Different representations of the Lorentz group in infinite-dimensional vector spaces are related to the types of particles (classification by spin and mass). These representations are linear. Are non-linear representations of groups used in physics? I would like to know examples of non-linear representations.

  • $\begingroup$ I thought that only linear actions of groups were considered to be representations. Are we using a different definition of "representation" here? $\endgroup$
    – Galen
    Mar 23, 2022 at 2:46

2 Answers 2


Yes they are. In particular in Quantum Field Theories non-linear representation of groups appear in the context of spontaneous symmetry breaking.

You start from a Lagrangian with a symmetry group $\mathcal G$ that is then broken to a subgroup $\mathcal H$. The new Lagrangian after symmetry breaking is then linearly invariant under $\mathcal H$, but it still contains the symmetry from the full group $\mathcal G$, it's just more hidden. In particular some fields will transform non-linearly.

As an explicit example (not the simplest, but for which I already have the results) consider a $SO(3)$ invariant theory with a field $\Phi$ that takes a vev $f$ in the $3$ direction, breaking the symmetry to $SO(2)$.

As usual, you parametrize $\Phi$ as a radial mode $\sigma$ plus two Goldstones modes $$ \Phi(x) = e^{i \frac{\sqrt{2}}{f} \Pi^i(x) T^i} (0,\,0,\,f+\sigma(x)) $$ where $i=1,2$ and $T^i$ are the "broken" generators.

The theory you get in terms of $\sigma$, $\Pi_1$ and $\Pi_2$ is still invariant under the full $SO(3)$ group:

the SO(2) subgroup just linearly mixes the two $\Pi$ fields, while transformations along the broken generators act as

$$ \Pi^i \to \Pi^i + |\Pi| \cot \frac{|\Pi|}{f}\alpha^i + \left( \frac{f}{|\Pi|} -\cot\frac{|\Pi|}{f} \right) (\alpha_j \Pi^j) \frac{\Pi^i}{|\Pi|} $$ where $\alpha$ are the parameters of the transformations and $|\Pi| = \sqrt{(\Pi^1)^2+(\Pi^2)^2}$

This transformation is clearly non-linear and in particular it implies a shift-symmetry under $\Pi \to \Pi + c$, whose consequence is also that the $\Pi$ must be massless.

Similar constructions can be also done for the Poincaré group. For example in 1405.7384 they show that the action for a point particle can be derived from a non-linear realization of the broken Poincaré group.

For more infos:

  • Weinberg QFT vol 2, the chapter on spontaneous symmetry breaking
  • CCWZ constructions (explained in Weinberg, but also reviewed in chapter 2 of 1506.01961 from which I took this example)
  • $\begingroup$ Isn't a "non-linear representation of a group" a contradiction in terms? I thought that representations of a group were linear actions of that group by definition. Can you clarify? $\endgroup$
    – Galen
    Mar 23, 2022 at 2:49
  • $\begingroup$ @DifferentialCovariance True. A representation is linear by definition. I've heard this called a "realization", meaning that it only concretely implements the product rule on some space. With this definition a representation is a linear realization. $\endgroup$
    – FrodCube
    Mar 23, 2022 at 9:21
  • $\begingroup$ Does this non-linear group realization come from the Baker-Campbell-Hausdorf formula for SO(3)? $\endgroup$
    – Craig
    Mar 20 at 2:05

In general, if $G$ is a group and $M$ is some set we can define a left $G$-action to be a map $\rho:G\times M\to M$ such that $$\rho(e,x)=x,\quad \forall x\in M,\\ \rho(g,\rho(h,x))=\rho(gh,x),\quad \forall g,h\in G, x\in M$$

where $e\in G$ is the identity. In general we just denote $\rho(g,x)=g\cdot x$ and then the two properties are $$e\cdot x=x,\quad \forall x\in M,\\ g\cdot(h\cdot x)=(gh)\cdot x,\quad \forall g,h\in G,x\in M.$$

A right $G$-action can be defined similarly. This is a very general concept in which the elements of a group are realized through the map $\rho$ as transformations acting in a particular set of objects. Notice there is no extra assumptions on $G$ or on the set of objects $M$.

It can be the case that $M$ is a vector space. In that case you could consider one specific case of this definition, in which the map $x\mapsto \rho(g,x)$ for $g\in G$ fixed is linear. In that case, the elements of $G$ are realized as linear transformations. This makes sense now because the set on which $G$ acts has more structure and we can bring that structure into play. This is nothing more than a group representation. In that case a group representation is one specific case of a group action.

And indeed there are cases in Physics where one considers actions which do not act linearly. Now suppose that $G$, apart from a group, is also a smooth manifold, such that the operations of multiplication and inversion are smooth. We then say that $G$ is a Lie group. Suppose further that $M$ is also a smooth manifold. In that case we may consider a left $G$-action $\rho:G\times M\to M$ which now is supposed to be smooth.

This finds its place in Physics for example in the subject of non-abelian gauge theory. In standard QFT when we have a non-abelian gauge theory with structure group $G$ one often just says that the gauge potential is just one adjoint-valued field $A_\mu^a(x)$. The fact is that there is more to it.

Indeed, given the structure group $G$ one may consider what one calls principal $G$-bundles over $M$. I won't dive into the technical definition here (if you want to get the details I suggest the book Modern Differential Geometry for Physicists by Isham), but I will just say that a principal $G$-bundle is one particular kind of smooth manifold $P$ with a projection map $\pi:P\to M$, which carries a $G$-action such that $M\simeq P/G$. Now, when we have a principal $G$-bundle we may talk about connections on said bundle.

When $P$ is trivial, meaning that $P\simeq M\times G$ with a trivial right $G$-action $(x,g)\cdot h=(x,gh)$ a connection can be specified by one adjoint-valued field $A_\mu^a(x)$. That is the standard QFT story. But when $P$ is non-trivial, the gauge field cannot be specified anymore by a single $A_\mu^a(x)$. In that case, the most accurate way of thinking of the gauge field is as a connection on a principal $G$-bundle. In particular, the topology of the principal bundle is in a sense part of what specifies the gauge field, inasmuch as in gravity the spacetime topology is part of what specifies the metric. Matter fields then are sections of the so-called associated bundles constructed by representations of $G$.

That story (which obviously cannot be told in full in a single answer) is just meant to call your attention to the fact that non-linear group actions are central to non-abelian gauge theory, since they enter the very definition of the main object in these theories: namely the principal $G$-bundle over which the gauge field is meant to be a connection.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.