Lecture 14: Spin glasses Outline: • the EA and SK models • heuristic theory • dynamics I: using random matrices • dynamics II: using MSR Random Ising model So far we dealt with “uniform systems” Jij was the same for all pairs of neighbours. Random Ising model So far we dealt with “uniform systems” Jij was the same for all pairs of neighbours. What if every Jij is picked (independently) from some distribution? Random Ising model So far we dealt with “uniform systems” Jij was the same for all pairs of neighbours. What if every Jij is picked (independently) from some distribution? We want to know the average of physical quantities (thermodynamic functions, correlation functions, etc) over the distribution of Jij’s. Random Ising model So far we dealt with “uniform systems” Jij was the same for all pairs of neighbours. What if every Jij is picked (independently) from some distribution? We want to know the average of physical quantities (thermodynamic functions, correlation functions, etc) over the distribution of Jij’s. Today: a simple model with <Jij> = 0 Random Ising model So far we dealt with “uniform systems” Jij was the same for all pairs of neighbours. What if every Jij is picked (independently) from some distribution? We want to know the average of physical quantities (thermodynamic functions, correlation functions, etc) over the distribution of Jij’s. Today: a simple model with <Jij> = 0: spin glass Simple model (Edwards-Anderson) Nearest-neighbour model with z neighbours J2 2 Jij av 0, Jij av z (Jij J ji ) Simple model (Edwards-Anderson) Nearest-neighbour model with z neighbours J2 2 Jij av 0, Jij av z (Jij J ji ) note averages over different “samples” (1 sample = 1 realization of choices of Jij’s for all pairs (ij) indicated by [ … ]av Simple model (Edwards-Anderson) Nearest-neighbour model with z neighbours J2 2 Jij av 0, Jij av z (Jij J ji ) note averages over different “samples” (1 sample = 1 realization of choices of Jij’s for all pairs (ij) indicated by [ … ]av E Jij Si S j hi Si ij i 12 Jij Si S j hi Si ij i Simple model (Edwards-Anderson) Nearest-neighbour model with z neighbours J2 2 Jij av 0, Jij av z (Jij J ji ) note averages over different “samples” (1 sample = 1 realization of choices of Jij’s for all pairs (ij) indicated by [ … ]av E Jij Si S j hi Si ij i 12 Jij Si S j hi Si ij i non-uniform J: anticipate nonuniform magnetization mi Si Sherrington-Kirkpatrick model Every spin is a neighbour of every other one: z = (N – 1) Sherrington-Kirkpatrick model Every spin is a neighbour of every other one: z = (N – 1) J 2 ij av J2 J2 N 1 N Sherrington-Kirkpatrick model Every spin is a neighbour of every other one: z = (N – 1) J 2 ij av J2 J2 N 1 N Mean field theory is exact for this model Sherrington-Kirkpatrick model Every spin is a neighbour of every other one: z = (N – 1) J 2 ij av J2 J2 N 1 N Mean field theory is exact for this model (but it is not simple) Heuristic mean field theory replace total field on Si, Heuristic mean field theory replace total field on Si, Hi Jij S j j Heuristic mean field theory replace total field on Si, Hi Jij S j j (take hi = 0) Heuristic mean field theory replace total field on Si, Hi Jij S j j by its mean (take hi = 0) Heuristic mean field theory replace total field on Si, Hi Jij S j by its mean Jij m j j j (take hi = 0) Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: j by its mean Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: mi tanh J ij m j j by its mean j Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: mi tanh J ij m j j no preference for mi > 0 or <0: by its mean j Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: mi tanh J ij m j j no preference for mi > 0 or <0: [mij]av = 0 by its mean j Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: mi tanh J ij m j j no preference for mi > 0 or <0: [mij]av = 0 by its mean j if there are local spontaneous magnetizations mi ≠ 0, measure them by the order parameter (Edwards-Anderson) Heuristic mean field theory replace total field on Si, Hi Jij S j (take hi = 0) Jij m j and calculate mj i as the average S of a single spin in field H: mi tanh J ij m j j no preference for mi > 0 or <0: [mij]av = 0 j by its mean if there are local spontaneous magnetizations mi ≠ 0, measure them by the order parameter (Edwards-Anderson) q mi2 av self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms => Hi is Gaussian self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms => Hi is Gaussian with variance 2 J m J J m m [J 2 ] [m 2 ] ij j ij ik j k ij av j av jk j j av self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms => Hi is Gaussian with variance 2 J m J J m m [J 2 ] [m 2 ] ij j ij ik j k ij av j av jk j j av [Jij2 ]av q J 2q j self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms => Hi is Gaussian with variance 2 J m J J m m [J 2 ] [m 2 ] ij j ij ik j k ij av j av jk j j av [Jij2 ]av q J 2q j q so H 2 tanh H exp 2 2 2J q 2J q dH 2 self-consistent calculation of q: To compute q: Hi is a sum of many (seemingly) independent terms => Hi is Gaussian with variance 2 J m J J m m [J 2 ] [m 2 ] ij j ij ik j k ij av j av jk j j av [Jij2 ]av q J 2q j q so H 2 tanh H exp 2 2 2J q 2J q dH 2 (solve for q) spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH 2 spin glass transition: q expand in β: H 2 tanh H exp 2 2 2J q 2J q dH 2 spin glass transition: q expand in β: q H 2 tanh H exp 2 2 2J q 2J q dH 2 H 2 H (H) L exp 2 2 2J q 2J q dH 1 3 3 2 spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH expand in β: q 2 dH 2J q 2 H 1 3 (H) L 3 2 H 2 exp 2 2J q 2 H 2 ( H) 3 (H) L exp 2 2 2J q 2J q dH 2 4 spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH expand in β: q 2 H 2 H (H) L exp 2 2 2J q 2J q 2 H 2 ( H) 3 (H) L exp 2 2 2J q 2J q dH dH 1 3 2 2 J 2q 23 3 4 J 4 q 2 L 2 3 4 spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH expand in β: q 2 H 2 H (H) L exp 2 2 2J q 2J q 2 H 2 ( H) 3 (H) L exp 2 2 2J q 2J q dH dH 1 3 2 2 J 2q 23 3 4 J 4 q 2 L critical temperature: Tc = J 2 3 4 spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH expand in β: q 2 H 2 H (H) L exp 2 2 2J q 2J q 2 H 2 ( H) 3 (H) L exp 2 2 2J q 2J q dH dH 1 3 2 2 3 4 2 J 2q 23 3 4 J 4 q 2 L critical temperature: Tc = J below Tc: J 2 2 1q 2q2 q Tc T spin glass transition: q H 2 tanh H exp 2 2 2J q 2J q dH expand in β: q 2 H 2 H (H) L exp 2 2 2J q 2J q 2 H 2 ( H) 3 (H) L exp 2 2 2J q 2J q dH dH 1 3 2 2 3 4 2 J 2q 23 3 4 J 4 q 2 L critical temperature: Tc = J below Tc: J 2 2 1q 2q2 q Tc T This heuristic theory is right up to this point, but wrong below Tc. the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, Hi Jij S j Jij m j mi Jij2 (1 m2j ) L j j j the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, was O(1/z). Hi Jij S j Jij m j mi Jij2 (1 m2j ) L j j j the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, Hi Jij S j Jij m j mi Jij2 (1 m2j ) L j j j was O(1/z). But here, the average of the 1st term is zero and you have to keep the second order term, the mean of which is of the order of the rms value of the first term. the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, Hi Jij S j Jij m j mi Jij2 (1 m2j ) L j j j was O(1/z). But here, the average of the 1st term is zero and you have to keep the second order term, the mean of which is of the order of the rms value of the first term. Thouless-Anderson-Palmer (TAP) equations): 2 2 2 mi tanh J ij m j mi J ij (1 m j ) hi j j the trouble below Tc In the ferromagnet, it was safe to approximate Hi Jij S j Jij m j j j because the next term in a systematic expansion in β, Hi Jij S j Jij m j mi Jij2 (1 m2j ) L j j j was O(1/z). But here, the average of the 1st term is zero and you have to keep the second order term, the mean of which is of the order of the rms value of the first term. Thouless-Anderson-Palmer (TAP) equations): 2 2 2 mi tanh J ij m j mi J ij (1 m j ) hi ______________ j j Onsager correction to mean field Dynamics (I: simple way) Glauber dynamics: Dynamics (I: simple way) Glauber dynamics: 0 dP({S},t) 1 2 1 Si tanhhi (t)P(S1 L Si L SN ) dt i 12 1 Si tanhhi (t)P(S1 L Si L SN ) i Dynamics (I: simple way) Glauber dynamics: 0 dP({S},t) 1 2 1 Si tanhhi (t)P(S1 L Si L SN ) dt i 12 1 Si tanhhi (t)P(S1 L Si L SN ) i recall we derived from this d Si (t) 0 Si (t) tanh Jij S j (t) dt j Dynamics (I: simple way) Glauber dynamics: 0 dP({S},t) 1 2 1 Si tanhhi (t)P(S1 L Si L SN ) dt i 12 1 Si tanhhi (t)P(S1 L Si L SN ) i recall we derived from this d Si (t) 0 Si (t) tanh Jij S j (t) dt j mean field: Dynamics (I: simple way) Glauber dynamics: 0 dP({S},t) 1 2 1 Si tanhhi (t)P(S1 L Si L SN ) dt i 12 1 Si tanhhi (t)P(S1 L Si L SN ) i recall we derived from this d Si (t) 0 Si (t) tanh Jij S j (t) dt j mean field: 0 dmi mi tanhHi dt Dynamics I (continued) 0 dmi mi tanhHi dt Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi Jij m j 2 mi Jij2 (1 q) hi j j Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) j Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) In basis where J is diagonal: j Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) j In basis where J is diagonal: dm 0 m 1 2J 2 J h dt Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) j In basis where J is diagonal: dm 0 m 1 2J 2 J h dt m ( ) susceptibility: h ( ) 1 i 0 2 J 2 J Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) j In basis where J is diagonal: dm 0 m 1 2J 2 J h dt m ( ) susceptibility: h ( ) 1 i 0 2 J 2 J instability (transition) reached when maximum eigenvalue Dynamics I (continued) dmi mi tanhH i dt linearize (above Tc): mi H i 0 use TAP: mi J ij m j 2 mi J ij2 (1 q) hi j j mi J ij m j mi 2 J 2 hi (q 0, T Tc ) j In basis where J is diagonal: dm 0 m 1 2J 2 J h dt m ( ) susceptibility: h ( ) 1 i 0 2 J 2 J instability (transition) reached when maximum eigenvalue 1 2 J 2 (J ) max eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 local susceptibility eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 1 1 ( ) ( ) ( ) local susceptibility ii N i N d() ( ) eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 1 1 ( ) ( ) ( ) d() ( ) local susceptibility ii N i N 2J 4J 2 2 d 2J 2J 1 i 0 2J 2 eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 1 1 ( ) ( ) ( ) d() ( ) local susceptibility ii N i N 2J 4J 2 2 d 2J 2J 1 i 0 2J 2 1 2J 4J 2 x 2 2 2 use dx y y 4J 2J yx eigenvalue spectrum of a random matrix For a dense random matrix with mean square element value J2/N, the eigenvalue density is “semicircular”: 1 (J ) 4J 2 J2 2J so (J ) max 2 c 1 1 1 ( ) ( ) ( ) d() ( ) local susceptibility ii N i N 2J 4J 2 2 d 2J 2J 1 i 0 2J 2 1 2J 4J 2 x 2 2 2 use dx y y 4J 2J yx with y 1 i 0 2J 2, x J critical slowing down ( ) 12 T 2 (1 i 0 ) 1 2 2 2 T (1 i ) 1 4T 0 critical slowing down ( ) 12 T 2 (1 i 0 ) 1 2 2 2 T (1 i ) 1 4T 0 (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 2 2 2 T (1 i ) 1 4T 0 (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1( ) T(1 i ) 2 2 2 T (1 i ) 1 4T 0 (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc (J = 1) critical slowing down critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc critical slowing down but note: for the softest mode (with eigenvalue 2J) (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc critical slowing down but note: for the softest mode (with eigenvalue 2J) 1 i 0 2 J 2 2J (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc critical slowing down but note: for the softest mode (with eigenvalue 2J) 1 i 0 2 J 2 2J (1 J) 2 i 0 (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc critical slowing down but note: for the softest mode (with eigenvalue 2J) 1 i 0 2 J 2 2J (1 J) 2 i 0 so its relaxation time diverges twice as strongly: (J = 1) critical slowing down ( ) 12 T 2 (1 i 0 ) 1 small ω: 1 ( ) T(1 i ), 2 2 2 T (1 i ) 1 4T 0 1 T Tc critical slowing down but note: for the softest mode (with eigenvalue 2J) 1 i 0 2 J 2 2J (1 J) 2 i 0 so its relaxation time diverges twice as strongly: (J = 1) 1 (T Tc ) 2 Dynamics II: using MSR Use a “soft-spin” SK model: Dynamics II: using MSR Use a “soft-spin” SK model: E[] 12 r0i2 14 u0i4 12 Jiji j hii i ij i Dynamics II: using MSR Use a “soft-spin” SK model: E[] r 14 u 12 Jiji j hii 2 1 2 0 i i 4 0 i ij i J 2 ij av J2 N Dynamics II: using MSR Use a “soft-spin” SK model: E[] r 14 u 12 Jiji j hii 2 1 2 0 i i 4 0 i ij J 2 ij av i Langevin dynamics: E[] i i (t) r0i u0i3 Jij j hi i (t) t i j J2 N Dynamics II: using MSR Use a “soft-spin” SK model: E[] r 14 u 12 Jiji j hii 2 1 2 0 i 4 0 i i ij J 2 ij av i Langevin dynamics: E[] i i (t) r0i u0i3 Jij j hi i (t) t i j Generating functional: J2 N Dynamics II: using MSR Use a “soft-spin” SK model: E[] r 14 u 12 Jiji j hii 2 1 2 0 i i 4 0 i ij J 2 ij av i J2 N Langevin dynamics: E[] i i (t) r0i u0i3 Jij j hi i (t) t i j Generating functional: Z[J,h, ] DDˆ exp S[,ˆ, J,h, ] Dynamics II: using MSR Use a “soft-spin” SK model: E[] r 14 u 12 Jiji j hii 2 1 2 0 i i 4 0 i ij J 2 ij av i J2 N Langevin dynamics: E[] i i (t) r0i u0i3 Jij j hi i (t) t i j Generating functional: Z[J,h, ] DDˆ exp S[,ˆ, J,h, ] 2 3 i ˆ, J,h, ] dt T ˆ i ˆ r u J h i S[, i i 0 i 0 i ij j i i i t i j averaging over the Jij expi dt(ˆ i j ˆ )J j i ij ij J 2 ˆ (t) ˆ (t') (t) (t') ˆ (t) ( t ) (t) ˆ ( t ) exp dt dt i i j j i i j j 2N ij averaging over the Jij expi dt(ˆ i j ˆ )J j i ij ij J 2 ˆ (t) ˆ (t') (t) (t') ˆ (t) ( t ) (t) ˆ ( t ) exp dt dt i i j j i i j j 2N ij The exponent contains 1 i ˆ (t'), C(t t ) i (t) i (t'), R(t t ) i (t) i N i N i averaging over the Jij expi dt(ˆ i j ˆ )J j i ij ij J 2 ˆ (t) ˆ (t') (t) (t') ˆ (t) ( t ) (t) ˆ ( t ) exp dt dt i i j j i i j j 2N ij The exponent contains 1 i ˆ (t'), C(t t ) i (t) i (t'), R(t t ) i (t) i N i N i so replace them in the exponent expi dt(ˆ i j ˆ )J j i ij ij 2 1 ˆ ˆ ˆ exp 2 J dt dt i (t)i (t')C(t t ) ii (t) i (t')R(t t ) i decoupling sites and introduce delta functions ˆ ˆ 1 DCDC exp dtdt C(t t )NC(t t ) i (t) i (t') i ˆ ˆ 1 DRDCRexp dtdt R(t t )NR(t t ) i i (t)i (t') i decoupling sites and introduce delta functions ˆ ˆ 1 DCDC exp dtdt C(t t )NC(t t ) i (t) i (t') i ˆ ˆ 1 DRDCRexp dtdt R(t t )NR(t t ) i i (t)i (t') i We are left with W Z[0,0,J]av DC DCˆ DRDRˆ expN dt dt Cˆ (t t )C(t t ) Rˆ (t t )R(t t ) ˆ expS [, ˆ, C, Cˆ ,R, Rˆ ] expN log DD loc (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) saddle-point equations: (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) saddle-point equations: wrt Cˆ : C(t t') (t)(t ) loc (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) saddle-point equations: wrt Cˆ : C(t t') (t) (t ) loc ˆ ( t) ˆ ( t ) wrt C : Cˆ (t t') 12 J 2 loc 0 (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) saddle-point equations: wrt Cˆ : C(t t') (t) ( t ) loc ˆ ( t) ˆ ( t ) wrt C : Cˆ (t t') 12 J 2 ˆ ( t ) wrt Rˆ : R(t t') i (t) loc loc 0 (almost there) where ˆ, C, Cˆ ,R, Rˆ ] Sloc [, dt Tˆ iˆ Ý r u 2 3 0 0 ˆ (t) ˆ (t ) iR(t t ) ˆ (t) (t ) 12 J 2 dt dt C(t t ) ˆ (t ) dt dt Cˆ (t t ) (t) (t ) iRˆ (t t ) (t) saddle-point equations: wrt Cˆ : C(t t') (t) (t ) loc ˆ ( t) ˆ ( t ) wrt C : Cˆ (t t') 12 J 2 ˆ ( t ) wrt Rˆ : R(t t') i (t) loc 0 loc ˆ (t) (t ) wrt R : Rˆ (t t ) 12 iJ 2 loc 12 J 2 R( t t) effective 1-spin problem: The average correlation and response functions are equal to those of a self-consistent single-spin problem with action effective 1-spin problem: The average correlation and response functions are equal to those of a self-consistent single-spin problem with action ˆ, C, Cˆ ,R, Rˆ ] dt T ˆ 2 i ˆ Ý r u 3 Sloc [, 0 0 ˆ (t)C(t t ) ˆ (t ) 12 J 2 dt dt ˆ ( t)R(t t ) ( t ) J 2 dt dt i effective 1-spin problem: The average correlation and response functions are equal to those of a self-consistent single-spin problem with action ˆ, C, Cˆ ,R, Rˆ ] dt T ˆ 2 i ˆ Ý r u 3 Sloc [, 0 0 ˆ (t)C(t t ) ˆ (t ) 12 J 2 dt dt ˆ ( t)R(t t ) ( t ) J 2 dt dt i describing a single spin effective 1-spin problem: The average correlation and response functions are equal to those of a self-consistent single-spin problem with action ˆ, C, Cˆ ,R, Rˆ ] dt T ˆ 2 i ˆ Ý r u 3 Sloc [, 0 0 ˆ (t)C(t t ) ˆ (t ) 12 J 2 dt dt ˆ ( t)R(t t ) ( t ) J 2 dt dt i describing a single spin subject to noise with correlation function 2Tδ(t – t’) +J2C(t - t’) effective 1-spin problem: The average correlation and response functions are equal to those of a self-consistent single-spin problem with action ˆ, C, Cˆ ,R, Rˆ ] dt T ˆ 2 i ˆ Ý r u 3 Sloc [, 0 0 ˆ (t)C(t t ) ˆ (t ) 12 J 2 dt dt ˆ ( t)R(t t ) ( t ) J 2 dt dt i describing a single spin subject to noise with correlation function 2Tδ(t – t’) +J2C(t - t’) and retarded self-interaction J2R(t - t’) local response function single effective spin obeys local response function single effective spin obeys t dS 3 2 r0S u0S J dt R(t t )S(t ) h(t) (t) dt local response function single effective spin obeys t dS 3 2 r0 S u0 S J dt R(t t )S( t ) h(t) (t) dt (t) ( t ) 2T (t t ) J 2C(t t ) local response function single effective spin obeys t dS 3 2 r0 S u0 S J dt R(t t )S( t ) h(t) (t) dt (t) ( t ) 2T (t t ) J 2C(t t ) Fourier transform (u0 = 0) local response function single effective spin obeys t dS 3 2 r0 S u0 S J dt R(t t )S( t ) h(t) (t) dt (t) ( t ) 2T (t t ) J 2C(t t ) Fourier transform (u0 = 0) iS() r0S() J 2R0 ()S() h() () local response function single effective spin obeys t dS 3 2 r0 S u0 S J dt R(t t )S( t ) h(t) (t) dt (t) ( t ) 2T (t t ) J 2C(t t ) Fourier transform (u0 = 0) iS() r0S() J 2R0 ()S() h() () response function (susceptibility) S( ) 1 R0 ( ) h( ) r0 J 2 R0 ( ) local response function single effective spin obeys t dS 3 2 r0 S u0 S J dt R(t t )S( t ) h(t) (t) dt (t) ( t ) 2T (t t ) J 2C(t t ) Fourier transform (u0 = 0) iS() r0S() J 2R0 ()S() h() () response function (susceptibility) S( ) 1 R0 ( ) h( ) r0 J 2 R0 ( ) (Can solve quadratic equation for R0 to find it explicitly) critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () compute R1( ) R01 ( ) 2 2 lim 1 J R0 (0)lim 0 0 (i ) 0 (i ) critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () compute R1 ( ) R01 ( ) 2 2 lim 1 J R0 (0)lim 0 0 (i ) 0 (i ) 1 J 2 R02 (0) critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () compute R1 ( ) R01 ( ) 2 2 lim 1 J R0 (0)lim 0 0 (i ) 0 (i ) 1 J 2 R02 (0) 1 1 J 2 R02 (0) 1 R(0) T critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () compute R1 ( ) R01 ( ) 2 2 lim 1 J R0 (0)lim 0 0 (i ) 0 (i ) 1 J 2 R02 (0) 1 1 J 2 R02 (0) 1 R(0) T critical slowing down at Tc = J critical slowing down at small ω, R0-1(ω) ~ 1 - iωτ from R01 ( ) r0 J 2 R0 () compute R1 ( ) R01 ( ) 2 2 lim 1 J R0 (0)lim 0 0 (i ) 0 (i ) 1 J 2 R02 (0) 1 1 J 2 R02 (0) 1 R(0) T critical slowing down at Tc = J (u0 > 0: perturbation theory does not change this qualitatively)