Suppose we wish to compute a predicted value for some continuous,
differentiable function f(t) at future time t=b, based only upon knowledge
of its value now at time t=a, which is f(a), and its first three
derivatives now, namely f'(a), f''(a), and f'''(a).
How would you compute such a prediction for the future value f(b)?
Give a formula or explain the sorts of terms that it would include and
why they appear.
2.
In practical computing problems with numerical data that is discretely
sampled at periodic points in time, derivatives defined
in terms of continuous limits must be replaced by discrete estimates. Provide
finite-difference expressions for each of the three derivatives
f'(t), f''(t), and f'''(t) on the discrete set of samples
...,f(ti), f(ti+1),f(ti+2), f(ti+3),...
assuming that this sequence of
samples are separated by a constant interval
.
3.
If you were trying to compute numerically the solution to a
continuous-time first-order differential equation
using the Runge-Kutta method of integration,
how would you expect the error of your solution
in the worst case to depend upon the stepsize ?
4.
Real-world signals are generally accompanied by noise. Using the
Differentiation Theorem of the Fourier Transform, comment upon how
noise is amplified as a function of its spectral composition when computing the
first-order, second-order, and third-order derivatives of the signal.