MAST90059 Stochastic Calculus with

Applications
Ai-hua Xia (Room 220, Richard Berry Building)

General info
2 lectures + 1 prac per week
Pracs from Friday 3rd March
Handouts will be posted on the LMS every Monday.
Office hours: Monday 4:00 p.m. 5:00 p.m., Friday
4:15 p.m. 5:15 p.m.

[Week 1 1]
 Three levels of learning
I see, I forget
I read, I know
I do, I understand

[Week 1 2]
 Topic 1. Review of infinitesimal calculus
1.1 Continuous and Differentiable Functions
A function g is called continuous at the point t0 if

g(t0 ) := g(t0 + t) g(t0 ) 0 as t 0.

If g is continuous at every point of its domain of

definition, it is simply called continuous.
g is differentiable at the point t0 if
g(t0 )
lim =C
t0 t
exists and we write C = g 0 (t0 ) and g(t0 ) Ct.
If g is differentiable at every point of its domain, it is
called differentiable.

[Week 1 3]
 Differentiability implies continuity but not vice versa.

(Mean value theorem) If f is continuous on [a, b] and

has a derivative on (a, b), there is a c (a, b) such that

f (b) f (a) = f 0 (c)(b a).

The function g(t) = t is not differentiable at 0.

[Week 1 4]
 Weierstrass 1872

[31/10/181519/2/1897]

for 0 t 2, the function

X cos(3n t)
f (t) =
2n
n=1

is cont but nowhere differentiable.

Continuity: uniformly convergence of continuous
functions.
Diff: roughly, diff. term by term, a divergent series!

[Week 1 5]
 Notations
f continuous: C function

f differentiable with a continuous derivative: C 1 function

f is twice differentiable with a continuous second

derivative: C 2 function.

[Week 1 6]
 Right and left continous

lim g(t) = g(t0 ) RC, lim g(t) = g(t0 ) LC

tt0 tt0

C implies LC and RC

LC version of g: g(t) = limst g(s)

RC version of g: g(t+) = limst g(s)

g is LC if g(t) = g(t) and g is RC if g(t) = g(t+).

Def. t is discontinuity of the first kind or a jump point if

both g(t+) and g(t) exist and are not equal. The jump
at t is defined as g(t) := g(t+) g(t). Any other
discontinuity is the second kind.

[Week 1 7]
 Example Classify the discontinuity at t = 0 for
sin(1/t) for t 6= 0,
f (t) =
0 for t = 0.

Theorem A function defined on an interval [a, b] can

have no more than countably many jumps.
jumps 6= discontinuities!

Theorem If f is differentiable with a finite derivative

f 0 (t) in an interval, then at all points f 0 (t) is either
continuous or has a discontinuity of the second kind.

[Week 1 8]
 Regular functions
Def A function is called regular if it does not have
discontinuities of the second kind, i.e. it has both left and
right limits at any point of the domain and has one-sided
limits at the boundaries.

Def D = D[0, T ]: the class of functions on [0, T ] which

are right continuous with left limits (or c`adl`ag in short).
The class includes C
The discontinuities of g D are jumps. g has no more
than countably many discontinuities.

g(t): in calculus, g(t) = g(t + t) g(t); in stochastic

calculus, it means the jump size of the jump at t!

[Week 1 9]
 1.2 Variation of a function
The variation of g over the interval [a, b] is
n
X
Vg ([a, b]) = sup |g(tni ) g(tni1 )|,
i=1

sup is over all partitions a = tn0 < tn1 < < tnn = b.
Adding points increases the sum, so
n
X
Vg ([a, b]) = lim
n
|g(tni ) g(tni1 )|
k{ti }k0
i=1

where k{tni }k = max1in (tni tni1 ) and the limit is over

all partitions.
If Vg ([a, b]) is finite, we say g is of finite variation on [a, b].
The variation function of g is Vg (t) = Vg ([0, t]).

[Week 1 10]
 Vg (t) is non-decreasing.

Def g is of finite variation if Vg (t) < for all t; g is

bounded variation if supt Vg (t) < .

Example If g is an increasing function, find Vg .

Example If g is decreasing, find Vg .

Example If g is differentiable with continuous derivative

0
Rt 0
g , g(0) = 0 and 0 |g (s)|ds < , then
Rt 0
Vg (t) = 0 |g (s)|ds.

Example g is pure jump c`

adl`
ag function, then
P
Vg (t) = 0st |g(s)|.

[Week 1 11]
 Example The function g(t) = t sin(1/t), t > 0 and
g(0) = 0 is continuous on [0, 1], differentiable at all points
except 0. It has infinite variation on [0, a], a > 0. Take
the partition at 1/(2k + /2), 1/(2k /2),
k = 1, 2, . . .
Theorem (Jordan Decomposition, cf Hahn Decom.) Any
g : [0, ) R of finite variation can be expressed as the
difference of two increasing functions g(t) = a(t) b(t).
One possible decomposition:

a(t) = Vg (t), b(t) = Vg (t) g(t).

Another:
1 1
g(t) = (Vg (t) + g(t)) (Vg (t) g(t)).
2 2

[Week 1 12]
 The sum, the difference and the product of functions of
finite variation are also of finite variation.

The ratio of functions of finite variation with

denominator having the modulus larger than a positive
constant is also of finite variation.

Theorem A finite variation function can have no more

than countably many discontinuities and all
discontinuities are jumps.
In fact, a finite variation function is differentiable
almost everywhere (w.r.t. the Lebesgue measure).

Theorem If g is continuous, g 0 exists and

R 0
|g (s)|ds < , then g is of finite variation.

[Week 1 13]
 Decomposition of a function
For a right-continuous increasing function g, let
X X
d
g (t) = (g(s)g(s)) = g(s), g c (t) = g(t)g d (t).
st 0<st

Then g d is a pure jump function (discontinuous part of g),

g c is continuous (continuous part of g) and g = g c + g d .

For any finite variation function g, it can be decomposed

into the sum of its discontinuous and continuous parts,
and the decomposition is unique in the sense that any
two such decompositions differ by a constant.

[Week 1 14]
 Quadratic variation
For a function g, its quadratic variation over the interval [0, t]
is
Xn
n n 2
[g](t) = limn
(g(ti ) g(t i1 )) ,
k{ti }k0
i=1
if exists, where the limit is taken over all partitions
0 = tn0 < tn1 < < tnn = t.

[g](t) is non-decreasing in t, hence of finite variation.

If g is continuous and of finite variation then its

quadratic variation is 0.

Standard calculus does not need quadratic variation.

[Week 1 15]
 Quadratic covariation
For functions f and g on [0, t], the quadratic covariation is
n1
X
[f, g](t) = lim
n
(f (tni+1 ) f (tni ))(g(tni+1 ) g(tni ))
k{ti }k0
i=0

if exists, where the limit is taken over all partitions {tni } of [0, t].
If f is continuous and g is of finite variation, then [f, g](t) = 0.
Covariation is symmetric: [f, g] = [g, f ].
Covariation is bilinear: [af + bg, h] = a[f, h] + b[g, h]
Polarization identity If f , g have quadratic variation, then
1
[f, g](t) = {[f + g, f + g](t) [f, f ](t) [g, g](t)}.
2
Hence [f, g] is of finite variation.

[Week 1 16]
 1.3 Riemann integral and Stielties integral

[Bernhard Riemann
17/09/182620/07/1866]

[Thomas Stieltjes
29/12/185631/12/1894]

[Week 1 17]
 The Riemann integral of f over [a, b] is defined as
Z b Xn
n n n
f (t)dt = lim
n
f ( i )(t i t i1 )
a k{ti }k
i=1

if the limit exists, where a = tn0 < tn1 < < tnn = b and
tni1 in tni .

If f is continuous or discontinuous at finitely many

Rb
points, a f (t)dt is well-defined.

[Week 1 18]
 The fundamental theorem of calculus
If f is differentiable on [a, b] and f 0 is Riemann integrable on
[a, b] then
Z b
f (b) f (a) = f 0 (s)ds.
a
It can NOT
be applied to discontinuous functions:
0 for 0 t < 1,
f (t) =
2 for 1 t 2

[Week 1 19]
 Stieltjes integral
The Stieltjes integral of f wrt a monotone fn g over (a, b] is
Z b Z b Xn
n n n
f dg := f (s)dg(s) := lim
n
f ( i )(g(ti ) g(ti1 )).
a a k{ti }k0
i=1

If g is of finite variation and

Z b Z b
|f (s)||dg(s)| := |f (s)|dVg (s) < ,
a a

then f is Stieltjes integrable wrt g and the integral is defined as

Z Z Z
f (s)dg(s) = f (s)dVg (s) f (s)d(Vg (t) g(t)).
(a,b] (a,b] (a,b]
Rb R
a
f (s)dg(s) =
f (s)dg(s) (a,b]
Rb R
a dg(s) = g(b) g(a), (a,b) dg(s) = g(b) g(0)

[Week 1 20]
 a.k.a. the Riemann-Stieltjes integral: if we take g(t) = t,
we get the Riemann integral.
0
Rt 0
If g exists and g(t) = g(0) + 0 g (s)ds, then
Z b Z b
f (s)dg(s) = f (s)g 0 (s)ds.
a a

If g is a pure jump fn with jumps at si (a, b] and jump

P
size hi , 1 i m, i.e. g(t) = si t hi , then
Z b m
X
f (t)dg(t) = f (si )hi .
a i=1

[Week 1 21]
 If f is Stieltjes integrable wrt g of finite variation, the
variation of the integral is
Z b Z b
VR b f (s)dg(s) ([a, b]) = |f (s)||dg(s)| = |f (s)|dVg (s).
a a a
R /2
Example Find 0 f (t)dg(t) when f (t) = sin t and
g(t) = 2t3 , t [0, /2];



0 for t < /6,

for /6 t < /4,

1
g(t) =


3 for /4 t < /3,


4.5 for /3 t /2.

[Week 1 22]
 Functions of infinite variation
If g is of infinite variation, then the limit
n
X
lim
n
f (in )(g(tni ) g(tni1 ))
k{ti }k0
i=1

may not exist for some continuous functions f .

In stochastic calculus, it is essential to work on such

integrals and well see how to overcome the difficulty
later.

[Week 1 23]
 Integration by parts
Let f and g be fns of finite variation, h(s) = h(s) h(s),
then

f (b)g(b) f (a)g(a)
Z b Z b X
= f (s)dg(s) + g(s)df (s) + f (s)g(s)
a a a<sb
Z b Z b
= f (s)dg(s) + g(s)df (s)
a a

where the sum is countable as there are at most countably

many jumps for a finite variation fn. If g is continuous, then
Z b Z b
f (b)g(b) f (a)g(a) = f (s)dg(s) + g(s)df (s).
a a

[Week 1 24]
 Change of variables
Let f C 1 , g be continuous and of finite variation, then
Z t Z g(t)
f (g(t)) f (g(0)) = f 0 (g(s))dg(s) = f 0 (u)du.
0 g(0)

If g has jumps and is right continuous then

f (g(t)) f (g(0))
Z t X 
0 0

= f (g(s))dg(s) + f (g(s)) f (g(s)) f (g(s))g(s) .
0 0<st

Its called It
os formula.

[Week 1 25]
 Lebesgues integration

[Henri Lebesgue 28/06/1875

26/07/1941]

Riemann and Stieltjes integrals dont preserve the monotone

convergence property.
Riemann sum is to divide the domain of integration into small
subintervals.

[Week 1 26]
 Leb sum is to divide the range of function [c, d] into small
subintervals c = y0 < y1 < < yn = d and
n1
X
yk length({t : yk f (t) < yk+1 }).
k=0

The limit of the Leb sums is the Leb integral.

It is more general than Riemann-Stieltjes integral and
preserves the convergence.
Can be generalized to an abstract probability space.

[Week 1 27]
 Differentials and integrals
f is differentiable, we write

df (t) = f 0 (t)dt.

If g is differentiable, then f (g(t)) is differentiable with the chain

rule
df (g(t)) = f 0 (g(t))g 0 (t)dt = f 0 (g(t))dg(t).

For f , w differentiable, we often need to solve for the differential

equation
df (t) = (t)dw(t)
or Z t
f (t) = f (0) + (s)dw(s).
0
When w is a non-differentiable function we need a stochastic
integral and then solve the stochastic integral equations.

[Week 1 28]
 Taylors formula
If f has derivatives up to order n + 1, then
n
X f (i) (x0 )
f (x) = f (x0 ) + (x x0 )i + Rn (x, x0 ),
i!
i=1

where f (i) is the ith derivative and R is the remainder

1
Rn (x, x0 ) = f (n+1) (n )(xx0 )n+1 , n (x0 x, x0 x).
(n + 1)!

We mostly need

0 1 00
f (x) = f (x0 )+f (x0 )(xx0 )+ f ()(xx0 )2 , (x0 x, x0 x).
2

[Week 1 29]
 Higher dimension
A function f (x1 , . . . , xn ) is differentiable at x = (x1 , . . . , xn ) if
n
X
f (x) = Ci xi + o(),
i=1
pPn
2 o()
where = i=1 (xi ) and lim0 = 0.
If f is differentiable at x, then it is differentiable wrt any
xi while the other coordinates are kept fixed and is called
partial derivative f /xi and
n
X f
df (x1 , . . . , xn ) = (x1 , . . . , xn )dxi .
xi
i=1

The existence of all partial derivatives does not guarantee

the differentiability.

[Week 1 30]
 If all partial derivatives exist and continuous, then f is
differentiable and Ci = f /xi .

If f : Rn R has continuous partial derivatives up to

order two, we say f C 2 .

Taylors formula If f C 2 , by considering a function of

one variable h(t) = f (x + tx) for 0 t 1, we have
n n n
X f 1 X X 2f
f (x+x) f (x)+ (x)dxi + (x+x)dxi dxj ,
xi 2 xi xj
i=1 i=1 j=1

where (0, 1) and dxi = xi .

[Week 1 31]
 Lipschitz and H
older conditions
f satisfies a H
older condition (H older continuous) of
order (0, 1] on [a, b] or R if there is a constant K > 0
such that

|f (x) f (y)| K|x y| , x, y [a, b] or R.

When = 1, its called Lipschitz condition.

If f is Holder continuous of order , then it is also Holder
continuous of order < .
These conditions are needed for solving differential
equations.

Example f (x) = x, x [0, ) is H older continuous
with = 12 but is not Lipschitz.

[Week 1 32]
 More terminologies
f is smooth on [a, b] if it is continuous on [a, b] and
possesses continuous derivative f 0 on (a, b) such that the
(finite) limits f 0 (a+) and f 0 (b) exist.

f is piecewise continuous on [a, b] if it is continuous on

[a, b] except possibly a finite number of points at which
right-hand and left-hand (finite) limits exist.

f is piecewise smooth on [a, b] if it is piecewise continuous

on [a, b], f 0 exists (except a finite number of points) and
is also piecewise continuous on [a, b].

[Week 1 33]
 Growth conditions
f satisfies the linear growth condition if

|f (x)| K(1 + |x|).

The condition says for large x, the growth is no more

than linear, for small x, it is bounded.

The condition is needed to ensure the existence and

uniqueness of differential equations.

[Week 1 34]
 Gronwalls inequality
Let f , g and h be non-negative on [0, T ], and suppose that
Z t
f (t) g(t) + h(s)f (s)ds, 0 t T.
0

Then
Z t Z t 
f (t) g(t) + h(s)g(s) exp h(u)du ds, 0 t T.
0 s

[Week 1 35]
 First order linear differential eq

dx(t)
+ g(t)x(t) = k(t).
dt
Linear in the unknown function and its derivative.

Integrating Factor Method Multiply both sides by

the integrating factor eG(t) , where G satisfies
G0 (t) = g(t), we can get
Z t 
x(t) = eG(t) eG(s) k(s) ds + x(0)eG(0)G(t) .
0

[Week 1 36]