1.2 The Vlasov part
1.2.2 Solutions of the Vlasov part
β«
πΎπβ’+
π+πΌπππΎπΌβ
β«
πΎπβ’β
π¦πΌπ+πΌ+ππΌ
πππΎπΌβ
β«
Ξ©
β«
R3
πΛπΌπ(0)ππ£ππ₯
for allπβΞ¨πβ’.
We explain in the following remark in what sense we can speak of traces ππΌ + of ππΌ
in a weak solution concept.
Remark 1.2.1. If Definition 1.1.1.(ii) is satisfied,π+πΌis the trace of ππΌin the following sense:
β’ As we have just seen, π+πΌis the restriction of ππΌtoπΎπ+β’if ππΌ βπΆ1 πΌπ
β’ΓΞ©ΓR3
.
β’ There is no otherπΛπΌ + βπΏ1
loc
πΎπ+β’
such that Definition 1.1.1.(ii) is satisfied as well, since for such πΛ+πΌwe have
β«
πΎπβ’+
π+πΌβ Λπ+πΌ
πππΎπΌ =0
for allπβπΆβ πΌπ
β’ΓΞ©ΓR3
with suppπβ [0, πβ’[ ΓΞ©ΓR3compact that vanish onπΎπββ’βͺπΎπ0β’. Consequently,πΛ+πΌ= π+πΌ.
1.2.2 Solutions of the Vlasov part
We give a brief introduction to the techniques and statements of Beals and Pro-topopescu [BP87], who used an approach via characteristics to tackle linear transport problems with initial-boundary conditions in a very general setting. Since we do not need the full statements of [BP87], we formulate those results in the way we will need them in our situation.
Throughout this subsection, letπ >0,Ξ© βR3 be an open, bounded set withπΆ1,π -boundary for someπ >0, andΞ£π B]0, π[ ΓΞ©ΓR3. Furthermore, letπbe a first order linear differential operator that is divergence free and whose coefficients are Lipschitz continuous onΞ£π. In accordance to our situation, we choose
πB ππ‘+
bπ£πΌΒ·ππ₯+πΉΒ·ππ£.
Thus, the assumptions aboutπhere reduce to two conditions onπΉ, namely, thatπΉis Lipschitz continuous onΞ£π and divergence free with respect to π£. We additionally assume thatπΉis bounded onΞ£π. By Lipschitz continuity ofπΉ, for each(π‘, π₯, π£) βΞ£π
there is a well-defined integral curveπ β¦β (π, π , π)(π , π‘, π₯, π£)satisfying π
ππ π=1, π
ππ π=πbπΌ, π
ππ π=πΉ(π , π , π), (π, π , π)(π‘, π‘, π₯, π£)=(π‘, π₯, π£).
This curve is defined as long as it remains inΞ£πand there is a corresponding maximal time intervalπΌ βRfor which it is defined. We define the length of this curve to be the
length of the maximal time interval for which the curve remains inΞ£π, that is to say, the length equalsπ +βπ βwhere
π +Bsup{π βπΌ | (π, π , π)(π , π‘, π₯, π£) βΞ£π}, π βBinf{π βπΌ | (π, π , π)(π , π‘, π₯, π£) βΞ£π}.
The next assumption is that there is a finite upper bound to all lengths of such integral curves. This condition is trivially satisfied in our caseΞ£π =]0, π[ ΓΞ©ΓR3sinceπis an upper bound. The last assumption is that each integral curve has a left and right limit point, i.e.,
π βπ limβ,π >π β(π, π , π)(π , π‘, π₯, π£),π βπ lim+,π
<π +(π, π , π)(π , π‘, π₯, π£) βΞ£π.
These limits, if they exist, have to be elements ofπΞ£π. For their existence it is sufficient that πΉ is bounded by some constant πΆ > 0 since then both πΒ€ and πΒ€ are bounded because of
πΒ€
=
πbπΌ
β€1,
πΒ€
=|πΉ(π , π , π)| β€πΆ.
Accordingly, we define π·πβ (π·π+) to be the subset of πΞ£π consisting of all such left (right) limits, often referred to as incoming (outgoing) sets. These sets are Borel sets since π·πβ (π·π+) is the image of the open set Ξ£π under the continuous function that maps a point inΞ£πto the left (right) limit point of the integral curve passing through this point. Note that possibly π·πΒ± are not disjoint and/or do not exhaust πΞ£π but bothπ·π+β©π·πβ andπΞ£π\ π·π+βͺπ·πβ
are negligible in the sense that the union of all associated integral curves inΞ£πhas Lebesgue measure zero.
We proceed with the definition of the test function space corresponding toπ. Definition 1.2.2. LetΞ¦ππbe the space of all measurable functionsπ:Ξ£πβRwith the following three properties:
(i) πis continuously differentiable along each integral curve.
(ii) πandππare bounded functions.
(iii) The support ofπis bounded and there is a positive lower bound to the lengths of the integral curves which meet the support ofπ.
Remark 1.2.3. β’ Here and in the following, the termπ β, where β β πΏ1
loc(Ξ£π), is in general to be understood as a distribution, i.e.,
(π β) π
=β
β«
Ξ£π
ππ‘π+
bπ£πΌΒ·ππ₯π+πΉΒ·ππ£πβ π(π‘, π₯, π£), πβπΆβπ (Ξ£π).
In Definition 1.2.2.(ii) or later in Definition 1.2.7.(i), this distribution is assumed to be given by a function onΞ£π.
β’ Because of Definition 1.2.2.(ii) and 1.2.2.(iii) we haveπ, ππβπΏπ(Ξ£π)for anyπβΞ¦ππ and 1β€ πβ€ β.
1.2 The Vlasov part 17
β’ Note that a functionπβ Ξ¦ππ only has to be continuously differentiable along each integral curve but may be discontinuous in other directions. Because of Defini-tion 1.2.2.(i) and 1.2.2.(ii) every π β Ξ¦ππ can be extended to be continuous at the endpoints of each integral curve.
SinceΞ¦ππdepends onπΉ, it cannot be suitable for the whole nonlinear system (VM), whereπΉis unknown. Thus, an important (technical) statement is that our test function spaceΞ¨πβ’, which is independent ofπΉ, belongs toΞ¦ππafter a cut-off in the time variable (ifπβ€πβ’). This is verified in the following two lemmas, where we follow the proof of [Guo93, Lemma 2.1.].
Lemma 1.2.4. (i) For any π > 0 there is a πΏ = πΏ(π) > 0 such that for all (π₯, π£) β ΛπΎβ satisfyingdist (π₯, π£),πΎΛ0
> πwe havebπ£πΌΒ·π(π₯) β€ βπΏ.
(ii) For anyπ >0there is anπ=π(π)>0such that for anyπ₯ βπΞ©,π¦ βR3we haveπ¦βΞ© if
π¦βπ₯
< πand π¦βπ₯
Β·π(π₯) β€ β π
π¦βπ₯ <0.
Proof. As for part 1.2.4.(i), suppose the contrary. Then we can find a π > 0 and a sequence(π₯π, π£π) β ΛπΎβ with dist (π₯π, π£π),πΎΛ0 > π
forπ β Nandbπ£π,πΌ Β·π(π₯π) β 0 for πβ β. Without loss of generality we can assume that(π£π)is bounded: If|π£π| β₯1 let π€π B |π£π£ππ|. Then,
0>π€bπ,πΌΒ·π(π₯π)=|
π€bπ,πΌ|cos(](
π€bπ,πΌ, π(π₯π))) β₯ |
bπ£π,πΌ|cos(](
bπ£π,πΌ, π(π₯π)))
=bπ£π,πΌΒ·π(π₯π) β0 forπβ βbecause of|
π€bπ,πΌ| β€ | bπ£π,πΌ|.
Therefore,(π₯π, π£π) βπΞ©ΓR3converges, after extracting a suitable subsequence, to some(π₯, π£) βπΞ©ΓR3. On the one hand, we have dist (π₯, π£),πΎΛ0β₯π, and on the other handbπ£πΌΒ·π(π₯)=0 which is a contradiction.
The proof of part 1.2.4.(ii) exploits that πΞ© is of class πΆ1,π . Suppose that the assertion does not hold, i.e., we can find aπ >0 and sequences(π₯π) β πΞ©, π¦π β
R3 with
π¦πβπ₯π
< 1π and π¦πβπ₯πΒ·π(π₯
π) β€ βπ π¦πβπ₯π
<0 butπ¦πβ Ξ©. We may assume that both sequences converge because of(π₯π) βπΞ©and π¦πβ
πΞ©+π΅1. The limits of both sequences have to be the same; we call the limitπ₯ βπΞ©. Sinceπ₯π+π‘ π¦πβπ₯πβ
Ξ© forπ‘ >0 small enough andπ¦π β Ξ©, there has to be aπ₯Λπ β π₯
π, π¦π
β©πΞ©. Obviously we have| Λπ₯πβπ₯π| < 1π and
( Λπ₯πβπ₯π) Β·π(π₯π)= π¦πβπ₯πΒ·π(π₯
π)| Λπ₯πβπ₯π| π¦πβπ₯π
β€ βπ| Λπ₯πβπ₯π|<0. (1.2.5) SinceπΞ©is compact andπ₯Λπ β π₯,π₯π β π₯forπ β β,π₯Λπ,π₯π, andπ₯ lie in the image of the same πΆ1,π -chart π:R2 β π β πΞ© if πis large enough. Let πΛπ B πβ1( Λπ₯π), ππ B πβ1(π₯π), andπ B πβ1(π₯). By continuity ofπβ1, both πΛπ
and ππ
converge to π. Thus, we may assume thatπΛπ, ππβπΎπ Bπ΅π π βπ
for suitableπ >0 and largeπ. We expand the left-hand side of (1.2.5) to get the estimate
|( Λπ₯πβπ₯π) Β·π(π₯π)|
=
. Together with (1.2.5) and the fact that πβ1 is Lipschitz continuous on π πΎπ
with some Lipschitz constant πΏπ,πΎ
π >0βsee proof belowβ, this yields for largeπ 0< π
But this contradicts πΛπβππ
β0 forπβ β.
So there remains to show the Lipschitz continuity ofπβ1onπ πΎπ
. This relies on the fact that, sinceπis a chart, the function
πΊ:πΎπΓππ΅1 βR, πΊ π,Λ πΏπ
is continuous and positive so that it is bounded from below by some positive constant π>0. Forπ₯, π₯Λ βπ πΎπ
also converge to the same limit due to the continuity ofπβ1. But this contradicts (1.2.6). Hence,πβ1is Lipschitz continuous on π πΎπ
and the proof is finished.
Lemma 1.2.5. For eachπ βΞ¨πwe haveπ
1.2 The Vlasov part 19 Sinceπ
Ξ£
π obviously satisfies Definition 1.2.2.(i) and 1.2.2.(ii), we only have to take care about Definition 1.2.2.(iii). First note that, since the support ofπ is compact in [0, π[ ΓΞ©ΓR3, there is a 0 β€ π 0 < πsuch that π(π‘, π₯, π£)= 0 forπ 0 β€ π‘ <π,π₯ β Ξ©, π£βR3.
We consider an integral curve which meets suppπ = suppπ
Ξ£π. This curve can be written asπ β¦β (π, π , π)(π , π β, π₯, π£)and remains inΞ£πfor a maximal time interval ]π β, π +[ β ]0, π[so that(π β, π₯, π£) βπ·πβ. Obviously it holds thatπ β β€ π 0. We have to find a positive lower bound for π +βπ βthat does not depend onπ β,π +,π₯, andπ£. In the following, letπ β ]π β, π +[.
Case 1. If
dist (π β, π₯, π£),suppπβ₯ π0
2
we can find an π such that(π, π , π)(π , π β, π₯, π£) β suppπ since the curve meets the support ofπ. By
πΒ€
β€1 and πΒ€
β€supΞ£
π|πΉ|we have π0
2 β€dist (π β, π₯, π£),suppπ β€ |(π, π , π)(π , π β, π₯, π£) β (π β, π₯, π£)|
β€ q
2+ kπΉk2β(π βπ β) so that π0
2
β
2+kπΉk2β is such a desired lower bound in this case.
Case 2. The more complicated case is
dist (π β, π₯, π£),suppπ
< π0 2. Since{π} ΓΞ©ΓR3andπΎπ+do not intersectπ·πβ, we have
π·πββπΎπββͺπΎπ0 βͺ
{0} ΓΞ©ΓR3 .
Clearly, it holds that(π β, π₯, π£)βπΎπ0 because of dist (π β, π₯, π£),suppπ
< π0
2 <π0 β€dist suppπ,πΎπ0
. If(π β, π₯, π£) β {0} ΓΞ©ΓR3we have
dist(π₯,πΞ©)=dist (π β, π₯, π£),{0} ΓπΞ©ΓR3
β₯dist suppπ,{0} ΓπΞ©ΓR3β
dist (π β, π₯, π£),suppπ
>π0βπ0 2 = π0
2 . Thus,π(π , π β, π₯, π£) βΞ©for 0β€π <minnπ
0 2 , πo
again because of πΒ€
β€1. Therefore, a positive lower bound to the length of the integral curve in this case is minnπ
0 2 , πo
.
Finally, suppose(π β, π₯, π£) βπΎπβ. First note that dist (π₯, π£),πΎΛ0
=dist (π β, π₯, π£),πΎπ0
β₯
dist suppπ,πΎπ0
β
dist (π β, π₯, π£),suppπ
>π0β π0 2 = π0
2 . (1.2.7)
LetπΏ=πΏπ
0 2
andπ=π πΏ2
according to Lemma 1.2.4. We claim that
πBmin (
πβπ 0,π
2, πΏ
9
2kπΉkπΏβ(Ξ£π;R3)+1 )
is such a positive lower bound (to the length of the integral curve) we search for.
Indeed, we firstly have[π β, π β+π] β [0, π]due toπ β β€π 0. Secondly, let π Bsup{π β ]π β, π β+π] |π(Λπ , π β, π₯, π£) βΞ©for allπ Λβ ]π β, π [}.
Because of
|π(π , π β, π₯, π£) βπ₯| β€π βπ β < π and
(π(π , π β, π₯, π£) βπ₯) Β·π(π₯)=
β« π
π β
πbπΌ(π, π β, π₯, π£)ππ
!
Β·π(π₯)
=(π βπ β)
bπ£πΌΒ·π(π₯) +
β« π
π β
β« π π β
π ππ πbπΌ
(π, π β, π₯, π£)ππππΒ·π(π₯)
β€ βπΏ(π βπ β) +9
2kπΉkπΏβ(Ξ£π;R3)Β·1
2(π βπ β)2 β€ βπΏ
2(π βπ β) β€ βπΏ
2|π(π , π β, π₯, π£) βπ₯| (which also impliesπ(π , π β, π₯, π£)β π₯ sinceβπΏ
2(π βπ β) <0) by (1.2.7) and
πbπ£πΌ,π ππ£π
β€
3 2, π, π = 1,2,3, we have π(π , π β, π₯, π£) β Ξ© and thus π = π β+π. This completes the proof.
We should remark that the three conditions onπβΞ¨πin (1.1.1) are really necessary:
Letπ >0 be small and, for simplicity, takeπΉ =0. Firstly, if we allow a test functionπ that does not vanish before timeπand has support onπΎπβ, we can find an integral curve enteringΞ£πonπΎπββ©suppπat timeπ β =πβπ. Secondly, if we allow a test function π with support onπΎπ0, then for some(π‘, π₯, π£) β πΎπ0βsuch that in a neighborhood of π₯ there are no common points of Ξ© and the tangent space of πΞ© at π₯βthe curves (π, π , π)(π ) = (π , π₯βππ(π₯) + (π βπ‘)
bπ£πΌ, π£), defined for all π β [0, π], will meet the support ofπ. Thirdly, if we allow a test functionπ with support on{0} ΓπΞ©ΓR3 we can find an integral curve meeting the support ofπ, (itsπ-coordinate) starting at time 0 nearπΞ©and leavingΞ©at timeπ. In all three cases, there will be no positive lower bound to the length of these curves.
1.2 The Vlasov part 21 Conversely, these restrictions cause no problems for later considerations. Firstly, we do not want to test a solution of (1.2.1) at timeπsince we are interested in an initial, and not final, value problem. Secondly, we only want to impose a boundary condition on πΎβ and not on πΎ0. Thirdly, proper initial data of the distribution function have to satisfy the boundary condition at time 0 a priori so that this property need not be tested, and{0} ΓπΞ©ΓR3is even a null set with respect toππΎπΌ.
We now proceed with some important results of [BP87]. There, the main idea is to use the βidentificationsβ
Ξ£π β(π , π§) |π§βπ·β
π,0<π <π(π§) , πβ π ππ ,
whereπ(π§)is the length of the integral curve corresponding toπ§. The first important result is the following property which is closely related to Greenβs identity; see [BP87, Proposition 7].
Proposition 1.2.6. There are two unique Borel measuresπΒ±onπ·Β±
πsuch that
β«
Ξ£π
πππ(π‘, π₯, π£)=
β«
π·+ππππ+β
β«
π·πβπππβ for allπβΞ¦ππ.
We have to define the space of functions in which we search for solutions of some initial-boundary problem.
Definition 1.2.7. For 1 β€ π < βletπΈπ(Ξ£π;π)be the space of functions π β πΏπ(Ξ£π) with the following two properties:
(i) π π βπΏπ(Ξ£π).
(ii) There is a trace of π onπ·Β±
π, i.e., a pair of functions πΒ± βπΏπ π·Β±
π, ππΒ±
satisfying the extended Greenβs identity
β«
Ξ£π
ππ π +π πππ(π‘, π₯, π£)
=
β«
π·+ π
π+πππ+β
β«
π·β π
πβπππβ
for allπβΞ¦ππ.
Note that a trace in the sense as stated above is unique and that all terms are well-defined according to Remark 1.2.3.
Lemma 1.2.8. Let 1 β€ π < β, π β πΈπ(Ξ£π;π) and π€ β πΆβ(Ξ£π) β©πΆ1π Ξ£π
. Then, π€ π βπΈπ(Ξ£π;π)and π€ πΒ±
=π€ πΒ±. Proof. Because of
π π€ π π
=β
β«
Ξ£π
ππ‘π+
bπ£πΌΒ·ππ₯π+πΉΒ·ππ£ππ€ π π(π‘, π₯, π£)
=β
β«
Ξ£π
ππ‘ π€π+
bπ£πΌΒ·ππ₯ π€π+πΉΒ·
ππ£ π€π π π(π‘, π₯, π£)
+
β«
Ξ£π
(ππ‘π€+
bπ£πΌΒ·ππ₯π€+πΉΒ·ππ£π€)πππ(π‘, π₯, π£)
=
β«
Ξ£π
π ππ€
ππ(π‘, π₯, π£) +
β«
Ξ£π
(ππ€)πππ(π‘, π₯, π£)
for any π β πΆβπ (Ξ£π), it holds thatπ π€ π
= π€π π + π ππ€ β πΏπ(Ξ£π). Now letπ β Ξ¦ππ. We haveπ€π β Ξ¦ππ since Definition 1.2.2.(i) and 1.2.2.(ii) are satisfied because of the regularity ofπ€and Definition 1.2.2.(iii) is satisfied because of supp π€π β
suppπ.
Thus, it holds that
β«
Ξ£π
ππ π€ π+π€ π π
ππ(π‘, π₯, π£)
=
β«
Ξ£π
π€ππ π +π π π€π π(π‘, π₯, π£)
=
β«
π·+ π
π+π€πππ+β
β«
π·β π
πβπ€πππβ,
which proves the assertion.
In the following it is convenient to splitπ·πΒ±as follows:
π·Β±πB (π‘, π₯, π£) βπ·πΒ±|0<π‘<π ,
π·0 B (π‘, π₯, π£) βπ·βπ |π‘=0 , π·ππ B (π‘, π₯, π£) βπ·π+ |π‘=π , so thatπ·β
π =π·βπβͺπ·0andπ·+
π =π·+πβͺπ·π
π. Note thatπ·0does not depend onπ(in the sense that any 0<πΛ <πyields the same setπ·0). According to this decomposition we writeππβ=ππβ
π·π
β,ππ0=ππβ π·
0,ππ+=ππ+ π·π
+,πππ=ππ+ π·π
π, πβ= πβ π·π
β, π0= πβ π·
0, π+= π+
π·π
+, and ππ= π+ π·π
π. We have
{0} ΓΞ©ΓR3 βπ·0 β {0} ΓΞ©ΓR3, {π} ΓΞ©ΓR3βπ·ππβ {π} ΓΞ©ΓR3, πΎπββπ·πβ βπΎπββͺπΎπ0, πΎπ+ βπ·+πβπΎπ+βͺπΎπ0.
Therefore, we can identify πΏπ-functions on π·0 (orπ·ππ) withπΏπ-functions onΞ©ΓR3 since
Ξ©ΓR3
\ Ξ©ΓR3
has(π₯, π£)-Lebesgue measure zero. Additionally, we may write π(0)and π(π)instead of π0and ππpointing out that we may evaluate π at time 0 andπin some sense.
For eachπβΞ¨πwe have
β«
Ξ£π
πππ(π‘, π₯, π£)=β
β«
Ξ©
β«
R3
π(0)ππ£ππ₯+
β« π
0
β«
πΞ©
β«
R3
πbπ£πΌΒ·π ππ£πππ₯ππ‘
=β
β«
Ξ©
β«
R3
π(0)ππ£ππ₯+
β«
πΎπ+
πππΎπΌβ
β«
πΎβπ
πππΎπΌ.
1.2 The Vlasov part 23 This shows thatππ0 =π(π₯, π£)onπ·0andππΒ± =ππΎπΌonπΎπΒ±. With an analog reasoning (consider test functions πΛ(π‘, π₯, π£) = π(πβπ‘, π₯, π£),π β Ξ¨π) we conclude that πππ = π(π₯, π£)onπ·ππas well.
We proceed with a definition of some properties of operators.
Definition 1.2.9. Let π be an operator between two function spaces on subsets of someRπ, whose first component we call time. πis called
(i) local in time ifπ(π’π£)=π’π(π£)for all continuous functionsπ’that only depend on time and all possibleπ£;
(ii) nonnegative ifπ(π£) β₯0 for allπ£ β₯0.
Now we are ready to state the following result regarding the unique solvability of linear transport problems with initial-boundary conditions; see [BP87, Proposition 1, Theorems 1 and 2].
Proposition 1.2.10. Let1 β€ π < β, β β πΏβ(Ξ£π),πΉ:Ξ£π β R be Lipschitz continuous, differentiable with respect toπ£, and divergence free with respect toπ£, andπ=ππ‘+
bπ£πΌΒ·ππ₯+πΉΒ·ππ£. (i) For all π βπΈπ(Ξ£π;π)we have
β«
π·π π
ππ
ππππ+
β«
π·π +
π+
πππ++π
β«
Ξ£π
β π
ππ(π‘, π₯, π£)
=
β«
π·0
π0
πππ0+
β«
π·πβ
πβ
πππβ+π
β«
Ξ£π
sign π π
πβ1(π+β)π π(π‘, π₯, π£). (1.2.8)
(ii) Let moreoverπ: πΏπ π·+π, ππ+ β πΏπ π·πβ, ππβ
be a bounded linear operator, that is local in time and has operator norm less than1, andπ0 βπΏπ(π·0),πββ πΏπ π·πβ, ππβ
. Then the problem
π π =0 onΞ£π, (1.2.9a)
π0 =π0 onπ·0, (1.2.9b)
πβ = ππ++πβ onπ·πβ (1.2.9c)
has a unique solutionπ βπΈπ(Ξ£π;π). Here,(1.2.9a)holds pointwise almost everywhere (cf. Definition 1.2.7.(i) and Remark 1.2.3), and(1.2.9b)and (1.2.9c)hold pointwise almost everywhere (with respect to the corresponding measures) in the sense of trace (cf. Definition 1.2.7.(ii)). Moreover, the solution is nonnegative ifπ, π0, and πβ are nonnegative.
Here and in the following, for functions the property βnonnegativeβ usually means
βnonnegative almost everywhereβ. We want to express, in some way, the theorem above in words that fit to our problem (1.2.1), that is to say, we should somehow replace π·0,π·πβ (and so on) byΞ©ΓR3,πΎπβ(and so on). Moreover, we search for solutions of (1.2.1) onπΌπ
β’ instead of solutions on some time interval [0, π]. To this end, we first have to define what we call a strong solution of (1.2.1). From now on, the force term πΉshall satisfy the following condition.
Condition 1.2.11. πΉ:πΌπ
β’ ΓΞ©ΓR3 βR3 is Lipschitz continuous and bounded onΞ£π
for anyπ β πΌπβ’, and moreover differentiable and divergence free (both) with respect toπ£.
Definition 1.2.12. Assume that 1 β€ π < β,π¦: πΏπ
lt
πΎπ+β’, ππΎπΌ
β πΏπ
lt
πΎπββ’, ππΎπΌ
, πΛ β πΏπ Ξ©ΓR3
, and π β πΏπ
lt
πΎπββ’
, ππΎπΌ
. We call a function π:πΌπ
β’ ΓΞ©ΓR3 βRa strong solution of (1.2.1) if:
(i) πππ βπΈπ(Ξ£π;π)for all 0<πβπΌπ
β’. (ii) For allπ βΞ¨πβ’it holds that
β« πβ’ 0
β«
Ξ©
β«
R3
ππ‘π+
bπ£πΌΒ·ππ₯π+πΉΒ·ππ£ππ ππ£ππ₯ππ‘
=
β«
πΎ+πβ’
π+πππΎπΌβ
β«
πΎβπβ’
π¦π++π
πππΎπΌβ
β«
Ξ©
β«
R3
πΛπ(0)ππ£ππ₯.
Note that, for each 0<π βπΌπ
β’, at first only a trace ofπππ is defined. By uniqueness, for another πΌπ
β’ 3 π0 β₯ π, the traces of ππ0π and πππ coincide on the common time interval[0, π]. Thus, we may write πΒ±, which is defined on all ofπΌπ
β’, and may drop the dependence on someπ.
Proposition 1.2.13. Let1 β€ π <β,πΉ satisfy Condition 1.2.11, andπ¦:πΏπ
lt
πΎπ+β’, ππΎπΌ
β πΏπ
lt
πΎπββ’, ππΎπΌ
be a linear operator, that is local in time and such that there is a0 β€ π0 < 1 satisfying
kπ¦πkπΏπ(πΎπβ,ππΎπΌ) β€π0kπkπΏπ(πΎπ+,ππΎπΌ) for all π β πΏπ
lt
πΎπ+β’, ππΎπΌ
, 0 < π β πΌπ
β’. Furthermore, let πΛ β πΏπ Ξ©ΓR3
and π β πΏπ
lt
πΎπββ’, ππΎπΌ
. Then:
(i) There is exactly one strong solution of (1.2.1)in the sense of Definition 1.2.12.
(ii) This solution is nonnegative ifπ¦, πΛ, andπare nonnegative.
Proof. Let 0<πβπΌπβ’and define
ππβ:π·πβ βR, ππβ(π‘, π₯, π£)=
(π(π‘, π₯, π£), (π‘, π₯, π£) βπΎπβ, 0, otherwise;
π0:π·0βR, π0(0, π₯, π£)= πΛ(π₯, π£).
Note that the latter definition makes sense since, as mentioned above,π·0 coincides with{0} ΓΞ©ΓR3up to a negligible set. We have
π
π
β
πΏπ(π·πβ,ππβ)= πππ
πΏπ(πΎβ,ππΎπΌ), π0
πΏπ(π·0)=
πΛ πΏπ(Ξ©ΓR3)
1.2 The Vlasov part 25 so that πβπβπΏπ π·πβ, ππΒ±
andπ0βπΏπ(π·0). Furthermore, forβ βπΏπ π·+π, ππ+ let β:πΎπ+β’ βR, β(π‘, π₯, π£)=
(β(π‘, π₯, π£), (π‘, π₯, π£) βπΎπ+, 0, otherwise and
ππβ:π·βπβR, (ππβ)(π‘, π₯, π£)=
( π¦β
(π‘, π₯, π£), (π‘, π₯, π£) βπΎπβ, 0, otherwise. Because of
kππβkπΏπ(π·πβ,ππβ)= πππ¦β
πΏπ(πΎπβ,ππΎπΌ) β€π0 ππβ
πΏπ(πΎπ+,ππΎπΌ)=π0kβkπΏπ(π·π+,ππ+) we conclude that ππ maps πΏπ π·+π, ππ+
to πΏπ π·βπ, ππβ
and has operator norm less than 1. Moreover,ππis local in time. Thus, by Proposition 1.2.10.(ii) there is a solution of (1.2.9) (withππ, π0, ππβ given). By uniqueness andπ·βπ β π·βπ0 forπ β€ π0, such a solution (and its trace) does not depend onπ, whence there is a function π such that πππ β πΈπ(Ξ£π;π)is the unique solution of (1.2.9) for any givenπ. Now takeπ β Ξ¨πβ’
and 0<π <πβ’such thatπ(π‘, π₯, π£)=0 ifπ‘ β₯π. By Lemma 1.2.5 we haveπ
Ξ£π βΞ¦ππ. Applying the definition of trace and using the properties ofπ, this leads to
β«
Ξ£πβ’
π πππ(π‘, π₯, π£)=
β«
Ξ£π
π πππ(π‘, π₯, π£)
=β
β«
π·0
ππππ0+
β«
π·+π
π+πππ+β
β«
π·βπ
πππ++ππβπππ
β
=β
β«
Ξ©
β«
R3
πΛπ(0)ππ£ππ₯+
β«
πΎ+π
π+πππΎπΌβ
β«
πΎπβ
πππ++πβπ πππΎπΌ
=β
β«
Ξ©
β«
R3
πΛπ(0)ππ£ππ₯+
β«
πΎ+πβ’
π+πππΎπΌβ
β«
πΎπβ’β
π¦π++ππππΎ
πΌ
and the proof of part 1.2.13.(i) is complete.
Part 1.2.13.(ii) follows from the fact thatππ,π0, andππβare nonnegative ifπ¦,πΛ, and πare nonnegative, and Proposition 1.2.10.(ii).
We now turn to the special situation thatπ¦=ππΎ, whereπΎis the reflection operator.
According to (0.2), πΎ π is defined for any function π (that is defined on a subset of RΓπΞ©ΓR3) and is self inverse, i.e., πΎ2 = id. Its restriction to πΏπ
lt
πΎπ+β’
, ππΎπΌ
yields an operator πΎ: πΏπ
lt
πΎπ+β’, ππΎπΌ
β πΏπ
lt
πΎπββ’, ππΎπΌ
for any 1 β€ π β€ β. Its inverse is πΎ:πΏπ
lt
πΎπββ’, ππΎπΌ
βπΏπ
lt
πΎπ+β’, ππΎπΌ
.
Lemma 1.2.14. For anyπ >0,1β€π <β, and anyπ, π β πΏβ πΎπβ
,πβ πΏπ πΎπ+, ππΎπΌ , and β βπΏπ πΎβπ, ππΎπΌ
we have
β«
πΎπβ
π|ππΎπ+β|πππΎπΌ=
β«
πΎπ+
πΎπ|ππΎπ+πΎ β|πππΎπΌ.
If additionallyπ0B kπkπΏβ(πΎπβ)<1andπ, π, π, βare nonnegative, the estimate
β«
πΎπβ
π|ππΎπ+β|πππΎπΌ β€ π0
β«
πΎ+π
(πΎπ)ππππΎπΌ+ (1βπ0)1βπ
β«
πΎπβ
π βπππΎπΌ
holds.
Proof. We compute
β«
πΎπβ
π|ππΎπ+β|πππΎπΌ
=
β
πΎπβ
π(π‘, π₯, π£)|π(π‘, π₯, π£)π(π‘, π₯, π£β2(π£Β·π(π₯))π(π₯)) +β(π‘, π₯, π£)|π|
bπ£πΌΒ·π(π₯)|ππ£πππ₯ππ‘
=
β
πΎπ+
π(π‘, π₯, π£β2(π£Β·π(π₯))π(π₯))|π(π‘, π₯, π£β2(π£Β·π(π₯))π(π₯))π(π‘, π₯, π£) +β(π‘, π₯, π£β2(π£Β·π(π₯))π(π₯))|π|β
bπ£πΌΒ·π(π₯)|ππ£πππ₯ππ‘
=
β«
πΎπ+
πΎπ|ππΎπ+πΎ β|πππΎπΌ.
using the change of variablesπ£ β¦βπ£β2(π£Β·π(π₯))π(π₯). Note that the determinant of the corresponding Jacobian equalsβ1 since the map is a reflection. As for the second statement, we estimate
β«
πΎπβ
π|ππΎπ+β|πππΎπΌ=
β«
πΎ+π
πΎπ|ππΎπ+πΎ β|πππΎπΌ
β€
β«
πΎ+π
πΎπ
π0π+ (1βπ0)(1βπ0)
β1πΎ β
π
ππΎπΌ
β€π0
β«
πΎπ+
(πΎπ)ππππΎπΌ+ (1βπ0)
β«
πΎπ+
πΎπ
(1βπ0)β1πΎ β
π
ππΎπΌ
=π0
β«
πΎπ+
(πΎπ)ππππΎπΌ+ (1βπ0)1βπ
β«
πΎπβ
π βπππΎπΌ
using the convexity of theπ-th power and the first statement.
1.2 The Vlasov part 27 Then there is a unique, nonnegative strong solution π β πΏβ
lt πΌπβ’; πΏ1β©πΏβ
in the sense that the conditions of Definition 1.2.12 are satisfied for all1β€π <β. Moreover, we have the estimates
(1βπ0)1π nonnegativeπΆ1-functionπ=π(π£)onR3that only depends on|π£|, is monotonically increasing in |π£|(i.e.,π(π£) =πΛ(|π£|)for some monotonically increasingπΛ β πΆ1(Rβ₯0)withπΛ0(0)=0), linear and local in time. We have
kπ¦πkπΏπ(πΎπβ,ππΎπΌ)=kππΎπkπΏπ(πΎπ+,ππΎπΌ) β€π0kπkπΏπ(πΎπ+,ππΎπΌ)
for allπβ πΏπ
lt
πΎπ+β’, ππΎπΌ
and 0<π βπΌπ
β’by Lemma 1.2.14. Thus, by Proposition 1.2.13 there is a unique solution for this π in the sense of Definition 1.2.12. Since π was arbitrary, it follows that, for all 1β€π <β, π βπΏπ
lt(Ξ£πβ’)and πΒ± βπΏπ
lt
πΎπΒ±β’, ππΎπΌ
, and all conditions of Definition 1.2.12 are satisfied.
Let 0 < π β πΌπ
β’ and recall that in the proof of Proposition 1.2.13 the solution on [0, π]was given by a solution of (1.2.9) withππ,π0,πβπ. Thus, πβ = πππ++ππβ =0 on π·βπ\πΎπβ. Applying Proposition 1.2.10.(i) (withβ =0), dropping negligible terms in (1.2.8), and using Lemma 1.2.14 we arrive at
β«
altogether and therefore πβ β πΏ1 that it is nonnegative and only depends on|π£|sinceπ½is nonnegative and only depends on|π£|. Proceeding similarly as before, we define πΛ(π‘, π₯, π£)Bπ½π(π£)π(π‘, π₯, π£) β₯0,
1.2 The Vlasov part 29 Taking the limitπβ0 does not cause any problem because we have
π½πβπ½ finite. Hence, (1.2.14) holds withπremoved. Next we insert the definition ofπ½and drop the terms where|π£| β₯π on the left-hand side to get
sinceπis monotonically increasing in|π£|. Note that it is important thatβπ½vanishes for |π£| > π . This proves (1.2.11) for 0 < π < β. Because of βπ β πΏπ R3;R3
is a bounded function.
As for (1.2.13), let 0<π β€ βand first derive the following key estimate:
β«
where we setπB β«
π΅π π£0πΌπ(π, π₯, π£)ππ£14
β [0,β[and used (1.2.10); ifπ=0, the second step makes no sense but clearly both the left-hand side and the right-hand side of (1.2.15) are zero in this case. Note that the integral on the left-hand side exists for almost all(π, π₯) βπΌπ
β’ ΓΞ©by Fubiniβs theorem and that the first estimate above holds trivially ifπ >π by π β₯0 and is an equality ifπβ€ π . Taking both sides of (1.2.15) to the power 43 and then integrating overΞ©yields (1.2.13).
Remark 1.2.16. TheπΏβ-spaces onπΎπΒ± with respect toππΎπΌ and the standard surface measure are the same and the respective norms coincide since null sets with respect to ππΎπΌ are null sets with respect to the standard surface measure and vice versa by bπ£πΌΒ·π(π₯)>0 (<0) onπΎπ+(πΎπβ). Consequently, from now on we will (mostly) not point out the measure in the denotation of suchπΏβ-spaces and simply writeπΏβ πΎπΒ±
.