T-invariant Decomposition and the Sign Problem in Quantum Monte Carlo Simulations Congjun Wu Kavli Institute for Theoretical Physics, UCSB Reference: Phys. Rev. B 71, 155115(2005); Phys. Rev. B 70, 220505(R) (2004); Phys. Rev. Lett. 91, 186402 (2003). 1 Collaborators • S. Capponi, Université Paul Sabatier, Toulouse, France. • S. C. Zhang, Stanford. Many thanks to D. Ceperley, D. Scalapino, J. Zaanen for helpful discussions. 2 Overview of numeric methods • Quantum many-body problems are hard to solve analytically because Hilbert spaces grow exponentially with sample size. No systematic, non-perturbative methods are available at high dimensions. • Exact diagonalization: up to very small sample size. • Density matrix renormalization group: restricted one dimensional systems. • Quantum Monte-Carlo (QMC) is the only scalable method with sufficient accuracy at D 2 . 3 Outline • A sufficient condition for the absence of the sign problem. • The conclusive demonstration of a 2D staggered ground state current phase in a bilayer model. • Physics of the staggered current state. • Applications in spin 3/2 Hubbard model. 4 Classical Monte-Carlo: Ising model • Probability distribution: H J iz w({ i }) exp[ H ({ i })] / Z z j ij • Observables: magnetization and susceptibility. M 1 N w{ }( i ) { } i 1 N2 w{ }( i ) { } 2 i • Metropolis sampling: 1. Start from a configuration {s} with probability w({s}). Get a trial configuration by flipping a spin. 2. Calculate acceptance ratio: r w({ trial}) / w({ }) . 3. If r>1, accept it; If r<1, accept it wit the probability of r. 5 Fermionic systems • Strongly correlated fermionic systems: electrons in solids, cold atoms, nuclear physics, lattice gauge theory, QCD. In particular, high Tc superconductivity: 2D Hubbard model in a square lattice. H t {ci c j h.c} nˆi U (nˆi 12 )( nˆi 12 ) ij i i • How to sample fermionic fields, which satisfy the anti-commutation relation? [c j , ci ] ij 6 Auxiliary Field QMC Blankenbecler, Scalapino, and Sugar. PRD 24, 2278 (1981) • Using path integral formalism, fermions are represented as Grassmann variables. • Transform Grassmann variables into probability. Probability: positive number Auxiliary field QMC Fermions: Grassmann number • Decouple interaction terms using Hubbard-Stratonovich (H-S) bosonic fields. • Integrate out fermions and the resulting fermion functional determinants work as statistical weights. 7 The Negative U Hubbard model(I) H t {ci c j h.c} nˆi | U | (nˆi 12 )( nˆi 12 ) ij i i • H-S decoupling in the density channel: 4-fermion interaction quadratic terms. H I | U | (ci ci ci ci )ni i • H-S decoupling becomes exact by integrating over fluctuations. Z Dc Dc exp d ci 0 DnDc Dc exp d { 0 i ci H K H I |U | (ni ( ) 1) 2 ci 2 } ci H K H I ( ) 8 The Negative U Hubbard model(II) • Integrating out fermions: det(I+B) as statistical weight. Z dn exp{ |U | 2 2 d ( n ( ) 1 ) det( I B)} i 0 i • B is the imaginary time evolution operator. B exp{ d H K H I ( )} 0 H I ( ) | U | ci, ( )ci , ( )n(i, ) i • Factorization of det(I+B): no sign problem. det( I B) det( I B ) det( I B ) | det( I B ) |2 0 9 The Positive U Hubbard model H t {ci c j h.c} nˆi U (nˆi 12 )( nˆi 12 ) ij i i • H-S decoupling in the spin channel. Z dn exp{ U2 d szi2 ( ) det( I B ) det( I B )} 0 i B exp{ d H K H I ( )} 0 H I ( ) U c i ( )c i c i ( )c i ( ) szi ( ) i • Half-filling in a bipartite lattice (=0). Particle-hole i transformation to spin down electron ci () ci . det( I B ) const det( I B ) no sign problem. 10 Antiferromagnetic Long Range Order at Half-filling AF structure factor S(p,p) as a function of =1/T for various lattice sizes. (White, Scalapino, et al, PRB 40, 506 (1989). 11 Pairing correlation at 1/8 filling small size results:4*4 lattice Pairing susceptibility in various channels. Solid symbols are full pairing correlations. Open symbols are RPA results. (White, Scalapino, et al, PRB 39, 839 (1989). 12 The sign (phase) problem!!! • Generally, the fermion functional determinants are not positive definite. Sampling with the absolute value of fermion functional determinants. O sign O / sign • Huge cancellation in the average of signs. • Statistical errors scale exponentially with the inverse of temperatures and the size of samples. • Finite size scaling and low temperature physics inaccessible. 13 A general criterion: symmetry principle • Need a general criterion independent of factorizibility of fermion determinants. The T (time-reversal) invariant decomposition. • Applicable in a wide class of multi-band and high models at any doping level and lattice geometry. The bi-layer spin ½ models : staggered current phase Reference: CW and S. C. Zhang cond-mat/0407272, to appear in Phys. Rev. B; C. Capponi, CW, and S. C. Zhang, Phys. Rev. B 70, 220505(R) (2004). 14 Digression: Time reversal symmetry • Kramers’ degeneracy in fermionic systems. T 2 1, T H T 1 H |f>, T|f> are degenerate Kramer doublets <f|T|f>=0. • Effects in condensed matter physics: Anderson theorem for superconductivity; Weak localization in disordered systems etc. 15 T-invariant decomposition CW and S. C. Zhang, to appear in PRB, cond-mat/0407272; E. Koonin et. al., Phys. Rep. 278 1, (1997) • Theorem: If there exists an anti-unitary transformation T T 2 1, TH K T 1 H K , TH I T 1 H I for any H-S field configuration, then det( I B) 0 Generalized Kramer’s degeneracy • I+B may not be Hermitian, and even not be diagonalizable. • Eigenvalues of I+B appear in complex conjugate pairs (l,l*. • If l is real, then it is doubly degenerate. det( I B) (l1 l1* )(l2 l*2 ) (ln l*n ) 0 • T may not be the physical time reversal operator. 16 Distribution of eigenvalues 17 The sign problem in spin 1/2 Hubbard model TnT 1 1 n, TST S • U<0: H-S decoupling in the density channel. T-invariant decomposition absence of the sign problem • U>0: H-S decoupling in the spin channel. Generally speaking, the sign problem appears. • The factorizibility of fermion determinants is not required. Validity at any doping level and lattice geometry. Application in multi-band, high spin models. 18 Outline • A sufficient condition for the absence of the sign problem. • The conclusive demonstration of a 2D staggered ground state current phase in a bilayer model. • Physics of the staggered current state. • Application in spin 3/2 Hubbard model. 19 The ground state staggered current phase • D-density wave: mechanism of the pseudogap in high Tc superconductivity? Chakravarty, et. al., PRB 63, 94503 (2000); Affleck and Marston, PRB 37, 3774 (1988); Lee and Wen, PRL 76, 503 (1996); • Staggered current phase in two-leg ladder systems. Bosonization+renormalization group: Lin, Balents and Fisher, PRB 58, (1998); Fjarestad and Marston, PRB 65, (2002); CW, Liu and Fradkin, PRB 68, (2003). Numerical method: Density matrix renormalization group: Marston et. al., PRL 89, 56404, (2002); U. Schollwöck et al., PRL 90, 186401, (2003). 20 Application: staggered current phase in a bilayer model • Conclusive results: Fermionic QMC simulations without the sign problem. • 2D staggered currents pattern: alternating sources and drains; curl free v.s. source free • T=Time reversal operation *flipping two layers top view d-density wave S. Capponi, C. Wu and S. C. Zhang, PRB 70, 220505 (R) (2004). 21 The bi-layer Scalapino-Zhang-Hanke Model D. Scalapino, S. C. Zhang, and W. Hanke, PRB 58, 443 (1998) c U t // t J V d H t // {ci c j d i d j h.c} t {ci d j h.c} n(i) ij i i J Sic Sid U (ni ,,c 12 )( ni ,,c 12 ) (c d ) V (ni ,c 1)(ni ,d 1) ij i i • U, V, J are interactions within the rung. • No inter-rung interaction. 22 T-invariant decoupling (Time-reversal*flip two layers) • T-invariant operators: total density, total density; bond AF, bond current. c nbond (i) 12 (di ci ci di ) n(i) (ci ci d i d i ) 1 nAF (i) 2 (ci ci di di ) ncurt (i) 2i (ci di di ci ) d • When g, g’, gc>0, T-invariant H-S decoupling absence of the sign problem. H SZH t // {ci c j d i d j h.c} t nbond (i) n(i) ij i 2 2 2 g {nbond (i) ncurt (i )} g n AF (i) g c (n(i) 2) 2 i i i i 3 J 3 4. g U V J , 4 g U V , 4 g c U 3V J 4 4 4 23 Fermionic auxiliary field QMC results at T=0K • The equal time staggered current-current correlations 1 J (r ) 2 ncurr (ri )ncurr (ri r ) L i iQr J (Q) e J (r ) Q (p , p ) r • Finite scaling of J(Q)/L2 v.s. 1/L. • True long range order: Ising-like order t // 1, t 0.1, U 0, V 0.5, J 2.0 S. Capponi, CW and S. C. Zhang, PRB 70, 220505 (R) (2004). 24 Outline • A sufficient condition for the absence of the sign problem. • The conclusive demonstration of a 2D staggered ground state current phase in a bilayer model. • Physics of the staggered current state. • Application in spin 3/2 Hubbard model. 25 Strong coupling analysis at half-filling • The largest energy scale J>>U,V. • Project out the three rung triplet states. + • Low energy singlet Hilbert space: doubly occupied states, rung singlet state. = E: U - V 34 J U 26 Pseudospin SU(2) algebra • The pseudospin SU(2) algebra v.s. the spin SU(2) algebra. c d ncurt (i) 2i (ci di di ci ) rung current nbd (i) 12 (di ci ci di ) bond strength Q(i) 12 (ci ci di di ) cdw • Pseudospin-1 representation. Q : 1; 0, • Rung current states 1 | up , down a ; E: U b V 34 J c U 1 2 i 2 27 Pseudospin-1 AF Heisenberg Hamiltonian • t// induces pseudospin exchange. t // t // H ex J pseudo {ncurt (i ) ncurt ( j ) ij nbd (i )nbd ( j ) Q (i )Q ( j )} • Anisotropic terms break SU(2) down to Z2 . c t d E t : uniform external field E : on - site anisotropy 1 H 2t nbd (i) E (Q 2 (i) ) 2 i E U (V 34 J ) 28 Competing phases • Neel order phases and rung singlet phases. staggered current CDW staggered bond order rung singlet 29 Competing phases • 2D spin-1 AF Heisenberg model has long range Neel order. t E 0 favors the easy plane of staggered current and CDW. favors the easy plane of staggered current and bond order. SU(2)Z2 the easy axis of the staggered current • Subtle conditions for the staggered current phase. t is too large polarized pseudospin along rung bond strength E is too large rung singlet state t 2z J p ( z 4), 0 E zJ p 30 Fermionic auxiliary field QMC results at T=0K • The equal time staggered current-current correlations 1 J (r ) 2 ncurr (ri )ncurr (ri r ) L i iQr J (Q) e J (r ) Q (p , p ) r • Finite scaling of J(Q)/L2 v.s. 1/L. • True long range order: Ising-like order t // 1, t 0.1, U 0, V 0.5, J 2.0 S. Capponi, CW and S. C. Zhang, PRB 70, 220505 (R) (2004). 31 Disappearance of the staggered current phase i) increase t ii) increase E U (V 34 J ) iii) increase doping 32 Outline • A sufficient condition for the absence of the sign problem. • The conclusive demonstration of a 2D staggered ground state current phase in a bilayer model. • Physics of the staggered current state. • Application in spin 3/2 Hubbard model. 33 The spin 3/2 Hubbard model H t{c c i, j , 3 1 ij , , 2 2 h.c.} ci, ci , i U 0 P00 (i ) P00 (i ) U 2 i P ( r ) P ( r ) 2m 2m i , m 2 , 1, 0 • The generic Hamiltonian with spin SU(2) symmetry. • F=0 (singlet), 2(quintet); m=-F,-F+1,…F. P (r ) 32 32 ; Fm | Fm 3 3 2 2 ; (r ) (r ) • Optical lattices with ultra-old atoms such as 132Cs, 9Be, 135Ba, 137Ba. 34 T-invariant decoupling in spin 3/2 model • T-invariant operators: density and spin nematics operators. a n(i) ci, ci , , na (i) ci, ci , • Five spin-nematics matrices = Dirac G matrices: a ija{Si S j S j Si } (a 1 5) (i, j 1 3) { a , b } 2 ab • Explicit SO(5) symmetric form: Wu, Hu and Zhang, PRL91, 186402 (2003). V, W>0 absence of the sign problem. H t { c c h . c .} c i, j , i, ci, ij , V 3U 0 5U 2 , 16 i W U2 U0 4 2 2 { V ( n ( i ) 2 ) Wn a (i )} i ,1 a 5 35 Application in spin 3/2 system 36 Summary • The “time-reversal” invariant decomposition criterion for the absence of the sign problem. • Applications: The bilayer spin 1/2 modelstaggered current phase. • Other applications: High spin Hubbard model; Model with bond interactions: staggered spin flux phase . 37