Trait ExactF64Conversion

Source

pub trait ExactF64Conversion {
    type Output;

    // Required methods
    fn try_to_f64(&self) -> Result<Self::Output, LaError>;
    fn to_rounded_f64(&self) -> Result<Self::Output, LaError>;
}

Expand description

Convert an already-computed exact result to finite binary64 output.

This extension trait is implemented for BigRational determinants and [BigRational; D] exact solutions. It lets callers retain the exact value, try the strict no-rounding contract, and recover with explicit rounding without repeating determinant evaluation or linear-system elimination. BigRational::new_raw values are interpreted by their mathematical quotient: denominator signs and common factors do not change the result. A zero denominator is rejected as UnrepresentableReason::NotFinite.

§Examples

use la_stack::prelude::*;

let matrix = Matrix::<2>::try_from_rows([
    [1.0 + f64::EPSILON, 0.0],
    [0.0, 1.0 - f64::EPSILON],
])?;
let exact = matrix.det_exact()?;
let rounded = match exact.try_to_f64() {
    Ok(value) => value,
    Err(error) if error.requires_rounding() => exact.to_rounded_f64()?,
    Err(error) => return Err(error),
};
assert_eq!(rounded.to_bits(), 1.0_f64.to_bits());

let system = Matrix::<1>::try_from_rows([[3.0]])?;
let rhs = Vector::<1>::try_new([3.0])?;
let exact_solution = system.solve_exact(rhs)?;
assert_eq!(exact_solution.try_to_f64()?.into_array(), [1.0]);

Required Associated Types§

Source

type Output

Finite binary64 output produced by the conversion.

Required Methods§

Source

fn try_to_f64(&self) -> Result<Self::Output, LaError>

Convert only when every exact value already has an exact finite binary64 representation.

The candidate conversion follows IEEE 754 round-to-nearest, ties-to-even, but this strict method returns it only when no rounding is required.

§Errors

Returns LaError::Unrepresentable with UnrepresentableReason::RequiresRounding when finite binary64 output would require rounding, or UnrepresentableReason::NotFinite when rounding cannot produce finite output. Exact solution errors include the first failing component index.

Source

fn to_rounded_f64(&self) -> Result<Self::Output, LaError>

Round the exact value to finite binary64 output.

Rounding follows IEEE 754 round-to-nearest, ties-to-even.

§Errors

Returns LaError::Unrepresentable with UnrepresentableReason::NotFinite when rounding cannot produce finite output. Exact solution errors include the first failing component index.