rbca 0.1.0

Recursive Binary Cosine Approximation

rbca

rbca (Recursive Binary Cosine Approximation) is a trigonometric approximation library with both symbolic and float APIs.

It computes cos, sin, and tan through recursive binary partitioning on angles.

The default APIs keep expressions in exact symbolic form (SymPy Expr) instead of forcing early floating-point evaluation. The _float APIs use float arithmetic internally for practical numeric calculation.

Current package version is 0.1.0.

Features

• Symbolic trig outputs using SymPy expressions
• Practical float trig outputs using float arithmetic
• Recursive midpoint-based approximation for angle intervals
• Angle normalization over full 0~360 degree input range
• Three API families:
  • calc_*_pure: mathematically pure symbolic recursion from 0 to 90
  • calc_*: symbolic recursion seeded with well-known anchor angles (0, 15, 30, 45, 60, 75)
  • calc_*_float: practical float calculation using the same anchor strategy as calc_*

Installation

Using pip

pip install rbca

Local development (this repository)

pip install -e .

Project requirements:

• Python >=3.10
• sympy>=1.10.0

Quick Start

from rbca import calc_cos, calc_cos_float, calc_sin, calc_sin_float, calc_tan
from sympy import sqrtdenest

depth = 5

cos_expr = calc_cos(37, depth)      # SymPy Expr
sin_expr = calc_sin(37, depth)      # SymPy Expr
tan_expr = calc_tan(37, depth)      # SymPy Expr

cos_value = calc_cos_float(37, depth)  # float
sin_value = calc_sin_float(37, depth)  # float

print(cos_expr)                     # symbolic expression
print(float(cos_expr))              # numeric projection
print(sqrtdenest(cos_expr))         # optional radical denesting
print(cos_value)                    # practical float result

Choosing an API Family

• Use calc_cos_pure, calc_sin_pure, and calc_tan_pure when you want the mathematically pure symbolic path.
  • This path starts from [0, 90] only.
• Use calc_cos, calc_sin, and calc_tan when you want symbolic results with a more practical anchored start.
  • This path uses anchor angles (0, 15, 30, 45, 60, 75, 90) to start from a narrower interval.
• Use calc_cos_float, calc_sin_float, and calc_tan_float when you want practical float results.
  • This path is implemented with float arithmetic internally.

Core API

• calc_cos(angle, depth) -> Expr

• calc_sin(angle, depth) -> Expr

• calc_tan(angle, depth) -> Expr

• calc_cos_pure(angle, depth) -> Expr

• calc_sin_pure(angle, depth) -> Expr

• calc_tan_pure(angle, depth) -> Expr

• calc_cos_float(angle, depth) -> float

• calc_sin_float(angle, depth) -> float

• calc_tan_float(angle, depth) -> float

For symbolic APIs, angle accepts Rational | int | float | str and is internally converted with sympy.Rational where needed.

Symbolic input note:

• float inputs are converted through Rational(float_value), which can preserve binary-float artifacts.
• For exact decimal intent, prefer str or Rational directly.
  • Example: use "0.1" instead of 0.1.

For float APIs, angle accepts int | float | str and is internally converted with float(angle).

depth is recursion depth:

• depth = 0: return one of the current interval endpoints
• larger depth: finer binary partitioning and generally tighter approximation

How It Works

1. Normalize input angle to first quadrant (0~90) and track output sign.
2. Choose initial bracket interval:
   • calc_cos_pure: [0, 90]
   • calc_cos: narrower interval between predefined anchor angles when possible.
   • calc_cos_float: same anchored interval strategy as calc_cos, but with float nodes and float values.
3. Recursively bisect the interval by midpoint angle.
4. Compute midpoint cosine via:
   • a product/half-angle style expression using endpoint cosine values.
5. Recurse until depth == 0 (or exact angle hit), then apply sign flip if needed.

Notes and Limitations

• This library is symbolic-first; expressions can grow quickly as depth increases.
• For angles where cosine is zero (e.g. 90 + 180k degrees), calc_tan and calc_tan_pure return sympy.zoo.
• For the same tangent singularities, calc_tan_float lets Python's ZeroDivisionError propagate.
• Inputs with symbolic free variables are rejected during normalization.

Maths

Condition:

$$ 0^\circ \le A,B \le 90^\circ $$

Using the half-angle formula,

$$ \cos\frac{A+B}{2}

\sqrt{\frac{1+\cos(A+B)}{2}} $$

Also, by the angle addition formula,

$$ \cos(A+B)

\cos A\cos B-\sin A\sin B $$

Since

$$ 0^\circ \le A,B \le 90^\circ $$

we can use

$$ \sin A=\sqrt{1-\cos^2 A} $$

and

$$ \sin B=\sqrt{1-\cos^2 B} $$

Therefore,

$$ \cos(A+B)

\cos A\cos B

\sqrt{(1-\cos^2 A)(1-\cos^2 B)} $$

Substituting this into the half-angle formula finally gives

$$ \boxed{ \cos\frac{A+B}{2}

\sqrt{ \frac{ 1+\cos A\cos B-\sqrt{(1-\cos^2 A)(1-\cos^2 B)} }{2} } } $$

License

MIT License. See LICENSE.