# Inverse function of $y=x+kx^3$

I want to invert the following function with respect to $x$:

$$f(x, k)=x+k x^3$$

where typical values for $x$ are between $0$ and $100$ and typical values for $k$ are between $-0.00005$ and $0.00005$. Further, it is known that:

$$k >-\frac{1}{3x^2}$$

therefore, the function should be invertible. The derivatives:

\begin{align}\frac{df}{dx}&=1+3kx^2\\ \frac{df}{dk} &= x^3\end{align}

are non-negative for all possible values of x and k. So the Function $f$ is monotonically increasing in both dimensions.

I need the function $g(y, k)$ so that $g(f(x, k), k) = x$ for all $x$ and all $k$.

What I have tried: The following function returns correct values for all $k>0$. For negative $k$ it does not work (returns complex numbers).

$$w = \sqrt[3]{\frac{y}{2 k} + \sqrt{\frac{y^2}{4 k^2} + \frac{1}{27 k^3}}}\\ g(y,k) = w - \frac{1}{3 kw}$$

-
u can try wolfram alpha, for reference. –  SHOBHIT GAUTAM Jul 21 '13 at 9:41
Well, I tried: wolframalpha.com/input/… still uses complex numbers all over the place and does not seem to be valid for positive k. –  DasKrümelmonster Jul 21 '13 at 11:02
The inverse function will be an odd function. Its power series might converge relatively quickly, and be reasonably accurate with a manageable number of terms. You appear to using approximate values anyway, is this a correct impression? –  Jyrki Lahtonen Jul 21 '13 at 12:13
The function comes from distortion correction. The coefficient k is found by camera calibration and in my program I need to convert image to world coordinates and vice versa. While the transformation is measured (and therefore, cannot be exact) I really need exact roundtrips. (i.e. the g(f(x, k), k) = x must be very accurate) –  DasKrümelmonster Jul 21 '13 at 12:35

Okay, i got it.

Using the second solution from WolframAlpha, and discarding the imaginary part, does the job for negative $k$. As both solutions do not work for $k=0$ I have the following solution:

function [ x ] = im2w( y, Kappa )
if abs(Kappa) < 1e-10
x = y / (1 + Kappa *  (y' * y));
elseif(Kappa > 0)
direction = y/norm(y);
w = nthroot(norm(y)/(2*Kappa) + sqrt((y'*y)/(4 * Kappa^2) + 1/(27*Kappa^3)), 3);
x = direction * (w - 1./(3*Kappa*w));
elseif(Kappa < 0)
direction = y/norm(y);
y = norm(y);
x_wa2 = (1-1i * sqrt(3))/(2^(2/3) * 3^(1/3) * (9 * Kappa^2 * y+sqrt(3) * sqrt(27 * Kappa^4 * y^2+4 * Kappa^3))^(1/3))-((1+1i * sqrt(3)) * (9 * Kappa^2 * y+sqrt(3) * sqrt(27 * Kappa^4 * y^2+4 * Kappa^3))^(1/3))/(2 * 2^(1/3) * 3^(2/3) * Kappa);

x = direction * real(x_wa2);
else
x = [0; 0]; % should not happen
end
end


This is Matlab Code (sorry) but gives results for all possible combinations for $k$ and $x$. The constraint for $k$: $$k > -\frac{4}{27 y^2}$$ Not sure whether I was wrong, or the difference is because I used x, wheres WA uses y.

-