Stirling's approximation

In mathematics, Stirling's approximation is an asymptotic approximation for factorials. It is a good approximation, leading to accurate results even for small values of. It is named after James Stirling, though a related but less precise result was first stated by Abraham de Moivre.
One way of stating the approximation involves the logarithm of the factorial:
where the big O notation means that, for all sufficiently large values of, the difference between and will be at most proportional to the logarithm of. In computer science applications such as the worst-case lower bound for comparison sorting, it is convenient to instead use the binary logarithm, giving the equivalent form
The error term in either base can be expressed more precisely as, corresponding to an approximate formula for the factorial itself,
Here the sign means that the two quantities are asymptotic, that is, their ratio tends to 1 as tends to infinity.

History

The formula was first discovered by Abraham de Moivre in 1721 in the form
De Moivre gave an approximate rational-number expression for the natural logarithm of the constant. Stirling's contribution in 1730 consisted of showing that the constant is precisely.

Derivation

The simplest version of Stirling's formula is
It can be quickly obtained by approximating the sum
with an integral:
The full formula, together with precise estimates of its error, can be derived as follows. Instead of approximating, one considers its natural logarithm, as this is a slowly varying function:
The right-hand side of this equation minus
is the approximation by the trapezoid rule of the integral
and the error in this approximation is given by the Euler–Maclaurin formula:
where is a Bernoulli number, and is the remainder term in the Euler–Maclaurin formula. Take limits to find that
Denote this limit as. Because the remainder in the Euler–Maclaurin formula satisfies
where big-O notation is used, combining the equations above yields the approximation formula in its logarithmic form:
Taking the exponential of both sides and choosing any positive integer, one obtains a formula involving an unknown quantity. For, the formula is
The quantity can be found by taking the limit on both sides as tends to infinity and using Wallis' product, which shows that. Therefore, one obtains Stirling's formula.

Alternative derivations

An alternative formula for using the gamma function is
. Rewriting and changing variables, one obtains
Applying Laplace's method one has
which recovers Stirling's formula:

Higher orders

Further corrections can also be obtained using Laplace's method. Stirling's formula to two orders is
From previous result, we know that, so we "peel off" this dominant term, then perform two changes of variables, to obtain:To verify this:.
Now the function is unimodal, with maximum value zero. Locally around zero, it looks like, which is why we are able to perform Laplace's method. In order to extend Laplace's method to higher orders, we perform another change of variables by. This equation cannot be solved in closed form, but it can be solved by serial expansion, which gives us. Now plug back to the equation to obtainnotice how we don't need to actually find, since it is cancelled out by the integral. Higher orders can be achieved by computing more terms in, which can be obtained programmatically.

Complex-analytic version

A complex-analysis version of this method is to consider as a Taylor coefficient of the exponential function, computed by Cauchy's integral formula as
This line integral can then be approximated using the saddle-point method with an appropriate choice of contour radius. The dominant portion of the integral near the saddle point is then approximated by a real integral and Laplace's method, while the remaining portion of the integral can be bounded above to give an error term.

Using the Central Limit Theorem and the Poisson distribution

An alternative version uses the fact that the Poisson distribution converges to a normal distribution by the Central Limit Theorem.
Since the Poisson distribution with parameter converges to a normal distribution with mean and variance, their density functions will be approximately the same:
Evaluating this expression at the mean, at which the approximation is particularly accurate, simplifies this expression to:
Taking logs then results in
which can easily be rearranged to give:
Evaluating at gives the usual, more precise form of Stirling's approximation.

Speed of convergence and error estimates

Stirling's formula is in fact the first approximation to the following series :
An explicit formula for the coefficients in this series was given by G. Nemes. Further terms are listed in the On-Line Encyclopedia of Integer Sequences as and. The first graph in this section shows the relative error vs., for 1 through all 5 terms listed above. gives the asymptotic formula for the coefficients:
which shows that it grows superexponentially, and that by the ratio test the radius of convergence is zero.
However, the representation obtained directly from the Euler–Maclaurin approximation, in which the correction term itself is the argument of the exponential function, converges much faster :
The th coefficient is directly calculated using Bernoulli numbers and
As, the error in the truncated series is asymptotically equal to the first omitted term. This is an example of an asymptotic expansion. It is not a convergent series; for any particular value of there are only so many terms of the series that improve accuracy, after which accuracy worsens. This is shown in the next graph, which shows the relative error versus the number of terms in the series, for larger numbers of terms. More precisely, let be the Stirling series to terms evaluated at . The graphs show
which, when small, is essentially the relative error.
Writing Stirling's series in the form
it is known that the error in truncating the series is always of the opposite sign and at most the same magnitude as the first omitted term.
Other bounds, due to Robbins, valid for all positive integers are
This upper bound corresponds to stopping the above series for after the term.
The lower bound is weaker than that obtained by stopping the series after the term. A looser version of this bound is that for all.

Stirling's formula for the gamma function

For all positive integers,
where denotes the gamma function.
However, the gamma function, unlike the factorial, is more broadly defined for all complex numbers other than non-positive integers; nevertheless, Stirling's formula may still be applied. If, then
Repeated integration by parts gives
where is the th Bernoulli number. The formula is valid for large enough in absolute value, when, where is positive, with an error term of. The corresponding approximation may now be written:
where the expansion is identical to that of Stirling's series above for, except that is replaced with.
A further application of this asymptotic expansion is for complex argument with constant. See for example the Stirling formula applied in of the Riemann–Siegel theta function on the straight line.

A convergent version of Stirling's formula

showed, in a letter to John Canton published by the Royal Society in 1763, that Stirling's formula did not give a convergent series. Obtaining a convergent version of Stirling's formula entails evaluating Binet's formula:
One way to do this is by means of a convergent series of inverted rising factorials. If
then
where
where denotes the Stirling numbers of the first kind. From this one obtains a version of Stirling's series
which converges when.
Stirling's formula may also be given in convergent form as
where

Versions suitable for calculators

The approximation
and its equivalent form
can be obtained by rearranging Stirling's extended formula and observing a coincidence between the resultant power series and the Taylor series expansion of the hyperbolic sine function. This approximation is good to more than 8 decimal digits for with a real part greater than 8. Robert H. Windschitl suggested it in 2002 for computing the gamma function with fair accuracy on calculators with limited program or register memory.
Gergő Nemes proposed in 2007 an approximation which gives the same number of exact digits as the Windschitl approximation but is much simpler:
or equivalently,
An alternative approximation for the gamma function stated by Srinivasa Ramanujan in Ramanujan's lost notebook is
for. The equivalent approximation for has an asymptotic error of and is given by
The approximation may be made precise by giving paired upper and lower bounds; one such inequality is