Optimization

Nonlinear programming:
One dimensional minimization methods
 Introduction
The basic philosophy of most of the numerical methods of
optimization is to produce a sequence of improved approximations
to the optimum according to the following scheme:

1. Start with an initial trial point Xi

2. Find a suitable direction Si (i=1 to start with) which points in the
general direction of the optimum
3. Find an appropriate step length i* for movement along the direction
Si
4. Obtain the new approximation Xi+1 as
X i 1  X i  *i S i
5. Test whether Xi+1 is optimum. If Xi+1 is optimum, stop the procedure.
Otherwise set a new i=i+1 and repeat step (2) onward.
Iterative Process of Optimization
 Introduction
• The iterative procedure indicated is valid for unconstrained as
well as constrained optimization problems.

• If f(X) is the objective function to be minimized, the problem

of determining i* reduces to finding the value i = i* that
minimizes f (Xi+1) = f (Xi+ i Si) = f (i ) for fixed values of Xi
and Si.

• Since f becomes a function of one variable i only, the methods

of finding i* in the previous slide are called one-dimensional
minimization methods.
 One dimensional minimization
methods
• Analytical methods (differential calculus methods)
• Numerical methods
– Elimination methods
• Unrestricted search
• Exhaustive search
• Dichotomous search
• Fibonacci method
• Golden section method
– Interpolation methods
• Requiring no derivatives (quadratic)
• Requiring derivatives
– Cubic
– Direct root
» Newton
» Quasi-Newton
» Secant
 One dimensional minimization
methods
Differential calculus methods:

• Analytical method

• Applicable to continuous, twice differentiable functions

• Calculation of the numerical value of the objective function is

virtually the last step of the process

• The optimal value of the objective function is calculated after

determining the optimal values of the decision variables
 One dimensional minimization
methods
Numerical methods:

• The values of the objective function are first found at various combinations
of the decision variables

• Conclusions are then drawn regarding the optimal solution

• Elimination methods can be used for the minimization of even

discontinuous functions

• The quadratic and cubic interpolation methods involve polynomial

approximations to the given function

• The direct root methods are root finding methods that can be considered to
be equivalent to quadratic interpolation
 Unimodal function
• A unimodal function is one that has only one peak (maximum)
or valley (minimum) in a given interval

• Thus a function of one variable is said to be unimodal if, given

that two values of the variable are on the same side of the
optimum, the one nearer the optimum gives the better
functional value (i.e., the smaller value in the case of a
minimization problem). This can be stated mathematically as
follows:
A function f (x) is unimodal if
– x1 < x2 < x* implies that f (x2) < f (x1) and
– x2 > x1 > x* implies that f (x1) < f (x2) where x* is the minimum point
 Unimodal function
• Examples of unimodal functions:

• Thus, a unimodal function can be a nondifferentiable or

even a discontinuous function

• If a function is known to be unimodal in a given range, the

interval in which the minimum lies can be narrowed down
provided that the function values are known at two different
values in the range.
 Unimodal function
• For example, consider the normalized interval [0,1] and two function
evaluations within the interval as shown:

• There are three possible outcomes:

– f1 < f2
– f1 > f2

– f1 = f2
 Unimodal function

• If the outcome is f1 < f2, the minimizing x can not lie to the
right of x2

• Thus, that part of the interval [x2,1] can be discarded and a

new small interval of uncertainty, [0, x2] results as shown in
the figure
 Unimodal function

• If the outcome is f (x1) > f (x2) , the interval [0, x1] can be
discarded to obtain a new smaller interval of uncertainty, [x1,
1].
 Unimodal function

• If f1 = f2 , intervals [0, x1] and [x2,1] can both be discarded to

obtain the new interval of uncertainty as [x1,x2]
 Unimodal function

• Furthermore, if one of the experiments (function evaluations in the

elimination method) remains within the new interval, as will be the
situation in Figs (a) and (b), only one other experiment need be placed
within the new interval in order that the process be repeated.

• In Fig (c), two more experiments are to be placed in the new interval in
order to find a reduced interval of uncertainty.
 Unimodal function

• The assumption of unimodality is made in all the elimination

techniques

• If a function is known to be multimodal (i.e., having several

valleys or peaks), the range of the function can be subdivided
into several parts and the function treated as a unimodal
function in each part.
 Elimination methods
In most practical problems, the optimum solution is known to lie within
restricted ranges of the design variables. In some cases, this range is not
known, and hence the search has to be made with no restrictions on the
values of the variables.

UNRESTRICTED SEARCH

• Search with fixed step size

• Search with accelerated step size

 Unrestricted Search
Search with fixed step size
• The most elementary approach for such a problem is to use a
fixed step size and move from an initial guess point in a
favorable direction (positive or negative).

• The step size used must be small in relation to the final

accuracy desired.

• Simple to implement

• Not efficient in many cases

 Unrestricted Search
Search with fixed step size
1. Start with an initial guess point, say, x1
2. Find f1 = f (x1)
3. Assuming a step size s, find x2=x1+s
4. Find f2 = f (x2)
5. If f2 < f1, and if the problem is one of minimization, the
assumption of unimodality indicates that the desired
minimum can not lie at x < x1. Hence the search can be
continued further along points x3, x4,….using the
unimodality assumption while testing each pair of
experiments. This procedure is continued until a point,
xi=x1+(i-1)s, shows an increase in the function value.
 Unrestricted Search
Search with fixed step size (cont’d)

6. The search is terminated at xi, and either xi or xi-1 can be

taken as the optimum point
7. Originally, if f1 < f2 , the search should be carried in the
reverse direction at points x-2, x-3,…., where x-j=x1- ( j-1 )s
8. If f2=f1 , the desired minimum lies in between x1 and x2, and
the minimum point can be taken as either x1 or x2.
9. If it happens that both f2 and f-2 are greater than f1, it implies
that the desired minimum will lie in the double interval
x-2 < x < x2
 Unrestricted Search
Search with accelerated step size
• Although the search with a fixed step size appears to be very
simple, its major limitation comes because of the
unrestricted nature of the region in which the minimum can
lie.

• For example, if the minimum point for a particular function

happens to be xopt=50,000 and in the absence of knowledge
about the location of the minimum, if x1 and s are chosen as
0.0 and 0.1, respectively, we have to evaluate the function
5,000,001 times to find the minimum point. This involves a
large amount of computational work.
 Unrestricted Search
Search with accelerated step size (cont’d)
• An obvious improvement can be achieved by increasing the
step size gradually until the minimum point is bracketed.
• A simple method consists of doubling the step size as long
as the move results in an improvement of the objective
function.
• One possibility is to reduce the step length after bracketing
the optimum in ( xi-1, xi). By starting either from xi-1 or xi, the
basic procedure can be applied with a reduced step size.
This procedure can be repeated until the bracketed interval
becomes sufficiently small.
 Example
Find the minimum of f = x (x-1.5) by starting from 0.0 with an initial step size of
0.05.
Solution:
The function value at x1 is f1=0.0. If we try to start moving in the negative x
direction, we find that x-2=-0.05 and f-2=0.0775. Since f-2>f1, the assumption of
unimodality indicates that the minimum can not lie toward the left of x-2. Thus,
we start moving in the positive x direction and obtain the following results:
i Value of s xi=x1+s fi = f (xi) Is fi >
fi-1
1 - 0.0 0.0 -
2 0.05 0.05 -0.0725 No
3 0.10 0.10 -0.140 No
4 0.20 0.20 -0.260 No
5 0.40 0.40 -0.440 No
6 0.8 0.80 -0.560 No
7 1.60 1.60 +0.160 Yes
 Example
Solution:
From these results, the optimum point can be seen to be xopt  x6=0.8.

In this case, the points x6 and x7 do not really bracket the minimum point but
provide information about it.

If a better approximation to the minimum is desired, the procedure can be

restarted from x5 with a smaller step size.
 Exhaustive search
• The exhaustive search method can be used to solve
problems where the interval in which the optimum is known
to lie is finite.

• Let xs and xf denote, respectively, the starting and final

points of the interval of uncertainty.

• The exhaustive search method consists of evaluating the

objective function at a predetermined number of equally
spaced points in the interval (xs, xf), and reducing the
interval of uncertainty using the assumption of unimodality.
 Exhaustive search
• Suppose that a function is defined on the interval (xs, xf), and
let it be evaluated at eight equally spaced interior points x1
to x8. The function value appears as:

• Thus, the minimum must lie, according to the assumption of

unimodality, between points x5 and x7. Thus the interval
(x5,x7) can be considered as the final interval of uncertainty.
 Exhaustive search
• In general, if the function is evaluated at n equally spaced points in
the original interval of uncertainty of length L0= xf - xs, and if the
optimum value of the function (among the n function values) turns
out to be at point xj, the final interval of uncertainty is given by:
2
Ln  x j 1  x j 1  L0
n 1
• The final interval of uncertainty obtainable for different number of
trials in the exhaustive search method is given below:

Number 2 3 4 5 6 … n

of trials
Ln/L0 2/3 2/4 2/5 2/6 2/7 … 2/(n+1)
 Exhaustive search
• Since the function is evaluated at all n points simultaneously, this
method can be called a simultaneous search method.

• This method is relatively inefficient compared to the sequential search

methods discussed next, where the information gained from the initial
trials is used in placing the subsequent experiments.
 Example
Find the minimum of f = x(x-1.5) in the interval (0.0,1.0) to within 10
% of the exact value.

Solution: If the middle point of the final interval of uncertainty is

taken as the approximate optimum point, the maximum deviation
could be 1/(n+1) times the initial interval of uncertainty. Thus, to find
the optimum within 10% of the exact value, we should have

1 1
 or n  9
n  1 10
 Example
By taking n = 9, the following function values can be calculated:

i 1 2 3 4 5 6 7 8 9

xi 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

fi=f(xi) -0.14 -0.26 -0.36 -0.44 -0.50 -0.54 -0.56 -0.56 -0.54

Since f7 = f8 , the assumption of unimodality gives the final interval

of uncertainty as L9= (0.7,0.8). By taking the middle point of L9 (i.e.,
0.75) as an approximation to the optimum point, we find that it is in
fact, the true optimum point.
 Dichotomous search
• The exhaustive search method is a simultaneous search method in
which all the experiments are conducted before any judgement is
made regarding the location of the optimum point.

• The dichotomous search method , as well as the Fibonacci and the

golden section methods discussed in subsequent sections, are
sequential search methods in which the result of any experiment
influences the location of the subsequent experiment.

• In the dichotomous search, two experiments are placed as close as

possible at the center of the interval of uncertainty.

• Based on the relative values of the objective function at the two

points, almost half of the interval of uncertainty is eliminated.
 Dichotomous search
• Let the positions of the two
experiments be given by:

L0 
x1  
2 2
L 
x2  0 
2 2

where  is a small positive number

chosen such that the two
experiments give significantly
different results.
 Dichotomous Search
• Then the new interval of uncertainty is given by (L0/2+/2).

• The building block of dichotomous search consists of conducting a pair

of experiments at the center of the current interval of uncertainty.

• The next pair of experiments is, therefore, conducted at the center of

the remaining interval of uncertainty.

• This results in the reduction of the interval of uncertainty by nearly a

factor of two.
 Dichotomous Search
• The intervals of uncertainty at the ends of different pairs of
experiments are given in the following table.
Number of 2 4 6

experiments

Final interval
1  L0     1  L0     
of uncertainty (L0+ )/2     
2 2  2 2 4 2 2

• In general, the final interval of uncertainty after conducting n

experiments (n even) is given by:
L0  1 
Ln  n / 2   1  n / 2 
2  2 
 Dichotomous Search
Example: Find the minimum of f = x(x-1.5) in the interval (0.0,1.0) to within
10% of the exact value.

Solution: The ratio of final to initial intervals of uncertainty is given by:

Ln 1   1 
 n / 2  1  n / 2 
L0 2 L0  2 

where  is a small quantity, say 0.001, and n is the number of experiments.

If the middle point of the final interval is taken as the optimum point, the
requirement can be stated as:
1 Ln 1

2 L0 10
i.e.
1   1  1
 1  n / 2  
2n / 2 L0  2  5
 Dichotomous Search
Solution: Since  = 0.001 and L0 = 1.0, we have

1 1  1  1
 1  n / 2  
2n / 2 1000  2  5
i.e.
999 1 995 999
 or 2 n/2
  5.0
1000 2 n / 2 5000 199

Since n has to be even, this inequality gives the minimum admissable value
of n as 6. The search is made as follows: The first two experiments are
made at:
L0 
x1    0.5  0.0005  0.4995
2 2
L 
x2  0   0.5  0.0005  0.5005
2 2
 Dichotomous Search
with the function values given by:
f1  f ( x1 )  0.4995(1.0005)  0.49975
f 2  f ( x2 )  0.5005(0.9995)  0.50025

Since f2 < f1, the new interval of uncertainty will be (0.4995,1.0). The
second pair of experiments is conducted at :
1.0  0.4995
x3  (0.4995  )  0.0005  0.74925
2
1.0  0.4995
x4  (0.4995  )  0.0005  0.75025
2
which gives the function values as:
f 3  f ( x3 )  0.74925(0.75075)  0.5624994375
f 4  f ( x4 )  0.75025(0.74975)  0.5624994375
 Dichotomous Search
Since f3 > f4 , we delete (0.4995,x3) and obtain the new interval of
uncertainty as:
(x3,1.0)=(0.74925,1.0)
The final set of experiments will be conducted at:

1.0  0.74925
x3  (0.74925  )  0.0005  0.874125
2
1.0  0.74925
x4  (0.74925  )  0.0005  0.875125
2
which gives the function values as:

f 5  f ( x5 )  0.874125(0.625875)  0.5470929844
f 6  f ( x6 )  0.875125(0.624875)  0.5468437342
 Dichotomous Search
Since f5 < f6 , the new interval of uncertainty is given by (x3, x6)
(0.74925,0.875125). The middle point of this interval can be taken as
optimum, and hence:
xopt  0.8121875
f opt  0.5586327148
 Interval halving method
In the interval halving method, exactly one half of the current
interval of uncertainty is deleted in every stage. It requires three
experiments in the first stage and two experiments in each
subsequent stage.

The procedure can be described by the following steps:

1. Divide the initial interval of uncertainty L0 = [a,b] into four equal
parts and label the middle point x0 and the quarter-interval points
x1 and x2.
2. Evaluate the function f(x) at the three interior points to obtain f1 =
f(x1), f0 = f(x0) and f2 = f(x2).
Interval halving method (cont’d)
3. (a) If f1 < f0 < f2 as shown in the figure, delete the interval
( x0,b), label x1 and x0 as the new x0 and b, respectively,
and go to step 4.
Interval halving method (cont’d)
3. (b) If f2 < f0 < f1 as shown in the figure, delete the interval
( a, x0), label x2 and x0 as the new x0 and a, respectively,
and go to step 4.
Interval halving method (cont’d)
3. (c) If f0 < f1 and f0 < f2 as shown in the figure, delete both
the intervals ( a, x1), and ( x2 ,b), label x1 and x2 as the
new a and b, respectively, and go to step 4.
Interval halving method (cont’d)
4. Test whether the new interval of uncertainty, L = b - a,
satisfies the convergence criterion L  ϵ where ϵ
is a small quantity. If the convergence criterion is
satisfied, stop the procedure. Otherwise, set the new L0
= L and go to step 1.

Remarks
1. In this method, the function value at the middle point of
the interval of uncertainty, f0, will be available in all the
stages except the first stage.
Interval halving method (cont’d)
Remarks
2. The interval of uncertainty remaining at the end of n
experiments ( n 3 and odd) is given by
( n 1) / 2
1
Ln    L0
2
 Example
Find the minimum of f = x (x-1.5) in the interval (0.0,1.0) to within 10% of the
exact value.

Solution: If the middle point of the final interval of uncertainty is taken as the
optimum point, the specified accuracy can be achieved if:
( n 1) / 2
1 L 1 L0
Ln  0 or   L0  (E1)
2 10 2 5
Since L0=1, Eq. (E1) gives

1 1
( n 1) / 2
 or 2(n -1)/2  5 (E2)
2 5
 Example
Solution: Since n has to be odd, inequality (E2) gives the minimum permissable
value of n as 7. With this value of n=7, the search is conducted as follows. The
first three experiments are placed at one-fourth points of the interval L0=[a=0,
b=1] as
x1  0.25, f1  0.25(1.25)  0.3125
x0  0.50, f 0  0.50(1.0)  0.5000
x2  0.75, f 0  0.75(0.75)  0.5625

Since f1 > f0 > f2, we delete the interval (a,x0) = (0.0,0.5), label x2 and x0 as the
new x0 and a so that a=0.5, x0=0.75, and b=1.0. By dividing the new interval of
uncertainty, L3=(0.5,1.0) into four equal parts, we obtain:

x1  0.625, f1  0.625(0.875)  0.546875

x0  0.750, f 0  0.750(0.750)  0.562500
x2  0.875, f 2  0.875(0.625)  0.546875
 Example
Solution: Since f1 > f0 and f2 > f0, we delete both the intervals (a,x1) and (x2,b),
and label x1, x0 and x2 as the new a,x0, and b, respectively. Thus, the new
interval of uncertainty will be L5=(0.625,0.875). Next, this interval is divided
into four equal parts to obtain:

x1  0.6875, f1  0.6875(0.8125)  0.558594

x0  0.75, f 0  0.75(0.75)  0.5625
x2  0.8125, f 0  0.8125(0.6875)  0.558594

Again we note that f1 > f0 and f2>f0, and hence we delete both the intervals
(a,x1) and (x2,b) to obtain the new interval of uncertainty as
L7=(0.6875,0.8125). By taking the middle point of this interval (L7) as
optimum, we obtain:
xopt  0.75, and f opt  0.5625
This solution happens to be the exact solution in this case.
 Fibonacci method
As stated earlier, the Fibonacci method can be used to find
the minimum of a function of one variable even if the
function is not continuous. The limitations of the method
are:

• The initial interval of uncertainty, in which the optimum

lies, has to be known.

• The function being optimized has to be unimodal in the

initial interval of uncertainty.
 Fibonacci method
The limitations of the method (cont’d):

• The exact optimum cannot be located in this method. Only

an interval known as the final interval of uncertainty will
be known. The final interval of uncertainty can be made as
small as desired by using more computations.

• The number of function evaluations to be used in the

search or the resolution required has to be specified before
hand.
 Fibonacci method
This method makes use of the sequence of Fibonacci
numbers, {Fn}, for placing the experiments. These numbers
are defined as:

F0  F1  1
Fn  Fn 1  Fn 2 , n  2,3,4,

which yield the sequence 1,1,2,3,5,8,13,21,34,55,89,...

 Fibonacci method
Procedure:
Let L0 be the initial interval of uncertainty defined by a x
 b and n be the total number of experiments to be
conducted. Define
Fn2
L 
*
2 L0
Fn

and place the first two experiments at points x1 and x2,

which are located at a distance of L2* from each end of L0.
 Fibonacci method
Procedure:
This gives
Fn  2
x1  a  L*2  a  L0
Fn
Fn  2 Fn 1
x2  b  L  b 
*
2 L0  a  L0
Fn Fn
Discard part of the interval by using the unimodality
assumption. Then there remains a smaller interval of
uncertainty L2 given by:
 F  F
L2  L0  L*2  L0 1  n2   n1 L0
 Fn  Fn
 Fibonacci method
Procedure:
The only experiment left in will be at a distance of
Fn2 F
L*2  L0  n2 L2
Fn Fn1
from one end and
Fn3 F
L2  L*2  L0  n3 L2
Fn Fn1
from the other end. Now place the third experiment in the interval
L2 so that the current two experiments are located at a distance of:
Fn 3 F
L*3  L0  n 3 L2
Fn Fn 1
 Fibonacci method
Procedure:
• This process of discarding a certain interval and placing a new
experiment in the remaining interval can be continued, so that the
location of the jth experiment and the interval of uncertainty at the end
of j experiments are, respectively, given by:

Fn  j
L 
*
j L j 1
Fn ( j 2)
Fn ( j 1)
Lj  L0
Fn
 Fibonacci method
Procedure:

• The ratio of the interval of uncertainty remaining after conducting

j of the n predetermined experiments to the initial interval of
uncertainty becomes:

Lj Fn ( j 1)

L0 Fn

and for j = n, we obtain

Ln F1 1
 
L0 Fn Fn
 Fibonacci method
• The ratio Ln/L0 will permit us to determine n, the required number
of experiments, to achieve any desired accuracy in locating the
optimum point.Table gives the reduction ratio in the interval of
uncertainty obtainable for different number of experiments.
 Fibonacci method
Position of the final experiment:
• In this method, the last experiment has to be placed with some
care. Equation
Fn  j
L 
*
j L j 1
Fn ( j  2 )

gives
L*n F 1
 0  for all n
Ln 1 F2 2

• Thus, after conducting n-1 experiments and discarding the

appropriate interval in each step, the remaining interval will
contain one experiment precisely at its middle point.
 Fibonacci method
Position of the final experiment:
• However, the final experiment, namely, the nth
experiment, is also to be placed at the center of the
present interval of uncertainty.
• That is, the position of the nth experiment will be the
same as that of ( n-1)th experiment, and this is true for
whatever value we choose for n.
• Since no new information can be gained by placing the
nth experiment exactly at the same location as that of
the (n-1)th experiment, we place the nth experiment
very close to the remaining valid experiment, as in the
case of the dichotomous search method.
 Fibonacci method
Example:
Minimize
f(x)=0.65-[0.75/(1+x2)]-0.65 x tan-1(1/x) in the interval [0,3]
by the Fibonacci method using n=6.
Solution: Here n=6 and L0=3.0, which yield:
Fn  2 5
L2 *  L0  (3.0)  1.153846
Fn 13
Thus, the positions of the first two experiments are given by
x1=1.153846 and x2=3.0-1.153846=1.846154 with f1=f(x1)=-
0.207270 and f2=f(x2)=-0.115843. Since f1 is less than f2, we
can delete the interval [x2,3] by using the unimodality
assumption.
 Fibonacci method
Solution:
 Fibonacci method
Solution:
The third experiment is placed at x3=0+ (x2-x1)=1.846154-
1.153846=0.692308, with the corresponding function value of f3=-
0.291364. Since f1 is greater than f3, we can delete the interval [x1,x2]
 Fibonacci method
Solution:
The next experiment is located at x4=0+ (x1-x3)=1.153846-
0.692308=0.461538, with f4=-0.309811. Noting that f4 is less than f3, we
can delete the interval [x3,x1]
 Fibonacci method
Solution:
The location of the next experiment can be obtained as x5=0+ (x3-
x4)=0.692308-0.461538=0.230770, with the corresponding objective
function value of f5=-0.263678. Since f4 is less than f3, we can delete the
interval [0,x5]
 Fibonacci method
Solution:
The final experiment is positioned at x6=x5+ (x3-x4)=0.230770+(0.692308-
0.461538)=0.461540 with f6=-0.309810. (Note that, theoretically, the
value of x6 should be same as that of x4; however,it is slightly different
from x4 due to the round off error). Since f6 > f4 , we delete the interval
[x6, x3] and obtain the final interval of uncertainty as L6 = [x5,
x6]=[0.230770,0.461540].
 Fibonacci method
Solution:
The ratio of the final to the initial interval of uncertainty is
L6 0.461540  0.230770
  0.076923
L0 3.0

This value can be compared with

Ln F1 1
 
L0 Fn Fn

which states that if n experiments (n=6) are planned, a resolution

no finer than 1/Fn= 1/F6=1/13=0.076923 can be expected from
the method.
 Golden Section Method
• The golden section method is same as the Fibonacci
method except that in the Fibonacci method, the total
number of experiments to be conducted has to be
specified before beginning the calculation, whereas this
is not required in the golden section method.
 Golden Section Method
• In the Fibonacci method, the location of the first two
experiments is determined by the total number of
experiments, n.

• In the golden section method, we start with the

assumption that we are going to conduct a large number
of experiments.

• Of course, the total number of experiments can be

decided during the computation.
 Golden Section Method
• The intervals of uncertainty remaining at the end of different number
of experiments can be computed as follows:
F
L2  lim N 1 L0
N  F
N

FN  2 F F
L3  lim L0  lim N  2 N 1 L0
N  F N  F
N N 1 FN
2
 FN 1 
 lim   L0
N 
 FN 

• This result can be generalized to obtain

k 1
 FN 1 
Lk  lim   L0
N 
 FN 
 Golden Section Method
Using the relation:
FN  FN 1  FN 2
We obtain, after dividing both sides by FN-1,
FN FN  2
 1
FN 1 FN 1

By defining a ratio  as
 FN 
  lim  
N  F
 N 1 
 Golden Section Method
The equation
FN FN  2
 1
FN 1 FN 1
can be expressed as:
1
 1

that is:
 2   1  0
 Golden Section Method
This gives the root =1.618, and hence the equation
k 1
F 
Lk  lim  N 1  L0
N 
 FN 
yields: 1
k 1

Lk    L0  (0.618) k 1 L0
 
2
F 
In the equation L3  lim  N 1  L0
N 
 FN 

the ratios FN-2/FN-1 and FN-1/FN have been taken to be

same for large values of N. The validity of this
assumption can be seen from the table:
Value of N 2 3 4 5 6 7 8 9 10 

Ratio FN-1/FN 0.5 0.667 0.6 0.625 0.6156 0.619 0.6177 0.6181 0.6184 0.618
 Golden Section Method
The ratio  has a historical background. Ancient Greek architects
believed that a building having the sides d and b satisfying the
relation
d b d
 
d b
will be having the most pleasing properties. It is also found in
Euclid’s geometry that the division of a line segment into two
unequal parts so that the ratio of the whole to the larger part is equal
to the ratio of the larger to the smaller, being known as the golden
section, or golden mean-thus the term golden section method.
Comparison of elimination methods
• The efficiency of an elimination method can be measured in terms of the ratio of the
final and the initial intervals of uncertainty, Ln/L0
• The values of this ratio achieved in various methods for a specified number of
experiments (n=5 and n=10) are compared in the Table below:

• It can be seen that the Fibonacci method is the most efficient method, followed by the
golden section method, in reducing the interval of uncertainty.
Comparison of elimination methods
• A similar observation can be made by considering the number of
experiments (or function evaluations) needed to achieve a specified
accuracy in various methods.

• The results are compared in the Table below for maximum permissable
errors of 0.1 and 0.01.

• It can be seen that to achieve any specified accuracy, the Fibonacci method
requires the least number of experiments, followed by the golden section
method.
 Interpolation methods
• The interpolation methods were originally developed as one
dimensional searches within multivariable optimization
techniques, and are generally more efficient than Fibonacci-type
approaches.

• The aim of all the one-dimensional minimization methods is to

find *, the smallest nonnegative value of , for which the
function
f ( )  f ( X  S )

attains a local minimum.

 Interpolation methods
• Hence if the original function f (X) is expressible as an explicit
function of xi (i=1,2,…,n), we can readily write the expression for
f () = f (X + S ) for any specified vector S, set

df
( )  0
d

and solve the above equation to find * in terms of X and S.

• However, in many practical problems, the function f ( ) can not

be expressed explicitly in terms of . In such cases, the
interpolation methods can be used to find the value of *.
 Quadratic Interpolation Method
• The quadratic interpolation method uses the function values only; hence it
is useful to find the minimizing step (*) of functions f (X) for which the
partial derivatives with respect to the variables xi are not available or
difficult to compute.
• This method finds the minimizing step length * in three stages:
– In the first stage, the S vector is normalized so that a step length of  = 1 is
acceptable.
– In the second stage, the function f () is approximated by a quadratic function
h() and the minimum, ~ *, of h() is found. If this is not sufficiently close to
the true minimum *, a third stage is used.
– In this stage, a new quadratic function h( )  a  b  c2 is used to approximate
f (), and a new value of ~ * is found. This procedure is continued until a~ *that
is sufficiently close to * is found.
 Quadratic Interpolation Method
• Stage 1: In this stage, the S vector is normalized as follows: Find
=max|si|, where si is the ith component of S and divide each
component of S by . Another method of normalization is to find
=(s12+ s22+ …+sn2 )1/2 and divide each component of S by .

• Stage 2: Let

h( )  a  b  c2
be the quadratic function used for approximating the function f (). It
is worth noting at this point that a quadratic is the lowest-order
polynomial for which a finite minimum can exist.
 Quadratic Interpolation Method
• Stage 2 cont’d: Let
dh
 b  2c  0
d
that is,
~ b
*  
2c
The sufficiency condition for the minimum of h () is that

d 2h
0
d ~*
2

that is,
c>0
 Quadratic Interpolation Method
Stage 2 cont’d:
• To evaluate the constants a, b, and c in the Equation

h( )  a  b  c2

we need to evaluate the function f () at three points.

• Let =A, =B, and =C be the points at which the function f () is
evaluated and let fA, fB and fC be the corresponding function values,
that is,
f A  a  bA  cA 2

f B  a  bB  cB 2
f C  a  bC  cC 2
 Quadratic Interpolation Method
Stage 2 cont’d:
• The solution of

f A  a  bA  cA2
f B  a  bB  cB 2
f C  a  bC  cC 2
gives
f A BC (C  B)  f B CA( A  C )  f C AB( B  A)
a
( A  B)( B  C )(C  A)
f A ( B 2  C 2 )  f B (C 2  A2 )  f C ( A2  B 2 )
b
( A  B)( B  C )(C  A)
f ( B  C )  f B (C  A)  f C ( A  B)
c A
( A  B)( B  C )(C  A)
 Quadratic Interpolation Method
Stage 2 cont’d:
• From equations
f A ( B 2  C 2 )  f B (C 2  A2 )  f C ( A2  B 2 )
~ b b
*   ( A  B)( B  C )(C  A)
2c f ( B  C )  f B (C  A)  f C ( A  B)
c A
( A  B)( B  C )(C  A)

the minimum of h () can be obtained as:

~* b f A ( B 2  C 2 )  f B (C 2  A2 )  f C ( A2  B 2 )
  
2c 2[ f A ( B  C )  f B (C  A)  f C ( A  B)]

provided that c is positive.

 Quadratic Interpolation Method
Stage 2 cont’d:
• To start with, for simplicity, the points, A, B and C can be chosen as 0, t, and 2t,
respectively, where t is a preselected trial step length.
• By this procedure, we can save one function evaluation since f A=f (=0) is generally
known from the previous iteration (of a multivariable search).
• For this case, the equations reduce to:

a  fA
4 f  3 f A  fC 4 f B  3 f A  fC
b B *  t
2t 4 f B  2 fC  2 f A
f  f A  2 fB
c C
2t 2

provided that
fC  f A  2 f B
c 2
0
2t
 Quadratic Interpolation Method
Stage 2 cont’d:
• The inequality
fC  f A  2 f B
c 2
0
2t
can be satisfied if
f A  fC
 fB
2
i.e., the function value fB should be
smaller than the average value of
fA and fC as shown in figure.
 Quadratic Interpolation Method
Stage 2 cont’d:
• The following procedure can be used not only to satisfy the inequality
f A  fC
 fB
2
~
but also to ensure that the minimum ~ * lies in the interval 0 <  *< 2t.
1. Assuming that fA = f (=0) and the initial step size t0 are known, evaluate
the function f at =t0 and obtain f1=f (=t0 ).
 Quadratic Interpolation Method
Stage 2 cont’d:
f A  fC
 fB
2
~
*
 Quadratic Interpolation Method
Stage 2 cont’d:
2. If f1 > fA is realized as shown in figure, set fC = f1 and evaluate the function f at
~
= t0 /2 and  * using the equation
4 f B  3 f A  fC
*  t
4 f B  2 fC  2 f A

with t= t0 / 2.
 Quadratic Interpolation Method
Stage 2 cont’d:
3. If f1 ≤ fA is realized as shown in figures, set fB = f1 and evaluate the function f at
= 2 t0 to find f2=f (= 2 t0 ). This may result in any of the equations shown
in the figure.
 Quadratic Interpolation Method
Stage 2 cont’d:
4. If f2 turns out to be greater than f1 as shown in the figures, set fC= f2 and
~
compute  * according to the equation below with t= t0.
4 f B  3 f A  fC
*  t
4 f B  2 fC  2 f A

5. If f2 turns out to be smaller than f1, set new f1= f2 and t= 2t0 and repeat steps 2
to 4 until we are able to find ~ *.
 Quadratic Interpolation Method
Stage 3: The ~ * found in Stage 2 is the minimum of the approximating
~
quadratic h() and we have to make sure that this  * is sufficiently
~
close to the true minimum * of f () before taking * =  * . Several
tests are possible to ascertain this.
~ ~ ~
• One possible test is to compare f ( *) with h( *) and consider  * a
sufficiently close good approximation if they differ not more than by a
small amount. This criterion can be stated as:
~ ~
h( *)  f ( *)
~  1
f ( *)
Quadratic Interpolation Method
Stage 3 cont’d:

Another possible test is to examine whether df /d is close to zero at ~ *.

Since the derivatives of f are not used in this method, we can use a finite-
difference formula for df /d and use the criterion:

~ ~ ~ ~
f ( *  *)  f ( *  *)
~  2
2 *

to stop the procedure. 1 and 2 are small numbers to be specified

depending on the accuracy desired.
 Quadratic Interpolation Method
Stage 3 cont’d: If the convergence criteria stated in equations
~ ~ ~ ~ ~ ~
h( *)  f ( *) f (  *   *)  f (  *   *)
~   ~  2
f ( *)
1
2 *
are not satisfied, a new quadratic function

h( )  a  b  c2

is used to approximate the function f ().

• To evaluate the constants a’, b’ and c’, the three best function values of
the current f A=f (=0), f B=f (=t0), f C=f (=2t0), and ~f  f (  ~*)
are to be used.
• This process of trying to fit another polynomial to obtain a better
~
approximation to  * is known as refitting the polynomial.
Quadratic Interpolation Method
Stage 3 cont’d: For refitting the quadratic, we consider all possible situations and select
~
the best three points of the present A, B, and C, and  * . There are four possibilities. The
best three points to be used in refitting in each case are given in the table.
 Quadratic Interpolation Method
Stage 3 cont’d:
 Quadratic Interpolation Method
~
Stage 3 cont’d: A new value of  * is computed by using the general
formula:
~* b f A ( B 2  C 2 )  f B (C 2  A2 )  f C ( A2  B 2 )
  
2c 2[ f A ( B  C )  f B (C  A)  f C ( A  B)]

~
If this  * does not satisfy the convergence criteria stated in
~ ~ ~ ~ ~ ~
h( *)  f ( *) f ( *  *)  f ( *  *)
~  1 ~  2
f ( *) 2 *

A new quadratic has to be refitted according to the scheme outlined in the

table.
 Cubic Interpolation Method
• The cubic interpolation method finds the minimizing step length ~ *
in four stages. It makes use of the derivative of the function f:
df d
f ( )   f ( X   S )  S T f ( X   S )
d d

• The first stage normalizes the S vector so that a step size =1 is
acceptable.

• The second stage establishes bounds on *, and the third stage finds
the value of * by approximating f () by a cubic polynomial h ().

~
• If the  * found in stage 3 does not satisfy the prescribed convergence
criteria, the cubic polynomial is refitted in the fourth stage.
 Cubic Interpolation Method
~
• *
Stage 1: Calculate =max|si|, where |si | is the absolute value
of the ith component of S and divide each component of S by .
Another method of normalization is to find =(s12+ s22+ …+sn2 )1/2 .
and divide each component of S by .
• Stage 2:To establish lower and upper bounds on the optimal
step size ~ *, we need to find two points A and B at which the
slope df / d has different signs. We know that at  = 0,
df
 ST f (X)  0
d  0

since S is presumed to be a direction of descent.(In this case,

the direction between the steepest descent and S will be less
than 90.
 Cubic Interpolation Method
• Stage 2 cont’d: Hence to start with, we can take A=0 and try to
find a point =B at which the slope df / d is positive. Point B
can be taken as the first value out of t0, 2t0, 4t0, 8t0…at which f’
is nonnegative, where t0 is a preassigned initial step size. It then
follows that * is bounded in the interval A ≤ * ≤ B.
 Cubic Interpolation Method
• Stage 3: If the cubic equation
h( )  a  b  c2  d3
is used to approximate the function f () between points A and B,
we need to find the values f A=f (=A), f A’=df/d  (=A), f B=f
(=B), f B’=df /d (=B) in order to evaluate the constants, a,b,c,
and d in the above equation. By assuming that A ≠0, we can derive
a general formula for ~ *. From the above equation, we have:
f A  a  bA  cA2  dA3
f B  a  bB  cB 2  dB 3

f A  b  2cA  3dA2

f B  b  2cB  3dB 2
 Cubic Interpolation Method
• Stage 3 cont’d: The equation
f A  a  bA  cA2  dA3
f B  a  bB  cB 2  dB 3

f A  b  2cA  3dA2

f B  b  2cB  3dB 2

can be solved to find the constants as:

1
a  f A  bA  cA2  dA3 with b 2
( B 2
f 
A  A 2
f B  2 ABZ )
(A - B)
1 1
c 2
[( A  B) Z  Bf A  Af B ] and d 2
(2Z  f A  f B ) where
(A - B) 3(A - B)
3( f A  f B )
Z  f A  f B
B A
 Cubic Interpolation Method
• Stage 3 cont’d: The necessary condition for the minimum
of h() given by the equation
h( )  a  b  c2  d3
is that
dh
 b  2c  3d2  0
d
that is
~  c  (c 2  3bd )1/ 2
* 
3d
 Cubic Interpolation Method
• Stage 3 cont’d: The application of the sufficiency
condition for the minimum of h() leads to the relation:
d 2h ~
~  2c  6 d *0
d2 *
 Cubic Interpolation Method
• Stage 3 cont’d: By substituting the expressions for b,c, and d given by
the equations
1 1
b 2
( B 2 f A  A2 f B  2 ABZ ) and c [( A  B) Z  Bf A  Af B ]
(A - B) (A - B) 2
1
d (2Z  f A  f B )
3(A - B) 2

into
~  c  (c 2  3bd )1/ 2
* 
3d

d 2h ~
~  2c  6 d *0
d2 *
 Cubic Interpolation Method
Stage 3 cont’d: We obtain:

~ f A  Z  Q
*  A ( B  A)
f A  f B  2Z
where
Q  ( Z 2  f A f B )1/ 2
2( B  A)( 2Z  f A  f B )( f A  Z  Q)
 2( B  A)( f A  Zf B  3Zf A  2Z 2 )
2

 2( B  A) f A f B  0
 Cubic Interpolation Method
Stage 3 cont’d: By specializing all the equations below:
1
a  f A  bA  cA2  dA3 with b 2
( B 2
f 
A  A 2
f B  2 ABZ )
(A - B)
1 1
c 2
[( A  B) Z  Bf A  Af B ] and d 2
(2Z  f A  f B ) where
(A - B) 3(A - B)
3( f A  f B )
Z  f A  f B
B A ~ f A  Z  Q
*  A ( B  A)
dh f A  f B  2Z
 b  2c  3d2  0
d where
that is Q  ( Z 2  f A f B )1/ 2
~  c  (c 2  3bd )1/ 2 2( B  A)( 2Z  f A  f B )( f A  Z  Q)
* 
3d
 2( B  A)( f A  Zf B  3Zf A  2Z 2 )
2

d 2h ~  2( B  A) f A f B  0
~  2c  6 d *0
d2 *
 Cubic Interpolation Method
Stage 3 cont’d:
For the case where A=0, we obtain:

1 1
a  fA b  f A c ( Z  f A ) d 2
(2Z  f A  f B )
B 3B
~ f A  Z  Q
*  B Q  ( Z 2  f A f B )1/ 2  0
f A  f B  2Z
3( f A  f B )
Z  f A  f B
B
 Cubic Interpolation Method
Stage 3 cont’d: The two values of ~ *in the equations:

~ f A  Z  Q
*  A ( B  A)
f A  f B  2 Z
~ f A  Z  Q
*  B
f A  f B  2 Z
correspond to the two possibilities for the vanishing of h’() [i.e., at a maximum of
h() and at a minimum]. To avoid imaginary values of Q, we should ensure the
satisfaction of the condition

Z 2  f A f B  0
in equation
Q  (Z 2  f A f B )1/ 2
 Cubic Interpolation Method
Stage 3 cont’d:
This inequality is satisfied automatically since A and B are
selected such that f’A <0 and f’B ≥0. Furthermore, the sufficiency
condition when A=0 requires that Q > 0, which is already
~
satisfied. Now, we compute  * using

f A  Z  Q
*  B
f A  f B  2Z
and proceed to the next stage.
 Cubic Interpolation Method
~
Stage 4: The value of  * found in stage 3 is the true minimum of h() and may not
be close to the minimum of f (). Hence the following convergence criteria
can be used before choosing *  ~ *

~ ~
h( *)  f ( *)
~  1
f ( *)
df
 S T f  2
d
~
~ *
*
where 1 and 2 are small numbers whose values depend on the
accuracy desired.
 Cubic Interpolation Method
Stage 4: The criterion
df
 ST f  2
d
~
~ *
*
can be stated in nondimensional form as

Sf
 2
S f ~
*

If the criteria in the above equation and the equation

~ ~
h( *)  f ( *)
~  1
f ( *)
are not satisfied, a new cubic equation can be used to approximate f () as follows:

h( )  a  b  c2  d 3

 Cubic Interpolation Method
Stage 4 cont’d: The constants a’, b’, c’ and d’ can be evaluated by using the function
and derivative values at the best two points out of the three points currently available:
A, B, and ~ *. Now the general formula given by the equation:

~ f A  Z  Q
*  A ( B  A)
f A  f B  2Z
is to be used for finding the optimal step size ~ *. If f (~*)  0 , the new points A and
~ ~
B are taken as  * and B, respectively; otherwise if f ( *)  0 , the new points A and B
are taken as A and ~ * and equations

~ ~
h( *)  f ( *) Sf
 1 and  2
~
f ( *) S f ~
*

are again used to test the convergence of ~ * . If convergence is achieved, ~ * is

taken as * and the procedure is stopped. Otherwise, the entire procedure is repeated
until the desired convergence is achieved.
 Example
Find the minimum of

f  5  53  20  5
By the cubic interpolation method.

Solution: Since this problem has not arisen during a multivariable

optimization process, we can skip stage 1. We take A=0, and find that:
df
(  A  0)  54  152  20  20  0
d  0

To find B at which df/d is nonnegative, we start with t0=0.4 and evaluate the
derivative at t0, 2t0, 4t0,……This gives
 Example
This gives:
f (t0  0.4)  5(0.4) 4  15(0.4) 2  20.0  22.272
f (2t0  0.8)  5(0.8) 4  15(0.8) 2  20.0  27.552
f (4t0  1.6)  5(1.6) 4  15(1.6) 2  20.0  25.632
f (8t0  3.2)  5(3.2) 4  15(3.2) 2  20.0  350.688

Thus, we find that:

A=0.0, fA = 5.0, f’A = -20.0,
B=3.2, fB = 113.0, f’B = 350.688,
A < * <B
 Example
Iteration I: To find the value of ~ * , and to test the convergence criteria, we first
compute Z and Q as:

3(5.0  113.0)
Z  20.0  350.688  229.588
3 .2
Q  [ 229.5882  ( 20.0)(350.688)]1/ 2  244.0
Hence
~  - 20.0  229.588  244.0 
 *  3.2   1.88 or - 0.1396
 - 20.0  350.688  459.176 
By discarding the negative value, we have :
~
 *  1.84
 Example
Convergence criterion: If ~ * is close to the true minimum *, then
~ ~
f ( *)  df ( *) / d
should be approximately zero. Since
f   54  152  20
~

f ( *)  5(1.84) 4  15(1.84) 2  20  13.0
~
Since this is not small, we go to the next iteration or refitting. As f ( *)  0 ,
we take A= ~ * and
~
f A  f ( *)  (1.84) 5  5(1.84) 3  20(1.84)  5  41.70
Thus,
A  1.84, f A  41.70, f A  13.0
B  3.2, f B  113.0, f B  350.688
~
A  * B
 Example
Iteration 2

3(41.7  113.0)
Z  13.00  350.688  3.312
(3.20  1.84)
Q  [( 3.312) 2  (13.0)(350.688)]1/ 2  67.5
Hence,
~ - 13.0 - 3.312  67.5
 *  1.84  (3.2  1.84)  2.05
- 13.0  350.688  6.624
Convergence Criteria
~
f ( *)  5.0(2.05) 4  15.0(2.05) 2  20.0  5.35
~ ~
Since this value is large, we go to the next iteration with B   *  2.05 [as f ( *)  0] and
f B  (2.05) 5  5.0(2.05) 3  20.0(2.05)  5.0  42.90
Thus,
A  1.84, f A  41.70, f A  13.00
B  2.05, f B  42.90, f B  5.35
A  *  B
 Example
Iteration 3 3(41.7  42.90)
Z  13  5.35  9.49
(2.05  1.84)
Q  [(9.49) 2  (13.0)(5.35)]1/ 2  12.61
Hence
~ - 13.0  9.49  12.61
 *  1.84  (2.05  1.84)  2.0086
- 13.0  5.35  18.98
Convergence criterion:
~
f ( *)  5.0(2.0086) 4  15.0(2.0086) 2  20.0  0.855

Assuming that this value is close to zero, we can stop the iterative process and take
~
*   *  2.0086
 Direct root methods
The necessary condition for f () to have a minimum of * is that

f ( )  0
Three root finding methods will be considered here:

• Newton method

• Quasi-Newton method

• Secant methods
 Newton method
Consider the quadratic approximation of the function f () at = i
using the Taylor’s series expansion:
1
f ( )  f (i )  f (i )(  i )  f (i )(  i ) 2
2

By setting the derivative of this equation equal to zero for the minimum
of f (), we obtain:

f ( )  f (i )  f (i )(  i )  0

If i denotes an approximation to the minimum of f (), the above

equation can be rearranged to obtain an improved approximation as:
 Newton method
f (i )
i 1  i 
f (i )
Thus, the Newton method is equivalent to using a quadratic
approximation for the function f () and applying the necessary
conditions.

The iterative process given by the above equation can be assumed to

have converged when the derivative, f’(i+1) is close to zero:

f (i 1 )  
where  is a small quantity
 Newton method
• FIGURE 5.18a sayfa 318
 Newton method
• If the starting point for the iterative process is not close to the true
solution *, the Newton iterative process may diverge as illustrated:
 Newton method
Remarks:

• The Newton method was originally developed by Newton for solving

nonlinear equations and later refined by Raphson, and hence the method is
also known as Newton-Raphson method in the literature of numerical
analysis.

• The method requires both the first- and second-order derivatives of f ().
 Example
Find the minimum of the function:
0.75 1 1
f ( )  0.65   0.65 tan
1  2 
Using the Newton-Raphson method with the starting point 1=0.1. Use =0.01 in
the equation
f (i 1 )  
where  is a small quantity
for checking the convergence.
 Example
Solution: The first and second derivatives of the function f () are given
by :
1.5 0.65 1 1
f ( )    0.65 tan
(1  2 ) 2 1  2 
1.5(1  32 ) 0.65(1  2 ) 0.65 2.8  3.22
f ( )    
(1   )
2 3
(1   )
2 2
1 2
(1  2 ) 3

Iteration 1
1=0.1, f (1) = -0.188197, f’ (1) = -0.744832, f’’ (1)=2.68659
f (1 )
2  1   0.377241
f (1 )
Convergence check: | f (2)| =|-0.138230| > 
 Example
Solution cont’d:
Iteration 2
f (2 ) = -0.303279, f’ (2) = -0.138230, f’’ (2) = 1.57296
f (2 )
3  2   0.465119
f (2 )
Convergence check: | f’(3)| =|-0.0179078| > 
Iteration 3
f (3 ) = -0.309881, f’ (3) = -0.0179078, f’’ (3) = 1.17126
f (3 )
4  3   0.480409
f (3 )
Convergence check: | f’(4)| =|-0.0005033| < 
Since the process has converged, the optimum solution is taken as *
4=0.480409
 Quasi-Newton Method
If the function minimized f () is not available in closed form or is
difficult to differentiate, the derivatives f’ () and f’’ () in the equation
f (i )
i 1  i 
f (i )

can be approximated by the finite difference formula as:

f (i   )  f (i   )
f (i ) 
2
f (i   )  2 f (i )  f (i   )

f (i ) 
2

where  is a small step size.

 Quasi-Newton Method
Substitution of:
f (i   )  f (i   )
f (i ) 
2
f (i   )  2 f (i )  f (i   )
f (i ) 
2

f (i )
into i 1  i 
f (i )

leads to
[ f (i   )  f (i   )]
i 1  i 
2[ f (i   )  2 f (i )  f (i   )]
 Quasi-Newton Method
This iterative process is known as the quasi-Newton method. To test the
convergence of the iterative process, the following criterion can be used:

f (i 1   )  f (i 1   )
f (i 1 )  
2
where a central difference formula has been used for evaluating the
derivative of f and  is a small quantity.

Remarks:
The equation
[ f (i   )  f (i   )]
i 1  i 
2[ f (i   )  2 f (i )  f (i   )]

requires the evaluation of the function at the points i+ and i - in
addition to i in each iteration.
 Example
Find the minimum of the function
0.75 1 1
f ( )  0.65   0.65 tan
1  2 

using the quasi-Newton method with the starting point 1=0.1 and the
step size =0.01 in central difference formulas. Use =0.01 in equation

f (i 1   )  f (i 1   )
f (i 1 )  
2

for checking the convergence.

 Example
Solution: Iteration I

  0.1,   0.01, f1  f (1 )  0.188197

f1  f (1   )  0.195512, f1  f (1   )  0.180615
 ( f1  f1 )
2  1   
 0.377882
2( f1  2 f1  f1 )
Convergence check :
f 2  f 2
f (2 )   0.137300  
2
 Example
Solution: Iteration 2

f 2  f (2 )  0.303368, f 2  f (2   )  0.304662

f 2  f (2   )  0.301916,
 ( f 2  f 2 )
3  2   
 0.465390
2( f 2  2 f 2  f 2 )
Convergence check :
f 3  f 3
f (3 )   0.017700  
2
 Example
Solution: Iteration 3

f 3  f (3 )  0.309885, f 3  f (3   )  0.310004

f 3  f (3   )  0.309650,
 ( f 3  f 3 )
4  3   0.480600
2( f 3  2 f 3  f 3 )
Convergence check :
f 4  f 4
f (4 )   0.000350  
2
Since the process has converged, we take the optimum solution as *  4  0.480600
 Secant method
• The secant method uses an equation similar to equation:

f ( )  f (i )  f (i )(  i )  0

as:
f ( )  f (i )  s(  i )  0

where s is the slope of the line connecting the two points (A, f’(A)) and (B,
f’(B)), where A and B denote two different approximations to the correct
solution, *. The slope s can be expressed as:

f ( B)  f ( A)
s
B A
 Secant method
The equation:

f ( )  f (i )  s(  i )  0

approximates the function f’() between A and B as a linear equation
(secant), and hence the solution of the above equation gives the new
approximation to the root of the f’() as:

f (i ) f ( A)( B  A)
i 1  i   A
s f ( B)  f ( A)

The iterative process given by the above equation is known as the secant
method. Since the secant approaches the second derivative of f () at A as
B approaches A, the secant method can also be considered as a quasi-
Newton method.
Secant method
 Secant method
• It can also be considered as a form of elimination technique since part of
the interval, (A,İ+1) in the figure is eliminated in every iteration.

The iterative process can be implemented by using the following step-by-

step procedure:

1. Set 1=A=0 and evaluate f’(A). The value of f’(A) will be negative.
Assume an initial trial step length t0.
2. Evaluate f’(t0).
3. If f’(t0)<0, set A= i=t0, f’(A)= f’(t0), new t0=2t0, and go to step 2.
4. If f’(t0)≥0, set B= t0, f’(B)= f’(t0), and go to step 5.
5. Find the new approximate solution of the problem as:

f ( A)( B  A)
i 1  A 
f ( B)  f ( A)
 Secant method
6. Test for convergence:

f (i 1 )  
where  is a small quantity. If the above equation is satisfied, take * i+1
and stop the procedure. Otherwise, go to step 7.

7. If f’(İ+1) ≥ 0, set new B= İ+1, f’(B) =f’(İ+1), i=i+1, and go to step 5.

8. If f’(İ+1) < 0, set new A= İ+1, f’(A) =f’(İ+1), i=i+1, and go to step 5.
 Secant method
Remarks:

1. The secant method is identical to assuming a linear equation for

f’(). This implies that the original function, f(), is approximated
by a quadratic equation.

2. In some cases, we may encounter a situation where the function

f’() varies very slowly with . This situation can be identified by
noticing that the point B remains unaltered for several consecutive
refits. Once such a situation is suspected, the convergence
process can be improved by taking the next value of i+1 as
(A+B)/2 instead of finding its value from
f ( A)( B  A)
i 1  A 
f ( B)  f ( A)
 Secant method
Example
0.75 1 1
Find the minimum of the function f ( )  0.65   0.65 tan
1  2 
using the secant method with an initial step size of t0=0.1, 1=0.0, and
=0.01.
Solution
1=A=0.0, t0=0.1, f’(A)=-1.02102, B=A+t0=0.1, f’(B)=-0.744832
Since f’(B)<0, we set new A=0.1, f’(A)=-0.744832, t0=2(0.1)=0.2,
B= 1+t0=0.2, and compute f’(B)=-0.490343. Since f’(B)<0, we set new
A=0.2, f’(A)=-0.490343, t0=2(0.2)=0.4, B= 1+t0=0.4, and compute f’(B)=-
0.103652. Since f’(B)<0, we set new A=0.4, f’(A)=-0.103652
t0=2(0.4)=0.8, B= 1+t0=0.8, and compute f’(B)=-0.180800. Since, f’(B)>0,
we proceed to find 2
 Secant method
Iteration 1
Since A=1=0.4, f’(A)=-0.103652, B=0.8, f’(B)= 0.180800, we compute
f ( A)( B  A)
2  A   0.545757
f ( B)  f ( A)
Convergence check:
f (2 )   0.0105789  
Iteration 2
Since f’(2)=+0.0105789 > 0, we set new A=0.4,f’(A)=-0.103652, B=
2=0.545757, f’(B)=f’(2)=0.0105789, and compute
f ( A)( B  A)
3  A   0.490632
f ( B)  f ( A)
Convergence check:
f (3 )   0.00151235  
Since the process has converged, the optimum solution is given by
*3=0.490632
 Practical Considerations
Sometimes, the Direct Root Methods such as the
Newton, Quasi-Newton and the Secant method or the
interpolation methods such as the quadratic and the
cubic interpolation methods may be:
• very slow to converge,
• may diverge
• may predict the minimum of the function f() outside the
initial interval of uncertainty, especially when the
interpolating polynomial is not representative of the
variation of the function being minimized.
In such cases, we can use the Fibonacci or the golden
section method to find the minimum.
 Practical Considerations
In some problems, it might prove to be more
efficient to combine several techniques. For
example:
• The unrestricted search with an accelerated step
size can be used to bracket the minimum and
then the Fibonacci or the golden section method
can be used to find the optimum point.
• In some cases, the Fibonacci or the golden
section method can be used in conjunction with
an interpolation method.
 Comparison of methods
• The Fibonacci method is the most efficient elimination technique in finding
the minimum of a function if the initial interval of uncertainty is known.

• In the absence of the initial interval of uncertainty, the quadratic

interpolation method or the quasi-Newton method is expected to be more
efficient when the derivatives of the function are not available.

• When the first derivatives of the function being minimized are available,
the cubic interpolation method or the secant method are expected to be
very efficient.

• On the other hand, if both the first and the second derivatives of the
function are available, the Newton method will be the most efficient one in
finding the optimal step length, *.