CHAPTER 2.

FINITE DIFFERENCES
Solution to Partial Differential Equations (PDEs)

Analytical solutions to PDEs are rare:

α (2 n +1)n π 2t
∂ p 1 ∂p
2
4 pinit ∞
(−1)n
(2n + 1)π x
∂x 2
=
α ∂t
⇒ p ( ) B1
x , t = p −
π
∑
n = 0 2n + 1
e 4 L2
cos
2L

More generally we seek a numerical approximation to the PDE:

• Usually we transform the PDE into algebraic equations which are easier
to solve
• We “discretize” the derivatives
• Assume constant pressure and properties over an area/block
Solutions for PDEs ∂ 2 p 1 ∂p
=
∂x 2
α ∂t

Analytical Solution Numerical Solution

Need to solve a complex PDE Need to solve algebraic equations

p ( x, t ) time step

Grid block

Continuous Solution P Discrete Solution

P

x
x
Finite Difference Approximations

• Finite differences are used to approximate the numerical value

of a derivative at a point, x.
∂f
f ′( x) =
∂x x

f(x)

• Begin by using a truncated Taylor Series approximation of a

function
1 1
f ( x + ∆x) = f ( x) + f ′( x)∆x + f ′′( x)∆x 2 + f "'( x)∆x3 + ...
2! 3!

• Various finite difference schemes can be employed, with varying

degrees of accuracy. Three common methods employed here:
• Forward difference
• Backward difference
• Centered difference
 Forward Difference Approximation
P(x) Tangent line
Secant line

∂P P( x + ∆x) − P( x)
≈
∂x forward ∆x

x x+∆x
x

• Derivative is defined as the “slope of the line tangent to the curve”

• This is difficult because you need two points to make a line

• Secant connects P(x) to P(x+∆x). Slope of the secant line is the “rise”
over the “run”

• Using smaller ∆x should give more accuracy to slope of the tangent

line, in fact they are the same in the limit that ∆x 0
Derivation of Forward Difference

Recall a Taylor Series:

∂P 1 ∂2 P 1 ∂3 P
P( x + ∆x) = P( x) + ∆x + ( ∆x ) 2
+ ( ∆x ) 3
+
∂x x 2! ∂x x
2
3! ∂x x
3

Re-arrange for the first derivative:

∂P P( x + ∆x) − P ( x) 1 ∂2 P 1 ∂3 P
= − ( ∆x ) + ( ∆x ) 2
+
∂x x ∆x 2! ∂x x
2
3! ∂x x
3

forward difference approximation truncation error

Forward approximation is as before; error is proportional to ∆x:

∂P P( x + ∆x) − P( x) P" P "'

≈ R= (∆x) + (∆x) 2 + = O(∆x)
∂x forward ∆x 2! 3!
Backward Difference Approximation for

Secant line
Tangent line
P(x)

∂P P( x) − P( x − ∆x)
≈
∂x backward ∆x

x-∆x x

• Secant now connects P(x) to P(x-∆x). Slope of the secant line is the
“rise” over the “run”

• Using smaller ∆x should give more accuracy to slope of the tangent

line, in fact they are the same in the limit that ∆x 0

• Backward difference is not necessarily “better” or “worse” than forward

difference
Derivation of Backward Difference

Even terms are now negative in Taylor Series:

∂P 1 ∂2 P 1 ∂3 P
P( x − ∆x) = P( x) − ∆x + ( ∆x ) 2
− ( ∆x ) 3
+
∂x x 2! ∂x x
2
3! ∂x x
3

Re-arrange for the first derivative:

∂P P( x) − P ( x − ∆x) 1 ∂2 P 1 ∂3 P
= + ( ∆x ) − ( ∆x ) 2
+
∂x x ∆x 2! ∂x x
2
3! ∂x x
3

backward difference approximation truncation error

Error in the backward difference approximation is proportional to ∆x, just

like the forward difference:

∂P P( x) − P ( x − ∆x) P" P "'

≈ R= (∆x) − (∆x) 2 + = O(∆x)
∂x backward ∆x 2! 3!
Central Difference Approximation

Tangent
P(x) line
Secant line

∂P P ( x + ∆x) − P( x − ∆x)
≈
∂x centered 2∆x

x-∆x x x+∆x

• Secant now connects P(x+∆x) to P(x-∆x). Slope of the secant line is the
“rise” over the “run”

• Using smaller ∆x should give more accuracy to slope of the tangent

line, in fact they are the same in the limit that ∆x 0

• Centered difference is potentially more accurate than forward or

backward difference
 Derivation of Central Difference Approximation

You can subtract forward and backward Taylor series to get centered

∂P 1 ∂2 P 1 ∂3 P
P( x + ∆x) = P( x) + ∆x + ( ∆x ) 2
+ ( ∆x ) 3
+
∂x x 2! ∂x x
2
3! ∂x x
3

∂P 1 ∂2 P 1 ∂3 P
− P( x − ∆x) = P( x) − ∆x + ( ∆x ) 2
− ( ∆x ) 3

∂x x 2! ∂x 2 x 3! ∂x 3 x

∂P 2 ∂3 P
P( x + ∆x) − P( x − ∆x) = 2 ∆x + ( ∆x ) 3
+ ...
∂x x 3! ∂x x
3

Rearranging gives the centered approximation with only O(∆x2) error!

∂P P( x + ∆x) − P( x − ∆x) P "'

≈ R=− (∆x) 2 + = O(∆x 2 )
∂x centered 2∆x 3!
What about 2nd Derivative?
• Can derive a forward or backward, 2nd derivative
• Here, we’ll derive a centered difference approximation
• Add forward and backward Taylor series together

∂P 1 ∂2 P 1 ∂3 P
P( x + ∆x) = P( x) + ∆x + ( ∆x ) 2
+ ( ∆x ) 3
+
∂x x 2! ∂x 2 x 3! ∂x 3 x

∂P 1 ∂2 P 1 ∂3 P
+ P( x − ∆x) = P ( x) − ∆x + (∆x) −
2
(∆x)3 + ...
∂x x 2! ∂x x
2
3! ∂x x
3

∂2 P
P( x + ∆x) + P( x − ∆x) = 2 P( x) + 2 (∆x) 2 + ...
∂x x

Centered difference approximation is O(∆x2)

P( x + ∆x) − P( x) P( x) − P( x − ∆x)
−
∂2 P P( x + ∆x) − 2 P ( x) + P ( x − ∆x) (∆x) (∆x)
≈ =
∂x 2 centered (∆x) 2 (∆x)
Summary of Important Finite Difference
Approximations

Derivative Finite Difference Approximation Type Error

∂P P ( t + ∆t ) − P ( t ) Forward Difference O ( ∆t )
∂t ∆t
∂P P ( t ) − P ( t − ∆t ) Backward Difference O ( ∆t )
∂t ∆t
∂P P ( t + ∆t ) − P ( t − ∆t ) Centered Difference O ( ∆t 2 )
∂t 2∆t
∂2 P P ( x + ∆x ) − 2 P ( x ) + P ( x − ∆x ) Centered Difference O ( ∆x 2 )
∂x 2 ∆x 2