Mean reflected Mckean-Vlasov stochastic differential equation

(i) By the definition of H(x,\nu) and the bi-Lipschitz assumption of h we have for any \nu \in \mathcal{P}_1(\mathbb{R})

and

(ii) For any \nu, \nu^\prime \in \mathcal{P}_1(\mathbb{R}), by definetion we have

The proof for the case of MR-SDE is given in [briand2020particles]. We present here the proof of the MR-MVSDE.

Step 1: Construct the solution (X,K) of (?).

Let us consider the set

And let \widetilde{X}_t\in \mathcal{C}^2 be a given process. We set

and define the function K_t by setting

The function K_t being given, let us define the process X_t by the formula

Step 2: Check K_t satisfies \mathbb{E}\left[h\left(X_t\right)\right] \geq 0 and \int_0^t \mathbb{E}\left[h\left(X_s\right)\right] d K_s=0.

Based on the definition of K_t, we have \mathbb{E}\left[ h(X_t) \right]\geq 0, K_t = G_0\left( \mathcal{L}_{\tilde{U}_t} \right), dK_t-a.e. and G_0(\mathcal{L}_{\tilde{U}_t})>0, dK_t-a.e. Then we obtain

Moreover, since h is continuous, we have \mathbb{E}\left[h\left(\tilde{U}_s+G_0\left(\mathcal{L}_{\tilde{U}_s}\right)\right)\right]=0 as soon as G_0\left(\mathcal{L}_{\tilde{U}_s}\right)>0, so that

Step 3: Show the solution map \Phi which associates to \tilde{X} to the solution X of (?) is a contraction.

Let \tilde{X}, \tilde{X}^\prime \in \mathcal{C}^2 and define \tilde{U}, K and \tilde{U}^\prime, K^\prime as above, using the some Brownian motion. We have from the Lipschitz assumption of b and \sigma, Cauchy-Schwarz inequality and Burkholder-Davis-Gundy (B-D-G) inequality

From the representation of K and Lemma 2.6, we have

Plugging (?) into (?), we have

Hence, there exists a positive \tilde{T}, depending on b, \sigma and h only, such that for all T<\tilde{T}, the map \Phi is a contraction. We first deduce the existence and uniqueness of the solution on [0, \tilde{T}] and then on \mathbb{R}^{+} by iterating the construction.

For given X_0\in \mathbb{R} and t_0\geq 0, let us set \mu_s : = \mathcal{L}_{X_s^{t_0,X_0}}. Then, by Lemma 2.3,

where, by (?)-(ii),

From the equation satisfied by u, we obtain

(i) We have

We first deal with K^p_T = \sup_{t\leq T}G_0(\mathcal{L}_{U_t}). Apply Lipschitz property of G_0, Lemma 2.6 and the definition of the Wasserstein metric \mathcal{W}_1, we have

since G_0(\mathcal{L}_{U_0}) = 0 as \mathbb{E}\left[ h(X_0) \right]\geq 0, and where U is defined by (3.2). Therefore

Plugging (?) into (Proof), we deduce that there exists a constant C_p\geq 0 such that

Apply Gronwall’s lemma, we deduce for all p\geq 2, there exists a constant C>0 such that

(ii) we observe that \nu \rightarrow G_0(\nu) is Lipschitz continuous, we have

so the result follows from standard computations.

(iii) we use the decomposition

As a result of (ii), we complete this proof.

Firstly, we show K is Lipschitz continuous. To do so, we prove that s\mapsto \bar{G}_0(\mu_s) is Lipschitz continuous on [0,T]. If s \leq t, we have

the first equality since H(\bar{G}_0(\mu_t),\mu_t) = 0.

Let \bar{A}_{x,\mu} be the partial operator of second order defined by

Apply Itô’s lemma, we have

The result follows that h has bounded derivatives and \sup_{t\leq T}|X_t|\in L^2 for each T>0. So from [briand2020particles], we deduce that K is Lipschitz continuous and has a bounded density on [0,T] for each T>0.

Then we show the density of dK is (3.6). From Assumption 2.4 and apply Itô’s lemma, for any s\leq t we can get

Then, we obtain

Using (?), Lemma 3.7 , and the proof of [briand2020particles][Proposition 2.7], we can get the density of dK w.r.t the Lebesgue measure is

Then, we complete the proof.

(i) is the [briandMEAN][Lemma 3].

Now, we prove (ii). Since the Assumptions 2.1, 2.3 and 2.4 hold., the mapping x\longmapsto \mathcal{A}_{\mathcal{L}_{X_t}}h(X_t) is continuous and for all x\in \mathbb{R}, there exists some universe constants C_1 and C_2 such that

Then, using (i), we deduce that t \rightarrow \mathbb{E}\mathcal{A}_{\mathcal{L}_{X_t}}h(X_t) is continuous.

Let \{\xi_i\}_{i=1}^n be a sequence in \mathbb{R} and \lim_{i\rightarrow \infty} \xi_i = \xi. We will show that

That means the mapping \Xi is continuous.

Apply the Lipschitz condition of b and \sigma, and Cauchy-Schwarz inequality, we have

and

Apply Gronwall’s lemma, we have

Letting n goes to infinity, we complete this proof.

Apply the Lipschitz condition of b^n and \sigma^n, and Cauchy-Schwarz inequality ,we have

where

Then, we control \sup_{t\leq T}\left|K^{n}_t - K_t\right|^2:

Apply Gronwall’s lemma, we have

Letting n go to infinity, we complete this proof.

Apply the Lipschitz condition of b and \sigma and Cauchy-Schwarz inequality, we have

where

Moreover,

Combining (Proof) and (Proof), we have

Since A^n_s+ \left\langle M^n \right\rangle_s is an increasing process, then according to the Stochastic Gronwall lemma, we obtain

using the assumptions of (A_t,A_t^n,M_t,M_t^n), we have

Therefore,

Let \tilde{b}(y,t) =b\left( y, \delta_{\psi_t}\right)+\sigma\left(y, \delta_{\psi_t}\right) h(t). Since b and \sigma satisfy assumption 2.1, and h \in L^2([0,T];\mathbb{R}), we can easily prove the existence and uniqueness of solution for ? through the standard argument (Picard iteration).

it is easy show (?) has unique solution.

According to Proposition 5.5 , there exists a measurable map \mathcal{G}^0:

such that for any h \in L^2([0,T],\mathbb{R})

Let

and

\mathcal{H}^M is endowed with the weak topology on L^2\left([0, T], \mathbb{R}\right). Then \mathcal{H}^M is a Polish space. For every \varepsilon > 0, there exists a measurable mapping \mathcal{G}^\varepsilon(\cdot) :C\left([0, T] ; \mathbb{R}\right) \rightarrow \mathcal{X} such that

and for any M > 0 and h^\varepsilon \in \tilde{\mathcal{H}}^M ,

From Girsanov’s Theorem, we know Z^\varepsilon is the solution of the following MR-MVSDE:

Following [liu2022large], we only need to verify the two claims:

• (LDP1) For every M<+\infty and any family \left\{h^n ; n \in \mathbb{N}\right\} \subset \mathcal{H}^M converging to some element h \in \mathcal{H}^M as n \rightarrow \infty

  \lim\limits_{n\rightarrow \infty} \sup_{t\in [0,T]}\left|\mathcal{G}^0\left( \int_0^{\cdot}h^n(s)ds \right) - \mathcal{G}^0\left( \int_0^{\cdot}h(s)ds \right)\right| = 0\,.

• (LDP2) For every M<+\infty and any family \left\{h^{\varepsilon} ; \varepsilon>0\right\} \subset \tilde{\mathcal{H}}^M and any \theta>0,

  \lim_{\varepsilon\rightarrow 0}\mathbb{P}\left[ \sup_{t\in [0,T]}\left|Z^\varepsilon_t - Y^{h^\varepsilon}_t\right| \geq \theta \right] = 0\,,

  where Z^{\varepsilon} is the unique solution to (?) and Y^{h^{\varepsilon}}=\mathcal{G}^0\left(\int_0^\cdot h^{\varepsilon}(s) \mathrm{d} s\right).

The verification of (LDP1) will be given in Proposition 5.7. (LDP2) will be established in Proposition 5.9.

For simplicity, we denote Y^{h_n} : = \mathcal{G}^0\left( \int_0^{\cdot}h^n(s)ds \right), Y^h = \mathcal{G}^0\left( \int_0^{\cdot}h(s)ds \right).

Let \kappa^n(t):= \sup_{r\in [0,t]}\left|Y^{h^n}(r) - Y^h(r)\right|. By the Lipschitz condition of b, we have

By the Lipschitz condition of \sigma, we have

By the linear growth of \sigma, we have

Since h^n(t) \rightarrow h(t) weakly in L^2\left( [0,T],\mathbb{R} \right) as n\rightarrow \infty, we have

Then, by the Cauchy-Schwarz inequality, we have

Similarly, we have, for any 0\leq t_1\leq t_2\leq T,

which means that the sequence \left \{ I^n_3:n\geq 1 \right \} is equi-continuous. By the Arzela-Ascoli theorem, we deduce that \left \{ I^n_3:n\geq 1 \right \} is relatively compact in \mathcal{C}\left( [0,T],\mathbb{R} \right).

We can deduce that for any t \in [0,T],

Together with the relative compactness of \left \{ I^n_3:n\geq 1 \right \} in \mathcal{C}\left( [0,T],\mathbb{R} \right), we can get

By Lemma 2.6 , we have

Combing (Proof), (Proof), (Proof), (Proof) and (?) together, we have

Then, by (?) and Gronwall’s lemma, we have

The proof is complete.

By the Lipschitz condition of b and \sigma, we have

Putting (?), (?) and (?) to (Proof) and applying the Gronwall’s lemma, we have

By the definition of \mathcal{W}_2\left( \mathcal{L}_{X^\varepsilon_t},\delta_{\psi_t} \right), we have following estimation

Thanking to Gronwall’s lemma, we have

By Chebyshev’s inequality, we only need to show that

Then, it follows that

And

By (Proof) and (Proof) and apply Gronwall’s lemma, we deduce that

Apply Gronwall’s lemma, Lemma 5.8 and letting \varepsilon \rightarrow \infty, we get

The proof is complete.

From the definition of P^*_t, we have P^*_t\mu_0 = \mu_t, P^*_t\nu_0 = \nu_t.

Apply Itô’s lemma to (X_t- Y_t)^2, we have

Since X_t and Y_t is the solution of (?) and h is bi-Lipschitz , we have \mathbb{E}\left[ X_t \right]\geq 0 and \mathbb{E}\left[ Y_t \right]\geq 0. Combining K^X_t and K^Y_t are non-decrease processes, we have

By the definition of solution (?), we obtain

Taking expectation of (Proof) and applying stochastic Fubini’s Theorem, we have the following estimation

Apply Assumption 2.5 and Gronwall’s lemma, we obtain

We first show that there exists a \mu \in \mathcal{P}_2(\mathbb{R}) such that

To this end, it suffices to show that \left\{P_t^* \delta_0\right\}_{t \geq 0} is a \mathcal{W}_{2}-Cauchy family when t \rightarrow \infty. Using the semigroup property of P_{t+s}^* and the exponential convergence, we have

where X^0_s is a solution of MR-MVSDE whose initial condition is \xi = 0. Then, we only need to show

It is true from Proposition 3.5. By (Proof), P^*_{t+s} = P^*_t P^*_s and

we obtain that

That means P^*_t \mu = \mu.

By Lemma 6.5 and Girsanov’s Theorem, d\mathbb{Q} := R_T d\mathbb{P} is a probability measure such that

is a Brownian motion. Then, (?) can be written to

Consider MR-MVSDE under \mathbb{Q} measure

By Theorem 3.2, we have \tilde{X}_t = Y_t for t\in [0,T]. So we deduce that

In particular,

Following [arnaudon2014equivalent], we can get for any f\in \mathcal{B}_b^+(\mathbb{R})

Repeating the proof of [wang2018distribution][Theorem 4.1 (ii)], we can obtain for any different \mu_0 and \nu_0\in \mathcal{P}_2(\mathbb{R}) and any f\in \mathcal{B}_b^+(\mathbb{R}),

and

Then, we complete this proof.

Let Y_t = X_t + \frac{tv}{T}, then

where

Let R_T = \exp\left \{ \int_0^T \eta_s dW_s - \frac{1}{2}\int_0^T |\eta_s|^2ds \right \}. Since \mathbb{E}\left[ \exp(\frac{1}{2}\int_0^T |\eta_s|^2ds) \right]< \infty, from Nokikov’s condition, we deduce that R_t is a martingale and \mathbb{E} R_T = 1. By the Girsanov’s Theorem, we can get

Applying Young’s inequality, we obtain

Then we have

and

By Assumption 2.1, for any n\geq 1, \left( R_{t\wedge\tau_n} \right)_{t\in [0,T]} is a uniformly integrable continuous martingale. Since \tau_n \rightarrow T as n\rightarrow \infty, by the martingale convergence Theorem, we only need to prove

We fix t \in [0,T) and n\geq 1. By Girsanov’s Theorem,

is a Brownian motion under the weighted probability \mathbb{Q}_{t,n}: = R_{t\wedge \tau_n} \mathbb{P}.

Under \mathbb{Q}_{t,n}, reformulating (?) and (?)

Apply Itô’s Lemma, we have

Using the same argument of proving Lemma 6.1, we have

Then, applying Young’s inequality, we have

By the exponential convergence (Lemma 6.1), we have

Then, we have

Therefore, (Proof) holds since t \in [0,T) and n\geq 1 are arbitrary.

Let us consider the set \mathcal{C}^2 = \left \{ X \text{ is } \mathcal{F}-\text{adapted continuous process}, \mathbb{E}\left( \sup_{0\leq t\leq T}|X_t|^2< \infty \right) \right \}. And let \widetilde{X}\in \mathcal{C}^2 be a given process. We set

and define the function K by setting

The function K being given, let us define the process X by the formula

Based on the definition of K, we have \mathbb{E}\left[ h(X_t) \right]\geq 0, K_t = G_0\left( \mathcal{L}_{\tilde{U}_t} \right), dK_t-a.e. and G_0(\mathcal{L}_{\tilde{U}_t})>0, dK_t-a.e. Then we obtain

Moreover, since h is continuous, we have \mathbb{E}\left[h\left(\tilde{U}_s+G_0\left(\mathcal{L}_{\tilde{U}_s}\right)\right)\right]=0 as soon as G_0\left(\mathcal{L}_{\tilde{U}_s}\right)>0, so that

Let \tilde{X}, \tilde{X}^\prime \in \mathcal{C}^2 and define \tilde{U}, K and \tilde{U}^\prime, K^\prime as above, using the some Brownian motion. We have from the Lipschitz assumption of b and \sigma, B-D-G inequality

From the representation of K and Lemma 2.6, we have

Plugging (?) into (?), and use the stochastic Gronwall lemma, we have

Hence, there exists a positive \tilde{T}, depending on b, \sigma and h only, such that for all T<\tilde{T}, the map \Phi is a contraction. We first deduce the existence and uniqueness of the solution on [0, \tilde{T}] and then on \mathbb{R}^{+} by iterating the construction.