---
title: 'Two-Variable Jordan Block: Concepts & Applications'
url: https://www.emergentmind.com/topics/two-variable-jordan-block
type: topic
---

# Two-Variable Jordan Block: Concepts & Applications

Searching arXiv for the provided topic and related papers to ground the article in the literature.
A two-variable Jordan block is not a single universally fixed object, but a family of closely related constructions that recur across operator theory, hydrodynamic-type PDEs, matrix theory, commutative algebra, and recent applied work. In the bidisk Hardy-space setting, it is precisely a doubly commuting quotient module of \(H^2(\mathbb{D}^2)\), and hence a tensor product of one-variable Jordan blocks. In parabolic systems of hydrodynamic type, it denotes a \(2\times 2\) non-diagonalizable coefficient matrix with a repeated eigenvalue and one generalized eigenvector. In finite-dimensional linear algebra, it appears as the canonical block \(J_2(\lambda)\), as a constituent of \(J_r\otimes J_s\), and as the organizing object in centralizer problems. Across these settings, the recurrent structure is non-semisimplicity: a repeated eigenvalue together with a nontrivial Jordan chain, or, in the bidisk case, a tensor-product analogue of that chain structure [1303.1041] [2212.01413] [1704.05837].

## 1. Terminological scope and common structural pattern

In the literature surveyed here, the phrase “two-variable Jordan block” is used in several mathematically distinct, but structurally related, ways.

| Context | Defining object | Salient feature |
|---|---|---|
| Hardy-module theory | \(\mathcal{Q}\subset H^2(\mathbb{D}^2)\) with doubly commuting compressed shifts | Tensor product \(\mathcal{Q}_{\Theta_1}\otimes \mathcal{Q}_{\Theta_2}\) |
| Hydrodynamic-type systems | \(A(u)=\begin{bmatrix}\lambda(u)&\mu(u)\\0&\lambda(u)\end{bmatrix}\) | Repeated speed, one eigenvector, one generalized eigenvector |
| Finite-dimensional matrix theory | \(J_2(\lambda)\), \(J_r\otimes J_s\), or centralizers of \(J_n\oplus J_n\) | Jordan chains, partitions, commuting nilpotents |
| Artinian algebra | Jordan type of \(m_\ell\) on \(k[x,y]/I\) | Block decomposition of multiplication by a linear form |
| Recent applications | Complex defective block in positional encoding; non-expansive number systems | Coupled oscillatory-polynomial modes or combinatorial length formulas |

This distribution of meanings suggests a stable conceptual core: Jordan-block behavior records the failure of diagonalizability, while the surrounding category determines whether that failure is encoded by quotient modules, PDE coefficients, nilpotent orbits, multiplication maps, or representation formulas. The most systematic two-variable classification in the operator-theoretic sense is the bidisk theorem for \(H^2(\mathbb{D}^2)\), where “two-variable Jordan block” is a precise term rather than a loose analogy [1303.1041].

## 2. Bidisk Hardy-space formulation

For the Hardy space on the bidisk,
\[
\|f\|_{H^2(\mathbb{D}^2)}^2=\sup_{0<r<1}\int_{\mathbb{T}^2}|f(r\zeta)|^2\,d\sigma(\zeta)<\infty,
\]
the coordinate multipliers \(M_{z_1}\) and \(M_{z_2}\) make \(H^2(\mathbb{D}^2)\) into a Hilbert module over \(\mathbb{C}[z_1,z_2]\). There is a canonical unitary module isomorphism
\[
U:\ H^2(\mathbb{D}^2)\longrightarrow H^2(\mathbb{D})\otimes H^2(\mathbb{D}),\qquad
U(z_1^{k_1}z_2^{k_2})=z^{k_1}\otimes z^{k_2},
\]
under which
\[
U\,M_{z_1}\,U^{-1}=M_z\otimes I,\qquad
U\,M_{z_2}\,U^{-1}=I\otimes M_z.
\]

If \(\Theta\) is an inner function on \(\mathbb{D}\), the one-variable Jordan block is
\[
\mathcal{Q}_\Theta:=H^2(\mathbb{D})/\Theta H^2(\mathbb{D}),
\]
unitarily equivalent to the model space \(K_\Theta=H^2(\mathbb{D})\ominus \Theta H^2(\mathbb{D})\). A quotient module \(\mathcal{Q}\subset H^2(\mathbb{D}^2)\) is called doubly commuting when the compressed coordinate shifts satisfy
\[
S_i^*S_j=S_jS_i^*\qquad (i\neq j).
\]

The central classification theorem specializes in two variables to the statement that a quotient module \(\mathcal{Q}\subset H^2(\mathbb{D}^2)\) is doubly commuting if and only if
\[
\mathcal{Q}=Q_1\otimes Q_2
\]
for quotient modules \(Q_1,Q_2\subset H^2(\mathbb{D})\), equivalently
\[
\mathcal{Q}=\mathcal{Q}_{\Theta_1}\otimes \mathcal{Q}_{\Theta_2},
\]
where each factor is either a one-variable Jordan block \(H^2(\mathbb{D})/\Theta_iH^2(\mathbb{D})\) for a one-variable inner function \(\Theta_i\), or the full Hardy module \(H^2(\mathbb{D})\) [1303.1041]. Under the tensor-product identification, the compressed shifts are
\[
S_1=C_z\otimes I,\qquad S_2=I\otimes C_z,
\]
so doubly commuting follows from
\[
S_1S_2^*=C_z\otimes C_z^*=S_2^*S_1.
\]

A standard finite-dimensional example takes \(\Theta_1(z_1)=z_1^m\) and \(\Theta_2(z_2)=B(z_2)\), where \(B\) is a finite Blaschke product of degree \(d\). Then
\[
Q_1=\mathcal{Q}_{z^m}\cong \operatorname{span}\{1,z,\dots,z^{m-1}\},\qquad
Q_2=\mathcal{Q}_{B}\cong K_B,
\]
and
\[
\mathcal{Q}=Q_1\otimes Q_2
\]
has dimension \(md\). On \(Q_1\), \(C_z\) is the truncated unilateral shift, nilpotent of order \(m\); on \(Q_2\), \(C_z\) is a finite-dimensional contraction determined by \(B\), nilpotent of order \(d\) when \(B=z^d\).

## 3. Invariant subspaces, co-doubly commuting submodules, and uniqueness

The bidisk theory extends beyond the quotient itself to the corresponding submodules. A submodule \(\mathcal{S}\subset H^2(\mathbb{D}^2)\) is called co-doubly commuting when the quotient \(H^2(\mathbb{D}^2)/\mathcal{S}\) is doubly commuting. In two variables, the Beurling-like classification states that \(\mathcal{S}\) is co-doubly commuting if and only if
\[
\mathcal{S}=\tilde{\Theta}_1H^2(\mathbb{D}^2)+\tilde{\Theta}_2H^2(\mathbb{D}^2),
\]
where
\[
\tilde{\Theta}_1(z_1,z_2)=\Theta_1(z_1),\qquad
\tilde{\Theta}_2(z_1,z_2)=\Theta_2(z_2),
\]
and the quotient is the two-variable Jordan block
\[
H^2(\mathbb{D}^2)/\mathcal{S}\cong \mathcal{Q}_{\Theta_1}\otimes \mathcal{Q}_{\Theta_2}.
\]
The associated projection formula is
\[
P_{\mathcal{Q}}=I-\big(I-M_{\tilde{\Theta}_1}M_{\tilde{\Theta}_1}^*\big)\big(I-M_{\tilde{\Theta}_2}M_{\tilde{\Theta}_2}^*\big)
\]
[1303.1041].

Recent work develops the invariant-subspace theory internal to the product model space \(K_{\theta_1}\otimes K_{\theta_2}\). With
\[
S_{\Theta_1}=S_{\theta_1}\otimes I,\qquad S_{\Theta_2}=I\otimes S_{\theta_2},
\]
a closed subspace \(M\subset K_{\theta_1}\otimes K_{\theta_2}\) is invariant and doubly commuting if and only if there exist inner functions \(\eta_i,\phi_i\) such that \(\theta_i=\eta_i\phi_i\) and
\[
M=\eta_1K_{\phi_1}\otimes \eta_2K_{\phi_2}.
\]
Equivalently, \(M=W_1\otimes W_2\), where each \(W_i\) is \(S_{\theta_i}\)-invariant in \(K_{\theta_i}\) and has the classical one-variable form \(W_i=\eta_iK_{\phi_i}\) [2605.25575].

The proof mechanism uses defect-operator identities. For any nonconstant inner \(\theta\),
\[
I_{K_\theta}-S_\theta^*S_\theta=T_z^*\theta\otimes T_z^*\theta,\qquad
I_{K_\theta}-S_\theta S_\theta^*=(1-\overline{\theta(0)}\theta)\otimes(1-\overline{\theta(0)}\theta).
\]
These identities, together with cyclicity of \(\{T_z^{*m}\theta:m\ge 1\}\) for \(S_\theta^*\), force reducing hulls to split as tensor products.

Several edge cases are rigid. If \(\Theta_i\) is unimodular, then \(\Theta_iH^2(\mathbb{D})=H^2(\mathbb{D})\) and \(\mathcal{Q}_{\Theta_i}=\{0\}\), so a unimodular factor collapses the two-variable Jordan block to the zero space. If \(\Theta\) is replaced by \(e^{i\phi}\Theta\), the corresponding submodule and quotient do not change, so the factorization is unique only up to unimodular constants. Factor uniqueness itself holds at the level of quotient modules: if \(\mathcal{Q}=Q_1\otimes Q_2=R_1\otimes R_2\), then \(Q_i=R_i\). By contrast, if one assumes only commutation and not double commutation, the tensor-product classification does not persist [1303.1041] [2605.25575].

## 4. Parabolic and Hamiltonian \(2\times 2\) Jordan-block systems

In hydrodynamic-type PDEs, a two-variable Jordan block is a quasilinear system
\[
u_t+A(u)u_x=0,\qquad
A(u)\sim J_2(\lambda(u))=
\begin{bmatrix}
\lambda(u)&1\\
0&\lambda(u)
\end{bmatrix},
\]
or, in concrete coordinates,
\[
A(u)=
\begin{bmatrix}
\lambda(u)&\mu(u)\\
0&\lambda(u)
\end{bmatrix}.
\]
Such a system has a repeated eigenvalue \(\lambda(u)\), one eigenvector, and one generalized eigenvector. In the Toeplitz-block parametrization, the general \(n\)-component system is \(A=\lambda(u)E+\sum_{i=1}^{n-1}\lambda_i(u)P^i\), and the \(2\times 2\) case reduces to \(A=\lambda(u)E+\lambda_1(u)P\) with \(P=\bigl[\begin{smallmatrix}0&1\\0&0\end{smallmatrix}\bigr]\) [2212.01413].

A central constraint is linear degeneracy. For a single \(2\times 2\) Jordan block, it reduces to
\[
\partial_{u^1}\lambda(u)=0.
\]
This condition is necessary for the existence of a first-order Dubrovin–Novikov Hamiltonian structure in Jordan-block systems, and in the \(2\times 2\) case it is also sufficient for the existence of a compatible quasilinear differential constraint [2212.01413] [2404.10101].

The two-variable equivalence theorem is particularly sharp. For systems
\[
u^1_t=\lambda(u^1,u^2)u^1_x+\mu(u^1,u^2)u^2_x,\qquad
u^2_t=\lambda(u^1,u^2)u^2_x,
\]
the following are equivalent: linear degeneracy, existence of a compatible quasilinear constraint, and existence of a Dubrovin–Novikov Hamiltonian structure. Any compatible first-order constraint is quasilinear and may be written
\[
u^2_x=\phi(x,t,u^1,u^2),
\]
while in the autonomous case one obtains
\[
\phi(u^1,u^2)=f(u^1)\exp\int \frac{\partial_{u^2}\lambda(u^1,u^2)}{\mu(u^1,u^2)}\,du^1
\]
[2404.10101].

The same non-diagonalizable structure appears in delta-functional reductions of the kinetic equation for dense soliton gas. For one block, in variables \((r,n)\),
\[
r_t=v(r,n)\,r_x+p(r,n)\,n_x,\qquad
n_t=v(r,n)\,n_x,
\]
with
\[
A(r,n)=
\begin{bmatrix}
v(r,n)&p(r,n)\\
0&v(r,n)
\end{bmatrix}.
\]
This system is linearly degenerate, \(\partial_r v=0\), and admits infinitely many local Hamiltonian structures. A parametric family of flat contravariant metrics is
\[
g^{11}=f_1(n)r^4+f_2(n)r^3,\qquad
g^{12}=g^{21}=f_2(n)r^2,\qquad
g^{22}=f_3(n)r^2,
\]
with \(f_1,f_2,f_3\) arbitrary functions of \(n\) [2212.01413].

For \(N\) delta-species, the reduction produces \(N\) decoupled \(2\times 2\) Jordan blocks
\[
r_a{}_t=v_a r_a{}_x+p_a\eta_a{}_x,\qquad
\eta_a{}_t=v_a\eta_a{}_x,
\]
and model-specific multi-Hamiltonian structures are worked out for KdV, sinh-Gordon, hard-rod, Lieb–Liniger, DNLS, and separable kernels. This places the two-variable Jordan block at the intersection of parabolicity, linear degeneracy, and Hamiltonian integrability [2212.01413].

## 5. Finite-dimensional matrix, partition, and centralizer theories

A different branch of the subject studies how Jordan blocks interact under tensor product and commutation. For \(J_r\) and \(J_s\) the full Jordan blocks with eigenvalue \(1\) over a field of characteristic \(p\), the Jordan canonical form of \(J_r\otimes J_s\) determines a partition
\[
\lambda(r,s,p)=(\lambda_1,\ldots,\lambda_r)
\]
of \(rs\), called the Jordan partition. Grouping equal parts yields the composition vector \(c(r,s,p)\). Glasby–Praeger–Xia proved periodicity and reflection in the second variable \(s\): if \(r\le p^\beta\), then
\[
c(r,s,p)=c(r,s+p^\beta,p),
\]
and within a period,
\[
c(r,p^\beta+i,p)=\operatorname{rev}\bigl(c(r,p^\beta+(p^\beta-i),p)\bigr).
\]
Barry determined the least period,
\[
L(r,p)=p^{\lceil \log_p r\rceil},
\]
and exhibited partial subperiodic and partial subreflective behavior via a six-case \(p\)-adic recursion for \(c(r,s,p)\) [1907.06519].

The finite-dimensional commuting theory becomes markedly richer when one starts from
\[
B\cong J_n\oplus J_n.
\]
Then
\[
C(B)=\{A\in M_{2n}(K):AB=BA\}
\]
has dimension \(4n\), and every element of \(C(B)\) is a \(2\times 2\) block matrix with each block a polynomial in \(J_n\) of degree \(<n\). Equivalently,
\[
C(B)\cong M_2\bigl(K[X]/(X^n)\bigr).
\]
Bogdanić–Đurić–Koljančić–Oblak–Šivic classify all nilpotent Jordan canonical forms that occur inside this centralizer. The admissible partitions are precisely nine infinite families \((P1)\)–\((P9)\); \((P1)\)–\((P7)\) occur over any field, while \((P8)\)–\((P9)\) may depend on the field and do occur over \(\mathbb{C}\) [2212.08425].

This classification shows a sharp contrast with the one-block case. For a single Jordan block, commuting nilpotents have almost rectangular shapes only. For \(J_n\oplus J_n\), one obtains almost rectangular partitions, mixed plateau families, and further families determined by valuation data \((z,l,m)\) and recursive rank formulas derived from
\[
C=
\begin{bmatrix}
0&1\\
c(X)&d(X)
\end{bmatrix},\qquad
s_{k+1}=d(X)s_k+c(X)s_{k-1}.
\]
The broader significance is that the presence of two equal Jordan blocks enlarges the centralizer from a truncated polynomial ring to a matrix algebra over that ring, and the orbit structure correspondingly expands [2212.08425].

## 6. Jordan type in two-variable Artinian algebras

For an Artinian algebra \(A\) and an element \(\ell\) of its maximal ideal, the multiplication map \(m_\ell\) is nilpotent, and its Jordan block decomposition defines the Jordan type \(P_\ell\). In the two-variable graded setting \(A=k[x,y]/I\), this becomes a fine invariant of the algebra and of the chosen linear form. A basic general inequality is
\[
P_{\ell,M}\le H(M)^\vee
\]
in dominance order, where \(H(M)^\vee\) is the conjugate partition of the Hilbert function. In codimension \(2\), if \(\operatorname{char}k=0\) or \(\operatorname{char}k\ge j_A\), a general \(\ell\in A_1\) is strong Lefschetz, so
\[
P_\ell=H(A)^\vee
\]
[1802.07383].

Height-two homogeneous complete intersections give a complete classification. Let
\[
A=R/I,\qquad R=k[x,y],\qquad I=(f,g),\qquad \deg f=d_1\le \deg g=d_2.
\]
Then the Hilbert function has the symmetric form \(T(d,k)\), where \(d=d_1\), \(k=d_2-d_1+1\), and the active Hessians are \(h^i(F)\) for \(i=0,\dots,d-2\), together with \(h^{d-1}(F)\) when \(k\ge 2\). The decisive fact is that the vanishing pattern of the active Hessians at a linear form \(\ell=ax+by\) determines the Jordan type completely. There is a one-to-one correspondence between complete-intersection Jordan type partitions and subsets of active Hessians that vanish at \(\ell\): \(2^d\) possibilities when \(k>1\), and \(2^{d-1}\) when \(k=1\) [1810.00716].

The partitions that occur are characterized numerically. Writing
\[
P=(p_1^{n_1},p_2^{n_2},\dots,p_t^{n_t}),\qquad p_1>\cdots>p_t,
\]
the complete-intersection criterion is
\[
p_{i-1}=n_{i-1}+n_i+p_i\qquad (2\le i\le t).
\]
Equivalently, among all partitions with diagonal lengths \(T(d,k)\), the CI Jordan type partitions are exactly those satisfying these equalities. They are also exactly the partitions with either \(d\) parts or \(d+k-1\) parts [1810.00716].

The strong Lefschetz case is the maximal one:
\[
P(\ell)=H(A)^\vee=(2d+k-2,\,2d+k-4,\,\dots,\,k).
\]
For the monomial complete intersection \(A=k[x,y]/(x^a,y^b)\), the generic Jordan type for \(\ell=x+y\) is
\[
P_\ell=(a+b-1,\ a+b-3,\ a+b-5,\ \dots,\ |a-b|+1),
\]
while the special choices \(\ell=x\) and \(\ell=y\) give
\[
P_x=a^b,\qquad P_y=b^a.
\]
In codimension \(2\), the Jordan degree type of a generic \(\ell\) has staircase form
\[
S_{\ell,A}=((p_1,0),(p_2,1),\dots,(p_s,s-1))
\]
[1802.07383].

This theory makes the phrase “two-variable Jordan block” concrete in commutative algebra: it becomes the block structure of multiplication by a linear form in a two-variable Artinian algebra, constrained by the Hilbert function, Hessians, and Lefschetz properties.

## 7. Computational, combinatorial, and modern applied directions

Near a defective matrix, the \(2\times 2\) Jordan block becomes a numerical target rather than a static normal form. A scalable algorithm computes the Jordan chain of a nearly defective matrix with a \(2\times 2\) Jordan block by first computing the corresponding two-dimensional invariant subspace and then solving the reduced \(2\times 2\) problem
\[
S=U^*AU.
\]
The method avoids dense SVDs, requires only shifted solves and small dense algebra, and produces an eigenvector and Jordan vector with \(O(\varepsilon)\) error when the matrix is \(O(\varepsilon)\) away from an exactly defective matrix; if a matrix derivative \(A'(\mu)\) is known, a one-step correction yields \(O(\varepsilon^2)\) accuracy. The motivating application is an electromagnetism problem with matrix size \(212^2\times 212^2\) [1704.05837].

A recent machine-learning adaptation treats a complex defective Jordan block as a relative positional encoding. With
\[
A=(-\gamma+i\omega)I+\eta N,\qquad
N=\begin{bmatrix}0&1\\0&0\end{bmatrix},\qquad N^2=0,
\]
the exact relative operator is
\[
J(d)=e^{(-\gamma+i\omega)d}(I+\eta d N),
\]
which yields the primitive functions
\[
e^{-\gamma d}\cos(\omega d),\quad e^{-\gamma d}\sin(\omega d),\quad
d e^{-\gamma d}\cos(\omega d),\quad d e^{-\gamma d}\sin(\omega d).
\]
This construction, termed Jordan-RoPE, couples phase and distance inside one defective representation rather than placing them in a direct sum. Exact, stabilized, and scaled-exact variants are distinguished by whether the one-parameter group law is preserved. On a small WikiText-103 byte language model, the scaled-exact variant improves over RoPE and direct-sum baselines within the Jordan family, while RoPE+ALiBi remains strongest overall [2605.04217].

A more arithmetic application considers the non-expansive number system generated by
\[
M=J_2(1)=
\begin{pmatrix}
1&1\\
0&1
\end{pmatrix},
\qquad
\mathcal{D}=\left\{
\binom{0}{1},\binom{0}{-1}
\right\}.
\]
Representations
\[
\binom{a}{b}=\sum_{i=0}^{k-1}M^i d_i
\]
reduce to signed sums
\[
b=\sum_{i=0}^{k-1}s_i,\qquad a=\sum_{i=0}^{k-1} i\,s_i,\qquad s_i\in\{\pm1\}.
\]
For \(b\ge 0\), the minimal length is
\[
k_{\min}=b+2\ell,
\]
where \(\ell\) is the smallest nonnegative integer satisfying
\[
\ell \equiv a-\frac{b(b-1)}{2}\pmod{2}
\]
and
\[
-\ell^2\le a-\frac{b(b-1)}{2}\le \ell^2+2b\ell.
\]
This yields an \(O(1)\) arithmetic test for minimal length and an \(O(k_{\min})\) constructive algorithm for a representation [2605.04990].

These developments show that the two-variable Jordan block is no longer confined to classical normal-form theory. It now functions as an organizing template for quotient modules, Hamiltonian degeneracies, modular partition phenomena, Lefschetz stratifications, scalable EP numerics, non-semisimple positional encodings, and combinatorial digit expansions. The persistent invariant across those domains is the same: a repeated eigenvalue together with a nontrivial Jordan-chain mechanism, realized either literally or through a tensor-product analogue.

Source: https://www.emergentmind.com/topics/two-variable-jordan-block