Results with short, advanced proofs or long, elementary proofs

Question

Recently I was preparing an undergrad-level proof of (a form of) the Jordan Curve Theorem, and I had forgotten just how much work is involved in it. The proof stored my head was just using Alexander duality plus some sanity-checks on the topology of the curve in question, which is a fine approach but does require a bit of algebraic topology machinery my audience didn't have access to. The more elementary proof (or at least the one I landed on) has a straightforward idea behind it, but turning that into a proper argument required slogging through quite a few details about polygons, regular neighborhoods, etc. Similarly, I couldn't help noticing just how much of a pain it is to prove the 2- and 3-dimensional versions of Stokes' theorem without some notion of manifolds, let alone the usual Stokes' theorem in some suitable setting.

Those are both elementary examples, but it got me thinking about the general topic of results that have very rough proofs from more elementary principles but have much clearer and smoother proofs with some more advanced background. Specifically, what are some examples of results from more advanced or narrow topics of mathematics can vastly simplify or explain in retrospect theorems that are encountered and proved laboriously in less specialized or more common areas of math? (If it helps clarify what I'm trying to get at, another example in my mind is May's "Concise Course in Algebraic Topology," which I think of as having the premise of, "So, now that you've gone through the standard intro algebraic topology course, here's what was secretly going on behind the scenes the whole time.")

Answer 1

The associativity of the group law on an elliptic curve can be proved in an elementary way by explicitly manipulating algebraic expressions, but this is not very enlightening. By using more advanced geometric ideas, one can prove associativity more conceptually.

Answer 2

A famous example is the Abel-Ruffini theorem.

In this video, Fields medallist Richard Borcherds introduces Galois Theory and its background, mentioning that the proof by Ruffini was quite cumbersome and not quite clear, while Abel's was neat and short:

Ruffini had a sort of 500 page proof of it, except no one's really quite sure whether it was a proof of it or not and they sort of suspect it wasn't, and a bit later Abel came along and gave a very clear 6-page proof of it.

Instead of more cumbersome, geometry-based traditional methods (see e.g. this enjoyable YouTube video by Veritasium covering the Italian history of the cubic equation), knowledge on Galois theory makes much easier to see that there is no solution to quintic equations and upwards, by the analogy that their corresponding Galois group is not solvable, i.e. can't be split into Abelian groups.

Answer 3

The famous example of this in my neck of the woods is Gelfand's slick proof, using Banach algebras (which should really be "Gelfand algebras"), of Wiener's theorem that the reciprocal of a nowhere-vanishing function on $\mathbb{T}$ with an absolutely summable Fourier series also has an absolutely summable Fourier series.

Answer 4

The number of spanning trees of an $n$-dimensional hypercube graph is $$2^{2^n-n-1}\prod_{j=1}^nj^{n\choose j}.$$ This is a quite tricky graph theory problem to try to prove directly. But it dies almost immediately to Kirchoff's matrix tree theorem, after computing the Laplacian eigenvalues using the discrete Radon transform.

Answer 5

Silver's proof that GCH cannot first fail at a singular cardinal of uncountable cofinality is a great example of this. The original one of Silver uses generic ultrapowers, so requires some nontrivial knowledge of forcing and comfortability with ultrapowers of models of set theory. For example, if we want to prove this for singular cardinals of cofinality $\omega_1$, then we take the forcing notion $P$ of stationary subsets of $\omega_1$, $G$ a $P$-generic over $V$, and so in $V[G]$ we can form the corresponding generic ultrapower $M$, from which we get an embedding $j:V\to M$. A careful, but quite short analysis gives the theorem. It's not too bad.

Alternatively, there is a longer proof that requires more bookkeeping, but requires only the most basic ideas about stationary sets and ultrafilters. A decent bit more work, but requires no knowledge of forcing, ultrapowers, working in different models of set theory, or embeddings.

For a little bit of context (though one would have to check the exact presentation), the proof using generic ultrapowers is less than a page in Jech, and the elementary proof is two and a half. Not an insane difference, but one I find memorable.

Answer 6

Resolution of singularity of curves (over algebraically closed fields). A somewhat long elementary proof is given e.g. in chapter 7 of Fulton's Algebraic Curves (which he made available online for free). Basically you use the primitive element theorem to reduce it to the planar case. And then for a planar curve you track the changes in multiplicity of a point under blow-ups via explicit computation. A more abstract approach, given for example in chapter I.6 of Hartshorne's, identifies a nonsingular curve with the set of discrete valuations of a function field, and shows that any projective morphism from an open subset of a nonsingular curve can be extended to the whole curve.

Answer 7

The original proof of the Stone-Weierstrass theorem was elementary enough that Stone could present it in a lengthy article in Mathematics Magazine. There is a much shorter proof due to de Branges that uses the Krein-Milman theorem, the Hahn-Banach theorem, and the Riesz representation theorem.

Answer 8

results that have very rough proofs from more elementary principles but have much clearer and smoother proofs with some more advanced background. Specifically, what are some examples of results from more advanced or narrow topics of mathematics can vastly simplify or explain in retrospect theorems that are encountered and proved laboriously in less specialized or more common areas of math

One alternative to proofs being long and complicated because they use only elementary methods is proofs being numerous because they use only elementary methods.

In this image,

• $AB$ is any chord of a parabola,
• $AC$ is parallel to the axis of the parabola,
• $BC$ is tangent to the parabola at $B.$

$D$ is the midpoint of $AC$ and we view it as the fulcrum of a lever $JB.$

$D$ is also the midpoint of $JB.$

$EH$ is any straight section of the triangle $ABC$ that is parallel to $AC.$

Proposition: If the weight of the segment $EH$ rests upon the lever at $G$, and the weight of the segment $EF$ rests at $J,$ then the lever is in equilibrium. (This is just the equation of the parabola.)

Therefore, if the weight of the region bounded by the triangle $ABC$ rests at its center of gravity $I$ and the weight of the region bounded by the curve and the chord $AB$ rests at $J,$ then the lever is in equilibrium.

Bottom-line conclusion: Since it is known that the center of gravity $I$ is one-third of the way from $D$ to $B,$ the area of the triangle is exactly three times the area of the region bounded by the curve and the chord.

Archimedes wrote that argument. Today we write $\int_0^1 x^2\,dx=\frac13$ and we use an antiderivative to show that. I don't think Archimedes used antiderivatives.

By similar methods Archimedes showed that the center of gravity of the northern hemisphere is $5/8$ of the way down from the north pole to the center of the earth. For this he considered the line through the center of the earth and the north pole to be a lever with its fulcrum at the north pole, and two of the infinitesimal weights that he had resting on it were a cross-section of the sphere parallel to the equator, and the cross-section in that same plane, of a cone whose apex is the north pole and whose base is bounded by the equator. (I don't remember the details but I think he let the cross-section of the sphere rest at its intersection with the lever and that smaller cross-section rest at a point whose distance above the north pole is the diameter of the sphere.)

In his "mechanical method" he proved 15 (thus "numerous" (to quote the term I used above)) propositions in geometry by doing things like these, finding areas and volumes and centers of gravity.

(I think maybe no one before Archimedes ever thought about centers of gravity.)

Answer 9

Perron-Frobenius theorem. Each of both version answers the question.

Weak version : If $A\in{\bf M}_n({\mathbb R})$ has non-negative entries, then the spectral radius $\rho(A)$ is an eigenvalue, associated with an eigenvector in ${\mathbb R}_+^n$.

The elementary proof is a consequence of of the strong version, applied to $A+\epsilon In$, passing to the limit as $\epsilon\to0+$. The short proof observes that the set $$K=\{x\in{\mathbb R}_+^n|\sum_ix_i=1,\,Ax\ge\rho(A)x\}$$ is non-void, convex, compact, and the application $$x\mapsto\frac1{\sum_i(Ax)_i}\,Ax$$ maps $K$ into itself (if the map is ill-defined then $\rho(A)=0$ and the result is trivial). Then apply Brouwer's theorem.

Strong version : If the entries are $>0$, then $\rho(A)$ is a simple eigenvalue and the eigenvector has $>0$ coordinates.

Short proof: the same map as above is a strict contraction of the simplex $$K=\{x\in{\mathbb R}_+^n|\sum_ix_i=1\}$$ for the Hilbert distance !

Answer 10

Brouwer's theorem is immediately obtained as a 'by-product' of the development of integration on manifolds. The 'elementary' alternative is to prove and use Sperner's Lemma, which (to me at least) seems a more tedious road.

Answer 11

I think a good example is the 1955/1958 Adian-Rabin theorem. This says that "given a finite presentation of a group, one can deduce almost nothing about the properties of the group". For example, the problem of deciding whether a group is trivial from its presentation is undecidable.

The first proof, by Adian, of this theorem is 60 pages long and uses many intricate -- but entirely elementary -- reasoning from combinatorial semigroup theory. By contrast, more modern proofs, which go via HNN-extensions, and which can arguably be called "advanced" (at the time), are much shorter, requiring little more than 2 pages to work out the full details (at least modulo the Novikov-Boone theorem).

Adian's proof, comprising 4 papers, can be found in English translation here. This also contains my summary of the articles and some historical background.

It is worth noting that both Adian's and Rabin's proofs use HNN-extensions implicitly, and the three proofs (Adian, Rabin, modern) are all the same in spirit -- the toolbox of general results available to a modern proof makes this more advanced as well as shorter.