crypto-ratio

Rational number arithmetic using crypto-bigint crypto-bigint.

Built for use in RISC Zero zkVM applications to leverage accelerated crypto-bigint precompiles with rational number operations.

Features

• Generic over integer width: Works with U64, U256, U512, U1024, U2048, and all crypto-bigint types up to U16384
• RISC Zero optimized: Designed to work with RISC Zero's accelerated crypto-bigint precompiles
• Performance-focused: Deferred reduction and smart heuristics minimize expensive GCD operations
• Overflow handling: Automatic fallback to wider types when operations overflow
• no_std compatible: Works in embedded, constrained, and zkVM environments

Why RISC Zero?

This library was built to enable efficient rational number arithmetic in RISC Zero's zkVM, where crypto-bigint operations are hardware-accelerated through precompiles. By building on crypto-bigint rather than num-bigint, this crate can leverage:

• Hardware acceleration for big integer operations in zkVM
• Reduced cycle counts for cryptographic workloads (U256 and U512 extrem fast)

Special thanks to the RISC Zero team for building an incredible zkVM platform and for their excellent documentation on accelerator usage.

Installation

Add to your Cargo.toml:

[dependencies]
crypto-ratio = "0.1"
crypto-bigint = "0.5"

For RISC Zero zkVM guests:

[dependencies]
crypto-ratio = "0.1"
crypto-bigint = "0.5"
risc0-zkvm = { version = "3.0.3" }

Quick Start

use crypto_ratio::{Ratio, RatioU512};
use crypto_bigint::U512;

// Create rationals
let a = RatioU512::from_u64(1, 2);  // 1/2
let b = RatioU512::from_u64(1, 3);  // 1/3

// Arithmetic operations
let sum = &a + &b;  // 5/6
let product = &a * &b;  // 1/6

// Explicit reduction when needed
let mut result = product;
result.normalize();

// Comparison
assert!(a > b);

// Float conversion
let r = RatioU512::from_float(0.75).unwrap();
assert_eq!(r.to_f64_approx(), 0.75);

RISC Zero zkVM Example

use crypto_ratio::RatioU512;
use risc0_zkvm::guest::env;

pub fn main() {
    // Read inputs from host
    let a_numer: u64 = env::read();
    let a_denom: u64 = env::read();
    let b_numer: u64 = env::read();
    let b_denom: u64 = env::read();
    
    // Perform rational arithmetic (leverages crypto-bigint precompiles)
    let a = RatioU512::from_u64(a_numer, a_denom);
    let b = RatioU512::from_u64(b_numer, b_denom);
    
    let mut result = a.mul(&b);
    result.normalize();
    
    // Commit result
    env::commit(&result.numer.to_le_bytes());
    env::commit(&result.denom.to_le_bytes());
}

Design Philosophy

Unreduced Operations by Default

Operations like multiplication and addition return unreduced results for performance. Call normalize() explicitly when reduction is needed:

use crypto_ratio::RatioU256;

let a = RatioU256::from_u64(2, 3);
let b = RatioU256::from_u64(3, 4);

// Unreduced: 6/12
let product = &a * &b;

// Reduced: 1/2
let mut reduced = product;
reduced.normalize();

Why? In loops and chained operations, automatic reduction uses too much cycles:

// Bad: Reduces 100 times (expensive!)
let mut sum = RatioU512::zero();
for i in 0..100 {
    let term = RatioU512::from_u64(i, 1000);
    sum = sum + term;  // Auto-reduce would GCD here
}

// Good: Reduces once at the end
let mut sum = RatioU512::zero();
for i in 0..100 {
    let term = RatioU512::from_u64(i, 1000);
    sum = sum.add(&term);  // Unreduced
}
sum.normalize();  // Single GCD at the end

Smart Reduction Heuristics

mul_reduced() provides automatic reduction using fast heuristics:

let a = RatioU512::from_u64(2, 3);
let b = RatioU512::from_u64(3, 4);

// Smart reduction uses:
// 1. Bit shifts for power-of-2 factors (~10ns)
// 2. Fast u64 GCD for small values (~50ns)
// 3. Skips GCD for large coprime values (~0ns)
let product = a.mul_reduced(&b);  // 1/2 (reduced)

Generic Over Integer Width

Choose the size that fits your precision and performance needs: Heads up: Performance decreases from U2048 onwards due to automatic wide types on overflow, fix coming soon.

use crypto_ratio::{Ratio, RatioU256, RatioU512, RatioU1024};
use crypto_bigint::{U256, U512, U1024};

// Small: Fast operations, limited range
let small = RatioU256::from_u64(1, 2);

// Medium: Good balance (recommended)
let medium = RatioU512::from_u64(1, 2);

// Large: Maximum precision
let large = RatioU1024::from_u64(1, 2);

// Custom size
let custom = Ratio::<U256>::from_u64(1, 2);

Overflow Handling

Operations automatically use wider types when needed:

use crypto_ratio::RatioU256;

// U256 operations may internally use U512
let a = RatioU256::from_u64(u64::MAX, 1);
let b = RatioU256::from_u64(u64::MAX, 1);

// This works! Uses U512 internally, reduces back to U256
let sum = a.add(&b);

Real-World Example: Taylor Series

Computing e^x using Taylor series expansion demonstrates the performance benefits:

use crypto_ratio::RatioU512;
use crypto_bigint::U512;

fn exp_approx(x: RatioU512, terms: usize) -> RatioU512 {
    let mut result = RatioU512::one();
    let mut term = RatioU512::one();
    let mut factorial = U512::ONE;
    
    for n in 1..terms {
        factorial = factorial.wrapping_mul(&U512::from_u64(n as u64));
        term = term.mul(&x);
        
        let next_term = term.div_by_uint(&factorial);
        result = result.add(&next_term);
        
        // Only reduce when needed (heuristic-based)
        if result.needs_reduction() {
            result.normalize();
        }
    }
    
    result.normalize();
    result
}

let x = RatioU512::from_float(0.5).unwrap();
let e_to_half = exp_approx(x, 10);
println!("e^0.5 ≈ {}", e_to_half.to_f64_approx());

Trade-offs

Performance vs Automatic Reduction

Integer Size Selection

Performance

Benchmark Overview

Typical performance on modern hardware (U512):

Running Tests

# Run all tests (2,300+ tests across 9 integer sizes)
cargo test

# Run tests for specific size
cargo test u256_tests
cargo test u512_tests

# Run specific test
cargo test test_addition_compatibility

# Run with output
cargo test -- --nocapture

# Run tests with all features
cargo test --all-features

Running Benchmarks

# Run all benchmarks
cargo bench

# Run benchmarks for specific size
cargo bench U256
cargo bench U512

# Run specific operation
cargo bench multiplication
cargo bench taylor_series

# Run specific size + operation
cargo bench U512/multiplication

# Generate detailed report with violin plots
cargo bench -- --save-baseline main

# Compare against baseline
cargo bench -- --baseline main

Benchmark results are saved in target/criterion/ with detailed HTML reports including:

• Violin plots showing distribution
• Regression analysis
• Historical comparison
• Statistical significance tests

Viewing Benchmark Reports

After running benchmarks, open the HTML report:

# On macOS
open target/criterion/report/index.html

# On Linux
xdg-open target/criterion/report/index.html

# On Windows
start target/criterion/report/index.html

API Overview

Construction

// From u64 (with reduction)
let r = RatioU512::from_u64(2, 3);

// From float
let r = RatioU512::from_float(0.5).unwrap();

// Without reduction
let r = RatioU512::new_raw(numer, denom, false);

// With reduction
let r = RatioU512::new(numer, denom);

Arithmetic

let a = RatioU512::from_u64(1, 2);
let b = RatioU512::from_u64(1, 3);

// Basic operations (unreduced)
let sum = &a + &b;
let diff = &a - &b;
let prod = &a * &b;
let quot = &a / &b;

// Negation
let neg = -a;

// Reciprocal
let recip = a.recip();

Reduction

let mut r = RatioU512::from_u64(6, 8);

// Explicit reduction
r.normalize();  // Now 3/4

// Conditional reduction
if r.needs_reduction() {
    r.normalize();
}

// Smart multiplication
let a = RatioU512::from_u64(2, 3);
let b = RatioU512::from_u64(3, 4);
let prod = a.mul_reduced(&b);  // 1/2 (reduced)

Comparison

let a = RatioU512::from_u64(1, 2);
let b = RatioU512::from_u64(1, 3);

assert!(a > b);
assert!(a >= b);
assert!(b < a);
assert!(a != b);

no_std Support

This crate is no_std compatible (requires alloc for GCD operations):

[dependencies]
crypto-ratio = { version = "0.1", default-features = false }

Perfect for:

• Embedded systems
• RISC Zero zkVM guests
• WebAssembly
• Constrained environments

Minimum Supported Rust Version (MSRV)

Rust 1.65 or later.

License

Licensed under either of:

• Apache License, Version 2.0 (LICENSE-APACHE or http://www.apache.org/licenses/LICENSE-2.0)
• MIT license (LICENSE-MIT or http://opensource.org/licenses/MIT)

at your option.

Contributing

Contributions are welcome! Please:

1. Fork the repository
2. Create a feature branch
3. Add tests for new functionality
4. Run cargo test and cargo bench
5. Ensure cargo clippy passes
6. Submit a pull request

Comparison with Alternatives

*GMP performance only available in native code, not in zkVM environments.

Choose crypto-ratio if you need: RISC Zero zkVM support, pure Rust, no_std support, control over reduction, or compile-time size selection.

Choose num-rational if you need: Simple API with automatic reduction, dynamic sizing, don't need zkVM support.

Choose rug if you need: Maximum performance in native environments, don't mind GMP dependency.

Acknowledgments

Built on crypto-bigint by RustCrypto.

Designed for RISC Zero zkVM with support for hardware-accelerated big integer operations.

Inspired by num-rational with focus on stable performance and flexibility.

Resources

• RISC Zero Documentation
• crypto-bigint Accelerators in RISC Zero
• API Documentation

Questions? Open an issue or discussion on GitHub.