---
title: Radon Hypergeometric Function (Radon HGF)
url: https://www.emergentmind.com/topics/radon-hypergeometric-function-radon-hgf
type: topic
---

# Radon Hypergeometric Function (Radon HGF)

Searching arXiv for recent papers on Radon hypergeometric functions and related Grassmannian constructions.
The Radon hypergeometric function (Radon HGF) is a multi-variable hypergeometric function on the Grassmannian \(\operatorname{Gr}(m,N)\), with \(N=rn\), defined as a Radon transform of a character of the universal covering group of a subgroup \(H_\lambda\subset GL(N)\) determined by a partition \(\lambda\) of \(n\). In Kimura’s formulation it includes confluent and non-confluent types, reduces to the Gelfand hypergeometric function when \(r=1\), and provides a unified realization of Hermitian matrix integral analogues of the Gauss hypergeometric function and its confluent family [2507.19048]. Subsequent work on contiguity relations shows that the Capelli identity and Cayley’s formula control the parameter-shift structure of Radon HGFs and yield recurrences for beta and gamma functions defined by Hermitian matrix integrals [2509.25900].

## 1. Geometric framework and Radon-transform construction

The ambient geometry is a Grassmannian double-fibration setting. With \(V=\mathbb{C}^N\), \(N=nr\), one considers
\[
M_1=\operatorname{Gr}(r,V),\qquad M_2=\operatorname{Gr}(m,V),
\]
together with the flag manifold
\[
\operatorname{Flag}(r,m,V)=\{(v_1,v_2)\mid v_1\subset v_2\subset V,\ \dim v_1=r,\ \dim v_2=m\}.
\]
The Radon transform is defined by pulling a function on \(M_1\) to the flag manifold, restricting it to a fiber \(\pi_2^{-1}(v)\simeq \operatorname{Gr}(r,m)\), and integrating along a suitable top-degree chain \(C(v)\) in that fiber. In homogeneous coordinates this becomes
\[
(\mathcal{R}f)(z)=\int f(tz)\,\tau(t),
\]
where \(z\) is a full-rank \(m\times N\) matrix representing a point of \(\operatorname{Gr}(m,N)\), \(t\in \operatorname{Mat}'(r,m)\), and \(\tau\) is a canonical top-degree form on \(\operatorname{Gr}(r,m)\) [2507.19048].

A useful matrix model identifies
\[
\operatorname{Gr}(m,V)\simeq GL(m)\backslash \operatorname{Mat}'(m,N),
\qquad
\operatorname{Gr}(r,m)\simeq GL(r)\backslash \operatorname{Mat}'(r,m).
\]
This model is the basis for the explicit formulas for the Radon HGF, for its differential operators, and for its covariance properties. It also makes clear that the external variables of the function are Grassmannian variables represented by full-rank matrices rather than local affine coordinates alone [2507.19048].

The terminology is not fully uniform across the 2025 literature. One paper works formally with Gelfand hypergeometric functions on \(\mathrm{GM}(2,N)\), but defines them as Radon transforms of characters of a maximal abelian subgroup \(H_\lambda\subset GL(N)\); in that setting the Radon-transform viewpoint is primary, and the case \(r=1\) is precisely the scalar limit of the later Radon HGF formalism [2506.11426].

## 2. The subgroup \(H_\lambda\), Jordan blocks, and characters

Let \(\lambda=(n_1,\dots,n_\ell)\) be a partition of \(n\). For each positive integer \(p\), the generalized Jordan group \(J_r(p)\subset GL(pr)\) is formed from block upper-triangular Toeplitz matrices, equivalently from the unit group of the truncated algebra \(R[w]/(w^p)\) with \(R=\operatorname{Mat}(r)\). The subgroup attached to \(\lambda\) is
\[
H_\lambda=\left\{\operatorname{diag}\big(h^{(1)},\dots,h^{(\ell)}\big)\mid h^{(j)}\in J_r(n_j)\right\}\subset GL(N),
\]
so \(H_\lambda\simeq J_r(n_1)\times\cdots\times J_r(n_\ell)\). In the non-confluent case \(\lambda=(1,\dots,1)\), one has
\[
H_{(1,\dots,1)}\simeq GL(r)^n,
\]
and for \(r=1\) this becomes the diagonal Cartan subgroup of \(GL(N)\) [2509.25900].

Characters of \(\widetilde{J_r(p)}\) are expressed using the logarithm of the unipotent part. Writing
\[
h=h_0\,\underline h,\qquad
\underline h\in J_r^\circ(p),\qquad
\log \underline h=\sum_{k=1}^{p-1}\theta_k(\underline h)\,w^k,
\]
a character has the form
\[
\chi_p(h;\alpha)
=
(\det h_0)^{\alpha_0}
\exp\!\left(\sum_{i=1}^{p-1}\alpha_i\,\operatorname{Tr}\theta_i(\underline h)\right),
\]
with \(\alpha=(\alpha_0,\dots,\alpha_{p-1})\in\mathbb{C}^p\). For \(H_\lambda\), the full character is a product over the blocks:
\[
\chi_\lambda(h;\alpha)=\prod_{k=1}^\ell \chi_{n_k}\big(h^{(k)};\alpha^{(k)}\big),
\qquad
\alpha=\big(\alpha^{(1)},\dots,\alpha^{(\ell)}\big)\in\mathbb{C}^n.
\]
In the non-confluent case this simplifies to
\[
\chi(h;\alpha)=\prod_{k=1}^n (\det h^{(k)})^{\alpha^{(k)}}.
\]
The parameters therefore record determinant exponents for non-confluent blocks and additional exponential-trace data for confluent blocks [2509.25900].

The standard assumptions are that \(\alpha_0^{(j)}\notin\mathbb{Z}\) for all \(j\), that \(\alpha_{n_j-1}^{(j)}\neq 0\) when \(n_j\ge 2\), and that
\[
\alpha_0^{(1)}+\cdots+\alpha_0^{(\ell)}=-m.
\]
The last condition ensures the correct \(GL(r)\)-weight for the integrand and is what allows the character factor to combine with the canonical form \(\tau(t)\) into a well-defined top form on \(\operatorname{Gr}(r,m)\) [2509.25900].

## 3. Definition of the Radon HGF and its analytic structure

Write \(z\in \operatorname{Mat}'(m,N)\) in blocks adapted to \(\lambda\),
\[
z=(z^{(1)},\dots,z^{(\ell)}),\qquad
z^{(j)}=(z_0^{(j)},\dots,z_{n_j-1}^{(j)}),\quad z_k^{(j)}\in\operatorname{Mat}(m,r),
\]
and define the Zariski-open set
\[
Z_\lambda=\left\{z\in \operatorname{Mat}'(m,N)\mid \operatorname{rank}z_0^{(j)}=r\ \text{for all }j\right\}.
\]
The integration space is
\[
T=\operatorname{Gr}(r,m)\simeq GL(r)\backslash \operatorname{Mat}'(r,m).
\]
If \(t=(t',t'')\) on the standard affine chart with \(\det t'\neq 0\) and \(u=(t')^{-1}t''\), the canonical top-degree form is
\[
\tau(t)=(\det t')^m\,du,
\qquad
du=\bigwedge_{i,j}du_{i,j},
\]
and it satisfies
\[
\tau(gt)=(\det g)^m\tau(t),\qquad g\in GL(r).
\]
This compensates exactly for the character weight imposed by \(\sum_j\alpha_0^{(j)}=-m\) [2507.19048].

For fixed \(z\), the branch divisors are
\[
S_z^{(j)}=\{[t]\in T\mid \det(tz_0^{(j)})=0\},
\qquad
X_z=T\setminus \bigcup_j S_z^{(j)}.
\]
The integrand may be written as
\[
\chi_\lambda(tz;\alpha)=f(t,z)\exp(g(t,z)),
\]
where
\[
f(t,z)=\prod_{j=1}^{\ell}\big(\det(tz_0^{(j)})\big)^{\alpha_0^{(j)}}
\]
determines a rank-one local system \(L_z\), and the exponential factor \(g(t,z)\) controls the irregular behavior in confluent cases through a family of supports \(\Phi_z\) in the sense of Pham. The Radon HGF is then
\[
F_\lambda(z;\alpha;C)=\int_{C(z)} \chi_\lambda(tz;\alpha)\,\tau(t),
\qquad
C(z)\in H^{\Phi_z}_{r(m-r)}(X_z;L_z).
\]
In affine coordinates this becomes
\[
F_\lambda(z;\alpha;C)
=
\int_{C(z)}
\prod_{k=1}^\ell
\big(\det(\vec uz_0^{(k)})\big)^{\alpha_0^{(k)}}
\exp\!\left(
\sum_{k=1}^\ell\sum_{i=1}^{n_k-1}
\alpha_i^{(k)}\operatorname{Tr}\theta_i(\underline{\vec uz}^{(k)})
\right)\,du,
\]
with \(\vec u=(1_r,u)\). In the non-confluent case all exponential terms disappear and one recovers the determinant-power integral familiar from Gelfand-type constructions [2507.19048].

Analytically, for a fixed sector of cycles the Radon HGF is multivalued holomorphic in \(z\) on a Zariski-open subset and meromorphic in the parameters \(\alpha\). The non-confluent branch behavior comes from determinantal divisors, whereas the confluent case introduces exponential growth and decay governed by the \(\theta_i\)-terms. This separation between algebraic monodromy and irregular behavior is one of the structural features distinguishing confluent from non-confluent Radon HGFs [2507.19048].

## 4. Differential equations, covariance, and the \(r=1\) Gelfand limit

The Radon HGF satisfies an overdetermined linear PDE system whose highest-order part expresses the image constraints of the Radon transform. For \(z=(z_{i,j})\in \operatorname{Mat}(m,N)\), let
\[
D_{i,j}=\frac{\partial}{\partial z_{i,j}},
\]
and for subsets \(I\subset [1,m]\), \(J\subset [1,N]\) with \(|I|=|J|=r+1\), define
\[
D_{I,J}=\det\bigl(D_{i_\mu,j_\nu}\bigr)_{1\le \mu,\nu\le r+1}.
\]
Then
\[
D_{I,J}F_\lambda=0
\]
for all such \(I,J\). These equations are the main equations characterizing the image of the Radon transform. They are complemented by first-order equations obtained by differentiating the covariance relations under \(H_\lambda\) and \(GL(m)\), so the full Radon hypergeometric system consists of these \((r+1)\)-st order determinantal equations together with Lie-algebraic covariance equations [2507.19048].

The covariance itself is explicit. For \(g\in GL(m)\) and \(h\in \widetilde{H_\lambda}\),
\[
F_\lambda(gz;\alpha;C)=(\det g)^{-r}F_\lambda(z;\alpha;\tilde C),
\qquad
F_\lambda(zh;\alpha;C)=\chi_\lambda(h;\alpha)\,F_\lambda(z;\alpha;C).
\]
This describes \(F_\lambda\) as a relative invariant under the \(GL(m)\times H_\lambda\)-action and is the basic source of the first-order differential relations [2507.19048].

When \(r=1\), the system reduces to the Gelfand hypergeometric system on a Grassmannian. In the \(\mathrm{GM}(2,N)\) setting, the second-order operators
\[
\square_{p,q}F
=
\bigl(\partial_{0,p}\partial_{1,q}-\partial_{1,p}\partial_{0,q}\bigr)F
=
0
\]
are presented as the part of the system that characterizes the image of the Radon transform, while additional first-order equations encode covariance under \(H_\lambda\) and \(GL(2)\). Restricting to a distinguished slice yields a hyperbolic subsystem, and the corresponding Gelfand HGF furnishes solutions of the 2-dimensional Toda-Hirota equation; the paper states that the contiguity relations play an important role in constructing the Laplace sequence and Bäcklund transformations [2506.11426].

A common misconception is to treat the Radon HGF merely as a reformulation of classical hypergeometric integrals. The differential-system viewpoint shows that its defining content is not only the integral representation but also the Grassmannian Radon-image condition encoded by the determinantal PDEs. In the scalar case this specializes to the classical Gelfand framework; for \(r>1\) it produces a higher-rank system rather than a direct tensor product of one-variable equations [2507.19048].

## 5. Contiguity relations, Capelli identity, and Cayley’s formula

The 2025 contiguity theory identifies exact parameter-shift operators for the Radon HGF. In the non-confluent case \(\lambda=(1,\dots,1)\), writing
\[
z=(z^{(1)},\dots,z^{(n)}),\qquad z^{(j)}\in \operatorname{Mat}(m,r),
\]
one introduces first-order operators
\[
L_{p,q}^{(i,j)}={}^t z_p^{(i)}\,\partial_q^{(j)},
\]
and the order-\(r\) operator
\[
L^{(i,j)}=\det\bigl(L_{p,q}^{(i,j)}\bigr)=\det\bigl({}^t z^{(i)}\,\partial^{(j)}\bigr).
\]
The contiguity relation is
\[
L^{(i,j)}F(z;\alpha)
=
b(\alpha^{(j)})\,
F\bigl(z;\alpha+e^{(i)}-e^{(j)}\bigr),
\qquad
b(s)=s(s+1)\cdots(s+r-1).
\]
Thus a differential operator of order \(r\) increments one parameter by \(1\) and decrements another by \(1\), with scalar factor given by the \(b\)-function [2509.25900].

For general \(\lambda\), the operator becomes
\[
L^{(i,j)}=\det\bigl({}^t z_0^{(i)}\,\partial_{n_j-1}^{(j)}\bigr).
\]
If \(n_j=1\), the same \(b\)-function appears:
\[
L^{(i,j)}F_\lambda(z;\alpha)
=
\alpha_0^{(j)}(\alpha_0^{(j)}+1)\cdots(\alpha_0^{(j)}+r-1)\,
F_\lambda\bigl(z;\alpha+e^{(i)}-e^{(j)}\bigr).
\]
If \(n_j\ge 2\), the confluent formula is
\[
L^{(i,j)}F_\lambda(z;\alpha)
=
\bigl(\alpha_{n_j-1}^{(j)}\bigr)^r
F_\lambda\bigl(z;\alpha+e^{(i)}-e^{(j)}\bigr).
\]
The shift always acts on the leading parameters by \(\alpha_0^{(i)}\mapsto \alpha_0^{(i)}+1\) and \(\alpha_0^{(j)}\mapsto \alpha_0^{(j)}-1\), while the scalar factor depends on whether the target block is non-confluent or confluent [2509.25900].

These formulas are derived from the Capelli identity
\[
\det\big(E'_{ij}+(r-j)\delta_{ij}\big)=\det(x_{ij})\,\det(\partial_{ij})
\]
and Cayley’s formula
\[
\det(\partial_{ij})\,(\det x)^s=b(s)(\det x)^{s-1},
\qquad
b(s)=s(s+1)\cdots(s+r-1).
\]
The role of the Capelli identity is to convert invariant differential operators into determinant-lowering operators, while Cayley’s formula supplies the scalar factor that appears in the shifted integral [2509.25900].

The same machinery yields contiguity relations for Hermitian matrix beta and gamma functions:
\[
B_r(a+1,b)
=
\frac{a(a-1)\cdots(a-r+1)}{(a+b)(a+b-1)\cdots(a+b-r+1)}\,B_r(a,b),
\]
\[
B_r(a,b+1)
=
\frac{b(b-1)\cdots(b-r+1)}{(a+b)(a+b-1)\cdots(a+b-r+1)}\,B_r(a,b),
\]
\[
\Gamma_r(a+1)=a(a-1)\cdots(a-r+1)\,\Gamma_r(a).
\]
These are obtained by realizing \(B_r(a,b)\) and \(\Gamma_r(a)\) as special Radon HGFs on \(\operatorname{Gr}(2r,3r)\) [2509.25900].

## 6. Weyl-group analogue and symmetry structure

A further structural layer is the Weyl-group analogue
\[
W_\lambda=N_G(H_\lambda)/H_\lambda,
\qquad G=GL(N),
\]
which governs symmetries of the Radon HGF. If
\[
\lambda=(\overbrace{n_1,\dots,n_1}^{p_1},\dots,\overbrace{n_s,\dots,n_s}^{p_s}),
\qquad n_1>\cdots>n_s>0,
\]
then
\[
W_\lambda
\simeq
\prod_{i=1}^s W_r(n_i)^{p_i}\rtimes P_i,
\]
where \(P_i\simeq \mathfrak{S}_{p_i}\) is the permutation group of equal-size Jordan blocks, and \(W_r(n_i)\) is the continuous factor identified with automorphisms of the truncated polynomial algebra \(\mathbb{C}[T]/(T^{n_i})\). In the non-confluent case \(\lambda=(1,\dots,1)\), one recovers
\[
W_{(1,\dots,1)}\simeq \mathfrak{S}_n
\]
[2510.23120].

The symmetry acts simultaneously on the Grassmannian variables and on the parameter vector. Writing the induced linear action on parameters as \(\rho(g)\), the main symmetry statement is
\[
F(zg,\alpha;C)=F\bigl(z,\alpha\cdot {}^t\rho(g);C\bigr),
\qquad g\in W_\lambda,
\]
and for the finite permutation part \(P_\sigma\in P\),
\[
F(zP_\sigma,\alpha;C)=F(z,\alpha;C).
\]
The continuous part therefore generates nontrivial linear transformations of the parameters, whereas the finite part produces genuine symmetry identities of the function [2510.23120].

One consequence is parameter reduction. The continuous subgroup can normalize higher confluent parameters so that, with the leading exponents fixed, each block may be reduced to a form
\[
(\alpha_0^{(i,k)},0,\dots,0,1).
\]
This explains why several matrix analogues in the confluent family carry fewer essential external parameters than the raw character formula might suggest. In the \(\operatorname{Gr}(2r,4r)\) applications, the same symmetry yields transformation formulas analogous to part of Kummer’s 24 solutions for the Gauss hypergeometric function and to Kummer’s first transformation for the confluent case [2510.23120].

## 7. Specializations, classical analogues, and related developments

A notable feature of the theory is that the Radon HGF organizes several scalar and matrix-valued special functions into a single Grassmannian framework. The following specializations are stated explicitly.

| Partition \(\lambda\) | Grassmannian | Specialization |
|---|---:|---|
| \((1,1,1)\) | \(\operatorname{Gr}(2r,3r)\) | Hermitian matrix beta function |
| \((2,1)\) | \(\operatorname{Gr}(2r,3r)\) | Hermitian matrix gamma function |
| \((3)\) | \(\operatorname{Gr}(2r,3r)\) | Gaussian matrix integral |
| \((1,1,1,1)\) | \(\operatorname{Gr}(2r,4r)\) | Gauss matrix HGF |
| \((2,1,1)\) | \(\operatorname{Gr}(2r,4r)\) | Kummer analogue |
| \((2,2)\) | \(\operatorname{Gr}(2r,4r)\) | Bessel analogue |
| \((3,1)\) | \(\operatorname{Gr}(2r,4r)\) | Hermite–Weber analogue |
| \((4)\) | \(\operatorname{Gr}(2r,4r)\) | Airy analogue |

For \(\operatorname{Gr}(2r,4r)\), the paper states that the Hermitian matrix integral analogues of the Gauss HGF and its confluent family are obtained in a unified manner from the partitions \((1,1,1,1)\), \((2,1,1)\), \((2,2)\), \((3,1)\), and \((4)\). This is one of the main points of the definition paper and places matrix hypergeometric functions of Faraut–Korányi and Muirhead inside the same Radon-transform construction [2507.19048].

The symmetry paper goes further by deriving matrix analogues of classical transformation formulas. For the Gauss analogue it obtains
\[
{}_2F_1(a,b,c;X)
=
\bigl(\det(1_r-X)\bigr)^{-b}\,
{}_2F_1\!\left(c-a,b,c;X(X-1_r)^{-1}\right),
\]
which is identified as the matrix analogue of a classical Kummer transformation. For the Kummer analogue it derives
\[
{}_1F_1(a,c;X)=\mathrm{etr}(X)\,{}_1F_1(c-a,c;-X).
\]
These formulas arise from the finite part of the Weyl-group analogue acting on normal forms in \(\operatorname{Gr}(2r,4r)\) [2510.23120].

At \(r=1\), the same general framework specializes to Gelfand hypergeometric functions on Grassmannians, including confluent and non-confluent types. On \(\mathrm{GM}(2,4)\), the partitions \((1,1,1,1)\), \((2,1,1)\), \((2,2)\), \((3,1)\), and \((4)\) correspond respectively to Gauss, Kummer, Bessel, Hermite–Weber, and Airy functions. The integrable-systems development shows that, on the slice \(X\subset Z_\lambda\), the Gelfand HGF enters the 2-dimensional Toda-Hirota equation through its contiguity operators and the Laplace sequence of the associated hyperbolic operators. This suggests a broader role for Radon-type Grassmannian hypergeometric functions in the interface between integral geometry, holonomic PDEs, and integrable systems [2506.11426].

Source: https://www.emergentmind.com/topics/radon-hypergeometric-function-radon-hgf