Tracer diffusion†

MM 677: Diffusion and Kinetics

M P Gururajan‡
28 February, 2023

‡ Department of Metallurgical Engineering and Materials Science

Indian Institute of Technology Bombay, Powai
Mumbai, Maharashtra 400076 INDIA

† This material is shared under Creative Commons license CC BY. See

https://creativecommons.org/licenses/ for details of the license.

1
 Tracer diffusion†
MM 677: Diffusion and Kinetics

M P Gururajan‡
28 February, 2023

In this tutorial, we will use solve two problems. One is the tracer diffusion data
from which we evaluate the diffusivity. The second is the Boltzmann-Matano
anlysis to obtain the composition dependent diffusivity. For this, we will take
the diffusion couple experimental data from the Cu-Zn system and carry out
the Matano analysis and the evaluation of concentration dependent diffusivity.

1 Tracer diffusion: diffusivity

The thin film solution is typically used to obtain the tracer diffusion coefficient.
Typically a thin film of radioactive tracer is plated on a cylinder and the diffusion
is allowed to take place for a set amount of time. Then, by sectioning the
sample and measuring the radioactivity, and comparing the results with thin
film solution, one can obtain the diffusivity of tracer in the given material.
In an experiment on oxygen tracer diffusion in Fe-doped SrTiO3 , a sample
annealed at 973 K for 86300 seconds, the depth and the normalised oxygen
isotope fraction are as given in Table 1.
This data given in the table is also available as an csv file named TracerDiffu-
sionData.csv
Using this data, evaluate the diffusivity by fitting a straight line to the logarithm
of concentration and square of distance. This is what we do in this session. Of
course, we can alsodetermine the diffusivity by fitting the data to thin film
solution, and compare the two diffusivities. This we will do in the next tutorial.

1.1 Read and plot the data

Let us first read the data, view it and plot it:

X <-read.csv("Data/TracerDiffusionData.csv")
View(X)
plot(X$x,X$c,xlab="Depth in cm", ylab="Isotope fraction",main="Oxygen tracer diffusion in Fe

2
 Distance (in cm Normalised fraction of iostopes

0.0002122 0.015616
0.0004751 0.015085
0.0007642 0.014471
0.0010528 0.01353
0.0014736 0.012754
0.0018941 0.011814
0.0022095 0.011119
0.0025514 0.010465
0.0029723 0.00977
0.0033933 0.009117
0.0037616 0.008463
0.004209 0.00781
0.0045509 0.007197
0.0049194 0.006666
0.005367 0.006136
0.0058674 0.005606
0.0063939 0.004912
0.0070523 0.004219
0.0078163 0.003567
0.008633 0.002875
0.0093183 0.002469
0.0102673 0.001982
0.0111371 0.001454
0.0119809 0.001131
0.0133787 0.00077
0.0148294 0.000491
0.0164914 0.000336
0.0185227 0.000183
0.0203169 0.000152
0.0218472 0.00016

Table 1: The tabulated values of distance versus isotope concentration. See R

A De Souza and M Martin, phys. stat. sol. (c) 4, No. 6, 1785–1801 (2007)
(DOI 10.1002/pssc.200675227)

3
 Oxygen tracer diffusion in Fe−doped SrTiO$_3$
0.015
0.010
Isotope fraction

0.005
0.000

0.000 0.005 0.010 0.015 0.020

Depth in cm

Figure 1: Plot of the depth (in cm) versus isotope fraction; oxygen tracer diffu-
sion in Fe-doped SrTiO3 , for a sample annealed at 973 K for 86300 seconds

4
Note that x is in cm and the normalised isotope fraction is a proxy for the
concentration.
Let us now consider the thin-film source solution. This solution is given by

M x2
c(x, t) = √ exp − (1)
4πDt 4Dt
Taking logarithm, we obtain

x2
 
M
log c(x, t) = log √ − (2)
4πDt 4Dt

1.2 Evaluation of D : x2 vs logc fit

Let us start with the fitting:

lnc <- log(X$c)

x2 <- (X$x)^2
fit <- lm(lnc~x2)
t = 86300
D = -1/(4*t*coefficients(fit)[2])

Let us now try to compare the fit and the data first:

fity = exp(coefficients(fit)[1])*exp(-(X$x)^2/(4*D*t))
plot(X$x,fity,type="l",ylim=c(0,0.016),xlab="Depth in cm", ylab="Isotope fraction",main="Oxy
points(X$x,X$c)

From the plot, it is clear that the fit is not great. This is because of a factor
called tracer surface exchange coefficient; the paper of De Souza and Martin
has the details about this coefficient and the correct form of solution to fit – in
their appendix. Even though we will not go into the details of this correction,
the tutorial does show how to fit the data to the thin-film source solution.
The second thing we need to do is to plot the residuals. Let us do that.

plot(X$x,X$c-fity,xlab="Depth in cm", ylab="Residuals",main="Residual plot for the fit")

The residuals show a pattern indicating that the errors are not normally dis-
tributed and are systematic.
Finally, we can also look at the fit paramters using the commands summary and
confint:

summary(fit)

##
## Call:
## lm(formula = lnc ~ x2)

5
 Oxygen tracer diffusion in Fe−doped SrTiO$_3$
0.015
0.010
Isotope fraction

0.005
0.000

0.000 0.005 0.010 0.015 0.020

Depth in cm

Figure 2: Plot of the data on depth (in cm) versus isotope fraction and the fit
to x2 versus log(c); oxygen tracer diffusion in Fe-doped SrTiO3 , for a sample
annealed at 973 K for 86300 seconds

6
 Residual plot for the fit
0.006
0.004
Residuals

0.002
0.000

0.000 0.005 0.010 0.015 0.020

Depth in cm

Figure 3: Plot of residuals of the fit

7
##
## Residuals:
## Min 1Q Median 3Q Max
## -0.55447 -0.33694 -0.03682 0.28201 1.09949
##
## Coefficients:
## Estimate Std. Error t value Pr(>|t|)
## (Intercept) -4.693e+00 9.104e-02 -51.55 <2e-16 ***
## x2 -1.078e+04 5.788e+02 -18.63 <2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.4037 on 28 degrees of freedom
## Multiple R-squared: 0.9253,Adjusted R-squared: 0.9227
## F-statistic: 347 on 1 and 28 DF, p-value: < 2.2e-16

confint(fit)

## 2.5 % 97.5 %
## (Intercept) -4.879962 -4.506975
## x2 -11967.903958 -9596.540958

We can see that the fit parameters are good even though the fit is not great from
our point of view – in terms of systematic errors and the tracer exchange coef-
ficient correction. So, in addition to statistical measures, it is always essential
to also look at the other aspects to evaluate the fit.

2 Diffusivity: non-linear regression of data

Let us now fit the data to thin-film source solution. This solution is given by

M x2
c(x, t) = √ exp − (3)
4πDt 4Dt
R allows for non-linear fit using the functional form. Knowing the solution for
D, we can give the initial guess for D. For M, we start with unity.

X <-read.csv("Data/TracerDiffusionData.csv")
c <- X$c
x <- X$x
t = 86300
M = 1
D = 2.5e-10
fit <- nls(c ~ (M/sqrt(4*pi*t*D))*exp(-(x^2)/(4*D*t)),
start = list(M=1,D=2.5e-10))
summary(fit)

8
##
## Formula: c ~ (M/sqrt(4 * pi * t * D)) * exp(-(x^2)/(4 * D * t))
##
## Parameters:
## Estimate Std. Error t value Pr(>|t|)
## M 1.485e-04 4.155e-06 35.73 < 2e-16 ***
## D 1.091e-10 7.357e-12 14.83 8.7e-15 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error: 0.0008602 on 28 degrees of freedom
##
## Number of iterations to convergence: 9
## Achieved convergence tolerance: 2.467e-06

fity = (coefficients(fit)[1]/sqrt(4*pi*t*coefficients(fit)[2]))*exp(-(x^2)/(4*coefficients(f
plot(X$x,fity,type="l",ylim=c(0,0.016),xlab="Depth in cm", ylab="Isotope fraction",main="Oxy
points(X$x,X$c)

9
 Oxygen tracer diffusion in Fe−doped SrTiO$_3$
0.015
0.010
Isotope fraction

0.005
0.000

0.000 0.005 0.010 0.015 0.020

Depth in cm

plot(x,c-fity,xlab="Position",ylab="Residual",main="Plot of residuals")

10
 Plot of residuals
0.0020
0.0015
0.0010
Residual

0.0005
0.0000
−0.0005
−0.0010

0.000 0.005 0.010 0.015 0.020

Position

Note that the diffusivity is the same order as earlier – but the number itself
is off by a factor or 2 or so. In this case also, the residuals are not normally
distributed; and, the data and the fit are not great. As indicated in the previous,
there are other physics that we have missed incorporating here. However, this
exercise still indicates how we can do non-linear least squares fitting to data.

11