cond-mat0202119/2d.tex
1: \documentstyle[aps,pre,epsf]{revtex}
2: 
3: \begin{document}
4: 
5: \twocolumn[\hsize\textwidth\columnwidth\hsize\csname
6: @twocolumnfalse\endcsname
7: 
8: \title{Stable propagation of an ordered array of cracks during directional drying}
9: \author{E. A. Jagla}
10: \address{Centro At\'omico Bariloche and Instituto Balseiro, 
11: Comisi\'on Nacional de Energ\'{\i}a At\'omica, (8400) Bariloche, Argentina}
12: \maketitle
13: \begin{abstract}
14: We study the appearance and evolution of an array of parallel 
15: cracks in a thin slab of material that is directionally dried, and show that 
16: the cracks penetrate the material uniformly
17: if the drying front is sufficiently sharp.
18: We also show that cracks have a tendency to become evenly spaced 
19: during the penetration.
20: The typical distance between cracks 
21: is mainly governed by the typical distance
22: of the pattern at the surface, and 
23: it is not modified during the penetration. 
24: Our results agree with recent experimental work, and can be 
25: extended to three dimensions to describe the properties 
26: of columnar polygonal patterns observed in some geological
27: formations.
28: \end{abstract}
29: 
30: 
31: %\newpage
32: 
33: \vskip2pc] \narrowtext
34: 
35: \section{Introduction}
36: 
37: Cracks appearing when a material shrinks are
38: common in everyday life. The most popular examples are probably 
39: the cracks appearing in
40: paint layers, and those in the surface of muds. 
41: In the case of paints, there is 
42: a superficial layer of material that shrinks on top of a substrate
43: to which it is attached. For muds, the superficial layer and the substrate
44: are the same material, but difference in the humidity concentration produces
45: the cracking at the surface.
46: In these two cases, the shrinking is due to changes in the humidity 
47: concentration within the sample, but it can also be due to the existence
48: of non-uniform temperature distributions, which produces stresses and 
49: generates cracking. The problem of surface fragmentation 
50: has been studied in the last
51: years, both theoretically\cite{theor} and experimentally\cite{exper}.
52: 
53: There are situations in which cracks appear at the surface 
54: of the material, and penetrate into the sample later on. A well known
55: example corresponds to the columnar fracturing of basaltic lava, 
56: seen in many different geographical locations\cite{basalts}.
57: A detailed description of this problem has only recently been 
58: foreshadowed\cite{jr0,jr1}, and still some points remain obscure.
59: The two dimensional equivalent of the columnar structure of basalts
60: corresponds to the case of a two dimensional material that dries
61: (or cools down) starting from a free edge. 
62: As in the three dimensional case, drying creates internal stresses 
63: in the material that generate cracks, first appearing at the free edge. 
64: As the drying front propagates to the interior the cracks also do so in turn.
65: There have been different experimental realizations of this phenomenon.
66: In one of them\cite{alan-lim} the material was a very thin layer of a colloidal dispersion
67: placed between two glasses, with one free edge, from which humidity can escape. 
68: In other experiments\cite{canad}, a slurry of Al$_2$O$_3$-water mixture 
69: was deposited onto a substrate, and
70: a glass was placed a few millimeters above it. A low, ultra dry $N_2$ gas breeze was
71: injected in the slot above the sample. The $N_2$ became saturated with water 
72: as it passed over the
73: material, and a rather sharp drying front propagates with time 
74: in the same direction than the air flow. A third realization corresponds
75: to the drying of thin films of aqueous silica sol-gel, where very nice patterns
76: have been observed\cite{hull}. In all these cases the propagation of a set
77: of parallel cracks has been observed.
78: 
79: A detailed theoretical description of this phenomenon is lacking, and
80: we will make an attempt in this direction for the two dimensional case. 
81: In the next section we will make a general description of the phenomenon, 
82: emphasizing the difference with a more standard fracture mechanics problem.
83: In Section III we present the calculations of elastic and crack energy
84: for the case in which the drying (or cooling) front is sharp. In Section IV
85: we discuss the conditions under which the crack front (formed by the
86: tips of all cracks) is stable, and progresses into the material when the
87: drying front advances.
88: In Section V we show that the array of cracks 
89: generated at the surface of the material becomes evenly spaced
90: when penetrates the sample, and that the typical width of the stripes is
91: determined mainly during the first stage of the process, near the surface.
92: In Section VI we briefly comment upon the effects of other drying
93: (or cooling) conditions than the one studied previously.
94: Section VII contains some implications of our work for the study of the
95: three dimensional case.
96: Finally, in Section VIII we summarize and conclude.
97: 
98: \section{Quasistatic fracture mechanics. Energetic and stress analysis}
99: 
100: One remarkable thing about the experiments cited in the previous 
101: section \cite{alan-lim,canad,hull} is the fact that a large number 
102: of cracks penetrate the sample in a coordinated and quasistatic manner, 
103: as the external conditions (the humidity profile) change. Then
104: this problem is qualitatively different of a typical problem in 
105: fracture mechanics, where the propagation of a crack is usually an
106: abrupt phenomenon and typically leads to the failure of the material.
107: In standard fracture mechanics, stress and energetic analysis
108: are two different ways of predicting the evolution of
109: the fracturing process\cite{broberg}. However, the equivalence of both
110: approaches is not clear\cite{setna}, particularly in cases in which
111: the cracks propagate at large speeds, which is almost always the case when
112: failure occurs.
113: 
114: The situation is different in our case. Cracks propagate
115: only because the external humidity profile changes. If the external conditions
116: were stationary, crack advance would be arrested. 
117: If we think on the configuration of the system as a point
118: in configuration space (the space spanned by all coordinates of all particles of 
119: the material), at each moment the system is at one minimum of the
120: energy landscape. As the humidity profiles changes the landscape changes itself,
121: the minimum on which the system is located shifts,
122: and the configuration of the system adapts so as to remain in the shifted
123: local minimum. This is what we understand as a quasistatic propagation
124: of cracks.\cite{nota10}
125: 
126: Under these conditions the propagation of cracks in a two dimensional
127: geometry can be studied by two
128: different but equivalent procedures:
129: Stress analysis consists on the calculation of the stress
130: intensity factors\cite{broberg} $K_I$ and $K_{II}$ of the opening and
131: shearing modes at the tips of 
132: the cracks present in the system. The propagation will occur 
133: in the direction along which $K_{II}=0$, which coincides with
134: that for which $K_I$ is maximum. Propagation actually occurs only if the
135: energy relieved by the advance of the cracks is enough to overcome
136: the fracture energy needed to elongate the cracks. In the quasistatic
137: case, these two energies will differ only infinitesimally. 
138: 
139: In the energetic procedure to calculate crack advance, the total energy 
140: (including elastic and crack energy) after virtual advances of the cracks
141: is calculated. Cracks will actually propagate 
142: only if this propagation reduces the total energy. Under quasistatic
143: advance the energy reduction during propagation is infinitesimally small,
144: and typically, there is only one possible direction of propagation for each crack,
145: all other propagation directions would produce an
146: increase of the total energy of the system. 
147: 
148: 
149: Both approaches are equivalent in the case of quasistatic advance of the cracks.
150: We will use stress analysis or energetic analysis according to
151: convenience in each case.
152: 
153: 
154: \section{Estimations of elastic and crack energy}
155: 
156: We will consider an isotropic and homogeneous material, and assume that linear 
157: elasticity can be applied\cite{elasticidad}. 
158: It is well known  that in this case the material possesses
159: only two independent elastic constants\cite{landau}. We will take as 
160: these two parameters the bulk modulus 
161: $C$ and the Poisson ratio $\nu$. The expansion or contraction
162: properties of the material are described by a humidity
163: expansion coefficient $\alpha$, which is formally equivalent to the 
164: thermal expansion coefficient, i.e., the relative change of
165: linear size  $\delta L/L$ of a piece of material after changing 
166: the humidity concentration $h$ by some quantity $\Delta h$ is given by
167: \begin{equation}
168: \delta L/L=\alpha \Delta h.
169: \end{equation}
170: 
171: The ideal situation we will address consists of a semi-infinite 
172: two-dimensional sample that is being dried from its surface. 
173: The drying 
174: will be considered to be non-homogeneous, and we will model it 
175: by a humidity profile that is given by some 
176: function $h(z,t)$ depending on depth $z$ within the material, and time $t$. 
177: The precise form of this function
178: will be specified later for different experimental situations, but it is 
179: important to point out that we do not consider the case in which the 
180: appearance of cracks in the system modifies itself the drying process.
181: Then our analysis applies more to cases as that described in \cite{canad} and
182: possibly \cite{hull}, but much less
183: to that in \cite{alan-lim} where evaporation of humidity through the cracks 
184: seems to be relevant(see also \cite{bahr}). Although in the experimental situation the material 
185: usually lies on top of a substrate, we will study the case in which there 
186: is no interaction with this substrate. In the experiments of Ref. \cite{canad}
187: this is achieved by introducing a layer of some slippery material
188: between the sample and the substrate. Then, all the stresses on the sample
189: are originated internally, and are due to the existence of a non-uniform 
190: humidity concentration.
191: 
192: It is known that in an isolated (namely, not clamped) piece of 
193: homogeneous material placed under a constant thermal (or humidity) gradient
194: all stresses vanish in linear elastic approximation \cite{lineart}. Under these
195: conditions cracks cannot appear at all, or if already present from the beginning
196: its propagation is completely halted. It is in fact 
197: crucial for the propagation of cracks that the humidity gradient it is not constant
198: within the sample.
199: 
200: We will take the edge of the semi-infinite sample as
201: the $z=0$ line, and $z>0$ in the interior. For convenience, we will also
202: refer to the edge as the `surface', and consider it to be horizontally
203: placed, in such a way that the cracks propagate down the material.
204: At $t=0$ the whole material is supposed to have 
205: a constant humidity concentration $h_0$. In this situation the material is unstressed.
206: Humidity concentration is assumed to decrease with 
207: time $h(z,t>0)\le h_0$. For any reasonable 
208: experimental realization of the
209: drying process occurring from the $z=0$ free surface, it is clear that 
210: at any time, well inside the material we should reach the original 
211: humidity concentration, i.e., $h(z\rightarrow \infty,t)=h_0$. Then 
212: the majority of the material is always at constant humidity $h_0$. 
213: In this region the sample must be unstressed, otherwise it will 
214: store an infinite amount
215: of energy. Then a boundary condition for our problem will be that 
216: stresses go to zero as $z$ go to infinity.
217: As a simplification of a possible experimental situation,
218: we will consider the case of an abrupt 
219: drying profile (see Fig. \ref{f0}),
220: namely $h(z,t)=h_0$ for $z>z_0(t)$, and $h(z,t)=h_1$ 
221: for $z<z_0(t)$, with $h_1<h_0$, and $z_0(t)$ being an increasing function 
222: of time. 
223: 
224: A given set of cracks will always correspond to a local minimum
225: of the total energy of the system.
226: %This is a reasonable model
227: %for the results on directional drying of Ref. \cite{canad}. 
228: The total energy is the sum of two well different parts. One is the fracture
229: energy, namely, the energy spent in the creation of all fractures present in the
230: sample. This is typically proportional to the total length of cracks in
231: the material.
232: The second part is the elastic energy stored in the sample.
233: %This is a very important characteristic of a given pattern of cracks 
234: %since it is crucial to identify whether the 
235: %advance of cracks will occur or not.
236: Let us suppose we have an evenly spaced
237: array of cracks, defining stripes of width $l$, which have penetrated down to
238: some distance $\bar z$, with the drying front being located 
239: at some position $z_0$.
240: 
241: The elastic energy $e_{\rm el}$ stored in the material per unit
242: of horizontal length must be proportional to the bulk modulus of the 
243: material $C$, and to the second power of the typical change in linear
244: density caused by the humidity gradient (this change being $\alpha \Delta h$, with
245: $\Delta h\equiv (h_0-h_1)$). This is at the basis of linear elasticity
246: theory. Using the fact that $l$
247: and $\bar z-z_0$ are the only two relevant lengths for this geometrical
248: configuration\cite{zz},
249: dimensional analysis allows to 
250: write down the following expression for $e_{\rm el}$ 
251: 
252: \begin{equation}
253: e_{\rm el}=(\alpha \Delta h)^2 C l {g} \left( (\bar z-z_0)/l\right),
254: \label {eelastica}
255: \end{equation}
256: where $g(x)$ is 
257: a dimensionless geometrical function\cite{poisson}. 
258: We have determined $g$ numerically
259: and the result is shown in Fig. \ref{f2} as a continuous line.
260: We can rationalize the general form of the function $g(x)$,
261: considering how the elastic energy is distributed along the $z$
262: direction. 
263: The total energy $e_{\rm el}$ is the integral over $z$ of 
264: the density of elastic energy $\delta e_{\rm el}(z)$.
265: In  Fig. \ref{f1} we show qualitatively the form of $\delta e_{\rm el}(z)$
266: for the cases 
267: $\bar z\ll z_0$, and $\bar z\gg z_0$. 
268: Let us consider first the case $\bar z\ll z_0$ (Fig. \ref{f1}(a)).
269: $\delta e_{\rm el}(z)$ is zero for $z> z_0$, since as we already 
270: discussed, the material has to be unstressed for $z\rightarrow \infty$. 
271: In the region $z_0>z>\bar z$ there is a rather constant energy density, 
272: associated to the change in humidity concentration, which cannot be compensated
273: by a change in density of the material since the material here is attached to
274: the part below $z_0$\cite{nota2}.
275: Around the position of the
276: crack front there is an increase of the stored elastic energy, 
277: which is associated to the elastic energy around the
278: tips of the cracks. For $z<\bar z$ the elastic energy density 
279: goes to zero, since here the existence of the cracks 
280: has allowed to relieve the elastic energy accumulated 
281: previously to the cracks formation. 
282: The linear dependence of
283: $g(x)$ as $x\rightarrow -\infty$ (i.e. for $\bar z\ll z_0$)
284: comes from the energy stored between 
285: $\bar z$ and $z_0$.
286: Let us consider now the case when $\bar z\gg z_0$ 
287: (Fig. \ref{f1}(b)). The elastic energy
288: becomes independent of $\bar z$, as the crack tips are in a region of material
289: that is unstressed. The constant value of $g(x)$ as $x\rightarrow +\infty$
290: comes from the energy stored in the 
291: independent stripes around the position of the humidity front, 
292: which in this limit is well behind the crack front. When $\bar z\simeq z_0$
293: there is a smooth crossover between the two limiting regimes.
294: 
295: To determine the actual position of the crack front in a realistic situation 
296: we will rely on the energetic argument. 
297: At any time during the drying process, the crack front will be located
298: at the position that minimizes the total energy of the system.
299: Equation (\ref{eelastica}) gives the elastic energy of a set of cracks
300: that have penetrated down to the position $\bar z$. 
301: In order to get the total energy, $e_{\rm el}$ 
302: has to be added to the energy cost of creating the cracks. This 
303: part, when measured
304: per unit of horizontal length, will be called the crack energy $e_{\rm ck}$, and
305: it is simply given in terms of the specific energy fracture of the
306: material $\eta$ in the form
307: 
308: 
309: \begin{equation}
310: e_{ck}=\eta \bar z/l 
311: \label {ecracks}
312: \end{equation}
313: In order to determine the value of $\bar z$ at which the fracture front prefers
314: to be located, we have to minimize the total energy 
315: $e_{\rm tot}=e_{\rm el}+e_{\rm ck}$.
316: The result we obtain is shown in Fig. \ref{f3}, where we plot the most convenient
317: position of the crack front  
318: as a function of the parameter $u\equiv\eta /[C(\alpha \Delta h)^2 l]$ .
319: As we see, for very small value of $u$, $x\equiv (\bar z-z_0)/l$ 
320: takes large and positive
321: values, i.e., the crack front is located well below the humidity front. This
322: is due to the negligible contribution of the crack energy compared to the 
323: elastic energy in this case. As $u$ increases $x$ decreases, crossing zero
324: (namely, the crack front coinciding with the humidity front) at $u\simeq 0.66$.
325: $x$ tends to $-\infty$  as $u$ approaches the limiting value $u_0\simeq 1.14$. 
326: In fact, for $u>u_0$ the crack energy is so high that cracks do not penetrate
327: the sample at all\cite{vidrio}. 
328: 
329: 
330: Under the conditions analyzed in this section, the crack front will be
331: located at a position such that $(\bar z-z_0)/l$ is given by the function plotted in
332: Fig. \ref{f3}. The crack front advances only due to changes in the position of the 
333: humidity front $z_0$, keeping always $\bar z-z_0$ as constant.
334: 
335: 
336: \section{Stability of the flat crack front}
337: 
338: The previous analysis has assumed that all cracks penetrate down to
339: a uniform depth $\bar z$, and has focused on what the value of $\bar z$ is, on
340: energetic grounds. It has to be complemented however with 
341: a stability analysis of the crack front. The necessity of this is clear
342: from the following example. If a material with an array of vertical cracks is 
343: loaded with a uniform horizontal stress (this situation can be
344: thought to be realized in our case
345: if $\bar z\ll z_0$), there will be 
346: typically a single crack that propagates and fractures the material. This
347: is a consequence of the fact that as soon as a single crack moves forward
348: a small distance, the stress at its tip increases, and those at the tips
349: of the other cracks decrease. This generates an unstable, rapid propagation
350: of a single crack. We will see that in our case, this can be compensated by the
351: fact that stresses decrease ahead of the humidity front, and this can 
352: stabilize a flat crack front.
353: 
354: The same kind of energetic arguments used in the previous section 
355: will be used to determine the stability conditions of the crack front.
356: Consider an evenly spaced set of cracks, labelled sequentially
357: by an index $j$, where now the tips of the cracks are at vertical positions $z^j$,
358: which can be slightly displaced from the mean position $\bar z$, i.e., 
359: $z^j=\bar z+\delta_ij$, with $\sum_j \delta_j=0$. The horizontal 
360: positions of the cracks are given by $x^j=jl$. The elastic energy of 
361: this configuration contains a term of the form (\ref{eelastica}), plus a correction
362: that can be expanded in powers of $\delta_j/l$. The first order term
363: in this expansion vanishes, as $\sum_j \delta_j=0$. The second order term can
364: be written in the form
365: \begin{equation}
366: \Delta e_{\rm el}=l^{-2}(a\sum_j \delta_j^2+b\sum_j \delta_j\delta_{j+1}
367: +c\sum_j \delta_j\delta_{j+2}+...)
368: \end{equation}
369: Successive term contain `interactions' between more distant cracks.
370: As the elastic energy is a local quantity, we expect $|a|<|b|<|c|<...~$.  
371: Any small displacement of the crack front, defined by the quantities
372: $\delta_j$ can be decomposed in a sum of `normal modes', by going
373: to the Fourier representation
374: \begin{equation}
375: \delta_j\equiv \int dk \tilde \delta_k \exp (ikx^j),
376: \end{equation}
377: and the energy decomposes into independent term, in the form
378: \begin{equation}
379: \Delta e_{\rm el}\sim \sum_k |\tilde \delta_k|^2 (a+b\cos (2\pi k/l)+c\cos (4\pi k/l)+...).
380: \end{equation}
381: In order for the flat crack front to be stable, $\Delta e_{\rm el}$ must be positive
382: for any choice of the $\delta _i$, and thus of the $\tilde \delta _k$.
383: This implies that 
384: \begin{equation}
385: (a+b\cos (2\pi k/l)+c\cos (4\pi k/l)+...)
386: \label{abc}
387: \end{equation}
388: must be positive for all $k$. In the limit of very small $k$, 
389: $\Delta e_{\rm el}$
390: is equivalent to a uniform advance of the crack front, and then is 
391: has to be a positive quantity, as the curvature of $e_{\rm el}(\bar z)$ is
392: always positive (see (\ref{eelastica}) and Fig. \ref{f2}).
393: On the other hand, It will be proportional to the 
394: sum of all coefficients in expression (\ref{abc}), i.e., 
395: $\Delta e_{\rm el}^{k\rightarrow 0}\sim a+b+c+...$.
396: Then it is clear that if an instability exists for some 
397: value of $k$, it will occur at $k=\pi/l$, where 
398: $\Delta e_{\rm el}^{k=\pi/l}\sim a-b+c-...~$.
399: Then we will analyze the stability of the flat crack front against 
400: a perturbation with $k=\pi/l$.
401: We took the equilibrium position of 
402: a crack front obtained in the previous
403: section (plotted in Fig. \ref{f3}) and calculated numerically 
404: the quadratic change in energy $\Delta e_{\rm el} = \chi(x) 
405: (\alpha \Delta h)^2 C  \delta ^2/l$ 
406: when a perturbation of $k=\pi/l$ and amplitude 
407: $\delta$ is introduced.
408: The dimensionless function $\chi (x)$ can be seen in Fig. \ref{f2} (dashed line).
409: We see that $\chi$ is positive 
410: (negative) for $x$ greater (lower) than $x_{\rm cr}\simeq -0.038$.
411: In this way we obtain that there are two regions in the plot of Fig. \ref{f3}.
412: The one at the left of $u_{\rm cr}\simeq 1.02$ corresponds to a stable situation. 
413: If the parameters 
414: of the system make the pattern of cracks to lie in that region, then the
415: time evolution of the drying process (i.e., the increase with time of $z_0$)
416: will produce a smooth advance of the crack front, keeping always the
417: same value of the distance $\bar z-z_0$ to the humidity front. 
418: At the right of $u_{\rm cr}$, the pattern is unstable. Should we have
419: one of those patterns at a given time, it will immediately propagate
420: forward some of its cracks (ideally, one of each two cracks), 
421: in order to reach a stable situation. 
422: This will imply in particular that some cracks will remain halted. The
423: further evolution of the crack front will correspond to a new crack pattern
424: with less cracks (i.e., with larger $l$) being propagated. 
425: %Note in fact
426: %in Fig. \ref{f3}  that a sufficiently large reduction of $l$ (i.e., stopping enough
427: %cracks) can increase $u$
428: %in such a way of shifting the pattern into the region of stability.
429: 
430: 
431: \section{Appearance and ordering of cracks}
432: 
433: Up to now we have assumed that a set of evenly spaced cracks exists,
434: and we focused on its stability conditions. We will study now how this
435: pattern can appear, starting with the process at the surface.
436: Cracks are not expected to appear evenly spaced at the surface.
437: In fact, it is known
438: in one-dimensional models of surface fragmentation \cite{1d} (that can 
439: be used to represent
440: the first stage of cracking of our two-dimensional
441: problem) that the distribution of fragment length is strongly 
442: dependent on the presence of small inhomogeneities in the material, 
443: making the fragment length distribution broad. But we will see that
444: as the superficial cracks penetrate the sample they become evenly
445: spaced. 
446: We will first discuss how the process of nucleating new cracks at the
447: surface is, and then argue that the cracks become evenly spaced as they penetrate
448: the sample.
449: 
450: \subsection{The appearance of cracks at the surface}
451: 
452: When the drying front penetrates the material from its surface, and before
453: the material gets cracked, the stresses can be very simply 
454: calculated.
455: The state of the system corresponds to the material being completely unstrained 
456: horizontally
457: (considering the unstrained state as that corresponding to the humidity value
458: $h_0$ that occurs well inside the material). 
459: Under this condition a uniform horizontal
460: stress $T_0$ appears for all $z<z_0$, 
461: which is simply calculated as $T_0=C\alpha \Delta h$. 
462: In order for the first crack to appear, this value has to overcome 
463: the uniform traction resistance of the material $T_r$ \cite{microck}: 
464: \begin{equation}
465: C\alpha \Delta h>T_r. 
466: \label{tr}
467: \end{equation}
468: If this relation is not satisfied, no cracks will appear whatsoever. 
469: Assuming that the relation (\ref{tr}) is satisfied, 
470: the first crack will nucleate at the surface. This crack
471: penetrates the sample as long as this penetration reduces the total energy 
472: of the system. In our case, the crack penetrates only 
473: down to a distance $d_0$ of the order of $z_0$, where the humidity front 
474: is located%
475: %\cite{b2}
476: . The horizontal stress at the surface is now zero at the position of the
477: crack, and increases as we move away from it, reaching the value $F_0$
478: at large distances of the crack. This means that new cracks will nucleate
479: away of the first one, in regions where relation (\ref{tr}) is still satisfied.
480: The number of cracks nucleated at the surface will be typically 
481: the minimum number that makes
482: the horizontal stress at every point of the surface
483: to be lower than $T_r$. The typical distance between cracks $l$
484: can then be estimated to be\cite{1d}
485: \begin{equation}
486: l\simeq d_0 f(T_0/T_r)
487: \end{equation}
488: where the geometrical function $f(T_0/T_r)$ goes to infinity when
489: $T_0\rightarrow T_r^+$, and is of order 1 for $T_0 \gg T_r$. 
490: Typically, $l$ is a few times $d_0$.
491:  
492: 
493: If we stick to the ideal sharp drying front we have 
494: been studying, the first crack appears for an infinitesimally
495: small value of $z_0$, the only restriction is that the relation
496: (\ref{tr}) is satisfied. Therefore, since the depth of the first crack $d_0$
497: is of the order of $z_0$, we should expect a very dense set 
498: of superficial cracks. However, this unrealistic situation is
499: removed if we note that any physical drying front will have some typical (finite) 
500: humidity gradient $\nabla h$. At the moment when relation (\ref{tr})
501: is first satisfied (with $\Delta h$ being now the difference between humidity 
502: concentrations right at the surface, and well inside the material), 
503: the first crack will penetrate down to a distance
504: \begin{equation}
505: d_0\simeq \Delta h/\nabla h=T_r/(C\alpha \nabla h),
506: \end{equation}
507: and the typical distance between cracks
508: at the surface will be a few times this distance. 
509: %If this value corresponds to a value of $u$ in
510: %Fig. \ref{f3} lying in the stable propagation region, then this $l$ will be the 
511: %typical width propagated to the interior of the material. If this  
512: %$l$ corresponds to a value of $u$ in the
513: %unstable region, then some cracks will remain stopped at the surface, and a pattern of
514: %width $\sim l_{\rm min}$ will propagate.
515: 
516: \subsection{How the distribution of cracks gets uniform}
517: 
518: The cracks that appear at the surface of the sample 
519: nucleate at positions that are strongly influenced
520: by the presence of small inhomogeneities in the sample \cite{1d}.
521: But a superficially uneven set of cracks has a tendency to become 
522: evenly spaced as it penetrates the sample.
523: The reason for this tendency can be understood on an 
524: energetic basis. For a fixed number of cracks the evenly spaced
525: configuration corresponds to the minimum of elastic energy.
526: It is then clear, using the energetic arguments, that this configuration
527: will be approached during the propagation process\cite{ronsin2}.
528: 
529: It is important to clarify the different effects 
530: that inhomogeneities have on the creation and propagation of cracks.
531: For the superficial layer, inhomogeneities are relevant, and responsible for the broad 
532: distribution of fragment lengths\cite{1d}, since at the beginning all the surface is 
533: uniformly stressed, and tiny differences in the properties of the material
534: will dictate which point of the surface will fail first.
535: However, once the array of cracks has been
536: defined at the surface, inhomogeneities play a secondary role, since new
537: cracks do not appear during penetration, and the elastic energy
538: is only very weakly dependent on the precise distribution
539: of defects. Then the evolution 
540: of the pattern is basically the same as if inhomogeneities
541: were absent.
542: 
543: The previous energetic arguments predict the trend towards evenly spaced cracks as
544: they penetrate the sample. 
545: Stress analysis allows to re-obtain and make this result
546: quantitative.
547: We have calculated the 
548: stresses that are present in the semi-infinite sample with a sharp
549: drying front, in the presence of a particular uneven set of parallel cracks. 
550: We took an infinite set of cracks separated sequentially by distances $l_1$ and $l_2$.
551: 
552: For this geometrical configuration we have determined 
553: numerically the direction of maximum
554: opening stress (mode I), at the tips of the fractures, and characterized it
555: by its angle $\theta$ with respect to the direction of forward advance. 
556: The first result is that $\theta$ always points in the direction of 
557: making the pattern more evenly spaced.
558: The actual value of $\theta$ is absolutely independent
559: of the values of $C$ and $\eta$ of the material, and also on the value $\Delta h$ of the 
560: step of the humidity front. It is in fact a quantity that only depends on the Poisson ratio
561: $\nu$ of the material and the values of $x\equiv (\bar z-z_0)/l$ and $(l_1-l_2)/(l_1+l_2)$. 
562: The results for the angle $\theta$, for a material with 
563: $\nu=1/3$ can be summarized as follows: $\theta$ is
564: rather independent of $x$, and within a maximum 10 \% error, it can be written
565: as 
566: \begin{equation}
567: \theta {\rm (deg)}\simeq 29 (l_1-l_2)/(l_1+l_2)\,
568: \label{ll}
569: \end{equation}
570: The value of $\theta$ vanishes for $l_1=l_2$, as in this case the pattern is actually
571: evenly spaced, and the cracks advance straightforwardly by symmetry. The limiting
572: case $l_1\gg l_2$ gives $\theta\simeq 29$ deg\cite{29}.
573: This is the maximum bending we can expect from a given crack, in the process of
574: uniformising the widths of the stripes. Note that as a consequence of the fact that
575: there are no typical lengths in (\ref{ll}),
576: the length needed for the pattern to become evenly spaced will
577: be proportional to the typical width of the stripes $l$.
578: 
579: Once the cracks have started to deviate in order to reduce the elastic energy as much as 
580: possible, the further detailed prediction of its evolution becomes more difficult, since now
581: we should calculate stresses ahead of a set of non parallel fractures (or alternatively,
582: the elastic energy of this nontrivial configurations of curved cracks). However,
583: the main conclusion that the pattern becomes evenly spaced in a distance of the order
584: of the stripes widths remains\cite{coupled}.
585: 
586: \section{Dependences on the characteristics of the drying front}
587: 
588: The sharp drying front is an idealization that is never exactly realized in
589: practice. For the experiments of Ref. \cite{canad}, for instance, a typical
590: distance over which humidity changes is expected. In other cases in which
591: the drying is through surface diffusion, the humidity profiles are still
592: smoother. 
593: As we have already discussed the stability reason of a flat crack front
594: in the case of a sharp humidity step is due to the rapid reduction of stresses
595: ahead of the crack front. Smoother humidity profiles will produce a weaker tendency
596: to generate a flat crack front, and it can even occur that a flat crack 
597: front is never stable
598: if the humidity profile is smooth enough. In this case a pattern of bifurcations
599: of the crack front has been predicted\cite{bb}.
600: Note that an additional stabilizing factor of the crack front exists when the drying
601: is favored by the presence of the cracks themselves\cite{bahr}, a case we have not addressed here.
602: 
603: \section{Implications for the three dimensional case}
604: 
605: The same phenomenon we have been discussing acquires novel characteristics in a three 
606: dimensional geometry. It has a beautiful realization in the geological 
607: formations named columnar basalts, which can be reproduced in a kitchen experiment
608: using corn-starch\cite{muler}. When an originally hot volcanic lava flow starts
609: to cool down (after solidification), the thermal stresses generate cracks at the
610: surface, which propagate progressively towards the interior of the igneous body.
611: There are many coincidences with the two dimensional case, and also some new features
612: \cite{basalts,jr0,jr1}.
613: Cracks are known to appear at the surface and at later times propagate to the interior.
614: At the surface the pattern of cracks is rather disordered, but becomes progressively
615: ordered as it penetrates the material. In three dimensions, the tendency to order
616: manifest in progressive tendency to form a polygonal pattern of cracks, in planes parallel
617: to the surface. Qualitatively this tendency to order can be understood on the same basis
618: as the two dimensional case, as a tendency of the system to reduce its total energy as
619: much as possible during the evolution. 
620: It poses however some interesting problems, because typically,
621: the expected perfect hexagonal pattern is not reached, but a collection of polygons
622: with different number of sides and areas. It has been recently demonstrated\cite{jr1}
623: that this is a consequence of the fact that the patterns starts being disordered
624: at the surface, and in the process of minimizing the energy it is not able to
625: reach the absolute minimum (namely, the honeycomb pattern) but is trapped in
626: a metastable minimum.
627: Based in the arguments for the two dimensional case, we can expect also in three dimensions
628: that the typical width of the columns is set by 
629: the typical distance between cracks at the surface,
630: and it is not modified with the further penetration. In fact this is what is
631: experimentally observed,
632: since columns are seen to keep its horizontal size over distances that reach
633: hundred times the width of the columns.
634: For the three dimensional case, the cooling through diffusion plus convection 
635: (and/or radiation)
636: at the surface is the most realistic model of cooling to consider. Although
637: this type of cooling can lead in two dimensions to the instability of the flat
638: crack front, this is not necessarily so in three dimensions, since in this case
639: all fractures form a connected
640: structure, and this generates an additional tendency to stabilize the flat crack front.
641: These issues and some additional ones particular of the 3D case will be discussed in 
642: a forthcoming publication.
643: 
644: \section{Summary and conclusions}
645: 
646: In this paper we have focused on a two dimensional material that cools
647: down or dessicates from one edge. We studied the conditions under which 
648: a set of cracks penetrate the sample during the process. We showed that
649: these cracks have a tendency to become evenly spaced, an analyze the conditions
650: for which all the tips of the cracks form a plane front. Our results
651: provide a quantitative framework to analyze recent experiments.
652: Extensions to the three dimensional case are expected to provide both a 
653: qualitatively similar scenario, and also some new features that will be fully 
654: developed elsewhere.
655: 
656: \section{Acknowledgments}
657: 
658: This work was financially supported by CONICET (Argentina). Partial support
659: from Fundaci\'on Antorchas is also acknowledged.
660: 
661: 
662: \begin{references}
663: 
664: 
665: 
666: 
667: \bibitem{theor} T. Hornig, I. M. Sokolov, and A. Blumen, Phys. Rev. E 
668: {\bf 54}, 4293 (1996); K. Leung and J. V. Andersen, Europhys. Lett. {\bf 38},
669: 589 (1997); S. Kitsunezaki, Phys. Rev. E, {\bf 60}, 6449 (1999).
670: 
671: \bibitem{exper}A. Groisman and E. Kaplan, Europhys. Lett. {\bf 25}, 415 (1994);
672: P. Meakin, Science {\bf 252}, 226 (1991); W. Korneta, S. K. Mendiratta, and
673: J. Menteiro, Phys. Rev. E {\bf 57}, 3142 (1998).
674: 
675: \bibitem{basalts} A. Holmes, {\em Principles of Physical Geology}, 3rd ed. (Wiley,
676: New York,  1978), pp. 1--3; J. Walker, Sci. Am. {\bf 255}(4), 178 (1986).
677: 
678: \bibitem{jr0}M. P. Ryan and C. G. Sammis, Geol. Soc. Am. Bull. {\bf 89},
679: 1295 (1978); J. M. DeGraff and A. Aydin, Geol. Soc. Am. Bull. {\bf 99},
680: 605 (1987); A. Aydin and J. M. DeGraff, Science {\bf 239}, 471 (1988);
681: P. Budkewitsch, P. Robin, J. Volcanol. Geotherm. Res. {\bf %
682: 59}, 219 (1994).
683: 
684: \bibitem{jr1} E. A. Jagla and A. G. Rojo, Phys. Rev. E {\bf ??} ???? (2002).
685: 
686: \bibitem{alan-lim}C. Allain, and L. Limat, Phys. Rev. Lett. {\bf 74}, 2981 (1995).
687: 
688: \bibitem{canad}K. A. Shorlin, J. R. de Bruyn, M. Grahan, and S. W. Morris,
689: Phys. Rev. E {\bf 61}, 6950 (2000).
690: 
691: \bibitem{hull}D. Hull and B. D. Caddock, J. Mat. Sci. {\bf 34}, 5707 (1999).
692: 
693: %\bibite{e-s}e-s
694: 
695: \bibitem{broberg}K. B. Broberg, {\it Cracks and Fracture}, (Academic, Cambridge, 1999)
696: 
697: \bibitem{setna}J. A. Hodgdon and J. P. Sethna, Phys. Rev. E {\bf 47},
698: 4831 (1993).
699: 
700: \bibitem{nota10}We are implicitly assuming here that the minimum in which 
701: the system is located does not destabilizes upon a change of the external
702: conditions.
703: 
704: 
705: \bibitem{elasticidad}To be more precise, we are assuming
706: that the size of the plastic region near the tips of the cracks
707: is always small compared to the distance between cracks. The
708: energy dissipated in the plastic region is accounted for in the
709: numerical value of the fracture energy $\eta$ defined below.
710: 
711: \bibitem{landau}L.D. Landau and E.M. Lifshitz, {\em Theory of Elasticity}, 3rd ed.
712: (Butterworth-Heinemann, Oxford, 1986).
713: 
714: \bibitem{bahr}T. Boeck, H.-A Bahr, S. Lampenscherfm and U. Bahr, Phys. Rev. E 
715: {\bf 59}, 1408 (1999).
716: 
717: \bibitem{lineart} B. A. Boley and J. H. Weiner, {\it Theory of thermal stresses},
718: (Wiley, New York, 1960).
719: 
720: \bibitem{zz}We are assuming here that $|\bar z-z_0|\ll z_0$, in such
721: a way that surface effects are negligible.
722: 
723: \bibitem{poisson}To simplify the presentation we will consider a material
724: with a fixed Poisson ratio $\nu$. In a general case, all constants and functions
725: introduced in the analysis must be considered to be also functions of $\nu$.
726: 
727: \bibitem{nota2} The material is vertically compressed in this region in such a 
728: way that vertical stress vanishes. However, it is the
729: impossibility of producing a horizontal compression what generates a horizontal
730: stress and then the finite value of this part of the 
731: elastic energy.
732: 
733: \bibitem{vidrio} From the results of Fig. \ref{f3} we can obtain the result that
734: thiner stripes will have their crack front more retracted compared
735: to thicker stripes (since $u\sim l^{-1}$), and that a minimum $l$ is necessary
736: for propagation. This is qualitatively
737: comparable to experimental results \cite{ronsin} 
738: showing that a single crack propagating in a glass plate of width $l$ between two
739: thermal baths has an equilibrium position that retracts as $l$ reduces, and also 
740: that the propagation is halted if $l$ is lower than some critical value.
741: 
742: 
743: \bibitem{ronsin}O. Ronsin, and B. Perrin, Phys. Rev E {\bf 58},
744: 7878 (1998).
745: 
746: \bibitem{1d}O. Morgenstern, I. M. Sokolov, and A. Blumen, Europhys. Lett. {\bf 22},
747: 487 (1993); J. Phys. A {\bf 26}, 4521 (1993).
748: 
749: \bibitem{microck}This is the traction resistance value for the bulk material, 
750: which of course takes into account the existence of defects (such as microcracks)
751: in the material.
752: 
753: \bibitem{ronsin2}The tendency to uniformise the stripe width $l$ is also consistent
754: with the experimentally observed stability of a crack propagating exactly at the middle
755: of a glass plate, driven by a temperature difference (see Ref. \cite{ronsin}).
756: 
757: \bibitem{29}This value nicely coincides with the expected half angle
758: for the infinitesimal branching of an original straight crack under mode I loading,
759: see Refs. \cite{broberg} and \cite{isida}.
760: 
761: \bibitem{isida}M. Isida and H. Noguchi, Int. J. Fract. {\bf 54}, 293 (1992).
762: 
763: \bibitem{coupled}During the process of uniformising an originally disordered
764: set of cracks, the crack front can deviate slightly from the sharp
765: form expected for evenly spaced cracks.
766: For instance, for a set of fractures with a typical width
767: of columns $l$ that depends smoothly on the horizontal coordinate, the fracture
768: front will be forwardly advanced at those position with largest $l$, and retracted
769: at positions with smallest $l$.
770: 
771: \bibitem{bb}H.-A Bahr, U. Bahr, and A. Petzold, Europhys. Lett. {\bf 19}, 485
772: (1992).
773: 
774: \bibitem{muler}G. M\"{u}ller, J. Geophys. Res. {\bf 103}, 15239 (1998).
775: 
776: \end{references}
777: 
778: \begin{figure}
779: \narrowtext
780: \epsfxsize=3.3truein
781: \vbox{\hskip 0.05truein
782: \epsffile{2d_f0.eps}}
783: \caption{Sketch of the process studied. There is a sharp drying profile at
784: depth $z_0$, with $z_0$ increasing slowly with time. As this occurs, the crack
785: front (located at $\bar z$)
786: penetrates more deeply into the material. The distance $\bar z-z_0$ (which can be
787: positive or negative, depending on the parameters) remains constant in time.}
788: \label{f0}
789: \end{figure}
790: 
791: 
792: \begin{figure}
793: \epsfxsize=3.3truein
794: \vbox{\hskip 0.05truein
795: \epsffile{2d_f2.eps}}
796: \caption{The functions $g(x)$ (continuous line) and $\chi (x)$ (dotted line)
797: for a material with Poisson ratio $\nu=1/3$. 
798: The values of $x$ for which $\chi (x)>0$
799: are those corresponding to a stable flat crack front.
800: } 
801: \label{f2}
802: \end{figure}
803: 
804: \begin{figure}
805: \narrowtext
806: \epsfxsize=3.3truein
807: \vbox{\hskip 0.05truein
808: \epsffile{2d_f1.eps}}
809: \caption{Sketch of the density of energy $\delta e_{\rm el}$ as a function of $z$
810: for a set of fractures
811: that has penetrated down to $\bar z<z_0$ (a), and $\bar z>z_0$ (b). The contraction
812: of the stripes is exaggerated for clarity.}
813: \label{f1}
814: \end{figure}
815: 
816: \begin{figure}
817: \narrowtext
818: \epsfxsize=3.3truein
819: \vbox{\hskip 0.05truein
820: \epsffile{2d_f3.eps}}
821: \caption{Position of the crack front [$x\equiv (\bar z-z_0)/l$] as a function of
822: $u\equiv \eta/[C(\alpha \Delta h)^2l]$, obtained by minimizing the
823: total energy of the system. The region at the left of $u_{\rm cr}$ 
824: corresponds to a stable situation in which all tips of the cracks form a stable
825: crack front, whereas at the right of this value the flat crack front is
826: unstable.}
827: \label{f3}
828: \end{figure}
829: 
830: 
831: 
832: \end{document}
833: