Abstract
We consider a stochastic perturbation of the Stefan problem. The noise is Brownian in time and smoothly correlated in space. We prove existence and uniqueness and characterize the domain of existence.
Similar content being viewed by others
Avoid common mistakes on your manuscript.
1 Introduction
Moving boundary problems are one of the important areas of partial differential equations. They describe a wide range of physically interesting phenomena where a system has two phases. However, since the boundary between these phases is defined implicitly, they provide deep mathematical challenges.
Our goal here is to study the effect of noise on the Stefan problem, which is one of the canonical moving boundary value problems. Fix a probability triple and assume that ζ is a random field which is Brownian in time but which is correlated in space (we will rigorously define ζ in Sect. 2). We consider the stochastic partial differential equation (SPDE)
The constant α∈ℝ is fixed (we shall later see why it is natural to include this term). We also assume that the initial condition u ∘∈C(ℝ) satisfies some specific properties:
-
u ∘≡0 on ℝ−, u ∘>0 on (0,∞), and \(\lim_{x\searrow 0}\tfrac{du_{\circ}}{dx}(x)\) exists.
-
u ∘ and its first three derivatives exist on (0,∞) and are square-integrable (on (0,∞)).
The last requirement in (1) means that the boundary between u≡0 and u>0 is exactly the graph of β.
In fact, it is not yet clear that (1) makes sense. Differential equations are pointwise statements. Stochastic differential equations are in fact shorthand representations of corresponding integral equations; pointwise statements typically do not make sense. It will take some work to restate the pointwise stochastic statement in the first line of (1) as a statement about stochastic integrals.
There has been fairly little written on the effect of noise on moving boundary problems (see [1, 4], and [15]; see also the work on the stochastic porous medium equation in [2, 6–8, 14]). We note here that the multiplicative term u in front of the dζ t places this work slightly outside of the purview of the theory of infinite-dimensional evolution equations with Gaussian perturbations. The multiplicative term is in fact a natural nonlinearity. It implies that there is only one interface; bubbles where u>0 cannot spontaneously nucleate where u=0, and conversely, bubbles where u=0 cannot spontaneously nucleate where u>0 (see Lemma 5.8).
Our work here follows on [15], where we investigated the structure of a moving boundary problem driven by a single Brownian motion. The moving boundary problem of interest there is a model of detonation. We here focus on the Stefan problem, and consider noise which is spatially correlated. We formulate several techniques which can (hopefully) be applied to a number of stochastic moving boundary value problems. In our particular case, where the randomness comes from a noise which is correlated in space and Brownian in time, several transformations (the transformations of Lemmas 4.3 and 4.4 and (25)) can transform the problem into a random nonlinear PDE (see (26)). All of these transformations are not in general available when the noise is more complicated, but most of the techniques we develop here should be. Secondly, the irregularity of the Brownian driving force requires some detailed analysis, no matter what perspective one takes: namely, in the analysis of Lemma 4.2 and the iterative bounds of Lemma 5.5.
2 The Noise
Let us start by constructing the noise process. Fix η∈C ∞(ℝ)∩L 2(ℝ) such that η (n)∈C ∞(ℝ)∩L 2(ℝ) for all n∈{1,2,3} and such that
Let W be a Brownian sheet. For t≥0 and x∈ℝ, define
Then ζ is a zero-mean Gaussian field with covariance structure given by
for all s and t in ℝ+ and x and y in ℝ. Thus for each x∈ℝ, t↦ζ t (x) is a Brownian motion, and for each t>0, x↦ζ t (x) is in C 2 with derivative given by
for all t≥0 and x∈ℝ for n∈{0,1,2}.
3 Weak Formulation
To see what we mean by (1), let us replace \(\dot{\zeta}_{t}(x)\) by a smooth function b:ℝ+×ℝ→ℝ; the Wong–Zakai result (cf. [13, Sect. 5.2D]) implies that this is a reasonable approximation of an SPDE with Stratonovich integration against the noise; we can then convert this into the desired SPDE with Ito integration. Namely, consider the PDE
where \(\alpha_{s} \overset {\mathrm {def}}{=}\alpha-\tfrac{1}{2}\) (we will see that this corresponds to the Stratonovich analog of (1)). This will be our starting point.
Let us see what a weak formulation looks like (see [10, Chap. 8]). Fix \(\varphi\in C_{c}^{\infty}(\mathbb {R}_{+}\times \mathbb {R})\). Assume that β ∘ is differentiable. Define
Differentiating, we get that
and we can use the fact that v(t,β ∘(t))=0 to delete the last term. We can also use the PDE for v for x>β ∘(t) to rewrite \(\tfrac{\partial v}{\partial t}\). Integrating by parts, we have that
Again, we use the fact that v(t,β ∘(t))=0, and we can also use the boundary condition on \(\tfrac{\partial v}{\partial x}\). Recombining things we get the standard formula that
Replacing b by our noise, we should have the following formulation: for any \(\varphi\in C^{\infty}_{c}(\mathbb {R}_{+}\times \mathbb {R})\) and any t>0,
The Ito formulation of this would be
Remark 3.1
The structure of the SPDE (1) is invariant under Ito and Stratonovich formulations; this is the motivation for including α in (1).
We can now formally define a weak solution of (1). In this definition, we allow for blowup. Define for all t≥0.
Definition 3.2
A weak solution of (1) is a nonnegative predictable path {u(t,⋅)∣0≤t<τ}⊂C(ℝ)∩L 1(ℝ), where τ is a predictable stopping time with respect to , such that for any \(\varphi\in C_{c}^{\infty}(\mathbb {R}_{+}\times \mathbb {R})\) and any finite stopping time τ′<τ,
and where
where β is a semimartingale.
Our main existence and uniqueness theorems are the following. The arguments leading up to these results will come together in Sect. 5.
Theorem 3.3
(Existence)
There exists a predictable path {u(t,⋅)∣0≤t<τ}⊂C(ℝ)∩L 1(ℝ) which satisfies (4), and u(t,⋅)∈C 1[β(t),∞) for all t∈[0,τ) and
Furthermore, if u(t,⋅)∈C 2[β(t),∞) for all t∈[0,τ), then it satisfies (5).
Proof
Combine Lemmas 5.9 and 5.10. □
We also have uniqueness.
Theorem 3.4
(Uniqueness)
Suppose that {u 1(t,⋅); 0≤t<τ 1} and {u 2(t,⋅); 0≤t<τ 2} are two solutions of (1). Assume that for i∈{1,2}, the map x↦u i (t,x+β i (t)) has three generalized square-integrable derivatives on (0,∞). Then u 1(t,⋅)=u 2(t,⋅) for 0≤t<min{τ 1,τ 2}.
Proof
The proof follows from Lemma 5.11. □
4 Regularity and a Transformation
The proof of Theorems 3.3 and 3.4 will hinge upon a transformation of (1) into a nonlinear integral equation on a fixed (as opposed to an implicitly defined) domain; we will address this in Sect. 4.2. First, however, let us make sure that we understand a bit about regularity; this will illuminate the assumptions needed.
4.1 Regularity
While regularity of moving boundary value problems is an incredibly challenging area (see [3]), we can make some headway. Namely, if we assume enough regularity for the boundary, we can get better control of the sense in which the boundary behavior holds.
We start by rewriting (4) using heat kernels. Define
where the second representation of p ± stems from the fact that p ∘ is even in its second argument. We then have that
as the relevant distinction between p + and p − is their behavior at x=0, namely,
for all \(n\in \mathbb {N}\overset {\mathrm {def}}{=}\{1,2,\ldots \}\), all t>0 and all y∈ℝ. This will come up in the arguments of Lemmas 4.2 and 4.3.
Let us next understand integration against ζ. Let {β(t)∣t≥0} be an ℝ-valued predictable and continuous function. Let {f(t)∣t≥0} be a second ℝ-valued predictable and continuous function, which is also bounded. We define
this being a limit in L 2. Due to (2),
is a Brownian motion. Therefore we can define a Brownian motion \(B^{x}_{t}\) for each fixed x as
Note also that dζ t (β(t)) is not the total derivative of ζ t (β(t)); i.e.,
To understand the total derivative, we must also include the spatial variation of ζ t :
This implies that
Lemma 4.1
Suppose that {u(t,⋅)∣0≤t<τ}⊂C(ℝ)∩C 1([β(t),∞)) is a weak solution of (1) with u(t,∞)=0 for 0≤t<τ. Suppose also that β is continuously differentiable and -adapted. Then u(t,x+β(t)) satisfies the integral equation
for all t<τ and x>0.
Proof
Fix x>0 and T>0. For t∈[0,τ∧T), define
Using Definition 3.2, we have
where
Then
We have here used the fact that u(t,β(t))=0. Thus
Now we consider \(A^{T}_{2}(t)\). Using the stochastic Fubini theorem [18, Theorem 2.6], we obtain
Combine things to get that
Now let T↘t to get the claimed result. □
Note that (9) is not an explicit formula for u since the right-hand side of (9) depends on u through β. Also, in order that u(t,∞)=0,0≤t<τ, it suffices that \(\{u(t,\cdot),\frac{\partial u}{\partial x}(t,\cdot)\}\subseteq L^{2}((\beta(t),\infty))\), which is guaranteed in the Picard iteration part. This can be seen by \(|f^{2}(\infty)|\le |f^{2}(M)|+\int_{M}^{\infty}2|ff'|\le|f^{2}(M)|+\int_{M}^{\infty}f^{2}+\int_{M}^{\infty}(f')^{2}\) for M>0, and liminf M→∞ f 2(M)=0.
The value of Lemma 4.1 is in that it implies that if β is continuously differentiable, then the Stefan boundary condition of (1) holds pointwise.
Lemma 4.2
Let {u(t,⋅)∣0≤t<τ} be a solution of (1) with u(t,∞)=0 for 0≤t<τ, and let u be continuously differentiable in x for x>β(t). Assume also that
If β is continuously differentiable, then
for all t∈[0,τ).
Proof
To see this, let us rewrite (9) in a slightly more convenient way. If {u(t,⋅)∣0≤t<τ} is a weak solution of (1) and 0<t<τ, set
Note that
Thus
and hence
Since \(\frac{\partial p_{+}}{\partial x}(t,0,y)=0\), we have
Next note that
Thus
Dominated convergence then implies that
Turning to A 3, we can integrate by parts and using the fact that u(t,β(t))=0, we get that
Since
for all T>0, we can use dominated convergence and (10) to get that
Finally, we consider A 4. Since p +(t,x,0)=2e αt p ∘(t,x), we get that
Therefore
Since β is continuously differentiable, dominated convergence ensures that
Collecting things together, we get the claim. □
4.2 A Transformation
The characterization of β given in (11) allows us to rewrite the moving boundary value problem in a more convenient way. The calculation which gives us some analytical traction is found in [16] (see also [10, Chap. 8]). Let us again return to our deterministic PDE (3). For all t≥0 and x∈ℝ, define \(\tilde{v}(t,x) = v(t,x+\beta_{\circ}(t))\); then \(v(t,x)= \tilde{v}(t,x-\beta_{\circ}(t))\). Assuming that β ∘ is differentiable, we have that for x>0 and t>0,
We can combine these equations and use the PDE for v to rewrite the evolution of \(\tilde{v}\) as
Inserting the boundary condition that \(-\varrho \dot{\beta}_{\circ}(t)=\lim_{x\searrow \beta_{\circ}(t)}\frac{\partial v}{\partial x}(t,x)\) back into (12), we have that
Replacing b by ζ and α s by α, we should be able to write down a nonlinear SPDE for \(\tilde{u}(t,x) \overset {\mathrm {def}}{=}u(t,x+\beta(t))\). We now get the following.
Lemma 4.3
Suppose that {u(t,⋅)∣0≤t<τ}⊂C(ℝ)∩L 1(ℝ) is a solution of (1) with u(t,∞)=0 such that u(t,⋅)∈C 1([β(t),∞)) and \(\tfrac{\partial u}{\partial x}(t,\cdot)\in L^{1}([\beta(t),\infty))\) for each 0≤t<τ and (10) holds. Suppose also that β is continuously differentiable and -adapted. Then \(\tilde{u}(t,x) = u(t,x+\beta(t))\) satisfies the integral equation
for all t∈[0,τ) and x>0 where
for all t∈[0,τ).
Proof
The proof is very similar to that of Lemma 4.1. Fix x>0 and T>0. For t∈[0,τ∧T), define
Using Definition 3.2, we have
where
Using (6) and the fact that p −(T−t,x,0)=0, we get that
We have here used the fact that u(t,β(t))=0. Combining the characterization of \(\dot{\beta}\) as in Lemma 4.2, we get that
Thus
Now we consider \(A^{T}_{2}(t)\). Again, using the stochastic Fubini theorem, we obtain
Combine things to get that
Now let t↗T to get the claimed result. □
We can also obtain an SPDE for \(\tilde{u}\). By (6) and (13), we have that for 0<t<τ and x>0,
Since p −(t,0,y)=0, we have the following SPDE:
We can also find a converse to Lemma 4.3.
Lemma 4.4
Suppose that \(\{\tilde{u}(t,\cdot)\mid 0\le t<\tau\}\subset C^{1}(\mathbb {R}_{+})\cap L^{1}(\mathbb {R}_{+})\) with \(\tilde{u}(t,\infty)=0\) for 0≤t<τ satisfies (13) and is positive. Set
and define
Then {u(t,⋅)∣0≤t<τ} is a weak solution of (1).
Proof
Fix \(\varphi\in C^{\infty}_{c}(\mathbb {R}_{+}\times \mathbb {R})\) and define for 0≤t<τ,
To see the evolution of A, we fix δ>0 and define
Then define
where
where
We also note that we can rewrite the evolution of β as
Thus
Similar calculations show that
and finally,
Adding these expressions together and using the definition of \(\tilde{u}_{\delta}\) and (13), we get that
By definition of p −, we conclude that u δ (s,β(s))=0. In addition, we also have that
Upon letting δ↘0 and rearranging things, we indeed get a weak solution of (1). □
5 A Picard Iteration
Our main task now is to show that we can indeed solve (13). The main complication is that (13) is fully nonlinear due to the presence of the \(\frac{1}{\varrho }\frac{\partial \tilde{u}}{\partial x}(t,0)\) term in the drift and the shift by β in the evaluation of the integral against ζ. If we turn off the noise, we can do this via semigroup theory as in [16]. The noise, however, complicates things, as we need to respect the rules of Ito integration and (unless we want to use more advanced theories of stochastic integrals) integrate against predictable functions.
Our approach will be to set up a functional framework in which we can use Picard-type iterations to show existence and uniqueness. As usual, \(C^{\infty}_{0}(\mathbb {R}_{+})\) is the collection of infinitely smooth functions on [0,∞) which asymptotically vanishes at infinity. Define next
in other words, \(C^{\infty}_{0,\mathrm {odd}}(\mathbb {R}_{+})\) are those elements of \(C^{\infty}_{0}(\mathbb {R}_{+})\) which can be extended to an odd element of C ∞(ℝ) (namely, consider the map y↦sgn (y)φ(|y|). For all \(\varphi\in C^{\infty}_{0}(\mathbb {R}_{+})\), define
Let H be the closure of \(C^{\infty}_{0}(\mathbb {R}_{+})\) with respect to ∥⋅∥ H and let H odd be the closure of \(C^{\infty}_{0,\mathrm {odd}}(\mathbb {R}_{+})\) with respect to ∥⋅∥ H . We also define
for all square-integrable functions on ℝ+. Of course H and H odd are Hilbert spaces (H is more commonly written as H 2; i.e., it is the collection of functions on ℝ+ which possess two weak square-integrable derivatives). The important aspect of H is the following fairly standard result.
Lemma 5.1
We have that H⊂C 1. More precisely, for any φ∈H, we have that
Finally, for i∈{0,1}, \(\varphi^{(i)}(0) \overset {\mathrm {def}}{=}\lim_{x\searrow 0} \varphi^{(i)}(x)\) is well-defined.
The proof is in Sect. 5.1.
We next need some truncation functions to control the effects of various nonlinearities. In particular, we need to prevent \(\frac{\partial \tilde{u}}{\partial x}(t,0)\) from becoming too big; Picard iterations in general allow only linear growth of various coefficients. Fix L>0, which we will use as a truncation parameter. Let Ψ L ∈C ∞(ℝ;[0,1]) be monotone decreasing such that Ψ L (x)=1 if |x|≤L and Ψ L (x)=0 if |x|≥L+1 (and thus |Ψ L |≤1). In other words, Ψ L is a cutoff function with support of width L+1.
Define
for all t>0 and x∈ℝ and recursively define
For each n∈ℕ, \(\{\tilde{u}^{L}_{n}(t,\cdot);\, t\ge 0\}\) is a well-defined, adapted, and continuous path in H odd.
To study (17), we will use the Dirichlet heat semigroup. For \(\varphi\in C^{\infty}_{0}(\mathbb {R}_{+})\), t>0, and x>0, define
Lemma 5.2
For each t>0, T t has a unique extension from \(C^{\infty}_{0}(\mathbb {R}_{+})\) to H such that T t H⊂H odd and such that ∥T t f∥ H ≤e αt∥f∥ H for all f∈H. Secondly, there is a K A >0 such that
for all f∈H odd∩C 3(ℝ+).
Again, we delay the proof until Sect. 5.1.
Another convenience will be to rewrite the ds part of (17). Set
for all ψ∈H. Then
for all n∈ℕ. For ψ and η in H, let us also define
Lemma 5.3
For each ψ and η in H, \((D\tilde{\varPsi }^{a}_{L})(\psi,\eta)\) is the Gâteaux derivative of \(\tilde{\varPsi }^{a}_{L}\) at ψ in the direction of η and similarly \((D\tilde{\varPsi }^{b}_{L})(\psi,\eta)\) is the Gâteaux derivative of \(\tilde{\varPsi }^{b}_{L}\) at ψ in the direction of η. Furthermore, there is a K B >0 such that
for all ψ and η in H and L>0.
Proof
The claim is straightforward. □
For each n∈ℕ, we now define \(\tilde{w}^{L}_{n}(t,x) \overset {\mathrm {def}}{=}\tilde{u}^{L}_{n+1}(t,x)-\tilde{u}^{L}_{n}(t,x)\) for all x≥0 and t≥0. Clearly \(\sup_{0\le t\le T}\mathbb {E}[\|\tilde{w}^{L}_{1}\|_{H}^{2}]<\infty\) for all T>0. We then write that
where
where \(\dot{\zeta}_{t}(x) = \frac{\partial \zeta_{t}}{\partial x}(x)\). Note that the \(\tilde{u}^{L}_{n}\)’s and \(\tilde{w}^{L}_{n}\)’s are all in H odd.
To bound \(A^{(n)}_{1}\) and \(A^{(n)}_{2}\), we use the fact that t −3/4 is locally integrable. More precisely,
for all t>0. Thus,
Similarly, we have that
To bound \(A^{(n)}_{3}\), \(A^{(n)}_{4}\), and \(A^{(n)}_{5}\), we first rewrite them. For z∈ℝ, define \(\underline {\eta }_{z}(y) \overset {\mathrm {def}}{=}\eta(y-z)\) for all y∈ℝ. Then
We will use the following bound on the interaction between η and the H-norm.
Lemma 5.4
There is a K>0 such that
for all f∈H and k∈{0,1}.
Proof
The structure of η ensures that there is an \(\hat{\eta}\in L^{2}(\mathbb {R})\) such that \(|\eta^{(n)}(x)|\le \hat{\eta}(x)\) for all n∈{0,1,2,3} and x∈ℝ. Thus, for all x∈ℝ, k∈{0,1}, and n∈{0,1,2},
Thus,
The claim follows. □
Fix T>0 and set \(K_{1}\overset {\mathrm {def}}{=}\exp(2|\alpha| T)\). Let us first bound \(A^{(n)}_{3}\). For k∈{0,1,2},
thus for 0≤t≤T,
The bound on \(A^{(n)}_{4}\) is similar:
Consequently for 0≤t≤T,
To bound \(A^{(n)}_{5}\), we first bound \(\beta^{L}_{n+1}-\beta^{L}_{n}\). We have that
For k∈{0,1,2}, we then have that
Hence for 0≤t≤T,
Lemma 5.5
For each T>0, we have that \(\sum_{n=1}^{\infty}\sup_{0\le t\le T}\mathbb {E}[\|\tilde{u}^{L}_{n+1}-\tilde{u}^{L}_{n}\|_{H}]<\infty\). Thus, ℙ-a.s., \(\tilde{u}^{L}(t,\cdot) \overset {\mathrm {def}}{=}\lim_{n\to \infty}\tilde{u}^{L}_{n}(t,\cdot)\) exists as a limit in C([0,T];H) and u L satisfies the integral equation
where
Proof
See also [18, Lemma 3.3]. Fixing T>0 we collect the above calculations to see that there is a K T,L >0 such that
for all t∈[0,T]. Iterating this, we get that
where B is the standard Beta function and thus that
To show that the terms on the right are summable, we use the ratio test. It suffices to show that
We calculate
This implies (19). The rest of the proof follows by Jensen’s inequality and standard calculations. □
We can finally show uniqueness.
Lemma 5.6
The solution of (18) is unique.
Proof
Let u 1 and u 2 be two solutions. Define \(\tilde{w}\overset {\mathrm {def}}{=}u_{1}-u_{2}\). By calculations as above we get that
We can iterate this inequality several times to get (cf. [18, Theorem 3.2])
We can now use Gronwall’s inequality. □
We can also show non-negativity. Define the random time
for L>0.
Lemma 5.7
The solution \(\tilde{u}^{L}(t,x)\) of (18) is nonnegative for 0≤t≤τ L and x≥0.
Proof
Let \(\tilde{u}^{L}_{\alpha}(t,x)\overset {\mathrm {def}}{=}\tilde{u}^{L}(t,x)e^{-\alpha t}\). Since we have that \(\varPsi _{L}(\|\tilde{u}^{L}(t,\cdot)\|_{H})=1\) for 0≤t≤τ L , from (18) we obtain that for 0≤t≤τ L ,
where \(B_{t}^{x}\) is defined as in (7). We will then follow the approach used in [5] to show non-negativity of \(\tilde{u}^{L}_{\alpha}\), which implies non-negativity of \(\tilde{u}^{L}\). To start, fix a nonnegative and nonincreasing η∈C ∞(ℝ) such that η(u)=2 if u≤−1 and η(u)=0 if u≥0. Define
Finally, define \(\varphi_{\varepsilon }(u) \overset {\mathrm {def}}{=}\varepsilon ^{2} \varphi(\frac{u}{\varepsilon })\) for all u∈ℝ. Fixing x∈ℝ+ and applying Ito’s formula to \(\{\varphi_{\varepsilon }(\tilde{u}^{L}_{\alpha}(t,x)): 0\le t<\tau_{L}\}\), we have that
Here φ ε (u ∘(x))=0 since u ∘≥0. Then (20) implies that
Next fix a nonincreasing ϖ∈C ∞(ℝ+) such that ϖ(x)=1 for x≤1 and ϖ(x)=0 for x≥2. For each N∈ℕ, define \(\varpi_{N}(x) \overset {\mathrm {def}}{=}\varpi(x/N)\). Let us now do several things. Let us multiply (21) with ϖ N . Let us then integrate in space, and finally take expectations. We get that
where
We need some bounds on φ ε . Define ∥⋅∥ C as the sup norm over ℝ. First note that \(\ddot{\varphi}(u) - 2\chi_{\mathbb {R}_{-}} = \eta(u)-2\chi_{\mathbb {R}_{-}}\). This implies that for all u∈ℝ,
the first bound is direct, and the second two follow by integration. Note also that since η is bounded,
Let us understand the behavior of the various terms of (22) as N→∞ and then ε→0. From the last bound of (23), we have that \(\lim_{\varepsilon \to 0}\varphi_{\varepsilon }(x)=x^{2}\chi_{\mathbb {R}_{-}}(x)\). Due to the last bound of (24), we can use dominated convergence and thus conclude that
We next consider \(A^{N,\varepsilon }_{1}(s)\). Integrating by parts and using the boundary conditions at x=0, we have that
The first term is nonpositive since η and ϖ are nonnegative. We can also see that
Thus
Thirdly, another integration by parts gives us that
and thus
Thus,
Let us finally bound \(A^{N,\varepsilon }_{3}\). Note that for every u∈ℝ,
In light of the first and last bounds of (24), we can use dominated convergence to see that
Combining things together, we finally get that
This implies the claimed result. □
Let us now see what happens as L↗∞. Define the random time
Let us also define
Lemma 5.8
(Positivity)
\(\tilde{u}(t,x)>0\) for all t≥0 and x>0.
Proof
We first define a transformation:
Since ζ t (x+β(t)) is an Ito process (recall (8)), by applying Ito’s formula and plugging into (14), we get
Since ζ t is smooth in space, we also have the following equalities:
Putting things together, we have that \(\tilde{u}^{*}(t,x)\) satisfies the random PDE
Now suppose there exists t 0>0,x 0>0 such that \(\tilde{u}(t_{0},x_{0})=\tilde{u}^{*}(t_{0},x_{0})=0\). In light of Lemma 5.7, since \(\tilde{u}^{*}(t,x)\ge 0, t\ge0, x\ge0\), we have that (t 0,x 0) is a minimum point of \(\tilde{u}^{*}(t,x)\). By applying Strong Parabolic Maximum Principle (cf. Theorem 2, p. 309, [17]) upon \(-\tilde{u}^{*}(t,x)\) on \(U_{t_{0},x_{0}}\overset {\mathrm {def}}{=}(0,t_{0}+1)\times[0,x_{0}+1]\) with M=0, we obtain that \(\tilde{u}^{*}(t,x)\equiv 0, (t,x)\in U_{t_{0},x_{0}}\), which contradicts the fact that u ∘(x)>0,x>0 and the continuity of \(\tilde{u}^{*}(t,x)\) at t=0. Therefore we have that \(\tilde{u}(t,x)>0,t\ge 0, x>0\). □
Lemma 5.9
We have that
Define u as in (15)–(16). Then {u(t,⋅)∣0≤t<τ} is a weak solution of (1).
Proof
Fixing L′>L we have from the uniqueness claim of Lemma 5.6 that \(\tilde{u}^{L'}(t,\cdot)=\tilde{u}^{L}(t,\cdot)\) for 0≤t≤τ L . Thus τ L′≥τ L for all L′>L, and so τ=lim L→∞ τ L =lim L→∞(τ L ∧L) and τ is predictable. We also have that \(\tilde{u}(t,\cdot) =\lim_{L\to \infty}\tilde{u}^{L}(t,\cdot)\) for 0≤t<τ. From this, Lemma 5.8, and Lemma 4.4, we conclude that {u(t,⋅)∣0≤t<τ} as defined by (15)–(16) indeed is a weak solution of (1). The characterization of \(\|\tilde{u}(t,\cdot)\|_{H}\) at τ− is obvious. □
In fact, we have a more explicit characterization of τ.
Lemma 5.10
We have that
Proof
For each L>0, define
Thus in fact \(\tau >\tau'_{L}\) and hence
Consequently,
Since \(\tau'_{L}< \tau\), we of course also have that \(\lim_{L\to \infty}\tau'_{L}\le \tau\). On the other hand, \(\|\tilde{u}(t,\cdot)\|_{H}\) may become large for many reasons other than \(|\tfrac{\partial \tilde{u}}{\partial x}(\tau'_{L},0)|\) becoming large, so necessarily \(\tau\le \lim_{L\to \infty}\tau'_{L}\). Putting things together, we get that \(\lim_{L\to \infty}\tau'_{L}=\tau\). The claimed result now follows. □
To finish things off, we prove uniqueness.
Lemma 5.11
(Uniqueness)
If \(\{\tilde{u}(t,\cdot)\mid 0\le t<\tau\}\subset H\) and \(\{\tilde{u}'(t,\cdot)\mid 0\le t<\tau'\}\subset H\) are two solutions of (13), then u(t,⋅)=u′(t,⋅) for 0≤t<min{τ,τ′}.
Proof
For each L>0, define
Then τ∧τ′=lim L→∞ σ L . We can use standard uniqueness theory to conclude that \(\tilde{u}\) and \(\tilde{u}'\) coincide on [0,σ L ], and we then let L↗∞. □
5.1 Proofs
We here give the delayed proofs. We start with the structural claims about H.
Proof of Lemma 5.1
The fact that H⊂C 1 is well known; [9]. Fix \(\varphi\in C^{\infty}_{0}(\mathbb {R}_{+})\), x∈(0,∞), and i∈{0,1}. We then have that
Thus
Of course we also have that
so the stated limits at x=0 exist. □
We next study {T t } t>0.
Proof of Lemma 5.2
The proof relies upon a combination of fairly standard calculations.
To begin, fix \(\varphi\in C_{0}^{\infty}(\mathbb {R}_{+})\) and define
Thus, u(t,x)=(T t φ)(x) for x>0, and since p ∘ is even in its second argument,
so in fact u(t,⋅) is odd. Thus we indeed have that \(\frac{\partial^{n} u}{\partial x^{n}}(t,0)=0\) for all even n∈ℕ; thus, T t φ∈H odd.
A standard calculation shows that T t is a contraction on H. Indeed, for each nonnegative integer n,
and thus
Summing these inequalities up for n∈{0,1,2,3}, we see that \(\|T_{t} \varphi\|^{2}_{H}\le \|\varphi\|^{2}_{H}\) for all \(\varphi\in C^{\infty}_{0}(\mathbb {R}_{+})\). This implies that T t is a contraction on \(C^{\infty}_{0}(\mathbb {R}_{+})\) and has the claimed extension.
To proceed, fix \(\varphi\in C^{\infty}_{0,\mathrm {odd}}(\mathbb {R}_{+})\). Note that thus y↦φ(|y|) is continuous. Define
We can now fairly easily conclude from (27) with n=0 that
Differentiating and integrating by parts as needed, we get that
We now note that there is a K>0 such that
for all t>0 and x∈ℝ. Thus,
Note now that
Combining this and (27) with n=0, we get that
Combine things together to get the last claim. □
6 Numerical Simulation
In this section, we will see from numerical simulations where the boundary is and how it is moving. In general, it is difficult to simulate the SPDE (1) directly since we need to find a solution of a stochastic heat equation and at the same time we need to trace the position of the moving boundary. Here we can avoid this difficulty since we have the explicit formula for the solution u in Lemma 4.4. That is,
where \((\tilde{u},\beta)\) is a solution of the SPDE
Therefore we first need to solve the SPDE (28) numerically in order to obtain the moving boundary β(t) and then the weak solution u(t,x). We first discretize space by using the explicit finite difference scheme. Here we can also approximate η by simple functions which converge to η in L 2(ℝ) (see [18]). As a result, we can have an approximation of ζ t (x). Note that ζ t (x) is a Brownian motion for each fixed x, however it is spatially correlated. Now we use the Euler–Maruyama type method to discretize time (see [11, 12]). Then we can get a numerical solution of (28). Since there is a stability issue for parabolic PDE, we note that Δt/(Δx)2<1/2, where Δt is a time step and Δx is a space step. Figure 1 is a simulation with initial condition
and α=0.5, ϱ=0.5. We can clearly see that there are two phases separated by the black line, which is the moving boundary, and how u is changing on the colored region where u>0. Furthermore, we can see that the boundary is moving left. This just follows from the positivity of the solution u(t,x) for x≥β(t) and the Stefan boundary condition of (1).
References
Barbu, V., Da Prato, G.: The two-phase stochastic Stefan problem. Probab. Theory Relat. Fields 124(4), 544–560 (2002)
Barbu, V., Da Prato, G., Röckner, M.: Existence of strong solutions for stochastic porous media equation under general monotonicity conditions. Ann. Probab. 37(2), 428–452 (2009)
Caffarelli, L., Salsa, S.: A Geometric Approach to Free Boundary Problems. Graduate Studies in Mathematics, vol. 68. American Mathematical Society, Providence (2005)
Caffarelli, L.A., Lee, K.-A., Mellet, A.: Homogenization and flame propagation in periodic excitable media: the asymptotic speed of propagation. Commun. Pure Appl. Math. 59(4), 501–525 (2006)
Chow, P.L.: Stochastic Partial Differential Equations. Chapman & Hall/CRC, Boca Raton (2007)
Da Prato, G., Röckner, M.: Invariant measures for a stochastic porous medium equation. In: Stochastic Analysis and Related Topics in Kyoto. Adv. Stud. Pure Math., vol. 41, pp. 13–29. Math. Soc. Japan, Tokyo (2004)
Da Prato, G., Röckner, M.: Weak solutions to stochastic porous media equations. J. Evol. Equ. 4(2), 249–271 (2004)
Da Prato, G., Röckner, M., Rozovskii, B.L., Wang, F.-Y.: Strong solutions of stochastic generalized porous media equations: Existence, uniqueness, and ergodicity. Commun. Partial Differ. Equ. 31(1–3), 277–291 (2006)
Evans, L.C.: Partial Differential Equations. American Mathematical Society, Providence (1998)
Friedman, A.: Partial Differential Equations of Parabolic Type. Prentice-Hall, Englewood Cliffs (1964)
Gaines, J.G.: Numerical experiments with S(P)DE’s. In: Stochastic Partial Differential Equations. London Math. Soc. Lecture Note Ser., vol. 216, pp. 55–71. Cambridge University Press, Cambridge (1995)
Higham, D.J.: An algorithmic introduction to numerical simulation of stochastic differential equations. SIAM Rev. 43(3), 525–546 (2001)
Karatzas, I., Shreve, S.E.: Brownian Motion and Stochastic Calculus, 2nd edn. Springer, New York (1991)
Kim, J.U.: On the stochastic porous medium equation. J. Differ. Equ. 220(1), 163–194 (2006)
Kim, K., Mueller, C., Sowers, R.B.: A stochastic moving boundary value problem. Illinois J. Math. (2011), to appear
Lunardi, A.: An introduction to parabolic moving boundary problems. In: Functional Analytic Methods for Evolution Equations. Lecture Notes in Math., vol. 1855, pp. 371–399. Springer, Berlin (2004)
McOwen, R.C.: Partial Differential Equations: Methods and Applications. Prentice-Hall, Upper Saddle (1996)
Walsh, J.B.: An introduction to stochastic partial differential equations. In: École d’été de probabilités de Saint-Flour, XIV–1984. Lecture Notes in Math., vol. 1180, pp. 265–439. Springer, Berlin (1986)
Author information
Authors and Affiliations
Corresponding author
Additional information
This work was supported by NSF grant DMS0705260. R.S. would like to thanks the Departments of Mathematics and Statistics of Stanford University for their hospitality in the Spring of 2010 during a sabbatical stay. The authors would like to thank the anonymous referee for his meticulous reading of the original version of this manuscript.
Rights and permissions
About this article
Cite this article
Kim, K., Zheng, Z. & Sowers, R.B. A Stochastic Stefan Problem. J Theor Probab 25, 1040–1080 (2012). https://doi.org/10.1007/s10959-011-0392-1
Received:
Revised:
Published:
Issue Date:
DOI: https://doi.org/10.1007/s10959-011-0392-1