---
title: 'Two-Modular Diagrams: A Survey'
url: https://www.emergentmind.com/topics/two-modular-diagrams
type: topic
---

# Two-Modular Diagrams: A Survey

Two-Modular Diagrams is not a single standardized term across mathematics and mathematical physics. In the literature represented here, it denotes, or is naturally used to denote, several distinct diagrammatic constructions: pairs of Dynkin diagrams producing modular generalized Nahm sums and characters of rational \(2\)d CFTs; \(2\)-loop modular Feynman diagrams on a torus; modular \(2\)-noncrossing or \(\sigma=2\) diagrams in RNA-related combinatorics; and modified \(2\)-modular diagrams of partitions related to Zeilberger’s KOH identity. In adjacent settings, the expression is also used interpretively for diagrammatic structures attached to modular representations, to degree-\(2\) level-\(2\) Siegel modular forms, and to prime-\(2\) topological modular forms [2604.00847, 1509.00363, 1003.2710, 1006.2881, 1101.1475, 1603.06506, 2605.13300, 1203.1696].

## 1. Terminological scope and recurring structure

A recurring pattern is that the qualifier “two” refers to different structural parameters in different subfields. In the theory of generalized Nahm sums it refers to a **pair** \((X,Y)\) of Dynkin diagrams. In genus-one superstring perturbation theory it refers to **two loops**. In RNA-related combinatorics it may refer either to **\(k=2\)** noncrossing diagrams or to **\(\sigma=2\)** modularity. In the partition-theoretic setting it refers to **\(N=2\)** labels in a modified modular diagram. In the Siegel-modular and chromatic-homotopy settings it refers to **degree \(2\), level \(2\)** or **prime \(2\), height \(2\)**, respectively [2604.00847, 1509.00363, 1003.2710, 1006.2881, 1101.1475, 2605.13300, 1203.1696].

This suggests that the expression is context-dependent rather than a universal term of art. The common feature is the use of diagrammatic data to package modular, homological, or representation-theoretic structure. The ambient subject therefore determines what the diagrams are, what “modular” means, and what kind of invariants are extracted from them.

A second recurring feature is that these constructions are not merely pictorial. In each setting, the diagrammatic language controls precise algebraic data: Kronecker products of Cartan matrices and root-length matrices in generalized Nahm sums, lattice sums and Laplace equations in torus amplitudes, generating functions and asymptotics in RNA combinatorics, or covariants and theta-theoretic constructions in genus-\(2\) modular forms [2604.00847, 1509.00363, 1003.2710, 2605.13300].

## 2. Pairs of Dynkin diagrams and modular generalized Nahm sums

In "Dynkin diagrams, generalized Nahm sums and 2d CFTs" [2604.00847], the phrase “two-modular diagram” does not appear verbatim, but the paper explicitly proposes the interpretation that a **two-modular diagram** is a pair \((X,Y)\) of Dynkin diagrams such that the associated generalized Nahm sum
\[
f_{A(X,Y),0,C(X,Y),D(X,Y)}(\tau)
\]
is a modular function. The construction starts from Cartan matrices \(C(X)\), \(C(Y)\), root-length matrices \(D(X)\), \(D(Y)\), and the Kronecker products
\[
A(X,Y)=C(X)\otimes C(Y)^{-1},\qquad D(X,Y)=D(X)\otimes D(Y).
\]
The generalized Nahm sum attached to \((X,Y)\) is then
\[
f_{A(X,Y),\,0,\,C(X,Y),\,D(X,Y)}(\tau)
=\sum_{n\in\mathbb{Z}_{\ge0}^{r(X)r(Y)}}
\frac{q^{\frac12 n^tA(X,Y)D(X,Y)n+C(X,Y)}}
{\prod_i (q^{d_i};q^{d_i})_{n_i}},
\]
where the diagonal entries \(d_i\) come from \(D(X,Y)\). The proposed central charge in the \(ABCDEFGT\) framework is
\[
c(X,Y)=\frac{\mathrm{tr}(D(X,Y))\,h(X)}{h(X)+h(Y)},\qquad
C(X,Y)=-\frac{c(X,Y)}{24}.
\]

The paper’s central conjecture extends the older \(ADET\) folklore conjecture to all finite-type Dynkin diagrams of type \(ABCDEFGT\): for every pair \((X,Y)\), the quadruple
\[
\bigl(A(X,Y),0,C(X,Y),D(X,Y)\bigr)
\]
is modular, meaning that the generalized Nahm sum is a modular function of weight \(0\). A duality
\[
(A,B,C,D)\mapsto \Big(A^{-1},A^{-1}B,\frac12 B^t(AD)^{-1}B-\frac{\mathrm{tr}(D)}{24}-C,D\Big)
\]
specializes, when \(B=0\), to the symmetry between \((X,Y)\) and \((Y,X)\). In this sense the modular structure is genuinely attached to the **pair**.

The CFT-theoretic significance is that modular Nahm sums with non-negative integer Fourier coefficients are natural candidates for characters of rational \(2\)d CFTs. The paper identifies many such sums with known characters. Two infinite families are especially explicit. For \((T_1,C_r)\), one has
\[
c(T_1,C_r)=\frac{3(r+1)}{2r+3},
\]
and the generalized Nahm sum is identified with the supersymmetric Virasoro minimal models \(\mathrm{SM}(4r+6,4)\). For \((T_1,D_r)\), one has
\[
c(T_1,D_r)=\frac{3r}{2r+1},
\]
and the associated CFT is \(\mathrm{SM}(8r+4,2)\). In the latter case the paper gives the explicit character identity
\[
f_{C(D_r)^{-1},\vec{0},-\frac{r}{8(2r+1)}}(\tau)
=
\chi^{\mathrm{SM}_{\rm eff}(8r+4,2)}_{\rm NS, j=1}
+\chi^{\mathrm{SM}_{\rm eff}(8r+4,2)}_{\rm NS, j=2r+1}
+2\,\chi^{\mathrm{SM}_{\rm eff}(8r+4,2)}_{\rm R, j=r+1}.
\]
The paper also notes that for rank \(\le 3\) generalized Nahm sums arising from Conjecture \(1.1\), there are \(35\) cases and \(28\) are now known to be modular [2604.00847].

## 3. Two-loop modular diagrams on a torus

In "Proof of a modular relation between 1-, 2- and 3-loop Feynman diagrams on a torus" [1509.00363], “two-modular diagrams” refers naturally to **\(2\)-loop modular diagrams/functions** on the torus. These arise in the low-energy expansion of the genus-one Type II four-graviton amplitude, where the coefficient functions are non-holomorphic modular functions of the torus modulus \(\tau\). The \(2\)-loop objects are the triple-sum modular functions
\[
C_{a,b,c}(\tau,\bar\tau),
\]
defined by momentum-conserving lattice sums with internal weight
\[
w=a+b+c.
\]

In this setting the relevant diagrams are vacuum Feynman diagrams on the torus with two independent momentum cycles. The associated modular function is
\[
C_{a,b,c}(\tau,\bar\tau)
=
\sum_{(m_r,n_r)}
\delta_{m_1+m_2+m_3,0}\,\delta_{n_1+n_2+n_3,0}
\prod_{r=1}^3
\frac{\tau_2^{a_r}}{\pi^{a_r}|m_r\tau+n_r|^{2a_r}},
\]
with \(a_1+a_2+a_3=w\). These functions are modular invariant but non-holomorphic, since the dependence on \(\tau\) enters through \(\tau_2\) and \(|m\tau+n|^2\).

The paper emphasizes the Laplace structure of these \(2\)-loop functions. Two low-weight identities are
\[
\Delta C_{1,1,1}=6E_3,\qquad
(\Delta-2)C_{2,1,1}=9E_4-E_2^2.
\]
Thus the \(2\)-loop objects are not Laplace eigenfunctions; instead their Laplacians are inhomogeneous combinations of \(1\)-loop Eisenstein series. At weight \(4\), the central theorem states
\[
D_4(\tau,\bar\tau)=24\,C_{2,1,1}(\tau,\bar\tau)+3\,E_2(\tau,\bar\tau)^2-18\,E_4(\tau,\bar\tau),
\]
equivalently
\[
F(\tau,\bar\tau)=D_4-24\,C_{2,1,1}-3E_2^2+18E_4=0.
\]
This shows that the \(3\)-loop modular function \(D_4\) is not independent: it is determined by the \(2\)-loop function \(C_{2,1,1}\) and \(1\)-loop Eisenstein series.

The role of two-modular diagrams is therefore structural. They are the essential intermediate depth in the hierarchy of modular graph functions appearing in genus-one string amplitudes. The paper further proves three new holomorphic modular identities in the course of establishing the weight-\(4\) relation, with \(\varphi_4(q)=\varphi_5(q)=\varphi_6(q)=0\) forced by modularity and cusp behavior [1509.00363].

## 4. Combinatorial meanings: modular \(2\)-noncrossing diagrams and \(\sigma=2\) modular diagrams

In RNA-related combinatorics there are two distinct but nearby meanings. In "Modular, \(k\)-noncrossing diagrams" [1003.2710], a **modular \(2\)-noncrossing diagram** is a diagram over \([n]\) in which no two arcs cross, every arc has length at least \(4\), and there are no isolated arcs. For \(k=2\), “\(2\)-noncrossing” means precisely noncrossing. The paper states that RNA secondary structures are exactly modular, \(2\)-noncrossing diagrams, so in this usage two-modular diagrams are classical RNA secondary structures with minimum arc length \(4\) and no isolated base pairs.

The enumeration is explicit. If \(Q_2(n)\) denotes the number of modular \(2\)-noncrossing diagrams over \(n\) vertices and
\[
\mathbf{Q}_2(z)=\sum_{n\ge0}Q_2(n)z^n,
\]
then
\[
\mathbf{Q}_2(z)=
\frac{1-z^2+z^4}{1-z-z^2+z^3+2z^4+z^6}\,
\mathbf{F}_2\!\left(
\frac{z^4-z^6+z^8}{(1-z-z^2+z^3+2z^4+z^6)^2}
\right),
\]
where \(\mathbf{F}_2(z)\) is the generating function of \(2\)-noncrossing matchings without isolated vertices. The asymptotic formula is
\[
Q_2(n)\sim 1.4848\, n^{-3/2}\,1.8489^{-n}.
\]
The paper derives this via symbolic and analytic combinatorics, using shapes, inflation into stems and stacks, and a supercritical composition analysis [1003.2710].

A different definition appears in "On the uniform generation of modular diagrams" [1006.2881]. There a **\(\sigma\)-modular** diagram is one in which every arc is contained in a stack of length at least \(\sigma\). Specializing to \(\sigma=2\), a **\(2\)-modular diagram** is a diagram in which every arc belongs to a stack of length at least \(2\). This is not the same as the “minimum arc length \(4\), no isolated arcs” condition above. The paper studies \(k\)-noncrossing, \(\sigma\)-modular diagrams through a bijection with \(*\)-tableaux and weighted cores. It proves an algorithmic uniform generation result: after \(O(n^k)\) preprocessing time, \(k\)-noncrossing, \(\sigma\)-modular diagrams can be generated in \(O(n)\) time and space. For \(\sigma=2\), the recursion for weighted-core counts specializes to
\[
\begin{aligned}
w^{2}_i(\lambda^r)
&=
w^{2}_{i-1}(\lambda_0^{r-1})
+\sum_{j=1}^{k-1}w^{2}_{i-1}(\lambda_{j+}^{r-1}) \\
&\quad+
\sum_{j=1}^{k-1}\sum_{s=2}^{\lfloor \frac{i+1}{2}\rfloor}
\sum_{\ell=1}^{\lfloor \frac{s}{2}\rfloor}
(-1)^{\ell-1}p(s,\ell,2)\,
w^{2}_{i-2s+1}(\lambda_{j-}^{r-1}),
\end{aligned}
\]
and the generation theorem remains uniform [1006.2881].

A common misconception is to conflate these two notions. In one paper, “modular” means **no isolated arcs and minimum arc length \(4\)**; in the other, \(\sigma=2\) means **every arc is contained in a stack of length at least \(2\)**. The objects are related by RNA-motivated combinatorics, but the definitions are not equivalent [1003.2710, 1006.2881].

## 5. \(2\)-modular diagrams of partitions and the KOH identity

In "The KOH terms and classes of unimodal \(N\)-modular diagrams" [1101.1475], the relevant objects are modified modular diagrams of partitions. An \(N\)-modular diagram of length \(k\) is the Ferrers diagram of a partition \(\lambda\), with an added zeroth column of length \(k\), each cell labeled by an integer in \(\{1,2,\dots,N\}\), all cells except the rightmost in each row labeled \(N\), and labels weakly decreasing down columns. When \(N=2\), one obtains a **\(2\)-modular diagram**: every cell is labeled \(1\) or \(2\), all non-rightmost cells are labeled \(2\), and rows of length \(0\) are allowed. The paper explicitly notes that \(2\)-modular diagrams are naturally in bijection with MacMahon diagrams, with label \(2\) corresponding to an unmarked cell and label \(1\) corresponding to a marked cell.

The paper uses these diagrams to interpret the summands in Zeilberger’s KOH identity, a reformulation of O’Hara’s proof of the unimodality of the Gaussian polynomial. For \(\lambda\vdash b\), the KOH term is
\[
F_\lambda(q)
=
q^{\,2\sum_{i\ge 1}\binom{\lambda_i}{2}}
\prod_{j\ge 1}
\begin{bmatrix}
j(a+2)-Y_{j-1}-Y_{j+1} \\
\lambda_j-\lambda_{j+1}
\end{bmatrix}_q,
\qquad
Y_i=\sum_{t=1}^i \lambda_t.
\]
The paper constructs two natural classes of modular diagrams whose generating functions are these KOH terms. One class uses \(\lambda_1\)-modular diagrams; the other uses \(p\)-modular diagrams built from data \((d_j,\gamma^{d_j})\).

The \(N=2\) specialization is especially concrete. For \(\lambda=(2^{m_2},1^{m_1})\) with \(2m_2+m_1=b\), the paper states that
\[
\frac{q^{3m_2+m_1}\big(1-q^{m_2(a-2)+1}\big)\big(1-q^{m_2(a-2)+am_1+1}\big)}{(1-q)^2}
\]
is the generating function of MacMahon \((2\)-modular) diagrams inside an \(a\times b\) rectangle satisfying four conditions: there are \(m_2\) unmarked rows and \(m_1+m_2\) marked rows; the sums of lengths of marked and unmarked rows obey explicit upper bounds; all unmarked rows have length at least \(2\); and within marked rows, and within unmarked rows, the difference between longest and shortest is at most \(1\) [1101.1475].

In this branch of the subject, two-modular diagrams are therefore not graph diagrams but **labeled Ferrers-type diagrams**. Their significance is that they provide a combinatorial model for symmetric unimodal polynomials arising from Gaussian coefficients and the KOH decomposition.

## 6. Related level-two and prime-two diagrammatics

Several further papers use the phrase only interpretively, but they reinforce the same pattern of context dependence. In "A new type of diagrams for modules" [1603.06506], the phrase “Two-Modular Diagrams” does not appear verbatim. The paper instead introduces **virtual categories** and **central tuned diagrams** for finitely generated modules, with specific applications suggested to the modular representations of finite groups of Lie type. The relation is indirect: if “two-modular” is read as characteristic \(2\) or as a comparison between two modular settings, then central tuned diagrams provide a refined diagrammatic language for radical and socle series, virtual simples, and \(\operatorname{Ext}^1\)-classes.

In "Tautological modular forms of level two and degree two" [2605.13300], the phrase is again interpretive rather than standard terminology. The paper studies vector-valued Siegel modular forms of **degree \(2\)** and **level \(2\)** via divisors on the projectivized Hodge bundle. The central diagrammatic structures are the moduli inclusions
\[
\mathcal{M}_2[2]\hookrightarrow \mathcal{A}_2[2],\qquad
\mathbb{P}(\mathbb{E})\xrightarrow{\varpi}\widetilde{\mathcal{A}_2[2]},
\]
the six divisors \(W_i\) coming from Weierstrass points, and the maps
\[
\mu:\mathcal{R}(\mathcal{A}_2[2])\to \mathcal{C}'(V^{\oplus 6}),
\qquad
\nu:\mathcal{C}'(V^{\oplus 6})\to \mathcal{R}(\mathcal{A}_2[2])[1/\chi_5].
\]
The paper shows how all vector-valued Siegel modular forms of level \(2\) and degree \(2\) can be constructed from tautological modular forms, theta gradients, and invariant theory. In this setting “two-modular” means degree-\(2\), level-\(2\), rather than any combinatorial graph notion.

In "Strictly commutative realizations of diagrams over the Steenrod algebra and topological modular forms at the prime \(2\)" [1203.1696], the explanatory discussion presents a structured picture of “two-modular diagrams” at **prime \(2\)** and **chromatic height \(2\)**. The central diagram is the strictly commutative square of \(\mathcal{E}_\infty\)-ring spectra
\[
\xymatrix{
tmf_{(2)} \ar[r]^{c}\ar[d]^{o} & ko_{(2)}\ar[d]^{\iota} \\
tmf_1(3)_{(2)} \ar[r]^{\tilde{c}} & ku_{(2)}
}
\]
lifting the classical diagram of modules over the mod-\(2\) Steenrod algebra
\[
\xymatrix{
\mathcal{A}^*/\!/\mathcal{A}(2) & \mathcal{A}^*/\!/\mathcal{A}(1)\ar[l] \\
\mathcal{A}^*/\!/E(2)\ar[u] & \mathcal{A}^*/\!/E(1)\ar[l]\ar[u].
}
\]
Here the “two” refers simultaneously to the prime and to chromatic height. The paper proves that the generalized \(BP\langle 2\rangle\)-like spectrum \(tmf_1(3)_{(2)}\) fits into this strictly commutative diagram and satisfies
\[
H^*(tmf_1(3);\mathbb{F}_2)\cong \mathcal{A}^*/\!/E(2).
\]

Across these related usages, the same label points to different kinds of structure: characteristic \(2\), degree \(2\), level \(2\), height \(2\), or a pair of inputs. This suggests that any serious use of the expression requires immediate disambiguation by field, notation, and source [1603.06506, 2605.13300, 1203.1696].

Source: https://www.emergentmind.com/topics/two-modular-diagrams