Linear Evolution Equations: Linear Parabolic PDE

AMIYA K PANI

INDUSTRIAL MATHEMATICS GROUP

DEPARTMENT OF MATHEMATICS
INDIAN INSTITUTE OF TECHNOLOGY BOMBAY
POWAI, MUMBAI-400076
akp@math.iitb.ac.in

1
1 Introduction

In these lectures, we discuss the existence and uniqueness of weak solution to the following
class of second order linear parabolic differential equations:
∂u
+ A(t)u = f (1.1)
∂t
and initial condition:
u(0) = u0 ,
where Ω is a bounded domain in Rn with smooth boundary ∂Ω, and T > 0 fixed. Further
f : Ω × (0, T ] → R and u0 : Ω → R are given functions in their respective domains of definition.
Here u = u(x, t) defined on Ω × (0, T ] is unknown and A(t) is an elliptic differential operator.
Note that we, at this stage, do not spell out the form of A and also the boundary condition.

Since we are dealing with functions in space-time domain, therefore, in the beginning,
we discuss Banach space valued distributions and function spaces. Then, we define weak
formulation and establish abstract theory for solvability of the weak formulation.

2 Banach Space Valued Distributions and Function Spaces

Since the problem (1.1) is defined on a space-time domain, we shall study some function spaces
defined on space-time domain.

Let X be a Banach space with k · kX . We now denote X-valued Lp spaces by Lp (0, T ; X),
which consists of all strongly measurable functions v : (0, T ] → X such that
Z T 1
kv(t)kpX dt < ∞, 1 ≤ p < ∞
p
kvkLp (0,T ;X) :=
0

and for p = ∞
kvkLp (0,T ;X) := essup0≤t≤T kv(t)kX < ∞.
The space C([0, T ]; X) consists of all continuous v : [0, T ] → X such that
kvkC([0,T ];X) := max kv(t)kX < ∞.
0≤t≤T

Definition 2.1 Weak Derivative: For u ∈ L1 (0, T ; X), v ∈ L1 (0, T ; X) is called its weak
derivative, that is, ut = v if
Z T Z T
φt (t)u(t)dt = − φ(t)v(t)dt
0 0

for all scalar test functions φ ∈ D(0, T ). Here, D(0, T ) is test space defined on (0, T ), that is,
it is the space of infinitely differentiable functions with compact support in (0, T ).

Definition 2.2 (Space-time Sobolev Space): The space W 1,p (0, T ; X) is defined as
n o
W 1,p (0, T ; X) := u ∈ Lp (0, T ; X) : ut exits and ut ∈ Lp (0, T ; X) .

2
On W 1,p (0, T ; X), define its norm as
Z T 1
(ku(t)kpX kut (t)kpX )dt
p
kukW 1,p (0,T ;X) := + , 1≤p<∞
0

and for p = ∞
kvkW 1,∞ (0,T ;X) := essup0≤t≤T (ku(t)kX + kut (t)kX ).
Hence forward for p = 2, we write

H 1 (0, T ; X) := W 1,2 (0, T ; X)

Below, we state without proof two theorems on calculus in an abstract space. For a proof, we
refer to pp. 286-288 of Evans [1]

Theorem 2.1 Let u ∈ W 1,p (0, T ; X), 1 ≤ p ≤ ∞. Then the followings hold:

(i) u ∈ C([0, T ]; X) after eventual modification on a set of measure zero.

Rt 0
(ii) u(t) = u(s) + s u (τ ) dτ 0 ≤ s ≤ t ≤ T.

(iii) Further,
max ku(t)kX ≤ C kukW 1,p (0,T ;X)
t∈[0,T ]

where C depends on T.

0 0
Note in Theorem 2.1, u and u ∈ Lp (0, T ; X). Now what can be said if u and u belong to
different spaces and the answer to this question can be founf from the results of the following
Theorem.

0 0
Theorem 2.2 Let u ∈ Lp (0, T ; V ) and ut ∈ Lp (0, T ; V ) where V is the dual space of V with
0 0
V ,→ H = H ⊂ V . Then, followings hold:

(i) u ∈ C([0, T ]; H) after possible modification on a set of measure zero.

(ii) The mapping t → ku(t)kH is absolutely continuous with

d
ku(t)k2H = 2hut (t), u(t)i for a.e. t ∈ [0, T ].
dt

(iii) Moreover, there is a positive constant C = C(T ) such that

max ku(t)kH ≤ C (ku(t)kL2 (0,T ;V ) + kut (t)kL2 (0,T ;V 0 ) ).

t∈[0,T ]

3
3 Abstract Formulation and Wellposedness
0
Given two separable Hilbert spaces H and V with dual H of H identified as H, consider the
Gelfand triplet
0 0
V ,→ H = H ,→ V (3.2)
0
where ,→ is continuous and dense embedding and V is the dual of V . We now denote by (·, ·)
0
an inner product in H and h·, ·i duality parring between V and V . Note that the following
0
relation holds for v ∈ H and w ∈ V

hv, wi = (v, w).

Below, we make the following assumptions:

0
(A1) A(t) ∈ L(V, V ) depends continuously on t ∈ [0, T ]

Now associate with A(t), a bilinear form on V given by

v, w 7→ a(t; v, w) for each t ∈ [0, T ].

which satisfies
a(t; v, w) = hA(t)v, wi. (3.3)
Further assume that the bilinear form satisfies the following Garding type inequality:

(A2) For v ∈ V there exist real constants α > 0 and β such that

hA(t)v, wi = a(t; v, w) ≥ αkvk2V − βkwk2H .

0
Now consider the following abstract evolution problem: For a given f ∈ L2 (0, T ; V ) and
0
u0 ∈ H find u ∈ L2 (0, T ; V ) with ut ∈ L2 (0, T ; V ) satisfying
du
+ A(t)u = f (t) in V 0 , for a.e t ∈ [0, T ], (3.4)
dt
with initial condition
u(0) = u0 . (3.5)
Below, we establish the main theorem on solvability of the problem (3.4)-(3.5).

Theorem 3.1 Let H, V and A(t) be as given above. Further, let assumptions (A1)-(A2) hold.
0
Then for a given f ∈ L2 (0, T ; V ) and u0 ∈ H, the problem (3.4)-(3.5) has a unique solution
0
u ∈ L2 (0, T ; V ) with ut ∈ L2 (0, T ; V ).

Proof: We shall first prove uniqueness. Assume that the solution is not unique, that is, u1 and
u2 are two distinct solutions of (3.4)-(3.5) with u1 6= u2 . Note, ui satisfies
dui
+ A(t)ui = f, (3.6)
dt

4
 ui (0) = u0 . (3.7)
With w = u1 − u2 , now u satisfies
dw
+ A(t)w = 0, (3.8)
dt
with
w(0) = 0. (3.9)
Taking duality between w and (3.8), we arrive at
dw
h , wi + hA(t)w, wi = 0
dt
Using (3.3) and (ii) of Theorem 2.2, we obtain
1d
kw(t)k2H + a(t; w, w) = 0. (3.10)
2 dt
Applying Garding type inequality for the bilinear form a(t; w, w) and find that (3.10) becomes
d
kw(t)k2H + 2αkw(t)k2V − 2βkw(t)k2H ≤ 0 (3.11)
dt
Using integrating factor e−2βt , we rewrite (3.11) as
d −2βt
(e kw(t)k2H ) + 2αe−2βt kw(t)k2V ≤ 0 (3.12)
dt
and hence, integrating with respect to t from 0 to t∗ , we obtain
Z t∗
−2βt∗ ∗ 2
e kw(t )kH + 2α e−2βs kw(s)k2V ds ≤ 0.
0

Therefore, w = 0, that is, u1 = u2 and it leads to a contradiction. Hence, the solution of

(3.4)-(3.5) is unique.

For existence, we use Bubnov-Galerkin method. Assume that {φ}∞ j=1 forms a basis of
Pthat for every m; {φ1 , φ2 , · · · , φm } are linearly independent and the linear
V in the sense
combinations m j=1 ξj φj , ξj ∈ R are dense in V.

For a fixed m, let Vm = span{φ1 , φ2 , · · · , φm } and let Pm be the orthogonal projection from
H onto Vm . We now seek a function um : [0, T ] → Vm of the form
m
X
um (t) := gjm (t)φj , (3.13)
j=1

where gjm ’s are chosen so that

d
( um (t), φk ) + a(t; um (t), φk ) = hf (t), φk i, 1≤k≤m (3.14)
dt
and
m
X
um (0) = Pm u0 := ξjm φj . (3.15)
j=1

5
with
m
X
Pm u0 := ξjm φj → u0 in H as m → ∞ (3.16)
j=1

The system (3.14)-(3.15) leads to a system of linear ODE and hence, by Picard’s theorem there
exists a unique solution to (3.14)-(3.15). Now, it remains to show that limm→∞ um (t) = u(t)
and the limiting function u is a solution of (3.4)-(3.5).

Multiply (3.14) by gkm (t) and summing over k, we arrive at

d
( um (t), um (t)) + a(t; um (t), um (t)) = (f (t), um (t))
dt
and hence,
1d
kum (t)k2H + a(t; um (t), um (t)) =if (t), um (t)i. (3.17)
2 dt
For hf (t), um (t)i, use Cauchy-Schwartz to arrive at

hf (t), um (t)i ≤ kf (t)kV 0 kum (t)kV (3.18)

1 2  2
Use Young’s inequality ab ≤ a + b a, b ≥ 0,  > 0 to (3.18) to find that
2 2
1 
hf (t), um (t)i ≤ kf (t)k2V 0 + kum (t)k2V (3.19)
2 2
On substituting (3.19) in (3.17) and for the bilinear form a(t; ·, ·), use Garding type inequality
with  = α, we obtain
d 1
kum (t)k2H + αkum (t)k2H ≤ kf (t)k2V 0 + 2βkum (t)k2H . (3.20)
dt α
Setting y(t) = kum (t)k2H and z(t) = kf (t)k2V 0 , we rewrite (3.20) as

d 1
y(t) ≤ z(t) + 2βy(t). (3.21)
dt α
Apply Gronwall’s inequality to obtain
Z t
1
y(t) ≤ e2βt (y(0) + z(s) ds). (3.22)
α 0

Note that
y(0) = kum (0)k2H ≤ Cku0 k2 .
Thus, we arrive at Z t
kum (t)k2H ≤ C(ku0 k + 2
kf (s)k2V 0 ds).
0
Taking maximum in time from 0 to T , we find that
Z T
max kum (t)k2H ≤ C(ku0 k + 2
kf (s)k2V 0 ds),
0≤t≤T 0

and hence,
kum k2L∞ (0,T ;H) ≤ C(ku0 k2 + kf k2L2 (0,T ;V 0 ) ). (3.23)

6
Again integrate (3.20) with respect to t from (0, T ] to obtain
Z T
2
kum kL2 (0,T ;V ) := kum (s)k2V ds ≤ C(T, α)(ku0 k2 + kf k2L2 (0,T ;V 0 ) ). (3.24)
0

As a consequence, the sequence {um } is bounded uniformly in the Hilbert space L2 (0, T ; V ).
By weak compactness, we can extract a subsequence called {uml } ⊂ L2 (0, T ; V ) such that

uml * u weakly in L2 (0, T ; V ). (3.25)

Let N be fixed, but arbitrary with ml > N . Note that (3.17) is valid with replacing m by ml .
Then multiply the resulting equation by ψ(t) where

ψ(t) ∈ C 1 [0, T ] with ψ(T ) = 0, (3.26)

and integrate over (0, T ]. Then, choose ψN = ψφN to obtain

Z T Z T
0
{−(uml (t), ψN (t)) + a(t; uml (t), ψN (t))}dt = hf (t), ψN (t)i dt + (u0ml , ψN (0)). (3.27)
0 0

By (3.25), we now pass the limit in (3.27) as ml → ∞ to find that

Z T Z T
0
{−(u(t), ψN (t)) + a(t; u(t), ψN (t))} dt = hf (t), ψN (t)i dt + (u0 , ψN (0)). (3.28)
0 0

Note that (3.28) holds for any ψ satisfying (3.26). Hence, the equation (3.28) makes sense if
ψ ∈ D(0, T ). With ψ ∈ D(0, T ), (3.28) reduce to
du
h (t), φN i + a(t; u(t), φN ) = hf (t), φN i. (3.29)
dt
0
Here, the derivative is taken in the sense of distribution, that is, in D (0, T ). Thus,
du
h (t), φN i + hA(t)u(t), φN i = hf (t), φN i. (3.30)
dt
Note that in (3.29), N can be arbitrary.

Since finite linear combinations of {φj } are dense in V , the equation (3.29) is valid for any
v ∈ V and hence, we arrive at
du 0
= −A(t)u + f in V , (3.31)
dt
du 0 0
as = −A(t)u + f ∈ L2 (0, T ; V ). Thus, u ∈ L2 (0, T ; V ) and ut ∈ L2 (0, T ; V ).
dt
In oder to obtain u(0) = u0 , note that from (3.27) holds true if ψN is replaced by ψ ∈
C 1 ([0, T ]; V ) and obtain
Z T Z T
{−(uml (t), ψt (t)) + a(t; uml (t), ψ(t))} dt = (f (t), ψ(t)) dt + (uml (0), ψ(0)). (3.32)
0 0

Taking limit as ml → ∞ we arrive at

Z T Z T
{−(u(t), ψt (t)) + a(t; u(t), ψ(t))} dt = (f (t), ψ(t)) dt + (u0 , ψ(0)). (3.33)
0 0

7
On the other hand multiply (3.31) by ψ and integrate to obtain
Z T Z T
{−(u(t), ψt (t)) + a(t; u(t), ψ(t))} dt = (f (t), ψ(t)) dt + (u(0), ψ(0)). (3.34)
0 0

Compare (3.33) with (3.34) to arrive at

(u0 , ψ(0)) = (u(0), ψ(0)).

Since ψ is arbitrary, we now obtain

u(0) = u0 ,
and this completes the rest of the proof.

3.1 Applications

Consider the following linear parabolic initial and boundary value problem: Find u(x, t) in
Ω × (0, ∞) such that

∂u
+ L(t)u = f, x ∈ QT := Ω × (0, T ], (3.35)
∂t
u(x, t) = 0, x ∈ ∂QT := ∂Ω × (0, T ], (3.36)
u(x, 0) = u0 , x ∈ Ω, (3.37)

where Ω is a bounded domain in Rd with smooth boundary ∂Ω, and

d d
X ∂  ∂φ  X ∂φ
Lφ := − ajk + bj + a0 φ.
∂xk ∂xj ∂xj
j,k=1 j=1

Below, we make reasonable assumptions on the coefficients, on f and u0 .

Assumptions. Assume that

(i) the elliptic operator L is elliptic in the sense that there is a positive constant α0 > 0 such
that
X d Xd
ajk ξj ξk ≥ α0 |ξj |2 ∀0 6= ξ ∈ Rd .
j,k=1 j=1

(ii) ajk ; bj , a0 ∈ L∞ (QT ) with ajk = akj .

(iii) f ∈ L2 (QT ) and u0 ∈ L2 (Ω).

We now associate with the elliptic operator L a bilinear form a(t, ·, ·) as

d Z d Z Z
X ∂v ∂w X ∂v
a(t; v, w) := ajk dx+ w dx+ v w dx, v, w ∈ H01 (Ω), a.e. t ∈ (0, T ].
Ω ∂xj ∂xk Ω ∂xj Ω
j,k=1 j=1

8
 To put our problem in the abstract framework like (3.4)-(3.5), we now associate with u(x, t)
a mapping u : (0, T ] −→ H01 (Ω) defined by
[u(t)](x) = u(x, t), x ∈ Ω, t ∈ [0, T ].
Essentially for fixed t in (0, T ], u(t) ∈ H01 (Ω). Similarly, f (x, t) can be defined as a map
f : [0, T ] −→ L2 (Ω) given by
[f (t)](x) = f (x, t), x ∈ Ω, t ∈ [0, T ].
Moreover, because of our assumption it is easy to check that the bilinear form satisfies the
following estimates:

• Boundedness. There is a positive constant M such that the biliear form is bounded in
the sense that for v, w ∈ H01 (Ω),
|a(t, v, w)| ≤ M kvkH01 (Ω) kwkH01 (Ω) .

• Garding type inequality. There exist two real constant α > 0 and β such that for
v ∈ H01 (Ω),
a(t; v, v) ≥ αkvk2H 1 (Ω) − βkvkL2 (Ω) .
0

Problem 3.1 Verify above two properties for the bilinear form.

Since for fixed t ∈ (0, T ] and fixed v ∈ H01 (Ω), the bilinear form a(t; v, ·) can be thought of
as a linear form on H01 (Ω), which is bounded because of the boundedness of the biliear form,
therefore, by Ritz-Representation theorem, we can associate with this an abstract bounded
linear operator A(t) : H01 (Ω) −→ H −1 (Ω) such that
hA(t)v, wi := a(t; v, w), w ∈ H01 (Ω), (3.38)
where H −1 (Ω) is the dual space of H01 (Ω), and h·, ·i is the duality pairing between H −1 (Ω) and
H01 (Ω).

Set V = H01 (Ω) and H = L2 (Ω). Note that V 0 = H −1 (Ω) is the dual space of V and it
is easy to check that V, H, V 0 forms a Gelfand triplet. Thus, one can write (3.35) -(3.37) in
abstract form as:
ut + A(t)u = f (t) in V 0
with u(0) = u0 ∈ H.

Note the using boundedness of the bilinear form, the hypothesis (A1) is satisfied and
further, due to Garding type inequality, the operator A(t) satisfies (A2). Therefore, we apply
Theorem 3.1 to discuss existence of a unique weak solution of (3.35)-(3.37).

References
[1] L. C. Evans, Partial Differential Equations, Graduate Studies in Mathematics, Vol. 19,
AMS, Providence, Rhode Island, 1998 ( Reprinted 2002).