nearby values always agree?

No. It says that differences exceeding each fixed tolerance become unlikely as the indices approach one another.

AI Glossary / Mathematics and optimization / Advanced

Continuity in probability

Definition

Continuity in probability is a property of a stochastic process: as an index approaches a fixed index, the corresponding random variables approach each other in probability. For a time-indexed real process X, it means P(|Xₛ−Xₜ|>ε) tends to zero as s tends to t, for every ε>0. It is also called stochastic continuity.

What it means

The definition asks whether a noticeable difference becomes unlikely when two specified times become close. It does not require every realized trajectory to be a continuous curve.

This distinction matters because a process can jump at random times while still being continuous in probability at every fixed time. A counting process offers an intuitive example: the probability of an event within a sufficiently short interval can become very small, although any realized event produces a jump.

“Continuous time” is another concept. It describes the set of allowable time indices, not how a process behaves as those indices approach one another.

How it works

Fix a time t and a positive tolerance ε. Examine the probability that Xₛ differs from Xₜ by more than that tolerance. The property holds at t when this probability approaches zero as s approaches t within the index domain. It holds for the process when it holds at every index.

The comparison uses the process's joint construction: Xₛ and Xₜ live on the same probability space. Merely showing that their separate distributions are identical or converge is insufficient. Independence between different times could preserve a substantial probability of disagreement despite unchanged marginal distributions.

Almost-surely continuous paths imply stochastic continuity, but the reverse implication does not hold in general. Stronger increment conditions can support stronger path-regularity results; they must be established separately rather than assumed from this definition.

Three practical examples

1. Arrival modeling: a Poisson count has jumps, but the probability of at least one arrival in a shrinking interval tends to zero, giving stochastic continuity.

2. Sensor modeling: a continuous-path noise model satisfies stochastic continuity, while smoothness or differentiability of paths requires additional properties.

3. Simulation review: analysts distinguish an assumption about short-interval probabilities from an assumption that every simulated trace should have no jumps.

Think of it this way

Think of checking whether a shop's door counter changes between two scheduled observations. As the observations get closer, a change becomes less likely under an ordinary random-arrival model. Yet whenever someone enters, the counter changes abruptly by one.

The analogy assumes a suitable arrival model. A scheduled mass entry at a known time could violate the same conclusion at that time. Stochastic continuity is a mathematical property to verify, not a universal feature of real event streams.

Worked example

Consider an original process on times [0,1]. Choose one random time U uniformly from [0,1], and define Xₜ=0 before U and Xₜ=1 at and after U. Every realized path has a jump at its own random U, apart from irrelevant endpoint qualifications.

Take fixed times t=0.4 and s=0.41. The two process values differ precisely when U lies in (0.4,0.41], an event with probability 0.01. With ε=0.5, P(|Xₛ−Xₜ|>ε)=0.01. At s=0.401 it becomes 0.001.

For any fixed t and nearby s within [0,1], this probability is |s−t| whenever 0<ε<1. It tends to zero. For ε≥1 the strict-exceedance probability is already zero. Hence the process is continuous in probability at every time, despite its jump paths.

The distinction is between choosing a fixed observation time before the random outcome and following the random jump time selected by that outcome. The definition does not state that one realized path must be continuous simultaneously at all times. This exact calculation is derived from the assumed uniform distribution; no simulation was run.

Illustrative case study

Fictional case: a service-operations team models the time at which an order switches from pending to completed. A reviewer sees a stochastic-continuity assumption and suggests smoothing every status trajectory to remove abrupt transitions.

The analyst explains that the status can change discontinuously while the probability of a change during a shrinking interval tends to zero. The team retains a jump-based representation and checks whether the actual process includes predictable batch-completion times. If it does, those scheduled jumps need explicit treatment rather than being hidden by a continuity label.

Common misconceptions and mistakes

Do not confuse stochastic continuity with continuous sample paths. The worked example separates them directly.

Do not infer the property from unchanged marginal distributions. For instance, independent fair binary values at distinct times disagree with probability one half regardless of how close the times are, so their common marginal distribution does not supply stochastic continuity.

Do not replace convergence in probability with almost-sure convergence without justification. Also distinguish a limit statement from a finite-time engineering tolerance: a probability tending to zero does not specify how small it is at the sampling interval your application actually uses.

Business use

The term helps teams specify stochastic models accurately and interpret assumptions in simulation or probabilistic learning research. It is not itself a forecasting algorithm or a guarantee of stable software behavior.

Ask which process, index domain and probability model are being discussed. If short-interval behavior matters operationally, request quantitative bounds or estimates at relevant intervals. Look for deterministic schedule changes, bursty arrivals and measurement artifacts that may conflict with the assumed model. Separate those empirical questions from the correctness of the mathematical definition.

Check your understanding

Why does the example remain stochastically continuous even though every ordinary trajectory has a jump?

Answer: For any fixed time, the probability that the random jump lies between that time and a nearby time shrinks with the interval length. Pathwise continuity is a stronger, different requirement.

Related terms

Stochastic process
Marginal distribution
convergence in probability
sample path
Poisson process
Brownian motion
almost-sure continuity
jump process

Learn next

compare the uniform-jump calculation with the Poisson probability 1−exp(−λh) of an arrival in an interval of length h. Then examine why an event scheduled at one fixed time changes the continuity conclusion there.

Sources

Sources: Harry Sapranidis Mantelos, “Some notes on Poisson processes,” definition of stochastic continuity and Lévy-process context read: https://warwick.ac.uk/fac/sci/maths/people/staff/sapranidis-mantelos/poisson_notes.pdf . The entry concerns convergence in probability of the jointly defined process, not merely convergence of marginal laws. Uniform-jump construction and fictional case are original; probabilities are calculated analytically.

Previous: Continuity correction Next: Continuous analytics