Weak error of finite element approximations of a nonlinear stochastic heat equation Adam Andersson (joint work with Stig Larsson) Department of Mathematical Sciences Chalmers University of Technology and University of Gothenburg April 19, 2012 1 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . Let β ∈ ( 12 , 1] and α ∈ [0, β) 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . Let β ∈ ( 12 , 1] and α ∈ [0, β) I X0 ∈ Ḣ α , 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . Let β ∈ ( 12 , 1] and α ∈ [0, β) I X0 ∈ Ḣ α , I f ∈ Cb2 (H, H), 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . Let β ∈ ( 12 , 1] and α ∈ [0, β) I X0 ∈ Ḣ α , I f ∈ Cb2 (H, H), I g ∈ Cb2 (Ḣ α , HS(U0 , Ḣ β−1 )), i.e., kA β−1 2 1 g (x)Q 2 kHS < ∞, ∀x ∈ Ḣ α , 2 / 11 A semi-linear stochastic heat equation The stochastic evolution equation ( dX (t) + [AX (t) − f (X (t))] dt = g (X (t)) dW (t), t ∈ (0, T ], X (0) = X0 . D ⊂ Rd convex polygonal domain. H = L2 (D), Q ∈ L(H) selfadjoint nonnegative operator, {W (t)}t≥0 an H-valued Q-Wiener process on (Ω, F, {Ft }t≥0 , P), U0 = Q 1/2 (H), L02 = HS(U0 , H), A = −∆, D(A) = H01 (D) ∩ H 2 (D). For γ ≥ 0, Ḣ γ = D(Aγ/2 ) and Ḣ −γ = (Ḣ γ )∗ . Let β ∈ ( 12 , 1] and α ∈ [0, β) I X0 ∈ Ḣ α , I f ∈ Cb2 (H, H), I g ∈ Cb2 (Ḣ α , HS(U0 , Ḣ β−1 )), i.e., kA I g 00 = 0. β−1 2 1 g (x)Q 2 kHS < ∞, ∀x ∈ Ḣ α , 2 / 11 Mild solution Let {E (t)}t≥0 be the analytic semigroup generated by −A. ∃! solution X ∈ C ([0, T ], L2 (Ω, Ḣ β )) to the mild equation Z t E (t − s)f (X (s)) ds X (t) = E (t)X0 + 0 Z t + E (t − s)g (X (s)) dW (s), t ∈ (0, T ]. 0 3 / 11 Approximation by the finite element method A discretized equation: ( dXh (t) + [Ah Xh (t) − Ph f (Xh (t))] dt = Ph g (Xh (t)) dW (t), t ∈ (0, T ] Xh (0) = Ph X0 . Finite element spaces {Sh }h∈(0,1] of continuous piecewise linear functions corresponding to a quasi-uniform family of triangulations of D. Ah is the discrete Laplacian satisfying hAh ψ, χiH = h∇ψ, ∇χiH , ∀ψ, χ ∈ Sh . Ph : H → Sh orthogonal projection w.r.t. h·, ·iH . 4 / 11 Mild solution of discretized equation Let {Eh (t)}t≥0 be the analytic semigroup generated by −Ah . For every h ∈ (0, 1] ∃! solution Xh ∈ C ([0, T ], L2 (Ω, Sh )) to the mild equation Z t Xh (t) = Eh (t)Ph X0 + Eh (t − s)Ph f (X (s)) ds 0 Z t + Eh (t − s)Ph g (X (s)) dW (s), t ∈ (0, T ]. 0 5 / 11 Malliavin integration by parts formula For all F ∈ D1,2 (H) and Φ ∈ L2 (Ω × [0, T ], L02 ) Z T D Z T E E F, Φ(t) dW (t) = E Tr{QΦ(t)∗ Dt F } dt 0 0 = hDF , ΦiL2 (Ω×[0,T ],L02 ) . In particular for u ∈ C 2 (H, R) Z D E ux (Xh (t)), 0 T Z E Φ(s) dW (s) = E T Tr{QΦ(s)∗ uxx (F )Ds Xh (t)} ds. 0 6 / 11 Malliavin integration by parts formula For all F ∈ D1,2 (H) and Φ ∈ L2 (Ω × [0, T ], L02 ) Z T D Z T E E F, Φ(t) dW (t) = E Tr{QΦ(t)∗ Dt F } dt 0 0 = hDF , ΦiL2 (Ω×[0,T ],L02 ) . In particular for u ∈ C 2 (H, R) Z D E ux (Xh (t)), T 0 Dsu Xh (t) T Z E Φ(s) dW (s) = E Tr{QΦ(s)∗ uxx (F )Ds Xh (t)} ds. 0 Z t = Eh (t − s)Ph σ(Xh (s))u + Eh (t − r )Ph f 0 (Xh (r ))Dsu Xh (r ) dr 0 Z t + Eh (t − r )Ph σ 0 (Xh (r )) · Dsu Xh dW (r ). 0 6 / 11 Main result Theorem For every test function Φ ∈ Cb2 (H, R) and > 0 there exist a C > 0 such that E[Φ(X (T )) − Φ(Xh (T ))] ≤ Ch2β− . 7 / 11 Main result Theorem For every test function Φ ∈ Cb2 (H, R) and > 0 there exist a C > 0 such that E[Φ(X (T )) − Φ(Xh (T ))] ≤ Ch2β− . Sketch of proof: Method by A. Debussche (2011). 7 / 11 Main result Theorem For every test function Φ ∈ Cb2 (H, R) and > 0 there exist a C > 0 such that E[Φ(X (T )) − Φ(Xh (T ))] ≤ Ch2β− . Sketch of proof: Method by A. Debussche (2011). Assume that f = 0 and X0 = 0. 7 / 11 Main result Theorem For every test function Φ ∈ Cb2 (H, R) and > 0 there exist a C > 0 such that E[Φ(X (T )) − Φ(Xh (T ))] ≤ Ch2β− . Sketch of proof: Method by A. Debussche (2011). Assume that f = 0 and X0 = 0. Let u(t, x) = EΦ(X (t, x)), t ∈ [0, T ] and x ∈ H where X (0, x) = x. Satisfies the Kolmogorov equation ut (t, x) + Lu(t, x) = 0, u(0, x) = Φ(x), t ∈ [0, T ), x ∈ H, x ∈ H. Here Lu(t, x) = hAx, ux (t, x)i − 1 Tr{g (x)Qg ∗ (x)uxx (t, x)}. 2 7 / 11 By Itô’s formula and the Kolmogorov equation E[Φ(X (T )) − Φ(Xh (T ))] = E[u(T , X0 ) − u(0, Xh (T ))] hZ T i =E ut (T − s, Xh (s)) + Lh u(T − s, Xh (s)) ds 0 T hZ =E (Lh − L)u(T − s, Xh (s)) ds i 0 T hZ =E h(Ah − A)Xh (s), ux (T − s, Xh (s))i ds 0 1 − 2 Z T Tr{(Ph g (Xh (s))Qg ∗ (Xh (s))Ph 0 −g (Xh (s))Qg ∗ (Xh (s)))uxx (t, Xh (s))} ds i = I +J 8 / 11 Z I ≤ E T Z = E T Z = E T 0 0 D E Ah Ph (I − Rh )ux (T − s, Xh (s)), Xh (s) ds (Rh = A−1 h Ph A) D ux (T − s, Xh (s)), Z s E (Ah Ph (I − Rh ))∗ Eh (s − r )Ph g (Xh (r )) dW (r ) ds 0 Z s n Tr Qg ∗ (Xh (r ))Ph Eh (s − r )Ah Ph (I − Rh ) 0 0 o uxx (T − s, Xh (s))Dr Xh (s) dr ds Z TZ s β−1 β−1 1−β 1 kQ 2 g ∗ (Xh (s))A 2 kHS kA 2 Ph Ah 2 kL(H) ≤E 0 × 0 kA1− h Eh (s × kA × kA 1+β− 2 β−1 2 ≤ Ch2β−3 Z 0 1−β+2 2 − r )Ph kL(H) kAh uxx (T − s, Xh (s))A 1−β 2 Ph (I − Rh )A− 1+β− 2 kL(H) kL(H) 1 2 Dr Xh (s)Q kHS dr ds T Z s 2− (T − s)− 2 (s − r )−1+ dr ds. 0 9 / 11 Estimates Let 0 ≤ s ≤ r ≤ 1. Then s r kAh2 Ph (I − Rh )A− 2 kL(H) ≤ Chr −s , h ∈ (0, 1]. 10 / 11 Estimates Let 0 ≤ s ≤ r ≤ 1. Then s r kAh2 Ph (I − Rh )A− 2 kL(H) ≤ Chr −s , h ∈ (0, 1]. Let γ ≥ 0. Then kAγ E (t)kL(H) + kAγh Eh (t)Ph kL(H) ≤ Ct −γ , ∀t > 0. 10 / 11 Estimates Let 0 ≤ s ≤ r ≤ 1. Then s r kAh2 Ph (I − Rh )A− 2 kL(H) ≤ Chr −s , h ∈ (0, 1]. Let γ ≥ 0. Then kAγ E (t)kL(H) + kAγh Eh (t)Ph kL(H) ≤ Ct −γ , ∀t > 0. Let λ ∈ [0, 1+β 2 ). Then kAλ ux (t, x)kL(H) ≤ Ct −λ |Φ|Cb1 , ∀t ∈ (0, T ], ∀x ∈ H. 10 / 11 Estimates Let 0 ≤ s ≤ r ≤ 1. Then s r kAh2 Ph (I − Rh )A− 2 kL(H) ≤ Chr −s , h ∈ (0, 1]. Let γ ≥ 0. Then kAγ E (t)kL(H) + kAγh Eh (t)Ph kL(H) ≤ Ct −γ , ∀t > 0. Let λ ∈ [0, 1+β 2 ). Then kAλ ux (t, x)kL(H) ≤ Ct −λ |Φ|Cb1 , ∀t ∈ (0, T ], ∀x ∈ H. Let λ, ρ ∈ [0, 1+β 2 ) be such that λ + ρ < 1. Then kAρ uxx (t, x)Aλ kL(H) ≤ Ct −(ρ+λ) |G |Cb2 , ∀t ∈ (0, T ], ∀x ∈ H. 10 / 11 Estimates Let 0 ≤ s ≤ r ≤ 1. Then s r kAh2 Ph (I − Rh )A− 2 kL(H) ≤ Chr −s , h ∈ (0, 1]. Let γ ≥ 0. Then kAγ E (t)kL(H) + kAγh Eh (t)Ph kL(H) ≤ Ct −γ , ∀t > 0. Let λ ∈ [0, 1+β 2 ). Then kAλ ux (t, x)kL(H) ≤ Ct −λ |Φ|Cb1 , ∀t ∈ (0, T ], ∀x ∈ H. Let λ, ρ ∈ [0, 1+β 2 ) be such that λ + ρ < 1. Then kAρ uxx (t, x)Aλ kL(H) ≤ Ct −(ρ+λ) |G |Cb2 , ∀t ∈ (0, T ], ∀x ∈ H. ∀t ∈ [0, T ], ∀s ∈ [0, t). Let γ ∈ [0, β). Then β−1+γ 2 E[kAh 1 Ds Xh (t)Q 2 k2HS ] ≤ C (t − s)−γ , 10 / 11 Thank you for your attention! 11 / 11