Vectorial Mechanics

1. Mathematical methods (2)

[email protected]

Dept. of Physics

v.2023-09-30

1 / 25
References

Xavier Jaén et al.

Fundamental Mechanics
UPC (2023)
Available online here, also in Spanish and Catalan
Chapter 1.5
Samuel J. Ling, Jeff Sanny and William Moebs
University Physics vol. 1
OpenStax, Rice University (2016)
Available online at https://openstax.org
Chapter 1.6

Note: Sources from the Wikipedia are abbreviated Wkpd.

2 / 25
Contents

1 Uncertainty in measurement

2 Linear regression

3 / 25
1. Uncertainty in measurement

1.1. Measurement and its uncertainty

Definition: Measurement
Measurement is the process by which we attach a number to a
physical property as the result of comparing with a standard
adopted as the unit.

All experimental measurements are affected by some variability

which can be related with the measured object, the instrument or
the observer.

As a consequence, the output of the measurement process is not

a number, but an interval.

4 / 25
1. Uncertainty in measurement

1.2. Accuracy, precision, error and uncertainty

It is generally assumed that a true value of the quantity being

measured, or measurand, exists. The true value is usually
unknown.

Accuracy is how close a measured value is to the true value. It is

a qualitative concept.

Precision is how well repeated independent measurements of the

same measurand agree among them. It is a qualitative concept.

5 / 25
 6 / 25

Source: Mr Evans’ Science Website at https://sites.google.com

1. Uncertainty in measurement
1. Uncertainty in measurement

Measurement error is the difference between a measured value

and the true value of the measurand. It is a quantitative concept.
It is usually unknown.

Measurement uncertainty is a quantitative estimate of a range of

values within which the true value of a measurand is likely to be.
It indicates the overall quality of the measurement.

Errors and uncertainties can be classified as random, or

statistical, and systematic.

7 / 25
1. Uncertainty in measurement
1.3. Reporting measurement results

A measurement process with repetition involves these steps:

• Take N readings xi , i = 1 . . . N .
• Discard clearly identifiable outliers, if any.
• Compute the average
1 X
x= xi
N i
• Calculate the associated random uncertainty u.

How to report the result of a measurement

The result of the measurement is expressed as

x±u

followed by the units. Example: 3.4 g ± 0.2 g or 3.4(2) g.

8 / 25
1. Uncertainty in measurement
The uncertainty u is seldom reported with more that one
significant figure. Example:
±0.2 is okay
±0.24 is okay but exceptional
±0.2432 is not correct.

Two figures are usual when the first digit is 1. Example:

±0.1 is okay
±0.14 is okay too.

The reported value of x must have a number of significant digits

such that the least significant of them is the last one affected by
±u. Example:
3.4 ± 0.1 is okay
3.41 ± 0.14 is okay too, but
3.41 ± 0.1 is incorrect.
9 / 25
1. Uncertainty in measurement

1.4. How is the random uncertainty calculated?

Compute the following quantities,

1 X 2
x2 = xi
N
i

N  2 
s2 = x − x2
N −1
r
s2
u = tN
N
where tN is from the following table,

10 / 25
1. Uncertainty in measurement

Student’s tN , 95% two-sided prob.

N t N t N t
2 12.71 14 2.160 26 2.060
3 4.303 15 2.145 27 2.056
4 3.182 16 2.131 28 2.052
5 2.776 17 2.120 29 2.048
6 2.571 18 2.110 30 2.045
7 2.447 19 2.101 40 2.023
8 2.365 20 2.093 50 2.010
9 2.306 21 2.086 60 2.001
10 2.262 22 2.080 80 1.990
11 2.228 23 2.074 100 1.984
12 2.201 24 2.069 120 1.980
13 2.179 25 2.064 ∞ 1.960

11 / 25
1. Uncertainty in measurement

Notice that for N larger than ∼10 the value tN ≈ 2.0 is an

acceptable approx for most practical purposes.

Under very general assumptions, it can be stated that the given

interval contains the true value of x with 95% probability.

1.5. Implicit uncertainty

If the uncertainty is not explicitly specified it can be assumed it is

implicit.

The least significant digit of a number is affected by ±1.

Example
2.137 ≡ 2.137 ± 0.001

12 / 25
1. Uncertainty in measurement
Numbers without decimal point can be assumed to be exact.
Example: 2π, 3x.

However, if the number has trailing zeroes (e.g., 12300) it must

be assumed that the zeroes are not significant.
Example: 12300 ≡ 1.23 × 104 .

Caution! Exceptions depending on the context

The rule about trailing zeroes does not apply in all contexts, e.g., in
accounting. Thus, if your bank account balance reads 12300 EUR,
that is exactly how much money you have.

Notice that a number expressed with trailing zeroes leads to

confusion. In such cases scientific notation is preferable.
Example: 1.230 × 104 instead of 12300.

13 / 25
1. Uncertainty in measurement
1.6. Propagation of uncertainties

For a product the result must contain the same number of

significant figures as the quantity with the least number of
significant figures.

Example: Rectangle
What is the area of a rectangle of dimensions 234.5 cm × 7.5 cm?
Answer: A = 234.5 × 7.5 = 1.8 × 103 cm2 .
(The product, 1758.75, given with only two significant digits.)

For a sum the result must contain the same number of decimal
places as the summand with least decimal places.
Example: 8034.323 + 5.21 = 8039.53
Example: 1300 − 243 = 1.1 × 103 , not 1057.
However, beware of the context, e.g., in accounting.
14 / 25
1. Uncertainty in measurement

These rules for sums and products are simplified recipes for
particular cases of a more general formula, as follows.

Propagation of uncertainties
Consider a variable y = f (x1 , x2 , . . . , xK ) which depends on the
uncorrelated variables xk . The uncertainty uy of y is related to the
individual uncertainties uk of each variable xk according to the
formula
X  ∂f 2
u2y = u2k
∂xk
k

If the xk ’s are correlated a more general formula involving the

correlation coefficients of each pair of variables must be
introduced. This formula is beyond the scope of this course.
15 / 25
Contents

1 Uncertainty in measurement

2 Linear regression

16 / 25
2. Linear regression

2.1. Least squares linear regression

What is the straight line that best fits a given set of {x, y} data?

Source: Wkpd by Sewaqu

17 / 25
2. Linear regression

Consider que quantities x and x2 defined before and, also,

1 X
y= yi
N
i

1 X 2
y2 = yi
N
i

1 X
xy = x i yi
N
i

18 / 25
2. Linear regression

Least squares linear regression

The straight line that best fits, in the least squares sense, a set of
{xi , yi } data is
y = mx + b
with
xy − x y
m= (slope)
x2 − x2
b = y − mx (intercept)

The correlation coefficient of the fit is defined as

xy − x y
r=q q r ∈ [−1, +1]
2
x −x 2 y2 − y2

19 / 25
2. Linear regression
The quantity r2 ∈ [0, 1] is called coefficient of determination.
The closer r2 is to unity, the better x’s and y’s are connected by
a linear relation.

The uncertaintites associated to slope and intercept are

s
m 1 − r2
um = t(N −1)
r N (N − 2)
p
ub = um x2

where t(N −1) is from the Student table presented before.

Sometimes it might be assumed that the dependence is a

proportionality, i.e. linear with null intercept,

y = mx
20 / 25
2. Linear regression

Least squares proportional regression

The straight line through the origin that best fits, in the least
squares sense, a set of {xi , yi } data is

y = mx

with
xy
m=
x2

The correlation coefficient of the fit is

xy
r=q q r ∈ [−1, +1]
x2 y 2

21 / 25
2. Linear regression

Note that these last formulas for m and r are identical to the
ones for the general linear fit replacing x and y by zero.

2.2. Examples

Linear regression finds application in a large variety of fields of

knowledge. Physics is one of them.

Depending on the nature of the problem being modeled, the

linear or the proportional regression formulas are preferable.

Three examples are presented below. One is suitable for linear

regression and one for proportional regression. In the last
example both methods are presented so that the degree of
accuracy and realism of each approach can be assessed.

22 / 25
2. Linear regression
Example: Real battery
An electrical battery can be modeled as a voltage source V with an
internal resistance r. If an external resistance R is connected to the
battery terminals the current I in the circuit is
1
V = (R + r) I ⇒ R=V −r
I
By varying R and measuring the corresponding I values a linear
regression
y = mx + b
can be made with the identifications

y=R x = 1/I m=V b = −r

That is, the slope yields V and the intercept is −r.

23 / 25
2. Linear regression

Example: Ideal gas

An ideal gas is characterized by the state eq.
1 1
P V = nRT or, equivalently, = V
P nRT
(R ≈ 8.31 J K−1 mol−1 ) By varying V keeping T constant and
measuring P a proportional regression

y = mx

can be made with the identifications

y = 1/P x=V m = (nRT )−1

Thus, the temperature can be inferred as T = (nRm)−1 .

24 / 25
2. Linear regression
Example: Massless spring vs. real spring
Consider a massless spring with a yet unknown constant k. Different
masses M are made to oscillate and the angular frequency ω is
measured. The relation
1 1
2
= M i.e. y = ω −2 x=M m = k −1
ω k
is amenable to proportional regression so that k can be obtained
from the slope m. But real springs have mass. The total oscillating
mass can be modeled as (M + M′ ) with M′ the unknown
contribution from the spring itself. Thus,

1 1 ′ 1 M′
= (M + M ) = M +
ω2 k k k
amenable to linear regression with slope 1/k and intercept M′ /k.

25 / 25