_{1}

^{*}

The study of the confluences of the roots of a given set of polynomials—root-pattern problem— does not appear to have been considered. We examine the situation, which leads us on to Young tableaux and tableaux representations. This in turn is found to be an aspect of multipartite partitions. We discover, and show, that partitions can be expressed algebraically and can be “differentiated” and “integrated”. We show a complete set of bipartite and tripartite partitions, indicating equivalences for the root-pattern problem, for select pairs and triples. Tables enumerating the number of bipartite and tripartite partitions, for small pairs and triples are given in an appendix.

We are interested in the “root-patterns” or confluences of the roots of a given set of polynomials―a topic one may have expected would have been studied in depth in the 19th century. However, apart from results on Resultants etc., there does not appear to have been much further development.

Motivation for consideration here arises from General Relativity where the classification of the Lanczos-Zund (3,1) spinor involves various confluences of the roots of two cubics [

It is shown here how the root-patterns problem becomes a problem in partitions. In this context, from bipartite partitions, an application to derivations of spinor factorizations in General Relativity has been made [

For example, given a quadratic and two cubics, a root-pattern may be indicated by ab, aad, bce where a, b are the roots of the quadratic and a, a, b and b, c, e are the roots of each of the cubics.

Three observations may be made. Firstly, the order of the polynomials is immaterial and so the root-pattern is also aad, ab, bce, or bce, aad, ab etc. Secondly, although we have written the roots of each polynomial in “ascending letter” order, the confluences of the roots or root-pattern is unchanged if we rearrange the letter order for each polynomial’s roots. Then the last root-pattern is also ebc, ada, ab. Thirdly, for any root-pattern, we can replace any letter by any other, unused, letter as the actual value of the root is unimportant. This then allows an interchange of letters, e.g., bce, aad, ab with the interchange

From an initial set S, a collection of elements of S where elements can be repeated, and shown juxtaposed, is just a list. A list with r elements is an r-element list and is the degree of the corresponding polynomial whose roots are the elements of the list or tuple. We define a root-pattern as just a collection of lists. Thus, the root-pattern ab, aad, bce is the collection of lists ab and aad and bce where the initial set is

The number of lists in the collection is the number of polynomials. The number of components in a list is the degree of the polynomial.

If we take two polynomials, we refer to the binary case, for three polynomials, the ternary case etc.

The three observations we made earlier may now be formalized as the following rules.

1) Any two polynomials can be interchanged. This translates to any two lists in a root-pattern can be interchanged.

2) Any rearrangement of the ordering of roots of a polynomial translates as a rearrangement of the ordering of the roots in the corresponding list for that polynomial in the root-pattern.

3) Any permutation of the letters (roots) in the set of polynomials translates to a permutation of the letters in the root pattern.

It is the commonality or confluence of roots in the various polynomials that we are interested in.

We define two root-patterns A and B to be equivalent if B can be obtained from A by any of the three rules; otherwise they are inequivalent.

For the ternary case, the root patterns ab, aad, bce and bc, bvb, xcy are equivalent but neither are equivalent to aad, bb, acc.

For a given set of polynomials the various root-patterns, that is the combinations of the various roots, which include common roots, is of interest to us.

What we would like to obtain is the enumeration and the consequent collection of (inequivalent) root-patterns of several polynomials. More specifically, given m_{1} linear forms, m_{2} quadratics, ・・・, m_{n} n-ics, enumerate and determine the set of inequivalent root-patterns. This is the “root-pattern’ problem.

Let us first consider a simple example.

Two QuadraticsThe roots of two quadratics (two lists of pairs,

There are 7 inequivalent cases which can be written as rows

Note, for example, that a case written as ab, cc (12, 33) is essentially the same as cc, ab (33, 12) since the order of the quadratics is immaterial. Then too cc, ab (33, 12) is also the same as case 4 aa, bc (11, 23), since the letters (and numbers) used do not matter.

It is seen that the roots of the two quadratics, displayed as rows, are instances of Young tableaux. We can always represent the roots of a set of polynomials as Young tableaux. For a polynomial of highest degree, say

As the order of roots of a polynomial in any row is immaterial we will take it that the numbers in each row of a Young tableau are always arranged in weakly increasing order.

A question arises as to whether every tableau can be ordered so that every row and column is weakly increasing. This is in fact not so. For any permutation of

cannot be displayed as one with weakly increasing columns, also allowing for the interchange of any rows.

We will take it that the numbers used in any tableau of n rows will be consecutive,

We now construct a different numbering on a tableau T. For a given T we form n-tuples or lists and there will be at most q of them. We will write the n-tuples or lists in sequence which we call the tableau representation. The

count the number of times 1 appears in the first row of the tableau, next the number of times 1 appears in the second row and so on, written as 110, the first 3-tuple or list. Then count the number of times 2 appears in the first row, next the number of times 2 appears in the second row and so on, to get 002 etc., so that finally we construct the tableau representation 110, 002, 120, 100 of T.

From this representation, the tableau can be reconstructed as follows.

1) The sum of the first components of each triple tells us the length of the first row; similarly for the second and third. Thus for T we have a 3-rowed tableau with row lengths

2) Since there are four triples there will be 4 different numbers 1, 2, 3, 4 used.

3) The first component of each 3-tuple tells us that the first row contains one 1, one 3 and one 4, so the first row is 134. Similarly the second row is 133 and the third row is 22. Putting these together, one row under another, in order, recovers T.

In the tableau representation of a set of polynomials, the

The 7 tableaux for two quadratics, namely

have the 7 tableaux representations

Partitions may be represented in two ways: for example the partitions of the number (4) are: 4, 3 + 1, 2 + 2, 2 + 1 + 1, 1 + 1 + 1 + 1 and a second way as:

so that we have

The general rule for partitions of a single variable is given by a generating function [

where the exponent of

For bipartite partitions the generating function for a set of pairs

Expanding as a power series in

We will use the first notation, however, and consider bipartite partitions here.

The pair (2,1) has the following 4 (bi)partitions:

We have used a delimiting comma here, rather than the usual summation sign, since we will now use the summation sign to express the partitions as a “sum”

Any (bipartite) partition is a tuple/list of parts: thus, the partition 11, 10 has parts 11 and 10. Each part is comprised of components; the part 10 has components 1 and 0. Partitions are then shown as their tableau representations. Thus we can talk of a partition or a tableau representation of a tableau, and construct a tableau that represents the partition1.

The + symbol used here needs more definition. Whilst it will be legitimate to write

We refer to the rule as the nil-addition of a partition.

The root-patterns of two quadratics, written (2,2) can then be shown as the sum of the 7 tableau representations, so that we now write

Note that the tableau representations

interchanging rows. The root patterns are equivalent for these two. Including the latter two equivalent tableaux representations to the 7 above, we have

This is exactly the expression of the pair (2,2) into its 9 bipartite partitions.

The 7 tableau representations (or corresponding partitions) for the roots of two quadratics are a subset of the full set of tableau representations of all partitions of (2,2).

It is convenient to call the sum of all partitions of a given number, pair, triple etc., its partition representation.

Algebraic RepresentationParts and hence partitions of a number, pair of numbers etc., can be represented algebraically. For the bipartite case let x represent the tuple 10 and y the tuple 01. Then put

Let us denote the concatenation of two tuples by a concatenation (circle) symbol

Besides the concatenation symbol

We define the following laws

where ac etc., is the ordinary (dot・) multiplication of two monomials.

It is easily seen that

Rules (2), (3) and (4) allow us to simplify some of the calculations that are used, by employing simplified rules. Thus with

using nil-addition. Most often we will therefore use rule 4′

Then with

The term

Rule 5 is a distributive law, and for simplified versions we have

Collating the rules that will be frequently used we have

Note that (5′) is derivable from (4′) and that the rhs of

In these shortened versions of the original laws it may be convenient to use the terminology of an integating operator instead of that of the extension operator, using the symbol i rather than

Let

The left table below shows an example of the use of the extension operator employed “algebraically”, with the rhs consisting of monomials, including concatenations of them. The right table shows the interpretation of these as tableau representations.

Example 1. Suppose we wish to obtain the tableau representation of (2,2) associated with the roots of two quadratics. It consists of the 9 partitions

We can construct the (2,2) partition representation from the algebraic representation of (2,1) The algebraic representation of

To get the (2,2) algebraic representation from the (2,1) algebraic representation we multiply it by 01, that is by the monomial y, using the extension operator

Taking each of the 4 terms separately we get, making simplifications and writing terms with

Adding these up we have the algebraic representation of (2,2)

Ignoring the irrelevant numerical coefficients (nil-addition) in this 9 term algebraic representation of (2,2) we have

In terms of the partition representation this is

The actual tableau corresponding to each term in these expressions is easily constructed.

In the algebraic representation, the terms

Thus we may consider the (2,2) partitioning as consisting of 7 inequivalent pairs. We may take this by defining an order (dominance) algebraically on the variables, say, x > y (thus ignoring the term

Alternatively we could “symmetrize” the expression and consider the set of symmetrised terms,

which are 7.

So really the problem of finding the number of inequivalent root-patterns of a set of polynomials (two quadratics here) is subsumed as that of determining the set of symmetrized partitions, a subset of all partitions corresponding to all tableau representations of the polynomials.

Partitons now being expressed algebraically, provide an opportunity to introduce “differentiation”.

If a and b are monomials in x (they must be of the form

In practice we may just differentiate a monomial and ignore any coefficients. The derivative operator can be extended to a partial derivative operator such as

Example 2. The tableau and algebraic partitioning of numbers (3) and (4) is

Ordinary differentiation of the latter (also using 2.) gives

which, ignoring coefficients, reduces to

The tableau and algebraic partitioning of the bivariate partitions (2,1) and (2,2) is

Differentiating the latter with respect to y gives

which all amounts to, ignoring coefficients,

precisely the partitioning of (2,1).

The process of differentiation in the example has exhibited the “downgrading” of partitions―from (4) to (3) and from (2,2) to (2,1). The reverse process of “integrating” or “upgrading” was performed in Example 1 in deriving the partitioning of (2,2) from (2,1).

The partition representations of low degree polynomials are displayed. Equivalent partitions are superscripted alike. The first-listed partition is the dominant one. Partitions of all different row lengths are necessarily all inequivalent. Partitions with equal row lengths will have partition equivalents. The algebraic representations can easily be constructed from the tableau representations.

There are 9 possible partitions with 7 being inequivalent.

As the polynomials are of different degrees all partitions are inequivalent. There are 16 inequivalent partitions.

The total number of partitions is 31. The number of inequivalent ones (here) is 21.

Other partitions, equivalent to some listed here, are:

The total number of partitions is 21. The number of inequivalent ones is 17.

It is easily seen that a partition is equivalent to a given partition if the

The total number of partitions is 26. The number of inequivalent ones is 20.

The total number of partitions is 66. The number of inequivalent ones is 51.

The table shows the number of bipartite partitions for

The tables show the number of tripartite partitions for