---
title: Young Generating Function
url: https://www.emergentmind.com/topics/young-generating-function
type: topic
---

# Young Generating Function

Searching arXiv for the cited paper and closely related work on Young generating functions and nearby terminology.
A Young generating function, in the representation-theoretic sense introduced in "Global fluctuations for standard Young tableaux," is the formal power series
\[
A_\rho(x_1,x_2,\ldots)=M_\rho(U_\infty)
\]
attached to a probability measure \(\rho\) on \(\mathbb Y_n\), the set of partitions of \(n\). It is designed to encode asymptotic information about random partitions and random standard Young tableaux through the logarithmic derivatives of \(A_\rho\) at the origin. In this framework, \(A_\rho\) plays the role of a characteristic function for random partitions: its first derivatives determine limiting transition measures and laws of large numbers, its second derivatives determine covariances and central limit theorems, and its multilevel extension leads to two-dimensional fluctuation results for height functions, whose limits are identified as a conditioned Gaussian Free Field [2507.18601].

## 1. Definition and representation-theoretic construction

The ambient combinatorial objects are Young diagrams, viewed in Russian notation. If \(\lambda\) is a partition, its boundary is a piecewise-linear curve \(\omega=\omega_\lambda\) embedded in the upper half-plane by the change of variables
\[
x=s-r,\qquad y=r+s,
\]
where \(r\) is the row index and \(s\) is the column index. The diagram is then viewed as a continuous diagram
\[
\sigma[\lambda](t)=\frac{\omega_\lambda(t)-|t|}{2}.
\]

For a probability measure \(\rho\) on \(\mathbb Y_n\), the construction begins with the associated character on the inductive limit \(S_\infty\) of symmetric groups,
\[
M_\rho(\cdot)=\sum_{\lambda\in\mathbb Y_n}\rho(\lambda)\frac{\chi_\lambda(\cdot)}{\dim(\lambda)},
\]
extended by zero off \(S_n\subset S_\infty\). One then chooses disjoint permutations \(\sigma[(k)^i]\in S_\infty\) that are products of \(i\) disjoint \(k\)-cycles, for \(i,k\ge 1\), and defines
\[
U_\infty=\prod_{k=1}^\infty \left(1+\sum_{i=1}^\infty n^{i(k-1)/2}\sigma[(k)^i]\frac{x_k^i}{i!}\right)
\]
as an element of \(R[S_\infty][\vec x]\), understood as the inverse limit of truncations at finite \(k\). The Young generating function is then
\[
A_\rho(x_1,x_2,\ldots):=M_\rho(U_\infty)\in R[\vec x].
\]
Since \(A_\rho(0,0,\ldots)=1\), the formal series \(\ln A_\rho\) is well-defined. Although \(U_\infty\) depends on the choice of disjoint representatives \(\sigma[(k)^i]\), \(A_\rho\) does not, because \(M_\rho\) is conjugation-invariant [2507.18601].

This construction is explicitly representation-theoretic. It packages the character values of the symmetric groups into a formal object that can be differentiated, and those derivatives recover asymptotic data of random partitions at the \(\sqrt n\) scale. The inverse-limit algebra \(R[\vec x]\) and the group ring \(R[S_\infty][\vec x]\) are therefore not auxiliary formalities; they are the setting in which characters, cumulants, and asymptotic moments are brought into a single calculus.

## 2. Transition measures, logarithmic derivatives, and free-probability content

The key analytic object attached to a partition \(\lambda\) is Kerov’s transition measure \(m_K[\lambda]\). For a continuous diagram \(\omega\) on \([a,b]\), it is characterized by the Markov–Krein correspondence
\[
\int_a^b (z-t)^{-1}\,dm_K[\omega](t)
=
z^{-1}\exp\int_a^b (t-z)^{-1}\,d\!\left(\frac{\omega(t)-|t|}{2}\right),
\qquad z\in\mathbb C\setminus[a,b].
\]
For discrete \(\lambda\), one has
\[
m_K[\lambda]=\sum_i \mu_i\delta_{x_i},
\]
where \(x_i\) are the minima of \(\omega_\lambda\) and \(\mu_i\) are the Plancherel growth weights of adding a box at \(x_i\).

The logarithmic derivatives of \(A_\rho\) recover scaled moments and cumulants of \(m_K[\lambda]\). First derivatives of \(\ln A_\rho\) along the variables \(x_k\) yield numbers \(c_k\), and these are assembled into
\[
F_\rho(z)=\sum_{k\ge 1} c_k z^{k-1}.
\]
Second derivatives yield numbers \(d_{i,j}\), assembled into
\[
Q_\rho(z,w)=\sum_{i,j\ge 1} d_{i,j} z^i w^j.
\]
Higher derivatives govern higher-order cumulants and Gaussianity criteria.

The paper identifies these quantities with the free-probability description of asymptotic transition measures. The coefficients \(c_k\) are the free cumulants of the rescaled limit of \(m_K[\lambda]\), and if \(C_\rho\) denotes the Stieltjes transform of the limiting transition measure, then
\[
zF_\rho(z)=C_\rho^{-1}(z)-\frac{1}{z}.
\]
In this sense, \(F_\rho\) is the Voiculescu \(R\)-transform in the variables used by the paper, while \(Q_\rho\) is the covariance-generating series controlling fluctuations [2507.18601].

This organization is the conceptual core of the theory. A measure on partitions is replaced by a formal series; the logarithm of that series is differentiated; those derivatives become free cumulants and covariance data; and the asymptotic geometry of Young diagrams is then reconstructed from \(F_\rho\) and \(Q_\rho\).

## 3. Law of large numbers, central limit theorem, and multilevel asymptotics

The asymptotic theory is formulated directly in terms of the derivatives of \(\ln A_{\rho_n}\). A sequence \(\rho_n\) on \(\mathbb Y_n\) is called **LLN-appropriate** if there exist numbers \(c_i\) such that, for each \(i\ge 1\),
\[
\lim_{n\to\infty}\partial_i\ln(A_{\rho_n})(\vec x)\big|_{\vec x=0}=c_i,
\]
and, for each \(r\ge 2\) and \(i_1,\ldots,i_r\),
\[
\lim_{n\to\infty}\partial_{i_1}\cdots\partial_{i_r}\ln(A_{\rho_n})(\vec x)\big|_{\vec x=0}=0.
\]
If
\[
X_k=n^{-k/2}\int_{\mathbb R}x^k\,m_K[\lambda](dx),
\]
then \(\rho_n\) satisfies a law of large numbers when \(\lim E[X_k]=a_k\) and all higher cumulants vanish asymptotically. Theorem 2.3 states that these two conditions are equivalent, and gives the limit moments as
\[
a_k=[z^{-1}]\frac{1}{k+1}\left(z^{-1}+zF_\rho(z)\right)^{k+1}.
\]

The central-limit regime is obtained by strengthening the derivative hypotheses. A sequence is **CLT-appropriate** if there exist \(c_i\) and \(d_{i,j}\) such that
\[
\lim_{n\to\infty}\partial_i\ln(A_{\rho_n})(0)=c_i,\qquad
\lim_{n\to\infty} n\,\partial_i\partial_j\ln(A_{\rho_n})(0)=d_{i,j},
\]
and
\[
\lim_{n\to\infty} n^{r/2}\partial_{i_1}\cdots\partial_{i_r}\ln(A_{\rho_n})(0)=0
\quad\text{for all }r\ge 3.
\]
If \(\rho_n\) satisfies a CLT, then
\[
\lim_{n\to\infty} n\,\kappa(X_k,X_{k'})=b_{k,k'},
\]
and higher scaled cumulants vanish. Theorem 2.6 states that CLT-appropriateness is equivalent to this central limit theorem, with
\[
a_k=[z^{-1}]\frac{1}{k+1}\left(z^{-1}+zF_\rho(z)\right)^{k+1},
\]
and
\[
\begin{aligned}
b_{k,k'}=[z^{-1}w^{-1}]&\Big\{
(z^{-1}+zF_\rho(z))^k (w^{-1}+wF_\rho(w))^{k'} \\
&\times\Big[
Q_\rho(z,w)-zw\,\partial_z\partial_w\Big(
zwF_\rho(z)F_\rho(w)
+\ln\Big(1-zw\,\frac{zF_\rho(z)-wF_\rho(w)}{z-w}\Big)
\Big)\Big]\Big\}.
\end{aligned}
\]

The theory is intrinsically multilevel. If increasing partitions are sampled via the branching rule
\[
p(\lambda^1,\ldots,\lambda^s)=\rho_n(\lambda^s)\prod_{t=1}^{s-1}p(\lambda^{t+1}\to\lambda^t),
\qquad
p(\mu\to\lambda)=\frac{\dim(\mu)\,\dim(\lambda\backslash\mu)}{\dim(\lambda)},
\]
then Theorem 3.1 gives a multilevel law of large numbers. For \(\alpha\in(0,1)\), \(n_\alpha=\lfloor \alpha n\rfloor\), and
\[
X_k^\alpha=n^{-k/2}\int x^k\,dm_K[\lambda^\alpha],
\]
one has
\[
\lim_{n\to\infty}E[X_k^\alpha]=a_k^\alpha,
\qquad
a_k^\alpha=[z^{-1}]\,\alpha^{-1}\frac{1}{k+1}\left(\alpha z^{-1}+zF_\rho(z)\right)^{k+1}.
\]
Theorem 3.2 gives the corresponding covariance formula,
\[
\begin{aligned}
b_{k,k'}^{\alpha,\alpha'}=[z^{-1}w^{-1}]&\Big\{
\frac{(\alpha z^{-1}+zF_\rho(z))^k(\alpha' w^{-1}+wF_\rho(w))^{k'}}{\alpha\alpha'} \\
&\times\Big[
Q_\rho(z,w)-zw\,\partial_z\partial_w\Big(
\frac{zw}{\max(\alpha,\alpha')}F_\rho(z)F_\rho(w) \\
&\hspace{4em}+ \ln\Big(1-zw\,\frac{zF_\rho(z)-wF_\rho(w)}{\max(\alpha,\alpha')(z-w)}\Big)
\Big)\Big]\Big\}.
\end{aligned}
\]
In particular, sublinear slices \(\alpha\to 0\) recover the semicircle limit and classical CLT [2507.18601].

## 4. Height functions and the conditioned Gaussian Free Field

For a standard Young tableau of shape \(\lambda\), the height function is
\[
H(x,t):=\frac{\lambda^{\lfloor t\rfloor}(x)-|x|}{2},
\]
the length of the diagonal \(x\) at time \(t\). In the multilevel setting, Theorem 3.1 implies the existence of a limiting surface
\[
H^\infty(x,t)=\lim_{n\to\infty} n^{-1/2}H(\sqrt n\,x,tn).
\]

To study fluctuations, the paper translates between transition measures, continuous diagrams, and co-transition measures, and linearizes Markov–Krein. For the continuous-diagram moments
\[
Y_k:=\int x^{k-1}\,d\sigma[\lambda],
\]
Lemma 4.1 expresses both \(\lim E[Y_k^\alpha]\) and \(\lim n\,\mathrm{Cov}(Y_k^\alpha,Y_{k'}^{\alpha'})\) in terms of \(F_\rho\) and \(Q_\rho\).

The limiting two-dimensional fluctuation field is not the unconditioned Gaussian Free Field. On the upper half-plane \(\mathfrak h=\{z\in\mathbb C:\operatorname{Im}z>0\}\), the covariance kernel is
\[
G(z,w)= -\frac{1}{2\pi}\ln\left|\frac{z-w}{z-\bar w}\right|
+\frac{\min(t(z),t(w))}{\pi}\operatorname{Im}(1/z)\operatorname{Im}(1/w),
\]
where \(t(z)\) is the multilevel parameter attached to \(z\) via a model-specific conformal map. The resulting Gaussian field \(\mathfrak c\) is identified as a Gaussian free field \(\mathfrak G\) conditioned by linear constraints along the level curves
\[
C_\alpha=\{z:t(z)=\alpha\}.
\]

Proposition 4.2 makes this precise. If \(K\) is the closed linear span of integrals of \(\mathfrak G\) against smooth test functions over the curves \(C_\alpha\), then
\[
\mathfrak c=\mathfrak G-P[\mathfrak G],
\]
where \(P\) is the orthogonal projection onto \(K\). The covariance of \(\mathfrak c\) is
\[
\operatorname{Cov}(\mathfrak c)=\operatorname{Cov}(\mathfrak G)-\operatorname{Cov}(P[\mathfrak G]),
\]
and the second term is exactly the conditioning term in \(G(z,w)\) [2507.18601].

The conditioned nature of the field is structurally important. It reflects the multilevel constraints inherited from Gelfand–Tsetlin branching, rather than a free Dirichlet field without extra linear conditions.

## 5. Principal models, examples, and technical machinery

The formalism is applied to three main classes of models, each with an explicit domain map and the same limiting fluctuation field.

| Model | Defining data | Limiting fluctuation statement |
|---|---|---|
| Plancherel growth process | \(p(\mu,\lambda)=\dim(\mu)/(|\mu|\dim(\lambda))\) if \(\mu\nearrow\lambda\) | \(\sqrt\pi[H(\sqrt n x,nt)-EH(\sqrt n x,nt)]\Rightarrow \mathfrak c(x,t)\) |
| Extreme characters of \(S_\infty\) | Thoma parameters \((\alpha,\beta)\) | \(\sqrt\pi[H(\sqrt n x,nt)-EH(\sqrt n x,nt)]\Rightarrow \mathfrak c(x,t)\) |
| Random SYT of fixed shape | \(\lambda^n\to \omega\) | \(\sqrt\pi[H(\sqrt n x,nt)-EH(\sqrt n x,nt)]\Rightarrow \mathfrak c(x,t)\) |

For Plancherel measure on \(\mathbb Y_n\), \(\rho(\lambda)=\dim(\lambda)^2/n!\), the associated character is the trivial character of \(S_\infty\), so
\[
A_\rho(x_1,x_2,\ldots)=\exp(x_1).
\]
Hence \(F_\rho(z)=1\), \(Q_\rho=0\), and the limit moments \(a_k\) are those of the semicircle law. The VKLS curve appears as the limit profile \(\omega(t)\), with
\[
\omega(t)=\frac{2}{\pi}\big(t\arcsin(t/2)+\sqrt{4-t^2}\big)\quad\text{for }|t|\le 2,
\qquad
\omega(t)=|t|\quad\text{otherwise}.
\]
In the growth process, the limiting liquid region is mapped to \(\mathfrak h\) by
\[
\Omega(x,y)=x/2+i\sqrt{y-x^2/4},
\]
and the level sets \(t=\alpha\) map to semicircles \(|z|^2=\alpha\).

For measures induced by extreme characters of \(S_\infty\), Thoma parameters \((\alpha,\beta)\) are scaled so that
\[
\frac1n\sum_{i\ge 1}\alpha_i(n)^{-1}\delta_{\sqrt n\,\alpha_i(n)}\Rightarrow \mathcal A,
\qquad
\frac1n\sum_{i\ge 1}\beta_i(n)^{-1}\delta_{\sqrt n\,\beta_i(n)}\Rightarrow \mathcal B,
\]
with \(\gamma(n)=1-\sum \alpha_i(n)-\sum \beta_i(n)\to \bar\gamma\). Then \(\rho_n\) is CLT-appropriate, and Proposition 4.4 gives
\[
\begin{aligned}
zF(z)=&\ z^{-2}C_{\mathcal A}(1/z)+z^{-2}C_{\mathcal B}(-1/z)
-\left(\int x\,\mathcal A(dx)-\int x\,\mathcal B(dx)\right) \\
&+\left(1-\int x^2\,\mathcal A(dx)-\int x^2\,\mathcal B(dx)\right)z.
\end{aligned}
\]
The inverse map \(z\mapsto (y_F,s_F)\) solves
\[
y_F=\alpha/z+zF(z),\qquad s_F=\alpha,
\]
and is a diffeomorphism onto its image. The Schur–Weyl specialization is obtained by taking \(n/D_n^2\to c^2\), \(\alpha_i(n)=1/D_n\) for \(i\le D_n\), \(\beta_i(n)=0\), \(\gamma(n)=0\), which gives
\[
F(z)=1+\sum_{k\ge 2} c^{k-1}z^{k-1}=1+\frac{cz}{1-cz}.
\]

For random standard Young tableaux of fixed shape, a deterministic sequence \(\lambda^n\) converges to a continuous diagram \(\omega\), and the domain map is determined by the Stieltjes transform \(C(z)\) of \(m_K[\omega]\). Proposition 4.6 states that for any \(y\in\mathbb R\) and \(\alpha\in[0,1]\), the equation
\[
1/z+(\alpha-1)/C(1/z)=y
\]
has at most one root \(z\in\mathfrak h\); the map \(z\mapsto (y_F(z),\hat s_F(z))\), with \(\hat s_F(z)=1/(1-\alpha)\), defines a diffeomorphism onto its image. In the square-shape example,
\[
C(z)=\frac{z}{z^2-1},
\]
and the level-line Stieltjes transforms are
\[
C_\alpha(z)=\frac{(2\alpha-1)z+\sqrt{z^2+4\alpha^2-4\alpha}}{2\alpha(z^2-1)}.
\]

The technical proof apparatus is built from several algebraic and asymptotic ingredients. The inverse-limit algebra \(R[\vec x]\) makes \(\ln A_\rho\) and \(\exp\) available at the formal level. Jucys–Murphy elements enter through central operators
\[
D_k=(n+1)^{-1}\operatorname{tr}\Gamma(n)^k,
\]
where \(\Gamma(n)\) is the adjacency matrix built from transpositions, and for characters \(M_\rho\),
\[
E\!\left[\int x^k\,dm_K[\lambda]\right]=[\mathfrak D_k M_\rho](e),
\qquad
\mathfrak D_k\chi(g)=\chi(D_k g).
\]
Theorem 5.1 expands \(D_k\) into central class sums with leading coefficients involving noncrossing partitions, while Theorem 5.2 and Lemmas 5.3–5.5 develop the Gelfand–Tsetlin algebra needed for multilevel cumulants. The resulting framework generalizes Schur generating functions of Bufetov–Gorin to integer partitions, recovers and extends free-probability laws for Plancherel-type measures, matches Kerov’s central limit theorem in the Plancherel case, and is distinct from determinantal approaches that require Poissonization and do not apply to these models [2507.18601].

## 6. Other meanings of the term in the literature

The phrase **Young generating function** is not unique to the representation-theoretic formal series \(A_\rho\). In the recent literature it also appears in several unrelated combinatorial senses.

In "Simple Generating Functions for Certain Young Tableaux with Periodic Walls," the relevant objects are ordinary generating functions for tableaux with horizontal walls. For the periodic building \(B_m^n\) of shape \(2\times(mn)\), the counting series
\[
\overline F_m(x)=\sum_{n\ge 0}\overline f_m(n)x^n
\]
satisfies
\[
\overline F_m(x)=\prod_{k=1}^{m}C(\xi^k x^{1/m})
=\exp\left(\sum_{n\ge 1}\binom{2mn-1}{mn-1}\frac{x^n}{n}\right),
\]
where \(C(x)\) is the Catalan generating function and \(\xi=e^{2\pi i/m}\) [2401.14627].

In "On the generating function for intervals in Young’s lattice," the central object is the multivariate series
\[
Q_k(x_1,\ldots,x_k,y)=\sum_{\lambda\in\Lambda(k)}P_\lambda(y)\,x^\lambda,
\]
where \(P_\lambda(y)=\sum_{\mu\in[\emptyset,\lambda]} y^{|\mu|}\). The main theorem states that \(Q_k\) satisfies a rational recursion and is therefore a rational function in \(x_1,\ldots,x_k,y\) [2107.09149].

In "Generating Functions for Inverted Semistandard Young Tableaux and Generalized Ballot Numbers," the basic generating function is
\[
G_{\lambda,\mu}(q)=\sum_{k\ge 0}T_{\lambda,\mu}(k)q^k,
\]
counting \(k\)-inverted semistandard Young tableaux of shape \(\lambda\) and content \(\mu\). A fixed-standardization generating function
\[
x_T(q)=\sum_{k\ge 0}|S_T^k(\lambda,\mu)|q^k
\]
is expressed as a product of \(q\)-numbers determined by the statistics \(dp\) and \(dp^\ast\) [1606.04869].

In "A new \(q\)-Selberg integral, Schur functions, and Young books," the generating function is the major-index enumerator
\[
G(q)=\sum_{B\in YB(n;r,s)} q^{\operatorname{maj}(B)}
\]
for Young books, and it is represented both by a Jackson integral and by a Schur-function expansion [1412.7914].

In "Derivatives, Eulerian polynomials and the \(g\)-indexes of Young tableaux," the expression “Young generating function” is used for tableau-weighted generating polynomials such as
\[
A_n(x)=\sum_{T\in \mathrm{SYT}(n)} G_T\,x^{n+1-\ell(\lambda(T))}
\]
and
\[
C_n(x)=\sum_{T\in \mathrm{SYT}(n)} G_T\,\lambda(T)!\,x^{n+1-2\ell(\lambda(T))},
\]
where the weights are \(g\)-indexes [2006.14064].

A useful way to read the terminology, therefore, is as context-dependent. In asymptotic representation theory, the Young generating function is the formal characteristic-function analogue \(A_\rho\). In enumerative combinatorics, the same phrase may denote an ordinary generating function, a \(q\)-generating function, or a tableau-weighted polynomial attached to a different class of Young-type objects.

Source: https://www.emergentmind.com/topics/young-generating-function