What Is a Padé Approximant?
A Padé approximant approximates a function not by a polynomial but by a ratio of polynomials:
R(x) = P(x)/Q(x), with deg P = m, deg Q = n, chosen so R matches the Taylor series of f through order xm+n.
It is built from the same Taylor coefficients as an ordinary polynomial approximation, but by allowing a denominator, it gains the ability to bend, level off, and even reproduce poles, features a polynomial fundamentally cannot capture. The payoff is dramatic: a Padé approximant frequently stays accurate far outside the radius of convergence of the original Taylor series, and it often achieves a given accuracy with far fewer coefficients. This calculator computes the Taylor coefficients of your function symbolically, solves the linear system for the [m/n] approximant, and compares its reach against the Taylor polynomial of the same total degree, both numerically and on a graph.
How to Use the Padé Approximant Calculator
Enter a function f(x) (analytic at 0) and the numerator and denominator degrees m and n. The steps list the Taylor coefficients through order m + n, solve the linear system for the denominator, back-substitute for the numerator, and display R(x) = P(x)/Q(x). A comparison table evaluates f, its Taylor polynomial of degree m + n, and the Padé approximant at several points, showing where each breaks down. The graph overlays all three. This extends the Taylor series calculator: Padé recycles the Taylor data into a more powerful rational form.
Worked Example
The [2/2] Padé approximant of ex at 0 is a classic. From the Taylor coefficients 1, 1, 1/2, 1/6, 1/24, solving the system yields
R(x) = (1 + x/2 + x²/12) / (1 − x/2 + x²/12)
This matches ex through the x⁴ term, exactly like the degree-4 Taylor polynomial, but its behavior away from 0 is far superior. At x = 3, e³ ≈ 20.09; the degree-4 Taylor polynomial gives 16.375 (off by 18%), while the [2/2] Padé gives 19.03 (off by 5%), using the same amount of information. The denominator lets R curve upward the way the exponential does, whereas a polynomial of that degree simply cannot keep up. The 1/(1+x) chip shows the extreme case: since the function is already a rational function, the [1/1] Padé reproduces it exactly, capturing the pole at x = −1 that no polynomial could ever represent.
Why Rational Beats Polynomial
The essential limitation of a Taylor polynomial is that polynomials must eventually run off to ±∞, and they cannot represent a pole (a place where the function blows up) or a horizontal asymptote (a place where it levels off). Yet these are exactly the features that make real functions interesting, and they are what limit a Taylor series' radius of convergence: the series of ln(1 + x) diverges beyond |x| = 1 because of the singularity at x = −1, and no number of polynomial terms can cross that barrier. A Padé approximant, by contrast, has a denominator whose roots can sit right at the function's poles, allowing the rational function to mimic the singularity and remain accurate on both sides of it. This is analytic continuation in disguise: Padé approximants can recover a function's values in regions where its Taylor series is useless. The construction solves a small linear system, the denominator coefficients are chosen to kill the Taylor series' terms beyond order m + n, and the numerator then matches the remaining terms. Because the fit uses a ratio, it can encode information about singularities implicitly, which is why Padé approximants are the tool of choice when a function's series converges slowly or diverges, and why they often outperform polynomials of far higher degree.
Common Mistakes to Avoid
- Expecting improvement near a spurious pole. Sometimes the denominator develops a root where the true function has none (a Froissart doublet), creating a fake singularity. Check the approximant's poles against the function's known behavior.
- Choosing degenerate degrees. For even or odd functions, some [m/n] systems are singular (the linear solve fails). Adjusting m or n by one usually resolves it; the tool flags such cases.
- Applying it where the function is not analytic at 0. Padé here expands at the origin, so f and its derivatives must exist at 0. For a function singular at 0, shift the expansion point first.
- Assuming more terms always help. Higher-order Padé approximants usually improve accuracy, but not monotonically everywhere, and numerical conditioning of the linear system worsens. Modest orders often give the best practical balance.
- Forgetting it still only matches locally. The approximant matches the Taylor series to order m + n at x = 0; far from 0 and far from any captured singularity, accuracy can still degrade. It extends reach, it does not make approximation global.
Real-World Applications
Padé approximants are a standard tool wherever series must be pushed beyond their comfort zone. In physics, perturbation series in quantum mechanics and quantum field theory are frequently divergent or slowly convergent, and Padé resummation extracts meaningful physical predictions from them, famously improving estimates of critical exponents in statistical mechanics and energy levels in quantum systems where naive term-by-term summation fails. Control theory uses Padé approximants to model time delays: the transfer function e−sτ of a pure delay is approximated by a rational function so that delay systems fit into the rational-transfer-function framework of classical control design.
Frequently Asked Questions
What is a Padé approximant?
A rational function P(x)/Q(x), with numerator degree m and denominator degree n, whose Taylor expansion matches that of a target function f through order xm+n. It uses the same Taylor data as a polynomial approximation but arranges it as a ratio for greater reach.
Why is a Padé approximant better than a Taylor polynomial?
The denominator lets it represent poles, asymptotes, and rapid growth that polynomials cannot. This often keeps it accurate far beyond the Taylor series' radius of convergence and matches a given accuracy with fewer coefficients, a form of analytic continuation from the same local data.
How is the approximant computed?
From the Taylor coefficients of f. The denominator coefficients are found by solving a small linear system that forces the product Q(x)f(x) − P(x) to vanish through order xm+n; then the numerator coefficients are read off. It is a linear algebra problem built on the Taylor series.
Can a Padé approximant be exact?
Yes, when the function is itself rational. The [1/1] Padé of 1/(1+x) reproduces it exactly, capturing the pole at x = −1. More generally, if f = P/Q with degrees within m and n, the [m/n] approximant recovers f identically, something no polynomial can do for a function with a pole.
What is Padé resummation?
Using Padé approximants to extract finite, meaningful values from a divergent or slowly convergent series. In physics, perturbation series that diverge term by term can yield accurate predictions when reorganized as Padé approximants, which encode the analytic structure the raw series obscures.
Why do some [m/n] choices fail?
The linear system for the denominator can be singular, especially for functions with symmetry (even or odd) that make certain coefficients vanish. This is a degenerate case; changing m or n by one usually produces a solvable, well-behaved approximant.
What is a spurious pole or Froissart doublet?
A pole of the approximant that does not correspond to any singularity of the true function, often paired with a nearby zero that nearly cancels it. These artifacts arise from noise or over-fitting and are a known pitfall; comparing the approximant's poles to the function's expected behavior detects them.
How does Padé approximation model time delays?
A pure time delay has transfer function e−sτ, which is not rational. Its Padé approximant is a ratio of polynomials in s that approximates the delay, letting delay systems be analyzed and controlled within the rational-transfer-function framework of classical control theory.
Does the approximant match the function globally?
No. It matches the Taylor series to order m + n at the expansion point and often extends accuracy across captured singularities, but far from the origin and far from any pole it reproduces, accuracy can still fade. Padé extends local approximation impressively; it does not make it global.
Where are Padé approximants used in software?
In the implementation of special functions (exponential, logarithm, trigonometric, gamma) in numerical libraries, where a rational approximation evaluates faster and more uniformly than a truncated Taylor series. They also underpin rational filter design in signal processing and model-order reduction in engineering.