Thermodynamic potentials from shifted boundary conditions

Transcription

1 L. Giusti ECT October 2011 p. 1/15 Thermodynamic potentials from shifted boundary conditions Leonardo Giusti University of Milano-Bicocca Based on: L. G. and H. B. Meyer PRL 106 (2011) , arxiv: and in preparation

2 L. Giusti ECT October 2011 p. 2/15 Outline Relation between the entropy density and the response of the system to the shift s = 1 T 2 lim 1 d 2 ln ({0,0, dz2 z}) z=0 X 0 Generalization to the specific heat Finite-size effects Extension to the lattice (and exploratory numerical study for SU(3) Yang Mills) Conclusions and outlook

3 L. Giusti ECT October 2011 p. 3/15 Momentum distribution from shifted boundary conditions The relative contribution to the partition function of states with momentum p is [T = 1/L 0 ] R(p) = Tr{e L 0Ĥ ˆP(p) } Tr{e L 0Ĥ} X 0 where ˆP (p) = 1 d 3 z e ip z e iˆp z, e iˆpz φ = φ z The momentum distribution can then be written as R(p) = 1 d 3 ip z (z) z e, φ(l 0,x) = φ(0,x + z) where (z) is the usual path integral but with shifted boundary conditions in time direction

4 L. Giusti ECT October 2011 p. 4/15 Cumulant generator As usual the generator of its cumulants is defined as e K(z) = 1 X p e ip z R(p) = e K(z) = (z) The momentum cumulants can then be written as ˆp 2n 1 1 ˆp 2n 2 2 ˆp 2n 3 3 c = ( 1)n 1 +n 2 +n 3 2n 1 z 2n 1 1 2n 2 z 2n 2 2 2n 3 z 2n 3 3» (z) ln z=0 In the continuum they equal the standard definition ˆp 2n 1 ˆp 2n 2 ˆp 2n 3 c = ( 1) n 1 +n 2 +n 3 T 01 T 03 c, T 0k (x 0 ) = d 3 x T 0k (x) and, being conn. corr. functions of the momentum charge, they are finite as they stand. The generator K(z) and the distribution R(p) are thus expected to be finite as well

5 L. Giusti ECT October 2011 p. 5/15 Momentum distribution from shifted boundaries on the lattice [Della Morte, LG 10; LG, Meyer 10] On the lattice a theory is invariant under a discrete group of translations only. It is still possible, however, to factorize the Hilbert space in sectors with definite conserved total momentum The momentum distribution is given by L a R(p) = a3 X z ip z (z) e where (z) is the usual PI but with (discrete) shifted boundary conditions. Since only physical states contribuite to it, R(p) is expected to converge to the continuum universal value without need for U renormalization The shifted boundary conditions allow us to define connected correlation functions of the momentum which do not require any U renormalization Their continuum limit satisfies the standard EMT WIs, which can be used to interpret the cumulants in terms of basic thermodynamic potentials

6 L. Giusti ECT October 2011 p. 6/15 Ward identities for two-point correlators of T 0k (I) In the continuum a judicious combination of WIs associated with translational invariance µ T µν (x) O 1... O n = nx i=1 O 1... δ x O i... O n, leads to (x 0 y 0, w k z k ) L 0 T 0k (x 0 ) T 0k (y) L k e T 0k (w k ) T 0k (z) = T 00 T kk where T 0k (x 0 ) = d 3 x T 0k (x), e T0k (x k ) = h Y dx ν i T 0k (x) ν k Note that: All operators at non-zero distance Number of EMT on the two sides different Trace component of EMT does not contribute On the lattice it can be imposed to fix the renormalization of T 0k

7 L. Giusti ECT October 2011 p. 7/15 Ward identities for two-point correlators of T 0k (II) In the thermodynamic limit the WI reads (y 0 x 0 ) L 0 T 0k (x 0 ) T 0k (y) = T 00 T kk By remembering that in the Euclidean ˆp k i T 0k, e = T 00, p = T kk = ˆp 2 k = T {e + p} = T 2 s In a finite box and for M 0 (M lightest screening mass) T 0k (x 0 ) T 0k (y) = T {e + p} + νmt2 2πL» M + 3T M e ML +... T i.e. leading finite-size effects are known functions of M, and are exponentially small in ML

8 L. Giusti ECT October 2011 p. 8/15 Entropy density from shifted boundary conditions By putting together the two formulas s = 1 T 2 lim 1 d 2 ln ({0, 0, dz2 z}) z=0 On the lattice the only difference is the discrete derivative s = 1 T 2 lim lim a 0» 2 ({0,0, n 2 za 2 ln nz a}) with n z being kept fixed when a 0 Note that: No ultraviolet renormalization Finite volume effects exponentially small Discretization effects O(a 2 ) once action improved

9 L. Giusti ECT October 2011 p. 9/15 Recursive relation for higher order cumulants There is more information in K(z). Again a judicious combination of WIs leads to T 0k (x 1 0 ) T 0k(x 2 0 )... T 0k(x 2n 0 ) c = T 00 (x 1 0 ) T kk(x 2 0 )... T 0k(x 2n 0 ) c + f.s.c. At finite temperature and volume L 0 T 00 (y 0 ) T 0k (x 1 0 )... T 0k(x 2n 0 ) c = L 0 L 0 T 0k (x 1 0)... T 0k (x 2n 0 ) c L 0 T kk (y 0 ) T 0k (x 1 0 )... T 0k(x 2n 0 ) c = n o L k + 2n T 0k (x 1 0)... T 0k (x 2n 0 ) c L k By combining these relations, and by taking the infinite volume limit T 0k (x 1 0 )... T 0k(x 2n 0 ) c = (2n 1) n 1 o T 0k (x 1 0 L 0 L )... T 0k(x 2n 2 0 ) c 0 Expression of finite-size corrections similar to the one of two-point corr. functions

10 L. Giusti ECT October 2011 p. 10/15 Specific heat Written for cumulants the recursive relation reads ˆp 2n z c = (2n 1) T 2 T n T ˆp 2n 2 z c o and analogously for mixed ones The definition of the specific heat implies c v = T T s = c v = 1 h ˆp 4 z c 3T 4 3 ˆp2 i z c T 2 and therefore c v = lim 1» 1 d 4 3T 4 dz d 2 T 2 dz 2 ln ({0, 0, z}) z=0 On the lattice the only difference are the discrete derivatives. Finite-size corrections are again known, and are exponentially suppressed

11 L. Giusti ECT October 2011 p. 11/15 Scale-invariant case If T is the only dimensionful parameter in the problem, the recursive relation implies ˆp 2n z c = c 2n T 2n+3 = c 2n = (n + 1) 4 (2n!) c 2 By using the moment-cumulant transformation, the generator of the cumulants reads K({0, 0, z}) = X n=1 n+1 ˆp2n z ( 1) c 2n! z 2n The series can be re-summed to obtain K({0,0, z}) = s 4 n 1 1 o (1 + z 2 T 2 ) 2 i.e. the entropy determines all the cumulants. The combination of scale and relativistic invariance fixes the functional form to be the one of the free case.

12 L. Giusti ECT October 2011 p. 12/15 Numerical algorithm for the cumulant generator (I) The most straightforward way for computing the cumulant generator is to rewrite it as (z) = n 1 Y i=0 (z, r i ) (z, r i+1 ) X 0 where a set of (n + 1) systems is designed so that the relevant phase spaces of successive path integrals overlap and that (z, r 0 ) = (z) and (z, r n ) = The path integrals of the interpolating systems are defined as (z, r) = DU DU 4,L0 /a 1 e S G[U,U 4,r] where U 4,L0 /a 1 is an extra (5 th ) temporal link assigned to each point of last time-slice

13 L. Giusti ECT October 2011 p. 13/15 Numerical algorithm for the cumulant generator (II) The action of the interpolating systems is S G [U, U 4, r] = S G [U] + β 3 (1 r) X n o ReTr U 0k (L 0 /a 1,x) U 4k (L 0 /a 1,x) x,k with the extra space-time plaquette given by U 4k (L 0 /a 1,x) = U 4 (L 0 /a 1,x) U k (0,x+z) U 4 (L 0/a 1,x + ˆk) U k (L 0/a 1,x) If we define the "reweighting observable as O[U, r i+1 ] = e S G[U,U 4,r i+1 ] S G [U,U 4,r i ] then (z, r i ) (z, r i+1 ) = O[U, r i+1] ri+1

14 L. Giusti ECT October 2011 p. 13/15 Numerical algorithm for the cumulant generator (II) The action of the interpolating systems is S G [U, U 4, r] = S G [U] + β 3 (1 r) X n o ReTr U 0k (L 0 /a 1,x) U 4k (L 0 /a 1,x) x,k with the extra space-time plaquette given by U 4k (L 0 /a 1,x) = U 4 (L 0 /a 1,x) U k (0,x+z) U 4 (L 0/a 1,x + ˆk) U k (L 0/a 1,x) On each lattice the entropy is finally given by s = 2 z 2 T 2 n 1 X i=0» (z, ri ) ln (z, r i+1 )

15 L. Giusti ECT October 2011 p. 14/15 Numerical results for the entropy Lat 6/g 2 0 L 0/a L/a K(z, a) 2K(z,a) z 2 T 5 7 (s/t 3 )=32π 2 /45 A (11) 5.10(3) A 1a (15) 5.089(19) s/t T=1.5 T c A (10) 4.98(4) A (8) 4.88(6) A (16) 5.12(19) A (17) 4.9(3) a 2 T 2

16 L. Giusti ECT October 2011 p. 14/15 Numerical results for the entropy Lat 6/g 2 0 L 0/a L/a K(z, a) 2K(z,a) z 2 T 5 7 (s/t 3 )=32π 2 /45 A (11) 5.10(3) A 1a (15) 5.089(19) s/t T=1.5 T c T=4.1 T c A (10) 4.98(4) A (8) 4.88(6) A (16) 5.12(19) A (17) 4.9(3) a 2 T 2 A linear extrapolation in a 2 gives s = 4.77 ± 0.08 ±?? T3 T s = 6.30 ± 0.09 ±?? T3 T = 1.5 T c = 4.1 T c Compatible with previous computations, but continuum extrapolation must be improved [Boyd et al. 96; Namekawa et al. 01] Lat 6/g 2 0 L 0/a L/a K(z, a) 2K(z,a) z 2 T 5 B (11) 6.58(3) B 1a (16) 6.684(20) B (15) 6.53(6) B (9) 6.40(6) B (17) 6.42(20) B (18) 6.1(3)

17 L. Giusti ECT October 2011 p. 14/15 Numerical results for the entropy Lat 6/g 2 0 L 0/a L/a K(z, a) 2K(z,a) z 2 T 5 7 (s/t 3 )=32π 2 /45 A (11) 5.10(3) A 1a (15) 5.089(19) s/t T=1.5 T c T=4.1 T c T=9.2 T c a 2 T 2 A (10) 4.98(4) A (8) 4.88(6) A (16) 5.12(19) A (17) 4.9(3) A linear extrapolation in a 2 gives s = 4.77 ± 0.08 ±?? T3 T s = 6.30 ± 0.09 ±?? T3 T = 1.5 T c = 4.1 T c Compatible with previous computations, but continuum extrapolation must be improved [Boyd et al. 96; Namekawa et al. 01] Lat 6/g 2 0 L 0/a L/a K(z, a) 2K(z,a) z 2 T 5 3 C (25) 7.18(3) C (4) 7.13(8) C (4) 6.94(12)

18 L. Giusti ECT October 2011 p. 15/15 Conclusions and outlook Correlation functions of total momentum fields can be related to derivatives of path integrals with shifted boundary conditions One of the applications is the computation of thermodynamic potentials, which can be connected to the cumulants via Ward identities of EMT. The entropy, for instance, is s = 1 T 2 lim 1 d 2 ln ({0, 0, dz2 z}) z=0 If lightest screening mass M 0, leading finite-size corrections exponentially small in ML On the lattice these formulas apply once the derivative is discretized and the continuum limit is taken. No additive (vac. subtraction) or multiplicative U renormalization is needed Same WIs allow for a non-perturbative renormalization of T 0k Feasibility study very promising even with a very simple-minded (expensive) algorithm