Documentos de Académico
Documentos de Profesional
Documentos de Cultura
1 and T-90 we get m(T)—>0). 3) The most negative eigenvalue of the Hessian near Te is equal to -c(T-Tc)'/9, while for the RS solution it is equal to -c(T-Te)’. So that, in a sense, the instability is reduced by a factor 9. The conclusion is that, although the considered one-step RSB solution has also tured out to be not perfect, itis @ much better approximation to the true solution than the RS one. Therefore one could try to move further and further in the chosen “direction” in the replica space. 4.3.3 Par ing Scheme The Parisi ansatz results from the following sequence. First divide the (nXn) matrix qup into (nm,)* blocks of (m:Xmy) matrices, Take the off-diagonal blocks to have all their elements equal to qo, Sub-divide the diagonal (m,Xm1) blocks into (my/m;)? sub- ablocks of (m:Xm,) matrices. Take the off-diagonal (mXm,) sub-blocks to have all their elements equal to qy. Subdivide further the (mjXm2) sub-blocks into (my/m3)*sub-sub- blocks of (m;Xm,) matrices. Take the off-diagonal (msXm,) sub-sub-blocks to have their elements equal to qz. And so on, sub-dividing diagonal blocks, with n2m12m2>...2mk>1, now take the limit n-¥0, reversing the inequalities to OGin (x); 0x Us bf Gterde) + HM 4) :where M is the magnetization per spin o See C20) M=[ ‘The statistical mechanics order parameter q is given by rh b= Jo qeudx en The fee energy i complicated and given for 0 by -Pf = ZBIT LI “28 ri Feud) a 1 (3m: ea ee ee H# Gos 3 de 22) where, for asiven as), Gly) i the oe of the partial differential equation 245 = -¥ 28 29 af 28-)°) 23) with tania condition > oF ae Gi. oe ay) Thereby the problem cah be formally reduced simply to that of solving a nonlinear differential equation.° alr 4.3.4 Replica. Symmeiy Breaking Solution : Near T, The general solution for 90 is unknown, but- ct 's possible te obtain cx exactly close to. Exparding the exponential in Fq.cB) ind performing the trace we cbtain . a F) ¥ 541 = Jing EBT CF Y- SHG LE Oe ) 22 1 ¢ +E foe fen ~F TOFD ceo ere 6s CT Go/y \ccording te the general scheme of RSB, We eacily et. : | & ¢ lim Maeag CF aS cm “M4 9, aan ~fdx9 bay (26) Ue An pi=len [agin + 3geo( ay Pew] Thus the free energy Com be written Ce 690m) =f ds [019% fs 2% -fxgex - ex LGeady ) 28) We vary it with respect te 90%) APIO — 4B - C9 tyrdy 2 gor f 'geardy +F Joo =o a 624)Because the signs sin the term XJijm; are random,” So both it and the readin field term ave of arder Unity. /n ferromagnets , in Contrast. the reaction field i's smaller than the cnonzero) mean molecular” field by @ facter 4y cory for finite range lattice mode/sp. : The TAP eqs, can be derived! by variation C265 m,e0) of @ Free energy functional he Fdmife- 25 Ty MM Z, hom fo fe eae CAM 3) +272, f+ f é "Inf; C+ m, )) ( ea tom: nleci-m.] It is instructive te leek i rameeg lan t74) eok at the Hessian mattix formed -from second derivatives of Frap Defining i ao Fae, P If cay is to eens ase Bin oo cy) Te San nsible ph i ; We cannet, have any weet e prea Saline, ea these T°Ci~ et Bo Fen = xb Zz; m?> ©Seo, Gc1 = |6| +0(6*) (395 The solution given by G9s.G3)-G0) . ¢38) +s sh Mn Figure. Jo It is alse straight —ferwored te include @ small magnetic = cS field H. The solution ic Ini Fe SU ef the arm C33)- 636) meno ' 3 H but Now ee = id Se a second “Plateau” regien appears As the field incveases. thie plateau rises, and once tt reaches the height of the Seco! plateau ¢ H=Haq) ~ +he only Solution 7s x- independent. That is- the ASB AicappOar and we are back te the Sk solution , Asa consequence of Gq.C38) We Coan Show shee S$, Yoodx =I + cHeer 7 exactly. seo. from cred 1.0 x a constant - below TF. A Constant Susceptibility below °° My 2. Te is Similay te many frelol -Covled Mmayretizatimn Mmeasuremeaty (3)44 TAP equations Ze In this sect, we discuss +he approaches toa Solution of the Sk model that have attempted to avoid -the replica “thick. Replicas Qppear, when we perform the average. over the disorder, Thouless. Andersen .a"d Palme (977) Suggested that one defer ths average te th end and first of all write down local MF quater Gor the site magnetizatims for a given set of bonds, The naive guess is that mM, =tonh(e@d; Tym;) F cn would be the appropriate Set ef equatiens. However, TAP realized that an extra € erm is necessary {or SG, even with imGinite Vanja, interactions, This is Known as the Onsager reactic field term. We have ea “mn, =tanh( acd, Hmsthy BS 3 ches : Bqsca ave N coupled nen lnear equatre 3 the local magnetizahons mM; . ae The third term describes the Contribuhen te a internal field from the spin Si itself. The ” at site i produces a MF mj; at site j-Wme induces a magnetization 23; ; qi at cite 5 and hen A mean field I; 7;;m, at site i.DidSerentiating this equation usth respect to % we Rind ! aol - 22900 -2f gewdy tegen =o ce, or g to =0 31> the last equation means that Gtx = Cont Which corresponds te the RS Selukom which Not be considered , Differentiating 30> agar iol yields Yow aie ¢32) [f We assume thet Qcxr is continucus . the Solution must be of *he form GOD = Ga asx s} 632) 9) = BR Ke ses, 4) Gm = FO. OS sXe Cas-9 Wheve 2, = 249 np OC aS: 24¢0) substituting 64s.€339-c2» ers 029) pivec #or[2 16] -2gea+ 2geu*) -#£ feot=q 636) m Gen) [ate] - 2fa+ 2a *) — #c0)%= o 37) t %=0 and te ves pectively , Frem c262-C4 'e see that either Gcor= G62 2 Which is just the W RS Selutt again. or A G=0 CH=0) aeet 216] =29¢0 -~2 903,: ‘cally The TAP solitions can be investigated ey 4 céose te the transition and A;30. Thes U =a ™m;=B5; Ty my +@hz- CBs) m 2 with Z)5;=NUI5 laves ; €4.@) is Solved by diagonalization af the Symmetric matrix Ty cq Jiy= Tl; with the eigenvalues Ty ane the peal orthogona Olgehfuactins ; Zs GND = Say’ XZ) salj> 238; We have the magnetizahins and Lie lels Wg = 2 Als; D hey hy a) Ard from (8) 2 m [i-BT, + 627 = Bh, cra) For N>ee the randem matriy Ty hag o distribut of eigenvalues PDD whic obeys a Semicirevlar lav Cle) with the lavgect eigenvalue gl = Paci aah 2 Ps Gapy Care Bayt cz) The Susceptibility A S duy a es aysmax a diverges for the largest eigenvalue J, at the freezing temperature Rd. Hence the TAP €75. Lead to a phase transition at the same Tp @s the Sk theory, Without the Onsager term the (ivergence would be at T=Z™*Zay, rhe TAP e9s. Can be Simplified if we replace. the hrc C5; Jay 5; ¢- m;9 =7°t-F) cts) with G2 tam} ce) the single-phase 5G order pammetar Gea 4 Simple estimate of the low-T limit of Ger) can be obtained from the entropy SCT) - nae have from the free energy es cl oa afm; = 2 2 2 S(t) = 9 o-2e*2;; Ty Cin, DC ) 4 a £2, feem,>Inlec 14M] + Cr M Mn Zer-msr] In the [imit Tro We havea Me>t( and the Secono| term vanishes, This ponds to ag) SCT=0) =- + wy* lim Br¢/- 9) : ‘he Condition af a he coe at Ta0 1s therefi Nhat -F vanish -fastey than ae as T> OoMC simulations indicate. Gen = (Ly » €T39) 0.9? This leads to sit tear T dependence of the Susceptibility kn asTeo. The const. o 's related te the disttibuhban per Of Ctetal> internal fields h; = = 3; Si;m; , TAP aa found numerically that pen) (s linear inh ; Pech) = # Civ ee (20) If we dating {ry Fa(dimiipcks cy) we have at low T dis I- Gen =f “Pky i -abehod na by {Cem homael _ Using 19 ana ae the TAP e7s. i'n the ny Bhi =Am; + tan hy ™; ra We aveain act y= beT *fer- -mdcam + tanh m) Cat he St idm €) TAP arqued that the physically relevant value Of & WAS the one which minimized ho; this Jives ec 2Clh2)* 2 1.665. On the other hand Bray aud Moore 6:¢7qQ> have propesed that te entire SG phase below the AT line is marginally stable, So that Eq Is Sati stiel os an equality Ths qv’ X= 1.810.The marginal stabilit ‘ hem 4 of TAP Solutions Cai be ch ee i Nemoto at gl (985.86) invest): a: ets ~ st ‘onary points of the TAP -free energy F{m;} Se 18 par inite ey fin Solutions wth (9, i whee e F)=0 as wel pe with IDF/#0. The latter turn out t, rs ee pe uth IVF/30 as Wace which stick : a oundary of the region of stability as defined 1 lowerest ergenvalve r"%~, In Figure | boundary is indicated & ae é : eee Y dished line, pf this stability Viwat yr"weutd he negative 2nd is oN permitted. Lis the Fiquye shows schematically 5% . ‘ the Erap a8 % function Of the. rs ence belongs 2 a, (owed of ‘ t eigenvalue Yh are, San 25 Tn case cap thave is astenle TAP case chy tH : : 4 minimum oth (0F20 solutin , WARrens in 3 : . : 1.6, mn to the pont ynizo fer Ne>ee. 1.0 marginally stad soddl. pome. — How many solutions of the TAP O95. 11980) .De Dominicis et a(C62)- Tanaka and Edwards C80) Aal Showed thet there ave an exponentially (AME nAmbCr, uchraso | ahove AT CNsCr.t)),, ecexp CIVOCHLT), oe ee are there? Bary ety 4.5 Physical interpretation of yeplicn My Symmetry breaking There ave many solutions of the TAP equations . Many minima should be separated by barvievs that diverge in the thermadynamics limit, otherwise therm fluctuakins would mix them tegether. Hence each minimum is & different thermodynamic state . For an infinite syttem it is necessary to distingutth between a within @ Single thermodynami. Cpure State and @ full Statistical mechanics CGibss) average , Which is a weighted Sum over all states. 45.1 Puve states (See 3.4) In pure stokes _ call the connected covvelaton functions Suchas < $; S$; 2B SS; S; > -<5:> $5 7- becomes zero at lavge distances in these states, 4.5.2 The phusteal order parameter . How can senile states fe distinguished one From another 7 oe i \t is natural to jntveduce, the concept of “distewce in the Space of states . ao exemple, dap = = cme-mf>> , he overlap between the two States is ap = rt Zaz mf 0 j -3- § 901s amonetonic one , then we Can introduce. he inverse function wep», onal g Kean be rewritten as follows : ol XCP) ce ct k q Me Cd ag F in the a hand . ‘ qiMs de Pop? AXxG2 So that , We 1 . Pep = ae ov yee =f Fde’Pea whieh aa the ancwer to the quection , what is the meaning of he funehion FO" ; The function *¢9) inverse te Jon gives the probability te find a pair ef pure States whteh would have On overlap not bigger than F. Fiqure Shows the form of Pq) derived from the Paris: lo in Reo at a temperature 24 % , hey with the vesults of ME simulation measurements on samples of several sézes, Py? 3.04.5.4 Ultvametricity The Peg) 15 not enough to understand what the tructure of the Space of pure States in SG paase . Ve need te know ‘the higher order covvelection propertve i the overlps of “the Stalles et us consider the function Peque4a » fs uhieh yould desevibe the joint statistics oF the overlaps & avbitrary three pure States, This function gives the probability that arbitrary three pure states of, 8 ard r would have their mutual over laps Gap. Gur aud Ger te be equal conrespendingly “te Feo9. ord ¥ PCQ1.4.95.) =L Pr hr Be, Bsrlav 1 CF Pale 5 Gr Gap) S45 farrier i i ab ¢. A poe We eed ysegF 5H -FP In the Pavis) parametrization [+ becomes PCR Mofo) = peg. re) SCS Ga-F1 t CP CH) PCH? 8.6% PISGar fy I+ Perr mtinns this result is very interesting ; It is not equal to zero only if ot least twe of the three overlaps are equal and their yolue is not bigger than the third one .Now we have arrived to the concept of the. ~ ultrametricity . As is Said to be ultrametrc if Uts metric has the property that -fer any three points A. b And c of this space the following inequality cs satisfied ; Aap $ MaxCdac Abe) . One of the Simplest ways to describe ultrametr space is in terms of the hierarchical tree, The ultrametric space i's =e OSsociated with the Set OF the endpoints of the tree. The distance between and two points of +thic space ’'s defined such that it depends Ee only on the number of = ime * Gene ratisns” in the “vertical” ‘ divection to that level of thettree at whieh thete two points have a common Ancestor. | | With thie definition of the distance the set of the endpoints of the tree is ultrametric . It is quite remarkable that such a highly disordered system as the $k SG should possess Such a vich Structure in its states , The uttrametricity of the phases Comes out ih this theory because we put itin the hierarchical Structuve of the 9%P matrix ¢ Parisi scheme