The ‘jack’ package: Jack and other symmetric polynomials

Jack, zonal, Schur, and other symmetric polynomials.

R-CMD-check R-CMD-check-valgrind

library(jack)

Schur polynomials have applications in combinatorics and zonal polynomials have applications in multivariate statistics. They are particular cases of Jack polynomials, which are some multivariate symmetric polynomials. The original purpose of this package was the evaluation and the computation in symbolic form of these polynomials. Now it contains much more stuff dealing with multivariate symmetric polynomials.

Breaking change in version 6.0.0

In version 6.0.0, each function whose name ended with the suffix CPP (JackCPP, JackPolCPP, etc.) has been renamed by removing this suffix, and the functions Jack, JackPol, etc. have been renamed by adding the suffix R to their name: JackR, JackPolR, etc. The reason of these changes is that a name like Jack is more appealing than JackCPP and it is more sensible to assign the more appealing names to the functions implemented with Rcpp since they are highly more efficient. The interest of the functions JackR, JackPolR, etc. is meager.

Getting the polynomials

The functions JackPol, ZonalPol, ZonalQPol and SchurPol respectively return the Jack polynomial, the zonal polynomial, the quaternionic zonal polynomial, and the Schur polynomial.

Each of these polynomials is given by a positive integer, the number of variables (the n argument), and an integer partition (the lambda argument); the Jack polynomial has a parameter in addition, the alpha argument, a number called the Jack parameter.

Actually there are four possible Jack polynomials for a given Jack parameter and a given integer partition: the \(J\)-polynomial, the \(P\)-polynomial, the \(Q\)-polynomial and the \(C\)-polynomial. You can specify which one you want with the which argument, which is set to "J" by default. These four polynomials differ only by a constant factor.

To get a Jack polynomial with JackPol, you have to supply the Jack parameter as a bigq rational number or anything coercible to a bigq number by an application of the as.bigq function of the gmp package, such as a character string representing a fraction, e.g. "2/5":

jpol <- JackPol(2, lambda = c(3, 1), alpha = "2/5")
jpol
## 98/25*x^3.y + 28/5*x^2.y^2 + 98/25*x.y^3

This is a qspray object, from the qspray package. Here is how you can evaluate this polynomial:

evalQspray(jpol, c("2", "3/2"))
## Big Rational ('bigq') :
## [1] 1239/10

It is also possible to convert a qspray polynomial to a function whose evaluation is performed by the Ryacas package:

jyacas <- as.function(jpol)

You can provide the values of the variables of this function as numbers or character strings:

jyacas(2, "3/2")
## [1] "1239/10"

You can even pass a variable name to this function:

jyacas("x", "x")
## [1] "(336*x^4)/25"

If you want to substitute a complex number to a variable, use a character string which represents this number, with I denoting the imaginary unit:

jyacas("2 + 2*I", "2/3 + I/4")
## [1] "Complex((-158921)/2160,101689/2160)"

It is also possible to evaluate a qspray polynomial for some complex values of the variables with evalQspray. You have to separate the real parts and the imaginary parts:

evalQspray(jpol, values_re = c("2", "2/3"), values_im = c("2", "1/4"))
## Big Rational ('bigq') object of length 2:
## [1] -158921/2160 101689/2160

Direct evaluation of the polynomials

If you just have to evaluate a Jack polynomial, you don’t need to resort to a qspray polynomial: you can use the functions Jack, Zonal, ZonalQ or Schur, which directly evaluate the polynomial; this is much more efficient than computing the qspray polynomial and then applying evalQspray.

Jack(c("2", "3/2"), lambda = c(3, 1), alpha = "2/5")
## Big Rational ('bigq') :
## [1] 1239/10

However, if you have to evaluate a Jack polynomial for several values, it could be better to resort to the qspray polynomial.

Skew Jack polynomials

As of version 6.0.0, the package is able to compute the skew Schur polynomials with the function SkewSchurPol, and the general skew Jack polynomial is available as of version 6.1.0 (function SkewJackPol).

Symbolic Jack parameter

As of version 6.0.0, it is possible to get a Jack polynomial with a symbolic Jack parameter in its coefficients, thanks to the symbolicQspray package.

( J <- JackSymPol(2, lambda = c(3, 1)) )
## { [ 2*a^2 + 4*a + 2 ] } * X^3.Y  +  { [ 4*a + 4 ] } * X^2.Y^2  +  { [ 2*a^2 + 4*a + 2 ] } * X.Y^3

This is a symbolicQspray object, from the symbolicQspray package.

A symbolicQspray object corresponds to a multivariate polynomial whose coefficients are fractions of polynomials with rational coefficients. The variables of these fractions of polynomials can be seen as some parameters. The Jack polynomials fit into this category: from their definition, their coefficients are fractions of polynomials in the Jack parameter. However you can see in the above output that for this example, the coefficients are polynomials in the Jack parameter (a): there’s no fraction. Actually this fact is always true for the Jack \(J\)-polynomials. This is an established fact and it is not obvious (it is a consequence of the Knop & Sahi formula).

You can substitute a value to the Jack parameter with the help of the substituteParameters function:

( J5 <- substituteParameters(J, 5) )
## 72*X^3.Y + 24*X^2.Y^2 + 72*X.Y^3
J5 == JackPol(2, lambda = c(3, 1), alpha = "5")
## [1] TRUE

Note that you can change the letters used to denote the variables. By default, the Jack parameter is denoted by a and the variables are denoted by X, Y, Z if there are no more than three variables, otherwise they are denoted by X1, X2, … Here is how to change these symbols:

showSymbolicQsprayOption(J, "a") <- "alpha"
showSymbolicQsprayOption(J, "X") <- "x"
J
## { [ 2*alpha^2 + 4*alpha + 2 ] } * x1^3.x2  +  { [ 4*alpha + 4 ] } * x1^2.x2^2  +  { [ 2*alpha^2 + 4*alpha + 2 ] } * x1.x2^3

If you want to have the variables denoted by x and y, do:

showSymbolicQsprayOption(J, "showMonomial") <- showMonomialXYZ(c("x", "y"))
J
## { [ 2*alpha^2 + 4*alpha + 2 ] } * x^3.y  +  { [ 4*alpha + 4 ] } * x^2.y^2  +  { [ 2*alpha^2 + 4*alpha + 2 ] } * x.y^3

The skew Jack polynomials with a symbolic Jack parameter are available too, as of version 6.1.0.

Compact expression of Jack polynomials

The expression of a Jack polynomial in the canonical basis can be long. Since these polynomials are symmetric, one can get a considerably shorter expression by writing the polynomial as a linear combination of the monomial symmetric polynomials. This is what the function compactSymmetricQspray does:

( J <- JackPol(3, lambda = c(4, 3, 1), alpha = "2") )
## 3888*x^4.y^3.z + 2592*x^4.y^2.z^2 + 3888*x^4.y.z^3 + 3888*x^3.y^4.z + 4752*x^3.y^3.z^2 + 4752*x^3.y^2.z^3 + 3888*x^3.y.z^4 + 2592*x^2.y^4.z^2 + 4752*x^2.y^3.z^3 + 2592*x^2.y^2.z^4 + 3888*x.y^4.z^3 + 3888*x.y^3.z^4
compactSymmetricQspray(J) |> cat()
## 3888*M[4, 3, 1] + 2592*M[4, 2, 2] + 4752*M[3, 3, 2]

The function compactSymmetricQspray is also applicable to a symbolicQspray object, like a Jack polynomial with symbolic Jack parameter.

It is easy to figure out what is a monomial symmetric polynomial: M[i, j, k] is the sum of all monomials x^p.y^q.z^r where (p, q, r) is a permutation of (i, j, k).

The “compact expression” of a Jack polynomial with n variables does not depend on n if n >= sum(lambda):

lambda <- c(3, 1)
alpha <- "3"
J4 <- JackPol(4, lambda, alpha)
J9 <- JackPol(9, lambda, alpha)
compactSymmetricQspray(J4) |> cat()
## 32*M[3, 1] + 16*M[2, 2] + 28*M[2, 1, 1] + 24*M[1, 1, 1, 1]
compactSymmetricQspray(J9) |> cat()
## 32*M[3, 1] + 16*M[2, 2] + 28*M[2, 1, 1] + 24*M[1, 1, 1, 1]

In fact I’m not sure the Jack polynomial makes sense when n < sum(lambda).

Hall inner product

The qspray package provides a function to compute the Hall inner product of two symmetric polynomials, namely HallInnerProduct. This is the generalized Hall inner product, the one with a parameter \(\alpha\). It is known that the Jack polynomials with parameter \(\alpha\) are orthogonal for the Hall inner product with parameter \(\alpha\). Let’s give a try:

alpha <- "3"
J1 <- JackPol(4, lambda = c(3, 1), alpha, which = "P")
J2 <- JackPol(4, lambda = c(2, 2), alpha, which = "P")
HallInnerProduct(J1, J2, alpha)
## Big Rational ('bigq') :
## [1] 0
HallInnerProduct(J1, J1, alpha)
## Big Rational ('bigq') :
## [1] 135/8
HallInnerProduct(J2, J2, alpha)
## Big Rational ('bigq') :
## [1] 63/5

If you set alpha=NULL in HallInnerProduct, you get the Hall inner product with symbolic parameter \(\alpha\):

HallInnerProduct(J1, J1, alpha = NULL)
## 3/128*alpha^4 + 1/4*alpha^3 + 63/128*alpha^2 + 81/64*alpha

This is a qspray object. The Hall inner product is always polynomial in \(\alpha\).

It is also possible to get the Hall inner product of two symbolicQspray polynomials. Take for example a Jack polynomial with symbolic parameter:

J <- JackSymPol(4, lambda = c(3, 1), which = "P")
showSymbolicQsprayOption(J, "a") <- "t"
HallInnerProduct(J, J, alpha = 2)
## [ 20*t^4 - 24*t^3 + 92*t^2 - 48*t + 104 ] %//% [ t^4 + 4*t^3 + 6*t^2 + 4*t + 1 ]

We use t to display the Jack parameter and not alpha so that there is no confusion between the Jack parameter and the parameter of the Hall product.

Now, what happens if we compute the symbolic Hall inner product of this Jack polynomial with itself, that is, if we run HallInnerProduct(J, J, alpha = NULL)? Let’s see:

( Hip <- HallInnerProduct(J, J, alpha = NULL) )
## { [ 6 ] %//% [ t^4 + 4*t^3 + 6*t^2 + 4*t + 1 ] } * alpha^4  +  { [ 9*t^2 - 6*t + 1 ] %//% [ t^4 + 4*t^3 + 6*t^2 + 4*t + 1 ] } * alpha^3  +  { [ 3*t^4 - 6*t^3 + 5*t^2 ] %//% [ t^4 + 4*t^3 + 6*t^2 + 4*t + 1 ] } * alpha^2  +  { [ 4*t^4 ] %//% [ t^4 + 4*t^3 + 6*t^2 + 4*t + 1 ] } * alpha

We get the Hall inner product of the Jack polynomial with itself, with two symbolic parameters: the Jack parameter displayed as t and the parameter of the Hall product displayed as alpha. This is a symbolicQspray object.

Now one could be interested in the symbolic Hall inner product of the Jack polynomial with itself for the case when the Jack parameter and the parameter of the Hall product coincide, that is, to set alpha=t in the symbolicQspray polynomial that we named Hip. One can get it as follows:

changeVariables(Hip, list(qlone(1)))
## [ 3*t^4 + t^3 ] %//% [ t^2 + 2*t + 1 ]

This is rather a trick. The changeVariables function allows to replace the variables of a symbolicQspray polynomial with the new variables given as a list in its second argument. The Hip polynomial has only one variable, alpha, and it has one parameter, t. This parameter t is the polynomial variable of the ratioOfQsprays coefficients of Hip. Technically this is a qspray object: this is qlone(1). So we provided list(qlone(1)) as the list of new variables. This corresponds to set alpha=t. The usage of the changeVariables is a bit deflected, because qlone(1) is not a new variable for Hip, this is a constant.

Laplace-Beltrami operator

Just to illustrate the possibilities of the packages involved in the jack package (qspray, ratioOfQsprays, symbolicQspray), let us check that the Jack polynomials are eigenpolynomials for the Laplace-Beltrami operator on the space of homogeneous symmetric polynomials.

LaplaceBeltrami <- function(qspray, alpha) {
  n <- numberOfVariables(qspray)
  derivatives1 <- lapply(seq_len(n), function(i) {
    derivQspray(qspray, i)
  })
  derivatives2 <- lapply(seq_len(n), function(i) {
    derivQspray(derivatives1[[i]], i)
  })
  x <- lapply(seq_len(n), qlone) # x_1, x_2, ..., x_n
  # first term
  out1 <- 0L
  for(i in seq_len(n)) {
    out1 <- out1 + alpha * x[[i]]^2 * derivatives2[[i]]
  }
  # second term
  out2 <- 0L
  for(i in seq_len(n)) {
    for(j in seq_len(n)) {
      if(i != j) {
        out2 <- out2 + x[[i]]^2 * derivatives1[[i]] / (x[[i]] - x[[j]])
      }
    }
  }
  # at this step, `out2` is a `ratioOfQsprays` object, because of the divisions
  # by `x[[i]] - x[[j]]`; but actually its denominator is 1 because of some
  # simplifications and then we extract its numerator to get a `qspray` object
  out2 <- getNumerator(out2)
  out1/2 + out2
}
alpha <- "3"
J <- JackPol(4, c(2, 2), alpha)
collinearQsprays(
  qspray1 = LaplaceBeltrami(J, alpha), 
  qspray2 = J
)
## [1] TRUE

Other symmetric polynomials

Many other symmetric multivariate polynomials have been introduced in version 6.1.0. Let’s see a couple of them.

Skew Jack polynomials

The skew Jack polynomials are now available. They generalize the skew Schur polynomials. In order to specify the skew integer partition \(\lambda/\mu\), one has to provide the outer partition \(\lambda\) and the inner partition \(\mu\). The skew Schur polynomial associated to some skew partition is the skew Jack \(P\)-polynomial with Jack parameter \(\alpha=1\) associated to the same skew partition:

n <- 3
lambda <- c(3, 3)
mu <- c(2, 1)
skewSchurPoly <- SkewSchurPol(n, lambda, mu)
skewJackPoly <- SkewJackPol(n, lambda, mu, alpha = 1, which = "P")
skewSchurPoly == skewJackPoly
## [1] TRUE

\(t\)-Schur polynomials

The \(t\)-Schur polynomials depend on a single parameter usually denoted by \(t\) and their coefficients are polynomials in this parameter. They yield the Schur polynomials when substituting \(t\) with \(0\):

n <- 3
lambda <- c(2, 2)
tSchurPoly <- tSchurPol(n, lambda)
substituteParameters(tSchurPoly, values = 0) == SchurPol(n, lambda)
## [1] TRUE

Hall-Littlewood polynomials

Similarly to the \(t\)-Schur polynomials, the Hall-Littlewood polynomials depend on a single parameter usually denoted by \(t\) and their coefficients are polynomials in this parameter. The Hall-Littlewood \(P\)-polynomials yield the Schur polynomials when substituting \(t\) with \(0\):

n <- 3
lambda <- c(2, 2)
hlPoly <- HallLittlewoodPol(n, lambda, which = "P")
substituteParameters(hlPoly, values = 0) == SchurPol(n, lambda)
## [1] TRUE

Macdonald polynomials

The Macdonald polynomials depend on two parameters usually denoted by \(q\) and \(t\). Their coefficients are not polynomials in \(q\) and \(t\) in general, they are ratios of polynomials in \(q\) and \(t\). These polynomials yield the Hall-Littlewood polynomials when substituting \(q\) with \(0\):

n <- 3
lambda <- c(2, 2)
macPoly <- MacdonaldPol(n, lambda)
hlPoly <- HallLittlewoodPol(n, lambda)
changeParameters(macPoly, list(0, qlone(1))) == hlPoly
## [1] TRUE

Kostka numbers and Kostka polynomials

The ordinary Kostka numbers are usually denoted by \(K_{\lambda,\mu}\) where \(\lambda\) and \(\mu\) denote two integer partitions. The Kostka number \(K_{\lambda,\mu}\) is then associated to the two integer partitions \(\lambda\) and \(\mu\), and it is the coefficient of the monomial symmetric polynomial \(m_{\mu}\) in the expression of the Schur polynomial \(s_{\lambda}\) as a linear combination of monomial symmetric polynomials. It is always a non-negative integer. It is possible to compute these Kostka numbers with the jack package. They are also available in the syt package. There is more in the jack package. Since the Schur polynomials are the Jack \(P\)-polynomials with Jack parameter \(\alpha=1\), one can more generally define the Kostka-Jack number \(K_{\lambda,\mu}(\alpha)\) as the coefficient of the monomial symmetric polynomial \(m_{\mu}\) in the expression of the Jack \(P\)-polynomial \(P_{\lambda}(\alpha)\) with Jack parameter \(\alpha\) as a linear combination of monomial symmetric polynomials. The jack package allows to compute these numbers. Note that I call them “Kostka-Jack numbers” here as well as in the documentation of the package but I don’t know whether this wording is standard (probably not).

The Kostka numbers are also generalized by the Kostka-Foulkes polynomials, or \(t\)-Kostka polynomials, which are provided in the jack package. These are univariate polynomials whose variable is denoted by \(t\), and their value at \(t=1\) are the Kostka numbers. These polynomials are used in the computation of the Hall-Littlewood polynomials.

Finally, the Kostka numbers are also generalized by the Kostka-Macdonald polynomials, or \(qt\)-Kostka polynomials, also provided in the jack package. Actually these polynomials even generalize the Kostka-Foulkes polynomials. They have two variables, denoted by \(q\) and \(t\), and one obtains the Kostka-Foulkes polynomials by replacing \(q\) with \(0\). Currently the Kostka-Macdonald polynomials are not used in the jack package.

The skew generalizations are also available in the jack package: skew Kostka-Jack numbers, skew Kostka-Foulkes polynomials, and skew Kostka-Macdonald polynomials.

References