SJÄLVSTÄNDIGA ARBETEN I MATEMATIK
MATEMATISKA INSTITUTIONEN, STOCKHOLMS UNIVERSITET
Sobolev norm estimates
of the time dependent Schrödinger equation
av
Erik Melander
2020 - No K2
Sobolev norm estimates
of the time dependent Schrödinger equation
Erik Melander
Självständigt arbete i matematik 15 högskolepoäng, grundnivå
Handledare: Salvador Rodriguez-Lopez
Contents
1 Introduction 2
2 Background 3
2.1 Schwartz space . . . 3
2.2 Fourier transform . . . 4
2.3 Tempered distributions . . . 13
2.3.1 Properties of tempered distributions . . . 14
2.3.2 Fourier transform of tempered distributions . . . 15
2.4 Pseudo differential operators and Sobolev spaces . . . 17
2.4.1 Pseudo differential operators . . . 17
2.4.2 Sobolev spaces . . . 18
2.5 Solving the Heat equation . . . 19
2.6 Solving the time dependent Schr¨odinger equation . . . 21
2.6.1 Regularity estimates for the free time dependent Schr¨odinger 24 2.6.2 The free fractional Schr¨odinger equation . . . 25
3 Regularity estimates for the free time dependent fractional Schr¨odinger equation 26 3.1 Estimate using the Plancherel theorem . . . 26
3.1.1 Replacing|ξ|a with a differentiable and injective function ϕ 29 3.2 Estimating using Pitt’s inequality . . . 31
3.2.1 Stronger estimate . . . 36
4 Final remarks 37
1 Introduction
This thesis presents regularity estimates for solutions to the free time dependent fractional Schr¨odinger equation with initial data using the theory of Fourier transforms. As such our main focus in this thesis are the functions Saf , where
pSafq rxsptq 1 2π
»8
8eixξei|ξ|atfppξqdξ for a ¥ 2. (1) These functions are solutions to the free time dependent fractional Schr¨odinger equation with initial data f , i.e. they satisfies that
iBtupx, tq pB2xqa{2upx, tq
with upx, 0q fpxq. As is indicated by the factor pf in the expression for Saf , these solutions are derived by the use of the Fourier transform, which we define as
fppξq :
»8
8eiξxfpxqdx.
The problem at hand is to find regularity estimates for Saf using the initial datum. To tackle this we restrict our focus to problems where we can guarantee the existence of the Fourier transform of our data function. This restriction, while seemingly severe, helps us establish tools and general theory for our esti- mates which we then can expand to hold for less restricted functions.
We assume throughout this text that our data function f is in the Schwartz space on R, denoted S pRq; a space of smooth functions whose derivatives de- crease more rapidly than any polynomial. This space is dense in LppRq for 1¤ p 8 and on this space the Fourier transform is a well defined continuous linear operator mapping S pRq onto itself and for such function we can apply the Plancherel theorem.
In the case of a 2, this is a solution to the partial differential equation called the free time dependent Schr¨odinger equation, which states thatBx2u iBtu [14].
We will derive a solution later in section 2.6 to the similar but time reversed equationBx2u iBtu.
The idea behind the regularity estimates comes from the fact that if you have an operator T and know thatkT fkB ¤ C kfkAfor some number C we can draw the conclusion that if f P A, then T f P B and as such we have gained knowledge about T f . For our estimates a number of different norms will be used, the main ones which are the LppRq-, 9HspRq- and L8pR, LppRqq-norms, where
kfkLppRq
»8
8|fpxq|p 1{p
for 1¤ p ¤ 8,
kfkH9spRq
»8
8|ξ|2s pfpξq2dξ 1{2
,
and
kfkL8pR,LppRqq sup
xPR
»8
8|fpx, tq|pdt 1{p
.
To acquire our result a recurring method throughout this thesis is to observe the effects of dilation of the normed function and different relationships between a function and its Fourier transform, such as the Plancherel theorem and Pitt’s inequality.
2 Background
In this text we will use the Fourier transform and its properties on functions in certain spaces to reach conclusions about differential equations. To make sure that all this makes sense we will start with the space of smooth functions such that they and all of their derivatives vanish faster than any polynomial at the infinities. This space is called the Schwartz space and on this the Fourier trans- form is a well defined bounded linear operator and has some useful properties for us. We will later discuss how to expand to functions beyond this space.
2.1 Schwartz space
The Schwartz space, denoted as S pRq, is the space of all smooth functions f :R Ñ C such that for all natural numbers α and N there is a constant Cα,N for which
|Bαfpxq| ¤ Cα,Np1 |x|qN uniformly inR [1, p.13].
Example 2.1. The Gaussian bell function Gpxq ex2{2belongs toS pRq which we see by noting that G is smooth and for all natural numbers M we have that
Gpxq ¤ CMp1 |x|qM (2)
for some number CM. And so, given positive integers α, N , there exists numbers Cα from the Leibniz product rule and CN by (2) above such that
|BαGpxq| ¤ Cα|p1 |x|qαGpxq|
¤ Cαp1 |x|qαCNp1 |x|qNα Cα,Np1 |x|qN, where Cα,N Cα CN, showing that GP S pRq.
We call the functions inS pRq Schwartz functions and equip the Schwartz space with the semi norms|f|m,S for mP N, where
|f|m,S : sup
|α| |β|¤msup
xPR
xαBβfpxq.
These semi norms are defined such as if|f|m,S 8 for all m P N0then f P S pRq and using them we define convergence inS pRq.
Definition 2.2. Let tfkuk S pRq for k P N be a sequence of functions in S pRq. We say that tfkuk converges to f inS pRq iff for all ε ¡ 0 and all m P N there is a KP N such that |fk f|m,S ε whenever k ¥ K.
We note that for p¥ 1 we have that S pRq LppRq. This is a direct result of the definition of the Schwartz space, since for f P S pRq there is a number C such that|fpxq| ¤ Cp1 |x|q2and so we have
kfkpLppRq
»8
8|fpxq|pdx¤ C
»8
8p1 |x|q2pdx 8.
A fundamental property of the Schwartz functions is that they vanish rapidly at infinity and an important subspace ofS pRq is C08pRq which is the space of smooth functions with compact support, i.e. vanish outside a compact set. Here we define the support of a function f , written as supppfq, by
supppfq : tx P R|fpxq 0u.
That is, the support of a function is the closure of the set of points on which the function is nonzero.
Example 2.3. The bump function ϕ defined by ϕpxq
#e1{p1x2q for|x| 1 0 for |x| ¥ 1
is a smooth function of compact support. Here supppϕq p1, 1q r1, 1s.
The space C08pRq has the useful property that it is dense in LppRq for 1 ¤ p 8 [15, p.13], and from this and the fact that S pRq LppRq we can draw the following conclusion.
Lemma 2.4. S pRq is dense in LppRq for 1 ¤ p 8.
And so, while we will restrict our results to a space of nicely behaving functions, it is feasible to expand our results to a wider space of functions by density.
2.2 Fourier transform
Our main focus on the Schwartz space is as mentioned the Fourier transform.
For functions in the Schwartz space we denote by pf the Fourier transform of a function f and define the transform as [1, p.9]
fppξq :
»8
8eiξxfpxqdx.
We can see that, given f P S pRq, we have a first indication why this definition makes sense by noting that
pfpξq
»8
8eiξxfpxqdx
¤»8
8|fpxq| dx 8.
Example 2.5. The Gaussian bell function Gpxq ex2{2is a Schwartz function as seen in example 2.1 and as such we can calculate its Fourier transform.
We have that Gpξq p
»8
8eiξxx2{2dx
»8
8epx iξq2{2ξ2{2dx
eξ2{2
»8
8epx iξq2{2dx eξ2{2
»
γ
ez2{2dz,
where γ is the curve inC parametrized by zptq t iξ for t P R. The integrand is analytic and vanishes as|z| Ñ 8 and so we get that
»
γ
ez2{2dz
»8
8ez2{2dz? 2π, where we use the known fact that ³8
8ex2dx?
π. This gives us that pGpξq eξ2{2?
2π?
2πGpξq.
Now, define the operator D as D iB. From the definition above we get the following properties for the Fourier transform.
Theorem 2.6. Let ϕP S pRq. Then
1. pαϕ βψqpξq α p{ ϕpξq β pψpξq for α, β P R 2. {pDnϕqpξq ξnϕpξqp
3. Dnϕpξq {p ppxqnϕqpξq 4. ϕpP S pRq
Proof. 1. From the linearity of integration we have that pαϕ βψqpξq {
»8
8eiξxpαϕpxq βψpxqqdx
α
»8
8eiξxϕpxqdx β
»8
8eiξxψpxqdx
α pϕpξq β pψpξq.
2. We begin by noting that since ϕP S pRq we have that
ξkeiξxDmϕpxq8
8 0 for all k, mP N0 and ξ P R. Using this, integration by parts now gives us that
pD{nϕqpξq
»8
8eiξxDnϕpxqdx ξn
»8
8eiξxϕpxq ξnϕpξq.p 3. By the Leibniz rule we have that
Dnϕpξq Dp n
»8
8eiξxϕpxqdx
»8
8Dneiξxϕpxqdx
»8
8eiξxpxqnϕpxqdx {ppxqnϕqpξq.
4. From Lemma 2.4 we have that if ϕP S pRq then
| pϕpξq|
»8
8eiξxϕpxqdx
¤»8
8|ϕpxq| dx kϕkL1pRq 8 and furthermore we can note that DnxmϕP S pRq for n, m P N0. Now, using 2. and 3. above, we have that
| pϕ|m,S sup
|α| |β|¤msup
ξPR
ξαDβϕpξqp
sup
|α| |β|¤msup
ξPR
ξαppxq{βϕqpξq
sup
|α| |β|¤msup
ξPR
pDα{pxqβϕqpξq
¤ sup
|α| |β|¤m
Dαpxqβϕ L1pRq 8.
Using the following boundedness lemma we can make an important observation about the Fourier transform.
Lemma 2.7. [1, p.14] For ϕP S pRq and m P N there is a constant Cm such that
| pϕ|m,S ¤ Cm|ϕ|m 2,S
Proof. Let ϕP S pRq, then kϕkL1pRq
»8
8|ϕpxq| dx
»8
8p1 |x|q2p1 |x|q2ϕpxqdx
¤ |ϕ|2,S
»8
8p1 |x|q2dx C |ϕ|2,S. Using this we have
| pϕ|0,S sup
ξPR| pϕpξq| sup
ξPR
»8
8eiξxϕpxqdx
¤ sup
ξPR
»8
8|ϕpxq| dx kϕkL1pRq¤ C |ϕ|2,S. Furthermore we have by theorem 2.6 that
ξαDβξϕpξq p pDαx{pxβqϕqpξq and by the Leibniz rule we have that
Dαxpxβqϕpxq ¤ ¸
k¤α
α k
pDkxβqpDαkϕpxqq ¤ Cα,βxβDαϕpxq.
Using this and taking the supremum over α and β, we have that sup
α β¤msup
ξPR
ξαDξβϕp ¤ sup
α β¤mCα,βDxαxβϕ
2,S
¤ sup
α β¤m 2sup
ξPRCα,βDαxxβϕ ¤ Cm|ϕ|m 2,S. That is, we have that | pϕ|m,S ¤ Cm|ϕ|m 2,S, and so we have that ϕp P S pRq whenever ϕP S pRq.
That is, for a Schwartz function its Fourier transform is bounded by the function itself which, together with the linearity of the transform, helps us establish that the transform is in fact continuous.
Theorem 2.8. Given tϕkuk S pRq such that ϕk Ñ ϕ in S pRq for some ϕ P S pRq as k Ñ 8, then xϕk Ñ pϕ in S pRq as k Ñ 8, i.e. the Fourier transform is continuous on S pRq.
Proof. Assume ϕk Ñ ϕ in S pRq as k Ñ 8, then pϕP S pRq and for all k P N we have thatϕpkP S pRq. Furthermore, we have from lemma 2.7 that there is a constant Cmsuch that
{pϕk ϕq
m,S ¤ Cm|ϕk ϕ|m 2,S. Now, given ε¡ 0 and m P N, there is a K P N such that
|ϕk ϕ|m 2,S ε Cm
whenever k¥ K. Thus, using the linearity of the transform, we have that
|xϕk pϕ|m,S {pϕk ϕq
m,S ¤ Cm|ϕk ϕ|m 2,S ε whenever k¥ K.
And so by Theorem 2.6, Lemma 2.7 and Lemma 2.8 we have that on S pRq the Fourier transform is a continuous linear operator mapping S pRq to S pRq.
Furthermore, it has the important property of transforming differentiation of a function to multiplication of its transform with a polynomial. This is useful in solving differential equations and we will later use this to define other spaces on which the concept of pseudo differential equations makes sense. Before this, however, the Fourier transform has other important properties we need.
We begin with translations, modulations and dilations, which we for a func- tion ϕ write as Tyϕ for a translation of ϕ by a factor y, where
pTyϕqpxq : ϕpx yq, and Myϕ for a modulation of ϕ by a factor y, where
pMyϕqpxq : eixyϕpxq,
and finally Dβϕ for dilation of ϕ by a factor β, where pDβϕqpxq : ϕpβxq.
These are connected by the Fourier transform and are useful tools in interpreting the transform and analysing its effect on functions.
Theorem 2.9. Given ϕP S pRq, then
1. {pTyϕqpξq eiξyϕpξq pMp yϕqpξq,p 2. {pMyϕqpξq pϕpξ yq pTyϕqpξq,p 3. zDβϕpξq β1ϕpp βξq β1pDβ1ϕqpξq.
Proof. All three parts relies entirely on a change of variables.
1. Setting t x y gives us that pT{yϕqpξq
»8
8eiξxpTyϕqpxqdx
»8
8eiξxϕpx yqdt
»8
8eiξptyqϕptqdt eiξy
»8
8eiξtϕptqdt
eiξyϕpξq pMp yϕqpξq.p 2. Setting t x y gives us that
pM{yϕqpξq
»8
8eiξxpMyϕqpxqdx
»8
8eiξxeixyϕpxqdx
»8
8eixpξyqϕpxqdx pϕpξ yq pTyϕqpξq.p 3. Setting y βx gives us that
pD{βϕqpξq
»8
8eiξxpDβϕqdx
»8
8eiβξβxϕpβxqdx
1 β
»8
8eiβξyϕpyqdy 1 βϕp
ξ β
.
We continue with convolutions. Given ϕ and ψ inS pRq the convolution ϕ ψ is defined as [12, p. 139]
pϕ ψqpxq :
»8
8ϕpx tqψptqdt.
For the Fourier transform we have that the transform of a convolution of two functions corresponds with a multiplication of their transforms.
Theorem 2.10. Given ϕ, ψ in S pRq, then pϕ ψqpξq p{ ϕpξq pψpξq
Proof. Using Fubini’s theorem [7, p. 86] and theorem 2.9 above we have that pϕ ψqpξq {
»8
8eiξxpϕ ψqpxqdx
»8
8eiξx
»8
8ϕpx tqψptqdt
dx
»8
8
»8
8eiξxpTtϕqpxqdx
ψptqdt
»8
8pMtϕqpξqψptqdtp
»8
8eiξtϕpξqψptqdt pp ϕpξq
»8
8eiξtψptqdt pϕpξq pψpξq.
As we will see it’s possible when using the Fourier transform to solve differen- tial equations to arrive at a transformed solution that is written as a product of two functions. Using the inverse transform we will soon define we can then use the result in Theorem 2.10 above to arrive at a solution, now written as a convolution.
We can note that while Theorem 2.6 gives us that the Fourier transform maps the Schwartz space to itself the transform does not reverse this action, i.e. in general pϕpxq ϕpxq. While this is somewhat inconvenient another useful prop-p erty of the Fourier transform is the existence of an inversion formula that gives us the inverse transform so that we can regain ϕ from ϕ, circumventing thisp problem. However, to show this we need the next two lemmas, the first called the adjoint or multiplication lemma.
Lemma 2.11. Let ϕ, ψP S pRq, then
»8
8ϕpxq pψpxqdx
»8
8ϕpxqψpxqdxp Proof. Using Fubini’s theorem we have that
»8
8ϕpxq pψpxqdx
»8
8ϕpxq
»8
8eixyψpyqdy
dx
»8
8
»8
8ϕpxqeixyψpyqdy
dx
»8
8
»8
8ϕpxqeixydx
ψpyqdy
»8
8ϕpyqψpyqdy.p
The second lemma, while not directly connected with the Fourier transform, describes a useful property of convolutions.
Theorem 2.12. Given ϕ P L1pRq such that ³8
8ϕpxqdx 1, let ϕεpxq ε1Dε1ϕpxq. Then for any f P LppRq, 1 ¤ p 8 we have that
f ϕεÑ f in LppRq as ε Ñ 0.
Proof. Using the change of variables τ ε1t, Fubini’s theorem and Minkowski’s integral inequality [5, p. 101] we have that
kf ϕε fkLppRq
»8
8|pf ϕεqpxq fpxq|pdx 1{p
»8
8
»8
8fpx tqϕεptqdt 1 fpxq
pdx 1{p
»8
8
»8
8fpx tqϕpε1tqε1dt 1 fpxq
pdx 1{p
»8
8
»8
8fpx ετqϕpτqdτ
»8
8ϕpτqdτ fpxq
pdx 1{p
»8
8
»8
8pfpx ετq fpxqqϕpτqdτ
pdx 1{p
¤
»8
8|ϕpτq|
»8
8|fpx ετq fpxq|pdx 1{p
dτ
»8
8|ϕpτq| kTετf fkLppRqdτ.
Now sincekTετf fkLppRq¤ 2 kfkLppRq andkTετf fkLppRq Ñ 0 as ε Ñ 0 we have by Lebesgue dominated convergence theorem thatkf ϕε fkLppRqÑ 0 as εÑ 0.
And so we can now state the Fourier inverse formula which gives us a tool to retrieving a transformed function.
Theorem 2.13. Let ϕP S pRq, then ϕpxq 1
2π
»8
8eiξxϕpξqdξ.p Remark: We denote in this formula the action of 2π1 ³8
8eiξxϕpξqdξ on ϕ in the right hand side above by the symbol _ so that the formula above can be written as
ϕpxq p pϕq_pxq.
Proof. Define ϕεby
ϕεpxq 1 2π
»8
8eiξxε2ξ2{2ϕpξqdξp
and let
gpξq eixξeε2ξ2{2 pMxDεGqpξq,
where G is the Gaussian bell function Gpξq eξ2{2as of example 2.5. Then, by Theorem 2.9 on G and since pGpξq ?
2πGpξq (by example 2.5), we have that
pgpζq ε1pTxDε1Gqpζq p ?
2πε1epζxq2{2ε2. Since
ϕεpxq 1 2π
»8
8gpξq pϕpξqdξ we now have by Lemma 2.11 that
ϕεpxq 1 2πε
»8
8pgpξqϕpξqdξ
?2πε1 2π
»8
8epξxq2{2ε2ϕpξqdξ
ϕ ε1Dε1
?G 2π
pxq.
Since³8
8?1
2πGpxqdx 1 we now have by Lemma 2.12 that in LppRq ϕεpxq Ñ ϕ as ε Ñ 0.
Finally, Lebesgue dominated convergence gives us that ϕεÑ 1
2π
»8
8eiξxϕpξqdξ as ε Ñ 0p and so we have that
ϕpxq 1 2π
»8
8eiξxϕpξqdξ.p
With this inversion formula, we now have that the Fourier transform is a con- tinuous bijective mapping on S pRq onto S pRq [12].
The next theorem is one of the major theorems for the Fourier transform and one of our main tools when establishing our future results.
Theorem 2.14 (Plancherel theorem). Given ϕP S pRq, then
?2πkϕkL2pRq kpϕkL2pRq.
Proof. We prove this by first proving a more general case, i.e. that
»8
8fpxqgpxqdx 2π
»8
8
fpξqpgpξqdξ.p
Using Fubini’s theorem and the Fourier inversion formula we have that
»8
8
fppξqpgpξqdξ
»8
8fpxq
»8
8eiξxpgpξqdξ
dx
»8
8fpxq
»8
8eiξxpgpξqdξ
dx 2π
»8
8fpxqgpxqdx.
and so for gpxq fpxq we get the Plancherel theorem.
This gives us a useful tool in analysing the solutions to differential equations as it allows us to switch between a function and its transform somewhat when analysing the regularity of the function. Furthermore we can make a direct observations from the Plancherel theorem and the fact that S pRq is dense in L2pRq.
Corollary 2.15. There is a unique continuous extension of the Fourier trans- form from S pRq to L2pRq, and as such the Plancherel theorem is extended to functions in L2pRq.
Proof. Given f P L2pRq, by Lemma 2.4 there is a sequence tϕkuk in S pRq converging to f in L2pRq. We have by Lemma 2.6 that tpϕkukis also a sequence in S pRq and by linearity of the transform and the Plancherel theorem we have that
k pϕm pϕnkL2pRq ϕm{ ϕn L2pRq?
2πkϕm ϕnkL2pRq,
and so we have that k pϕm pϕnkL2pRq Ñ 0 as m, n Ñ 8. By the completeness of L2pRq we now have that there is a function gϕin L2pRq such that pϕk Ñ gϕ
in L2pRq.
Now, given another sequence tψkuk in S pRq such that ψk Ñ f in L2pRq we have by the same argument as above that there is a function gψ in L2pRq such that pψkÑ gψ in L2pRq. Furthermore, by the triangle inequality and Plancherel theorem, we have that
pϕn pψn
L2pRq ?
2πkϕn ψnkL2pRq
¤?
2πpkϕn fkL2pRq kf ψnkL2pRqq Ñ 0 as n Ñ 8 and so
kgϕ gψkL2pRq¤ kgϕ pϕmkL2pRq
pψn gψ
L2pRq
pϕm pψm
L2pRq. Letting m, nÑ 8 now gives us that
kgϕ gψkL2pRq 0
i.e. gϕ gψin L2pRq. We now have that given f in L2pRq there exists a unique functionFf in L2pRq such that for every sequence tϕkuk in S pRq converging to f in L2pRq the sequence tpϕkuk inS pRq converges to Ff in L2pRq.
Finally, we have that
kFfkL2pRq¤ kpϕk FfkL2pRq k pϕkkL2pRq
kpϕk FfkL2pRq ?1
2πkϕkkL2pRq
and so, by letting kÑ 8 above we have that kFfkL2pRq¤ 1
?2πkfkL2pRq.
Furthermore
kfkL2pRq¤ kϕk fkL2pRq kϕkkL2pRq
kϕk fkL2pRq
?2πk pϕkkL2pRq.
and so, by once again letting kÑ 8 we have that kfkL2pRq¤?
2πkFfkL2pRq. That is, for f in L2pRq we have that
kfkL2pRq?
2πkFfkL2pRq.
This gives us that there is a extensionF for the Fourier transform from S pRq to L2pRq such that for ϕ P S pRq we have that Fϕ pϕ. As such we will from now on also useF to indicate the Fourier transform on functions in L2pRq.
While we now have a space of nicely behaving functions for which the Fourier transform makes sense and has a number of useful properties, not all functions are nicely behaving, e.g. we can’t calculate a Fourier transform of or even dif- ferentiate all functions, and so to make the transform a bit more useful the restriction toS pRq need to be relaxed, why the next step is needed.
2.3 Tempered distributions
To generalize the idea of the Fourier transform, we introduce the space of tem- pered distributions. This space, denotedS1pRq, is the dual space of S pRq, i.e.
the set of all linear and continuous mappings T :S pRq Ñ C in the sense of that f is a tempered distribution if its linear and there exists a constant C and an interger k such that|fpϕq| ¤ |ϕ|k,S for all ϕP S pRq [2, p. 18]. This allows us to define the Fourier transform on objects in a wider sense than before while retaining and expanding on the core properties of the transform.
Example 2.16. The Dirac delta function, δ0, is classically defined to be such that δ0pxq 0 if x 0, and ³8
8δ0pxqdx 1, i.e. it spikes at zero enough to
give its integral a value of 1. This is not actually a function, but it is a tempered distribution and as such defined as
δ0pϕq : ϕp0q for all ϕ P S pRq .
To see that δ0 is indeed a tempered distribution we have that for all ϕ P S pRq and natural numbers k we have that
|δ0pϕq| |ϕp0q| ¤ |ϕ|k,S.
There is a more general case δx, the translation of δ0 by a factor x, which gives us
δxpϕq : ϕpxq for all ϕ P S pRq .
A core property of this space is that a function f that behaves nicely with all functions in S pRq (called test functions) can be identified with a tempered distribution Tf by the duality product [1]
Tfpϕq xTf, ϕy :
»8
8fpxqϕpxqdx, for all ϕ P S pRq .
This is linear by the linearity of integration and continuous by the Lebesgue con- vergence theorem. The important niceness of f here is that this integral above need to make sense. Indeed if f P L1pRq is at most of polynomial growth (i.e.
p1 |x|qNfpxq P L1pRq for some N P N), then Tf is a tempered distribution[1].
Remark: For functions f we will write xf, ϕy instead of xTf, ϕy when iden- tifying f as a tempered distribution and there is no risk of ambiguity.
To see that this duality product does indeed give us a tempered distribution we can note that given f P L1pRq of at most polynomial growth of degree k (as described above), then we have that there is a number D such that p1 | |qkf L1pRq ¤ D. Furthermore, given ϕ P S pRq we have that for k P N there exists a number Ck such that sup
xPRp1 |x|qkϕpxq ¤ Ck|ϕ|k,S. We now have that
|xf, ϕy|
»8
8fptqϕptqdt
»8
8p1 |t|qkfptqp1 |t|qkϕptqdt
¤ sup
xPRp1 |x|qkϕpxq
»8
8p1 |t|qkfptqdt
¤ sup
xPRp1 |x|qkϕpxq p1 | |qkf L1pRq¤ D Ck|ϕ|k,S for all ϕP S pRq.
2.3.1 Properties of tempered distributions
The tempered distribution behaves linearly [6, p. 6], i.e. for λ, µ P C, f, g P S1pRq and ϕ P S pRq we have
xλf µg, ϕy αxf, ϕy µxg, ϕy
and we define what we mean by differentiation of a tempered distribution by:
Definition 2.17. If f P S1pRq, then the derivative of order m of f, Bmf , is xBkf, ϕy p1qkxf, Bkϕy
Example 2.18. This definition coincides with the classical differentiation of differentiable functions. Let f be a differentiable function such that its derivative is polynomially bounded. Then, using integration by parts, we get that
xBTf, ϕy xf, Bϕy
»8
8fpxqϕ1pxqdx
hkkkkkkkkikkkkkkkkj0
f1pxqϕpxq8
8
»8
8f1pxqϕpxqdx xTf1, ϕy
and so we se that the differentiation of the tempered distribution defined by f correspond to the tempered distribution defined by the derivative of f , as we would expect if we want to avoid ambiguity.
This definition of differentiation has the benefit of moving the differentiation to the smooth test functions, thus not only giving us a way to differentiate tem- pered distributions but also expand up on which functions that are differentiable (as identified as a tempered distribution). As such, since they relax the criteria of differentiation, these derivatives are called distributional derivatives.
Example 2.19. The function
fpxq
#
1, if n¤ x ¤ n 1 0, else
has the distributional derivative δn δn 1 since we have that xBf, ϕy
»8
8fpxqϕ1pxqdx
»n 1 n
ϕ1pxqdx ϕpnq ϕpn 1q
xδn, ϕy xδn 1, ϕy xδn δn 1, ϕy for all test functions ϕ.
2.3.2 Fourier transform of tempered distributions
One of the main reasons for us to introduce tempered distributions is to expand the use of the Fourier transform. To do this we define the Fourier transform of f P S1pRq as the distribution identified through passing the transform over to the test functions. That is, given f P S1pRq, we define the Fourier transform of f , by [1]
xFrfs, ϕy : xf, Frϕsy, for all ϕ P S pRq and, similarly, for the inverse
xF1rfs, ϕy : xf, F1rϕsy, for all ϕ P S pRq .
Since our test functions are Schwartz functions this definition lets us keep prop- erties of the Fourier transform while allowing us to calculate a Fourier transform of any tempered distribution [1, p .24], which contains functions that are not necessarily compatible with the first idea of the Fourier transform, such as the constant functions.
Example 2.20. The Fourier transform of the Dirac delta function δ0: We have that
xFrδ0s, ϕy xδ0,Frϕsy Frϕsp0q
»8
81 ϕpxqdx x1, ϕy and soFrδ0s is identified with the constant function 1.
Example 2.21. The fourier transform of the constant function fpxq 1:
We have that
xFr1s, ϕy x1, Frϕsy
»8
8ϕpξqdξ ϕp0q xδp 0, ϕy and soFr1s is the δ0 distribution.
The Fourier transform keeps the property of being a linear and continuous operator in S1pRq in the sense of the following lemma given by Bahoiuri et al [2, prop. 1.23].
Lemma 2.22. Given a sequencetfkuk inS1pRq that converges to f in S1pRq, thentFrfksuk converges toFrfs.
As stated before, an important tool of the Fourier transform is that of convo- lutions and for tempered distributions we define them by once again passing along the operation to the test functions. As such we can define a convolution between a distribution and a Schwartz function as follows.
Let f P S1pRq and g P S pRq, then f g P S1pRq is defined by xf g, ϕy : xf, ˜g ϕy for all ϕ P S pRq where ˜gpxq : gpxq for all x P R.
Example 2.23. Convolution with δ0. Let gP S pRq then for all ϕ P S pRq pδ0 gqpϕq δ0p˜g ϕq p˜g ϕqp0q
»8
8gpx 0qϕpxqdx gpϕq and so as distributions we have that δ0 g g.
As we will see later, if we use the Fourier transform to acquire solutions to a differential equation it, it is possible to get a solution in the form of a convolution with certain functions (called good kernels [12, p. 139]). To help us establish these solution we observe a property of some sequences of distributions that has a delta-like behaviour which, together with the property of δ0 we saw in the example above, will be of use for us.
Lemma 2.24. [6, p.12] Let βpxq be a nonnegative function with³8
8βpxqdx 1 and let
βεpxq 1 εβpx
εq.
Then βεpϕq Ñ δ0pϕq as ε Ñ 0 for all ϕ P S pRq.
Proof. We have that
βεpϕq δ0pϕq
»8
8βεpxqϕpxqdx ϕp0q
1 ε
»8
8βpx
εqϕpxqdx ϕp0q
»8
8βpxqϕpεxqdx
»8
8βpxqdxϕp0q
»8
8βpxqpϕpεxq ϕp0qqdx
Now since |ϕpεxq ϕp0q| is bounded and ϕpεxq Ñ ϕp0q as ε Ñ 0 for all x P R we have from Lebesgue convergence theorem [7, p. 54] that
»8
8βpxqpϕpεxq ϕp0qqdx
¤»8
8βpxq |ϕpεxq ϕp0q| dx Ñ 0 as ε Ñ 0.
2.4 Pseudo differential operators and Sobolev spaces
A function f onR is said to be locally LppRq, 1 ¤ p ¤ 8, written as LplocpRq, if
»
K
|fpxq|pdx 1{p
8
for all compact K R [1, p. 204]. Clearly every function in LppRq is in LplocpRq while the reverse is not true.
Example 2.25. On R, every nonzero constant function is in LplocpRq but not in LppRq.
Example 2.26. OnR, every continuous function is in LplocpRq by the existence of upper and lower bound on compact intervals for such functions.
Example 2.27. The function fpxq x1 is in LplocpRzt0uq.
2.4.1 Pseudo differential operators
We defined the differential operator D acting on a function f as Df : iBf and as such we had the property zpDϕq ξ pϕ for ϕP S pRq. Using this we can draw
the conclusion that Dϕ pξ pϕq_ by the Fourier inversion theorem (Theorem 2.13). Extending on this, for the linear differential operator Ppx, Dq °
kakDk we define the symbol Ppx, ξq by replacing Dk with ξ i.e.
Ppx, ξq ¸
k
akξk
and as such we have that
pP px, Dqϕqpxq pP px, ξq pϕq_pxq
That is, we can represent a differential operator with a polynomial by the use of the inverse Fourier transform. To further extend upon this we can define the class of operators called pseudo differential operators Ppx, Dq by using a wider set of signs Ppx, ξq than polynomials, where as before
Ppx, Dqϕpxq : 1 2π
»8
8eiξxPpx, ξq pϕpξqdξ.
Using this idea of pseudo differential operators we can create an understanding of our results in this thesis by applying this to the following spaces.
2.4.2 Sobolev spaces
A function f :R Ñ R is called homogenous of degree s P R if fptxq tsfpxq for all t¡ 0 and x 0 [1, p. 34]. By this definition, the function p1 |ξ|qs is not homogenous (also called inhomogenous) while the function|ξ|s is homogeneous of degree s. From this, for s P R, the space of tempered distributions f such that pf P L2locpRq and
kfk2HspRq:
»8
8p1 |ξ|2qs pfpξq2dξ 8
is called the inhomogenous Sobolev space HspRq and the space of tempered distributions f such that pf P L1locpRq and
kfk2H9spRq:
»8
8|ξ|2s pfpξq2dξ 8
is called the homogenous Sobolev space 9HspRq[2, p. 38,25]. These two spaces, while quite similar in their definitions, are not the same space but are linked in that for positive s we have that HspRq 9HspRq and for negative s the reverse is true i.e. 9HspRq HspRq. We will in this thesis concentrate our findings to Schwartz functions, which are in all 9HspRq, and use the homogenous Sobolev norm to acquire our results. As such we will try to understand how to interpret the norm.
We can begin by noting that by Plancherel theorem we have that kfk2H9spRq
»8
8|ξ|2s pfpξq2dξ ξsfp
L2pRq¤?
2πkDsfkL2pRq
where Ds is the pseudo differential operator represented by the symbol ξs, i.e.
Dsf F1rξsfps, and so we have that
f P 9HspRq ðñ Dsf P L2pRq .
As such, we can interpret this in the following way. If s k is a positive integer and f P 9HkpRq then the derivative of order k of f is in L2pRq. So for positive integers k we have that 9HkpRq is the space of distributions whose kth derivative is in L2pRq.
Now, what happens if s is negative? We start with the case s 1 and by noting that
f P 9H1pRq ðñ pF : |ξ|1fpP L2pRq .
To get an understanding of what this means we need to understand F , the inverse of pF defined above. Firstly we note that
DFpxq pξ pFq_pxq pξ |ξ|1fpq_pxq psgnpξq pfq_pxq.
Now define σpξq : sgnpξq and let G : σpDqF . Then G P L2pRq and DGpxq pξσpξq pFq_pxq pσpξq2fpq_pxq fpxq.
And so 9H1pRq consists of those distributions that are a distributional deriva- tive of a function in L2.
This reasoning can be expanded on for when s is a negative integers.
If we want to understand what happens when s in a non-integer number then the idea is to expand on these results to pseudo differential operators. This is, however, outside the scope of this thesis.
2.5 Solving the Heat equation
We have now established a theory we can apply to find solutions to some partial differential equations. At times this requires assuming that the function and its initial data are in S1pRq. To help with some calculations of our solutions we have the following useful lemma.
Lemma 2.28. [2, p.23] Let z be a nonzero complex number with nonnegative real part and let fzpxq ezx2 for xP R. Then fzP S pRq if Repzq ¡ 0 and
F rfzs pξq
?π
?zeξ24z
where?
z |z|1{2eiparg zq{2 and arg zP rπ{2, π{2s.
For z above with Repzq 0, the result holds in a distributional way.
Proof. Firstly, assume that Repzq ¡ 0 and let z a bi with a ¡ 0. Then fzpxq eax2eibx2 and so |fz| ¤ |Gapxq|, where Gapxq eax2, a Gaussian bell function. Since Ga P S pRq by Example 2.1 and |Bαf| ¤ Cαp1 |x|qα|fz|
we have that|fz|m,S ¤ |Ga|m,S 8, and so fzP S pRq.
If Impzq 0 (i.e. z is a positive real number), then we have that F rfzs pξq
»8
8eiξxfzpxqdx
»8
8eiξxzx2dx
eξ24z
»8
8ezpx i2zxq2dx eξ24z
»
γ1
ezw2dw,
where γ1 is the line in C parametrized as wpsq s iξ2z2, 8 s 8. The function ezw2 is analytical onC and vanishes as |w| Ñ 8. Cauchy’s integral theorem [10, p. 187] now gives us that we can change the integration over γ1 to integrating over γ2, the curve parametrized by wpsq s, 8 s 8 (i.e. the real line inC), giving us that
»
γ1
ezw2dw
»
γ2
ezw2dw
»8
8ezs2ds
?π
?z, and thus we have that
F rfzs pξq
?π
?zeξ24z.
Both F rfzs and ??πzeξ24z are analytical (w.r.t. z) on D tz P C : Repzq ¡ 0u and since they are equal on the part of D that are strictly real then they are equal on all of D [10, p. 297].
Now let z s.t. Repzq 0, i.e. z it for t P R, and let tznun D be a sequence s.t. zn Ñ it as n Ñ 8, then by Lebesgue’s dominated convergence theorem [7, p. 54] we have for all ϕP S pRq that
nlimÑ8
»8
8eznx2ϕpxqdx
»8
8eitx2ϕpxqdx and that
nlimÑ8
»8
8
?π
?zne4znξ2 ϕpxqdx
»8
8
?π
?ite4itξ2ϕpxqdx.
Now, sinceF
eznx2
pξq ??zπ
ne4znξ2 for each znwe have by Lemma 2.22 that as tempered distributions
F rfzs pξq
?π
?zeξ24z.
The heat equation is a partial differential equation that describes how heat disperses over time and is what inspired Joseph Fourier to establish the theory that grew into Fourier analysis. In one space dimension with initial heat given by f P S1pRq the equation is as follows [12, p. 146].
#Btupx, tq B2xupx, tq upx, 0q fpxq
To solve this assume that for all t we have upx, tq P S1pRq. We can now use the Fourier transform on this equation (with respect to x) and get
FrBtuspξ, tq FrBx2uspξ, tq
and so, using the theory of differentiation and the Fourier transform from The- orem 2.6 on the right hand side and Leibniz integral rule on the left hand side we have that
Btpupξ, tq ξ2pupξ, tq and thus that
pupξ, tq Apξqeξ2t.
This gives us that pupξ, 0q Apξq and from the initial condition we have that pupξ, 0q pfpξq, so our solution becomes
pupξ, tq eξ2tfppξq.
Now for t¡ 0, by Lemma 2.28 above with z t we have Htpxq : F1
eξ2t
pxq 1 2π
»8
8eiξxeξ2tdξ
1 2π
»8
8eiηxeη2tdη 1 2πF
eξ2t
pxq ex24t
?4πt
and so, by Theorem 2.10, we get our solution as upx, tq pHt fqpxq
and we are left with confirming that upx, tq Ñ fpxq as t Ñ 0.
Now, we note that if βpxq ?1πex2 then ³8
8βpxqdx 1 and since Htpxq
?1
4tβp?x4tq we have that Htpxq Ñ δ0as tÑ 0 by lemma 2.24. And so we have, for fixed x, that inS1pRq
u Ht f Ñ δ0 f f as t Ñ 0.
2.6 Solving the time dependent Schr¨ odinger equation
Our main focus in this thesis is the equation proposed by Erwin Schr¨odinger in the early 20th century to describe a wave equation for matter. The need for this equation came from the discoveries in particle physics where, under certain conditions a particle has a wavelike behaviour[4]. The equation presented by Schr¨odinger was
i~Btupx, tq p ~2
2m∆ Vpx, tqqupx, tq
where x P Rn, t ¡ 0 [3]. Solutions to this equation describes probabilities of particles being located in certain regions. We will now consider the case of one space dimension so that the Laplacian ∆ of the space variables is B2x and we
will omit the positive physical terms ~ and 2m (for clarity) and assume that the potentiality term V is zero. As such the free time dependent Schr¨odinger equation in one space dimension [13] with initial datum fP S1pRq states that
#iBtupx, tq B2xupx, tq
upx, 0q fpxq. (3)
To solve this we begin a similar approach as for the heat equation above and using the Fourier transform (with respect to x) gives us that
Btpupξ, tq iξ2pupξ, tq and so our solution becomes
pupξ, tq pgtpξq pfpξq wheregptpξq eitξ2.
First we note that we have for finite t¡ 0 and a P N that |Bξagpt| ¤ Cap1 |ξ|qa and given hP S pRq then for any α, N P N there is a Cα,N such that
|Bξαh| ¤ Cα,Np1 |ξ|qN and so there is a Cα,N such that
¸
a¤α
|Baξh| ¤ Cα,Np1 |ξ|qNα
thus, for any α, N P N, Leibniz rule gives us
|Bαξp pgthq| ¸
a bα a,b¥0
Ca,b|pBaξgptqpBbξhq|
¤ p1 |ξ|qα ¸
0¤b¤α
Cb|Bξbh| ¤ Cα,Np1 |ξ|qαp1 |ξ|qαN
¤ Cα,Np1 |ξ|qN
and so for all hP S pRq we have pgthP S pRq.
We have that
| pgt 1| eitξ2 1
»1 0
eitsξ2dspitξ2q
¤ t|ξ|2
and, for t 1, we have for α ¥ 1 that |Bξαgpt| ¤ Cαtp1 |ξ|qα. Given m P N
then, using the same method as above for the sum, we have that
| pgth h|m,S sup
ξPR sup
α β¤m||ξ|βBαξp pgtpξqhpξq hpξqq|
sup
ξPR sup
α β¤m|ξ|β|p pgt 1qBαξh ¸
a bα a¡0, b¥0
Ca,bpBξagptqpBξbhq|
¤ sup
ξPR sup
α β¤m|ξ|βt|ξ|2|Bξαh| tp1 |ξ|qα ¸
0¤b¤α
Cb|Bbξh|
¤ sup
ξPR sup
α β¤mtCα,β|ξ|β 2p1 |ξ|qβ3 tCαp1 |ξ|qαp1 |ξ|qα1
¤ sup
ξPR sup
α β¤mtCα,βp1 |ξ|q1 tCαp1 |ξ|q1¤ tD and so for all hP S pRq we have | pgth h|m,S Ñ 0 as t Ñ 0.
Now, given f P S pRq then pf P S pRq and since pupξ, tq pgtpξq pfpξq and pupξ, 0q fppξq we have from our previous result, with h pf , that pupξ, tq Ñ pfpξq in S pRq as t Ñ 0. Since F1 is continuous, we have upx, tq F1r pgtfpspxq Ñ F1r pfspxq fpxq as t Ñ 0. That is, u Ñ f in S pRq as t Ñ 0.
Let ˜fpxq fpxq, then ˜gt gt and, from the results above, for all ϕ P S pRq we have gt ϕ Ñ ϕ in S pRq as t Ñ 0 and so, using convolutions in S1pRq, we have
pgt fqpϕq fp˜gt ϕq fpgt ϕq.
So by the continuity of f and the results above we have that fp˜gt ϕq Ñ fpϕq as t Ñ 0
i.e.
gt f Ñ f as t Ñ 0 in S1pRq . Now, if f P L2pRq, then, using Plancherel
kukL2pRq p2πq1{2kpukL2pRq p2πq1{2 pgtfp L2pRq
p2πq1{2 pf
L2pRq kfkL2pRq
and so for t¡ 0 we have u P L2pRq.
Finally, using Plancherel again, we have
ku fkL2pRq p2πq1{2 pu pf L2pRq
p2πq1{2 p pgt 1q pf
L2pRq
and so, since gpt 1 Ñ 0 as t Ñ 0, Lebesgue dominated convergence theorem gives us thatku fkL2pRqÑ 0 as t Ñ 0, i.e. u Ñ f in L2pRq as t Ñ 0.
We can now summarize our solution as upx, tq pgt fqpxq
»8
8gtpx yqfpyqdy (4)