1.

Significant figures and Scientific notation

Some of you may already be familiar with the concept of significant figures: The numbers that
carry meaningful significance to the measurement resolution of a value. Proper significant
figure use is essential in order to derive meaning from data.

What Are significant figures?

Significant figures are the number of digits in a value that tell us the scale and precision of a
value. Scale refers to the magnitude of a system, such as its physical size or its duration. The
significant figures of a value include:

1. All non-zero digits

2. Any zero digits between non-zero digits
3. Zero digits following a non-zero digit and preceding the decimal point
4. Trailing zero digits that come after the decimal point

We will shortly find that the only exceptions to these rules are trailing zeros in numbers with no
decimal points Let’s see some examples:

• You measure 25.4 degrees Celsius from a digital thermometer. All digits are nonzero
digits, so all digits are significant, the final significant value is 25.4
• On a cold day, you use the same thermometer and measure 04.7 degrees. In this case
there are only 2 significant figures since the leading 0 is not significant
• The voltage through a wire was measured using a multimeter as 0.14 V, this was quoted
in a report as 140 mV. In this case only the first 2 digits are significant, since the trailing
zero does not precede a decimal point.
• You are measuring the fluctuations from a 1 mA current through a wire. The peak
current you measure is 1.0035 mA. All zero digits in this value are between non-zero
digits, hence all digits are significant.
• You measure the same current fluctuations, but this time you use a lower resolution
ammeter. The value displayed on the screen is 1.00. In this case, the only zero digits
come after the decimal point, hence the value has 3 significant figures.
• You now apply a current of 100 mA through the wire. The first ammeter reads 100.00
mA, in this case all 5 digits are significant since 2 of them come follow a non-zero digit
and precede the decimal point whilst the last 2 are trailing zeros that come after the
decimal point.

In the lab we often use significant figures to determine the precision of a measurement made
using a piece of equipment. It also helps us determine how to write a value in scientific
notation.

Scientific notation

A common use case for significant figures is scientific notation. Scientific notation is a
convenient way of writing numbers when they vary over many orders of magnitude. As an
example, the number 251,000,000 written in scientific notion is 2.51 × 108. A key feature of this
way of writing a number is that we begin by writing the first significant figure of the number (in
this case 2) and then the decimal point with all subsequent significant figures written after the

1
decimal point. The second key feature of scientific notation is the × 108, where the power of 10
communicates the order of magnitude of the original number (in our case 8). It should become
clear as you start using scientific notation that if we were to multiply out the number written in
scientific notation, we get back to the expanded way to represent that number. A useful feature
of writing in scientific notation is that we instantly know the precision of a measurement.

The order of magnitude is also a useful tool for estimation. If you calculate the approximate
order of magnitude you expect a result to be, you can easily determine whether or not the result
is reasonable. If the resulting value of your experiment is of the wrong order of magnitude there
might be a problem with your experiment or your assumptions. The order of magnitude
technically refers to an approximation of the base 10 logarithm of a quantity. Practically, it is
often thought of as the power of 10 used when expressing a value in scientific notation and it is
sufficient to use this definition in the lab. So the value 1.86 × 106 has an order of magnitude of 6,
the value 200 has an order of magnitude of 2 and 0.03 has an order of magnitude of -2.

THE EXCEPTION

Let’s go back to our current through a wire example. We use our low resolution ammeter to
measure a 100 mA current through a wire. In this case the display will read 100 mA and there is
only 1 significant figure so one might be tempted to write 1 × 102 however, this would imply that
the true value of the number lies anywhere between 50 and 149, since both produce 1 × 102
when rounded to the nearest 100. Instead we would write 1.00 × 102 since we know our
ammeter is capable of measuring the difference between 100 mA and 101 mA.

2. Observation and Uncertainty

Experimental observation is the foundation of Physics. Experiments allow us to explore the

world around us in a systematic way, and identify underlying relationships that allow us to
better understand our observations. The process of discovery in Physics is often circular:
effects are observed in experiment, theory is developed to explain the observations, and then
experiments are conducted to confirm the theory. Sometimes the first step might be missing,
with theory predicting something and experiment confirming, but either way, experiment is the
ultimate test of the ‘truth’ of a physical theory.

In the first year lab we have the opportunity to explore some fundamental theories in Physics.
Sometimes we might start with observations and then use theory to explain the observations,
while other times we might start with some theory and seek to identify the conditions under
which it applies.

Measurements and observations fall broadly into two categories, and we will refer to these
terms in the experimental labs:

1. Qualitative measurements – these are measurements we record using words. They are
observations or descriptions. Eg. ‘The ruler was made of flexible plastic. Deformation or
changes in temperature may therefore influence its accuracy’.

2
 2. Quantitative measurements – these are measurements we record using numbers. E.g.
a table of measurements of every person’s height in a lab class. Almost always you will
need to assign units to quantitative measurements. In Physics we generally use the SI
system (kg, m, s, A, K, mol, cd). You should use this system of measurement by default
unless you can justify using another.

The factors that can cause uncertainty in a measurement can be divided into three main
categories 1:
• Measurement resolution
• Random fluctuations (or “random errors”)
• Systematic errors
Note, these uncertainties are inherent in an experiment and so are distinguished from mistakes
that are due to carelessness or inexperience (e.g. using a 50 g mass when you meant to use a 100
g mass, or reading the numbers on a measurement display incorrectly). If you make a mistake in
your procedure, you shouldn’t record the result until you have corrected the mistake. We can all
make mistakes, and we should do our best to pick them up, but what we are talking about when
we refer to uncertainty is the intrinsic uncertainty in a measurement, even when you do
everything well.

Measurement resolution: This is usually the display resolution of the instrument, but it is
possible to reduce the measurement resolution below the display resolution with the
experimental method.

If we are measuring some quantity X using an instrument, the display resolution is the smallest
variation ΔX that the instrument can detect.
• On an analogue instrument (e.g. a ruler) the display resolution is the minimum distance
between two ticks of the scale (e.g. a standard ruler has a display resolution ΔX = 1 mm).
• On a digital instrument the display resolution is the unit value of the least significant digit
(e.g. for a digital scale with reading 43.4 g, the display resolution is ΔX = 0.1 g).
We shall take the uncertainty due to measurement resolution, δXmax, as the maximum
resolution uncertainty δXmax = ΔX/2. We would therefore report the scale measurement as (43.4
± 0.05) g. Note, if we instead wished to use a resolution uncertainty in line with how we treat
random fluctuations, we would use 𝛿𝛿𝑋𝑋𝑟𝑟𝑟𝑟𝑟𝑟 = ∆𝑋𝑋/√12 (see Fornasini p47). In the labs you will
quote the maximum uncertainty δXmax for measurement resolution uncertainty.

Example 3.19 (Fornasini p35): A time interval τ is measured by means of a digital stopwatch. A
value 34.27 s is read on the display. The unit least significant digit value is 0.01 s. We should quote
the measurement as τ = (34.27 ± 0.005) s. Note, we are

Example 3.20 (Fornasini p36). The height of a paper sheet is measured by a ruler which has
resolution ∆X = 1 mm. The end of the sheet is equidistant from the ticks corresponding to 296 and
297 mm. We can take 296.5 mm as the central value. We should report the measurement as H
= (296.5 ± 0.5) mm.

1
Further detail on content presented in the Introduction can be found in: P. Fornasini, The Uncertainty in
Physical Measurements, Springer (2008) [Available on Canvas]

3
Random fluctuations: A second cause of uncertainty in a measurement is due to random
fluctuations, which historically have often been called “random errors”. If we repeat a
measurement multiple times, there are two possible outcomes:
a) All the measurements fall within the same resolution interval (e.g. always measuring 29
mm for the length of a piece of paper using a ruler with resolution interval ΔX = 1 mm).
b) The different measurements fall randomly into different resolution intervals (e.g.
recording the time taken for a ball to fall from a bench to the floor, using a stopwatch with
resolution interval ΔX = 0.01 s, but recording values [0.56 s, 0.61 s, 0.55 s]).
In case a) any random fluctuations are below the display resolution, so we report the uncertainty
in the measurement as δX = ΔX/2, as before. In case b) the variation could be due to reading
errors (e.g. inadequacy of the eye for separating between values) or due to background noise (e.g.
due to vibrations in the environment) or due to some other element of the measurement system
(for instance, the reaction time of the experimenter). Determining the uncertainty δX in the
presence of random fluctuations in case b) is more involved and we need some theory for this.

Consider the situation in which we have obtained N different measurements of a quantity X,

such that we have a list of values denoted here as [x1, …, xN]. The measured value we should
quote for X is the mean of these N values μ, i.e.

∑𝑁𝑁
𝑖𝑖=1 𝑥𝑥𝑖𝑖
𝜇𝜇 = (1)
𝑁𝑁
To quantify the uncertainty in this value for the mean ideally we would conduct all the
measurements again to get another value for the mean. If we do this again and again we will
end up with a distribution of values for the mean, and from this we can estimate the uncertainty
in the mean. We can side-step these additional experiments by using the measured values we
used in (1) to estimate the limiting standard deviation σ (the Greek symbol ‘sigma’):

1
𝜎𝜎 = � ∑𝑁𝑁
𝑖𝑖=1(𝑥𝑥𝑖𝑖 − 𝜇𝜇 )
2 (2)
𝑁𝑁−1

The uncertainty in the mean (1), known as the standard uncertainty or the standard error of
the mean σe, can then be estimated to be:
𝜎𝜎
𝜎𝜎𝑒𝑒 = (3)
√𝑁𝑁
We can now report a measurement for the quantity X in terms of the central value for the X
quantity, which we shall denote X0, and the uncertainty in this quantity δX i.e. 𝑋𝑋 = 𝑋𝑋0 ± 𝛿𝛿𝛿𝛿
with:

𝑋𝑋0 ≅ 𝜇𝜇, 𝛿𝛿𝛿𝛿 ≅ 𝜎𝜎𝑒𝑒 (4)

These results come from statistical theory, which you may encounter in your maths classes, or
you can read the relevant Sections in Fornasini (e.g. look at p57).

Example: Three measurements of a mass m yield values of (20.5 ± 0.1) g, (20.3 ± 0.1) g, (20.6 ±
0.1) g. We can use the Excel function AVERAGE() to find the mean and STDEV() to find the limiting
standard deviation. We report the mass as m = (20.47 ± 0.09) g. Note, the uncertainty in this
mass is less than the uncertainty due to measurement resolution, 0.1g.

4
Systematic errors: these typically modify a measurement by the same amount whenever the
measurement is repeated. There are a broad range of possible sources of systematic error, but
for our purposes we consider two main types:
a) Systematic error due to an instrument
b) Systematic error due to measurement procedure
Case a) is the situation in which some intrinsic bias exists in an instrument, for example a scale
that has been wrongly calibrated, so that all measurements are 50 g heavier than they should be.
Case b) instead involves a systematic error due to the procedure, and can be much more varied
in nature. For instance, if a stopwatch is being used by an experimenter, there may be some
reaction time, leading to a delayed start in recording the time (in this case uncertainty in the
measurement may have a mix of both random and systematic components). We should look for
possible sources of systematic error when we are conducting experiments.

One systematic error commonly encountered when measuring by eye is parallax error. Parallax
is the difference in the apparent position of an object depending on the line of sight (see figure 1).
Its effects can be reduced by attempting to align the line of sight at 90 degrees with the ruler.

Fig. 1 An illustration of how parallax error can affect a measurement.

FINAL IMPORTANT POINTS: Whatever the sources of the uncertainty, the uncertainty when
reported is always expressed to a maximum of two significant figures. When rounding, the
least significant digit of the quantity should be of the same order as the least significant digit
of the uncertainty.

3. Uncertainty Propagation

We often have the situation where the quantity we are interested in is indirectly found from some
functional relationship with quantities that are directly measured. For instance, we may wish to
find the acceleration due to gravity g by measuring the period of a pendulum T and its length l and
then using the relation 𝑔𝑔 = 4𝜋𝜋 2 𝑙𝑙/𝑇𝑇 2 . The challenge in this situation is to estimate the uncertainty
in the calculated value for g given uncertainties in T and l. To do this we need to “propagate the
uncertainty” from the measured quantities to the calculated quantity. To do this, we rewrite the
problem in a general way. Let the quantities X, Y, Z, … be directly measured, while the quantity Q
is indirectly measured through the functional relation Q = f(X,Y,Z,…); for example Q = X + Y or Q =
XY/Z. To proceed we shall assume that the measurements of X, Y, Z, … are statistically
independent, i.e. deviations in one quantity have no connection to deviations in another quantity.

5
We shall present the results, but illustrate the method with the simplest example of a quantity
that depends on only one measured quantity, for instance Q = f(X). If we have reported a value
for X as X = X0 ± δX then the central value for Q is approximated as Q0 = f(X0). Using Taylor’s
theorem 2 we can find the uncertainty in Q due to the uncertainty 𝛿𝛿𝛿𝛿 and the measured quantity
X0:
𝑑𝑑𝑑𝑑
𝑄𝑄 = 𝑄𝑄0 ± 𝛿𝛿𝛿𝛿 ≅ 𝑓𝑓(𝑋𝑋0 ) ± � � 𝛿𝛿𝛿𝛿 (5)
𝑑𝑑𝑑𝑑 𝑋𝑋=𝑋𝑋0

So given Q = f(X) we see that:

𝑑𝑑𝑑𝑑
𝑄𝑄0 ≅ 𝑓𝑓(𝑋𝑋0 ), 𝛿𝛿𝛿𝛿 ≅ � � 𝛿𝛿𝛿𝛿 (6)
𝑑𝑑𝑑𝑑 𝑋𝑋=𝑋𝑋0

Returning to the general problem, if Q = f(X,Y,Z,…) then the uncertainty in Q, δQ, is given by:

𝜕𝜕𝜕𝜕 2 𝜕𝜕𝜕𝜕 2 𝜕𝜕𝜕𝜕 2

𝛿𝛿𝛿𝛿 ≅ �� � (𝛿𝛿𝛿𝛿)2 + � � (𝛿𝛿𝛿𝛿)2 + � � (𝛿𝛿𝛿𝛿)2 + ⋯ (7)
𝜕𝜕𝜕𝜕 𝑋𝑋=𝑋𝑋0 𝜕𝜕𝜕𝜕 𝑌𝑌=𝑌𝑌0 𝜕𝜕𝜕𝜕 𝑍𝑍=𝑍𝑍0

Note … refers to additional terms as needed. This formula can be simplified (using the Schwarz
inequality, for those interested) to find the maximum uncertainty in Q, denoted (δQ)max:
𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕 𝜕𝜕𝜕𝜕
(𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚 ≅ � � 𝛿𝛿𝛿𝛿 + � � 𝛿𝛿𝛿𝛿 + � � 𝛿𝛿𝛿𝛿 + ⋯ (8)
𝜕𝜕𝜕𝜕 𝑋𝑋=𝑋𝑋0 𝜕𝜕𝜕𝜕 𝑌𝑌=𝑌𝑌0 𝜕𝜕𝜕𝜕 𝑍𝑍=𝑍𝑍0

𝜕𝜕𝜕𝜕
Note, 𝜕𝜕 signifies partial differentiation, tells us to take the derivative of the function Q with
𝜕𝜕𝜕𝜕
respect to X, assuming everything else is held constant.

Relationship between Central value of Uncertainty ±𝛿𝛿𝛿𝛿 is Maximum

𝑄𝑄 and quantities X and Q, Q0, given calculated using: Uncertainty
Y central values X0 (𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚
and Y0
Multiplying by the
constant a (with 𝑄𝑄0 ≅ 𝑎𝑎𝑋𝑋0 𝛿𝛿𝛿𝛿 ≅ |𝑎𝑎|𝛿𝛿𝛿𝛿 (𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚 ≅ |𝑎𝑎|𝛿𝛿𝛿𝛿
negligible uncertainty):
𝑄𝑄 = 𝑎𝑎𝑎𝑎
Adding or subtracting:
𝑄𝑄 = 𝑎𝑎𝑎𝑎 + 𝑏𝑏𝑏𝑏 𝑄𝑄0 ≅ 𝑎𝑎𝑋𝑋0 + 𝑏𝑏𝑌𝑌0 𝛿𝛿𝛿𝛿 ≅ �𝑎𝑎2 (𝛿𝛿𝛿𝛿)2 + 𝑏𝑏 2 (𝛿𝛿𝛿𝛿)2 (𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚
𝑄𝑄 = 𝑎𝑎𝑎𝑎 − 𝑏𝑏𝑏𝑏 𝑄𝑄0 ≅ 𝑎𝑎𝑋𝑋0 − 𝑏𝑏𝑌𝑌0 ≅ |𝑎𝑎|𝛿𝛿𝛿𝛿 + |𝑏𝑏|𝛿𝛿𝛿𝛿

Multiplying or dividing:
𝑄𝑄 = 𝑋𝑋𝑋𝑋 𝑄𝑄0 ≅ 𝑋𝑋0 𝑌𝑌0 𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2 (𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚
𝑄𝑄 = 𝑋𝑋/𝑌𝑌 𝑄𝑄0 ≅ 𝑋𝑋0 /𝑌𝑌0 � � ≅� � +� � |𝑄𝑄0 |
𝑄𝑄0 𝑋𝑋0 𝑌𝑌0
𝛿𝛿𝛿𝛿 𝛿𝛿𝛿𝛿
≅ +
|𝑋𝑋0 | |𝑌𝑌0 |
Raising to a power:
𝑄𝑄 = 𝑋𝑋 𝑛𝑛 𝑄𝑄0 ≅ 𝑋𝑋0𝑛𝑛 𝛿𝛿𝛿𝛿 𝛿𝛿𝛿𝛿 (𝛿𝛿𝛿𝛿)𝑚𝑚𝑚𝑚𝑚𝑚 𝛿𝛿𝛿𝛿
≅ 𝑛𝑛 ≅ 𝑛𝑛
|𝑄𝑄0 | |𝑋𝑋0 | |𝑄𝑄0 | |𝑋𝑋0 |

Table 1: Summary of rules, implemented using Eq. (7) [column 3] and (8) [column 4]

2
You don’t need to know this at this stage, but it is a beautiful mathematical theorem, used throughout
Physics.

6
Example: A student wants to calculate the weight (W) of a ball using the formula for weight: 𝑊𝑊 =
𝑚𝑚𝑚𝑚. She measures the mass of the ball to be 𝑚𝑚 = (1.3 ± 0.2) 𝑘𝑘𝑘𝑘. She looks up the acceleration
due to gravity, g, and finds that it has a value of 𝑔𝑔 = 9.81 𝑚𝑚/𝑠𝑠. Since this is a constant that has
been measured with advanced lab equipment and techniques, she assumes the uncertainty on
this value is very small compared to the uncertainty on her measurements. She thus treats g as
a constant. She calculates the weight of the ball as 𝑊𝑊 = 𝑚𝑚𝑚𝑚 = 1.3 𝑘𝑘𝑘𝑘 × 9.81 𝑚𝑚/𝑠𝑠 2 = 12.75 𝑁𝑁.
She now needs to calculate the uncertainty on this quantity. Since she is multiplying a
measured quantity (𝑚𝑚) by a constant (𝑔𝑔), she looks at the first row of the table, and uses the
rule δQ = |𝑎𝑎|δX. For her quantities, this becomes δ𝑊𝑊 = δ𝑚𝑚 × 𝑔𝑔 = 0.2 𝑘𝑘𝑘𝑘 × 9.81 𝑚𝑚/𝑠𝑠2 = 2 𝑁𝑁. She
thus writes down her result for the weight of the ball as 𝑊𝑊 = (13 ± 2)𝑁𝑁.

Example: The student now wants to find the density of the ball in the previous example. She
knows the mass is 𝑚𝑚 = (1.3 ± 0.2) 𝑘𝑘𝑘𝑘 and measures the volume to be 𝑉𝑉 = (0.15 ± 0.04) 𝑚𝑚3 . She
uses the formula for density (𝜌𝜌), 𝜌𝜌 = 𝑚𝑚/𝑉𝑉 to find that the density is 𝜌𝜌 = 1.3 𝑘𝑘𝑘𝑘 /0.15 𝑚𝑚3 = 8.67
𝑘𝑘𝑘𝑘/𝑚𝑚3. Since she has divided two quantities with uncertainties to find the density, she finds the
uncertainty in the density by using the formula in the third row of the table:
𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2
� � ≅� � +� �
𝑄𝑄0 𝑋𝑋0 𝑌𝑌0
𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2 𝛿𝛿𝛿𝛿 2
For her quantities, this becomes � � = � � + � � which leads to 𝛿𝛿𝛿𝛿 = 8.67 ×
𝜌𝜌0 𝑚𝑚0 𝑉𝑉0
0.2 2 0.04 2
��� � + � � � = 2.7𝑘𝑘𝑘𝑘/𝑚𝑚3 . She thus writes down her final result as 𝜌𝜌 = (9 ± 3) 𝑘𝑘𝑘𝑘/𝑚𝑚3 .
1.3 0.15

Note, if the measurements of X and Y are not statistically independent then we can’t use Eq.
(7) or the third column of Table 1 to propagate uncertainties. However, Eq. (8) and column 4
still hold, so we can use these.

Fornasini rules of thumb and examples (p209 – 210)

7
Note m* is the mean of the measurements, i.e. the value X 0.