Numerical Methods For American Options

S.C.Benbow

August 22, 2005

 Abstract

The problem of solving the Black-Scholes equation for the valuation of Amer-
ican options is tackled using a Crank-Nicolson finite difference method for-
mulated in a Lagrangian frame.

We firstly introduce a transformation to convert the Black-Scholes equation

into a dimensionless constant coefficient forward equation. We then formu-
late the equation is the moving Lagrangian reference frame. The problem of
solving the Black-Scholes equation for American options is treated as a free
boundary problem, where we must determine both the value of the option,
and also when the option should be exercised. We introduce a new method
for locating the moving boundary. The equation is then solved using a finite
difference Crank-Nicolson method.
A monitor function is introduced to increase the resolution of the method
close to the exercise boundary, and the method is modified to accommodate
this. We then attempt to solve the problem using a finite element method
and compare the accuracy of the two approaches.
Acknowledgements

I would like to thank Prof. M.J.Baines for his time and invaluable help with
this thesis. I have thoroughly enjoyed working with him, this dissertation has
given me an insight into the true nature of research and provided me with
motivation to continue with my studies. I would also like to acknowledge my
colleagues on the Msc course and the other members of the academic staff in
the Department of mathematics at the University of Reading.
This year has been both enjoyable and challenging. I would like to thank my
family, especially me wife, for all of their support and hard work in the last
twelve months in allowing me to study at Reading and take full advantage
of this wonderful opportunity.

1
Contents

1 Introduction 4
1.1 The Problem . . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2 Aims . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6

2 Background 8
2.1 Model for Asset Prices . . . . . . . . . . . . . . . . . . . . . . 8
2.2 Black-Scholes Model . . . . . . . . . . . . . . . . . . . . . . . 10
2.3 Black-Scholes For European Option . . . . . . . . . . . . . . . 13
2.4 Modification to the model . . . . . . . . . . . . . . . . . . . . 14
2.5 American Option . . . . . . . . . . . . . . . . . . . . . . . . . 18
2.6 American Call with Dividends . . . . . . . . . . . . . . . . . . 20
2.7 General Analysis of Call with Dividends . . . . . . . . . . . . 21

3 Transformation 25
3.1 American Options PDE . . . . . . . . . . . . . . . . . . . . . 25
3.2 Transformation to Diffusion Equation . . . . . . . . . . . . . . 26

4 Numerical Methods 28
4.1 Finite Difference Based Front Tracking Method . . . . . . . . 28
4.2 Invertbility . . . . . . . . . . . . . . . . . . . . . . . . . . . . 35

2
 4.3 Stability Analysis . . . . . . . . . . . . . . . . . . . . . . . . . 36
4.4 Local Analysis of the Free Boundary . . . . . . . . . . . . . . 38
4.5 Derivative Boundary Conditions . . . . . . . . . . . . . . . . 40
4.6 Description of Algorithm 1 . . . . . . . . . . . . . . . . . . . . 41
4.7 Algorithm 2 . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
4.8 Invertbility-method two . . . . . . . . . . . . . . . . . . . . . 46
4.9 Description of Algorithm 2 . . . . . . . . . . . . . . . . . . . . 46

5 Finite Element Method 49

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
5.2 Weak Form . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
5.3 Finite Elements Algorithm . . . . . . . . . . . . . . . . . . . . 53

6 Results 55
6.1 Finite Differences . . . . . . . . . . . . . . . . . . . . . . . . . 55
6.2 Finite Element Algoritm . . . . . . . . . . . . . . . . . . . . . 63

7 conclusion 65

8 Reference 67
.1 Program 1 - pt1.f90 . . . . . . . . . . . . . . . . . . . . . . . . 69
.2 Program 2 - pt2.f90 . . . . . . . . . . . . . . . . . . . . . . . . 84
.3 Program 3 - pt3.f90 . . . . . . . . . . . . . . . . . . . . . . . . 95

3
Chapter 1

Introduction

1.1 The Problem

The simplest financial option, a European call option, is a contract with the
following conditions

• At a prescribed time in the future, known as the expiry date the holder
of the option may do the following;

• purchase a prescribed asset, known as the underlying for a

• prescribed amount, known as the exercise price or strike price

For the holder of the option, the contract is a right but not an obligation.
The other party to the contract, the individual who is known as the writer
does have a potential obligation, he must sell the asset if the holder chooses
to buy it.

The option confers to its holder a right without an obligation, and there-
fore has intrinsic value.

4
The main concerns in the valuation of options are:

• How much would one pay for this right, i.e. what is the value of an
option?

• How can the writer minimize the risk associated with his obligation?

Options have become extremely popular recently, primarily because they are
attractive to investors, both for speculation and for hedging, and because
there is now a systematic way to determine how much they are worth.

We let E denote the exercise price, i.e the cost of purchasing the option,
and S(T ) denote the underlying asset price at the expiry date. At expiry if
S(T ) > E then the holder of the call option may buy that asset for E and
then immediately sell it in the market for S(T ), gaining an amount S(T )−E.
Conversely, if E ≥ S(T ) then the holder gains nothing. The value of the call
option may therefore be expressed as

C = max(S(T ) − E, 0) (1.1.1)

Plotting S(T ) on the x-axis and C on the y-axis gives a payoff diagram as
shown in 1.1.

American options have the additional feature that exercise is permitted at

any time during the life of the option. This is in contrast to a European
option which may only be exercised at expiry. Since the American option
gives its holder greater rights than the European option, via the right of early

5
 Figure 1.1: Payoff Diagram for a European Call

exercise, potentially it has a higher value. This report will focus on the valu-
ation of American options, which are mathematically more interesting than
their European counterparts since they can be interpreted as free boundary
problems.
The American option valuation problem can be shown to be uniquely spec-
ified by a series of constraints, which are similar to those of an ’obstacle’
problem. Since we do not know the location of the free boundary Sf (t) a
priori we are lacking one piece of information compared with the European
options. Not only must a value be assigned to the option but, we must also
determine when it is best to exercise the option.

1.2 Aims

The aims of this dissertation are to produce an accurate method for the val-
uation of American call options. The optimal exercise boundary is modeled

6
by a moving boundary. The time dependent boundary point is physically in-
terpreted as the division between two regions, one where we should hold the
option, and the other where we should exercise. This point is known as the
optimal exercise price. We seek a method which will determine accurately
the location of this free boundary, and furthermore provide a corresponding
valuation for the option at discrete time steps up until the expiry date.
The conventional approach is to transform the Black-Scholes equation into a
dimensionless parabolic equation and then discretise the problem using nu-
merical methods to find a solution1 . We propose a new method of solution,
in which the equations are discretised on a moving grid. Since the optimal
exercise boundary is time dependent, and found to increase in time for the
case of a call, we also propose a new technique for expanding the domain.
Finally, we intend to solve same the problem using a finite element approach,
and compare the accuracy between the two schemes.

1
see K.N PANAZOPOULOS ET AL

7
Chapter 2

Background

2.1 Model for Asset Prices

In order to value an option we must develop a mathematical description

of how the underlying asset behaves. The price of an asset is a measure
of investors confidence. Although an oversimplification, it is reasonable to
assume that the market responds instantaneously to external influences. Fur-
thermore asset prices must move randomly because of the efficient market
hypothesis. This can be stated as

• The past history is fully reflected in the present price, which does not
hold any further information

• Markets respond immediately to any new information about an asset

With these two assumptions, the asset price is said to follow a Markov pro-
cess

8
Definition 1. A Markov process is a particular type of stochastic process
where only the present value of a variable is relevant in predicting the future.
The past history of the variable, and the way in which the present has emerged
from the past are irrelevant

Suppose that at time t the asset price is S. Consider a small subsequent

time interval dt, during which S changes to S + dS. We decompose this
return into two parts, one component that is deterministic, comparable to
the return on a risk-free investment. This gives contribution to the return
dS/S
µdt (2.1.1)

µ is a measure of the average rate of growth of the asset price, also known
as drift.

The second contribution to dS/S models the random change in the asset
price in response to external effects. It is represented by a random sample
drawn from a normal distribution with mean zero. It adds contribution

σdX (2.1.2)

σ is known as the volitility, which measures the standard deviation of the

returns.

Putting the two together yields the Stochastic Differential Equation

dS
= σdX + µdt (2.1.3)
S
This is the basic mathematical representation for generating asset prices.
The term dX, which contains the random element which is a feature of asset

9
prices, is known as a Wiener process, it has the following properties:

• dX is a random variable, drawn from a normal distribution

• dX has a mean of zero and a variance dt

This may be expressed as

√
dX = φ dt (2.1.4)

φ is a random variable drawn from a standardized normal distribution, with

zero mean and variance of one. The probability distribution function is given
by
1 1 2
√ e− 2 φ (2.1.5)
2π
A generalized Wiener process with drift µ and variance σ is shown in figure
2.1

2.2 Black-Scholes Model

The key question concerning the valuation of Options is: what is an option
worth at time t=0? The problem is to systematically determine a fair value
for an option, at the time at which the contract is entered into. Before we
proceed with any analysis, there are several assumptions that we must first
make

• The asset price follows a log normal random walk

• The risk-free interest rate r and the asset volatility σ are known func-
tions of time over the life of the option.

10
 • There are no associated transaction costs

• The underlying asset pays no dividends during the life of the option.

• There are no arbitrage possibilities

• Trading of the underlying asset can take place continuously

We now look for a function V (S, t) that gives the option value for any asset
price S ≥ 0 and at any time 0 ≤ t ≤ T . In this setting, V (S0 , 0) is the
required time-zero option value. We further assume that such a function
exists and is smooth in both variables. Ito’s Lemma provides us with a
derivative chain rule for stochastic functions; i.e. if f = f (W, t) where W is
some stochastic function.

df 1 d2 f
df = (σSdX + µSdt) + σ 2 S 2 2 dt (2.2.1)
dS 2 dS

Using 2.2.1 we can write

1 2 2 ∂ 2V
µ ¶
dV ∂V ∂V
dV = σS dX + µS + σ S + dt (2.2.2)
dS ∂S 2 ∂S 2 ∂t

This gives the random walk followed by V .

We now construct a portfolio consisting of one option and a proportion −∆

of the underlying asset. The value of the portfolio is

Π = V − ∆S (2.2.3)

The change in the value of this portfolio in one time-step is

dΠ = dV − ∆dS (2.2.4)

11
Combining 2.1.3, 2.2.2 and 2.2.3 we find that Π follows the random walk

1 2 2 ∂2
µ ¶ µ ¶
∂V ∂V ∂V
dΠ = σS − ∆ dX + µS + σ S + − µ∆S dt (2.2.5)
∂S ∂S 2 ∂S 2 ∂t
∂V
We can eliminate the random component by choosing ∆ = ∂S
. This results
in a portfolio whose increment is wholly deterministic

∂2V
µ ¶
∂V 1
dΠ = + σ2S 2 2 dt (2.2.6)
∂t 2 ∂S

The return on on an amount Π invested in a risk less asset would see a growth
of rΠdt in a time dt. If the right hand side of 2.2.6 were greater than this
amount, an arbitrager could make a guaranteed risk less profit by borrowing
an amount Π to invest in the portfolio. Conversely, if the right-hand side of
2.2.6 were less than rΠdt then the arbitrager would make a risk less, no cost,
instantaneous profit.

Thus we have
∂2V
µ ¶
∂V 1
rΠdt = + σ2S 2 2 dt (2.2.7)
∂t 2 ∂S

∂V
Substituting 2.2.3 and ∆ = ∂S
into 2.2.7 and dividing by dt we arrive at the
Black-Scholes partial differential equation

∂V 1 2 2 ∂2V ∂V
+ σ S + rS − rV = 0 (2.2.8)
∂t 2 ∂S 2 ∂S

Any derivative security whose price depends only on the current value of S
and on t, which is paid for up-front, must satisfy the Black-Scholes equation.

12
2.3 Black-Scholes For European Option

The Black-Scholes equation 2.2.8 is a backward parabolic equation. We must

therefore impose boundary conditions to ensure a unique solution. We must
impose two conditions on S, and one on t. For example we may specify that
V (S, t) = Va (t) on S = a and also that V (S, t) = Vb (t) on S = b.

Since the equation is backward in time we must also impose a final condition
such as

V (S, t) = VT (S) on t=T (2.3.1)

Where VT is a known function of time.

In the case of European options, in particular the European call, we de-

note the call value by C(S, t), with exercise price E and expiry date T .

At time t = T , the value of the call is known for certainty to be

C(S, T ) = max(S − E, 0) (2.3.2)

This is the final condition.

If S = 0 at expiry then the payoff is zero, the call option is therefore worthless,
even if there is still a period of time until expiry.

C(0, t) = 0 (2.3.3)

As the asset price increases, it will become more likely that the option will
be exercised and the actual magnitude of the exercise price becomes less

13
important. We can therefore write that as S → ∞ the value of the option
becomes that of the asset

C(S, t) S as S → ∞ (2.3.4)

For the European options, without the possibility of early exercise 2.2.8 can
be solved exactly to give the Black-Scholes value of the call option.

Assuming that the interest rate and the volatility are constant, the explicit
solution for the European call is

C(S, t) = SN (d1 ) − Ee−r(T −t) N (d2 ) (2.3.5)

N is a cumulative distribution function for a standardized normal random

variable, given by
Z x
1 1 2
N (x) = √ e− 2 y dy (2.3.6)
2π −∞
log(S/E) + (r + 21 σ 2 )(T − t)
d1 = √ (2.3.7)
σ T −t
log(S/E) + (r − 21 σ 2 )(T − t)
d2 = √ (2.3.8)
σ T −t

Figure 2.3 below shows a European call value C(S, t) as a function of S for
several values of time to expiry, with r = 0.1 and σ = 0.2

2.4 Modification to the model

In this dissertation we attempt to model the price of American options based

on dividend paying assets. The model introduced in the previous section
makes the simplification that no dividends are paid. We now consider the

14
 Figure 2.1: American Call Value Prior to expiry

effect on the options price when dividend payment is incorporated into the
model.
When assets pay out dividends, the price of an option on a underlying asset is
affected by the payments. A modification must be made to the black-scholes
equation.

In modeling dividends we must ask two questions:

• When and how often are dividend payments made?

• How large are the dividend payments?

The amounts paid as dividends may be modeled as either deterministic or

stochastic. In this dissertation we consider only those equities with dividends
whose amount and timing is known at the start of the options life.

Suppose that in time dt the underlying asset pays out a dividend D0 Sdt
where D0 is a constant. The payment is independent of time, but dependent

15
on the stock price S. The Dividend Yield is defined as the proportion of the
asset price that is paid out per unit time in this way.

Arbitrage considerations show that in each time-step dt, the asset price must
fall by the amount of the dividend payment, in addition to the usual fluctu-
ations. The random walk of the asset price 2.1.3 is modified to become

dS = σSdX + (µ − D0 )Sdt (2.4.1)

Considering the effect of the dividend payments on our hedged portfolio, we

receive and amount D0 Sdt for every asset held, and since we hold −∆ of the
underlying, the portfolio changes by an amount

−D0 S∆dt (2.4.2)

Adding 2.4.2 to 2.2.4 we arrive at

dΠ = dV − ∆dS − D0 S∆dt (2.4.3)

Following the same analysis as previously we obtain

∂V 1 ∂2V ∂V
+ σ 2 S 2 2 + (r − D0 )S − rV = 0 (2.4.4)
∂t 2 ∂S ∂S
The only change to the boundary conditions is that

C(S, t) Se−D0 (T −t) as S → ∞ (2.4.5)

The value of a European call with dividends can be shown to be

C(S, t) = e−D0 (T −t) SN (d10 ) − Ee−r(T −t) N (d20 ) (2.4.6)

Where
log(S/E) + (r − D0 + 12 σ 2 )(T − t)
d10 = √ (2.4.7)
σ T −t
√
d20 = d1 − σ T − t (2.4.8)

16
Figure 2.2: European Option Values with (lower) and without (lower) divi-
dends

17
2.5 American Option

American Options have the important additional feature that early exercise
is permitted at any time during the life of the option.

Definition 2. An American Call Option gives its holder the right, but not
the obligation, to purchase from the writer a prescribed asset for a prescribed
price at any time between the start date and a prescribed expiry date in the
future.

The formulae in section 2.3 and 2.4 do not necessarily agree with the
value of American options. The ability to exercise the option at any time
extends to the owner additional rights, and thus the American option has
potentially a higher value.

If S lies in this range so that P (S, t) < max(E − S, 0) and we exercise

the option, there is an obvious arbitrage opportunity. We could buy the as-
set in the market immediately for S and at the same time buy the option for
P ; if we then exercised the option by selling the asset for E we make a risk
free profit of E − P − S.

This opportunity would not last long before the value of the option was
pushed up by the demand of the arbitragers. We must therefore conclude
that when early exercise is permitted we must impose the constraint

V (S, t) ≥ max(S − E, 0) (2.5.1)

American and European options must therefore have different values.

18
In the case of American options there are some values of S for which it
is optimal from the holders point of view to exercise the American option. If
this were not the case the option would have the same value as the European
option, the Black-Scholes equation would hold for all S.

The valuation of an American option is therefore more complicated than

its European counterpart since we have to determine not only the option
value but also, for each value of S, whether or not it should be exercised.

This is what is known as a free boundary problem. At each time t there

is a particular value of S which marks the boundary between two regions: to
one side one should hold the option and to the other side one should exercise
it.

We denote this value, which varies with time, by Sf (t), and refer to it as
the optimal exercise price.

As we have already observed, since we do not know Sf a priori unlike the cor-
responding European problem, we do not know where to apply the boundary
conditions, and for this reason, the problem is called a free boundary problem.

An American option valuation can be shown to be uniquely specified by

a set of constraints

• the option value must be greater than or equal to the payoff function

• the Black-Scholes equation is replaced by an inequality

19
 • the option value must be a continuous function of S

• the option delta (slope) must be continuous

2.6 American Call with Dividends

From section 2.2 the value C(S, t) of an American option satisfies

∂C 1 2 2 ∂ 2 C ∂C
+ σ S 2
+ (r − D0 )S − rC = 0 (2.6.1)
∂t 2 ∂S ∂S
This holds as long as exercise is not optimal. The payoff condition is

C(S, T ) = max(S − E, 0) (2.6.2)

Also, since the option may be exercised at any time, we have that

C(S, t) ≥ max(S − E, 0) (2.6.3)

Along the optimal exercise boundary S = Sf (t)

∂C
C(Sf (t), t) = Sf (t) − E (Sf (t), t) = 1 (2.6.4)
∂S
If the optimal exercise boundary exists then 2.6.1 is valid only while C(S, t) >
max(S −E, 0) since max(S −E, 0) is not a solution of the Black-Scholes equa-
tion.

2.6.1 can be replaced by an inequality

∂C 1 2 2 ∂ 2 C ∂C
+ σ S + (r − D 0 )S − rC ≤ 0 (2.6.5)
∂t 2 ∂S 2 ∂S
The inequality holds only if C(S, t) > max(S − E, 0). If early exercise
if optimal, it is because the option would be less valuable than if it were
exercised immediately and the funds deposited in an interest paying bank
account.

20
2.7 General Analysis of Call with Dividends

We can simplify the Black-Scholes equation with dividend payments by as-

suming that the interest rate and the dividend payments satisfy r > D0 > 0.
We can then make equations 2.6.1,2.6.2 and 2.6.4 dimensionless and reduce
2.6.1 to a constant coefficient forward equation1 .

We now also subtract off the payoff S − E for the call value C(S, t).

τ
S = Eex , t = T − 1 2 , C(S, t) = S − E + Ec(x, τ ) (2.7.1)
2
σ

the result is
∂c ∂ 2c ∂c
= + (k´− 1) − kc + f (x) (2.7.2)
∂τ ∂x2 ∂x
for −∞ < x, ∞ and τ > 0. The function c(x, 0),the initial profile, is given
by
c(x, 0) = max(1 − ex , 0) (2.7.3)

A graph of c(x, 0) is shown in 2.7. The two parameters k and ḱ are given by

r (r − D0 )
k= 1 2, ḱ = 1 2 (2.7.4)
2
σ 2
σ

The function f is given by

f (x) = (ḱ − k)ex + k (2.7.5)

Assuming that the free boundary does exist, x = xf (t), at this boundary we
have
∂c
c(xf (τ ), τ ) = (xf (τ ), τ ) = 0 (2.7.6)
∂x
1
We follow the same analysis as Wilmott,Mathematics of Financial Derivatives,Chapter
7.7

21
 Figure 2.3: c(x,0)

We now have the constraint that c ≥ max(1 − ex , 0). The behavior of f (x),
the consumption/replenishment term, is critical to the behavior of the free
boundary. A graph of f (x) is shown in figure 2.7. f (x) is positive when
x < x0 where
x0 = log(k/(k − ḱ)) = log(r/D0 ) > 0 (2.7.7)

For x ≥ x0 the function is negative. If we suppose that no free boundary

existed and consider the initial data c(x, 0) for positive values of x. For x > 0

∂c(x, 0) ∂ 2 c(x, 0)
c(x, 0) = = =0 (2.7.8)
∂x ∂x2

From equation 2.7.2 at expiry we have

∂c
= f (x) (2.7.9)
∂τ

For 0 < x < x0 , f (x) > 0 and thus c is positive. If x > x0 then f (x) < 0
and c will be negative. We have the constraint that c > 0 for x > 0 thus the
latter does not satisfy this constraint.

22
 Figure 2.4: f(x)

If we hold the option in x > x0 the option falls below its intrinsic value
and the constraint is broken.

We must therefore take xf (0) = x0 since this is the only point consistent
with c(xf (0), 0) = 0
Figures 2.7 and 2.7 show the values of c(x, τ ) in the dimensionless diffusion
setting, and the original C(S, t) at times prior to expiry.

23
Figure 2.5: Local Solution c(x, τ )

Figure 2.6: Option Value C(S, t)

24
Chapter 3

Transformation

3.1 American Options PDE

In this section we introduce a transformation for the valuation of American

calls. The Black-Scholes equation modeling the price of a dividend paying
asset V , under deterministic yield D, volatility σ and interest rate r may be
written as
1
Vt + σ 2 S 2 Vss + (r − D)SVs − rV = 0 (3.1.1)
2
Here S denotes the underlying asset on which the call option is written. The
early exercise feature of the American option results in an optimal exercise
boundary problem, which in the PDE setting is treated as a free boundary
problem.
We denoted the free boundary with B̃(t). The domain of equation 3.1.1 is

25
 ˜ × [0, T ). The boundary conditions are given below
(0, B(t))

V (S, T ) = max(S − K, 0), S ∈ (0, B̃(T )) (3.1.2)

rK
B̃(T ) = max(K, ), (3.1.3)
D
V (0, t) = 0 (3.1.4)

V (B̃(t), t) = B̃(t) − K (3.1.5)

Vs (B̃(t), t) = 1 (3.1.6)

3.2 Transformation to Diffusion Equation

The transformation we apply, a detailed derivation can be found in Panza-

opoulos,Houstis and Kortesis(1997), reduces the Black-Scholes equation to a
diffusion equation. The benefit of doing this is that the diffusion equation is
a far simpler and less cluttered equation than the Black-Scholes. It is then
a simpler matter to find exact solutions to the diffusion equation and then
convert back to financial variables. Letting k1 and k2 be defined as follows

2r 2(r − D)
k1 = k2 = (3.2.1)
σ2 σ2

We introduce the following transformations

1 2
τ = σ (T − t) (3.2.2)
2
x = log(S/K) + (K2 − 1)τ (3.2.3)
˜ )/K) + (k2 − 1)τ
B(τ ) = log(B(τ (3.2.4)
ek1 τ
u(x, τ ) = (V (S, t) − S + K) (3.2.5)
K

Equation 2.6.5 becomes

uτ = uxx + g(x, τ ) (3.2.6)

26
Where
g(x, τ ) = ek1 τ ((k2 − k1 )ex−(k2 −1)τ + k1 ) (3.2.7)

This is the principle equation which we will attempt to solve using nu-
merical methods in the following section. The domain of equation 3.2.6
is (−∞, B(0)) × (0, 12 σ 2 T ). The boundary conditions for the American call
become

u(x, 0) = max(1 − ex , 0), x ∈ (−∞, B(0)) (3.2.8)

r
B(0) = max(0, log ) (3.2.9)
D
limn→−∞ u(x, τ ) = ek1 τ (1 − e(x−(k2 −1)τ ) (3.2.10)

u(B(τ ), τ ) = 0 (3.2.11)

ux (B(τ ), τ ) = 0 (3.2.12)

27
Chapter 4

Numerical Methods

4.1 Finite Difference Based Front Tracking

Method

In the PDE 3.2.6 u = u(x, t) is defined in a fixed frame of reference with co-
ordinate x and time t. The differential operator L1 involves space derivatives
only.

Instead of working in the fixed(Eulerian) frame it is possible to take a La-

grangian viewpoint in which x is taken to be a moving coordinate x(t). We
then have a time-dependent mapping from a fixed set of reference coordi-
nates, e.g. a = x(0).

If we now define an invertible mapping between the fixed coordinates a and

1
Lu = u̇ − ẋux − uxx

28
the moving coordinates x at time t

x = x̂(a, t) (4.1.1)

We have
u(x, t) = u(x̂(a, t), t) = û(a, t) (4.1.2)

where û and x̂ are Eulerian.

Applying the chain rule to 4.1.2 gives

∂ û ∂ x̂ ∂u ∂u
= · + (4.1.3)
∂t ∂t ∂ x̂ ∂t

From equation 3.2.6 we have that uτ = uxx + g. Substituting into equation

4.1.3 yields
∂ û ∂ x̂ ∂u ∂ 2 u
= · + + g(x, τ ) (4.1.4)
∂t ∂t ∂x ∂x2
This is the time dependent equation, the solution of which gives the price for
the American call option.

We now discretise the problem using finite difference methods. We let N

denote the number dividing the interval of S into equally spaced subinter-
vals.

Si = iδS, i = 0, · · · , N (4.1.5)
B(τ ) − x −
δS = (4.1.6)
N

L denotes the number dividing the time interval such that

τj = jδτ, j = 0, · · · , L (4.1.7)
1 2
δτ = σ T /L (4.1.8)
2

29
The grid used for this numerical scheme is shown in Figure 4.1.

Figure 4.1: Mesh for the finite difference approximation

∂U
For an interior point (i, j) on the grid, ∂S
is approximated by a central
difference formula
j j
∂U ∂Ui+1 − Ui−1
≈ (4.1.9)
∂S 2δS

∂U
To approximate the time derivative ∂τ
we use a forward difference approxi-
mation
∂U U j+1 − Uij
≈ i (4.1.10)
∂τ δτ
2
The second spatial derivative, ∂∂SU2 is approximated by
j
∂ 2U Ui−1 − 2Uij + Ui+1
j
≈ (4.1.11)
∂S 2 (δS)2

Finally, we approximate the ’nodal velocity’, Ṡ by

∂S Sij+1 − Sij
≈ (4.1.12)
∂τ δτ

30
We first discretised the PDE 4.1.4
à ! à !
j+1
Uij+1 − Uij Ui−1 − 2Uij + Ui+1
j+1 j
Ui−1 − 2Uij + Ui+1
j
= θ1 + θ2
δτ δS 2 δS 2
"Ã !#
Uij − Ui−1
j
Sij+1 − Sij
+
δS δτ
+ θ3 Gj+1 + θ4 Gji
¡ ¢
i (4.1.13)

For 1 ≤ j ≤ J − 1 and 1 ≤ n ≤ N − 1. The parameters θi control the

implicitness of the scheme.
For consistency we must have

θ1 + θ2 = θ3 + θ 4 = 1 (4.1.14)

As time increases the domain expands with B(τ ). The grid is appropriately
expanded by first determining the position of the free boundary then dividing
the domain into equal linearly spaced grid points. i.e if we let xj+1
N denote
the position of the free boundary, xf (τ ), then the grid points at the j + 1
time step are defined by xj+1 = x− + Ni xj+1
¡ ¢
N −x where i = 1, 2, . . . , N .
−
i

Differentiating gives the relation

i
ẋi = (ẋN ) (4.1.15)
N

We use this equation to determine the velocity of each nodal point.

θ-Weighted Finite Difference Discretization

1
For θ = 0 the discretization is explicit, for θ = 2
we have the Crank-Nicolson
scheme, and for θ = 1 the method is implicit. In this dissertation we look
1
only at the θ = 2
case.

31
Re-arranging equation 4.1.13 we obtain

Uij+1 − Uij = αi θ1 Ui−1

£ ¡ j+1
− 2Uij+1 + Ui+1
j+1
¡ j
− 2Uij + Ui+1
j
¢ ¢¤
+ θ2 Ui−1

+ βi θ3 Gj+1 + θ4 Gji + γi Uij − Ui−1

j
¢ ¡ j+1
XN − XNj
£ ¤ £¡ ¢¤
i

Where
δτ i
αi = (δS)2
> 0, βi = 2k > 0, γi = N δS
>0
Rearranging 4.1.16 we are left with

j+1
ci Ui−1 + ai Uij+1 + bi Ui+1
j+1
+ fi (Uij − Ui−1
j
)XNj+1 = ći Ui−1
j
+ ái Uij + b́i Ui+1
j

+ f´i (Uij − Ui−1

j
)XNj + ei Gj+1
i + éi Gji

where

ci = −αi θ1 ći = αi θ2
ai = 1 + 2αi θ1 ái = 1 − 2αi θ2
bi = −αi θ1 b́i = αi θ2 (4.1.16)

ei = 2θ1 βi éi = 2θ2 βi

fi = θ1 γi f´i = θ2 γi

The problem is then reduced to solving the system of equations

~ j+1 = BU j + d~
T U j+1 + βX (4.1.17)
N

In order to find the location of free boundary at each successive time step,
we require one more piece of information. This is given by the derivative
∂C(B(τ ),τ )
boundary conditions 4.1.17. The condition ∂x
= 0 gives one extra
equation, namely uN −1 = uN . Writing this as a matrix equation with ??, we
have the equations

~ N = B~uj + d~
T ~uj+1 + βx (4.1.18)

hT ~u = 0

32
Where the components of T ,B,d and β are given by

 
2 + 2r −r 0 ··· 0 
 
 −r 2 + 2r −r 
 

T =  0 . . . . . .

 −r 0 
 .
. .

 . .. ..
 . −r 

 
0 ··· 0 −r 2 + 2r
(4.1.19)
 
2 − 2r r 0 ··· 0 
 
 r 2 − 2r r 
 

B =  . .
.. ..

 0 r 0 

 .
... ...

 .
 . r 

 
0 ··· 0 r 2 − 2r
(4.1.20)
j j
u −u
di = 1
( − −
) i i−1 i j
2 g(x +ih,j∆τ )+g(x +ih,(j+1)∆τ ) −( xi −xi−1 )( N )xN

j j
u −u
βi = − i
N
i
j
i−1
j
x −x
i i−1

(4.1.21)

hT = 0 0 · · · · · · − 1 1 (4.1.22)

33
To simplify the notation we absorb the known quantity Buj into the d vector.
Writing this matrix equation explicitly we now have
   
  uj1 d´1
2 − 2r −r 0 ··· 0 −β1   
 ´ 

 j


  u2   d2 
 −r 2 − 2r −r . . . −β    
2   .. 
   . 
 .. 
 . 


 0 ... ... .. 
−r 0 .  . 
 
  . 
. . 
 .  =  . 

 . ... ... .. 
  
 .. −r .    
 ´ 
 u j
d
 
  N −1   N −1 
 0 ··· 0 −r 2 − 2r −βN −1  
   
   xj+1

N

  0 
 
0 0 · · · −1 1 0    
0 0
This may be written symbolically as
~
   

T β~ ~u d´
    
=   (4.1.23)
    
 
    
~hT 0 j+1
xN 0
Equation 4.7.5 can now be re-arranged to solve for xN

~ N = d~
T ~u + βx
~hT ~u = 0
~ ~
³ ´
⇒ ~u = T −1
d´ − βx −N
~
³ ´
⇒ ~hT T −1 d´ − T −1 βx
~ N = 0
˜
hT T −1 d´
⇒ xN = (4.1.24)
hT T −1 β̃
We have therefore defined a method for locating the free boundary xf (τ ) at
each successive time step of the algorithm. Once xj+1
N has been calculated,
we may determine the velocity at which the nodes move using the simple
equation
i
ẋi = ẋN (4.1.25)
N

34
We may then substitute into equation 4.7.5 to obtain

~ ~ j+1
T ~u = d´ − βxN (4.1.26)

Solving this equation is straightforward, and is accomplished using a tridi-

agonal solver, where
~ ~
~u = T −1 (d´ − βxN) (4.1.27)

4.2 Invertbility

We must ensure that in the case of explicit schemes, e.g. Crank-Nicolson,

that the matrix T is indeed invertible. To be able to solve equation 4.1.27
we must be able to find the inverse T −1 ; we consider the following definition

Definition 3. A tridiagonal matrix A is said to be strictly diagonally domi-

nant (s.d.d) if and only if
|ai | > |ci | + |bi |

The a′i s are the coefficients along the diagonal and the c′i s and the b′i s
are the coefficients on the lower and upper diagonal respectively. Then the
matrix A is non-singular.
1
If we consider the matrix T in the Crank-Nicolson, θ = 2
scheme, the coeffi-
cients are

ci = −αi θ1

ai = 1 + 2αi θ1

bi = −αi θ1

Clearly
1 + 2αi θ1 > 2αi θ1 ⇒ |ai | > ci | + |bi |

35
Therefore the matrix T is s.d.d and invertible

4.3 Stability Analysis

In this section we analyse the problem of stability of the finite difference

calculations that are used to solve equation 3.2.6.
¯
Let Qj and Rj be two solutions of the system of equations AU j+1 = BU j + d,
that have the same inhomogeneous term d¯ but with different initial data Q0
and R0 . Their difference U j = Qj − Rj satisfies the homogeneous system of
equations and stability is achieved by establishing that

AU j+1 = BU j and ⇒ °W j ° ≤ K °W 0 °
° ° ° °

If the constant K is such that |K| ≤ 1 then the scheme is said to be stable.
Fourier or von Neumann’s method is the most precise and useful tool for
studying stability in the l2 norm. The Fourier method expresses the initial
values at the mesh points along t = 0 in terms of a finite Fourier series, then
considers the growth of a function that reduces to this series for t = 0 by a
seperation of variables method.
Fourier stability analysis can be restrictive however, since it can only be
applied to linear problems with constant coefficients and periodic boundary
conditions. The problem we are considering, u̇ − ux ẋ = uxx + g, is not a
linear equation. We can still proceed however, by ’linearising’ the problem,
and applying the stability condition locally at every interior point of the
domain.
We begin by making the substitution Unj = λn eikjδS and Xnj = ξn eikjδS . The

36
numerical scheme is given by
( )
j+1 j+1 j+1
uj+1
i − uj
i u i−1 − 2u i + u i+1
= θ + g(x− + ih, (m + 1)k)
k h2
( )
uji−1 − 2uji + uji+1
+ (1 − θ) + g(x− + ih, mk)
h2
uji − uji−1 i xj+1
N − xjN
+ (4.3.1)
h N k

uji −uji−1 i xj+1

N −xN
j
The term involving h N k
may be linearised by freezing the ux
uji −uji−1
term h
. We also drop the known g function. We may then write

eikjδS − λn eikjδS
µ ¶
λn+1 =
k
1 λn+1 eik(j−1)δS − 2λn+1 eikjδS + λn+1 eik(j+1)δS
µ ¶
=
2 h2
1 λn eik(j−1)δS − 2λn eikjδS + λn eik(j+1)δS
µ ¶
+
2 h2
ξn+1 eijkδS − ξn eijkδS
+ βi (4.3.2)
k

multiplying through by k and eikjδS gives

k ¡ ikδS −ikδS
¢ k ¡ ikδS −ikδS
¢
λn+1 −λn = λ n+1 e − e − 2 + λ n e − e − 2 +βi (ξn+1 − ξn )
2h2 2h2

Writing eikδS as 2cos(kδS), using the identity cos(kδS) = 1 − 2sin2 ( kδS

2
) and
k
writing h2
= µ we obtain

ikδS ikδS
λn+1 (1 − 2µsin2 ( )) = λn (1 − 2µsin2 ( )) + (ξn+1 − ξn ) (4.3.3)
2 2

Which holds for N − 1 interior equations.

We also have the boundary conditions that u̇N − u̇N −1 = 0 which provides
one extra equation
λn − eikδS λn−1 = 0 (4.3.4)

37
We therefore have two equations in two unknowns allowing us to solve for
amplification factors λ and ξ.

4.4 Local Analysis of the Free Boundary

A graph of the initial date profile is given in figure 2.7. The domain of equa-
tion 3.2.6 is (−∞, B(0)) × (0, 21 σ 2 T ). Furthermore, from section 2.7 we know
where the free boundary, xf (0), must start. Clearly, the is a discrepancy be-
tween the points where the initial data falls to zero,i.e. x0N and the position
of xf (0).( see fig 4.4).
In moving the curve from x = 0 to x = xf (0) the finite difference algorithm

Figure 4.2: Initial Data Curve and Free Boundary

becomes unstable. We must find another way to advance the curve for the
first time step of the algorithm. Once this has been achieved, the algorithm
can then be used to progress the curve.

To see how the free boundary x = xf (τ ) initially moves away from xf (0)

38
we find an asymptotic solution that is valid close to expiry2 .

Restricting our analysis small values of τ and for x close to x0 we expand

f (x) by a Taylor series about x0 .

f (x) = f (x0 ) + f´(x0 )(x − x0 ) + O((x − x0 ))2 (4.4.1)

(x − x0 )f´(x0 ) = −k(x − x0 )

We assume an approximate local solution c(x , τ )that satisfies

∂c ∂2c
= − k(x − x0 ) (4.4.2)
∂τ ∂x2

with

∂c
c= =0 (4.4.3)
∂x

Taken on

x = xf (τ ) (4.4.4)

xf (0) = x0 (4.4.5)

This local problem can be solved exactly ( see appendix ). The similarity
solution in terms of the variable

(x − x0 )
ξ= √ (4.4.6)
τ

is of the form
c = τ 3/2 c∗ (ξ) (4.4.7)

(note about cstar) We also try a free boundary of the form

√
xf (τ ) = x0 + ξ0 τ (4.4.8)
2
See Wilmott, Mathematics of Financial Derivatives, section 7.7.2

39
Where ξ0 is a constant taken to be 0.9034....

This is approximation to the free boundary motion that we use to expand

the grid for the initial time step, i.e. for i = 1. The method described in
section 4.1 is then used for successive time steps, up until expiry.

4.5 Derivative Boundary Conditions

Equation 4.1.17 gives the condition that the derivative at the moving bound-
ary must be zero. In finite difference notation this can be written

uN −1 = uN (4.5.1)

Since we also have the boundary condition u(B(τ ), τ ) = 0 this implies that
uN −1 is also zero. A graph of this data is shown in figure 4.5 below.
The existence of a discontinuity in the data, where the function falls to zero,

Figure 4.3: Derivative Conditions

causes instability close to the leading edge of the curve. This leads to poor

40
results in the region local to the moving boundary point.
In an effort to overcome this, we replace the zero derivative requirement by
an approximation. We seek a function of the form

ay 2 + by = 0 (4.5.2)

to approximate the curve close to xf . Taking the derivative of this function

gives

2ay + b = 0 (4.5.3)

The derivaitve is known to be zero at y = 0 which implies that b = 0 and

hence our curve is modeled by the function ay 2 . We have the condition that

ah2 = uN −1

a(2h)2 = uN −2 (4.5.4)

We therefore have a relation that

uN −2 = 4uN −1 (4.5.5)

4.6 Description of Algorithm 1

The initial data profile presents two difficulties from a numerical perspective.
Firstly, the algorithm is found to be unstable when applied to the function
max(1−exp(x), 0). This problem is overcome by utilizing the approximation
to the free boundary derived in section 4.4. The second problem arises due
to the discontinuous data curve that results from imposing the boundary

41
 Figure 4.4: Approximation to the Derivative

conditions 4.1.17. At the moving boundary,xN , the form of the solution is

approximated by a parabola .
We now present a brief overview of the algorithm

Set Initial Conditions u0 , B 0 , x0

Approximate the derivative at xN by ’parabola’
For j = 1 DO
√
- Set the Velocity of xN to ξ0 τ

- Rescale x grid points

- Solve equation 4.1.26 for u1

- Set τ = τ + ∆τ

END DO
For j = 2 to N − 1 DO

42
 - Solve equation 4.1.24 for xiN

- Rescale x grid points

- Approximate the derivative at xN by ’parabola’

- Solve equation 4.1.26 for uji

- Set τ = τ + ∆τ

END DO

4.7 Algorithm 2

To improve the accuracy of the algorithm used in section 4.1 we now introduce
a monitor function, the effect of which is to increase the number of grid points
in the local region(s) where the curve is changing rapidly. Likewise, regions
where the data is varying less rapidly will be assigned fewer grid points.
Our aim is to increase the resolution close to the moving boundary point β(τ ).
To understand the significance of this, we make a transformation back to
financial variables. Here, the moving boundary represents a division between
two points. Points to the left of the moving boundary represent asset prices
for which it would be unprofitable to enter into the option, whereas points
to the right are asset prices for which the option would be profitable. It is
therefore essential to gain an accurate approximation for this point.
The monitor function is described by
µ ¶
(xf (τ ) − x− )
xi = xf (τ ) − (N − i)2 (4.7.1)
N2

43
This function gives the position of each nodal point in terms of the free
boundary. The velocity of the nodes may be calculated from 4.7.1 as

(N − i)2
µ ¶
ẋi = 1− x˙f (τ ) (4.7.2)
N2

Algorithm 1 uses equal linearly spaced intervals. The introduction of the

Figure 4.5: Grid Spacing

monitor function leads to a parabolic spacing of the grid points, see fig 4.7.
Since the x− spacing is no longer constant between grid points, the Crank-
Nicolson method used in the previous algorithm is no longer valid. In order
to apply a finite difference discretization we introduce a Lagrange polynomial
to approximate the second derivative.

∂ 2f 2f (x0 ) 2f (x1 ) 2f (x2 )

2
= + + (4.7.3)
∂x (x0 − x1 )(x0 )(x2 ) (x1 − x0 )(x1 − x2 ) (x2 − x0 )(x2 − x1 )

The lagrange polynomial is degree two on the support {x0 , x1 , x2 } for the
function u(x). Replacing the difference approximation to the second deriva-

44
tive in equation gives
( j+1 j+1
)
j+1 j 2u j+1 2u
u −u i−1 2u i+1
i i =θ i
∆τ j+1 j+1 j+1 j+1 + j+1 j+1 j+1 j+1 + j+1 j+1 j+1 j+1
(x −x )(x −x ) (x −x )(x −x ) (x −x )(x −x )
i−1 i i−1 i+1 i i−1 i i+1 i+1 i−1 i+1 i
( j j
)
2u j 2u
i−1 2u i+1
+(1−θ) j j j j + j j i j j + j j j j
(x −x )(x −x ) (x −x )(x −x ) (x −x )(x −x )
i−1 i i−1 i+1 i i−1 i i i+1 i−1 i+1 i
j j µ ¶ j+1 j
u −u (N −i)2 x −x
+θ(g(x− +ih,j∆τ ))+(1−θ)(g(x− +ih,(j+1)∆τ ))+ i
j
i−1
j 1− 2
N
∆τ
N
x −x N
i i−1

As before this may be written as a matrix equation

~ N = d~´
T ~u + βx (4.7.4)

hT ~u = 0

´ β and hT are given by

Where Where the components of T , d,

 
k k
1 − α´2 − α´3 ··· ··· 0 
 
 −k k k
 α´2 1 − α´3 − α´4


 ... ... ... .. 
T = 
 0 . 
 (4.7.5)
 . ..
... ... ...

 .
 . .


 
0 ... ... − αNk´−2 1 − k
αN´−1

d´i = k j
(
k
αi−1 ui−1 + 1+ αi )uji + αi+1
k uj
i+1
}

à ! j j
(N −i)2 u −u
βi = − 1−
N2
i
j
i−1
j
x −x
i i−1

hT = 0 0 · · · · · · − 1 1 (4.7.6)

Where

αi−1 =(xji−1 −xji )(xji−1 −xji+1 )

αi =(xji −xji−1 )(xji −xji )

αi+1 =(xji+1 −xji−1 )(xji+1 −xji ) (4.7.7)

45
Writing this matrix equation explicitly we have
    
1 − αk´2 − αk´3 0 ··· −β1 uj1 d´1
    
 −k k k ´
  j   
 α´1 1 − α´3 − α´4 · · · −β2   u2   d2 
   

 0 ... ... ... ..  . 
  .. 
 . 
 .. 
.
 .  =  . 
    
 . ... ... ... ..
 .. .   ..   .. 
    
    
´ 
 0 ... ... − αNk´−2 1 − αNk´−1 −βN −1  uN −1  dN −1 
  j 
    
0 0 ··· ··· −1 1 0 xN 0
The solution method for xN remains unchanged. We solve for xN as before,
~ ~
and express the solution as ~u = T −1 (d´ − βxN ).

4.8 Invertbility-method two

We proceed as previously outlined in section 4.2

k k k
ci = − ai = (1 − )bi = −
άi−1 άi άi+1
Again, we need to ensure |ai | > |bi | + |ci |
k k k
⇒ (1 − )>− −
άi άi−1 άi+1
This is restriction imposed on the grid spacing

4.9 Description of Algorithm 2

The introduction of the monitor function, reducing the spatial interval near
the moving boundary, modifies the T matrix given in equation 4.7.6. The ’r’

46
 k
values, h2
, are no longer a constant and must be evaluated at each pair of
nodes within the domain. In our modified algorithm, the T matrix contains
k
entries α
where α is equivalent to h in our previous method when the domain
was divided into linear elements. Clearly as the spacing becomes smaller, the
k k
magnitude of α
becomes larger. When the ration α
becomes larger than 1 the
solution method is found to become unstable. To rectify this we precondition
the matrices as follows;
Let D be the diagonal of T . Multiply both sides of equation 4.1.26 by D−1
to obtain
D−1 T ~u = D−1 d~ − D−1 βx
~ j+1
N (4.9.1)

To simplify the notation we let T D−1 = T́ and D−1 d~ − D−1 βx

~ j+1 = f´. We
N

then solve the equation

u = T́ −1 f´ (4.9.2)

We now present a brief description of the algorithm

Set Initial Conditions u0 , B 0 , x0
Approximate the derivative at xN by ’parabola’
For j = 1 DO

√
- Set the Velocity of xN to ξ0 τ

- Rescale x grid points using the monitor function

- Calculate βi and construct T matrix

- Multiply equation 4.1.26 by D−1

- Solve equation 4.9.2 for u1

47
 - Set τ = τ + ∆τ

END DO
For j = 2 to N − 1 DO

- Solve equation 4.1.24 for xiN

- Rescale x grid points using the monitor function

- Approximate the derivative at xN by ’parabola’

- Calculate βi and construct T matrix

- Multiply equation 4.1.26 by D−1

- Solve equation 4.9.2 for uji

- Set τ = τ + ∆τ

END DO

48
Chapter 5

Finite Element Method

5.1 Introduction

In this section we attempt to find a solution to the American call problem us-
ing a finite element method. The previous approach has been to replace the
continuous operation of differentiation with the discrete operation of finite
differences. We then reformulate the equations in terms of finite differences.
The fundamental idea of the finite element method is the replacement of con-
tinuous functions by piecewise approximations. The most elementary choice
of basis functions is the piecewise linear polynomials, this is the approxima-
tion we shall use in this report.
We divide the interval (−∞, B(0)) into linearly spaced elements. The parti-
tion is defined by

£ ¤
x− = x0 , x1 , · · · , xj−1 , xj , xj+1 , · · · , xN +1 = xf (5.1.1)

49
Define the basis functions ( ’hat’ functions ) as follows




 0 : 0 ≤ x ≤ xi−1

 x−xi−1 : xi < x ≤ xi+1


xi −xi−1
φi (x) =
xi+1 −x
 xi+1 −xi : xi < x ≤ xi+1





0 : xi+1 < x ≤ 1



The functions φi are piecewise linear, the derivatives φ́i are constant on
(xi , xi+1 ) for each i = 0, 1, · · · , n.




 0 : 0 ≤ x ≤ xi−1


 1

: xi < x ≤ xi+1
hi−1
φ́i (x) =
−1



 hi
: xi < x ≤ xi+1


0 : xi+1 < x ≤ 1



Figure 5.1: Hat Function

Finally let
N
X +1
U (x) = Ui φi (x) (5.1.2)
i=0
be a trial function, once differentiable between each xi . This is the piecewise
linear Finite Dimensional Representation.

50
5.2 Weak Form

The weak form of the differential equation 3.2.6 is found as follows.

Let φi (x) be a set of test functions where φi (x) ∈ C 1 . Multiply 3.2.6 by φi
and integrate.
Z xi+1 (τ ) Z xi+1 (τ )
φi ut dx = φi (uxx + g) dx (5.2.1)
xi−1 (τ ) xi−1 (τ )
(5.2.2)

Since the grid is not fixed

xi+1 (τ ) xi+1· ¸x (τ ) Z B(τ )
d dx i+1
Z Z
φi udx = φi ut + φi u Let θ = udx
dt xi−1 (τ ) xi−1 (τ ) dt xi−1 (τ ) A

Now θ, the area under the curve, is not constant in time, since we have a
source term g(x, τ ) that adds and subtracts ’mass’.
Z xi+1 (t)
Let φi udx = ci θ (5.2.3)
xi−1 (t)
P
be our new monitor function, where ci is a fraction such that i ci θ = 1. See
dψ
?? below We now define ẋ to be eqaul to dx
, the ’velocity’ potential.
At interior points, 1 ≤ i ≤ N − 1
xi+1 xi+1 (t) xi+1 (t)
dφi dφi dψ
Z Z Z
− (t) ux dx + φi gdx − u = ci θ̇ (5.2.4)
xi−1 (t) dx xi−1 (t) xi−1 (t) dx dx

At i = 0
x2 (t) x2 x2 (t)
dφ1 dφ1
Z Z Z
xi+1 (t)
−ux |x=x− − ux dx+ φ1 gdx+φi uẋ |xi−1 (t) − uẋdx = c1 θ̇
x0 (t) dx x0 x0 (t) dx
(5.2.5)
At i = N
xN (t) xN (t)
dφN −1
Z Z
− ux dx + φN −1 gdx (5.2.6)
xN −1 (t) dx xN −2 (t)

51
 Figure 5.2: Monitor Function

Equation 5.2.4 may be written as a system of equations for ψ and θ̇

Ḱψ + θ̇c = g − Ku (5.2.7)

Where
xi+1
dφi dφj
Z
dx = K´ij
xi−1 dx dx
Z xi+1
dφi dφj
u dx = Kij
xi−1 dx dx
Z xi+1
φi gdx = Gi
xi−1
Z xi+1
φi φj dx = Mij (5.2.8)
xi−1

the tridiagonal matrix K is known as the stiffness matrix, while M is known

as the mass matrix.
In order to solve 5.2.9 we need one extra piece of information. This is provided

52
 dψ
by the boundary condition ẋ = 0 at x = x− . This implies that dx
= 0 at
x = x− , i.e. ψ0 = ψ1 . Writing this as a matrix equation we have
    
Ḱ ~c ~
ψ f~
    
 =   (5.2.9)
    

    
~hT
0 θ̇ 0

The problem is then to solve

ψ ~˙
~ = Ḱ −1 (~g − K~u − θ) (5.2.10)

From 5.2.9 we have that

~ = 0
~hT ψ

~ = ~hT Ḱ −1~g + ~hT Ḱ −1 K~u − θ~˙~hT Ḱ −1~c

⇒ ~hT ψ
˙ ~hT Ḱ −1~g − ~hT Ḱ −1 K~u
⇒ θ~ = (5.2.11)
~hT Ḱ −1~c

5.3 Finite Elements Algorithm

In this section we provide an outline of the steps involved in calculating the

Finite Element solution of the American Call problem , 4.1.4. In order to
evaluate the stiffness matrix K´ij we approximate the function u by a linear
interpolant between each node i.e.
µ ¶ µ ¶
xi+1 − x x − xi
U = ui + ui+1 (5.3.1)
xi+1 − xi xi+1 − xi

The steps of the algoritm are outlined below Set Initial Conditions u0 , B 0 , x0
For j = 1to N − 1 DO
For i = 1 to N − 1 DO

- Calculate gi

53
 - Calculate Kij

- Calculate Ḱij using linear interpolant

˙ ~hT Ḱ −1~g −~hT Ḱ −1 K~

- Set θ~ = ~hT Ḱ −1~c
u

~˙c
~ = ~g − K~u − θ~
- Solve Ḱ ψ

dψ
- Calculate nodal velocities ẋ by setting ẋ = dx

- Integrate ẋ to find new node positions xj+1

i

- Evaluate M using new node positions

- Solve M~u = ~c

END DO
END DO
The ẋ values are integrated using a Runge-Kutta order four method, ui+1 =
ui + 61 (k1 + 2k2 + 2k3 + h4 ).
Elements of the K matrix are calculated exactly whilst the entries in g vector
and Ḱ matrix are evaluated using a numerical quadrature. ( n = 2: Simpsons
Rule ). Both the K and Ḱ matrices are tridiagonal, equation 5.2.10 is found
using a tridiagonal solver. ( Thomas Algorithm )

54
Chapter 6

Results

6.1 Finite Differences

In this section we present the numerical results of the finite difference method.
We assume the following parameters

K = 10

T = {0.25, 0.5, 1.0}

r = {0.03, 0.06, 0.1}

σ = {0.2, 0.4, 0.6}

D = {0.8r, 1.0r, 1.2r} (6.1.1)

The smaller values of σ represent lower volatility of the underlying assets,

while values of T represent short, medium and long term call options.
We first make the transformation back to financial variables, taking ui to be
our numerical solution to the diffusion problem, we find the solution to our

55
 Figure 6.1: {T = 1, σ = 0.2, r = 0.03, D = 0.8r}

option value as follows

S = expx−(k2 −1)τ K

t = −((2τ )σ 2 − T )
u(x, τ )K
V (S, t) = +S−K (6.1.2)
expk1 τ

We apply the numerical scheme for the three combinations of the parameters
given in 6.1.1, time to expiry is taken as 1.
Figures 6.1,6.1 and 6.1 show the data curves plotted in the dimensionless
σ2
( Diffusion ) setting. A time step of ∆τ = 2
/100 is taken,the spatial
increment,∆h, is 0.001. It is clear from figures that as we vary the pa-
rameters the is a recognisable change in the profile of the curve. The contour
of the curve is determined predominantly by the source term g which itself
is a function of the interest and dividend parameters.

56
 0.8

0.7

0.6

0.5

u(x,t) 0.4

0.3

0.2

0.1

0
−1.5 −1 −0.5 0 0.5 1
x

Figure 6.2: {T = 1, σ = 0.4, r = 0.06, D = 1.0r}

From the figures we see that choosing a larger interest rate and dividend pa-
rameter causes the free boundary to move further along the x−axis, as would
be expected. Shown below is a plot of the ’source’ term g in the diffusion
equation 3.2.6

This function represents a consumption term for x < 0 and a replenish-

ment term for x > 0. Looking more closely at the function u(x, τ ) in the
dimensionless setting (see fig 6.1) we can see for x > 0 this function has
the effect of decreasing u(x, τ ), drawing out the moving boundary, whilst for
x < 0 the function adds mass to the equation, causing the curve to increase
with time.
Figures 6.1,6.1 and 6.1 show the data once it has been transformed back
into financial variables. The small movement of the free boundary in the
dimensionless field is now emphasized. The moving boundary, which now
represents the point at which we should exercise the option, can be seen

57
 0.8

0.7

0.6

0.5
u(x,t)

0.4

0.3

0.2

0.1

0
−1.5 −1 −0.5 0 0.5 1
x

Figure 6.3: {T = 1, σ = 0.6, r = 0.1, D = 1.2r}

0.8

0.6

0.4
g(x,t)

0.2

−0.2

−4 −3 −2 −1 0 1 2
x

Figure 6.4: G − function {T = 1, σ = 0.6, r = 0.1, D = 1.2r}

58
 1.4

1.2

0.8

0.6

0.4

0.2

0
−4 −3.5 −3 −2.5 −2 −1.5 −1 −0.5 0 0.5 1

Figure 6.5: U (x, τ )

moving to the right. This is as we would expect. As the time to expiry

decreases, there is less time for the price of the underlying asset to move,
and thus less potential for the option to become profitable. Consequently
the boundary which determines when we should ’hold’ or ’sell’ the option
shifts to the right.
Table 6.1 gives values a comparison of values for a call option for a num-
ber of values of the ratio S/K. We take 100 time steps and the parameters:
K = 10, σ = 0.3, D = 0.02, r = 0.04, T = 0.25.
The Benchmark solution is the solution given by the binomial method for
2500 time steps. The money less ration, S/K, is taken in the range 0.7 − 1.3.
The columns IFT,MLII and LC are included for comparison. They represent
other PDE methods that have been proposed for the pricing of American
options. We also show the relative error between the option value given by
BENCH and our approximation.

59
 20

18

16

14

12
C(S,t)

10

0 5 10 15 20 25
S

Figure 6.6: {T = 1, σ = 0.2, r = 0.03, D = 0.8r}

25

20

15
C(S,t)

10

0 5 10 15 20 25 30
S

Figure 6.7: {T = 1, σ = 0.4, r = 0.06, D = 1.0r}

60
 30

25

20

C(S,t)
15

10

0 5 10 15 20 25 30 35
S

Figure 6.8: {T = 1, σ = 0.6, r = 0.1, D = 1.2r}

Table 6.1: Prices of Call options

S/K FD IFT MLII LC BENCH Error

Call Options
0.70000 0.003261 0.004025 0.003977 0.004057 0.003988 0.182291
0.75454 0.013500 0.016526 0.016450 0.016577 0.016450 0.178977
0.80909 0.047274 0.051412 0.051360 0.051353 0.051353 0.079435
0.91818 0.117828 0.127973 0.128069 0.127895 0.127893 0.078698
0.97272 0.443699 0.481072 0.481230 0.480072 0.486390 0.087771
1.02727 0.714487 0.775478 0.775818 0.775283 0.775587 0.787790
1.08181 1.056226 1.142981 1.143191 1.142845 1.142953 0.075885
1.13636 1.414501 1.569793 1.569901 1.569956 1.569856 0.099026
1.19090 1.883253 2.040556 2.040556 2.040546 2.040508 0.770665
1.24554 2.364591 2.540893 2.540893 2.541063 2.540901 0.069388
1.30000 2.973134 3.060005 3.060005 3.059990 3.059931 0.029398

61
The ratio of S/K represents the value of the option, for a given value of S
when the strike price is taken to be 10. The term ’moneyness’ represents the
fact that when the ratio S/K < 1 the value of the underlying stock is less
than the price of the option, and when the fraction S/K > 1 the asset price
has risen above the strike price. Referring to table 6.1 we see that as the vale
of the underlying stock approaches the strike price the option value begins
to increase, and once the stock price actually exceeds the strike, the value of
option rises quickly.
The numerical results from the finite difference method are displayed in col-
umn ’FD’. The values generated by our method are approximately 25 smaller
than the benchmark values. This suggest that our approximation to the free
boundary in the dimensionless setting is poor.

62
 −40

−60

−80

−100

−120

−140

−160
−0.05 −0.04 −0.03 −0.02 −0.01 0

Figure 6.9: Velocity Potential

6.2 Finite Element Algoritm

The results provided by the finite elements algorithm proved to be very poor.
The method was unsuccessful in accurately modeling the motion of the free
boundary; after a very small number of time steps the velocity of the nodes in
the region close to the moving boundary became too large by several orders
of magnitude.
In the finite element code, we calculate a velocity potential Ψ, where Ψ =
Σψφi , and then rescale the grid by calculating the nodal velocity as equal to
dΨi
the gradient of the velocity potential. (ẋ = dt
. It is then a simple matter of
integrating this velocity to obtain the new grid points. A graph of a typical
values of Ψvsx is shown below. Ultimately, determining the new node points
rests on solving the system of equations

Ḱψ = f´

The Ḱ matrix is given by equation 5.2.8. Because of the nature of the u

function, the diagonal entries of the Ḱ matrix ranged from approximately

63
40 in size, down to 0.01. This matrix was found to be ill-conditioned, which
lead to poor results for ψ.
In an effort to overcome this we take the same approach as section ??, pre-
multiplying by D−1 where the D matrix is the sum of the diagonal matrices.
This did not solve the problem of the steep gradient close to the free boundary
however, and due to time constraints we were unable to obtain any accurate
results from this method.

64
Chapter 7

conclusion

The finite difference method applied to the American Call problem produced
valuations to within −30% of the BENCH value. The undervaluing of the
option is a result of the algorithms inability to accurately resolve the true po-
sition of the moving boundary at each time step. The algorithm was found
to be unstable when applying the derivative boundary conditions directly,
hence we saught to approximate the derivative by the form of a parabola.
Although this improved stability, we believe that this reduces accuracy of
the method in determing the moving boundary, the velocity is consistently
over approximated.
The introduction of the monitor function improved results, further work is
required in analysing the effect of approximation to the boundary conditions,
there are also several further monitor functions that can be implemented. A
comparison of these methods would be the next step in our research.
We have devised a new method for the valuation of American options as free
boundary problems, transforming the equation to a Lagrangian frame, and
are also the first to introduce the Shur method for locating the free bound-

65
ary. ( To our Knowledge ) Due to time limitations we were unsuccessful in
applying the finite element algorithm to the problem. We were unable to re-
solve the problem caused by the ill-conditioned matrix when evaluating the
velocity potentials.
Concluding, further work is required in formulating the finite element ap-
proach to this problem and finding an alternative method of calculating the
velocity potential . A comparison could then be made between the accu-
racy of the finite difference method we have presented, and the finite element
method that was originally proposed.

66
Chapter 8

Reference

1 M.J.Baines A moving mesh finite element algorithm for the adaptive

solution of time-dependent partial differential equations with moving
boundaries, Applied Numerical Mathematics 54(2005)450-469

2 F.Black & M.Scholes, The pricing and Corportate Liabilities, Journal

of Political Economy, Vol. 81, 1973

3 R.L.Burden Numerical Analysis,Brooks/Cole 1997

4 L.Clewlow & C.Strickland,Implementing Derivatives Models,John Wi-

ley & Sons,1998

5 P.Garrcia-Navarro & A.Priestley, A Conservative and Shape-Preserving

Semi-Lagrangian method for the solution of Shallow Water Equations.
Numerical Analysis Report 6/93, University of Reading, 1993

6 D.J.Higham An Introduction to Financial Option Valuation. Cam-

bridge, 2004.

67
 7 J.Hull Options,Futures, and Other Derivatives. Prentice-Hall Interna-
tional, 2003

8 A. Priestley, A Quasi-Conservative Version of the Semi-Lagrangian Ad-

vection Scheme. Numerical Analysis Report 2/92, University of Read-
ing, 1992.

9 K.Singh, A Comparison of Numerical Schemes for Pricing Bond Op-

tions. Msc Thesis, University of Reading, 1998

10 S.Smith, Submitted PhD thesis, University of Reading, 1996

11 G.D.Smith,Numerical Solution of Partial Differential Equations. Ox-

ford, 1985

12 I.M.Smith,Programming in Fortran 90,John Wiley & Sons, 1995

13 P.Wilmott The Mathematics of Finacial Derivatives. Cambridge 2002.

68
 .1 Program 1 - pt1.f90
PROGRAM AMERICAN OPTIONS707
DOUBLE PRECISION : : K, NEGINFINITY ,XN, Delt , T f i n a l , t , B, h , A, C, DelTau ,
Domain , Tau ,SOUT
DOUBLE PRECISION,ALLOCATABLE: : U ( : ) ,UTEMP( : ) ,RHS ( : ) , Z ( : ) ,Y ( : ) ,HT ( : ) ,V ( : ) ,
X ( : ) ,U2D ( : , : ) ,X2D ( : , : ) , S2D ( : , : ) ,V2D ( : , : )
INTEGER: : IOS , j , N, L , i
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.5D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

8 PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ CRANK NICOLSON SOLULTION ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ DIFFUSION EQUATION ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’
16 PRINT∗

NEGINFINITY=−30.0D0
T f i n a l =(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 ) ∗ 1 . 0 D0

Domain=NEGINFINITY
N=800
24 L=NINT ( ( T f i n a l / ( Domain/N) ∗ ∗ 2 ) )
print ∗ ,L

k=( T f i n a l /L)

DelTau=( T f i n a l ) /L

B=0.0D0
32 XN=B

69
 ALLOCATE(U( 0 :N) ,UTEMP( 0 :N) ,RHS ( 1 : N−1) ,V( 1 : N−1) , Z ( 1 : N−1) ,Y( 1 : N−1) ,HT( 1 : N
−1)&
,X( 0 :N) ,X2D ( 0 : L , 0 : N) ,U2D ( 0 : L , 0 : N) ,&
S2D ( 0 : L , 0 : N) ,V2D ( 0 : L , 0 : N) )

U=0.0D0
UTEMP=0.0D0
40 RHS=0.0D0
V=0.0D0
Z=0.0D0
Y=0.0D0
HT=0.0D0
t =0.0D0
S2D=0.0
V2D=0.0
48

OPEN(UNIT=13 ,FILE=”x . dat ” ,IOSTAT=IOS )

OPEN(UNIT=14 ,FILE=”U. dat ” ,IOSTAT=IOS )
OPEN(UNIT=15 ,FILE=” x1 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=16 ,FILE=”U1 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=17 ,FILE=” x2 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=18 ,FILE=”U2 . dat ” ,IOSTAT=IOS )
56 OPEN(UNIT=19 ,FILE=” x3 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=20 ,FILE=”U3 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=21 ,FILE=” x4 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=22 ,FILE=”U4 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=23 ,FILE=” x5 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=24 ,FILE=”U5 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=25 ,FILE=” x6 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=26 ,FILE=”U6 . dat ” ,IOSTAT=IOS )
64 OPEN(UNIT=27 ,FILE=” x7 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=28 ,FILE=”U7 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=29 ,FILE=” x8 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=30 ,FILE=”U8 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=31 ,FILE=” x9 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=32 ,FILE=”U9 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=33 ,FILE=” x10 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=34 ,FILE=” u10 . dat ” ,IOSTAT=IOS )

70
 72 OPEN(UNIT=35 ,FILE=” x0 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=36 ,FILE=” u0 . dat ” ,IOSTAT=IOS )

OPEN(UNIT=37 ,FILE=” S2 . dat ” ,IOSTAT=IOS )

OPEN(UNIT=38 ,FILE=”V2 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=39 ,FILE=” S3 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=40 ,FILE=”V3 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=41 ,FILE=” S4 . dat ” ,IOSTAT=IOS )
80 OPEN(UNIT=42 ,FILE=”V4 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=43 ,FILE=” S5 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=44 ,FILE=”V5 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=45 ,FILE=” S6 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=46 ,FILE=”V6 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=47 ,FILE=” S7 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=48 ,FILE=”V7 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=49 ,FILE=” S8 . dat ” ,IOSTAT=IOS )
88 OPEN(UNIT=50 ,FILE=”V8 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=51 ,FILE=” S9 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=52 ,FILE=”V9 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=53 ,FILE=” S10 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=54 ,FILE=”V10 . dat ” ,IOSTAT=IOS )

OPEN(UNIT=60 ,FILE=”x2D . dat ” ,IOSTAT=IOS )

OPEN(UNIT=61 ,FILE=”u2D . dat ” ,IOSTAT=IOS )
96

OPEN(UNIT=62 ,FILE=”S . dat ” ,IOSTAT=IOS )

OPEN(UNIT=63 ,FILE=”V. dat ” ,IOSTAT=IOS )

OPEN(UNIT=64 ,FILE=” S1 . dat ” ,IOSTAT=IOS )

OPEN(UNIT=65 ,FILE=”V1 . dat ” ,IOSTAT=IOS )

104 OPEN(UNIT=70 ,FILE=”SOUT. dat ” ,IOSTAT=IOS )

OPEN(UNIT=71 ,FILE=”STIME . dat ” ,IOSTAT=IOS )

IF ( IOS/=0) THEN

71
 PRINT∗ , ’ E r r o r Occured in Opening The Output File ’
112 STOP
END IF

! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
!∗ MAIN PROGRAM ∗
!∗ ∗
! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗

120 j =0
DO i =0 ,N

X( i )=NEGINFINITY+(REAL( i ) /REAL(N) ) ∗ (B−NEGINFINITY)

X2D( j , i )=X( i )

END DO
128

DO i =0 ,N

WRITE(UNIT=13 ,FMT= ’( E12 . 6 ) ’ )X( i )

WRITE(UNIT=60 ,FMT= ’(4251 E21 . 6 ) ’ )X2D( j , i )

END DO

136 CALL BOUNDARY CONDITIONS(N, U, X, U2D, j , L)

DO i =1 ,(N−4)
HT( i ) =0.0D0
END DO

HT(N−3) =9.0D0 / 4 . 0 D0
HT(N−2)=−1.0D0
144 HT(N−1) =4.0D0

j =0
Tau =(0.5D0∗ ( Sigma ∗ ∗ 2 . 0 D0) ) ∗ ( 1 . 0 D0) ∗ (REAL( j ) /REAL(L) )
h=(B−NEGINFINITY) /N

72
 CALL CRANK NICOLSON(N, k , U, RHS, B, X, t , h , DelTau , Tau )
152

XN=XN+0.9034∗SQRT( ( 1 . 0 ∗ ( Sigma ∗ ∗ 2 . 0 D0) ) ∗ ( 1 . 0 D0) ∗ ( 1 . 0 D0/REAL(L) ) )

B=XN

DO i =0 ,N
X( i )=NEGINFINITY+(REAL( i ) /REAL(N) ) ∗ (B−NEGINFINITY)
END DO

160 DO i =0 ,N
WRITE(UNIT=13 ,FMT= ’( E12 . 6 ) ’ )X( i )
END DO

CALL TRIDIAG SOL(N, RHS, U, k , h , NEGINFINITY , X, U2D, j , L , S2D , V2D, t f i n a l )

DO j =1 ,L

168 h=(B−NEGINFINITY) /N

Tau=(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 1 . 0 D0) ∗ (REAL( j ) /REAL(L) )

CALL CRANK NICOLSON(N, k , U, RHS, B, X, t , h , DelTau , Tau )

CALL VEC(N, U, V, h )
CALL TRIDIAG SOL1 (N, RHS, Z , k , h )
176 CALL TRIDIAG SOL2 (N, V, Y, k , h )

CALL XN SOLVE(N, Y, Z ,HT,XN)

DO i =1 ,N−1

RHS( i )=RHS( i )−V( i ) ∗XN

184 END DO

h=(XN−NEGINFINITY) /N

73
 DO i =0 ,N

X( i )=NEGINFINITY+(REAL( i ) /REAL(N) ) ∗ (XN−NEGINFINITY)

192

X2D( j , i )=X( i )

END DO

DO i =0 ,N

200 WRITE(UNIT=13 ,FMT= ’( E12 . 6 ) ’ )X( i )

WRITE(UNIT=60 ,FMT= ’(4251 E21 . 6 ) ’ )X2D( j , i )

END DO

CALL TRIDIAG SOL(N, RHS, U, k , h , NEGINFINITY , X, U2D, j , L , S2D , V2D, t f i n a l )

208 B=XN
t=T f i n a l −(2.0∗ Tau/ sigma ∗ ∗ 2 )
SOUT=(exp (XN+(1−K2) ∗ tau ) ) ∗KK
WRITE(UNIT=70 ,FMT= ’( F12 . 6 ) ’ )SOUT
WRITE(UNIT=71 ,FMT= ’( F12 . 6 ) ’ ) Tau

216 END DO

WRITE(UNIT=15 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =5 ,5) , i =0 ,N)

WRITE(UNIT=16 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =5 ,5) , i =0 ,N)
WRITE(UNIT=17 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =10 ,10) , i =0 ,N)
WRITE(UNIT=18 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =10 ,10) , i =0 ,N)
WRITE(UNIT=19 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =20 ,20) , i =0 ,N)
WRITE(UNIT=20 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =20 ,20) , i =0 ,N)
224 WRITE(UNIT=21 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =30 ,30) , i =0 ,N)
WRITE(UNIT=22 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =30 ,30) , i =0 ,N)
WRITE(UNIT=23 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =40 ,40) , i =0 ,N)
WRITE(UNIT=24 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =40 ,40) , i =0 ,N)

74
 WRITE(UNIT=25 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =50 ,50) , i =0 ,N)
WRITE(UNIT=26 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =50 ,50) , i =0 ,N)
WRITE(UNIT=27 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =60 ,60) , i =0 ,N)
WRITE(UNIT=28 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =60 ,60) , i =0 ,N)
232 WRITE(UNIT=29 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =70 ,70) , i =0 ,N)
WRITE(UNIT=30 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =70 ,70) , i =0 ,N)
WRITE(UNIT=31 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =80 ,80) , i =0 ,N)
WRITE(UNIT=32 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =80 ,80) , i =0 ,N)
WRITE(UNIT=33 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =89 ,89) , i =0 ,N)
WRITE(UNIT=34 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =89 ,89) , i =0 ,N)
WRITE(UNIT=35 ,FMT= ’( F12 . 6 ) ’ ) ( (X2D( j , i ) , j =89 ,89) , i =0 ,N)
WRITE(UNIT=36 ,FMT= ’( F12 . 6 ) ’ ) ( (U2D( j , i ) , j =89 ,89) , i =0 ,N)
240

WRITE(UNIT=37 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =5 ,5) , i =0 ,N)

WRITE(UNIT=38 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =5 ,5) , i =0 ,N)
WRITE(UNIT=39 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =10 ,10) , i =0 ,N)
WRITE(UNIT=40 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =10 ,10) , i =0 ,N)
WRITE(UNIT=41 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =15 ,15) , i =0 ,N)
248 WRITE(UNIT=42 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =15 ,15) , i =0 ,N)
WRITE(UNIT=43 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =20 ,20) , i =0 ,N)
WRITE(UNIT=44 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =20 ,20) , i =0 ,N)
WRITE(UNIT=45 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =30 ,30) , i =0 ,N)
WRITE(UNIT=46 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =30 ,30) , i =0 ,N)
WRITE(UNIT=47 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =40 ,40) , i =0 ,N)
WRITE(UNIT=48 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =40 ,40) , i =0 ,N)
WRITE(UNIT=49 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =50 ,50) , i =0 ,N)
256 WRITE(UNIT=50 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =50 ,50) , i =0 ,N)
WRITE(UNIT=51 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =60 ,60) , i =0 ,N)
WRITE(UNIT=52 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =60 ,60) , i =0 ,N)
WRITE(UNIT=53 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =70 ,70) , i =0 ,N)
WRITE(UNIT=54 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =70 ,70) , i =0 ,N)

WRITE(UNIT=64 ,FMT= ’( F12 . 6 ) ’ ) ( ( S2D ( j , i ) , j =80 ,80) , i =0 ,N)

WRITE(UNIT=65 ,FMT= ’( F12 . 6 ) ’ ) ( (V2D( j , i ) , j =80 ,80) , i =0 ,N)
264

75
 CLOSE(UNIT=11)
CLOSE(UNIT=12)
CLOSE(UNIT=13)
CLOSE(UNIT=14)
272 CLOSE(UNIT=15)
CLOSE(UNIT=16)
CLOSE(UNIT=17)
CLOSE(UNIT=18)
CLOSE(UNIT=19)
CLOSE(UNIT=20)
CLOSE(UNIT=21)
CLOSE(UNIT=22)
280 CLOSE(UNIT=23)
CLOSE(UNIT=24)
CLOSE(UNIT=25)
CLOSE(UNIT=26)
CLOSE(UNIT=27)
CLOSE(UNIT=28)
CLOSE(UNIT=29)
CLOSE(UNIT=30)
288

CLOSE(UNIT=31)
CLOSE(UNIT=32)
CLOSE(UNIT=33)
CLOSE(UNIT=34)
CLOSE(UNIT=35)
CLOSE(UNIT=36)
CLOSE(UNIT=37)
296 CLOSE(UNIT=38)
CLOSE(UNIT=39)
CLOSE(UNIT=40)
CLOSE(UNIT=41)
CLOSE(UNIT=42)
CLOSE(UNIT=43)
CLOSE(UNIT=44)
CLOSE(UNIT=45)
304 CLOSE(UNIT=46)
CLOSE(UNIT=47)

76
 CLOSE(UNIT=48)
CLOSE(UNIT=49)

CLOSE(UNIT=64)
CLOSE(UNIT=65)

312

END PROGRAM AMERICAN OPTIONS707

! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
!∗ ∗
!∗ FUNCTIONS AND SUBROUTINES ∗
!∗ ∗
320 ! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗

SUBROUTINE BOUNDARY CONDITIONS(N, U, X, U2D, j , L)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(OUT) : : U
328 DOUBLE PRECISION,DIMENSION( 0 : L , 0 : N) ,INTENT(OUT) : : U2D
INTEGER: : i
INTEGER,INTENT(IN) : : j , L

DO i =0 ,N

U( i )=max ( 1 . 0 D0−EXP(X( I ) ) , 0 . 0 D0)

336

U2D( j , i )=U( i )

END DO

DO i =0 ,N

344 WRITE(UNIT=14 ,FMT= ’( E12 . 6 ) ’ )U( i )

77
 WRITE(UNIT=61 ,FMT= ’(4251 E21 . 6 ) ’ )U2D( j , i )

END DO

END SUBROUTINE BOUNDARY CONDITIONS

SUBROUTINE CRANK NICOLSON(N, k , U, RHS, B, X, t , h , DelTau , Tau )

352 IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,INTENT(IN) : : h , k , B, t , DelTau , Tau
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U,X
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : RHS
DOUBLE PRECISION,EXTERNAL: : G
DOUBLE PRECISION : : r , TEST
INTEGER: : i
360

r=k / ( h ∗ ∗ 2 . 0 D0)

DO i =1 ,N−1

RHS( i )=r ∗U( i −1) +(2.0D0−2.0D0∗ r ) ∗U( i )+r ∗U( i +1) +1.0∗ k ∗ (G(X( i ) , Tau )+G(X(
i ) , Tau+DelTau ) )&
+2.0D0∗ (REAL( i ) /REAL(N) ) ∗ ( (U( i )−U( i −1) ) /h ) ∗(−B)

368 END DO

END SUBROUTINE CRANK NICOLSON

SUBROUTINE VEC(N, U, V, h )
IMPLICIT NONE
INTEGER,INTENT(IN) : : N
376 DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : V
DOUBLE PRECISION,INTENT(IN) : : h

INTEGER: : i

78
 DO i =1 ,N−1
384

V( i ) =−2.0D0∗ (REAL( i ) /REAL(N) ) ∗ (U( i )−U( i −1) ) /h

END DO

END SUBROUTINE VEC

DOUBLE PRECISION FUNCTION G( x , Tau )

392 IMPLICIT NONE
DOUBLE PRECISION,INTENT(IN) : : x , Tau
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.5D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

G=(EXP(K1∗Tau ) ) ∗ ( ( ( K2−K1) ∗EXP( x−(K2−1.0D0) ∗Tau ) )+K1)

END FUNCTION G
400

SUBROUTINE TRIDIAG SOL(N, RHS, U, k , h , NEGINFINITY , X, U2D, j , L , S2D , V2D, t f i n a l )

IMPLICIT NONE
DOUBLE PRECISION,INTENT(IN) : : NEGINFINITY , h , k , t f i n a l
INTEGER,INTENT(IN) : : N, j , L
DOUBLE PRECISION,DIMENSION( 0 :N) : : Alpha , s , y
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(OUT) : : U
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : RHS
408 DOUBLE PRECISION,DIMENSION( 0 : L , 0 : N) ,INTENT(INOUT) : : U2D
DOUBLE PRECISION,DIMENSION( 0 : L , 0 : N) ,INTENT(INOUT) : : S2D , V2D
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) : : RES, SS
DOUBLE PRECISION : : a , b , c , Tau , r , Sv , t
INTEGER: : i , e

DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.5D0 ,D=0.8D0∗ I r ,KK=10.0D0

416 DOUBLE PRECISION,PARAMETER: : K1=(2.0D0∗ I r ) / ( sigma ∗ ∗ 2 . 0 D0) ,K2=(2.0D0∗ ( I r −D
) ) / ( sigma ∗ ∗ 2 . 0 D0)

Alpha =0.0
s =0.0

79
 y =0.0

r=k / ( h ∗ ∗ 2 )

424 a=r
b =(2.0D0+2.0D0∗ r )
c=r
Alpha ( 0 )=b
S ( 0 )=RHS( 1 )

Tau=(( sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 1 . 0 D0) ∗ (REAL( j ) /REAL(L) )

432

y ( 0 )=exp (K1∗Tau ) ∗ ( 1 . 0 D0−exp (X( 0 ) −(K2−1.0D0) ∗ tau ) )

DO i =1 ,(N−3)

Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )

S ( i )=RHS( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )
440

END DO

y (N−3)=S (N−3)/ Alpha (N−3)

y (N−2) =(4.0D0 / 9 . 0 D0) ∗y (N−3)
y (N−1) =(1.0D0 / 4 . 0 D0) ∗y (N−2)
y (N) =0.0

448 DO i =(N−4) ,1 , −1
y ( i ) =( s ( i )+c ∗y ( i +1) ) / Alpha ( i )
END DO

Sv =7.0D0

DO i =0 ,N
t=T f i n a l −(2.0∗ Tau/ sigma ∗ ∗ 2 )
456 U( i )=y ( i )

80
 END DO
DO i =0 ,N

SS ( i ) =(exp ( x ( i ) +(1−K2) ∗ tau ) ) ∗KK

464

S2D ( j , i )=SS ( i )
RES( i ) =((U( i ) ∗KK) / ( exp (K1∗ ( tau ) ) ) )+SS ( i )−KK

V2D( j , i )=RES( i )

472 U2D( j , i )=U( i )

END DO

DO i =0 ,N

WRITE(UNIT=62 ,FMT= ’( E12 . 6 ) ’ ) SS ( i )

480 WRITE(UNIT=63 ,FMT= ’( E12 . 6 ) ’ )RES( i )

WRITE(UNIT=61 ,FMT= ’(4251 E21 . 6 ) ’ )U2D( j , i )

WRITE(UNIT=14 ,FMT= ’( E12 . 6 ) ’ )U( i )

END DO

488

END SUBROUTINE TRIDIAG SOL

SUBROUTINE TRIDIAG SOL1 (N, RHS, Z , k , h )

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , S , y
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : Z
496 DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : RHS
DOUBLE PRECISION,INTENT(IN) : : k , h

81
 DOUBLE PRECISION : : a , b , c , r
INTEGER: : i

s =0.0
y =0.0
504

r=k / ( h ∗ ∗ 2 . 0 D0)

a=( r )
b =(2.0D0+2.0D0∗ r )
c=( r )
Alpha ( 1 )=b
S ( 1 )=RHS( 1 )
512

DO i =2 ,(N−1)

Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )

S ( i )=RHS( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

520 END DO

y (N−1)=S (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

y ( i ) =(S ( i )+c ∗y ( i +1) ) / Alpha ( i )

528 END DO

DO i =1 ,N−1

Z ( i )=y ( i )

END DO

536

82
 END SUBROUTINE TRIDIAG SOL1

SUBROUTINE TRIDIAG SOL2 (N, V, Y, k , h )

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,INTENT(IN) : : k , h
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , s ,U
544 DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : Y
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : V
DOUBLE PRECISION : : a , b , c , r
INTEGER: : i

s =0.0
552 Y=0.0
U=0.0

r=k / ( h ∗ ∗ 2 . 0 D0)

a=( r )
b =(2.0D0+2.0D0∗ r )
c=( r )
560

Alpha ( 1 )=b
S ( 1 )=V( 1 )

DO i =2 ,N−1

Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )

S ( i )=V( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )
568

END DO

U(N−1)=s (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

U( i ) =(S ( i )+c ∗U( i +1) ) / Alpha ( i )

83
576

END DO

DO i =1 ,N−1

Y( i )=U( i )

END DO
584

END SUBROUTINE TRIDIAG SOL2

SUBROUTINE XN SOLVE(N, Y, Z , HT,XN)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : Y, Z ,HT
592 DOUBLE PRECISION,INTENT(OUT) : : XN
DOUBLE PRECISION : : A, B
INTEGER: : i

A=0.0
B=0.0
DO i =1 ,N−1

600 A=A+Ht ( i ) ∗Z ( i )
B=B+Ht ( i ) ∗Y( i )

END DO

XN=A/B

END SUBROUTINE XN SOLVE

.2 Program 2 - pt2.f90
PROGRAM movingr
DOUBLE PRECISION : : K, NEGINFINITY ,XN, Delt , T f i n a l , t , h , A, C,KN, BN, l e f t ,
domain

84
 DOUBLE PRECISION,ALLOCATABLE: : U ( : ) ,UTEMP( : ) ,RHS ( : ) , Z ( : ) ,Y ( : ) ,HT ( : ) ,V ( : ) ,
X( : ) , r1 ( : ) , r2 ( : ) , r3 ( : )
INTEGER: : IOS , j , N, L , i
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

8 PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ CRANK NICOLSON SOLULTION ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ DIFFUSION EQUATION ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’
16 PRINT∗

NEGINFINITY=−30.0D0
T f i n a l =(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 ) ∗ 1 . 0

Domain=NEGINFINITY
24 N=300
L=NINT ( ( T f i n a l / ( Domain/REAL(N) ) ∗ ∗ 2 ) )
PRINT∗ , ( T f i n a l / ( Domain/REAL(N) ) ∗ ∗ 2 )

k=T f i n a l /L
L e f t =−20.0D0
N=100
32 t =0.0D0
D e l t =1.0D0/L
BN=0.0D0
KN=((BN−LEFT) /N∗ ∗ 2 . 0 D0)

85
 ALLOCATE(U( 0 :N) ,UTEMP( 0 :N) ,RHS ( 1 : N−1) ,V( 1 : N−1) , Z ( 1 : N−1) ,Y( 1 : N−1) ,HT( 1 : N
−1) ,X( 0 :N) ,&
40 r 1 ( 1 : N−1) , r 2 ( 1 : N−1) , r 3 ( 1 : N−1) )

U=0.0D0
UTEMP=0.0D0
RHS=0.0D0
V=0.0D0
Z=0.0D0
Y=0.0D0
48 HT=0.0D0
t =0.0D0

OPEN(UNIT=11 ,FILE=”x . dat ” ,IOSTAT=IOS )

OPEN(UNIT=12 ,FILE=”U. dat ” ,IOSTAT=IOS )

56 IF ( IOS/=0) THEN
PRINT∗ , ’ E r r o r Occured in Opening The Output File ’
STOP
END IF

! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
!∗ ∗
!∗ MAIN PROGRAM ∗
64 !∗ ∗
! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗

X( 0 )=LEFT
WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X( 0 )
DO i =1 ,N−1
72 X( i )=BN−(KN∗ (N−i ) ∗ ∗ 2 . 0 D0)
WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X( i )
END DO
X(N)=BN
WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X(N)

86
 CALL BOUNDARY CONDITIONS(N, U,X)

80 DO i =1 ,(N−4)
HT( i ) =0.0D0
END DO

HT(N−3) =9.0D0 / 4 . 0 D0
HT(N−2)=−1.0D0
HT(N−1) =4.0D0

88 t =1.0D0
XN=BN
DO j =0 ,L

t=t−d e l t

tau =(( sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 1 . 0 D0−t )

96 DO i =1 ,(N−1)
r 1 ( i ) =(( x ( i −1)−x ( i ) ) ∗ ( x ( i −1)−x ( i +1) ) )
r 2 ( i ) =(( x ( i )−x ( i −1) ) ∗ ( x ( i )−x ( i +1) ) )
r 3 ( i ) =(( x ( i +1)−x ( i −1) ) ∗ ( x ( i +1)−x ( i ) ) )
END DO

CALL CRANK NICOLSON(N, k , U, RHS, X, t , h , d e l t , r1 , r2 , r3 ,XN)

104 CALL VEC(N, U, V, h ,X)
CALL TRIDIAG SOL1 (N, RHS, Z , k , r1 , r2 , r 3 )
CALL TRIDIAG SOL2 (N, V, Y, k , h , r1 , r2 , r 3 )

CALL XN SOLVE(N, Y, Z ,HT,XN)

BN=XN

112

DO i =1 ,N−1
RHS( i )=RHS( i )−V( i ) ∗XN
END DO

87
 KN=((BN−LEFT) /N∗ ∗ 2 . 0 D0)

X( 0 )=LEFT
120 WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X( 0 )
DO i =1 ,N−1
X( i )=BN−(KN∗ (N−i ) ∗ ∗ 2 . 0 D0)
WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X( i )
END DO

X(N)=BN
WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X(N)
128

DO i =1 ,(N−1)
r 1 ( i ) =(( x ( i −1)−x ( i ) ) ∗ ( x ( i −1)−x ( i +1) ) )
r 2 ( i ) =(( x ( i )−x ( i −1) ) ∗ ( x ( i )−x ( i +1) ) )
r 3 ( i ) =(( x ( i +1)−x ( i −1) ) ∗ ( x ( i +1)−x ( i ) ) )

END DO

136 CALL TRIDIAG SOL(N, RHS, U, t , k , NEGINFINITY , r1 , r2 , r3 , x ) ! Solve for

new U

END DO

CLOSE(UNIT=11)
144 CLOSE(UNIT=12)

END PROGRAM movingr

! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
!∗ ∗
!∗ FUNCTIONS AND SUBROUTINES ∗
!∗ ∗
152 ! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗

88
 SUBROUTINE BOUNDARY CONDITIONS(N, U,X)
IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(OUT) : : U
160 INTEGER: : i

DO i =0 ,N
U( i )=max ( 1 . 0 D0−EXP(X( I ) ) , 0 . 0 D0)
WRITE(UNIT=12 ,FMT= ’( E12 . 6 ) ’ )U( i )
END DO

END SUBROUTINE BOUNDARY CONDITIONS

168

SUBROUTINE CRANK NICOLSON(N, k , U, RHS, X, t , h , d e l t , r1 , r2 , r3 ,XN)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,INTENT(IN) : : h , k , t , d e l t ,XN
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U,X
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : r1 , r2 , r 3
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : RHS
176 DOUBLE PRECISION,EXTERNAL: : G
DOUBLE PRECISION : : TEST, tp
INTEGER: : i

tp=t−d e l t

DO i =1 ,N−1
184 RHS( i ) =2.0∗ r 1 ( i ) ∗U( i −1) +(2.0D0+2.0∗ r 2 ( i ) ) ∗U( i ) +2.0∗ r 3 ( i ) ∗U( i +1)+(k ) ∗ (
G(X( i ) , t )+G(X( i ) , tp ) )&
+(1.0D0−(((N−i ) ∗ ∗ 2 . 0 D0) /N∗ ∗ 2 . 0 D0) ) ∗ ( (U( i )−U( i −1) ) / ( x ( i )−x ( i −1) ) ) ∗(−XN
) ∗2.0
! PRINT∗ , i ,RHS( i )

END DO

END SUBROUTINE CRANK NICOLSON

89
192 SUBROUTINE VEC(N, U, V, h ,X)
IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U,X
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : V
DOUBLE PRECISION,INTENT(IN) : : h

INTEGER: : i
200

DO i =1 ,N−1
V( i ) = −(2.0D0−(((N−i ) ∗ ∗ 2 . 0 D0) /N∗ ∗ 2 . 0 D0) ) ∗ (U( i )−U( i −1) ) / ( x ( i )−x ( i −1) )
END DO

END SUBROUTINE VEC

208 DOUBLE PRECISION FUNCTION G( x , t )

IMPLICIT NONE
DOUBLE PRECISION,INTENT(IN) : : x , t
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)
DOUBLE PRECISION : : Tau

Tau =(0.5D0∗ ( Sigma ∗ ∗ 2 . 0 D0) ) ∗ ( 1 . 0 D0−t )

216 G=(EXP(K1∗Tau ) ) ∗ ( ( ( K2−K1) ∗EXP( x−(K2−1.0D0) ∗Tau ) )+K1)

END FUNCTION G

SUBROUTINE TRIDIAG SOL(N, RHS, U, t , k , NEGINFINITY , r1 , r2 , r3 , x )

IMPLICIT NONE
DOUBLE PRECISION,INTENT(IN) : : NEGINFINITY , k , t
INTEGER,INTENT(IN) : : N
224 DOUBLE PRECISION,DIMENSION( 0 :N) : : Alpha , s , y
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(OUT) : : U
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : RHS, r1 , r2 , r 3
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : x
DOUBLE PRECISION : : a , b , c , tau , r

90
 INTEGER: : i

DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0

232 DOUBLE PRECISION,PARAMETER: : K1=(2.0D0∗ I r ) / ( sigma ∗ ∗ 2 . 0 D0) ,K2=(2.0D0∗ ( I r −D
) ) / ( sigma ∗ ∗ 2 . 0 D0)

Alpha =0.0D0
s =0.0D0
y =0.0D0

a =2.0∗ r 1 ( 2 )
240 b =(2.0D0−2.0∗ r 2 ( 2 ) )
c =2.0∗ r 3 ( 2 )
Alpha ( 1 )=b
S ( 1 )=RHS( 1 )

tau =(( sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 1 . 0 D0−t )

248

DO i =2 ,(N−3)

a =2.0∗ r 1 ( i )
b =(2.0D0−2.0∗ r 2 ( i ) )
c =2.0D0∗ r 3 ( i )
Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )
S ( i )=RHS( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )
256

END DO

y (N−3)=S (N−3)/ Alpha (N−3)

y (N−2) =(4.0D0 / 9 . 0 D0) ∗y (N−3)
y (N−1) =(1.0D0 / 4 . 0 D0) ∗y (N−2)
y (N) =0.0D0

264

DO i =(N−4) ,3 , −1
y ( i ) =( s ( i )+c ∗y ( i +1) ) / Alpha ( i )

91
 END DO

DO i =0 ,N
U( i )=y ( i )
272 WRITE(UNIT=12 ,FMT= ’( E12 . 6 ) ’ )U( i )
Print ∗ ,U( i )
END DO

END SUBROUTINE TRIDIAG SOL

SUBROUTINE TRIDIAG SOL1 (N, RHS, Z , k , r1 , r2 , r 3 )

280 IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , S ,Q
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : Z
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : RHS, r1 , r2 , r 3
DOUBLE PRECISION,INTENT(IN) : : k
DOUBLE PRECISION : : a , b , c , r
INTEGER: : i
288

s =0.0D0

a =2.0∗ r 1 ( 1 )
b =(2.0D0−2.0∗ r 2 ( 1 ) )
c =2.0D0∗ r 3 ( 1 )
296

Alpha ( 1 )=b
S ( 1 )=RHS( 1 )

DO i =2 ,(N−1)

a =2.0∗ r 1 ( i )
304 b =(2.0D0−2.0∗ r 2 ( i ) )
c =2.0∗ r 3 ( i )

92
 Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )
S ( i )=RHS( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

END DO

312 Q(N−1)=S (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

Q( i ) =(S ( i )+c ∗Q( i +1) ) / Alpha ( i )

END DO

320 DO i =1 ,N−1

Z ( i )=Q( i )

END DO

END SUBROUTINE TRIDIAG SOL1

328

SUBROUTINE TRIDIAG SOL2 (N, V, Y, k , h , r1 , r2 , r 3 )

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,INTENT(IN) : : k , h
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , s ,D
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : Y
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : V, r1 , r2 , r 3
336 DOUBLE PRECISION : : a , b , c , r
INTEGER: : i

s =0.0
Y=0.0

344 a =2.0∗ r 1 ( 1 )

93
 b=2.0 −2.0∗ r 2 ( 1 )
c =2.0∗ r 3 ( 1 )

Alpha ( 1 )=b
S ( 1 )=V( 1 )

DO i =2 ,N−1
352

a =2.0∗ r 1 ( i )
b=2.0D0−2.0∗ r 2 ( i )
c =2.0∗ r 3 ( i )

Alpha ( i )=b−(a ∗ c / Alpha ( i −1) )

S ( i )=V( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

360 END DO

D(N−1)=s (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

D( i ) =(S ( i )+c ∗D( i +1) ) / Alpha ( i )

368 END DO

DO i =1 ,N−1

Y( i )=D( i )

END DO

376

END SUBROUTINE TRIDIAG SOL2

SUBROUTINE XN SOLVE(N, Y, Z , HT,XN)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : Y, Z ,HT
DOUBLE PRECISION,INTENT(OUT) : : XN

94
384 DOUBLE PRECISION : : A, B
INTEGER: : i

A=0.0D0
B=0.0D0
DO i =1 ,N−1

A=A+Ht ( i ) ∗Z ( i )
392 B=B+Ht ( i ) ∗Y( i )

END DO

XN=A/B

END SUBROUTINE XN SOLVE

.3 Program 3 - pt3.f90
PROGRAM FINITE ELEMENTS V5n
DOUBLE PRECISION : : k , NEGINFINITY , Delt , T f i n a l , t , h ,SUMG1,SUMG2, Theta1 ,
ThetaNew , tau , SumC, Domain ,GINTERGRAL, DERIV,SUMG
DOUBLE PRECISION,ALLOCATABLE: : U ( : ) , Z ( : ) ,Y ( : ) ,X ( : ) ,G ( : ) ,KM( : , : ) ,KP ( : , : ) ,
Phi ( : ) , h V e c t o r ( : ) ,dPdX ( : ) , Cvec ( : ) , ThetaV ( : ) , f ( : ) ,&
W( : ) ,HT ( : ) ,RHS ( : ) ,M( : , : ) ,RHS1 ( : )
INTEGER: : IOS , j , N, L , i , q
DOUBLE PRECISION,EXTERNAL: : Uprime
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0
8 DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ FINITE ELEMENT SOLULTION ∗’
PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ DIFFUSION EQUATION ∗’
PRINT∗ , ’ ∗ ∗’
16 PRINT∗ , ’ ∗ ∗’
PRINT∗ , ’ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ∗ ’

95
 PRINT∗

NEGINFINITY=−0.05D0
T f i n a l =(Sigma ∗ ∗ 2 . 0 ) / 2 . 0 D0

Domain=NEGINFINITY
24 N=100
L=NINT ( ( T f i n a l / ( Domain/N) ∗ ∗ 2 ) )

k=( T f i n a l /L)

D e l t =( T f i n a l ) /L

32

t =0.0D0
GINTERGRAL=0.0
Theta1 =0.0D0
B=0.0D0
f =0.0D0
40 RHS=0.0D0

ALLOCATE(U( 0 :N+1) ,G( 0 :N) , Z ( 1 : N−1) ,Y( 1 :N) ,X( 0 :N+1) ,KM( 1 : N, 0 : N+1) ,KP( 1 : N
− 1 , 0 :N) ,&
Phi ( 0 :N) , h v e c t o r ( 0 :N) ,dPdX ( 0 :N) , Cvec ( 1 : N−1)&
, ThetaV ( 0 :N) , f ( 1 : N−1) ,W( 0 :N) ,HT( 1 :N) ,RHS ( 1 : N−1) ,M( 1 : N− 1 , 0 :N) ,RHS1 ( 1 : N
−1) )

SumC=0.0D0
U=0.0D0
48 Y=0.0D0
Z=0.0D0
Y=0.0D0
KM=0.0D0
KP=0.0D0
t =0.0D0
f =0.0D0

96
 M=0.0D0
56

OPEN(UNIT=11 ,FILE=”x . dat ” ,IOSTAT=IOS )

OPEN(UNIT=12 ,FILE=”U. dat ” ,IOSTAT=IOS )
OPEN(UNIT=13 ,FILE=”K. dat ” ,IOSTAT=IOS )
OPEN(UNIT=14 ,FILE=”KP. dat ” ,IOSTAT=IOS )
OPEN(UNIT=15 ,FILE=”G. dat ” ,IOSTAT=IOS )
OPEN(UNIT=16 ,FILE=” Phi . dat ” ,IOSTAT=IOS )
64 OPEN(UNIT=17 ,FILE=”M. dat ” ,IOSTAT=IOS )
OPEN(UNIT=18 ,FILE=”C . dat ” ,IOSTAT=IOS )
OPEN(UNIT=19 ,FILE=”SumTheta . dat ” ,IOSTAT=IOS )
OPEN(UNIT=20 ,FILE=”Y. dat ” ,IOSTAT=IOS )
OPEN(UNIT=21 ,FILE=”dPdX . dat ” ,IOSTAT=IOS )
OPEN(UNIT=22 ,FILE=”NewTheta . dat ” ,IOSTAT=IOS )
OPEN(UNIT=23 ,FILE=”Z . dat ” ,IOSTAT=IOS )
OPEN(UNIT=24 ,FILE=”W. dat ” ,IOSTAT=IOS )
72 OPEN(UNIT=25 ,FILE=”G1 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=26 ,FILE=”G2 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=27 ,FILE=”Y1 . dat ” ,IOSTAT=IOS )
OPEN(UNIT=28 ,FILE=” kk . dat ” ,IOSTAT=IOS )
OPEN(UNIT=29 ,FILE=” S t h e t a . dat ” ,IOSTAT=IOS )
OPEN(UNIT=30 ,FILE=” Yold . dat ” ,IOSTAT=IOS )
OPEN(UNIT=31 ,FILE=”MassMul . dat ” ,IOSTAT=IOS )
IF ( IOS/=0) THEN
80 PRINT∗ , ’ E r r o r Occured in Opening The Output File ’
STOP
END IF

! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗
!∗ ∗
!∗ MAIN PROGRAM ∗
!∗ ∗
88 ! ∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗∗

WRITE(UNIT=28 ,FMT= ’( E12 . 6 ) ’ ) k

DO i =0 ,N

97
 X( i )=NEGINFINITY+(REAL( i ) /REAL(N) ) ∗ (B−NEGINFINITY)

96 WRITE(UNIT=11 ,FMT= ’( E12 . 6 ) ’ )X( i )

END DO

CALL INITIAL DATA (N, U,X)

104

HT( 1 ) =1.0D0
HT( 2 ) =−1.0D0

DO i =3 ,N

HT( i ) =0.0D0
112

END DO

t =1.0D0

DO i =0 ,N−1

120 CALL Theta (X, ThetaV , N, i , t ,U)

Theta1=Theta1+ThetaV ( i )

WRITE(UNIT=19 ,FMT= ’( E12 . 6 ) ’ ) Theta1

END DO
WRITE(UNIT=29 ,FMT= ’( E12 . 6 ) ’ ) Theta1
128

DO q=1 ,1

GINTEGRAL=0.0

98
 Tau=(( sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 1 . 0 D0) ∗ (REAL( q ) /REAL(L) )

136 DO i =1 ,N−1

CALL C v e c t o r (X, Cvec , N, i , t , U, Theta1 )

WRITE(UNIT=18 ,FMT= ’( E12 . 6 ) ’ ) Cvec ( i )

SumC=SumC+Cvec ( i )

144 END DO

DO i =1 ,N−1

CALL SIMPSONS1(X,SUMG1, N, i , q , L)

CALL SIMPSONS2(X,SUMG2, N, i , q , L)
152

G( i )=SUMG1+SUMG2

WRITE(UNIT=15 ,FMT= ’( E12 . 6 ) ’ )G( i )

END DO

G(N)=G(N−1)
160

DO i =1 ,N−1

CALL MASS MATRIX( i , N, X,KM)

CALL MASS MATRIX PRIME( i , N, X, KP,U)

END DO
168 KP( 1 , 1 )=KP( 1 , 1 ) / 2 . 0 D0

RHS1 ( 1 )=U( 1 ) ∗KM( 1 , 1 )+U( 2 ) ∗KM( 1 , 2 )

99
 DO j =2 ,N−2

176 RHS1( j )=U( j −1)∗KM( j , j −1)+U( j ) ∗KM( j , j )+U( j +1)∗KM( j , j +1)

WRITE(UNIT=31 ,FMT= ’( E14 . 8 ) ’ ) RHS1( j )

END DO

RHS1(N−1)=U(N−2)∗KM(N−1 ,N−2)+U(N−1)∗KM(N−1 ,N−1)

DO j =1 ,N−1
184

Y( j )=G( j )−RHS1( j )

WRITE(UNIT=27 ,FMT= ’( E14 . 8 ) ’ )Y( j )

END DO

192 DO j =1 ,N−1

f ( j )=Y( j )

END DO

200 DO i =1 ,N

h v e c t o r ( i )=X( i )−X( i −1)

END DO

208 CALL TRIDIAG SOL1 (N, f , Z , k ,KP)

CALL TRIDIAG SOL2 (N, Cvec ,W, k ,KP)

100
 CALL THETA SOLVE(N,W, Z , HT, ThetaNew )
WRITE(UNIT=22 ,FMT= ’( E14 . 8 ) ’ ) ThetaNew

216 DO j =1 ,N−1

WRITE(UNIT=30 ,FMT= ’( E14 . 8 ) ’ )Y( j )

END DO

Do i =1 ,N−1

224

CALL I n t e g r a t e G (X,SUMG, N, i , q , L)

GINTERGRAL=GINTERGRAL+SUMG

END DO

DERIV=−UPrime (NEGINFINITY , Tau )+GINTERGRAL

232

PRINT∗ ,THETANEW, ’ New Theta ’

DO j =1 ,N−1

240 Y( j )=Y( j ) −(THETANEW) ∗ Cvec ( j )

WRITE(UNIT=20 ,FMT= ’( E14 . 8 ) ’ )Y( j )

END DO

CALL TRIDIAG SOL(N, Y, KP, Phi )

248

DO i =1 ,N

101
 IF ( i /=N) THEN

dPdX( i ) =(((X( i )−X( i −1) ) ∗ ( Phi ( i )−Phi ( i −1) ) +(X( i +1)−X( i ) ) ∗ ( Phi (
i +1)−&
Phi ( i ) ) ) ) / ( (X( i )−X( i −1) ) +(X( i +1)−X( i ) ) )

256

ELSE

dPdX( i ) =(Phi ( i )−Phi ( i −1) ) / (X( i )−X( i −1) )

END IF

END DO
264 dPdX ( 0 ) =0.0D0

DO i =0 ,N

CALL EULERS(N, X, h v e c t o r , k , dPdX)

END DO

272 DO i =0 ,N

END DO

DO i =1 ,N−1

CALL M MATRIX( i , N, X,M)

280

END DO

DO i =1 ,N−1

RHS( i )=ThetaNew∗ Cvec ( i )

102
288

END DO

CALL TRIDIAG SOL3 (N,M, U, RHS, tau , NEGINFINITY)

DO i =0 ,N

! PRINT∗ , i ,U( i )
296 !WRITE(UNIT=12 ,FMT= ’(E12 . 6 ) ’ )U( i )
END DO
! Theta1=ThetaNew
END DO

WRITE(UNIT=13 ,FMT= ’(101 F14 . 8 ) ’ ) ( ( (KM( j , i ) ) , i =0 ,N) , j =1 ,N−1)

WRITE(UNIT=14 ,FMT= ’(101 F14 . 8 ) ’ ) ( ( (KP( j , i ) ) , i =0 ,N) , j =1 ,N−1)
WRITE(UNIT=21 ,FMT= ’( E12 . 6 ) ’ ) ( ( dPdX( i ) ) , i =0 ,N)
304 WRITE(UNIT=17 ,FMT= ’(101 F14 . 8 ) ’ ) ( ( (M( j , i ) ) , i =0 ,N) , j =1 ,N−1)

CLOSE(UNIT=11)
CLOSE(UNIT=12)
CLOSE(UNIT=13)
CLOSE(UNIT=14)
CLOSE(UNIT=15)
CLOSE(UNIT=16)
312 CLOSE(UNIT=17)
CLOSE(UNIT=18)
CLOSE(UNIT=19)
CLOSE(UNIT=20)
CLOSE(UNIT=21)
CLOSE(UNIT=22)
CLOSE(UNIT=23)
CLOSE(UNIT=24)
320 CLOSE(UNIT=25)
CLOSE(UNIT=26)
CLOSE(UNIT=27)
CLOSE(UNIT=28)
CLOSE(UNIT=29)
CLOSE(UNIT=30)
CLOSE(UNIT=31)

103
328 END PROGRAM FINITE ELEMENTS V5n

SUBROUTINE INITIAL DATA (N, U,X)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(OUT) : : U
INTEGER: : i
336

DO i =0 ,N

U( i )=max ( 1 . 0 D0−EXP(X( I ) ) , 0 . 0 D0)

WRITE(UNIT=12 ,FMT= ’( E12 . 6 ) ’ )U( i )

END DO
344

END SUBROUTINE INITIAL DATA

SUBROUTINE I n t e g r a t e G (X,SUMG, N, i , q , L)
IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i , q , L
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(IN) : : X
DOUBLE PRECISION,INTENT(OUT) : : SUMG
352 DOUBLE PRECISION : : h , XI0 , XI2 , XI1 , a , b , XI ,SUM, Tau
DOUBLE PRECISION,EXTERNAL: : GF
DOUBLE PRECISION,PARAMETER: : Sigma =0.4D0
INTEGER: : k ,M

Tau=(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 0 . 5 D0) ∗ (REAL( q ) /REAL(L) )

SUMG=0.0
360

M=INT (N∗ 2 0 0 )
a=X( i −1)
b=X( i )

h=ABS( (X( i )−X( i −1) ) ) /M

104
 XI0=GF( b , Tau )+GF( a , Tau )
368 XI1 =0.0D0
XI2 =0.0D0

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

376

XI2=XI2+GF( XI , Tau )
ELSE
XI1=XI1+GF( XI , Tau )

END IF
END DO

384 SUMG=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

END SUBROUTINE I n t e g r a t e G

SUBROUTINE SIMPSONS1(X,SUMG1, N, i , q , L)
392 IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i , q , L
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(IN) : : X
DOUBLE PRECISION,INTENT(OUT) : : SUMG1
DOUBLE PRECISION : : h , XI0 , XI2 , XI1 , a , b , XI ,SUM, Tau
DOUBLE PRECISION,EXTERNAL: : GF
DOUBLE PRECISION,PARAMETER: : Sigma =0.4D0
INTEGER: : k ,M
400

Tau=(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 0 . 5 D0) ∗ (REAL( q ) /REAL(L) )

M=N∗200
a=X( i −1)

105
 b=X( i )

h=ABS( (X( i )−X( i −1) ) ) /M

408

XI0=(b−X( i −1) ) ∗GF( b , Tau ) +(a−X( i −1) ) ∗GF( a , Tau )

XI1 =0.0D0
XI2 =0.0D0

DO k=1 ,M−1

XI=a+k∗h
416

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+(XI−X( i −1) ) ∗GF( XI , Tau )

ELSE
XI1=XI1+(XI−X( i −1) ) ∗GF( XI , Tau )

END IF
424 END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUMG1=SUM∗ ( 1 . 0 D0/ (X( i )−X( i −1) ) )

WRITE(UNIT=25 ,FMT= ’( E12 . 6 ) ’ )SUMG1

432 END SUBROUTINE SIMPSONS1

SUBROUTINE SIMPSONS2(X,SUMG2, N, i , q , L)
IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i , q , L
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(IN) : : X
DOUBLE PRECISION,INTENT(OUT) : : SUMG2
DOUBLE PRECISION : : h , XI0 , XI2 , XI1 , a , XI , b ,SUM, Tau
440 DOUBLE PRECISION,PARAMETER: : Sigma =0.4
DOUBLE PRECISION,EXTERNAL: : GF
INTEGER: : k ,M

106
 Tau=(( Sigma ∗ ∗ 2 . 0 D0) / 2 . 0 D0) ∗ ( 0 . 5 D0) ∗ (REAL( q ) /REAL(L) )

a=X( i )
b=X( i +1)
448 M=N∗200
h=ABS( (X( i +1)−X( i ) ) ) /M

XI0=(X( i +1)−a ) ∗GF( a , Tau ) +(X( i +1)−b ) ∗GF( b , Tau )

XI1 =0.0D0
XI2 =0.0D0

DO k=1 ,M−1
456

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+(X( i +1)−XI ) ∗GF( XI , Tau )

ELSE
XI1=XI1+(X( i +1)−XI ) ∗GF( XI , Tau )
464

END IF
END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUMG2=SUM∗ ( 1 . 0 D0/ (X( i +1)−X( i ) ) )

WRITE(UNIT=26 ,FMT= ’( E12 . 6 ) ’ )SUMG2

472

END SUBROUTINE SIMPSONS2

DOUBLE PRECISION FUNCTION GF( x , tau )

480 IMPLICIT NONE
DOUBLE PRECISION,INTENT(IN) : : x , tau
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0

107
 DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

488 GF=(EXP(K1∗Tau ) ) ∗ ( ( ( K2−K1) ∗EXP( x−(K2−1.0D0) ∗Tau ) )+K1)

END FUNCTION GF

DOUBLE PRECISION FUNCTION UPrime ( x , Tau )

IMPLICIT NONE
496 DOUBLE PRECISION,INTENT(IN) : : x , Tau
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

UPrime=−exp (K1∗Tau ) ∗ ( exp ( x−(K2−1)∗Tau ) )

504

END FUNCTION Uprime

SUBROUTINE MASS MATRIX( i , N, X,KM)

IMPLICIT NONE
512 INTEGER,INTENT(IN) : : N, i
DOUBLE PRECISION,DIMENSION( 0 :N+1) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 1 : N, 0 : N+1) ,INTENT(INOUT) : :KM
DOUBLE PRECISION : : SUM1, SUM2, SUM3, SUM4,SUM
DOUBLE PRECISION,EXTERNAL: : f u n c
DOUBLE PRECISION : : a , b , h , XI , XI1 , XI2 , XI0
INTEGER: : k ,M

108
520 a=X( i −1)
b=X( i )
M=INT (N∗ 2 0 0 )
h=ABS( (X( i )−X( i −1) ) ) /M

XI0=f u n c ( a )+f u n c ( b )

XI1 =0.0D0
528 XI2 =0.0D0
XI =0.0D0

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

536

XI2=XI2+f u n c ( XI )
ELSE
XI1=XI1+f u n c ( XI )

END IF
END DO

544 SUM=h ∗ ( ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0)

SUM1= ( ( 1 . 0D0/ (X( i )−X( i −1) ) ) )

a=X( i )
b=X( i +1)

552 h=ABS( (X( i +1)−X( i ) ) ) /M

XI0=f u n c ( a )+f u n c ( b )
XI1 =0.0D0
XI2 =0.0D0
XI =0.0

109
 DO k=1 ,M−1
560

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+f u n c ( XI )
ELSE
XI1=XI1+f u n c ( XI )
568

END IF

END DO

SUM=(h ) ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM2=(1.0D0/ (X( i +1)−X( i ) ) )

576 KM( i , i )=SUM1+SUM2

a=X( i )
b=X( i +1)

h=ABS( (X( i +1)−X( i ) ) ) /M

XI0=f u n c ( a )+f u n c ( b )
584 XI1 =0.0D0
XI2 =0.0D0
XI =0.0

DO k=1 ,M−1

XI=a+k∗h

592 IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+f u n c ( XI )
ELSE
XI1=XI1+f u n c ( XI )

110
 END IF

600 END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM3= −1.0∗((1.0D0/ (X( i +1)−X( i ) ) ∗ ∗ 2 ) ) ∗SUM
KM( i , i +1)=−1.0D0 ∗ ( 1 . 0 D0/ (X( i +1)−X( i ) ) )

a=X( i −1)
b=X( i )
608

h=ABS( (X( i )−X( i −1) ) ) /M

XI0=f u n c ( a )+f u n c ( b )
XI1 =0.0D0
XI2 =0.0D0
XI =0.0D0

616 DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+f u n c ( XI )
ELSE
624 XI1=XI1+f u n c ( XI )

END IF

END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM4=−1.0D0 ∗ ( ( 1 . 0 D0/ (X( i )−X( i −1) ) ) ∗ ∗ 2 ) ∗SUM
632 KM( i , i −1)=−1.0D0 ∗ ( 1 . 0 D0/ (X( i )−X( i −1) ) )

END SUBROUTINE MASS MATRIX

111
 DOUBLE PRECISION FUNCTION f u n c ( x )
IMPLICIT NONE
640 DOUBLE PRECISION,INTENT(IN) : : x

f u n c =1.0D0

END FUNCTION f u n c

DOUBLE PRECISION FUNCTION UINTER1(UN,UM, UP, Xin , i , a , b )

IMPLICIT NONE
648 INTEGER,INTENT(IN) : : i
DOUBLE PRECISION,INTENT(IN) : : UN,UM,UP
DOUBLE PRECISION,INTENT(IN) : : a , b
DOUBLE PRECISION,INTENT(IN) : : Xin

UINTER1=UM∗ ( Xin−a ) / ( b−a )+UN∗ ( ( b−Xin ) / ( b−a ) )

END FUNCTION UINTER1

656

DOUBLE PRECISION FUNCTION UINTER2(UN,UM, UP, Xin , i , a , b )

IMPLICIT NONE
INTEGER,INTENT(IN) : : i
DOUBLE PRECISION,INTENT(IN) : : UN,UM,UP
DOUBLE PRECISION,INTENT(IN) : : a , b
DOUBLE PRECISION,INTENT(IN) : : Xin

664 UINTER2=UM∗ ( b−Xin ) / ( b−a )+UP∗ ( Xin−a ) / ( b−a )

END FUNCTION UINTER2

DOUBLE PRECISION FUNCTION UINTER(UP,UM, Xin , i , a , b )

IMPLICIT NONE
INTEGER,INTENT(IN) : : i
DOUBLE PRECISION,INTENT(IN) : : UP,UM
672 DOUBLE PRECISION,INTENT(IN) : : a , b
DOUBLE PRECISION,INTENT(IN) : : Xin

UINTER=UM∗ ( b−Xin ) / ( b−a )+UP∗ ( ( Xin−a ) / ( b−a ) )

112
 END FUNCTION UINTER

SUBROUTINE MASS MATRIX PRIME( i , N, X, KP,U)

680 IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U
DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(INOUT) : : KP
DOUBLE PRECISION : : SUM1, SUM2, SUM3, SUM4, e , f , g

DOUBLE PRECISION : : a , b , h , XI , XI1 , XI2 , XI0 ,UN,UM, UP, TEST,SUM

688 DOUBLE PRECISION,EXTERNAL: : UINTER1 , UINTER2
INTEGER: : k ,M

UN=U( i −1)
UM=U( i )
UP=U( i +1)

a=X( i −1)
696 b=X( i )
M=N∗200
h=ABS( (X( i )−X( i −1) ) ) /M

XI0=UINTER1(UN,UM, UP, a , i , a , b )+UINTER1(UN,UM, UP, b , i , a , b )

XI1 =0.0D0
XI2 =0.0D0
XI =0.0
704

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+UINTER1(UN,UM, UP, XI , i , a , b )
712 ELSE
XI1=XI1+UINTER1(UN,UM, UP, XI , i , a , b )

113
 END IF
END DO

SUM=(h ) ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM1=SUM∗ ( ( 1 . 0 D0/ (X( i )−X( i −1) ) ) ∗ ∗ 2 . 0 D0)
720

b=X( i +1)
a=X( i )

h=(X( i +1)−X( i ) ) /M
XI0=UINTER2(UN,UM, UP, a , i , a , b )+UINTER2(UN,UM, UP, b , i , a , b )
XI1 =0.0D0
XI2 =0.0D0
728 XI =0.0D0

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

736 XI2=XI2+UINTER2(UN,UM, UP, XI , i , a , b )

ELSE
XI1=XI1+UINTER2(UN,UM, UP, XI , i , a , b )

END IF

END DO

744 SUM=(h ) ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM2=SUM∗ ( ( 1 . 0 D0/ (X( i +1)−X( i ) ) ) ∗ ∗ 2 . 0 D0)

KP( i , i )=SUM1+SUM2
f=KP( i , i )

b=X( i +1)
a=X( i )
752

h=ABS( (X( i +1)−X( i ) ) ) /M

114
 XI0=UINTER2(UN,UM, UP, a , i , a , b )+UINTER2(UN,UM, UP, b , i , a , b )
XI1 =0.0D0
XI2 =0.0D0
XI =0.0D0

760 DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+UINTER2(UN,UM, UP, XI , i , a , b )
ELSE
768 XI1=XI1+UINTER2(UN,UM, UP, XI , i , a , b )

END IF

END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM3=SUM∗( −1.0D0) ∗ ( ( 1 . 0 D0/ (X( i +1)−X( i ) ) ) ∗ ∗ 2 . 0 D0)
776 KP( i , i +1)=SUM3
g=KP( i , i +1)

b=X( i )
a=X( i −1)

h=ABS( (X( i )−X( i −1) ) ) /M

784 XI0=UINTER1(UN,UM, UP, a , i , a , b )+UINTER1(UN,UM, UP, b , i , a , b )

XI1 =0.0D0
XI2 =0.0D0
XI =0.0D0

DO k=1 ,M−1

XI=a+k∗h
792

115
 IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+UINTER1(UN,UM, UP, XI , i , a , b )
TEST=UINTER1(UN,UM, UP, XI , i , a , b )

ELSE
XI1=XI1+UINTER1(UN,UM, UP, XI , i , a , b )
800

END IF

END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM4=SUM∗ ( ( − 1 . 0D0) ∗ ( ( 1 . 0 D0/ (X( i )−X( i −1) ) ) ∗ ∗ 2 . 0 D0) )
KP( i , i −1)=SUM4
808 e=KP( i , i −1)

END SUBROUTINE MASS MATRIX PRIME

SUBROUTINE TRIDIAG SOL(N, Y, KP, Phi )

IMPLICIT NONE
816 INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 :N) : : Alpha , S , C
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(INOUT) : : Phi
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : Y
DOUBLE PRECISION,DIMENSION( 0 :N) : : Z
DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(IN) : : KP
DOUBLE PRECISION : : a , b
INTEGER: : i
824

S=0.0D0
a=−KP( 1 , 0 )
b=KP( 1 , 1 )
c ( 1 )=−KP( 1 , 2 )

Alpha ( 1 )=b

116
832 S ( 1 )=Y( 1 )

DO i =2 ,(N−1)

a=−KP( i , i −1)
b=KP( i , i )
c ( i )=−KP( i , i +1)
840

Alpha ( i )=b−(a ∗ c ( i −1)/ Alpha ( i −1) )

S ( i )=Y( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

END DO

Z (N−1)=S (N−1)/ Alpha (N−1)

848 DO i =(N−2) ,1 , −1

Z ( i ) =(S ( i )+c ( i ) ∗Z ( i +1) ) / Alpha ( i )

END DO
Z (N)=Z (N−1)
Z ( 0 )=Z ( 1 )

856 DO i =0 ,N

Phi ( i )=Z ( i )

WRITE(UNIT=16 ,FMT= ’( E12 . 6 ) ’ ) Phi ( i )

864

END DO

END SUBROUTINE TRIDIAG SOL

SUBROUTINE EULERS(N, X, h v e c t o r , k , dPdX)

117
 IMPLICIT NONE
872 INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(INOUT) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : dPdX
DOUBLE PRECISION,INTENT(IN) : : k
DOUBLE PRECISION,DIMENSION( 1 ,N−1) ,INTENT(IN) : : h v e c t o r
INTEGER: : i

DO i =0 ,N
880

X( i )=X( i )+k∗dPdX( i )

END DO

END SUBROUTINE EULERS

888

SUBROUTINE C v e c t o r (X, Cvec , N, i , t , U, Theta1 )

IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i
DOUBLE PRECISION,INTENT(IN) : : t
DOUBLE PRECISION,INTENT(IN) : : Theta1
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U
896 DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(INOUT) : : Cvec
DOUBLE PRECISION,EXTERNAL: : UINTER1 , UINTER2
DOUBLE PRECISION : : h , XI0 , XI2 , XI1 , a , XI , SUMC1, SUMC2,UN,UM, UP, b ,SUM, TEST
INTEGER: : k ,M

M=N∗300
a=X( i −1)
b=X( i )
904 h=ABS( (X( i )−X( i −1) ) ) /M

UN=U( i −1)
UM=U( i )
UP=U( i +1)

118
 XI0=(a−X( i −1) ) ∗UINTER1(UN,UM, UP, a , i , a , b ) +(b−X( i −1) ) ∗UINTER1(UN,UM, UP
, b , i , a , b)
912 XI1 =0.0D0
XI2 =0.0D0

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

920

XI2=XI2+(XI−X( i −1) ) ∗UINTER1(UN,UM, UP, XI , i , a , b )

ELSE
XI1=XI1+(XI−X( i −1) ) ∗UINTER1(UN,UM, UP, XI , i , a , b )

END IF
END DO
928

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUMC1=(1.0D0/ (X( i )−X( i −1) ) ) ∗SUM

b=X( i +1)
a=X( i )
936

h=ABS( (X( i +1)−X( i ) ) ) /M

XI0=(X( i +1)−a ) ∗UINTER2(UN,UM, UP, a , i , a , b ) +(X( i +1)−b ) ∗UINTER2(UN,UM, UP

, b , i , a , b)
XI1 =0.0D0
XI2 =0.0D0
944

DO k=1 ,M−1

119
 XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2+(X( i +1)−XI ) ∗UINTER2(UN,UM, UP, XI , i , a , b )

952 ELSE
XI1=XI1+(X( i +1)−XI ) ∗UINTER2(UN,UM, UP, XI , i , a , b )

TEST=UINTER2(UN,UM, UP, XI , i , a , b )

END IF
END DO

960 SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUMC2=(1.0D0/ (X( i +1)−X( i ) ) ) ∗SUM

Cvec ( i ) =(SUMC1+SUMC2) / Theta1

END SUBROUTINE C v e c t o r

968

SUBROUTINE Theta (X, ThetaV , N, i , t ,U)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i
DOUBLE PRECISION,INTENT(IN) : : t
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : U
976 DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(INOUT) : : ThetaV
DOUBLE PRECISION,EXTERNAL: : UINTER
DOUBLE PRECISION : : h , XI0 , XI2 , XI1 , a , XI , SUMC1, UP,UM, b , t e s t
INTEGER: : k ,M

M=(N∗ 3 0 0 0 )
a=X( i )
b=X( i +1)
984 h=ABS( (X( i +1)−X( i ) ) ) /M

120
 UP=U( i +1)
UM=U( i )

XI0=UINTER(UP,UM, a , i , a , b )+UINTER(UP,UM, b , i , a , b )
992 XI1 =0.0D0
XI2 =0.0D0

DO k=1 ,M−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

1000

XI2=XI2+UINTER(UP,UM, XI , i , a , b )

ELSE
XI1=XI1+UINTER(UP,UM, XI , i , a , b )

END IF
END DO
1008

SUMC1=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

ThetaV ( i )=SUMC1

END SUBROUTINE Theta

SUBROUTINE TRIDIAG SOL1 (N, f , Z , k ,KP)

1016 IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , S , y , C
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : : Z
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : f
DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(IN) : : KP
DOUBLE PRECISION,INTENT(IN) : : k
DOUBLE PRECISION : : a , b , r
1024 INTEGER: : i

121
 s =0.0D0
y =0.0D0

1032 a=−KP( 1 , 0 )
b=KP( 1 , 1 )
c ( 1 )=−KP( 1 , 2 )

Alpha ( 1 )=b
S ( 1 )=f ( 1 )

1040 DO i =2 ,(N−1)

a=−KP( i , i −1)
b=KP( i , i )
c ( i )=−KP( i , i +1)

Alpha ( i )=b−(a ∗ c ( i −1)/ Alpha ( i −1) )

1048 S ( i )=f ( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

END DO

y (N−1)=S (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

1056 y ( i ) =(S ( i )+c ( i ) ∗y ( i +1) ) / Alpha ( i )

END DO

DO i =1 ,N−1

Z ( i )=y ( i )
WRITE(UNIT=23 ,FMT= ’( E12 . 6 ) ’ ) Z ( i )

122
1064 END DO

END SUBROUTINE TRIDIAG SOL1

SUBROUTINE TRIDIAG SOL2 (N, Cvec ,W, k ,KP)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
1072 DOUBLE PRECISION,INTENT(IN) : : k
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Alpha , s , U, C
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(OUT) : :W
DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(IN) : : KP
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : Cvec
DOUBLE PRECISION : : a , b , r
INTEGER: : i

1080

s =0.0D0
W=0.0D0
U=0.0D0

1088 a=−KP( 1 , 0 )
b=KP( 1 , 1 )
c ( 1 )=−KP( 1 , 2 )

Alpha ( 1 )=b
S ( 1 ) =(Cvec ( 1 ) /k )

DO i =2 ,N−1
1096

a=−KP( i , i −1)
b=KP( i , i )
c ( i )=−KP( i , i +1)

Alpha ( i )=b−(a ∗ c ( i −1)/ Alpha ( i −1) )

S ( i ) =(Cvec ( i ) /k ) +(a ∗S ( i −1)/ Alpha ( i −1) )

123
1104 END DO

U(N−1)=s (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1

U( i ) =(S ( i )+C( i ) ∗U( i +1) ) / Alpha ( i )

1112 END DO

DO i =1 ,N−1

W( i )=U( i )
WRITE(UNIT=24 ,FMT= ’( E12 . 6 ) ’ )W( i )
END DO

1120

END SUBROUTINE TRIDIAG SOL2

SUBROUTINE THETA SOLVE(N,W, Z ,HT, ThetaNew )

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : :W, Z ,HT
DOUBLE PRECISION,INTENT(OUT) : : ThetaNew
1128 DOUBLE PRECISION : : A, B
INTEGER: : i

A=0.0D0
B=0.0D0
DO i =1 ,N−1

A=A+Ht ( i ) ∗Z ( i )
1136 B=B+Ht ( i ) ∗W( i )

END DO

ThetaNew=A/B

124
 END SUBROUTINE THETA SOLVE

1144 SUBROUTINE TRIDIAG SOL3 (N,M, U, RHS, tau , NEGINFINITY)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N
DOUBLE PRECISION,DIMENSION( 1 :N) : : Alpha , S , C
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(INOUT) : : U
DOUBLE PRECISION,DIMENSION( 1 : N−1) ,INTENT(IN) : : RHS
DOUBLE PRECISION,DIMENSION( 1 : N−1) : : Z
DOUBLE PRECISION,INTENT(IN) : : tau , NEGINFINITY
1152 DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(IN) : :M
DOUBLE PRECISION,PARAMETER: : I r =0.03D0 , Sigma =0.2D0 ,D=0.8D0∗ I r ,KK=10.0D0
DOUBLE PRECISION,PARAMETER: : K1=2.0D0∗ I r / ( Sigma ∗ ∗ 2 . 0 D0) ,K2=2.0D0∗ ( I r −D) / (
sigma ∗ ∗ 2 . 0 D0)

DOUBLE PRECISION : : a , b
INTEGER: : i

1160 S=0.0D0
a=M( 1 , 0 )
b=M( 1 , 1 )
c ( 1 )=M( 1 , 2 )
Alpha =0.0D0

Alpha ( 1 )=b
S ( 1 )=RHS( 1 )
1168

DO i =2 ,(N−1)

a=M( i , i −1)
b=M( i , i )
c ( i )=M( i , i +1)

1176 Alpha ( i )=b−(a ∗ c ( i −1)/ Alpha ( i −1) )

S ( i )=RHS( i ) +(a ∗S ( i −1)/ Alpha ( i −1) )

END DO

125
 Z (N−1)=S (N−1)/ Alpha (N−1)

DO i =(N−2) ,1 , −1
1184

Z ( i ) =(S ( i )+c ( i ) ∗Z ( i +1) ) / Alpha ( i )

END DO

DO i =1 ,N−1

U( i )=Z ( i )
1192

END DO
U( 0 )=exp (K1∗ tau ) ∗ ( 1 . 0 D0−exp (NEGINFINITY−(K2−1.0D0) ∗ tau ) )
U(N) =0.0

END SUBROUTINE TRIDIAG SOL3

1200

SUBROUTINE M MATRIX( i , N, X,M)

IMPLICIT NONE
INTEGER,INTENT(IN) : : N, i
DOUBLE PRECISION,DIMENSION( 0 :N) ,INTENT(IN) : : X
DOUBLE PRECISION,DIMENSION( 1 : N− 1 , 0 :N) ,INTENT(INOUT) : :M
DOUBLE PRECISION : : SUM1, SUM2, SUM3, SUM4,SUM
1208 DOUBLE PRECISION,EXTERNAL: : f u n c
DOUBLE PRECISION : : a , b , h , XI , XI1 , XI2 , XI0
INTEGER: : k , MI

a=X( i −1)
b=X( i )
MI=i n t (N∗ 2 0 0 )
h=ABS( (X( i )−X( i −1) ) ) /MI
1216

XI0=(b−a ) ∗∗2+(a−a ) ∗∗2

XI1 =0.0D0

126
 XI2 =0.0D0
XI =0.0

DO k=1 ,MI−1

1224 XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2 +((XI−a ) ∗ ∗ 2 )
ELSE
XI1=XI1 +((XI−a ) ∗ ∗ 2 )

1232 END IF
END DO

SUM=(h ) ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM1=((( b∗∗3−a ∗ ∗ 3 ) / 3 )+b ∗ ( a ∗ ∗ 2 )−a ∗ ( b ∗ ∗ 2 ) ) / ( (X( i )−X( i −1) ) ∗ ∗ 2 )

a=X( i )
b=X( i +1)
1240

h=ABS( (X( i +1)−X( i ) ) ) /MI

XI0=(b−b ) ∗∗2+(b−a ) ∗∗2

XI1 =0.0D0
XI2 =0.0D0
XI =0.0

1248 DO k=1 ,MI−1

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2 +((b−XI ) ∗ ∗ 2 )
ELSE
1256 XI1=XI1 +((b−XI ) ∗ ∗ 2 )

127
 END IF

END DO

SUM=(h ) ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM2=((( b∗∗3−a ∗ ∗ 3 ) / 3 . 0 D0) +(( a ∗ ∗ 2 ) ∗b−a ∗ ( b ∗ ∗ 2 ) ) ) / ( (X( i +1)−X( i ) ) ∗ ∗ 2 )
1264

M( i , i )=SUM1+SUM2

a=X( i )
b=X( i +1)

1272 h=ABS( (X( i +1)−X( i ) ) ) /MI

XI0=(b−b ) ∗ ( b−a ) +(b−a ) ∗ ( a−a )

XI1 =0.0D0
XI2 =0.0D0
XI =0.0

DO k=1 ,MI−1
1280

XI=a+k∗h

IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2 +((b−XI ) ∗ ( XI−a ) )

ELSE
XI1=XI1 +((b−XI ) ∗ ( XI−a ) )
1288

END IF

END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM3=SUM∗ ( ( 1 . 0 D0/ (X( i +1)−X( i ) ) ) ∗ ∗ 2 )
M( i , i +1)=−1.0D0 ∗ ( ( ( b∗∗3−a ∗ ∗ 3 ) / 6 . 0 D0) +(( a ∗ ∗ 2 ) ∗b−a ∗ ( b ∗ ∗ 2 ) ) / 2 . 0 D0) / ( (X(
i +1)−X( i ) ) ∗ ∗ 2 )

128
1296

a=X( i −1)
b=X( i )

h=ABS( (X( i )−X( i −1) ) ) /MI

XI0=(b−a ) ∗ ( a−a ) +(b−b ) ∗ ( b−a )

1304 XI1 =0.0D0
XI2 =0.0D0
XI =0.0

DO k=1 ,MI−1

XI=a+k∗h

1312 IF (MOD( k , i n t ( 2 ) ) /=0) THEN

XI2=XI2 +((b−XI ) ∗ ( XI−a ) )

ELSE
XI1=XI1 +((b−XI ) ∗ ( XI−a ) )

END IF

1320 END DO

SUM=h ∗ ( XI0 +2.0D0∗ XI2 +4.0∗ XI1 ) / 3 . 0 D0

SUM4= ( ( 1 . 0D0/ (X( i )−X( i −1) ) ) ∗ ∗ 2 ) ∗SUM
M( i , i −1)=−1.0D0 ∗ ( ( ( b∗∗3−a ∗ ∗ 3 ) / 6 . 0 D0) +(( a ∗ ∗ 2 ) ∗b−a ∗ ( b ∗ ∗ 2 ) ) / 2 . 0 D0) / ( (X(
i +1)−X( i ) ) ∗ ∗ 2 )

END SUBROUTINE M MATRIX

129