flexsolve 0.2.14

Flexible function solvers

Contents

• What is flexsolve?

• Installation

• Documentation

• Bug reports

• License information

• Citation

What is flexsolve?

flexsolve presents a flexible set of function solvers by defining alternative tolerance conditions for accepting a solution. These solvers also implement methods like Wegstein and Aitken-Steffensen acceleration to reach solutions quicker.

Installation

Get the latest version of flexsolve from PyPI. If you have an installation of Python with pip, simple install it with:

$ pip install flexsolve

To get the git version, run:

$ git clone git://github.com/yoelcortes/flexsolve

Documentation

Flexsolve solvers can solve a variety of specifications:

• Solve x where f(x) = x:

  • fixed_point: Simple fixed point iteration.

  • wegstein: Wegstein’s accelerated iteration method.

  • aitken: Aitken-Steffensen accelerated iteration method.

• Solve x where f(x) = 0 and x0 < x < x1:

  • bisection: Simple bisection method

  • false_position: Simple false position method.

  • IQ_interpolation: Quadratic interpolation solver (similar to scipy.optimize.brentq)

  • bounded_wegstein: False position method with Wegstein acceleration.

  • bounded_aitken: False position method with Aitken-Steffensen acceleration.

• Solve x where f(x) = 0:

  • secant: Simple secant method.

  • wegstein_secant: Secant method with Wegstein acceleration.

  • aitken_secant: Secant method with Aitken acceleration.

Parameters for each solver are pretty consitent and straight forward:

• f: objective function in the form of f(x, *args).

• x: Root guess. Solver begins the iteration by evaluating f(x).

• x0, x1: Root bracket. Solution must lie within x0 and x1.

• xtol=1e-8: Solver stops when the root lies within xtol.

• ytol=5e-8: Solver stops when the f(x) lies within ytol of the root.

• yval=0: Root offset. Solver will find x where f(x) = yval.

• args=(): Arguments to pass to f.

Here are some exmples using flexsolve’s Profiler object to test and compare different solvers. In the graphs, the points are the solver iterations and the lines represent f(x). The lines and points are offset to make them more visible (so all the points are actually on the same curve!). The shaded area is just to help us relate the points to the curve (not an actual interval):

>>> import flexsolve as flx
>>> from scipy import optimize as opt
>>> x0, x1 = [-5, 5]
>>> f = lambda x: x**3 - 40 + 2*x
>>> p = flx.Profiler(f) # When called, it returns f(x) and saves the results.
>>> opt.brentq(p, x0, x1, xtol=1e-8)
3.225240462778411
>>> p.archive('[Scipy] Brent-Q') # Save/archive results with given name
>>> opt.brenth(p, x0, x1)
3.2252404627917794
>>> p.archive('[Scipy] Brent-H')
>>> flx.IQ_interpolation(p, x0, x1)
3.225240462796626
>>> p.archive('IQ-interpolation')
>>> flx.bounded_wegstein(p, x0, x1)
3.225240462790051
>>> p.archive('Wegstein')
>>> x_aitken = flx.bounded_aitken(p, x0, x1)
3.2252404627883218
>>> p.archive('Aitken')
>>> flx.false_position(p, x0, x1)
3.225240462687035
>>> p.archive('False position')
>>> p.plot(r'$f(x) = 0 = x^3 + 2 \cdot x - 40$ where $-5 < x < 5$')

>>> p = flx.Profiler(f)
>>> x_guess = -5
>>> flx.wegstein_secant(p, x_guess)
3.22524046279178
>>> p.archive('Wegstein')
>>> flx.aitken_secant(p, x_guess)
3.22524046279178
>>> p.archive('Aitken')
>>> flx.secant(p, x_guess)
3.2252404627918057
>>> p.archive('Secant')
>>> opt.newton(p, x_guess)
3.2252404627918065
>>> p.archive('[Scipy] Newton')
>>> p.plot(r'$f(x) = 0 = x^3 + 2 \cdot x - 40$')

>>> # Note that x = 40/x^2 - 2/x is the same
>>> # objective function as x**3 - 40 + 2*x = 0
>>> f = lambda x: 40/x**2 - 2/x
>>> p = flx.Profiler(f)
>>> x_guess = 5.
>>> flx.wegstein(p, x_guess)
3.2252404626726996
>>> p.archive('Wegstein')
>>> flx.aitken(p, x_guess)
3.2252404627250075
>>> p.archive('Aitken')
>>> p.plot(r'$f(x) = x = \frac{40}{x^2} - \frac{2}{x}$',
...        markbounds=False)
>>> # Fixed-point iteration is non-convergent for this equation,
>>> # so we do not include it here

If your project is need for speed, you can speed up calculations in flexsolve using the speed_up() method, which works by jit compiling computationally-heavy algorithms in flexsolve. The following example benchmarks flexsolve’s speed with and without compiling:

>>> import flexsolve as flx
>>> f = lambda x: x**3 - 40 + 2*x
>>> # Time solver without compiling
>>> %timeit flx.IQ_interpolation(f, -5, 5)
38.3 µs ± 4.7 µs per loop (mean ± std. dev. of 7 runs, 10000 loops each)
>>> flx.speed_up() # This is the only line we need to run to speed up flexsolve
>>> # First run is slower because it need to compile
>>> x = flx.IQ_interpolation(f, -5, 5)
>>> # Time solver after compiling
>>> %timeit flx.IQ_interpolation(f, -5, 5)
11.3 µs ± 156 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

The iterative methods for solving f(x) = x (e.g. fixed-point, Wegstain, Aitken) are capable of solving multi-dimensional problems. Simply make sure x is an array and f(x) returns an array with the same dimensions. In fact, the The Biorefinery Simulation and Techno-Economic Analysis Modules (BioSTEAM) uses flexsolve to solve many chemical engineering problems, including process recycle stream flow rates and vapor-liquid equilibrium compositions.

Bug reports

To report bugs, please use the eqsolvers’s Bug Tracker at:

https://github.com/yoelcortes/flexsolve

License information

See LICENSE.txt for information on the terms & conditions for usage of this software, and a DISCLAIMER OF ALL WARRANTIES.

Although not required by the eqsolvers license, if it is convenient for you, please cite eqsolvers if used in your work. Please also consider contributing any changes you make back, and benefit the community.

Citation

To cite eqsolvers in publications use:

Yoel Cortes-Pena (2019). flexsolve: Flexible function solvers. https://github.com/yoelcortes/flexsolve