Unitary conformal bootstrap Aninda Sinha CHEP, Indian Institute of Science 1 Saturday, 23 January 16 2 Saturday, 23 January 16 2 Saturday, 23 January 16 dictionary meaning Saturday, 23 January 16 dictionary meaning Saturday, 23 January 16 General philosophy • No Lagrangian • No Feynman diagrams • No regularization, RG etc. • Only conformal symmetry and some physics inputs. Saturday, 23 January 16 Outline Contents • Epsilon expansion review • Modern bootstrap review (small sampling) • A new approach d>2, no supersymmetry Saturday, 23 January 16 • Based on 1502.01437, 1504.0072 with Apratim Kaviraj and Kallol Sen • 1510.07770 with Kallol Sen • 16xx.xxxxx with Rajesh Gopakumar, Apratim Kaviraj, Joao Penedones and Kallol Sen Saturday, 23 January 16 Epsilon expansion: Review Epsilon-expansion; Wilson;Wilson-Fisher;Polyakov;.Mack.....Rychkov, Tan O(N) model Z d 4 ✏ ⇥ i 2 x (@µ ) + ( i i 2 ) ⇤ • Wilson-Fisher fixed point. • 3d Ising model (critical point of boiling water) N = 1, ✏ = 1 • • 1 N = 1, ✏ = 2 2d Ising model c=2 XY model N = 2, ✏ = 1 7 Saturday, 23 January 16 8 Saturday, 23 January 16 • State of the art is epsilon^5 by Kleinert et al. • Needs ~135 diagrams. Possibility of mistakes. • Any way the series is asymptotic. 9 Saturday, 23 January 16 hep-th/9503230 cf: Critical properties of phi4 theories by Kleinert et al. 539 pages. 10 Saturday, 23 January 16 = d 2 2 N +2 2 N +2 3N + 14 + ✏ + 6 2 2 4(N + 8) 4(N + 8) (N + 8)2 Ising d = 2 ! 0.11 1 actual = = 0.125 8 N=1 d = 3 ! 0.519 numerics ⇡ 0.518 XY N=2 d = 3 ! 0.52, 1 3 ✏ 4 experiments ⇡ 0.521 expt ⇡ 0.506 Numerical results are from bootstrap based on methods pioneered by Rattazzi, Rychkov, Vichi and Tonni and used by El Showk et al. Often quoted as most accurate numerical estimates for the 3d Ising model at criticality. 11 Saturday, 23 January 16 2 =d Ising N +2 N +2 2 2+ ✏+ (13N + 44)✏ 3 N +8 2(N + 8) d = 2 ! 1.136 actual = 1 d = 3 ! 1.45 XY seems good but!! d = 3 ! 1.54 ↵=2 5 ! O(✏ ) ! Saturday, 23 January 16 d d numerics ⇡ 1.41 expts ⇡ 1.41 expt ⇡ 1.51 2 0.055 0.004 12 expt ⇡ 0.013 International space station superfluid He experiment discrepancy! “most precise” 1403.4545, El Showk et al 13 Saturday, 23 January 16 Numerics are not yet accurate to resolve the disagreement with experiments Kos, Poland, Simmons-Duffin,Vichi, 2015 14 Saturday, 23 January 16 Numerics are not yet accurate to resolve the disagreement with experiments Kos, Poland, Simmons-Duffin,Vichi, 2015 14 Saturday, 23 January 16 Higher spin operators: Wilson-Kogut ` ( r )=d N +2 2 2+`+ 2✏ (1 2(N + 8) 2 ✏ N =1! ( r )= (1 54 ` 6 ) `(` + 1) 6 ) `(` + 1) For large spin, we can use analytic bootstrap since the blocks are known in this limit for any dimension [Kaviraj, Sen, AS, 2015]! Find precise agreement with this! 15 Saturday, 23 January 16 Numerical bounds in fractional dimensions 1309.5089, El-Showk et al 16 Saturday, 23 January 16 Unitarity based approach • Polyakov in 1974 suggested a Lagrangian free approach to criticality based on unitarity. This approach has not been examined carefully in the literature (at all, although it keeps getting cited in modern times). • This approach gave the correct leading order (in epsilon) anomalous dimensions for certain operators. • The general equations proved too hard to solve and this program was abandoned. Saturday, 23 January 16 • In fact the bootstrap approach pioneered by Migdal, Polyakov received quite a bit of criticism from Wilson himself who overwhelmingly favoured his RG based approach which of course had several physics advantages. Saturday, 23 January 16 Wilson-Nobel lecture 1982 19 Saturday, 23 January 16 • However, the epsilon expansion approach which was quite successful faces issues. • The series is asymptotic and resummation methods are needed. • Let’s see what Polyakov has to say about the RG! Saturday, 23 January 16 Polyakov interview 2003. Source: Rychkov, 2011 21 Saturday, 23 January 16 • Clearly the question arises: Can we do better? cf classification like Minimal models for 2d CFTs. Saturday, 23 January 16 • In fact recently we have shown [Sen, AS 2015] that using Polyakov’s approach one can easily reproduce the epsilon^2 results without calculating any Feynman diagrams. • We are developing this further and hopefully the full power of this method will be elucidated soon. Saturday, 23 January 16 • Let me briefly review what can be done analytically using existing techniques. Saturday, 23 January 16 Quick review of modern bootstrap “directchannel” Saturday, 23 January 16 “crossedchannel” Quick review of modern bootstrap “directchannel” Saturday, 23 January 16 “crossedchannel” Quick review of modern bootstrap “directchannel” Saturday, 23 January 16 “crossedchannel” even spin Quick review of modern bootstrap “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` Saturday, 23 January 16 “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` Quick review of modern bootstrap Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` Saturday, 23 January 16 “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` Quick review of modern bootstrap Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` u= Saturday, 23 January 16 2 2 x12 x34 x224 x213 , v= “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` 2 2 x14 x23 x224 x213 Conformal cross ratios Quick review of modern bootstrap Crossing Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` u= Saturday, 23 January 16 2 2 x12 x34 x224 x213 , v= u$v “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` 2 2 x14 x23 x224 x213 Conformal cross ratios Quick review of modern bootstrap Crossing Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` u= Twist ⌧= Saturday, 23 January 16 ` 2 2 x12 x34 x224 x213 , v= u$v “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` 2 2 x14 x23 x224 x213 Conformal cross ratios Quick review of modern bootstrap Crossing Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` u= Twist ⌧= Saturday, 23 January 16 ` 2 2 x12 x34 x224 x213 , P⌧,` v= u$v “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` 2 2 x14 x23 x224 x213 Conformal cross ratios OPE x OPE Quick review of modern bootstrap Crossing Can only be reproduced upon considering large spin operators on the RHS “directchannel” X u 1+ P⌧,` g⌧,` (u, v) = ( ) v ⌧,` u= Twist ⌧= ` 2 2 x12 x34 x224 x213 , v= “crossedeven spin ✓channel” ◆ X 1+ P⌧,` g⌧,` (v, u) ⌧,` 2 2 x14 x23 x224 x213 Conformal cross ratios OPE x OPE P⌧,` g⌧,` (u, v) Saturday, 23 January 16 u$v Dolan, Osborn; Blocks Saturday, 23 January 16 Closed form expressions for conformal blocks are known only in even dimensions. Saturday, 23 January 16 Dolan, Osborn; Closed form expressions for conformal blocks are known only in even dimensions. However, simplifications occur in certain limits Saturday, 23 January 16 Dolan, Osborn; Fitzpatrick et al; Komargodski, Zhiboedov Closed form expressions for conformal blocks are known only in even dimensions. However, simplifications occur in certain limits ` 1 u ⌧ 1, v < 1 Saturday, 23 January 16 In the crossed channel we interchange u, v Dolan, Osborn; Fitzpatrick et al; Komargodski, Zhiboedov Closed form expressions for conformal blocks are known only in even dimensions. Dolan, Osborn; However, simplifications occur in certain limits ` 1 u ⌧ 1, v < 1 (d) g⌧,` (u, v) = u (1 “factorizes” Saturday, 23 January 16 ⌧ 2 In the crossed channel we interchange u, v ⌧ ⌧ v) 2 F1 ( + `, + `, ⌧ + 2`, 1 2 2 ` Fitzpatrick et al; Komargodski, Zhiboedov v)F (d) (⌧, u) (twist,spin,v) x (twist, u, d) Closed form expressions for conformal blocks are known only in even dimensions. Dolan, Osborn; However, simplifications occur in certain limits ` 1 u ⌧ 1, v < 1 (d) g⌧,` (u, v) = u (1 “factorizes” Saturday, 23 January 16 ⌧ 2 In the crossed channel we interchange u, v ⌧ ⌧ v) 2 F1 ( + `, + `, ⌧ + 2`, 1 2 2 ` Fitzpatrick et al; Komargodski, Zhiboedov v)F (d) (⌧, u) (twist,spin,v) x (twist, u, d) Closed form expressions for conformal blocks are known only in even dimensions. Dolan, Osborn; However, simplifications occur in certain limits ` 1 u ⌧ 1, v < 1 (d) g⌧,` (u, v) = u (1 “factorizes” Saturday, 23 January 16 ⌧ 2 In the crossed channel we interchange u, v ⌧ ⌧ v) 2 F1 ( + `, + `, ⌧ + 2`, 1 2 2 ` Fitzpatrick et al; Komargodski, Zhiboedov v)F (d) (⌧, u) (twist,spin,v) x (twist, u, d) Saturday, 23 January 16 Recursion relations for blocks in any dimension Saturday, 23 January 16 F (d) (⌧, v) = 2 (1 v) d 2 2 2 F1 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), ⌧ d + 2, v . (2.6) The general recursion relation, relating the conformal blocks for d dimensions to those in d 2 dimensions are worked out in section 5 of [1]. We write down the relation for the crossed conformal Recursion relations for blocks in any dimension blocks for convenience, ✓ z̄ z (1 z)(1 z̄) ◆2 g (d) ,` (v, u) =g 4(` 2)(d + ` 3) (d 2) g 2,` (v, u) (d + 2` 4)(d + 2` 2) 4(d 3)(d 2) ( + `)2 (d 2) g ,`+2 (v, u) d 2 2)(d 2 ) 16( + ` 1)( + ` + 1) (d + ` 4)(d + ` 3)(d + ` 2)2 (d 2) g ,` (v, u) . 4(d + 2` 4)(d + 2` 2)(d + ` 3)(d + ` 1) (2.7) (d 2) 2,` (v, u) In the limit when ` ! 1 at fixed ⌧ = relation simplifies to, ✓ 1 v v ◆2 (d) `, and for z ! 0 and z̄ = 1 (d 2) 4,`+2 (v, u) g⌧,` (v, u) = g⌧ (d 2) 2,` (v, u) g⌧ + 2 Saturday, 23 January 16 (d 16(d ⌧ v + O(z), the above 1 (d 2) g (v, u) 16 ⌧ 2,`+2 ⌧ 2)2 (d g⌧,` 3)(d ⌧ 1) (2.8) 2) (v, u) . F (d) (⌧, v) = 2 (1 v) d 2 2 2 F1 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), ⌧ d + 2, v . (2.6) The general recursion relation, relating the conformal blocks for d dimensions to those in d 2 dimensions are worked out in section 5 of [1]. We write down the relation for the crossed conformal Recursion relations for blocks in any dimension blocks for convenience, ✓ z̄ z (1 z)(1 z̄) ◆2 g (d) ,` (v, u) Crossed channel =g 4(` 2)(d + ` 3) (d 2) g 2,` (v, u) (d + 2` 4)(d + 2` 2) 4(d 3)(d 2) ( + `)2 (d 2) g ,`+2 (v, u) d 2 2)(d 2 ) 16( + ` 1)( + ` + 1) (d + ` 4)(d + ` 3)(d + ` 2)2 (d 2) g ,` (v, u) . 4(d + 2` 4)(d + 2` 2)(d + ` 3)(d + ` 1) (2.7) (d 2) 2,` (v, u) In the limit when ` ! 1 at fixed ⌧ = relation simplifies to, ✓ 1 v v ◆2 (d) `, and for z ! 0 and z̄ = 1 (d 2) 4,`+2 (v, u) g⌧,` (v, u) = g⌧ (d 2) 2,` (v, u) g⌧ + 2 Saturday, 23 January 16 (d 16(d ⌧ v + O(z), the above 1 (d 2) g (v, u) 16 ⌧ 2,`+2 ⌧ 2)2 (d g⌧,` 3)(d ⌧ 1) (2.8) 2) (v, u) . Dolan, Osborn; F (d) (⌧, v) = 2 (1 v) d 2 2 2 F1 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), ⌧ d + 2, v . (2.6) The general recursion relation, relating the conformal blocks for d dimensions to those in d 2 dimensions are worked out in section 5 of [1]. We write down the relation for the crossed conformal Recursion relations for blocks in any dimension blocks for convenience, ✓ z̄ z (1 z)(1 z̄) ◆2 g (d) ,` (v, u) Crossed channel =g 4(` 2)(d + ` 3) (d 2) g 2,` (v, u) (d + 2` 4)(d + 2` 2) 4(d 3)(d 2) ( + `)2 (d 2) g ,`+2 (v, u) d 2 2)(d 2 ) 16( + ` 1)( + ` + 1) (d + ` 4)(d + ` 3)(d + ` 2)2 (d 2) g ,` (v, u) . 4(d + 2` 4)(d + 2` 2)(d + ` 3)(d + ` 1) (2.7) (d 2) 2,` (v, u) Solution to recursion relations `, and for zin ! 0closed and z̄ = 1 form v + O(z), the above relation simplifies to, in even d. known only In the limit when ` ! 1 at fixed ⌧ = ✓ 1 v v ◆2 (d) (d 2) 4,`+2 (v, u) g⌧,` (v, u) = g⌧ (d 2) 2,` (v, u) g⌧ + 2 Saturday, 23 January 16 (d 16(d ⌧ 1 (d 2) g (v, u) 16 ⌧ 2,`+2 ⌧ 2)2 (d g⌧,` 3)(d ⌧ 1) (2.8) 2) (v, u) . Dolan, Osborn; F (d) (⌧, v) = 2 (1 v) d 2 2 2 F1 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), ⌧ d + 2, v . (2.6) The general recursion relation, relating the conformal blocks for d dimensions to those in d 2 dimensions are worked out in section 5 of [1]. We write down the relation for the crossed conformal Recursion relations for blocks in any dimension blocks for convenience, ✓ z̄ z (1 z)(1 z̄) ◆2 g (d) ,` (v, u) Crossed channel =g 4(` 2)(d + ` 3) (d 2) g 2,` (v, u) (d + 2` 4)(d + 2` 2) 4(d 3)(d 2) ( + `)2 (d 2) g ,`+2 (v, u) d 2 2)(d 2 ) 16( + ` 1)( + ` + 1) (d + ` 4)(d + ` 3)(d + ` 2)2 (d 2) g ,` (v, u) . 4(d + 2` 4)(d + 2` 2)(d + ` 3)(d + ` 1) (2.7) (d 2) 2,` (v, u) Solution to recursion relations `, and for zin ! 0closed and z̄ = 1 form v + O(z), the above relation simplifies to, in even d. known only In the limit when ` ! 1 at fixed ⌧ = ✓ 1 v ◆2 (d) (d 2) 4,`+2 (v, u) (d 2) 2,` (v, u) v In the large spin limit and u ⌧ 1, v (d + relation simplifies. 16(d ⌧ g⌧,` (v, u) = g⌧ g⌧ 2 Saturday, 23 January 16 1 (d 2) 16 <g⌧12,`+2(v, u) ⌧ 2)2 (d g⌧,` 3)(d ⌧ 1) the recursion(2.8) 2) (v, u) . Dolan, Osborn; F (d) (⌧, v) = 2 (1 v) d 2 2 2 F1 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), ⌧ d + 2, v . (2.6) The general recursion relation, relating the conformal blocks for d dimensions to those in d 2 dimensions are worked out in section 5 of [1]. We write down the relation for the crossed conformal Recursion relations for blocks in any dimension blocks for convenience, ✓ z̄ z (1 z)(1 z̄) ◆2 g (d) ,` (v, u) =g 4(` 2)(d + ` 3) (d 2) g 2,` (v, u) (d + 2` 4)(d + 2` 2) 4(d 3)(d 2) ( + `)2 (d 2) g ,`+2 (v, u) d 2 2)(d 2 ) 16( + ` 1)( + ` + 1) (d + ` 4)(d + ` 3)(d + ` 2)2 (d 2) g ,` (v, u) . 4(d + 2` 4)(d + 2` 2)(d + ` 3)(d + ` 1) (2.7) (d 2) 2,` (v, u) Crossed channel Dolan, Osborn; Solution to recursion relations in closed form relation simplifies to, in even d. known only In thethe limitansatz when `for !1 fixed ⌧ = form `, and for zconformal ! 0 and z̄ blocks = 1 v at + O(z), above by, Inserting theatfactorized of the largethe ` given at ` ✓ 1 v ◆2 (d) g⌧,` (v, u) (d) g(d⌧,`2)(v, u) = k(d2` (1 2) ⌧ 2 u)v 1 F(d(d)2)(⌧, v) , In the large spin limit and (dand ⌧ noticing 2) 1 and u ⌧ 1 into the above recursion +relation that k(v,2(`+2) (1 g u) . relation simplifies. 16(d ⌧ 3)(d ⌧ 1) (d) v = g⌧ 4,`+2 (v, u) (v, u) 2,` (v, u) u⌧ 1, v 16 <g⌧12,`+2 the recursion(2.8) g⌧ 2 (d 2) ⌧,` we arrive at the following recursion relation satisfied by the functions F 2 (1 v) F (d) (⌧, v) = 16F (d 2) (⌧ (2.9) (d 2) 4, v) 2vF 2 (⌧ (d 2, v)+ 16(d ⌧ u) = 24 k2` (1 u), (⌧, v) given in [2], ⌧ 2)2 3)(d ⌧ 1) v 2 F (d 2) (⌧, v) . (2.10) As we have explicitly checked, the solutions in (2.6) satisfy these recursion relations for general d dimensions. Saturday, 23 January 16 New results from bootstrap Saturday, 23 January 16 New results from bootstrap Gauss Hypergeometric Saturday, 23 January 16 New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 2 F1 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Saturday, 23 January 16 ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 2 F1 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order Saturday, 23 January 16 ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 2 F1 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order 1 ⇡ (function of u) ⇥ v ⌧ /2 Saturday, 23 January 16 (1 v) F (d) (⌧, v) ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 2 F1 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order 1 ⇡ (function of u) ⇥ v ⌧ /2 Saturday, 23 January 16 (1 v) F (d) (⌧, v) Needs large ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 2 F1 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order 1 ⇡ (function of u) ⇥ v ⌧ /2 To match powers of v, we must have Saturday, 23 January 16 (1 v) F (d) (⌧, v) Needs large ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 F1 2 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order 1 ⇡ (function of u) ⇥ v ⌧ /2 To match powers of v, we must have ⌧ =2 Fitzpatrick et al; Komargodski, Zhiboedov Saturday, 23 January 16 + 2n (1 v) F (d) (⌧, v) Needs large ◆ New results from bootstrap Gauss Hypergeometric F (d) (⌧, v) = 2 (1 ⌧ v) d 2 2 F1 2 ✓ 1 (⌧ 2 1 d + 2), (⌧ 2 d + 2), (⌧ d + 2), v Kaviraj, Sen, AS Bootstrap equation demands at leading order 1 ⇡ (function of u) ⇥ v ⌧ /2 To match powers of v, we must have ⌧ =2 Fitzpatrick et al; Komargodski, Zhiboedov Saturday, 23 January 16 + 2n (1 v) F (d) (⌧, v) Needs large Same as what appears in MFT. OPE’s known. ◆ • Can there be any universal results? • The answer surprisingly appears to be yes. • The anomalous dimension of large spin, large twist operators appears to be universal. ` Saturday, 23 January 16 1 General dimensions Assume minimal twist for stress tensor exchange d-2 Kaviraj, Sen, AS With some effort this can be derived analytically in all d. In terms of cT: <Stress Stress>~cT Comment: For Twist 2 Parisi-Callan-Gross theorem/Nachtmann theorem Saturday, 23 January 16 Universality at large twist Saturday, 23 January 16 Universality at large twist Plots in diverse spacetime dimensions for various conformal dimensions of the seed scalar. Asymptotes indicate same intercept independent of conformal dimension. Saturday, 23 January 16 • Do not need large N, not tied with gauge/ gravity duality. Should hold for any CFT. • Can also reproduce exactly from AdS/CFT hinting at a universal sector both in gravity and CFT. Saturday, 23 January 16 New game ala Polyakov • The algorithm for the new game is the following (see Apratim’s talk). • First we try to derive a spectral function that gives the “standard” conformal blocks. • We find that this spectral function leads to non-convergent behaviour for the 4pt fn at large complex spectral parameter. Saturday, 23 January 16 • To fix this and applying Liouville’s theorem we find that we need to add a factor that is a square of a gamma function with poles at the location of double field operators. • The resulting block is different from the usual block and leads to anomalous log and power law singularities. Saturday, 23 January 16 • Demanding that these vanish gives a set of algebraic equations. • Solving the equations in an epsilon expansion gives readily the epsilon^2 results that we obtain using standard QFT methods. • Note that our approach bypasses Feynman diagrams and usual regularization, renormalization. Saturday, 23 January 16 • It will be nice to explore these equations further and try to get a better analytic handle. • One surprise that we had was that the spectral function approach is similar to what we would do in AdS/CFT via Witten diagrams! • I expect a lot of other surprises to be in store! Saturday, 23 January 16 Saturday, 23 January 16 Saturday, 23 January 16 Unitarity approach ala Polyakov • Due to lack of time, I will explain this method first in words before showing equations. • When we compute CFT four point functions (of scalar fields for concreteness), there are two things to keep in mind. • First, we can do it using OPEs. • Second we can use the completeness of Saturday, 23 January 16 states to compute the imaginary part of the four point function and then use dispersion relations. [recall one loop diagrams become imaginary when intermediate states become onshell] • It turns out that the unitarity based method gives terms (log and regular) that are absent in the OPE (algebraic) approach. • For consistency we have to cancel these terms. • We get algebraic equations involving OPE and the conformal dimensions. NB: no cross ratios in these equations! • Can we solve these equations? Saturday, 23 January 16 • Why can this be advantageous over modern methods? • For starters modern methods do not provide analytic results that easily--only for some special sectors. No results on epsilon expansion using modern techniques. • As we will see there is a close connection with Witten diagrams. So there is a lot to explore. • May be string theory will be “inevitable.” Saturday, 23 January 16 Sen, AS conf. inv. fixes 3pt Unknowns: OPE coefficients, conformal dimensions Use dispersion relations to get full function Saturday, 23 January 16 • It turns out that cancellation of offending terms demanding only quadratic scalar exchange is sufficient to fix the anomalous dimensions of the exchange operator upto 2nd order in epsilon. [Sen, AS]* • It involves solving 2 algebraic equations which fix the OPE coefficients and the anomalous dimensions. * Saturday, 23 January 16 “the paper is very nice, congratulations on being the first people in the world who actually managed to understand Polyakov’s computation in that paper and to go further. “.....Slava Rychkov • What about other exchange operators? • What about higher orders? • Is the epsilon expansion an expansion in terms of the spin of the exchange operators? • To answer these questions we found it more convenient to use the framework of Witten diagrams and Mellin space. Saturday, 23 January 16 Mellin space bootstrap work in progress with A. Kaviraj, J. Penedones and K. Sen • We can think of CFT correlation functions as scattering amplitudes in AdS space. • Mack in 2009 taught us that it is sometimes simpler to use Mellin space. This was further explained and emphasised by Joao Penedones. • With Apratim, Joao and Kallol we have reformulated (almost) the unitarity approach in Mellin space. Saturday, 23 January 16 Correlation function in Mellin space Mack, 2009 n X ij =0 ij = ji ii = i i=1 A = h (x1 ) (x2 ) (x3 ) (x4 )i Saturday, 23 January 16 Unitary amplitudes from Witten diagrams CPW decomp. spectral fn standard blocks Polyakov ignores this piece spin-J Witten diagram Factorization necessary to reproduce OPE 1. spectral fn is known. 2. need to fix norm. for F Saturday, 23 January 16 s channel: spin 0 exchange Projector s channel Symanzik star formula Mellin variables Barnes’ first lemma x1 = 0, x2 = r, x3 = R, x4 = R + r Saturday, 23 January 16 0 Shadow piece s-channel a’s NB: no logs as of now. x1 = 0, x2 = r, x3 = R, x4 = R + r Saturday, 23 January 16 0 R2 r2 , r02 t/u channel log comes from double poles in the Gamma function The t/u channel result is independent of spin. We will get a sum over spins. So we have fixed the F’s. Saturday, 23 January 16 • It turns out that once we put in the the spectral function as in the Witten diagram CPW, s-channel leads to both regular as well as log terms arising from double poles. • So now all channels have regular and log terms which are absent in the algebraic amplitude. • DEMAND THAT THESE CANCEL Saturday, 23 January 16 ✏ 1 2 =1 + ✏ 2 108 = 2 + ✏ + 1 ✏2 fixed by conf. symm. unknown + =0 agrees with 2 loop Feynman. Can easily extend to O(n). Saturday, 23 January 16 Complete set of equations for spin-0 s-channel exchange OPE coeffs log term regular term d h= 2 Saturday, 23 January 16 Pole structure for spin-0 exchange in the s-channel Saturday, 23 January 16 • Observe that in an epsilon expansion 1/ epsilon^3 and 1/epsilon^2 terms in the equations arise only from quadratic scalar operator exchange. • This means that we will get the correct anomalous dimension for this operator upto epsilon^2 by solving a pair of simultaneous equations! Find precise agreement with Feynman diagram approach. Saturday, 23 January 16 • Higher order scalars will contribute in the equations from 1/epsilon (so will contribute to the epsilon^3 order in the anom. dim.) • Non-zero spins will contribute from epsilon^0 order and hence will contribute from epsilon^4 order in the anom. dim.) • A better understanding is in progress. • Our findings give a different perspective why the epsilon expansion was useful. • Where is string theory? Problem for 2016-. Saturday, 23 January 16 Thank you and seasons greetings! Saturday, 23 January 16