Here’s a calculation that I’ve had to do a few times. I’m writing it up here for my future reference and for the benefit of anybody else who needs to do the same calculation.

The Jacobi sn function is doubly periodic: it has one period as you move along the real axis and another period as you move along the imaginary axis. Its values are determined by a fundamental rectangle, and here we solve for a parameter 0 ≤ m < 1 that yields a desired length, or width, or aspect ratio for this rectangle.

You cannot specify both the length and width of the fundamental rectangle independently: you only have one degree of freedom. You can specify its length, or width, or aspect ratio.

This post focuses on sn, then explains how to modify the results for the other Jacobi elliptic functions—cn, dn, sc, etc.—at the end.

Given period

A parameter value m corresponds to a horizontal period of 4K(m) and a vertical period of 2K(1-m) where K is the complete elliptic integral of the first kind.

K(m) is an increasing function from [0, 1) to [π/2, ∞). So the minimum horizontal period is 2π and the minimum vertical period is π.

Given aspect ratio

The ratio of the horizontal period to the vertical period is given by

r(m) = 4K(m) / 2K(1-m) = 2K(m)/K(1-m).

The function r is an increasing function from [0, 1) to [0, ∞), so any aspect ratio is possible. This is important, for example, in conformal mapping. You cannot specify the length and width of the fundamental period independently, but you can solve for the parameter m that gives the aspect ratio you want, then follow it by a dilation to then get the length and width you want.

Numerical computation

If you need to find m efficiently or to high precision, see [1]. The code here simple and good enough for my purposes, and hopefully good enough for yours.

We need to either invert K or r, both of which are monotone functions, and so we can use the bisection method to find the value of m we’re after. The bisection method is not the most efficient, but it’s simple and very robust.

Python code

    from numpy import pi, linspace
    from scipy.special import ellipk, ellipj
    from scipy.optimize import bisect
    
    def invert_ellipk(K):
        "Given K, find m such that K(m) = K."
        assert(K >= pi/2)
        if K == pi/2:
            return 0
    
        # Find a bracket [a, b] for m
        a = 0
        b = 0.5
        y = ellipk(b)
        while y < K:
            b = 0.5*(b + 1)
            y = ellipk(b)
        return bisect(lambda m: ellipk(m)-K, a, b)
    
    def invert_horizontal_period(T):
        "Find parameter m such that sn(z, m) has real period T."
        return invert_ellipk(T/4)
    
    def invert_vertical_period(T):
        "Find parameter m such that sn(z, m) has imaginary period T."    
        return invert_ellipk(T/2)
    
    def invert_r(r):
        "Find parameter m such that the fundamental rectangle of sn(z, m) has aspect ratio r."
    
        ratio = lambda m: 2*ellipk(m)/ellipk(1-m)
        
        if r == 2:
            return 1/2
    
        a, b = 0.5, 0.5
        while ratio(b) < r: b = 0.5*(1 + b) while ratio(a) > r:
            a /= 2
        return bisect(lambda m: r - ratio(m), a, b)

To test the code, let’s find m such that sn(z, m) has period 5 along the real axis.

    import matplotlib.pyplot as plt

    T = 16
    m = invert_horizontal_period(T)
    print(4*ellipk(m))
    assert(abs(4*ellipk(m) - T) < 1e-8)
    x = linspace(0, 2*T, 200)
    plt.plot(x, sn(x, m))

This produces a plot that does have period 16.

Other Jacobi elliptic functions

There are 12 Jacobi elliptic functions, denoted with all pairs of {s, c, d, n} such that the two letters are different. The periods of the functions fall into three groups.

The functions sn, ns, cd, and dc all have real period 4K(m) and imaginary period 2K(1-m).

The functions cn, nc, ds, and sd all have real period 4K(m) and imaginary period 4K(1-m).

The functions cs, sc, dn, and nd all have real period 2K(m) and imaginary period 4K(1-m).

You can find plots of fundamental rectangles of these 12 functions here.

[1] Toshio Fukushima. Numerical computation of inverse complete elliptic integrals of the first and second kinds. Journal of Computational and Applied Mathematics. 249 (2013) 37–50.

In the previous post we wanted to find a value of n such that f(n) = 10¹² where

and we took a rough guess n = 200. Turns out f(200) ≈ 4 × 10¹² and that was good enough for our purposes. But what if we wanted to solve f(n) = x accurately?

We will work our way up to solving f(n) = 10¹² exactly using the Lambert W function.

Lambert’s W function

If you can solve one equation involving exponential functions you can bootstrap your solution solve a lot more equations.

The Lambert W function is defined to be the function W(x) that maps each x to a solution of the equation

w exp(w) = x.

This function is implemented Python under scipy.special.lambertw. Let’s see how we could use it solve similar equations to the one that define it.

Basic bootstrapping

Given a constant c we can solve

cw exp(w) = x

simply by solving

w exp(w) = x/c,

which says w = W(x/c).

And we can solve

w exp(cw) = x

by setting y = cw and solving

y exp(y) = cx.

and so cw = W(cx) and w = W(cx)/c.

Combining the two approaches above shows that the solution to

aw exp(bw) = x

is

w = W(bx/a) / b.

We can solve

exp(w) / w = x

by taking the reciprocal of both sides and applying the result above with a = 1 and b = -1.

More generally, we can solve

w^c exp(w) = x

by raising both sides to the power 1/c; the reciprocal the special case c = 1.

Similarly, we can solve

w exp(w^c) = x

by setting y = w^c , changing the problem to

y^-1/c exp(y) = x.

Putting it all together

We’ve now found how to deal with constants and exponents on w, inside and outside the exponential function. We now have all the elements to solve our original problem.

We can solve the general equation

with

The equation at the top of the post corresponds to a = 1/√48, b = -1, c = π √(2/3), and d = 1/2.

We can code this up in Python as follows.

    from scipy.special import lambertw

    def soln(a, b, c, d, x):
        t = (x/a)**(d/b) *c*d/b
        return (lambertw(t)*b/(c*d))**(1/d) 

We can verify that our solution really is a solution by running it though

    from numpy import exp, pi

    def f(a, b, c, d, w):
        return a*w**b * exp(c*w**d)

to make sure we get our x back.

When we run

    a = 0.25*3**-0.5
    b = -1
    c = pi*(2/3)**0.5
    d = 0.5
    x = 10**12

    w = soln(a, b, c, d, x)
    print(f(a, b, c, d, w))

we do indeed get x back.

    a = 0.25*3**-0.5
    b = -1
    c = pi*(2/3)**0.5
    d = 0.5

    w = soln(a, b, c, d, 10)
    print(f(a, b, c, d, w))

When we take a look at the solution w, we get 1.443377079584187e-13. In other words, we get a number near zero. But our initial guess was w = 200. What went wrong?

Nothing went wrong. We got a solution to our equation. It just wasn’t the solution we expected.

The Lambert W function has multiple branches when viewed as a function on complex numbers. It turns out the solution we were expecting is on the branch SciPy denotes with k = -1. If we add this to the call to lambertw above we get the solution 183.85249773880142 which is more in line with what we expected.

More Lambert W posts

• Lambert W strikes again
• Giant components
• Tower of powers
• Approximate inverse of gamma function

The Bessel functions J and Y are analogous to sine and cosine. Bessel functions come up in polar coordinates the way sines and cosines come up in rectangular coordinates. There are J and Y functions of various orders, conventionally written with a subscript ν.

I recently ran across a curious set of relations between these functions and their derivatives. Let’s start with the following terms defined by 2 by 2 determinants.

There are a lot of symmetries in the definitions above. First, every term has a subscript ν, so you can ignore those for now.

Second, every determinant has J‘s on the top row and Y‘s on the bottom.

Third, every determinant has a‘s in the first column and b‘s in the second.

And finally, the primes, indicating derivatives, have the following pattern.

Now that we have all these definitions in place, there are several identities that the p’s, q’s, r’s, and s‘s satisfy. Some of them depend on the orders, and that’s why we included the ν subscripts. You can find various relations in A&S equation 9.1.32, but I wanted to point out one that’s particularly elegant, A&S equation 9.1.34:

This equation, a determinant of terms defined by determinants, reduces to a simple form that depends only on where the Bessel functions are evaluated, i.e. a and b, but not on their order ν.

Python example

The Python code below will show how to call Bessel functions and their derivatives and will illustrate the equations above.

    from math import pi
    from scipy.special import jv, yv, jvp, yvp
    
    def example(n, a, b):
    
        ja = jv(n, a)
        jb = jv(n, b)
        ya = yv(n, a)
        yb = yv(n, b)
        
        jap = jvp(n, a)
        jbp = jvp(n, b)
        yap = yvp(n, a)
        ybp = yvp(n, b)
        
        p = ja  * yb  - jb  * ya
        q = ja  * ybp - jbp * ya
        r = jap * yb  - jb  * yap
        s = jap * ybp - jbp * yap
        
        print(p*s - q*r)
        print(4/(pi*pi*a*b))
    
    example(3, 16, 21)

This prints

    0.0012062045671706876
    0.0012062045671706878

The two results differ slightly in the last decimal place due to rounding error.

The order of a Bessel function can be any real number [1], and the arguments can be any complex number. Here’s another example with more general arguments.

    example(3.14, 16-3j, 21+1.2j)    

This prints

    (0.0011738907214344785 + 0.00015140286689885318j)
    (0.0011738907214345800 + 0.00015140286689880630j)

with the real and imaginary parts agreeing to 15 decimal places.

Related posts

• Denver airport, Weierstrass, and A&S
• Fourier-Bessel series
• Bessel function crossings

[1] You can define Bessel functions with complex order, but SciPy doesn’t support that.