Notes

Abhinav Pradeep
October 24, 2023

1 Multivariate Limits
1.1 Continuity in calculus of one variable
To prove f is continuous at c,

lim f = lim− f
x→c+ x→c
+
For c ,

∀ϵ > 0, ∃δ > 0 s.t for 0 < x − c < δ

|f (x) − L| < ϵ
Analogously, for c− ,

∀ϵ > 0, ∃δ > 0 s.t for − δ < x − c < 0

|f (x) − L| < ϵ

1.2 Limits in calculus of one variable

f :R→R
To prove that limx→c f (x) = L

∀ϵ > 0, ∃δ > 0 s.t for 0 < |x − c| < δ

|f (x) − L| < ϵ

1.3 Euclidean distance in R2

For P, Q ∈ R2 , where P = (P1 , P2 ), Q = (Q1 , Q2 )
p
D(P, Q) = (P1 − Q1 )2 + (P1 − Q1 )2

1
1.4 Euclidean distance in Rn
For P, Q ∈ Rn , where P = (P1 , P2 , P3 ...Pn ), Q = (Q1 , Q2 , Q3 ...Qn )
v
u n
uX
D(P, Q) = t (Pi − Qi )2
i=1

1.5 Convergence of a sequence Rn

NOTE: SEQUENCES ARE DENOTED WITH SUPERSCRIPT

Consider (xi )i∈N ∈ Rn , limi→∞ (xi )i∈N = X ∈ Rn if

∀ϵ > 0, ∃I ∈ N s.t ∀i ≥ I,

D(xi , X) < ϵ

1.6 Continuity of multivariate functions

f : Rn → Rm
To prove f is continuous at X,

∀ sequences (xi )i∈N such that:

lim (xi )i∈N = X

i→∞

we have that:

lim f (xi ) = f (X)

i→∞

Similar process for

f : D → Rm
Where

D ⊆ Rn
Considering the set of all sequences that converge at X is equivalent to considering the set of all
directions possible ∈ Rn or ∈ D that X can be approached from. Hence, where the single variable case
considers approaches to X + and X − , the multivariate case considers approaches from all directions
possible in abstract Rn
1.7 Limits of multivariate functions
f : Rn → Rm
To prove

lim f (x) = L
x→C

∀ϵ > 0, ∃δ > 0 s.t for 0 < D(x, C) < δ

D(f (x), L) < ϵ

1.8 Limit laws for multivariate functions

Generalized triangle inequality

1.9 Smoothness

2 Multivariate Derivatives
NOTE: || : Rn → R where for V ∈ Rn , |V | = D(0, V )

2.1 Differentiability
f : Rn → Rm is differentiable at X ∈ Rn if ∃T such that:

|f (x + h) − f (x) − T h|
lim =0
h→0 |h|
Where h ∈ Rn

2.2 Directional Derivatives

For f : Rn → R and f differentiable on Rn , derivative in the direction of unit vector V ∈ Rn is defined
as:

|f (x + V h) − f (x)|
DV f (x) = lim
h→0 |h|
Where h ∈ Rn

When extended to f : Rn → Rm , where f can we written as f = (f1 , f2 , f3 , ...fm ), it can be proved

that DV f exists iff DV fk exists ∀k ∈ [1, m]. In this case,

DV f = (DV f1 , DV f2 , DV f3 , ...DV fm )
2.3 Partial Derivatives
Partial derivatives are a special case of DV f where V is a standard basis vector such as ei , ej , ek ... etc.
In this case it is written as:
∂f
Di f ⇔
∂xi

2.4 Total Differentiability

2.5 Total Derivative
For f : Rn → Rm the total derivative can be defined:

2.5.1 As an n × m matrix
 
D1 f
 D2 f 
Df =  .. 
 
 . 
Dn f
Which forms :
 
D1 f1 D1 f2 D1 f3 ... D1 fm
 D2 f1 D2 f2 D2 f3 ... D2 fm 
Df =  ..
 
.. .. .. .. 
 . . . . . 
Dn f1 Dn f2 Dn f3 ... Dn fm

2.5.2 As an m × n matrix
 
Df = (D1 f )T (D2 f )T (D3 f )T ... (Dn f )T
Which forms :
 
D1 f1 D2 f1 D3 f1 ... Dn f1
 D1 f2 D2 f2 D3 f2 ... Dn f2 
Df =  ..
 
.. .. .. .. 
 . . . . . 
D1 fm D2 fm D3 fm ... Dn fm
Both forms are valid?(ish) but m × n form is preferred

In such case Df is also called the Jacobian matrix Jf (⃗x) where ⃗x ∈ Rn :

 
D1 f1 (⃗x) D2 f1 (⃗x) D3 f1 (⃗x) ... Dn f1 (⃗x)
 D1 f2 (⃗x) D2 f2 (⃗x) D3 f2 (⃗x) ... Dn f2 (⃗x) 
Jf (⃗x) = 
 
.. .. .. .. .. 
 . . . . . 
D1 fm (⃗x) D2 fm (⃗x) D3 fm (⃗x) ... Dn fm (⃗x)

For f : Rn → R the total derivative, which is called the gradient ∇, can be defined as:

∇f = (D1 f, D2 f, D3 f, ...Dn f )
2.6 Derivative Laws
2.6.1 Linearity
Consider f : Rn → Rm , g : Rn → Rm which are totally differentiable at x ∈ Rn

f + g : Rn → Rm
Is totally differentiable at x and

D(f (x) + g(x)) = Df (x) + Dg(x)

Alternatively, in Jacobian notation,

Jf +g (x) = Jf (x) + Jg (x)

λf : Rn → Rm , λ ∈ R
Is totally differentiable at x and

D(λf (x)) = λDf (x)

Alternatively, in Jacobian notation,

Jλf (x) = λJf (x)

2.6.2 Chain rule

2.7 Higher Order Derivatives

2.8 Using ∇ := ( δxδ 1 , δxδ 1 , ... δxδ n )
There are three ways the operator ∇ can act:
1. On a scalar function f : Rn → R : ∇f (the gradient);
2. On a vector function f : Rn → Rn , via the dot product: ∇ · f (the divergence);
3. On a vector function f : Rn → Rn , via the cross product: ∇ × f (the curl)

2.8.1 Gradient: (∇f ) : Rn → Rn

For f : Rn → R
The gradient ∇f points in the direction of maximum increase of the function f .

The magnitude |∇f | gives the slope (rate of increase) along this maximal direction

2.8.2 Divergence: (∇ · f ) : Rn → R
Identifies sources and sinks:
2.8.3 Curl: (∇ × f ) : Rn → Rn

3 Tangent Plane
Consider f : R2 → R
Let z = f (x, y)

The plane tangent to z = f (x, y) at some point P⃗ = (a, b, f (a, b)) is of the form:

⃗n(⃗r) = ⃗n(P⃗ )
Where ⃗n is the vector normal to z = f (x, y) at P⃗ . This would be:

⃗n = (1, 0, Dx f ) × (0, 1, Dy f )
Where (1, 0, Dx f ) denotes how much f (x, y) changes in the x direction and (0, 1, Dy f ) denotes how
much f (x, y) changes in the y direction.

Alternatively, a (1, 0, Dx f ) and (0, 1, Dy f ) can be obtained by considering F : R2 → R3 as:

 
x
F (x, y) =  y 
f (x, y)
This communicates the same idea as original f (x, y). In this case,

Dx F = (1, 0, Dx f )
Dy F = (0, 1, Dy f )
Either way,
 
i j k
⃗n = det 1 0 Dx f 
0 1 Dy f

⃗n = −(Dx f )i − (Dy f )j + k

⃗n = (−Dx f, −Dy f, 1)
Let ⃗r = (x, y, z)

⃗n(⃗r) = ⃗n(P⃗ )

(−Dx f, −Dy f, 1)(x, y, z) = (−Dx f, −Dy f, 1)(a, b, f (a, b))

−(Dx f )x − (Dy f )y + z = −(Dx f )a − (Dy f )b + f (a, b)

z = (Dx f )x − (Dx f )a + (Dy f )y − (Dy f )b + f (a, b)

z = (Dx f )(x − a) + (Dy f )(y − b) + f (a, b)

4 Approximations: Taylor series

4.1 Multi-index notation
n-dimensional multi-index:

α ∈ Nn0
ie. n-dimensional vector/n-tuple of natural numbers + 0. Elements denoted by:

α = (α1 , α2 , α3 , ...αn )

4.1.1 Rules
For n-dimensional α
n
X
|α| = αi
i=1
n
Y
α! = αi !
i=1
n
Y
n
For x ∈ R , x = α
xαi i
i=1

4.1.2 Applications to derivtives

For f : Rn → R
∂f
Where Di f ⇔ ∂xi
,

α ∂ |α| f
D f= n
Q αi
i=1 ∂xi
4.2 General Taylor Series for Rn → R
For f : Rn → R
X Dα f (x)
T (h) = · hα
α!
|α|≤k

Where the summations can alternatively be written as matrix products. For example, consider T (h)
of order 2:
X Dα f (x)
T2 (h) = · hα
α!
|α|≤2

X Dα f (x) X Dα f (x) X Dα f (x)

T2 (h) = · hα + · hα + · hα
α! α! α!
|α|=0 |α|=1 |α|=2

X Dα f (x) X Dα f (x)
T2 (h) = f (x) + · hα + · hα
α! α!
|α|=1 |α|=2

Only one possibility for |α| = 1: α is an n-tuple containing one entry with value 1 and all other entries
are 0. In such case, α! = 1
X X Dα f (x)
T2 (h) = f (x) + Dα f (x) · hα + · hα
α!
|α|=1 |α|=2

Summing over all such α is equivalently:

n
X X Dα f (x)
T2 (h) = f (x) + Di f (x) · hj + · hα
i=1
α!
|α|=2

This summation is equivalently:

X Dα f (x)
T2 (h) = f (x) + Jf (x) · h + · hα
α!
|α|=2

Or, as f : Rn → R
X Dα f (x)
T2 (h) = f (x) + ∇f · h + · hα
α!
|α|=2

α of for that satisfy |α| = 2 are of the form:

Contains only one 2:

α = (0, 0, 0, ...2, ...0, 0, 0)

Contains two 1:

α = (0, 1, 0, ...1, ...0, 0, 0)

Summing over case 1 is equivalent to:
 n
1X
Di Di f (x) · h2i
2 i=1
Summing over case 2 is equivalent to:
n
1X
Dj Di f (x) · hi · hj
2 i̸=j

The 12 factor is done as ’commutativity’ (ie. Dj Di = Di Dj ) means that ni̸=j double counts.
P
α
Summing over |α|=2 D α!f (x) · hα requires summing over both case 1 and 2:
P

X Dα f (x) n n
1X 1X
· hα = Di Di f (x) · h2i + Dj Di f (x) · hi · hj
α! 2 i=1 2 i̸=j
|α|=2

X Dα f (x) n
α 1X
·h = Di Dj f (x) · hi · hj
α! 2 i,j=1
|α|=2

Hence,
n
1X
T2 (h) = f (x) + ∇f · h + Di Dj f (x) · hi · hj
2 i,j=1

Which when re-written as:

n
1X
T2 (h) = f (x) + ∇f · h + hi · Di Dj f (x) · hj
2 i,j=1

Is ’evidently’ equivalent to:

1
T2 (h) = f (x) + ∇f · h + hT Hf (x)h
2

4.3 R2 → R
For some z = f (x, y), given P = f (a, b)
Before everything,
X Dα f (x)
T (x + δ) = · δα
α!
|α|≤k

Let h = x + δ ⇔ δ = (h − x)
X Dα f (x)
T (h) = · (h − x)α
α!
|α|≤k

To calculate first order Taylor approximation centered at P,

z ≈ T1 (x, y) = f (a, b) + ∇f (a, b) · (x − a, y − b)

∇f (a, b) must be given.
 ∇f (a, b) = (Dx f (a, b), Dy f (a, b))

z ≈ T1 (x, y) = f (a, b) + (Dx f (a, b), Dy f (a, b)) · (x − a, y − b)

z ≈ T1 (x, y) = f (a, b) + Dx f (a, b)x − Dx f (a, b)a + Dy f (a, b)y − Dy f (a, b)b

z ≈ T1 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)

Second order Taylor approximation centered at P:

1
z ≈ T2 (x, y) = f (a, b) + ∇f · (x − a, y − b) + (x − a, y − b)T Hf (x)(x − a, y − b)
2
From before,

1
z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b) + (x − a, y − b)T Hf (x)(x − a, y − b)
2
To calculate T2 (x), Hf (x) must be given:

Hf (x) = D∇f (x)

∇f (x) : R2 → R2
 
Dx (∇f (x))x Dy (∇f (x))x
Hf (x) =
Dx (∇f (x))y Dy (∇f (x))y
 
Dx Dx f (x) Dy Dx f (x)
Hf (x) =
Dx Dy f (x) Dy Dy f (x)

z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+

 
1 T Dx Dx f (x) Dy Dx f (x)
2
(x − a, y − b) (x − a, y − b) (1)
Dx Dy f (x) Dy Dy f (x)

z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+

    
1 Dx Dx f (x) Dy Dx f (x) x−a
(x − a, y − b) (2)
2 Dx Dy f (x) Dy Dy f (x) y−b

z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+

 
1 x−a
(Dx Dx f (x)(x − a) + Dx Dy (y − b), Dy Dx f (x)(x − a) + Dy Dy (y − b)) (3)
2 y−b
 z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+
1
2
(Dx Dx f (x)(x − a)2 + Dx Dy (y − b)(x − a) + Dy Dx f (x)(x − a)(y − b) + Dy Dy (y − b)2 ) (4)

’commutativity’ of D:

z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+

1
2
(Dx Dx f (x)(x − a)2 + 2Dx Dy f (x)(x − a)(y − b) + Dy Dy (y − b)2 ) (5)

z ≈ T2 (x, y) = f (a, b) + Dx f (a, b)(x − a) + Dy f (a, b)(y − b)+

1
D D f (x)(x − a)2 + Dx Dy f (x)(x − a)(y − b) + 12 Dy Dy (y − b)2 (6)
2 x x

5 Optimization on f : Rn → R
5.1 Definition of local extrema
Consider f : Rn → R
Define Bϵ (x) = {a ∈ Rn |D(x, a) < ϵ}

If f has a local maximum at X, then ∃ϵ > 0 such that:

f (X) ≥ f (x) ∀x ∈ Bϵ (X)

If f has a local minimum at X, then ∃ϵ > 0 such that:

f (X) ≤ f (x) ∀x ∈ Bϵ (X)

Isolated extrema remove the equals.

5.2 Definition of local extrema in terms of ∇f

Consider f : Rn → R

If f has local extrema at X,

 
0
0
∇f (x) =  .. 
 
.
0
If f has a local maximum at X,

Hf (X) = Negative definite

If f has a local minimum at X,
 Hf (X) = Positive definite
If f has a saddle point at X,

Hf (X) = Indefinite

5.3 Definite and indefinite matrices

Definition:
M is positive definite iff:

xT M x > 0 ∀x ∈ Rn
M is negative definite iff:

xT M x < 0 ∀x ∈ Rn
M is indefinite iff:

xT M x = 0 ∀x ∈ Rn

5.3.1 Regarding eigenvalues

M is positive definite if and only if all of its eigenvalues are positive.
M is negative definite if and only if all of its eigenvalues are negative
M is indefinite if and only if it has both positive and negative eigenvalues.

5.4 Constrained optimization and Lagrange multipliers

The problem:
Given:

f : Rn → R

gi : Rn → R, i ∈ [1, m]
Optimize f with respect to constraints gi = 0 ∀i ∈ [1, m]

gi = 0 forms a contour or level curve. From below proof, it is evident the ∇gi (x) ⊥ (gi = 0)@x:

Consider r(t) which satisfies

gi (r(t)) = 0

d
gi (r(t)) = 0
dt
d
∇gi · r(t) = 0 − (1)
dt
QED

Next consider f (r(t)). f (r(t)) need not be a level curve.

However, it is intuitive and a fact that the minimums and maximums of f (r(t)) will occur
where f (r(t)) is flat.

That is, where

d
f (r(t)) = 0
dt
d
∇f · r(t) = 0 − (2)
dt
Putting (1) and (2) together allows for a formulation that is independent of parameterization:
Maximums and minimums occur when:
m
X
∇f = λi ∇gi
i=1

With the added initial constraints of:

gi = 0 ∀i ∈ [1, m]

6 Line, Surface and Volume Integrals

7 Taxonomy of Differential Equations:
7.1 Derivatives:
Ordinary
Partial

7.2 Order:
1st order: Only first derivatives
2nd order: Only second derivatives
..
.
Nth order

7.3 Linearity:
Linear:
n
X di f (x)
ai (x) = b(x)
i=1
dxi
That is it can be written as:
 n
!
X di
ai (x) i f (x) = b(x)
i=1
dx
Non-Linear:
Function can not be written as some operator acting on f (x).

f (x)n

di f (x) dj f (x)
·
dxi dxj

g(f (x))
Where g is not linear.

7.4 Auxillary conditions:

Initial Value Problems(IVP):

di+1 f df d2 f d3 f di f
= F (x, , , , ... )
dxi+1 dx dx2 dx3 dxi
Given

df (x0 ) d2 f (x0 ) di f (x0 )

f (x0 ) = v0 , = v1 , = v2 , ... = vi
dx dx2 dxi

Boundary Value Problems(BVP):

7.5 Autonomous
Derivative is independent of the dependant variable. Only depends on function itself.

7.6 Homogeneous
Lack of any free terms.
8 First-Order Differential Equations
General first order ODE is:

y ′ = f (y, x)

8.1 Linear
General solution to:

p(x)y ′ + q(x)y = r(x)

Which can be rewritten as:

y ′ + Q(x)y = R(x)
R
Qdx
Multiply both sides by e
R R R
Qdx ′ Qdx Qdx
e y + Q(x)e y=e R(x)
By the inverse chan rule,
d  R Qdx  R
e y = e Qdx R(x)
dx
R
Z R
Qdx
e y = e Qdx R(x)dx
Z
1 R
Qdx
y= R
Qdx
e R(x)dx
e

8.2 Non-Linear Separable

Rewrite

y ′ = f (y, x)
As

M (y, x) + N (y, x)y ′ = 0

In the case that

M (x) + N (y)y ′ = 0

dy
M (x) + N (y) =0
dx

M (x)dx + N (y)dy = 0
Then the solution can be implicitly written as
Z Z
N (y)dy = M (x)dx
8.3 Existence and Uniqueness Theorems
8.4 Exact Differential Equations
Consider a linear first order differential equation of the form:

M (x, y) + N (x, y)y ′ = 0

If M (x, y) and N (x, y) are both continuous in some rectangular region R and,

∂M (x, y) ∂N (x, y)
=
∂x ∂y
Then it is exact in R. To find y, find f (x, y) such that:

∂f (x, y)
= M (x, y)
∂x
∂f (x, y)
= N (x, y)
∂y
When integrating M (x, y) w.r.t x, +h becomes +h(y). Contours of f (x, y) are the implicit expression
of y. That is:

f (x, y) = C
Where C ∈ R is determined by initial values.

8.5 Integrating Factors to convert non-exact to exact

M (x, y) + N (x, y)y ′ = 0
Where

∂M (x, y) ∂N (x, y)
̸=
∂x ∂y
Goal is to find µ(x, y) such that

µ(x, y)M (x, y) + µ(x, y)N (x, y)y ′ = 0

is exact. That is:

∂ (µ(x, y)M (x, y)) ∂ (µ(x, y)N (x, y))

=
∂x ∂y
By the product rule,

∂µ(x, y) ∂M (x, y) ∂µ(x, y) ∂N (x, y)

M (x, y) + µ(x, y) = N (x, y) + µ(x, y)
∂x ∂x ∂y ∂y
∂µ(x, y) ∂M (x, y) ∂µ(x, y) ∂N (x, y)
M (x, y) + µ(x, y) − N (x, y) − µ(x, y) =0
∂x ∂x ∂y ∂y
 
∂µ(x, y) ∂µ(x, y) ∂N (x, y) ∂M (x, y)
M (x, y) − N (x, y) − µ(x, y) + =0
∂x ∂y ∂y ∂x
Written in simpler notation,

µx (x, y)M (x, y) − µy (x, y)N (x, y) − µ(x, y) (Ny (x, y) + Mx (x, y)) = 0
The only way this simplifies is:

µ(x, y) = µ(x) µ(x, y) = µ(y)

dµ(x)
dx
M (x, y) − µ(x) (Ny (x, y) + Mx (x, y)) = 0 − dµ(y)
dy
N (x, y) − µ(y) (Ny (x, y) + Mx (x, y)) = 0

dµ(x) µ(x)(Ny (x,y)+Mx (x,y)) dµ(y) −µ(y)(Ny (x,y)+Mx (x,y))

dx
= M (x,y) dy
= N (x,y)

(Ny (x,y)+Mx (x,y)) (Ny (x,y)+Mx (x,y))

Solvable iff M (x,y)
is a function of x only. Solvable iff M (x,y)
is a function of y only.

8.6 Difference Functions

9 Second-Order Differential Equations

General form:

y ′′ = f (y, y ′ , x)

9.1 Homogenous Linear with Constant Coefficients, Charecteristic Equa-

tions
ay ′′ + by ′ + cy = 0
Taking the anstaz eλx , the two values of λ can be determined by solving:

aλ2 + bλ + c = 0

9.2 Imaginary Solutions

Given some imaginary solution:

e(a+bi)x = eax ebxi

= eax (cos(bx) + sin(bx)i)

= eax cos(bx) + eax sin(bx)i

If e(a+bi)x is a solution then so is ℜ e(a+bi)x and ℑ e(a+bi)x .
Hence,

y = c1 eax cos(bx) + c2 eax sin(bx)

9.3 Singular Root, Reduction of Order
Given that the characteristic equation yields only root and therefore one solution y1 . Set y2 (x) =
v(x)y1 (x)

9.4 Wronskian
The Wronskian W (y1 , y2 )(x) is defined as
 
y1 (x) y2 (x)
W (y1 , y2 )(x) = det ′
y1 (x) y2′ (x)

W (y1 , y2 )(x) = y1 (x)y2′ (x) − y2 (x)y1′ (x)

If the Wronskian is non-zero for some t, y1 and y2 are linearly independent and hence can be added
up to with a specific c1 and c2 to solve an IV P .

9.5 Non-Homogeneous Linear with Constant Coefficients

ay ′′ + by ′ + cy = p(x)
Particular solution must be determined by anstaz. Below are general rules for anstaz:

If

ay ′′ + by ′ + cy = AeBx
Assume y = λeBx

If

ay ′′ + by ′ + cy = A cos(Bt)
or

ay ′′ + by ′ + cy = A sin(Bt)
Assume y = λ1 sin(Bt) + λ2 cos(Bt)

If

ay ′′ + by ′ + cy = A cos(Bt)eCx
or

ay ′′ + by ′ + cy = A sin(Bt)eCx
Assume y = λ1 eCx sin(Bt) + λ2 eCx cos(Bt). In general:
The general solution of the above equation can be written as:

y = c1 y1 (t) + c2 y2 (t) + Y (t)

y1 and y2 are solutions of the homogenous equation ay ′′ + by ′ + cy = 0. Y is a particular solution.

9.6 Series Solutions

Refresher of taylor series of a function f about point x0 :
∞
X f (n) (x0 )
f (x) = (x − x0 )n
n=0
n!
Let,

f (n) (x0 )
an = (x − x0 )n
n!
Radius of convergence:

an+1
lim = c|x − x0 |
n→∞ an
Where radius of convergence is:
1
|x − x0 | ≤
c
Consider two series given by:
∞
X
s1 = bn (x − x0 )n
n=1

Which converges to f (x) for some radius |x − x0 | ≤ l, and

∞
X
s2 = cn (x − x0 )n
n=1
Which converges to g(x) for some radius |x − x0 | ≤ l

Consider that:

f (x) ± g(x)

f (x)g(x)

f (x)
g(x)
Differentiation rules:

∞
X
y= an (x − x0 )n
n=1
∞
′ d X
y = an (x − x0 )n
dx n=1
The equation:

P (x)y ′′ + Q(x)y ′ + R(x)y = 0

Can be solved/approximated at an ordinary point x0 (x0 such that P (x0 ) ̸= 0). In such case the above
can be rewritten as:

y ′′ + p(x)y ′ + q(x)y = 0
THe solution takes the form:

y = a0 y1 + a1 y2
Where a0 and a1 are arbitrary and depend on the initial conditions. y1 and y2 are power series. The
function is analytic at x = x0 where the R(y) ≥ min (R(p), R(q)).

To solve for y, write y and its derivatives as taylor series. Then solve for the coefficients. Change of
index and the above derived rules for operations of power series will be useful. Find a relation (explicit
or recursive) for the coefficients of y1 in terms of a0 and do the same for y2 in terms of a1 .

9.7 Non-Homogeneous Linear with Function Coefficients: General solu-

tion
To find the general solution to a second order linear differential equation

y ′′ + p(x)y ′ + q(x)y = r(x)

Consider the correspond homogenous equation:

y ′′ + p(x)y ′ + q(x)y = 0
Solve the above using series solutions to find that:

y = a0 y1 + a1 y2
General solution of the non homogenous equation will be of the form,

y = a0 y1 + a1 y2 + Y
Let Y be of the form:

Y = u1 (x)y1 + u2 (x)y2
It can be shown that:
Z
y2 (x)r(x)
u1 (x) = − dx
W (y1 , y2 )(x)
Z
y1 (x)r(x)
u2 (x) = dx
W (y1 , y2 )(x)
Hence, the solution can be written as:
Z Z
y2 (x)r(x) y1 (x)r(x)
y = a0 y1 + a1 y2 − y1 dx + y2 dx
W (y1 , y2 )(x) W (y1 , y2 )(x)