9
$\begingroup$

In many areas of math, I've been surprised at how much research, past and present, focuses on second order 'things'. Examples:

  1. Number theory: quadratic reciprocity, quadratic number fields
  2. Analysis: Second-order PDE's, Lagrangian equations, Newton's laws
  3. Geometry: conic sections, quadric surfaces, quadratic forms
  4. Topology: Surfaces
  5. Calculus: Second-derivative test

And many more. Introductory PDE classes and number theory classes may spend the entire semester on quadratic things.

Why is this so common? Is it because we can't handle third-order things, or is it because second order things are more important/common than higher-order things?

$\endgroup$
  • 3
    $\begingroup$ After the first-order case, the second-order case (rather than say the seventeenth order case) seems a natural place to continue. Are you asking why interest declines after the second-order case (if indeed this should be the case, which I much doubt it does in many cases)? Of course in cases where first-order is the only thing ever studied, it is unlikely to be called first-order. $\endgroup$ – Marc van Leeuwen Jan 14 '14 at 12:52
  • 1
    $\begingroup$ My naive thought is that "dimension one" is easy, and "dimension two" is the most easy among all that difficult dimensions greater than $1$. $\endgroup$ – Matemáticos Chibchas Jan 14 '14 at 13:03
  • 3
    $\begingroup$ So in the last 3000 years mathematicians haven't really counted past 2 in most areas? $\endgroup$ – Brian Rushton Jan 14 '14 at 13:07
  • $\begingroup$ Calibrate your surprise against the prevalence of studying second-order things against the prevalence of studying first-order things! Or even against zeroth-order things! $\endgroup$ – user14972 Jan 14 '14 at 13:51
  • 2
    $\begingroup$ Because every $n>1$ is a second order for $n-1$. Long live induction!! :-) $\endgroup$ – Asaf Karagila Jan 14 '14 at 14:50
6
$\begingroup$

My numbering below does not correspond to the original numbering in the question.

  1. In many areas of math homogeneous polynomials appear. Diagonal homogeneous polynomials, like diagonal matrices, have a particular simple form. Most matrices are not diagonal, and there are some important theorems in linear algebra that give conditions under which a matrix can be diagonalized by a linear change of variables (spectral theorems). For homogeneous polynomials, those of degree $1$ are diagonal by definition and it's a basic theorem that all homogeneous polynomials of degree $2$ over any field (not of characteristic 2) can be put into diagonal form by a suitable linear change of variables over that field. This is false in every degree greater than $2$. See https://mathoverflow.net/questions/77702/transformation-of-a-cubic-form for an example of a cubic form that is not diagonalizable over the rational numbers. Because a quadratic form in two variables can be diagonalized by completing the square, you could think of diagonalizability of quadratic forms in more than two variables as a generalized form of completing the square, but I think a better way to look at this is geometric: the reason quadratic forms (outside of characteristic $2$) can be diagonalized over their field of definition is because of the existence of reflections in orthogonal groups. That leads in to our next topic.

  2. In group theory, a rich source of groups comes from orthogonal groups of quadratic forms. There is no problem in extending this definition to higher degree: for any homogeneous polynomial $f(\mathbf x)$ in $n$ variables over a field $K$, define the orthogonal group of $f$ over $K$ to be all $A \in {\mathrm GL}_n(K)$ such that $f(A\mathbf x) = f(\mathbf x)$. Jordan proved that for $f$ of degree greater than 2 and "nonsingular" in a suitable sense, and $K$ of characteristic 0, the orthogonal group of $f$ over $K$ is finite. For instance, if $f = x_1^d + \cdots + x_n^d$ and $d > 2$ then the orthogonal group of $f$ over the complex numbers consists of the permutations of the $n$ variables, multiplication of each variable by a $d$th root of unity in $\mathbf C$, and compositions of these operations. For me, this finiteness of orthogonal groups in degree above $2$ is a vague reason why classical geometry is governed by degree $2$ objects. Without an infinite orthogonal group there are not enough interesting transformations, at least algebraically. Moving from algebra to analysis, the Morse lemma provides an important role for quadratic forms in topology.

  3. In physics, I've always thought that most important differential equations are second-order essentially because Newton's laws connect the physical concept of force to acceleration, which is a second derivative. Newton's first law shows force in general couldn't be directly related to velocity, a first derivative, and the next derivative turns out to be what we need. There are some interesting differential equations of physical interest are higher than second-order, such as the KdV equation.

  4. Quadratic fields are popular in number theory because they are computationally tractable and can be described in a more or less uniform way. One nice feature of quadratic fields is that every order in a quadratic field can be described in the computationally concrete form ${\mathbf Z}[\alpha]$ but this is false for any number field of degree greater than $2$: if $[K:\mathbf Q] > 2$ then there are orders in $K$ that are not of the form ${\mathbf Z}[\alpha]$. Also, in any number field $K$ of degree greater than 2 there are infinitely many ${\mathbf Z}$-lattices in $K$ that are not invertible as a fractional ideal for their ring of multipliers. That is, if $L$ is a ${\mathbf Z}$-lattice in $K$ then $L$ is a fractional ideal for the ring $R(L) = \{x \in K : xL \subset L\}$ and infinitely many such $L$ are not invertible $R(L)$-fractional ideals. That never happens for $\mathbf Z$-lattices in quadratic fields and is related to the fact that all orders in quadratic fields have the form ${\mathbf Z}[\alpha]$. Historically, quadratic fields were first studied in number theory as an outgrowth of the study of quadratic forms, and this ties in somewhat with the first item above.

  5. The field $\mathbf R$ has only one proper finite extension, namely $\mathbf C$, and this is responsible for Hamilton's quaternions being the only noncommutative finite-dimensional $\mathbf R$-central division algebra. If you study division rings over other fields, then the simplest nontrivial examples are in degree 4 (necessarily: the dimension over the center is a perfect square if it is finite), and these are the quaternion algebras. Their description is closely related to the quadratic extensions of the field. To describe, concretely, a division ring having dimension greater than 4 over its center means you're dealing with an object of dimension at least 9, and you can appreciate that this is going to be more complicated if you want to study it computationally. (There is a general theory of division rings, or more broadly central simple algebras, with quaternion algebras being the simplest interesting examples in the same way that quadratic fields are the simplest interesting field extensions.)

  6. Here is a minor, but interesting, reason that 2 is special: the only finite extension of $\mathbf R$ has degree $2$, and there is a theorem of Artin and Schreier that if $k$ is any non-algebraically closed field such that its algebraic closure is a finite extension, then the algebraic closure must be a quadratic extension and $k$ must look in some sense like the real numbers ($k$ has to be a real-closed field).

$\endgroup$
  • $\begingroup$ There are higher-degree forms with continuous automorphism groups, e.g., the determinant has $SL_n$. Jordan's theorem assumes the hypersurface is smooth, (reduced), projective and in characteristic 0. Some of the exceptional Lie groups can be realized as automorphisms of tri-linear forms. $\endgroup$ – zyx Jan 15 '14 at 0:53
  • $\begingroup$ @zyx: True. I did write in my answer that Jordan's theorem is for homogeneous polynomials that are "nonsingular", and I didn't want to say anything further on that. Thanks for pointing out a basic example without that hypothesis where indeed there are lots of elements preserving the polynomial. $\endgroup$ – KCd Jan 15 '14 at 1:00
  • $\begingroup$ The "nonsingular" had scrolled off the screen when the comment was written, so I actually did not realize that. The point though was not that (2.) is wrong (it's a very relevant fact), but that from 19th century point of view there isn't a reason to see the smooth/projective cases as fundamental, since the whole subject of Invariant Theory was based on the existence of high degree polynomials with a large linear symmetry group. (The trilinear forms for exceptional groups are not polynomials in any Sym^3(V), but there are polynomial realizations of much higher degree using invariant theory). $\endgroup$ – zyx Jan 15 '14 at 1:45
2
$\begingroup$

I recall a remark by a leading physicist to the effect that most of the important differential equations in physics are second order, so the "two" seems to be built into the structure of the universe. This appears to be confirmed by planetary orbits which have of course been conics (i.e. quadratic curves) since Kepler.

$\endgroup$
  • $\begingroup$ The physics student in me agrees, and adds that order two equations are where you get waves. I am not sure this is a mathematical reason (beyond the fact that mathematics and physics were studied as one subject until they split). $\endgroup$ – kleineg Jan 14 '14 at 18:30
  • $\begingroup$ Can you find any mathematical thing that isn't built into the universe? $\endgroup$ – PyRulez Apr 5 '15 at 14:39
  • 1
    $\begingroup$ @PyRulez, yes: the set of natural numbers. $\endgroup$ – Mikhail Katz Dec 8 '15 at 20:22
2
$\begingroup$

For some of those things, $2$ is special. e.g.:

  • Quadratic reciprocity works better than higher-order reciprocity laws because $\mathbb{Z}$ has a square root of unity
  • Quadratic number fields are always Galois over the rationals, simplistic unit groups, and the class number has unusual behavior. And the fact $\mathbb{Z}$ has a square root of unity is again relevant.
  • Conic sections are just lines in disguise, which you'll see if you study projective geometry
  • Quadratic forms satisfy the Hasse principle, which dramatically simplifies their theory

For many other things, $2$ is simply a small number -- by the time you've worked out the details to attack $3$, you can develop tools that would work for all numbers.

$\endgroup$
  • $\begingroup$ Recall also R.K. Guy's Strong Law of Small Numbers $\endgroup$ – Bill Dubuque Jan 14 '14 at 14:20
  • $\begingroup$ The Hasse principle for solutions of the forms holds because the degree 2 is sufficiently small relative to the number of variables, so this is more a case of "2 is simply a small number". There is the Hasse Minkowski theorem as a local-global principle, but this is unavailable in higher degree for a much more basic reason: the automorphism groups of higher degree forms usually are finite, and when not there is an exceptional situation studied using the extra structure and not a general theory of degree $n$ forms. $\endgroup$ – zyx Jan 15 '14 at 2:07
2
$\begingroup$

The Pythagorean Theorem probably has as much to do with the presence of squares in so many places. The square of distance can be written as a sum of squares of distances in orthogonal directions. Whole careers have been built around generalizations of the theorem of Pythagoras.

Then there's energy in Physics. Sums of squares, and conservation laws connect this subject with an abstract geometry. Independence of work along a path requires inverse square force laws, and the potentials which arise satisfy Laplace's equation. Laplace's equation gives rise to second order partial differential equations, whose solutions can often be built from mutually orthogonal pieces, which brings back sums of squares, and which leads to generalized geometry again. So we're back to Pythagoras.

Conics and other things of that type seem to be natural just because they're the next thing after linear. Second order approximation for multivariate functions can be studied through symmetric matrices where one has orthogonal eigenvector decompositions and sums (and differences) of squares in the right basis; so it's familiar. Cubic is just too hard.

$\endgroup$
2
$\begingroup$

With PDEs I imagine that, historically, a lot of it has less to do with things being second-order and more to do with the fact that second-order is where you start seeing periodic solutions. There isn't enough structure in first order systems to get very interesting dynamics in general, but once you get to second order you start seeing harmonic oscillators. Since so much of what we want to look at revolves around the idea of periodic structure, it makes sense that second-order systems show up as the simplest model.

$\endgroup$

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.