To approximate f ( x ) = ( 1 − x 2 ) 1 2 f(x) = \left( 1 - x^{2} \right)^{\frac{1}{2}} f ( x ) = ( 1 − x 2 ) 2 1 near x = 1 x=1 x = 1 , a standard Taylor series expansion centered at x = 1 x=1 x = 1 fails because f ′ ( x ) f^\prime(x) f ′ ( x ) involves a term of ( 1 − x 2 ) − 1 2 \left( 1 - x^{2} \right)^{-\frac{1}{2}} ( 1 − x 2 ) − 2 1 , which is undefined at the boundary.
Instead, we utilize a Puiseux series, which allows for expansions involving fractional powers.
Puiseux series
We begin by substituting x = 1 − t x = 1 - t x = 1 − t into the function:
f ( 1 − t ) = 1 − ( 1 − t ) 2 = 2 t − t 2 = 2 t ( 1 − t 2 ) 1 2 . \begin{align}\textstyle
f(1-t) = \sqrt{1 - (1 - t)^{2}} = \sqrt{2t - t^{2}} = \sqrt{2t} \left( 1 - \frac{t}{2} \right)^{\frac{1}{2}}.
\end{align} f ( 1 − t ) = 1 − ( 1 − t ) 2 = 2 t − t 2 = 2 t ( 1 − 2 t ) 2 1 .
Using the generalized binomial theorem, ( 1 + u ) α = ∑ n = 0 ∞ ( α n ) u n (1 + u)^{\alpha} = \sum_{n=0}^{\infty} \binom{\alpha}{n} u^{n} ( 1 + u ) α = ∑ n = 0 ∞ ( n α ) u n for ∣ u ∣ < 1 \lvert u \rvert < 1 ∣ u ∣ < 1 , we expand it for ∣ t ∣ < 2 \lvert t \rvert < 2 ∣ t ∣ < 2 :
f ( 1 − t ) = 2 t ( 1 − t 4 − t 2 32 − t 3 128 − ⋯ ) = 2 t 1 2 − 2 4 t 3 2 − 2 32 t 5 2 − 2 128 t 7 2 − ⋯ . \begin{align}\textstyle
f(1 - t)
&=\textstyle \sqrt{2t} \left( 1 - \frac{t}{4} - \frac{t^{2}}{32} - \frac{t^{3}}{128} - \cdots \right) \\
&=\textstyle \sqrt{2} t^{\frac{1}{2}} - \frac{\sqrt{2}}{4}t^{\frac{3}{2}} - \frac{\sqrt{2}}{32}t^{\frac{5}{2}} - \frac{\sqrt{2}}{128}t^{\frac{7}{2}} - \cdots.
\end{align} f ( 1 − t ) = 2 t ( 1 − 4 t − 32 t 2 − 128 t 3 − ⋯ ) = 2 t 2 1 − 4 2 t 2 3 − 32 2 t 2 5 − 128 2 t 2 7 − ⋯ .
The behavior of the function near the singularity depends on the direction of approach:
For x < 1 x < 1 x < 1 , the positive t = 1 − x t = 1 - x t = 1 − x yields a strictly real-valued series:
f ( x ) = 2 ( 1 − x ) 1 2 − 2 4 ( 1 − x ) 3 2 − 2 32 ( 1 − x ) 5 2 − 2 128 ( 1 − x ) 7 2 − ⋯ . \begin{align}\textstyle
f(x) = \sqrt{2} (1-x)^{\frac{1}{2}} - \frac{\sqrt{2}}{4}(1-x)^{\frac{3}{2}} - \frac{\sqrt{2}}{32}(1-x)^{\frac{5}{2}} - \frac{\sqrt{2}}{128}(1-x)^{\frac{7}{2}} - \cdots.
\end{align} f ( x ) = 2 ( 1 − x ) 2 1 − 4 2 ( 1 − x ) 2 3 − 32 2 ( 1 − x ) 2 5 − 128 2 ( 1 − x ) 2 7 − ⋯ .
For x > 1 x > 1 x > 1 , letting t = ( x − 1 ) e i π t = (x - 1)e^{i\pi} t = ( x − 1 ) e iπ , where t 1 2 = i ( x − 1 ) 1 2 t^{\frac{1}{2}} = i (x-1)^{\frac{1}{2}} t 2 1 = i ( x − 1 ) 2 1 , yields an imaginary-valued series:
f ( x ) = i [ 2 ( x − 1 ) 1 2 + 2 4 ( x − 1 ) 3 2 − 2 32 ( x − 1 ) 5 2 + 2 128 ( x − 1 ) 7 2 − ⋯ ] . \begin{align}\textstyle
f(x) = i \left[
\sqrt{2} (x-1)^{\frac{1}{2}} + \frac{\sqrt{2}}{4}(x-1)^{\frac{3}{2}} - \frac{\sqrt{2}}{32}(x-1)^{\frac{5}{2}} + \frac{\sqrt{2}}{128}(x-1)^{\frac{7}{2}} - \cdots
\right].
\end{align} f ( x ) = i [ 2 ( x − 1 ) 2 1 + 4 2 ( x − 1 ) 2 3 − 32 2 ( x − 1 ) 2 5 + 128 2 ( x − 1 ) 2 7 − ⋯ ] .