1: \chapter[Correlation between fragility and vibrational properties]
2: {Correlation between fragility of the liquid and the vibrational
3: properties of its glass}
4: \label{chap3}
5:
6: \section{Introduction}
7:
8: The identification of the microscopic details that, in a given
9: glass former, determine the temperature dependence of the
10: viscosity, and thus the value of the fragility, is a long standing
11: issue in the physics of supercooled liquids and glassy state.
12: Large numerical and theoretical effort has been devoted to the
13: attempt to relate the fragility to the specific interparticle
14: interactions (e.~g. strong glasses are often characterized by
15: highly directional covalent bonds, while the fragile one have more
16: or less isotropic interactions).
17: The phenomenological relevance of the concept of fragility relies on the
18: correlations that have been found between this index and other
19: properties of glass-forming liquids. Examples of these
20: correlations are the specific heat jump at $T_g$
21: (see Eq.~(\ref{frag-DCp}) and \cite{MA01}),
22: the degree of stretching in the
23: non-exponential decay of the correlation functions in the liquid
24: close to $T_g$ \cite{Ngai}, the visibility of the Boson peak at
25: the glass transition temperature \cite{Sokolov}, or the temperature
26: behavior of the shear elastic modulus in the supercooled liquid state
27: \cite{Dyre}. Recently a strong correlation between fragility of the liquid and
28: vibrational properties of its glass has been found~\cite{TS}.
29:
30: \subsection{Fragility and number of states}
31:
32: Recently, the attention has
33: been focused on the possible relation existing between
34: fragility and the properties of the (free) energy landscape,
35: more specifically the (free) energy distribution of the minima
36: and the properties of the basins of attraction of such minima.
37: A key point is the validity of
38: the Adam-Gibbs relation (\ref{AGrel}):
39: \begin{equation}
40: \label{AG}
41: \t(T) = \t_\infty \exp \left( \frac{\cal E}{T \Sigma(T)} \right) \ .
42: \end{equation}
43: By using the
44: Adam-Gibbs relation, one could expect to relate fragility to the
45: properties of $\Sigma(T)$, i.e., to the distribution of basins in
46: the phase space of the system. For example, many authors proposed
47: that fragile systems should have an higher number of states, \ie
48: a larger complexity, with respect to strong
49: ones~\cite{An95,Sp99,Sa01,DS01}.
50:
51: However, this possibility is
52: frustrated by the lack of knowledge on the parameter ${\cal E}$.
53: Some theories attempting to compute ${\cal E}$,
54: summarized in section~\ref{sec1:instantons}, appeared only
55: recently\footnote{In particular, the papers \cite{Fr05,DSW05} appeared
56: after this work was completed so their results were not known at the
57: time this calculation was performed.} and in general the theory
58: of the Adam--Gibbs relation is still at an early stage
59: of development.
60:
61: Unfortunately, even if a model for $\Sigma(T)$ is chosen, so the total number
62: of states is fixed, one can obtain the whole
63: range of experimentally observed fragilities by varying
64: ${\cal E}$: fragility is related to $\Si(T)$ by
65: Eq.~(\ref{frag-DCp}), but the value of $T_g$ depends strongly on $\EE$.
66: More specifically, in \cite{noifrag} it was observed that for
67: a large class of models for $\Sigma(T)$ - where $\Sigma(T)$ is a
68: concave function of $T$ that vanishes at a given temperature
69: $T_K$ and assumes its maximum $\Si_\io$ at high temperature
70: (``Gaussian-like models'') - the relevant parameter that actually
71: determines the fragility is
72: \begin{equation}
73: \label{Ddef} D=\frac{\cal E}{T_K \Si_\io} \ .
74: \end{equation}
75: For example, if $\Si(T)=\Si_\io \left( 1 - \frac{T_K}{T} \right)$ - the form
76: that is commonly used to fit experimental data, see Eq.~(\ref{ScFIT}) and
77: Fig.~\ref{fig1:Sconf} -
78: is substituted in Eq.~(\ref{AG}) and the fragility is calculated from
79: Eq.~(\ref{mAdef}), one gets
80: \beq
81: \frac{m_A}{17} = 17 \log 10 \, D^{-1} + 1 \ .
82: \eeq
83: Thus, fragility appears to be determined by the ratio between
84: ${\cal E}$ (measured in units of $k_B T_K$) and the total number
85: of states $\Si_\io/k_B$; it is related to both the distribution
86: of minima (through $\Si_\io$) and the characteristic of the
87: transition path between them (through $\cal E$). The relation
88: between fragility and phase space properties can be even more
89: complicated, in those cases where the function $\Sigma(T)$ does
90: not belong to the Gaussian class.
91:
92: \subsection{Fragility and vibrational properties of the glass}
93:
94: \begin{figure}[t]
95: \centering
96: \includegraphics[width=.45\textwidth,angle=0]{fig-ch3/scopignofig2.eps}
97: \includegraphics[width=.45\textwidth,angle=0]{fig-ch3/scopignofig3.eps}
98: \vskip-4cm
99: \caption[Correlation between $\a$ and the fragility from IXS]
100: {From~\cite{TS}: (Left) The inverse
101: nonergodicity factor $f(Q^*,T)^{-1}$ for
102: $Q^* = 2 \, {\rm nm}^{-1}$ and for three
103: different substances as a function of $T/T_g$. In the inset, the
104: wave vector dependence of $f(Q,T)$ is shown at fixed temperature $T^*$ to
105: demonstrate that $f(Q^*,T) \sim \lim_{Q \to 0}f(Q,T)$.
106: (Right) Correlation plot of the
107: fragility and the index $\a$ defined in Eq.~(\ref{alphaTS}).}
108: \label{fig3:tullio}
109: \end{figure}
110:
111: Recently a strong correlation between fragility
112: and the vibrational properties of the glass at low temperatures
113: has been found \cite{TS}.
114: The nonergodicity factor $f(k,T)$ defined in Eq.~(\ref{nonergdyn}) was
115: measured, by mean of inelastic X-rays scattering,
116: in the (nonequilibrium) glass phase after a quench from high
117: temperature. In Fig.~\ref{fig3:tullio} the temperature dependence of
118: $f(k \sim 0,T)$ is shown\footnote{In~\cite{TS} the wave vector was indicated
119: by $Q$ instead of $k$, so this notation is used in Fig.~\ref{fig3:tullio}}.
120: It is found that $f(k\sim 0,T)^{-1}$ is approximately linear in $T/T_g$ and
121: an index $\a$ is defined as the slope of the curves in Fig.~\ref{fig3:tullio}:
122: \begin{equation}
123: \label{alphaTS}
124: \alpha(T_g) =
125: \lim_{k \rightarrow 0} \left. \frac{d [f(k,T)]^{-1}}{d (T/T_g)} \right|_{T=0} \ .
126: \end{equation}
127: The index $\a$ has an explicit dependence on $T_g$
128: (as $\frac{d}{d(T/T_g)}=T_g \frac{d}{dT}$). Moreover, it depends on $T_g$ also
129: because depending on the value of $T_g$ (\ie on the experimental time scale)
130: different states are selected: in particular, the states that are selected are
131: the {\it equilibrium} states at $T=T_g$, as exactly at this value of temperature
132: the system falls out of equilibrium.
133: The order parameter $f(k,T)$ of a given state
134: is a measure of the volume of this state in the phase space, and can in principle
135: depend on the state in which the sistem is frozen below $T_g$.
136: In particular, in~\cite{TS} an expression of $\a$ in terms
137: of the harmonic vibrational properties of the states (eigenmodes of
138: the disordered structure), was derived: and the eigenmodes will depend on the
139: particular structure in which the system is frozen, that is, on the
140: {\it equilibrium structure} at $T=T_g$.
141:
142: In the right panel of Fig.~\ref{fig3:tullio} it is shown
143: that the index $\a$ is strongly correlated with the fragility
144: index $m_A$. This finding implies the existence of a
145: relation between three features of the energy landscape: the energy of the
146: minima, the transition paths between them (that together determine
147: the fragility) and the Hessian matrix, evaluated at the minima
148: themselves, that fixes the vibrational properties.
149:
150:
151: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
152: %%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
153:
154: \section{Fragility in mean field $p$-spin models}
155:
156: As discussed in the first chapter, mean field models such as the
157: disordered $p$-spin model provide an useful framework to understanding
158: many aspects of the glass transition. Using the arguments of sections
159: \ref{sec1:BBargument} and \ref{sec1:instantons} one can relate the
160: quantities appearing in the Adam--Gibbs relation, \ie $\Si(T)$ and
161: $\EE (T)$, to the mean field potential $V(q,T)$ that is expected to
162: describe short range models at the {\it local} level.
163: One can then investigate the $p$-spin models as solvable models of ``glass'',
164: where the distribution of minima is ``Gaussian-like'' as in real structural
165: glasses, and both the vibrational properties of the minima and the energy
166: barrier $\EE(T)$ can be analytically estimated. The $p$-spin models are ``Gaussian-like'',
167: in the sense that their complexity - even if the distribution of states is not
168: exactly Gaussian - is known to be a concave function of the temperature,
169: that vanishes at $T_K$ and assumes its maximum at $T_d$, without any inflection point
170: in between, see~\cite{CS95} and Fig.~\ref{fig1:Scqualit}, yielding a form very
171: similar to Eq.~(\ref{ScFIT}) for $\Si(T)$.
172:
173: In this chapter, both the spherical and Ising version of the $p$-spin model
174: will be investigated, in order to
175: check whether one can reproduce the correlation between fragility of the
176: liquid and the vibrational properties of its glass found in
177: \cite{TS} by studying the geometry of the phase space of these models.
178: Moreover the question of the existence of a correlation between fragility and
179: number of states \cite{Sp99,Sa01,DS01} will be addressed.
180:
181: The mean field models will be considered as models for the {\it local}
182: properties of a short range glass, as indicated by the arguments discussed
183: in sections \ref{sec1:BBargument} and \ref{sec1:instantons}. Then, the existence
184: of the dynamical transition will be ignored, being an artifact of mean field,
185: and it will be assumed that it is possible to equilibrate the system below
186: $T_d$ with a relaxation time following the Adam--Gibbs relation (\ref{AG})
187: with $\EE$ and $\Si$ determined by the mean field potential $V(q,T)$.
188: As will be clarified below, the fragility of the models can be varied by
189: varying the parameter $p$.
190:
191:
192: \subsection{Definition of the relevant observables}
193: \label{sec:definitions}
194:
195: \noindent
196: It is useful to summarize the definition of all the quantities that will be
197: computed, that are listed below:
198: \begin{equation}
199: \nonumber
200: \begin{array}{ll}
201: T_K & \text{Thermodynamical transition temperature} \\
202: T_g & \text{Glass transition temperature} \\
203: T_d & \text{Dynamical transition temperature} \\
204: \Sigma(T_g) & \text{Complexity at $T_g$} \\
205: m(T_g) & \text{Fragility} \\
206: \alpha(T_g) & \text{``Volume'' of the equilibrium states at $T_g$} \\
207: {\cal E}(T_g) & \text{``Barrier height'' at $T_g$}
208: \end{array}
209: \end{equation}
210: Setting $k_B=1$, all the above quantities are either dimensionless or have the
211: dimension of an energy; in the $p$-spin model -as usual in classical spin
212: models- a natural energy scale
213: $J$ appears as the strenght of the couplings between spins. Thus, if one
214: additionally sets $J=1$, all the quantities become dimensionless.
215:
216: \subsubsection{Temperatures}
217:
218: From the two replica potential $V(q,T)$ discussed in section
219: \ref{sec1:potentialmethod} the complexity $T\Si(T)=V(q_{min}(T),T)-V(0,T)$ and
220: the barrier height $\EE(T)=V(q_{max}(T),T)-V(q_{min}(T),T)$ are extracted as
221: functions of the temperature.
222: Then, the thermodynamical transition temperature $T_K$ is defined as the temperature where
223: the complexity vanishes: $\Sigma(T_K)=0$,
224: and the value of $V$ at the secondary minimum becomes equal to zero
225: (see Fig.~\ref{fig_1}).
226: The dynamical transition temperature $T_d$ is the temperature at which the metastable
227: minimum first appears, so it is defined by $\EE(T)=0$.
228:
229: Then, an Adam--Gibbs like relation in which $\EE(T)$ plays the role of the energy
230: barrier and $T\Si(T)$ of the configurational entropy is
231: considered\footnote{The argument of section \ref{sec1:instantons} predict an
232: Adam--Gibbs relation where $(T\Si)^2$ enters in the denominator and
233: $\EE^{\frac{3}{2}}$ in the numerator. However, the exponents will be neglected
234: as their robustness is still a matter of debate.}.
235: Starting from the Adam--Gibbs relation
236: (\ref{AG}) one defines $T_g$ by $\t(T_g)/\t_\infty =$ const, or, equivalently, by
237: \begin{equation}\label{defTg}
238: \frac{{\cal E}(T_g)}{T_g \Sigma(T_g)} = {\cal C} \ .
239: \end{equation}
240: The value of the constant ${\cal C}$ determines the value of $T_g$.
241: It is arbitrary because proportionality factors have always been neglected,
242: so it will be fixed in order to
243: obtain reasonable values for the fragility,
244: $m_A/17 \sim 1 \div 10$, as observed in experiments, see Fig.~\ref{fig3:tullio}.
245: It will turn out that the analysis is not strictly dependent on the
246: value of $T_g$ (and of $\CC$), the behavior of the various quantities at $T_g$
247: being representative, as will be shown, of a general trend observed
248: at all temperatures $T\in [T_K,T_d]$ by varying $p$.
249: Different choices of the constant $\CC$ change only quantitatively
250: the results, while the qualitative picture stays the same.
251:
252:
253: \subsubsection{Complexity, barrier heights and fragility}
254:
255: Given a definition of $T_g$, the complexity at $T_g$ is simply $\Sigma(T_g)$
256: and the barrier height ${\cal E}(T_g)$: clearly, these two quantities are related
257: by Eq.~(\ref{defTg}).
258: Knowing the complexity as a function of the temperature,
259: the fragility can be defined as in Eq.~(\ref{frag-DCp}). To simplify the notations,
260: the factor $17$ entering Eq.~(\ref{frag-DCp}) will be neglected in the following
261: and the fragility defined as:
262: \begin{equation}
263: \label{fragilita}
264: m(T_g)=1+T_g \frac{\Sigma'(T_g)}{\Sigma(T_g)} \ .
265: \end{equation}
266: The latter definition is very useful in a mean field context as - once a definition of
267: $T_g$ has been chosen - it involves only the complexity,
268: that is a well-defined quantity in mean field models. It is equivalent to the usual Angell
269: definition of fragility if $\eta_\infty$ does not depend strongly on the material,
270: and the Adam-Gibbs relation is assumed
271: to be valid~\cite{noifrag}. This definition of fragility has been shown to be correlated to
272: the fragility defined from the relaxation time using experimental data in~\cite{MA01}.
273:
274: \subsubsection{Volume of the states}
275:
276: The index $\alpha$ defined in \cite{TS} can be replaced by other equivalent
277: - equivalent meaning positively correlated -
278: definitions. An useful equivalent definition of $\alpha$ is
279: \begin{equation}
280: \alpha(T_g) = \lim_{k \rightarrow 0} \Big[ 1 - f(k,T_g) \Big] \ .
281: \end{equation}
282: As one can easily check observing Fig.~\ref{fig3:tullio}, this definition is equivalent
283: to Eq.~\ref{alphaTS} if the curves of $f(k,T)$ as function of $T$ for different materials
284: do not intersect.
285:
286: The quantity $f(k,T)$ (in the low-$k$ limit) can be identified in spin models with
287: the self-overlap of the states as discussed in section \ref{sec1:orderparameter}.
288: Thus, one can define
289: \begin{equation}
290: \label{alphaGP}
291: \alpha(T_g) = 1 - q(T_g) \ ,
292: \end{equation}
293: where $q(T_g)$ is the self-overlap of the equilibrium states at $T_g$, i.e., the value of
294: $q$ where $V(q,T)$ has the secondary minimum at $T=T_g$ (see Fig.~\ref{fig_1}).
295:
296: As the self-overlap of the states is related to their volume in phase space (high overlap
297: corresponding to small states), a small value of $\alpha$ corresponds to
298: small-volume states, while a big value of $\alpha$ corresponds to large-volume states.
299: In this sense, $\alpha(T_g)$ will be called ``volume of the equilibrium states at $T_g$''.
300: Note that a similar identification has been discussed in \cite{TS}: indeed, from
301: Eq.~(7) of
302: \cite{TS}\footnote{Due to a misprint in Eq.(7) of \cite{TS} the power $-1$ has to be disregarded.}
303: one can see that $\alpha$ is related to the curvatures of the minima
304: of the potential (in the harmonic approximation), and that small curvatures (large volume)
305: correspond to large $\alpha$, while high curvatures (small volume) correspond to small $\alpha$.
306: This is consistent with the equivalence of the definition of $\alpha$ given in \cite{TS}
307: and the one adopted here.
308:
309: \subsubsection{Summary of the definitions}
310:
311: To conclude this section, it is useful to give
312: a short summary of all the definition discussed above.
313: Calling $q_{min}(T)$ the value of $q$ where $V(q,T)$ has the secondary minimum,
314: and $q_{max}(T)$ the value of $q$ where $V(q,T)$ has a maximum, the definitions are:
315: \begin{equation}
316: \nonumber
317: \begin{array}{lcl}
318: \Sigma(T) & = & \big[ V(q_{min}(T),T) - V(0,T) \big]/T \\
319: {\cal E}(T) & = & V(q_{max}(T),T)-V(q_{min}(T),T) \\
320: T_K & : & \Sigma(T_K)=0 \\
321: T_g & : & \frac{{\cal E}(T_g)}{T_g \Sigma(T_g)} = {\cal C} \\
322: T_d & : & \EE(T_d)=0 \\
323: m(T_g) & = & 1+T_g \frac{\Sigma'(T_g)}{\Sigma(T_g)} \\
324: \alpha(T_g) & = & 1- q_{min}(T_g)\\
325: \end{array}
326: \end{equation}
327: The constant ${\cal C}$ has to be chosen in order for the fragility to be in the
328: experimentally observed range, $m \sim 1\div 10$.
329:
330:
331: \subsection{Spherical $p$-spin model}
332: \label{sec:sferico}
333:
334: \begin{figure}[t]
335: \includegraphics[width=.5\textwidth,angle=0]{fig-ch3/fig2.eps}
336: \includegraphics[width=.5\textwidth,angle=0]{fig-ch3/fig4.eps}
337: \caption[Transition temperatures, fragility and volume of the states for the spherical $p$-spin]
338: {(Left) Thermodynamic transition temperature $T_K$, glass transition temperature $T_g$
339: and dynamical transition temperature $T_d$, and (right) fragility $m(T_g)$,
340: configurational entropy $\Sigma(T_g)$, ``volume''
341: of the equilibrium states $\alpha(T_g)$ and barrier height ${\cal E}(T_g)$ for the $p$-spin
342: spherical model as a function of $p-2$.
343: }
344: \label{fig_2}
345: \end{figure}
346:
347: The full expression for $V(q,T)$ in the $p$-spin spherical model has been computed
348: in~\cite{FP95,BFP97}.
349: However, a simplified expression can be used when the value of
350: $V(q,T)$ {\it on its stationary points} is considered:
351: \begin{equation}
352: \label{Vq}
353: V(q,T) - F(T) = -\frac{\beta}{4} q^p - \frac{T}{2} \log (1-q) - \frac{Tq}{2} \ .
354: \end{equation}
355: This function has been shown in section~\ref{sec1:compreplica} to coincide with
356: the correct $V(q,T)$ on each stationary point of $V(q,T)$.
357: If one is interested only in the value of $V(q,T)$ on its stationary points, the use of the
358: correct $V(q,T)$ calculated in \cite{FP95,BFP97} or
359: of the one given by Eq.~(\ref{Vq}) gives exactly the same result.
360:
361: Note that, while the model is defined only for integer $p$, Eq.~(\ref{Vq}) makes sense also
362: for real $p$; therefore the behavior of the different quantities for any real
363: $p\geq2$ can be investigated. In particular, the $p\rightarrow2$ limit is interesting
364: being related to
365: a diverging fragility ($T_d\rightarrow T_K$) and to the discontinuous {\sc 1rsb} transition
366: becoming a continuous one.
367:
368: \subsubsection{Temperatures}
369:
370: From Eq.~(\ref{Vq}) one can compute the three temperatures $T_K$, $T_g$ and $T_d$ as
371: functions of $p$. Their behavior is reported in Fig.~\ref{fig_2}. For $p\sim2$,
372: the difference between $T_K$ and $T_g$ is very small, therefore the system is very
373: fragile; for $p\rightarrow\infty$
374: the Kauzmann temperature approaches zero (as $1/\sqrt{\log p}$),
375: while the glass transition temperature remains finite.
376: The system therefore becomes stronger and stronger on increasing $p$.
377:
378: \begin{figure}[t]
379: \centering
380: %\vspace{.05cm}
381: \includegraphics[width=.6\textwidth,angle=0]{fig-ch3/fig3.eps}
382: \caption[Angell plot of the complexity for the spherical $p$-spin]
383: {Scaled plot of the complexity, $\Sigma(T_g)/\Sigma(T)$, as a function of $T_g/T$ for the $p$-spin
384: spherical model at different values of $p$. The figure has to be compared with
385: Fig.~\ref{fig1:Sconf_scaled}; in both figures
386: fragility is the slope of the curves in $T_g/T=1$.
387: The system becomes stronger on increasing $p$.}
388: \label{fig_3}
389: \end{figure}
390:
391: \subsubsection{Complexity and fragility}
392:
393: The same observation can be made more quantitative by considering an ``Angell plot'' for the
394: complexity~\cite{MA01}:
395: in Fig.~\ref{fig_3} the complexity $\Sigma(T)$ is plotted as a function of
396: the temperature, for different values of $p$. The choice of the particular scaling that appears
397: in Fig.~\ref{fig_3} has been made in order to make a close correspondence with
398: Fig.~\ref{fig1:Sconf_scaled}, extracted from \cite{MA01}.
399: The curves for different values of $p$ are ordered from bottom to top.
400: The same behavior is observed in glass formers of different fragility. Indeed, the index
401: of fragility defined in Eq.~(\ref{fragilita}) is exactly one plus the slope in $T_g/T=1$
402: of the curves in Fig.~\ref{fig_3}:
403: \begin{equation}
404: m(T_g) =1+T_g \frac{\Sigma'(T_g)}{\Sigma(T_g)}
405: = 1 + \left. \frac{d [\Sigma(T_g)/\Sigma(T)]}{d [T_g/T]} \right|_{T=T_g} \ .
406: \end{equation}
407: The fragility index $m$ is shown in Fig.~\ref{fig_2} as a function of $p$. It is
408: a decreasing function of $p$. Its values are in the range observed for experimental system due
409: to the (arbitrary) choice of the constant ${\cal C}$ appearing in Eq.~(\ref{defTg}),
410: ${\cal C}=0.1$.
411: In Fig.~\ref{fig_2} $\Sigma(T_g)$ is also reported as a function of $p$. It is an
412: increasing function of $p$, that diverges as $\log p$ for $p \rightarrow \infty$:
413: thus, the number of states in this system is a decreasing function
414: of the fragility\footnote{A review of some results on the correlation between fragility
415: and number of states can be found in~\cite{noifrag}.}.
416:
417: \subsubsection{Barrier heights and volume of the states}
418:
419: In Fig.~\ref{fig_2} the barrier height ${\cal E}(T_g)$ is also reported as a function of $p$,
420: together with the index $\alpha(T_g)=1-q(T_g)$ that was called ``volume'' of the equilibrium
421: states at $T_g$. In this model the states become smaller on increasing $p$,
422: while the barriers separating them increase.
423: The correlations between these quantities will be discussed in section~\ref{correlazioni},
424: where a geometric description of the evolution of the phase space of this model
425: at different $p$ will be proposed, that relates fragility to geometric properties
426: of the phase space.
427:
428: \subsection{Ising $p$-spin model}
429: \label{sec:ising}
430:
431: The Ising $p$-spin model is another popular model for the study
432: of the glass transition \cite{KTW87a,KTW87b,Derrida}.
433: Its Hamiltonian is given by Eq.~(\ref{Hpspin}), where the variables $\sigma_i$ are Ising spins,
434: $\sigma_i = \pm 1$, and the spherical constraint is absent.
435: For the Ising $p$-spin model, the two-replica potential $V(q,T)$ is given by
436: \begin{equation}
437: V(q,T)-F(T)=\beta \frac{p-1}{4} q^p + \beta \frac{p}{4} q^{p-1}
438: - \frac{\int {\cal D}z \cosh(\Lambda z) \log \cosh (\Lambda z)}{\int {\cal D}z \cosh(\Lambda z)} \ ,
439: \end{equation}
440: where ${\cal D}z = \exp(-z^2/2) \ dz$, and $\Lambda^2 = \beta^2 \frac{p}{2} q^{p-1}$.
441:
442: Note that the total number of states in the Ising $p$-spin model cannot be greater than $2^N$
443: (the total number of configurations), and hence $\Sigma(T) \leq \log 2$,
444: while in the spherical model $\Sigma(T_g)$
445: diverges as $\log p$ for $p \rightarrow \infty$, as previously discussed.
446:
447: \subsubsection{Temperatures}
448:
449: \begin{figure}[t]
450: \includegraphics[width=.5\textwidth,angle=0]{fig-ch3/fig5.eps}
451: \includegraphics[width=.5\textwidth,angle=0]{fig-ch3/fig6.eps}
452: \caption[Transition temperatures, fragility and volume of the states for the Ising $p$-spin]
453: {(Left) Thermodynamic transition temperature $T_K$, glass transition temperature $T_g$
454: and dynamical transition temperature $T_d$ and (right) fragility $m(T_g)$,
455: configurational entropy $\Sigma(T_g)$, ``volume'' of the equilibrium
456: states $\alpha(T_g)$ and barrier height ${\cal E}(T_g)$ for the $p$-spin Ising model as a function
457: of $p-2$.}
458: \label{fig_5}
459: \end{figure}
460:
461: The first consequence of this difference is observed when studying the transition temperatures
462: as functions of $p$ (see Fig.~\ref{fig_5}).
463: Indeed, as in the spherical model, $T_K \sim T_g$ for $p \sim 2$,
464: and $T_g \gg T_K$ for $p \rightarrow \infty$. But, in this model, $T_K$ tends to a finite value
465: at large $p$, while $T_g$ and $T_d$ diverge.
466: This behavior can be understood recalling that for a ``Gaussian-like'' model one has
467: $T_K \sim 1/\sqrt{\Si_\io}$, $\Si_\io$ being the total number of
468: states, i.e. the maximum of $\Sigma(T)$ \cite{noifrag}.
469:
470: \subsubsection{Complexity and geometric properties of the phase space}
471:
472: The ``Angell plot'' for the complexity of the Ising $p$-spin model looks very similar to the
473: one of the spherical model, Fig.~\ref{fig_3}, so it is not useful to report it.
474:
475: Having fixed an appropriate value for the constant ${\cal C}$ in Eq.~(\ref{defTg})
476: (${\cal C}=0.02$, different from the value chosen in the spherical case),
477: the behavior of the fragility as a function of $p$ is also very similar to the one of the
478: spherical model. The same behavior is found
479: for the other quantities under study, as one can deduce from
480: a comparison of Fig.~\ref{fig_5} and Fig.~\ref{fig_2}, the main difference being the discussed
481: behavior of $\Sigma(T_g)$ at large $p$.
482:
483: \subsubsection{Vibrational properties and volume of the states}
484:
485: Another relevant difference between the spherical and the Ising model is that, in the latter,
486: harmonic vibrations are not present (the variables being discrete): we have
487: $q(T) \rightarrow 1$ exponentially for $T \sim 0$, and the definition of $\alpha$
488: via Eq.~(\ref{alphaTS})
489: gives $\alpha = 0$ for all $p$. However, the definition given in Eq.~(\ref{alphaGP}) and used in
490: these calculations gives a reasonable result also in absence of harmonic vibrations.
491:
492: \section{Correlations between different properties of the phase space}
493: \label{correlazioni}
494:
495: In this section the correlations between the quantities under study will be investigated,
496: trying to relate fragility to the phase space geometry.
497: The results will be compared with the general consideration of~\cite{noifrag},
498: and with the experimental results of~\cite{TS}.
499:
500: \subsection{Fragility and volume of the states}
501:
502: In~\cite{TS} it has been established that fragility is positively correlated with the index
503: $\alpha$ defined in section~\ref{sec:definitions}.
504: In other words, {\it fragile systems have large basins while strong systems have small basins}.
505: In Fig.~\ref{fig_7} the fragility $m$ is plotted as a function of $\alpha$ parametrically in $p$
506: for the investigated systems. The curve $m(\alpha)$ is very similar for
507: the two models - remember that the only adjustable parameter is the constant ${\cal C}$
508: in Eq.~(\ref{defTg}).
509: By comparison with Fig.~\ref{fig3:tullio},
510: one can conclude that the model has a behavior similar to the one of real systems.
511: Surprisingly, also the linear correlation between $m$ and $\alpha$ is reproduced for
512: $\alpha \leq 0.4$.
513: Thus, mean field $p$-spin models are able to describe the relation between fragility and
514: the volume of the basins visited around $T_g$ found in~\cite{TS}.
515:
516: \subsection{Fragility, barrier heights and number of states}
517:
518: \begin{figure}[t]
519: \centering
520: %\vspace{.05cm}
521: \includegraphics[width=.6\textwidth,angle=0]{fig-ch3/fig7.eps}
522: \caption[Correlation between fragility and $\alpha$ for $p$-spin models]
523: {Fragility versus $\alpha$ for the spherical and Ising $p$-spin models.
524: The curve is very similar for the two models,
525: and is consistent with the correlation found in \cite{TS}, see Fig.~\ref{fig3:tullio}.
526: The linear correlation is reproduced for $\alpha \leq 0.4$.}
527: \label{fig_7}
528: \end{figure}
529:
530: It has been conjectured that fragile systems have a larger number of states than strong ones,
531: even if the total number of states is not an experimentally accessible quantity and numerical
532: simulations give contradictory results~\cite{Sp99,Sa01,DS01}.
533: However, in the models considered here the behavior is exactly the opposite. In Fig.~\ref{fig_8}
534: $\Sigma(T_g)$ is reported as a function of the fragility: the total number of states
535: is a decreasing function of the fragility.
536:
537: This point was discussed in detail in \cite{noifrag}, where the possibility
538: of correlating fragility with the total number of states for general models of $\Sigma(T)$ was
539: discussed, assuming the validity of the Adam-Gibbs relation, Eq.~(\ref{AG}).
540: The conclusion was that the knowledge of the distribution of states is not enough to determine
541: the fragility. Indeed, the relevant
542: parameter, for a general ``Gaussian-like'' distribution of states, is
543: \begin{equation}
544: \label{D}
545: D = \frac{{\cal E}(T_g)}{T_K \Sigma(T_g)} \ .
546: \end{equation}
547: Note that in Eq.~(\ref{D}) ${\cal E}$ has to be evaluated at $T=T_g$ because in the considered models
548: the barrier height ${\cal E}$ is a $T$-dependent quantity, while in the Adam-Gibbs relation
549: it is usually assumed to be a constant. However, the Adam-Gibbs relation
550: has been tested around $T_g$, therefore, to a good approximation, one can fix ${\cal E}$ to be a
551: constant equal to its $T=T_g$ value.
552: The parameter $D$ is inversely proportional to the fragility $m$: therefore $m \sim \Sigma/{\cal E}$,
553: and fragility turns out not to be simply correlated to the total number of states.
554: If the ``barrier heights'' grow faster than the total number of states,
555: fragility can be a decreasing function of $\Sigma$:
556: this is indeed the case in the considered models.
557: \begin{figure}[t]
558: \centering
559: %\vspace{.05cm}
560: \includegraphics[width=.6\textwidth,angle=0]{fig-ch3/fig8.eps}
561: \caption[Correlation between fragility and number of states in $p$-spin models]
562: {Total number of states (represented by the complexity at $T_g$) as a function of
563: the fragility $m$ for the $p$-spin models.}
564: \label{fig_8}
565: \end{figure}
566: Indeed, from Fig.~\ref{fig_2} and Fig.~\ref{fig_5}, one observes that the barrier height is an
567: increasing function of $p$.
568: Using Eq.~(\ref{defTg}), Eq.~(\ref{D}) can be written as
569: \begin{equation}
570: D = {\cal C} \frac{T_g}{T_K} \ .
571: \end{equation}
572: Therefore, from Fig.~\ref{fig_2} and Fig.~\ref{fig_5}, $D$ is an increasing
573: function of $p$ that diverges for $p \rightarrow \infty$, as the ratio $T_g/T_K$ increase on
574: increasing $p$ for both models.
575: Thus, in the considered models the height of the barriers (in units of $T_K$)
576: increases faster than the total number of states. This explains why one observes an inverse
577: correlation between fragility and the total number of states, as discussed above and
578: in~\cite{noifrag}.
579:
580: \subsection{A geometric picture of the phase space}
581:
582: \begin{figure}[t]
583: \centering
584: %\vspace{.05cm}
585: \includegraphics[width=.6\textwidth,angle=0]{fig-ch3/fig9.eps}
586: \caption[Sketch of the evolution with $p$ of the $p$-spin free energy landscape]
587: {Sketch of the evolution with $p$ of the $p$-spin free energy landscape: at small $p$
588: there is a small number of states of large volume separated by low barriers; at high $p$
589: there is a large number of states of small volume separated by high barriers. The height
590: of the barriers increase faster than the number of states: thus, fragility is a decreasing
591: function of $p$.}
592: \label{fig_9}
593: \end{figure}
594:
595: The information obtained in the previous sections can be collected in
596: a geometric picture of the evolution with $p$ of the $p$-spin model free energy landscape.
597: Indeed, on increasing $p$: \\
598: {\it i)} The total number of states increases. \\
599: {\it ii)} The volume of the states decreases ($\alpha$ decreases). \\
600: {\it iii)} The height of the barriers between states increases. \\
601: Thus, we get the picture of a landscape where, on increasing $p$, a great number of small
602: states with very high curvatures and separated by very high barriers appear: a sketch of this
603: evolution is given in Fig.~\ref{fig_9}.
604: The behavior of the fragility in this situation is related to the behavior of $\Sigma/{\cal E}$,
605: the ratio between number of states and height of the barriers between them: in these models,
606: it turns out that ${\cal E}$ increase faster than $\Sigma$, and the fragility is a decreasing
607: function of $p$.
608:
609: This behavior is consistent with the fact that fragility turns out to be
610: positively correlated with the ``volume'' of the states as measured by $\alpha$. Indeed,
611: if, on the contrary, the barrier height grew slower than the total number of states
612: (equivalently, if $m$ would be positively correlated with the total number of states),
613: there should be also an inverse correlation between $m$ and $\alpha$, in disagreement with what
614: is experimentally observed.
615:
616: In the $p\rightarrow 2$ limit, where the fragility becomes infinite, the second
617: derivative with respect to $q$ of the potential $V(q,T)$ calculated in
618: $q=0$ and $T=T_K=T_d$ vanishes (see Fig.~\ref{fig_1})
619: and the so-called spin glass susceptibility
620: diverges at the critical temperature. In other words when the fragility becomes
621: infinite soft modes appear at the critical temperature supporting the
622: previously presented physical picture.
623:
624: Note that the outlined picture is valid for ``Gaussian-like'' models, i.e., models where
625: the complexity is a concave function of the temperature that vanishes at $T_K$ without any
626: inflection point. These models seem to describe correctly the distribution of basins in real
627: systems only for relatively high fragilities. The behavior of the complexity (or configurational
628: entropy, or excess entropy) as a function of temperature for very strong systems is still an
629: open problem, see \eg~\cite{SPS01,MBLSSTZ04}; our discussion may not apply to these systems.
630:
631: The main prediction of this analysis is that the total number of states $\Si_\io$
632: and the Adam-Gibbs barrier ${\cal E}$ should both be decreasing functions of the
633: fragility. This prediction disagrees with the statement of \cite{Sp99,Sa01,DS01} that
634: fragile systems should have an higher number of states with respect to strong ones.
635: A critical analysis of this works can be found in \cite{noifrag}. Unfortunately,
636: existing data are not sufficient to strictly test this prediction; the excess entropy
637: is available only for few experimental systems, and numerical simulations are performed
638: in a temperature range where the fragility of the investigated systems is approximately
639: the same. Hopefully this predictions will be tested in the future.
640:
641: