Introduction: How to Minimize Error Propagation in Experiments

INTRODUCTION: HOW TO MINIMIZE ERROR

PROPAGATION IN EXPERIMENTS

Students in laboratory periods in physics should apply basic techniques on how to

conduct an experiment, particularly on how to take data; how to minimize and estimate
experimental uncertainty; how to combine uncertainties; and, how to analyze data and draw
conclusions.

I.1. How to take data – Experimental data usually consists of recording one or more measured
quantities while some other controllable quantity is changed systematically. Data to be
gathered should be guided by one or more explicit questions the students are seeking to
answer in the experiment. These questions are clearly written down in the laboratory manuals.
In order to answer these questions the students must be thoroughly familiar with the
experimental topic, which may mean consulting a text other than the laboratory manual.
Observations and procedures in the experiment from which the students will draw some
conclusions should be kept together in a neat and concise format.

I.2. How to minimize and estimate experimental uncertainty – Data gathered in laboratory
measurements are never exact. The error or uncertainty in any measured quantity should be
minimized. Uncertainties or errors in measurement could be due to blunders, systematic errors,
random errors and limitations in the measuring instruments. Constantly checking and
correcting experiment procedures, instrument readings, data recordings and data analysis
could significantly minimize blunders or mistakes. If one measurement seems inconsistent with
the other measurements, a mistake should be suspected. Errors that are repeated throughout
an entire set of measurements are termed systematic errors that could be due to instrumental
defects, constant bias in making a measurement, and misunderstanding of the quantity to be
measured. These systematic errors could be minimized by checking/comparing an instrument
against a similar instrument known to be in good working order, sound preparation and
understanding of the quantity to be measured, and having a second person perform some of
the measurements and look for systematic differences. Uncontrollable fluctuations in the
behavior of the system being measured and the measuring instrument themselves and
limitations in the precision of the instruments used could bring about random errors. For
estimating instrumental uncertainties, least counts of all measuring instruments along with an
estimated instrumental uncertainty should be recorded. The best estimate of a quantity based
upon repeatedly measured data can be obtained from the average or mean value, X, of the
data.

Average or Mean is given as

− 1 N
X =  Xi (1)
N i =1

The symbol X is used to denote the quantity being measured, and the subscript i labels
particular measurement (e.g., X2 means the second measurement), N is the number of
measurements and the symbol  means “add up all the measurements.” The spread in the data
can be best estimated by the universe standard deviation,  (sigma), which is calculated as

v|Page
 Introduction: How to Minimize Error Propagation in Experiments

1/ 2
1 N − 2
 =    X i − X   (2)
 N i =1   

The average experimental value then can be written down accurately as X   .

In most cases, the average experimental value, X, is compared to the currently accepted
value or standard value of the same quantity. Critical comparison of these two values generally
means calculating the percentage difference between the standard value and the experimental
value.

| Standard value - Experimental value |

% difference = x 100 (3)
Standard value

I.3. Data Analysis and Conclusions - In most of the experiments, the students will try to
determine the relationship between two variables (e.g., x & y). For two sets of measured
quantities (e.g., x i & y i) expected to have linear relationship, determination of the slope of this
relationship will yield the value of a desired physical quantity. The Method of Least Squares is
an analytical method for obtaining the most reliable slope from a set of linearized experimental
data. According to this method, the most probable value of a quantity to be determined is
obtained from a set of measurements by choosing the value, which minimizes the sum of the
squares of the deviations of these measurements. If two variables x and y are known to be
related by a linear equation of the form y = mx + b where m is the slope of the line and b
is the y - intercept, then the most probable values of m and b can be obtained from

 N   N  N 
N   xi yi  −   xi   yi 
m =  i =1   i =1  i =1 
2
(4)
 N
2 
N

N   xi  −   xi  = 
 i =1   i 

 N  N 2   N  N 
  i   i    i i   xi 
y x − x y
b =  i =1  i =1   i =1 2  i =1  (5)
 N   N 
N   xi2  −   xi  = 
 i =1   i =1 

The standard deviations of the values of m and b are given by the square roots of the variances
(2), where :
N 2
m =
2
(6)


 n

 2   xi2 
 b2 =  i =1  (7)


2
 N   N 
 = N   xi2  −   xi  (8)
 i =1   i =1 

vi | P a g e
 Introduction: How to Minimize Error Propagation in Experiments

1  N 2
2 =  
N − 2  i =1
( mxi + b − yi ) 

(9)

Some of the experiments in the course will involve the application of the least squares
procedure.

Example:

A student is studying electrical current and potential differences. He connects cells of the
battery across the wire and measures the potential difference between the negative end and
various positions along the wire. Assume the uncertainty in each potential measurement to be
0.05 V. Apply the method of least squares to find the line that would describe the way in
which the voltage V varies as a function of position x along the wire.

Table 1 shows the gathered data of the potential difference (V) as a function of position along
a current-carrying wire.

Potential Fitted Pot.

Point Position Difference Difference
xi2 xiVi
Number xi (cm) Vi (V) a+ bx

1 10.0 0.37 100 3.70 0.33

2 20.0 0.58 400 11.60 0.60
3 30.0 0.83 900 24.90 0.86
4 40.0 1.15 1600 46.00 1.12
5 50.0 1.36 2500 68.00 1.38
6 60.0 1.62 3600 97.20 1.64
7 70.0 1.90 4900 133.00 1.91
8 80.0 2.18 6400 174.40 2.17
9 90.0 2.45 8100 220.50 2.43

xi = 450.0 Vi=12.44 xi2=28500 xiVi =779.30

 = Nxi − (xi ) = (9 x28,500) − (450) = 54,000

2 2 2

The y-intercept is

b=
(xi
2
Vi − xi xiVi
=
)
(28,500x12.44 − 450.0 x779.30) = 0.0714
 54,000

The slope of the line is

(N iVi − xi Vi ) (9 x779.30 − 450.0 x12.44)

m= = = 0.0262
 54,000

The equation of the regression line is y = 0.0714 + 0.0262 x

vii | P a g e