Level–Rank Duality in Gauge Theories

• Level–rank duality is a correspondence that exchanges the rank of a gauge group with the level of its current algebra, clarifying dualities in conformal field theories and knot invariants.
• It connects diverse domains—including topological quantum field theory, vertex operator algebras, and combinatorial crystal bases—leading to practical insights in knot theory and quantum computation.
• The duality is realized via explicit transformations in WZW and Chern–Simons theories with refined counterterm adjustments, ensuring matching modular data and observables across dual models.

Level–rank duality is a family of correspondences in which the rank of a gauge group or affine Lie algebra and the level of the current algebra or Chern–Simons coupling are exchanged. In its classical form, the central example is $SU(N)_k \leftrightarrow SU(k)_N$, but the same phrase also denotes analogues for symplectic and orthogonal theories, affine type $A$ crystals, vertex operator algebras, non-abelian theta functions on higher-rank Prym varieties, and blocks of cyclotomic Hecke algebras attached to finite reductive groups (Soroush, 2015, Ostrik et al., 2020, Gerber, 2018, Baier et al., 2024).

1. Foundational formulation in WZW and Chern–Simons theory

In the WZW formulation, level–rank duality states that the primary fields and their correlation functions in the $SU(N)_k$ and $SU(k)_N$ models are in one-to-one correspondence, and via the standard relation between WZW conformal blocks and Chern–Simons Hilbert spaces this becomes a duality between observables of two different Chern–Simons theories whose roles of level and rank are exchanged (Soroush, 2015). For Chern–Simons theory on $S^3$, this correspondence takes the explicit form

$$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$

where $K$ is a knot, $\tilde K$ its mirror, and $\lambda^t$ is the transposed Young diagram.

A refined three-dimensional formulation requires attention to background gauge fields, spin$_c$ connections, and gravitational counterterms. In that setting, level/rank duality is an exact equivalence of TQFTs only after adding specific background Chern–Simons counterterms, so that currents, anomalies, framing data, and line operators match. Representative dualities include $A$0, $A$1, and the further $A$2–$A$3 dualities involving shifted Abelian levels (Hsin et al., 2016). A related point is that the duality is often naturally a statement about spin TQFTs, or about non-spin theories tensored with an almost trivial spin sector carrying a transparent spin-$A$4 line.

At the level of knot and link invariants, the duality is encoded in relations among modular $A$5-matrices, fusion coefficients, braid matrices, and quantum dimensions. For $A$6, the root of unity $A$7 is invariant under exchanging $A$8 and $A$9, which is one reason braid-theoretic data admit level–rank reformulations (Schnitzer, 2021). A recurrent misconception is that the duality is always realized by naive transposition of every Young tableau. In fact, for $SU(N)_k$0 it is one-to-one on cominimal equivalence classes rather than on individual representations, and in several settings quotients by simple-current actions are essential (Schnitzer, 2015).

2. Knot invariants and the topological-string realization

For knot theory, the most important concrete consequence is the mirror relation for colored HOMFLY invariants,

$SU(N)_k$1

with $SU(N)_k$2 invariant under $SU(N)_k$3 and $SU(N)_k$4 sent to $SU(N)_k$5 (Soroush, 2015). In this form, level–rank duality becomes a statement that mirror reflection of the knot and transposition of the representation correspond to exchanging level and rank in Chern–Simons theory.

The same relation has a worldsheet interpretation in topological string theory. Under the large $SU(N)_k$6 duality between $SU(N)_k$7 Chern–Simons theory on $SU(N)_k$8 and the closed A-model on the resolved conifold, Wilson loops along a knot $SU(N)_k$9 correspond to open-string amplitudes with a knot Lagrangian $SU(k)_N$0. Writing the open amplitudes as generating series for open Gromov–Witten invariants, level–rank duality yields the identity

$SU(k)_N$1

which the paper interprets as orientation reversal of the Lagrangian brane together with degree reflection under $SU(k)_N$2 (Soroush, 2015).

This A-model picture has a B-model counterpart. In the non-toric knot setting, the mirror curve is identified with the augmentation polynomial $SU(k)_N$3 from knot contact homology, and the mirror knot is obtained by

$SU(k)_N$4

Disk, annulus, and higher-genus amplitudes then transform compatibly with the same degree-reflection rule, and the topological recursion built from the augmentation polynomial and the physical annulus kernel reproduces the full worldsheet identity (Soroush, 2015).

This suggests that level–rank duality is not merely a boundary RCFT symmetry. A plausible implication is that it organizes an entire package of knot-theoretic, symplectic, and mirror-geometric data: colored HOMFLY polynomials, augmentation varieties, and open Gromov–Witten invariants transform in parallel under the same exchange $SU(k)_N$5, $SU(k)_N$6, and boundary-orientation reversal.

3. Categorical, VOA, and boundary-CFT manifestations

In tensor-categorical form, one of the sharpest results is the symplectic duality

$SU(k)_N$7

sending the simple object indexed by $SU(k)_N$8 to the object indexed by $SU(k)_N$9, where $S^3$0 is the transpose of the complement in the $S^3$1 rectangle. After the standard $S^3$2-modification of the braiding, this becomes a genuine braided equivalence $S^3$3 (Ostrik et al., 2020). One proof uses the conformal embedding $S^3$4; the other uses classification of braided fusion categories of type $S^3$5 by braiding eigenvalues.

In the VOA setting for types $S^3$6 and $S^3$7, level–rank duality appears as a commutant construction. The commutant of the diagonally embedded $S^3$8 inside $S^3$9 is realized as a fixed-point subalgebra of a VOA based on $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$0 at level $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$1, or of a simple-current extension thereof, associated with a finite abelian group. In this sense, exchanging the tensor-power multiplicity and the orthogonal rank yields a version of level–rank duality in orthogonal coset theory (Jiang et al., 2017).

Boundary CFT furnishes another manifestation. For WZW models on a circle and for untwisted and twisted D-branes, the left–right entanglement entropy is an explicit functional of modular $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$2-matrix data. Level–rank duality then implies precise relations for the finite part of the entropy. For $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$3, the finite part is exactly invariant for any Cardy state; for $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$4, it differs by a universal $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$5 shift for generic Cardy states, while twisted branes in $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$6 acquire a $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$7 shift (Schnitzer, 2015).

For orthogonal WZW branes, the duality is more selective. Untwisted $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$8 D-branes corresponding to tensor representations are mapped to untwisted $$\langle W_{\lambda}(K)\rangle_{SU(N)_k}\longleftrightarrow \langle W_{\lambda^t}(\tilde K)\rangle_{SU(k)_N},$$9 D-branes, while for $K$0-twisted sectors of $K$1 the paper proves a duality only for the spinor twisted branes. In both cases, the spectrum of an open string ending on the branes is isomorphic to the spectrum on the level–rank-dual branes, but the orthogonal story is partial rather than uniform across all representations (0706.1957).

4. Combinatorial, crystal, and Hecke-algebra realizations

In affine type $K$2, level–rank duality acts on Fock spaces by exchanging the number of multipartition components and the affine rank. Gerber formulates a combinatorial bijection

$K$3

sending an $K$4-multipartition with charge $K$5 to an $K$6-multipartition with dual charge $K$7, and realizes the duality on abaci by infinite periodic stacking followed by a $K$8 change of viewpoint (Gerber, 2018). In that framework, a central theorem states that an $K$9-abacus is $\tilde K$0-cylindric if and only if its level–rank dual is a source in the $\tilde K$1-crystal; the FLOTW condition is strengthened by requiring that the dual be simultaneously a source in the Heisenberg crystal (Gerber, 2018).

This gives cylindricity a representation-theoretic meaning. What had been a combinatorial condition on multipartitions becomes the highest-weight condition for the dual affine crystal. A plausible implication is that level–rank duality here is less a symmetry between two fixed models than a transport mechanism between different commuting crystal structures.

A recent finite-group extension places the same phenomenon inside unipotent representation theory of finite reductive groups. For $\tilde K$2, $\tilde K$3-cuspidal pairs $\tilde K$4 yield Harish–Chandra series parametrized by partitions with fixed $\tilde K$5-core, while the associated relative Weyl group is $\tilde K$6, whose irreducibles are indexed by $\tilde K$7-multipartitions. After specializing Broué–Malle cyclotomic Hecke algebras to roots of unity, the intersections of $\tilde K$8- and $\tilde K$9-series produce blocks related by Uglov’s level–rank bijection between charged multipartitions, recovering an avatar of the duality studied by Frenkel, Uglov, Chuang–Miyachi, and others (Trinh et al., 9 Aug 2025).

5. Brane, holographic, and Chern–Simons–matter derivations

String-theoretic derivations emphasize that level–rank duality can arise as the infrared shadow of higher-dimensional brane dynamics. In a non-supersymmetric type IIB configuration with fivebranes, an $\lambda^t$0-plane, and anti-D3 branes, swapping the fivebranes produces a three-dimensional non-supersymmetric Seiberg duality whose IR limit is the symplectic level–rank pair

$\lambda^t$1

with an analogous unitary duality

$\lambda^t$2

The scalar in the brane system acquires positive mass-squared and decouples, while integrating out fermions generates the level shifts needed for the duality (Armoni et al., 2014).

A more systematic field-theoretic account couples Chern–Simons theories to background fields, spin$\lambda^t$3 connections, and the metric, and shows that finite counterterms are required for exact duality. This produces refined level/rank dualities for pure Chern–Simons theories and, by adding matter, a web of bosonization, boson/boson, and fermion/fermion dualities in three dimensions (Hsin et al., 2016). In that framework, level–rank duality is best viewed as an equivalence of full quantum field theories, not merely of abstract modular tensor categories.

A holographic realization in type IIB starts from D3-branes compactified on a circle with an axion gradient inducing a boundary $\lambda^t$4 Chern–Simons term. Probe D7-branes at the tip of the cigar geometry support a $\lambda^t$5 worldvolume theory, and careful treatment of RR two- and six-form boundary conditions reproduces dual pairs such as $\lambda^t$6 and $\lambda^t$7. The full $\lambda^t$8 orbit is realized through electric/magnetic duality of the RR potentials (Argurio et al., 2022).

Lens-space partition functions supply a stringent test of these refined statements. Using supersymmetric localization and a supersymmetry-preserving regularization, the lens-space partition function acquires a background-dependent topological phase. With an explicit phase factor $\lambda^t$9, the partition functions on $_c$0 exhibit perfect agreement, including the total phase, for standard and spin level–rank dual pairs such as $_c$1, $_c$2, and $_c$3 (Kubo et al., 2021).

6. Entanglement, quantum computation, and recent geometric extensions

Level–rank duality also acts on information-theoretic and computational structures extracted from Chern–Simons theory. In the topological quantum-computation setting, the pair $_c$4 has identical Jones representations at $_c$5, and the dimensions of the relevant conformal-block spaces match: $_c$6 on the $_c$7 side correspond to

$_c$8

on the $_c$9 side. Consequently, the universal topological quantum computer based on $A$00 implies a qutrit-based universal model built from $A$01 (Schnitzer, 2018).

A related construction identifies generalized Pauli operators across quotiented level–rank pairs such as $A$02 and $A$03, at least when the induced qudit dimension $A$04 is odd. This yields one-to-one maps between graph states, hypergraph states, and magic states built from the same Pauli data on both sides of the duality (Schnitzer, 2021). This suggests that level–rank duality can transport not only modular data and link invariants but also computational resource states.

A geometric extension replaces ordinary Jacobians and moduli of bundles by higher-rank Prym varieties associated with an unramified double cover $A$05. In this anti-invariant setting, the authors construct a Prym–Hitchin connection on bundles of non-abelian theta functions, identify it with the twisted WZW connection on twisted conformal blocks, and formulate an anti-invariant strange-duality morphism exchanging rank and level. They prove the duality at level one and show that at all levels the duality respects the flat connections (Baier et al., 2024).

Across these realizations, the shared content is the interchange of rank and level together with a precise control of the structures attached to that interchange: modular data, fusion, crystals, Hecke blocks, Hitchin-type connections, knot invariants, topological string amplitudes, and quantum-information constructions. What varies from context to context is the mechanism. In some settings the duality is a braid-reversing equivalence, in others a source condition in a crystal, a commutant/orbifold relation for VOAs, a mirror symmetry for knot amplitudes, or a flat pairing of non-abelian theta bundles.