1 Definition and basic properties

Skorokhod space is a function space designed for paths that may have jumps. It is central in probability theory because many stochastic processes do not have continuous sample paths. The space is usually denoted by \(D([0,T],E)\) or a similar notation, where the functions take values in a metric space \(E\) and are defined on a time interval such as \([0,T]\) or \([0,\infty)\).

The defining feature of Skorokhod space is that its elements are càdlàg functions: they are right-continuous and have left limits at every point in the domain. This makes the space suitable for studying evolving systems whose trajectories can change abruptly while still retaining enough regularity for analysis.

1.1 Càdlàg functions

A càdlàg function has two key properties. First, at each time point, the value from the right agrees with the function value. Second, approaching the point from the left yields a finite limit, even if the function itself may jump at that instant. Such functions naturally model events that occur suddenly, such as arrivals, shocks, or threshold crossings.

The càdlàg condition is weaker than continuity but strong enough to control path behavior. It ensures that jumps are isolated in a precise sense and that the trajectory remains well behaved under many analytic operations.

1.2 The space of paths

The full path space depends on the time interval and the target space. In many applications, the target is the real line or \(\mathbb{R}^d\), but more general metric spaces are also used. The collection of all càdlàg paths forms the underlying set of Skorokhod space, while the topology determines how paths are compared.

This path-space viewpoint allows random processes to be treated as random elements taking values in a function space. As a result, problems about convergence of stochastic processes can be reduced to questions in topology and measure theory.

1.3 Examples and non-examples

Typical examples include step functions, sample paths of jump processes, and trajectories of counting processes. A process that remains constant except for occasional jumps is a standard instance of a càdlàg path.

Non-examples include functions with essential discontinuities, functions lacking left limits at some points, and paths with oscillations that prevent a left limit from existing. Continuous functions are also càdlàg, but they represent only a special subclass.

1.4 Relationship to continuous function spaces

The space of continuous functions is contained in Skorokhod space, but the latter is broader and better suited to jump phenomena. When all relevant paths are continuous, one often works instead with a uniform topology on a continuous path space. Skorokhod space extends this framework by accommodating discontinuities while preserving a usable notion of convergence.

2 Skorokhod topologies

The usefulness of Skorokhod space comes largely from its topologies, which compare functions not only by their values but also by allowing small distortions in time. This feature is essential when jumps in two trajectories occur at nearly the same moments but not exactly simultaneously.

2.1 The J1 topology

The J1 topology is the most widely used Skorokhod topology. It permits time reparameterizations that align jump times, making it well suited for convergence of stochastic processes with isolated discontinuities.

2.1.1 Time-change functions

In the J1 setting, two paths are compared after applying increasing, continuous time-change functions that are close to the identity. These time changes allow one to shift jump locations slightly so that corresponding features of two paths can be matched more effectively.

The admissible transformations are constrained enough to avoid excessive distortion. They preserve the overall chronological order while permitting small local adjustments.

2.1.2 Convergence in the J1 metric

Convergence in the J1 topology requires that both the time change and the path values become close. Intuitively, one path can be stretched or compressed slightly in time so that it nearly overlays the other. This is stronger than pointwise convergence and more flexible than uniform convergence on discontinuous paths.

A sequence may converge in J1 even when jump times do not match exactly, provided the mismatches vanish under appropriate time adjustments. This makes the topology especially effective for limit theorems involving jumps.

2.2 The M1 topology

The M1 topology is another important Skorokhod topology. It is generally weaker than J1 and is particularly useful when paths have multiple jumps that may merge in the limit or when convergence occurs through monotone or cumulative effects.

Instead of matching jumps as precisely as J1 does, M1 allows a broader comparison based on completed graphs of paths. This often yields convergence in settings where J1 convergence fails.

2.3 Comparison of topologies

The J1 and M1 topologies serve different analytical needs. J1 is typically preferred when jumps remain distinct and their timing matters, while M1 is useful when the limit process may arise from the aggregation of smaller discontinuities.

2.3.1 Relative strength and applications

J1 is stronger than M1 in the sense that J1 convergence implies M1 convergence under standard settings, but not conversely. In practice, J1 is common in classical functional limit theorems, whereas M1 appears in applications involving monotone processes, queueing models, and certain sums of jumps.

2.3.2 Topological properties

Both topologies are designed to support probability theory on path spaces, but they differ in the behavior of continuity, compactness, and convergence criteria. The choice of topology affects which mappings are continuous and which process approximations are valid.

3 Metric and topological structure

Skorokhod space is often equipped with a metric that generates the chosen topology. This turns the set of càdlàg functions into a structured metric space, allowing the use of tools from analysis and probability.

3.1 Skorokhod metrics

The standard metrics encode both value differences and time distortions. They compare paths by minimizing over admissible time-change functions, so that nearby trajectories need not coincide at identical time points. This is the defining mechanism behind the topology.

Different variants of the metric correspond to different topologies, especially J1 and M1. Although the formulas are technical, their purpose is to capture convergence of jumps in a stable and flexible way.

3.2 Completeness and separability

Under the usual assumptions, Skorokhod space is complete and separable. Completeness means that Cauchy sequences of paths converge to a limit in the space, while separability means that a countable dense subset exists. These properties are valuable for measure-theoretic arguments and limit theorems.

3.3 Polish space properties

When equipped with the J1 topology on a suitable domain and target space, Skorokhod space is often Polish, meaning complete and separable as a metric space. Polish structure is highly desirable in probability theory because it supports weak convergence, regular conditional probabilities, and standard tightness criteria.

3.4 Compactness criteria

Compactness in Skorokhod space is more subtle than in finite-dimensional spaces. Criteria usually involve control of oscillations, jump behavior, and uniform boundedness on compact time intervals. Such conditions help determine when families of stochastic processes are relatively compact and therefore admit convergent subsequences.

4 Convergence in Skorokhod space

A major reason for using Skorokhod space is to formulate convergence of stochastic processes in a rigorous way. It provides a natural framework for understanding how distributions of trajectories behave in the limit.

4.1 Weak convergence of stochastic processes

Weak convergence in Skorokhod space means convergence in distribution of random paths. Rather than comparing individual sample paths directly, one compares the induced probability laws on the function space. This is the standard language for many functional limit results.

The topology is chosen so that many approximating processes converge even when their jumps are slightly misaligned. That makes the framework much more realistic than a pointwise approach.

4.2 Tightness conditions

Tightness is a key step in proving weak convergence. It ensures that a family of probability measures on Skorokhod space does not escape to infinity and has subsequences that converge in distribution. In practice, tightness is established through bounds on increments, oscillations, and modulus-of-continuity-type estimates adapted to càdlàg paths.

4.3 Continuous mapping theorem

The continuous mapping theorem applies to Skorokhod space when the transformation of paths is continuous with respect to the chosen topology. This principle allows one to deduce convergence of derived stochastic quantities from convergence of the original processes.

Because many functionals of trajectories are discontinuous at certain jump configurations, verifying continuity may require care. When it applies, however, the theorem is a powerful tool for transferring limit results.

4.4 Functional limit theorems

Functional limit theorems describe convergence of entire processes rather than single random variables. Classical examples include convergence of rescaled random walks to Brownian motion or jump processes to Lévy-type limits. Skorokhod space provides the natural environment for such statements, especially when paths are not continuous.

5 Applications in probability theory

Skorokhod space is widely used because it offers a unified language for many stochastic models. Its flexibility makes it especially suitable for systems with sudden changes or cumulative arrival phenomena.

5.1 Markov processes

Many Markov processes have càdlàg sample paths, particularly those with jump transitions. Skorokhod space allows these processes to be studied through their trajectory laws, facilitating analysis of convergence, generators, and long-term behavior.

5.2 Lévy processes

Lévy processes are a central class of stochastic processes with stationary independent increments and càdlàg paths. Skorokhod space is the standard path space for these processes, since their trajectories often include jumps of varying size and frequency.

5.3 Queueing theory

In queueing theory, customer arrivals, service completions, and workload changes are naturally modeled by càdlàg processes. Skorokhod space is used to express convergence of queue length and waiting time processes under scaling limits, especially in heavy-traffic analysis.

5.4 Random walks and invariance principles

Random walks can be embedded into path space and rescaled to study their limiting behavior. Invariance principles often show that a sequence of properly normalized walks converges to a continuous or jump process in Skorokhod space. The topology is essential when discrete jumps persist in the limit or when jump times need to be aligned asymptotically.

Several variants of Skorokhod space are used to adapt the theory to more general settings. These extensions preserve the central idea of working with càdlàg trajectories while broadening the range of possible applications.

6.1 Multidimensional Skorokhod space

When the state space is \(\mathbb{R}^d\), one obtains multidimensional Skorokhod space. This is important for vector-valued processes, where each coordinate may jump independently or jointly. The topology is adapted componentwise while still governing the combined path behavior.

6.2 Infinite-dimensional extensions

Some applications involve paths taking values in function spaces or other infinite-dimensional objects. In such cases, one studies Skorokhod-type spaces with values in Banach or more general metric spaces. These extensions support modern stochastic analysis in areas such as interacting systems and random fields.

6.3 Alternative path-space constructions

Other path-space frameworks include spaces of continuous functions, regulated functions, or paths with weaker regularity conditions. These alternatives may be preferable when continuity, bounded variation, or monotonicity plays a special role. Skorokhod space remains the most standard choice for jump processes because of its balance between flexibility and structure.

6.4 Skorokhod representation theorem

The Skorokhod representation theorem is related in name and theme, though it concerns probabilistic coupling rather than the path space itself. Under suitable conditions, it states that weakly convergent probability measures can be represented on a common probability space so that the corresponding random variables converge almost surely. This result is highly influential in limit theory and often complements work done in Skorokhod space.

7 Historical background

The development of Skorokhod space arose from the need to analyze stochastic processes with discontinuous sample paths. Its history is closely tied to the growth of modern probability theory and the study of functional convergence.

7.1 Anatoliy Skorokhod

Anatoliy Skorokhod was a prominent mathematician whose work had a major impact on probability theory, stochastic processes, and functional limit theorems. The spaces and topologies bearing his name reflect his contributions to the rigorous study of jump processes and convergence of random trajectories.

7.2 Development of path-space methods

Path-space methods emerged as probability theory moved beyond finite-dimensional distributions. Researchers sought ways to describe entire trajectories and their limits, especially for processes with discontinuities. Skorokhod’s ideas helped establish a framework in which random functions could be treated as objects in a metric space.

7.3 Influence on modern stochastic analysis

Skorokhod space has become a standard tool in modern stochastic analysis. It underlies many arguments in weak convergence, stochastic-process approximation, and the study of discontinuous dynamics. Its influence extends across pure and applied probability, where pathwise reasoning is essential for understanding complex random systems.