Classroom Note Radioactivity Half Lives Considered as Data

The numbers in the rst column of Table cover an unusually wide range some orders of magnitude They provide quite a dramatic example of the e ectiveness of data transformation one that I think will be unfamiliar to most statisticians The numbers in the rst column of Table are half lives of a certain set of radioactive nuclides I will now give a few sentences of introduction to radioactivity One process of radioactive decay is the emission of an particle This reduces the mass number of the nucleus that is the total number of protons and neutrons by Another process is by the emission of a particle This does not change the mass number of the nucleus Consequently there are four series of radioactive nuclides with mass numbers that respectively have remainders and when divided by In the rst column of Table are the half lives of emitting nuclides in the rst of these series starting from Fm an isotope of Fermium

The 11 numbers in the rst column of Table 1 cover an unusually wide range, some 25 orders of magnitude.They provide quite a dramatic example of the e ectiveness of data transformation, one that I think will be unfamiliar to most statisticians.
The numbers in the rst column of Table 1 are half-lives of a certain set of radioactive n uclides.I will now g i v e a few sentences of introduction to radioactivity.One process of radioactive decay is the emission of an particle.This reduces the mass number of the nucleus (that is, the total number of protons and neutrons) by 4. Another process is by the emission of a particle.This does not change the mass numberofthenucleus.Consequently, there are four series of radioactive nuclides | with mass numbers that respectively have remainders 0, 1, 2, and 3 when divided by 4 .In the rst column of Table 1 are the half-lives of -emitting nuclides in the rst of these series, starting from 252 100 Fm (an isotope of Fermium  Box-and-whisker plot of log(half-life).(The \box" extends from the lower quartile to the upper quartile, and thus accounts for the middle 50% of observations, and the \whiskers" extend out as far as the smallest and the largest observations.There being 11 observations, the quartiles have been taken to be the 3rd and the 9th smallest.) ;10 ;5 0 5 10 15 20 Figure 2 .Cumulative distribution (plotted as the z-transformation) of log(half-life).q q q q q q q q q q q ;10 ;5 that has a half-life of about 1 day), and proceeding by w ay o f 232 90 Th (an isotope of Thorium that has a half-life of about 10 10 years), to end with the stable 208 82 Pb (an isotope of Lead).There are some -emitters in the series, which h a ve been omitted from this table.It should also be mentioned that one might question exactly what nuclides should be included in this series.For example: (i) The series is usually considered to start with 232 90 T h I h a ve t a k en it back as far as 252 100 Fm, but not as far as 256 102 No. (ii) Some nuclides decay b y b o t h -emission and -emission.However, details like these are unimportant for present purposes.
It is di cult to get to grips with a set of numbers that vary so much from each other, and it might seem a lost cause to try to discover any regularity or law applicable to them.But we shall see that this is overly pessimistic.Column 2 of Table 1 shows the logarithms of the half-lives.If one did not know t h e s e w ere logarithms to base 10, one would consider them and their box-and-whisker plot (Figure 1) to be very ordinary-looking.
Can the log-half-lives be described by some well-known statistical distribution, such as the normal?A standard way of proceeding is to plot z against log(half-life), as this will be approximately linear in such a case here, z is de ned via (z) = cumulative proportion of observations, where is the cumulative normal integral.C o l u m n 3 o f T able 1 shows the rank of the observations, from smallest to largest, and column 4 shows the value of z that corresponds to rank/12.(It is common to use i=(n +1), not i=n, as the \plotting position" of the ith smallest observation out of a total of n.) Figure 2 shows that this plot is indeed approximately linear thus we can conclude that the half-lives have a p p r o ximately a log-normal distribution.(The Weibull distribution, with the same general shape as the log-normal, is also a good t.) A common disadvantage of transforming data is that there is a loss of interpretability | in this case, time is a familiar variable, but it is by no means obvious what the meaning of log(time) is.Happily, in this example there is an interpretation of log(time), in terms of a theory of nuclear decay b y -emission.This theory postulates the quantum-mechanical tunnelling of an -particle through the Coulomb barrier formed by the attractive strong-interaction forces and the repulsive electrostatic forces.See, for example, 1] (pp.115{119) or 2] (pp.551{553).The energy of the -particles emitted by a g i v en nuclide is just as characteristic of the decay process as the half-life is, and the theory referred to predicts a relation between these two things this relationship will be close in the sense of there being a high correlation, and strong in the sense that a small change in energy leads to a large change in half-life.The relationship is one of linearity between log(half-life) and energy ;1=2 .(And some reasons are known why the relationship is not exact.)So, instead of the distribution of half-life, let us look at the distribution of energy ;1=2 .Table 2 shows the values of this.The log-normal distribution of halflife that we had in Table 1, extending over 25 orders of magnitude, is re ecting a normal distribution of energy ;1=2 that extends over a fraction of an order of magnitude, see Figure 3. (Incidentally, I make no claim that it is only the ; 1 T. P. HUTCHINSON Figure 3 .Cumulative distribution (plotted as the z-transformation) of energy ;1=2 .0:30 0:35 0:40 0:45 0:50 q q q q q q q q q q q ;1:5 0:30 0:35 0:40 0:45 0:50 q q q q q q q q q q q ;10 0 10 ;1:383 several other powers, also.)For this set of nuclides, the scatterplot of energy ;1=2 and log(half-life) is shown as Figure 4 the correlation is 0.94.I need hardly say that there is much more that could be done with such data as this | di erent s e t s o f n uclides could be considered, and distributions other than the log-normal could be used as the standard.However, I think there is a fundamental limitation on this line of work, which is that it is usual to think of half-lives and disintegration energies as constants, not as realisations of some random process.It would therefore be misguided to ask why they should exhibit particular statistical features (such as a log-normal or a Weibull distribution), or imagine a stochastic mechanism by w h i c h they are generated.Presuming that this is right, it would be meaningless to test, for example, whether the half-lives in the four disintegration series (these were mentioned in the second paragraph of this paper) are from the same distribution or not | there is no population and no procedure of random sampling.

Table 2 .
Disintegration energies of the nuclides in Table1.