Arithmetic-geometric Mean

Share:

Description

The arithmetic-geometric mean of real or complex numbers.

Usage

1
agmean(a, b)

Arguments

a, b

vectors of real or complex numbers of the same length (or scalars).

Details

The arithmetic-geometric mean is defined as the common limit of the two sequences a_{n+1} = (a_n + b_n)/2 and b_{n+1} = √(a_n b_n).

When used for negative or complex numbers, the complex square root function is applied.

Value

Returns a list with compoinents: agm a vector of arithmetic-geometric means, component-wise, niter the number of iterations, and prec the overall estimated precision.

Note

Gauss discovered that elliptic integrals can be effectively computed via the arithmetic-geometric mean (see example below), for example:

\int_0^{π/2} \frac{dt}{√{1 - m^2 sin^2(t)}} = \frac{(a+b) π}{4 \cdot agm(a,b)}

where m = (a-b)/(a+b)

References

http://mathworld.wolfram.com/Arithmetic-GeometricMean.html

See Also

Arithmetic, geometric, and harmonic mean.

Examples

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
##  Accuracy test: Gauss constant
1/agmean(1, sqrt(2))$agm - 0.834626841674073186  # 1.11e-16 < eps = 2.22e-16

## Gauss' AGM-based computation of \pi
a <- 1.0
b <- 1.0/sqrt(2)
s <- 0.5
d <- 1L
while (abs(a-b) > eps()) {
    t <- a
    a <- (a + b)*0.5
    b <- sqrt(t*b)
    c <- (a-t)*(a-t)
    d <- 2L * d
    s <- s - d*c
}
approx_pi <- (a+b)^2 / s / 2.0
abs(approx_pi - pi)             # 8.881784e-16 in 4 iterations

##  Example: Approximate elliptic integral
N <- 20
m <- seq(0, 1, len = N+1)[1:N]
E <- numeric(N)
for (i in 1:N) {
    f <- function(t) 1/sqrt(1 - m[i]^2 * sin(t)^2)
    E[i] <- quad(f, 0, pi/2)
}
A <- numeric(2*N-1)
a <- 1
b <- a * (1-m) / (m+1)

## Not run: 
plot(m, E, main = "Elliptic Integrals vs. arith.-geom. Mean")
lines(m, (a+b)*pi / 4 / agmean(a, b)$agm, col="blue")
grid()
## End(Not run)

Want to suggest features or report bugs for rdrr.io? Use the GitHub issue tracker.