1.

0 THEORY OF ERRORS WITH APPLICATIONS SURVEYING

MEASUREMENTS AND COMPUTATIONS
Introduction
The basic task in surveying is the establishment of three-dimensional control. And this is
usually achieved by linear and angular measurements. Such measurement must inevitably
contain errors. Thus statistical techniques must be employed not only to distribute these
errors but also to assess the accuracy and confidence limits of a final accepted value.

1.1 Definition of Terms

1. True Value (T) – This is the true value (t) of a measurement; but it can never be
found or if it is found, we would never know that we have found it.
2. True Error (𝜀𝑥 )- This is defined to be the difference between the true value T and
the observed value 𝑥 , that is,
𝑇 − 𝑥 = 𝜀𝑥
It can never be found from a single observed value of a quantity.
̅̅̅. This is the value which is most likely than any
3. Most Probable Value - MPV (𝒙)
other to be the true value. It is often taken to be the true value. It is often taken to be
the arithmetic mean of the series of observations of a quantity provided that the
observations are independent of each other and have equal reliability.
4. Residual (V). It is defined to be difference between the MPV (𝑥̅ ) and the observed
value (x).
𝑥̅ − 𝑥 = 𝑉
5. Weight (P). The weight P of an observation is a measure of trustworthiness (or
reliability) with respect to other observations. It is usually expressed a number. If an
observation is more reliable than the other, we assign more weight to it. Ideally the
weight P of an observation is the reciprocal of the square of the standard error, that
is:
1
𝑃=
𝜎2
6. Accuracy and Precision. If a quantity is measured several times, the degree of
agreement within the measurements is the precision of the set. Thus the residual are
small, the observations are precise and vice-versa.

1
 The accuracy of the set is the difference between the most probable value and the true
value. Hence in most cases, the accuracy can never be found.

Fig. 1.1: Accuracy and precision of measurements.

1st Measurement- accurate and precise; 2nd Measurement- not accurate but precise
3rd Measurement- not accurate, not precise; 4th measurement- accurate but not
precise.
7. Relative Error (Rx). It is a measure of the error relative to the size of the
measurement. For instance, a distance of 10m may be measured with an error of
±1mm, whilst a distance of 100m may also be measured with an error of ±1mm.
Although the error is the same in both cases. The second error is believed to be more
accurate than the first one. To allow for this, the term relative error may be used
where
𝜀𝑥
𝑅𝑥 =
𝑥
Thus in the first case, x=10m, 𝜀𝑥 = ±1𝑚𝑚
1
∴ 𝑅𝑥 =
10,000

While in the second case, x=100m, 𝜀𝑥 = ±1𝑚𝑚

1
𝑅𝑥 =
100,000
Clearly illustrating the distinction; multiplying the relative error by 100 gives the
percentage error, that is, 𝑅𝑥 x 100.
Relative error is extremely defined and is commonly used in expressing the accuracy of
linear errors, e.g. the relative closing error of a traverse is usually expressed this way. The

2
definition is clearly not applicable to expressing the accuracy to which an angle is
measured.

1.2 Types of Error

There are three major types of errors
a) Mistakes or blunders (Gross errors)
b) Systematic error
c) Random or accidental error

(A) Mistakes or Blunders

These often result from carelessness, tiredness, fatigue or lack of experience of
a surveyor. Typical examples are:
i. Reading six (6) on a levelling staff as nine (9);
ii. Sighting a wrong target in a round of angles
iii. Transposition of numbers, e.g. writing 3457 instead of 3475. Mistakes are the
largest of the errors likely to occur therefore great cares must be taken to avoid
them.

(B) Systematic Errors

These arise from some physical phenomenon such as:
1. Refraction of light;
2. Variation in the speed of electromagnetic waves through the
atmosphere;
3. Expansion or contraction of steel tapes due to temperature variations
4. Systematic errors can also arise from psychological tendency on the part
of the observer. This is known as “Personal Equation” in Astronomy.
Systematic errors can also arise from instrumental adjustment.
The magnitude of systematic error is often difficult to assess and equally troublesome
to remove. However, very often the values of these errors are estimated and applied as
correction to the measured quantity.

(C) Random or Accidental Errors

3
These are the smallest errors. They have discrepancies which remain when all other error
sources have been eliminated as far as possible. These errors follow the Law of Probability
and as such can be analysed and adjusted using statistical techniques. It must be
emphasized that these errors alone are adjustable by statistical method.

2.0 PROPAGATION OF SYSTEMATIC AND RANDOM ERRORS

2.1 Propagation of Systematic Errors

Consider a base line of n-bays each with a systematic error e1, e2, e3, … en. The
resultant systematic error e of the whole base will be given by

𝐸 = 𝑒1 + 𝑒2 + 𝑒3 + ⋯ + 𝑒𝑛 = ∑ 𝑒

If these errors are assumed to be equal and of the same signs, then
𝑒1 = 𝑒2 = 𝑒3 … 𝑒𝑛 = 𝑒
∴ 𝐸 = +𝑛𝑒 𝑜𝑟 − 𝑛𝑒
For convenience, the magnitude of the probable systematic error is considered in
terms of its square. i.e.
𝐸 2 = 𝑛2 𝑒 2
Thus eliminating the +ve and –ve signs.
If the lengths of the bays are 𝑙1 , 𝑙2 , 𝑙3 … 𝑙𝑛 respectively, the relative systematic
𝑒1 𝑒2 𝑒3 𝑒𝑛
errors (or proportional systematic errors) are , 𝑙 ,𝑙 …
𝑙1 2 3 𝑙𝑛

Let the relative errors equals

1 1 1 1
, , 𝑘 … 𝑘 respectively
𝑘1 𝑘2 3 𝑛

Then
𝑙1 𝑙2 𝑙3 𝑙𝑛
𝑒1 = , 𝑒2 = , 𝑒3 = … 𝑒𝑛 =
𝑘1 𝑘2 𝑘3 𝑘𝑛
Then the error of the whole base is
𝑙1 𝑙2 𝑙3 𝑙𝑛
𝐸 = ∑𝑒 = + + +⋯+
𝑘1 𝑘2 𝑘3 𝑘𝑛

4
2.2 Propagation of Random Errors
In theory, only random errors are left after mistakes and systematic errors have been
removed. In practice, however, some systematic errors would remain.
In the study of random errors, we are concerned with probability. Hence, propagation of
random errors is studied through statistical analysis of results to find the probability of the
occurrence of the errors of various magnitudes. For such an analysis to be valid, a very
large number of errors is usually sampled.
Normal Distribution Curve or Gaussian Curve (or simply Normal Curve) is used in this
analysis. From this curve, the magnitude of the mean square error (or standard deviation),
the probable error and other indices of precision can be determined.

5
3.0 STATISTICAL ANALYSIS OF OBSERVATIONS AND DERIVED
PARAMETERS.

3.1 Normal Curve

Consider a length of 29.42m measured with a tape and correct to ±0.05m.
The range of these measurements to be from 29.37m to 29.47m, giving 11
possibilities to 0.01m interval for the answer. If the next bay was measured in
the same way, there would again be eleven possibilities. Thus, the correct value
for the sum of the two bays would lie between 11 x 11 =121 possibilities, and
the range of the sum would be 2 x ±0.05m i.e. between -0.10m and +0.10m.
Now the error of -0.10m can only occur once when both bays have an error
of -0.05m; similarly, +0.10m will occur once. Consider an error of -0.08, this
can occur in three ways (-0.05 and -0.03; -0.04 and -0.04;-0.03 and -0.05). By
applying this procedure through the whole range, Table 3.1 can be produced.
The lower half of this table is simply a repetition of the upper half. If the
decimal probabilities are added together, they are equal to 1.000. If the above
results are plotted as error against probability, the histogram in Fig. 2 is
obtained, the errors being represented by rectangles. Then in the limit, the
histogram is simply the superimposed curve.

6
Table 3.1: Table Showing Probability Distribution of Occurence
Error Occurrence Probability
-0.10 1⁄
1 121 = 0.0083
-0.09 2⁄
2 121 =0.0165
-0.08 3⁄
3 121 =0.0248
-0.07 4⁄
4 121 =0.0331
-0.06 5⁄
5 121 =0.0413
-0.05 6⁄
6 121 =0.0496
-0.04 7⁄
7 121 =0.0579
-0.03 8⁄
8 121 =0.0661
-0.02 9⁄
9 121 =0.0744
-0.01 10⁄
10 121 = 0.0826
0.00 11⁄
11 121 =0.0909
0.01 10⁄
10 121 = 0.0826
0.02 9⁄
9 121 = 0.0744
0.03 8⁄
8 121 =0.0661
0.04 7⁄
7 121 =0.0579
0.05 6⁄
6 121 =0.0496
0.06 5⁄
5 121 =0.0413
0.07 4⁄
4 121 =0.0331
0.08 3⁄
3 121 =0.0248
0.09 2⁄
2 121 =0.0165
0.10 1⁄
1 121 =0.0083

7
The superimposed curve is called the Normal Probability Curve or simply the
Normal Curve. The area under it represents the probability that the error must
lie between ±0.01m and is thus equal 1.000 (certainty) as shown in Table 3.1.
More typical bell-shaped probability curve are shown Fig 3.2

The tall-thin curve indicates small scatter and thus high precision, while the
flatten curve indicates large scatter and thus low precision.
Dispersion of the normal curve reveals the law of random error as follows:
i. Small errors are more frequent than large errors
ii. Positive and negative errors are equally likely to occur
iii. Very large errors do not occur.

8
3.2 Properties of the Normal Curve

−𝟏⁄ 𝟐 𝟐
𝒚=𝝅 𝟐 𝒉𝒆−𝒉 𝒙

1. Height, h of the normal curve indicates the precision of the set of

observations. The more precise an observation is, the higher will be the
curve.
ℎ
Height of the curve = 1⁄
𝜋 2

2. If the tape used in the experiment of section 2.0 was faulty due to poor
calibration, it may produce a set of results with the same symmetrical
distribution as in the experiment of 2.0. But the whole set would spread
about a position that was incorrect (see Fig 3.3).
The amount by which the axis of the most probable value fails to
coincide with x=0 is the systematic error of the set of observations (a).
In theory, it is often assumed that a=0 by the use of such phrases as
“assuming that there is no systematic errors” and thus the graph of the
distribution of residual is taken to be coincidence with the error curve.
3. Indices of precision. It is important to be able to assess the precision of
a set of observations. Some of the standard used for doing this are
i. mean square error or standard error 𝜎
ii. Probable error 𝜖
iii. Average error 𝜂
On the normal curve these indices are shown as below

9
 The area under the normal curve bounded by the point of inflexion
represents the standard error or the mean square error. It establishes the limit
between which 68.3% of the value of the set should fall.
1
𝜎 = ± ℎ√𝜋 (from the curve)

The probable error 𝜖 reps 50% of the area under the curve. It is
2
approximately 3 𝜎

The average error 𝜂 is defined to be the mean of all the error taken without
regard to the signs. i.e.
2
𝜖 = 0.6745𝜎 ≈ 𝜎 ………………………….(3.1)
3
∑|𝑥|
𝜂= ……………………………………(3.2)
𝑛
1
𝜂= 1 (on a normal curve)
ℎ𝜋 2

The relationships between average error 𝜂 and the other quantities are:
𝜂 = 0.7979 𝜎 = 1.183 𝜖…………………………(3.3)
Hence, a statistical analysis of observations, the three statistical quantities used are
the standard error 𝜎, average error 𝜂 and the probable error 𝜖. The probable error was at
one time frequently used in surveying but not so now. Although, average error is simpler
to apply, 𝜎 is more commonly used since it gives a better estimate of precision for a
relatively small number of observations.

3.3 Standard Error or Mean Square Error

The standard error of a single observation is defined as
1
𝜀 2 1 + 𝜀 2 2 +𝜀 2 2 +⋯𝜀 2 𝑛 2
𝜎𝑠 = ( ) ……………………(3.4)
𝑛

10
 1⁄
∑ 𝜀2 2
𝜎𝑠 = ( ) …………………………(3.5)
𝑛

Where 𝜀 = true error

However, when the standard error is computed from the residuals V, the Bessel’s correction
is applied by dividing equations 3.5 by n-1 instead of n. Thus, standard error of a single
observation is
1⁄
∑ 𝑣2 2
𝜎𝑠 = ± ( 𝑛−1 ) …………………………(3.6)

In practice, 𝜎𝑠 means that 68.3% of the values of a set of observations should fall within
the limit set by the value 𝜎𝑠 on a normal curve.
2𝜎𝑠 represents 95% confidence limit
3𝜎𝑠 represents 99.7% confidence limit
3.29𝜎𝑠 represents 99.9% confidence limit
The 3.29𝜎𝑠 standard is often for the rejection of doubtful observations. As the probability
of making an error is greater than 3.29𝜎𝑠 is 1:1000, then any observation whose residual
is greater than 3.29𝜎𝑠 should be rejected and the whole set recomputed without it. If 𝜎𝑠 =
±1.2 , then any observation which differs from most probable value (mpv) by more than
3.95 should be rejected. However, this rejection criterion is only applicable to a large
sample. For smaller samples, the following factors should be used.
No of measurements Factor
4 1.5𝜎𝑠
6 2.0
8 2.3
10 2.5
20 3.0
50 3.5

Other Useful Standards

(a) Standard error of the arithmetic mean
1⁄
∑ 𝑣2 2 𝜎𝑠
𝜎𝑚 = ± (𝑛(𝑛−1)) =±
√𝑛

(b) Standard error of a single weighted observation

1⁄
∑ 𝑤𝑣2 2
𝜎𝑠𝑤 = ±( )
𝑛−1

11
 (c) Standard error of the weighted arithmetic mean
1⁄
∑ 𝑚𝑣2 2 𝜎𝑠𝑤
𝜎𝑚𝑤 = ± ((𝑛−1) ∑ 𝑤) =± 1
(∑ 𝑤) ⁄2

(d) Standard error of the difference between means

1⁄
𝜎1−2 = (𝜎1 2 + 𝜎2 2 ) 2

Equation (d) is useful for comparing different means to see if they are representatives of a
common value. If the difference between two means is greater than 2𝜎1−2 , it may be
regarded as significantly different.

3.4 AVERAGE ERROR (𝜂)

∑|𝑥| ∑|𝜀|
Average Error 𝜂 = or 𝜂 =
𝑛 𝑛

Where n is the number of observations. For instance, if the errors are +4, -4, +2, -
12
2, the average error, 𝜂 = =3
4

3.5 PROBABLE ERROR (𝜖)

𝜖 = 0.6745𝜎 = 0.8453𝜂

12
 Example
The triangular closures listed below were produced by two observers A and B
using similar equipment and methods. By an analysis of average error, standard
error and probable error of each set of results, assess the relative quality of the
observations produced by each surveyor.
𝜀A 𝜀B
+10 +5
-1 -3
+2 +4
-9 -4
+1 -6
-1 -2
+6 +1
-1 +7
+2 -3
-7 +5
Solution
∑ 𝜀A𝟐 = 278

∑ 𝜀B𝟐 = 190

∑|𝜀| 40
𝜂𝐴 = =10 = 4
𝑛

∑|𝜀| 40
𝜂𝐵 = =10 = 4
𝑛

Hence,
𝜂
𝜎𝐴 = = 5
0.7979
𝜎𝐴 = 𝜎𝐵
𝜖 = 0.6745𝜎 = 3.4

Since we have known the value of the true error,

1⁄
1⁄ 2
∑ 𝜀2 2 ∑ 𝑣2
𝜎𝑠 = ( ) = ± ( 𝑛−1 )
𝑛

13
But computing 𝜎𝐴 𝑎𝑛𝑑 𝜎𝐵 from the formula,
1⁄
∑ 𝜀2 2
𝜎𝑠 = ( )
𝑛
For A,
1⁄
278 2
𝜎𝐴 = ± ( 10 ) = ±5.3

Probable error𝜖𝐴 = 0.6745 𝜎 = ±3.6

For B,
1⁄
190 2
𝜎𝐵 = ± ( 10 ) = ±4.4 (more precise)

𝜖𝐵 = 0.6745 𝜎𝐵 = = ±3.0

Since 𝜎𝐵 is smaller than 𝜎𝐴 , it can be said that these observations are slightly
better than A’s, contrary to the impressions created by Average Error 𝜂, thus 𝜎
gives better estimate of accuracy than 𝜂.

Assignment
A certain angle was measured 10 times by observers A and B with the following
results, all measurements being equally reliable ( i.e. of the same unit weight).
The results obtained are as follows:
A B
74 36 18 74 36 10
74 36 20 74 36 21
74 36 15 74 36 25
74 36 21 74 36 08
74 36 24 74 36 15
74 36 16 74 36 20
74 36 22 74 36 28
74 36 17 74 36 11
74 36 19 74 36 18
74 36 13 74 36 24

14
Determine:
a. The standard error and the probable error for a single observation for each
observer
b. Standard error of the mean for each observer.
Note that the square of the standard deviation 𝜎 2 is called the variance and is
accepted as the fundamental measure of the precision as it is always positive.

15
4.0 REVIEW OF MATRIX ALGEBRA AND MATRICES
Matrix Notations
The extensive numerical computations which are necessary for adjustments
by the methods of least squares are usually carried out by matrix algebra.
4.1 Matrix Notations
Matrices are often denoted by bold-faced A. when it is necessary to indicate
how many rows and columns a matrix contains. These dimensions of a
matrix are written below the main symbol A. For example, a matrix A has m-
rows and n-columns.

𝑎11 𝑎12 … 𝑎1𝑛

𝑎21 𝑎22 … 𝑎2𝑛
𝐴=[ ⋮ ⋮ ⋮ ⋮ ]
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑛

𝑎11 , 𝑎12 , …𝑎𝑚𝑛 are called elements of the matrix.

The transpose of matrix A is written as 𝐴𝑇 . The rows and columns of A have
been interchanged. For example,
1 3 5
𝐴=[ ]
6 8 10

1 6
𝐴𝑇 = [3 8 ]
5 10
Inverse matrix of A is denoted by 𝐴−1

4.2 Special Matrices

Some of the special matrices include
1. Vector
2. Triangular
3. Diagonal
4. Identity matrix
5. Orthogonal
6. Determinant
7. Square
8. Adjugate (Adjoint)

16
 9. Partitioned
10. Singular matrix
1. Vector- This is a matrix which has only one column (column vector) or one
row (row vector). The elements of a vector are called components. We shall
denote vectors by small Roman Letters, l.
𝑙1
𝑙2
𝒍𝒎,𝟏 = 𝑙3 or 𝒍𝑻 = [𝑙1 𝑙2 𝑙3 … 𝑙𝑚 ]
⋮
[𝑙 𝑚 ]

2. Square Matrix- In a square matrix, the number of columns equals number of

rows (i.e. m = n).
𝑎11 𝑎12 … 𝑎1𝑚
𝑎21 𝑎22 … 𝑎2𝑚
𝐴𝑚𝑥𝑛 =[ ⋮ ⋮ ⋮ ⋮ ]
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑚

2 4 3
𝑨 = [3 4 4]
1 3 1
A square matrix is symmetrical if the elements which are situated on either side
of the main diagonal (from upper left hand corner to lower right hand corner) are
equal.
1 4 2
𝑨 = [4 2 6]
2 6 3

A symmetric matrix is identical with its own transpose 𝑨𝑻 = 𝑨 if 𝑎𝑖𝑗 = 𝑎𝑗𝑖

For instance, Normal Equations are usually squared matrices.

3. Triangular- An upper triangular matrix, U is a square matrix all of whose

elements below the principal diagonal (main) are zero.
3 1 2
𝑼𝟑𝒙𝟑 = [0 4 3]
0 0 1

17
 3 0 0
𝑳𝟑𝒙𝟑 = [1 4 0]
2 3 1
Similarly, the lower triangular matrix, L, all the elements above the main
diagonal are zero. Lower and upper triangular matrices are very useful in forward
substitution and backward substitution respectively. For example a set of
equation which can be written as 𝐿𝑥 = 𝐾, is very easily solved for the first line
𝐿11 𝑥1 = 𝐾1

𝐾1
𝑥1 =
𝐿11
Substituting these in the next line,
𝐿21 𝑥1 + 𝐿22 𝑥2 = 𝐾2 , gives 𝑥2 and so on.
𝐿11 0 0 𝑥1 𝑘1
[𝐿21 𝐿22 𝑥
0 ] . [ 2 ] = [𝑘2 ]
𝐿31 𝐿32 𝐿33 𝑥3 𝑘3

2 0 𝑥1 4
[ ] . [𝑥 ] = [ ]
3 1 2 0

2𝑥1 = 4
𝑥1 = 2
3𝑥1 + 𝑥2 = 0
6 + 𝑥2 = 0
∴ 𝑥2 = −6
The process is known as forward substitution. Equations of the form
𝑈𝑥 = 𝑘 are similarly solved. These processes are useful in the solution of Normal
Equations (e. g. Cholesky decomposition).

4. Diagonal Matrix is square matrix in which all the elements outside the main
diagonal are zero.
3 0 0
𝑫 = [0 2 0]
0 0 4

5. Identity Matrix (I) is a diagonal matrix, all elements of which are 1.

18
 𝟏 𝟎
𝑰=[ ]
𝟎 𝟏

6. Orthogonal Matrix- a square matrix is defined to be orthogonal if both,

i. the sum of the squares of the elements of every row and of every column
equals 1
ii. the sum of the product of the corresponding elements of every pair of
rows or columns is zero.
It follows that if A is orthogonal, then 𝐴. 𝐴𝑇 = 𝐴𝑇 . 𝐴 = 𝐼
1 0 0 sin 𝜃 − cos 𝜃 0
[0 1 0] . [cos 𝜃 sin 𝜃 0] = 𝑅3
0 0 1 0 0 1
7. Determinant- to every square matrix, their corresponding scalar is known as its
determinant, written as |𝑨| 𝑜𝑟 det 𝑨
𝑎11 𝑎12 … 𝑎1𝑚
𝑎 𝑎22 … 𝑎2𝑚
|𝑨|𝑚𝑥𝑚 = [ ⋮21 ⋮ ⋮ ⋮ ]
𝑎𝑚1 𝑎𝑚2 … 𝑎𝑚𝑛
1st order det. A= 𝑎11 , the only element
2nd order det|𝑨| is
𝑎11 𝑎12
|𝑎 𝑎22 | = 𝑎11 𝑎22 − 𝑎12 𝑎21
21

3rd order det|𝑨| is

3 1 3
|1 2 6|
4 8 1
= 3(−46) − 1(−23) + 3(0)
= −115

8. Adjugate Matrix: the adjugate matrix of a square matrix written as Adj. A are
𝑇
sometimes called adjoint, is defined by 𝐴𝑑𝑗 𝐴 = [𝐴𝑖𝑗 ] = [𝐴𝑗𝑖 ] where 𝐴𝑖𝑗 is a
co-factor of matrix.
𝐴𝑖𝑗 = (−1)𝑖+𝑗 Minor
In other words, any adjoint A, is the transpose of the matrix whose elements are
the co-factors of the corresponding elements of A.

19
 3 1 2
𝑁 = [1 3 2]
2 2 4

8 𝑇
0 −4
𝐴𝑑𝑗 𝑁 = [𝑁𝑖𝑗 ] = [ 0 8 −4]
−4 −4 8
|𝐴| 0 … 0
0 |𝐴| … 0
𝐴 × 𝐴𝑑𝑗 𝐴 = [ ] = |𝐴| × 𝐼
⋮ ⋮ ⋱ 0
0 0 … |𝐴|
One would get determinant at the diagonal whence
𝐴𝑑𝑗 𝐴
|𝐴|
= 𝐴−1 …………….(4.1)

It can be argued then that

1
𝑥 = 𝐴−1 𝑘 = |𝐴| (𝐴𝑑𝑗 𝐴)𝑘 ………………………..(4.2)

9. Partitioned Matrix
An extensive matrix with many rows and columns can be divided by horizontal
and vertical lines into partitioned matrix. Inversely, smaller matrices can be
united into a composed matrix. In matrix operation, the partitioned matrix can
be dealt with as elements of composed matrix.
A matrix A whose elements are 𝑎11 to 𝑎𝑚𝑛 may, for example, be partitioned as
below:
𝐴 𝐴12 𝐴13 𝐴14
[ 11 ]
𝐴21 𝐴22 𝐴23 𝐴24
𝐴11, 𝐴12 , … 𝐴24 are sub matrices. Partitioned matrices may be added and
multiplied by the rules of matrix algebra, using some matrices as if they were
single elements A, B, C and D.
1 5 9 13 17 21
[2 6 10 14 18 22]
3 7 11 15 19 23
4 8 12 16 20 24
10. Singular Matrix
If det A = 0, equation 4.2 shows that x is infinite or indeterminate
1
𝑥 = 𝐴−1 𝑘 = (𝐴𝑑𝑗 𝐴)𝐾
|𝐴|

20
 A matrix whose determinant is 0 has no inverse by definition and is said to be
singular.

4.3 Matrix Operations

4.3.1 Addition and Subtraction of Matrices
Addition and subtraction of matrix are possible when all matrices
have the same dimension number (i.e. of the same order)
𝑨𝒎𝒙𝒏 + 𝑩𝒎𝒙𝒏 = 𝑪𝒎𝒙𝒏 when 𝑎𝑖𝑗 + 𝑏𝑖𝑗 = 𝑐𝑖𝑗
1 2 1 2 2 4
[ ]+[ ]=[ ]
3 4 3 1 6 5

4.3.2 Product of Two Matrices

A matrix B of order 𝑚2 × 𝑛2 can be multiplied by another matrix A of order
𝑚1 × 𝑛1 if 𝑛1 = 𝑚2 = (𝑠𝑎𝑦 𝑙). They are then said to be conformable in the order
A x B. Then their product A x B or AB say C. is defined to be a matrix in as
many rows as A and has many columns as B and each element of C is defined as
𝐶𝑖𝑗 = 𝑎𝑖1 𝑏1𝑗 + 𝑎𝑖2 𝑏2𝑗 + ⋯ + 𝑎𝑖𝑙 𝑏𝑙𝑗

So every element of 𝐶𝑖𝑗 of a matrix C is the sum of l product, formed by

multiplying successive element of row i of A by the corresponding element of
column j of B
1 2 (2 × 1) + (1 × 2) (3 × 1) + (2 × 2) 4 7
2 3
𝐴𝐵 = [4 1] . [ ] = [(2 × 4) + (1 × 1) (3 × 4) + (2 × 1)] = [9 14]
1 2
3 2 (2 × 3) + (1 × 2) (3 × 3) + (2 × 2) 8 13

21
 B is said to be pre-multiplied by A to give AB while A is post-multiplied by
B. Matrix multiplication is commutative in the following regards:
1. Zero matrix: 𝑎 × 0 = 0 × 𝑎 = 0
2. AB = BA if both A and B are of order (1 x 1)
3. Unit matrix: 𝐴 × 𝐼 = 𝐼 × 𝐴 = 𝐴 where I is an identity matrix
4. Scalar matrix: 𝐴 × 𝐾 = 𝐾 × 𝐴. In 1-4 it has been assumed that A is a
square matrix
5. 𝐴𝑚 × 𝐴𝑛 = 𝐴𝑛 × 𝐴𝑚 = 𝐴𝑚+𝑛 where A is square matrix, m and n are
whole numbers +ve or –ve.

4.3.3 Transposition and Inversion

The transpose of an 𝑚 × 𝑛 matrix A is an 𝑛 × 𝑚 matrix 𝐴𝑇 is the same as the
𝑇
corresponding row of A. So, (𝑎𝑖𝑗 ) = 𝑎𝑗𝑖
𝑎11 𝑎12 𝑇
𝑎11 𝑎21
𝐴 = [𝑎 𝑎22 ] , 𝐴 = [𝑎12 𝑎22 ]
21

(𝑎21 )𝑇 = 𝑎12

If
2 −1
2 3 4 𝑇
𝐴=[ ] , 𝐴 = [3 0 ]
−1 0 3
4 3

The transpose of (𝐴)𝑇 = (𝐴𝑇 )𝑇 = 𝐴

4.3.4 The Inverse or Reciprocal of a Matrix

Let 𝐴−1 be defined to be a matrix such that 𝐴𝐴−1 = 𝐼
𝐴−1 exists provided A is square and not singular. 𝐴−1 is of the same order as A

22
 𝐴−1 (𝐴𝐴−1 ) = 𝐴−1 𝐼 = 𝐴−1
𝐴−1 𝐴 = 𝐼 = 𝐴𝐴−1
𝐴−1 𝐴 = 𝐼 = 𝐴𝐴−1
(𝐴𝐵)−1 =𝐵 −1 𝐴−1
(𝐴𝐵𝐶)−1 =𝐶 −1 𝐵 −1 𝐴−1 (Each is a square matrix)
(𝐴 + 𝐵)𝑇 = 𝐴𝑇 + 𝐵 𝑇 (Associative law)

4.3.5 Matrix Division

Multiplication by the inverse is the equivalent of division 𝐴−1 𝐵 is the result of
dividing matrix B by A
𝐵
= 𝐴−1 𝐵
𝐴
Note that it is only possible if A is squared, non-singular and conformable
𝐴−1 𝐴𝑚 = 𝐴𝑚−1
If A is symmetric, 𝐴−1 is symmetric also.

Home Work
Use the method of adjugate matrix N to invert the following normal matrix
3 1 2
𝑁 = [1 3 2]
2 2 4

23
Solution
𝐴𝑑𝑗 𝑁
𝑁 −1 =
det 𝑁
1st Step, to check if det 𝑁 ≠ 0
det 𝑁 = 3(−8) − 1(0) + 2(−4)
det 𝑁 = 24 − 8 = 16
det 𝑁 ≠ 0, ∴ 𝐴−1 𝑒𝑥𝑖𝑠𝑡𝑠.
𝑪𝟏𝟏 = (−𝟏)𝟏+𝟏 |𝑵𝟏𝟏 | = 𝟏(𝟖) = 𝟖
𝑪𝟏𝟐 = (−𝟏)𝟑 |𝑵𝟏𝟐 | = −𝟏(𝟎) = 𝟎
𝑪𝟏𝟑 = (−𝟏)𝟒 |𝑵𝟏𝟑 | = 𝟏(−𝟒) = −𝟒
⋮= ⋮ =⋮ =⋮
𝑪𝟑𝟑 = (−𝟏)𝟔 |𝑵𝟑𝟑 | = 𝟏(𝟖) = 𝟖
8 0 −4
𝑪𝒊𝒋 = [ 0 8 −4]
−4 −4 8

8 0 −4
𝒂𝒅𝒋 𝑵 = 𝑪𝒊𝒋 𝑻 = [ 0 8 −4]
−4 −4 8

−𝟏
1 8 0 −4
∴𝑵 = [0 8 −4]
16
−4 −4 8

Check

16 0 0 1 0 0
1
= 16 [ 0 16 0 ] = [0 1 0 ]
0 0 16 0 0 1

24
5.0 THEORY OF LEAST SQUARES
The purpose of least squares adjustment is to produce the most probable value
(mpv) of a set of variants for which additional values, often called “Redundant
Measures” have been taken.
For a group of equally weighted observation, the fundamental principle that is
enforced in least square adjustment is that the sum of the squares of the residuals
is minimized. This principle which has been developed from the equation of
−𝟏⁄ 𝟐 𝟐
normal distribution curve is given by (𝒚 = 𝒉𝝅 𝟐 𝒆−𝒉 𝜺 ) provides most
probable values (mpv) for the adjusted quantities.
Two basic methods exist for the adjustment of observation by the Least Squares
techniques. They are:
a) Parametric Method (Observation Equations) and
b) Method of Correlates (Condition Equations)
The basic arguments as to which method to adopt for a given problem hinge
on the following:
i. Using observation equations, there are as many normal equations as
there are unknowns.
ii. Using condition equation, there are as many normal equation as
there are conditions.
As the bulk of the work lies in the solution of the normal equation and increases as
square of their number, then in certain circumstances, the use of condition equation
will reduce the number of normal equations and therefore be the method to use.
However, apart from figural adjustments in which condition equation method is
used, the observation equation method is used both in Surveying and
Photogrammetry.

5.1 THEORY OF LEAST SQUARES

The equation for the probability curve is given by
−1⁄ 2 2
𝑦 = ℎ𝜋 2 𝑒 −ℎ 𝜀

where e = the exponential function

h= the index of a precision
y= the probability of occurrence of an error 𝜺
−1
Let 𝜋 2 = A

25
then
2𝜀2
𝑦 = 𝐴ℎ𝑒 −ℎ ……………………………………….(5.1)
Differentiating with respect to h:
𝜕𝑦 2 2 2 2
= 𝐴{𝑒 −ℎ 𝜀 + ℎ(−2ℎ𝜀 2 𝑒 −ℎ 𝜀 )}
𝜕ℎ

𝜕𝑦 2𝜀2
= 𝐴𝑒 −ℎ (1 − 2ℎ2 𝜀 2 )…………………………..(5.2)
𝜕ℎ

For maximum y,
𝜕𝑦
= 0, i.e. 1 − 2ℎ2 𝜀 2 = 0
𝜕ℎ
1
∴ 𝜀 2 = 2ℎ2 …………………………………….(5.3)

Considering errors
1 1 1
𝜀 21 + 𝜀 22 + … + 𝜀 2𝑛 = 2
+ 2 +⋯+ 2
2ℎ 1 2ℎ 2 2ℎ 𝑛
then
1
∑ 𝜀2 = ∑
2ℎ2
and since h represents precision, the accuracy of the observations will increase as
1
h increases. However, as h increases, decreases, thus maximum accuracy will
2ℎ2

be achieved when
1
∑ = ∑ 𝜀 2 = 𝑎 𝑚𝑖𝑛𝑖𝑚𝑢𝑚 ……………………………(5.4)
2ℎ2

If it was “the m.p.v. of the quantity is the one for which the sum of the squares of
the errors (the residuals) is a minimum.” This is the principle of Least Square”
Suppose a group of m equally weighed measurements were taken having residuals
𝑉1 , 𝑉2 , 𝑉3 , … , 𝑉𝑚 , then in equation form, the fundamental condition of least squares
is expressed as
2 2 2 2
∑𝑚 2
𝑖=1(𝑉𝑖) = 𝑉1 + 𝑉2 + 𝑉3 + ⋯ 𝑉𝑚 = 𝑚𝑖𝑛𝑖𝑚𝑢𝑚……………(5.5)

Some basic assumptions which underlie least squares theorem are that the number
of observations being adjusted is large and the frequency distribution of the errors is
normal. Although these basic assumptions are not always met, least squares
adjustment still provides the most rigorous error treatment available. Hence it has
become very popular and important in many areas of surveying.
Advantages derived from the use of least squares adjustment include:

26
 i. It yields the M.P.V. of the unknowns
ii. It enables precisions of adjusted quantities to be determined
iii. It reveals the presence of large errors and mistakes, so that steps can be taken
to eliminate them.

5.2 APPLYING LEAST SQUARES

In the observation equation method of least squares adjustments, observation
equations are written which relate measured values to their residual errors and the
unknown parameters. One observation equation is written for each measurement.
For a unique solution, the number of equations must equal the number of unknowns.
If redundant (excess) observations are made, then more observation equations can
be written than one needed for a unique solution, and MPV of the unknowns can
be determined by the method of least squares. For a group of equally weighted
observations, an equation for each residual error is obtained from each observation
equation. The residuals are squared and added to obtain the function

∑(𝑉𝑖)2

To linearize the function, partial derivatives are taken with respect to each unknown
variable, and said to be equal to zero. This yields a set of equations called Normal
Equations, which are equal in number to the number of the unknowns. The normal
equations are solved to obtain most probable values of the unknown.
In summary, the following steps are adopted in least square:
i. Form observation equations or condition equations for the residuals
ii. Square the residuals to obtain
𝑚

∑(𝑉𝑖)2
𝑖=1

iii. Obtain normal equations by minimizing (or linearizing) the functions in (ii)
iv. Solve normal equations to obtain the Most Probable Values.

6.0 LINEARIZATION OF FUNCTIONS

In least square adjustments, Equation (5.5) is very important i.e.
𝑚

∑(𝑉𝑖)2 = 𝑉1 2 + 𝑉2 2 + 𝑉3 2 + ⋯ 𝑉𝑚 2 = 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
𝑖=1

27
 If the equation were linear or of the form
𝑓𝑎 (𝑣) = 𝑎1 𝑣1 + 𝑎2 𝑣2 + 𝑎3 𝑣3 + ⋯ + 𝑎𝑛 𝑣𝑣 + 𝑙𝑎
𝑓𝑏 (𝑣) = 𝑏1 𝑣1 + 𝑏2 𝑣2 + 𝑏3 𝑣3 + ⋯ + 𝑏𝑛 𝑣𝑣 + 𝑙𝑏 ………(5.6)
Where the coefficients a, b and the last terms la, lb are constant, the solution would
be easy. Therefore, we shall have to linearize equation (5.5) before the values of the
residuals can be obtained from the adjustment.
To linearize a function, we compute the values of the partial derivatives for the observed
quantities (in the case of condition equation) so that
𝜕𝑓𝑎 𝜕𝑓𝑏
𝑎𝑖 = ; 𝑏𝑖 = ……………………..(5.7)
𝜕𝑣𝑖 𝜕𝑣𝑖

For example, in the case of observation equation method

Let 𝑃 = ∑ 𝑣 2
Then the condition for making it minimum equals
𝜕𝑃 𝜕𝑃 𝜕𝑃
= 𝜕𝑣 = 𝜕𝑣 = 0…………………….…….(5.8)
𝜕𝑣1 2 3

In the condition equation method of adjustment (where correlates are used) the least square
condition is applied as a function of the residuals 𝑉𝑖 in a slightly different way from the
observation equation method.
𝐹 = ∑𝑚 2
𝑖=1(𝑉𝑖) − 2 𝐾1 (𝑎1 𝑣1 + ⋯ + 𝑎𝑛 𝑣𝑛 + 𝑞1 ) − 2𝐾2 (𝑏1 𝑣1 + ⋯ + 𝑏𝑛 𝑣𝑛 + 𝑞2 ) −

2𝐾3 (𝑐1 𝑣1 + ⋯ + 𝑐𝑛 𝑣𝑛 + 𝑞3 )……………………(5.9)

where 𝐾1 , 𝐾2 𝑎𝑛𝑑 𝐾3 𝑎𝑟𝑒 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑠

Examples
1. Given that the observation equations for a round of angles at a station are:
𝑥 = 𝑉1
𝑦 = 𝑉2
𝑧 = 𝑉3
𝑥 + 𝑦 + 3 = 𝑉4
𝑦 + 𝑧 + 1 = 𝑉5

28
Apply the least squares condition to linearize it
Solution
𝑚

∑(𝑉𝑖)2 = 𝑉1 2 + 𝑉2 2 + 𝑉3 2 + ⋯ 𝑉𝑚 2 = 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
𝑖=1

then,
𝑃 = 𝑥 2 + 𝑦 2 + 𝑧 2 + (𝑥 + 𝑦 + 3)2 + (𝑦 + 𝑧 + 1)2 = 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
To linearize P
𝜕𝑃
= 2𝑥 + 2(𝑥 + 𝑦 + 3) = 0
𝜕𝑥
= 2𝑥 + 𝑦 + 3 = 0

𝜕𝑃
= 2𝑦 + 2(𝑥 + 𝑦 + 3) + 2(𝑦 + 𝑧 + 1) = 0
𝜕𝑦
= 𝑥 + 3𝑦 + 𝑧 + 4 = 0

𝜕𝑃
= 2𝑧 + 2(𝑦 + 𝑧 + 1) = 0
𝜕𝑧
= 𝑦 + 2𝑧 + 1 = 0
Hence,
2𝑥 + 𝑦 + 3 = 0
𝑥 + 3𝑦 + 𝑧 + 4 = 0 …………………..(5.10)
𝑦 + 2𝑧 + 1 = 0
The linearized function given in equation (5.10) is the Normal Equation

29
 2. Given the condition equations
𝑣1 + 𝑣2 − 𝑣4 + 3 = 0
𝑣2 + 𝑣3 − 𝑣5 − 2 = 0 5.11
𝑣1 + 𝑣2 + 𝑣3 − 𝑣6 − 1 = 0
Apply least squares condition
Solution
The least square condition to apply is in equation (5.9)
𝐹 = 𝑉1 2 + 𝑉2 2 + 𝑉3 2 + 𝑉4 2 + 𝑉5 2 + 𝑉6 2 − 2𝐾1 (𝑣1 + 𝑣2 − 𝑣4 + 3) − 2𝐾2 (𝑣2 + 𝑣3 −
𝑣5 − 2) − 2𝐾3 (𝑣1 + 𝑣2 + 𝑣3 − 𝑣6 − 1)……………………(5.12)
where 𝐾1 , 𝐾2 𝑎𝑛𝑑 𝐾3 𝑎𝑟𝑒 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑠
To linearize (5.12), differentiate P with respect to each variable in turn and equate
to zero
𝜕𝑃
= 2𝑣1 − 2𝑘1 − 2𝑘3 = 0
𝜕𝑣1

𝜕𝑃
= 2𝑣2 − 2𝑘1 − 2𝑘2 − 2𝑘3 = 0
𝜕𝑣2
𝜕𝑃
= 2𝑣3 − 2𝑘2 − 2𝑘3 = 0
𝜕𝑣3
𝜕𝑃
= 2𝑣4 + 2𝑘1 = 0
𝜕𝑣4
𝜕𝑃
= 2𝑣5 + 2𝑘2 = 0
𝜕𝑣5
𝜕𝑃
= 2𝑣6 + 2𝑘3 = 0
𝜕𝑣5

From equation 5.13

𝑣1 = 𝑘1 + 𝑘3
𝑣2 = 𝑘1 + 𝑘2 + 𝑘3
𝑣3 = 𝑘2 + 𝑘3
𝑣4 = −𝑘1
𝑣5 = −𝑘2
𝑣6 = −𝑘3
When (5.14) is substituted back into (5.11), Normal Equations of correlates are
obtained in equation (5.15)
3𝑘1 + 𝑘2 + 2𝑘3 + 3 = 0
𝑘1 + 3𝑘2 + 2𝑘3 − 2 = 0
2𝑘1 + 2𝑘2 + 4𝑘3 − 1 = 0

30
7.0 CLASSICAL AND MATRIX ALGEBRA FORMATION OF
OBSERVATION AND CONDITION EQUATIONS
7.1 Classical Formation of Observation Equations
The aim of field observations is to produce the true or at least the most
probable value (MPV) of that measurements. Provided there are no
systematic errors, the random errors can be distributed using the observation
equation method of least squares adjustments.
The procedure usually adopted is to assume an approximate value for the
quantity and by least squares ascertain the correction to that approximate
value that would produce the MPV. It follows that if the value assumed is as
close as possible to the MPV, then the size of the correction will be
correspondingly smaller. A simple station adjustment would be solved to
illustrate the technique.

Example 7.1

The diagram above illustrates the observations taken to surrounding stations from
O, the following mean values being recorded:
𝑥 = 25° 18 30
𝑦 = 40° 20 25
𝑥 = 30° 30 35
𝑥 + 𝑦 = 65° 38 52
𝑦 + 𝑧 = 70° 51 02
𝑥 + 𝑦 + 𝑧 = 96° 09 31
The last three observations are redundant observations. Use a least squares
adjustment to find the most probable values for x, y and z.

31
 Solution
Step 1: Assume value (𝑥 𝑜 , 𝑦 𝑜 , 𝑧 𝑜 ) for the required quantities.
𝑥 𝑜 = 25° 18 30
𝑦 𝑜 = 40° 20 25
𝑧 𝑜 = 30° 30 35

Step 2: Formulate the observation equations. By applying a correction C to the

assumed values, one obtains the M.P.V.
𝑥 𝑜 + 𝐶1 = 𝑚. 𝑝. 𝑣 𝑜𝑓 𝑥
𝑦 𝑜 + 𝐶2 = 𝑚. 𝑝. 𝑣 𝑜𝑓 𝑦
𝑧 𝑜 + 𝐶3 = 𝑚. 𝑝. 𝑣 𝑜𝑓 𝑧
Denoting the observed values as 𝑙 𝑜𝑏𝑠 , then
(𝑥 𝑜 + 𝐶1 ) − 𝑙 𝑜𝑏𝑠 𝑥 = 𝑉1 (error in angle x)
i.e.
𝑀𝑃𝑉 − 𝑂𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 = 𝑅𝑒𝑠𝑖𝑑𝑢𝑎𝑙 𝑒𝑟𝑟𝑜𝑟
Thus substituting the assumed values from step 1 and the observed values, one gets:
25° 18  30 + 𝐶1 − 25° 18  30 = 𝑉1
∴ 𝐶1 = 𝑉1
Similarly,
𝐶2 = 𝑉2
𝐶3 = 𝑉3
Also,
(𝑥 𝑜 + 𝐶1 ) + (𝑦 𝑜 + 𝐶2 ) − 𝑙(𝑥+𝑦) = 𝑉4
i.e.
(25° 18  30 + 𝐶1 ) + (40° 20 25 + 𝐶2 ) − 65° 38 52 = 𝑉4
∴ 𝐶1 + 𝐶2 + 3 = 𝑉4
Similarly, for (y+z):
(40° 20 25 + 𝐶2 ) + (30° 30 35 + 𝐶3 ) − 70° 51 02 = 𝑉5
∴ 𝐶2 + 𝐶3 − 2 = 𝑉5
and for (x+y+z):
(25° 18  30 + 𝐶1 ) + (40° 20 25 + 𝐶2 ) + (30° 30 35 + 𝐶3 ) − 96° 09 31 =𝑉6
∴ 𝐶1 + 𝐶2 + 𝐶3 − 1 = 𝑉6These equations below are called the observation
equations:

32
 𝐶1 = 𝑉1
𝐶2 = 𝑉2
𝐶3 = 𝑉3
𝐶1 + 𝐶2 + 3 = 𝑉4
𝐶2 + 𝐶3 − 2 = 𝑉5
𝐶1 + 𝐶2 + 𝐶3 − 1 = 𝑉6

Step 3: Formulate the normal equations by applying the least squares condition.

∑ 𝑉 2 = ∑(𝑉. 𝑉) = 𝑎 𝑚𝑖𝑛𝑖𝑚𝑢𝑚

This can also be expressed in terms of the corrections

i.e.

∑(𝐶. 𝐶) = 𝑎 𝑚𝑖𝑛𝑖𝑚𝑢𝑚

If 𝑃 = ∑(𝐶. 𝐶), the condition for making it a minimum becomes:

𝜕𝑃 𝜕𝑃 𝜕𝑃
= = =0
𝜕𝐶1 𝜕𝐶2 𝜕𝐶3
From the observation equations:
𝑃 = 𝐶1 2 + 𝐶2 2 + 𝐶3 2 + (𝐶1 + 𝐶2 + 3)2 + (𝐶2 + 𝐶3 − 2)2
+ (𝐶1 + 𝐶2 + 𝐶3 − 1)2
Therefore,
𝜕𝑃
= 2𝐶1 + 2(𝐶1 + 𝐶2 + 3) + 2(𝐶1 + 𝐶2 + 𝐶3 − 1)
𝜕𝐶1
Equating to zero, eliminating the unwanted factor of 2 and collecting like terms:
3𝐶1 + 2𝐶2 + 𝐶3 + 2 = 0
Similarly,
𝜕𝑃
= 2𝐶2 + 2(𝐶1 + 𝐶2 + 3) + 2(𝐶2 + 𝐶3 − 2) + 2(𝐶1 + 𝐶2 + 𝐶3 − 1) = 0
𝜕𝐶2
= 4𝐶1 + 8𝐶2 + 4𝐶3 = 0
= 𝐶1 + 2𝐶2 + 𝐶3 = 0

𝜕𝑃
= 2𝐶3 + 2(𝐶2 + 𝐶3 − 2) + 2(𝐶1 + 𝐶2 + 𝐶3 − 1) = 0
𝜕𝐶3
= 2𝐶1 + 4𝐶2 + 6𝐶3 − 6 = 0
= 𝐶1 + 2𝐶2 + 3𝐶3 − 3 = 0

33
 These are the Normal Equations and are summarized as follows:
3𝐶1 + 2𝐶2 + 𝐶3 = −2
𝐶1 + 2𝐶2 + 𝐶3 = 0
𝐶1 + 2𝐶2 + 3𝐶3 = 3
The above equations are solved by method of substitution giving:
𝑪𝟏 = −𝟏 , 𝑪𝟐 = −𝟎. 𝟐𝟓 𝒂𝒏𝒅 𝑪𝟑 = 𝟏. 𝟓
These values are now applied to the assumed values to give the M.P.V.
𝑚. 𝑝. 𝑣 𝑜𝑓 𝑥 = 𝑥 𝑜 + 𝐶1 = 25° 18 30 − 1 = 25° 18 29. 00
𝑚. 𝑝. 𝑣 𝑜𝑓 𝑦 = 𝑦 𝑜 + 𝐶2 = 40° 20 25 − 0. 25 = 40° 20 24. 75
𝑚. 𝑝. 𝑣 𝑜𝑓 𝑧 = 𝑧 𝑜 + 𝐶3 = 30° 30 35 + 1. 5 = 30° 30 36. 50

7.11 General Equations for Observation Equations

Expressing the observation equations in general terms:
𝑎1 𝐶1 + 𝑏1 𝐶2 + 𝑐1 𝐶3 − 𝑙1 = 𝑉1
𝑎2 𝐶1 + 𝑏2 𝐶2 + 𝑐2 𝐶3 − 𝑙2 = 𝑉2
⋮
𝑎𝑛 𝐶1 + 𝑏𝑛 𝐶2 + 𝑐𝑛 𝐶3 − 𝑙𝑛 = 𝑉𝑛
that is
𝑀𝑃𝑉 − 𝑂𝑏𝑠𝑒𝑟𝑣𝑒𝑑 𝑣𝑎𝑙𝑢𝑒 = 𝑅𝑒𝑠𝑖𝑑𝑢𝑎𝑙 𝑒𝑟𝑟𝑜𝑟
From least squares ∑(𝑉. 𝑉) = 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
Thus squaring 𝑉1gives:

𝑉1 2 = 𝑎1 2 𝐶1 2 + 2𝑎1 𝑏1 𝐶1 𝐶2 + 2𝑎1 𝑐1 𝐶1 𝐶3 − 2𝑎1 𝑙1 𝐶1 + 𝑏1 2 𝐶2 2 + 2𝑏1 𝑐1 𝐶2 𝐶3 − 2𝑏1 𝑙1 𝐶2

+ 𝑐1 2 𝐶3 2 − 2𝑐1 𝑙1 𝐶3 + 𝑙1 2

Repeating for 𝑉2 , … , 𝑉𝑛 will only change the coefficients to 𝑎2 , 𝑏2 , 𝑐2 and 𝑎𝑛 , 𝑏𝑛 , 𝑐𝑛 . Thus

summing the results and expressing the sum of the squares in the usual manner, (i.e.

∑ 𝑉 2 𝑎𝑠 ∑(𝑉. 𝑉),

One gets

∑(𝑉. 𝑉) = ∑(𝑎. 𝑎) 𝐶1 2 + 2 ∑(𝑎. 𝑏)𝐶1 𝐶2 + 2 ∑(𝑎. 𝑐)𝐶1 𝐶3 − 2 ∑(𝑎. 𝑙) 𝐶1

+ ∑(𝑎. 𝑏)𝐶2 2 + 2 ∑(𝑏. 𝑐) 𝐶2 𝐶3 − 2 ∑(𝑏. 𝑙)𝐶2 + ∑(𝑐. 𝑐) 𝐶3 2

− 2 ∑(𝑐. 𝑙)𝐶3 + ∑(𝑙. 𝑙)

34
As ∑(𝑉. 𝑉) = 𝑓(𝐶1 , 𝐶2 , 𝐶3 , differentiate and equate to zero for a minimum.
𝜕𝑓
= 2 ∑(𝑎. 𝑎) 𝐶1 + 2 ∑(𝑎. 𝑏) 𝐶2 + 2 ∑(𝑎. 𝑐) 𝐶3 − 2 ∑(𝑎. 𝑙) = 0
𝜕𝐶1

𝜕𝑓
= 2 ∑(𝑎. 𝑏) 𝐶1 + 2 ∑(𝑏. 𝑏) 𝐶2 + 2 ∑(𝑏. 𝑐) 𝐶3 − 2 ∑(𝑏. 𝑙) = 0
𝜕𝐶2

𝜕𝑓
= 2 ∑(𝑎. 𝑐) 𝐶1 + 2 ∑(𝑏. 𝑐) 𝐶2 + 2 ∑(𝑐. 𝑐) 𝐶3 − 2 ∑(𝑐. 𝑙) = 0
𝜕𝐶3
These reduce to the general form for normal equations as follows:
[𝑎𝑎]𝐶1 + [𝑎𝑏]𝐶2 + [𝑎𝑐]𝐶3 = [𝑎𝑙]
[𝑎𝑏]𝐶1 + [𝑏𝑏]𝐶2 + [𝑏𝑐]𝐶3 = [𝑏𝑙]
[𝑎𝑐]𝐶1 + [𝑏𝑐]𝐶2 + [𝑐𝑐]𝐶3 = [𝑐𝑙]

7.12 Use of the General Equation

Consider the example in Section 7.1 again
𝐶1 = 𝑉1
𝐶2 = 𝑉2
𝐶3 = 𝑉3
𝐶1 + 𝐶2 + 3 = 𝑉4
𝐶2 + 𝐶3 − 2 = 𝑉5
𝐶1 + 𝐶2 + 𝐶3 − 1 = 𝑉6

𝐶1 = 0
𝐶2 = 0
𝐶3 = 0
𝐶1 + 𝐶2 = −3
𝐶2 + 𝐶3 = 2
𝐶1 + 𝐶2 + 𝐶3 = 1
Written in the general form, the above becomes
𝑎𝐶1 = 0
𝑏𝐶2 = 0
𝑐𝐶3 = 0
𝑎𝐶1 + 𝑏𝐶2 = −3

35
𝑏𝐶2 + 𝑐𝐶3 = 2
𝑎𝐶1 + 𝑏𝐶2 + 𝑐𝐶3 = 1
Note that 𝑎 = 𝑏 = 𝑐 = 1
a b c l
1 0 0 0
0 1 0 0
0 0 1 0
1 1 0 -3
0 1 1 2
1 1 1 1
From above, the product and summation of columns gives
[𝑎𝑎] = 3 [𝑎𝑏] = 2 [𝑎𝑐] = 1 [𝑎𝑙] = −2
[𝑏𝑏] = 4 [𝑏𝑐] = 2 [𝑏𝑙] = 0
[𝑐𝑐] = 3 [𝑐𝑙] = 3
Sub in equation (7.1)
3𝐶1 + 2𝐶2 + 𝐶3 = −2
2𝐶1 + 4𝐶2 + 2𝐶3 = 0
𝐶1 + 2𝐶2 + 3𝐶3 = 3
Homework
Consider the following three equally weighted measurements taken between
points A, B and C in the diagram below:

𝑥 + 𝑦 = 3.0
𝑥 = 1.5
𝑦 = 1.4
Obtain the least squares solution for the most probable values of x and y using
classical observation equation method.

36
 7.2 Classical Formation of Condition Equations
The condition equation method has its greatest advantage when the number of unknowns
is by far greater than the number of conditions to be satisfied. To reduce the number of
equations, the method is generally used in conjunction with “correlates”.
A correlate is any number by which the condition equations are multiplied to form the
correlate normal equations. The value of the correlate is unknown until the solution of the
correlate normal equations, but their relationship to the correction is unknown.

Example 7.2
Use the condition equation method to solve the Example 7.1

The diagram above illustrates the observations taken to surrounding stations from
O, the following mean values being recorded:
𝑥 = 25° 18 30
𝑦 = 40° 20 25
𝑥 = 30° 30 35
(𝑥 + 𝑦) = 65° 38 52
(𝑦 + 𝑧) = 70° 51 02
(𝑥 + 𝑦 + 𝑧) = 96° 09 31

37
From the diagram, the conditions of adjustment are as follows:
𝑥 + 𝑦 = (𝑥 + 𝑦)
𝑦 + 𝑧 = (𝑦 + 𝑧)
𝑥 + 𝑦 + 𝑧 = (𝑥 + 𝑦 + 𝑧)
Note: It is important to use just the number of conditions. A rule to decide the correct
number of directly observed quantities minus number of independent unknowns equals
the number of conditions.
In the above case;
Number of directly observed quantities = 6
Number of independent unknowns = 3 (i.e. x, y, z)
Therefore, the number of conditions = 3
However, these conditions are only true for the MPV, thus corrections 𝑉1 , 𝑉2 , … , 𝑉6 are
applied for measured quantities in equation (7.2a)
𝑥 + 𝑦 = (𝑥 + 𝑦)
𝑦 + 𝑧 = (𝑦 + 𝑧)
𝑥 + 𝑦 + 𝑧 = (𝑥 + 𝑦 + 𝑧)
Applying correction to each observed value

𝑥 + 𝑉1 + 𝑦 + 𝑉2 = (𝑥 + 𝑦) + 𝑉4
𝑦 + 𝑉2 + 𝑧 + 𝑉3 = (𝑦 + 𝑧) + 𝑉5
𝑥 + 𝑉1 + 𝑦 + 𝑉2 + 𝑧 + 𝑉3 = (𝑥 + 𝑦 + 𝑧) + 𝑉6
Substituting the observed values into equation (7.3a),
𝑥 + 𝑉1 + 𝑦 + 𝑉2 = (𝑥 + 𝑦) + 𝑉4
25° 18 30 + 𝑉1 + 40° 20 25 + 𝑉2 = 65° 38 52 + 𝑉4
Resulting into 1st condition equation:
𝑉1 + 𝑉2 − 𝑉4 + 3 = 0………(7.4a)
Similarly, for the two remaining conditions,
40° 20 25 + 𝑉2 + 30° 30 35 + 𝑉3 = 70° 51 02 + 𝑉5
𝑉2 + 𝑉3 − 𝑉5 − 2 = 0………(7.5a)
and
25° 18 30 + 𝑉1 + 40° 20 25 + 𝑉2 + 30° 30 35 + 𝑉3 = 96° 09 31 + 𝑉6
𝑉1 + 𝑉2 + 𝑉3 − 𝑉6 − 1 = 0………(7.6a)
First, multiply the condition equations in (7.4a-7.6a) by the correlates 𝐾1 , 𝐾2 𝑎𝑛𝑑 𝐾3 :

38
 𝑉1 + 𝑉2 − 𝑉4 + 3 = 0
𝑉2 + 𝑉3 − 𝑉5 − 2 = 0
𝑉1 + 𝑉2 + 𝑉3 − 𝑉6 − 1 = 0
Resolving further by applying the least squares condition:
𝑚

𝐹 = ∑(𝑉𝑖)2 − 2 𝐾1 (𝑎1 𝑣1 + ⋯ + 𝑎𝑛 𝑣𝑛 + 𝑞1 ) − 2𝐾2 (𝑏1 𝑣1 + ⋯ + 𝑏𝑛 𝑣𝑛 + 𝑞2 )

𝑖=1

− 2𝐾3 (𝑐1 𝑣1 + ⋯ + 𝑐𝑛 𝑣𝑛 + 𝑞3 )
where 𝐾1 , 𝐾2 𝑎𝑛𝑑 𝐾3 𝑎𝑟𝑒 𝑐𝑜𝑟𝑟𝑒𝑙𝑎𝑡𝑒𝑠
Differentiating w.r.t. each variable and equating it to zero and substituting the correlate
functions into the condition equations as on page 30, the Normal Equations of Correlates
are obtained as:
3𝑘1 + 𝑘2 + 2𝑘3 + 3 = 0
𝑘1 + 3𝑘2 + 2𝑘3 − 2 = 0
2𝑘1 + 2𝑘2 + 4𝑘3 − 1 = 0
Note the symmetry in the above equation; normal equation should always be square and
symmetrical.

Solution
3 1 2 −3
𝐴 = [1 3 2] = [ 2 ]
2 2 4 1

𝐴𝑑𝑗
𝐴−1 =
|𝐴|
|𝐴| = 16

Finding the co-factors:

8 0 −4
[0 8 −4]
−4 −4 8
Finding the transpose of the cofactor:
8 0 −4
[0 8 −4]
−4 −4 8

39
 𝐴𝑑𝑗
𝐴−1 =
|𝐴|

8 0 −4
[0 8 −4]
−4 −4 8
16

1⁄ 0 −1⁄
2 4 −3 𝑘1
𝐴−1 = 0 1⁄ −1⁄ = [ 2 ] = [𝑘2 ]
2 4
−1 −1⁄ 1⁄ 1 𝑘3
[ ⁄4 4 2]

On solution, 𝐾1 = −1.75, 𝐾2 = 0.75 𝑎𝑛𝑑 𝐾3 = 0.75

These values are now substituted in equation (5.14) on page 30,
𝑣1 = 𝑘1 + 𝑘3 = −1
𝑣2 = 𝑘1 + 𝑘2 + 𝑘3 = −0.25
𝑣3 = 𝑘2 + 𝑘3 = 1.50
𝑣4 = −𝑘1 = 1.75
𝑣5 = −𝑘2 = −0.75
𝑣6 = −𝑘3 = −0.75
These corrections are now applied to the appropriate observed values to give the MPVs
𝑥 = 25° 18 30 + 𝑉1 = 25° 18 29
𝑦 = 40° 20 25 + 𝑉2 = 40° 20 24.75
𝑥 = 30° 30 35 +𝑉3 = 30° 30 36.50
(𝑥 + 𝑦) = 65° 38 52+𝑉4=65° 38 53.75
(𝑦 + 𝑧) = 70° 51 02+𝑉5= 70° 51 1.25
(𝑥 + 𝑦 + 𝑧) = 96° 09 31+𝑉6= 96° 09 30.25

7.2.1 Classical Formation of Condition Equations

Writing the condition equations in a general form:
𝑎1 𝑉1 + 𝑎2 𝑉2 + ⋯ + 𝑎𝑛 𝑉𝑛 + 𝑞1 = 0
𝑏1 𝑉1 + 𝑏2 𝑉2 + ⋯ + 𝑏𝑛 𝑉𝑛 + 𝑞2 = 0
𝑐1 𝑉1 + 𝑐2 𝑉2 + ⋯ + 𝑐𝑛 𝑉𝑛 + 𝑞3 = 0

40
Applying the least squares condition, we obtain:
𝐹 = 𝑉1 2 + 𝑉2 2 + ⋯ + 𝑉𝑛 2 − 2𝐾1 (𝑎1 𝑣1 + 𝑎2 𝑣2 + ⋯ + 𝑎𝑛 𝑣𝑛 + 𝑞1 )
− 2𝐾2 (𝑏1 𝑣1 + 𝑏2 𝑣2 + ⋯ + 𝑏𝑛 𝑣𝑛 + 𝑞2 )
− 2𝐾3 (𝑐1 𝑣1 + 𝑐2 𝑣2 + ⋯ + 𝑐𝑛 𝑣𝑛 + 𝑞3 ) = 𝑎 𝑚𝑖𝑛𝑖𝑚𝑢𝑚
Differentiating each variable in turn and equating to zero,
𝜕𝐹
= 2𝑣1 − 2𝑘1 𝑎1 − 2𝑘2 𝑏1 − 2𝑘3 𝑐1 = 0
𝜕𝑣1
𝜕𝐹
= 2𝑣2 − 2𝑘1 𝑎2 − 2𝑘2 𝑏2 − 2𝑘3 𝑐2 = 0
𝜕𝑣2
:
:
𝜕𝐹
= 2𝑣𝑛 − 2𝑘1 𝑎𝑛 − 2𝑘2 𝑏𝑛 − 2𝑘3 𝑐𝑛 = 0
𝜕𝑣𝑛
The above equation reduced to:
𝑣1 = 𝑘1 𝑎1 + 𝑘2 𝑏1 + 𝑘3 𝑐1
𝑣2 = 𝑘1 𝑎2 + 𝑘2 𝑏2 + 𝑘3 𝑐2
:
:
𝑣𝑛 = 𝑘1 𝑎𝑛 + 𝑘2 𝑏𝑛 + 𝑘3 𝑐𝑛
Substituting these values into the original condition equations and substituting K for k
simply to emphasize the format, gives the general form for correlate normal equations:
𝐾1 [𝑎𝑎] + 𝐾2 [𝑎𝑏] + 𝐾3 [𝑎𝑐] + 𝑞1 = 0
𝐾1 [𝑎𝑏] + 𝐾2 [𝑏𝑏] + 𝐾3 [𝑏𝑐] + 𝑞2 = 0
𝐾1 [𝑎𝑐] + 𝐾2 [𝑏𝑐] + 𝐾3 [𝑐𝑐] + 𝑞3 = 0
Note the symmetry of the equation; note also that these equations are identical with the
previous ones (7.1) derived for the observation equation method with the correction C
replaced by K.

7.2.2 Use of General Equations

The equation in (7.8) can be used mechanically to produce the normal equation direct from
condition equations. For the condition equation below
𝑣1 + 𝑣2 − 𝑣4 + 3 = 0
𝑣2 + 𝑣3 − 𝑣5 − 2 = 0
𝑣1 + 𝑣2 + 𝑣3 − 𝑣6 − 1 = 0

41
 Multiply equation by the correlate k, it then becomes
𝑘1 (𝑣1 + 𝑣2 − 𝑣4 + 3) = 0………………….. (7.9)
𝑘2 (𝑣2 + 𝑣3 − 𝑣5 − 2) = 0………………(7.10)
𝑘3 (𝑣1 + 𝑣2 + 𝑣3 − 𝑣6 − 1) = 0……………(7.11)
From the original derivation (see section 7.2.1), it can be seen that all the coefficients of
equation 7.9 are “a”, of 7.10 are “b” and of 7.11 are “c”. The equation will therefore be re-
written purely to facilitate the explanation of the method.
𝑘1 (𝑎1 𝑣1 + 𝑎2 𝑣2 − 𝑎4 𝑣4 + 3) = 0………………….. (7.12)
𝑘2 (𝑏2 𝑣2 + 𝑏3 𝑣3 − 𝑏5 𝑣5 − 2) = 0……………...……(7.13)
𝑘3 (𝑐1 𝑣1 + 𝑐2 𝑣2 + 𝑐3 𝑣3 − 𝑐6 𝑣6 − 1) = 0……………(7.14)
where a= b = c = 1 in this case.
Forming normal equation using classical/mechanical method
𝐾1 [𝑎𝑎] + 𝐾2 [𝑎𝑏] + 𝐾3 [𝑎𝑐] + 𝑞1 = 0
𝐾1 [𝑎𝑏] + 𝐾2 [𝑏𝑏] + 𝐾3 [𝑏𝑐] + 𝑞2 = 0
𝐾1 [𝑎𝑐] + 𝐾2 [𝑏𝑐] + 𝐾3 [𝑐𝑐] + 𝑞3 = 0

V1 V2 V3 V4 V5 V6 q
K1 a 1 1 0 -1 0 0 3
K2 b 0 1 1 0 -1 0 -2
K3 c 1 1 1 0 0 -1 -1

3𝑘1 + 𝑘2 + 2𝑘3 + 3 = 0
𝑘1 + 3𝑘2 + 2𝑘3 − 2 = 0
2𝑘1 + 2𝑘2 + 4𝑘3 − 1 = 0

7.3 Matrix Algebra Formation of Observation Equations

𝑎1 𝐶1 + 𝑏1 𝐶2 + 𝑐1 𝐶3 − 𝑙1 = 𝑉1
𝑎2 𝐶1 + 𝑏2 𝐶2 + 𝑐2 𝐶3 − 𝑙2 = 𝑉2
⋮
𝑎𝑛 𝐶1 + 𝑏𝑛 𝐶2 + 𝑐𝑛 𝐶3 − 𝑙𝑛 = 𝑉𝑛

In summary,
𝐴𝑋 − 𝐿 = 𝑉

OR

𝐴𝑋 = 𝐿 + 𝑉

42
Expressing the equation in matrix formula

𝑎1 𝑏1 𝑐1 𝐶1
𝐴 = [ 𝑎2 𝑏2 𝑐2 ] 𝑋 = [𝐶2 ]
𝑎𝑛 𝑏𝑛 𝑐𝑛 𝐶3
Design Matrix Matrix of Unknown

𝑙1 𝑉1
𝑙2 𝑉2
𝑳 = 𝑙3 𝑽 = 𝑉3
⋮ ⋮
[𝑙 𝑛 ] [𝑉 𝑛 ]
Observation Matrix Residual Matrix

Reversible form of Observation Equation

[𝑎𝑎]𝐶1 + [𝑎𝑏]𝐶2 + [𝑎𝑐]𝐶3 = [𝑎𝑙]
[𝑎𝑏]𝐶1 + [𝑏𝑏]𝐶2 + [𝑏𝑐]𝐶3 = [𝑏𝑙]
[𝑎𝑐]𝐶1 + [𝑏𝑐]𝐶2 + [𝑐𝑐]𝐶3 = [𝑐𝑙]
From the above, normal matrix is formed

𝐴𝑇 𝐴𝐶 = 𝐴𝑇 𝐿

𝐴𝑇 𝐴𝑋 = 𝐴𝑇 𝐿 ……………..(7.17)

𝑁𝑋 = 𝐴𝑇 𝐿 = 𝑈

𝑙1
𝑎1 𝑎2 𝑎𝑛 𝑎1 𝑏1 𝑐1 𝑎1 𝑎2 𝑎𝑛 𝑙 2
[𝑏1 𝑏2 𝑏𝑛 ] . [ 𝑎2 𝑏2 𝑐2 ] × 𝐶 = [ 𝑏1 𝑏2 𝑏𝑛 ] . 𝑙3
𝑐1 𝑐2 𝑐𝑛 𝑎𝑛 𝑏𝑛 𝑐𝑛 𝑐1 𝑐2 𝑐𝑛 ⋮
[𝑙 𝑛 ]

𝑨𝑻 𝑨 𝑪 = 𝑨𝑻 𝑳

In developing matrix equations for least squares computations, analogue will remain to
the algebraic approach given in section 7.11
First of all, the observation equation (7.0) may be represented in matrix form as
𝐴𝑋 = 𝐿 + 𝑉
It will be noted that normal equations (7.1) are obtained as follows
𝐴𝑇 𝐴𝑋 = 𝐴𝑇 𝐿
In the above equation (7.17) is the matrix of normal equation of the unknown. A is called
the design matrix pre-multiplying both sides of equation (7.17) by (𝐴𝑇 𝐴)−1 and reducing
we have

43
 (𝐴𝑇 𝐴)−1 (𝐴𝑇 𝐴)𝑋 = (𝐴𝑇 𝐴)−1 𝐴𝑇 𝐿
𝐼𝑋 = (𝐴𝑇 𝐴)−1 𝐴𝑇 𝐿
𝑋 = (𝐴𝑇 𝐴)−1 𝐴𝑇 𝐿
𝑋 = 𝑁 −1 𝑈……….(7.18)
where
𝑁 = 𝐴𝑇 𝐴= Normal matrix
𝑈 = 𝐴𝑇 𝐿
In the above reduction, I is the identity matrix. Equation (7.18) is the basic least squares
matrix equation for equally weighted observations.
Note: Sometimes 𝐴𝑇 𝐴 is called the normal matrix or N-matrix

Example
Solve the following observation equations using matrix method.
𝑥 + 𝑦 = 3.0
𝑥 = 1.5
𝑦 = 1.4
Solution
The above observation equations may be expressed in matrix form as follows:
𝐴𝑋 = 𝐿 + 𝑉
where
1 1 𝑥
𝐴 = [1 0] 𝑋 = [𝑦 ]
0 1

3.0 𝑉1
𝐿 = [1.5] 𝑉 = [𝑉2 ]
1.4 𝑉3

𝐴𝑇 𝐴𝑋 = 𝐴𝑇 𝐿
1 1
1 1 0 2 1
𝐴𝑇 𝐴 = [ ] . [1 0] = [ ]
1 0 1 1 2
0 1
3.0
1 1 0 4.5
𝐴𝑇 𝐿 = [ ] . [1.5] = [ ]
1 0 1 4.4
1.4

44
 𝐴𝑇 𝐴𝑋 = 𝐴𝑇 𝐿
2 1 𝑥 4.5
[ ] . [𝑦 ] = [ ]
1 2 4.4
Normal equation is therefore
2𝑥 + 𝑦 = 4.5
𝑥 + 2𝑦 = 4.4
Solving simultaneously,
𝒙 = 𝟏. 𝟓𝟑𝟑, 𝒚 = 𝟏. 𝟒𝟑𝟑

OR
𝐴𝑇 𝐴𝑋 = 𝐴𝑇 𝐿

2 1 𝑥 4.5
[ ].[ ] = [ ]
1 2 𝑦 4.4

𝐴𝑑𝑗 𝑁
𝑁 −1 = = (𝐴𝑇 𝐴)−1
det 𝑁

2 −1
[− ]
1 2
3

2⁄ 1
[ 3 − ⁄3] . [4.5] = [𝑥 ]
−1⁄ 2⁄ 4.4 𝑦
3 3

𝒙 = 𝟏. 𝟓𝟑𝟑, 𝒚 = 𝟏. 𝟒𝟑𝟑

7.4 Matrix Algebra Formation of Condition Equation

The general condition equations in eqn (7.8a) can be re-written as
𝑎1 𝑣1 + 𝑎2 𝑣2 + ⋯ + 𝑎𝑛 𝑣𝑛 = −𝑞1
𝑏1 𝑣1 + 𝑏2 𝑣2 + ⋯ + 𝑏𝑛 𝑣𝑛 = −𝑞2
𝑐1 𝑣1 + 𝑐2 𝑣2 + ⋯ + 𝑐𝑛 𝑣𝑛 = −𝑞3
The matrix rotation for equation (7.19) is BV = Q ….(7.20)
where

45
 𝑉1
𝑎1 𝑎2 𝑎𝑛 𝑉2 −𝑞1
𝐵 = [𝑏1 𝑏2 𝑏𝑛 ] 𝑽 = 𝑉3 𝑄 = [−𝑞2 ]
𝑐1 𝑐2 𝑐𝑛 ⋮ −𝑞3
[𝑉 𝑛 ]

The normal equations are obtained from equation 7.8 as follows

𝐵𝐵 𝑇 𝐾 = 𝑄 or 𝑁𝐾 = 𝑄
where
𝐵𝐵 𝑇 = 𝑁
𝐾1
𝐾
𝐾 = [ 2]
𝐾3
The solution is
𝐾 = (𝐵𝐵𝑇 )−1 𝑄
= 𝑁 −1 𝑄 …………..(7.22)
Then, V’s are given by
𝑉 = 𝐵𝑇 𝐾
𝑉 = 𝐵 𝑇 (𝐵𝐵 𝑇 )−1 𝑄
𝑉 = 𝐵 𝑇 𝑁 −1 𝑄 ……….(7.23)
Example
Use the matrix method to solve the following condition equations.
𝑣1 + 𝑣2 − 𝑣4 + 3 = 0
𝑣2 + 𝑣3 − 𝑣5 − 2 = 0
𝑣1 + 𝑣2 + 𝑣3 − 𝑣6 − 1 = 0

where
1 1 0 −1 0 0 −3
𝐵 = [0 1 1 0 −1 0 ] , 𝑄=[ 2 ]
1 1 1 0 0 −1 1
𝑉1
𝑉2
𝐾1
𝑉
𝑽= 3 Let 𝐊 = [𝐾2 ]
𝑉4
𝐾3
𝑉5
[𝑉6 ]
Form Normal equations of correlates from (7.21)

46
 3 1 2
𝑁 = 𝐵𝐵 𝑇 = [1 3 2]
2 2 4
1 8 0 −4
𝑁 −1 = (𝐵𝐵 𝑇 )−1 = [0 8 −4]
16
−4 −4 8

𝐾 = 𝑁 −1 𝑄
1⁄ 0 −1⁄ −7⁄
2 4 −3 4
𝐾= 0 1⁄ −1⁄ . [ 2 ] = 3⁄
2 4 4
−1⁄ −1⁄ 1⁄ 1 3⁄
[ 4 4 2] [ 4]

−1
𝑉1 1 0 1 −1⁄4
𝑉2 1 1 1 −7⁄4 3⁄
𝑉3 0 1 1 2
𝑉= = . 3⁄4 = 7
𝑉4 −1 0 0 ⁄4
𝑉5 0 −1 3
0 [ ⁄4 ] −3⁄
[𝑉6 ] [ 0 0 −1] 4
−3⁄
[ 4]
𝑉 = 𝐵𝑇 𝐾

Questions
Check Pages 135-177 of Surveying Problem Solving Textbook for related problems.

47
OBJECTIVE TYPE QUESTIONS
1. Theory of errors is applied to minimize
(a) the gross errors.
(b) the systematic errors.
(c) the random errors.
(d) all the above.
2. Most probable value of a quantity is equal to
(a) observed value + correction.
(b) the observed value – correction.
(c) the true value + correction.
(d) the true value – correction.
3. The method of least squares of determining the most probable value of a quantity is based upon the
criterion that
(a) S Correction2 = a minimum.
(b) S Error2 = a minimum.
(c) S (Weight × correction)2 = a minimum.
(d) S Residual2 = a minimum.
4. If the observations of a quantity contains systematic and random errors, the most probable value of the
quantity is obtained by
(a) removing the systematic and random errors from the observations.
(b) removing the systematic errors and minimizing the residuals from the observations.
(c) removing the random errors and minimizing the systematic errors from the observations.
(d) minimizing the systematic and random errors from the observations.
5. The most probable value of a quantity is the quantity which is nearest to
(a) the true value of the quantity.
(b) the true value of the quantity ± standard deviation.
(c) the true value of the quantity ± probable error.
(d) the observed value of the quantity ± weight of the observation.
6. The theory of least squares is used in
(a) the method of differences.
(b) in the normal equation method.
(c) the method of correlates.
(d) all the above.
7. In a braced quadrilateral the number of conditions required to be satisfied for adjustment
excluding the condition imposed by least squares theory, is
(a)2.
(b)3.
(c)4.
(d)5.
8. The spherical excess for a triangle of area 200 sq km is approximately
(a) 0.5″ .
(b) 1.0″ .
(c) 1.5″ .
(d) 2.0″ .
9. Correlate is the unknown multiplier used to determine the most probable values by multiplying
it with
(a) normal equation.
(b) observation equation.
(c) condition equation.
(d) condition imposed by the least squares theory .
10. Station adjustment of observation means
(a) making sum of the angles observed around a station equal to 360°.
(b) checking the permanent adjustment of the instrument at every station.
(c) adjusting the instrument so that it is exactly over the station.
(d) shifting the station location to make it intervisible from other stations.
ANSWERS

48
1. (c) 2. (a) 3. (d) 4. (b) 5. (a) 6. (d)
7. (c) 8. (b) 9. (c) 10. (a)

49