- 1 ++ # MöbiusNumber

0.1 + 0.2 = 0.3. Exactly.

from mobius_number import M

>>> 0.1 + 0.2 == 0.3
False

>>> M("0.1") + M("0.2") == M("0.3")
True

A number type that carries its own correction. No rounding error. No epsilon comparisons. No workarounds.

Install

pip install mobius-number

The Problem

Every computer in the world gets this wrong:

>>> 0.1 + 0.2
0.30000000000000004

>>> 0.1 + 0.2 == 0.3
False

This has been true since 1985 when IEEE 754 was published. The reason: computers store numbers in base 2. The number 0.1 in binary is a repeating fraction — like 1/3 in decimal (0.333...). It gets rounded. Then 0.2 gets rounded. The rounding errors stack.

Every workaround — arbitrary precision, interval arithmetic, posit numbers — tries to fix the math layer while leaving the foundation untouched: the number is stored in the form the transistor speaks, and the human adapts.

What if the representation served the number instead?

The Idea

DNA has two strands. Every base carries its complement — A pairs with T, G pairs with C. There is no excess. The complement consumes what the original doesn't cover.

A Möbius strip has one surface. Traversing the full loop covers both "sides" and returns to the origin.

A MöbiusNumber stores two strands:

• The binary strand — float64, hardware-fast, carries rounding error
• The rational strand — exact Fraction, no loss, no repeating

They are not two separate representations. They are one object. The binary strand is the shadow. The rational strand is the substance. On collapse, the rational governs.

CURRENT (IEEE 754):
  Store 0.1 → 0.1000000000000000055511...
  Residue:     0.0000000000000000055511...  ← THROWN AWAY

MöbiusNumber:
  Binary strand:   0.1000000000000000055511...
  Rational strand: 1/10
  Anti (residue):  exact_value − binary_value
  Collapse:        the rational governs → 0.1 exactly

Arithmetic propagates both strands. When you add two MöbiusNumbers, the rationals add exactly. The floats add approximately. The error exists but is never consulted for truth. The anti-strand is always present, always correct, and annihilates the rounding error on collapse.

Proof

$ python -c "from mobius_number import M; print(M('0.1') + M('0.2') == M('0.3'))"
True

The Famous Failures — All Fixed

Test	IEEE 754	MöbiusNumber
0.1 + 0.2 == 0.3	False	True
(1/49) * 49 == 1	False	True
(1 + 1e-16) - 1	0.0 (total loss)	1e-16 (exact)
0.1 * 10 == 1.0	True¹	True
$10 / 3 * 3 == $10	True¹	True
0.01 * 100 == 1.0	True¹	True
(1/7) * 7 == 1	False	True
0.001 added 1000×	False	True

¹ IEEE 754 gets these by luck — the rounding errors happen to cancel. The MöbiusNumber gets them by construction.

Strand Anatomy

from mobius_number import M

n = M("0.1")
print(n.diagnose())
# {
#   'binary_strand': 0.1,
#   'rational_strand': '1/10',
#   'residue': '-1/180143985094819840',
#   'residue_float': -5.55e-18,
#   'collapsed': 0.1
# }

The residue is the anti-strand — the exact complement of the binary error. It exists. It is never discarded. On collapse, the Möbius strip closes.

Usage

from mobius_number import M

# Basic arithmetic
a = M("0.1")
b = M("0.2")
c = a + b          # M('3/10')
c.collapse()        # 0.3

# Financial
price = M("19.99")
tax = price * M("0.0825")
total = price + tax  # Exact

# Comparison — the rational governs
M("0.1") + M("0.2") == M("0.3")  # True — always

# Interop with plain numbers
M("0.5") + 1        # M('3/2')
3 * M("0.1")         # M('3/10')

# Inspect the strands
n = M("0.1")
n.approx             # 0.1 (the float — fast, lossy)
n.exact              # Fraction(1, 10) (the truth)
n.residue            # Fraction(-1, 180143985094819840)

Why Not Just Use Fraction?

You can. Python's fractions.Fraction gives exact rational arithmetic. But:

1. Speed — MöbiusNumber carries a float for fast approximate work. Use .approx in hot loops, .collapse() when you need truth.
2. Drop-in intent — M("0.1") reads like a number. Fraction("0.1") reads like a workaround.
3. The conceptual point — the number and its correction are one object. The anti-strand is not a separate operation. It is intrinsic. Like DNA. Like a Möbius strip.

How It Works

Every MöbiusNumber is internally:

(_approx: float, _exact: Fraction)

• Construction from string: M("0.1") → _exact = Fraction("0.1") = 1/10; _approx = float(1/10)
• Construction from float: M(0.1) → recovers the rational intent via limit_denominator
• Arithmetic: both strands propagate independently through +, -, *, /, **
• Comparison: always uses _exact — the rational strand governs all equality and ordering
• Collapse: float(_exact) — the Möbius traversal returns the exact value

No external dependencies. Pure Python. Works on 3.9+.

The Name

A Möbius strip is a surface with one side. If you trace a line along it, you cover both "sides" and return to the origin having traversed the whole thing. There is no front and back — only one continuous surface.

A MöbiusNumber is a number with one identity. The binary approximation and the exact rational are not two things — they are one object that, when fully traversed, resolves to the truth. The representation IS the correction.

Author

Jay Carpenter — SECS Research

See Also

• mobius-constant — Exact irrational constants (sqrt(2)**2 == 2, by construction)
• mobius-integer — Dual-strand integer: machine i64 + exact BigInt (Rust)
• mobius-units — Fundamental constants from the eigenvalue tower — one measurement derives them all

Same pattern. Same anatomy. Same fix. Different domain.

License

MIT