Math Cheatsheet Newton’s Method Some Linearized Models:

Compiled by Eigen-π (https://www.linkedin.com/in/islamibr29/).

We are searching for a solution starting by an initial guess: x0 • Exponential Model:
bx
Licensed under CC BY-NC-SA 4.0. Please share comments, suggestions, y = ae
f (xi )
and errors at http://github.com/islamibr/today_i_learned. xi+1 = xi − We can write:
f ′ (xi )
ln y = ln a + bx
Last Updated May 28, 2024 Stopping Criterion Y = a0 + a1 X
• The required number of iterations is reached. • Power Model:
Solving a Linear System of Equations • Or |xi+1 − xi | < ϵ, where ϵ is the maximum error bound. y = ax
b

a11 x1 + a12 x2 + a13 x3 = b1 We can write:
Relative Percentage Error

a21 x1 + a22 x2 + a23 x3 = b2 log y = log a + b log x

a x + a x + a x = b xi+1 − xi
31 1 32 2 33 3 3 RP E = × 100% Y = a0 + a1 X
xi+1
Jacobi Method • Growth Rate Model:
ax
(0) (0)
Initial values: x1 , x2 , x3
(0) Linear Regression y=
b+x
(n) (n)
Given the Data Set We can write:
1 x+b

(n+1) b1 −a12 x2 −a13 x3

x = x y =
 1 a11
y ax


(n) (n)
(n+1) b2 −a21 x1 −a23 x3
x2 = a22 1 1 b 1

 (n) (n) = +
y a ax

 (n+1)
 b3 −a31 x1 −a32 x2
x3 = a33 • Number of points = n Y = a0 + a1 X
Gauss-Seidel Method • We need to find a straight line that best fits the given data:
(0) (0)
Initial values: x1 , x2 , x3
(0)
y = a0 + a1 x Numerical Differentiation
• To find a0 and a1 :
 (n)
b1 −a12 x2 −a13 x3
(n) Finite Difference of First Derivative
(n+1)
x1
 = x y x2 xy (y − a0 − a1 x)2 (y − ȳ)2
• Forward:
 a11

 (n+1) (n) f (xi+1 ) − f (xi )
(n+1) b2 −a21 x1 −a23 x3 ′
x2 = a22
f (xi ) = + O(h)

 (n+1) (n+1)
h
 (n+1)
 b3 −a31 x1 −a32 x2
x3 = a33 • Backward:
• Sum: ′ f (xi ) − f (xi−1 )
f (xi ) = + O(h)
Diagonally Dominant Matrix
X X X 2
X
x y x xy Sr St h
• Diagonally dominant matrix: • Central:
X Solve ′ f (xi+1 ) − f (xi−1 ) 2
|aii | ≥ |aij | f (xi ) = + O(h )
2h
n
! n
X X
i̸=j na0 + xi a1 = yi
• Strictly diagonally dominant matrix: i=1 i=1 Finite Difference of Second Derivative
n
! n
! n
X X X 2
X • Central:
|aii | > |aij | xi a0 + xi a1 = xi yi
i̸=j i=1 i=1 i=1 ′′ f (xi+1 ) − 2f (xi ) + f (xi−1 ) 2
f (xi ) = + O(h )
h2
• Convergence of the Jacobi & Gauss-Seidel Methods: Coefficient of Determination
To guarantee the convergence of these two methods, the
2 St − Sr Numerical Integration
coefficient matrix must be strictly diagonally dominant. r =
St
• Required to find an approximate value for the definite integral:
Solving a Nonlinear Equation Correlation Coefficient Z b
f (x) = 0 s
I = f (x)dx
St − Sr a
r=
Bisection Method St
• Step size = h
We are searching for a solution within an interval [a, b]. Mean of y Values
ai +bi
• Number of segments = n
1. Get xi = n
!
2 1 X
ȳ = yi b−a
2. Get f (xi ), f (ai ), f (bi ) n h=
i=1 n
3. If f (xi )f (ai ) > 0 (same sign), then the new interval is [xi , bi ],
otherwise it’s [ai , xi ] • For the model of the form:
x a a+h a + 2h ... a + kh ... b
4. Repeat f (y) = a0 + a1 g(x)
y y0 y1 y2 ... yk ... yn
Maximum error bound: We can use the substitution
b−a Trapezoidal Rule
ϵ= Y = f (y) and X = g(x)
2n
Then perform the linear regression to the model: n−1
!
 
b−a
 h X
n = log2 I ≈ y0 + yn + 2 yk
Y = a0 + a1 X 2
ϵ k=1
Simpson’s 1/3 Rule Bessel Function of the 1st kind Complex Analysis
n = even ∞
  X (−1)k (x/2)n+2k Complex Number
Jn (x) =
h
n−1
X n−2
X  k=0
k! Γ(n + k + 1) z = x + iy
I ≈ y0 + yn + 4 yk + 2 yk 
3  
nd
k=1 k=2
k: odd k: even
Bessel Function of the 2 kind Polar Form
cos(nπ)Jn (x) − J−n (x) z = r(cos θ + i sin θ)
Solution of Differential Equations Yn (x) =
sin(nπ)
Consider the differential equation:
From Cartesian to Polar:
′
y = F (x, y), at x = x0 , y = y0
Properties p −1
 
y
n
r= x2 + y 2 θ = tan
J−n (x) = (−1) Jn (x) for n = positive integer x
with the initial condition: r
y0 = f (x0 ) J1/2 (x) =
2
sin x
From Polar to Cartesian:
πx
x = r cos(θ) y = r sin(θ)
Step size is denoted as h. r
2
J−1/2 (x) = cos x
Euler’s Method πx Euler’s Formula
iθ
yn+1 = yn + h F (xn , yn ) Bessel Recurrence Relations e = cos(θ) + i sin(θ)
iθ
2n ∴ z = re
Heun’s Method (Improved Euler’s Method) Jn−1 (x) + Jn+1 (x) = Jn (x)
x
Predictor: ′ Power of Complex Number (De Moivre’s Theo-
∗
yn+1 = yn + h F (xn , yn ) Jn−1 (x) − Jn+1 (x) = 2Jn (x)
rem)
Corrector: d n
n
x Jn (x) = x Jn−1 (x)
h dx Case 1: Integer Power (n ∈ Z)
∗

yn+1 = yn + F (xn , yn ) + F (xn+1 , yn+1 ) d  −n 
−n n
2 x Jn (x) = −x Jn+1 (x) (cos(θ) + i sin(θ)) = cos(nθ) + i sin(nθ)
dx
Gamma Function
 
Case 2: Rational Power n = pq
Integrals containing Bessel Functions
Definition d
Z
(cos(θ) + i sin(θ))
p/q
= cos(φ) + i sin(φ)
n
n n n
Z ∞ x Jn (x) = x Jn−1 (x) ⇒ x Jn−1 (x) dx = x Jn (x) + c
n−1 −x dx
Γ(n) = x e dx pθ + 2πk
0 φ= k = 0, 1, 2, . . . , (q − 1)
d  −n 
−n q
x Jn (x) = −x Jn+1 (x)
Properties dx
Differentiability and Derivative:
Z ∞
−x2
Z
2n−1 −n −n
Γ(n) = 2 e x dx ⇒ x Jn+1 (x) dx = −x Jn (x) + c If the Cauchy-Riemann equations are satisfied, the function f (z) is
0 differentiable and the derivative can be obtained as:
√
 
1
Γ
2
= π General Integral ′
f (z) =
dw
=
∂w
=
1 ∂w
Z
m+n dz ∂x i ∂y
Γ(n) = (n − 1)Γ(n − 1) m
x Jn (x) dx
Γ(n) = (n − 1)! For n = Positive integer 2 Analyticity:
Choose the more suitable rule:
• If the Cauchy-Riemann equations are satisfied at all points in a
Beta Function Odd domain D, then the function f (z) is analytic in D.
Definition • If the function f (z) is analytic at all points of the complex
If m > n: Z plane, we call it an entire function.
Z 1 m m
n−1 m−1 x Jn−1 (x) dx = x Jn (x) + c
B(n, m) = x (1 − x) dx • if f (z) is not differentiable at z = n ( n is called singular
0
point), then we call it analytic except at z = n
Properties Even • If the Cauchy-Riemann equations are not satisfied, then f (z) is
Z π/2 If m < n: not analytic in any domain, so we can say f (z) is nowhere
2n−1 2m−1
B(n, m) = 2 (cos θ) (sin θ) dθ analytic.
Z
m m
0 x Jn+1 (x) dx = −x Jn (x) + c
B(n, m) = B(m, n)
The result is in terms of 0x Jn (x) dx.
R
Γ(n)Γ(m)
B(n, m) =
Γ(n + m)
Complex Differentiation
Bessel Differential Equation For the function:
2 ′′ ′ 2 2
x y + xy + (x − n )y = 0 w = f (z) = u(x, y) + iv(x, y)
Solution Cauchy-Riemann equations:
y(x) = c1 Jn (x) + c2 J−n (x) for n ∈
/ {0, 1, 2, 3, . . .}
∂u ∂v ∂u ∂v
y(x) = c1 Jn (x) + c2 Yn (x) for n ∈ {0, 1, 2, 3, . . .} = and =−
∂x ∂y ∂y ∂x