cond-mat0005286/tci.tex
1: \documentstyle[aps,prl,epsf]{revtex} 
2: \begin{document}
3: \title{Edge critical behaviour of the 2-dimensional
4: tri-critical Ising model}
5: \author{Ian Affleck}
6: \address{Institute for Theoretical Physics, University of
7: California, Santa Barbara, CA93106-4030, U.S.A. and 
8: Canadian Institute for Advanced
9: Research  and Department of Physics and Astronomy, University of 
10: British 
11: Columbia, 
12: Vancouver, B.C., Canada, V6T 1Z1}
13: \maketitle
14: \begin{abstract}
15: Using previous results from boundary conformal field theory
16: and integrability, a phase diagram is derived for the 2 dimensional 
17: Ising model at its bulk tri-critical point as a function of boundary
18: magnetic field and boundary spin-coupling constant.  A boundary 
19: tri-critical point separates phases where the spins on the boundary
20: are ordered or disordered.  In the latter
21: range of coupling constant, there is a non-zero critical field
22: where the magnetization is singular.  In the former range, as the
23: temperature is lowered, the boundary undergoes a first order
24: transition while the bulk simultaneously undergoes a second order transition.
25: \end{abstract}
26: Conformal field theory has led to many exact results on 2-dimensional 
27: dimensional  critical pheomena both with regard to bulk behaviour
28: and edge or boundary behaviour.  (For a review see \onlinecite{DiF}.)
29: Assuming the bulk system is at a critical point, one can consider
30: critical behaviour at the boundary as a function of various fields
31: and interactions applied near the boundary.  In general, various
32: boundary phases and critical points exist for a given bulk
33: critical point.  These models can be used to describe either
34: 2-dimensional classical systems at bulk critical points or
35: else semi-infinite quantum chains at zero temperature.  Some of
36: these latter systems find experimental application to strongly
37: correlated electron impurity problems.    While the boundary
38: phase diagram of the critical Ising model is well understood,\cite{Cardy} 
39: surprisingly, this is not so for the next simplest case,
40: the tri-critical Ising model. 
41: Six conformally invariant boundary conditions
42: have been constructed using the fusion method by Cardy\cite{Cardy} which
43: should correspond to boundary critical points.  Certain integrable
44: renormalization group (RG) flows between these critical points have
45: been constructed by Chim \cite{Chim}.  The purpose of this note is simply
46: to connect the points with a phase diagram written in terms of
47: microscopic parameters.  This is shown schematically in
48: Fig. 1.  The most surprising conclusion is perhaps the
49: existence of a phase, in zero magnetic field,
50:  where the spins on the boundary exhibit long range
51: order while those in the bulk do not. This is the 
52: physical interpretation of a boundary
53: condition for which the corresponding boundary state is
54: a sum  of boundary states corresponding to a spin up
55: or spin down boundary condition.  
56: \begin{figure}
57: \epsfxsize=8 cm
58: \centerline{\epsffile{pd.eps}}
59: \caption{Schematic phase diagram of the boundary tri-critical
60: Ising model. Arrows 
61: indicate direction
62: of RG flows as the length scale is increased.  Along the thick
63: line the spins on the boundary are ordered.}
64: \label{fig:pd}
65: \end{figure}
66: 
67: We first briefly review the simpler case of the ordinary 
68: critical Ising model.  In that case, Cardy identified only
69: 3 conformally invariant boundary conditions corresponding
70: to spin up, spin down and free.  There are no relevant
71: boundary operators at the spin up/down critical points,
72: indicating that they are stable against the addition to the Hamiltonian
73: of arbitrary perturbations located near the boundary.   
74: On the other hand, the free boundary condition has one
75: relevant operator, of dimension $x=1/2$.  (Boundary operators are
76: relevant if they have scaling dimension $x<1$.)  The corresponding
77: relevant coupling constant is naturally interpreted as a boundary
78: magnetic field.  By standard scaling arguments we expect the
79: boundary magnetization, to scale with boundary field as:
80: \begin{equation} |m| \propto |H|^{x/(1-x)}\propto |H|.\label{ordis}
81: \end{equation}
82: This behaviour is expected to be independent of the details of
83: the spin coupling constants near the boundary.  (Only the
84: constant of proportionality in Eq. (\ref{ordis}) will depend on
85: these coupling constants, not the exponent.) No additional
86: critical points are expected at any finite field.  
87: The renormalization
88: group flow between free and fixed b.c.'s (boundary conditions)
89: was shown to be integrable by Ghoshal and Zamalodchikov in a
90: pioneering paper \cite{Ghoshal}.
91: 
92: The phase diagram for the tri-critical Ising
93: model is more interesting.  This model can be defined as a
94: spin-1 Ising model with a crystal field term which favours
95: the  $S=0$ state over $S=\pm 1$ or equivalently a diluted
96: spin-1/2 Ising model.  The (classical) Hamiltonian is:
97: \begin{equation}
98:  H=-J\sum_{<i,j>}S_iS_j +\mu\sum_iS_i^2,\ \  (S_i=-1,0,1).
99:  \label{bulkHam}\end{equation}
100:  The schematic (bulk) phase diagram is drawn in Fig. 2.  There
101:  is a second order phase transition line in the Ising
102:  universality class and also a first order phase
103:  transition line separating phases with unbroken and broken symmetry.  These
104:  lines join at a tri-critical point.  By an approximate
105:  transfer matrix mapping, one can show that this classical
106:  model exhibits the same critical behaviour as a quantum
107:  chain at T=0.  This model has the Hamiltonian:
108:  \begin{equation}
109:  H=-\sum_i[S^z_iS^z_{i+1}-D(S^z_i)^2+H_TS^x_i].\end{equation}
110:  $S^a_j$ now label quantum S=1 operators.  The transverse field
111:   $H_T$ now controls the  temperature in
112:   the classical model.
113:   While the ordinary Ising critical point corresponds to the
114:   simplest conformal field theory with central charge, c=1/2,
115:   the tri-critical point corresponds to the next unitary
116:   minimal model with c=7/10.
117:   
118:   \begin{figure}
119: \epsfxsize=10 cm
120: \centerline{\epsffile{bulk.eps}}
121: \caption{Schematic bulk phase diagram of the spin-1 Ising model of Eq. 
122: (\protect\ref{bulkHam}).}
123: \label{fig:bulk}
124: \end{figure}
125:   
126:   We now wish to consider the classical model on a semi-infinite half-plane
127:   with a boundary as shown in Fig. 3. 
128:    Although various microscopic boundary
129:   interactions could be considered, for our purposes it is
130:   enough to consider a boundary field, $H$ and a modified interaction,
131:   $J_b$ along the boundary. These couplings are indicated in
132:   Figure 3.  In the corresponding quantum chain, the Hamiltonian is:
133:   \begin{equation}
134:   H=-\sum_{i=0}^\infty S^z_iS^z_{i+1}+
135:   \sum_{i=1}^\infty [-H_TS^x_i+D(S^z_i)^2]-H_{Tb}S^x_0
136:   +D_b(S^z_0)^2-HS^z_0.
137:   \end{equation}
138:   Roughly speaking, increasing the boundary interaction, $J_b$, in
139:   the classical model corresponds to decreasing $|H_{Tb}|$ 
140:    and $D_b$ in the 
141:   quantum model thus enhancing the tendency for the spins
142:   to order at the boundary.
143:   
144:   \begin{figure}
145: \epsfxsize=10 cm
146: \centerline{\epsffile{couplings.eps}}
147: \caption{Couplings and field for the boundary tri-critical Ising model.
148: The bulk parameters, $J$ and $\mu$ are adjusted to the tri-critical point.
149: The magnetic field is applied only at the boundary.}
150: \label{fig:couplings}
151: \end{figure}
152:   
153:   The phase diagram in Figure 1 can be deduced rather straightforwardly
154:   from the properties of the 6 conformally invariant boundary
155:   states found by Cardy and discussed by Chim, and  from the 
156:   integrable RG flows discussed by Chim.  We first review these
157:   boundary states.  Using the fusion approach, one finds
158:   6 boundary states corresponding to the 6 primary fields in
159:   the (bulk) tri-critical Ising model.  Their physical properties
160:   have been elucidated, to some extent, by Chim, and we use
161:   his notation for them.  There are 2 states corresponding
162:   to spin up and spin down boundary conditions, $(\pm )$.  There are
163:   2 more boundary conditions, $(0\pm )$ which also break the $Z_2$ symmetry
164:   but appear to have the spins at the boundary only partially
165:   polarized.  [Chim actually labelled the negative polarization
166:   b.c. $(-0)$ rather than $(0-)$ but we prefer the latter
167:   notation.]  $(0)$ is the free boundary condition \cite{Saleur}.
168:     Finally there is
169:   one more b.c. which does not break the $Z_2$ symmetry and is
170:   labelled $(d)$ (for degenerate).  The correspondance between
171:   the fusion label and the physical label for the corresponding
172:   boundary states is:
173:   \begin{eqnarray}
174:   |\tilde 0>&=&|(-)>\nonumber \\
175:   |\tilde {3\over 2}>&=&|(+)>\nonumber \\
176:   |\tilde {1\over 10}>&=&|(0-)>\nonumber \\
177:   |\tilde {3\over 5}>&=&|(0+)>\nonumber \\
178:   |\tilde {7\over 16}>&=&|(0)>\nonumber \\
179:   |\tilde {3\over 80}>&=&|(d)>.
180:   \end{eqnarray}
181:   By the standard fusion rules, the (primary) boundary operator content
182:   with the boundary condition corresponding to the state
183:   $|\tilde a>$ is the set of operators appearing in the (bulk)
184:   operator product expansion (OPE) of ${\cal O}_a\times {\cal O}_a$.
185:   It thus follows that the $(\pm )$ b.c.'s admit no relevant
186:   operators; they are completely stable. (The only operator appearing
187:   in the OPE is the identity operator which just corresponds to
188:   the possibility of adding a c-number to the quantum Hamiltonian,
189:   having no effect on the critical behaviour.)
190:    The only primary
191:   operator at the free b.c. $(0)$ has dimension 3/2.  This follows
192:   from the OPE
193:   \begin{equation}\sigma '\times \sigma '=I + \epsilon '',\end{equation}
194:   where $\sigma '$ and $\epsilon ''$ are the primary fields of
195:   dimension 7/16 and 3/2 respectively.  Thus the free b.c. is
196:   also a stable fixed point!  This is a somewhat surprising result
197:   since it implies that adding a boundary magnetic field does not
198:   destabilize the free fixed point in the tri-critical Ising model, unlike
199:   what happens in the ordinary Ising model.  The partially polarized
200:   boundary conditions $(0\pm )$ have one relevant boundary operator
201:   with $x=3/5$.  Thus, there should be one unstable direction and
202:   one stable direction in the RG flow in the $(J_b,H)$ plane at the
203:   corresponding fixed points.  The
204:   $(d)$ b.c. has two relevant boundary operators of dimension
205:   $x=1/10$ and $3/5$, so both directions should be unstable
206:   at this fixed point.  Finally, it is important to consider the
207:   boundary condition labelled $(+) \& (-)$ by Chim.  The corresponding
208:   boundary state is
209:   \begin{equation}
210:   |(+) \& (-)>= |(+)>+|(-)>=|\tilde 0>+|\tilde {3\over 2}>.\end{equation}
211:   The corresponding OPE is:
212:   \begin{equation}
213:   [I+\epsilon '']\times [I+\epsilon '']=2[I+\epsilon ''].\end{equation}
214:   Thus there should be only one relevant boundary operator, of dimension
215:   $x=0$, at this critical point (disgarding the identity operator
216:   which is always present and has no effect). The presence of a
217:   non-trivial boundary operator with dimension 0 is the hallmark
218:   of an ordered phase, or equivalently 
219:   a first order phase transition with magnetic field.  It is natural to associate
220:   this operator with a boundary magnetic field.  The usual scaling
221:   law,
222:   \begin{equation}
223:   |m|\propto |H|^{x/(1-x)},\end{equation}
224:   implies $|m|\propto |H|^0$, i.e. a discontinuous jump in $m$ as
225:   $H$ passes through 0.  This in turn implies long range order
226:   in zero field.  It is also noteworthy that there are no
227:   additional relevant operators with the $  (+) \& (-)$ b.c.  Thus
228:   we expect it is stable against small variations of $J_b$ at $H=0$.
229:   This is different than the other combination
230:   $(0+)\& (0-)$ which has, in addition to an $x=0$ boundary operator,
231:   2 other relevant boundary operators with $x=2/5$.  It is
232:   also different than the situation in the ordinary Ising
233:   model where the combination of spin up and down gives 3 relevant
234:   boundary operators.   The identification of a boundary state
235:   which is a sum of two or more other boundary states with long range
236:   order was also made in the context of a critical line separating
237:   two semi-infinite Ising planes \cite{Oshikawa} and in the boundary
238:   3-state Potts model \cite{Affleck}.  However, in both those cases 
239:   it is an unstable fixed point, even in zero field.  The somewhat
240:   unusual feature of the tri-critical Ising model is that the broken
241:   symmetry phase is stable.  Related phenomena also occur in
242:   quantum Brownian motion on a triangular lattice \cite{tri}.
243:   
244:   It is now a relatively straightforward matter to connect the points
245:   to obtain the schematic phas diagram of Fig. 1. 
246:    Several comments
247:   about this phase diagram are in order. The flows from $(d)$ to
248:   $(0)$ and $(+)\&(-)$ and from $(0+)$ to $(0)$ and $(+)$
249:   are integrable. Since there is only one relevant operator at
250:   the $(0\pm )$ fixed points there must be lines in the $(J_b,H)$
251:   plane which flow towards them.  These lines must end at the
252:   tri-critical point $(d)$ since it is the only fixed point
253:   with two relevant operators.  
254:   The values
255:   of $J_b$ at the fixed points are in general unknown 
256:   (and are presumably {\it not} equal at 5 of the fixed 
257:   points as drawn in Fig. 1.).  However, we do expect that
258:   $J_b$ is smaller at $(0)$ than at $(d)$. It is natural to
259:   place the $(+)\&(-)$ fixed point at $J_b=\infty$ since there
260:   the spins along the boundary are perfectly ordered.  The values of
261:   $J_b$ at the $(\pm )$ and $(0\pm )$ fixed points are relatively
262:   insignificant and simply correspond to points where the leading
263:   irrelevant coupling constant vanishes.  Although a boundary
264:   field is irrelevant in the central phase of the phase diagram 
265:   this {\it does not} imply that the boundary magnetization is
266:   zero in the presence of a non-zero field.  Irrelevant operators
267:   will still lead to a non-zero magnetization which should be
268:   an {\it analytic} function of $H$, thus being linear at small
269:   $H$.  
270:   At the $(0\pm )$ fixed points, we expect $m(H)$ to be
271:   singular, behaving as:
272:   \begin{equation}
273:   m-m_c- a(H-H_c)\to b_{\pm}|H-H_c|^{x/(1-x)}=b_{\pm}|H-H_c|^{3/2},
274:   \end{equation}
275:   since $x=3/5$. Here the amplitudes, $b_{\pm}$ are presumably
276:   different for $H>H_c$ and $H<H_c$.  The linear term, $\propto a$,
277:   is non-singular.  
278:    This is a relatively mild singularity since
279:   both $m$ and its first derivative remain finite and continuous,
280:   while the second derivative diverges. 
281:   The shape
282:   of the phase boundary near $(d)$ is determined from the scaling
283:   dimension of  $H$ and $J_b-J_c$ to be:
284:   \begin{equation}
285:   J_c-J_b(H)\propto |H|^{4/9}.\end{equation}
286:   At the tri-critical point $(d)$, 
287:   \begin{equation}
288:   |m|\propto |H|^{1/9},\end{equation}
289:   and for $J_b>J_c$, $m$ has a first order jump at $H=0$.
290:   
291:   These results are all consistent with the ``g-theorem''\cite{Affleck1}
292:    that states
293:   that the groundstate degeneracy, $g$, always decreases during
294:   a boundary RG flow.  The $g$ values are given by:
295:   \begin{eqnarray}
296:   g_{(\pm)}&=&C \nonumber \\
297:   g_{(0\pm )}&=&C\eta^2 \nonumber \\
298:   g_{(0)}&=&\sqrt{2}C \nonumber \\
299:   g_{(d)}&=&\sqrt{2}\eta^2C\nonumber \\
300:   g_{(+)\&(-)}&=&2C,\end{eqnarray}
301:   where:
302:   \begin{equation}
303:   C=\sqrt{\sin {\pi \over 5}\over \sqrt{5}},\ \  \eta = 
304:   \sqrt{\sin {2\pi \over 5}\over \sin {\pi \over 5}}.
305:   \end{equation}
306:   Noting that $\eta^2\approx 1.61803>\sqrt{2}$
307:   we see that all
308:   RG flows in Fig. 1 are consistent with the $g$ theorem.  This was
309:   already observed by Chim in the special cases of the integrable flows.  
310:   
311:   The existence of the ordered line may seem somewhat surprising since
312:    there is long range order at finite temperature along the 
313:    (one dimensional) boundary,
314:   for $J_b>J_c$, even though the (two dimensional) bulk is disordered (or,
315:   more accurately, is sitting at a critical point separating
316:   ordered and disordered phases).  This is
317:   surely reasonable at $J_b\to \infty$ 
318:   but may be harder to swallow for finite $J_b$. 
319:   In the quantum chain context this behaviour is not so unfamiliar.  
320:   We may think of the $(+)\& (-)$ fixed point as corresponding
321:   to $H_{Tb}=0$.  In this limit $S^z_0$ commutes with the 
322:   Hamiltonian
323:   and there are 2 degenerate groundstates, with $S^z_0=\pm 1$. 
324:   (We assume that $D_b$ is sufficiently large and negative that
325:   these states have lower energy than the one with $S^z_0=0$.)
326:   These groundstates have non-zero, equal
327:   and opposite values of the magnetization, localized near the boundary.
328:     Applying an
329:   infinitesimal boundary field picks out one of these two groundstates,
330:   leading to the discontinuity.
331:   The above RG analysis
332:   implies that a small transverse boundary field is {\it irrelevant}
333:   so the jump in the magnetization at $H=0$ should persist for a range
334:   of non-zero $H_{Tb}$.  A somewhat related phenomena occurs in the
335:   {\it ferromagnetic} Kondo problem.  The Kondo coupling constant
336:   is irrelevant in this case so that the impurity spin decouples
337:   from the conduction electrons at the stable fixed point.  Since
338:   the impurity spin operator then commutes with the Hamiltonian, 
339:   there are degenerate groundstates and a discontinuous magnetization
340:   in an applied field.  
341:   
342:   It is interesting to consider varying $T$ through the bulk phase transition
343:   (with the Hamiltonian held fixed) .  Referring to Fig. 2, we see that for
344:   large $\mu$ the bulk transition is continuous and  in the usual Ising
345:   universality class.  In this case the boundary also orders continuously. 
346:   (This follows from the fact that the $(+)\&(-)$ fixed point is unstable, even
347:   in zero field, in the ordinary Ising model.)   On the other hand, for smaller
348:   $\mu$ the bulk transition is first order.  We then expect the boundary
349:   transition to also be first order since  critical behaviour of the boundary
350:   at a non-zero $T$ is presumably impossible unless the bulk is also critical. 
351:   Now consider what happens if $\mu$ is adjusted to its critical value so that
352:   the bulk transition is in the tri-critical universality class.  In this case
353:   there are two possibilities for the boundary transition, depending on $J_b$. 
354:   When $J_b<J_c$, the boundary transition is also second order.  However, when 
355:   $J_b>J_c$, the boundary undergoes a type of first order transition while the
356:   bulk undergoes a second order transition.  This follows from observing that,
357:   infinitesimally below the critical temperature,  the boundary magnetization
358:   is finite whereas the bulk magnetization is infinitesmal.  On the other hand,
359:   infinitisimally above $T_c$, the correlation lengths in both bulk and
360:   boundary are presumably diverging together.  We may understand the
361:   possibility of the boundary having a first order transition at the bulk
362:   tri-critical point as being connected with the fact that the bulk system is
363:   at the end of a first order transition line. When the bulk transition is
364:   ``almost first order'' it becomes possible for the boundary  transition to be
365:   truly first order.
366: 
367:   It was recently observed \cite{Recknagel} that the integrable RG flow from 
368:    $(d)$ to $(+)\&(-)$ 
369:   has a generalization to all the minimal models
370:   with diagonal partition functions, which may be thought of as
371:   increasingly multi-critical Ising models and that, in all cases, the
372:   $(+)\&(-)$ fixed point is stable except for a dimension 0 operator.  
373:   This implies that all these models have an ordered phase at zero field, as
374:   discussed here for the tri-critical case. 
375:   
376:   I would like to thank M. Oshikawa and H. Saleur for numerous 
377:   useful discussions on related subjects and H. Saleur for
378:   drawing my attention to Ref. \onlinecite{Recknagel}.   
379:   This research was supported by NSF grant PHY-94-07194 and
380:   by NSERC of Canada.
381:   
382:   \begin{references}
383:   \bibitem{DiF} P. Di Francesco, P. Mathieu and C. S\'en\'echal,
384:   {\it Conformal Field Theory}, Springer, Berlin, 1997.
385:   \bibitem{Cardy} J.L. Cardy, Nucl. Phys. {\bf B324}, 481 (1989).
386:   \bibitem{Chim} L. Chim, J. Mod. Phys. {\bf A11}, 4491 (1996).
387:   \bibitem{Ghoshal} S. Ghoshal and A. B. Zamolodchikov,
388:   Int. J. Mod. Phys. {\bf A9}, 3841  (1994).
389:   \bibitem{Saleur} H. Saleur and M. Bauer, Nucl. Phys. {\bf B320}, 591 (1989).
390:   \bibitem{Oshikawa} M. Oshikawa and I. Affleck, Nucl. Phys.
391:   {\bf B352}, 849 (1997).
392:   \bibitem{Affleck} I. Affleck, M. Oshikawa and H. Saleur,
393:   J. Phys. {\bf A31}, 5287 (1998).
394:   \bibitem{tri} H. Yi and C.L. Kane, Phys. Rev. {\bf B57}, R5597 (1998);
395:   I. Affleck, M. Oshikawa and H. Saleur, in preparation.
396:   \bibitem{Affleck1} I. Affleck and A.W.W.  Ludwig, Phys. Rev. Lett.
397:    {\bf 67}, 161 (1991).
398:    \bibitem{Recknagel} A. Recknagel, D. Roggenkamp and V. Schomerus,
399:    hep-th/0003110.
400:    \end{references}
401:   
402:   
403:   
404:   
405:   
406:   
407:   
408:   
409:   
410:  
411:   
412: 
413: 
414: \end{document}
415: