An introduction to the IM/ODE correspondance
Clare Dunning
University of Kent
Integrability in Low Dimensional Systems, MATRIX, July 2017
ODE/IM Correspondence
The same functional relations are satisfied by
I spectral determinants of Ordinary Differential Equations
I T and Q functions in Integrable Models
Consequently
eigenvalues of the ODE = the Bethe roots of the quantum IM
Apply IM techniques to ODEs and vice versa
ODE/IM Correspondence
The same functional relations are satisfied by
I spectral determinants of Ordinary Differential Equations
I T and Q functions in Integrable Models
Consequently
eigenvalues of the ODE = the Bethe roots of the quantum IM
Apply IM techniques to ODEs and vice versa
ODE/IM Correspondence
The same functional relations are satisfied by
I spectral determinants of Ordinary Differential Equations
I T and Q functions in Integrable Models
Consequently
eigenvalues of the ODE = the Bethe roots of the quantum IM
Apply IM techniques to ODEs and vice versa
Which IMs and ODEs?
Simplest example
XXZ spin chain
6-vertex model ←→(− d2
dx2 + x2M)ψ(x) = Eψ(x)
c ≤ 1 conformal field theory
Outline
I An integrable model: the 6-vertex model
I A differential equation: the cubic oscillator
I The IM/ODE correspondence (massless)
I The IM/ODE correspondence (massive)
I Summary
Square ice in nature?
Square ice in Nature
One atom thin layer of water between two layers of graphene
Square ice in graphene nanocapillariesG. Algara-Siller, O. Lehtinen, F. C. Wang, R. R. Nair, U. Kaiser,H. A. Wu, A. K. Geim I. V. Grigorieva, Nature 519 (2015) 443
Square ice in Nature. Perhaphs not?
One atom thin layer of water between layers of graphene
The observation of square ice in graphene questionedW. Zhou, K. Yin, C. Wang, Y. Zhang, T. Xu, A. Borisevich, L.Sun, J C Idrobo, M.F. Chisholm, S.T. Pantelides, R.F Klie andA.R. Lupini, Nature 528 (2015) E1
Square ice rules
Arrange water molecules on an N by N ′ lattice
I Oxygen atom at each site
I Two hydrogen ions bind to O by strong covalent bonds
I Two hydrogen ions bind to O by weak hydrogen bonds
Tetrahedral ice Square ice
Square ice
I Replace covalent bonds with down/left arrows
I Replace hydrogen bonds with up/right arrows
I Assume periodic boundary conditions
Six possible configurations of arrows at each vertex
6-vertex modelAllow probabilities of each configuration to vary:
W
[→↑↑→
]= W
[←↓↓←
]= sin(η + iν)
W
[→↓↓→
]= W
[←↑↑←
]= sin(η − iν)
W
[→↑↓←
]= W
[←↓↑→
]= sin(2η)
Pictorial representation of Boltzmann weight
��������
α
α
νββW
αβ β
α(ν )
where anisotropy η is fixed and spectral parameter ν varies
Partition function Z
Z =∑
configs
∏sites
W[· ·· ·
]
Transfer matrix
Tαα′αα (ν) =
∑{βi}
W
[β1α′1α1β2
](ν)W
[β2α′2α2β3
](ν) . . .W
[βNα′NαNβ1
](ν)
or representing this 2N by 2N matrix pictorially
ν
α α αα
β1β2 β3 4β βNβ1
ν ν ν(ν)T
α 1 α α
1 2
3
N
N
3
2
{βi }Σ
α
��������
��������
��������
��������
ThenZ = Trace
[TN′]
Setting ∑αα′
Tαα′αα Ψ
(j)αα′ = tj(ν)Ψ(j)
αα .
Free energy per site in the limit N ′ →∞ can be obtained as
f = − 1NN′ logZ = − 1
NN′ log Trace[TN′]∼ − 1
N log t0
Baxter’s TQ approach
Since [T(ν),T(ν ′)] = 0 we solve for each tj(ν) independently
Baxter introduced an auxiliary function q(ν) such that
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Properties
I t(ν) and q(ν) are entire functions
I t(ν) and q(ν) are iπ-periodic functions
Suppose q(ν1) = q(ν2) · · · = q(νn) = 0. Then
q(ν) =n∏
l=1
sinh(ν − νl)
Baxter’s TQ approach
Since [T(ν),T(ν ′)] = 0 we solve for each tj(ν) independently
Baxter introduced an auxiliary function q(ν) such that
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Properties
I t(ν) and q(ν) are entire functions
I t(ν) and q(ν) are iπ-periodic functions
Suppose q(ν1) = q(ν2) · · · = q(νn) = 0. Then
q(ν) =n∏
l=1
sinh(ν − νl)
Baxter’s TQ approach
Since [T(ν),T(ν ′)] = 0 we solve for each tj(ν) independently
Baxter introduced an auxiliary function q(ν) such that
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Properties
I t(ν) and q(ν) are entire functions
I t(ν) and q(ν) are iπ-periodic functions
Suppose q(ν1) = q(ν2) · · · = q(νn) = 0. Then
q(ν) =n∏
l=1
sinh(ν − νl)
Baxter’s TQ approach
Since [T(ν),T(ν ′)] = 0 we solve for each tj(ν) independently
Baxter introduced an auxiliary function q(ν) such that
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Properties
I t(ν) and q(ν) are entire functions
I t(ν) and q(ν) are iπ-periodic functions
Suppose q(ν1) = q(ν2) · · · = q(νn) = 0. Then
q(ν) =n∏
l=1
sinh(ν − νl)
TQ relation → BAE
Set ν = νi in
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Rearranging we have
q(νi − 2iη)
q(νi + 2iη)= −sin(η + iν)N(νi , η)
sin(η − iν)N
The Bethe ansatz equations
(−1)nn∏
l=1
sinh(2iη − νi + νl)
sinh(2iη − νl + νi )= −sin(η + iν)N
sin(η − iν)N, i = 1 . . . n
TQ relation → BAE
Set ν = νi in
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Rearranging we have
q(νi − 2iη)
q(νi + 2iη)= −sin(η + iν)N(νi , η)
sin(η − iν)N
The Bethe ansatz equations
(−1)nn∏
l=1
sinh(2iη − νi + νl)
sinh(2iη − νl + νi )= −sin(η + iν)N
sin(η − iν)N, i = 1 . . . n
TQ relation → BAE
Set ν = νi in
t(ν)q(ν) = sin(η + iν)Nq(ν + 2iη) + sin(η − iν)Nq(ν − 2iη)
Rearranging we have
q(νi − 2iη)
q(νi + 2iη)= −sin(η + iν)N(νi , η)
sin(η − iν)N
The Bethe ansatz equations
(−1)nn∏
l=1
sinh(2iη − νi + νl)
sinh(2iη − νl + νi )= −sin(η + iν)N
sin(η − iν)N, i = 1 . . . n
Which solution of BAE?
The ground state eigenvalue t0(ν) has N/2 distinct, real roots
ν ν ν ν ν ν ν ν0 1 2 3 4 5 6 7
ν
6-vertex model and the XXZ model
The transfer matrix eigenvectors of the XXZ model
HXXZ = −1
2
N∑j=1
(σxj σ
xj+1 + σyj σ
yj+1 − cos 2η σzj σ
zj+1
)coincide with those of the 6-vertex model
HXXZ = −i sin 2ηd
dνlnT(ν)
∣∣∣ν=−iη
− 1
2cos 2η I⊗N .
Continuum limit of the 6-vertex model
Take N →∞, the lattice spacing d → 0 with Nd finite and scale νappropriately to find
ln t0(N) = −f N +πceff
6N+ . . .
TQ relation becomes
t(E )q(E ) = e iφq(ω2E ) + e−iφq(ω−2E )
where ω = −e−2iη and q(El) = 0 implies
∞∏l=1
(El − ω2Ei
El − ω−2Ei
)= −e−2iφ , i = 1 . . .∞
Have also sneaked in a twist φ in the periodic boundary conditions
Continuum limit of the 6-vertex model
Take N →∞, the lattice spacing d → 0 with Nd finite and scale νappropriately to find
ln t0(N) = −f N +πceff
6N+ . . .
TQ relation becomes
t(E )q(E ) = e iφq(ω2E ) + e−iφq(ω−2E )
where ω = −e−2iη and q(El) = 0 implies
∞∏l=1
(El − ω2Ei
El − ω−2Ei
)= −e−2iφ , i = 1 . . .∞
Have also sneaked in a twist φ in the periodic boundary conditions
Continuum limit of the 6-vertex model
Take N →∞, the lattice spacing d → 0 with Nd finite and scale νappropriately to find
ln t0(N) = −f N +πceff
6N+ . . .
TQ relation becomes
t(E )q(E ) = e iφq(ω2E ) + e−iφq(ω−2E )
where ω = −e−2iη and q(El) = 0 implies
∞∏l=1
(El − ω2Ei
El − ω−2Ei
)= −e−2iφ , i = 1 . . .∞
Have also sneaked in a twist φ in the periodic boundary conditions
Ordinary differential equations
We shall study several eigenproblems associated with the ODE[− d2
dx2+ x2M
]y(x ,E ) = E y(x ,E ) , M ≥ 1
Simple example: P(x) = x3 − E
Consider the WKB approximation for large-|x | when M = 3/2
ψ(x) ∼ 1
P1/4(x)exp
(±∫ x√
P(t) dt
), |x | → ∞
Thus there are two asymptotic behaviours as x →∞ :
ψ±(x) ∼ x−3/4 exp
(±2
5x
52
)
Call solution that grows for large-x as dominant and the solutionthat decays as subdominant
E belongs to the spectrum iff the subdominant solution asx → −∞ is also subdominant as x →∞
Simple example: P(x) = x3 − E
Consider the WKB approximation for large-|x | when M = 3/2
ψ(x) ∼ 1
P1/4(x)exp
(±∫ x√
P(t) dt
), |x | → ∞
Thus there are two asymptotic behaviours as x →∞ :
ψ±(x) ∼ x−3/4 exp
(±2
5x
52
)
Call solution that grows for large-x as dominant and the solutionthat decays as subdominant
E belongs to the spectrum iff the subdominant solution asx → −∞ is also subdominant as x →∞
Simple example: P(x) = x3 − E
Consider the WKB approximation for large-|x | when M = 3/2
ψ(x) ∼ 1
P1/4(x)exp
(±∫ x√
P(t) dt
), |x | → ∞
Thus there are two asymptotic behaviours as x →∞ :
ψ±(x) ∼ x−3/4 exp
(±2
5x
52
)
Call solution that grows for large-x as dominant and the solutionthat decays as subdominant
E belongs to the spectrum iff the subdominant solution asx → −∞ is also subdominant as x →∞
WKB expansion does not hold near x such that P(x) = 0 socannot simply continue x from −∞ to ∞
Instead we can take x to be complex and continue x throughcomplex values from large negative real x to large positive real x
We need to take into account the Stokes phenomenon where thedominant component of an asymptotic solution, if nonzero, canhide a ‘discontinuous’ change in the size of its subdominantcomponent as x varies past a Stokes line
Stokes phenomenon
Set x = −iρe iθ with ρ, θ ∈ R then
ψ±(x) ∼ ρ−3/4 exp
(±2
5e i
5θ2 ρ
52
)ρ→∞
The dominant and subdominant solutions can be distinguished forall θ except when
<e(e i52θ) = 0
As θ sweeps past
θ =π
5± 2πn
5, n = 0, 1, 2
previously subdominant solutions swap to be dominant and viceversa.
Stokes phenomenon
Set x = −iρe iθ with ρ, θ ∈ R then
ψ±(x) ∼ ρ−3/4 exp
(±2
5e i
5θ2 ρ
52
)ρ→∞
The dominant and subdominant solutions can be distinguished forall θ except when
<e(e i52θ) = 0
As θ sweeps past
θ =π
5± 2πn
5, n = 0, 1, 2
previously subdominant solutions swap to be dominant and viceversa.
Stokes phenomenon
Set x = −iρe iθ with ρ, θ ∈ R then
ψ±(x) ∼ ρ−3/4 exp
(±2
5e i
5θ2 ρ
52
)ρ→∞
The dominant and subdominant solutions can be distinguished forall θ except when
<e(e i52θ) = 0
As θ sweeps past
θ =π
5± 2πn
5, n = 0, 1, 2
previously subdominant solutions swap to be dominant and viceversa.
Stokes sectorsThe complex plane is thus split into five Stokes sectors
Sk =
∣∣∣∣arg(x)− 2πk
5
∣∣∣∣ < π
5
separated by the anti-Stokes lines along which both WKB solutionsoscillate
.
x
S2
S1
0S
S−1
S−2
TQ relations from ODEs
Consider [− d2
dx2+ x3
]y(x ,E ) = E y(x ,E )
The ODE has a unique solution [Sibuya (1970s)]
I y is an entire function of x and E
I For |x | → ∞ with |arg x | < 3π/5
y ∼ 1√2ix−3/4 exp
[−2
5x5/2
]
Family of solutions
There is a family of solutions
yk(x ,E , l) = ωk/2y(ω−kx , ω2kE , l) ω = exp(2πi/5)
such that
I yk exists and is an entire function of x and E
I yk is subdominant in the Stokes sector Sk , and is dominant inSk−1 and Sk+1 .
MoreoverW [yk , yk+1] := yk y
′k+1 − y ′k yk+1 = 1
implies yk , yk+1 are linearly independent solutions
Family of solutions
There is a family of solutions
yk(x ,E , l) = ωk/2y(ω−kx , ω2kE , l) ω = exp(2πi/5)
such that
I yk exists and is an entire function of x and E
I yk is subdominant in the Stokes sector Sk , and is dominant inSk−1 and Sk+1 .
MoreoverW [yk , yk+1] := yk y
′k+1 − y ′k yk+1 = 1
implies yk , yk+1 are linearly independent solutions
Family of solutions
There is a family of solutions
yk(x ,E , l) = ωk/2y(ω−kx , ω2kE , l) ω = exp(2πi/5)
such that
I yk exists and is an entire function of x and E
I yk is subdominant in the Stokes sector Sk , and is dominant inSk−1 and Sk+1 .
MoreoverW [yk , yk+1] := yk y
′k+1 − y ′k yk+1 = 1
implies yk , yk+1 are linearly independent solutions
Expand y−1 in basis of y0 ≡ y , y1:
y−1(x ,E ) = C (E ) y0(x ,E ) + C (E ) y1(x ,E )
The Stokes multipliers C (E ) and C (E ) are
and
C (E ) =W [y−1, y0]
W [y1, y0]= −W [y−1, y0]
W [y0, y1]= −1
Thus the Stokes relation is
C (E )y(x ,E ) = ω−1/2 y(ωx , ω−2E ) + ω1/2 y(ω−2x , ω2E )
Expand y−1 in basis of y0 ≡ y , y1:
y−1(x ,E ) = C (E ) y0(x ,E ) + C (E ) y1(x ,E )
The Stokes multipliers C (E ) and C (E ) are
and
C (E ) =W [y−1, y0]
W [y1, y0]= −W [y−1, y0]
W [y0, y1]= −1
Thus the Stokes relation is
C (E )y(x ,E ) = ω−1/2 y(ωx , ω−2E ) + ω1/2 y(ω−2x , ω2E )
Expand y−1 in basis of y0 ≡ y , y1:
y−1(x ,E ) = C (E ) y0(x ,E ) + C (E ) y1(x ,E )
The Stokes multipliers C (E ) and C (E ) are
and
C (E ) =W [y−1, y0]
W [y1, y0]= −W [y−1, y0]
W [y0, y1]= −1
Thus the Stokes relation is
C (E )y(x ,E ) = ω−1/2 y(ωx , ω−2E ) + ω1/2 y(ω−2x , ω2E )
TQ relations
If we x = 0 and D(E ) = y(0,E ) we have
C (E )D(E ) = ω−1/2D(ω−2E ) + ω1/2D(ω2E )
which we can compare with
t(E )q(E ) = e−iφq(ω−2E ) + e iφq(ω2E )
TQ relations
If we x = 0 and D(E ) = y(0,E ) we have
C (E )D(E ) = ω−1/2D(ω−2E ) + ω1/2D(ω2E )
which we can compare with
t(E )q(E ) = e−iφq(ω−2E ) + e iφq(ω2E )
What are C and D?
They are spectral determinants–functions that vanish at theeigenvalues of an eigenvalue problem
We see thatD(Ek) = 0 ⇐⇒ y(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Dirichlet boundary condition at x = 0
I ψ(x) ∈ L2(R+)
What are C and D?
They are spectral determinants–functions that vanish at theeigenvalues of an eigenvalue problem
We see thatD(Ek) = 0 ⇐⇒ y(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Dirichlet boundary condition at x = 0
I ψ(x) ∈ L2(R+)
What are C and D?
They are spectral determinants–functions that vanish at theeigenvalues of an eigenvalue problem
We see thatD(Ek) = 0 ⇐⇒ y(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Dirichlet boundary condition at x = 0
I ψ(x) ∈ L2(R+)
What are C and D?
They are spectral determinants–functions that vanish at theeigenvalues of an eigenvalue problem
We see thatD(Ek) = 0 ⇐⇒ y(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Dirichlet boundary condition at x = 0
I ψ(x) ∈ L2(R+)
Could also set D(E ) = y ′(0,E ) to obtain
C (E )D(E ) = ω1/2D(ω−2E ) + ω−1/2D(ω2E )
We see thatD(Ek) = 0 ⇐⇒ y ′(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Neumann boundary condition at x = 0
I ψ(x) ∈ L2(R+)
Could also set D(E ) = y ′(0,E ) to obtain
C (E )D(E ) = ω1/2D(ω−2E ) + ω−1/2D(ω2E )
We see thatD(Ek) = 0 ⇐⇒ y ′(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Neumann boundary condition at x = 0
I ψ(x) ∈ L2(R+)
Could also set D(E ) = y ′(0,E ) to obtain
C (E )D(E ) = ω1/2D(ω−2E ) + ω−1/2D(ω2E )
We see thatD(Ek) = 0 ⇐⇒ y ′(0,Ek) = 0
and by definition
y(x ,E )→ 0 as |x | → ∞ in S0
Therefore zeros of D(E ) are the eigenvalues of the ODE with
I Neumann boundary condition at x = 0
I ψ(x) ∈ L2(R+)
C is also a spectral determinant
SinceC (ek) = 0 ⇐⇒ W [y−1(x , ek), y1(x , ek)] = 0
which impliesy−1(x , ek) ∝ y1(x , ek)
This means there is a solution of the ODE with E = ek that decayssimultaneously in S−1 and S1
This is equivalent to a spectral problem in PT -symmetric quantummechanics:[
− d2
dx2+ ix3
]y(x ,E ) = −e y(x ,E ) , y ∈ L2(R)
C is also a spectral determinant
SinceC (ek) = 0 ⇐⇒ W [y−1(x , ek), y1(x , ek)] = 0
which impliesy−1(x , ek) ∝ y1(x , ek)
This means there is a solution of the ODE with E = ek that decayssimultaneously in S−1 and S1
This is equivalent to a spectral problem in PT -symmetric quantummechanics:[
− d2
dx2+ ix3
]y(x ,E ) = −e y(x ,E ) , y ∈ L2(R)
C is also a spectral determinant
SinceC (ek) = 0 ⇐⇒ W [y−1(x , ek), y1(x , ek)] = 0
which impliesy−1(x , ek) ∝ y1(x , ek)
This means there is a solution of the ODE with E = ek that decayssimultaneously in S−1 and S1
This is equivalent to a spectral problem in PT -symmetric quantummechanics:[
− d2
dx2+ ix3
]y(x ,E ) = −e y(x ,E ) , y ∈ L2(R)
Three eigenproblems
Solve Hψ = Eψ for H = p2 + x3 subject to
1. ψ(0) = 0 and ψ ∈ L2(R+)
2. ψ′(0) = 0 and ψ ∈ L2(R+)
3. ψ ∈ L2(C) where C is complex contour from S−1 to S1
.
x
S2
S1
0S
S−1
S−2
Bethe ansatz equations
Setting E = Ej where
D(E ) = D(0)∞∏i=0
(1− E
Ej
)
Then evaluating the TQ relation at E = Ej
C (Ej)D(Ej) = ω−1/2D(ω−2Ej) + ω1/2D(ω2Ej)
implies the Bethe ansatz equations
∞∏k=1
Ek − ω2Ej
Ek − ω−2Ej= −e
−2iπ5 j = 1, 2, . . .
Bethe ansatz equations
Setting E = Ej where
D(E ) = D(0)∞∏i=0
(1− E
Ej
)
Then evaluating the TQ relation at E = Ej
C (Ej)D(Ej) = ω−1/2D(ω−2Ej) + ω1/2D(ω2Ej)
implies the Bethe ansatz equations
∞∏k=1
Ek − ω2Ej
Ek − ω−2Ej= −e
−2iπ5 j = 1, 2, . . .
The generalised problem
The eigenvalues {Ej}(− d2
dx2+ x2M +
l(l + 1)
x2
)ψ(x ,E , l) = Eψ(x ,E , l) ψ ∈ L2(R+)
satisfy the Bethe ansatz equations
∞∏k=1
Ek − q2Ej
Ek − q−2Ej= −e
iπ(2l+1)M+1 j = 1, 2, . . .
Many eigenvalue problems
T1 1/2
C
T
x
IM/ODE dictionary
6-vertex model with twistφ = −(2l + 1)π/(2M+2)
Schrodinger equation withpotential x2M + l(l + 1)/x2
Spectral parameter ↔ Energy
Anisotropy ↔ Degree of potential
Twist parameter ↔ Angular momentum
Transfer matrix ↔ The Stokes multiplier C (E )
(Fused) transfermatrices
↔ Lateral spectral problems
defined at |x |=∞
q functions ↔ Radial spectral problems
linking |x |=∞ and |x |=0
Many generalisations
I Perk-Schultz model/hairpin model of boundary interaction
I Spin-j su(2) quantum chains in thermodynamic limit and theboundary parafermionic sinh-Gordon model
I vertex models with Lie algebra symmetry (simply andnon-simply laced)
I finite spin-j XXZ quantum chains at ∆ = ±1/2
I Coqblin-Schrieffer model
I Circular Brane model
I Paperclip models
I Finite spin-1/2 XYZ quantum chain
I...
Bazhanov, Dorey, Dunning, Hibberd, Khoroshkin, Lukyanov, Mangazeev, Masoero, Raimondo, Suzuki, Tateo,Tsvelik, Valeri, Vitchev, Zamolodchikov,Zamolodchikov . . .
Massive ODE/IM correspondance
Following work of Gaiotto, Moore & Neitzke, Lukyanov &Zamolodchikov established the correspondence for the quantumsine-Gordon field theory on a finite cylinder
Instead of an ODE, start with a classical integrable model, amodification of the sinh-Gordon model, and study its linear problem
∂z∂zη − e2η + p(z)p(z)e−2η = 0
where p(z) = z2M − s2M
Several of the people above + Adamopoulou, Faldella, Ito, Locke, Negro have generalised this idea
Summary