Error Analysis – PHY F110 - Physics

Laboratory
Dr. Kannan Ramaswamy
BITS Pilani Department of Physics
Hyderabad Campus
Error Analysis-Topics

• Why are errors important?

• Reporting Experimental Data
• Types of Errors
• Error propagation
• Linear least square fit
• Model question for the Comprehensive lab test
• Good practices

BITS Pilani, Hyderabad Campus

 Why are errors important?
Two measurements of body temperature before and after a
drug is administered are 38.2 0C and 38.4 0C
Is the temperature rise significant?
It depends on the associated errors!

(38.20.01)C and (38.4 0.01)C Significant

(38.20.5)C and (38.4 0.5)C Not Significant

BITS Pilani, Hyderabad Campus

Reporting Experimental Data
Number of digits in the Examples of reporting
experimental result
1.65±0.05

Significant Figures 2.500±0.003

(6.02214179±0.000 000 30)*1023 mol-1

Accuracy of the result

BITS Pilani, Hyderabad Campus

Significant figures
Don’t over quote results to a level inconsistent with
the error!

Examples of incorrect reporting

9.82 ± 0.02385
10.0 ± 2
Examples of correct reporting
9.82 ± 0.02
10.0 ± 2.0

BITS Pilani, Hyderabad Campus

 Quiz on significant digits

Number of significant digits in: Answer

1.65 3
2.500 4
6.02214179
9
5.00 3
1
0.0005
0.500 3
500 I do not know!!

BITS Pilani, Hyderabad Campus

 Types of errors

Accidental or intended errors

Can be avoided by being careful

Systematic errors

Erroneous calibration Bias of the experimenter

Incorrect measurement
technique
• Systematic errors do not average to zero
• Estimation of systematic errors require patience
BITS Pilani, Hyderabad Campus
 Type of errors

Random errors

Fluctuations in the initial Limited precision of the

conditions measuring apparatus

Random fluctuations
due to thermal motions

BITS Pilani, Hyderabad Campus

 Random vs systematic errors

Random errors only

True value

Random + systematic

•A result is said to be accurate if it is relatively free

from systematic error
•A result is said to be precise if the random error is
small
Precision and Accuracy
Read-off error Analog Instruments

2±1A
Read-off error Digital Instruments

12.88 ± 0.005 V

Error is half of last precision digit

 Estimating reading errors

16 17

•Linear scale (e.g. a ruler)

•Need to estimate precision with which measurement
can be made
•May be a subjective choice
16.70.1
Handling random errors
Statistical Analysis
What do we do to get the measured value when there is
random error in the experiment?

Make large number of measurements

of the experimental parameter, Q
N
1
Q
N
 Q , i  1,2...N
1
i

 Q  Q  , i  1,2...N
N
1 2
d i
N 1

Q Qd Result of the

measurement
BITS Pilani, Hyderabad Campus
Standard Deviation Time-period

Trial, i 1 2 3 4 5
ti (s) 3.9 3.5 3.7 3.4 3.5

1
Best estimate of T: t 
N
t
i
i  3.6s

 t  t   0.2s
1
Best estimate of error in T: t 
2
i
N i

T = 3.6 ± 0.2 s
Standard Deviation Normal Distribution

Frequency

ti

Six sigma principle – Motorola 1986

Handling random errors

If you cannot do several experiments, what will you do?

Estimate the error using the least count of

the measuring device

Q  least count Result of the

measurement

BITS Pilani, Hyderabad Campus

Error propagation
Typically, we need to use a formula to get the final result
How to combine errors from
various measurements?
Suppose, z = x + y, x and y
being the experimentally measured
quantities
 z z 2
Error in z is z   ( x)  ( y ) 
2

 x y 
z  x  y
2 2

Δx and Δy are the absolute random errors

in x and y
BITS Pilani, Hyderabad Campus
Error propagation

Typically, we need to use a formula to get the final result

How to combine errors from
various measurements?
Function Pr opagate Error
z  ax  b z  ax
1

z  x   y  
2 2 2
x
z  cxy; c       
y z  x   y  

z x
z  cx a a
z x
z
z  sin x  x cot x
z

BITS Pilani, Hyderabad Campus

Treatment of errors in formulae

Z  A B C  Z    A    B    C 
2 2 2 2

 Z   A   B   C 
2 2 2 2

Z  ABC or AB / C etc        
 Z   A   B   C 
 Z   A   B   C 
2 2 2 2

ZAB Cn m l
   n  m   l 
 Z   A   B   C 
A
Z  ln A Z 
A
Z
Z  exp A  A
Z
Example of error calculation

A r 2
Where r=(50.5)m
A=78.5398m2
r 0.5
  0.1
r 5
 A   r 
2 2

 
  2   (2  0.1) 2
 0.04
 A   r 
A
  0.2 hence A  0.2  A A  0.2  (78.5398m 2 )  16m 2
A

Hence final result is A=(7916)m2

Example of error calculation
l
  2 l=(2.50.1)m, g=(9.80.2)ms-2
g
2
    1 l   1 g 
2 2

=3.1735s       
   2 l   2 g 
2 2
 1 0.1   1 0.2  4
  
      5.04  10
 2 2.5   2 9.8 
/=0.022 hence =0.022x3.1735=0.070
=(3.170.07)s
 Least square fit method

Consider a straight line equation with non-zero intercept

y  mx  c

S    yi  mxi  c 
2

This will be a minimum for best fit

Because
S is the square of the error (difference), between the expected value of
y for a particular value of x, and the actual value.

BITS Pilani, Hyderabad Campus

Least square fit method
Equation with non-zero intercept
y  mx  c
Minimizing S    yi  mxi  c 
2

S S
 0 and 0
m c

m
  
xi  x yi
c  y  mx

 xi  x
2


You can use the above eqns. in the coupled pendu-

lum experiment
BITS Pilani, Hyderabad Campus
Least square fit method
Equation with zero intercept

y  mx
Minimizing S    yi  mxi 
2

S
0
m

m
 xy i i

x
2
i

You can use the above eqns. in Newton’s

Rings and diffraction grating experiments
BITS Pilani, Hyderabad Campus
 Least square fit method: An
example Spring experiment
Xi(gm) Yi (cm) Xi-Xav (cm) (Xi-Xav)Yi (cm2) (Xi-Xav)2
52 1.5 -100 -150 10000
102 6.4 -50 -320 2500
152 12.4 0 0 0
202 18.4 50 920 2500
252 24.5 100 2450 10000

m = Σ (Xi-Xav)Yi/Σ(Xi-Xav)2 c = yav-mxav= 12.6 – 0.116*152

= -5
m = 2900/(25000) = 0.116
Now generate the fit line using y = mx + c
BITS Pilani, Hyderabad Campus
Least square fit method: An
example
250

200

150

100
Y

50

0
-50 0 50 100 150 200

-50
X
BITS Pilani, Hyderabad Campus
Least square fit: Non-linear
functions
How will you do least square fit for non-linear functions like

y  ax ? b

Convert them into linear form first

log y  log a  b log x

Then proceed as before

If you plot this on a log-log graph you will get a

straight line!

BITS Pilani, Hyderabad Campus

Log Log Graph

(300, 20000)

BITS Pilani, Hyderabad Campus

Semi-log graphs & log graphs.
Semi-log graph (shown below) is useful if we want to fit
the data to a relation of the form y = log(x)
0.97

100

0.90

60
0.80

0.70

0.60

0.50

0 0.40

1
0.30
2 4 6 10 20 40 100 200 1000
0.20

Point shown here is x = 3, y = 40

0.10

0.07
1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1 2 3 4 5 6 7 8 9 1
Semi-log is useful for relations of the type

C  C o exp(  E / kT )
Taking log on both sides of this equation and
changing the base to 10

E
log C  log C o 
2.303kT
 Vernier scale:

0 5 10

1 5 10 15 20

For most Vernier scales, n divisions on the vernier

scale coincide with n-1 divisions on main scale
For most Vernier scales, n divisions on the vernier
scale coincide with n-1 divisions on main scale
Model Questions

Obtain Newton’s rings using the given apparatus and

measure the diameter of the 4th , 8th and 10th dark rings.
Use this data of the diameter, to determine the average
radius of curvature of the given lens (Use Least Square Fit).
[Wavelength of the light = 5893 Å]
Marks : Setting - up : 20 marks, Readings : 20 marks,
Graph 10 marks, Calculation : 10 marks

BITS Pilani, Hyderabad Campus

Good Practices

• Come well prepared

• Do not panic
• Remember all the relevant equations
• Do the experiments honestly

All the best

BITS Pilani, Hyderabad Campus