QEPPaper
QEPPaper
QEPPaper
KELLER VANDEBOGERT
1. Introduction
1
Consider a Banach space X and let f : D → X and u : G → X,
where D and G are real intervals. A is a bounded or unbounded linear
operator whose domain is densely defined in X. We then consider:
du
+ A(t)u = f (t)
dt
The above is referred to as an evolution equation. We can impose an
initial condition, say, u(0) = x, and this evolution becomes what is
known as The Abstract Cauchy Problem. In general, there are different
methods of solution for these types of problems, but no single method
always seems to work. This is where semigroups can be used.
Consider an operator, T (t), which can be thought of as an ”evolu-
tion” operator. T (t) applied to u(t0 ) will have the following effect:
T (t)u(t0 ) = u(t0 + t)
u(t) = T (t)x
apply our evolution operator in succession, it would just have the effect
of shifting our time twice. More precisely,
We have just plunged into the world of semigroups. Put very simply, a
semigroup is merely a group without the inverse or identity property.
In the language of what we have just presented, we have:
and
T (s)T (t) = T (s + t)
T (h)x − x
Ax = lim
h→0 h
This definition is actually quite natural if you just consider the defi-
nition of the derivative of T (t), and then utilize the properties of semi-
groups. This consideration leads us to our first lemma.
d
T (t)x = AT (t)x = T (t)Ax
dt
d T (t + h)x − T (t)x
T (t)x = lim
dt h→0 h
T (t)T (h)x − T (t)x
= lim
(2.1) h→0 h
T (h)x − x
= T (t) lim
h→0 h
= T (t)Ax
Also,
T (t)x − T (t − h)x T (h)x − x
lim = lim T (t − h)
h→0 h h→0 h
= T (t)Ax
Commutativity follows immediately by the commutativity of addi-
tion. Namely,
Note that we proved equality for h tending from both the left and
the right, since we are only working in the strong operator topology.
We now want to introduce the concept of an analytic semigroup. As
we shall see, analytic semigroups are a restriction on the set of C0 semi-
groups, and this class of semigroups in fact provides better regularity
of solutions for PDE’s.
du
(3.1) + A(t)u = f (t)
dt
where u(0) = x, and A can be an either bounded or unbounded linear
operator.
If we examined the homogeneous case of (3.1), it is possible to pose
a rather naive solution. Namely, if
du
+ Au = 0
dt
Then,
u(t) = e−At x
d
(3.2) (T (t)x) + A(t)T (t)x = 0
dt
With (3.2) and Definition 2.2, we see that if −A is an infinitesimal
generator of {T (t)}, then, symbolically, we have a solution. Comparing
this with our ”naive” solution, this implies that T (t)x = e−At x.
Interestingly, all of the operations line up. Assuming that -A is our
infinitesimal generator, we see:
And other properties are readily verified. Also, we note that the
above properties are merely based off of the assumption that e−At will
act the same as the regular exponential function, which we intend to
prove. It is extremely important to note that the operation between the
above expressions is not multiplication. This is an arbitrary operation,
and because of this, the above relations are not trivial.
To make sense of this, we will have to look at e−At in a different way.
Where L−1 (.) is the inverse Laplace Transform. From our first pos-
sible definition, we see
Proof. Firstly, ||A|| < ∞ since our operator is bounded. We first show
that T (s)T (t) = T (s + t).
∞ ∞ ∞
X (tA)i X (sA)j X ((s + t)A)n
T (s)T (t) = eAt eAs = =
i=0
i! j=0
j! n=0
n!
∞ ∞
X (tA)n X (t||A||)n
||T (t) − I|| = || || 6 = et||A|| − 1
n=1
n! n=1
n!
Letting t → 0+ , we see that the norm tends to 0.
Thus, for bounded operators we see that this is in fact well defined.
However, not all operators are bounded. Here, we will state the the-
orem of Hille-Yosida without proof, since it is well beyond the scope
of the paper(see [2]). However, this theorem gives a very broad char-
acterization of linear operators which are infinitesimal operators of C0
semigroups.
8 KELLER VANDEBOGERT
Note in the below proof we will use T (t) and e−At interchangeably.
Proof. Define
ˆ
−tA 1
(4.1) e = eλt R(λ; −A)dλ
2πi Γ
ˆ ˆ
1 λ0 s 0 0 1 0
eλ s f R(λ0 ; −A) dλ0
f e R(λ ; −A)dλ =
2πi Γ0 2πi Γ0
ˆ ˆ
1 λ0 s 0
0 1
eλs f R(λ; −A) dλ
e f R(λ ; −A) dλ =
2πi Γ0 2πi Γ
We now consider:
ˆ ˆ
−tA −sA 1 0
e e = 2
eλt+λ s R(λ; −A)R(λ0 ; −A)dλdλ0
(2πi) Γ Γ0
Employing the resolvent equation, this becomes:
ˆ ˆ
1 λt+λ0 s 1
e R(λ; −A) − R(λ ; −A) dλdλ0
0
(2πi)2 Γ Γ0 λ0 − λ
Since Γ lies to the left of Γ0 , λ 6= λ0 when integrating over Γ. With
this and Fubini’s theorem, we can reduce the integrand.
ˆ ˆ
1 λt+λ0 s 1
e R(λ; −A) − R(λ ; −A) dλdλ0
0
(2πi)2 Γ Γ0 λ0 − λ
10 KELLER VANDEBOGERT
ˆ ˆ ˆ ˆ
1 λt+λ0 s 1 0 1 0 1
= 2
e 0
R(λ; −A)dλdλ − 2
eλt+λ s 0 R(λ0 ; −A)dλdλ0
(2πi) Γ Γ0 λ −λ (2πi) Γ Γ0 λ −λ
The second term in the above expression is 0 with the use of Cauchy’s
theorem. The first expression simplifies to:
ˆ
1
eλ(t+s) R(λ; −A)dλ = e−(t+s)A = e−tA e−sA
2πi Γ
And we have proved the semigroup property. We still need to show that
this is a C0 semigroup satisfying the additional conditions of Definition
2.4.
Let > 0. If we can show that T (t) has an analytic continuation
along any curve in the sector ∆φ− , then T (t) has an analytic contin-
uation in all of ∆φ . Thus, we consider a curve Γ in the sector ∆φ− .
Since A is of type (φ, M ),
λ C|t|
; −A)|| 6
||R(
|t| |λ|
0
Where C = C(). Letting λ = |t|λ, we can scale our contour such that
Γ0 = |t|Γ. Then, since Γ0 will not contain any new singularities:
ˆ ˆ
−tA 1 1 0
e = λt
e R(λ; −A)dλ = eλ arg(t) R(λ0 /|t|; −A)dλ0 /|t|
2πi Γ 2πi Γ
Then,
ˆ
−tA 0
(4.2) ||e || 6 C |eλ arg(t) ||dλ0 |/|λ0 | 6 C
Γ
ˆ
−tA 1 0
Ae = eλ arg(t) AR(λ0 /|t|; −A)dλ0 /|t|
2πi Γ
Rewrite this as:
ANALYTIC SEMIGROUPS AND APPLICATIONS 11
ˆ ˆ
1 λ0 arg(t)
0
0 0 λ0 arg(t) 0 0 0
e λ /|t|I+A R(λ /|t|; −A)dλ /|t|− e λ /|t|R(λ /|t|; −A)dλ /|t|
2πi Γ Γ
ˆ
−tA 1 0 C
(4.3) ||Ae || 6 eλ arg(t) |λ0 | ∗ ||R(λ0 /|t|; −A)|| ∗ |dλ0 |/|t| 6
|t| Γ |t|
With this, we know that T (t) and AT (t) are bounded operators. If
dT (t)x
we can show that dt
= −AT (t)x, then this means that T (t) can be
locally represented by a power series in the uniform operator topology.
By the principle of analytic continuation, it will then be possible to
extend T (t) to all of ∆φ . Using (4.1):
ˆ
d 1
T (t) = λeλt R(λ; −A)dλ
dt 2πi Γ
Rewrite λR(λ; −A) as I − AR(λ; −A):
ˆ ˆ
1 λt λt
e dλ − A e R(λ; −A)dλ
2πi Γ Γ
The first integral vanishes since the integrand is holomorphic, and we
thus see:
ˆ
d 1
(4.4) T (t)x = −A eλt R(λ; −A)xdλ = −AT (t)x
dt 2πi Γ
ˆ t
−At
lim e x − x = lim −Ae−Aτ dτ
t→0+ t→0 0
Since we’ve already shown the integrand is bounded, this clearly tends
to 0, so we do have a C0 semigroup.
Finally, we merely have to show that −A does indeed generate every
T (t). By definition of infinitesimal generator and (4.4), we have:
ˆ t
e−At x − x 1
=− e−Aτ dτ Ax → −Ax
t t 0
Then for any u0 ∈ X and for any f (t) that is uniformly Hölder
continuous of exponent β in [0, t0 ], there exists a unique solution u(t)
of the Cauchy problem. Furthermore, the solution is given by:
ˆ t
u(t) = U (t, 0)u0 + U (t, s)f (s)ds
0
du
Where U (t, τ ) is a fundamental solution of dt
+ A(t)u = 0.
(5.1) ut = uxx
u(x, 0) = f
∂bu
+ ω2ub=0
∂t
u
b(ω, 0) = fb(ω)
2
b = fbe−ω t
u
14 KELLER VANDEBOGERT
ˆ ∞
1 2
(5.2) u(x, t) = eiωx fbe−ω t dω
2π −∞
ˆ ∞
1 2 /4t
(5.3) u(x, t) = √ f (y)e−(x−y) dy
4πt −∞
2 2
Where we’ve used the fact that e−ω t is the Fourier transform of √ 1 e−x /4t .
4πt
1 −x2 /4t
Kt (x) = √ e
4πt
Then, by denoting ∗ as the operation of convolution, the solution (5.3)
can be rewritten as:
u(x, t) = Kt ∗ f
6. Conclusion
high level results from the theory of semigroups and PDE’s were pre-
sented and an illustration of the application of semigroup theory was
given in the last section by means of the solution of the heat equation
in one dimension.
References
[1] Semigroups of Linear Operators. Sheree L. Levarge.
[2] Partial Differential Equations. Avner Friedman. 91-130.
[3] Semigroups of Linear Operators and Applications to Partial Differential Equa-
tions. A. Pazy. 100-121.