Monte Carlo studies of the spontaneous rotational symmetry breaking in dimensionally reduced super Yang-Mills models

Transcription

1 Monte Carlo studies of the spontaneous rotational symmetry breaking in dimensionally reduced super Yang-Mills models K. N. Anagnostopoulos 1 T. Azuma 2 J. Nishimura 3 1 Department of Physics, National Technical University of Athens 2 Institute for Fundamental Sciences, Setsunan University 3 KEK Theory Center and SOKENDAI, Tsukuba Crete 2013: Heraclion, June 14, 2013

2 Outline

3 Outline

4 IIB Matrix Model: Overview A non-perturbative definition of string theory in the large N limit A theory with only one scale, possibility to dynamically choose a unique vacuum Dynamical emergence of space time and matter content Dynamical compactification of extra dimensions Tackle cosmological questions, like expansion of dimensional space time, resolution of cosmic singularity

5 The IKKT or IIB Matrix Model [Ishibashi,Kawai,Kitazawa,Tsuchiya hep-th/ ] Z = da dψ e is S = 1 4g 2 tr ([A µ, A ν ][A µ, A ν ]) 1 2g }{{} 2 tr (Ψ α(cγ µ ) αβ [A µ, Ψ β ]). }{{} =S B =S F A µ (µ = 0,..., 9), Ψ α (α = 1,..., 16) (10D Majorana-Weyl spinor), (A µ ) ij, (Ψ α ) ij, i, j = 1,..., N hermitian matrices. manifest SO(9,1) symmetry and SU(N) gauge invariance N = 2 Supersymmetry

6 Relation to String Theory Matrix regularization of IIB string action in the large N limit: S Schild = d 2 σ g ( 1 4 {X µ(σ), X ν (σ)} ) 2 Ψ(σ)CΓµ {X µ (σ), Ψ(σ)} X µ (σ) (A µ ) ij {, } ı [, ] Ψ α (σ) (Ψ α ) ij d 2 σ g tr non-commutative world sheet block structure in matrices second quantized string theory reproduce interaction between D-branes at one loop level loop equation for Wilson loops light cone IIB string field theory: w(c) = trp exp [ ı C kµ A µ ] Ψ [k( )] [Fukuma, Kawai, Kitazawa, Tsuchiya ( 97)]

7 N = 2 Supersymmetry { δ (1) A µ = iɛ 1 CΓ µ Ψ and bosonic symmetry δ (1) Ψ = i 2 Γµν [A µ, A ν ]ɛ 1 { δ (2) A µ = 0 δ (2) Ψ = ɛ 2 1 Generators: Q (1), Q (2), P µ resp. { δ (1) A µ = c µ 1 δ (1) Ψ = 0 Q (1) = Q (1) + Q (2), Q (1) = i(q (1) Q (2) ) [ɛ 1 C Q (i), ɛ 2 C Q (j) ] = 2δ ij ɛ 1 CΓ µ ɛ 2 P µ Identify as D = 10, N = 2 SUSY and P µ as translations eigenvalues of A µ D = 10 space-time coordinates [Aoki et al hep-th/ ]

8 Space Time Interpretation 6d (small) 4d (large) non-commutative space time [Iso,Kawai 99, Ambjørn,KNA,Bietenholz,Hotta,Nishimura et al 00] possibility of dynamical compactification of extra dimensions possibility of built-in mechanism that generates (3 + 1)-dim space-time

9 Emergence of (3 + 1) dimensional spacetime of Lorentzian model: no sign problem, introduce large scale cutoffs in tr(a 2 0) and tr(a 2 i )[Kim,Nishimura,Tsuchiya ] dynamical time from A 0 (SUSY crucial) expanding 3+1 universe after a critical time classical, expanding solutions at late times [Kim,Nishimura,Tsuchiya , ] local field theory as fluctuations around classical solns representing commutative space-time [ ] constructively realize chiral fermions at finite-n by imposing conditions on extra dims [ ]

10 Euclidean Model A 0 ia 10 Γ 0 iγ 10 SO(10) rotational symmetry Finite: quantum effects, despite flat directions [Krauth,Nicolai,Staudacher 98, Austing, Wheater 01] Gaussian Expansion Method Calculations [Nishimura,Sugino 02, Kawai et.al. 03,06] show that: [Nishimura,Okubo,Sugino ] d = 3 configurations have lowest free energy extent of the shrunken dimensions r is independent of d the extent of the large dimensions R depends on d so that the 10 dimensional volume is a finite constant and independent of d: R d r 10 d = l 10 the ratio R/r remains finite in the large N limit Dynamical compactification by SSB of SO(10) SO(3)

11 6 dimensional Euclidean model Need a simpler model to study the above results using Monte Carlo simulations D = 4 studied before, no SSB of SO(4) [Ambjørn,KNA,Bietenholz,Hotta,Nishimura et al 00] D = 6 the simplest model with SSB of SO(6): Z = da dψ d ψ e Sb Sf S b = 1 4g 2 tr[a µ, A ν ] 2 S f = 1 2g 2 tr ( ψα (Γ µ ) αβ [A µ, ψ β ] ) A µ are N N, hermitian, traceless, vectors w.r.t. SO(6) ψ α, ψ α are N N, grassmannian entries, Weyl spinors w.r.t. SO(6) Similar to D = 10: SO(6) rotational symmetry, N = 2 SUSY, SU(N) symmetry

12 Dynamical Compactification Order Parameter T µν = 1 N tr (A µa ν ) Eigenvalues of T µν : λ n, n = 1,..., 6 λ 1 λ 2... λ 10 Extended d dimensions if e.g. in the large N limit λ 1 =... = λ d R 2 Shrunk (6 d) dimensions if e.g. λ d+1 =... = λ 10 r 2 SSB of SO(6) invariance SO(6) SO(d)

13 Gaussian Expansion Method (GEM) - Improved Mean Field Approximation [P.M.Stevenson 81], [Kabat,Lifschytz,Lowe 00-02], [Nishimura,Sugino 01],[Kawai et.al. 02] a systematic expansion method to study non perturbative effects introduce Gaussian S 0 [M µ, A αβ ] where M µ, A αβ parameters S = (S + S 0 ) S 0 = (S b + S f + S 0 ) S 0 S 0 [M µ, A αβ ] = M µ tr(a 2 µ) + A αβ tr( ψ α ψ β ) expand S = S 0 + ɛ S b + ɛ S f w.r.t ɛ replace M (1 ɛ)m, A (1 ɛ)a reorganize series, truncate, set ɛ = 1 look for plateaux in parameter space (M µ, A αβ ), in practice by solving F F = 0 = 0 M µ A αβ

14 GEM results[aoyama, Nishimura, Okubo (arxiv: )] parameter space is very large: simplify by considering SO(d) invariant ansätze, 2 d 5 λ 1 SO(d) =... = λ d SO(d) = (R d ) 2 compute free energy and observables at solutions in the large N limit compare free energy of ansätze, minimum free energy for the d = 3 ansatz, i.e. conclude SO(6) SO(3) The extent of the shrunken dimensions r ( compactification scale ) is independent of d The extent of the large dimensions R depends on d so that the 6 dimensional volume is a finite constant and independent of d: The ratio R d /r is finite (in units where g N = 1) R d r 6 d = l 6 r l

15 Outline

16 Monte Carlo Simulation[KNA, Azuma, Nishimura (unpublished)] Integrate out fermions first: Z = dad ψdψ e Sb Sf = da e Sb Z f [A] Z f [A] = d ψdψ e Sf = det M Monte Carlo simulations hard due to the strong complex action problem det M = det M e ıγ is Complex

17 The Algorithm : ignore the phase e ıγ Z 0 = da e S0 S 0 = S b log det M Simulate using Rational Hybrid Monte Carlo: use rational approximation Q x 1/2 a k a 0 + x + b k k=1 increased accuracy and range of x requires higher Q coefficients a k and b k computed using Remez algorithm [E. Remez 34, Clark and Kennedy 05 github.com/mikeaclark/algremez]

18 The Algorithm Define D = M M det D 1/2 = det M, then we can approximate det D 1/2 df df e SPF[F,F,A] (F α ) ij pseudofermions where { } Q S PF [F, F, A] = tr a 0 F F + a k F (D + b k ) 1 F k=1 spectrum of D determines Q rescale A, F to adjust spectrum to desired range

19 The Algorithm H = 1 2 trπ2 + tr Π Π + Seff [F, F, A] where S eff [F, F, A] = S 0 [A] + S PF [F, F, A] Π µ ij = (Π ) µ ji, Π α ij canonical momenta of ((A µ ) ij, (F α ) ij ) d Πd Π dπdfdf da e H = dfdf dae Seff τ-evolution according to eom preserve H: dπ µ dτ da µ dτ = H Π µ = Π µ, = H A µ = S eff A µ, df β dτ d Π β dτ = H Π β = Π β, = H F β = S eff F β

20 The Algorithm Discretize eom: τ f = N τ τ Discretization errors: H O( τ 2 ). To maintain detailed balance condition use a Metropolis accept/reject decision. Acceptance rate depends on H, tune parameters in order to maximize acc. ratio and minimize autocorrelation times. Main part of computational effort: terms (D + b k ) 1 F. Replace by solutions χ of (D + b k )χ k = F Use conjugate gradient method for the smallest of b k s. (O(N 3 ) ops if cleverly done) Use multimass Krylov solvers for other b k (O(Q) gain). Conjugate gradient method needs O(N 2 ) iterations to converge. (instead of O(1) in typical LQCD)

21 <λ n > n=1 n=2 n=3 n=4 n=5 n= /N In the large N limit λ 1 0 =... = λ 6 0 = l consistent with GEM result No SO(6) SSB phase fluctuations are important in inducing SSB as expected

22 Z = da e S0 e ıγ Z f [A] = det M e ıγ no ordinary Monte Carlo importance sampling possible: not a positive definite probability measure A serious and important technical problem Lattice QCD at high T/finite µ [ ] Lattice QCD with θ-vacua [ ] Real time QFT [hep-lat/ ] Electron structure calculation [PRL 71(93)1148, J.Chem.Phys 102, ,6219] Repulsive Hubbard model [PRB 41(90) 9301] Nuclear shell model [Phys.Repts. 278(97)1] Polymer theory [Phys.Repts. 336(00)167]

23 Possible approach: use the phase quenched model Z 0 = da e S0 : λ n = λ n e ıγ 0 e ıγ 0 e ıγ 0 decreases as e N2 f Z/Z 0, f > 0. Need O(e cn2 ) statistics for given accuracy goal. Overlap problem: distribution of sampled configs in Z 0 has exponentially small overlap with Z Dominant configurations determined by competition of entropy, action and phase fluctuations.

24 Factorization Method [KNA, Nishimura 01] λ n λ n = λ n 0 λ n 0 1, deviation from 1 is the effect of the phase Consider the distribution functions 6 6 ρ(x 1,..., x 6 ) = δ(x k λ k ) ρ (0) (x 1,..., x 6 ) = δ(x k λ k ) k=1 Consider the ensemble Z x1,...,x 6 = da e S0[A] 6 δ(x k λ k ) k=1 then ρ(x 1,..., x 6 ) = 1 C ρ(0) (x 1,..., x 6 ) w(x 1,..., x 6 ) where w(x 1,..., x 6 ) = e iγ x1,...,x 6 C = e iγ 0 not needed in the calculation. k=1 0

25 Factorization Method λ n = 6 dx k x n ρ(x 1,..., x 6 ) k=1 In the large-n limit, dominating configs determined by minimum of the free energy : F(x 1,..., x 6 ) = 1 N 2 log ρ(x 1,..., x 6 ) = 1 N 2 log ρ(0) (x 1,..., x 6 ) 1 N 2 log w(x 1,..., x 6 ) + 1 N 2 log C The minimum is determined by solutions of 1 N 2 log ρ (0) (x 1,..., x 6 ) = 1 x n x n N 2 log w(x 1,..., x 6 ) for n = 1,..., 6

26 Factorization Method 1 N 2 log ρ (0) (x 1,..., x 6 ) = 1 x n x n N 2 log w(x 1,..., x 6 ) for n = 1,..., 6 each function has a well defined large-n limit dominating solution can be used as an estimator of λ n no need to know ρ(x 1,..., x 6 ) everywhere to compute λ n RHS has complex action problem but scales fast with increasing N extrapolation to larger N errors do not propagate exponentially with N as with a naive large N extrapolation

27 Factorization Method key in using the method: find the right observables to constrain determine the ones that are strongly correlated with the phase expectation values of all others computed at the saddle point solution: no sign problem! [KNA,Azuma,Nishimura , ] d-dimensional configs: d = 6 det M C, d = 5 det M R, (R + dominates at large N) d = 4, 3 det M R +, d 2 det M = 0 phase is stationary w.r.t. perturbations around d < 6 configs [Nishimura, Vernizzi 00] strong evidence that λ 1,..., λ 6 found to be the only ones strongly correlated with the phase: our choice for studying their distribution functions [ ] Strong complex phase fluctuations play central role in the SSB mechanism [Nishimura, Vernizzi 00, KNA, Nishimura 01]

28 Simplifications hard to solve the saddle point equations in full 6D parameter space we study SO(d) symmetric vacua 2 d 5, compare to GEM x 1 =... = x d > 1 > x d+1 =... = x 6 we find that large evs, when sufficiently large, decorrelate from the phase omit large evs from ρ(x 1,..., x 6 ) we find that small evs to acquire the same value in the large-n limit omit smallest evs from ρ(x 1,..., x 6 ) Therefore, in order to study the SO(d) vacuum, consider only ρ(x d+1 )

29 Observables We take n = d + 1 for the SO(d) vacuum Define w n (x) = e iγ n,x w.r.t Z n,x = da e S0[A] δ(x λ n ) Define ρ (0) n (x) = δ(x λ n ) 0 Let x n be the solution to the saddle point equation 1 N 2 f n (0) (x) 1 N 2 d dx log ρ(0) n in the x < 1 region. Then we define the estimator (x) = d 1 dx N 2 log w n(x) λ n SO(d) = x n, n = d + 1

30 Observables Given x n we also use the estimators λ k SO(d) = λ k n, xn Compute free energy F SO(d) = 1 x n 1 N 2 f (0) n (x)dx 1 N 2 log w n( x n ), where n = d + 1 By computing F SO(d) for different d we can in principle determine the true vacuum

31 We simulate the system Z n,v = da e S0[A] V(λn[A]), V(z) = 1 γ (z ξ)2 2 γ large enough e V δ(x λ n ) in practice, we make sure that results are independent of γ study the distribution function ρ n,v (x) = δ(x λ n ) n,v ρ (0) n (x) exp { V (x λ n 0 )}

32 position of the peak of ρ n,v (x) solution of 0 = d dx log ρ n,v(x) = f (0) n (x) λ n 0 V (x λ n 0 ) we take the peak sharp and use we define the estimators w n (x p ) = cos Γ n,v, x p = λ n n,v f (0) n (x p ) = λ n 0 V ( λ n n,v ) = γ λ n 0 ( λ n n,v ξ). γ too small, distribution of λ n wide, large error in λ n n,v γ too large, small error in λ n n,v propagates by factor of γ to f n (0) (x p ) ( λ n n,v ξ 1/γ

33 It is possible to compute f n (0) (x), w n (x) for x suppressed by many orders of magnitude in Z 0 w n (x) hard due to the complex action problem, but scales for small enough N 1 Φ n (x) = lim N N 2 log w n(x) f n (0) (x), w n (x) computed by interpolation or fits. Fitting functions determined by simple scaling arguments for small x We find that the function f n (0) (x) scales as 1 N f n (0) (x) for x 0.4, but as 1 N f (0) 2 n (x) for smaller x. Need to subtract the O(1/N) finite size effects in the calculations.

34 Outline

35 λ n f 3 (0) (x)/n 2 - g3 (x)/(nx) N=24 N=32 -Φ 3 (x) f 4 (0) (x)/n 2 - g4 (x)/(nx) N=24 N=32 -Φ 4 (x) x x Compute the solution to 1 N f (0) 2 n (x) = Φ (x) (after subtracting finite size effects): Compare to the GEM result r 2 /l /0.627 = λ 3 SO(2) = x 3 = 0.31(1) λ 4 SO(3) = x 4 = 0.35(1)

36 λ n f 5 (0) (x)/n 2 - g5 (x)/(nx) N=24 N=32 -Φ 5 (x) f 6 (0) (x)/n 2 - g6 (x)/(nx) N=24 N=32 -Φ 6 (x) x x Compute the solution to 1 N f (0) 2 n (x) = Φ (x) (after subtracting finite size effects): Compare to the GEM result r 2 /l /0.627 = λ 5 SO(4) = x 5 = 0.34(2) λ 6 SO(5) = x 6 = 0.36(3)

37 Constant volume property L n 2 (x) x n=3 n=4 n=5 n=6 λ k SO(d), k n = d + 1, is estimated from λ k xp = λ k n,v In order to minimize the finite size effects, we compute ( 6 ) 1 6 Ln(x) 2 = λ k n,x k=1 and find that L 2 n(x) l for 0.5 < x < 1

38 Free Energy log w n (x)/n n= n=4 n=5 n= x Hard! 1 After subtracting finite size effects, we fit N f (0) 2 n (x) = p n e qnx. Attempt e.g. to substitute in F SO(d) = 1 1 x n N f (0) 2 n (x)dx 1 N log w 2 n ( x n ) for x n Still working!! TBA...

39 Outline

40 Simulation from first principles 6D version of IIB matrix model Complex action problem very strong, use factorization method successfully Computed numerically the maxima of λ n distributions and estimated λ n for SO(d) vacua Large-N and small-x scaling properties of distribution functions play important role in the calculation Short distance, non-perturbative, dynamics of eigenvalues of matrices A play crucial role in determining r are consistent with GEM prediction R d r 6 d = l 6, r , l Consistent with the GEM scenario of dynamical compactification with SSB of SO(6) SO(3) Consistent with (euclidean) spacetime having volume independent of d and R/r finite

41