## Simple Proof of Tokuyama’s Formula

Tokuyama’s Formula is a combinatorial result that expresses the product of the deformed Weyl denominator and the Schur polynomial as a sum over strict Gelfand-Tsetlin patterns. This result implies Gelfand’s parametrization of the Schur polynomial, Weyl’s Character Formula, and Stanley’s formula on Hall-Littlewood polynomials — all for ; also, the formula is related to alternating sign matrices. Moreover, Tokuyama’s formula gives a combinatorial evaluation of certain matrix coefficients of representations of -adic groups.

A *Gelfand-Tsetlin pattern* (or GT-pattern for short) with the top row is a triangular array of numbers

such that the betweenness condition is satisfied for all . In other words, each entry in the pattern lies between two entries above it. A Gelfand-Tsetlin pattern is called *strict* if rows are strictly decreasing, that is, for all . Denote by and the sets of GT patterns and strict GT patterns with the top row , respectively.

We call entry *left-leaning* if . In other words, the entry equals to the entry above on the left. Similarly, is *right-leaning* if . We call *generic* if it’s not left- or right-leaning.

Let and be the number of left- and right-leaning entries in GT-pattern , respectively. Let be the number of generic entries. Also, let be the -th row sum, and , the difference of sums of consequent rows. Now we can state

**Theorem**. (Tokuyama, 1988). Let be a partition of length . We have

where , and is the Schur polynomial.

The formula expresses a weighted sum over strict Gelfand-Tsetlin patterns in the closed form as the product of the deformed Weyl denominator and the Schur polynomial. Note that the usual Weyl denominator for is obtained by setting .

**Example**. Let’s see Tokuyama’s formula in action. Let . Then . In the picture above you can see all the patterns with the corresponding contribution to the sum. Yellow entries are left-leaning, blue entries are generic.

Now, if we sum all the contributions, we get

Beautiful. Before going to the proof of Tokuyama’s formula, let’s have a look at its multiple corollaries.

**Corollary**. (Gelfand’s parametrization) Let be a partition of length . Then

**Sketch of the proof**. Set , then non-zero terms in the sum are exactly strict Gelfand-Tsetlin patterns with the top row with no right-leaning terms. They are in bijection with all Gelfand-Tsetlin patterns with the top row . QED.

**Corollary**. (Weyl’s Character Formula). Let be a partition of length . Let be functions from the bialternant formula of Jacobi. Then

**Sketch of the proof**. Set , then non-zero terms in the sum are exactly strict Gelfand-Tsetlin patterns with the top row with no generic terms. They are in bijection with the permutation group . The sum becomes the deformed Vandermonde determinant since equals to the number of inversions in . QED.

In a similar fashion, by setting , one can prove Stanley’s formula for expressing Hall-Littlewood polynomial as a sum over strict Gelfand-Tsetlin patterns. See the details in Section 3.3. Another connection is with alternating sign matrices: the sum in Tokuyama’s formula can be interpreted as a sum over alternating sign matrices which gives a generalization of the Weyl’s Character Formula where the sum is taken just over permutations (which are a subset of the alternating sign matrices). See Section 3.4. for details.

The original proof in terms of characters of tensor products of highest weight representations by Tokuyama is a little hard to read. A better proof can be given in terms of solvable lattice models, but I will show the following short proof by Daniel Bump which is expressed in a more familiar language.

Denote . Note that in a strict GT pattern each row is strictly decreasing, and so it can be written as for some partition of length . Recall that partition of length interleaves partition of length if

Note that each row in a GT pattern interleaves the next row. In a strict GT pattern every row interleaves above it.

**Idea of the proof**. Denote the left and right sides of Tokuyama’s Formula by

We will show the following branching rules for and :

where partitions are such that interleaves with , and interleaves with ; also, is a vertical strip. Then we show that , and by induction we are done.

**Proof**. We start with . We consider the contribution of patterns with top row and second row . Since interleaves we have , or . Removing the top row from produces a pattern with the top row and collecting these produces with a coefficient equal to . Thus, we proved the branching rules for .

Next, consider . Recall the classical branching rules for the Schur polynomial:

where partitions of length interleaves with . We also have

where is the -th elementary symmetric polynomial. Then, replace and in to get that equals

Recall Pieri’s formula in the form , where each is such that is a -vertical strip, that is, or , and exactly times. Collecting everything together, we get that equals to

with as above. Next, we notice that and , and that interleaves since interleaves and is a vertical strip. Hence, we can write the branching rules for as follows:

Finally, we have to prove that

(1)

where interleaves and is a vertical strip. The conditions on can be written as and or . Hence, we can write the right side in terms of individual as follows:

Now, if , then a pattern has a left-leaning entry in the -th position; if , a pattern has a right-leaning entry; and in the remaining cases the entry is generic. In the left-leaning case, must be , and in the right-leaning case, must be . In the generic case, can be either or . Thus,

It proves (1), and we are done. QED.

Tokuyama’s formula is related to the representation theory of -adic groups as follows. Let , where is a local non-archimedean field with the uniformizer and the cardinality of the residue field . By Casselman-Shalika formula, the value of the spherical Whittaker function on the diagonal element is given by

which up to a constant equals the right side of Tokuyama’s formula when . In other words, Tokuyama’s formula can be interpreted as a combinatorial evaluation of the spherical Whittaker function. Now, the Casselman-Shalika formula holds for other groups beside evaluating the spherical Whittaker function as the product of the deformed Weyl denominator and the corresponding character.

For example, for the symplectic group , the formula, up to a constant, takes the form

A natural question is whether the formula has a combinatorial evaluation in terms of GT-patterns for the symplectic (and other) groups. For the symplectic group, the answer is affirmative and is achieved by Dmitriy Ivanov using the solvable lattice models. No short proof like the one presented above is known.

For the orthogonal groups the answer is still unknown. In other words, it is unknown how to represent

as a weighted sum over orthogonal GT patterns.

Note that the idea of the short proof above doesn’t work (at least I tried and failed!) for symplectic or orthogonal groups: both the branching rules and Pieri’s rules become not multiplicity-free within the type.