Normat.

Nordisk matematisk tidskrift 61:2–4, 133–152 Published in March 2019

Werner Fenchel, a pioneer in convexity theory and

a migrant scientist
Christer Oscar Kiselman

Abstract. Werner Fenchel was a pioneer in introducing duality in convexity theory.

He got his PhD in Berlin, was forced to leave Germany, moved to Denmark, and
lived during almost two years in Sweden. In a private letter he explained his views
on the development of duality in convexity theory and the many terms that have
been used to describe it. The background for his move from Germany is sketched.

1. Introduction
Werner Fenchel (1905–1988) was a pioneer in convexity theory and in particular
the use of duality there. When asked about his views on the many terms used to
express this duality he described in a private letter (1977) the whole development
from Legendre and onwards, as well as his preferences concerning the choice of
terms. The background for his leaving Germany and moving to Denmark and later
to Sweden is sketched in Section 12.

2. Convex sets and convex functions

In a vector space E over the field R of real numbers—think of E as the two-
dimensional plane, identified with R2 , or three-dimensional space, identified with
R3 —we define, given two points a and b, the segment with endpoints a and b:

[a, b] = {(1 − t)a + tb; t ∈ R, 0 6 t 6 1}.

A subset A of E is said to be convex if

{a, b} ⊂ A implies that [a, b] ⊂ A.

The modern theory of convex sets starts with the work of Hermann Minkowski
(1864–1909); see his book (1910), a large part of which was published in 1896.
When calculating with functions, it will be convenient to allow infinite values,
thus to let functions take values in R ∪ {−∞} ∪ {+∞}, adding two infinities to R.
This is the set of extended real numbers, and will be denoted by R! or [−∞, +∞].
It is ordered so that −∞ < x < +∞ for all x ∈ R.
The operation of addition,

R × R 3 (x, y) 7→ x + y ∈ R,
2 Christer Oscar Kiselman

will cause difficulties when we try to define sums like (+∞) + (−∞). A convenient
solution, pioneered by Moreau (1963; 1966–1967:9) is to extend it in two different
·
ways to operations R! × R! → R! , viz. as upper addition x + y, defined as +∞ if
one of the terms is equal to +∞, and lower addition x + y, defined as −∞ if one
·
of the terms is equal to −∞. (Some authors allow only values in ]−∞, +∞], but
then the rules for avoiding −∞ become complicated: if f is admissible, maybe −f
is not; the infimum of an admissible function may not be admissible, etc.)
It is convenient to define convex functions with the help of convex sets. We
define the finite epigraph of a function f : E → R! by

epiF (f ) = {(x, t) ∈ E × R; t > f (x)} ⊂ E × R.

Then a function f : E → R! is defined to be convex when epiF (f ) is convex.

In the other direction we note that a set A is convex if and only its indicator
function indA is convex. Here indA : E → R! is defined to take the value 0 in A
and +∞ in its complement. We have epiF (indA ) = A × [0, +∞[.
So far, there is no duality.
The book by Bonnesen and Fenchel (1934) is mainly concerned with convex
bodies; it mentions the supporting function but does not go into the duality theory.
Also—remarkably—Fenchel’s survey article (1983), a translation of a talk given in
1973, does not mention duality.
It is tempting to believe that convex sets possess some kind of regularity. But
this is not so. Let W be any subset of Rn , as irregular as you like, for instance a
set which is not measurable in the sense of Lebesgue (if you believe in the axiom
of choice). Then the set

{x ∈ Rn ; kxk2 < 1} ∪ {x ∈ W ; kxk2 = 1},

where we use the Euclidean norm k · k2 , is convex.

3. Duality in convexity theory

The definition of a convex set is done from the inside and does not need duality,
but there is also a definition from the outside, which requires an introduction of
duality.
We introduce the algebraic dual of E, which is the set of all linear forms ξ : E →
R and is denoted by E ∗ . An element of E ∗ is thus a function ξ : E → R such that
ξ(x + ty) = ξ(x) + tξ(y) for all elements x, y of E and all real numbers t. We
can now speak of hyperplanes in E: they are given by an equation ξ(x) = c for a
nonzero element ξ of E ∗ and a constant c ∈ R, as well as affine functions on E;
they are of the form f (x) = ξ(x) + c.
If E has a topology, we also speak about the dual of E, meaning the set of
all continuous linear forms on E and denoted by E 0 . If E is equal to Rn with
the usual topology, we have E 0 = E ∗ . Every linear form can then be written as
x 7→ ξ · x = ξ1 x1 + · · · + ξn xn for some vector (ξ1 , . . . , ξn ) ∈ Rn .
By duality in convexity theory I mean any consideration involving the dual or
the algebraic dual to a given space. A typical example is that of a norm k · k0 dual
 Werner Fenchel, a pioneer in convexity theory 3

to a given norm k · k and defined by

kξk0 = sup ξ(x), ξ ∈ E0.

x∈E
kxk61

If A is a closed convex subset of Rn , then it can be described as the intersection

of a family of closed half spaces. Here a closed half space is of the form

{x ∈ Rn ; ξ · x 6 t}

for some ξ ∈ Rn r {0} and some real number t. This result is essentially the
Hahn–Banach theorem in the finite-dimensional case. It follows that if A is an
open convex subset of Rn it can be described as the intersection of a family of
open half spaces. An open half space is of the form {x ∈ Rn ; ξ · x < t}.
In general we need half spaces of a more general kind. Let us say that Y is a
refined half space if it is convex and satisfies

{x ∈ Rn ; ξ · x < t} ⊂ Y ⊂ {x ∈ Rn ; ξ · x 6 t}.

This means that Y is the union of the open half space and a convex subset Aξ of
the hyperplane given by the equation ξ · x = t. Then any convex subset of Rn can
be described as the intersection of a family of refined half spaces Yξ , ξ of norm 1,
taking Aξ as the set of points y in A which satisfy ξ · y = t. (Cf. the notion of
refined digital hyperplane in my paper 2004:456, Definition 6.2.)

4. The Legendre transformation

The Legendre transformation is named for Adrien-Marie Legendre (1752–1833),
who introduced it in classial mechanics to go from the Lagrangian1 to the Hamil-
tonian2 formulation: the Hamiltonian is, in modern language, the Legendre trans-
form of the Lagrangian (see, e.g., Goldstein 1950:217).
If f : Rn → R is a differentiable mapping with gradient grad f : Rn → Rn ,
we may take ξ = (grad f )(x) as a new independent variable, thus looking at the
inverse
(grad f )(x) = ξ 7→ x = (grad f )−1 (ξ).
This is in general a multivalued mapping, which, in Legendre’s time, was no big
deal. The Legendre transform of f is the mapping

Rn 3 ξ 7→ ξ · (grad f )−1 (ξ) − f ((grad f )−1 (ξ)) = ξ · x − f (x).

(See Hiriart-Urruty & Lemaréchal 2001:209.) If f is strictly convex, then x is

uniquely determined by ξ, and we arrive at the Fenchel transform (see Section 6
below). If f is convex, the point x is not uniquely determined by ξ: the expression
ξ · x − f (x) can attain its maximum for several choices of x.
1 Refers to Joseph-Louis Lagrange (1736–1813).
2 Refers to William Rowan Hamilton (1805–1865).
4 Christer Oscar Kiselman

5. Tropicalization
Tropicalization means, roughly speaking, to replace a sum or an integral by a
supremum. A simple example is the lp -norm,
n
X
1/p
kxkp = |xj |p , x ∈ Rn , 1 6 p < +∞,
j=1

which becomes

1/p
kxk∞ = sup |xj |p = sup |xj |
j=1,...,n j=1,...,n

when the sum is replaced by the supremum. We shall see that the Fenchel trans-
formation is a case in point.
Tropical geometry is now a field of growing interest; see, e.g., Richter-Gebert
et al. (2005).

6. The Fenchel transformation

The Fenchel transform f˜ of a function f : Rn → R! is defined as

f˜(ξ) = sup ξ · x − f (x) , ξ ∈ Rn .

x∈Rn

Clearly ξ · x − f (x) 6 f˜(ξ), which can be written as

·
ξ · x 6 f (x) + f˜(ξ), (ξ, x) ∈ Rn × Rn ,

called Fenchel’s inequality. It follows that the second transform f˜ satisfies f˜ 6 f .

The operation f 7→ f˜ is a cleistomorphism (closure operator) if we define the
order between functions by inclusion of their finite epigraphs, which is actually a
very natural thing to do. This is because +∞ corresponds to vacuum since its finite
epigraph is empty, while −∞ corresponds to an infinitely dense black hole—in the
model for material objects, the density is e−f .
On the other hand, if we order functions by the relation f (x) 6 g(x) for all x,
then the operation is an anoiktomorphism. In this way, the Fenchel transformation
enters the field of mathematical morphology with its complete lattices, ethmomor-
phisms, cleistomorphisms and anoiktomorphisms.
The equality f˜ = f holds if and only if f is convex, lower semicontinuous, and
takes the value −∞ only if it is −∞ everywhere. Let us denote by CVX0 (Rn ) the
family of all functions with these three properties. The restriction of the Fenchel
transformation to CVX0 (Rn ) is thus an involution, i.e., it is equal to its own inverse.
Artstein-Avidan & Milman (2009:662) have proved that any decreasing involution
CVX0 (Rn ) → CVX0 (Rn ) is, up to some obvious modifications, the restriction to
CVX0 (Rn ) of the Fenchel transformation. Thus the Fenchel transformation is,
essentially, unique among all transformations with the indicated properties. While
this is, intuitively, hardly surprising, it is a very nice property and a convincing
witness to the importance of Fenchel’s work.
 Werner Fenchel, a pioneer in convexity theory 5

If f is defined on an arbitrary vector space E, we define the Fenchel transform

on its algebraic dual E ∗ :

f˜(ξ) = sup ξ(x) − f (x) , ξ ∈ E∗.

x∈E

Again,
·
ξ(x) 6 f (x) + f˜(ξ), (ξ, x) ∈ E ∗ × E.
We can apply the transformation a second time, but then it is convenient to choose
first an arbitrary subspace Ξ of E ∗ , for instance Ξ = E 0 , and define

f˜(x) = sup ξ(x) − f˜(ξ) ,

x ∈ E.
ξ∈Ξ

The Fenchel transformation f 7→ f˜ is a tropical counterpart of the Laplace

transformation. The Laplace transform of a function g : [0, +∞[ → [0, +∞[ is
Z ∞
(L g)(ξ) = g(x)e−ξx dx, ξ ∈ R.
0

If we replace the integral by a supremum and take the logarithm, we get

(Ltrop g)(ξ) = sup(log g(x) − ξx) = f˜(−ξ), ξ ∈ R, f = − log g.

x

7. The supporting function

A special case of the Fenchel transform is obtained when the function is the in-
dicator function of some set: f = indA . Then f˜ is the supporting function of A,
denoted by HA :
HA (ξ) = sup ξ · x, ξ ∈ Rn .
x∈A

This function was introduced already by Minkowski (1903:448).

However, conversely, a Fenchel transform f˜ is also a supporting function if we
go up in dimension. We equip Rn × R with the inner product

(ξ, τ ) · (x, t) = ξ · x + τ t, (ξ, τ ), (x, t) ∈ Rn × R.

Then
f˜(ξ) = HepiF (f ) (ξ, −1), ξ ∈ Rn .

8. Galois correspondences
Let X and Y be two ordered sets. A Galois correspondence3 is a pair (F, G) of
decreasing mappings F : X → Y and G : Y → X such that G ◦ F is larger than the
identity in X and F ◦ G is larger than the identity in Y . Taking both F and G
as the Fenchel transformation, we see that the two form a Galois correspondence,
provided we order the functions using the inclusion relation between the finite
epigraphs. (See, e.g., my paper 2010: Section 4.)
3 Named for Évariste Galois (1811–1832).
6 Christer Oscar Kiselman

Moritz Werner Fenchel (1905 May 03 — 1988 January 24)

9. Werner Fenchel
Fenchel was born on 1905 May 03 in Berlin. He obtained his doctoral degree
in 1928 at the Friedrich-Wilhelms-Universität in Berlin with Ludwig Bieberbach
(1886–1982) as advisor (Segal 2003:410).4 A paper based on his thesis and enti-
tled “Über Krümmung und Windung geschlossener Raumkurven” appeared in the
Mathematische Annalen in 1929.
After his PhD, Werner Fenchel became an assistant at Göttingen University. He
was dismissed from this position in 1933. He accepted an invitation to Copenhagen
and arrived to Denmark in the summer of 1933. (Jessen 1987:89.)
In December 1933 he married Käte Sperling (1905–1983), who was also a mathe-
matician, and who had been ousted from a position as high-school teacher in Berlin.
In Denmark, Werner Fenchel was first supported by various foundations until he
4 Bieberbach joined the National Socialist Lecturers Association in November 1933 and a lit-

tle later became the representative of this organization at the University of Berlin (Mehrtens
1987:220). He joined the Sturmabteilung (SA) on 1934 November 05 and the Nationalsozialistische
Deutsche Arbeiterpartei (NSDAP) on 1937 May 01 (Bieberbach’s personal file in the archive at
Humboldt University, signature UK B 220, folio 64; information provided by Reinhard Siegmund-
Schultze on 2016 January 26). In a lecture in 1934 he justified the Nazi-organized boycott of
Edmund Landau’s classes (Mehrtens 1987:227). However, in 1928 he could still serve as advisor
for Fenchel: “no one had ever observed antisemitic tendencies in Bieberbach before the summer
of 1933” (Schappacher 1998:130). For more on Bieberbach, see Segal (2003).
 Werner Fenchel, a pioneer in convexity theory 7

got a teaching position in 1938 and became a lecturer at Copenhagen University

in 1942. (Jessen 1987:89.)
As it turned out, Copenhagen was not sufficiently far away: on 1940 April 09,
Germany invaded Denmark. However, during the first years of the occupation,
Jews were not deported from Denmark; deportations started in October 1943. A
majority of all Jews in Denmark could escape to Sweden, although some were
deported to Theresienstadt (Jacoby 2015). Werner Fenchel came to Sweden in
October 1943 (Jessen 1987:89).
A Danish school, den danske Skole i Lund, started on 1943 November 15 in Lund
in southern Sweden and functioned during 19 months, to the end of the war. Werner
Fenchel and many others taught there, for example Harald Bohr. (Siegmund-
Schultze 2009:107, 136.) Also his wife Käte Fenchel taught at the Danish school,
and Werner Fenchel lectured also at Lund University (Jessen 1987:89).
In 1950/51 Werner Fenchel was at the Institute for Advanced Study in Prince-
ton, NJ, and met Harold W. Kuhn (1925–2014) and Albert W. Tucker (1905–1995)
there:
It must therefore have been very exciting for Kuhn and Tucker when they became
aware that Werner Fenchel from Copenhagen University, who was the leading expert
on convexity at the time, was visiting the Institute for Advanced Study in Princeton
as part of a sabbatical year in the USA in 1950/51. Tucker invited Fenchel to give
a series of lectures on the theory of convexity at the mathematics department at
Princeton University within his ONR project. (Kjeldsen 2010:3250)
Fenchel wrote notes (1951) based on his lectures in Princeton, expanding his short
paper (1949):
Fenchel was motivated by problems and connections in pure mathematics, and his
lecture notes became highly influential for further developments within the theory
of convexity, especially through the work of R. T. Rockafellar [50].5 (Kjeldsen
2010:3251)
Werner Fenchel died on 1988 January 24.

10. Werner Fenchel’s letter

I wrote a letter (in Swedish) to Werner Fenchel on 1977 February 24, asking him
about the history of duality in convexity theory and the many terms used in that
theory. He replied with a letter dated 1977 March 07 (3 pages; in Danish).6 He
reported in detail on the development of convexity theory and the duality which
had been known under so many different names, and where he himself had been a
pioneer. I translate it into English below.

5 Refers to Rockafellar (1970). This book the author “dedicated to Fenchel, as honorary co-

author” (Rockafellar (1970:viii).

6 For the benefit of young readers, let me mention that Fenchel’s letter as well as mine were

written on paper using a kind of mechanical device called “typewriter,” which had “arms” or
“bars” equipped with profiles in the form of letters, upper case letters being above the corre-
A
sponding lower case letter, like this: . A tiny moment before an arm hit the paper, a textile
a
ribbon, slightly moistened with a kind of ink, jumped up in front of the paper and because of
the pressure from the arm left a mark on the paper, very much looking like a letter or symbol,
although often a little fuzzy.
8 Christer Oscar Kiselman

Werner Fenchel lecturing in Menton in the middle of the sixties

p q
professor w. fenchel 7. 3. 1977
sønderengen 110
2860 søborg

Dear Kiselman.
Thanks for the letter! It is not easy for me to answer. I have again after
many years looked at the relevant literature, which has delayed my answer. I
briefly report on the history of the matter.
The classical Legendre transformation was introduced purely formally by
Legendre. That it is an involution and closely related to duality was, however,
as far as I know, first discovered by Monge somewhat later. One shall of course
keep the name Legendre transformation.
W. H. Young has in a small paper from 1912 proved the following. Let ϕ(x)
be positive and differentiable with ϕ0 (x) > 0 for x > 0, and let ψ(y) be the
inverse of ϕ. Then we have with b = ϕ(a),
Z x Z y
xy − ab 6 ϕ(ξ)dξ + ψ(η)dη.
a b
Rx Ry
If we put f (x) = a ϕ(ξ)dξ and f˜(y) = ab + b ψ(η)dη, we easily see that f 7→ f˜
is a special case of the classical Legendre transformation, which Young, however,
did not notice, even though this is clear from his proof. He is only interested in
the inequality.
The next contribution to the topic is due to Z. W. Birnbaum & W. Or-
licz, 1931. They consider convex functions f (x) which are defined for x ∈ R
and satisfy f (0) = 0, f (x) > 0 for x > 0, f (−x) = f (x), f (x)/x → 0 as x → 0,
 Werner Fenchel, a pioneer in convexity theory 9

Werner Fenchel in the late sixties or early seventies

f (x)/x → ∞ as x → ∞. Here the “complementary” function f˜ to f is defined

by

(∗) f˜(y) = sup (xy −f (x)),

x∈R

and it is shown that f˜ = f . The authors also show that f (x) = 0 ϕ(ξ)dξ and
Rx
Ry
f˜(y) = 0 ψ(η)dη, where ϕ is non-decreasing and continuous to the right, and
ψ is its inverse, conveniently defined with the same properties.
In 1939 there appeared a somewhat imprecise note by S. Mandelbrojt,7 where
he defined f˜ by (∗) for an arbitrary convex function of one variable, but without
investigating under which conditions the transformation f 7→ f˜ is an involution.
It is clear that he did not know about Birnbaum & Orlicz.
When I wrote my paper “On conjugate convex functions” in 1949, I also did
not know about Birnbaum & Orlicz. I define there f˜ for an arbitrary convex
function defined in a convex subset C of Rn by

f˜(y) = sup (hx, yi − f (x))

x∈C

and find a necessary and sufficient condition for the equality f˜ = f to hold. The
inequality
7 That his note is imprecise, as Fenchel writes, is due to the fact that Mandelbrojt assumes

real values of the functions without noticing that, if f has real values and does not grow faster
than all linear functions, then f˜(y) will not be real valued; it will take the value +∞ when y or
−y is large.
10 Christer Oscar Kiselman

(∗∗) hx, yi 6 f (x) + f˜(y)

is a trivial consequence. I also show that f 7→ f˜ under suitable differentiability
hypotheses is the Legendre transformation. In Lecture Notes, Princeton 1951,
I have given a more systematic treatment of the transformation and several
variants. A further development is due to R. T. Rockafellar from 1963.
Generalizations to infinite-dimensional spaces have been undertaken inde-
pendently by J. J. Moreau, 1962, A. Brøndsted, 1964, and U. Dieter, 1965.
In the bibliography in Rockafellar’s book “Convex Analysis” all the men-
tioned papers are listed.
Now to the question of the name! I admit that the situation is rather chaotic.
That I called f˜ the function conjugated to f depends on the fact that I considered
the relation as a generalization of the “conjugated expononents” of classical
analysis xp /p and y q /q with 1/p + 1/q = 1. This was also the starting point
for Young and Birnbaum–Orlicz. But the connection with that has become
very weak and the name is therefore not adequate. Rockafellar and Brøndsted
have retained my terminology. Moreau calls f˜ the function dual to f . But in
the special case considered by Birnbaum and Orlicz, he talks about the Young
dual. The inequality (∗∗) is called Young’s inequality by several authors. This
is OK, but I cannot see any reason for calling f 7→ f˜ the Young transformation.
If you want to include a personal name, it seems Legendre is the closest at
hand, even though it does not create the correct association. The “maximum
transformation”8 I have not seen. That is a possibility. I do not want to add
a new name, but nevertheless say, that if I now should give the transformation
one, I would let me be guided by the analogy and the connection with polarity
between convex sets (in dual vector spaces) and for instance call it parabolic
polarity.
With kind regards

Werner Fenchel
x y
I thanked Fenchel in a letter dated 1977 April 02 and told him that Christer Borell
and several other persons in Uppsala had discussed it, and that we liked very much
the term parabolic polarity, which I had not seen before. However, I remarked that
it is rather unwieldy to call f˜ the function parabolically polar to f .
I later met Werner Fenchel at a conference in Uppsala and could continue the
discussion on the history of duality and possible terms.

11. A plethora of terms

As Fenchel wrote concerning names: “I admit that the situation is rather chaotic.”
Since the Fenchel transformation is also an example of a Galois correspondence,
where the situation concerning concepts and the terms used for them is even more
8 Mentioned in my earlier letter to Fenchel; the term was used by Bellman & Karush (1963).
 Werner Fenchel, a pioneer in convexity theory 11

chaotic, this story can be continued: terms like residual mapping, adjunction, upper
adjoint, and lower adjoint occur. (See my paper 2010, Section 4). The reason for
this state of affairs is pretty clear: the concepts are indeed fundamental in nature,
but they are approached from many different directions, often by researchers who
are not aware of earlier work.
In Section 6 above I have used the terms that I prefer now. However, several
terms have been used for the main concepts involved here: for the supporting
function of a set; for a Fenchel transform f˜; for the Fenchel transformation f 7→ f˜;
·
and for Fenchel’s inequality ξ · x 6 f (x) + f˜(ξ). I list a few of them below.

• le Gendre [Legendre] (1789; Lû le 1er Septembre 1787): “Au lieu de considérer
z, p, q comme des fonctions de x & y, rien n’empêche de regarder x, y, z
comme des fonctions de p & q; [. . . ]” (p. 315). This is the very beginning of
what we now call the Legendre transformation. Here it is about a function z
of two variables x and y with p = ∂z/∂x and q = ∂z/∂y. The symmetry or
duality between z, p, q and x, y, z is evident. Later in the article the author
considers functions of more than two variables.

• Minkowski 1896: in Minkowski (1910) appear the terms “Stützebene” (p. 13);
“nirgends concave Fläche” (p. 35), meaning a convex surface; “überall convexe
Fläche” (p. 38), meaning a strictly convex surface. The part containing these
terms was published already in 1896.

• Minkowski 1897: What is now known as “Minkowski addition” appears for

instance in a paper published in 1897, reproduced in Minkowski (1911a:108).

• Minkowski (1903): “konvexer Körper” (p. 447); “Stützebenfunktion” (p. 448).

• Minkowski (1911b): “konvexer Körper” (p. 131); “Stützebene” (p. 136).

• Birnbaum & Orlicz (1931): “konjugierte Potenzen” (p. 2); “Die N -Funktion
N (v) heißt konjugiert” (p. 6); “komplementäre Funktion” (p. 8).

• Bonnesen & Fenchel (1934): “Die Stützfunktion” (p. 23).

• Mandelbrojt (1939): “qu’on peut appeler la fonction convexe associée à f (x)”

(p. 977).

• Fenchel (1949): The two functions are called “conjugate” to each other.

• Goldstein (1950): “the Legendre transformation” (p. 215).

• Landau & Lifshitz (1960): “Legendre’s transformation” (p. 131).

• Moreau (1962): “la fonction convexe duale de f ” (p. 2897).

• Bellman & Karush (1963): “maximum convolution” (p. 67); “maximum trans-
form” (p. 68).

• Hörmander (1963): “Legendretransformen (konjugerade funktionen)” (p. 2.4).

• Valentine (1963): “support functional” (p. 473).

12 Christer Oscar Kiselman

• Moreau (1966–1967): “fonction polaire de h” (p. 33); “une paire de fonctions

duales (ou que chacune est la duale de l’autre) si chacune est la fonction
polaire de l’autre” (p. 45).

• Abraham & Marsden (1967): “a Legendre transformation” (p. 122).

• Ioffe & Tihomirov (1968): “konvolci” (for infimal convolution);

“preobrazovanie nga ili soprenna ” ‘Young’s transformation or
conjugate’; “neravenstvo nga” ‘Young’s inequality’ (p. 55).

• Rockafellar (1970): “This f ∗ is called the conjugate of f ” (p. 104); “sup-

port function” (p. 112); “subgradient” (p. 214); “subdifferential” (p. 215); “The
classical Legendre transformation for differentiable functions defines a corre-
spondence which, for convex functions, is intimately connected with the con-
jugacy correspondence” (p. 251); “The Legendre conjugate,” “the Legendre
transformation” (p. 256).

• Kiselman (1978): “The Legendre transformation.”

• Desloge (1982): “Legendre transformation” (p. 920).

• Fenchel (1983): “support function” (p. 127).

• Hörmander (1983): “supporting function” (p. 105); “The quadratic forms

[. . . ] are said to be dual” (p. 206); “dual cone” (p. 257).

• Moreau (1993, in a private letter to Hiriart-Urruty): “C’est moi qui ai intro-

duit le terme sous-gradient (et sa traduction anglaise subgradient).”

• Hiriart-Urruty & Lemaréchal (1993): “Theory of Conjugate Functions”

(p. 36); “the so-called Fenchel correspondence, and is closely related to the
Legendre transform” (p. 38); “Young–Fenchel inequality” (p. 38).

• Hörmander (1994): “Conjugate convex functions (Legendre transforms)”

(p. 16); “The Legendre transform (also called the conjugate function)” (p. 17);
“Then the Legendre transform (= conjugate function = Fenchel transform)”
(p. 67); “supporting function” (p. 69).

• Singer (1997): “subdifferential” (p. 19); “Fenchel–Young inequality” (p. 254);

“Fenchel–Rockafellar theorem” (p. 273); “Fenchel–Moreau theorem” (p. 433);

• Gao (2000): “Legendre transformation” (p. 27); “Legendre conjugate transfor-

mation” (p. 27); “Legendre Duality Theorem” (p. 29); “Legendre duality pair”
(p. 30); “Legendre conjugate functions” (p. 30); “Legendre–Fenchel transfor-
mation” (p. 32); “Fenchel transformation” (p. 32); “Fenchel-conjugate func-
tion” (p. 32); “Fenchel–Young inequality” (p. 32); “Fenchel-conjugate” (p. 32);
“Fenchel conjugate pair” (p. 234); “sub-differential” (p. 236); “sub-gradients”
(p. 236).

• Hiriart-Urruty & Lemaréchal (2001): “Legendre transform” (p. 209) for the
original Legendre transform; “The conjugate of a function”; “conjugacy op-
eration”; “the Legendre–Fenchel transform” (p. 211).
 Werner Fenchel, a pioneer in convexity theory 13

• Ehrenpreis (2003): “Legendre transform (sometimes called Young conjugate)”

(p. 153).

• Murota (2003): “Legendre–Fenchel transform” (p. 10); “Legendre–Fenchel

transformation” (p. 10); “Fenchel duality” (p. 12); “subdifferential” (p. 80);
“subgradient” (p. 80); “convex conjugate” (p. 81).

• Kiselman (2015): “The Fenchel transformation”; “the Fenchel transform”;

“Fenchel’s inequality” (p. 305).

In 1977, I still called the transformation the Legendre transformation, following my

advisor Lars Hörmander (1931–2012), who, with some emphasis, insisted that the
Fenchel transformation is the same as the Legendre transformation.
I later switched to calling it the Fenchel transformation. The motivation for this
is that I am now of the opinion that the results of Legendre belong to differential
geometry, whereas Fenchel’s seminal work belongs to the theory of ordered sets
and complete lattices, which has a very different flavor.
As Fenchel wrote, the name Legendre does not create the correct association.
Rockafellar (1970:251–260) distingishes between the conjugacy correspondence, the
Legendre conjugate, and the Legendre transformation. As I mentioned in Section
6 above, the Fenchel transformation is a tropicalization of the Fourier or Laplace
transformation and does not invoke differentiability as the Legendre theory does.
It is indeed an early example of the tropicalization of mathematics, which is now
a very active field of research.
The terms transform for f˜ and transformation for f 7→ f˜ are very practical and
follow a well-known pattern: the Fourier transformation, the Laplace transforma-
tion, the Radon transformation, the Fenchel transformation, . . .

12. Why Werner Fenchel had to leave Germany

Many scientists in Germany lost their jobs after the Nazi Machtübernahme in 1933.
Fenchel was one of them. It seems appropriate to describe in some detail this
development, which is of interest also because of the general decline of science and
culture in Germany. See also Gordin (2015: Chapter 7, Unspeakable).
As described in my paper (2019), German mathematics and physics were world
leading in many respects during the nineteenth century and the beginning of the
twentieth century. Scientists in many other countries, including Sweden, wrote in
German, but after 1945, this became rare in Sweden.
The reasons behind the transition from German to English in scientific writing
are often considered to be Germany’s defeat in the Second World War and the
subsequent rise of the USA to the world’s leading country in engineering and natural
science. This opinion needs to be qualified.

12.1. Science education in Germany 1933–1939

A factor which, according to my opinion, has not been given sufficient attention, is
how drastic the loss in German science was already before the start of the Second
World War. This can be illustrated by a few figures.
14 Christer Oscar Kiselman

In the year 1932, the number of students in mathematics, insurance mathe-

matics, and physics in Germany was 7,139. Just before the war, in the beginning
of 1939, this number had gone down to 1,270, a decrease by 82 per cent in seven
years (Mehrtens 1985:85). In Göttingen, where, as Laurent Schwartz wrote in his
memoirs (1997:79), the greatest scientists of the world resided, the number of math-
ematics students went down from 432 till 37, a decrease by 91 per cent during the
same period.
We can compare with how Harald Bohr (1887–1951) described Göttingen in an
earlier epoch:
While Göttingen was in many ways a provincial town, calm and peaceful, the richest
scientific life flourished there. A spirit of genuine international brotherhood of a
rare intensity reigned there among the many young mathematicians who came from
nearly all over the world to make a pilgrimage to Göttingen, bound together as they
were by their common interest in and love for their science. (Bohr 1952:xxi–xxii)
The importance of Göttingen as a center is further discussed by Thomas Schøtt:
Göttingen and Bohr’s orientation toward this place is not an exceptional instance.
It is one of a recurring pattern of how the scientific community fits into the various
national societies according to a center and periphery theory [. . . ] (Schøtt 1979:86)
However, the center of a science is not a fixed point. It can move.

12.2. The law of 1933

Among the decisions causing this disaster in higher education and research was the
Gesetz zur Wiederherstellung des Berufsbeamtentums ‘Law for the Restoration of
the Professional Civil Service’, enacted by the government (not by the Reichstag)
on 1933 April 07, little more than two months after the Machtübernahme, and
signed by Der Reichskanzler Adolf Hitler, Der Reichsminister des Innern Frick,
and Der Reichsminister der Finanzen Graf Schwerin von Krosigk.
As the law was first drafted by the Interior Minister Wilhelm Frick (1877–
1946; Reichsminister des Innern 1933–1943), all Beamte nicht arischer Abstam-
mung ‘civil servants of non-Aryan descent’, as well as several other categories, were
to be fired immediately at the Reich, Länder, and Municipal levels of government.
However, the President of Germany, Paul von Hindenburg (1847–1934; Reichs-
präsident 1925–1934) objected to the bill until it had been amended to exclude
three classes of civil servants from the ban.
The cardinal provision in the law is Article 3:
§3
(1) Beamte, die nicht arischer Abstammung sind, sind in den Ruhestand (§§ 8 ff.)
zu versetzen; soweit es sich um Ehrenbeamte handelt, sind sie aus dem Amtsverhält-
nis zu entlassen. (Gesetz zur Wiederherstellung des Berufsbeamtentums)
Civil servants who are not of Aryan descent are to be put into the state of retire-
ment (§§ 8 et seq.); in so far as it is about a commission of trust, they are to be
dismissed from service. (My translation)
The law contained provisions for reducing the pensions of these retired civil ser-
vants.
The second paragraph of Article 3 contained certain exceptions, imposed by
President Hindenburg. However, the first and sixth articles opened for a broader
application:
 Werner Fenchel, a pioneer in convexity theory 15

§1
(1) Zur Wiederherstellung eines nationalen Berufsbeamtentums und zur Verein-
fachung der Verwaltung können Beamte nach Maßgabe der folgenden Bestimmungen
aus dem Amt entlassen werden, auch wenn die nach dem geltenden Recht hierfür
erforderlichen Voraussetzungen nicht vorliegen. (Gesetz zur Wiederherstellung des
Berufsbeamtentums)
For the restoration of a national professional civil service and in order to simplify
administration, civil servants can, in accordance with the following provisions, be
dismissed even if the required legal conditions are not fullfilled. (My translation)

§6
Zur Vereinfachung der Verwaltung können Beamte in den Ruhestand versetzt
werden, auch wenn sie noch nicht dienstunfähig sind. [. . . ] (Gesetz zur Wiederher-
stellung des Berufsbeamtentums)
In order to simplify administration, employees can be transferred to the state
of retirement, even if they are not yet unsuitable for service. [. . . ] (My translation)

12.3. Hilbert
David Hilbert (1862–1943) was the founder of the theory of Hilbert spaces and
known for having presented in Paris on 1900 August 08 ten of a total of twenty-
three problems that were to keep mathematicians busy for a long time. He was a
professor at Göttingen and is reported to have summed up the situation there soon
after Hitler’s Machtübernahme thus:
Sitting next to the Nazis’ newly appointed minister of education at a banquet, he
was asked, “And how is mathematics in Göttingen now that it has been freed of the
Jewish influence?”
“Mathematics in Göttingen?” Hilbert replied. “There is really none any more.”
(Reid 1970:205)
Reinhard Siegmund-Schultze (personal communication 2014 August 19) remarks,
however, that there is no documentation showing that Hilbert actually pronounced
these words in the presence of the minister of education.

12.4. Einstein, Weyl, Fenchel, and many others

Albert Einstein (1879–1955) resigned his position at the Preußische Akademie
der Wissenschaften in a letter dated 1933 March 28, before he was ousted (Pais
1982:450).9 He accepted an offer (one of many) from the Institute for Advanced
Study in Princeton, NJ. Most dismissed scientists could of course not find some-
thing similar.
Hermann Weyl (1885–1955) succeeded Hilbert in Göttingen in 1930. He was
not Jewish, but his wife Friederike Bertha Helene Weyl (1893–1948) was. The law
said nothing about non-Aryan descent of spouses, but, as noted above, it opened
9 In spite of Einstein’s own decision to leave, a caricature exhibited at the Einstein Museum in

Bern (a part of the Musée d’histoire de Berne, Bernisches Historisches Museum) shows Einstein
being kicked down a staircase by a boot, mentioning that he mistakingly thought he was a
Prussian: “Der Hausknecht der Deutschen Gesandtschaft in Brüssel wurde beauftragt, einen dort
herumlungernden Asiaten von der Wahnvorstellung, er sei ein Preuße, zu heilen.” ‘The house
servant of the German embassy in Brussels got the commission to free an Asian tramp from his
misconceived idea that he was a Prussian.’
16 Christer Oscar Kiselman

wider possibilites based on “Vereinfachung der Verwaltung.” Weyl did not want
to leave Germany and therefore declined an early offer to join the Institute for
Advanced Study, but later changed his mind and accepted a second offer in 1933.
Another victim of this “Vereinfachung” was Michael A. Sadowsky (1902–1967),
who got his doctorate at the Technische Hochschule Berlin in 1927 and later lost his
teaching permit there: “In 1933 Michael Sadowsky’s teaching permit was revoked
because his wife was of Jewish descent. However, the legal base10 for such a decision
was established only in 1937.” (Brüning et al. 1998:6; see also Siegmund-Schultze
2009:37–38).
Among the mathematicians forced to retire or who were dismissed were Otto
Blumenthal (1876–1944), Edmund Landau (1877–1938),11 Felix Bernstein (1878–
1956), Emmy Noether (1882–1935), Max Born (1882–1970), Richard Courant (1888–
1972), and Stefan Bergman (1895–1977).
Of all these scientists, I came to know one personally: Werner Fenchel.

12.5. Zentralblatt and Mathematical Reviews

A reviewing journal called Zentralblatt für Mathematik und ihre Grenzgebiete was
founded in 1931 by Otto Neugebauer (1899–1990). It provided reviews and ab-
stracts for articles in pure and applied mathematics. In 1938, the publisher,
Springer-Verlag, asked for written assurances that no Jews would act as review-
ers (Jackson 1997:330). A number of reviewers in England and the United States
resigned in protest (Lehmer 1988:265).
Neugebauer left the Zentralblatt in 1938 when the name of a Jewish editor (Levi-
Civita12 ) was, without further notice, eliminated from the title page. (Mehrtens
1987:217)
The Zentralblatt no longer registered all mathematical publications. Something had
to be done. A new reviewing journal, named Mathematical Reviews, was started in
the United States. It was founded in 1940 by the same Otto Neugebauer, now at
Brown University. (Nowadays most users rely on its online version MathSciNet.)

12.6. The end of a glorious era

It is clear that the glorious era for mathematics in Germany ended well before the
beginning of the war—and independently of the outcome of the war as well as of
the rise of the United States.

12.7. Hermann Minkowski

In another epoch, people moved to Germany rather than out of it: Hermann
Minkowski, who was born in the Kovno Governorate in the Russian Empire (now
Kaunas, Lithuania), moved as a young boy with his family to Königsberg in 1872
10 “Legal base” . . . of a very special kind.
11 Harald Bohr, who worked with Landau in Göttingen, wrote: “Landau made a strong im-
pression on everyone who came into contact with him. His baroque sense of humour and his
exceptional vitality, characterized equally his scientific research and his teaching. His thinking
was amazingly quick and sure and his standards for precision and exactitute of exposition were
absolute and inexorable.” (Bohr 1952:xxii).
12 Tullio Levi Civita (1873–1941).
 Werner Fenchel, a pioneer in convexity theory 17

to avoid persecution. He became a brilliant mathematician, an early pioneer of

convexity theory, and later taught in Bonn, Königsberg, and Zürich (where Ein-
stein studied for him in the Eidgenössische Polytechnische Schule), and finally in
Göttingen.

Acknowledgments

Tom Fenchel, son of Käte Fenchel, née Sperling, and Werner Fenchel, allowed me
to publish a translation of his father’s letter. In addition, he kindly sent me three
photos of his father and allowed them to be published here. And it was Tinne Hoff
Kjeldsen who brought me into contact with Tom!
Reinhard Siegmund-Schultze provided important historical information, and
Seidon Alsaody, Sharon Rider, and Arild Stubhaug made valuable comments to
earlier versions of the manuscript.
Michael D. Gordin encouraged me to write this paper separately from my ear-
lier paper on language choice in doctoral theses presented at Uppsala University
and in the journal Acta Mathematica, and sent me important comments on both
manuscripts.
Staffan Rodhe sent me a link to the seminal article by le Gendre [Legendre]
(1789). Jean-Baptiste Hiriart-Urruty sent encouraging comments, drew my at-
tention to the nice result of Shiri Artstein-Avidan and Vitali Milman (2009), and
sent me a copy of a letter from Jean-Jacques Moreau. Juliusz Brzezinski sent me
comments and corrections.
For all this invaluable help I am most grateful.

References

Abraham, Ralph; Marsden, Jerrold E. 1967. Foundations of Mechanics. New York, NY;
Amsterdam: W. A. Benjamin, Inc.
Artstein-Avidan, Shiri; Milman, Vitali. 2009. The concept of duality in convex analysis,
and the characterization of the Legendre transform. Ann. of Math. 169, 661–674.
Begehr, H. G. W.; Koch, H.; Kramer, J.; Schappacher, N.; Thiele, E.-J., Eds. 1998.
Mathematics in Berlin. Berlin et al.: Birkhäuser Verlag.
Bellman, Richard; Karush, William. 1963. On the maximum transform. J. Math. Anal.
Appl. 6, 67–74.
Birnbaum, Z. W.; Orlicz, W. 1931. Über die Verallgemeinerung des Begriffes der zuein-
ander konjugierten Potenzen. Studia Mathematica 3, No. 1, 1–67.
Bohr, Harald. 1952. Harald Bohr. Collected Mathmatical Works. Volume I. Copenhagen:
Dansk Matematisk Forening.
Bonnesen, T.; Fenchel, W. 1934. Theorie der konvexen Körper. Berlin: Verlag von Julius
Springer.
Brüning, Jochen; Ferus, Dirk; Siegmund-Schultze, Reinhard. 1998. Terror and Ex-
ile. Persecution and Expulsion of Mathematicians from Berlin between 1933 and
1945. An Exhibition on the Occasion of the International Congress of Mathemati-
cians. Technische Universität Berlin; August 19 to 27, 1998. Berlin: Deutsche
Mathematiker-Vereinigung, DMV.
Desloge, Edward A. 1982. Classical Mechanics. New York, NY, et al.: John Wiley &
Sons.
18 Christer Oscar Kiselman

Ehrenpreis, Leon. 2003. The Universality of the Radon Transform. Oxford: Clarendon
Press.
Fenchel, W. 1949. On conjugate convex functions. Canadian J. Math. 1, 73–77.
Fenchel, Werner. 1951. Convex cones, sets, and functions. Lecture Notes, Department of
Mathematics, Princeton University.
Fenchel, W. 1983. Convexity through the ages. In: Gruber, Peter M.; Wills, Jörg M.,
Eds. 1983. Convexity and Its Applications, pp. 120–130. Basel et al.: Birkhäuser
Verlag.
Gao, David Yang. 2000. Duality Principles in Nonconvex Systems: Theory, Methods and
Applications. Dordrecht et al.: Kluwer Academic Publishers.
Gesetz zur Wiederherstellung des Berufsbeamtentums. Available at
www.documentarchiv.de/ns/beamtenges.html. Accessed 2018 May 09.
Goldstein, Herbert. 1950. Classical mechanics. Reading, MA: Addison-Wesley Publishing
Company, Inc.
Gordin, Michael D. 2015. Scientific Babel: How Science Was Done Before and After
Global English. Chicago; London: The University of Chicago Press.
Hiriart-Urruty, Jean-Baptiste; Lemaréchal, Claude. 1993. Convex Analysis and Mini-
mization Algorithms I: Fundamentals. Berlin et al.: Springer.
Hiriart-Urruty, Jean-Baptiste; Lemaréchal, Claude. 2001. Fundamentals of Convex Anal-
ysis. Berlin et al.: Springer.
Hörmander, Lars. 1963. Konvexa och subharmoniska funktioner. Seminarier, Stock-
holms universitet ht 63 [Convex and subharmonic functions. Seminars, Stockholm
University, Fall Term 1963].
Hörmander, Lars. 1983. The Analysis of Linear Partial Differential Operators I. Distri-
bution Theory and Fourier Analysis. Berlin et al.: Springer-Verlag.
Hörmander, Lars. 1994. Notions of Convexity. Boston et al.: Birkhäuser.
Ioffe, A. D.; Tihomirov, V. M. Dvostvennost~ vypuklyh funkci i kstre-
mal~nye zadaqi. Uspehi matematiqeskih nauk. t. XXIII, vyp. 6 (144), 51–
116.
Jackson, Allyn. 1997. Chinese Acrobatics, an Old-Time Brewery, and the “Much Needed
Gap”: The Life of Mathematical Reviews. Notices Amer. Math. Soc. 44 (3), 330–337.
Jacoby, Ruth. 2015. Så skapades den makabra bilden av nazisternas ”mönsterläger” [Thus
was created the macabre image of the Nazis’ “Model Camp”]. Dagens Nyheter 2015
May 10.
Jessen, Børge. 1987. Matematiker blandt de tyske emigranter: Otto Neugebauer, Werner
og Käte Fenchel, Herbert Busemann, Willy Feller [Mathematicians among the Ger-
man emigrants: Otto Neugebauer, Werner and Käte Fenchel, Herbert Busemann,
Willy Feller]. In: Steffensen, Steffen. På flugt fra nazismen. Tysksprogede emi-
granter i Danmark efter 1933, pp. 86–91. Edited by Willy Dähnhardt and Birgit S.
Nielsen. Second edition. Copenhagen: C. A. Reitzels Forlag.
Kiselman, Christer O. 1978. The partial Legendre transformation for plurisubharmonic
functions. Invent. Math. 49, No. 2, 137–148.
Kiselman, Christer O. 2004. Convex functions on discrete sets. In: Combinatorial Image
Analysis. 10th International Workshop, IWCIA 2004; Auckland, New Zealand, De-
cember 1–3, 2004; Proceedings. R. Klette and J. Zunic, Eds., pp. 443–457. Lecture
Notes in Computer Science 3322.
Kiselman, Christer O. 2010. Inverses and quotients of mappings between ordered sets.
Image and Vision Computing 28, 1429–1442.
 Werner Fenchel, a pioneer in convexity theory 19

Kiselman, Christer O. 2015. Estimates for solutions to discrete convolution equations.

Mathematika 61, issue 02, 295–308.
Kiselman, Christer O. 2019. Language choice in scientific writing: The case of mathe-
matics at Uppsala University and a Nordic journal. Normat. Nordisk matematisk
tidskrift 61, No. 2, 111–132.
Kjeldsen, Tinne Hoff. 2010. History of convexity and mathematical programming: Con-
nections and relationships in two episodes of research in pure and applied mathe-
matics of the 20th century. In: Proceedings of the International Congress of Math-
ematicians, Hyderabad, August 19–27, 2010, volume 4, pp. 3233–3257. Hindustan
Book Agency.
Klee, Victor L., Ed. 1963. Convexity. Proceedings of a Symposium in Pure Mathematics,
Volume VII. Providence, RI: American Mathematical Society.
Landau, L[ev] D[avidovič].; Lifshitz, E[vgeniı̆] M[ihaı̆lovič]. 1960. Mechanics. Translated
by J. B. Sykes & J. S. Bell. Oxford et al.: Pergamon Press. [Russian orginal from
1958.]
le Gendre [Legendre]. 1789. Mémoire sur l’intégration de quelques équations aux dif-
férences partielles. Lû le 1er Septembre 1787. Histoire de l’Académie royale des
Sciences. Année M. DCCLXXXVII. Avec les Mémoires de Mathématique & de
Physique de la même année, pp. 309–351. A Paris, de l’Imprimerie royale
M. DCCLXXXIX.
Lehmer, D. H. 1988. A Half Century of Reviewing. In: Duren, Peter, Ed., with the
assistance of Richard A. Askey and Uta C. Merzbach. A Century of Mathematics in
America, vol. 1, pp. 265–266. Providence, RI: American Mathematical Society.
Mandelbrojt, Szolem. 1939. Sur les fonctions convexes. Comptes Rendus hebdomadaires
des Séances de l’Académie des Sciences 209, 977–978. Séance du 27 décembre 1939.
Mehrtens, Herbert. 1985. Die ,,Gleichschaltung“ der mathematischen Gesellschaften im
national-sozialistischen Deutschland. Jahrbuch Überblicke Mathematik 1985, vol. 18,
pp. 83–103. Mannheim et al.: Bibliographisches Institut.
Mehrtens, Herbert. 1987. Ludwig Bieberbach and “Deutsche Mathematik”. In: Phillips,
Esther R. Studies in the History of Mathematics, pp. 195–241. Studies in Mathemat-
ics, volume 26. Washington, DC: The Mathematical Association of America.
Minkowski, Hermann. 1903. Volumen und Oberfläche. Math. Ann. 57, 447–495. [Also in
Minkowski (1911a:230–276).]
Minkowski, Hermann. 1910. Geometrie der Zahlen. Leipzig; Berlin: B. G. Teubner. [The
first 240 pages were published in 1896.]
Minkowski, Hermann. 1911a. Gesammelte Abhandlungen. Volume 2. Leipzig; Berlin: B.
G. Teubner.
Minkowski, Hermann. 1911b. Theorie der konvexen Körper, insbesondere Begründung
ihres Oberflächenbegriffs. In: Minkowski (1911a:131–229).
Moreau, Jean-Jacques. 1962. Fonctions convexes duales et points proximaux dans un
espace hilbertien. C. R. Acad. Sci. Paris 255, 2897–2899.
Moreau, Jean-Jacques. 1963. Remarques sur les fonctions à valeurs dans [−∞, +∞]
définies sur un demi-groupe. C. R. Acad. Sci. Paris 257, 3107–3109.
Moreau, J. J. 1966–1967. Fonctionelles convexes. Séminaire Jean Leray. Sur les équations
aux dérivées partielles, no 2, pp. 1–108.
Moreau, J. J. 1993 March 26. A private letter to Jean-Baptiste Hiriart-Urruty.
Murota, Kazuo. 2003. Discrete Convex Analysis. Philadelphia: Society for Industrial
and Applied Mathematics.
20 Christer Oscar Kiselman

Pais, Abraham. 1982. ‘Subtle is the Lord . . . ’: The Science and Life of Albert Einstein.
Oxford et al.: Oxford University Press.
Reid, Constance. 1970. Hilbert. Berlin et al.: Springer-Verlag.
Richter-Gebert, Jürgen; Sturmfels, Bernd; Theobald, Thorsten. 2005. First steps in
tropical geometry. In: Idempotent Mathematics and Mathematical Physics, pp. 289–
317, Contemp. Math., 377. Providence, RI: American Mathematical Society.
Rockafellar, R. Tyrrell. 1970. Convex Analysis. Princeton, NJ: Princeton University
Press.
Schappacher, Norbert. 1998. The Nazi era: the Berlin way of politicizing mathematics.
In: Begehr et al. (1998:127–136).
Schøtt, Thomas. 1979. Fundamental research in small countries: Mathematics in Den-
mark 1928–1977. Aarhus: Aarhus Universitet, Matematisk institut.
Schwartz, Laurent. 1997. Un mathématicien aux prises avec le siècle. Paris: Éditions
Odile Jacob.
Segal, Sanford L. 2003. Mathematicians under the Nazis. Princeton, NJ; Oxford: Prince-
ton University Press.
Siegmund-Schultze, Reinhard. 2009. Mathematicians Fleeing from Nazi Germany: Indi-
vidual Fates and Global Impact. Princeton, NJ; Oxford: Princeton University Press.
Singer, Ivan. 1997. Abstract Convex Analysis. New York, NY, et al.: John Wiley & Sons,
Inc.
Valentine, F. A. 1963. The dual cone and Helly type theorems. In: Klee, Ed. (1963:473–
493).

Uppsala University, Department of Information Technology

Paper address: P. O. Box 337, SE-751 05 Uppsala
Amber addresses: kiselman@it.uu.se, christer@kiselman.eu
URL: www.cb.uu.se/˜kiselman

Submitted 2016 March 30; in revised form 2016 April 29 and 2016 December 21.
Accepted for publication in 2016. Published in March 2019.