Abstract
We study a stochastic velocity tracking problem for the 2D-Navier–Stokes equations perturbed by a multiplicative Gaussian noise. From a physical point of view, the control acts through a boundary injection/suction device with uncertainty, modeled by stochastic non-homogeneous Navier-slip boundary conditions. We show the existence and uniqueness of the solution to the state equation, and prove the existence of an optimal solution to the control problem.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Optimization problems of evolutionary systems under uncertainty are of major relevance in technology, sciences and everyday life. The initial step in addressing the problem relies on its mathematical formulation. Once the problem has been formulated, appropriate theoretical arguments and computational methods are applied to study the existence and uniqueness of the solution, which will be ultimately determined or approximated.
In the literature, a wide range of optimization problems has been addressed.
We would like to mention the contribution [30] in applying the Multivariate Adaptive Regression Splines (MARS) model to systems of discretized stochastic differential equations (describing behavioral reactions and human sentiments), in order to estimate the relevant parameters in the equations that fit real-world historical data (see also [47]). Later [38], the authors successful apply the Robust Optimization algorithms including parametric uncertainties in MARS models, solving real market problems.
Furthermore, we must say that a strategic combination of statistical regression and optimization methods can impact knowledge in Medical Sciences with strong consequences in real life, referring to the Newroscience application, we mention [45] and references therein.
Here we focus on a so-called continuous-time stochastic optimal control problem, which is based on a dynamic law (typically a stochastic differential equation) that governs the temporal evolution of the state process y that we aim to control. Then practical control actions on the physical system are modeled by incorporating the appropriate control variable u into the dynamic law. A key step in analyzing the control problem is to show that an imput u will produce a single output \(y^u\); in mathematical words this step corresponds to showing the well-posedness of the state equation. The goal or desired performance corresponds to maximizing (or minimizing) an objective (cost) functional \(J=J(t, u,y^u)\).
Important applications of optimal control theory can also be found in mathematical finance [26, 36, 48]. Namely to solve investor allocation problems or Merton problems; let us mention for instance the recent work [21] (and references therein) in the context of continuous-time stochastic volatility, and [4, 40] where the authors consider adjusted to reality Hybrid Models with jumps and regime switches. We also address to the reader the paper [5] concerning the optimal management of defined contribution pension funds under the effect of inflation, mortality and uncertainty. In this framework, the investor or the manager can be seen as a player who chooses the optimal strategy to obtain the best profit. More envolved optimization problems arise when multiple players (cooperating or competing with each other) choose their own strategies to achieve their own optimal individual profits. Game theory provides the appropriate framework to tackle such complex problems (see for instance [37, 39, 41, 42]).
It is worth to mention that continuous-time optimal control problems can be classified according to the control type: feedback or not feedback. A feedback control reads \(u(t)=U(t,y(t))\), where U(t, y) is a policy function defined on the state space. The theory for finding an optimal policy function relies on studying the Hamilton-Jacobi-Bellman (HJB) equation. If the state space is \({\mathbb {R}}^d\), the HJB equation is a deterministic nonlinear partial differential equation or integro-partial differential equation on \({\mathbb {R}}^d\). When the evolutionary law is a stochastic partial differential equation, the state space is infinite dimensional, and the study of a feedback control problem is a delicate issue. Even the definition of the HJB equation is far from being well understood. We refer to [10, 22, 34], where the authors proved the existence of \(\epsilon \)-optimal feedback controls.
We must emphasize that the difficulty of solving an optimal control problem feedback/not feedback strongly depends on the structure of the stochastic differential equation, which can be linear or nonlinear, finite or infinite dimensional. Furthermore, the noise can be of finite or infinite dimensional, continuous in time or with jumps, requiring a specific analysis accordingly.
In the present article, the evolutionary law is a nonlinear stochastic partial differential equation that describes the evolution of a velocity field of a Newtonian fluid under the action of stochastic random forces and filling a two-dimensional bounded domain \({ {\mathcal {O}}}\subset {\mathbb {R}}^{2}\). The control actions are imposed on the boundary of the container and are not of feedback type. We aim to show the well-posedness of the state equation and the existence of the optimal solution. In forthcoming works we intend to apply Pontryagin’s maximum principle to deduce first-order optimality conditions and second-order sufficient optimality conditions.
More precisely, we consider the Navier–Stokes equations with non-homogeneous Navier-slip boundary conditions
where \({\textbf{y}}={\textbf{y}}(t,{\textbf{x}})\) is the 2D-velocity random field, \(\pi =\pi (t,{\textbf{x}})\) is the pressure, \(\nu >0\) is the viscosity and \({\textbf{y}}_{0}\) is the initial condition that verifies
Here
is the rate-of-strain tensor; \({\textbf{n}}\) is the external unit normal to the boundary \(\Gamma \in C^{2}\) of the domain \({{\mathcal {O}}}\) and \(\varvec{\tau } \) is the tangent unit vector to \(\Gamma ,\) such that \(({\textbf{n}},\varvec{\tau } ) \) forms a standard orientation in \({\mathbb {R}}^{2}.\) The positive constant \(\alpha \) is the so-called friction coefficient. The quantity a corresponds to the inflow and outflow fluid through \(\Gamma \), satisfying the compatibility condition
This condition means that the quantity of inflow fluid should coincide with the quantity of outflow fluid. The boundary functions a and b will be considered as the control variables for the physical system (1.1). The term \({\textbf{G}}(t,{\textbf{y}})\,{ {\mathcal {W}}}_{t}\) is a multiplicative white noise.
The main goal of this paper is to control the solution of the system (1.1) by the boundary condition (a, b), which is a predictable stochastic process belonging to the space \({\mathcal {A}}\) of admissible controls to be defined in Sect. 4. The cost functional is given by
where \({\textbf{y}}_{d}\in L_{2}(\Omega \times {{\mathcal {O}}}_{T})\) is a desired target field and \(\lambda _{1},\lambda _{2}>0.\) We aim to control the random velocity field \({\textbf{y}}\), defined as the solution of the Stochastic Navier–Stokes equations, through minimization of the cost functional (1.4). More precisely, our goal is to solve the following problem
Let us mention that boundary control of fluid flows is of main importance in several branches of the industry, for instance, in the aviation industry. Extensive research has been carried out concerning the implementation of injection-suction devices to control the motion of the fluids (see [8, 9]). On the other hand, rotating flow is critically important across a wide range of scientific, engineering and product applications, providing design and modeling capability for diverse products such as jet engines, pumps, food production and vacuum cleaners, as well as geophysical flows. The control problem for deterministic Newtonian and non-Newtonian flows, has been widely studied in the literature (see [13,14,15, 23, 28, 29, 44]). However, it is well known that the study of turbulent flows, where small random disturbances produce strong macroscopic effects, requires a statistical approach. Recently, special attention has been devoted to stochastic optimal control problems, where control is exerted by a distributed mechanical force (see [10, 16, 22, 43]). To the best of our knowledge, this is the first paper where the boundary control problem is addressed for stochastic Navier–Stokes equations under Navier-slip boundary conditions.
The plan of the present paper is as follows. In Sect. 2, we present the general setting, by introducing the appropriate functional spaces and some necessary classical inequalities. Section 3 deals with the well-posedness of the state equations. In Sect. 4, we show the existence of an optimal solution to the control problem.
2 General Setting
Let X be a real Banach space endowed with the norm \(\left\| \cdot \right\| _{X}.\) We denote \(L_{p}(0,T;X)\) as the space of X-valued measurable \(p-\)integrable functions defined on [0, T] for \(p\geqslant 1\).
For \(p,r\geqslant 1\), let \(L_{p}(\Omega ,L_{r}(0,T;X))\) be the space of the processes \({\textbf{v}}={\textbf{v}}(\omega ,t)\) with values in X defined on \( \ \Omega \times [0,T],\) adapted to the filtration \(\left\{ {\mathcal {F}} _{t}\right\} _{t\in [0,T]}\), and endowed with the norms
and
where \({\mathbb {E}}\) is the mathematical expectation with respect to the probability measure P. As usual, in the notation for processes \({\textbf{v}}= {\textbf{v}}(\omega ,t)\), we generally omit the dependence on \(\omega \in \Omega .\)
We define the spaces
We denote \((\cdot ,\cdot )\) as the inner product in \(L_{2}({\mathcal {O}})\) and \(\Vert \cdot \Vert _{2}\) as the associated norm. The norms in the spaces \( L_{p}({\mathcal {O}})\) and \(H^{p}({\mathcal {O}})\) are denoted by \(\Vert \cdot \Vert _{p}\) and \(\Vert \cdot \Vert _{H^{p}}\). On the space V, we consider the following inner product
and the corresponding norm \(\Vert {\textbf{v}}\Vert _{V}=\sqrt{\left( \textbf{v },{\textbf{v}}\right) _{V}}.\)
Throughout the article, we often use the continuous embedding results
Let us introduce the notation
We notice that for any vector \({\textbf{v}}\in V\) we have \({\textbf{v}}_{ {\mathcal {O}}}=0,\) since
Using it and the results that can be found on the p. 62, 69 of [32], p. 125 of [35], and on the p. 16–20 of [46], we formulate the next lemma.‘
Lemma 2.1
For any \({\textbf{v}}\in H^{1}({\mathcal {O}})\) and any \(q\geqslant 2\), the Gagliano-Nirenberg-Sobolev inequality
and the trace interpolation inequality
are valid. Moreover, any \({\textbf{v}}\in V\) satisfies Korn’s inequality
that is the norms \(\Vert \cdot \Vert _{H^{1}}\) and \(\Vert \cdot \Vert _{V}\) are equivalent.
Remark 2.2
We should mention that throughout the article, we will represent by C a generic constant that can assume different values from line to line. These constants C will depend mainly of the physical constants \(\nu ,\) \(\alpha , \) the domain \({\mathcal {O}}\), a given time \(T>0\).
Now, we state a formula that can be derived easily via integration by parts
which holds for any \({\textbf{v}}\in H^{2}({\mathcal {O}})\) and \({\textbf{z}}\in V\). Let us assume that \({\textbf{v}}\) satisfies Navier-slip boundary condition (1.1), then we have
In what follows we will frequently use
that is a particular case of Young’s inequality
For a vector
we introduce the norm and the absolute value of the inner product of \( {\textbf{h}}\) with a fixed \({\textbf{v}}\in H\) as
Assume that the stochastic noise is represented by
where \({\textbf{G}}(t,{\textbf{y}})=(G^{1}(t,{\textbf{y}}),\dots ,G^{m}(t,\textbf{y }))\) has suitable growth assumptions, as defined in the following, and \({ {\mathcal {W}}}_{t}=({{\mathcal {W}}}_{t}^{1},\dots ,{{\mathcal {W}}}_{t}^{m})\) is a standard \({\mathbb {R}}^{m}\)-valued Wiener process defined on a complete probability space \((\Omega ,{\mathcal {F}},P)\) endowed with a filtration \( \left\{ {\mathcal {F}}_{t}\right\} _{t\in [0,T]}\). We assume that \( {\mathcal {F}}_{0}\) contains every P-null subset of \(\Omega \).
Let \({\textbf{G}}(t,{\textbf{y}}):[0,T]\times H\rightarrow H^{m}\) be Lipschitz on \({\textbf{y}}\) and satisfy the linear growth
for some positive constant K.
Let us define the space of functions \({\mathcal {H}}_{p}(\Gamma )=\left\{ (a,b):||(a,b)||_{{\mathcal {H}}_{p}(\Gamma )}<+\infty \right\} \) with the norm
In this work, we consider the data a, b and \({\textbf{u}}_{0}\) belong to the following Banach spaces
In addition, we assume that (a, b) is a pair of predictable stochastic processes.
3 State Equation
This section is devoted to the study of the state equation. We use the variatioal approach to show the existence and the uniqueness of solution, and deduce appropriete estimates to study the control problem.
Since, we are considering non-homogeneous boundary conditions, we first introduce a suitable change of variables based on the solution of the non-homogeneous linear Stokes equation, which allows to write the state in terms of a vector field satisfying a homogeneous Navier-slip boundary condition.
Lemma 3.1
Let (a, b) be a given pair of functions satisfying (2.12). Then there exists a unique solution
of the Stokes problem with the non-homogeneous Navier-slip boundary condition
such that
In particular, we have
Proof
Let us introduce the function \({\textbf{c}}=\nabla h,\) where h is the unique solution of the system
with \(\int _{\Gamma }h\,\,d\mathbf {\gamma }=0.\) Theorem 1.10, p. 15 in [27] implies that the function \({\textbf{c}}\) satisfies the estimates
where the constant \(C_{p}\) depends on p, \(2<p<\infty \).
Let us consider the following Stokes problem
with \(\widetilde{b}=b-\left[ 2D(\mathbf {{\textbf{c}}})\,{\textbf{n}}+\alpha \mathbf {{\textbf{c}}}\right] \cdot {\varvec{\tau }\in W_{p}^{-\frac{1}{p}}(\Gamma ) }\) by (3.4) and Lemma 2.4 in [2]. Using Theorem 2.1 in [1], we have that there exists a unique solution \({\textbf{b}}\) of this Stokes problem such that
Due to the regularity (2.12) and the estimates (3.4)–(3.5), we conclude that the system (3.2) has the unique solution \(\mathbf { a=b+c}\), satisfying the first estimate in (3.3). The second one in (3.3) is a direct consequence of the embeedings \(W_{2}^{1}(0,T) \hookrightarrow C([0,T])\) and \(W_{p}^{1}({{\mathcal {O}}})\hookrightarrow C( \overline{{{\mathcal {O}}}}),\) since \(2<p<+\infty .\) \(\square \)
With the help of the solution of the non-homogeneous Stokes equation, we introduce the notion of solution to the state system (1.1).
Definition 3.2
Let the data (a, b) and \({\textbf{u}}_{0}\) satisfy the regularity (2.12), and \({\textbf{a}}\) be the corresponding solution of (3.2). A stochastic process \({\textbf{y}}={\textbf{u}}+{\textbf{a}}\) with \( {\textbf{u}}\in C([0,T];H)\cap L_{2}(0,T;V),\quad P\)-a.e. in \(\Omega ,\) is a strong (in the stochastic sense) solution of (1.1) with \({\textbf{y}} _{0}={\textbf{u}}_{0}+{\textbf{a}}(0)\) if P-a.e. in \(\Omega \) the following equation holds
where the stochastic integral is defined by
The existence of solution for the system (1.1)–(1.2) will be shown by Galerkin’s method. Since the injection operator \(I:V\rightarrow H\) is a compact operator, there exists a basis \(\{{\textbf{e}}_{i}\}\subset V\) of eigenfunctions verifying the property
which is an orthonormal basis for H, and the corresponding sequence \( \{\lambda _{i}\}\) of eigenvalues verifies \(\lambda _{i}>0\), \(\forall i\in {\mathbb {N}}\) and \(\lambda _{i}\rightarrow \infty \) as \(i\rightarrow \infty .\) For the details we refer to Theorem 1, p. 355, of [25]. Moreover the ellipticity of the equation (3.7) and the regularity \(\Gamma \in C^{2}\) imply that \(\{{\textbf{e}}_{i}\}\subset C^{2}({{\mathcal {O}}})\cap V\).
For any fixed \(n\in {\mathbb {N}}\), we consider the subspace \(V_{n}=\textrm{span }\,\{{\textbf{e}}_{1},\ldots ,{\textbf{e}}_{n}\}\) of V. Taking into account the relation (2.7), the approximate finite dimensional problem is: for\(\ P\)-a.e. in \(\Omega \,\) to find \({\textbf{y}}_{n}\) in the form
as the solution of the following finite dimensional stochastic differential equation
where \({\textbf{u}}_{n,0}=\sum _{j=1}^{n}\left( {\textbf{u}}_{0},{\textbf{e}} _{j}\right) {\textbf{e}}_{j}\) is the orthogonal projection of \({\textbf{u}} _{0}\in H\) into the space \(V_{n}.\) From the Parseval’s identity we infer that
The equation (3.8) defines a system of n stochastic ordinary differential equations with locally Lipschitz nonlinearities. Hence, there exists a local-in-time adapted solution \({\textbf{u}}_{n}\in C([0,T_{n}];V_{n}) \) by classical results [31]. The next lemma will establish uniform estimates, which guarantee that \({\textbf{u}}_{n}\) is a global-in-time solution.
Lemma 3.3
Let the data (a, b) and \({\textbf{u}}_{0}\) satisfy the regularity (2.12). Then the system (3.8) has a solution \( {\textbf{y}}_{n}={\textbf{u}}_{n}+{\textbf{a}}\), such that
Moreover, there exists a positive constant \(C_{0}\), such that for the function
and any \(t\in [0,T]\), the following estimate holds
where
and the positive constants \(C_{0}\) and C are independent of the parameter n, which may depend on the regularity of the boundary \(\Gamma \) and the physical constants \(\nu \) and \(\alpha \).
Proof
Let \(\xi _{0}\) be the function defined by (3.10) with a constant \(C_{0}\) to be concretized later on (see expression (3.15) below). For each \(n\in {\mathbb {N}}\), let us set
and consider the sequence \(\{\tau _{N}^{n}\}\)\(_{{N\in {\mathbb {N}}}}\) of the stopping times defined by
Taking \(\varvec{\varphi }={\textbf{e}}_{i}\) for each \(i=1,\dots ,n\) in the equation (3.8) and using \({\textbf{y}}_{n}={\textbf{u}}_{n}+{\textbf{a}}\), we obtain
Step 1. Estimate in the space H up to \(\tau _{N}^{n}\). The Itô formula gives
where the absolute value in the last term is defined by (2.10). Summing these equalities over \(i=1,\dots ,n,\) we obtain
Considering Young’s inequality (2.8) for an appropriate \(\varepsilon >0\), the inequalities (2.3)–(2.5) and the regularities (3.1), (3.3), we estimate the terms \(I_{1},\) \(I_{2}\) and \( I_{3}\). Namely
where A is defined by (3.12). A similar reasoning gives
and
where we used the assumption (2.11). Gathering the previous estimates, we obtain the existence of a positive constant \(C_{0},\) such that
Taking the function \(\xi _{0}\) as in (3.10), thanks to (3.14)- (3.15), the application of Itô’s formula yields
Therefore, we can write
Now, considering the sequence \((\tau _{N}^{n})\) of the stopping times introduced in (3.13) and using (2.11), the Burkholder-Davis-Gundy inequality gives
For \(t\in [0,T]\), we first take the supremun of the relation (3.16) for \(s\in [0,\tau _{N}^{n}\wedge t]\), next we take the expectation and incorporate the previous estimate of the stochastic term with \(\varepsilon =\frac{1}{2}\). Then considering (3.9), we deduce
Hence, the function
fulfills the Gronwall type inequality
which implies
Step 2. The limit transition as \(N\rightarrow \infty \). From (3.17) we have
for some constant C independent of N and n. Let us fix \(n\in {\mathbb {N}} \). Since \({\textbf{u}}_{n}\in C([0,T_{n}];V_{n}),\) we have \(\ g(\tau _{N}^{n})\geqslant N\) and
which implies that \(P\left( \tau _{N}^{n}<T\right) \rightarrow 0\) as \( N\rightarrow \infty .\) This means that \(\tau _{N}^{n}\rightarrow T\) in probability as \(N\rightarrow \infty \). Then, there exists a subsequence \( \{\tau _{N_{k}}^{n}\}\) of \(\{\tau _{N}^{n}\}\) (which may depend on n) such that
Since \(\tau _{N_{k}}^{n}\leqslant T_{n}\leqslant T\), we deduce that \(T_{n}=T\), hence \({\textbf{y}}_{n}={\textbf{u}}_{n}+{\textbf{a}}\) is a global-in-time solution of the stochastic differential equation (3.8). In addition, for each fixed \(n\in {\mathbb {N}}\), the sequence \(\left\{ \tau _{N}^{n}\right\} \) is monotone on N, therefore we can apply the monotone convergence theorem in order to pass to the limit in the inequality (3.17) as \(N\rightarrow \infty \), thereby deducing the estimate (3.11). \(\square \)
In the next lemma, by assuming a better integrability for the initial data, we improve the integrability properties for the solution \({\textbf{y}}_{n}\) of problem (3.8).
Lemma 3.4
Let the data (a, b) and \({\textbf{u}}_{0}\) satisfy the regularity (2.12). In addition we assume
Then, the solution \({\textbf{y}}_{n}={\textbf{u}}_{n}+{\textbf{a}}\) of problem (3.8) has the regularity
such that
where the function \(\xi _{0}\) is defined in (3.10),
and C is a positive constant, being independent of n.
Proof
Taking the square on both sides of the inequality (3.16) and the supremum on \(s\in [0,\tau _{N}^{n}\wedge t]\) with \(\tau _{N}^{n}\) defined by (3.13), we infer that
where B is defined by (3.21). Therefore taking the expectation in this inequality and applying the Burkholder-Davis-Gundy inequality
we obtain
Using Gronwall’s inequality, we deduce that
Arguing as in the proof of Lemma 3.3, there exists a monotone subsequence \(\{\tau _{N_{k}}^{n}\}\) of \(\{\tau _{N}^{n}\}\), which converges to T a.e. \(\omega \in \Omega \), as \(k\rightarrow \infty \). Thus, applying the monotone convergence theorem, we can pass to the limit in (3.22) as \(k\rightarrow \infty \), in order to deduce the estimate (3.20). \(\square \)
Theorem 3.5
Let the data (a, b) and \({\textbf{u}}_{0}\) satisfy the regularity (2.12) and (3.19). Then there exists, a unique strong solution \({\textbf{y}}={\textbf{u}}+{\textbf{a}}\) to the system (1.1)– (1.2), such that
and for any \(t\in [0,T],\) the following estimates hold
where the functions \(\xi _{0}\) and A, B are defined by (3.10) and (3.12), (3.21), respectively. Here C is a positive constant that is independent of n.
Proof
The proof is splitted into three steps.
Step 1. Convergence related to the projection operator. Let \( P_{n}:V\rightarrow V_{n}\) be the orthogonal projection defined by
where \(\{\widetilde{{\textbf{e}}}_{j}=\frac{1}{\sqrt{\lambda _{j}}}{\textbf{e}} _{j}\}_{j=1}^{\infty }\) is the orthonormal basis of V. By Parseval’s identity, for any \({\textbf{v}}\in V\) we have
Considering an arbitrary \({\textbf{z}}\in L_{s}(\Omega \times (0,T);V)\) for some \(s\geqslant 1\), we have
which are valid P-a.e. \(\omega \in \Omega \) and a.e. \(t\in (0,T).\) Hence, Lebesgue’s dominated convergence theorem implies that for any \({\textbf{z}}\in L_{s}(\Omega \times (0,T);V)\), we have
Step 2. Passage to the limit in the weak sense.
Let us define\(~f_{0}(t)=C_{0}\left( ||(a,b)||_{{\mathcal {H}}_{p}(\Gamma )}^{2}+1\right) \). Since
by (2.12), there exists a positive constant \(K(\omega ),\) which depends only on \(\omega \in \Omega \backslash A\) and satisfies
The estimates (3.11) and (3.20) give that
for some constant C that is independent of the index n. These uniform estimates imply
where \(V^{\prime }\) denotes the topological dual of the space V. The uniform estimates (3.28) ensures the existence of a suitable subsequence \({\textbf{u}}_{n}\), which is indexed by the same index n to simplify the notation, and a function \({\textbf{u}},\) such that
Moreover, we have
by (3.26). The limit function \({\textbf{u}}\) satisfies the estimates (3.23), (3.24) by the lower semicontinuity of integral in \(L_{2}\) and \(L_{4}\) spaces.
Considering (2.11) and (3.29), there exist some operators \(B^{*}(t)\) and \({\textbf{G}}^{*}(t)\) such that
Since \({\textbf{y}}_{n}\) solves the equation (3.8), then using Itô’s formula, we infer that
that is, the following integral equation holds
Denoting
the following differential equation holds
We notice that due to the properties of the stochastic integral and the assumption (2.11), we have
Now, we pass to the limit in the equation (3.33) in the distributional sense. Namely multiplying the equation (3.34) by the test funtion \( \theta (t)\eta (\omega )\), with \(\theta \in C^{\infty }([0,T])\) with compact support and \(\eta \in L_{2}(\Omega )\), and passing to the limit, we derive
Therefore \(\frac{\partial {\textbf{h}}}{\partial t}\in L_{2}(\Omega \times (0,T);\left( H^{1}({\mathcal {O}})\right) ^{*})\). Since \({\textbf{h}}\in L_{2}(\Omega \times (0,T);H^{1}({\mathcal {O}}))\), we infer that \({\textbf{h}}\in L_{2}(\Omega ;C([0,T];L_{2}({\mathcal {O}}))\) by the Aubin-Lions embeeding result [3, 46]. Taking into account the continuity property of the stochastic integral, we conclude that \(\xi _{0}^{2}{\textbf{y}}\in L_{2}(\Omega ;C([0,T];L_{2}({\mathcal {O}})).\) In addition
where the index \(\omega \) means that we are considering \(L_{2}(\Omega )\times L_{2}({\mathcal {O}})\) endowed with the weak topology. Hence, we have
Now, we multiply the equation (3.33) by an arbitrary \(\eta \in L_{2}(\Omega )\) and take the expectation, we derive
Applying (3.30)–(3.32) and (3.35), we pass to the limit \( n\rightarrow \infty \) in this equality and deduce
Since \(\eta \in L_{2}(\Omega )\) is arbitrary, the following equation holds
for any \(t\in [0,T]\) and P-a.e. in \(\Omega ,\) that is
Moreover if we use Itô’s formula
we derive that the limit function \({\textbf{y}}\) in the form \({\textbf{y}}= {\textbf{u}}+{\textbf{a}}\) with
satisfies P-a.e. in \(\Omega \) the stochastic differential equation
and \({\textbf{y}}(0)={\textbf{y}}_{0}.\)
Step 3. Deduction of strong convergence as \(n\rightarrow \infty .\) In order to prove that the limit process \({\textbf{y}}\) satisfies the equation (3.6), we adapt the methods in [10]. Writing \( {\textbf{y}}={\textbf{u}}+{\textbf{a}}\), \({\textbf{y}}_{n}={\textbf{u}}_{n}+{\textbf{a}}\) and taking the difference of the equations (3.8) and (3.37) with \( \varvec{\varphi }={\textbf{e}}_{i}\in V_{n}\), \(i=1,\ldots ,n\), we deduce
Then the Itô’s formula yields
Summing over \(i=1,\dots ,n,\) we derive
Standard computations give
In addition, using (2.3), (2.4), (3.3) and Theorem 4.47, p. 210, of [24], we show the existence of a constant \(C_{2}\), verifying the relation
On the other hand, Hölder’s inequality gives
and
The last term \(A_{3}\) will be considered later on.
Denoting
we have
The standard relation \(x^{2}=(x-y)^{2}-y^{2}+2xy\;\) allows to write
From (2.11) and (3.25)\(_{1}\), we have
then for the fixed constant \(C_{3}=2K\) it follows that
The positive constants \(C_{2}\) and \(C_{3}\) in (3.40) and (3.44) are independent of n.
We notice that with the help of the convergence results (3.26), (3.30)–(3.32), and performing a suitable limit trfansition in the equation (3.39), as \(n\rightarrow \infty ,\) we can verify that all terms on the right-hand side of the equality (3.39) containing \(P_{n} {\textbf{u}}-{\textbf{u}}\) will vanish; however, terms that contain \(P_{n}{\textbf{u}}-{\textbf{u}}_{n}\) will remain. Fortunately, these terms can be eliminated by introducing the auxiliary function
with \(\widetilde{f}(t)=C_{3}+\max (3C_{0},C_{2})(1+||(a,b)||_{{\mathcal {H}} _{p}(\Gamma )}^{2}+\left\| {\textbf{u}}\right\| _{V}^{2}).\)
Applying Itô’s formula to the equality (3.39) and using the definition (3.45) of \(\widetilde{\xi }\), we obtain
Writing this equation in the integral form, taking the expectation, and applying the estimates (3.40), (3.44), we deduce that
Next, we will show that the right-hand side of this inequality tends to zero as \(n\rightarrow \infty \).
Considering the estimate (3.41) and using \(\widetilde{\xi }\leqslant \xi _{0}^{3}\) on (0, T), then we deduce that
Using (2.3) for \(q=4\), we have
by the estimates (3.23)–(3.24). Applying similar calculations we can show that there exists a constant C, such that
that is
For the term \(J_{2},\) using the estimate (3.42), we can show that
Therefore we get that the terms \(J_{i},\) \(i=1,2,\) converge to zero as \( n\rightarrow \infty \) by (3.31).
The convergences of (3.30) and (3.31) show that
The operator \(\xi _{0}^{2}A_{3}=\xi _{0}^{2}\left( ({\textbf{y}}\cdot \nabla ) {\textbf{y}}-B^{*}\right) \) belongs to \(L_{2}(\Omega \times (0,T);V^{\prime })\) by (3.29) and (3.32), thus
Due to (3.31), we have
Due to the convergence results (3.26), (3.30), (3.31), (3.32) and (3.43), we obtain
that implies
After combining all the convergence results, we obtain the following strong convergences
for \(t\in (0,T)\), which combined with (3.31), imply
In addition, considering (2.11), we conclude
Since \(\widetilde{\xi }\) is strictly positive, we infer that
From (3.32) and (3.47), it follows that \(\widetilde{\xi }(t)( {\textbf{y}}\cdot \nabla ){\textbf{y}}=\widetilde{\xi }(t){\textbf{B}}^{*}(t)\ \) a.e. in \(\Omega \times (0,T),\) that implies
Considering the identities (3.48), ( 3.49), the equation (3.36) reads
The uniqueness of the solution \({\textbf{y}}\) follows from the stability result established in the next theorem. \(\square \)
Let us denote by \(\widehat{\mathbf {\varphi }}=\mathbf {\varphi }_{1}-\mathbf { \varphi }_{2}\) the diference of two given functions \(\mathbf {\varphi }_{1}, \mathbf {\varphi }_{2}.\)
Theorem 3.6
Let us consider \({\textbf{y}}_{1}={\textbf{u}}_{1}+{\textbf{a}}_{1},\)\({\textbf{y}}_{2}={\textbf{u}}_{2}+{\textbf{a}}_{2}\) with
two solutions of (1.1), satisfying the estimates (3.23), (3.24) with two corresponding boundary conditions \(a_{1},\;b_{1}\), \( a_{2},\) \(b_{2}\) and the initial conditions
Then there exist a strictly positive function \(f_{1}(t)\in L_{1}(0,T)\quad P\) -a.e. in \(\Omega ,\) depending only on the data, such that the following estimate
is valid with the function \(\xi _{1}\) defined as
Proof
The proof follows the same reasoning as the proof of Theorem 3.5. \(\square \)
4 Solution to the Control Problem
This section studies the existence of an optimal solution to the optimal control problem \(({\mathcal {P}}).\) We intend to control the solution of the system (1.1) by boundary values (a, b), which belongs to the space \({\mathcal {A}}\) of admissible controls defined as a compact subset of \(L_{2}(\Omega \times (0,T);\mathcal {H }_{p}(\Gamma ))\) verifying an exponential integrability condition. More precisely, we assume that there exists a constant \(\lambda >0\) such that
Remark 4.1
We notice that given a control pair \((a,b)\in {\mathcal {A}}\), the corresponding state \({\textbf{y}}={\textbf{u}}+{\textbf{a}}\) defined as the solution of the state equation (3.6) belongs to \(L_{2}(\Omega \times (0,T)\times {\mathcal {O}})\). Namely, considering the auxiliar function \(\xi _{0}\) introduced in (3.10), and the estimates (3.23), (3.24), Hölder’s inequality gives
Therefore, the cost functional (1.4) is well defined for every \( (a,b)\in {\mathcal {A}}.\)
Now, we write one of the main result of the article, which establishes the existence of a solution for the optimal control problem \(({\mathcal {P}})\).
Theorem 4.1
Assume that (a, b) and \({\textbf{y}}_{0}\) verify the regularity (2.12), (3.19), such that (a, b) belongs to the space \({\mathcal {A}}\). Then there exists at least one solution for the optimal control problem \(({\mathcal {P}}).\)
Proof
Let us consider a minimizing sequence
of the cost functional J, namely
and \({\textbf{y}}_{n}\) is the weak solution of the system (1.1) for the sequence \((a_{n},b_{n})\in {\mathcal {A}}\).
Due to the compactness of \({\mathcal {A}}\), there exists a subsequence, still indexed by n, such that
From Theorem 4.9., p. 94, of [11], there exists a subsequence of \( \left( a_{n},b_{n}\right) \), still denoted by \(\left( a_{n},b_{n}\right) \), and a function \(h\in L_{2}(\Omega \times (0,T))\) such that
Considering the function \(h=h(t)\), let us introduce the following weight
If we replace \(a,b,{\textbf{a}}\) by \(a_{n},b_{n},{\textbf{a}}_{n}\), respectively in the relations (3.2), (3.3), then, taking into account the estimates (3.23), (3.24), we conclude that the sequence \(\textbf{ u}_{n}={\textbf{y}}_{n}-{\textbf{a}}_{n}\), \(n\in {\mathbb {N}},\) satisfies the estimates
for any \(t\in [0,T]\), where the constants C are independent of n. Therefore there exists a subsequence, still indexed by n, such that
In addition, the following uniform estimate holds
Hence there exist operators \(B^{*}\) and \(G^{*}\) such that
Arguments already used in Step 2 of the proof of Theorem 3.5 allow to pass to the limit equation (4.2) in the distributional sense, as \( n\rightarrow \infty \), to obtain
Writing \({\textbf{y}}={\textbf{u}}+{\textbf{a}}\), \({\textbf{y}}_{n}={\textbf{u}}_{n}+ {\textbf{a}}_{n}\) and doing the difference between (4.2) and (4.10) with \(\varvec{\varphi }={\textbf{e}}_{i}\), \(i\in {\mathbb {N}}\), we deduce
which holds for any element of the basis \(\left\{ {\textbf{e}}_{i}\right\} .\)
By applying Itô’s formula, the equation (4.11) gives
Summing over the index \(i\in {\mathbb {N}}\), we derive
We write
With the help of (2.3), (2.4), (3.3) and (4.6), we deduce the following estimates
where the function h in (4.13) is given by (4.4).
Setting
and using the same arguments as in the deductions of (3.44) by taking \( C_{3}=2K\), we infer that
The positive constants \(C_{2}\) and \(C_{3}\) in (4.13) and (4.15) are independent of n, and they may depend on the data.
Let us consider the function
Now, by applying Itô’s formula to the equality (4.12), the definition (4.16) of \(\widehat{\xi }\), we obtain
Therefore, writing the inequality (4.17) in the integral form, taking the expectation, and incorporating the estimates (4.13), (4.15 ), we infer that
In the following, we show that the right-hand side of this inequality tends to zero as \(n\rightarrow \infty \). The Hölder inequality, (2.12), (3.3) and \(\widehat{\xi }\leqslant \xi _{h}^{2}\) yield
Considering the estimate (4.6) and using that \(\widehat{\xi } \leqslant \xi _{h}^{2}\) on (0, T), we deduce that
where we used the following uniform estimates with respect to the parameter n
and
For the term \(J_{2},\) using Hölder’s inequality, (4.1), (4.3) and (4.6), we can show that
Therefore, the terms \(J_{1},\) \(J_{2}\) converge to zero as \(n\rightarrow \infty \).
The convergence (4.7) shows that
The operator \(\xi _{h}^{2}B_{3}=\xi _{h}^{2}\left( ({\textbf{y}}\cdot \nabla ) {\textbf{y}}-B^{*}\right) \) belongs to \(L_{2}(\Omega \times (0,T);{ V^{\prime }})\) by (4.8), thus (4.9) implies
The term
Due to the convergence results (4.9) and (4.14), we obtain
which implies
Gathering the convergence results for \(J_{i}\), \(i=0,\dots ,5\), and passing to the limit in the inequality (4.18), we deduce the following strong convergences
for \(t\in (0,T).\) In addition, we obtain
then
On the other hand, from (4.9) and (4.20), we infer that \( \widehat{\xi }(t)({\textbf{y}}\cdot \nabla ){\textbf{y}}=\widehat{\xi }(t) {\textbf{B}}^{*}(t)\) a.e. in \(\Omega \times (0,T),\) that implies
Considering the identifications (4.21)–(4.22), the equation (4.10) reads
for any \(\varvec{\varphi }\in V.\) Therefore \({\textbf{y}}\) is the solution of the state equation, corresponding to the control pair (a, b).
Taking into account the lower semicontinuity of the cost functional, the strong convergence (4.20) and Remark 4.1, we infer that
which implies
hence the triplet \((a,b,{\textbf{y}})\) is a solution to the control problem \(( {\mathcal {P}})\). \(\square \)
5 Conclusion and Discussion
This work adresses an optimal control problem for the evolution of a viscous incompressible Newtonian fluid filling a two-dimensional bounded domain, under the action of random forces modeled by a multiplicative Gaussian noise. We prove the existence and uniqueness of the solution to the stochastic state equation and establish the existence of an optimal control. The control is exerted at the boundary through the physical non-homogeneous Navier-slip boundary conditions.
Let us emphasise that the studies in the literature [12, 17,18,19,20] turn out that the non-homogeneous Navier-slip boundary conditions are compatible with the inviscid limit transition of the viscous state, then we expect that our approach will be relevant to control the evolution of turbulent flows typically associated with high Reynolds number (or small viscosity).
We should mention that the most results in the literature on the optimal control of fluid flows are of deterministic nature. The control of a stochastic system is much more involved and there are few results available in the literature. We refer the articles [6, 7, 10, 33], where the authors solved tracking control problems in 2D and 3D, respectively. In these works, the control variables act in the interior of the domain. Recently in [49], the authors studied a stochastic boundary control problem for the deterministic steady Navier–Stokes equations, where the stochastic control is imposed on the boundary by a stochastic non-homogeneous Dirichlet boundary condition.
In a forthcoming paper, we intend to deduce the first-order necessary optimality conditions and analyse the second-order sufficient conditions, which are important for implementing numerical methods to determine the optimal boundary control.
References
Acevedo, P., Amrouche, C., Conca, C., Ghosh, A.: Stokes and Navier–Stokes equations with Navier boundary condition. J. Differ. Equ. 285(5), 258–320 (2021)
Amrouche, C., Rejaiba, A.: Stationary Stokes equations with friction slip boundary conditions. Monografías Matem áticas, García de Galdeano 39, 23–32 (2014)
Aubin, J.-P.: Approximation of Elliptic Boundary-Value Problems. Wiley, New York-Toronto (1972)
Azevedo, N., Pinheiro, D., Pinheiro, S.: Dynamic programming for semi-Markov modulated SDEs. Optimization 71(8), 2315–2342 (2022)
Baltas, I., Dopierala, L., Kolodziejczyk, K., Szczepański, M., Weber, G.W., Yannacopoulos, A.N.: O ptimal management of defined contribution pension funds under the effect of inflation, mortality and uncertainty. Eur. J. Oper. Res. 298(3), 1162–1174 (2022)
Benner, P., Trautwein, C.: Optimal control problems constrained by the stochastic Navier–Stokes equations with multiplicative Levy noise. Math. Nachr. 292(7), 1444–1461 (2019)
Benner, P., Trautwein, C.: A stochastic maximum principle for control problems constrained by the stochastic Navier–Stokes equations. Appl. Math. Optim. 84, 1001–1054 (2021)
Black T.L., Sarnecki A.J.: The Turbulent Boundary Layer with Suction or Injection. Aeronautical Research Council Reports and Memoranda 3387 (October, 1958), London (1965)
Braslow A.L.: A History of Suction-Type Laminar-Flow Control with Emphasis on Flight Research. NASA History Division (1999)
Breckner H.I.: Approximation and optimal control of the stochastic Navier–Stokes equations. Ph.D. Thesis, Halle (Saale) (1999)
Brezis, H.: Functional Analysis, Sobolev Spaces and Partial Differential Equations. Springer (2011)
Chemetov, N.V., Antontsev, S.N.: Euler equations with non-homogeneous Navier slip boundary condition. Physica D 237, 92–105 (2008)
Chemetov, N.V., Cipriano, F.: Well-posedness of stochastic second grade fluids. J. Math. Anal. Appl. 454, 585–616 (2017)
Chemetov, N.V., Cipriano, F.: Uniqueness for optimal control problems of two-dimensional second grade fuids. Elect. J. Differ. Equ. 2022(22), 1–12 (2022)
Chemetov, N.V., Cipriano, F.: Injection–suction control for two-dimensional Navier–Stokes equations with slippage. SIAM J. Control. Optim. 56(2), 1253–1281 (2018)
Chemetov, N.V., Cipriano, F.: Optimal control for two-dimensional stochastic second grade fluids. Stoch. Process. Appl. 128(8), 2710–2749 (2018)
Chemetov, N.V., Cipriano, F.: Boundary layer problem: Navier–Stokes equations and Euler equations. Nonlinear Anal. Real World Appl. 14(6), 2091–2104 (2013)
Chemetov, N.V., Cipriano, F.: The inviscid limit for the Navier–Stokes equations with slip condition on permeable walls. J. Nonlinear Sci. 23(5), 731–750 (2013)
Chemetov, N.V., Cipriano, F.: Inviscid limit for Navier–Stokes equations in domains with permeable boundaries. Appl. Math. Lett. 33, 6–11 (2014)
Chemetov, N.V., Cipriano, F., Gavrilyuk, S.: Shallow water model for the lake with friction and penetration. Math. Methods Appl. Sci. 33(6), 687–703 (2010)
Cipriano, F., Martins, N.F.M., Pereira, D.: Optimal portfolio for the \(\alpha \)-Hypergeometric stochastic volatility model. SIAM J. Finan. Math. 12(1), 226–253 (2021)
Cipriano, F., Pereira, D.: On the existence of optimal and \(\epsilon -\)optimal feedback controls for stochastic second grade fluids. J. Math. Anal. Appl. 475, 1956–1977 (2019)
Coron, J.M.: On the controllability of the \(2D\) incompressible Navier–Stokes equations with the Navier-slip boundary conditions. ESAIM Control Optim. Calc. Var. 1, 35–75 (1996)
Demengel, F., Demengel, G.: Functional the Theory of Elliptic Partial Differential Equations. Springer, London (2012)
Evans L.C.: Partial Differential Equations. AMS, Graduate Studies in Mathematics. 19 (1998)
Fleming, W., Soner, H.: Controlled Markov Processes and Viscosity Solutions, 2nd edn. Springer, New York (2006)
Girault, V., Raviart, P.-A.: Finite Element Methods for Navier–Stokes Equations, Theory and Algorithms. Springer, Berlin (1986)
Gunzburger, M., Hou, L., Svobodny, T.: Analysis and finite element approximation of optimal control problems for the stationary Navier–Stokes equations with Dirichlet controls. Modél. Math. Anal. Num. 25, 711–748 (1991)
Gunzburger, M., Manservisi, S.: The velocity tracking problem for Navier–Stokes flows with boundary control. SIAM J. Control. Optim. 30, 594–634 (2000)
Kalaycı, B., Özmen, A., Weber, G.M.: Mutual relevance of investor sentiment and finance by modeling coupled stochastic systems with MARS. Ann. Oper. Res. 295, 183–206 (2020)
Karatzas, I., Shreve, S.E.: Brownian Motion and Stochastic Calculus. Springer, GTM (1987)
Ladyzhenskaya O.A., Solonnikov V.A., Uraltseva N.N.: Linear and quasilinear equations of parabolic type. Translations of mathematical monographs. 23. Providence, Amer. Math. Society (1968)
Lisei, H.: A minimum principle for the stochastic Navier–Stokes equation. Stud. Univ. Babes-Bolyai Math. 45(2), 37–65 (2000)
Lisei, H.: Existence of optimal and Epsilon-optimal controls for the stochastic Navier–Stokes equation. Nonlinear Anal. 51, 95–118 (2002)
Nirenberg, L.: On elliptic partial differential equations. Ann. Scuola Norm. Sup. Pisa. 13(3), 115–162 (1959)
Oksendal, B., Sulem, A.: Applied Stochastic Control of Jump Diffusions. Springer, Berlin (2005)
Özcan, İ, Śledziński, J.D., Alparslan Gök, S.Z., Butlewski, M., Weber, G.W.: Mathematical encouragement of companies to cooperate by using cooperative games with fuzzy approach. J. Ind. Manag. Optim. 19(10), 7180–7195 (2023)
Özmen, A., Zinchenko, Y., Weber, G.M.: Robust multivariate adaptive regression splines under cross-polytope uncertainty: an application in a natural gasmarket. Ann. Oper. Res. 324, 1337–1367 (2023)
Savku, E.: A stochastic control approach for constrained stochastic differential games with jumps and regimes. Mathematics 11(14), 3043 (2023)
Savku E., Azevedo N., Weber G.W.: Optimal control of stochastic hybrid models in the framework of regime switches. In: Pinto A., Zilberman D. (editors), Modeling, Dynamics, Optimization and Bioeconomics II, DGS 2014, Cham. (Springer Proceedings in Mathematics & Statistics vol. 195), Switzerland, Springer, 371–387 (2017)
Savku E., Weber G.W.: A Regime-Switching Model with Applications to Finance: Markovian and Non-Markovian Cases. In: Dynamic Economic Problems with Regime Switches. Berlin-Heidelberg, Germany, Springer, 287-309 (2021)
Savku, E., Weber, G.W.: Stochastic differential games for optimal investment problems in a Markov regime-switching jump-diffusion market. Ann. Oper. Res. 312, 1171–1196 (2020)
Sritharan, S.S.: An Introduction to Deterministic and Stochastic Control of Viscous Flow, Optimal Control of Viscous Flow, pp. 1–42. SIAM, Philadelphia (1998)
Tahraoui, Y., Cipriano, F.: Optimal control of two dimensional third grade fluids. J. Math. Anal. Appl. 523(2), 127032 (2023)
Taylan, P., Yerlikaya-Özkurt, F., Uçak, B.B., Weber, G.W.: A new outlier detection method based on convex optimization: application to diagnosis of Parkinson’s disease. J. Appl. Stat. 48(13–15), 2421–2440 (2021)
Temam, R.: Navier–Stokes Equations, Theory and Numerical Analysis. AMS Chelsea Publishing, Providence (2001)
Weber, G.W., çavuşoǧlu Z., Özmen A.: Predicting Default probabilities in emerging markets by new conic generalized partial linear models and their optimization. Optimization 61(4), 443–457 (2012)
Young, J., Zhou, X.: Stochastic Controls: Hamiltonian Systems and HJB Equations. Springer, New York (1999)
Zhao W., Gunzburger M.: Stochastic collocation method for stochastic optimal boundary control of the Navier–Stokes equations. Appl. Math. Optim. 87, 6 (2023). https://doi.org/10.1007/s00245-022-09910-y
Acknowledgements
A substantial part of this work was developed during N.V. Chemetov’s visit to the NOVAMath Research Center. He would like to thank the NOVAMath for the financial support (through the projects UIDB/00297/2020 and UIDP/00297/2020) and the very good working conditions. The work of N.V. Chemetov was also supported by FAPESP (Fundação de Amparo à Pesquisa do Estado de São Paulo), project 2021/03758-8, “Mathematical problems in fluid dynamics”. The work of F. Cipriano is funded by national funds through the FCT - Funda ção para a Ciência e a Tecnologia, I.P., under the scope of the projects UIDB/00297/2020 and UIDP/00297/2020 (Center for Mathematics and Applications).
Funding
Open access funding provided by FCT|FCCN (b-on).
Author information
Authors and Affiliations
Corresponding author
Additional information
Communicated by Giancarlo Bigi.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.
Rights and permissions
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
About this article
Cite this article
Chemetov, N., Cipriano, F. A Boundary Control Problem for Stochastic 2D-Navier–Stokes Equations. J Optim Theory Appl (2024). https://doi.org/10.1007/s10957-024-02416-3
Received:
Accepted:
Published:
DOI: https://doi.org/10.1007/s10957-024-02416-3