Physics Part IA

Cambridge University, 2012-2013

Notes for Experimental Physics

Dimensional analysis:
The dimension of a quantity specifies the nature of that quantity. It can be
presented in two forms specific and abstract:
Specific dimensions: Commonly referred to as units. The standard specific
dimensions of mass, length and time are kg, m and s, respectively.
Abstract dimensions: These do not specify the particular unit in which a
value is given, only its fundamental nature. The abstract dimensions of
mass, length and time are expressed as [M], [L] and [G], respectively.
Dimensional analysis:
Since dimensions specify the nature of a quantity, it is possible to deduce how
given quantities are related to other quantities by comparing their dimensions.
This is called dimensional analysis:

Example: Determining how the period

mass

and length

of a simple pendulum depends on

l :
T

Abstract dimension of

This necessitates the presence of another, constant quantity in the

expression which incorporates units of time this is the gravitational

T m l g

is [M] and

is [L]

g , which has abstract dimension [L][T]-2

constant,
Expressing

is [T],

generally in terms of

m ,l

and

g :

[ T ] =[ M ] [ L ] ( [ L ][ T ]

) =[ M ] [ L ] + [ T ]2

=0,2 =1 =
1

T l 2 g

1
1
1
=0 =
2
2
2

1
2

l
g

Limitations of dimensional analysis:

Dimensional analysis can obviously only be applied to quantities which
have dimensions
It therefore cannot give the constant of proportionality in an equation,
since such constants are dimensionless numbers
The relationship between quantities obtained via dimensional analysis will
not be unique if some or all of the constituent quantities can be combined
to give a dimensionless quantity. Such a dimensionless quantity could be
multiplied into the equation or excluded without affecting the overall

Physics Part IA
Cambridge University, 2012-2013
dimension, and hence there exist an infinite number of possibilities for the
form of such an equation

Experimental error and uncertainty:

The final result of any experiment is always expressed in the form
where

is the value of the result obtained and

A A ,

is its associated

uncertainty. Uncertainty is caused by error, which is intrinsic to every

experimental procedure, and can be one of two types random or systematic.

Physics Part IA
Cambridge University, 2012-2013
Types of error:
Random error: The effect of natural fluctuations in measured values from the
central/mean value:
Equally likely to be positive or negative
Can be detected by analysing the spread of data
Can be reduced by repeating measurements to reduce the spread of the
data
Generally affects the precision of the result
Systematic error: The effect of an experimental arrangement which differs
from that assumed:
Not equally likely to be positive or negative
May vary or remain constant throughout the experiment
Not a result of the natural spread of data, hence cannot be reduced by
repeating measurements
If its source and size are identified, it can be removed from the final result
Systematic error generally affects the accuracy of the result
Distribution of experimental data:
By the Central Limit Theorem, experimental data collected with a large number
of trials is always normally-distributed, thus follows a Gaussian distribution of the
form:
2

f ( x )=

Where

1
e
2 2

( x )
2
2

is the mean of the distribution and

is its standard deviation. This

leads to the following properties of experimental data:

The mean obtained from the data

mean of the distribution

is an approximation of the true

, such that

x
, thus

The maximum value of the distribution is at

is the best

indication of the true value represented by the data

Data are therefore always represented by

Experimental uncertainty is the uncertainty in this mean value,

for all trials conducted

mean

Result=x mean
Where
and

mean

and

mean

are always given to the same number of decimal places,

is given to two significant figures if it begins with a 1 or a 2, and one

significant figure otherwise.

Obtaining the value of

from experimental data:

Physics Part IA
Cambridge University, 2012-2013
Since experimental data

usually have discrete values

x , the mean is

simply given by:

n

1
x = x i
n i=1
Where
trial.

is the number of trials conducted and

x i is the result of the

i th

Physics Part IA
Cambridge University, 2012-2013

mean

Obtaining the value of

from experimental data:

The error in the mean can be obtained from the unbiased form of the standard
deviation in the data,

2
^ , where unbiased refers to the factor of

which is introduced to account for the fact that

true mean of the distribution,
N

^ 2=

( n1n )

is only an estimate of the

( x ix )2 ( x ix )2
i=1

^ 2

Relating

n/(n1)

Var ( X )=Var

to

= i=1

n1

mean :

X 1+ X 2+ + X n
1
= 2 [ Var ( X 1 ) +Var ( X 2 ) + +Var ( X n ) ]
n
n

1
1
^ 2
Var ( X ) = 2 ( ^ 2+ ^ 2+ + ^ 2) = 2 ( n ^ 2) =
n
n
n

mean=

2
^
n

The result of an experiment is therefore given by:

Result=x

Where

^
n

is the number of trials conducted. It can therefore be seen that by

increasing the number of trials, the precision of the experimental result can be
increased, although the increase in precision becomes negligible for large

n .

Combining experimental errors:

The finite approximation to an exact differential
an error in the quantity

f . The error

the errors in its constituent quantities

depends on two others,

can be used to represent

can then be expressed in terms of

x 1 , x 2 x n . For a quantity which

f (x , y ) , errors can be combined according to two

differential statements of equivalence:

First statement:

Second statement:

Physics Part IA
Cambridge University, 2012-2013

f =

f
f
x+
y
x
y

By the Central Limit Theorem, distributions

which depend upon more than one variable
approach Gaussian distributions very quickly,
thus combinations of their standard
deviations/errors are Gaussian:

f f x f y
=
+
f x f y f

Where

f x
,
f x

and

fractional errors in

f x N ( fx , ( f x ) 2) , f y N ( f y , ( f y )2 )

y
y
f ,

are

y , respectively.

and

f =f x +f y :

If

( f )2=( f x )2 + ( f y )2
2

( f ) =

f
f
x +
y
x
y

)(

f
f
=
f
x

x
f
+
f
y

y
f

( ) ( )( ) ( )( )

The four principal laws used to combine errors are derived from these two
statements:

f =x n

( Z =A

):

f f x
x
x
=
=n x n1 n =n
f x f
x
x

f
x Z
A
=n
=n
f
x
Z
A

f =kx

or

f =x k :

f =x y :
2
2
( f )2= f ( x )2+ f ( x )2
x
x

( )

( )

( f )2=( x )2+ ( y )2 ( Z)2=( A )2+ ( B )2

f =xy

f=

or

x
y :

f f x
x x
=
=k =
f x f
kx x

( ) ( )( ) ( )( )

f f x x
=
=
f x f
x

f x Z A
=
=
f
x
Z
A

f 2 f
=
f
x

x 2 f
+
f
x
2

f
x
y
= y2
+ x2
f
xy
xy

y
f

( ) ( ) ( )

f 2 x 2 y 2
Z 2 A 2 B
=
+

=
+
f
x
y
Z
A
B

( ) ( )( ) ( ) ( )( )

Physics Part IA
Cambridge University, 2012-2013

Techniques for eliminating systematic errors:

Experimental procedure can be optimised to eliminate systematic errors in four
main ways:
Difference methods: The difference between two measurements is taken to
eliminate the effects of an extraneous, constant variable present in both
measurements.
For example, when measuring the force due to gravity acting on a ball as it rolls
down an incline, measurements can be taken at two different incline angles,

and

2 . One measurement can then be subtracted from the other to

eliminate the effect of friction:

m a1=mg sin 1F f , ma 2=mg sin 2F f

m ( a1 a2 ) =mg(sin 1sin 2 )
Zero offsets: The equipment is adjusted so that values are always measured from
zero. If the equipment has no zero offset error, this can be as simple as pressing
the tare button on an electronic balance, however if the equipment has an
intrinsic zero offset error, this must be eliminated manually as follows:
Take two measurements for which the error is equal and opposite
Average the measurements
Variation with time: Measurements are taken in an appropriate sequence to
identify and hence mitigate the effects of environmental change with time, or
equipment drift with time:
measured
values
true
value

measured
values 1
measured
values 2
true
value

Measurements taken in sequence from

first to last variation of equipment or
environmental conditions with time
causes experimental drift, such that
the gradient is steeper than it should
be.

Measurements taken in two groups

first the odd data points from first to
last, then the even data points from
first to last. Here, the experimental drift
can be seen and can be eliminated by
averaging the gradients.

Calibrating equipment: Systematically inaccurate equipment can be calibrated

through comparing it to more accurate standard equipment:
For example, a wooden ruler should be checked against a more accurate
standard metal ruler. It is necessary to check the distance actually measured
using the wooden ruler rather than its total length, as the inaccuracy

Physics Part IA
Cambridge University, 2012-2013
accumulates over the distance measured, rather than remaining constant for the
length of the ruler.

Physics Part IA
Cambridge University, 2012-2013

Tables and graphs:

Tables and graphs are drawn/plotted according to the following specific
conventions:
Tables:
Numbers in tables should be pure/dimensionless numbers
Column headings should therefore also represent pure numbers, hence
have the format measurement/unit
If scientific notation is used, it should be written in the column headings,
rather than beside the numbers themselves, such that the headings
become measurement/(

10

unit)

Values calculated in tables should be given to an accuracy of one more

significant figure than the actual measured values, to ensure that no
precision is lost. Obviously the final result would then be rounded to the
correct number of significant figures.

Graphs (computer-generated):
Theory should be formulated such that a straight line is expected, in which
the gradient and intercepts may provide quantitative information
The dependent variable is plotted on the y-axis, and the independent
variable on the x-axis
Origin needs only to be included if it is being tested by the theory
Data points should be roughly evenly-spaced along the x-axis in the final
graph. The values of the independent variable at which measurements are
taken must therefore be chosen appropriately so that the x-axis values for
the corresponding straight line are evenly-spaced. For example, the graph
of the relationship

y x 3 will be

independent variable values should be

values on the x-axis are

x 3 . Thus the

against
1
3

1
3

1
3

1
3

1 , 2 ,3 , 4 such that the

1,2,3,4

Graphs (hand-drawn):

Points should be plotted as

If the size of the cross does not represent the error bars, this needs to be
stated on the graph
When drawing the trendline, data should be evenly-distributed on either
side of the line, and the line should go through at least 2/3 of the error
bars
The error bars should be annotated as deviations due to uncertainties

rather than

or

Techniques of data analysis:

Certain techniques, such as linear regression analysis, are used to analyse
collected data. Other techniques, such as the method of exact fractions, are used
to collect data to a required level of accuracy.
Linear regression analysis:

Physics Part IA
Cambridge University, 2012-2013
Linear regression analyses can be conducted on a calculator or computer to fit a
trendline to a set of data, and hence determine its gradient and the associated
uncertainty in this gradient:
The computer generates a line

Ax + B , and then using an algorithm,

numerically minimises the value of:

n

( y i( A x i+ B ) )2
i=1

Where

yi

is the data point value for

x i and

A xi + B

is the value given by

the trendline. The deviations are squared such that negative deviations do not
negate positive deviations.

Physics Part IA
Cambridge University, 2012-2013
Method of exact fractions:
The method of exact fractions is an iterative method used to obtain a high level
of precision for measurements taken using relatively rudimentary apparatus,
such as a manually-operated stopwatch. This method removes the need to take
very large measurements, or measurements over very long time periods, in
order to reduce their fractional error. This greatly reduces the probability of
human error in taking these measurements, and also greatly improves efficiency.
Use of exact fractions to measure the period of a pendulum,

to an accuracy

of 1/2000:
The method of exact fractions is best demonstrated using an example. In this
example, the timing device used is a stopwatch with standard error

0.1 s.

Thus, if the method of exact fractions were not used, the time over which the
swings of the pendulum were counted would have to be at least:

t
1
0.1
1

t 200 s
t 2000
t 2000
Where personal timing error is not taken into consideration.
Applying the method of exact fractions:
1. Obtain a first estimate of the period:
Time a small number of swings several times and use these
measurements to find the initial estimate of the period
timing error

T 1 , and the

mean = / n . The timing error should be used to estimate

the true time over which the pendulum should swing in order to reduce its
fractional error to 1/2000, as opposed to an estimate like that above.
2. Use the first estimate
estimates of the period

T1

to obtain progressively more accurate

T2 , T3

etc.:

a. Time an unknown, integer number of swings for a progression of

increasing time intervals up to the time interval calculated in (1).
b. Divide the times (beginning with the shortest) by the first estimate
of the period to approximate the number of swings (rounding the
values obtained to the nearest integer).
c. If a given approximation is close to an integer, the accuracy is good
enough for the current estimate of the period to be used for the
next timing value. If the number has drifted from an integer value,
however, the value should be rounded to the nearest integer, then
the time interval from which it was calculated should be divided by
this integer value to obtain a new estimate for the period
d. The new estimate for the period should then be used to estimate
the number of swings for subsequent time periods, until accuracy
begins to fall again
e. Continue the process until an accurate integer number of swings
has been obtained for all the measured time intervals

Physics Part IA
Cambridge University, 2012-2013
3. Compare a series of short time intervals timed at the beginning of
the timing stage to an equivalent series of long time intervals
timed at the end:
a. Subtract the shorter time intervals from their equivalent longer time
intervals
b. Divide each by the last estimate of the period obtained
c. Round all calculated values to the nearest integer (they should all
be very near to the same integer value)
d. Divide each time period by this integer value to obtain a series of
estimates of the period
e. Average these estimates and calculate
final, accurate estimate of the period!

mean = / n . This is the