Neumann - A Python Library for Applied Mathematics in Physical Sciences#

Important

Neumann is in the early stages of it’s lifetime, and some concepts may change in the future. If you want long-term stability, wait until version 1.0, which is planned to be released if the core concepts all seem to sit. Nonetheless, the library is well tested with a coverage value above 90%, so if something catches your eye use it with confidence, just don’t forget to pin down the version of Neumann in your ‘requirements.txt’.

Features#

  • Linear Algebra

    • A mechanism that guarantees to maintain the property of objectivity of tensorial quantities.

    • A ReferenceFrame class for all kinds of frames, and dedicated RectangularFrame and CartesianFrame classes as special cases, all NumPy compliant.

    • NumPy compliant classes like Tensor and Vector to handle various kinds of tensorial quantities efficiently.

    • A JaggedArray and a Numba-jittable csr_matrix to handle sparse data.

  • Operations Research

    • Classes to define and solve linear and nonlinear optimization problems.

      • A LinearProgrammingProblem class to define and solve any kind of linear optimization problem.

      • A BinaryGeneticAlgorithm class to tackle more complicated optimization problems.

  • Graph Theory

    • Algorithms to calculate rooted level structures and pseudo peripheral nodes of a networkx graph, which are useful if you want to minimize the bandwidth of sparse symmetrix matrices.

Important

Be aware, that the library uses JIT-compilation through Numba, and as a result, first calls to these functions may take longer, but pay off in the long run.

Tip

Create a minimal example that best describes a typical usage scenario of yours and name it like ‘compile.py’ or so. You can integrate this file into your CI/CD workflow.

A Quick Guide#

Installation#

Neumann can be installed (either in a virtual enviroment or globally) from PyPI using pip on Python >= 3.7:

>>> pip install neumann

or chechkout with the following command using GitHub CLI

>>> gh repo clone dewloosh/Neumann

and install from source by typing

>>> python install setup.py

Motivating Examples#

Linear Algebra#

Define a reference frame $\mathbf{B}$ relative to the frame $\mathbf{A}$:

>>> from neumann.linalg import ReferenceFrame, Vector, Tensor
>>> A = ReferenceFrame(name='A', axes=np.eye(3))
>>> B = A.orient_new('Body', [0, 0, 90*np.pi/180], 'XYZ', name='B')

Get the DCM matrix of the transformation between two frames:

>>> B.dcm(target=A)

Define a vector $\mathbf{v}$ in frame $\mathbf{A}$ and show the components of it in frame $\mathbf{B}$:

>>> v = Vector([0.0, 1.0, 0.0], frame=A)
>>> v.show(B)

Define the same vector in frame $\mathbf{B}$:

>>> v = Vector(v.show(B), frame=B)
>>> v.show(A)

Linear Programming#

Solve the following Linear Programming Problem (LPP) with one unique solution:

>>> from neumann.optimize import LinearProgrammingProblem as LPP
>>> from neumann.function import Function, Equality
>>> import sympy as sy
>>> variables = ['x1', 'x2', 'x3', 'x4']
>>> x1, x2, x3, x4 = syms = sy.symbols(variables, positive=True)
>>> obj1 = Function(3*x1 + 9*x3 + x2 + x4, variables=syms)
>>> eq11 = Equality(x1 + 2*x3 + x4 - 4, variables=syms)
>>> eq12 = Equality(x2 + x3 - x4 - 2, variables=syms)
>>> problem = LPP(cost=obj1, constraints=[eq11, eq12], variables=syms)
>>> problem.solve()['x']
array([0., 6., 0., 4.])

NonLinear Programming#

Find the minimizer of the Rosenbrock function:

>>> from neumann.optimize import BinaryGeneticAlgorithm
>>> def Rosenbrock(x):
...     a, b = 1, 100
...     return (a-x[0])**2 + b*(x[1]-x[0]**2)**2
>>> ranges = [[-10, 10], [-10, 10]]
>>> BGA = BinaryGeneticAlgorithm(Rosenbrock, ranges, length=12, nPop=200)
>>> BGA.solve()
...

Indices and tables#