BigFloatLibrary 4.0.2

BigFloat Library for C#

BigFloat is a C# struct/class library for arbitrary-precision floating-point numbers, backed by System.Numerics.BigInteger. It combines a flexible mantissa, a wide exponent range, and a 32-bit guard-bit region to deliver high-precision arithmetic for very large or very small values.

Ryan Scott White / MIT License
Updated: December 1st, 2025
Current BigFloat version: 4.0.0

• GitHub repo: https://github.com/SunsetQuest/BigFloat
• Documentation site: https://bigfloat.org
• NuGet package (BigFloatLibrary): https://www.nuget.org/packages/BigFloatLibrary
• Original CodeProject article: “A BigFloat Library in C#”

---

Contents

• What is BigFloat?
• Key features
• Use cases
• Requirements
• Installation
• Quick start
• Core concepts: representation & guard bits
• Precision, accuracy, and guard-bit notation
• Accuracy & precision APIs (2025+)
• Math functions & constants
• Working with very large / very small numbers
• Project layout
• Background & related work
• License

---

What is BigFloat?

BigFloat is an arbitrary-precision floating-point type implemented as a readonly partial struct in the BigFloatLibrary namespace. Internally, a BigFloat value is stored as:

• A BigInteger mantissa (_mantissa)
• An integer binary scale (Scale)
• A cached bit count (_size), which includes guard bits

Mathematically, values are represented in the form:

value = mantissa × 2^(Scale - GuardBits)

Most users can treat BigFloat as “a very large double with configurable precision,” but it exposes more information and control than standard IEEE types:

• You can inspect the effective precision (Size)
• You can tune how many effective working bits you want for integers and conversions
• You can grow or shrink the working accuracy explicitly without silently losing information

---

Key features

Arbitrary precision

• Mantissa backed by System.Numerics.BigInteger with a design target of up to billions of bits of precision (limited by memory and patience).

Guard bits for stability

• A fixed 32 hidden guard bits live in the least-significant portion of the mantissa. These bits are not counted as “in-precision,” but they carry extra context so chained operations round more accurately.

Binary-first design

• BigFloat is fundamentally base-2; decimal formatting/parsing is layered on top.
• Decimal input like 0.1 is converted to repeating binary internally; the API makes these trade-offs explicit rather than hiding them.

Accuracy & precision controls (2025 refresh)

Modern APIs are available for explicitly managing the accuracy “budget”:

• AdjustAccuracy, SetAccuracy
• SetPrecisionWithRound, AdjustPrecision (and compatibility helpers around ExtendPrecision / ReducePrecision)
• Rounding helpers like ToNearestInt and truncation helpers that preserve accuracy metadata

Rich operator and helper set

• Full arithmetic and comparison operators (+ - * / %, ==, <, >, etc.)
• Constructors from int, long, ulong, double, decimal, BigInteger, and string
• Conversion helpers such as FitsInADouble, explicit casts to primitive numeric types, and formatting with standard and custom format strings

Math functions & algorithms

• Extended math library (exponential, log, roots, powers, trig, etc.) implemented in BigFloatMath and related partials.
• High-performance algorithms such as Newton-Plus-style square root, Karatsuba multiplication, and Burnikel–Ziegler-style division for very large operands.

Constants library

• Constants.cs exposes fundamental constants such as π and e with configurable accuracy.
• Pre-computed decimal value files in the values folder extend constants up to roughly 1,000,000 digits.

---

Use cases

BigFloat is designed for scenarios where IEEE double/decimal are not accurate enough:

• Scientific and engineering simulations that accumulate error over long compute chains
• Financial or actuarial models where rounding behavior must be controlled and auditable
• Cryptographic or number-theoretic experimentation (huge exponents, large primes)
• Teaching / research into floating-point and arbitrary-precision arithmetic

---

Requirements

BigFloat follows the modern .NET toolchain.

Runtime / SDK

• .NET 8.0 SDK or later (the library targets net8.0, net9.0 or net10.0).

Language

• C# 12 or later (the version shipped with the .NET 8 SDK at the time of writing).

Dependencies

• System.Numerics.BigInteger (ships with .NET) – no external library dependencies.

Check your SDKs with:

dotnet --list-sdks

and confirm a 8.x, 9.x, or 10.x entry is available.

---

Installation

Option 1 – NuGet (recommended)

Add the library via NuGet:

dotnet add package BigFloatLibrary

or via PackageReference:

<ItemGroup>
  <PackageReference Include="BigFloatLibrary" Version="*" />
</ItemGroup>

Refer to NuGet for the current package version (the README describes the 4.0.0 library APIs; NuGet version numbers may differ). (NuGet)

Option 2 – From source

1. Clone the repository:

   git clone https://github.com/SunsetQuest/BigFloat.git
   

2. Add the BigFloatLibrary project to your solution.

3. Reference it from your main project.

4. Optionally include Constants.cs and the values folder for extended constant precision. (GitHub)

---

Quick start

A minimal console application using BigFloat:

using System;
using BigFloatLibrary;

class Program
{
    static void Main()
    {
        // Standard double loses precision
        double d = 0.1 + 0.2;
        Console.WriteLine($"Double: 0.1 + 0.2 = {d}");
        Console.WriteLine($"Equals 0.3? {d == 0.3}");

        // BigFloat maintains a stable representation
        BigFloat x = new("0.1");
        BigFloat y = new("0.2");
        BigFloat sum = x + y;

        Console.WriteLine();
        Console.WriteLine($"BigFloat: {x} + {y} = {sum}");
        Console.WriteLine($"Equals 0.3? {sum == new BigFloat(\"0.3\")}");

        // Use a constant and perform higher-precision math
        BigFloat radius = new("10.123456789");
        BigFloat pi = Constants.Fundamental.Pi;
        BigFloat area = pi * radius * radius;

        Console.WriteLine();
        Console.WriteLine($"Area with radius {radius} ≈ {area}");
        Console.WriteLine($"Nearest integer (guard-bit aware): {BigFloat.ToNearestInt(area)}");
    }
}

The examples in the documentation site’s Getting Started and Examples pages are kept in sync with the current APIs and are suitable for copy/paste into your projects. (bigfloat.org)

---

Core concepts: representation & guard bits

DataBits, Scale, and Size

Internally, each BigFloat has three important fields/properties: (GitHub)

• DataBits (_mantissa : BigInteger) Holds the signed binary representation, including guard bits.

• Scale (int) A binary offset that effectively slides the radix point left or right. Positive scale moves the point left (larger values), negative scale moves it right (more fractional bits).

• Size (int) The number of in-precision bits, excluding guard bits (SizeWithGuardBits includes them). For zero, size is zero.

The public API exposes these either directly or via helper properties/methods (Size, Scale, BinaryExponent, etc.).

GuardBits

BigFloat reserves:

public const int GuardBits = 32;

as a fixed number of least-significant bits dedicated to guard precision. (GitHub)

These bits:

• Are not counted as in-precision, but they retain “extra evidence” from intermediate calculations.
• Make long chains of additions, multiplications, and transcendental operations behave more predictably than a pure fixed-precision representation.

Conceptually:

precise_bits | guard_bits

For example (conceptually):

101.01100|110011001100...   ≈ 5.4
100.01001|100110011001...   ≈ 4.3
---------------------------------
1001.1011|001100110011...   ≈ 9.7

If we discarded the right side of the |, the final rounding decision would be based on much less information. With guard bits preserved, BigFloat can round closer to the “true” value in the final representation. (GitHub)

---

Precision, accuracy, and guard-bit notation

“X” digits and | separator

The library uses several conventions to show what is in-precision versus out-of-precision: (GitHub)

• Decimal outputs may show trailing X characters (e.g., 232XXXXXXX) where digits are beyond the guaranteed precision.
• Binary diagnostics can include a | between in-precision bits and guard bits.
• Some parsing/formatting helpers accept or emit the | separator to round-trip accuracy hints.

For example:

// "123.456|789" - '789' bits are treated as guard / accuracy hints
BigFloat value = BigFloat.Parse("123.456|789");

Decimal vs. binary precision

Most decimal fractions cannot be represented exactly in binary:

• 5.4 → 101.0110011001100… (repeating)
• 4.3 → 100.0100110011001… (repeating)
• 0.25 → 0.01 (exact) (GitHub)

BigFloat is explicit about this:

• Conversions are always done in binary.
• Guard bits exist specifically to preserve more of the repeating tail when it is helpful.
• Precision and accuracy APIs let you control how many effective bits you want to keep.

---

Accuracy & precision APIs (2025+)

The 2025 releases focus on making precision control explicit and less error-prone. New APIs are described in the docs site’s Accuracy Controls section. (bigfloat.org)

Growing or shrinking accuracy

BigFloat x = new("123.456");

// Add 64 working bits without changing the represented value
BigFloat wider = BigFloat.AdjustAccuracy(x, +64);

// Remove 32 bits and round before dropping them
BigFloat trimmed = BigFloat.AdjustPrecision(x, -32);

// Set a specific accuracy target (including guard bits)
BigFloat withBudget = BigFloat.SetAccuracy(x, 256);
Console.WriteLine(withBudget.Size); // around 256 - GuardBits

Use these helpers instead of manually manipulating scale or trying to emulate precision changes yourself.

Rounding and truncation

BigFloat v = new("9876.54321");

// Round to nearest integer value without burning all guard bits
int nearest = BigFloat.ToNearestInt(v);

// Truncate to an integer but keep accuracy metadata in the BigFloat
BigFloat integer = v.TruncateToIntegerKeepingAccuracy();

// Reduce size with rounding at the cut point
BigFloat compact = BigFloat.SetPrecisionWithRound(v, 80);

The older ExtendPrecision / ReducePrecision helpers are still present for backward compatibility, but the newer APIs are preferred because they encode intent (accuracy vs. bare precision) more clearly. (bigfloat.org)

---

Math functions & constants

Math functions

The core struct is extended via partial classes:

• BigFloatMath.cs – higher-order math (log, exp, powers, trig, roots, etc.) (GitHub)
• BigFloatRoundShiftTruncate.cs – shifting, rounding, truncation, splitting
• BigFloatParsing.cs – string/Span parsing, binary/decimal utilities
• BigFloatStringsAndSpans.cs – string/Span formatting

These modules use algorithms such as:

• Newton-Plus variants for square root on large inputs
• Karatsuba multiplication and Burnikel–Ziegler-style division once operands cross certain thresholds (GitHub)

Constants

Constants.cs exposes a hierarchy of constants collections, e.g.:

using BigFloatLibrary;

var constants = new BigFloat.Constants(
    requestedAccuracyInBits: 1000,
    onInsufficientBitsThenSetToZero: true,
    cutOnTrailingZero: true);

BigFloat pi = constants.Pi;
BigFloat e  = constants.E;

• Constants.cs provides constants with thousands of decimal digits.
• Text files in the values folder can be included to extend certain constants close to 1,000,000 digits. (GitHub)

---

Working with very large / very small numbers

Basic arithmetic

BigFloat a = new("123456789.012345678901234");
BigFloat b = new(1234.56789012345678); // from double

BigFloat sum       = a + b;
BigFloat diff      = a - b;
BigFloat product   = a * b;
BigFloat quotient  = a / b;

Console.WriteLine($"Sum: {sum}");
Console.WriteLine($"Difference: {diff}");
Console.WriteLine($"Product: {product}");
Console.WriteLine($"Quotient: {quotient}");

The examples in the current README show how results may end with X characters in decimal form when you have exceeded the guaranteed precision range for that value. (GitHub)

Comparisons and base-10 vs. base-2 intuition

Because BigFloat is base-2, decimal rounding intuition can be misleading. For example:

BigFloat num1 = new("12345.6789");
BigFloat num2 = new("12345.67896");

bool equal   = num1 == num2;
bool bigger  = num1 > num2;

Even though 12345.6789 and 12345.67896 are different in base-10, their binary encodings may compare equal at the available precision. The README includes extended examples of this behavior and how guard bits mitigate surprises by preserving more of the repeating tail. (GitHub)

Extreme scales

BigFloat is designed to represent values with very large positive or negative exponents:

BigFloat large      = new("1234e+7");
BigFloat veryLarge  = new("1e+300");
BigFloat verySmall  = new("1e-300");

• Very large values naturally move towards exponential notation (1 * 10^300 style strings) for readability.
• Very small values may output as long rows of zeros followed by a small non-zero tail; the guard bits and accuracy metadata still track how meaningful that tail is. (GitHub)

---

Project layout

The repository is organized roughly as follows: (GitHub)

• BigFloatLibrary/

  • Core BigFloat struct (BigFloat.cs)

  • Partial classes:

    • BigFloatCompareTo.cs
    • BigFloatExtended.cs
    • BigFloatMath.cs
    • BigFloatParsing.cs
    • BigFloatRandom.cs
    • BigFloatRoundShiftTruncate.cs
    • BigFloatStringsAndSpans.cs

  • Constants.cs and supporting constant builder/catalog files

• BigFloat.Tests/

  • Unit tests covering arithmetic, conversions, parsing, rounding, and edge cases.
  • Use dotnet test at the solution level to exercise the test suite.

• BigFloatPlayground/

  • Small exploratory console projects for:

    • Benchmarking (Benchmarks.cs)
    • Showcases / sample scenarios
    • Experimental testing of new APIs or algorithms (GitHub)

• ChangeLog.md

  • High-level history of releases and notable changes (consult this file for detailed version history, including 4.0.0 notes).

• BigFloatSpecification.md

  • A deeper technical specification describing the format, invariants, and accuracy rules for BigFloat values.

For curated, user-oriented documentation (architecture notes, examples, and API overviews), refer to https://bigfloat.org. (bigfloat.org)

---

Background & related work

BigFloat grew out of earlier work on high-precision numerics and the Newton-Plus square-root algorithm for large BigInteger values. (GitHub)

• Original write-up: “A BigFloat Library in C#” on CodeProject.
• Companion project: NewtonPlus – Fast BigInteger and BigFloat Square Root, which explores performance-optimized sqrt for very large integers and shows how the algorithm is adapted for BigFloat. (GitHub)

BigFloat is designed as an educational tool as much as a production-quality numeric type: the source is heavily commented, the specification is public, and the tests and playground make it easier to experiment with tricky numerics.

---

License

This project is distributed under the MIT License. See the MIT License file in this repository for details. (GitHub)