To be clear: What I am actually talking about is a nonlinear operator on a finitely generated vector space V with dimension $d(V)\;\in \mathbb{N}>1$. I can think of several nonlinear operators on such a vector space but none of them have the requisite properties of a group. In particular, but not exclusively, are there any such nonlinear operator groups that meet the definition of a Lie Group.

  • 1
    $\begingroup$ Are you asking "are there any non-linear operators from $V$ to $V$ that are group homomorphisms?" or are you asking "are there any collections of non-linear operators from $V$ to $V$ that form a group?" ? $\endgroup$ May 25 '13 at 20:40
  • $\begingroup$ I'm asking if there are any collections of nonlinear operators from V to V that form a group. $\endgroup$
    – Mr X
    May 25 '13 at 20:51
  • $\begingroup$ Does "nonlinear operator" have any particular meaning here I'm not aware of, or is it just an arbitrary map (of sets)? Pick your favourite three points in V, say a, b and c, and permute them around. (This is just a non-linear action of $S_3$ on V - you can do this with any group you like on any V that has enough points.) $\endgroup$
    – Billy
    May 25 '13 at 21:18
  • $\begingroup$ Yes it does, billy. If V is our n-dimensional vector space,f is a mapping from V to V, and A & B are any 2 vectors in V,then f is linear iff f(A+B)=f(A)+f(B) which is called the superposition principle. A nonlinear map f from V to V is one where f(A+B)!=f(A)+f(B). $\endgroup$
    – Mr X
    May 25 '13 at 21:52
  • $\begingroup$ Now posted on MathOverflow. $\endgroup$ Jun 4 '13 at 13:44

As some guy once said, dividing operators into linear and nonlinear is like dividing the world into bananas and nonbananas. If you're ignoring the vector space structure on $V$ then it's just a set, and you're just asking for a group action. It is easy to write down group actions, even Lie group actions (see homogeneous space), on a vector space that just don't happen to be linear. Probably the simplest example is $\mathbb{R}$ acting on itself by translation.

  • $\begingroup$ Based on Mr X's clarification in the comments, the $\mathbb{R}$ action on itself by left translation is "linear". (Mr X, perhaps nonstandardly, doesn't mention scalar multiplication!). That said, examples still abound. Pick your favorite nonlinear automorphism $f$ (in whatever category you wish, so long as the category contains the linear maps). Then conjugating $Gl(V)$ by $f$ will givea group of nonlinear things. $\endgroup$ May 26 '13 at 4:42
  • $\begingroup$ Er... no it's not? $\endgroup$ May 26 '13 at 4:45
  • $\begingroup$ Wow - I really shouldn't be doing math after midnight I apologize! I can delete my whole rant if you'd prefer. (And that would be my preference, but whatever.) $\endgroup$ May 26 '13 at 5:06
  • $\begingroup$ Thanks, Jason DeVito! And Quiaochu Yuan, I was asking for examples of nonlinear transformations on a vector space $V^{n}\;$ where n > 1. I'm aware that a nonlinear maps of the form $f:\mathbb{R}^{1} \to \mathbb{R}^{1} \;$ form a translation group. $\endgroup$
    – Mr X
    Jun 5 '13 at 14:36
  • $\begingroup$ Examples are just as easy to write down in the higher-dimensional case. Jason DeVito's comment works in all dimensions. $\endgroup$ Jun 5 '13 at 18:18

Oh I've got one....

Let A by a non-singular, invertible square matrix whose elements are components of vectors in $V^{n}$. Now consider the operator $T(\hat{v}) = e^{A\hat{v}}, \; \;\forall \hat{v} \in V^{n}$. So if $\alpha,\beta$ are any such operators with corresponding invertible non-singular matrices (A,B) respectively, than the binary group operation is $\phi(\alpha,\beta) = e^{AB\hat{v}} = (e^{A\hat{v}})^{B}$ where AB is the product matrix A*B. The inverse operator is $T^{-1}(\hat{v}) = A^{-1} ln(\hat{v})$.The identity mapping is $\hat{I}(\hat{v})= e^{ln(\hat{v})}=\hat{v}$.


How about the following construction. Let $M \in R_{d}(G)$ be a $d$ dimensional representation of some Lie group (e.g. $SL(2,R)$ and $V$ is some $d$ dimensional vector space. Now define: $$ f(x,M,\epsilon)= \frac{M x}{(1+\epsilon w.x)} $$ where the vector $w$ is chosen so that $w.M= w$ for all $M\in SL(2,\mathbb{R})$. Such vectors exist, for instance, if $d=4$ and $R_4(G)=R_2(G) \otimes R_2(G)$ then $w=\begin{pmatrix} 0 \\ 1\\ -1\\ 0 \end{pmatrix}$ is one such vector. It is now not too difficult to convince oneself that: $$ \begin{eqnarray} f(x,I,0) = x \\ f(f(x,M_1,\epsilon_1),M_2,\epsilon_2) = f(x,M_1 M_2,\epsilon_1+\epsilon_2) \end{eqnarray}$$ so that the mapping $f$ is nonlinear and satisfies the group property.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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