## Trapezoidal Rule
Approximates the integral of a function by summing up areas of trapezoids.
- Implicit, because $u^{n+1}$ appears on both sides
- Second-order accurate in time
Given an ODE: $\frac{du}{dt}=f(t,u)$
The trapezoidal rule updates the solution as:
$u^{n+1}=u^n+\frac{\Delta t}{2}\left[f(t^n,u^n)+f(t^{n+1},u^{n+1})\right]$
**Single interval:**
$\int_a^b f(x) \, dx \approx \frac{h}{2} \left[ f(a) + f(b) \right]$
**Composite trapezoidal rule:**
- $n$: number of subintervals
- $h = \frac{b - a}{n}$
- $x_i = a + i h$
$\int_a^b f(x)\, dx \approx \frac{h}{2} \left[ f(x_0) + 2 \sum_{i=1}^{n-1} f(x_i) + f(x_n) \right]$
Accuracy: $\mathcal{O}(h^2)$, improves with smaller $h$ (finer grid)
## Romberg Integration
Improves the accuracy of the trapezoidal rule using **Richardson extrapolation**.
- Trapezoidal rule: error $\mathcal{O}(h^2)$
- Romberg integration: applies Richardson recursively for $\mathcal{O}(h^4)$, $\mathcal{O}(h^6)$, etc.
- Efficient for **smooth** functions when high accuracy is needed
1. Compute trapezoidal estimates with step sizes $h$, $h/2$, $h/4$, etc.
2. Extrapolate to cancel $\mathcal{O}(h^2)$, $\mathcal{O}(h^4)$ errors and so on.
Let $R_{i,0}$ be the trapezoidal estimate using $2^i$ intervals.
$R_{i,j} = \frac{4^j R_{i,j-1} - R_{i-1,j-1}}{4^j - 1}$
Each $j$ increases the order of accuracy by 2.
**Romberg table structure:**
| $i \backslash j$ | 0 (trapezoid) | 1 (Rich. 1) | 2 (Rich. 2) |
|------------------|---------------|-------------|-------------|
| 0 | $R_{0,0}$ | | |
| 1 | $R_{1,0}$ | $R_{1,1}$ | |
| 2 | $R_{2,0}$ | $R_{2,1}$ | $R_{2,2}$ |
**Stopping criterion:**
$|R_{i,j} - R_{i,j-1}| < \text{tolerance}$
## Crank-Nicolson
Time integration scheme that is the trapezoidal rule applied to spacially discretized PDEs.
- implicit
- 2nd order accurate in time
Example: 1D diffusion equation:
$\frac{\partial u}{\partial t}=\kappa\frac{\partial^2 u}{\partial x^2}$
Discretize space (e.g., finite differences), yielding:
$\frac{du}{dt}=\kappa D u$
Apply the trapezoidal rule:
$u^{n+1}=u^n+\frac{\Delta t}{2}\kappa\left(Du^n+Du^{n+1}\right)$
Rearranged into matrix form:
$\left(I-\frac{\Delta t}{2}\kappa D\right)u^{n+1}=\left(I+\frac{\Delta t}{2}\kappa D\right)u^n$
## ADI Schemes
An Alternating Direction Implicit method is a time-stepping scheme that is:
- Implicit in time (unconditionally stable)
- Often second-order accurate in time
- Only tridiagonal (1D) solves required per step
- Great for 2D heat/diffusion problems
$\frac{\partial u}{\partial t} = \kappa\left(\frac{\partial^2 u}{\partial x^2} + \frac{\partial^2 u}{\partial y^2}\right)$
A full time step is split into two half steps:
1. First half-step (implicit in $x$, explicit in $y$):
$u^* = u^n + \frac{\Delta t}{2}\kappa\left(\frac{\partial^2 u^*}{\partial x^2} + \frac{\partial^2 u^n}{\partial y^2}\right)$
2. Second half-step (implicit in $y$, explicit in $x$):
$u^{n+1} = u^* + \frac{\Delta t}{2}\kappa\left(\frac{\partial^2 u^*}{\partial x^2} + \frac{\partial^2 u^{n+1}}{\partial y^2}\right)$
Each step only requires solving a 1D tridiagonal system — much cheaper than solving a full 2D system.
## Initial Value Problems
After discretizing the spatial derivatives we obtain a (coupled) system of (nonlinear) ODEs
for IVPs, developed with a single equations
similar methods to solving steady problems by iterations
linear diff eq's can be solved analytically
non-linear equations require numerical solutions
Forward difference is bad because truncation error is always the same sign
### Numerical IVP Solving
Given: $\frac{dy}{dt} = f(t, y), \quad y(t_0) = y_0$
Use Taylor expansion to step forward:
$y(t+h) = y(t) + h f(t, y) + \frac{h^2}{2} f_t + \cdots$
Where $f_t = \frac{d}{dt} f(t, y) = \frac{\partial f}{\partial t} + \frac{\partial f}{\partial y} \cdot f$
#### First-Order (Euler)
$y_{n+1} = y_n + h f(t_n, y_n)$
- Truncates after 1st term
- Local truncation error: $\mathcal{O}(h^2)$
- Global error: $\mathcal{O}(h)$
#### Second-Order Taylor
$y_{n+1} = y_n + h f(t_n, y_n) + \frac{h^2}{2} f_t(t_n, y_n)$
- More accurate but needs derivative of $f$