Calculus › Section J · Infinite series › Topic J.9
Representing functions as power series
One geometric series, plus the rules of calculus, manufactures power series for logarithms, arctangents and functions no antiderivative can touch. The trick is that inside its radius, a power series behaves exactly like a polynomial.
One known representation, many targets
You know one function that is a power series: the geometric series says
That single fact, combined with substitution, differentiation and integration, is a factory. The plan of this topic is to start from \(\frac{1}{1-x}\), bend it into shapes like \(\frac{1}{1+x^2}\), and then integrate or differentiate the series term by term to reach functions such as \(\ln(1+x)\) and \(\arctan x\), whose series would be miserable to build any other way. Every result must travel with its interval of validity: a series represents a function only where it converges.
Substitution: bending the master series
The geometric formula holds for any expression in place of \(x\), as long as its absolute value is under 1: the ratio \(r\) can be a whole expression. Replace \(x\) by \(-x\) and get
Replace \(x\) by \(-x^2\) and get \(\frac{1}{1 + x^2} = 1 - x^2 + x^4 - x^6 + \cdots\), the seed for the arctangent below. Each substitution changes the convergence condition the same way: the series for \(\frac{1}{1+x^2}\) needs \(|x^2| < 1\), that is \(|x| < 1\). The pattern to remember: a denominator \(1 - (\text{something})\) becomes a geometric series in that something, valid where the something is less than 1 in size.
Term-by-term calculus
Here is the licence that makes the factory run. Inside its radius of convergence, a power series can be differentiated and integrated term by term:
and both results have the same radius as the original. Differentiate or integrate every term, and the radius does not budge; only the endpoints can change hands. This is exactly how finite polynomials behave, and inside the radius the series genuinely acts like one. The endpoint caveat matters: integration tends to gain endpoints (extra convergence), differentiation tends to lose them, so after any operation the endpoints deserve a fresh check.
Worked example: the series for ln(1 + x)
Integrate the series for \(\frac{1}{1+x}\) term by term from 0 to \(x\):
which is \(\sum_{n=1}^{\infty} \frac{(-1)^{n+1}}{n} x^n\). The radius is still 1, but now the endpoints: at \(x = 1\) the series becomes the alternating harmonic series, which converges to \(\ln 2\), exactly matching the famous sum from the alternating series topic. At \(x = -1\) it becomes \(\sum \frac{-1}{n}\), the negative harmonic series, which diverges. Integration gained an endpoint: the interval is \((-1, 1]\). A series this clean for a logarithm is a genuine prize.
Worked example: the series for arctan x
Start from \(\frac{1}{1+x^2} = 1 - x^2 + x^4 - x^6 + \cdots\), valid for \(|x| < 1\), and integrate from 0 to \(x\):
in sigma form \(\sum_{n=0}^{\infty} \frac{(-1)^n x^{2n+1}}{2n+1}\). Check the endpoints once more. At \(x = 1\) the series is \(1 - \frac{1}{3} + \frac{1}{5} - \cdots\), an alternating series whose term sizes shrink to 0, so it converges; at \(x = -1\) the same is true with the signs flipped. So the interval is \([-1, 1]\): integration gained both endpoints. This series is how \(\pi\) was computed for centuries of record attempts, since \(\arctan 1 = \frac{\pi}{4}\).
Worked example: differentiating instead
Differentiate the master series and you meet its own derivative:
a series whose coefficients are \(1, 2, 3, 4, \dots\), growing linearly. The differentiation traded the constant coefficients for counting numbers, and the endpoint structure shifted the other way: the differentiated series diverges at both \(x = 1\) (terms grow) and \(x = -1\) (terms alternate with sizes growing), so the interval is exactly \((-1, 1)\). Note the symmetry with the integration examples: here a function with a vertical asymptote at \(x = 1\) sits behind the series, and no endpoint rescue is available.
Applied: computing a logarithm by hand
Series representations are calculators. Estimate \(\ln(1.1)\) using the series \(\ln(1+x) = x - \frac{x^2}{2} + \frac{x^3}{3} - \cdots\) with \(x = 0.1\):
The series alternates with shrinking term sizes, so the alternating series error bound applies: the error is at most the first omitted term, \(\frac{0.1^4}{4} = 0.000025\). The true value is \(\ln(1.1) = 0.095310\ldots\), and indeed the gap is about \(0.000023\), inside the promise. Three terms, four decimal places of accuracy: near the centre of its interval, a power series is a precision instrument.
Common mistakes to avoid
The error with the widest blast radius is quoting a series without its interval: \(\frac{1}{1+x}\) equals \(1 - x + x^2 - \cdots\) only for \(|x| < 1\), and using it elsewhere produces confident nonsense. Second, when substituting, track powers carefully: replacing \(x\) by \(-x^2\) produces only even powers with alternating signs, and \(\frac{1}{1 - x^2}\) produces \(1 + x^2 + x^4 + \cdots\), never odd powers. Third, do integrate term by term but never forget the effect on endpoints: the \(\ln(1+x)\) and \(\arctan x\) series gained endpoints that the derivative versions do not have. Fourth, an indefinite term-by-term integration needs its constant \(C\); starting the integral at 0, as in the worked examples, pins the constant down automatically.
Key takeaways
The geometric series \(\frac{1}{1-x} = \sum x^n\) for \(|x| < 1\) generates representations: substitute to handle denominators \(1 \pm (\text{expression})\), then differentiate or integrate term by term, since a power series behaves like a polynomial inside its radius and keeps that radius under both operations. The landmark results are \(\ln(1+x) = \sum \frac{(-1)^{n+1} x^n}{n}\) on \((-1, 1]\) and \(\arctan x = \sum \frac{(-1)^n x^{2n+1}}{2n+1}\) on \([-1, 1]\), both born by integration, and \(\frac{1}{(1-x)^2} = \sum n x^{n-1}\) by differentiation. Every representation carries its interval of validity, endpoints rechecked after each operation, and inside that interval the series is the function.
Flashcards
Flashcard
1/(1 - x) = Σ xⁿ for |x| < 1
Test your understanding
Question 1 of 5
What is the power series for 1/(1 + x)?
Question 2 of 5
What are the first three nonzero terms of the series for ln(1 + x)?
Question 3 of 5
What is the series for arctan x?
Question 4 of 5
What series results from differentiating 1/(1 - x) = Σ xⁿ?
Question 5 of 5
Using its power series, ln(1.1) ≈ 0.1 - 0.005 + 0.000333. Why is the error at most 0.000025?
Standards: 10 (BC)
