Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities...

39
Surgery of quantum graphs: eigenvalues and heat kernels Delio Mugnolo Discrete Laplacians Quantum graph Laplacians Spectral inequalities Heat kernels Surgery of quantum graphs: eigenvalues and heat kernels Delio Mugnolo University of Hagen March 22, 2018 1 / 39

Transcript of Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities...

Page 1: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Surgery of quantum graphs:eigenvalues and heat kernels

Delio Mugnolo

University of Hagen

March 22, 2018

1 / 39

Page 2: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

2 / 39

Page 3: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Braess’ paradox

In unfavorable cases, an extension of the roadnetwork may result in longer running times.

3 / 39

Page 4: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Braess’ paradox

In unfavorable cases, an extension of the roadnetwork may result in longer running times.

Figure: D. Braess 1968

4 / 39

Page 5: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Braess’ metatheorem

Networks don’t necessarily become more efficient bymerely getting denser and denser!

In the context of quantum graphs:

▸ Exner-Jex 2012

▸ Kurasov-Malenova-Naboko 2013

5 / 39

Page 6: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Outline

▸ Discrete graphs

▸ Quantum graphs

▸ Heat kernels

6 / 39

Page 7: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Let G = (V,E) be a loopless (finite) graph (possibly amultigraph, no loops!).Regard f ∶ V → C as a function on G: the natural Hilbertspace is `2(V) ≃ CV.

Upon chosing an arbitrary orientation, consider the incidencematrix I = (ιve):

ιve ∶=⎧⎪⎪⎪⎨⎪⎪⎪⎩

−1 if v is initial endpoint of e,+1 if v is terminal endpoint of e,0 otherwise

ThenLG ∶= IIT

defines a (bounded) self-adjoint, positive semidefiniteoperator on `2(V): the discrete Laplacian (Kirchhoff 1847,Brooks-Smith-Stone-Tutte 1940, Anderson-Morley 1971).

7 / 39

Page 8: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Let G = (V,E) be a loopless (finite) graph (possibly amultigraph, no loops!).Regard f ∶ V → C as a function on G: the natural Hilbertspace is `2(V) ≃ CV.

Upon chosing an arbitrary orientation, consider the incidencematrix I = (ιve):

ιve ∶=⎧⎪⎪⎪⎨⎪⎪⎪⎩

−1 if v is initial endpoint of e,+1 if v is terminal endpoint of e,0 otherwise

ThenLG ∶= IIT

defines a (bounded) self-adjoint, positive semidefiniteoperator on `2(V): the discrete Laplacian (Kirchhoff 1847,Brooks-Smith-Stone-Tutte 1940, Anderson-Morley 1971).

8 / 39

Page 9: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Spectral graph theory

The spectrum {0 = λ0, λ1, . . . , λ∣V∣−1} of LG carries muchinteresting information:

▸ multiplicity of 0 in σ(LG) = # of connectedcomponents of G

▸ the higher the spectral gap

λ1(LG) = inff ∈`2(V)f ⊥1

∥IT f ∥2`2(E)

∥f ∥2`2(V)

the more connected G (λ1(LG) is also called algebraicconnectivity of G).

▸ Eigenfunctions associated with λ1(LG), λ2(LG) inducevery reasonable drawings of planar graphs

▸ . . . (see e.g. Spielman 2007, Spielman 2010)

9 / 39

Page 10: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

“Qualitative” spectral graph theory

Idea: relate spectral quantities of LG with somecombinatorial quantities of G.

Example

▸ Fiedler 1973: η ≥ λ1(LG) ≥ 2η (1 − cos π∣V∣)

▸ Dodziuk 1984, Alon-Milman 1985:2h ≥ λ1(LG) ≥ h2

2 degmax

▸ McKay 1988, Chung 1989, Mohar 1991:degmax(∣V∣−1)(2 D−∣V∣+1)+ ≥ λ1(LG) ≥ 4

∣V∣D

(η edge connectivity, h Cheeger constant of G,degmax maximal degree, D diameter)

10 / 39

Page 11: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Inserting edges tends to lower λ1(LG), but. . .

Figure: C. Maas 1987

. . . sometimes λ1(LG) = λ1(LG′)

11 / 39

Page 12: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Quantum graphs

A quantum graph (or metric graph, or network) G isobtained by associating an interval (0, `e) of length `e witheach edge e of G = (V,E).

Natural Hilbert space on G:

L2(G) ∶=⊕e∈E

L2(0, `e)

12 / 39

Page 13: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

No boundary conditions can be imposed on functions inL2(G), so functions in L2(G) do not see thecombinatorics of G.

Introduce

C(G) ∶= {f ∈⊕e∈E

C [0, `e] ∶ f is continuous in each v ∈ V}

and

H1(G) ∶= {f = (fe)e∈E ∈ C(G) ∶ fe ∈ H1(0, `e) ∀e ∈ E}

13 / 39

Page 14: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

No boundary conditions can be imposed on functions inL2(G), so functions in L2(G) do not see thecombinatorics of G.

Introduce

C(G) ∶= {f ∈⊕e∈E

C [0, `e] ∶ f is continuous in each v ∈ V}

and

H1(G) ∶= {f = (fe)e∈E ∈ C(G) ∶ fe ∈ H1(0, `e) ∀e ∈ E}

14 / 39

Page 15: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Spectral gap of quantum graphs

In analogy with

λ1(LG) = inff ∈`2(V)f ⊥1

∥IT f ∥2`2(E)

∥f ∥2`2(V)

and λ1(∆) = inff ∈H1(0,1)

f ⊥1

∥f ′∥2L2(0,1)

∥f ∥2L2(0,1)

consider

λ1(∆G) ∶= inff ∈H1(G)

f ⊥1

∥f ′∥2L2(G)

∥f ∥2L2(G)

λ1(∆G) is the spectral gap of ∆G , the self-adjoint, positivesemidefinite operator on L2(G) associated with

a(f ) ∶=∑e∈E∫

`e

0∣f ′∣2, f ∈ H1(G)

15 / 39

Page 16: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Spectral gap of quantum graphs

In analogy with

λ1(LG) = inff ∈`2(V)f ⊥1

∥IT f ∥2`2(E)

∥f ∥2`2(V)

and λ1(∆) = inff ∈H1(0,1)

f ⊥1

∥f ′∥2L2(0,1)

∥f ∥2L2(0,1)

consider

λ1(∆G) ∶= inff ∈H1(G)

f ⊥1

∥f ′∥2L2(G)

∥f ∥2L2(G)

λ1(∆G) is the spectral gap of ∆G , the self-adjoint, positivesemidefinite operator on L2(G) associated with

a(f ) ∶=∑e∈E∫

`e

0∣f ′∣2, f ∈ H1(G)

16 / 39

Page 17: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

A secular equation yielding all eigenvalues of ∆G is known(von Below 1985, Kottos-Smilansky 1997) but is useless inpractice.

17 / 39

Page 18: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Spectral estimates for quantum graphs

Proposition (Nicaise 1987)

▸ λ1(∆G) ≥ π2

L2

▸ λ1(∆G) ≥ h2

4

Nicaise’ first estimate is sharp: it is attained for path graphs.Also the second estimate can be shown to be sharp, adapting an

example invented by Buser.

(L total length, h Cheeger constant of G)

(Many upper estimates by Kennedy-Kurasov-Malenova-M.2016; Band-Levy 2016; Rohleder 2016; Ariturk 2016)

18 / 39

Page 19: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proof of Nicaise’ isoperimetric inequality #1▸ Cut through all cycles of G to turn it into a tree G′:

since H1(G) ⊂ H1(G′), this lowers the Rayleighquotient.

▸ Consider the double cover G′(2) of the tree G′ (i.e.,

double each edge): the total length is 2L.▸ Each f ∈ H1(G) induces f(2) ∈ H1(G′(2)) canonically:

their Rayleigh quotients coincide.▸ For f(2) ∈ H1(G′(2)) define g ∈ H1(G′′) (≡cycle of length 2L)

by suitably dropping a few continuity conditions in thevertices (transverse the graph like in a Eulerian cycle).

v0

⇒ H1(G′(2)) can be naturally embedded in H1(G′′)19 / 39

Page 20: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proof of Nicaise’ isoperimetric inequality #2

▸ If f is an eigenfunction for λ1(∆G), g is still a testfunction for G′′ ⇒ λ1(∆G′′) is a lower bound forλ1(∆G)

▸ λ1(∆G′′) is just the spectral gap (=lowest nonzero

eigenvalue) of the second derivative on (0,2L) with

periodic boundary conditions, i.e. 4π2

(2L)2 .

20 / 39

Page 21: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Nicaise’ isoperimetric inequality has beenrediscovered/reproved by

▸ Solomyak (2002)

▸ Friedlander (2005)

▸ Berkolaiko-Kennedy-Kurasov-M. (2017)

▸ Berkolaiko-Kennedy-Kurasov-M. (2018)

21 / 39

Page 22: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proposition (Kurasov-Naboko 2014, Band-Levy 2016)

Nicaise isoperimetric inequality can be improved to

λ1(∆G) ≥4π2

L2

▸ if G is Eulerian;

▸ if, more generally, G is doubly edge connected.

22 / 39

Page 23: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proposition (Kurasov-Naboko 2014, Band-Levy 2016)

Nicaise isoperimetric inequality can be improved to

λ1(∆G) ≥4π2

L2

▸ if G is Eulerian;

▸ if, more generally, G is doubly edge connected.

23 / 39

Page 24: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Kennedy–Kurasov–Malenova–M. 2016,Kennedy-M. 2016,Berkolaiko–Kennedy–Kurasov–M. 2017

estimates on λ1(∆G) upper lower

total length 7 π2

L2

total length + # edges π2E2

L2 7

diameter + # vertices π2(V+1)2

D2 7

diameter + # edges 4π2E2

D2π2

D2E2

diameter + total length π2(4L−3D)D3

1DL

Cheeger constant π2h2E2

4h2

4

edge connectivity + total length 7 ( ηπL+`max(η−2)+ )

2

24 / 39

Page 25: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proposition (Nicaise 1987;Berkolaiko-Kennedy-Kurasov-M 2017)

▸ λ0(∆G) ≥ π2

4L2 if there are Dirichlet vertices.

▸ λ0(∆G) ≥ π2

D2 if G is a tree and all leaves are Dirichlet.

▸ λ0(∆G) ≥ π2

L2 if there are Dirichlet vertices and thegraph is doubly connected.

25 / 39

Page 26: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Surgery of quantum graphs - #1

Proposition (Berkolaiko-Kennedy-Kurasov-M)

λk(∆G) ≥ λk(∆G′) for all k

if G′ is obtained from G by

▸ cutting through two previously identified vertices;

Gv0 v2

v1 v2

v3

v4 G′

▸ attaching a pendant (e.g., an edge) to one vertex;

▸ lengthening an edge

▸ replacing k parallel edges by one edge of same totallength (“unfolding G”)

G G′26 / 39

Page 27: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Surgery of quantum graphs - #2

Proposition (Berkolaiko-Kennedy-Kurasov-M)

Given any eigenfunction ψ for λ1(∆G), then

λ1(∆G) ≥ λ1(G′)

if G′ is obtained from G by

▸ deleting an edge where ψ vanishes identically;

▸ inserting an edge of length ≥ π√λ1(∆G)

between any v ,w ;

▸ inserting an edge between any v ,w s.t. ψ(v) = ψ(w);

▸ replacing k parallel edges (along which ψ is growing) bym ≤ k equilateral edges of same total length;

G G′

▸ cutting a subgraph H with ψ(H) ⊂ [0, κ] and attachingany graph of same length at vertices where ψ is ≥ κ.

27 / 39

Page 28: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Surgery of quantum graphs - #2

Proposition (Berkolaiko-Kennedy-Kurasov-M)

Given any eigenfunction ψ for λ1(∆G), then

λ1(∆G) ≥ λ1(G′)

if G′ is obtained from G by

▸ deleting an edge where ψ vanishes identically;

▸ inserting an edge of length ≥ π√λ1(∆G)

between any v ,w ;

▸ inserting an edge between any v ,w s.t. ψ(v) = ψ(w);

▸ replacing k parallel edges (along which ψ is growing) bym ≤ k equilateral edges of same total length;

G G′

▸ cutting a subgraph H with ψ(H) ⊂ [0, κ] and attachingany graph of same length at vertices where ψ is ≥ κ.

28 / 39

Page 29: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Comparison principles for quantum graphs

Proposition (Berkolaiko-Kennedy-Kurasov-M 2017)

▸ The lasso graph with loop length S has lowest λ1(∆G)among all graphs with circumference S .

v− v+

▸ The dumbbell graph with handle length D and loopslength (L −D)/2 has lowest λ1(∆G) among all graphswith total length L and diameter D.

v− v+

▸ The dumbbell graph with handle length L−V and loopslength V /2 has lowest λ1(∆G) among all graphs withdoubly connected part of length V .

29 / 39

Page 30: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Diffusion on graphs and quantum graphs

Laplacians on graphs and quantum graphs are associatedwith Dirichlet forms:

▸ Beurling-Deny 1959, Keller-Lenz 2010 (discrete graphs)

▸ Kramar Fijavz-M-Sikolya 2007 (quantum graphs)

⇒ e−tLG

and et∆G are submarkovian semigroups.

30 / 39

Page 31: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Parabolic Braess’ paradox

Because

sup∥f ∥L2=1

∥e−t∆G f − ∫Gf ∥L2 = o(e−tλ1(∆G))

surgery principle can be interpreted for (connected) quantumgraphs: e.g.

▸ By identifying vertices of a quantum graph,convergence to equilibrium becomes faster.

▸ If there exist two vertices v,w and an eigenfunction ψ(on G) for λ1 s.t. ψ(v) = ψ(w), then diffusion onG′ ∶= G ∪ vw converges to equilibrium more slowly thanon G.

(For comparison: diffusion on G′ ∶= G ∪ vw converges toequilibrium more quickly than on G.)

31 / 39

Page 32: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

What about diffusion between pairs of vertices?Diffusing from a to z is faster on

Diffusing from b to c is faster on

32 / 39

Page 33: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Domination: the right notion?Domination of a semigroup (etA)t≥0 by another positivesemigroup (etB)t≥0 means that

∣etAf ∣ ≤ etB ∣f ∣ for all t ≥ 0and all f ≥ 0;

Examples:

▸ (et∆N )t≥0 dominates (et∆D)t≥0;

▸ (e−t(∆N)2)t≥0 does not dominate (et∆N )t≥0;

Proposition (Ouhabaz 1996)

Let a ∼ A, b ∼ B be Dirichlet forms on H, D(a) ⊂ D(b).TFAE:

▸ etA ≤ etB ;

▸ Re a(u, v) ≥ b(∣u∣, ∣v ∣) for all u ∈ D(a) and v ∈ D(b)such that ∣v ∣ ≤ ∣u∣; and additionally, D(a) is an ideal ofD(b).

33 / 39

Page 34: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Domination: the right notion?Domination of a semigroup (etA)t≥0 by another positivesemigroup (etB)t≥0 means that

∣etAf ∣ ≤ etB ∣f ∣ for all t ≥ 0and all f ≥ 0;

Examples:

▸ (et∆N )t≥0 dominates (et∆D)t≥0;

▸ (e−t(∆N)2)t≥0 does not dominate (et∆N )t≥0;

Proposition (Ouhabaz 1996)

Let a ∼ A, b ∼ B be Dirichlet forms on H, D(a) ⊂ D(b).TFAE:

▸ etA ≤ etB ;

▸ Re a(u, v) ≥ b(∣u∣, ∣v ∣) for all u ∈ D(a) and v ∈ D(b)such that ∣v ∣ ≤ ∣u∣; and additionally, D(a) is an ideal ofD(b).

34 / 39

Page 35: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proposition

▸ If G is subgraph of G′, then e−tLG /≤ e−tL

G′ /≤ e−tLG

▸ If G′ is obtained from G by cutting through twopreviously identified vertices, then

e−tLG /≤ e−tL

G′ /≤ e−tLG

.

In particular, there is no domination (in either direction) fordiffusion on spanning (either combinatorial or quantum)trees.

35 / 39

Page 36: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Interwoven semigroups

Given f > 0, (etAf )t∈[0,∞), (etB f )t∈[0,∞) are interwoven if forall t ∈ [0,∞) and some t1, t2 ≥ t one has

et1B f > et1Af and et2Af > et2B f .

Theorem (Gluck-M 2018)

Let (X , µ) be a finite measure space and A,B be distinctself-adjoint operators on L2(X , µ). If (etA)t≥0, (etB)t≥0 arepositive and map L2(X ) into L∞(X ), then TFAE:

▸ s(A) = s(B)▸ there exists 0 < f ∈ L2 such that the orbits (etAf )t≥0

and (etB f )t≥0 are interwoven.

36 / 39

Page 37: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Interwoven semigroups

Given f > 0, (etAf )t∈[0,∞), (etB f )t∈[0,∞) are interwoven if forall t ∈ [0,∞) and some t1, t2 ≥ t one has

et1B f > et1Af and et2Af > et2B f .

Theorem (Gluck-M 2018)

Let (X , µ) be a finite measure space and A,B be distinctself-adjoint operators on L2(X , µ). If (etA)t≥0, (etB)t≥0 arepositive and map L2(X ) into L∞(X ), then TFAE:

▸ s(A) = s(B)▸ there exists 0 < f ∈ L2 such that the orbits (etAf )t≥0

and (etB f )t≥0 are interwoven.

37 / 39

Page 38: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Proposition

▸ If G,G′ are two graphs on V, then there exists

0 < f ∈ `2(V) such that (e−tLGf )t≥0 and (e−tLG′

f )t≥0

are interwoven.

▸ If G,G′ are two quantum graphs of same total length,then there exists 0 < u ∈ L2(G) such that (e−t∆Gu)t≥0

and (e−t∆G′u)t≥0 are interwoven.

38 / 39

Page 39: Surgery of quantum graphs: eigenvalues and heat kernels · 2018-03-26 · Spectral inequalities Heat kernels Spectral estimates for quantum graphs Proposition (Nicaise 1987) L 1(G)≥ˇ

Surgery ofquantum graphs:eigenvalues and

heat kernels

Delio Mugnolo

DiscreteLaplacians

Quantum graphLaplacians

Spectralinequalities

Heat kernels

Thank you for your attention!

39 / 39