A Proof of the Continued Fraction Identity $-\pi/4 = {\rm K}_{n=1}^{\infty}\bigl((n-1)^2\,/\,{-(2n-1)}\bigr)$

This paper provides a self-contained analytic proof of the continued fraction identity $-\pi/4 = {\rm K}_{n=1}^{\infty}\bigl((n-1)^2\,/\,{-(2n-1)}\bigr)$ by transforming the classical Gauss continued fraction for $\arctan(z)$ evaluated at $z=-1$ and demonstrates its super-exponential convergence advantage over the Gregory–Leibniz series.

The Gist

Imagine you are trying to measure the exact length of a circle's curve (specifically, a quarter of it, which relates to the number $\pi$). For centuries, mathematicians have had different "recipes" to calculate this number.

This paper is essentially a detective story where the author, Chao Wang, solves a mystery about a very specific, strange-looking recipe for $-\pi/4$ that was discovered by a computer algorithm (the "Ramanujan Machine").

Here is the breakdown of the paper using simple analogies:

1. The Mystery: A Weird Recipe

Mathematicians love writing numbers as Continued Fractions. Think of a continued fraction like a Russian nesting doll or a set of stairs going down forever. You start with a number, add a fraction, then add another fraction to the bottom of that, and so on.

There is a famous, well-behaved recipe for $\pi/4$ discovered by Gauss (a mathematical giant). It looks like this:
$$\frac{1}{1 + \frac{1^2}{3 + \frac{2^2}{5 + \frac{3^2}{7 + \dots}}}}$$
This recipe works perfectly. Everyone knows it.

But then, a computer project called the Ramanujan Machine found a different recipe that looked very similar but had a twist: every number in the bottom row (the denominators) had a negative sign.
$$\frac{1}{-1 + \frac{1^2}{-3 + \frac{2^2}{-5 + \frac{3^2}{-7 + \dots}}}}$$
This recipe claimed to equal $-\pi/4$.

The Problem: Computers are great at guessing, but they aren't great at proving. Just because a computer says "this equals $-\pi/4$" doesn't mean it's mathematically true. The math community needed a human to prove it.

2. The Solution: The "Magic Mirror" Transformation

Chao Wang's paper provides the proof. He doesn't invent a new, complicated math theory. Instead, he uses a clever trick called an Equivalence Transformation.

Think of the two recipes (the famous one and the weird negative one) as two different languages describing the same story.

• Recipe A (Gauss): Uses positive numbers.
• Recipe B (The Mystery): Uses negative numbers.

Wang shows that you can turn Recipe A into Recipe B using a simple "magic mirror" (a mathematical operation). He takes the famous recipe and multiplies every single denominator by $-1$.

• If you flip the sign of the bottom numbers, you must also flip the sign of the top numbers to keep the value the same.
• He shows that if you do this flip, the famous recipe for $\pi/4$ transforms perfectly into the weird recipe for $-\pi/4$.

The Analogy: Imagine you have a photo of a house (Recipe A). If you take a photo of the same house but view it in a mirror (Recipe B), the house looks different (left becomes right), but it is still the exact same house. Wang proved that the "mirror image" of the famous $\pi$ recipe is exactly the strange negative recipe found by the computer.

3. Why Does It Work? (The Safety Net)

You might ask, "Is it safe to flip all those signs? Does the infinite staircase still hold together?"

Wang explains that the original recipe (Gauss's) is mathematically "solid." It converges, meaning if you keep adding more layers to the stairs, you get closer and closer to the true answer without falling off. Because the two recipes are just mirror images of each other, if the original one is solid, the new one is solid too.

4. The Speed Test: Why This Matters

The paper ends with a race.

• The Old Way (Gregory-Leibniz Series): This is the most famous way to calculate $\pi$. It's like walking up a hill one tiny step at a time. It works, but it's incredibly slow. To get 10 digits of accuracy, you need millions of steps.
• The New Way (This Continued Fraction): This is like taking a helicopter. It zooms to the answer.

The paper shows a table proving that this new recipe reaches high precision in just a few steps. It's not just a curiosity; it's a super-fast calculator for $\pi$.

Summary

• The Goal: Prove that a weird, computer-discovered formula for $-\pi/4$ is actually true.
• The Method: Show that this formula is just a "sign-flipped" version of a famous, already-proven formula by Gauss.
• The Result: The formula is proven true, and it turns out to be an incredibly fast way to calculate $\pi$ compared to older methods.

In short, the author took a computer's "wild guess," showed it was just a familiar friend wearing a disguise, and proved that the disguise didn't change the identity of the number.

Technical Summary

Here is a detailed technical summary of the paper "A Proof of the Continued Fraction Identity $-\pi/4 = \mathbf{K}_{n=1}^{\infty} \frac{(n-1)^2}{-(2n-1)}$" by Chao Wang.

1. Problem Statement

The paper addresses the mathematical verification of a specific continued fraction identity conjectured by the Ramanujan Machine project. The identity in question is:
$$-\frac{\pi}{4} = \cfrac{1}{-1 + \cfrac{1^2}{-3 + \cfrac{2^2}{-5 + \cfrac{3^2}{-7 + \ddots}}}}$$
This can be formally expressed as a continued fraction $K(a_n/b_n)$ where:

• Numerators ($a_n$): $a_1 = 1$, and $a_n = (n-1)^2$ for $n \geq 2$.
• Denominators ($b_n$): $b_n = -(2n-1)$ for all $n \geq 1$.

While this identity was discovered algorithmically, a rigorous analytic proof was required to confirm its validity and convergence. The paper distinguishes this identity from other similar Ramanujan Machine discoveries (e.g., those with $b_n = -(3n-2)$), noting that despite converging to the same value, they are structurally distinct.

2. Methodology

The author employs a two-step analytic approach relying on classical hypergeometric function theory and the theory of continued fraction equivalence transformations.

Step 1: Connection to the Gauss Continued Fraction

The proof begins by recalling the classical Gauss continued fraction for the arctangent function, $\arctan(z)$. For $|z| \leq 1$ ($z \neq \pm i$), the expansion is:
$$\arctan(z) = \cfrac{z}{1 + \cfrac{1^2 z^2}{3 + \cfrac{2^2 z^2}{5 + \cfrac{3^2 z^2}{7 + \ddots}}}}$$
The author evaluates this at $z = -1$.

• Convergence Justification: The value is derived from the hypergeometric function $_2F_1(\frac{1}{2}, 1; \frac{3}{2}; -1)$. Using the Euler integral representation and the Gauss–Kummer theorem, the author establishes that the series converges absolutely at the boundary point $z=-1$ because $\text{Re}(c-a-b) = 0 > -1$.
• Resulting Value: Substituting $z=-1$ yields:
  $$\arctan(-1) = -\frac{\pi}{4} = \cfrac{-1}{1 + \cfrac{1^2}{3 + \cfrac{2^2}{5 + \cfrac{3^2}{7 + \ddots}}}}$$

Step 2: Equivalence Transformation

The paper utilizes the concept of equivalence transformations for continued fractions. Two continued fractions are equivalent if they can be transformed into one another via a sequence of non-zero scalars $\{r_n\}$ without changing their value.

• The author defines a constant transformation sequence: $r_n = -1$ for all $n \geq 1$.
• Applying the transformation rules:
  • $\tilde{a}_1 = r_1 a_1 = (-1)(-1) = 1$
  • $\tilde{b}_n = r_n b_n = (-1)(2n-1) = -(2n-1)$
  • $\tilde{a}_n = r_n r_{n-1} a_n = (-1)(-1)(n-1)^2 = (n-1)^2$
• This transformation maps the Gauss form (with positive denominators) directly to the target identity (with negative denominators), proving they share the same value.

3. Key Contributions

• Rigorous Proof: Provides the first self-contained analytic proof of the specific continued fraction identity for $-\pi/4$ found in the Ramanujan Machine literature.
• Structural Insight: Demonstrates that the complex-looking identity is merely a sign-flipped version of the classical Gauss continued fraction for $\arctan(-1)$. The "non-canonical" appearance (negative denominators) is shown to be a simple equivalence transformation rather than a fundamentally new mathematical object.
• Convergence Analysis: Confirms absolute convergence using the properties of the underlying hypergeometric series and the limit-periodic nature of the continued fraction coefficients.

4. Results

• Theorem 3.3: The paper formally proves that the continued fraction converges and its value is exactly $-\pi/4$.
• Numerical Comparison: The author compares the convergence rate of this continued fraction against the Gregory–Leibniz series (the standard alternating harmonic series for $\pi/4$).
  • Gregory–Leibniz: Converges linearly ($O(n^{-1})$). At $n=20$, the error is $\approx 10^{-2}$.
  • Target Continued Fraction: Exhibits super-exponential acceleration. At $n=20$, the error drops to the double-precision floor ($\approx 10^{-16}$).
  • The ratio of errors ($S_n / F_n$) grows exponentially, reaching a factor of $2.14 \times 10^{13}$at$n=20$.

5. Significance

• Algorithmic Verification: The work validates the output of algorithmic discovery tools (like the Ramanujan Machine) using established mathematical theory, bridging the gap between computational conjecture and analytic proof.
• Efficiency: The results highlight the computational superiority of Gauss-type continued fractions over standard power series for approximating transcendental numbers like $\pi$. The super-exponential convergence makes this identity highly efficient for high-precision calculations.
• Simplicity: The proof relies on elementary transformations and classical theorems (Gauss, Kummer, Euler), demonstrating that deep mathematical identities often have simple structural explanations.

In conclusion, Chao Wang successfully demystifies a complex algorithmic discovery by linking it to the classical theory of hypergeometric functions, proving that the identity is a direct consequence of the Gauss continued fraction for $\arctan(-1)$ under a simple sign transformation.