solana-program-library/libraries/math/src/precise_number.rs

570 lines
20 KiB
Rust

//! Defines PreciseNumber, a U256 wrapper with float-like operations
use crate::uint::U256;
// Allows for easy swapping between different internal representations
type InnerUint = U256;
/// The representation of the number one as a precise number as 10^12
pub const ONE: u128 = 1_000_000_000_000;
/// Struct encapsulating a fixed-point number that allows for decimal calculations
#[derive(Clone, Debug, PartialEq)]
pub struct PreciseNumber {
/// Wrapper over the inner value, which is multiplied by ONE
pub value: InnerUint,
}
/// The precise-number 1 as a InnerUint
fn one() -> InnerUint {
InnerUint::from(ONE)
}
/// The number 0 as a PreciseNumber, used for easier calculations.
fn zero() -> InnerUint {
InnerUint::from(0)
}
impl PreciseNumber {
/// Correction to apply to avoid truncation errors on division. Since
/// integer operations will always floor the result, we artifically bump it
/// up by one half to get the expect result.
fn rounding_correction() -> InnerUint {
InnerUint::from(ONE / 2)
}
/// Desired precision for the correction factor applied during each
/// iteration of checked_pow_approximation. Once the correction factor is
/// smaller than this number, or we reach the maxmium number of iterations,
/// the calculation ends.
fn precision() -> InnerUint {
InnerUint::from(100)
}
fn zero() -> Self {
Self { value: zero() }
}
fn one() -> Self {
Self { value: one() }
}
/// Maximum number iterations to apply on checked_pow_approximation.
const MAX_APPROXIMATION_ITERATIONS: u128 = 100;
/// Minimum base allowed when calculating exponents in checked_pow_fraction
/// and checked_pow_approximation. This simply avoids 0 as a base.
fn min_pow_base() -> InnerUint {
InnerUint::from(1)
}
/// Maximum base allowed when calculating exponents in checked_pow_fraction
/// and checked_pow_approximation. The calculation use a Taylor Series
/// approxmation around 1, which converges for bases between 0 and 2. See
/// https://en.wikipedia.org/wiki/Binomial_series#Conditions_for_convergence
/// for more information.
fn max_pow_base() -> InnerUint {
InnerUint::from(2 * ONE)
}
/// Create a precise number from an imprecise u128, should always succeed
pub fn new(value: u128) -> Option<Self> {
let value = InnerUint::from(value).checked_mul(one())?;
Some(Self { value })
}
/// Convert a precise number back to u128
pub fn to_imprecise(&self) -> Option<u128> {
self.value
.checked_add(Self::rounding_correction())?
.checked_div(one())
.map(|v| v.as_u128())
}
/// Checks that two PreciseNumbers are equal within some tolerance
pub fn almost_eq(&self, rhs: &Self, precision: InnerUint) -> bool {
let (difference, _) = self.unsigned_sub(rhs);
difference.value < precision
}
/// Checks that a number is less than another
pub fn less_than(&self, rhs: &Self) -> bool {
self.value < rhs.value
}
/// Checks that a number is greater than another
pub fn greater_than(&self, rhs: &Self) -> bool {
self.value > rhs.value
}
/// Checks that a number is less than another
pub fn less_than_or_equal(&self, rhs: &Self) -> bool {
self.value <= rhs.value
}
/// Checks that a number is greater than another
pub fn greater_than_or_equal(&self, rhs: &Self) -> bool {
self.value >= rhs.value
}
/// Floors a precise value to a precision of ONE
pub fn floor(&self) -> Option<Self> {
let value = self.value.checked_div(one())?.checked_mul(one())?;
Some(Self { value })
}
/// Ceiling a precise value to a precision of ONE
pub fn ceiling(&self) -> Option<Self> {
let value = self
.value
.checked_add(one().checked_sub(InnerUint::from(1))?)?
.checked_div(one())?
.checked_mul(one())?;
Some(Self { value })
}
/// Performs a checked division on two precise numbers
pub fn checked_div(&self, rhs: &Self) -> Option<Self> {
if *rhs == Self::zero() {
return None;
}
match self.value.checked_mul(one()) {
Some(v) => {
let value = v
.checked_add(Self::rounding_correction())?
.checked_div(rhs.value)?;
Some(Self { value })
}
None => {
let value = self
.value
.checked_add(Self::rounding_correction())?
.checked_div(rhs.value)?
.checked_mul(one())?;
Some(Self { value })
}
}
}
/// Performs a multiplication on two precise numbers
pub fn checked_mul(&self, rhs: &Self) -> Option<Self> {
match self.value.checked_mul(rhs.value) {
Some(v) => {
let value = v
.checked_add(Self::rounding_correction())?
.checked_div(one())?;
Some(Self { value })
}
None => {
let value = if self.value >= rhs.value {
self.value.checked_div(one())?.checked_mul(rhs.value)?
} else {
rhs.value.checked_div(one())?.checked_mul(self.value)?
};
Some(Self { value })
}
}
}
/// Performs addition of two precise numbers
pub fn checked_add(&self, rhs: &Self) -> Option<Self> {
let value = self.value.checked_add(rhs.value)?;
Some(Self { value })
}
/// Subtracts the argument from self
pub fn checked_sub(&self, rhs: &Self) -> Option<Self> {
let value = self.value.checked_sub(rhs.value)?;
Some(Self { value })
}
/// Performs a subtraction, returning the result and whether the result is negative
pub fn unsigned_sub(&self, rhs: &Self) -> (Self, bool) {
match self.value.checked_sub(rhs.value) {
None => {
let value = rhs.value.checked_sub(self.value).unwrap();
(Self { value }, true)
}
Some(value) => (Self { value }, false),
}
}
/// Performs pow on a precise number
pub fn checked_pow(&self, exponent: u128) -> Option<Self> {
// For odd powers, start with a multiplication by base since we halve the
// exponent at the start
let value = if exponent.checked_rem(2)? == 0 {
one()
} else {
self.value
};
let mut result = Self { value };
// To minimize the number of operations, we keep squaring the base, and
// only push to the result on odd exponents, like a binary decomposition
// of the exponent.
let mut squared_base = self.clone();
let mut current_exponent = exponent.checked_div(2)?;
while current_exponent != 0 {
squared_base = squared_base.checked_mul(&squared_base)?;
// For odd exponents, "push" the base onto the value
if current_exponent.checked_rem(2)? != 0 {
result = result.checked_mul(&squared_base)?;
}
current_exponent = current_exponent.checked_div(2)?;
}
Some(result)
}
/// Approximate the nth root of a number using a Taylor Series around 1 on
/// x ^ n, where 0 < n < 1, result is a precise number.
/// Refine the guess for each term, using:
/// 1 2
/// f(x) = f(a) + f'(a) * (x - a) + --- * f''(a) * (x - a) + ...
/// 2!
/// For x ^ n, this gives:
/// n n n-1 1 n-2 2
/// x = a + n * a (x - a) + --- * n * (n - 1) a (x - a) + ...
/// 2!
///
/// More simply, this means refining the term at each iteration with:
///
/// t_k+1 = t_k * (x - a) * (n + 1 - k) / k
///
/// where a = 1, n = power, x = precise_num
/// NOTE: this function is private because its accurate range and precision
/// have not been estbalished.
fn checked_pow_approximation(&self, exponent: &Self, max_iterations: u128) -> Option<Self> {
assert!(self.value >= Self::min_pow_base());
assert!(self.value <= Self::max_pow_base());
let one = Self::one();
if *exponent == Self::zero() {
return Some(one);
}
let mut precise_guess = one.clone();
let mut term = precise_guess.clone();
let (x_minus_a, x_minus_a_negative) = self.unsigned_sub(&precise_guess);
let exponent_plus_one = exponent.checked_add(&one)?;
let mut negative = false;
for k in 1..max_iterations {
let k = Self::new(k)?;
let (current_exponent, current_exponent_negative) = exponent_plus_one.unsigned_sub(&k);
term = term.checked_mul(&current_exponent)?;
term = term.checked_mul(&x_minus_a)?;
term = term.checked_div(&k)?;
if term.value < Self::precision() {
break;
}
if x_minus_a_negative {
negative = !negative;
}
if current_exponent_negative {
negative = !negative;
}
if negative {
precise_guess = precise_guess.checked_sub(&term)?;
} else {
precise_guess = precise_guess.checked_add(&term)?;
}
}
Some(precise_guess)
}
/// Get the power of a number, where the exponent is expressed as a fraction
/// (numerator / denominator)
/// NOTE: this function is private because its accurate range and precision
/// have not been estbalished.
#[allow(dead_code)]
fn checked_pow_fraction(&self, exponent: &Self) -> Option<Self> {
assert!(self.value >= Self::min_pow_base());
assert!(self.value <= Self::max_pow_base());
let whole_exponent = exponent.floor()?;
let precise_whole = self.checked_pow(whole_exponent.to_imprecise()?)?;
let (remainder_exponent, negative) = exponent.unsigned_sub(&whole_exponent);
assert!(!negative);
if remainder_exponent.value == InnerUint::from(0) {
return Some(precise_whole);
}
let precise_remainder = self
.checked_pow_approximation(&remainder_exponent, Self::MAX_APPROXIMATION_ITERATIONS)?;
precise_whole.checked_mul(&precise_remainder)
}
/// Approximate the nth root of a number using Newton's method
/// https://en.wikipedia.org/wiki/Newton%27s_method
/// NOTE: this function is private because its accurate range and precision
/// have not been established.
fn newtonian_root_approximation(
&self,
root: &Self,
mut guess: Self,
iterations: u128,
) -> Option<Self> {
let zero = Self::zero();
if *self == zero {
return Some(zero);
}
if *root == zero {
return None;
}
let one = Self::new(1)?;
let root_minus_one = root.checked_sub(&one)?;
let root_minus_one_whole = root_minus_one.to_imprecise()?;
let mut last_guess = guess.clone();
let precision = Self::precision();
for _ in 0..iterations {
// x_k+1 = ((n - 1) * x_k + A / (x_k ^ (n - 1))) / n
let first_term = root_minus_one.checked_mul(&guess)?;
let power = guess.checked_pow(root_minus_one_whole);
let second_term = match power {
Some(num) => self.checked_div(&num)?,
None => Self::new(0)?,
};
guess = first_term.checked_add(&second_term)?.checked_div(&root)?;
if last_guess.almost_eq(&guess, precision) {
break;
} else {
last_guess = guess.clone();
}
}
Some(guess)
}
/// Based on testing around the limits, this base is the smallest value that
/// provides an epsilon 11 digits
fn minimum_sqrt_base() -> Self {
Self {
value: InnerUint::from(0),
}
}
/// Based on testing around the limits, this base is the smallest value that
/// provides an epsilon of 11 digits
fn maximum_sqrt_base() -> Self {
Self::new(std::u128::MAX).unwrap()
}
/// Approximate the square root using Newton's method. Based on testing,
/// this provides a precision of 11 digits for inputs between 0 and u128::MAX
pub fn sqrt(&self) -> Option<Self> {
if self.less_than(&Self::minimum_sqrt_base())
|| self.greater_than(&Self::maximum_sqrt_base())
{
return None;
}
let two = PreciseNumber::new(2)?;
let one = PreciseNumber::new(1)?;
// A good initial guess is the average of the interval that contains the
// input number. For all numbers, that will be between 1 and the given number.
let guess = self.checked_add(&one)?.checked_div(&two)?;
self.newtonian_root_approximation(&two, guess, Self::MAX_APPROXIMATION_ITERATIONS)
}
}
#[cfg(test)]
mod tests {
use super::*;
use proptest::prelude::*;
fn check_pow_approximation(base: InnerUint, exponent: InnerUint, expected: InnerUint) {
let precision = InnerUint::from(5_000_000); // correct to at least 3 decimal places
let base = PreciseNumber { value: base };
let exponent = PreciseNumber { value: exponent };
let root = base
.checked_pow_approximation(&exponent, PreciseNumber::MAX_APPROXIMATION_ITERATIONS)
.unwrap();
let expected = PreciseNumber { value: expected };
assert!(root.almost_eq(&expected, precision));
}
#[test]
fn test_root_approximation() {
let one = one();
// square root
check_pow_approximation(one / 4, one / 2, one / 2); // 1/2
check_pow_approximation(one * 11 / 10, one / 2, InnerUint::from(1_048808848161u128)); // 1.048808848161
// 5th root
check_pow_approximation(one * 4 / 5, one * 2 / 5, InnerUint::from(914610103850u128));
// 0.91461010385
// 10th root
check_pow_approximation(one / 2, one * 4 / 50, InnerUint::from(946057646730u128));
// 0.94605764673
}
fn check_pow_fraction(
base: InnerUint,
exponent: InnerUint,
expected: InnerUint,
precision: InnerUint,
) {
let base = PreciseNumber { value: base };
let exponent = PreciseNumber { value: exponent };
let power = base.checked_pow_fraction(&exponent).unwrap();
let expected = PreciseNumber { value: expected };
assert!(power.almost_eq(&expected, precision));
}
#[test]
fn test_pow_fraction() {
let one = one();
let precision = InnerUint::from(50_000_000); // correct to at least 3 decimal places
let less_precision = precision * 1_000; // correct to at least 1 decimal place
check_pow_fraction(one, one, one, precision);
check_pow_fraction(
one * 20 / 13,
one * 50 / 3,
InnerUint::from(1312_534484739100u128),
precision,
); // 1312.5344847391
check_pow_fraction(one * 2 / 7, one * 49 / 4, InnerUint::from(2163), precision);
check_pow_fraction(
one * 5000 / 5100,
one / 9,
InnerUint::from(997802126900u128),
precision,
); // 0.99780212695
// results get less accurate as the base gets further from 1, so allow
// for a greater margin of error
check_pow_fraction(
one * 2,
one * 27 / 5,
InnerUint::from(42_224253144700u128),
less_precision,
); // 42.2242531447
check_pow_fraction(
one * 18 / 10,
one * 11 / 3,
InnerUint::from(8_629769290500u128),
less_precision,
); // 8.629769290
}
#[test]
fn test_newtonian_approximation() {
// square root
let test = PreciseNumber::new(9).unwrap();
let nth_root = PreciseNumber::new(2).unwrap();
let guess = test.checked_div(&nth_root).unwrap();
let root = test
.newtonian_root_approximation(
&nth_root,
guess,
PreciseNumber::MAX_APPROXIMATION_ITERATIONS,
)
.unwrap()
.to_imprecise()
.unwrap();
assert_eq!(root, 3); // actually 3
let test = PreciseNumber::new(101).unwrap();
let nth_root = PreciseNumber::new(2).unwrap();
let guess = test.checked_div(&nth_root).unwrap();
let root = test
.newtonian_root_approximation(
&nth_root,
guess,
PreciseNumber::MAX_APPROXIMATION_ITERATIONS,
)
.unwrap()
.to_imprecise()
.unwrap();
assert_eq!(root, 10); // actually 10.049875
let test = PreciseNumber::new(1_000_000_000).unwrap();
let nth_root = PreciseNumber::new(2).unwrap();
let guess = test.checked_div(&nth_root).unwrap();
let root = test
.newtonian_root_approximation(
&nth_root,
guess,
PreciseNumber::MAX_APPROXIMATION_ITERATIONS,
)
.unwrap()
.to_imprecise()
.unwrap();
assert_eq!(root, 31_623); // actually 31622.7766
// 5th root
let test = PreciseNumber::new(500).unwrap();
let nth_root = PreciseNumber::new(5).unwrap();
let guess = test.checked_div(&nth_root).unwrap();
let root = test
.newtonian_root_approximation(
&nth_root,
guess,
PreciseNumber::MAX_APPROXIMATION_ITERATIONS,
)
.unwrap()
.to_imprecise()
.unwrap();
assert_eq!(root, 3); // actually 3.46572422
}
fn check_square_root(check: &PreciseNumber) {
let epsilon = PreciseNumber {
value: InnerUint::from(10),
}; // correct within 11 decimals
let one = PreciseNumber::one();
let one_plus_epsilon = one.checked_add(&epsilon).unwrap();
let one_minus_epsilon = one.checked_sub(&epsilon).unwrap();
let approximate_root = check.sqrt().unwrap();
let lower_bound = approximate_root
.checked_mul(&one_minus_epsilon)
.unwrap()
.checked_pow(2)
.unwrap();
let upper_bound = approximate_root
.checked_mul(&one_plus_epsilon)
.unwrap()
.checked_pow(2)
.unwrap();
assert!(check.less_than_or_equal(&upper_bound));
assert!(check.greater_than_or_equal(&lower_bound));
}
#[test]
fn test_square_root_min_max() {
let test_roots = [
PreciseNumber::minimum_sqrt_base(),
PreciseNumber::maximum_sqrt_base(),
];
for i in test_roots.iter() {
check_square_root(i);
}
}
#[test]
fn test_floor() {
let whole_number = PreciseNumber::new(2).unwrap();
let mut decimal_number = PreciseNumber::new(2).unwrap();
decimal_number.value += InnerUint::from(1);
let floor = decimal_number.floor().unwrap();
let floor_again = floor.floor().unwrap();
assert_eq!(whole_number.value, floor.value);
assert_eq!(whole_number.value, floor_again.value);
}
#[test]
fn test_ceiling() {
let whole_number = PreciseNumber::new(2).unwrap();
let mut decimal_number = PreciseNumber::new(2).unwrap();
decimal_number.value -= InnerUint::from(1);
let ceiling = decimal_number.ceiling().unwrap();
let ceiling_again = ceiling.ceiling().unwrap();
assert_eq!(whole_number.value, ceiling.value);
assert_eq!(whole_number.value, ceiling_again.value);
}
proptest! {
#[test]
fn test_square_root(a in 0..u128::MAX) {
let a = PreciseNumber { value: InnerUint::from(a) };
check_square_root(&a);
}
}
}