5
$\begingroup$

The following construction is standard, and it deserves a name.

Suppose we need to construct a diffeomorphism from a manifold $M$ to itself with some additional properties. Observe that the flow $\phi^t$ for any reasonable vector field $v(t)$ on $M$ defines a diffeomorphism for any $t$. So it remains to find a vector field $v(t)$ such that the flow $\phi^1$ satisfies your properties.

Do you know a name for it?

If not, what would be a good name?

$\endgroup$
8
  • 6
    $\begingroup$ Physicists call this type of thing "infinitesimal generator". $\endgroup$ Apr 10, 2020 at 19:50
  • 1
    $\begingroup$ @WillieWong, usually not, but sometimes it is useful. $\endgroup$ Apr 10, 2020 at 21:35
  • 3
    $\begingroup$ I only ask because after I upvoted @TobiasFritz's comment, I begin to doubt whether the usual use of the "infinitesimal generator" allows for $v$ to be time dependent. $\endgroup$ Apr 10, 2020 at 21:57
  • 3
    $\begingroup$ You see people give these flows various names. I use this construction in the time-dependent context. My vector fields are usually defined by an isotopy of a submanifold, you then extend the velocity vector field of the submanifold to the ambient manifold. So I just call it isotopy extension. But I see other people call them "point pushing" or "circle pushing" maps depending on what the submanifold might be. Your context is a little more general than my own, though. $\endgroup$ Apr 11, 2020 at 0:19
  • 3
    $\begingroup$ I would say "Moser trick", since it reminds me of the trick explained below for volume forms. There is still not too much named after Moser, so I don't think it would confuse anyone. "Euler trick" or "Arnold trick" or "Newton trick" might confuse people. $\endgroup$
    – Ben McKay
    Apr 11, 2020 at 14:14

2 Answers 2

6
$\begingroup$

This is often called "path method". In particular one has classically the Moser path method.

Let us prove for example (after Moser) that given a closed oriented $n$-manifold $M$ and two volume forms $\omega_0$, $\omega_1$ on $M$ with the same integral, there exists a diffeomorphism $f$ of $M$ such that $$f^*(\omega_1)=\omega_0$$ To this end, the path method considers $$\omega_t=(1-t)\omega_0+t\omega_1$$ and looks for a time-dependant vector field $X_t$ whose flow $\phi_t$ satisfies the condition $$\phi_t^*(\omega_t)=\omega_0$$ (Then, for $t=1$, $f=\phi^1$ will work)

The condition holds trivially for $t=0$. Deriving this condition with respect to $t$, one finds $$\phi_t^*(L_{X_t}\omega_t+\omega_1-\omega_0)=0$$ where $L$ is the Lie derivative. This amounts to $$L_{X_t}\omega_t=\omega_0-\omega_1$$ i.e. by Cartan's formula $$d\iota_{X_t}\omega_t=\omega_0-\omega_1$$ Since $\omega_0$ and $\omega_1$ have the same integral, they are cohomologous: one has a $(n-1)$-form $\alpha$ on $M$ such that $$\omega_0-\omega_1=d\alpha$$ Hence it is enough that $$\iota_{X_t}\omega_t=\alpha$$ The end of the argument is purely (multi)linear algebra: this last equation admits for every time $t$ and at every point $x$ a unique solution $X_t(x)$ since $\omega_t(x)$ is a nonzero $n$-form on $T_xM$. An analogous method applies to symplectic forms, nonsingular closed $1$-forms and contact forms. The path method also proves the $MJ^2$ Lemma (Lemma 3.2 in F. Laudenbach, A proof of Reidemeister-Singer's theorem by Cerf's methods. Ann. Fac. Sci. Toulouse Math. (6) 23 (2014), 1, 197–221, Arxiv 1202.1130); and in particular the Morse Lemma.

$\endgroup$
4
  • 1
    $\begingroup$ What is the $MJ^2$ Lemma? $\endgroup$
    – abx
    Apr 11, 2020 at 9:19
  • $\begingroup$ Thank you, but maybe "Moser trick" is better than "path method" (?) $\endgroup$ Apr 11, 2020 at 16:34
  • $\begingroup$ As you like it, but to me, this deserves the name of "method", being more interesting, more general, and involving more the nature of things, than a mere "trick". $\endgroup$ Apr 12, 2020 at 16:32
  • $\begingroup$ why "path" --- say "flow method" would be better. $\endgroup$ Apr 13, 2020 at 5:36
0
$\begingroup$

This is a hugely studied field in the area of medical image processing. Goes by name of (large) diffeomorphic registration or diffeomorphic mapping

https://en.m.wikipedia.org/wiki/Large_deformation_diffeomorphic_metric_mapping

$\endgroup$

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.