1 Introduction to Sample Paths
A sample path is one complete realization of a stochastic process—an entire timeline of values produced when the underlying randomness is fixed. If a process describes how a system behaves “in general,” a sample path records what happens in one particular experiment.
1.1 Intuition: one realization vs. the whole process
A stochastic process specifies a family of random variables indexed by time (or another parameter). Each random variable is uncertain, but the process also captures structure across indices (for example, correlations or regularity). A sample path is what those random variables become once the random outcome is realized. Conceptually, it is the difference between describing “typical behavior” and observing “this particular outcome.”
1.2 Formal definition for stochastic processes
Let \((X_t)_{t\in I}\) be a stochastic process, where \(I\) is an index set (commonly time). A sample path can be viewed as a function \(t\mapsto X_t(\omega)\) for some outcome \(\omega\) in the underlying probability space. Thus a path is not merely a sequence of observations; it is the induced function over the whole index set (or over the portion considered).
1.3 Sample path vs. sample point vs. trajectory
Terminology varies across fields. In probability theory, sample point often refers to an individual outcome \(\omega\) in the probability space. A trajectory is frequently used in physics, finance, and dynamical systems to mean the realized evolution of a state variable; it matches the idea of a sample path. In stochastic analysis, “sample path” emphasizes the function-valued nature of the realization, while “trajectory” may be used more informally for continuous evolutions.
1.4 Index sets and time parameterization
Paths depend on how the process is indexed. Time can be discrete (e.g., \(t=0,1,2,\dots\)) or continuous (e.g., \(t\ge 0\)). In addition, index sets need not be literal time intervals; they can be multidimensional parameters or subsets of \(\mathbb{R}\). The parameterization affects both the description of regularity (such as continuity) and the definitions of path-based events.
2 Construction and Representation
Sample paths arise from combining a stochastic process with a particular element of its probability space. The main construction issues are identifying the randomness source, defining which functions represent the outcome, and ensuring the relevant mappings are measurable.
2.1 Underlying probability space and randomness source
A stochastic process is defined on a probability space \((\Omega,\mathcal{F},\mathbb{P})\). For each index \(t\), the map \(\omega\mapsto X_t(\omega)\) is measurable, ensuring that events such as \(\{X_t\in A\}\) are well-defined.
2.1.1 Events generated by a path
Once \(\omega\) is fixed, the path \(t\mapsto X_t(\omega)\) determines many events. For example, the event “the path stays above zero on \([0,T]\)” can be written as \(\{X_t>0\ \text{for all } t\in[0,T]\}\), which is a subset of \(\Omega\). Such events are then assigned probabilities under \(\mathbb{P}\).
2.1.2 Measurability of the mapping “outcome → path”
To treat sample paths as random elements in a function space, one considers the mapping \(\Phi:\Omega\to \mathsf{E}\) defined by \(\Phi(\omega)= (t\mapsto X_t(\omega))\), where \(\mathsf{E}\) is a suitable space of functions. Measurability of \(\Phi\) ensures that probabilities of sets of paths are meaningful. The choice of \(\mathsf{E}\) and its \(\sigma\)-algebra (often generated by evaluation maps) is crucial for rigorous statements.
2.2 Functions and notation for paths
A common notation treats a sample path as a function \(x(\cdot)\) satisfying \(x(t)=X_t(\omega)\). When the process is clear from context, writers may suppress \(\omega\) and simply denote a realization by \((X_t)_{t\in I}\) “on that outcome.” In formal developments, one distinguishes carefully between the random function and its realization.
2.3 Discrete-time sample paths
In discrete time, an index set might be \(I=\mathbb{N}\) or \(\{0,1,\dots\}\). A sample path is then typically a sequence \((X_0(\omega),X_1(\omega),\dots)\). Regularity questions translate into properties of sequences, such as boundedness, hitting a level, or eventual behavior.
2.4 Continuous-time sample paths
In continuous time, a path is a function on an interval. Since functions may fail to be continuous, one often classifies realizations into function spaces that permit the appropriate singularities. For example, càdlàg paths allow right-continuity with left limits, capturing jump behavior common in many stochastic models.
3 Path Properties and Regularity
Many of the deepest ideas in stochastic processes depend on the fine structure of sample paths. Regularity properties are used to define events, establish convergence, and compare different models.
3.1 Continuity, càdlàg, and càglàd paths
A path’s regularity determines what kinds of events are stable under limits and what can be inferred from approximations.
3.1.1 Definitions and common conventions
A function \(x(t)\) is continuous if it has no discontinuities on its domain. A càdlàg function is right-continuous and has left limits at each time; càglàd is left-continuous with right limits. These conventions reflect different typical behaviors, such as jumps occurring in one direction relative to the time point.
3.1.2 Visualizing jump behavior
Discontinuities in a path correspond to sudden changes in the process. In càdlàg paths, jumps appear as “vertical gaps” where the value at time \(t\) matches the right-hand limit, while the left-hand behavior approaches a possibly different number. This viewpoint is often used when interpreting processes with events like arrivals or switching regimes.
3.2 Differentiability and absolute continuity (where applicable)
Differentiability is rarer in general stochastic processes, particularly those with jumps or rough sample paths. In models where the paths are sufficiently regular (for instance, certain diffusion settings under stronger assumptions), one may discuss almost sure differentiability or absolute continuity, often linked to the existence of densities for increments over time.
3.3 Boundedness and growth conditions
Pathwise growth conditions describe how large the process gets over time. Examples include almost sure boundedness on compact intervals, sublinear growth as time increases, or moment-based growth constraints. These properties are essential when studying maxima, escape times, and long-run behavior.
3.4 Hitting, crossing, and staying events on a path
Path-based events include:
- Hitting: the path reaches a particular level at some time.
- Crossing: the path transitions from below a level to above it (or vice versa).
- Staying: the path remains above (or below) a threshold over a time interval.
Such events depend sensitively on regularity—especially at points where jumps occur.
4 Sample Path Events and Almost Sure Statements
A major theme in probability theory is the relationship between events defined via entire paths and statements that hold with probability one.
4.1 Tail events and long-run behavior
Tail events are events determined by the process’s behavior “far out” in the index set, typically as time tends to infinity. Path language is natural here: for instance, “eventually the path stays above a line” depends on the asymptotic portion of the trajectory. Tail events often interact with zero-one laws and ergodic-type principles.
4.2 Almost sure properties phrased via paths
An almost sure property means it holds for all outcomes except a set of probability zero. When phrased in terms of paths, it becomes a statement like: “with probability one, the realized path is càdlàg” or “with probability one, the path satisfies a particular boundary behavior.” The exceptional trajectories form a null set of outcomes.
4.3 Sets of paths and probability of path classes
Instead of describing events at fixed times, one can consider collections of paths, such as:
- all paths with finitely many jumps on \([0,T]\),
- paths whose running maximum stays below a threshold,
- paths belonging to a specific regularity class.
These path classes correspond to measurable subsets of the function space where paths live, allowing probability statements about whole trajectories.
4.4 Null sets of paths and exceptional trajectories
Even when a property fails for some trajectories, it may still hold almost surely if the failing set is negligible under the governing probability measure. In practice, visualizations can misleadingly suggest that exceptions are common; almost sure statements clarify that exceptions may exist but are rare in a probabilistic sense.
5 Examples of Sample Paths
Examples illustrate how sample paths behave under different stochastic mechanisms and regularity regimes.
5.1 Deterministic processes (degenerate sample paths)
A deterministic process is a special case where all randomness is absent: \(X_t(\omega)\) is the same for every \(\omega\). Then every sample path coincides, producing a degenerate situation where pathwise variability disappears. This serves as a baseline for understanding how randomness introduces multiplicity of trajectories.
5.2 Random walks and their realizations
A random walk defines increments that are independent and identically distributed over discrete times. Each realization is a discrete sequence that moves by random steps. Sample-path questions include recurrence versus transience, hitting levels, and fluctuations of the partial sums.
5.3 Poisson process paths
For a Poisson process, sample paths are step functions that increase by jumps of size one at random times. On any finite interval, only finitely many jumps occur, and the waiting times between jumps are random. The path perspective emphasizes the event structure: arrivals are visible as upward discontinuities.
5.4 Brownian motion sample paths
Brownian motion has continuous sample paths with highly irregular behavior. While it is continuous everywhere, it is almost surely nowhere differentiable and exhibits fractal-like fluctuations. Many classical properties—such as the distribution of maxima or hitting probabilities—are derived from the path structure rather than from any single time point.
5.5 Martingale sample path intuition (high level)
Martingales capture “fair game” behavior across time in expectation, but their sample paths can still fluctuate dramatically. Intuitively, a martingale’s conditional mean does not drift, yet the path can wander widely. Path-based results often connect martingales to stopping times, maximal inequalities, and almost sure convergence.
6 Distributions Induced by Sample Paths
To study sample paths probabilistically beyond the level of individual events, one introduces measures on spaces of paths.
6.1 Pushforward measures on path space
Given a measurable mapping \(\Phi(\omega)\) from outcomes to paths, the pushforward measure describes the induced distribution on path space. If \(\mathbb{P}\) governs \(\omega\), then the distribution of paths is \(\mathbb{P}\circ \Phi^{-1}\). This formalizes the idea that “the process is a random element” of a function space.
6.2 Finite-dimensional distributions vs. path distribution
A process is determined by its finite-dimensional distributions (joint laws of \((X_{t_1},\dots,X_{t_n})\)) under appropriate conditions, but not always in complete generality. The full path distribution retains information about the entire trajectory, including dependence across times that cannot be reconstructed from finitely many observations in some settings.
6.3 Consistency and projective perspectives
Finite-dimensional distributions must satisfy consistency: marginals of higher-dimensional distributions agree with those of lower-dimensional ones. This compatibility underpins projective constructions and allows one to study whether a consistent family of distributions can extend to a distribution on a full path space.
6.4 Path space viewpoints (function space formulations)
Function space formulations treat sample paths as elements of spaces such as \(C([0,T])\), \(D([0,T])\) for càdlàg functions, or other Banach/metric structures. Choosing the topology or metric on this space influences what it means for processes to converge and which limiting results apply.
7 Convergence and Limits via Paths
Convergence in stochastic processes is naturally expressed in terms of the behavior of sample paths, both at fixed times and across the entire timeline.
7.1 Convergence at fixed times (finite-dimensional convergence)
Finite-dimensional convergence means that for any finite set of times, the joint distributions of the process values converge. This captures agreement at those time points but does not automatically control how paths behave between them.
7.2 Convergence of processes in path-space metrics
To obtain convergence that respects whole trajectories, one uses a metric or topology on path space and studies convergence of random elements there.
7.2.1 Uniform-on-compact convergence intuition
Uniform-on-compact convergence focuses on the maximum deviation over time intervals \([0,T]\). When paths are continuous, this notion often aligns well with classical uniform convergence and yields strong control over the entire sample trajectory on finite horizons.
7.2.2 Skorokhod-type notions for jump processes
For jump processes, uniform convergence can be too strict because small time shifts can cause large discrepancies at discontinuities. Skorokhod-type approaches allow time changes that align jumps more effectively, producing a topology suited to càdlàg paths.
7.3 Almost sure convergence along sample paths
Almost sure convergence requires that, on nearly every outcome, the realized paths converge pointwise or under a chosen topology. This is stronger than convergence in distribution and is useful when constructing strong approximations or proving stability of pathwise algorithms.
7.4 Tightness and why it matters for path limits
Tightness is a condition ensuring that a family of process laws does not “escape” to infinity in path space. Intuitively, it guarantees relative compactness, enabling extraction of convergent subsequences. Tightness is a key ingredient in functional central limit theorems and invariance principles.
8 Markovian and Dependence Viewpoints
Sample paths also clarify dependence structure, especially for Markov processes where future behavior depends on the present state and sometimes the path history in a limited way.
8.1 Markov property expressed through path segments
The Markov property states that, given the present state, the future is independent of the past. In path terms, this can be expressed via segments of the trajectory: the distribution of the post-\(t\) evolution depends only on \(X_t\), not on the earlier values.
8.2 Stopping times as path-dependent random times
A stopping time is a random time determined by information available up to that time. In a path view, it is defined by a rule that inspects the realized trajectory and decides when a condition is met (for example, when the path first crosses a threshold).
8.3 Filtrations generated by sample paths
The filtration records which events are observable as time progresses. When defined from a process, it is generated by the information carried by the sample path up to each index. This connects pathwise information flow to martingale and Markov properties.
8.4 Strong vs. weak formulations (path-related intuition)
Different formulations of a stochastic model may emphasize either:
- the path constructed on a probability space with specific driving randomness (strong viewpoint), or
- the distributional behavior of the process (weak viewpoint).
Path language helps distinguish whether one is concerned with almost sure realization under a particular construction or with matching distributions in law.
9 Practical Computation and Simulation
Even though sample paths are mathematical objects, computation relies on approximations, numerical schemes, and empirical diagnostics based on simulated trajectories.
9.1 Simulating discrete-time sample paths
Discrete-time simulations generate sample paths directly by iterating the update rule. Random increments are sampled from the prescribed distribution, producing many trajectories whose empirical frequencies estimate probabilities of path events.
9.2 Generating continuous-time approximations
For continuous-time processes, exact simulation may be possible for certain models, but often one uses time discretization. Approximations produce piecewise-constant or piecewise-linear paths, and analysis focuses on whether these approximations converge to the true sample path distribution under a chosen topology.
9.3 Estimating path functionals (e.g., maxima)
Many quantities depend on the entire path, such as running maxima, cumulative sums, occupation times, or first-passage times. Simulation estimates these path functionals by applying the functional to each generated trajectory and aggregating the results statistically.
9.4 Visual diagnostics and uncertainty from finite samples
Plots of simulated paths help assess qualitative features like jump frequency or roughness, but they do not by themselves quantify uncertainty. Confidence intervals and variance estimates are needed because any finite set of trajectories provides only an approximation of the underlying law.
10 Common Pitfalls and Clarifications
Several misunderstandings arise when moving between the general process viewpoint and the single-trajectory viewpoint.
10.1 Confusing a process with a single trajectory
A stochastic process describes a family of random variables; it is not identical to any one realized path. Statements about a process typically mean they hold in probability, in distribution, or almost surely across outcomes, not that every drawn path behaves the same way.
10.2 Measurability and “path events” misunderstandings
When events refer to whole trajectories, measurability is not automatic. Rigorous work requires specifying the path space and ensuring the event corresponds to a measurable subset. Intuition can fail when the path space or topology is not chosen carefully.
10.3 Dependence on the choice of index set
Properties can change when the index set is altered. Sampling a process only at discrete times may hide behavior occurring between observations, and different index choices affect definitions of continuity, jumps, and convergence.
10.4 Interpreting “almost surely” when visualizing paths
Seeing a property fail once in a few plots can be misleading, since “almost surely” does not mean “always.” A null set of exceptional trajectories may still appear in small simulations due to randomness in the sampling of outcomes.