Tensor Models and Group Field Theory

Tensor models turn index contractions into gluing instructions, while group field theories add group-valued geometric data to the same combinatorial machinery. Their Feynman graphs can be dual to cellular complexes and, for suitable interactions, to simplicial spin-foam histories. The exact combinatorial statement is powerful: an interaction fixes which faces are identified. It does not imply that the dominant large-$N$ graphs form a smooth four-dimensional Lorentzian spacetime or obey Einstein dynamics.

Helpful background. Flat-Spacetime Quantum Field Theory supplies Feynman expansions and renormalization; Einstein Equations supplies the target effective action; Quantum Information supplies tensors and entanglement; Quantum Gravity supplies evidence standards; Causality and Global Structure supplies topology and signature; and Computational Relativity supplies finite-size and phase diagnostics.

Contractions as colored gluings

Let $T_{a_1\ldots a_d}$ be a rank-$d$ complex tensor, with each index transforming under its own $U(N)$ factor. In the invariant single-tensor formulation, an interaction contracts indices color by color between $T$ and $\bar T$; its bipartite contraction graph is often called a bubble. A related simplicial colored formulation uses $d+1$ distinct field species, an interaction containing one of each species, and its complex conjugate. The two conventions share stranded Feynman combinatorics but should not be conflated.

For example, the rank-three simplicial colored interaction

$$I_{\mathrm{tet}}= \sum_{\substack{a,b,c;\,d,e,f}} T^{(0)}_{abc}\, T^{(1)}_{ade}\, T^{(2)}_{fbe}\, T^{(3)}_{fdc} +\text{complex conjugate}$$

shows the rule directly. Every pair of the four colored tensors shares exactly one index: $a,b,c,d,e,f$ label the six pairings. The contraction graph is the complete graph $K_4$, dual to the four triangular faces of a tetrahedron. Propagators identify a tensor with its conjugate color by color, gluing tetrahedra along matched faces. Changing the contraction pattern changes the cellular complex even when the polynomial degree is unchanged.

Rank two is the instructive limit: a matrix propagator has two strands, and ribbon Feynman graphs define cellular decompositions of orientable surfaces. Cubic vertices are dual to triangles; other polynomial interactions generally give other polygons. At higher rank, coloring controls identifications and makes a topological expansion possible, but arbitrary graphs may represent pseudomanifolds or singular gluings rather than manifolds.

The large-N expansion

With the interaction and covariance scaled appropriately, a connected vacuum graph $\mathcal G$ of a rank-$d$ colored tensor model has amplitude

$$\mathcal A_{\mathcal G}\propto N^{\,d-\frac{2}{(d-1)!}\omega(\mathcal G)},$$

where the nonnegative Gurau degree $\omega$ generalizes ribbon-graph genus. Gurau derives this arbitrary-rank organization in Gurau 2012, Eq. (16), Definition 2, and Lemma 4. Graphs with $\omega=0$ dominate as $N\to\infty$; in the simplest invariant models, their recursive melonic structure and critical behavior are established in Bonzom et al. 2011, §§ III–V and Propositions 3–5.

This is a controlled combinatorial result, not yet a gravitational continuum. In basic models the melonic continuum has branched-polymer-like properties rather than a four-dimensional extended geometry. For the quartic melonic model class in dimension $d\lt6$, a correlated double-scaling limit retains selected subleading graph families Bonzom et al. 2014, §§ IV.A–IV.C. Other interactions and multicritical limits require their own continuum observables and universality analysis. A large number of simplices does not by itself make a local metric.

Group field theory adds geometry

A rank-$d$ group field is a function

$$\varphi:G^d\longrightarrow\mathbb C, \qquad \varphi(g_1h,\ldots,g_dh)=\varphi(g_1,\ldots,g_d),$$

where diagonal right invariance imposes a closure or gauge constraint. Its action has a kinetic kernel and nonlocal interaction,

$$S[\varphi,\bar\varphi] =\int\bar\varphi\,K\,\varphi +\lambda\int V\,\varphi^{d+1} +\bar\lambda\int\bar V\,\bar\varphi^{d+1}.$$

For $d=4$, each field quantum can represent a tetrahedron and a five-field simplicial interaction glues five tetrahedra into a four-simplex. Harmonic expansion on $G$ turns group data into representation labels and intertwiners. With suitable kernels and simplicity restrictions, a GFT Feynman amplitude reproduces a spin-foam or lattice-BF amplitude on the generated two-complex Oriti 2006, §§2–4.

This relation has two directions. A chosen spin-foam vertex can define a GFT interaction, so the perturbative expansion sums complexes built from that vertex. Conversely, GFT renormalization treats changing couplings and operators under scale transformations. Neither direction proves that the perturbative sum converges or that its continuum phase is Einsteinian.

Condensate states posit a collective mean field $\sigma(g_I)=\langle\widehat\varphi(g_I)\rangle$ and derive Gross–Pitaevskii-like equations. With isotropy, weak interactions, and a relational matter clock, selected models yield effective Friedmann equations and bounce-like solutions Oriti, Sindoni, and Wilson-Ewing 2016, §§3–5. These are many-body truncations. Connectivity fluctuations, anisotropies, interactions, the clock, and the map to local metric observables remain part of the approximation.

Results and open questions as of 2026-08-20

Programme or position	Strongest result and source	Regime, adverse result, or missing step
Colored invariant tensor models	Exact $1/N$ organization by Gurau degree and a melonic leading critical sector; Gurau 2012, Eq. (16), Definition 2, and Lemma 4, Bonzom et al. 2011, §§ III–V and Propositions 3–5	Basic leading continuum is branched-polymer-like, not demonstrated four-dimensional Einstein geometry
Double-scaled tensor models	Selected subleading graphs survive correlated $N$ and coupling limits in the quartic melonic class for $d\lt6$; Bonzom et al. 2014, §§ IV.A–IV.C	The summability range and result are model-specific; signature, causality, and local gravitational observables remain unspecified
Simplicial group field theory	Feynman diagrams generate cellular complexes and amplitudes can equal declared BF/spin-foam models; Oriti 2006, §§2–4	Perturbative sum, simplicity sectors, renormalization, and continuum limit remain open
Tensorial GFT renormalization	Renormalizable examples and nontrivial flow structures exist for selected groups and kernels; Carrozza 2016, §§3–5	Fixed points and exponents are model- and truncation-dependent; relation to four-dimensional Lorentzian gravity is conditional
GFT condensate cosmology	Effective Friedmann and bounce-like dynamics arise in controlled homogeneous condensate truncations; Oriti, Sindoni, and Wilson-Ewing 2016, §§3–5	Symmetry reduction, clock, weak-correlation ansatz, and microscopic-state interpretation prevent a generic full-theory inference

The permissive view treats the $1/N$ expansion, spin-foam generating functional, renormalization flow, and condensate hydrodynamics as successive scales of one many-body quantum geometry. The conservative view accepts each mechanism but asks for one controlled trajectory connecting microscopic couplings to a universal phase with causal four-dimensional observables. A skeptical view takes melonic or branched-polymer dominance as evidence that simple interactions select the wrong geometry. The continuum relation remains open; parallel mathematical control does not make all extrapolations equally strong.

An adverse check is to replace one repeated index in $I_{\mathrm{tet}}$ by a fresh index. One pair of triangles is no longer identified, the $K_4$ contraction graph loses an edge, and the interaction no longer encodes a closed tetrahedron. Thus combinatorics determines gluing exactly, but geometry depends on choosing and dynamically weighting the right combinatorics.

The conclusions could change after a controlled large-$N$/continuum limit yields extended four-dimensional causal geometry, a GFT fixed point survives enlarged operators and regulators, condensate predictions survive connectivity and anisotropy corrections, a no-go result excludes the relevant universality class, or evidence after 2026-08-20 changes the comparison.

Exercises

In $I_{\mathrm{tet}}$, list which pair of tensors shares each index. Then replace the final $d$ in $T^{(3)}_{fdc}$ by a new index $h$ and identify what gluing fails.

Solution

The six pairings are

$$a:(0,1),\quad b:(0,2),\quad c:(0,3),\quad d:(1,3),\quad e:(1,2),\quad f:(2,3).$$

Every pair among $0,1,2,3$ occurs once, so the contraction graph is $K_4$. Replacing $T^{(3)}_{fdc}$ by $T^{(3)}_{fhc}$ leaves $d$ appearing only in tensor 1 and $h$ only in tensor 3; the $(1,3)$ contraction disappears and both indices are open unless additional tensors absorb them. The polynomial may still be written, but it is no longer the closed colored invariant dual to a tetrahedron. This local check establishes the gluing rule, not the continuum phase selected by a sum of such vertices.

References

• Bonzom, Valentin, Razvan Gurau, Aldo Riello, and Vincent Rivasseau. “Critical Behavior of Colored Tensor Models in the Large $N$ Limit.” Nuclear Physics B 853 (2011): 174–195. DOI. Open PDF.
• Bonzom, Valentin, Razvan Gurau, James P. Ryan, and Adrian Tanasa. “The Double Scaling Limit of Random Tensor Models.” Journal of High Energy Physics 2014 (2014): 051. DOI. Open PDF.
• Carrozza, Sylvain. “Flowing in Group Field Theory Space: A Review.” SIGMA 12 (2016): 070. Open review.
• Gurau, Razvan. “The Complete $1/N$ Expansion of Colored Tensor Models in Arbitrary Dimension.” Annales Henri Poincaré 13 (2012): 399–423. DOI. Open PDF.
• Oriti, Daniele, Lorenzo Sindoni, and Edward Wilson-Ewing. “Emergent Friedmann Dynamics with a Quantum Bounce from Quantum Gravity Condensates.” Classical and Quantum Gravity 33 (2016): 224001. Open manuscript.
• Oriti, Daniele. “The Group Field Theory Approach to Quantum Gravity.” In Approaches to Quantum Gravity (2009). Open review.