---
title: Doubled Geometry in String Theory
url: https://www.emergentmind.com/topics/doubled-geometry-framework
type: topic
---

# Doubled Geometry in String Theory

Searching arXiv for the cited doubled-geometry papers to ground the article in the literature and verify identifiers.
tool call: arxiv_search({"query":"id:1106.6291 OR id:0902.4032 OR id:1212.1736 OR id:1401.1311 OR id:1402.2513 OR id:1106.1174 OR id:1909.12348 OR id:1303.6727 OR id:1903.01243 OR id:1804.00744", "max_results": 10, "sort_by": "relevance"})
Doubled geometry is an \(O(d,d)\)-covariant framework for the NS–NS sector of string theory in which the conventional target-space data \((G,B,H=dB)\) are encoded on a doubled space, and physical spacetime is recovered only after a choice of polarization. In the fully doubled version used by Hull and Reid-Edwards and tested in WZW models, the doubled target is a quotient \(\mathcal{X}_{2d}=\Gamma\backslash \mathcal{G}_{2d}\) equipped with a split-signature bilinear form and a generalized metric; in the torus-fibered version, only the torus fiber is doubled. The framework was developed to geometrize T-duality, T-fold monodromies, and more general flux backgrounds, and it provides a common language for geometric compactifications, locally non-geometric T-folds, and backgrounds that are not even locally geometric [0902.4032, 1106.6291].

## 1. Formal definition of the doubled target space

In the Hull–Reid-Edwards setup, the doubled space is a \(2d\)-dimensional Lie-group quotient
\[
\mathcal{X}_{2d}=\Gamma\backslash \mathcal{G}_{2d},
\]
where \(\mathcal{G}_{2d}\) has Lie algebra generators \(T_M\) and dual left-invariant one-forms \(\mathcal{P}^M\) obeying
\[
[T_M,T_N]=t_{MN}{}^P T_P,\qquad
d\mathcal{P}^M+\frac12\,t_{NP}{}^M\,\mathcal{P}^N\wedge\mathcal{P}^P=0.
\]
A constant \(O(d,d)\) metric \(L_{MN}\) of signature \((d,d)\) is required,
\[
ds^2_{(d,d)}=L_{MN}\,\mathcal{P}^M\mathcal{P}^N,
\]
and invariance of \(L\) implies that \(t_{MNP}:=t_{MN}{}^Q L_{QP}\) is totally antisymmetric. The doubled space also carries a Riemannian metric
\[
ds^2_{\text{Riem}}=\mathcal{M}_{MN}\,\mathcal{P}^M\mathcal{P}^N,
\]
with \(\mathcal{M}\) an \(O(d,d)\)-compatible generalized metric. In a polarization basis \(T_M=(Z_m,X^m)\), it takes the standard form
\[
\mathcal{M}_{MN}=
\begin{pmatrix}
g_{mn}+b_{mp}g^{pq}b_{qn} & b_{mp}g^{pn}\\
-g^{mp}b_{pn} & g^{mn}
\end{pmatrix},
\]
so that the ordinary metric and \(B\)-field are extracted from the doubled data [1106.6291].

A polarization is a choice of maximally isotropic \(d\)-dimensional subbundle with respect to \(L\). In the corresponding basis, the Lie algebra decomposes into the flux data
\[
\begin{aligned}
[Z_m,Z_n] &= K_{mnp}X^p + f_{mn}{}^p Z_p,\\
[Z_m,X^n] &= f_{mp}{}^n X^p + Q_m{}^{np} Z_p,\\
[X^m,X^n] &= Q_p{}^{mn} X^p + R^{mnp} Z_p.
\end{aligned}
\]
These \(K,f,Q,R\) coefficients encode NS–NS twisting in a polarization-dependent way. If \(R=0\), the \(X^m\) generate a subgroup \(G_{\text{pol}}\subset \mathcal{G}_{2d}\), and the physical target can be obtained as the quotient
\[
X_d = G_{\text{pol}}\backslash \mathcal{X}_{2d}.
\]
Local physical one-forms \(\mathfrak{p}^m\) and \(\mathfrak{q}_m\) then allow reconstruction of the physical fields via
\[
ds^2_{\text{phys}}=G_{mn}(x)\,\mathfrak{p}^m\mathfrak{p}^n,
\]
and
\[
H=dB_{\text{phys}}-d(L_{mn}\,\mathfrak{p}^m\wedge \mathfrak{q}^n)-\mathcal{K},
\qquad
\mathcal{K}=\frac16\,t_{MNP}\,\mathcal{P}^M\wedge\mathcal{P}^N\wedge\mathcal{P}^P.
\]
This is the basic mechanism by which doubled geometry recovers ordinary backgrounds.

## 2. Relation to T-folds, generalized geometry, and double field theory

Three frameworks are explicitly identified for the study of non-geometric backgrounds: the T-fold construction, Hitchin’s generalized geometry, and fully doubled geometry. In the T-fold description, only a torus fiber is doubled. One replaces a \(T^n\) fiber by a doubled \(T^{2n}\) with coordinates \((x^i,\tilde x_i)\), an \(O(n,n)\)-invariant metric
\[
L_{IJ}=
\begin{pmatrix}
0 & \delta^i{}_j\\
\delta_i{}^j & 0
\end{pmatrix},
\]
and a generalized metric \(H_{IJ}(y)\) that packages the ordinary \(G\) and \(B\) fields. A polarization selects the physical \(T^n\subset T^{2n}\), and different polarizations related by \(O(n,n;\mathbb Z)\) give T-dual descriptions [1106.6291].

Generalized geometry instead works on the generalized tangent bundle \(T\oplus T^*\) of an ordinary manifold. It has a natural \(O(d,d)\) structure and a local flux decomposition mirroring doubled geometry:
\[
K_{mnp},\quad f_{mn}{}^p,\quad Q_m{}^{np},\quad R^{mnp}.
\]
In this setting the generalized torsion acts on differential forms with degrees shifted by \(+3,+1,-1,-3\), respectively. The formal similarity with doubled geometry is exact at the level of the flux algebra, but generalized geometry presumes a conventional manifold, so it is intrinsically local when \(Q\neq 0\), and it ceases to provide even a local sigma-model description when \(R\neq 0\) [1106.6291].

The double field theory literature systematized these structures in basis-independent form. An invariant geometry of DFT was formulated with generalized connection, torsion, and curvature defined without reference to a particular basis; the resulting generalized Riemann tensor contains the conventional Ricci tensor and scalar curvature but not the full Riemann tensor, which already indicates a structural difference from ordinary Riemannian geometry [1212.1736]. Global analyses of finite generalized diffeomorphisms then showed that a “generalised manifold” in the \(O(d,d)\) setting is an ordinary smooth manifold with a global \(O(d,d)\) metric, while the generalized metric carries gerbe patching data; triple overlaps are trivial on coordinates but non-trivial on fields, including the generalized metric [1401.1311].

## 3. Polarization, fluxes, and global structure

The central geometric operation in doubled geometry is not coordinate elimination but polarization. The same doubled group manifold can encode conventional geometry, a T-fold, or an essentially doubled background depending on how its maximally isotropic subspaces are selected. In the group-manifold formulation, the obstruction to a global spacetime description is therefore not doubling itself but the interplay among the polarization, the discrete quotient \(\Gamma\), and the flux decomposition [0902.4032].

A standard misconception is that nonzero \(Q\)-flux automatically implies nongeometry. In the WZW analysis, \(Q\neq 0\) is present even when the physical background is globally geometric. The actual obstruction is whether the polarization subgroup is preserved by the discrete identifications: when \(R=0\), the quotient \(G_{\text{pol}}\backslash \mathcal{X}_{2d}\) exists globally only if \(\Gamma\) acts by automorphisms preserving the \(X^m\) subalgebra. Schulz formulates this as the requirement that \(Q\) be \(\Gamma\)-invariant. This sharply separates algebraic flux data from the global existence of a physical spacetime [1106.6291].

The nilmanifold reduction literature makes the same distinction in a classification with three cases. If the dual generators \(X^m\) form a subgroup preserved by the discrete quotient, the result is a geometric background. If they form a subgroup only locally, the result is a T-fold patched by \(O(d,d;\mathbb Z)\) transformations. If they fail to close to a subalgebra, no local quotient exists and the background is “essentially doubled”; this is the characteristic \(R\)-flux situation [1909.12348]. This suggests a precise hierarchy: geometric backgrounds are special polarizations of the doubled space, T-folds require patchwise polarizations, and \(R\)-flux backgrounds have no local physical slice at all.

A further generalization replaces the constant defining \(O(d,d)\) metric by a curved split-signature metric \(H_{MN}(X)\) on the doubled space. In that construction, any \((d,d)\)-signature metric defines a local \(O(d,d)\) structure, generalized diffeomorphisms close with covariant derivatives \(D_M\), and the section condition becomes
\[
H^{MN}D_M\otimes D_N=0.
\]
The flat \(\eta_{MN}\) of standard DFT is then a special case. This sharpens the distinction between the defining doubled metric \(H_{MN}\) and the DFT generalized metric \(\mathcal H_{MN}\), which encodes the physical metric and \(B\)-field after a section is chosen [1402.2513].

## 4. WZW realization: the \(SU(2)\) model and general groups

The \(SU(2)\) WZW model at large level \(n\) provides a controlled CFT realization of doubled geometry. Semiclassically it is string theory on \(S^3\) with
\[
r^2=n\alpha',\qquad
ds^2_{\text{phys}}=r^2\,ds^2_{S^3},\qquad
H=2n\alpha'\,\omega_{S^3},
\]
or equivalently, in left-invariant one-forms \(\lambda^m\),
\[
ds^2_{\text{phys}}=r^2\,\delta_{mn}\lambda^m\lambda^n,\qquad
H=n\alpha'\,\epsilon_{mnp}\lambda^m\wedge\lambda^n\wedge\lambda^p.
\]
The same background admits a T-fold description in which \(S^3\) is viewed as a Hopf fibration with a doubled Hopf circle, and a fully doubled description in which
\[
\mathcal{X}_6=\Gamma\backslash (SU(2)_1\times SU(2)_2),
\qquad
\pi(g_1,g_2)=g_1g_2^{-1}.
\]
In the diagonal polarization basis
\[
Z_m=\tfrac12(T_m^{(2)}+T_m^{(1)}),\qquad
X^m=\tfrac12(T_m^{(2)}-T_m^{(1)}),
\]
the doubled algebra has
\[
K_{mnp}=Q_{mnp}=2C_{mnp},\qquad
f_{mn}{}^p=R^{mnp}=0,
\]
so the doubled space carries nonzero \(K\)- and \(Q\)-flux but vanishing \(f\)- and \(R\)-flux [1106.6291].

Applying the Hull–Reid-Edwards formalism to this doubled group reproduces the physical metric and \(H\)-flux exactly. At the WZW point, with modulus \(r^2=n\alpha'\) and \(b=0\), the doubled Riemannian metric is
\[
ds^2_{\text{Riem}} = n\, d_{mn}(p^m p^n + q^m q^n),
\]
while the physical metric extracted from the chosen polarization becomes
\[
ds^2_{\text{phys}} = r^2\, d_{mn}\,\Lambda_{\text{phys}}^m\Lambda_{\text{phys}}^n.
\]
The flux computation yields
\[
H=dB_{\text{phys}} + C_{mnp}\,\Lambda_{\text{phys}}^m\wedge\Lambda_{\text{phys}}^n\wedge\Lambda_{\text{phys}}^p,
\]
which for \(SU(2)\) reduces to \(2n\alpha'\,\omega_{S^3}\). The same formalism also reproduces the effective scalar potential and the stabilization condition
\[
r^2=n\alpha',
\]
matching both sigma-model beta functions and exact CFT analysis [1106.6291].

The construction extends from \(SU(2)\) to any compact semisimple WZW group \(G\), again with doubled group
\[
\mathcal{G}_{2d}=G_1\times G_2,\qquad
\pi(g_1,g_2)=g_1g_2^{-1},
\]
and the same algebraic pattern
\[
K_{mnp}=Q_{mnp}=2C_{mnp},\qquad f=R=0.
\]
The physical metric and flux then take the universal form
\[
ds^2_{\text{phys}}=r^2\,d_{mn}\Lambda_{\text{phys}}^m\Lambda_{\text{phys}}^n,\qquad
H=C_{mnp}\,\Lambda_{\text{phys}}^m\wedge\Lambda_{\text{phys}}^n\wedge\Lambda_{\text{phys}}^p.
\]

## 5. Worldsheet formulations, dualities, and branes

The worldsheet implementation of doubled geometry is most explicit on doubled twisted tori. There the sigma model is written directly on the doubled target, together with a self-duality constraint that halves the doubled degrees of freedom. Gauging a maximally isotropic subgroup associated with a chosen polarization yields the conventional sigma model on the physical quotient. This construction accommodates reductions with duality twists, T-folds, and a class of flux compactifications, and it extends to cases where no local geometric target exists [0902.4032].

Nilmanifold reductions furnish a detailed family of examples. A nilmanifold \(N=G/\Gamma\) is doubled to \(\mathcal M=\mathcal G/\tilde\Gamma\) with \(\mathcal G=T^*G\), and different polarizations recover geometric nilmanifolds, tori with \(H\)-flux, T-folds, and \(R\)-flux backgrounds. When these nilmanifolds are fibered over an interval, the doubled geometry of the nilmanifold factor unifies special-holonomy spaces with intersecting brane solutions, exotic branes, and essentially doubled spaces; the various duals arise as different slices of the same doubled manifold [1909.12348].

The KK-monopole/NS5 system provides a complementary test because localization breaks the isometry required by standard Buscher rules. In doubled geometry, the compact direction \(\theta\) and its dual \(\tilde\theta\) are treated symmetrically. The formalism reproduces the smeared T-dual pair by choosing opposite polarizations, and it extends to the localized case through a conserved doubled current even when one side has broken isometry. The resulting KK background is localized in winding space: the physical fields depend on the dual coordinate, so the background is non-geometric from the ordinary spacetime viewpoint but natural in the doubled one [1106.1174].

D-branes are also naturally described in the doubled framework. In fully doubled WZW models, classical D-branes wrap maximal isotropic submanifolds \(G_b=bG_{\text{diag}}b^{-1}\subset G_1\times G_2\), and their projections reproduce the conjugacy classes and twisted conjugacy classes that support WZW branes. The corresponding worldsheet boundary condition
\[
J^m(z)=(\mathrm{Ad}_b)^m{}_n\,\bar J^n(\bar z)
\]
arises geometrically from the doubled embedding [1106.6291]. This suggests a general interpretation of branes as geometric objects in doubled space even when their projections are non-geometric.

## 6. Later developments and scope of the framework

Subsequent work broadened doubled geometry in several directions. A basis-independent geometry for DFT defined generalized torsion and curvature without reference to a particular frame, while a teleparallel reformulation then replaced curvature by a flat but torsionful Weitzenböck connection compatible with both the generalized metric and the \(O_{D,D}\) structure; the DFT action is recovered exactly from quadratic invariants built from generalized torsion [1303.6727, 1303.6727]. In the \(\alpha'\)-corrected setting, T-duality remains unchanged while diffeomorphisms and \(b\)-field gauge transformations are deformed; the resulting gauge algebra is an \(\alpha'\)-deformation of the duality-covariantized Courant bracket, and the gravitational field is promoted from a constrained generalized metric to an unconstrained double metric [1303.6727].

More algebraic realizations also clarify the framework. A \(0+1\)-dimensional rigid-rotator model on \(SU(2)\) and its Poisson–Lie dual on \(SB(2,\mathbb C)\) can be unified on the Drinfel’d double \(SL(2,\mathbb C)\), where an \(O(3,3)\) metric, a generalized metric, and an almost para-Hermitian structure make the doubled geometry entirely explicit. Gauging different symmetry subgroups of the doubled model recovers the original and dual systems, making section choice concrete in a finite-dimensional setting [1804.00744]. The same logic extends to the \(SU(2)\) principal chiral model, whose phase space carries a Born geometry and whose parent action on \(SL(2,\mathbb C)\) yields the \(SU(2)\) and \(SB(2,\mathbb C)\) sigma models as different polarizations [1903.01243].

A separate brane-based line of work probes doubled geometry through wrapping rules rather than sigma models. There, consistency of the solitonic sector with the doubled fundamental sector leads to a dual wrapping rule and suggests generalized Kaluza–Klein monopoles encoded by mixed-symmetry fields \(D_{6+n,n}\). This is not a complete doubled formalism, but it uses brane spectra to constrain the same T-duality-covariant structure [1106.0212].

A distinct usage of the term also appears in non-product spectral geometry, where a “doubled geometry” means a two-sheeted spectral triple with two metrics and an off-diagonal Dirac operator, leading to modified bimetric theories. This is mathematically separate from the \(O(d,d)\)-covariant string-theoretic framework described above [2202.03765]. The coexistence of these usages makes terminology potentially ambiguous, but in string theory and related arXiv literature the phrase usually denotes the \(O(d,d)\)-based formalism in which doubled coordinates, generalized metrics, and polarization encode T-duality and non-geometric flux backgrounds.

Source: https://www.emergentmind.com/topics/doubled-geometry-framework