Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues...

48
Introduction Power-Ginibre decomposition Other ensembles Powers of Ginibre Eigenvalues The Laundry Machine Effect Guillaume Dubach Courant Institute, NYU Northeast Probability Seminar November 15th, 2018 Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 1 / 21

Transcript of Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues...

Page 1: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Powers of Ginibre EigenvaluesThe Laundry Machine Effect

Guillaume DubachCourant Institute, NYU

Northeast Probability SeminarNovember 15th, 2018

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 1 / 21

Page 2: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Complex Ginibre EnsembleGinibre ensemble: Gin(N), N ×N matrix G , with i.i.d. entries

Gi ,jd=NC (0, Id) .

Eigenvalues are almost surely distinct : G =P∆P−1.∆= Diag(λ1, . . . ,λN) with density proportional to∏

i<j|λi −λj |2e−

∑Ni=1 |λi |2

with respect to the Lebesgue measure on C.

Convergence of the empirical measure of the scaled eigenvalues tothe uniform measure on D=D(0,1).

N∑k=1

δ λkpN

d→ 1π1D.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 2 / 21

Page 3: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Complex Ginibre EnsembleGinibre ensemble: Gin(N), N ×N matrix G , with i.i.d. entries

Gi ,jd=NC (0, Id) .

Eigenvalues are almost surely distinct : G =P∆P−1.

∆= Diag(λ1, . . . ,λN) with density proportional to∏i<j

|λi −λj |2e−∑N

i=1 |λi |2

with respect to the Lebesgue measure on C.

Convergence of the empirical measure of the scaled eigenvalues tothe uniform measure on D=D(0,1).

N∑k=1

δ λkpN

d→ 1π1D.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 2 / 21

Page 4: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Complex Ginibre EnsembleGinibre ensemble: Gin(N), N ×N matrix G , with i.i.d. entries

Gi ,jd=NC (0, Id) .

Eigenvalues are almost surely distinct : G =P∆P−1.∆= Diag(λ1, . . . ,λN) with density proportional to∏

i<j|λi −λj |2e−

∑Ni=1 |λi |2

with respect to the Lebesgue measure on C.

Convergence of the empirical measure of the scaled eigenvalues tothe uniform measure on D=D(0,1).

N∑k=1

δ λkpN

d→ 1π1D.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 2 / 21

Page 5: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Complex Ginibre EnsembleGinibre ensemble: Gin(N), N ×N matrix G , with i.i.d. entries

Gi ,jd=NC (0, Id) .

Eigenvalues are almost surely distinct : G =P∆P−1.∆= Diag(λ1, . . . ,λN) with density proportional to∏

i<j|λi −λj |2e−

∑Ni=1 |λi |2

with respect to the Lebesgue measure on C.

Convergence of the empirical measure of the scaled eigenvalues tothe uniform measure on D=D(0,1).

N∑k=1

δ λkpN

d→ 1π1D.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 2 / 21

Page 6: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Circular Law : Ginibre points vs. uniform points

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 3 / 21

Page 7: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Definition (Gamma variables)

For any α> 0, γα is a real variable with density

1Γ(α)

tα−1e−t1R+ .

Theorem (Kostlan ’92){|λ1|2, . . . , |λN |2} d= {γ1, . . . ,γN

}, where the gamma variables are

independent, with parameters 1,2, . . . ,N.

One consequence : Gumbel fluctuations for max(|λi |2). (Rider ’03,Chafaï-Péché ’ 14)

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 4 / 21

Page 8: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Definition (Gamma variables)

For any α> 0, γα is a real variable with density

1Γ(α)

tα−1e−t1R+ .

Theorem (Kostlan ’92){|λ1|2, . . . , |λN |2} d= {γ1, . . . ,γN

}, where the gamma variables are

independent, with parameters 1,2, . . . ,N.

One consequence : Gumbel fluctuations for max(|λi |2). (Rider ’03,Chafaï-Péché ’ 14)

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 4 / 21

Page 9: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Definition (Gamma variables)

For any α> 0, γα is a real variable with density

1Γ(α)

tα−1e−t1R+ .

Theorem (Kostlan ’92){|λ1|2, . . . , |λN |2} d= {γ1, . . . ,γN

}, where the gamma variables are

independent, with parameters 1,2, . . . ,N.

One consequence : Gumbel fluctuations for max(|λi |2). (Rider ’03,Chafaï-Péché ’ 14)

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 4 / 21

Page 10: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 5 / 21

Page 11: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Repulsion between eigenvalues

(Term∏

i<j |λi −λj |2 in the joint density.)

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 6 / 21

Page 12: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 7 / 21

Page 13: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Kostlan’s Theorem

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 8 / 21

Page 14: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

High powersPower map : πM(z)= zM . Image of Gin(N) under πM ?

Theorem (Hough, Krishnapur, Peres, Virág ’06)

For any integer M ≥N, the following equality in distribution holds:{λM1 , . . . ,λMN

}d= {Z1, . . . ,ZN }

where the variables Zk are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 9 / 21

Page 15: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

High powersPower map : πM(z)= zM . Image of Gin(N) under πM ?

Theorem (Hough, Krishnapur, Peres, Virág ’06)

For any integer M ≥N, the following equality in distribution holds:{λM1 , . . . ,λMN

}d= {Z1, . . . ,ZN }

where the variables Zk are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 9 / 21

Page 16: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

High powersPower map : πM(z)= zM . Image of Gin(N) under πM ?

Theorem (Hough, Krishnapur, Peres, Virág ’06)

For any integer M ≥N, the following equality in distribution holds:{λM1 , . . . ,λMN

}d= {Z1, . . . ,ZN }

where the variables Zk are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 9 / 21

Page 17: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Rains for the CUE

CUE(N) : unitary matrix chosen uniformly on UN(C).

Eigenvaluedensity proportional to: ∏

k<l|e iθk −e iθl |2

with respect to the Lebesgue measure on the unit circle.

Theorem (Rains ’03)

For any M ≥ 1,

CUE(N)Md=

N⋃k=1

CUE

(⌈N −k

M

⌉).

where the CUE blocks in the right hand side are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 10 / 21

Page 18: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Rains for the CUE

CUE(N) : unitary matrix chosen uniformly on UN(C). Eigenvaluedensity proportional to: ∏

k<l|e iθk −e iθl |2

with respect to the Lebesgue measure on the unit circle.

Theorem (Rains ’03)

For any M ≥ 1,

CUE(N)Md=

N⋃k=1

CUE

(⌈N −k

M

⌉).

where the CUE blocks in the right hand side are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 10 / 21

Page 19: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Rains for the CUE

CUE(N) : unitary matrix chosen uniformly on UN(C). Eigenvaluedensity proportional to: ∏

k<l|e iθk −e iθl |2

with respect to the Lebesgue measure on the unit circle.

Theorem (Rains ’03)

For any M ≥ 1,

CUE(N)Md=

N⋃k=1

CUE

(⌈N −k

M

⌉).

where the CUE blocks in the right hand side are independent.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 10 / 21

Page 20: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre blocks

For fixed M ≤N and k ∈ [[1,M]], we consider the arithmeticprogressions

Ik :={i ∈ [[1,N]] | i ≡ k [M]

}.

DefinitionThe Power-Ginibre distribution Gin(N ,M ,k) is the point processindexed by Ik with joint density proportional to∏

i<ji ,j∈Ik

|zi −zj |2∏i∈Ik

|zi |2(k−M)

M e−|zi |2/M

dm(zi ).

Example: Gin(5,2,1) consists of {z1,z3,z5} with the above density.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 11 / 21

Page 21: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre blocks

For fixed M ≤N and k ∈ [[1,M]], we consider the arithmeticprogressions

Ik :={i ∈ [[1,N]] | i ≡ k [M]

}.

DefinitionThe Power-Ginibre distribution Gin(N ,M ,k) is the point processindexed by Ik with joint density proportional to∏

i<ji ,j∈Ik

|zi −zj |2∏i∈Ik

|zi |2(k−M)

M e−|zi |2/M

dm(zi ).

Example: Gin(5,2,1) consists of {z1,z3,z5} with the above density.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 11 / 21

Page 22: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre blocks

For fixed M ≤N and k ∈ [[1,M]], we consider the arithmeticprogressions

Ik :={i ∈ [[1,N]] | i ≡ k [M]

}.

DefinitionThe Power-Ginibre distribution Gin(N ,M ,k) is the point processindexed by Ik with joint density proportional to∏

i<ji ,j∈Ik

|zi −zj |2∏i∈Ik

|zi |2(k−M)

M e−|zi |2/M

dm(zi ).

Example: Gin(5,2,1) consists of {z1,z3,z5} with the above density.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 11 / 21

Page 23: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre decomposition

Theorem (D. ’18)

The images of Ginibre eigenvalues under πM are distributed like asuperposition of M independent Power-Ginibre point processes.

Gin(N)Md=

M⋃k=1

Gin(N ,M ,k).

d= t t·· ·t︸ ︷︷ ︸M independent blocks

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 12 / 21

Page 24: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre decomposition

Theorem (D. ’18)

The images of Ginibre eigenvalues under πM are distributed like asuperposition of M independent Power-Ginibre point processes.

Gin(N)Md=

M⋃k=1

Gin(N ,M ,k).

d= t t·· ·t︸ ︷︷ ︸M independent blocks

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 12 / 21

Page 25: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre decomposition

Theorem (D. ’18)

The images of Ginibre eigenvalues under πM are distributed like asuperposition of M independent Power-Ginibre point processes.

Gin(N)Md=

M⋃k=1

Gin(N ,M ,k).

d=

t t·· ·t︸ ︷︷ ︸M independent blocks

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 12 / 21

Page 26: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre decomposition

Theorem (D. ’18)

The images of Ginibre eigenvalues under πM are distributed like asuperposition of M independent Power-Ginibre point processes.

Gin(N)Md=

M⋃k=1

Gin(N ,M ,k).

d= t

t·· ·t︸ ︷︷ ︸M independent blocks

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 12 / 21

Page 27: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Power-Ginibre decomposition

Theorem (D. ’18)

The images of Ginibre eigenvalues under πM are distributed like asuperposition of M independent Power-Ginibre point processes.

Gin(N)Md=

M⋃k=1

Gin(N ,M ,k).

d= t t·· ·t︸ ︷︷ ︸M independent blocks

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 12 / 21

Page 28: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Repulsion breaking

Gin(N)

πM : z 7→ zM

Gin(N)M

Gin(N,M,2)

Gin(N,M,1)Gin(N,M,1)

Gin(N,M,M)

(Superposition)

Gin(N)M

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 13 / 21

Page 29: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Twisted circular law

M fixed, N →∞. The pushforward of the uniform distribution onthe disk has density 1

πM |z |2/M−2.

• Convergence to the twisted circular law holds for every block.• Second order: i.i.d. Gaussian fluctuations coherent with GFF.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 14 / 21

Page 30: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Twisted circular law

M fixed, N →∞. The pushforward of the uniform distribution onthe disk has density 1

πM |z |2/M−2.

• Convergence to the twisted circular law holds for every block.• Second order: i.i.d. Gaussian fluctuations coherent with GFF.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 14 / 21

Page 31: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Twisted circular law

M fixed, N →∞. The pushforward of the uniform distribution onthe disk has density 1

πM |z |2/M−2.

• Convergence to the twisted circular law holds for every block.

• Second order: i.i.d. Gaussian fluctuations coherent with GFF.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 14 / 21

Page 32: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Twisted circular law

M fixed, N →∞. The pushforward of the uniform distribution onthe disk has density 1

πM |z |2/M−2.

• Convergence to the twisted circular law holds for every block.• Second order: i.i.d. Gaussian fluctuations coherent with GFF.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 14 / 21

Page 33: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Proof : main tool

Lemma (Andréief 1883)

Let (E ,E ,ν) be a measure space.For any functions (φi ,ψi )

Ni=1 in L2(ν),

1N!

∫EN

det(φi (λj)) det(ψi (λj)) dνN(λ)= det(fi ,j)

where fi ,j =∫Eφi (λ)ψj(λ)dν(λ).

With φi (z)= z i−1,ψi (z)= z i−1, this gives a formula for thestatistics E

(∏Ni=1P(λi )

)with any polynomial P .

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 15 / 21

Page 34: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Proof : main tool

Lemma (Andréief 1883)

Let (E ,E ,ν) be a measure space.For any functions (φi ,ψi )

Ni=1 in L2(ν),

1N!

∫EN

det(φi (λj)) det(ψi (λj)) dνN(λ)= det(fi ,j)

where fi ,j =∫Eφi (λ)ψj(λ)dν(λ).

With φi (z)= z i−1,ψi (z)= z i−1, this gives a formula for thestatistics E

(∏Ni=1P(λi )

)with any polynomial P .

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 15 / 21

Page 35: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

General 2D β-ensembles

These results do not depend on the Gaussian reference measure.

The same is true for the joint density∏i<j

|λi −λj |βe−∑N

i=1V (|λi |2)

with β= 2 and any suitable V .

For instance, Kostlan’s theorem holds with generalized Gammavariable (Density tα−1e−V (t) on R+).

Conditional results available for any β= 2p,p ∈N.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 16 / 21

Page 36: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

General 2D β-ensembles

These results do not depend on the Gaussian reference measure.

The same is true for the joint density∏i<j

|λi −λj |βe−∑N

i=1V (|λi |2)

with β= 2 and any suitable V .

For instance, Kostlan’s theorem holds with generalized Gammavariable (Density tα−1e−V (t) on R+).

Conditional results available for any β= 2p,p ∈N.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 16 / 21

Page 37: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

General 2D β-ensembles

These results do not depend on the Gaussian reference measure.

The same is true for the joint density∏i<j

|λi −λj |βe−∑N

i=1V (|λi |2)

with β= 2 and any suitable V .

For instance, Kostlan’s theorem holds with generalized Gammavariable (Density tα−1e−V (t) on R+).

Conditional results available for any β= 2p,p ∈N.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 16 / 21

Page 38: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

General 2D β-ensembles

These results do not depend on the Gaussian reference measure.

The same is true for the joint density∏i<j

|λi −λj |βe−∑N

i=1V (|λi |2)

with β= 2 and any suitable V .

For instance, Kostlan’s theorem holds with generalized Gammavariable (Density tα−1e−V (t) on R+).

Conditional results available for any β= 2p,p ∈N.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 16 / 21

Page 39: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Figure: Quaternionic ’eigenvalues’.

B Artistic view : actual product may vary.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 17 / 21

Page 40: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Figure: Quaternionic ’eigenvalues’.

B Artistic view : actual product may vary.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 17 / 21

Page 41: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

QGE: eigenvalues (λ1,λ1, . . . ,λN ,λN) with joint density∏i<j

|λi −λj |2∏i≤j

|λi −λj |2e−∑ |λi |2dµN(λ).

Note that :∏i<j

|λi −λj |2∏i≤j

|λi −λj |2 =∆2N(λ1, . . . ,λN ,λ1, . . . ,λN)N∏i=1

(λi −λi ).

∣∣∣∣∣∣∣∣∣∣1 . . . 1 1 . . . 1λ1 . . . λN λ1 . . . λN...

......

...

λ2N−11 . . . λ2N−1

N λ2N−11 . . . λ

2N−1N

∣∣∣∣∣∣∣∣∣∣

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 18 / 21

Page 42: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

QGE: eigenvalues (λ1,λ1, . . . ,λN ,λN) with joint density∏i<j

|λi −λj |2∏i≤j

|λi −λj |2e−∑ |λi |2dµN(λ).

Note that :∏i<j

|λi −λj |2∏i≤j

|λi −λj |2 =∆2N(λ1, . . . ,λN ,λ1, . . . ,λN)N∏i=1

(λi −λi ).

∣∣∣∣∣∣∣∣∣∣1 . . . 1 1 . . . 1λ1 . . . λN λ1 . . . λN...

......

...

λ2N−11 . . . λ2N−1

N λ2N−11 . . . λ

2N−1N

∣∣∣∣∣∣∣∣∣∣

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 18 / 21

Page 43: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

QGE: eigenvalues (λ1,λ1, . . . ,λN ,λN) with joint density∏i<j

|λi −λj |2∏i≤j

|λi −λj |2e−∑ |λi |2dµN(λ).

Note that :∏i<j

|λi −λj |2∏i≤j

|λi −λj |2 =∆2N(λ1, . . . ,λN ,λ1, . . . ,λN)N∏i=1

(λi −λi ).

∣∣∣∣∣∣∣∣∣∣1 . . . 1 1 . . . 1λ1 . . . λN λ1 . . . λN...

......

...

λ2N−11 . . . λ2N−1

N λ2N−11 . . . λ

2N−1N

∣∣∣∣∣∣∣∣∣∣

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 18 / 21

Page 44: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Proposition (De Bruijn)

Let (E ,E ,ν) be a measure space.For any functions (φi ,ψi )

2Ni=1 in L2(ν),∫

ENdet(φi (λj) | ψi (λj)) dνN(λ)=N!2NPf(fi ,j)

2Ni ,j=1

wherefi ,j =

∫Eφi (λ)ψj(λ)dν(λ).

where Pf is the pfaffian.

(PfM)2 = det(M−Mt

2

).

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 19 / 21

Page 45: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Proposition (De Bruijn)

Let (E ,E ,ν) be a measure space.For any functions (φi ,ψi )

2Ni=1 in L2(ν),∫

ENdet(φi (λj) | ψi (λj)) dνN(λ)=N!2NPf(fi ,j)

2Ni ,j=1

wherefi ,j =

∫Eφi (λ)ψj(λ)dν(λ).

where Pf is the pfaffian. (PfM)2 = det(M−Mt

2

).

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 19 / 21

Page 46: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Theorem (Rider ’03)

For QGE,{|λ1|2, . . . , |λN |2} d= {

γ2, . . . ,γ2N}

.

Theorem (D. ’18)

For any integer M ≥ 2N,{λM1 ,λ1

M, . . . ,λMN ,λN

M}d=

{Z1,Z 1, . . . ,ZN ,ZN

}where the Zi ’s are independent variables.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 20 / 21

Page 47: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Quaternionic Ginibre Ensemble

Theorem (Rider ’03)

For QGE,{|λ1|2, . . . , |λN |2} d= {

γ2, . . . ,γ2N}

.

Theorem (D. ’18)

For any integer M ≥ 2N,{λM1 ,λ1

M, . . . ,λMN ,λN

M}d=

{Z1,Z 1, . . . ,ZN ,ZN

}where the Zi ’s are independent variables.

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 20 / 21

Page 48: Powers of Ginibre Eigenvaluesdubach/Beamer_Ginibre_Powers.pdfPowers of Ginibre Eigenvalues TheLaundryMachineEffect GuillaumeDubach CourantInstitute,NYU NortheastProbabilitySeminar

Introduction Power-Ginibre decomposition Other ensembles

Thank you for your attention !

Guillaume Dubach Courant Institute, NYU Powers of Ginibre Eigenvalues 21 / 21