Introduction to Matrix Models and Non-Critical String Theory
Akikazu Hashimoto
Taipei, July 2005
These lectures are meant to provide pedagogical
introduction to continuum and matrix model formulation of non-critical string theory
They typically describe strings in 1+0 or 1+1 dimensions with a linear dilaton. They are important because:
• Exactly Solvable
• Context to explore non-perturbative issues (open/closed duality)
• Appears to describe a sector of critical string theory
• Have many connections to topics in mathematics
The subject is reviewed extensively in the literature
• Brezin, Wadia (Collection of Reprints)
• Di Francesco, Ginsparg, Zinn-Justin
• Dijkgraaf
• Ginsparg and Moore
• Klebanov
• Morozov
• Seiberg, Shih
Warning: “These theories are exactly solvable. That does not necessarily mean we understand them.”
Plan
1) Review of Liouville Theory and Matrix Model 2) Methods of Orthogonal Polynomials and KdV heirarchy
3) Branes and open/closed string duality
Non-critical string theory
Bosonic string in d dimensions Z =
Z Dg DX
Vol(diff)e−4πα01 R d2ξ
√ggµν∂µXi∂νXi
, i = 0...d − 1 Measure:
Z
Dge−12|δg2| = 1, |δg2| = Z
d2ξ√
g(gacgbd − 2gabgcd)δgabδgcd Z
DXe−|δX2| = 1, |δX2| = Z
d2ξ√
gδXiδXi
Action and measure are diffeomorphism invariant
Gauge fix by chosing conformal gauge g = eϕg(τ )ˆ
by introducing ghosts Z
DbD¯bDcD¯c e−Sgh Sgh = Z
d2ξ√
gbzz∇z¯cz−bz ¯¯z∇zcz¯ Under g → eσg, the action is invariant, but the measure transforms
DgXDg(gh) → DeσgXDeσg(gh) = e24πα0d−26 SL(σ,g)DgXDg(gh)
SL = Z
dξ2√
g 1
2gab∂aσ∂bσ + Rσ + µeσ
This is the conformal anomaly which is cancelled by chosing d = 26
Alternatively, make ϕ dynamical Z
Dϕe−4πα01 R d2ξ
√gˆ(25−d12 gˆab∂aϕ∂bϕ+25−d6 Rϕˆ )
Then δˆg = (ξ)ˆg, δϕ = −(ξ) is a symmetry
Rescale
q25−d
12 ϕ → ϕ Z
Dϕe−4πα01 R d2ξ
√gˆ(gˆab∂aϕ∂bϕ+Q ˆRϕ)
where
Q =
r25 − d 3 The stress tensor
T = −1
2∂ϕ∂ϕ + Q
2 ∂2ϕ
T (z)T (w) ∼ 1 2
c
(z − w)4 + ..., c = 1 + 3Q2
Finally, the µeγϕ term adjusted so that the operator has dimension
[eγϕ] = −1
2γ(γ − Q) = 1 So the bottom line
Z
Dϕe−4πα01 R d2ξ
√gˆ(gˆab∂aϕ∂bϕ+Q ˆRϕ+µeγϕ)
γ = 1
√12(√
25 − d ∓ √
1 − d) = Q
2 ± 1 2
pQ2 − 8
pick branch γ = 1
√12(√
25 − d − √
1 − d) = Q
2 − 1 2
pQ2 − 8 so that in d → −∞ limit, γ → 0, so that
Q = 2/γ − γ → 2/γ, giving rise to “classical” Liouville theory
S = Z
d2ξp ˆ g
ˆ
gab∂aϕ∂bϕ + Q ˆRϕ + µ
γ2eγϕ
which is Weyl invariant under ˆg → e2ρg, ϕ → ϕ −ˆ γ2ρ
These can be coupled to d free bosons (or CFT with central charge c) so that
cmatter + cliouv − 26 = 0
c ≤ 1 and c ≥ 25 picked out because γ is real. At c = 25 the signature of ϕ coordinate flips. We will focus on
c ≤ 1 for which ϕ coordinate is Eucledean.
• c < 1: 1 linear dilaton dimension
• c = 1: 1 linear dilaton dimension, one flat direction
linear dilaton: c.f. NS5-brane
X
φ
µeγϕ marginal tachyon condensate no X for c < 1, c = 1 is a barrier.
Other interesting example: (p, q) Minimal CFT
O11 · · · O1(p−1)
O21 O2(p−1)
... ...
O(q−1)1 · · · O(q−1)(p−1)
c = 1 − 6(p − q)2 pq
• (p, q) = (2, 3): c = 0 pure gravity
• (p, q) = (3, 4): c = 1/2 Ising Model
• (p, q) = (2, 5): c = −22/5 Yang-Lee
• Other (p, q): critical limits of various statistical models
[Or,s] = (pr − qs)2 − (p − q)2 4pq
• (3,2): 0 0
• (4,3):
1
2
1
16 0 0 161 12
• (5,2): 0 −15 −15 0
Gravitational dressing
[eαϕO] = 1 [O] = ∆0, [eαϕ] = −1
2α(α − Q), Q =
r25 − d so 3
α = 1
√12
h√
25 − d − p
1 − d + 24∆0 i α < Q
2 , Seiberg Bound
Some simple observables: String Susceptibility Γ Let
Z(A) = Z
DϕDX e−Sδ
Z
d2ξp ˆ
geγϕ − A
which for large A scales as
Z(A) ∼ A(Γ−2)χ/2−1
String suceptibility is this scaling exponent.
Simple scaling argument: Dϕ invariant under ϕ → ϕ + ρ/γ
Under this transformation Q
8π Z
d2ξp ˆ
g ˆRϕ → Q 8π
Z
d2ξp ˆ
g ˆRϕ + Q 8π
ρ
γd2ξp ˆ g ˆR so
Z[A] =
Z
DϕDXe−SeQρχ2γ δ
Z
d2ξp ˆ
geγϕ+ρ − A
=
Z
DϕDXe−SeQρχ2γ −ρδ
Z
d2ξp ˆ
geγϕ − e−ρA
= eQρχ2γ −ρZ[e−ρA] = A−2γ−1Qχ Z[1], and so Γ = 2 − Q
γ = 1 12
h
(d − 1) − p
(d − 25)(d − 1)i
Along similar lines, one can define scaling dimension ∆r,s
1 Z[A]
Z
DϕDXe−Sδ
Z
d2ξp ˆ
geγϕ − A
Z
d2ξp ˆ
geαr,sϕOr,s
For ∆0 = (pr−qs)4pq2−(p−q)2 and c = 1 − 6(p−q)pq 2, α
γ = 1 − ∆rs = p − q − |pr − qs|
2q
Physical observables are correlation functions of BRST cohomology
For L ⊗ M (p, q),
H = {Oneαnϕ}, On : matter, ghosts, Liouville αn
γ = p + q − n
2q , n > 1, 6= 0 mod q
BRST cohomology has infinite element even though number of primary matter fields were finite.
Matrix Model
Theory of random matrces: (applications in disordered systems and quantum chaos)
Z = Z
dM e−TrV (M ) M is hermitan N × N matrix. Measure
DM = Y
i
dMii Y
i<j
dMijdMij∗ invariant under M → U M U†.
Parametrizing
M = U†ΛU, dM = dU Y
i
dλi∆(λ)2 where
Vandermonde : ∆(λ) = Y
i<j
(λi − λj)
One way to see this. Let dU = idT U . Line element:
TrdM 2 = X
i
dMii2 + X
i<j
dMij∗ dMij
= Tr U†(dΛ + i[Λ, dT ])U )2
= X
i
dλ2i + X
i<j
(λi − λj)2|dTij|2 analgoue of dx2 + dy2 = dr2 + r2dΩ2.
Simplest Matrix Model: Gaussian ensumble Z =
Z
dM e−TrM2 = Z
dmi Y
i<j
(mi − mj)2e−Pi m2i
=
Z
dmie− Pi m2i+2 log(mi−mj)
Ensumble of N particles in harmonic oscillator potential with Logarighmic repulsion
Large N limit described in terms of the eigenvalue density ρ(m).
S = − Z
dm m2ρ(m)+
Z
dm dm0 log(m−m0)ρ(m)ρ(m0) Equation of motion
δS = −m2 + 2 Z
dm0 log(m − m0)ρ(m0) = 0 and differentiating with respect to m,
m = Z
dm0 1
m − m0ρ(m0)
Solution:
ρ(m) = 1 π
p2N − m2 Wigner semi-circule distribution
ρ
m
One can also consider more general potential (BIPZ) e.g. V (M ) = 1
2M2 + gM4 so that equation of motion becomes
1
2λ + 2gλ3 = Z
dλ0 1
λ − λ0ρ(λ0) solved by
ρ(λ) = 1 π
1
2 + 4ga2 + 2gλ2 N
p4a2N − λ2, 12ga4 + a2 − 1 = 0
Such generalization is interesting because the matrix action
Z
dM e−TrM 22 +gM 33!
can be represented in Feynman expansion
↔ triangulation of 2D surface
1/N expansion ↔ genus expansion g → gc ↔ continuum limit
Resolvent: useful computational tool Z =
Z
dM e−N V (M )
Change integration variable M → M + M −z1 Under this transformation
δdM = −Tr
1
M − z
Tr
1
M − z
dM δeN V (M ) = −N V 0(M )
M − z
So we arrive at an identity hTr
1
M − z
Tr
1
M − z
i + hN V 0(M )
M − z i = 0 In the large N limit, one can factorize
hTr
1
M − z
ihTr
1
M − z
i + hN V 0(M )
M − z i = 0 Schwinger-Dyson equation. A little rewriting:
hTr
1 M − z
ihTr
1 M − z
i + N V 0(z)h 1
M − zi = −hN V 0(M ) − N V 0(z)
M − z i
= N2f (z), Polynomial order p − 2
Let’s denote
R(z) = 1
N Tr 1 M − z Then
R(z)2 + R(z)V 0(z) = f (z) For the simple case of V (M ) = M 2/2,
R2 + zR(z) = c or
R(z) = −z + √
4c + z2
2 = −z + √
z2 − 4 2
• Constant c fixed by requirement that R(z) = 1/z + O(z−2)
• R(z) imaginary for −2 < z < 2.
ρ(z) = 1 2π
p4 − z2
ρ
m
For the interacting cubic theory V = M2
2 + gM3 one has
R(z) = −V 0(z) + pV 0(z)2 + 4f 2
f = cz + d, and the descriminant is quartic in z. Arrange f so that descriminiant has two real roots and one
double root.
With this ansatz, one has
R(z) = −z + 3gz2
2 + (1 + 3g(a + b) + 3gz)p(z − 2a)(z − 2b) 2
where
3g(a − b)2+ 2(a + b)(1 + 3g(a + b)) = 0, (b − a)2(1 + 6g(a + b)) = 4 so that
R(z) = 1
z + O(z−2)
This determines the eigenvalue distribution ρ(z) and it solves the equation of motion.
F(0) =
Z 2b 2a
dλρ(λ) 1
2λ2 + gλ3
− 1 2
Z
dλdλ0ρ(λ)ρ(λ0) ln(λ − λ0)
= −1 3
σ(3σ2 + 6σ + 2)
(1 + σ)(1 + 2σ)2 + 1
2 ln(1 + 2σ) where
σ = 3g(a + b) is a solution of
18g2 − σ(1 + σ)(1 + 2σ) with a bit more massaging
F(0) = − X
k
1 2
(72g2)k (k + 2)!
Γ(3k/2)
Γ(k/2 + 1) ≈
r 2
3πk7(108√
3g2)2k
Now, series
X kΓ−3 g gc
2k
= (gc − g)2−Γ The area scales like
A = g∂g∂ F(0)
F(0) ∼ 1 gc − g so
Z ∼ AΓ−2 and we can read off
Γ = −7
2 + 3 = −1 2
in agreement with Γ = 2 − Q
γ = 1 12
h
(d − 1) − p
(d − 25)(d − 1) i
for d = 0.
All this was for pure 2D gravity (c = 0). What if we are interested in adding conformal matter?
Consider multi-matrix model Tr1
2A2 + 1
2B2 + gA3 + gB3 + cAB
Ising model (c = 1/2 theory)
Can also consider Z
dt 1
2(∂tM )2 + 1
2M2 + gM3 (c = 1)
Also consider adding higher order terms in the potential V (M ) = 1
2M2 + g3M3 + g4M4 + ...
• Different discretization generically lead to same continuum limit (universality)
• Fine tuning coupling gives rise to new critical behavior (multi-criticality)
Double Scaling Limit
If one also compute 1/N corrections, F ∼ X
χ
X
n
Nχn(Γ−2)χ/2−1(g/gc)n ∼ (gc − g)(2−Γ)χ/2Nχ so scale g → gc keeping
κ = (gc − g)(2−Γ)/2N = fixed Then
F = κ2F(0) + F(1) + κ−2F(2) + ...
Method of Orthogonal Polynomials
Z = Z
dM e−TrV (M ) =
N
Y
i=1
dλ ∆(λ)2e− Pi V (λi) Because of anti-symmetry
∆(λ) = det(λj−1i ) = det
1 λ1 λ21 · · · λN −11 1 λ2 λ22 · · · λN −12
... ... ... ... ...
1 λN −1 λ2N −1 · · · λN −1N −1
Now, define a set of polynomials
Pn(λ) = λn + . . . satisfying orthogonality relation
Z ∞
−∞
e−V (λ)Pn(λ)Pm(λ) = hnδmn Then,
∆(λ) = det(λj−1i ) = det(Pj−1(λi))
and
Z = N !
N −1
Y
i=0
hi = N !hN0
N −1
Y
i=1
fkN −k, fk = hk hk−1
Finding Pn (and therefore hn) is a finite procedure, so Z can be computed exactly
If interest is in large N limit, 1
N2F = 1 N
X
1 − k N
ln fn ∼
Z 1 0
dξ(1 − ξ) ln(f (ξ)) where ξ = k/N .
Need fn for large n.
This can be gotten from studying the recursion relation
λPn(λ) =
n+1
X
i=0
ciPi(λ) but in fact
λPn(λ) = Pn+1 + rnPn−1 because
Z
λPn(λ)Pi(λ)e−V (λ) = 0 for i < n − 1.
Now,
Z
e−V (λ)(Pn(λ)λ)Pn−1(λ) = rnhn−1
= Z
e−V (λ)Pn(λ)(λPn−1) = hn so
rn = hn
hn−1 = fn
Similarly,
Z
e−V (Pn0(λ)λ)Pn(λ) =
Z
e−V (nPn(λ) + . . .)Pn(λ) = nhn
= Z
e−V Pn0(λ)(λPn(λ)) =
Z
e−V Pn0(λ)rnPn−1(λ)
= − Z
Pn(λ)(rne−V Pn−1)0
=
Z
rnPn(λ)Pn−1(λ)e−V V 0(λ)
− Z
Pn(λ)rne−V Pn−10 so
nhn = rn Z
e−V V 0(λ)Pn(λ)Pn−1(λ)
To apply these structures, consider for simplicity a potential with even terms only
V = 1 2g
λ2 + λ4
N + b λ6 N2
Then
gV 0 = λ + 2λ3
N + 3b λ5 N2 Insert this in
nhn = rn Z
e−V V 0(λ)Pn(λ)Pn−1(λ)
λ : Pn(λ) %
&
Pn+1 rnPn−1 so
gn = rn + 2
N rn(rn−1 + rn + rn+1) + 3b
n2( 10 rrr terms) In the large n limit,
ξ = n
N , r(ξ) = rn N
and
gξ = r + 6r2 + 30br3 ≡ W (r) In general, if
V (λ) = 1
2gapλ2p then
W (r) = ap(2p − 1)!
(p − 1)!2 rp
For generic W (r),
gξ = W (r) = gc + 1
2W 00(rc)(r − rc)2 + ...
Then r − rc ∼ (gc − ξg)−Γ with Γ = −1/2 so r = rc + (gc − ξg)−Γ, so that
1
N2F =
Z 1 0
dξ(1 − ξ)f (ξ)
∼
Z 1 0
dξ (1 − ξ)(gc − ξg)−Γ ∼ (gc − g)−Γ+2 Agree with Γ for pure gravity computed earlier.
Multi-criticality In general
W (r) = gc + c(r − rc)2 → Γ = −1 2 W (r) = gc + c(r − rc)3 → Γ = −1
... 3
W (r) = gc + c(r − rc)m → Γ = − 1 m Γ = 2 − Q
γmin = 2
1 − p − q, (p, q) = (2l + 1, 2)
Now, if one is interested in higher-genus contributions, gξ = W (r) + 2r(ξ)(r(ξ + ) + r(ξ − ) − 2r(ξ)) where = 1/N . Now scale
gc − ξg = a2z r − rc = au(z)
N = a−5/2 g − gc = κ−4/5a2
⇒ z = u(z)2 − 1
3u00(z), u(κ−4/5) = Z00(κ−4/5)
This is the KdV equation.
Can be solved perturbatively u = z1/2(1 − X
k
ukz−5/2k)
= z1/2
1 − 1
24z−5/2 − 49
1152z−5 − 1225
6912z−15/2 + ...
and computes the genus expansion of Z
Summary of Orthogonal Polynomials
Z = Z
dM e−TrV (M ) =
Z N Y
i=1
dλ ∆(λ)e− Pi V (λi)∆(λ)
|N i =
N −1
Y
i=0
det(Pj−1(λi)) Z = hN |S|N i
S = Snmb†nbm, Smn = δmnhn
KdV Heirarchy
Normalize Πn so that Z
dλe−V ΠnΠm = δnm and define Qnm by
λΠn = rhn+1
hn Πn + rnrhn−1
hn Πn−1
= √
rn + 1Πn+1 + √
rnΠn−1 ≡ QnmΠm Qnm = Qmn
Along similar lines, define
∂
∂λΠn = AnmΠm
which has [Q, A] by definitnion. No particular symmetry 0 =
Z
dλ ∂
∂λΠnΠme−V = (Anm + Amn − V 0)ΠnΠme−V
⇒ A + AT = V 0(Q) P = A − 1
2V 0(Q) = 1
2(A − AT) is antisymmetric [P, Q] = 1
In the double scaling limit, Qnm becomes a differential operator
Anticipate scaling
r(ξ) = rc + a2u(z) Then
Q = 2rc1/2 + a2 rc1/2
(u + rcκ2∂z2) ∼ d2 + u P = d3 + 3
4{u, d} Cubic in d
1 = [P, Q] = 3
4u2 + 1 4u00
0
⇒ KdV (2,3) model:
P = (Q3/2)+
Orthogonal Polynomials, Lax Pairs, etc generlizes Multi-matrix model
Z =
Z n Y
a=1
dMa exp [−TrVa(Ma) + caMaMa+1]
...
are also solvable.
Key identity: Itzykson-Zuber integral Z
dAeTrV (A)+cAB
= Z
dai∆(a)
∆(b)e− Pi V (ai)+caibi then
Z =
Z n Y
a=1
dMa exp [−TrVa(Ma) + caMaMa+1]
=
Z n Y
a=1
dλa∆(λ1)e−S(λa)∆(λn)
Define biorthogonal polynomials Z n
Y
a=1
Πi(λ1) ˜Πj(λn)e−Va(λa)+caλaλa+1 = δnm from which one derives
Q = dq + {vq−2(z), dq−2} + {vq−4(z), dq−4} + . . . v0(z) and adjust V ’s such that
P = (Q)p/q+
P = (Qp/q)+, [P, Q] = 1 defines differential equation for u(z)
One can also turn on “coupling” tn P → P + 1
q
X
n
ntn(Qn/q−1)+ Generalized KdV flow equation
∂
∂tnQ = [(Qn/q)+, Q]
Solve for u(z, ti) = F00(z, ti)
τ = Z = e−F
is called the τ -function: compute correlators
Expectation value of generic single trace operator Trf (M )
computes insertion of integrated lowest dimension operator. Fine tune for higher dimension operators
∂
∂tnQ = [(Qn/q)+, Q]
αn ∼ p + q − n q
γ ∼ αp−1,q−1 = 2
⇒ αn
γ = p + q − n 2q
compare with BRST cohomology
Alternative Matrix formulation of KdV flow eF (Λ) =
R dM exp h
−Tr12ΛM2 + iM63 i R dM exp −Tr12ΛM2
define
ti(Λ) = −(2i − 1)!!TrΛ−2i−1 Expand in small ti
ln τ = F = t30
6 + t1
24 + t30t1
6 + 1
24t0t2 + t21
48 + ...
So we have
Double Scaled Matrix Model (gauge theory) l
Non-critical string theory (gravity theory) l
Kontsevich Matrix Model (gauge theory)
Can they be thought of as analogues of AdS/CFT in any way?
Think about D-branes
• Matrix point of view
1
M TrΦM
In double scaling limit 1
M ΦM = 1
M (2rc + a2
√rcQ)M scale
M = 2rc` a2
Then 1
M ΦM = 1
`e`Q ⇐ 1
LTreLΦ Laplace transform
Z
dL e−xL 1
LTreLΦ = Tr log(x − Φ)
Differentiate wrt x
R(x) = Tr 1 x − Φ is the resolvent
Interpret as insertion of boundary cosmologial constant Z
d2ξp ˆ g
ˆ
gab∂aϕ∂bϕ + Q ˆRϕ + µeγϕ
+
Z
∂Σ
Kϕ+µBeγϕ/2 These are branes considered by FZZ/T
What does the resolvent measure?
R(x) = Tr 1
Φ − x = X
i
1 λi − x
force due to log interaction with other Eigenvalues y ≡ V 0(x) + 2R(x) = Effective Force
Z
ydx = Effective Potential R(x) = −V 0(x) + pV 0(x)2 + 4f
2
For example for cubic potential theory
y2 = (1 double root and pair of single root)
In the continuum limit g2 → 1/108√
3 the double root approach the cut
....
In this limit, one obtains
T2(y) = 2y2 − 1 = x(4x2 − 3) = T3(x)
for the (2,3) model. Along limilar lines, (2, 2l + 1) model gives rise to a cut and l stationary points
T2(y) = T2l+1(x)
....
One arrives at a following global picture of 1-point function of FZZT brane
x x
CFT side: one has the Liouville Boundary State
|µBi = Γ(1 + 2iP b)Γ(1 + 2iP/b) cos(2πσP )
21/4(−2iπP ) µ−iP b|P i µB
√µ = cosh πbσ, b2 = q p These branes are semi-localized
Ψ(ϕ) = hϕ|µBi = e−µBebϕ
µB φ=−(1/b)log( )
CFT and Matrix Model agree e.g. annulus Z =
Z
dτ ZghostZLiovilleZmatter ZLiouville(τ ) = hl1|e−τ (L0+ ¯L0)|l2i
Z(l1, l2) ∼ X
k=1
k sin(πk/q)Kk
q(l1)Ik
q(l2) Small l2 limit: loops = P
BRST cohomology
FZZT expectation value probe target space (as function of µB)
=−(1/b)log( )
....
x
µB φ
This picture is strictly perturbative
Nice geometrical picture (Seiberg-Shih)
• tachyon backgrounds deforming the Reimann sufrace preserving the singularity
• adding ZZ-brane opens the root into a cut
Ignores non-perturbative effect such as tunneling of eigenvalues (ZZ-branes)
Non-perturbatively
W (x) = log(Φ − x) heW (x)i = det(Φ − x) Equivalent to adding fundamental matter
Z
d ¯χdχeχ(Φ−x)χ¯
To be concrete, pick a simple model: Gaussian potential heW (x)i =
Z
dΦd ¯χdχ e−2g1 Φ2+ ¯χ(x−Φ)χ
heW (x)i =
Z
dΦd ¯χdχ e−2g1 Φ2+ ¯χ(x−Φ)χ
=
Z
d ¯χdχex ¯χχ−g2( ¯χχ)2
= 1
2πg Z
d ¯χdχds ex ¯χχ−2g1 s2+is ¯χχ
= 1
2πg Z
ds (x + is)Ne−2g1 s2
= g 2
N/2
HN(x/p
2g)
Hermite polynomial is an orthogonal polynomial for Gaussian measure
det(Φ − x) = detij(λi−1j )
∆(λ) = detij(Pi−1(λj))
∆(λ) where i = 1..(N + 1), λN = x So
hdet(Φ − x)i =
Z Y
dλ det
ij (λi−1j )∆(λ)e−2g1 λ2 = PN(x) FZZT is probing the wavefunction of fermion at the top of fermi-surface
Recursion relation
λPn(λ) = √
rn+1Pn+1(λ) + √
rnPn−1 asymptotes to
Qψ(z, λ) = ∂2
∂z2 − z
ψ(z, λ) = λψ(z, λ) in the double scaling limit.
Baker-Akheizer function
Go back to 1
2πg Z
ds (x + is)Ne−2g1 s2 = 1 2πg
Z
ds e−2g1 s2+N log(x+is)
and scale
g = 3, N = −3, s = i + ˜s, x = 2 + 2x˜ 1
2π Z
d˜s e−i
˜ s3
3 +˜s˜x
= Ai(˜x)
• This is the famous Airy function
• This is the famous Kontsevich 1 × 1 matrix model
Multi-FZZT amplitude generalizes this to the matrix Airy integral
h
n
Y
a=1
det(Φ − xa)i = Z
dSe−iTr
S3
3 +SX
Gaussian matrix model corresponds to (p, q) = (2, 1).
No conformal content: topological gravity. c = −2
• Gaiotto and Rastelli: Open SFT of FZZT in
topological (2,1) theory is the Kontsevich matrix integral Rank n of OSFT is precisely the number of FZZT branes (not N )
• Kontsevich: This integral computes topological closed string amplitudes
This is AdS/CFT correspondence
OSFT is simple because the theory was topological (much like the duality of Gopakumar-Vafa)
• Airy function: Non-perturbative FZZT amplitude
• The function is entire
• Multi-sheeted structure of FZZT moduli-space is lost at the non-perturbative level
How did this happen?
Stoke’s phenomenon
Ai(˜x) = 1 2π
Z
d˜s e−i
˜ s3
3 +˜s˜x
solution of
∂2
∂z2 − z
f (z) = 0
Two solutions: Ai(z) and Bi(z). Different s contour:
different linear combination of homogeneous solution.
Pick the solution which gives rise to Ai(z) (decay for positive real z)
Airy integral has three saddle points
Steepest descent contour hits only one of the saddles
0pi/12 7pi/12 9pi/12 11pi/12 12pi/12
3
1 1
0
4 2
3 3 2
4 0 1
2
3 4
0 2
1 0
4 0
4 3
2 1
But as z is taken off axis, different saddle points appear and disappear (along the steepest descent contour)
The locus on parameter space where contributing saddles re-arrange themselves is called “stoke’s line”
The branch cut is lost behind the Stoke’s line
Matrix model is powerful enough to address these non-perturbative issues
• Perturbatively, many possible vacua
• non-perturbative FZZT calculation is blind to this, except
• # of stationary point must be even for wave function to decay properly
• (2, 2l + 1) model is well defined non-perturbatively only for l even.
What are the analogues of all these ideas for c = 1 or ˆ
c = 1.
What are the analogues of all these ideas for (p, q)
Open SFT
S = Z
Ψ ∗ QΨ + Ψ ∗ Ψ ∗ Ψ
+
Chern-Simions
Z
AdA + 2
3A3
s
back