What recent programmes to alter highly-entrenched mathematical terminology have succeeded, and under what conditions do they tend to succeed or fail?

Question

I think we all occasionally come across terminology that we'd like to see supplanted (e.g. by something more systematic). What I'd like to know is, under what circumstances is it reasonable to believe that such a fight is winnable?

Question. What recent programmes to alter highly-entrenched mathematical terminology have succeeded, and under what conditions do they tend to succeed or fail?

Definitions.

• Recent = In the past fifty years.
• Highly-entrenched = The literature at the time had the old terminology written all over it.
• Succeeded = The new terminology is now used almost exclusively in research papers.

Answer 1

Although just beyond your 50-year scope, this may be of interest. Among the series $\mathsf A_n, \mathsf B_n, \mathsf C_n, \mathsf D_n$ in the Cartan-Killing classification of simple Lie groups, everyone (I believe) always agreed to call $\mathsf A_n$ the special linear group, $\mathbf{SL}(n)$, and $\mathsf B_n$ and $\mathsf D_n$ the special orthogonal groups, $\mathbf{SO}(2n+1)$ and $\mathbf{SO}(2n)$.

But $\mathsf C_n$? Jordan in his Traité des substitutions (1870) called it (or rather its product with dilations) the abelian group, because of its role in Hermite's "important investigations on the transformation of abelian functions"; p.172:

It is clear that if two [linear] substitutions $S, S'$ multiply $\varphi\ [=x_1\eta_1-\xi_1y_1+\dots+x_n\eta_n-\xi_ny_n]$ respectively by constant integers $m, m'$, $SS'$ will multiply $\varphi$ by the constant integer $mm'$. Hence the sought substitutions form a group. We will call it the abelian group, and its substitutions abelian.

This was well entrenched by the time Dickson wrote his Linear groups (1901); p.89:

A linear homogeneous substitution on $2m$ indices (...) is called Abelian if (...) it leaves formally invariant up to a factor (belonging to the field) the bilinear function $$ \varphi\equiv\sum_{i=1}^m \begin{vmatrix} \xi_{i1}&\eta_{i1}\\ \xi_{i2}&\eta_{i2} \end{vmatrix}. $$ The totality of such substitutions constitutes a group called the general Abelian linear group ${}^2)$ $GA(2m,p^n)$. These of its substitutions which leave $\varphi$ absolutely invariant form the special Abelian linear group $SA(2m,p^n)$.

${}^2)$ To distinguish these groups from the ordinary Abelian , i.e. commutative, groups, we prefix the adjective linear. The Abelian linear group is not commutative in general.

On the other hand, Sophus Lie and his school called $\mathsf C_n$ the linear complex group because it consists of symmetries of Plücker's linear line complex (a degree 1 hypersurface in the 4-dimensional space of affine lines in $\mathbf R^3$; Plücker (1866), p.341: "The latin word complexus, which means an intertwining, an inter-crossing, has seemed to us very appropriate to express the new idea we are presenting here. For lack of a better term, we ask for permission to introduce it in the mathematical language.") Thus for instance one finds in Lie and Engel's Transformationsgruppen, vol. II (1890), p.522:

One can say that the Pfaffian equation (73) represents a linear complex of the space $x'_1\cdots x'_n$, $y'_1\cdots y'_n$, $z'$. The group (72) should therefore be called the projective linear complex group.

Needless to say, both proposed names came into conflict with other spreading usages of these words. Hence Weyl's famous footnote in The classical groups (1938), p.165:

$$\text{CHAPTER VI}$$ $$\textbf{THE SYMPLECTIC GROUP}$$ $$\text{1. Vector Invariants of the Symplectic Group}^*$$ * The name "complex group" formerly advocated by me in allusion to line complexes, as these are defined by the vanishing of antisymmetric bilinear forms, has become more and more embarrassing through collision with the word "complex" in the connotation of complex number. I therefore propose to replace it by the corresponding Greek adjective "symplectic." Dickson calls the group the "Abelian linear group" in homage to Abel who first studied it.

Weyl's change was (obviously) highly successful. Ironically, Plücker's linear line complex term had won over Chasles' proposed focal system (1837). If it hadn't, today's symplectic geometers would probably all be doing "focal geometry"!

Answer 2

The subject known for decades as recursion theory, studying the class of recursive functions and the recursively enumerable (r.e.) sets and degrees, is now known almost universally, especially amongst the newer generation, as computability theory, studying the computable functions and the computably enumerable (c.e.) sets and degrees. This change, led by Bob Soare, is remarkable in that it was not merely a change in terminology for an isolated mathematical construction or idea, but a change in the name of the entire subject, a wholesale recasting of entire schemes of terminology to focus on what was all along the actual focus, namely, the concept of computability, rather than specifically recursion. The change was strongly resisted at first, and even lampooned, but in about a decade, it has triumphed.

Here are some of the elements that may have been important for success:

• The arguments for change were very strong. The recursion theory terminology grew out of Gödel's treatment of the primitive recursive functions, which grew into the class of recursive functions with the addition of unbounded search, and the subject became known as recursion theory. But in fact, the theme all along, even for Gödel, was computability, and to cap it off one doesn't even need recursion once one has unbounded search. So computability theory is simply a more accurate, honest name.

• Bob Soare was a tireless advocate for the change, and spent several years giving talks, writing articles and generally making the argument that the subject should be called computability theory. At first, many people argued against the change, but Soare persisted.

• Soare was a major figure an leader in the field, who wrote the standard textbook from which everyone had learned the subject. This text was called, "Recursively enumerable sets and degrees," using the old terminology, but Soare preferred to recast it as "Computably enumerable sets and degrees."

Answer 3

An example of a failure to change notation is the movement by Eilenberg, Jacobson, Herstein and others to replace function notation $f(x)$ with $xf$ and then have the composition $fg$ mean first do $f$ and then $g$. The notation has the advantage that diagrams $X\xrightarrow{f}Y\xrightarrow{g} Z$ don't have to be flipped around. It also has the property that if you want to use left modules, then you don't have to use opposite algebras when proving the Wedderburn theorem.

It seems that this notation caught on for a while in Algebra and is quite popular in permutation group theory, automata theory and semigroup theory. GAP uses this convention.

But I think it has nowadays mostly burned out and perhaps this has contributed to why few people still use Herstein's book.

Answer 4

Roger Penrose's abstract index notation for tensors is a relatively modest example, but I think it fits all the criteria of the question. Around 1952, Penrose invented a personal graphical notation for tensors and tensor operations such as contraction and covariant derivatives. It's been described as "fornicating ostriches;" variations on it are referred to by terms such as "birdtracks." It was manifestly coordinate-independent (in fact it didn't even use letters of the alphabet). Penrose continued to use it for his own purposes, but didn't publish a description of it until [Penrose 1971].

However, he also realized, apparently some time later, that it was possible to make up a notation that looked almost the same as traditional coordinate-dependent index-gymnastics notation (which Einstein invented ca. 1914) but that was in fact isomorphic to his coordinate-independent diagrammatic notation. Very little changed about the way the notation was written; the main changes were to the semantics. In old-style notation, the usual convention was that a Greek index ranged over the values 0 to 3, referring to four coordinates, while a Latin index ranged over only the spacelike indices. The new convention is that Latin indices are abstract, Greek ones concrete. In a notation like $p^c=T^{cd}u_d$, $d$ is a bound symbol and $c$ is not bound. Neither $c$ nor $d$ takes on specific values like 2 that refer to a particular coordinate system. This equation is simply a way of typesetting the following manifestly coordinate-independent diagram:

Here the square is $p$, the rectangle is $T$, and the circle is $u$. The abstract indices are to be interpreted simply as instructions for how to hook up the plumbing of the diagram, like instructions that say, "insert tab A in slot B."

Abstract index notation was popularized by Penrose, Geroch, and Rindler in the 1970s, and is now nearly universal among relativists. It won because (1) it looked familiar; (2) it was superior to index-gymnastics notation because it could express things in a manifestly coordinate-independent way; and (3) it was capable of conveniently and compactly representing tensor calculations that would have been too cumbersome to express using "mathematician-style" coordinate-independent notation.

Penrose, "Applications of negative dimensional tensors," in Combinatorial Mathematics and its Applications, ed. D.J.A. Welsh, Academic Press, 1971

Answer 5

From around 1900 to 1970, there was a highly-entrenched practice to write maps as $f:x\to y$ with an arrow between argument and value (e.g. Weyl 1913, p. 54).

Starting in the 1930s, a programme arose to write them instead as $f:X\to Y$, with the arrow now between domain and codomain (e.g. Eilenberg-Steenrod 1952, p. 4).

These conflict: does $f:X\to X\cup\{X\}$ mean von Neumann’s successor map, or an unspecified map between these sets? This can get bad when dealing with sheafs, or localization.

In the late 1950s, some bourbakists (Serre and Grothendieck?) attempted to lift the ambiguity by writing argument-to-value with what they called a tortillon : $x\rightsquigarrow y$. This didn't catch on.

Around early 1964, more bourbakists (Serre and Borel?) tried again, this time with what they called a poussoir : $x\mapsto y$. This was a hit, and almost entirely displaced the old $\to$ in research papers.

Lesson: enroll Borel?

Answer 6

In most cases that I am familiar with, successful changes to terminology were accompanied by other more basic innovations. E.g. Grothendieck's language became standard in algebraic geometry in the late 1950's, because he successfully rewrote the foundations of the subject.

An example of a proposed change for its own sake is the word "contrahomology" for "cohomology". I guess because it's contravariant. This occurs in the book by Hilton and Wylie on algebraic topology written in 1960. Happily, it never caught on.

Answer 7

A few examples:

Regarding Grothendieck's theory of schemes: These were called preschemes at first, while the word 'scheme' was reserved for separated (pre)schemes. ( Preschemes and schemes).

An example from (combinatorial) topology: Nowadays the terminology simplicial set is universally used for a bunch of sets indexed over the nonnegative numbers connected by face and degeneracy operators. The basic idea is from Eilenberg, who called simplicial sets without degeneracies "semisimplicial complexes" as they were a generalization of simplicial complexes. Later, Eilenberg and Zilbert called today's simplicial sets "complete semisimplicial complexes" or "c.s.s. complexes". Later, some people used the term "semisimplicial complex" instead of "complete semisimplicial complex" as the complete ones turned out to be more important. At some point it changed then to simplicial set. Some people now call Eilenberg's original semisimplicial complexes "semisimplicial sets" (or "preseimplicial sets") - one has to be very careful if one compares to older literature!

Originally, Quillen distinguished between model categories and closed model categories. But everyone just worked with the latter notion. In books by Hovey and Hirschhorn around 2000 the adjective "closed" was finally dropped and so one can speak nowadays just of a model category and everyone knows what one means.

One could also mention some terminology specific to German: Spektralfolgen became Spektralsequenzen and Fernparallelismus became Zusammenhang, both, I think, by the influence of backtranslating from English.

Answer 8

The statement that elliptic curves are modular was (I think) universally known as the Weil Conjecture back in 1970 or so when I first learned about it. Nowadays, it's generally known as the Shimura-Taniyama-Weil Conjecture (or, the Modularity Theorem). I don't know enough about the history to discuss the conditions responsible for the success of the proposed change.

Answer 9

I'm pretty sure it was less than 50 years ago that what are now called compact topological spaces were known as "bicompact", and "compact" meant what is now called "sequentially compact".