Unique continuation principle and its absence on continuum space, on - - PowerPoint PPT Presentation

unique continuation principle and its absence on
SMART_READER_LITE
LIVE PREVIEW

Unique continuation principle and its absence on continuum space, on - - PowerPoint PPT Presentation

1 Unique continuation principle and its absence on continuum space, on lattices, and on quantum graphs Ivan Veseli c (TU Dortmund) on joint works joint with Daniel Lenz, Ivica Naki c, Norbert Peyerimhoff, Olaf Post, Constanza


slide-1
SLIDE 1

1

Unique continuation principle and its absence on continuum space, on lattices, and on quantum graphs

Ivan Veseli´ c (TU Dortmund)

  • n joint works joint with Daniel Lenz, Ivica Naki´

c, Norbert Peyerimhoff, Olaf Post, Constanza Rojas-Molina, Matthias T¨ aufer, Martin Tautenhahn OTIND conference, Wien 2016

slide-2
SLIDE 2

2

Unique continuation principle

Everyday encounter with unique continuation problem

slide-3
SLIDE 3

3

Unique continuation principle

Everyday encounter with unique continuation problem

slide-4
SLIDE 4

4

Unique continuation principle

Everyday encounter with unique continuation problem Possibly with fatal consequences

slide-5
SLIDE 5

5

Introduce properties of function classes

Let Ω ⊂ Rd be open, F ⊂ {f : Ω → C measurable}. The class F has the (weak) unique continuation property, if for all f ∈ F ∃ W ⊂ Ω non empty and open, such that f ≡ 0 on W ⇒ f ≡ 0. E.g. holomorphic functions f : C → C have to this property due to the local uniqueness theorem (analytic continuation). strong unique continuation property, if for all f ∈ F ∃ x0 ∈ Ω ∀ N > 0 lim

ǫ→0 ǫ−N

  • B(x0,ǫ)

|f |dx = 0 ⇒ f ≡ 0.

slide-6
SLIDE 6

6

Introduce properties of function classes

Let Ω ⊂ Rd be open, F ⊂ {f : Ω → C measurable}. The class F has the (weak) unique continuation property, if for all f ∈ F ∃ W ⊂ Ω non empty and open, such that f ≡ 0 on W ⇒ f ≡ 0. E.g. holomorphic functions f : C → C have to this property due to the local uniqueness theorem (analytic continuation). strong unique continuation property, if for all f ∈ F ∃ x0 ∈ Ω ∀ N > 0 lim

ǫ→0 ǫ−N

  • B(x0,ǫ)

|f |dx = 0 ⇒ f ≡ 0. vanishing order (at most) M > 0, if for all f ∈ F, f = 0 for each x0 ∈ Ω0 there exist a constant C > 0 and a radius ǫ0 > 0 such that CǫM ≤

  • B(x0,ǫ)

|f |dx for all ǫ ∈ (0, ǫ0). These properties are local, i.e. only information in an arbitrarily small ball around x0 are required.

slide-7
SLIDE 7

7

(Non)uniform vanishing order For k ∈ N let fk : C → C, z → zk : fk holomorphic ⇒ strong unique continuation property For large k, however, fk vanishes arbitrarily fast at z0 = 0. Similarly for harmonic functions gk = ℜfk.

slide-8
SLIDE 8

8

(Non)uniform vanishing order For k ∈ N let fk : C → C, z → zk : fk holomorphic ⇒ strong unique continuation property For large k, however, fk vanishes arbitrarily fast at z0 = 0. Similarly for harmonic functions gk = ℜfk. Thus, for F = {f : C → C : f holomorphic} there is no M > 0 such that F has vanishing order (at most) M. Growth vs. vanishing Let f : C → C be holomorphic growing slower at ∞ than vanishing at 0, i.e. lim inf

ǫց0 m(ǫ) · m(1/ǫ) = 0 where m(r) := max |z|=r|f (z)|.

Then f ≡ 0. Interplay of local and global properties!

slide-9
SLIDE 9

9

Hadamard’s three circle Theorem Let r1 < r2 < r3. Let f be a holomorphic function in a neighbourhood of the annulus r1 ≤ |z| ≤ r3 and denote m(r) := max|z|=r|f (z)|. Then log r3 r1

  • log m(r2) ≤ log

r3 r2

  • log m(r1) + log

r2 r1

  • log m(r3).

Thus log r → log m(r) is a convex function. R iR

slide-10
SLIDE 10

10

Hadamard’s three circle Theorem Let r1 < r2 < r3. Let f be a holomorphic function in a neighbourhood of the annulus r1 ≤ |z| ≤ r3 and denote m(r) := max|z|=r|f (z)|. Then log r3 r1

  • log m(r2) ≤ log

r3 r2

  • log m(r1) + log

r2 r1

  • log m(r3).

Thus log r → log m(r) is a convex function. R iR

slide-11
SLIDE 11

11

Hadamard’s three circle Theorem Let r1 < r2 < r3. Let f be a holomorphic function in a neighbourhood of the annulus r1 ≤ |z| ≤ r3 and denote m(r) := max|z|=r|f (z)|. Then log r3 r1

  • log m(r2) ≤ log

r3 r2

  • log m(r1) + log

r2 r1

  • log m(r3).

Thus log r → log m(r) is a convex function. R iR

slide-12
SLIDE 12

12

Main Result: Scale free quantitative unique continuation estimate

for spectral projectors of a Schr¨

  • dinger operator on a cube.

Setting: geometry infinite discrete set Z = (xj)j∈Zd ⊂ Rd,

slide-13
SLIDE 13

13

Main Result: Scale free quantitative unique continuation estimate

for spectral projectors of a Schr¨

  • dinger operator on a cube.

Setting: geometry infinite discrete set Z = (xj)j∈Zd ⊂ Rd, radius δ ∈ (0, 1/2) such that B(xj, δ) ⊂ Λ1 + j = [−1/2, 1/2]d + j δ-neighbourhood of Z: S = S(δ) = B(Z, δ) =

j∈Zd B(xj, δ)

slide-14
SLIDE 14

14

Main Result: Scale free quantitative unique continuation estimate

for spectral projectors of a Schr¨

  • dinger operator on a cube.

Setting: geometry infinite discrete set Z = (xj)j∈Zd ⊂ Rd, radius δ ∈ (0, 1/2) such that B(xj, δ) ⊂ Λ1 + j = [−1/2, 1/2]d + j δ-neighbourhood of Z: S = S(δ) = B(Z, δ) =

j∈Zd B(xj, δ)

For L ∈ N, set ΛL = [−L/2, L/2]d, SL(δ) = ΛL ∩ S(δ), and characteristic function WL = χSL(δ).

slide-15
SLIDE 15

15

Main Result: Scale free quantitative unique continuation estimate

Setting: Schr¨

  • dinger operator

For bounded, measurable potential V : Rd → R define s.a. Schr¨

  • dinger operator

HL = −∆ + V on ΛL, L ∈ N, with Dirichlet, Neumann, or periodic b.c. and spectral projection χ(−∞,E](HL), for energy E > 0. Theorem [Naki´ c, T¨ aufer, Tautenhahn, & Veseli´ c 15, 16] Let δ, Z, L, SL(δ), V , HL, E as above. Then exists K depending only on dimension d such that χ(−∞,E](HL) WL χ(−∞,E](HL) ≥ δ

K

  • 1+

√ E+V 2/3

  • χ(−∞,E](HL)

where WL = χSL(δ). Bound is equivalent to

  • SL(δ)

|ψ|2 ≥ δ

K

  • 1+

√ E+V 2/3

ΛL

|ψ|2 for any ψ =

  • Ek ≤E

αkψk ∈ ran χ(−∞,E](HL)

slide-16
SLIDE 16

16

Concepts and methods for proof of main result.

slide-17
SLIDE 17

17

Carleman estimates

allow to deduce ucp (after some calculations) whole zoo of Carleman estimates exists many with abstract weight functions (satisfying H¨

  • rmanders subellipticity condition)

we want explicit estimate, thus explicit weight function We start with a formulation of [Bourgain, Kenig 05] since this has given crucial stimulus to the theory of random Schr¨

  • dinger operators.
slide-18
SLIDE 18

18

Carleman estimate as formulated in [Bourgain, Kenig 05]

Weight function φ: [0, ∞) → [0, ∞) φ(r) = r exp

r 1 − e−t t dt

  • w : Rd → [0, ∞), w(x) = φ(|x|)

r 1/φ(r) ⇒ ∀ r ∈ (0, 1) : r/3 ≤ φ(r) ≤ r Theorem [Bourgain & Kenig 05] There are constants C1(d) and C2(d) ∈ [1, ∞) s. t. for all α ≥ C1 and real valued f ∈ C 2(B(0, 1)) with compact support in B(0, 1) \ {0} we have α3

  • w−1−2α f 2 dx ≤ C2
  • w2−2α (∆f )2 dx
slide-19
SLIDE 19

19

Carleman estimate as formulated in [Bourgain, Kenig 05]

Weight function φ: [0, ∞) → [0, ∞) φ(r) = r exp

r 1 − e−t t dt

  • w : Rd → [0, ∞), w(x) = φ(|x|)

r 1/φ(r) ⇒ ∀ r ∈ (0, 1) : r/3 ≤ φ(r) ≤ r Theorem [Bourgain & Kenig 05] There are constants C1(d) and C2(d) ∈ [1, ∞) s. t. for all α ≥ C1 and real valued f ∈ C 2(B(0, 1)) with compact support in B(0, 1) \ {0} we have α3

  • w−1−2α f 2 dx ≤ C2
  • w2−2α (∆f )2 dx

Possible to scale the inequality to a ball of radius ρ and extend to Sobolev space H2.

slide-20
SLIDE 20

20

Consequences/Applications of Carleman estimates: Three annuli inequality

A1

A2 A3 A1 A1 A2 A2 A3 A3 Insert χ × ψ = cut-off × eigenfunction in Carleman inequality (e.g. [Bourgain, Kenig 05]) to get three annuli inequality α3

  • A2

w2−2α|ψ|2

  • A1

w2−2α|ψ|2 +

  • A3

w2−2α|ψ|2. profile of cut-off function χ

slide-21
SLIDE 21

21

Compare to Hadamard’s three circle theorem

Carleman inequality (e.g. [Bourgain, Kenig 05]) implies three annuli inequality α3

  • A2

w2−2α|ψ|2

  • A1

w2−2α|ψ|2 +

  • A3

w2−2α|ψ|2. annuli instead of circles (no worries about regularity and trace map) middle annulus controlled by inner annulus and outer annulus integral over outer annulus plays (typically) the role of a global bound/ bound ’at infinity’/ a priori bound

slide-22
SLIDE 22

22

Unique continuation on combinatorial and quantum graphs?

slide-23
SLIDE 23

23

Half space with unique continuation for discrete Schr¨

  • dinger operators
slide-24
SLIDE 24

24

Half space with unique continuation for discrete Schr¨

  • dinger operators

ψ known e.g. ψ = 0 Hψ = Eψ ψ unknown

slide-25
SLIDE 25

25

Half space with unique continuation for discrete Schr¨

  • dinger operators

ψ known e.g. ψ = 0 Hψ = Eψ ψ unknown

inequality with one unknown; solvable

slide-26
SLIDE 26

26

Half space with unique continuation for discrete Schr¨

  • dinger operators

ψ known e.g. ψ = 0 Hψ = Eψ ψ unknown

inequality with one unknown; solvable

slide-27
SLIDE 27

27

Half space with unique continuation for discrete Schr¨

  • dinger operators

ψ known e.g. ψ = 0 Hψ = Eψ ψ known also here

slide-28
SLIDE 28

28

Half space without unique continuation for discrete Schr¨

  • dinger operators

ψ known Hψ = Eψ ψ unknown

slide-29
SLIDE 29

29

Half space without unique continuation for discrete Schr¨

  • dinger operators

ψ known Hψ = Eψ ψ unknown

inequality with two unknowns; unsolvable

slide-30
SLIDE 30

30

Half space without unique continuation for discrete Schr¨

  • dinger operators

ψ known Hψ = Eψ ψ unknown

? ? ?

No unique continuation!

slide-31
SLIDE 31

31

Outside-in unique continuation holds

ψ known Hψ = Eψ ψ unknown

slide-32
SLIDE 32

32

Outside-in unique continuation holds

ψ known Hψ = Eψ

slide-33
SLIDE 33

33

Inside-out unique continuation does not hold

ψ unknown Hψ = Eψ ψ known

slide-34
SLIDE 34

34

Inside-out unique continuation does not hold

ψ unknown Hψ = Eψ ψ known

? ? ? ?

slide-35
SLIDE 35

35

Kagome lattice: infinite planar 4-regular planar graph X ⊂ C

w1 w2

1

Each vertex ∈ unique upside triangle.

2

Identify the lower left vertex of upside triangle with the origin in C its other two vertices with w1 = 1 and w2 = eπi/3 ∈ C.

3

Vertex set of X V (X) = (2Zw1 + 2Zw2) · ∪ (w1 + 2Zw1 + 2Zw2) · ∪ (w2 + 2Zw1 + 2Zw2). is Z2-periodic.

4

The the planar graph X ⊂ C defines a metric graph.

5

Graph tiles plan with triangles and hexagons.

slide-36
SLIDE 36

36

Some results of Lenz, Peyerimhoff, Post, & Ves. 09

  • n spectrum and IDS of the combinatorial Laplacian on Kagome lattice

Eigenfunction supported in one hexagon For an arbitrary hexagon H ⊂ X with vertices {u0, u1, . . . , u5} there exists a centre w0 ∈ C of H such that {u0, u1, . . . , u5} = { w0 + ekπi/3 | k = 0, 1, . . . , 5 }. The function FH : V → {−1, 0, 1} FH(v) :=

  • 0,

if v ∈ V \ {u0, . . . , u5}, (−1)k, if v = w0 + ekπi/3, (1) satisfies ∆combFH(v) = 1 deg(v)

  • w∼v

(FH(v) − FH(w)) = 3 2 FH(v).

slide-37
SLIDE 37

37

LPPV on spectrum of combinatorial Kagome Laplacian

ℓ2 and finitely supported eigenfunctions

1

If f ∈ ℓ2(V ) satisfies ∆combf = µf , then µ = 3/2.

2

If F ∈ ℓ2(V ) satisfies ∆combF = µF and has finite support then µ = 3/2 and F is a linear combination of finitely many eigenfunctions FH.

3

Let Hi (i = 1, . . . , k) be a collection of distinct hexagons. Then the set FH1, . . . , FHk is linearly independent.

4

The space of ℓ2(V )-eigenfunctions to the eigenvalue 3/2 is spanned by compactly supported eigenfunctions. Whole spectrum Denote by σac(∆comb) and σpp(∆comb) the absolutely continuous and point spectrum of ∆comb on the Z2-periodic Kagome graph. Then σac(∆comb) =

  • 0, 3

2

  • and

σpp(∆comb) = 3 2

  • .
slide-38
SLIDE 38

38

LPPV on IDS of combinatorial Kagome Laplacian

Integrated density of states/Spectral distribution function Let Ncomb be the IDS of the Z2-periodic operator ∆comb, given by Ncomb(µ) = 1 3 Tr[1QPcomb((−∞, µ])], where Tr is the trace and Pcomb denotes the spectral projection. Then

1

Ncomb vanishes on (−∞, 0],

2

is continuous on R \ {3/2},

3

has a jump of size 1/3 at µ = 3/2,

4

Ncomb is strictly monotone increasing on [0, 3/2], and

5

Ncomb(µ) = 1 for µ ≥ 3/2.

slide-39
SLIDE 39

39

Spectrum and IDS of the periodic Kirchhoff Laplacian

Quantum graph with Kirchhoff (or free) vertex conditions: function is continuous at the vertex and the sum of all derivatives add up to zero. Spectrum of Kirchhoff Laplacian ∆0 on equilateral Kagome metric graph Let σpp and σac denote the point spectrum and absolutely continuous spectrum and σcomp denote the spectrum given by the compactly supported eigenfunctions. Then σcomp(−∆0) = σpp(−∆0) = 2k + 2 3 2 π2

  • k ∈ Z
  • k2π2

k ∈ N

  • and

σac(−∆0) =

  • 0,

2 3 2 π2 ∪

  • k∈N
  • 2k − 2

3 2 π2,

  • 2k + 2

3 2 π2 .

slide-40
SLIDE 40

40

LPPV on IDS of Kagome quantum graph

IDS N0 of the negative Kirchhoff Laplacian −∆0 on the metric Kagome graph N0(λ) = 1 vol(F) Tr[1FP0((−∞, λ])], where Tr is the trace, F the fundamental domain, and P0 denotes the spectral projection.

1

All discontinuities of N0 : R → [0, ∞) are

1

at λ = (2k + 2

3 )2π2, k ∈ Z, with jumps of size 1 6 , 2

at λ = k2π2, k ∈ N, with jumps of size 1

2 . 2

N0 is strictly monotone increasing on the absolutely continuous spectrum σac(−∆0) and

3

N0 is constant on the complement of σ(−∆0).

slide-41
SLIDE 41

41

General correspondence for combinatorial and quantum graphs eigenvalues of compactly supported eigenfucntions = jumps of IDS See e.g.

1

Klassert, Lenz, & Stollmann 03,

2

  • Ves. 05,

3

Gruber, Lenz & Ves. 07,

4

Lenz & Ves. 09,

5

Pogorzelski, Schwarzenberger, & Seifert 13

slide-42
SLIDE 42

42

Advertisement with kind permission of the organisers: 5th Najman conference on Spectral Theory and Differential Equations 10th to 15th September 2017 Opatija, Croatia Previous conferences in the series are describe at: http://web.math.pmf.unizg.hr/najman conference (or simply google ”Najman conference”) There will appear detailed information on the 5th Najman conference shortly. Organisers: Jussi Behrndt, Luka Grubiˇ si´ c, Ivica Naki´ c, Ivan Veseli´ c

Thank you for your attention!