In this chapter we revise some basic concepts from stochastic analysis. We begin with the properties of stochastic processes before defining the stochastic integral in the sense of Itô. The third section is concerned with the chain rule in stochastic integration which is known as Itô's formula. Finally we present classical as well as more recent results on stochastic ordinary differential equations. These will be used in the finite dimensional approximation of stochastic PDEs in chapter 10.
Stochastic analysis; Stochastic processes; Stochastic integration; Stochastic differential equations; Itô's formula
We consider random variables on a probability space . Let be a filtration such that for . A real-valued stochastic process is a set of random variables on with values in . A stochastic process can be interpreted as a function of t and ω, where t can be interpreted as time. For fixed the mapping is called path or trajectory of X. We follow the presentation from [100] where the interested reader may also find details of the proofs.
The most important process is the Wiener process.
The aim of this section is to define stochastic integrals of the form
Here M is a square integrable martingale, X a stochastic process and . Throughout the section we assume that -a.s. Moreover, we suppose that M is a quadratically integrable -adapted martingale where is a filtration which satisfies the usual conditions (see Definition 8.1.6). A process could be of unbounded variation in any finite subinterval of . Hence integrals of the form (8.2.1) cannot be defined pointwise in . However, M has finite quadratic variation given by the continuous and increasing process (see Theorem 8.1.31). Due to this fact, the stochastic integral can be defined with respect to continuous integrable martingales M for an appropriate class of integrands X.
The definition of the stochastic integral goes back to Itô. He studied the case where M is a Wiener process. His students Kunita and Watanabe considered the general case . In the following we have a look at the class of integrands which are allowed in (8.2.1). We define a measure on by
We call two -adapted stochastic processes and equivalent with respect to M, if -a.e. This leads to the following equivalence relation: for a -adapted process X we define
provided the right-hand-side exists. So is the -norm of X as a function of with respect to the measure . We define the equivalence relation
Our definition of the stochastic integral will imply that and coincide provided X and Y are equivalent.
For we define as the space of processes with for all and a.e. and set
A process can be identified with a process only defined on . In particular we have that is a closed subspace of the Hilbert space
In order to define the stochastic integral for we have to approximate the elements of in an appropriate way by step processes, i.e. by processes in . This can be done thanks to the following theorem.
One of the most important tools in stochastic analysis is Itô's Lemma. It is a chain-rule for paths of stochastic processes. In contrast to the deterministic case it can only be interpreted as an integral equation because the stochastic processes we are interested in (for instance the Wiener process) are in general not differentiable.
In this section we are concerned with stochastic differential equations. We seek a real-valued process on a probability space with filtration such that
which holds true -a.s. and for all . Here W is a Wiener process with respect to . The functions are assumed to be continuous. As in Remark 8.3.20, equation (8.4.11)1 is only an abbreviation for the integral equation
There are two different concepts of solutions to (8.4.12).
i) We talk about strong solution (in the probabilistic sense) if the solution exists on a given probability space with a given Wiener process W. A strong solution exists for a given initial datum and there holds a.s.
ii) We talk about weak solution (in the probabilistic sense) or martingale solution if there is a probability space and a Wiener process such that (8.4.12) holds true. The solution is usually written as
This means that when seeking a weak solution, constructing the probability space (and the Wiener process on it) is part of the problem. A solution typically exists for a given initial law and we have . Even if an initial datum is given it might live on a different probability space. Hence and can only coincide in law.
The existence of a strong solution in the sense of Theorem 8.4.35 is classical, see e.g. [12] and [82,83]. If the assumptions on the coefficients are weakened, strong solutions might not exist, see [17]. In this case we can only hope for a weak solution. We refer to [95] for a nice proof and further references.
The stochastic ODEs which appear later all have strong solutions. However, the concept of martingale solutions will be important for the SPDEs.
The stochastic ODEs we considered so far have two drawbacks. First, we need vector valued processes and, secondly, we have to weaken the assumptions on the drift μ (Lipschitz-continuity in X and linear growth is too strong). Everything in this chapter can be obviously extended to the multi-dimensional setting. Here a standard Wiener process in is a vector valued stochastic process and each of its components is a real valued Wiener process (recall Definition 8.1.4). Moreover, the components are independent. Getting rid of the assumed Lipschitz continuity is more difficult. Now seek a -valued process on a probability space with filtration such that
Here W is a standard -valued Wiener process with respect to and is some initial datum. The functions
are continuous in for each fixed , . Moreover, they are assumed to be progressively measurable. The application in Chapter 10 requires weaker assumptions as in the classical existence theorems mentioned above. In our application we only have local Lipschitz continuity of μ. Fortunately, some more recent results apply. In the following we state the assumptions which are in fact a special case of the assumptions in [124, Thm. 3.1.1.].
(A1) We assume that the following integrability condition on μ for all
(A2) μ is weakly coercive, i.e. for all we have that
where is -adapted.
(A3) μ is locally weakly monotone, i.e. for all and all the following holds
(A4) Σ is Lipschitz continuous, i.e. for all and all the following holds
3.129.194.106