Quantitative Analysis
Parallel Processing
Numerical Analysis
C++ Multithreading
Python for Excel
Python Utilities

I. Wavelet calculations.
II. Calculation of approximation spaces in one dimension.
III. Calculation of approximation spaces in one dimension II.
IV. One dimensional problems.
1. Decomposition of payoff function in one dimension. Adaptive multiscaled approximation.
2. Constructing wavelet basis with Dirichlet boundary conditions.
3. Accelerated calculation of Gram matrix.
4. Adapting wavelet basis to arbitrary interval.
5. Solving one dimensional elliptic PDEs.
6. Discontinuous Galerkin technique II.
7. Solving one dimensional Black PDE.
8. Solving one dimensional mean reverting equation.
V. Stochastic optimization in one dimension.
VI. Scalar product in N-dimensions.
VII. Wavelet transform of payoff function in N-dimensions.
VIII. Solving N-dimensional PDEs.
Downloads. Index. Contents.

Decomposition of payoff function in one dimension. Adaptive multiscaled approximation.

he scripts basis3.py, basis3a.py, project.py in the directory OTSProjects/wavelets2 contain the following derivations.

We would like to obtain a decomposition for a function MATH with respect to a wavelet basis adapted to the interval MATH .

We start from results of the procedure ( Calculation of approximation spaces in one dimension II. ) We have the sets of functions MATH We adapt these functions to the interval MATH by applying the operation MATH to each function: MATH normalize them to MATH . This is done in basis3.py and basis3a.py.

In the script project.py we initially calculate MATH where the $d_{0}$ is the minimal scale that satisfies the condition ( Sufficiently fine scale 2 ) and the projection is taken in MATH .

We form the set MATH for some fixed $D>d_{0}$ . Then we repeatedly do MATH until $\#B$ reaches a given number.

This experiment yields the following conclusions:

1. Approximation is significantly better then anything that may be obtained by classical finite elements or low order wavelets.

2. Floating point errors break procedure for reasonably large $D$ if we use the Poly class. We successfully remove the difficulty using the localized piecewise polynomial representation of the functions MATH covered in the section ( Manipulation of localized piecewise polynomial functions ).

3. Accelerated version of the procedure involves taking several functions of the same scale on the step MATH The number of functions decreases with scale and increases with required precision or $\#B$ . Another way to accelerate the procedure is to gradually increase the set $S$ based on results of previous steps.

One does not need to calculate wavelet decomposition at every evaluation because of scaling and translation properties of the wavelet basis. The decomposition may assembled almost instantly and with perfect precision from precalculated data.

One dime payoff residue figure
Plot of $u-U$ for $\#B=60$ executed using LPoly.

The script dproject.py contains a similar experiment performed in MATH -space.

Note that the resulting recipe is opposite to the recipe of the section ( Adaptive approximation ). Indeed, the wavelet basis approximates perfectly the linear part of the function $u\left( x\right) $ and has difficulty around the singularity. However, it is around the singularity where the function $u\left( x\right) $ has the smallest variation.

Downloads. Index. Contents.

Copyright 2007