See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/8952691

Effect of a columnar defect on the shape of slow-combustion fronts

Article in Physical Review E · December 2003

DOI: 10.1103/PhysRevE.68.051103 · Source: PubMed

CITATIONS READS
22 60

7 authors, including:

Markko Myllys J. Maunuksela

University of Jyväskylä University of Jyväskylä
45 PUBLICATIONS 1,183 CITATIONS 18 PUBLICATIONS 395 CITATIONS

SEE PROFILE SEE PROFILE

Juha Merikoski J. Timonen

University of Jyväskylä University of Jyväskylä
50 PUBLICATIONS 1,202 CITATIONS 367 PUBLICATIONS 8,656 CITATIONS

SEE PROFILE SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Optical Solitons View project

Collapse in BEC View project

All content following this page was uploaded by Juha Merikoski on 26 May 2014.

The user has requested enhancement of the downloaded file.

 Effect of a columnar defect on the shape of slow-combustion fronts
M. Myllys, J. Maunuksela, J. Merikoski, and J. Timonen
Department of Physics, P.O. Box 35, FIN-40014 University of Jyväskylä, Finland

V. K. Horváth
Department of Biological Physics, Eötvös University, 1117 Budapest, Hungary

M. Ha and M. den Nijs

Department of Physics, University of Washington, Seattle, WA 98195, USA

We report experimental results for the behavior of slow-combustion fronts in the presence of a
columnar defect with excess or reduced driving, and compare them with those of mean-field theory.
We also compare them with simulation results for an analogous problem of driven flow of particles
arXiv:cond-mat/0307231v1 10 Jul 2003

with hard-core repulsion (ASEP) and a single defect bond with a different hopping probability. The
difference in the shape of the front profiles for excess vs. reduced driving in the defect, clearly
demonstrates the existence of a KPZ-type of nonlinear term in the effective evolution equation for
the slow-combustion fronts. We also find that slow-combustion fronts display a faceted form for
large enough excess driving, and that there is a corresponding increase then in the average front
speed. This increase in the average front speed disappears at a non-zero excess driving in agreement
with the simulated behavior of the ASEP model.

PACS numbers: 64.60.Ht, 05.40.-a, 05.70.Ln

I. INTRODUCTION be expected to be generic for nonequilibrium interfaces.

One can demonstrate the presence of this term indirectly
Nonequilibrium interfaces that display interesting scal- by using, e.g., inverse schemes able to infer the (partial)
ing properties are quite common in physical (crystal differential equation that governs the observed stochas-
growth, fluid penetration into porous media, etc.), chem- tic evolution of interfaces [12], but there is also a way to
ical (reaction fronts), as well as biological (growing bac- produce a directly observable effect on the shape of the
terial colonies) systems. The dynamics of these systems interface, due to this term.
have long been thought to be generically described by the This method for observing the operation of the non-
Kardar-Parisi-Zhang (KPZ) equation [1], or some other linear term was suggested already for some while ago by
equation of motion in the same universality class [2]. In Wolf and Tang [13]. They considered the effect of colum-
two space dimensions in particular (one-dimensional in- nar defects, columnar in two space dimensions on which
terfaces) when exact solutions are available, the scaling case we concentrate here, and found that there is a clear
properties of the KPZ equation are well understood. The ’asymmetry’ between the shapes of the fronts that prop-
same is not, however, true for the experimental observa- agate in the presence of an advancing (excess driving)
tions of scaling of interfaces. Typically one has found and a retarding (reduced driving) defect. This asymme-
a roughening exponent clearly higher than that for the try is a direct consequence of the nonlinear term in the
KPZ equation [2]. Various explanations have been sug- KPZ equation. For a positive coefficient in this term,
gested as for why the KPZ scaling has not generally applicable to slow-combustion fronts, the noise-averaged
been found, and most of the time correlated and/or non- front should be faceted with a forward-pointing trian-
Gaussian noise have been the prime suspects [2, 3]. gular shape around an advancing defect, with a height
Recent experiments on slow-combustion fronts propa- proportional asymptotically to the width of the sample
gating in paper [4, 5], and on flux fronts penetrating a in the case of a defect in the middle of the sample, or to
high-Tc thin-film superconductor [6], have provided new the basic period in the case of periodic boundary condi-
insight into this problem [7]. It indeed appears that tions. In the case of a retarding defect, the shape of the
short-range-correlated noise, quenched and dynamical, front should not be faceted, and the (negative) height of
with possibly at the same time a non-Gaussian ampli- the deformation in the profile should be proportional, ac-
tude distribution for small time differences, induce an cording to this mean-field theory, to the logarithm of the
additional time and an additional length scale, beyond basic period. Despite its apparent simplicity, this kind of
which KPZ scaling can only be observed. Despite these experiment has never been performed.
recent advances, it would still be worth while to demon- As is well known, the BCSOS interface model in which
strate the existence of the nonlinear term, as introduced the nearest-neighbor heights are restricted to differ only
by Kardar, Parisi and Zhang [8], and essential for the by ±1, displays KPZ behaviour, and is on the other
KPZ dynamics, directly from the observed fronts. This hand equivalent to a driven flow of particles (hopping
would in essence prove that KPZ type of dynamics, in- rate p) with hard-core repulsive interactions (ASEP) [2].
cluding possibly effects of nontrivial noise, can indeed A columnar defect in an interface model corresponds to
 2

a fixed slow or fast bond (hopping rate rp with r < 1 sium nitrate determines the average speed of the fronts,
or r > 1, respectively) in the ASEP model. A faceted so it serves as the control parameter of the problem.
interface corresponds to a traffic jam of infinite length It is, however, quite difficult to accurately regulate
in the thermodynamic limit behind the slow bond. An- the amount of potassium nitrate absorbed in the sam-
other related question is the detailed shape of the den- ple. This means that it is difficult to produce samples
sity/interface profile. The mean-field theory of [13] pre- with exactly the same base concentration, and the same
dicts an infinite queue for all r < 1 and no queue (a concentration difference between the base paper and the
logarithmic decay of the density profile) for r > 1, i.e., columnar defect. Therefore, the statistics we get for any
rc = 1. Janowsky and Lebowitz [9] considered the to- fixed difference in the concentration is not quite as good
tally asymmetric TASEP model with a slow bond, but as we would hope. They are adequate for the main fea-
concentrated mainly on the shock wave fluctuations far tures of the fronts but not for such details as, e.g., ac-
away from the slow bond, and apparently did not con- curate forms of the front profiles. They are also good
sider the faceting/queueing transition (they had a phase enough for a quantitative analysis of changes in the front
diagram with rc = 1). The same model was also con- speed.
sidered by Kolomeisky [10], but he did not consider the The samples were typically 20 cm (width) by 40 cm,
faceting/queueing transition either. and the columnar defect (vertical stripe) in the middle
Kandel and Mukamel considered a somewhat different of the sample was 1.0 cm wide. The defect cannot in our
model, which is supposed to be in the same universality case be too narrow as fluctuations in the slow-combustion
class, and proposed [14] that the faceting/queueing tran- process would then tend to wipe out its effect. Too wide
sition should take place at an rc < 1. Their simulation a stripe would on the other hand cause effects due to its
data were not, however, conclusive. nonzero width, which are unwarranted. We also used
In slow-combustion experiments the detailed shape of simulations with a discretized KPZ equation to check
the front profile is difficult to determine, and thereby also that the ratio 1 cm to 20 cm should not cause addi-
the disappearance of faceting. Faceting is however re- tional effects [15]. The length of the samples was in
lated to increased front speed, also in the thermodynamic most cases adequate for achieving stationary behavior,
limit, and this is an easier observable. For possible non- and only those results are used here where saturation of
faceted fronts, which would correspond to rc < r < 1, an the profile was evident.
increased front speed would only be a finite-size effect, When analyzing the front profiles, the stripe was re-
as also the decreased front speed in the case of a retard- moved from the data, as well as about 6 mm from both
ing defect corresponding to r > 1. Notice that the ef- boundaries of the samples. As the system is symmetric
fective nonlinear term is positive in the slow-combustion across the stripe in the middle, the observed front profiles
experiments, while it is negative in the ASEP models. were also symmetrized for better statistics.
Therefore, an advancing (retarding) columnar defect in
As reported already before [4, 5], fluctuations in the
the first case corresponds to a slow (fast) bond in the
slow-combustion fronts in paper are noticeable. For ex-
latter case.
treme values of the potassium-nitrate concentration there
appear problems with pinning (low concentration) or lo-
cal avalanche type of bursts (high concentration) in the
II. EXPERIMENTAL DETAILS fronts. Also, very small values of the concentration differ-
ence between the base paper and the defect stripe could
The equipment we use in slow-combustion experiments not be used, as fluctuations then completely masked the
has been described elsewhere [4, 5], so it suffices to say effect of the defect. These problems were noticeable for
here that samples were ’burned’ in a chamber with con- retarding defects in particular. In the data reported here,
trollable conditions and that the video signal of propagat- concentration varied between 0.265 and 0.61 gm−2 in the
ing fronts was compressed and stored on-line on a com- base paper, between 0.1 and 1.05 gm−2 in the stripe, and
puter. The spatial resolution of the set up was 120 µm, the concentration difference varied between 0.06 and 0.49
and the time resolution was 0.1 s. For the samples we gm−2 on the positive side (25 burns), and −0.197 and
used the lens-paper grade (Whatman) we have used also −0.478 gm−2 on the negative side (19 burns). Because
previously [5]. Lens paper was now used to speed up the of the practical restrictions and fluctuation effects, the
experiments. number of successful burns was relatively small.
As slow-combustion fronts do not propagate in paper
without adding an oxygen source for maintaining the
chemical reaction involved, we added as before [4, 5] a
small amount of potassium nitrate in the samples. This III. RESULTS
method also allows for a relatively easy way to produce
advancing and retarding columnar defects. By using We will need the dependence on potassium-nitrate con-
masks it is straightforward to produce a narrow (ver- centration of the front velocity below so we consider it
tical) stripe with a smaller or an additional amount of first. It is useful to begin with a discussion of the accu-
potassium nitrate. The average concentration of potas- racy of the front-velocity determination.
 3

0.6 12.0

0.5 11.0
10.0
0.4

v [mm/s]
9.0
P(v)

0.3
8.0
0.2 7.0
0.1 6.0
0.0 5.0
4 5 6 7 8 9 10 11 12 13 0.0 0.2 0.4 0.6 0.8 1.0
2
v [mm/s] C(KNO3) [g/m ]

FIG. 1: Velocity distributions for slow-combustion fronts with FIG. 2: Front velocity as a function of potassium-nitrate con-
potassium-nitrate concentrations 0.34 gm−2 (full line), and centration. Measured points are denoted by circles and the
0.536 gm−2 (dashed line). Velocities are determined for a line is a linear fit to these points.
time difference of 2 s.

the potassium-nitrate concentration, and these data are

A. Front velocity shown in Fig. 2 together with a linear fit to the measured
points.
Lens paper is thin so that variations in its mass density The dependence on potassium-nitrate concentration of
and dynamical effects such as (possibly turbulent) con- the front velocity is not expected to be linear especially
vection around the combustion front are both expected near the pinning limit, but, for the concentration range
to give a contribution to the effective noise. Noise ampli- shown here, it is well approximated by a linear behavior,
tude is consequently relatively large [4, 5], and therefore which is also more convenient for the subsequent analy-
also velocity distribution of a propagating front can be sis. We find that, with a linear fit to the data, the front
expected to be broad. We show in Fig. 1 that distri- velocity v is given on the average by
bution for two different values of the potassium-nitrate
concentration.
v = 4.2C + 6.2, (1)
It is evident from this figure that, even though the
average velocities in these two cases are relatively well
where C is the potassium-nitrate concentration, and v is
separated and easily distinguishable, the velocity distri-
in units mms−1 when C is expressed in gm−2 .
butions have a big overlap. Together with the limited
Before showing the measured front profiles in the pres-
statistics for any fixed value for (the difference in) the
ence of a columnar defect, let us consider, in order to
potassium-nitrate concentration, these broad distribu-
make later comparisons more transparent, what is ex-
tions mean that some variation can be expected to occur
pected from the mean-field solution as reported in Ref.
in the measured average velocities (velocity differences)
[13].
of the fronts.
In the presence of a columnar defect, we should, e.g.,
determine the change in the average front speed arising B. Mean-field prediction
from the defect. This can be accomplished by analyzing
separately the undeformed ’flat’ part of the fronts, and
the part of the front profile affected by the presence of We assume that the time evolution of the fronts h(x, t)
the defect. Determination of the average speed of the flat is governed by the KPZ equation ∂h/∂t = ν∇2 h +
λ 2
fronts is done in the transient (wrt profile shape) phase 2 (∇h) + κ + η(x, t), where η(x, t) describes white noise
in which the (growing) width of the deformed profile is with delta-function correlations in space and time, and
still less than the width of the sample. In this phase the the driving term contains the idealized
P defect as a delta-
flat part of the front is already in the saturated regime function contribution, κ = κ0 + κ1 n δ(x − L/2 + nL).
with constant average velocity. The average speed of the For simplicity we assume here as in [13] periodic bound-
deformed profile is determined in a later phase in which ary conditions. If we average over noise in the KPZ equa-
the width of the deformed part of the profile essentially tion, and denote H(x, t) ≡ hh(x, t)i, we find that
coincides with the sample width.
We have also determined the average front velocity ∂H(x, t) λ λ
for 122 individual burns for a fairly broad interval in = νH ′′ + (H ′ )2 + h(∇δh)2 i + κ, (2)
∂t 2 2
 4

in which H ′ ≡ H ′ (x, t) denotes the spatial derivative 40

of H, and δh ≡ h(x, t) − H(x, t) describes fluctuations
around the noise-averaged profile. A corresponding equa-
30
tion can be derived for δh [13]. 20
As δh should not depend (locally) on H(x, t) nor

< h> [mm]

on H ′ (x, t), and only local interdependence between δh 10
and the noise averaged profile can be assumed to ap-
pear, one would then expect [13] that in leading order 0
h(∇δh)2 i = a0 + a2 H ′′ (x, t), with a0 and a2 some con- -10
stants. This assumption will make Eq. (2) closed so that
it can be solved without further reference to the fluctu- -20
ations. The delta-function contribution in the driving
term will induce cusps in H(x) at x = L/2 − nL, and the -30
solution of Eq. (2) is therefore equivalent to solving the -40
equation -100 -80 -60 -40 -20 0 20 40 60 80 100
x [mm]
∂H(x, t) λ
= νe H ′′ + (H ′ )2 + κe (3)
∂t 2
FIG. 3: Average front profiles with a columnar defect for
in the interval −L/2 ≤ x ≤ L/2, with boundary condi- ∆C = ±0.33 (upper and lower profile, respectively).
tions H(L/2) = H(−L/2) and H ′ (±L/2) = ±s. Here
we have defined the effective (renormalized) parameters
νe ≡ ν + λa2 /2 and κe ≡ κ0 + λa0 /2, and the magnitude C. Measured front profiles
of the slope of the front at the defects is s ≡ κ1 /2νe .
Equation (3) is the well-known Burgers equation [16]
which can be solved in closed form in one space dimen- In the above mean-field theory, δv ≡ κ1 /L is the dif-
sions. It is useful to express it first in dimensionless ference between the front velocity (driving) inside the
form, which can be achieved with transformations H = defect and outside the defect. In the slow-combustion
H0 H̃, x = H0 x̃/s, t = 2H0 t̃/(λs2 ), with H0 ≡ 2νe /λ the experiments this velocity difference is regulated by
internal length scale of the system. the potassium-nitrate concentration so that now δv =
We look for a stationary solution of this equation in 4.2∆C, where the numerical factor comes from the linear
the form H̃(x̃, t̃) = (κe + sgn(κ1 )q 2 )t̃ + ln(f (qx̃)), where fit given by Eq. (1), and ∆C ≡ Cdefect − Cbase is the con-
sgn(z) is the sign of z, and we have already used the centration difference. This means that the scaling factor
Hopf transformation in the spatial part of the ansatz to s is given by s = 4.2L|∆C|/νe . Without as yet knowing
remove the nonlinearity from the equation for f . We find the actual value of νe needed for evaluating the size of s
that [13] and H0 , reasonable estimates, based on the results from
the inverse method solution for the effective equation of
! motion [12], indicate that the slow-combustion fronts are
q L̃ not necessarily in the strictly asymptotic regime: we ex-
f (z) = cosh(z), q tanh = 1, (4)
2 pect that L̃ > 1 but not by a very big margin. Notice
that the size of L̃ is now regulated by ∆C as the width L
for advancing defects (κ1 > 0), and of the samples is held fixed. Despite the achievable val-
! ues of L̃, we can expect to clearly see the ’asymmetry’ in
q L̃ the heights of the front profiles for different signs of the
f (z) = cos(z), q tan = 1, (5) concentration difference. In Fig. 3 we show the averaged
2
(and symmetrized) front profiles for ∆C = ±0.33.
for retarding defects (κ1 < 0). Here L̃ ≡ sL/H0 is It is indeed evident that there is a clear difference in
the dimensionless width of the system, and z ≡ qx̃. the heights of the front profiles around advancing and re-
Asymptotically, for L̃ ≫ 1 (and λ > 0), the profile tarding defects. By following the transient time evolution
around an advancing defect is a forward-pointing tri- of the fronts, we could also see a clear difference there.
angle with sides that have slopes ±s, and with height For ∆C > 0, when a triangular deformation was formed
∆H+ ≡ H(L/2) − H(0) ≃ sL/2. The asymptotic profile after a while around the central stripe, its height and base
around a retarding defect is given by H0 ln|cos(πx/L)| so length grew with a more or less constant velocity until
that ∆H− ≃ −H0 ln(sL/πH0 ). The magnitude of H+ the base length reached the width of the sample, while
thus grows linearly with L (or s) while that of H− only the slopes of the sides of this triangle remained roughly
grows logarithmically with L (or s). This asymmetry is constant. For ∆C < 0 on the other hand, the height
a direct consequence of the nonlinear term that enhances of the deformation saturated much faster even though it
the deformation in the former case but reduces it in the also grew more or less linearly in time in the beginning,
latter case. and the base length of the deformation reached the sam-
 5

1.2
1.0
0.8

H/(L/2)
0.6
0.4
0.2
0.0
-0.2
-0.4
-0.4 -0.2 0.0 0.2 0.4
2
C(KNO3) [g/m ]

FIG. 5: The absolute value of the average LH slope of the

deformed front profile around a columnar defect as a function
of difference in the potassium-nitrate concentration. The full
line is a fit by the mean-field solution with Eqs. (4) and (5).

system.
We show in Fig. 5 the experimentally determined val-
ues for k+ and k− together with the fit by the mean-field
solution using Eqs. (4) and (5).
Fits to the data were not very sensitive to the actual
value of parameter A so that the correlation coefficient
FIG. 4: Successive fronts with a time difference of 0.5 s for did not change much even if A was changed in a relatively
concentration difference ∆C = 0.327. Also marked are the large interval. If the ∆C > 0 and ∆C < 0 data were fit-
stripe with the excess concentration of potassium nitrate, the ted separately without any restrictions on the two param-
fronts between which the average profile is determined (thick eters, these fits had also a tendency to produce somewhat
lines), the height of the final profile, and the average shape of different values for the two cases. As the signal-to-noise
the profile. ratio is better for the ∆C > 0 data, we fixed A such that
it was between the two separately fitted values but closer
to the one from the unrestricted two-parameter fit to the
ple width at the same time. This transient behavior will ∆C > 0 data, and in the interval within which the quality
be analyzed in more detail below. The qualitative behav- of this fit was essentially unchanged: A ≃ 0.3. Thereafter
ior for the ∆C > 0 case is clearly visible in Fig. 4 which an unrestricted one-parameter fit to the whole data was
shows the successive fronts with a time difference of 0.5 used to find the value for B. In this way we found that
s for ∆C = 0.327. B ≃ 2.5.
A more quantitative comparison between the mean-
The fitted values for parameters A and B allow now an
field solution for the noise averaged front and the ob-
served slow-combustion fronts can also be made. For estimation of two physical parameters, the ’renormalized’
this purpose we have found it convenient to consider in- diffusion coefficient νe and the coefficient of the nonlinear
term, λ. We thus find that νe ≃ 144 mm2 s−1 , and λ ≃
stead of the profile heights ∆H± the average slopes of
the left-hand (LH) sides of the profiles (c.f. Fig. 3), 5.6 mms−1 . In the estimate for νe we used an ’effective’
sample width Leff ≃ 180 mm, a bit smaller than the
k± ≡ 2∆H± /L. As we do not expect to be in the strictly
asymptotic regime, we have used the full transcendental width 202 mm of the actual sample, due to the width
of the defect stripe and to allowing for some boundary
equations for q in Eqs. (4) and (5) above when fitting
the observed k± with the mean-field result. effects. By other methods we have found previously that
λ ≃ 4.1 − 5.1 mms−1 [12], so that the value found here is
The average slopes, as functions of concentration dif- fairly close to these previous estimates.
ference ∆C, will now depend on two parameters, A ≡
H0 /L and B ≡ 2.1L/νe, which are used to fit the mea- In view of the unavoidable fluctuations in the measured
sured slopes. From the fitted values for these parameters averaged slopes, we find the fits to the measured points
we can then estimate the coefficients λ and νe for this by the mean-field solution to be quite reasonable.
 6

D. Defect-induced change in front velocity and the 0.5

queueing transition
0.4

(V-V0)/V0 , (J0-J)/J0
As already discussed above, the mean-field solution 0.3
predicts a faceting or queueing transition at ∆C = 0.
Above this transition (∆C > 0), the average front ve- 0.2
locity is increased due to the presence of an advancing
columnar defect, and below this transition the change in 0.1
front velocity should vanish for large enough L̃. For neg-
ative ∆C the change in velocity is negative, and should 0.0
decrease in magnitude with increasing |∆C|. According
to Kandel and Mukamel [14], this transition should ap- -0.1
pear at a ∆Ccr > 0. -0.5 0.0 0.5 1.0
As the numerical data of [14] is not decisive, we have (C-C0)/C0 , 1-r
done [11] simulations on a totally asymmetric ASEP
model with a fixed defect bond with hopping rate rp in
FIG. 6: Scaled velocity change of slow-combustion fronts due
the middle of the system, while the hopping rate at the to a columnar defect as a function of (C − C0 )/C0 (⋆), and
other bonds was p. Open boundary conditions were im- scaled current change in the totally asymmetric ASEP model
posed such that the hopping-in rate at the left boundary due to a defect bond as a function of (p − rp)/p = 1 − r (◦).
was αp, and the hopping-out rate at the right bound- The full line connects the latter data points.
ary was βp. In what follows we only consider the case
α = β = p = 1/2.
This model shows [11] a queueing transition at r = wipe out the whole effect, and the system is then not in
rc = 0.80 ± 0.02. In addition, the density profile dis- the ’asymptotic regime’. These results indicate, however,
plays a qualitatively similar asymmetry between the slow that there indeed is a faceting (queueing) transition at a
and fast defect-bond cases as the mean-field solution non-zero ∆Ccr (∆pcr , i.e., rc 6= 1).
for the KPZ fronts between the advancing and retard-
ing columnar-defect cases. In the thermodynamic limit
the deformation stays non-zero only in the faceted phase E. Transient behavior
above the transition. The density profiles in both the
faceted and non-faceted phases also display interesting In addition to the stationary profiles analyzed above,
power-law tails. it is also possible, as already indicated, to study the tran-
As the detailed shapes of the front profiles are difficult sient profiles, i.e., how the defect-induced profiles grow
to determine experimentally, we only compare the results at the initial phases of the process. The transient be-
for the dependence of the average front velocity V ≡ hvi havior of the profile around an advancing column is par-
(current J ≡ hji) on the potassium nitrate concentra- ticularly simple. The Burgers equation Eq. (3) admits
tion C (hopping rate p). In this comparison dimension- in this case a solution of exactly the same shape as the
less variables are used, (V − V0 )/V0 for the change in stationary solution, which grows linearly in time until its
the front velocity, and similarly for the current but for baseline reaches the width of the sample. Such a ’self-
reversed sign as an advancing defect corresponds to a similar’ transient does not exist in the case of negative
slow bond, ∆C/C0 for the potassium-nitrate concentra- ∆C, so analytical results for transient behavior are then
tion difference, and ∆p/p = 1−r for the hopping-rate dif- difficult to find. In the non-asymptotic regime L̃ ≪ 1
ference. Differences are all determined between the value one can however show that the situation is symmetric,
with or at the defect and the value elsewhere or without ∆H− ≃ −∆H+ . One would thus expect that, at least
the defect. In this way no fitting is involved in the com- in our case when L̃ is not particularly large, the height
parison. Obviously the actual driving force is not known |∆H− | would also grow initially (at least nearly) linearly
exactly for the slow-combustion fronts, but the observed in time.
linear dependence well above the pinning transition be- The expected transient behavior for ∆C > 0 is already
tween the potassium-nitrate concentration and the front (qualitatively) evident from Fig. 4 above. More quanti-
velocity suggests that the dimensionless difference can be tatively the transient time evolution of the height of the
reliably used in this kind of comparison. deformed profile can be analyzed, e.g., by plotting H(2t)
This comparison of the slow-combustion experiment against H(t). For a linear time evolution the former value
and the totally asymmetric ASEP model results is shown is twice the latter. In Fig. 7 we show this plot, averaged
in Fig. 6. It is evident from this figure that agreement over 32 individual burns, including both signs of ∆C.
between the two results is reasonable as there is no fit- The initial transient behavior is approximately linear
ting involved. There are still fairly large fluctuations in in time for both cases. For ∆C > 0 the trend continues
the experimental data, and it is not possible to have re- nearly linear until saturation sets in when the width of
sults for very small values of ∆C as fluctuations tend to the profile equals the width of the sample. For ∆C < 0
 7

60
50
......... lution. For negative velocity difference a nearly linear
behavior in time was observed initially, followed perhaps
40
.... by a regime of nonlinear time evolution before saturation.
30
.... Fitting the average height (or equivalently the average
H(2t) [mm]

20
..... slopes of the sides) of the profile with the mean-field so-
10
..... lutions provided us with estimates for the effective ’diffu-
sion constant’ νe and the coefficient of the nonlinear term,
0
...... λ. The latter parameter can also be determined from the
-10
-20
........ slope dependence of the local front velocity [5, 12], or by
applying an inverse method on the observed fronts [12].
-30 The value found here for λ is fairly close to these previ-
-40 ous estimates, and we find this level of agreement very
-30 -20 -10 0 10 20 30 40 reasonable in view of the rather large fluctuations in the
H(t) [mm] present data.
One should, however, notice that the λ measured for a
sample depends on the potassium-nitrate concentration
FIG. 7: ∆H(2t) as a function of ∆H(t) averaged over 32
burns. Positive values correspond to ∆C > 0 and negative in that sample, and that the average potassium-nitrate
values to ∆C < 0. The full line is ∆H(2t) = 2∆H(t). concentration was not the same in the samples used in the
experiments. We did not take this variation into account,
as it can be assumed to give a small effect in comparison
the behavior is quite similar except that saturation takes with the other experimental uncertainties, so that the
place earlier. There is also some indication that, in this present estimate represents an ’average’ value.
case, the growth of ∆H becomes nonlinear in time al- The effective diffusion coefficient νe contains, in ad-
ready before saturation, but the quality of the data does dition to the bare diffusion coefficient of the original
not allow for a decisive conclusion on this. KPZ equation, an unknown renormalization factor due
to noise-induced fluctuations around the average front
profile. We cannot thus get an estimate for the ’bare’
IV. DISCUSSION AND CONCLUSIONS diffusion coefficient ν, which can be estimated by other
means [12]. However, we can conclude that the noise-
The difference in the amplitude (height), and perhaps induced renormalization of νe appears to be sizable.
not so clearly in the shape, of the front in the slow- The position and nature of the faceting (queueing)
combustion experiments, caused by a columnar defect transition in interfaces affected by a columnar defect (in
with excess or reduced driving, respectively, was clearly the ASEP model by a defected bond), has been a long-
demonstrated. The behavior of the height of the de- standing problem. The agreement found here between
formed profile, and the qualitative shape of the profile slow-combustion experiments with a columnar defect and
in the case of excess driving, were also reasonably well the related TASEP model results, indicates that this
explained by the mean-field solution of Ref. [13]. The transition is indeed at a non-zero value of the respective
asymptotic shape of the profile in the case of negative ve- control parameter. No scaling properties of the transition
locity difference could not be unequivocally determined could be analyzed at this stage, but the TASEP model
as fluctuations are more important in this case of rela- results also indicate that this transition is continuous. It
tively small amplitude of the profile. The reduced, in remains an experimental challenge to analyze this tran-
comparison with the case of excess driving, height of the sition in more detail.
profile was very evident. In the case of positive velocity The authors gratefully acknowledge support by the
difference the transient behavior of the profile, i.e., the Academy of Finland (MaDaMe Programme and Project
growth of the defect induced deformation in the profile No. 44875), and fruitful discussions with David Mukamel
shape, could as well be explained by the mean-field so- and Joachim Krug.

[1] M. Kardar, G. Parisi, and Y.-C. Zhang, Phys. Rev. Lett. Rev. Lett. 79, 1515 (1997); M. Myllys, J. Maunuksela, M.
56, 889 (1986). J. Alava, T. Ala-Nissila, and J. Timonen, ibid 84, 1946
[2] For a review, see e.g. T. Halpin-Healy and Y.-C. Zhang, (2000).
Phys. Rep. 254, 215 (1995). [5] M. Myllys, J. Maunuksela, M. Alava, T. Ala-Nissila, J.
[3] V. K. Horváth, F. Family, and T. Vicsek, Phys. Rev. Lett. Merikoski, and J. Timonen, Phys. Rev. E 64, 036101
67, 3207 (1991). (2001).
[4] J. Maunuksela, M. Myllys, O.-P. Kähkönen, J. Timo- [6] R. Surdeanu, R. J. Wijngaarden, E. Visser, J. M. Hui-
nen, N. Provatas, M. J. Alava, and T. Ala-Nissila, Phys. jbregtse, J. Rector, B. Dam, and R. Griessen, Phys. Rev.
 8

Lett. 83, 2054 (1999). Kärkkäinen, M. S. Welling, and R. J. Wijngaarden, Eur.

[7] For an early similar observation, see V. K. Horváth, F. Phys. J. B 33, 193 (2003) .
Family, and T. Vicsek, J. Phys. A 24, L25 (1991). [13] D. E. Wolf and L.-H. Tang, Phys. Rev. Lett. 65, 1591
[8] See also D. Forster, D. R. Nelson, and M. J. Stephen, (1990).
Phys. Rev. A 16, 732 (1977). [14] D. Kandel and D. Mukamel, Europhys. Lett. 20, 325
[9] S. A. Janowsky and J. L. Lebowitz, Phys. Rev. A 45, 618 (1992).
(1992); J. Stat. Phys. 77, 35 (1994). [15] M. Myllys, J. Maunuksela, J. Merikoski, and J. Timonen,
[10] A. B. Kolomeisky, J. Phys. A: Math. Gen. 31, 1153 unpublished.
(1998). [16] J. M. Burgers, The Nonlinear Diffusion Equation (Riedel,
[11] M. Ha, M. den Nijs, and J. Timonen, unpublished. Boston, 1974) .
[12] J. Maunuksela, M. Myllys, J. Merikoski, J. Timonen, T.

View publication stats