Limit Operators Getting your hands on the essentials. Marko Lindner - - PowerPoint PPT Presentation

limit operators
SMART_READER_LITE
LIVE PREVIEW

Limit Operators Getting your hands on the essentials. Marko Lindner - - PowerPoint PPT Presentation

Limit Operators Getting your hands on the essentials. Marko Lindner 14-18 August 2017 TUHH Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 1 / 44 ...with the help of... This talk is based on joint work


slide-1
SLIDE 1

Limit Operators

Getting your hands on the essentials.

Marko Lindner 14-18 August 2017 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 1 / 44

slide-2
SLIDE 2

...with the help of...

This talk is based on joint work with Markus Seidel, Zwickau Raffael Hagger, Hannover Hagen S¨

  • ding, TU Hamburg

Simon Chandler-Wilde, Reading Bernd Silbermann, Chemnitz TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 2 / 44

slide-3
SLIDE 3

Table of Contents

1

The essentials

2

Limit operators

3

Stability of approximation methods

4

The Fibonacci Hamiltonian TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 3 / 44

slide-4
SLIDE 4

Table of Contents

1

The essentials

2

Limit operators

3

Stability of approximation methods

4

The Fibonacci Hamiltonian TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 4 / 44

slide-5
SLIDE 5

The operators and their matrices

For a bounded linear operator A : X → X

  • n a Banach space X, choose a basis in X and represent A as an infinite matrix.

Sometimes it is convenient to number the basis elements over the integers Z (rather than the naturals N), leading to a bi-infinite matrix: A =         ... . . . . . . . . . ... · · · a-1,-1 a-1,0 a-1,1 · · · · · · a 0,-1 a 0, 0 a 0, 1 · · · · · · a 1,-1 a 1, 0 a 1, 1 · · · ... . . . . . . . . . ...         We will mostly think of banded matrices A with uniformly bounded entries: sup |aij| < ∞, so A acts as a bounded linear operator on ℓp(Z), p ∈ [1, ∞]. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 5 / 44

slide-6
SLIDE 6

Compact operators & the Calkin algebra

For a Banach space X, put L(X) = the set (Banach algebra) of all bounded linear operators X → X, K(X) = the set of all compact operators X → X (closed ideal in L(X)). Then one can form the factor algebra The Calkin algebra L(X)/K(X) = {A + K(X) : A ∈ L(X)}. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 6 / 44

slide-7
SLIDE 7

Compact operators & the Calkin algebra

For a Banach space X, put L(X) = the set (Banach algebra) of all bounded linear operators X → X, K(X) = the set of all compact operators X → X (closed ideal in L(X)). Then one can form the factor algebra The Calkin algebra L(X)/K(X) = {A + K(X) : A ∈ L(X)}. More specifically, for X = ℓp(Z), let BO(X) = the set (algebra) of all operators X → X with a band matrix, BDO(X) = the norm closure (Banach algebra) of BO(X). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 6 / 44

slide-8
SLIDE 8

Operator classes graphically

⊂ = BO(X)

band operators

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 7 / 44

slide-9
SLIDE 9

Operator classes graphically

⊂ = BO(X)

band operators

K(X) =

compact operators

⊂ = BDO(X)

band-dominated

  • perators

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 7 / 44

slide-10
SLIDE 10

Operator classes graphically

⊂ = BO(X)

band operators

K(X) =

compact operators

⊂ = BDO(X)

band-dominated

  • perators

BDO(X)/K(X) = / = TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 7 / 44

slide-11
SLIDE 11

The essentials

Let A ∈ L(X). Definition: essential norm Aess := A + K(X) = inf{A + K : K ∈ K(X)} = dist (A, K(X)) TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 8 / 44

slide-12
SLIDE 12

The essentials

Let A ∈ L(X). Definition: essential norm Aess := A + K(X) = inf{A + K : K ∈ K(X)} = dist (A, K(X)) Definition: Fredholmness A is Fredholm (“essentially invertible”) iff A + K(X) is invertible in L(X)/K(X). A is Fredholm iff its kernel and cokernel have finite dimension. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 8 / 44

slide-13
SLIDE 13

The essentials

Let A ∈ L(X). Definition: essential norm Aess := A + K(X) = inf{A + K : K ∈ K(X)} = dist (A, K(X)) Definition: Fredholmness A is Fredholm (“essentially invertible”) iff A + K(X) is invertible in L(X)/K(X). A is Fredholm iff its kernel and cokernel have finite dimension. Definition: Essential spectrum specessA := specL(X)/K(X)(A + K(X)) = {λ ∈ C : A − λI is not Fredholm} TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 8 / 44

slide-14
SLIDE 14

Essential spectrum and norm: diagonal examples

For A =               ... 3 5 3 5 3 3 ...               , we clearly have spec A = {3, 5}, A = 5 but specessA = {3}, Aess = 3. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 9 / 44

slide-15
SLIDE 15

Essential spectrum and norm: diagonal examples

For A =               ... 3 5 3 5 3 3 ...               , we clearly have spec A = {3, 5}, A = 5 but specessA = {3}, Aess = 3. The spectral value 5 is not essential (“not visible at ∞”). A − 5I is not invertible but still Fredholm (kernel and cokernel have finite dimension). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 9 / 44

slide-16
SLIDE 16

Essential spectrum and norm: diagonal examples

For A =           ... 3 + ε−1 3 + ε0 3 + ε1 3 + ε2 ...           , with positive εn such that εn → 0 as n → ±∞, we have spec A = {3 + εn : n ∈ Z} ∪ {3}, A = max{3 + εn : n ∈ Z}. The spectral value 3 is no eigenvalue but still in the spectrum. A − 3I is injective but has no bounded inverse (not Fredholm, range not closed). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 10 / 44

slide-17
SLIDE 17

Essential spectrum and norm: diagonal examples

For A =           ... 3 + ε−1 3 + ε0 3 + ε1 3 + ε2 ...           , with positive εn such that εn → 0 as n → ±∞, we have spec A = {3 + εn : n ∈ Z} ∪ {3}, A = max{3 + εn : n ∈ Z}. The spectral value 3 is no eigenvalue but still in the spectrum. A − 3I is injective but has no bounded inverse (not Fredholm, range not closed). It holds specessA = {3}, Aess = 3. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 10 / 44

slide-18
SLIDE 18

Essential spectrum and norm: diagonal examples

For A =               ... 3 + ε−1 5 + ε−1 3 + ε0 5 + ε0 3 + ε1 5 + ε1 ...               , with positive εn such that εn → 0 as n → ±∞, we have spec A = {3 + εn, 5 + εn : n ∈ Z} ∪ {3, 5}, A = max{3 + εn, 5 + εn : n ∈ Z}. The spectral values 3 and 5 are no eigenvalues of A. A − 3I and A − 5I are injective but have no bounded inverse (range not closed). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 11 / 44

slide-19
SLIDE 19

Essential spectrum and norm: diagonal examples

For A =               ... 3 + ε−1 5 + ε−1 3 + ε0 5 + ε0 3 + ε1 5 + ε1 ...               , with positive εn such that εn → 0 as n → ±∞, we have spec A = {3 + εn, 5 + εn : n ∈ Z} ∪ {3, 5}, A = max{3 + εn, 5 + εn : n ∈ Z}. The spectral values 3 and 5 are no eigenvalues of A. A − 3I and A − 5I are injective but have no bounded inverse (range not closed). It holds specessA = {3, 5}, Aess = 5. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 11 / 44

slide-20
SLIDE 20

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z. In other words: λ ∈ specessA ⇐ ⇒ ∃n1, n2, · · · → ±∞ : ank ,nk → λ. Moreover, Aess = the largest (in modulus) partial limit = lim sup |an,n|. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 12 / 44

slide-21
SLIDE 21

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z. In other words: λ ∈ specessA ⇐ ⇒ ∃n1, n2, · · · → ±∞ : ank ,nk → λ. Moreover, Aess = the largest (in modulus) partial limit = lim sup |an,n|. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 13 / 44

slide-22
SLIDE 22

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z. In other words: λ ∈ specessA ⇐ ⇒ ∃n1, n2, · · · → ±∞ : ank ,nk → λ. Moreover, Aess = the largest (in modulus) partial limit = lim sup |an,n|. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 14 / 44

slide-23
SLIDE 23

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z Aess = the largest (in modulus) partial limit = lim sup |an,n|. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 15 / 44

slide-24
SLIDE 24

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z Aess = the largest (in modulus) partial limit = lim sup |an,n|. The whole coset A + K(X) ∈ L(X)/K(X) is encoded in the partial limits of (an,n)n∈Z. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 15 / 44

slide-25
SLIDE 25

Essential spectrum and norm: a general diagonal matrix

For a general (bounded) diagonal matrix A =           ... a−1,−1 a0,0 a1,1 a2,2 ...           , it holds that specessA = the set of all partial limits of the sequence (an,n)n∈Z Aess = the largest (in modulus) partial limit = lim sup |an,n|. The whole coset A + K(X) ∈ L(X)/K(X) is encoded in the partial limits of (an,n)n∈Z. Restricting consideration to diagonal matrices, the Calkin algebra is Ldiag(X)/Kdiag(X) ∼ = ℓ∞/c0. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 15 / 44

slide-26
SLIDE 26

Table of Contents

1

The essentials

2

Limit operators

3

Stability of approximation methods

4

The Fibonacci Hamiltonian TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 16 / 44

slide-27
SLIDE 27

From diagonal to band-dominated matrices

For A ∈ BDO(X), the coset A + K(X) is still determined by the asymptotics of A = (ai,j) at infinity. Again, take a sequence n1, n2, · · · → ±∞ and follow the entries ank ,nk as k → ∞. (1) New: Now also the context of the entries (1) is important. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 17 / 44

slide-28
SLIDE 28

From diagonal to band-dominated matrices

For A ∈ BDO(X), the coset A + K(X) is still determined by the asymptotics of A = (ai,j) at infinity. Again, take a sequence n1, n2, · · · → ±∞ and follow the entries ank ,nk as k → ∞. (1) New: Now also the context of the entries (1) is important. Not only the sequence (1) itself shall converge but also its neighbour entries: ank +i,nk +j →: bi,j ∀i, j ∈ Z. The existence of such sequences h = (nk) is gua- ranteed by the Bolzano-Weierstrass theorem. Definition: limit operator The operator with matrix B = (bi,j)i,j∈Z is called limit operator of A w.r.t. the sequence h. We write Ah for B and σop(A) for the set of all Ah. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 17 / 44

slide-29
SLIDE 29

Limit operators: Time for examples

A periodic matrix: σop

  • =
  • all shifts of
  • TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 18 / 44

slide-30
SLIDE 30

Limit operators: Time for examples

A periodic matrix: σop

  • =
  • all shifts of
  • Simple but non-periodic:

σop

  • =
  • all shifts of

,

  • TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 18 / 44

slide-31
SLIDE 31

Limit operators: Time for examples

Discrete Schr¨

  • dinger operator in 1D

(Ax)n = xn−1 + v(n)xn + xn+1, n ∈ Z with a bounded potential v ∈ ℓ∞(Z). The matrix looks like this A =               ... ... ... v−2 1 1 v−1 1 1 v0 1 1 v1 1 1 v2 ... ... ...               TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 19 / 44

slide-32
SLIDE 32

Limit operators: Time for examples

Discrete Schr¨

  • dinger operator in 1D

(Ax)n = xn−1 + v(n)xn + xn+1, n ∈ Z with a bounded potential v ∈ ℓ∞(Z). The matrix looks like this A =               ... ... ... v−2 1 1 v−1 1 1 v0 1 1 v1 1 1 v2 ... ... ...               Limit op’s of A: (Bx)n = xn−1 + w(n)xn + xn+1, n ∈ Z with a potential w “locally representing v at infinity”. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 19 / 44

slide-33
SLIDE 33

Example: Discrete Schr¨

  • dinger operator

So it is enough to look at the potential v: Example 1: locally constant potential v = (· · · , β, β, β, β

  • 4

, α, α, α

3

, β, β

  • 2

, α

  • 1

, β, β

  • 2

, α, α, α

3

, β, β, β, β

  • 4

, · · · ), α = β TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 20 / 44

slide-34
SLIDE 34

Example: Discrete Schr¨

  • dinger operator

So it is enough to look at the potential v: Example 1: locally constant potential v = (· · · , β, β, β, β

  • 4

, α, α, α

3

, β, β

  • 2

, α

  • 1

, β, β

  • 2

, α, α, α

3

, β, β, β, β

  • 4

, · · · ), α = β ⇒ 4 limop’s: w = (· · · , α, α, α, α, α, α, · · · ) w = (· · · , β, β, β, β, β, β, · · · ) w = (· · · , α, α, α, β, β, β, · · · ) w = (· · · , β, β, β, α, α, α, · · · ) ...and shifts of the latter two. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 20 / 44

slide-35
SLIDE 35

Example: Discrete Schr¨

  • dinger operator

So it is enough to look at the potential v: Example 1: locally constant potential v = (· · · , β, β, β, β

  • 4

, α, α, α

3

, β, β

  • 2

, α

  • 1

, β, β

  • 2

, α, α, α

3

, β, β, β, β

  • 4

, · · · ), α = β ⇒ 4 limop’s: w = (· · · , α, α, α, α, α, α, · · · ) w = (· · · , β, β, β, β, β, β, · · · ) w = (· · · , α, α, α, β, β, β, · · · ) w = (· · · , β, β, β, α, α, α, · · · ) ...and shifts of the latter two. Example 2: slowly oscillating potential v(n + 1) − v(n) → 0, n → ∞ e.g. v(n) = cos

  • |n|.

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 20 / 44

slide-36
SLIDE 36

Example: Discrete Schr¨

  • dinger operator

So it is enough to look at the potential v: Example 1: locally constant potential v = (· · · , β, β, β, β

  • 4

, α, α, α

3

, β, β

  • 2

, α

  • 1

, β, β

  • 2

, α, α, α

3

, β, β, β, β

  • 4

, · · · ), α = β ⇒ 4 limop’s: w = (· · · , α, α, α, α, α, α, · · · ) w = (· · · , β, β, β, β, β, β, · · · ) w = (· · · , α, α, α, β, β, β, · · · ) w = (· · · , β, β, β, α, α, α, · · · ) ...and shifts of the latter two. Example 2: slowly oscillating potential v(n + 1) − v(n) → 0, n → ∞ e.g. v(n) = cos

  • |n|.

⇒ Limop’s (all constant): w(n) ≡ a, a ∈ v(∞) TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 20 / 44

slide-37
SLIDE 37

Example: Discrete Schr¨

  • dinger operator

Example 3: random (actually pseudo-ergodic) potential Take random (i.i.d.) samples v(n) from a compact set V ⊂ C.

a.s.

= ⇒ The infinite “word” (· · · , v(−1), v(0), v(1), · · · ) contains every finite word over V as a subword (up to arbitrary accuracy ε > 0). [pseudo-ergodic, Davies 2001] TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 21 / 44

slide-38
SLIDE 38

Example: Discrete Schr¨

  • dinger operator

Example 3: random (actually pseudo-ergodic) potential Take random (i.i.d.) samples v(n) from a compact set V ⊂ C.

a.s.

= ⇒ The infinite “word” (· · · , v(−1), v(0), v(1), · · · ) contains every finite word over V as a subword (up to arbitrary accuracy ε > 0). [pseudo-ergodic, Davies 2001] ⇒ lots of limop’s: all functions w : Z → V TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 21 / 44

slide-39
SLIDE 39

Example: Discrete Schr¨

  • dinger operator

Example 3: random (actually pseudo-ergodic) potential Take random (i.i.d.) samples v(n) from a compact set V ⊂ C.

a.s.

= ⇒ The infinite “word” (· · · , v(−1), v(0), v(1), · · · ) contains every finite word over V as a subword (up to arbitrary accuracy ε > 0). [pseudo-ergodic, Davies 2001] ⇒ lots of limop’s: all functions w : Z → V Example 4: (almost-)periodic potential v(n) = cos(nα), n ∈ Z Case 1: α = p

q 2π ∈ πQ (periodic)

⇒ q limop’s: wk(n) = cos((n + k)α), k = 1, ..., q TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 21 / 44

slide-40
SLIDE 40

Example: Discrete Schr¨

  • dinger operator

Example 3: random (actually pseudo-ergodic) potential Take random (i.i.d.) samples v(n) from a compact set V ⊂ C.

a.s.

= ⇒ The infinite “word” (· · · , v(−1), v(0), v(1), · · · ) contains every finite word over V as a subword (up to arbitrary accuracy ε > 0). [pseudo-ergodic, Davies 2001] ⇒ lots of limop’s: all functions w : Z → V Example 4: (almost-)periodic potential v(n) = cos(nα), n ∈ Z Case 1: α = p

q 2π ∈ πQ (periodic)

⇒ q limop’s: wk(n) = cos((n + k)α), k = 1, ..., q Case 2: α ∈ πQ (almost-periodic, see Almost-Mathieu operator) ⇒ ∞-many limop’s: wθ(n) = cos(nα + θ), θ ∈ [0, 2π) TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 21 / 44

slide-41
SLIDE 41

Limit operators: The definition revisited

For each n ∈ Z, define the n-shift on X = ℓp(Z) via Sn : x → y with xi = yi+n. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 22 / 44

slide-42
SLIDE 42

Limit operators: The definition revisited

For each n ∈ Z, define the n-shift on X = ℓp(Z) via Sn : x → y with xi = yi+n. Then, for h = (n1, n2, ...) with |nk| → ∞, one has (S−nk ASnk )i,j = Ai+nk ,j+nk , i, j ∈ Z, so that the limit operator Ah of A ∈ BDO(X) equals Ah = lim

k→∞ S−nk ASnk ,

the limit taken in the strong topology (pointwise convergence on X). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 22 / 44

slide-43
SLIDE 43

Limit operators: The definition revisited

For each n ∈ Z, define the n-shift on X = ℓp(Z) via Sn : x → y with xi = yi+n. Then, for h = (n1, n2, ...) with |nk| → ∞, one has (S−nk ASnk )i,j = Ai+nk ,j+nk , i, j ∈ Z, so that the limit operator Ah of A ∈ BDO(X) equals Ah = lim

k→∞ S−nk ASnk ,

the limit taken in the strong topology (pointwise convergence on X). In this sense, the set σop(A) of all limit operators of A is the set of all partial limits of the

  • perator sequence

(S−nASn)n∈Z. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 22 / 44

slide-44
SLIDE 44

Limit operators: The definition revisited

The very same can be done with X = ℓp(Zd) and A ∈ BDO(X). Again, the set σop(A) of all limit operators of A is the set of all partial limits of the

  • perator “sequence”

(S−nASn)n∈Zd

  • r, likewise, of the function

fA : n ∈ Zd → S−nASn ∈ BDO(X). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 23 / 44

slide-45
SLIDE 45

Limit operators: The definition revisited

The very same can be done with X = ℓp(Zd) and A ∈ BDO(X). Again, the set σop(A) of all limit operators of A is the set of all partial limits of the

  • perator “sequence”

(S−nASn)n∈Zd

  • r, likewise, of the function

fA : n ∈ Zd → S−nASn ∈ BDO(X). Take a suitable compactification of Zd. Extend the function fA continuously to it. Evaluate fA at the boundary ∂Zd ⇒ limit operators of A One can enumerate the limit operators of A by the elements of ∂Zd (rather than by sequences h = (n1, n2, . . . ) for which S−nk ASnk converges). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 23 / 44

slide-46
SLIDE 46

Limit operators: The definition revisited

The very same can be done with X = ℓp(Zd) and A ∈ BDO(X). Again, the set σop(A) of all limit operators of A is the set of all partial limits of the

  • perator “sequence”

(S−nASn)n∈Zd

  • r, likewise, of the function

fA : n ∈ Zd → S−nASn ∈ BDO(X). Take a suitable compactification of Zd. Extend the function fA continuously to it. Evaluate fA at the boundary ∂Zd ⇒ limit operators of A One can enumerate the limit operators of A by the elements of ∂Zd (rather than by sequences h = (n1, n2, . . . ) for which S−nk ASnk converges). These ideas can be extended from ℓp(Zd) [Rabinovich, Roch, Silbermann 1998] to ℓp(G) for finitely generated discrete groups G [Roe 2005] ℓp(X) for strongly discrete metric spaces X [Spakula & Willett 2014] Lp(X, µ) for fairly general metric spaces X and measures µ [Hagger & Seifert 2017+x] TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 23 / 44

slide-47
SLIDE 47

Limit operators: The definition revisited

So we have New enumeration (independent of A) of the limit operators of A σop(A) = {Ah : h = (n1, n2, . . . ) in Zd with |nk| → ∞ s.t. lim S−nk ASnk exists} = {Ag : g ∈ ∂Zd} Now one can add or multiply two instances of σop(A) elementwise and get σop(A + B) = σop(A) + σop(B), σop(AB) = σop(A)σop(B), σop(αA) = ασop(A). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 24 / 44

slide-48
SLIDE 48

Limit operators: The definition revisited

So we have New enumeration (independent of A) of the limit operators of A σop(A) = {Ah : h = (n1, n2, . . . ) in Zd with |nk| → ∞ s.t. lim S−nk ASnk exists} = {Ag : g ∈ ∂Zd} Now one can add or multiply two instances of σop(A) elementwise and get σop(A + B) = σop(A) + σop(B), σop(AB) = σop(A)σop(B), σop(αA) = ασop(A). In short: The map ϕ : A → σop(A), BDO(X) → ℓ∞(∂Zd, BDO(X)) is an algebra homomorphism. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 24 / 44

slide-49
SLIDE 49

Limit operators: The definition revisited

So we have New enumeration (independent of A) of the limit operators of A σop(A) = {Ah : h = (n1, n2, . . . ) in Zd with |nk| → ∞ s.t. lim S−nk ASnk exists} = {Ag : g ∈ ∂Zd} Now one can add or multiply two instances of σop(A) elementwise and get σop(A + B) = σop(A) + σop(B), σop(AB) = σop(A)σop(B), σop(αA) = ασop(A). In short: The map ϕ : A → σop(A), BDO(X) → ℓ∞(∂Zd, BDO(X)) is an algebra homomorphism. Key observation The kernel of that homomorphism ϕ : A → σop(A) is K(X). So A + K(X) → σop(A) is an isomorphism BDO(X)/ ker ϕ → im ϕ. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 24 / 44

slide-50
SLIDE 50

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-51
SLIDE 51

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah are invertible TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-52
SLIDE 52

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah are invertible B is a Φ-regulariser of A B + K(X) = [A + K(X)]−1 Bh = A−1

h

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-53
SLIDE 53

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah are invertible B is a Φ-regulariser of A B + K(X) = [A + K(X)]−1 Bh = A−1

h

essential spectrum

specessA specL(X)/K(X)(A + K(X)) ∪h spec Ah TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-54
SLIDE 54

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah are invertible B is a Φ-regulariser of A B + K(X) = [A + K(X)]−1 Bh = A−1

h

essential spectrum

specessA specL(X)/K(X)(A + K(X)) ∪h spec Ah

essential pseudospectrum

specε

essA

specε

L(X)/K(X)(A + K(X))

∪h specεAh TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-55
SLIDE 55

Limit operators and our essentials

The result is an identification A + K(X) ∼ = σop(A) for all A ∈ BDO(X) – with the following consequences: A A + K(X) σop(A)

essential norm

Aess A + K(X)L(X)/K(X) maxh Ah A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah are invertible B is a Φ-regulariser of A B + K(X) = [A + K(X)]−1 Bh = A−1

h

essential spectrum

specessA specL(X)/K(X)(A + K(X)) ∪h spec Ah

essential pseudospectrum

specε

essA

specε

L(X)/K(X)(A + K(X))

∪h specεAh

For A ∈ BO(X) and d = 1:

A is Fredholm A + K(X) invertible in L(X)/

K(X)

all Ah injective on ℓ∞(Z) specessA specL(X)/K(X)(A + K(X)) ∪h spec∞

pointAh

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 25 / 44

slide-56
SLIDE 56

Limit operators and our essentials

A + K(X) ∼ = σop(A) A σop(A)

essential norm

Aess maxh Ah

[Hagger, ML, Seidel 2016] [Lange, Rabinovich 1985+] [Rabinovich, Roch, Silbermann 1998+]

A is Fredholm all Ah are invertible

[ML, Silbermann 2003], [ML 2003+] [Chandler-Wilde, ML 2007] [Seidel, ML 2014]

B is a Φ-regulariser of A Bh = A−1

h [Seidel 2013]

essential spectrum

specessA ∪h spec Ah

[Seidel, ML 2014]

essential pseudospectrum

specε

essA

∪h specεAh

[Hagger, ML, Seidel 2016]

For A ∈ BO(X) and d = 1:

A is Fredholm all Ah injective on ℓ∞(Z)

[Chandler-Wilde, ML 2008]

specessA ∪h spec∞

pointAh

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 26 / 44

slide-57
SLIDE 57

Self-similar operators

Definition: self-similar operator We say that A ∈ BDO(X) is self-similar if A ∈ σop(A). Roughly speaking, this means that A contains a copy of itself, at infinity. Each pattern that you see once in A, you will see infinitely often in A. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 27 / 44

slide-58
SLIDE 58

Self-similar operators

Definition: self-similar operator We say that A ∈ BDO(X) is self-similar if A ∈ σop(A). Roughly speaking, this means that A contains a copy of itself, at infinity. Each pattern that you see once in A, you will see infinitely often in A. But then, by the above, Aess = A A is Fredholm ⇐ ⇒ A is invertible specessA = spec A specε

essA

= specεA TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 27 / 44

slide-59
SLIDE 59

Self-similar operators

Definition: self-similar operator We say that A ∈ BDO(X) is self-similar if A ∈ σop(A). Roughly speaking, this means that A contains a copy of itself, at infinity. Each pattern that you see once in A, you will see infinitely often in A. But then, by the above, Aess = A A is Fredholm ⇐ ⇒ A is invertible specessA = spec A specε

essA

= specεA “essential stuff = real stuff.” TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 27 / 44

slide-60
SLIDE 60

Table of Contents

1

The essentials

2

Limit operators

3

Stability of approximation methods

4

The Fibonacci Hamiltonian TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 28 / 44

slide-61
SLIDE 61

The finite section method

Task: Find an approximate solution of the equation Ax = b. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 29 / 44

slide-62
SLIDE 62

The finite section method

Task: Find an approximate solution of the equation Ax = b. Idea: Approximate A by growing but finite square submatrices An and, assuming that A is invertible, hope that also the inverses A−1

n

exist, at least for sufficiently large n, and that they converge to the inverse of A, i.e. A−1

n

→ A−1, TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 29 / 44

slide-63
SLIDE 63

The finite section method

Task: Find an approximate solution of the equation Ax = b. Idea: Approximate A by growing but finite square submatrices An and, assuming that A is invertible, hope that also the inverses A−1

n

exist, at least for sufficiently large n, and that they converge to the inverse of A, i.e. A−1

n

→ A−1, It turns out: This “hope” will come true iff the sequence (An) is stable, meaning that all An with sufficiently large n are invertible and supn≥n0 A−1

n < ∞.

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 29 / 44

slide-64
SLIDE 64

Stability is just another “essential”

The sequence (An) is stable ⇐ ⇒ D := Diag(A1, A2, . . . ) is Fredholm. This brings us back to limit operators of D – and hence of A. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 30 / 44

slide-65
SLIDE 65

Following the corners as they move out to infinity

In the end we have to follow the two “corners” (semi-infinite matrices) aln,ln · · · . . . ...

  • and

... . . . · · · arn,rn

  • f An as n → ∞ and find (partial) limits of these matrix sequences:

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 31 / 44

slide-66
SLIDE 66

Following the corners as they move out to infinity

In the end we have to follow the two “corners” (semi-infinite matrices) aln,ln · · · . . . ...

  • and

... . . . · · · arn,rn

  • f An as n → ∞ and find (partial) limits of these matrix sequences:

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 31 / 44

slide-67
SLIDE 67

Following the corners as they move out to infinity

In the end we have to follow the two “corners” (semi-infinite matrices) aln,ln · · · . . . ...

  • and

... . . . · · · arn,rn

  • f An as n → ∞ and find (partial) limits of these matrix sequences:

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 31 / 44

slide-68
SLIDE 68

Following the corners as they move out to infinity

In the end we have to follow the two “corners” (semi-infinite matrices) aln,ln · · · . . . ...

  • and

... . . . · · · arn,rn

  • f An as n → ∞ and find (partial) limits of these matrix sequences:

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 31 / 44

slide-69
SLIDE 69

Table of Contents

1

The essentials

2

Limit operators

3

Stability of approximation methods

4

The Fibonacci Hamiltonian TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 32 / 44

slide-70
SLIDE 70

The Fibonacci Hamiltonian

The Fibonacci Hamiltonian is a particular discrete Schr¨

  • dinger operator in 1D:

(Ax)n = xn−1 + vnxn + xn+1, n ∈ Z. So, again, the matrix looks like this A =               ... ... ... v−2 1 1 v−1 1 1 v0 1 1 v1 1 1 v2 ... ... ...               . The potential v only assumes the values 0 and 1 – but in a very interesting pattern. 50 letters of the Fibonacci word (“quasiperiodic”) . . . 10110101101101011010110110101101101011010110110101 . . . TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 33 / 44

slide-71
SLIDE 71

Fibonacci and his rabbit population

time population count 1 2 3 4 5 6 7 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 34 / 44

slide-72
SLIDE 72

When rabbits become numbers

time population count 1 1 1 2 10 2 3 101 3 4 10110 5 5 10110101 8 6 1011010110110 13 7 101101011011010110101 21 8 1011010110110101101011011010110110 34 9 1011010110110101101011011010110110101101011011010110101 55 . . . . . . . . . Three equivalent constructions of the Fibonacci word 0 → 1, 1 → 10; fk+1 := fkfk−1; vn = χ[1−α,1)(nα mod 1), α =

2 1+ √ 5

The last formula is also used to define vn for all n ∈ Z. (⇒ bi-infinite Fibonacci word) TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 35 / 44

slide-73
SLIDE 73

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-74
SLIDE 74

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-75
SLIDE 75

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-76
SLIDE 76

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-77
SLIDE 77

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-78
SLIDE 78

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-79
SLIDE 79

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-80
SLIDE 80

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-81
SLIDE 81

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-82
SLIDE 82

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 1 1 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-83
SLIDE 83

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 1 1 · · · TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-84
SLIDE 84

The “mod 1” rotation formula and limit operators

vn = χ[1−α,1)( nα mod 1 ), n ∈ Z, α =

2 1+ √ 5

(“golden mean”) n . . . −3 −2 −1 1 2 3 4 5 6 7 8 · · · vn 1 1 1 1 1 · · · TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 36 / 44

slide-85
SLIDE 85

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-86
SLIDE 86

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 2 01, 10, 11 3 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-87
SLIDE 87

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 2 01, 10, 11 3 3 010, 011, 101, 110 4 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-88
SLIDE 88

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 2 01, 10, 11 3 3 010, 011, 101, 110 4 4 0101, 0110, 1010, 1011, 1101 5 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-89
SLIDE 89

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 2 01, 10, 11 3 3 010, 011, 101, 110 4 4 0101, 0110, 1010, 1011, 1101 5 . . . . . . n · · · n + 1 Interesting feature: Very moderate (in fact: minimal) growth, compared to 2n. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-90
SLIDE 90

Fibonacci word: subword complexity

In an infinite random word over the alphabet {0, 1} you can find (almost surely) all 2n subwords of length n. In contrast: How many can you find in the Fibonacci word v = · · · 1011010110110101101011011010110110101101011011010110101 · · · ? List of subwords of length n length subwords count 1 0, 1 2 2 01, 10, 11 3 3 010, 011, 101, 110 4 4 0101, 0110, 1010, 1011, 1101 5 . . . . . . n · · · n + 1 Interesting feature: Very moderate (in fact: minimal) growth, compared to 2n. One can show: The main diagonal of every limit operator of A has the same list of subwords! TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 37 / 44

slide-91
SLIDE 91

Limit operators and their subwords

Let v = · · · 10110101101101011010110110101101101011 · · · be the Fibonacci word, A = S−1 + Mv + S1 be the Fibonacci Hamiltonian, TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 38 / 44

slide-92
SLIDE 92

Limit operators and their subwords

Let v = · · · 10110101101101011010110110101101101011 · · · be the Fibonacci word, A = S−1 + Mv + S1 be the Fibonacci Hamiltonian, h = (n1, n2, . . . ) be a sequence in Z with nk → ±∞ and limit operator Ah = S−1 + Mvh + S1 with vh = limk→∞ S−nk v (pointwise). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 38 / 44

slide-93
SLIDE 93

Limit operators and their subwords

Let v = · · · 10110101101101011010110110101101101011 · · · be the Fibonacci word, A = S−1 + Mv + S1 be the Fibonacci Hamiltonian, h = (n1, n2, . . . ) be a sequence in Z with nk → ±∞ and limit operator Ah = S−1 + Mvh + S1 with vh = limk→∞ S−nk v (pointwise). The main diagonal of the limit operator Ah has the same subwords as that of A. w ≺ v ⇐ ⇒ w ≺ vh TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 38 / 44

slide-94
SLIDE 94

Limit operators and their subwords

Let v = · · · 10110101101101011010110110101101101011 · · · be the Fibonacci word, A = S−1 + Mv + S1 be the Fibonacci Hamiltonian, h = (n1, n2, . . . ) be a sequence in Z with nk → ±∞ and limit operator Ah = S−1 + Mvh + S1 with vh = limk→∞ S−nk v (pointwise). The main diagonal of the limit operator Ah has the same subwords as that of A. w ≺ v ⇐ ⇒ w ≺ vh ⇐ w ≺ vh = ⇒ w ≺ S−nk v for large k = ⇒ w ≺ v. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 38 / 44

slide-95
SLIDE 95

Limit operators and their subwords

Let v = · · · 10110101101101011010110110101101101011 · · · be the Fibonacci word, A = S−1 + Mv + S1 be the Fibonacci Hamiltonian, h = (n1, n2, . . . ) be a sequence in Z with nk → ±∞ and limit operator Ah = S−1 + Mvh + S1 with vh = limk→∞ S−nk v (pointwise). The main diagonal of the limit operator Ah has the same subwords as that of A. w ≺ v ⇐ ⇒ w ≺ vh ⇐ w ≺ vh = ⇒ w ≺ S−nk v for large k = ⇒ w ≺ v. ⇒ Let w ≺ v, say (w.l.o.g.) w ≺ v+. = ⇒ Every S−nk v contains w in a Fn+1-neighbourhood of zero. = ⇒ Every limit potential vh contains w in a Fn+1-neighbourhood of zero. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 38 / 44

slide-96
SLIDE 96

Limit operators and the “mod 1” rotation formula

For the Fibonacci Hamiltonian A = S−1 + Mv + S1, one gets σop(A) =

  • S−1 + Mvθ + S1, S−1 + Mwθ + S1 : θ ∈ [0, 1)
  • ,

where v θ

n

:= χ[1−α,1)(θ + nα mod 1), w θ

n

:= χ(1−α,1](θ + nα mod 1), n ∈ Z. In particular, A ∈ σop(A); so A is self-similar. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 39 / 44

slide-97
SLIDE 97

What do we want from the Fibonacci Hamiltonian?

A lot is known of the spectrum of A; it is a Cantor set on the real line

  • f Lebesgue measure zero,

there is no point spectrum (w.r.t. ℓ2) in fact, the spectrum is purely singular continuous... TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 40 / 44

slide-98
SLIDE 98

What do we want from the Fibonacci Hamiltonian?

A lot is known of the spectrum of A; it is a Cantor set on the real line

  • f Lebesgue measure zero,

there is no point spectrum (w.r.t. ℓ2) in fact, the spectrum is purely singular continuous... Our focus: Applicability of the FSM with arbitrary cut-off points. We show this via invertibility of B, B+ and B− for all B ∈ σop(A), including B = A (i.e. 0 is not in the spectrum of any of these operators). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 40 / 44

slide-99
SLIDE 99

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) A is injective on ℓ2 A∗ is injective on ℓ2 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-100
SLIDE 100

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are invertible on ℓ2 A is injective on ℓ2 A∗ is injective on ℓ2 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-101
SLIDE 101

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are injective on ℓ∞ A is injective on ℓ2 A∗ is injective on ℓ2 TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-102
SLIDE 102

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are injective on ℓ∞ A is injective on ℓ2 ⇐ = A is injective on ℓ∞ A∗ is injective on ℓ2 ⇐ = A = A∗ is injective on ℓ∞ TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-103
SLIDE 103

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are injective on ℓ∞ A is injective on ℓ2 ⇐ = A is injective on ℓ∞ A∗ is injective on ℓ2 ⇐ = A = A∗ is injective on ℓ∞ Similarly: For invertibility of all B+ and B− it is enough to show their injectivity on ℓ∞ (since all are Fredholm and self-adjoint). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-104
SLIDE 104

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are injective on ℓ∞ A is injective on ℓ2 ⇐ = A is injective on ℓ∞ A∗ is injective on ℓ2 ⇐ = A = A∗ is injective on ℓ∞ Similarly: For invertibility of all B+ and B− it is enough to show their injectivity on ℓ∞ (since all are Fredholm and self-adjoint). We demonstrate this for B = A (so that B+ = A+): Let A+x = o, i.e.            1 1 1 1 1 1 1 1 1 1 1 1 1 1 ... ... ...                       x1 x2 x3 x4 x5 x6 . . .            =            . . .            . In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-105
SLIDE 105

Sketch of proof

To show that A is invertible on ℓ2 (hence on any ℓp), we show that A is Fredholm (⇒ closed range) ⇐ ⇒ all B ∈ σop(A) are injective on ℓ∞ A is injective on ℓ2 ⇐ = A is injective on ℓ∞ A∗ is injective on ℓ2 ⇐ = A = A∗ is injective on ℓ∞ Similarly: For invertibility of all B+ and B− it is enough to show their injectivity on ℓ∞ (since all are Fredholm and self-adjoint). We demonstrate this for B = A (so that B+ = A+): Let A+x = o, i.e.            1 1 1 1 1 1 1 1 1 1 1 1 1 1 ... ... ...                       x1 x2 x3 x4 x5 x6 . . .            =            . . .            . In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . . TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 41 / 44

slide-106
SLIDE 106

Sketch of proof

In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . . TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 42 / 44

slide-107
SLIDE 107

Sketch of proof

In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . .

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 vn 1 1 1 1 1 1 1 1 1 1 1 1 xn 1 −1 −1 2 −1 −2 3 2 −5 3 5 −8 3 8 −11 −8 19 −11 −19 30

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 42 / 44

slide-108
SLIDE 108

Sketch of proof

In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . .

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 vn 1 1 1 1 1 1 1 1 1 1 1 1 xn 1 −1 −1 2 −1 −2 3 2 −5 3 5 −8 3 8 −11 −8 19 −11 −19 30

· · · = ⇒ x ∈ ℓ∞ = ⇒ A+ is injective on ℓ∞ TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 42 / 44

slide-109
SLIDE 109

Sketch of proof

In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . .

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 vn 1 1 1 1 1 1 1 1 1 1 1 1 xn 1 −1 −1 2 −1 −2 3 2 −5 3 5 −8 3 8 −11 −8 19 −11 −19 30

· · · = ⇒ x ∈ ℓ∞ = ⇒ A+ is injective on ℓ∞ Similarly: B+ and B− are injective on ℓ∞ for all B ∈ σop(A). = ⇒ B+ and B− are invertible for all B ∈ σop(A). TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 42 / 44

slide-110
SLIDE 110

Sketch of proof

In short: xn−1 + vnxn + xn+1 = 0 for all n ∈ N. Starting with x0 = 0 and x1 = 1 (w.l.o.g.) this is a 2-term recurrence for x2, x3, . . .

n 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 vn 1 1 1 1 1 1 1 1 1 1 1 1 xn 1 −1 −1 2 −1 −2 3 2 −5 3 5 −8 3 8 −11 −8 19 −11 −19 30

· · · = ⇒ x ∈ ℓ∞ = ⇒ A+ is injective on ℓ∞ Similarly: B+ and B− are injective on ℓ∞ for all B ∈ σop(A). = ⇒ B+ and B− are invertible for all B ∈ σop(A). Theorem (ML, S¨

  • ding 2016)

The FSM is applicable, with arbirtrary cut-off points, to A and also to A+. A−1

n

→ A−1 A−1

+,n → A−1 +

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 42 / 44

slide-111
SLIDE 111

Thank you!

TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 43 / 44

slide-112
SLIDE 112

Literature

S.N. Chandler-Wilde and M. Lindner: Sufficiency of Favards condition for a class of band-dominated operators...,

  • J. Functional Analysis, 254 (2008).
  • M. Lindner and M. Seidel:

An affirmative answer to a core issue on limit operators,

  • J. Functional Analysis, 267 (2014).
  • R. Hagger, M. Lindner and M. Seidel:

Essential pseudospectra and essential norms of band-dominated operators,

  • J. Mathematical Analysis and Applications, 437 (2016).
  • V. Rabinovich, S. Roch and B. Silbermann:

Limit Operators and Their Applications in Operator Theory, Birkh¨ auser 2004.

  • M. Lindner:

Infinite Matrices and their Finite Sections, Birkh¨ auser 2006. TUHH

Marko Lindner LimOps: Your hands on the essentials 28th IWOTA, Aug 2017, Chemnitz 44 / 44