Numerical Methods

Change of Sign Methods
Sometimes there is no easy way of finding the roots of an equation by
factorising (e.g
x3 – 7x + 3 = 0 )
Alternative is to look at graph to find the interval where the roots lie (e.g
between 2 and 3)
Then try plugging in values of x between that interval and wait for the
answer to change sign (from + to – or visa versa)
Repeat until gradually you gain appropriate levels of accuracy (enough
decimal places)
Can go wrong:
If a repeated root occurs (so it touches the x-axis and never goes
below)
If there is a discontinuity in the graph (if there’s a change of sign
without a root)
Fixed Point Iteration
Rearrange f(x) = 0 into x = g(x) and solve to find roots (if done incorrectly
can mean converging to different root or not converging at all)
This is essentially splitting f(x) into the lines y = x and x = g(x)
This means the point on the graph where y = x and x = g(x) meet is the
same x-value as the root of the equation
f(x) = 0

You can do this by gradually gaining more accuracy from a start point:

The sequence converges on a root of the equation, providing x0 is a close
enough approximation and the curve is not too steep close to the root
The gradient of the curve close to the root must be between -1 and 1

Newton-Raphson Method
Based on evaluating the gradient of the tangent to the curve y = f(x) at

Repeating this causes convergence in the same way as fixed point iteration
This method rarely fails, but if the approximation is close to a turning point
then iterations may diverge or converge to a different root (as
f (x) will be
very small)
The Trapezium Rule
Not all functions can be integrated
For such functions approximate values for a definite integral can be found
by using the trapezium rule
This involves dividing the area under the graph into a number of trapezia
and summing the areas of these trapezia
Using more trapezia will give a more accurate value
This method can give overestimates or underestimates:
If the curve is always concave upwards for the region being
considered then this method will give an overestimate (trapezia will
always lie partially above curve)
If the curve is always concave downwards for the region being
considered then this method will give an underestimate (trapezia will
always lie partially below curve)
If there is a point of inflection in region being considered, cannot
easily tell if overestimate or underestimate

Upper and Lower Bounds
Trapezium rule gives approximate value for an integral
Can often tell from shape of curve if over or underestimate, but not easy to
tell how accurate the estimate is
Using rectangles instead of trapezia (if the curve has no turning point) can
be useful
Rectangles that are either above or below the curve can give a lower or
upper bound