MathFlow

A Pythonic Interface for Symbolic and Numerical Mathematics

MathFlow bridges the gap between symbolic mathematics (SymPy) and numerical computations (NumPy/SciPy), offering a unified interface that maintains mathematical rigor while providing practical tools for real-world problems.

Ready to revolutionize your mathematical computing workflow?

pip install mathflow

Have Questions? Take a look at the Q&A: Questions & Answers: Addressing Potential Concerns

Key Features

Operative Closure: Mathematical operations return new Expression objects by default
Mutability Control: Choose between immutable (default) and mutable expressions for different workflows
Seamless Numerical Integration: Every symbolic expression has a .n attribute providing numerical methods without manual lambdification (uses cached lambdified expression when needed)
Enhanced Printing: Flexible output formatting through the .print attribute (LaTeX, pretty printing, code generation)
Signal System: Qt-like signals for tracking expression mutations and clones, enabling reactive programming
Automatic Type Conversions: Seamlessly and automatically converts between internal Poly and Expr representations based on context
Lightweight: ~0.5 MB itself, ~100 MB including dependencies
Fully backward compatible: Seamlessly integrate SymPy and MathFlow in the same script. All methods that work on SymPy Expr or Poly objects work on MathFlow objects
Exploratory: Full IDE support, enabling easy tool finding and minimizing the learning curve.

Quick Start

from mathflow import Expression, Polynomial, Rational

# Create expressions naturally
f = Expression("2x^2 + 3x + \frac{1}{2}")  # latex is automatically parsed
g = Expression("sin(x) + cos(x)")

# Automatic operative closure - operations return new objects of the same type
h = f + g  # f and g remain unchanged
hprime = h.diff()  # hprime is still an Expression object

# Numerical evaluation made easy
result = f(2.5)  # Numerically evaluate at x = 2.5

# Use the .n attribute to access fast numerical methods
numerical_roots = f.n.all_roots()
# Call f's n-prefixed methods to use variable precision numerical methods
precise_roots = f.nsolve_all(prec=50)  # 50 digits of accuracy

# quick and easy printing
f.print()
f.print('latex')
f.print('mathematica_code')
# or
print(f.print.latex())    # LaTeX output
print(f.print.ccode())    # c code output

Numerical Computing

MathFlow excels at bridging symbolic and numerical mathematics:

f = Expression("x^3 - 2x^2 + x - 1")

# Root finding
all_roots = f.n.all_roots(bounds=(-5, 5))
specific_root = f.nsolve_all(bounds=(-5, 5), prec=50)  # High-precision solve

# Numerical calculus
derivative_func = f.n.derivative_lambda(df_order=2)  # 2nd derivative numerical function  
integral_result = f.n.integrate(-1, 1)               # Definite integral  

# Optimization
minimum = f.n.minimize(bounds=[(-2, 2)])

Core Classes

graph BT
    %% Visual hierarchy with correct arrow semantics
    sympy.Expr
    sympy.Poly
    BaseExpression
    Expression
    Polynomial
    Rational
    Function
    
    %% Arrows still point child → parent
    BaseExpression -.-> sympy.Expr
    BaseExpression -.-> sympy.Poly
    Expression --> n
    n --> |lambdified numerical representation| A["f(x)"]
    n -.-> scipy.numerical_methods
    Expression --> BaseExpression
    Polynomial --> Expression
    Rational --> Expression
    Function --> |Only an Alias| Expression

Diagram Notes

Dotted arrows mean "proxy to".

Additional methods are not shown, only the core structure.

Expression

The primary class for general symbolic expressions with numerical and printing capabilities.

# Create from string with natural notation
expr = Expression("2x^2 + ln(|x-1|)")

# Or from SymPy objects
from sympy import sin, cos
expr = Expression(sin(x) + cos(x))

# Symbolic operations
derivative = expr.diff(x)
expanded = expr.expand()

# Numerical methods via .n attribute
roots = expr.n.all_roots(bounds=(-10, 10))
integral = expr.n.quad(0, 1)  # Numerical integration

Polynomial

Specialized Expression subclass with polynomial-specific functionality.

# Create from coefficients (ascending order by default)
poly = Polynomial.from_coeffs([1, 2, 3])  # 1 + 2x + 3x²

# Create from roots
poly = Polynomial.from_roots([1, 2, 3])   # (x-1)(x-2)(x-3)  

# Access polynomial-specific methods
coeffs = poly.all_coeffs()
degree = poly.degree()
roots = poly.n.all_poly_roots()  # Optimized polynomial root finding

Rational

For rational functions (quotients of polynomials).

# Create from numerator and denominator coefficients  
rational = Rational.from_coeffs([1, 2], [1, 1, 1])  # (1 + 2x)/(1 + x + x²)  
  
# Access numerator and denominator as Expression objects  
num = rational.numerator  
den = rational.denominator  
  
# Partial fraction decomposition
poles = den.n.all_roots()  # numerically find the poles.
pf = rational.partial_fractions(x)

Advanced Features

Mutability Control

# Immutable (default)
f = Expression("x^2")
g = f + 1  # f unchanged, g is new Expression object

# Mutable Mode (f is modified in-place)
f = Expression("(x-1)^2", mutable=True)
f += 1
f.expand()

Flexible String Parsing

# Natural mathematical notation
expr = Expression("2x^2 + ln(|x-1|)")

# LaTeX input support
expr = Expression(r"\frac{x^2+1}{x-1}")

# Implicit multiplication
expr = Expression("2x sin(x)")

Signal System for Reactive Programming

def on_change(expr):  # runs whenever an operation changes f
    print(f"Expression changed to: {expr}")  
f = Expression("x^2")  
f.on_self_mutated.connect(on_change)  
f.on_self_cloned.connect(on_change)
f += 1  # on_change() is called

Padé Approximations (Useful in Engineering Applications)

# High-quality Padé approximants for function approximation
f = Expression("exp(x)")
pade_approx = f.pade(m=3, n=3, x0=0)  # [3/3] Padé approximant around x=0
  
# Multiple backends available (returns Expression by default, but you can change return type)
pade_fast = f.pade(3, 3, backend='mpmath')      # Fast numerical
pade_exact = f.pade(3, 3, backend='symbolic')   # Exact symbolic
pade_verbose = f.pade(3, 3, backend='verbose')  # Educational output

When backend='verbose', each step is displayed as the Padé approximation is computed.

f = Expression("sqrt(x)")
p = f.pade(2, 2, x0=1, backend='verbose')

For example, the above displays this:

Step 1. Create rational function with numerator P and denominator Q, each with unknown coefficients:
(a0 + a1*h + a2*h^2)/(1 + b1*h + b2*h^2)

Step 2. Equate the rational function to the taylor series A so that the unknown coefficients may be solved:
(a0 + a1*h + a2*h^2)/(1 + b1*h + b2*h^2)  =  1 + 1/2*h + -1/8*h^2 + 1/16*h^3 + -5/128*h^4

Step 3. Multiply the rhs by the denominator of the lhs to get the equation in the form P = QA:
a0 + a1*h + a2*h^2  =  (1 + b1*h + b2*h^2) (1 + 1/2*h + -1/8*h^2 + 1/16*h^3 + -5/128*h^4)

Step 4. Expand the RHS by performing discrete convolution on the coefficient vectors of Q and A (using a table):
	Q's coeffs = [1, b1, b2]
	A's coeffs = [1, 1/2, -1/8, 1/16, -5/128]
╭────┬─────┬────────┬───────────┬─────────┬─────────────╮
│    │ 1   │ 1/2    │ -1/8      │ 1/16    │ -5/128      │
├────┼─────┼────────┼───────────┼─────────┼─────────────┤
│ 1  │ 1   │ 1*1/2  │ 1*(-1/8)  │ 1*1/16  │ 1*(-5/128)  │
├────┼─────┼────────┼───────────┼─────────┼─────────────┤
│ b1 │ b1  │ b1*1/2 │ b1*(-1/8) │ b1*1/16 │ b1*(-5/128) │
├────┼─────┼────────┼───────────┼─────────┼─────────────┤
│ b2 │ b2  │ b2*1/2 │ b2*(-1/8) │ b2*1/16 │ b2*(-5/128) │
╰────┴─────┴────────┴───────────┴─────────┴─────────────╯

Step 5. Get the sum of the anti-diagonals from the above table to form the new coeffs (only as many terms as unknown coeffs we need to solve for, in this case 5):
╭────────┬──────────────────────╮
│ Term   │ Coeff                │
├────────┼──────────────────────┤
│ h^0    │ 1                    │
├────────┼──────────────────────┤
│ h^1    │ b1 + 1/2             │
├────────┼──────────────────────┤
│ h^2    │ b1/2 + b2 - 1/8      │
├────────┼──────────────────────┤
│ h^3    │ -b1/8 + b2/2 + 1/16  │
├────────┼──────────────────────┤
│ h^4    │ b1/16 - b2/8 - 5/128 │
╰────────┴──────────────────────╯

Step 6. Use these coefficients to setup a system of equations:
a0 = 1
a1 = b1 + 1/2
a2 = b1/2 + b2 - 1/8
 0 = -b1/8 + b2/2 + 1/16
 0 = b1/16 - b2/8 - 5/128

Step 7. Solving the above system yields:
a0 = 1
a1 = 5/4
a2 = 5/16
b1 = 3/4
b2 = 1/16

Step 8. Substituting these values back into the original rational function yields:
(1 + 5/4*h + 5/16*h^2)/(1 + 3/4*h + 1/16*h^2)

Step 9. `h` may be substituted for the original (x-c), and then expanded. `c` is the point at which both the Taylor series and the Padé approximation are centered at:
(1 + 5/4*(x-c) + 5/16*(x-c)^2)/(1 + 3/4*(x-c) + 1/16*(x-c)^2)

Design Philosophy

MathFlow follows several key principles:

Intuitive API: Mathematical operations should feel natural in Python, providing an "exploratory" experience
Performance: Automatic optimizations (Horner's method, efficient algorithms, automatic numerical computing when needed)
Flexibility: Support both functional and OOP programming styles
Extensibility: Easy integration with other mathematical libraries
Type Safety: Comprehensive type and method hints for full I

MathFlow

Install / Use

README

MathFlow

Key Features

Quick Start

Numerical Computing

Core Classes

Expression

Polynomial

Rational

Advanced Features

Mutability Control

Flexible String Parsing

Signal System for Reactive Programming

Padé Approximations (Useful in Engineering Applications)

Design Philosophy