Some inverse source problems in semilinear fractional PDEs
Marián Slodicka
Ghent University (Belgium)Faculty of Engineering and Architecture
Research Group forNumerical Analysis and Mathematical Modeling
[email protected]://cage.ugent.be/~ms
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 1 / 43
Outline
1 Inverse source problems
2 Fractional calculus
3 ISP for time-fractional parabolic equation
4 Time-fractional hyperbolic ISP
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 2 / 43
Inverse source problems
Introduction
ν Γ
Ω
ΓD
N
Example: A general linear parabolic PDE with mixed BCs
ut +∇ · (−Adif∇u − aconu) + asouu = f +∇ · f div in Ωu = gDir on ΓD
(−Adif∇u − aconu)Tν − gRobu = gNeu on ΓN
u(x ,0) = u0(x) in Ω
Game of IPsWhat is known/unknown?Additional dataWell-posednessHow to reconstruct missing data?
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 3 / 43
Inverse source problems
ISPs in the literature
Consider a PDE of the type Lu(t , x) = f (t , x).Literature research gives many of papers devoted to the recovery of
f (x) based on e.g. on final measurement [Rundell, 1980, Cannon, 1968,Prilepko and Solov’ev, 1988, Solov’ev, 1990, Isakov, 1990, Farcas and Lesnic, 2006,Hasanov, 2007, Johansson and Lesnic, 2007a, Johansson and Lesnic, 2007b]. . .
f (t) based on local or non-local measurement.[Prilepko et al., 2000, Hasanov, 2011, Yang et al., 2011,Hasanov and Slodicka, 2013, Slodicka, 2013, Hazanee et al., 2013, Hasanov, 2011,Hasanov and Pektas, 2013]. . .
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 4 / 43
Inverse source problems
Example spectral analysis . . .Let us consider the following homogeneous problem in Ω = (0,1)
−u′′(x) = f (x) x ∈ Ω,u(0) = u(1) = 0. (1)
We denote by A : D(A)→ X the second order differential operator, where
A = − d2
dx2 , D(A) = H2(Ω) ∩ H10 (Ω)
and X = L2(Ω). The spectrum σ(A) of the operator A consists of the eigenvalues
λn = π2n2, n ∈ N.
The corresponding eigenfunctions have the form
en(x) =√
2 sin(nπx), n ∈ N.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 5 / 43
Inverse source problems
. . . Example spectral analysis
The set of all eigenfunctions is an orthonormal complete system in L2(Ω). Thus
(en,em) =
∫Ω
en(x)em(x) dx = δn,m1.
Each function f ∈ L2(Ω) can be written as
f =∞∑i=1
(f ,ei )ei ,
where (·, ·) is the inner product in X
(f ,ei ) =
∫Ω
f (x)ei (x) dx .
1This is the Kronecker symbol δn,m = 1 if n = m, else δn,m = 0.M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 6 / 43
Inverse source problems
Example ISP . . .
IS Problem: Find (u(t , x),h(t)) such that
ut − u′′(x) = h(t)f (x) x ∈ Ω = (0,1),u(0) = u(1) = 0
u(0, x) = 0u(t , x0) = m(t) x0 ∈ Ω
(2)
Question: Is the solution unique?Answer: If yes, then m(t) = 0 =⇒ (u(t , x),h(t)) = (0,0).
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 7 / 43
Inverse source problems
. . . Example ISP . . .Take an eigenfunction w of the operator Au = −u′′, i.e. Aw = λw . Set f = w . Seek the solution uin the form u(t , x) = α(t)w(x). From the PDE we get
[α′(t) + λα(t)] w(x) = h(t)w(x)
Thus α solvesα′(t) + λα(t) = h(t); α(0) = 0.
Therefore
α(t) =
∫ t
0e−λ(t−s)h(s) ds.
Non uniqueness: If x0 is a zero point of w(x), i.e. w(x0) = 0, which implies m(t) = 0, but we haveat least 2 solutions
(u,h) = (0,0), (u,h) =
(w(x)
∫ t
0e−λ(t−s)h(s) ds,h(t)
).
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 8 / 43
Inverse source problems
. . . Example . . .
Bad choice of a measurement point: Zero point of an eigenfunction.How many eigenfunctions do I have?: ∞Recall that
en(x) =√
2 sin(nπx), n ∈ N.
What are all zero poins of all eigenfunctions?Solving
0 = sin(nπx) =⇒ nπx = mπ m ∈ Z
we getx =
mn.
Zero points of all eigenfunctions are dense in Ω.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 9 / 43
Inverse source problems
. . . Example
Which condition ensures the uniqueness for the ISP? A nonlocal one
(u(t),1)ω :=
∫ω
u(t , x) dx = m(t) ω ⊂ Ω, f ∈ C10 (Ω).
Multiply PDE by 1 and integrate over ω to see
(∂tu(t),1)ω − (u′′(t),1)ω = h(t)(f ,1)ω=⇒h(t) = . . .
Multiply PDE by −u′′ and integrate over Ω to see (m(t) = 0 by uniqueness)
12∂t ‖u′(t)‖
2+ ‖u′′(t)‖2
=(u′′(t),1)ω
(f ,1)ω(f ,u′′) =
−(u′′(t),1)ω(f ,1)ω
(f ′,u′)
Young’s inequality12∂t ‖u′(t)‖
2+ (1− ε) ‖u′′(t)‖2 ≤ Cε ‖u′(t)‖
2.
Grönwall’s lemma implies the uniqueness.M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 10 / 43
Inverse source problems
Rothe’s method as a semigroup
Example 1 ( D(A) = H2(Ω) ∩ H10 (Ω), t ∈ [0,T ],u(0) = u0)
Idea for x ∈ R: ex = limn→∞
(1 +
xn
)n. Set τ = T
n .
continue problem Rothe’s method
∂tu + Au = f (u) δui + Aui =ui−ui−1
τ + Aui = f (ui−1)
S(t) := e−At Sτ (t) := (I + τA)tτ
u(t) = S(t)u0 +
∫ t
0S(t − s)f (u(s)) ds ui = Sτ (ti )u0 +
i−1∑k=0
Sτ (ti − tk )f (uk )τ
Semigroups S(t) := e−At , Sτ (t) := (I + τA)tτ
Error ‖ui − u(ti )‖ ≤ C(∥∥Aβu0
∥∥)τmin1,β
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 11 / 43
Fractional calculus
Fractional derivatives . . .What do we need?
D0 = I, Aditivity DαDβ = Dα+β , for any α, β ∈ R+
Restriction of the fractional operator to natural numbers coincides with the classical derivative
Dα =dα
dxαfor α ∈ N
Assume that α ≥ 0, x > aRiemann-Liouville
(Dαa y) (x) :=
(DnIn−α
a y)
(x)
=1
Γ(n − α)
(ddx
)n ∫ x
a
y(t)(x − t)α−n+1 dt
Caputo (CDα
a y)
(x) :=(In−αa Dny
)(x)
=1
Γ(n − α)
∫ x
a
y (n)(t)(x − t)α−n+1 dt
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 12 / 43
Fractional calculus
. . . Fractional derivatives
Theorem 2 (Relationship)
Let α ≥ 0, y ∈ ACn([a,b]). Then
(CDα
a y)
(x) = (Dαa y) (x)−
n−1∑k=0
y (k)(a)
Γ(k − α + 1)(x − a)k−α.
PropertiesAditivity DαDβ = Dα+β , for any α, β ∈ R+
Caputo = Sturm-Liouville if
y (k)(a) = 0 for k = 0, . . . ,n − 1
How to get a priori estimates?
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 13 / 43
Fractional calculus
Example of a convolution kernelAssume v ∈ C[0,∞) and T > 0. It holds
ddt
[∫ t
0e−(t−s)v(s) ds
]= v(t)−
∫ t
0e−(t−s)v(s) ds.
Soddt
[∫ t
0e−(t−s)v(s) ds
]2
= 2
[∫ t
0e−(t−s)v(s) ds
][v(t)−
∫ t
0e−(t−s)v(s) ds
].
Integration in time over [0,T ] gives
2∫ T
0v(t)
∫ t
0e−(t−s)v(s) ds dt =
[∫ T
0e−(T−s)v(s) ds
]2
+
∫ T
0
[∫ t
0e−(t−s)v(s) ds
]2
dt≥ 0.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 14 / 43
Fractional calculus
Positive kernels [Nohel and Shea, 1976]
Let a(t) ∈ L1loc(0,∞) is of positive type if∫ T
0v(t)
∫ t
0v(ξ)a(t − ξ)dξ dt ≥ 0
for any v ∈ C[0,∞) and any T > 0.
A real function a(t) is strongly positive if there exists η > 0 such thatb(t) = a(t)− ηe−t is of a positive type.
Let a(t) ∈ L1loc(0,∞) be not constant, ≥ 0, nonincreasing, convex and
such that da′(t) is not a purely singular measure. Then a(t) is stronglypositive.
In particular, twice-differentiable a(t) satisfying
(−1)k a(k)(t) ≥ 0; 0 < t <∞, k = 0,1,2; a′ 6≡ 0
are strongly positive.M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 15 / 43
Fractional calculus
Caputo derivative as a convolutionDefine (Riemann-Liouville kernel)
gβ(t) :=tβ−1
Γ(β), t > 0, β > 0
which is strongly positive definite.
By K ∗ u we denote the usual convolution in time, namely
(K ∗ u(x))(t) =
∫ t
0K (t − s)u(x , s) ds.
The Caputo fractional derivative can be also rewritten as a convolution with gβ
∂αt v(t) =∂αv∂tα
:=
(g1−α ∗ ∂tv) (t), α ∈ (0,1)(g2−α ∗ ∂ttv) (t), α ∈ (1,2)
∂tv(t), α = 1
How to get a priori estimates?M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 16 / 43
Fractional calculus
Zacher’s lemma
Let H be a real Hilbert space with a scalar product (·, ·)H with the corresponding norm ‖·‖H .
[Zacher, 2010, Lemma 2.3.2], [Zacher, 2008, Zacher, 2013] proved the following identity
( ddt (k ∗ v)(t), v(t)
)H = 1
2ddt
(k ∗‖v‖2
H
)(t) + 1
2 k(t) ‖v(t)‖2H
+ 12
∫ t
0[−k ′(s)] ‖v(t)− v(t − s)‖2
H ds a.e. t ∈ (0,T ),
which is valid for any k ∈ H1,1([0,T ]) and each v ∈ L2([0,T ],H).
The assumption k ∈ H1,1([0,T ]) is too strong. What now?
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 17 / 43
Fractional calculus
Crucial lemma . . .
[Slodicka and Šišková, 2016] CAMWAThe following crucial lemma which will play a central role in the proofs.
Lemma 3
Let H be a real Hilbert space with a scalar product (·, ·)H and the corresponding norm ‖·‖H .Assume T > 0, g ∈ L1(0,T ), g′ ∈ L1,loc(0,T ), g′ ≤ 0, g ≥ 0. If v : [0,T ]→ H such that v(0) ∈ H,v ∈ H1((0,T ),H) then∫ ξ
0
(ddt
(g∗ v) (t), v(t))
Hdt ≥ 1
2
(g∗‖v‖2
H
)(ξ) + 1
2
∫ ξ
0g(t) ‖v(t)‖2
H dt
≥ g(T )
2
∫ ξ
0‖v(t)‖2
H dt
for any ξ ∈ [0,T ].
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 18 / 43
Fractional calculus
. . . Crucial lemma
Proof.
Start from the Zacher’s lemma and replace k by gn(s) := minn,g(s).It holds
g′n(s) ≤ 0, gn(s)→ g(s) a.e. in [0,T ].
Integrate in timePass to the limit for n→∞.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 19 / 43
ISP for time-fractional parabolic equation
Inverse source problem . . .Differential operator Ω ⊂ Rd , t ∈ [0,T ]
L(x , t)u = ∇ · (−A(x , t)∇u − b(x , t)u) + c(t)u,A(x , t) = (ai,j (x , t))i,j=1,...,d ,
b(x , t) = (b1(x , t), . . . ,bd (x , t)).
Governing PDE
(g1−β ∗ ∂tu(x)) (t) + L(x , t)u(x , t) = h(t)f (x) +
∫ t
0F (x , s,u(x , s)) ds, (3)
where g1−β denotes the Riemann-Liouville kernel
g1−β(t) =t−β
Γ(1− β), t > 0, 0 < β < 1
IC & BCu(x ,0) = u0(x), x ∈ Ω
(−A(x , t)∇u(x , t)− b(x , t)u(x , t)) · ν = g(x , t) (x , t) ∈ Γ× (0,T ).(4)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 20 / 43
ISP for time-fractional parabolic equation
. . . Inverse source problem
Find (u(x , t),h(t)) obeying (3), (4).The unknown time-dependent function h(t) will be determined from the following additionalmeasurement
m(t) =
∫Ω
u(x , t) dx = (u(t),1) , t ∈ [0,T ]. (5)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 21 / 43
ISP for time-fractional parabolic equation
Variational framework . . .
We associate a bilinear form L with the differential operator L as follows
(Lu, ϕ) = L (u, ϕ) + (g, ϕ)Γ , ∀ϕ ∈ H1(Ω),
i.e.L(t) (u(t), ϕ) = (A(t)∇u(t) + b(t)u(t),∇ϕ) + c(t) (u(t), ϕ) .
Throughout the paper we assume that
ai,j ,bi : Ω× [0,T ]→ R, |ai,j |+ |bi | ≤ C, i , j = 1, . . . ,d ,0 ≤ c(t) ≤ C, ∀t ∈ [0,T ],
L(t) (ϕ,ϕ) ≥ C0 ‖∇ϕ‖2, ∀ϕ ∈ H1(Ω), ∀t ∈ [0,T ].
(6)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 22 / 43
ISP for time-fractional parabolic equation
. . . Variational frameworkIntegrate (3) over Ω, apply Green’s theorem and use (5) to get
(g1−β ∗m′) (t) + c(t)m(t) = h(t) (f ,1)− (g(t),1)Γ +
∫ t
0(F (s,u(s)),1) ds. (MP)
Assuming that (f ,1) 6= 0 we have
h(t) =
(g1−β ∗m′) (t) + c(t)m(t) + (g(t),1)Γ −∫ t
0(F (s,u(s)),1) ds
(f ,1). (7)
The variational formulation of (3) and (4) reads as
((g1−β ∗ ∂tu) (t), ϕ) + L(t) (u(t), ϕ)
= h(t) (f , ϕ) +
(∫ t
0F (s,u(s)) ds, ϕ
)− (g(t), ϕ)Γ
(P)
for any ϕ ∈ H1(Ω), a.a. t ∈ [0,T ] and u(0) = u0.M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 23 / 43
ISP for time-fractional parabolic equation
Uniqueness
Theorem 4
Let f ,u0 ∈ L2(Ω),∫
Ωf 6= 0, m ∈ C1([0,T ]), F be a global Lipschitz continuous function in all
variables. Assume (6) and g ∈ C([0,T ], Γ).Then there exists at most one solution (u,h) to the (P), (MP) obeyingu ∈ C
([0,T ],L2(Ω)
)∩ L∞
((0,T ),H1(Ω)
)with ∂tu ∈ L2
((0,T ),L2(Ω)
), h ∈ C([0,T ]).
Proof.
Suppose that (ui ,hi ) for i = 1,2 solve (P), (MP).Set u = u1 − u2 and h = h1 − h2.Subtract the corresponding variational formulations from each other.Set ϕ = u(t) in variational formulation and integrate in time over (0, ξ).Use crucial lemma.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 24 / 43
ISP for time-fractional parabolic equation
Time discretization, Discrete convolutionEquidistant time-partitioning of [0,T ] with a step τ = T/n, for any n ∈ N. Set ti = iτ and denote
zi = z(ti ), δzi =zi − zi−1
τ.
Let us define the discrete convolution in time as follows
(K ∗ v)i :=i∑
k=1
Ki+1−k vkτ.
Please note that this definition allows blow up of K at t = 0. An easy calculation yields
δ (K ∗ v)i =(K ∗ v)i − (K ∗ v)i−1
τ= K1vi +
i−1∑k=1
δKi+1−k vkτ, i ≥ 1 (8)
as (K ∗ v)0 := 0. Similarly we may write
δ (K ∗ v)i = Kiv0 +i∑
k=1
δvk Ki+1−kτ = Kiv0 + (K ∗ δv)i , i ≥ 1. (9)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 25 / 43
ISP for time-fractional parabolic equation
Discrete crucial lemma
The following technical lemma is a discrete analogy of Lemma 3. It plays a central role byestablishing a priori estimates for ui and hi .
Lemma 5
Let vii∈N and Kii∈N be sequences of real numbers. Assume that K decreases, i.e. Ki ≤ Ki−1for any i. Then
2δ (K ∗ v)i vi ≥ δ(K ∗ v2)
i + Kiv2i , i ∈ N.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 26 / 43
ISP for time-fractional parabolic equation
Discrete problem
Consider a system with unknowns (ui ,hi ) for i = 1, . . . ,n. At time ti we approximate (P) by
((g1−β ∗ δu)i , ϕ
)+ Li (ui , ϕ) = hi (f , ϕ) +
(i∑
k=1
F (tk ,uk−1)τ, ϕ
)− (gi , ϕ)Γ (DPi)
and (MP) by
(g1−β ∗m′)i + cimi = hi (f ,1) +
(i∑
k=1
F (tk ,uk−1)τ,1
)− (gi ,1)Γ . (DMPi)
Considering uk−1 in the argument of F makes (DPi) linear in ui .The decoupling of ui and hi has been achieved by considering uk−1 in (DMPi).For a given i ∈ 1, . . . ,n solve first (DMPi) and then (DPi). Then increase i to i + 1.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 27 / 43
ISP for time-fractional parabolic equation
Existence of (ui ,hi)
Lemma 6
Let f ,u0 ∈ L2(Ω),∫
Ωf 6= 0, m ∈ C1([0,T ]), F be a global Lipschitz continuous function in all
variables. Assume (6) and g ∈ C([0,T ], Γ). Then for each i ∈ 1, . . . ,n there exists a uniquecouple (ui ,hi ) ∈ H1(Ω)× R solving (DPi) and (DMPi).
Proof.Resolving (DMPi) for hi we get
hi =(g1−β ∗m′)i + cimi + (gi ,1)Γ −
(∑ik=1 F (tk ,uk−1)τ,1
)(f ,1)
∈ R. (10)
Use Lax-Milgram lemma for (DPi).
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 28 / 43
ISP for time-fractional parabolic equation
Stability analysis . . .
Introduce the following notation
(g1−β ∗‖u‖2
)j
=
j∑k=1
g1−β(tj+1−k ) ‖uk‖2τ.
Lemma 7
Let the assumptions of Lemma 6 be fulfilled. Then there exist positive constants C and τ0 suchthat for any 0 < τ < τ0 we have
(i) max1≤j≤n
(g1−β ∗‖u‖2
)j
+n∑
i=1
g1−β(ti ) ‖ui‖2τ +
n∑i=1
‖ui‖2H1(Ω) τ ≤ C,
(ii) max1≤j≤n
|hj | ≤ C.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 29 / 43
ISP for time-fractional parabolic equation
. . . Stability analysis . . .
Compatibility condition: Assume that the (3) is fulfilled at t = 0, i.e. (P) holds true for t = 0.Therefore we may also put t = 0 in (MP), which allows us to define h0 as follows
h0 =c0m0 + (g0,1)Γ
(f ,1). (11)
Lemma 8
Let the assumptions of Lemma 6 be fulfilled. Moreover assume (11), u0 ∈ H1(Ω), g ∈ C1([0,T ], Γ),m ∈ C2([0,T ]), ∂tc ∈ L∞(0,T ) and ∂tai,j , ∂tbi ∈ L∞(Ω× (0,T )) for all i , j = 1, . . . ,d. Then thereexist positive constants C and τ0 such that for any 0 < τ < τ0 we have
(i) max1≤j≤n
(g1−β ∗‖δu‖2
)j
+n∑
i=1
g1−β(ti ) ‖δui‖2τ +
n∑i=1
‖δui‖2H1(Ω) τ ≤ C,
(ii) |δhi | ≤ C + Ct−βi for any i = 1, . . . ,n.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 30 / 43
ISP for time-fractional parabolic equation
Existence of a solution
Theorem 9
Let f ∈ L2(Ω), u0 ∈ H1(Ω),∫
Ωf 6= 0, m ∈ C2([0,T ]), and g ∈ C1([0,T ], Γ). Suppose that F is a
global Lipschitz continuous function in all variables. Assume (6), (11), ∂tc ∈ L∞[0,T ] and∂tai,j , ∂tbi ∈ L∞(Ω× (0,T )) for all i , j = 1, . . . ,d.Then there exists a solution (u,h) to the (P), (MP) obeying u ∈ C
([0,T ],H1(Ω)
)with
∂tu ∈ L2((0,T ),H1(Ω)
), h ∈ C([0,T ]).
Proof.Cauchy, Young inequalities; Lebesgue dominated theoremConvergence (functional analysis)
Noisy data Regularization m ≈ mε ∈ C2
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 31 / 43
ISP for time-fractional parabolic equation
Example . . .
We consider problem (P)-(MP) for Ω = (0.5,3), T = 3 and β = 0.5 with
L (u, ϕ) = (∇u,∇ϕ) ,f (x) = sin x ,
F (x , t ,u) = −4tu exp(
1− u2
sin2 x
),
along with the initial and boundary conditions
u0(x) = 2 sin x ,g(0.5, t) = (t2 + 2) cos 1
2 ,
g(3, t) = (t2 + 2) cos 3,
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 32 / 43
ISP for time-fractional parabolic equation
. . . Example . . .
where the time-dependent measurement is
m(t) =
(cos
12− cos 3
)(t2 + 2
).
One can easily verify that functions
u(x , t) =(t2 + 2
)sin x
andh(t) =
83√π
t32 + t2 − exp
(1− (t2 − 2)2)+ e−3 + 2
solve the given problem.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 33 / 43
ISP for time-fractional parabolic equation
. . . Example . . .
Discretization parametersΩ is uniformly divided into 50 subintervalsThe solution ui is calculated using a finite element method with Lagrange polynomials of thesecond order used as basis functions.Calculations were made several times for various values of τ .
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 34 / 43
ISP for time-fractional parabolic equation
. . . Example . . .
Figure: Decay of maximal relative error
(a) Logarithm of maximal relative error in time of h fordifferent values of τ . Slope of the line is 0.39529.
(b) Logarithm of maximal relative error in time of u fordifferent values of τ . Slope of the line is 0.99983.M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 35 / 43
Time-fractional hyperbolic ISP
Inverse source problem . . .
[Šišková and Slodicka, 2017] APNUMGoverning PDE (x ∈ Ω ⊂ Rd , t ∈ (0,T ))
(g2−β ∗ ∂ttu(x)) (t)−∆u(x , t) = h(t)f (x) + F (x , t ,u(x , t)) (12)
BC & ICsu(x ,0) = u0(x), x ∈ Ω,
∂tu(x ,0) = v0(x), x ∈ Ω,−∇u(x , t) · ν = g(x , t), (x , t) ∈ Γ× (0,T ),
(13)
The unknown time-dependent function h(t) will be determined from the following additionalmeasurement ∫
Ω
u(x , t)ω(x) dx = m(t), t ∈ [0,T ], (14)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 36 / 43
Time-fractional hyperbolic ISP
Variational setting
Multiplying (12) by the function ω, integrating over Ω, applying the Green theorem and using (14),we obtain
(g2−β ∗m′′) (t) + (∇u(t),∇ω) = h(t) (f , ω)− (g(t), ω)Γ + (F (t ,u(t)), ω) . (MP)
Similarly multiplying (12) by a function ϕ ∈ H1(Ω) and using Green’s theorem, we obtain thevariational formulation of (12) and (13)
((g2−β ∗ ∂ttu) (t), ϕ) + (∇u(t),∇ϕ) = h(t) (f , ϕ) + (F (t ,u(t)), ϕ)− (g(t), ϕ)Γ , (P)
for any ϕ ∈ H1(Ω), a.a. t ∈ [0,T ] and u(0) = u0, ∂tu(0) = v0.The relations (P) and (MP) represent the variational formulation of the ISP (12), (13) and (14).
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 37 / 43
Time-fractional hyperbolic ISP
Time discretization
The discrete convolution is defined by
(K ∗ v)i :=i∑
k=1
Ki+1−k vkτ,
note that by this definition we avoided problems with a blow up if K has a singularity at t = 0. Thenwe can calculate a difference for the discrete convolution as follows
δ (K ∗ v)i =(K ∗ v)i − (K ∗ v)i−1
τ= K1vi +
i−1∑k=1
δKi+1−k vkτ, i ≥ 1, (15)
as(K ∗ v)0 := 0
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 38 / 43
Time-fractional hyperbolic ISP
Schema
On the i−th time-layer we approximate the solution of (P),(MP) by (ui ,hi ), which solves((g2−β ∗ δ2u
)i , ϕ)
+ (∇ui ,∇ϕ) = hi (f , ϕ) + (F (ti ,ui−1), ϕ)− (gi , ϕ)Γ , (DPi)
for ϕ ∈ H1(Ω), with δu0 := v0 and
(g2−β ∗m′′)i + (∇ui−1,∇ω) = hi (f , ω) + (F (ti ,ui−1), ω)− (gi , ω)Γ . (DMPi)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 39 / 43
Time-fractional hyperbolic ISP
Solvability of the ISP
Theorem 10
Let f ∈ L2(Ω), u0, v0, ω ∈ H1(Ω),∫
Ωfω 6= 0, m ∈ C3([0,T ]), and g ∈ C2([0,T ], Γ). Suppose that F
is a global Lipschitz continuous function in all variables and (11) holds true.Then there exists a solution (u,h) to the (P), (MP) obeying u ∈ C
([0,T ],H1(Ω)
)with
∂tu ∈ C([0,T ],L2(Ω)
)∩ L∞
((0,T ),H1(Ω)
), ∂ttu ∈ L2
((0,T ),L2(Ω)
)and h ∈ C([0,T ]).
Theorem 11
Let f , v0 ∈ L2(Ω),u0, ω ∈ H1(Ω),∫
Ωfω 6= 0, m ∈ C2([0,T ]), F be a global Lipschitz continuous
function in all variables and g ∈ C([0,T ], Γ). Then there exists at most one solution (u,h) to the(P), (MP) obeying u ∈ C
([0,T ],H1(Ω)
), ∂tu ∈ C
([0,T ],L2(Ω)
)∩ L2
((0,T ),H1(Ω)
)with
∂ttu ∈ L2((0,T ),L2(Ω)
)and h ∈ C([0,T ]).
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 40 / 43
Time-fractional hyperbolic ISP
Fractional Dynamical BC
[Šišková and Slodicka, 2018] CAMWA
(g2−β ∗ ∂ttu(x)) (t)−∆u(x , t) = h(t)f (x), x ∈ Ω, t ∈ (0,T ), (16)
The equation (16) is accompanied with the following initial and boundary conditions
u(x ,0) = u0(x), x ∈ Ω,∂tu(x ,0) = v0(x), x ∈ Ω,
u(x , t) = 0, (x , t) ∈ ΓD × (0,T ),− (g2−β ∗ ∂ttu(x)) (t)−∇u(x , t) · ν = σ(x , t), (x , t) ∈ ΓN × (0,T ),
(17)
ISP: Find the couple (u,h) obeying∫Ω
u(x , t)ω(x) dx = m(t), t ∈ [0,T ], (18)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 41 / 43
Time-fractional hyperbolic ISP
Evolutionary boundary condition
The dynamical BCs model non-perfect contact. (They are not very common in the mathematicalliterature.)
They appear in many mathematical models including heat transfer in a solid in contact with amoving fluid, thermo-elasticity, diffusion phenomena, problems in fluid dynamics, etc. (see[Escher, 1993, Igbida and Kirane, 2002, Chill et al., 2006] and the references therein).ν × E = ν × (∂tB (H)× ν) [Vrábel’ and Slodicka, 2012]∂tβ(u) + (∇u + b(u)) · ν = g [Su, 1993]∂tβ(u)−∆Γu + u +∇u · ν = g [Vrábel’ and Slodicka, 2013]
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 42 / 43
Time-fractional hyperbolic ISP
Boundary measurement
[Šišková and Slodicka, 2019] JCAM(g2−β ∗ ∂ttu(x)) (t)−∆u(x , t) = h(t)f (x) + F (x , t ,u(x , t)),
u(x ,0) = u0(x),∂tu(x ,0) = v0(x),
−∇u(x , t) · ν = γ(x , t)
(19)
The Inverse Source Problem (ISP) we are interested in here consists of identifying a couple(u(x , t),h(t)) obeying (19) and∫
Γ
u(x , t)ω(x)dS = m(t), t ∈ [0,T ], (20)
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Cannon, J. (1968).Determination of an unknown heat source from overspecified boundary data.SIAM J. Numer. Anal., 5:275–286.
Chill, R., Fašangová, E., and Prüss, J. (2006).Convergence to steady states of solutions of the Cahn–Hilliard and Caginalp equations with dynamicboundary conditions.Mathematische Nachrichten, 279(13-14):1448–1462.
Escher, J. (1993).Quasilinear parabolic systems with dynamical boundary conditions.Communications in partial differential equations, 18(7-8):1309–1364.
Farcas, A. and Lesnic, D. (2006).The boundary-element method for the determination of a heat source dependent on one variable.J. Eng. Math., 54:375–388.
Hasanov, A. (2007).Simultaneous determination of source terms in a linear parabolic problem from the finaloverdetermination: Weak solution approach.J. Math. Anal. Appl., 330:766–779.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Hasanov, A. (2011).An inverse source problem with single Dirichlet type measured output data for a linear parabolicequation.Appl. Math. Lett., 24(7):1269–1273.
Hasanov, A. and Pektas, B. (2013).Identification of an unknown time-dependent heat source from overspecified dirichlet boundary data byconjugate gradient method.Computers & Mathematics with Applications, 65(1):42–57.
Hasanov, A. and Slodicka, M. (2013).An analysis of inverse source problems with final time measured output data for the heat conductionequation: A semigroup approach.Applied Mathematics Letters, 26(2):207–214.
Hazanee, A., Ismailov, M., Lesnic, D., and Kerimov, N. (2013).An inverse time-dependent source problem for the heat equation.Appl. Numer. Math., 69:13–33.
Igbida, N. and Kirane, M. (2002).M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
A degenerate diffusion problem with dynamical boundary conditions diffusion problem with dynamicalboundary conditions.Mathematische Annalen, 323(2):377–396.
Isakov, V. (1990).Inverse source problems.Mathematical Surveys and Monographs, 34. Providence, RI: American Mathematical Society (AMS). xiv,193 p. .
Johansson, T. and Lesnic, D. (2007a).Determination of a spacewise dependent heat source.J. Comput. Appl. Math., 209:66–80.
Johansson, T. and Lesnic, D. (2007b).A variational method for identifying a spacewise dependent heat source.IMA J. Appl. Math., 72:748–760.
Nohel, J. and Shea, D. (1976).Frequency domain methods for volterra equations.Advances in Mathematics, 22(3):278 – 304.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Prilepko, A., Orlovsky, D., and Vasin, I. (2000).Methods for solving inverse problems in mathematical physics, volume 222 of Monographs andtextbooks in pure and applied mathematics.Marcel Dekker, Inc., New York-Basel.
Prilepko, A. I. and Solov’ev, V. V. (1988).Solvability theorems and Rothe’s method for inverse problems for a parabolic equation.Differential Equations, 23:1341–1349.
Rundell, W. (1980).Determination of an unknown non-homogeneous term in a linear partial differential equation fromoverspecified boundary data.Applicable Analysis, 10:231–242.
Slodicka, M. (2013).A source identification problem in linear parabolic problems: Semigroup approach.Journal of Inverse and III-posed Problems, 21(4):579–600.DOI: 10.1515/jip-2012-0070.
Slodicka, M. and Šišková, K. (2016).An inverse source problem in a semilinear time-fractional diffusion equation.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Computers & Mathematics with Applications, 72:1655–1669.
Solov’ev, V. V. (1990).Solvability of the inverse problems of finding a source, using overdetermination on the upper base for aparabolic equation.Differential Equations, 25:1114–1119.
Su, N. (1993).Multidimensional degenerate diffusion problem with evolutionary bopundary condition: Existence,uniqueness and approximation.International Series of Numerical Mathematics, 114:165–178.
Vrábel’ V. and Slodicka, M. (2012).An eddy current problem with a nonlinear evolution boundary condition.J. Math. Anal. and Appl., 387(1):267–283.
Vrábel’ V. and Slodicka, M. (2013).Nonlinear parabolic equation with a dynamical boundary condition of diffusive type.Applied Mathematics and Computation, 222:372–380.DOI:10.1016/j.amc.2013.07.057.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Šišková, K. and Slodicka, M. (2017).Recognition of a time-dependent source in a time-fractional wave equation.Applied Numerical Mathematics, 121(1–17).
Šišková, K. and Slodicka, M. (2018).A source identification problem in a time-fractional wave equation with a dynamical boundary condition.Computers & Mathematics with Applications, 75(12):4337–4354.
Šišková, K. and Slodicka, M. (2019).Identification of a source in a fractional wave equation from a boundary measurement.JCAM, 349:172–186.
Yang, L., Dehghan, M., Yu, J.-N., and Luo, G.-W. (2011).Inverse problem of time-dependent heat sources numerical reconstruction.Math. Comput. Simul., 81(8):1656–1672.
Zacher, R. (2008).Boundedness of weak solutions to evolutionary partial integro-differential equations with discontinuouscoefficients.Journal of Mathematical Analysis and Applications, 348(1):137 – 149.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43
References
Zacher, R. (2010).De Giorgi-Nash-Moser estimates for evolutionary partial integro-differential equations.Halle, Univ., Naturwissenschaftliche Fakultät III, Habilitationsschrift.
Zacher, R. (2013).A weak Harnack inequality for fractional evolution equations with discontinuous coefficients.Annali Della Scuola Normale Superiore Di Pisa-Classe Di Scienze, 12(4):903–940.
M. Slodicka (Ghent University) Some inverse source problems in semilinear fractional PDEs 43 / 43