17
$\begingroup$

Every year I teach an introductory class on the differential geometry of surfaces, including numerical aspects (e.g., how to solve PDEs on surfaces). Historically this class has included an introduction to exterior calculus, describing (for example) differential operators on curved surfaces in terms of the exterior derivative and the Hodge star.

I teach the class this way because that's how I myself learned differential geometry, but am starting to wonder how much value this approach really has for the students. Exterior calculus takes considerable effort to absorb, and students are often left wondering why we don't use simpler, more conventional language to discuss the same ideas.

Take the gradient, for example. One could say that the gradient of a scalar field $\phi$ on a curved surface is the unique vector field $\nabla\phi$ satisfying

$$ \langle \nabla\phi, u \rangle = D_u \phi, $$

where $\langle \cdot, \cdot \rangle$ is the Riemannian metric and $D_{\cdot}$ is the directional derivative. From there, it takes additional work to introduce the exterior derivative $d$ and the sharp operator $\sharp$ so that one can write the ever so slightly more succinct relationship

$$ \nabla \phi = (d\phi)^\sharp. $$

But what is really gained by having this slightly more succinct expression?

My question is an earnest one. I actually want to know, especially from people who do surface theory, why they bother to use exterior calculus and not a more elementary language. Or to put it another way:

  • Are there clear examples where it's beneficial to use exterior calculus, as opposed to simpler language (e.g., writing everything explicitly in terms of the inner product)?
  • How pervasive is exterior calculus in surface theory? E.g., roughly what percent of new results about surfaces would one estimate are expressed in the language of differential forms?
  • Likewise, how much are new results being written in a more classical language?
  • What did the historical development look like? I.e., why has exterior calculus become more (or less) popular over time?

In other words, since this is not my main area, I am trying to get some broad cultural knowledge about how pervasive this tool is, and how likely it is to benefit my students.

Thanks!

$\endgroup$
4
  • $\begingroup$ In my experience, the main point of contact between differential forms and Riemannian geometry is when one wants to integrate quantities related to curvature (e.g. Gauss-Bonnet). Expressing this directly in terms of the metric is hard in general because the curvature tensor is so complicated, but for surfaces it is very simple (the tangent space at each point contains only one plane). So my suspicion is that differential forms are not essential in the geometry of surfaces, but I would be delighted to be proven wrong. $\endgroup$ Mar 8, 2014 at 0:51
  • $\begingroup$ (Of course, differential forms may be more natural if one hopes to explore the topology of surfaces, thanks to de Rham cohomology.) $\endgroup$ Mar 8, 2014 at 0:52
  • 1
    $\begingroup$ Sorry, one last comment, which is really an elaboration on my first comment. One reason that differential forms are used to define classical operators like the gradient, curl, and divergence is that it allows you to deduce the classical integral theorems of vector calculus from the much more general Stokes' theorem for differential forms; for instance, the divergence theorem is just Stokes' theorem for $2$-forms on closed surfaces in $\mathbb{R}^3$ (via the Hodge operator). $\endgroup$ Mar 8, 2014 at 1:02
  • 2
    $\begingroup$ I have a book by Alois Svec called Global Differential Geometry of Surfaces. The preface is one page; he says "In my book, I am using the machinery of E. Cartan's calculus. It should be equivalent to the tensor calculus; nevertheless, using it I get better results (but, honestly, sometimes it is too complicated)." Anyway, the style is a simplified style of exterior differential systems; as Chern said, if you have an equation, differentiate it. $\endgroup$
    – Will Jagy
    Mar 8, 2014 at 2:53

2 Answers 2

18
$\begingroup$

Well, there's really not a whole lot more to say beyond what Deane already wrote. He certainly hit the main points, but maybe I can expand a bit on what he wrote and comment on my own experience over the years both learning and teaching differential geometry.

Maybe I should say a little bit about my own education: Originally, I learned differential geometry from O'Neill's Elementary Differential Geometry, which introduces differential forms and makes very good use of them. I went on to read a series of papers in differential geometry that used differential forms extensively, particularly many beautiful papers of S.-s. Chern and his students, and his postdoctoral advisor, the great Élie Cartan. So differential forms became very natural to me, but when I started teaching differential geometry, I found that they were always a little bit of a barrier to students, who had to take some time getting used to them. I experimented with teaching the curves and surfaces course without them, using, for example, do Carmo's wonderful book Differential Geometry of Curves and Surfaces, which avoids differential forms in favor of a classical vector calculus in local coordinates approach and manages to cover a lot of great material from the classical literature. Ultimately, though, I became convinced that this was not an efficient way to go, and have reverted to teaching the curves and surfaces course using differential forms, but spending a little extra time at the beginning to redevelop vector calculus using forms, which has its own benefits.

Your example using differential forms to get a minor improvement in notation does not give any indication of the efficiency of actually using differential forms (mainly because you haven't really used the exterior derivative). Here's a better example: The construction of the Gauss curvature invariant of a metric $g$ on a surface. Locally write $g = {\omega_1}^2 +{\omega_2}^2$ where $\omega_1$ and $\omega_2$ are a ($g$-orthonormal) basis of $1$-forms. There is then a unique $1$-form $\phi$ that satisfies $\mathrm{d}\omega_1 = -\phi\wedge\omega_2$ and $\mathrm{d}\omega_2 = \phi\wedge\omega_1$, and there is a unique function $K$ such that $\mathrm{d}\phi = K\ \omega_1\wedge\omega_2$. Then $K$ clearly depends on $2$ derivatives of the coframing $\omega_i$, but it turns out to depend only $g$ and not on the choice of coframing. Here is why: If $g = {\bar\omega_1}^2 +{\bar\omega_2}^2$, then $\bar\omega_1 = \cos\theta\ \omega_1 + \sin\theta\ \omega_2$ and $\bar\omega_2 = \pm(-\sin\theta\ \omega_1 + \cos\theta\ \omega_2)$ for some function $\theta$ on the domain of the coframing. Then a simple computation yields $\bar\phi = \pm(\phi+\mathrm{d}\theta )$, so $$\bar K\ \bar\omega_1\wedge\bar\omega_2 = \mathrm{d}\bar\phi = \pm \mathrm{d}\phi = \pm K\ \omega_1\wedge\omega_2 = K\ \bar\omega_1\wedge\bar\omega_2\,, $$ so $\bar K = K$ depends only on the metric $g$. (It's hard to imagine any construction of $K$ and proof that it is invariantly defined that starts with writing $$ g = E(x,y)\ \mathrm{d}x^2 + 2F(x,y)\ \mathrm{d}x\mathrm{d}y + G(x,y)\ \mathrm{d}y^2 $$ and goes through the process of defining the Christoffel symbols and then $K(x,y)$ and then proving that the result doesn't depend on the choice of coordinates that isn't considerably longer than this.)

Moreover, you get easy proofs of fundamental results: For example: If $K$ vanishes identically, then $\mathrm{d}\phi = 0$, so, locally $\phi = -\mathrm{d}\theta$ for some function $\theta$. Using this to define $\bar\omega_i$ as above, we get $\mathrm{d}\bar\omega_i = 0$, so $\bar\omega_i = \mathrm{d}x_i$ for some local functions $x_1$ and $x_2$, so $g = {\mathrm{d}x_1}^2 + {\mathrm{d}x_2}^2$, i.e., $g$ is locally flat. Similarly, there are easy proofs that $K\equiv 1$ implies that $g$ is locally isometric to the unit $2$-sphere, and lots of other such results.

Of course, as Deane pointed out, it was Cartan who advocated and popularized the use of differential forms in surface theory and throughout differential geometry. Beginning in the 1890s and continuing through the 1940s, he wrote many highly influential papers and books using differential forms that effectively demonstrated their efficacy. The number of results that he was able to derive using them and the efficiency with which he did it is still astonishing today, and it made an enormous impression on the differential geometers of the time. Throughout the 20th century, his followers continued to develop and apply the techniques in higher dimensions, but also in classical surface theory. Many results from the classical literature were rewritten in forms language and extended. A typical (and beautiful) example is the paper by Chern and Terng, An analogue of Bäcklund's theorem in affine geometry (Rocky Mountain Journal of Mathematics (1980), 105–124, in which they give a much shorter proof of Bäcklund's theorem (about Euclidean surfaces with $K=-1$) using differential forms and discover a corresponding version in affine geometry. (One can give lots of examples of this kind, of course.)

I have no way of reliably estimating the percentage of current research papers about curves and surfaces that use differential forms, but it is fairly high, simply because the language is very efficient; I can say that 100% of my own papers, even the curves and surfaces papers, use differential forms. I can't even imagine translating some of them out into classical vector calculus notation without them becoming much longer and essential unintelligible.

The great efficiency of using differential forms accounts for their popularity; it's a pervasive language throughout differential geometry now. The resistance to introducing differential forms is that there seems to be an extra level of abstraction beyond ordinary vector calculus, but part of that is caused by the common feeling that you have to introduce tensors and all kinds of abstract concepts in order to define them. However, that's not really so if they are introduced the right way. Once students understand that $1$-forms measure velocities, $2$-forms measure areas, $3$-forms measure volumes, etc., they learn the rules pretty quickly and come to appreciate that the exterior algebra keeps track of what are otherwise messy formulas involving determinants and cross-products, etc. Anyone who has struggled to memorize the formulas for div, grad, and curl in cylindrical and spherical coordinates can appreciate the simplicity of the exterior derivative.

$\endgroup$
12
$\begingroup$

We all should wait for Robert Bryant to answer this question since he's by far the most qualified person to do so.

But let me just say that differential forms are useful not only in global differential geometry but also local differential geometry. One reason for this is that the Maurer-Cartan equations for a Lie group have a beautiful clean formulation using the exterior derivative. This in turns leads very naturally and easily to the structure equations for the geometry of a Lie group and more generally manifolds with geometric structures based on the Lie group. The basic technique is to restrict to an appropriate bundle of frames (bases of tangent vectors) that reflect the geometry. On such a frame bundle there are canonically defined differential forms satisfying structure equations that come from the Maurer-Cartan equations. These in turn lead to a straightforward way to identify local invariants of the geometric structure and the equations they satisfy. Then if you want to make assumptions about these invariants and explore the consequences, this is often easier in this approach because one does not have to worry about artifacts arising from local co-ordinates such as Christoffel symbols.

This overall approach was developed by Elie Cartan, who called it "moving frames". It was used quite effectively by Chern and subsequently Robert Bryant.

One of my favorite papers about this is:

Griffiths, P. On Cartan's method of Lie groups and moving frames as applied to uniqueness and existence questions in differential geometry. Duke Math. J. 41 (1974), 775–814.

Finally, the elementary textbook by O'Neill does teach the differential geometry of surfaces in 3-space using moving frames. It is initially more abstract and difficult than other books such as the one by Do Carmo, but if you have motivated students who are wiling to take it on, they will be amply rewarded later in the course.

$\endgroup$
7
  • 4
    $\begingroup$ Well, I'm touched by the vote of confidence, Deane, but I'm at a weekend meeting in Chicago that doesn't give me much time to write the thoughtful response that this (series of) question(s) require[s]. Having taught curves and surfaces using differential forms and having also taught it without, I think I understand the issues and the OP's concerns, but (naturally) I do still come down on the side of introducing differential forms, even for the undergraduate curves and surfaces course. In fact, I believe that vector calculus itself could be taught effectively using differential forms. $\endgroup$ Mar 8, 2014 at 14:14
  • $\begingroup$ I agree with Robert's view on teaching vector calculus using differential forms. There is even a book by Harold Edwards, Advanced Calculus: A Differential Forms Approach, that does this. $\endgroup$
    – Deane Yang
    Mar 8, 2014 at 16:51
  • $\begingroup$ Thank you Deane; thank you Robert. These answers are in line with what was my fuzzy view of the culture, and it's very nice to have these concrete reference points and examples. (I am also very tempted to use this proof of the theorema egregium in my class!) Thanks again. $\endgroup$ Mar 9, 2014 at 14:20
  • 3
    $\begingroup$ I am now teaching a grad. course Manifolds II' here at UCSC. 3/4 of the students did not have undergrad d.g. We focus on forms. Some of my arguments in favor of learning forms at a young age are contained in W. Burke's ``Applied Differential Geometry', in Cartan's Les systemes differentiaels exterieurs...', in the forms proof of Gauss-Bonnet, and Moser's deformation method for proving Darboux and that two vol. forms yielding the same vol. on a cpt manifold are diffeomorphic. $\endgroup$ Mar 13, 2014 at 5:55
  • $\begingroup$ Thanks Richard. Do you have a pointer to a good proof of Gauss-Bonnet using forms? I typically have the students prove the discrete analog (via angle defect), but have so far not found a nice, simple proof of the smooth version that uses forms. $\endgroup$ Mar 13, 2014 at 13:30

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.