Last time we took a look at how we could define copulas to represent the dependency between random variables by summing the results of a generator function φ applied to the results of their cumulative distribution functions, or CDFs, and then applying the inverse of that function φ^-1 to that sum.
These are known as Archimedean copulas and are valid whenever φ is strictly decreasing over the interval [0,1], equal to zero when its argument equals one and have n^th derivatives that are non-negative over that interval when n is even and non-positive when it is odd, for n up to the number of random variables.
Whilst such copulas are relatively easy to implement we saw that their densities are a rather trickier job, in contrast to Gaussian copulas where the reverse is true. In this post we shall see how to draw random vectors from Archimedean copulas which is also much more difficult than doing so from Gaussian copulas.

Greetings Sir R-----! Might I suggest that you take one of these spiced beef pies and a mug of mulled cider to stave off this winter chill? And perhaps a wager to fire the blood?

Good man! Good man!

I propose a game that ever puts me in mind of my ill-fated expedition to recover for the glory of the Empress of Russia the priceless Amulet of Yendor from the very depths of Hell.

About a year and a half ago we saw how we could use Gaussian copulas to define dependencies between the elements of a vector valued multivariate random variable whose elements, when considered in isolation, were governed by arbitrary cumulative distribution functions, known as marginals. Whilst Gaussian copulas are quite flexible, they can't represent every possible dependency between those elements and in this post we shall take a look at some others defined by the Archimedean family of copulas.

The Baron's latest wager set Sir R----- the task of surpassing his score before he reached eight points as they each cast an eight sided die, each adding one point to their score should the roll of their die be less than or equal to it. The cost to play for Sir R------ was one coin and he should have had a prize of five coins had he succeeded.

A key observation when figuring the fairness of this wager is that if both Sir R----- and the Baron cast greater than their present score then the state of play remains unchanged. We may therefore ignore such outcomes, provided that we adjust the probabilities of those that we have not to reflect the fact that we have done so.

We have spent a few months looking at how we might interpolate between sets of points (x_i, y_i), where the x_i are known as nodes and the y_i as values, to approximate values of y for values of x between the nodes, either by connecting them with straight lines or with cubic curves.
Last time, in preparation for interpolating between multidimensional vector nodes, we implemented the ak.grid type to store ticks on a set of axes and map their intersections to ak.vector objects to represent such nodes arranged at the corners of hyperdimensional rectangular cuboids.
With this in place we're ready to take a look at one of the simplest multidimensional interpolation schemes; multilinear interpolation.

For several months now my fellow students and I have been exploring ℓ-space, being the set of infinite dimensional vectors whose elements are the powers of the prime factors of the roots of rational numbers, which we chanced upon whilst attempting to define a rational valued logarithmic function for such numbers.
We have seen how we might define functions of roots of rationals employing the magnitude of their associated ℓ-space vectors and that the iterative computation of such functions may yield cyclical sequences, although we conspicuously failed to figure a tidy mathematical rule governing their lengths.
The magnitude is not the only operation of linear algebra that we might bring to bear upon such roots, however, and we have lately busied ourselves investigating another.

Over the last few months we have been taking a look at algorithms for interpolating over a set of points (x_i,y_i) in order to approximate values of y between the nodes x_i. We began with linear interpolation which connects the points with straight lines and is perhaps the simplest interpolation algorithm. Then we moved on to cubic spline interpolation which yields a smooth curve by specifying gradients at the nodes and fitting cubic polynomials between them that match both their values and their gradients. Next we saw how this could result in curves that change from increasing to decreasing, or vice versa, between the nodes and how we could fix this problem by adjusting those gradients.
I concluded by noting that, even with this improvement, the shape of a cubic spline interpolation is governed by choices that are not uniquely determined by the points themselves and that linear interpolation is consequently a more mathematically appropriate scheme, which is why I chose to generalise it to other arithmetic types for y, like complex numbers or matrices, but not to similarly generalise cubic spline interpolation.

The obvious next question is whether or not we can also generalise the nodes to other arithmetic types; in particular to vectors so that we can interpolate between nodes in more than one dimension.

Sir R-----! I must say that it is a relief to have the company of a fellow nobleman in these distressing times. That I have had to sell not one, but two of my several hundred antiquities to settle the burden of tax that this oppressive democracy has put upon me, simply to enrich slugabeds I might add, is quite intolerable!

Come, let us drown our sorrows whilst we still have the means to do so and engage in a little sport to raise our spirits.

I have a fancy for a game that I used to play when I was the Russian ambassador to the Rose Tree Valley commune. Founded by the philosopher queen Zway Remington as a haven for downtrodden wealthy industrialists, it was the purest of pure meritocracies; no handouts to the idle labouring classes there!

We have seen how it is possible to smoothly interpolate between a set of points (x_i, y_i), with the x_i known as nodes and the y_i as values, by specifying the gradients g_i at the nodes and calculating values between adjacent pairs using the uniquely defined cubic polynomials that match the values and gradients at them.
We have also seen how extrapolating such polynomials beyond the first and last nodes can yield less than satisfactory results, which we fixed by specifying the first and last gradients and then adding new first and last nodes to ensure that the first and last polynomials would represent straight lines.
Now we shall see how cubic spline interpolation can break down rather more dramatically and how we might fix it.

Recall that the Baron's game is comprised of taking turns to place dominoes on a six by six grid of squares with each domino covering a pair of squares. At no turn was a player allowed to place a domino such that it created an oddly-numbered region of empty squares and Sir R----- was to be victorious if, at the end of play, the lines running between the ranks and files of the board were each and every one straddled by at least one domino.