“Vektorfelder in n-dimensionalen Mannigfaltigkeiten,” Math. Ann. 96 (1926/27), 225-250.

Vector fields in n-dimensional manifolds

by

Heinz Hopf in Berlin

Translated by D. H. Delphenich

___________

§§
Poincaré has proved that in general it is not possible to attach a tangent vector to each
point of a continuously differentiable, closed, boundaryless surface of genus p in such a
way that the resulting vector field is everywhere continuous. He has shown that the sum
of the “indices” of the singularities that thus appear has the value 2 – 2p, from which it
follows that for p ≠ 1 discontinuities must always be present 1). Brouwer has extended
this theorem to n-spheres. Here, as well, the sum of the indices of the singularities is
independent of the special choice of vector field; it is 2 for even-dimensional spheres and
0 for the spheres of odd dimension 2). These facts may follow from a roughly
simultaneous unproved theorem of Hadamard that is an obvious generalization the work
of Brouwer on the subject, that for any n-dimensional, closed, boundaryless manifold that
lies in (n + k)-dimensional (k ≥ 1) Euclidian space the sum of the indices of a tangential
vector field is a topological invariant of the manifold, such that, e.g., for the
determination of the numbers that Brouwer gave for spheres, the consideration of special
vector fields suffices 3). (As Herr Brouwer has informed me, the work of Brouwer and
Hadamard came about piecemeal from an exchange of ideas between the two authors.)
During an examination of the curvatura integra for closed hypersurfaces, I arrived at
a proof of the theorem proposed by Hadamard for the case of k = 1 4); since, as he
likewise discussed, not every n-dimensional closed manifold can be regularly embedded
in the (n + 1)-dimensional Euclidian space, he therefore treated only a special case of the
previous assertion.
In the present work, it will now be proved completely. The theorem will thus be
sharpened in two directions: the one – inessential – sharpening consists in the fact that
one can always make an embedding of the manifold in a space of higher dimension,
which comes about easily for a suitable definition of vector field, in particular, the
interpretation of a vector field as a “small transformation.” The second, however, will be

1
) Sur les courbes définies par les équations différentielles, 3. parties, chap. 13, Journ. de Math. (4) 1
(1885).
2
) Über Abbilding von Mannigfaltigkeiten, Math. Ann. 71 (dated July 1910).
3
) Note sur quelques applications de l’indice de Kronecker in Tannery, Introduction à la théorie des
fonctions d’une variable II, 2nd ed. (1910), no. 42. – In this, the work of Poincaré, Dyck, and Brouwer was
cited; in the questions that arose in the treatises of these three authors, Poincaré and Brouwer treated the
special cases mentioned above, while Dyck indeed proved different versions of the theorems, but not the
theorem formulated by Hadamard.
4
) Über die Curvatura integra geschlossener Hyperflächen, Math. Ann. (1925).
 Vector fields in n-dimension manifolds 2

the one that actually makes the sum of the indices that appear to be a topological
invariant: It is equal to the Euler characteristic of the manifold, which was already to be
expected from its present determination in the special cases. Thus, singularity-free vector
fields are possible only when the characteristic is 0. The question arises whether
conversely a singularity-free vector field can always be constructed in the case of
vanishing characteristic, thus in the case of a closed, boundaryless manifold of odd
dimension 5). This question is answered in the affirmative by showing that the desired
construction comes down to the solution of a certain “boundary-value problem for vector
distributions” that I have treated in connection with other things 6). One of the
consequences of these facts is the theorem: “A manifold admits arbitrarily small fixed-
point-free transformations into itself when and only when its characteristic has the value
0.” In particular, any boundaryless, closed manifold of odd dimension admits
transformations of that sort, while that is never the case for manifolds of even dimension,
in general.
A comparatively broad class of spaces (§§ 1, 2) must be employed for the discussion
of the – mostly known – concepts and facts that concern complexes, manifolds, and their
representations. The connection between the index sum of the singularities of a vector
field and Euler characteristic will be essentially treated in § 3; thus, the (n – 1)-
dimensional structure that leads one back to the proofs for the n-dimensional manifolds is
no longer a manifold, but a “complex:” the boundary complex of the manifold. This
situation makes it necessary that in a complex one can no longer speak of the continuity
of a vector distribution, so one must introduce a new concept: that of the “complex-
continuous vector field.” In § 4, a proof of the auxiliary construction that was made in §
3 will be added, and in § 5 the theorem will be given its ultimate formulation; It will be
regarded, in the aforementioned way, as a fixed-point theorem for small transformations
and conversely, on the basis of the solubility of the “boundary-value problem” in the
likewise aforementioned way; furthermore, it will be shown that the numbers that appear
as the “total curvatures” of closed hypersurfaces 4) can be interpreted as Euler
characteristics in many cases.

§ 1.

Complexes and their representations.

1. In ordinary n-dimensional space, let βn simplexes Tvnn [vn = 1, …, βn] be given; let
their k-dimensional boundary simplexes be denoted by Tvkk [vk = 1, …, βk]. The Tvnn
define a “complex representation” Dn if, between the points of certain Tvnn , which will be
said to be “linked to each other,” associations of the following sort exist:
Let T1n , T2n be linked to each other; there are then two simplexes T1k , T2k [0 ≤ k ≤ n]
that belong to T1n , T2n whose points are related in a one-to-one and continuous way, such

5
) See, e.g., B. H. Tietze, Über die topologischen Invarianten mehrdimensionale Mannigfaltigkeiten,
Wiener Monatsch. für Math. u. Phys. 19 (1908), § 8.
6
) Abbildungsklassen n-dimensionaler Mannigfaltigkeiten, Math. Ann. 96.
 Vector fields in n-dimension manifolds 3

that each T1 p [0 ≤ p ≤ k] of T1k corresponds to a T2p of T2k , while two points A1 , A2 of T1n ,
T2n that do not belong to T1k , T2k are not associated with each other. This association is
transitive − i.e.: if, on the one hand, A1 , A2, and on the other hand, A2 , A3 are associated
points of T1n , T2n ( T2n , T3n , resp.) then A1 and A3 are associated with each other.
As a result of transitivity, we can, for each p [0 ≤ p ≤ n], divide the βp simplexes Tv pp
into αp groups g λpp [λp = 1, …, αp; 1 ≤ αp ≤ βp], such that the Tp that belong to gp are
associated with each other, and analogously the points A can be collected into the groups
a. We call the groups a the “points” and the groups g λpp , the “simplexes” of the
“complex Cn that is represented by Dn,” and say that two points (simplexes, resp.) of Dn
that belong to the same group are “identical in Cn.”

2. If one has β1k = β 2k for each k for two complex representations D1n , Dn2 and one
does not distinguish them with regard to the groupings g λk k of their simplexes, but only
with regard to the point associations within the simplexes Tνkk , then we call them
“isomorphic;” two complexes C1n , C2n that can be represented isomorphically by D1n , Dn2 ,
resp. , may be mapped to each other in a one-to-one and continuous way such that k-
dimensional complexes correspond to each other, as is prescribed by the isomorphism 7),
and we consider them to be indistinguishable from each other.
To each representation Dn there is an “affine” representation that is isomorphic to it −
i.e., one such that the maps between two associated simplexes to each other are affine; in
order to obtain such a representation, one must, for every two simplexes Tνkk , perform
only such affine maps that are uniquely determined by the corners under the association
prescribed by means of Dn.
A representation Dn is called “reduced” when αn = βn in it – i.e., when associations
are given only for boundary points, but not for interior points of Tνnn . One may “reduce”
any representation by omitting certain Tνnn , and we regard the complex represented by the
reduced complex as not being distinct from the original one. In general, we shall focus
on reduced, affine, complex representations in the sequel.

3. The (n − 1)-dimensional boundary simplexes Tνnn−−11 of Dn define, by maintaining

the association prescribed for Dn, an (n – 1)-dimensional complex representation Dn−1. If
Dn is affine then Dn−1 is also affine, although, in general, Dn−1 is also not reduced for a
reduced Dn. We call the complex C n−1 that is represented by Dn−1 the “boundary
complex” of C n.

7
) H. Kneser, Die Topologie der Mannigfaltigkeiten (Anhang), Jahresbericht der Deutsch. Math. Ver. 34,
1 – 4, Heft (1925). – There, only manifolds were considered, so the validity of the argument remains
unchanged for complexes.
 Vector fields in n-dimension manifolds 4

4. If one subdivides each Tνnn of Dn into finitely many sub-simplexes in such a way
that the various Tνkk [1 ≤ k ≤ n] of the resulting decomposition, as long as they are
associated with each other, are “identical in Cn” with each other then what results “by
subdivision” of Dn (Cn, resp.) is a representation D1n of a complex C1n . As is well-
known, Cn and C1n have the same “Euler characteristic;” in the above notation for Cn this
n
is defined to be ∑ (−1) α
k =0
k k
. Under the decomposition carried out on the Tνnn what results

at the same time by subdivision of Dn−1 (Cn−1, resp.) is a representation D1n −1 of the
boundary complex C1n −1 of C1n .
Let An be an affine representation. Thus, any representation that results from
subdivision A1n is also affine. One can perform an arbitrarily dense subdivision of a
given affine representation An as follows: Let m be an arbitrarily large whole number.
One divides each edge Tν11 into m equal parts, and through each such point A, one
intersects planar spaces that are parallel to those T n−1 that belong to the same T n, but do
not include A. In this way, each Tk will be divided into finitely many arbitrarily small
convex polyhedra Pk, and these decompositions of Tk are “identical in Cn” with each
other. One now further divides the Pk into simplexes, and still further, while observing
the association that is present, such that a subdivision of Cn results 8). – Thus, the
following remark is important for a later application: We refer to two polyhedra as
indistinguishable from each other in “shape and position” when they can be converted
into each other by means of a dilatation and a translation – Thus, in an (x1, …, xν)-
coordinate system, by a transformation xν′ = c xν + aν [ν = 1, …, n] – then in shape and
position only finitely many polyhedra come into consideration for the Pn, independently
of m. In fact, if we introduce an affine coordinate system into – e.g. – T1n , whose sides
are T1n −1 , …, Tnn+−11 , such that the corner that is opposite the side Tnn+−11 is the null point, the
edges are the axes through it, and the remaining n corners are the unit points on the axes,
then a Pn that belongs to T1n is a part of a “parallelepiped” Π whose edges are parallel
and proportional to the unit line segments of the coordinate system – namely, of length
1/m − thus, we have a structure that is independent of shape and position of m; indeed, Pn
is one of the pieces of Π that one obtains when one intersects the planar space through
each corner of Π that is parallel to Tnn+−11 , which are likewise determined in shape and
position from now on. Now, let this decomposition of this Pn into simplexes in shape and
position likewise be prescribed from now on 9). – This consequence is true for each
individual Tνnn ; thus, one shows that one can give a representation A1n of An by an
arbitrarily dense subdivision (i.e., a subdivision with arbitrarily large m), whose

8
) Hadamard, loc. cit., no. 10, footnote 2).
9
) One links the center of mass of each Pk [2 ≤ k ≤ n] with each corner of Pk and with the center of mass
of each Pl [2 ≤ l < k] that belongs to the boundary of Pk.
 Vector fields in n-dimension manifolds 5

simplexes are henceforth restricted in regard to shape and position to finitely many given
possible cases, which are determined from An alone.

5. Let T1n be a simplex of Dn and let T1n − k [k ≥ 1] be a boundary simplex of T1n .

T1n − k belongs to k simplexes Tκn − k [κ = 1, …, k]; the planar (n – k)-dimensional space En−k
that includes the T1n − k is the intersection of the k (n – 1)-dimensional planar spaces Eκn −1
that include the Tκn − k . Each Eκn −1 decomposes the n-dimensional space into two parts:
We call the one that contains T1n the “positive side” of Eκn −1 . We call the intersection of
the positive sides of Eκn −1 [κ = 1, …, k] the “interior” of the “k-fold angle Wkn −1 ” defined
by Eκn −1 whose vertex is En−k; the interior, when one includes the boundary, is the “closed
angle space” Wkn . (Thus, one understands W1n to mean the positive half of the space that
is determined by a En−1.) Each Wkn will be bounded by k closed angle spaces Wkn−−11 ,
which belong to the representation Dn−1 of the boundary complex Cn−1 defined by Dn.

6. Let An be a reduced affine representation of Cn, An−1, the associated affine (non-
reduced) representation of the boundary complex Cn−1, and A1n−1 , a reduced affine
representation of Cn−1 in a planar space Fn−1. Let E1n −1 be the planar space that includes
the boundary simplex T1n −1 of An, P1, a point of T1n −1 , and w1, a ray of E1n −1 that emanates
from P1 . If T2n −1 , …, Trn −1 are boundary simplexes of An that are identical with T1n −1 in Cn
and P2, …, Pr are the points of them that are identical with P1 then the rays w2, …, wr
that begin at Pρ and lie in the Eρn −1 are defined by means of the affine association between
the Tρn −1 [ρ = 1, …, r] that are included in Eρn −1 . The r rays defined in An correspond in
A1n−1 , by means of the affine and transitive association, to precisely one w* of Fn−1, which
emanates from the point p of the simplex tn−1 of A1n−1 that corresponds to P1, which is the
image of the Tρn −1 . If P1 and w1 simultaneously belong to many (n − 1)-dimensional
boundary simplexes T n−1 of An then the ray w1 and the rays w2 , …, wm [m ≥ r] of An that
are identical with them in Cn correspond to many rays Fn−1, which therefore all lie in
boundary spaces of A1n−1 and are mapped to each other by means of the affine and
transitive relation between the boundary spaces A1n−1 .

7. Let k ≥ 1, let Tn−k be a boundary simplex of An, P, a point of Tn−k, En−k, the planar
space that includes Tn−k, Wkn , the k-fold angle that belongs to En−k as a vertex, u, a ray that
is based at P and directed into the interior of Wkn , u , the ray diametrically opposite to u,
and e2, a two-dimensional half-plane spanned by u and u . e2 intersects each of the k
boundary spaces Eκn −1 [κ = 1, …, k] that include En −k in a ray wκ . If Tκn −1 are the
 Vector fields in n-dimension manifolds 6

boundary simplexes that belong to the Eκn −1 then each Tκn −1 corresponds to a simplex tκn −1
of A1n−1 , and on each of them there is a tκn − k that is the image of Tn−k, and to each tκn − k there
is an angle ( wkn−−11 )κ of A1n−1 ; at each tκn − k there is an image point pκ of P, and each wκ
corresponds to a ray wκ∗ of Pn−1 based at pκ . We consider the directions of these wκ∗
more closely; there are two cases to distinguish:
I. (Main case): e2 has only the point P in common with any two of the Eκn −1 ; each of
the rays wκ then belong to only one Tκn −1 ; thus, no wκ∗ lies in a (n – 2)-dimensional
boundary space of A1n−1 . If one rotates u in e2 into the position u then let w1 be the first
intersection with an Eκn −1 ; w1 is then the only wκ that belongs to the boundary of Wkn ,
since all other wκ point to the exterior of Wkn . Thus, w1∗ points to the interior of ( wkn−−11 )1 ,
while every other wκ∗ is directed to the exterior of its ( wkn−−11 )κ .
II. (Boundary case): e2 has, in addition to P, another point in common with some of
the Eκn −1 , hence, a ray; thus, not all of the wκ are distinct from each other. The k rays wκ
can be collected into i groups (i < k) in such a way that the rays of one group overlap in a
ray w′j [j = 1, …, i]. For the w′j , the facts that were established for the wi in case I
remain correct. If w1′ is the first intersection of the rotated ray u with a Eκn −1 and w1′ is
identical with only one wκ then the result of the argument in case I remain unchanged,
that of the wκ∗ [κ = 1, …, k] precisely one of them − namely, w1∗ − points into the interior
of its ( wkn−−11 )1 , and every other wκ∗ is directed to the exterior of its (wkn−−11 ) . By
comparison, if w1′ is identical with some wκ then this fact must be modified in such a
way that certain wκ∗ – say, w1∗ , …, w∗m (namely, the ones that correspond to w1′ ) – have
boundaries that belong to their ( wkn−−11 )κ , and indeed in such a way that they are mapped to
each other by means of the affine, transitive association defined in A1n−1 , while every
other wκ∗ [κ = m + 1, …, k] points to the exterior of its ( wkn−−11 )κ .
Before we utilize the facts thus established we must first consider a special complex.

§ 2.

Manifolds and their representations

1. A corner Tν00 of a reduced representation of Cn is called a regular corner when the

Tνkk [k = 1, …, n] that contain it, along with the points in Cn that are identical to it, are
associated with each other like the adjacent simplexes and boundary simplexes of a
certain simplex star of the n-dimensional Cartesian space. Thus, we understand a
simplex star to mean an element Sn composed of finitely many simplexes in such a way
that all simplexes of a corner have A in common, while all other corners lie on a sphere
 Vector fields in n-dimension manifolds 7

around A 10); Tν00 is called and “interior corner” or a “boundary corner” according to
whether A lies in the interior or on the boundary of Sn.
A complex that possesses only regular corners – whether interior or boundary corners
– and is, in addition, “connected,” – i.e., one in which one can get from any T1n to any
other T2n along a chain that links each Tn to the ones that follow – is called a (closed)
“manifold” Mn. If Mn has only interior corners then one calls it “boundaryless” 11); if Mn
also has boundary corners then all “boundary points” define a finite number of closed
boundaryless (n – 1)-dimensional manifolds 12); thus, a point is called a boundary point
of Mn when it belongs to a boundary simplex such that under each association of
simplexes, the simplexes of a simplex star Sn corresponds to a simplex constructed from
the boundary points of Sn .
A complex whose representation D1n comes about by subdividing a representation Dn
of a manifold Mn is, as would follow from the definition, itself a manifold. It gives us
nothing different from Mn.

2. We consider the simultaneous map of some simplexes of a representation in Mn

that are bound together to a subset of an element in Cartesian space: First, let Tνnn be the
simplexes of an affine representation An of Mn, T00 , a corner, and S 0n , the associated
simplex star. The association that exists between the k-simplexes Z ρk (0 ≤ k ≤ n) of S 0n ,
on the one hand, and the simplexes Tνkk , on the other, as long as it is defined, may then be
refined into a map in which one carries out that uniquely defined affine map from each
simplex Z ρk of S 0n and the Tνkk that is associated with it under the association of the
corners of Z ρk to those of Tνkk ; in this way, S 0n will be mapped in a one-to-one and
continuous manner to that subset Σ0n of Mn that is represent in An by all of the corners T00
or a simplex Ti n that contains the corner Ti 0 that is identical to it in Mn.

3. The subset Σ0n of Mn that is thus constructed in a piece of Cartesian space includes
all simplexes that define the neighborhood of point, namely, the ones represented by T00 ;
we now seek an analogous map of the entire neighborhood of a simplex of a
representation of Mn; we define:
An affine representation A1n of Mn is called a “neighborhood representation” when
each of its simplexes T0n gives rise to an element E0n of ordinary space with the following
properties: If Ω n0 is the “simplicial neighborhood of T0n ” – i.e., the subset of Mn that is
represented by the simplexes Ti n [i = 1, …, m] in A1n that are linked to T0n − then E0n can

10
) This definition of simplex star deviates inessentially from the one that was given by Brouwer in the
reference cited in 2).
11
) Obviously, Mn then has nothing but “interior” points in the ordinary sense; on this, cf., the report of
H. Kneser cited in 7).
12
) Hadamard, loc. cit., no. 16.
 Vector fields in n-dimension manifolds 8

be decomposed into m + 1 simplexes z0,n i [i = 0, 1, …, m] and mapped to Ω n0 in a one-to-

one and continuous manner, such that z0,n i is affinely related to Ti n [i = 0, 1, …, m] 13).
We show that one can present a neighborhood representation for each Mn: Let An be
the aforementioned affine representation, relative to which one has been given the Sνn0
and Σνn 0 for the map described for a single v0 [v0 = 1, …, β0]. We present a representation
A1n of Mn by subdivision by dividing each one-dimensional edge Tν11 into n + 1 equal
parts: Lay a planar (n – 1)-dimensional space that is parallel to the faces Tνnn−−11 through the
dividing points, and decompose the resulting convex polyhedron into simplexes. If t0n is
a simplex of the representation A1n and, say, T0n is the simplex of An that belongs to t0n
then there is an (n – 1)-dimensional face of T0n that has no point in common with t0n . In
fact, if we introduce (as in § 1.4) an affine coordinate system ξ1, …, ξn into T0n , whose
null point is the corner of T0n that also intersects the faces T1n −1 , …, Tnn −1 , whose axes are
the edges that emanate from the null point, and whose unit points on the axes are the
remaining corners of T0n then the coordinates of each point of the simplexes of the
representation A1n that has a point in common with each of the faces T1n −1 , …, Tnn −1
satisfy the inequalities:

n
1
ξi ≤
n +1
[i = 1, …, n]; ∑ξ
i =1
i < 1;

this simplex thus possesses no point in common with the latter side Tnn+−11 of T0n , which is
n
defined by the equation ∑ξ
i =1
i = 1. Thus, to t0n there is a face of T0n – e.g., T0n −1 − that t0n

has no point in common with. If T00 is the corner point of T0n that is opposite to T0n −1
and S 0n is the simplex star that belongs to T00 then one can clarify the aforementioned
one-to-one and continuous piecewise affine relation between S 0n and the simplexes of An
that contain the corner that is identical with T00 in Mn in t0n , as well as in any simplex of
A1n that is linked to t0n and is therefore in the “simplicial neighborhood” Ω n0 of t0n – i.e.,
A1n is a neighborhood representation.
We can now directly employ the aforementioned simplexes zµn n , µ n in place of the tµnn
for the representation of Mn, and thus, when we henceforth set zµn n , µ n = Tµnn , in order to
revert to our previous notation, obtain a neighborhood representation that is as follows:

13 n n n
) In general, zν 1 ,ν 2 then denotes the sub-simplex that is the image of Tν 2 when the element Eν1
n
represents the simplex neighborhood of Tν1 ,
 Vector fields in n-dimension manifolds 9

14
To each simplex Tµnn ), one attaches simplexes zµn n ,i [i = 1, …, mµ n ] to those boundary
simplexes that do not represent any boundary point of Mn, which, together with Tµnn ,
define an element Eµnn , the one-to-one image of the simplex neighborhood Ωnµ n of Tµnn in
Mn; thus, any two simplexes zµn n , µ n , zµn n , µ n , that belong to two different elements
1 2

Eµ n , Eµ n , and which likewise correspond to the piece of Mn that is represented by Tµnn =

n n
1 2

zµ n , µ n by the mediation of Mn, are affinely mapped to each other.

n

4. This “distinguished neighborhood representation” of Mn, which we would again

like to denote by An, is suitable for the investigation of certain transformations of Mn:
A single-valued continuous map of Mn onto itself or a subset of itself is called a
“neighborhood transformation” of Mn, relative to An, when each point of Mn that is
represented by a point of Tµnn goes to a point of the simplex neighborhood Ωnµ n of Tµnn .
For example, suppose that the transformations fi are a series of transformations f1, f2,
… that converge uniformly to the identity in all of Mn as neighborhood transformations,
relative to any arbitrary distinguished neighborhood An with a certain index that is
independent of An; we occasionally express this by saying that an “arbitrarily small
transformation” of Mn is a neighborhood transformation, relative to any distinguished
normal neighborhood.
If f is a neighborhood transformation relative to An then one defines, in a unique
manner, a single-valued and continuous map f µ n of each simplex Tµnn onto a point set
that belongs to the element Eµnn . We assume that f has at most finitely many fixed points
and that they correspond to only interior points of Tµnn . If we now attach to each point P
of Tµnn the vector v(P) that points to the point f µ n ( P) then this vector field B is, in a
certain sense, single-valued and continuous in all of Mn, except for the fixed points 15). It
has, by the use of the notations of § 1, the following properties, among others:
A. B is single-valued and continuous on each individual Tµnn [µn = 1, …, βn], except
for at most finitely many points that lie in its interior.
B. Let P0 be a boundary point of T0n that belongs to a boundary simplex T0n − k [1 ≤ k
≤ n]. Let Tρn − k [ρ = 1, …, r] be the boundary simplexes of the other Tµnn that are identical
to T0n − k , Pρ , the points of Tρn − k that are identical with P0 , and let (Wkn ) ρ [ρ = 1, …, r],
the k-fold angle whose vertex is Tρn − k . Then, one of the following two cases will appear:

14
) Thus, µn now denotes an index that runs from 1 to αn, just as νn did in § 1.
15
) From on, we shall, unless expressly stated to the contrary, consider only the continuity of direction,
but not the length of the vectors; zero loci of the vector field then amount to singularities.
 Vector fields in n-dimension manifolds 10

I. (Main case): Of the r + 1 vectors v(Pρ), precisely one of them points to the
interior of its (Wkn ) ρ , while all others are directed to the exteriors of their (Wkn ) ρ .
II. (Boundary case): Some of the v(Pρ) are attached to the boundaries of their
(Wkn ) ρ , and are constructed by means of the affine and transitive associations that exist
between the boundary spaces, while the remaining v(Pρ) point to the exteriors of their
(Wkn ) ρ .

As one knows, case II appears when and only when P0 and f0(P0) belong to the same
boundary simplex.

§ 3.

Complex-continuous vector fields

In the formulation of properties A and B of the vector field B that was given in the
conclusion of the previous paragraph, no use was made of the fact that we have a
neighborhood representation of a manifold before us. If we have a reduced affine
representation An of an arbitrary complex Cn then none of the aforementioned properties
will become meaningless when we replace Mn with Cn. We may therefore define:
An association B of vectors v(P) to the points P of the reduced, affine representation
A of the complex Cn is called a “complex-continuous vector field on Cn (relative to An)”
n

when it satisfies the requirements A and B. [See the “Appendix” at the conclusion of this
paper.]

1. Of the properties of complex-continuous vector fields that we will be occupied

with in the sequel, let us first establish: If A1n is a complex representation that comes
about by subdivision of An then B is also complex-continuous relative to A1n , assuming
that no singular point of B lies on a boundary simplex of the representation A1n . One
convinces oneself of the validity of this assertion by establishing that B has property B,
not only, as assumed, on the boundaries of the representation A1n , but also on the new
boundaries that come about by subdivision, on which B is continuous in the ordinary
sense.

2. A second important property of the complex-continuous vector fields concerns the

“projection of the complex-continuous vector field B onto the boundary complex.” One
understands this to mean: Cn, An, and B have the meanings as all along, but let An−1 be
the non-reduced representation of the boundary complex Cn−1 that is defined for An, let
A1n−1 be a reduced affine representation of Cn−1, let Tνkk [k = 0, …, n; vk = 1, …, βk; βn =
αn] be the simplexes An, and let tλk [k = 0, …, n − 1; λk = 1, …, γk; γn−1 = α n−1] be the
k
 Vector fields in n-dimension manifolds 11

simplexes of A1n−1 . On the boundary of Tνnn , let there be given a field Uν n of vectors u(P)
with the following properties:
a) u(P) is directed into the interior of Tνnn .
b) If P lies on a Tn−2 then the directions u(P) and v(P) do not agree.
c) There are at most finitely many points P at which the directions of u(P) and v(P)
do agree.
We shall postpone for a moment the discussion of whether such vector fields Uν n
always exist.
On each Tn−1 of Tνnn , we now focus on the points P at which v(P) is either directed to
the positive side of the planar space En−1 that contains T n−1 or lies in En−1 – in which v(P)
thus belongs to the “closed angle space W1n ” in question – and project these v(P) from
u(P) onto E n−1; i.e., we present those vectors w(P) at which En−1 will be intersected by
the half-plane e2 that is spanned by the u(P), v(P), and the vector u ( P) thus, the stated
sequence of vectors in e2 is always the following one: u, v, w, u . This construction will
be possible only at the points P considered at which u(P) and v(P) agree, which are at
most finite in number. The vector w(P) now corresponds to either (cf., § 1.6) precisely
one vector w* in A1n −1 or it corresponds to several w* that lie in the boundary spaces of
A1n −1 and are affinely mapped to each other. We call the totality W* of the vectors w*
thus produced in A1n −1 a “projection of the field B” and assert that it represents a
complex-continuous vector field on Cn−1. In fact: That W* possesses the properties A and
B that are characteristic of complex-continuous vector fields follows from described
construction of W*, as well as the fact that the demand B on B is fulfilled for k = 1, in
particular. That W* possesses property B for every k* ≤ n − 1 is obtained from the fact
that B possesses this property for every k = k* + 1, as well as the behavior of projected
vectors that was discussed in § 1.7, by which, in particular, w* belongs to its closed angle
space wkn−−11 when and only when w is the first intersection of the vector u rotated into e2
with a boundary space En−1 of Wkn , and therefore, when v belongs to the closed angle
space .

3. We now relate the indices of the singularities of B to the indices of the

singularities of W*. Let sν n be the sum of the indices of those singularities of B that lie
αn
in Tνnn and let sn = ∑ sν n then be the sum of the indices of all singularities of B;
ν =1
n

furthermore, let sn−1 be the sum of all indices of all singularities of W*, let aν n be the sum
of the coincidence indices 4) of the two maps of the boundary of Tνnn onto the sphere of
directions, which will be mediated by Uν n and the boundary field Bν n associated with B
 Vector fields in n-dimension manifolds 12

αn
(in this sequence!), and let a = ∑ aν n be the sum of all of these coincidence indices.
ν n =1

Now, the number a may be determined in two ways: A singularity of W* exists where
and only where Uν n and Bν n have coincidence loci. The index of such a coincidence is
equal to the index of the singularity of the field of projected vectors w, and thus, also
equal to the index of the singularity of W*, assuming that one orients the (n – 1)-
dimensional boundary space En−1 that the point considered belongs to in such a way that a
positively oriented system of axes of En−1, together with a vector of Uν n as the last axis,
defines a negative system of axes for n-dimensional space 16); in our case, however, the
indicatrix of E n−1 is determined to be the boundary indicatrix of Tνnn ; i.e., an n-fold
system of axes that is defined in the manner just described is positively oriented 2). It
follows that the coincidence index of Uν n and Bν n is equal and opposite to the index of
the singularity of W* at the corresponding point, and is therefore:

(1) a = − sn−1.

αn
On the other hand, a = ∑ aν n is to be determined in the following way: aν n is the sum of
ν n =1

the coincidence indices of maps of the boundary of Tνnn onto the sphere of directions that
is mediated by Uν n and Bν n . The map mediated by Uν n has the degree (−1)n, since all
vectors u(P) are directed into the interior of Tνnn , and it is thus continuous when one
establishes that its starting points go to vectors that point to a fixed interior point. The
map mediated by Bν n has the degree sν n . Thus, one has the equation 4):

(2) aν n = (−1)n−1 ⋅ (−1)n + sν n = − 1 + sν n ,

and from this, what follows upon summing is a second value for a:

αn
(3) a= ∑ aν n = − α n + s n.
ν n =1

Comparing the two values of a gives:

(4) s n = α n – sn−1 .

4. We now begin the proof of the following theorem:

16
) For the proof, cf., the paper cited in 4) of § 1.
 Vector fields in n-dimension manifolds 13

Theorem I. The index sum of the singularities of a complex-continuous vector field

on Cn is equal to the Euler characteristic of Cn multiplied by (−1)n.

We give the proof by going from n – 1 to n.

First, let n = 1. Cn = C1 is therefore a system of α 1 line segments whose corners are

composed of α 0 groups; the corners that belong to the one group are identical in C1 and
represent a point of this complex. (We can think of this identification as being carried
out in – say – three-dimensional space by fastening them together.) The complex-
continuous vector field consists of vectors that lie in the lines that the line segments
belong to and possess singularities in the interior of the line segments with the index sum
s1. At each of the α 0 points of the complex that are represented by the β 0 corners of the
line segment it exhibits precisely one line segment that is directed into its interior. Thus,
if – a 17) is the number of all corner vectors that are directed into the interior of the line
segments then one has:
(1*) a = − α 0.

We determine a in a second way, in which we consider each of the line segments Tν11
individually: A singular location for the 1-dimensional vector field B is – in a reasonable
application of the definitions that pertained to n dimensions – to be understood as having
the index + 1 in the event that all of the vectors in its neighborhood point outward, the
index – 1 in the event that all of the vectors in its neighborhood point inward, and the
index 0 in the event that all vectors in its neighborhood have the same direction (and the
singularity is therefore removable). Singularities with other indices do not occur for n =
1. Let sν 1 be the sum of the indices of all singularities of B on Tν11 and let − aν 1 the
number of corner vectors that point into the interior of Tν11 ; one then has sν 1 = − 1, 0, or +
1, according to whether − aν 1 = 2, 1, or 0, resp. In any case, one thus has:

(2*) aν 1 = − 1 + sν 1 .
Summing gives:
(3*) a = α1 + s1,

and from this, it follows, by comparison with (1*), that:

(4*) s1 = α1 – α0 = − (α0 – α1).

For n = 1, this is the relation that we asserted in our theorem. We now assume that it
has been proved for n − 1. Then, if Cn is a complex and B is a complex-continuous
vector field on it, such that one can construct a vector field Uν n with the properties a), b),
c) stated above in 2, then, since W* is complex-continuous and the theorem is true for the
boundary complex Cn−1, since one must have:
17
) Notations and signs are chosen by specifying the agreement with the n-dimensional case.
 Vector fields in n-dimension manifolds 14

n −1
sn−1
= (−1) n−1
⋅ ∑ (−1) α
k =0
k k
,

the stated relation follows from (4):

n −1 n −1
(5) sn = αn – (−1)n−1 ∑ (−1) k α k = (−1)n ∑ (−1) k α k .
k =0 k =0

Thus, we do not know whether one can always construct the field Uν n . However, since
the complex that arises by subdivision of Cn has the same Euler characteristic as Cn,
Theorem I is proved completely, as long as the validity of the following Lemma is
proved, which will happen in the next paragraph:

If An is a reduced, affine representation of the complex Cn and B is a complex-

continuous vector field on it then one can, by subdividing An, present a representation Bn
of An and a complex-continuous vector field P in Bn whose singularities are identical
with those of B relative to position and index, in such a way that a vector field Uν n can
be constructed in each n-dimensional simplex tλnn of Bn that possesses the properties a),
b), c) relative to P.

§ 4.

Completion of the proof of the theorem on the index sum of the

singularities of a complex-continuous vector field

In order to preserve Bn and P in the desired manner, we first remove the vectors of
B that are based in the interiors of the simplexes Tνnn of An, and replace them with a new
vector field P that has the same boundary field Bν n and the same singularities with the
same indices as B, but is analytic in a certain neighborhood Q(Pρ) of the singular point
Pρ – naturally, it is itself removed; the fact that there is such a P will be shown in another
18
place ). P is complex-continuous on An, since has the same boundary field as the
complex-continuous field B; P is therefore (from § 3.1) also complex-continuous in
each complex representation Bn that arises by subdivision of An, as long as none of the
singular points lie on a boundary simplex of Bn. Now, if γ is an arbitrary positive
number then we present a representation Bn(γ) by subdivision of An that fulfills the
following conditions, except for the aforementioned consideration of the singular loci:

18
) § 5, problem 4, supplement to the work cited in 6).
 Vector fields in n-dimension manifolds 15

Bn(γ) is a sufficiently fine decomposition that 1. Each simplex tn of Bn(γ) that

contains a singular point Pρ lies completely in the analytic neighborhood Q(Pρ). 2. The
fluctuation of the vector direction of B at each tn that does not lie completely in a Q(Pρ)
is smaller than γ; when 1 is already satisfied, condition 2 is always fulfilled by further
subdivision, as a result of the uniform continuity of P outside of Q(Pρ). 3. Bn(γ) shall
have the property that each of the simplexes tn coincide with one of finitely many
simplexes in shape and position (§ 1.4), which is henceforth determined by An; that the
fulfillment of 3 is compatible with an arbitrary refinement of the subdivision was shown
in § 1.4.
We now prove that for a sufficiently small γ one can attach a vector field U λ n to the
boundaries of the tλnn in the desired manner. In order to determine such a γ, we first focus
on a simplex τ ρn ; let Eνn −1 [ν = 1, …, n + 1] be the planar spaces that bound τ ρn , whose
positive sides are defined as in § 1.5. We understand the phrase “the negative star of
directions σρ of τ ρn ” to mean a system of n + 1 unit vectors an attached to a fixed point O
of space, which are directed such that an [ν = 1, …, n + 1] does not point to the positive
side of Eνn −1 ; thus, it either points to the negative side of Eνn −1 or it is parallel to Eνn −1 .
The σρ define an (n − 1) ⋅ (n + 1)-dimensional closed set Sρ . Among the 12 n ⋅ (n + 1)
angles between each two directions of a σρ there is a largest one m(σρ); thus, angle
quantities must be measured so that they always lie between 0 and π, inclusive. m(σρ) is
always positive. If one had m(σρ) = 0 then that would mean that all vectors aν of a σρ
overlap in a single vector α, and that this vector a would be directed to the positive side
for no Eνn −1 ; however, that is impossible, since an oriented line that is parallel to a and
goes through an interior point of τ ρn is directed to the positive side of that Eνn −1 , through
which it enters τ ρn . One therefore always has m(σρ) > 0. On the other hand, m(σρ) is a
continuous function on the closed set Sρ so it attains its lower limit γρ at some point;
hence, one also has γρ > 0.
We now define γ to be the smallest of the r numbers γ1 , …, γr and must prove that
one can construct a vector field U with the properties a), b), c) (cf., § 3), after establishing
the subdivision Bn(γ) either (Case α) on each simplex tn, on whose boundary 19
) the
fluctuation of P is smaller than γ, or also (Case β) on each simplex tn, on whose
boundary P is analytic.
We begin with Case α: t0n thus has the property that the angle between any two
vectors that are attached to the part P0 of P that is found on its boundary is smaller than
γ; we then assert that there is, among its boundary spaces F1n −1 , …, Fnn+−11 , at least one of
them such that all vectors of P0 are directed to its positive side. Otherwise, this would

19
) It suffices to consider P on the boundary of the tn.
 Vector fields in n-dimension manifolds 16

allow one to define a negative direction star σ for t0n from the vectors of P0, and this
would likewise be a negative direction star for that τ ρn that agrees with t0n in shape and
position; one would then have m(σ) ≥ γρ ≥ γ, contrary to the fact that the fluctuation of P0
is less than γ. Therefore, let all vectors of P0 be directed to – say – the positive side of
F1n −1 . Let A be an interior point of the boundary simplexes t1n −1 of t0n that belong to
F1n −1 and let g be a ray that emanates from A and is directed into the interior of t0n ; let
t2n −1 , …, tnn+−11 be the remaining (n – 1)-dimensional boundary simplexes of t0n , let P0 be
the part of P0 that belongs to it, M, the (possibly empty) set of points of g in which g will
be cut out from the rays determined by the vectors of P0 . A does not belong to M, since
otherwise the ray of P0 that contains A would not be directed to the positive side of
F1n −1 . M is, however, closed; thus, there are points on g in the interior of t0n that do not
belong to M; let B be such a point. We now next define the field U0 to be constructed on
the boundary of t0n from the t2n −1 , …, tnn+−11 by the demand that these vectors all go through
B. It then certainly fulfills conditions a), b), c) there if it is everywhere directed to the
interior of t0n and has no point of coincidence at all with P0 . We must now construct U0
at the interior points of the simplexes t1n −1 , on whose boundary, it is already established.
If we consider U0 as pointing to the positive side of F1n −1 on this boundary and P0 as
pointing to the positive side of F1n −1 on all of t0n then we can determine U0 at the interior
points of t1n −1 by the following prescription: If P is an interior point of t1n −1 that is
different from A then let P be the intersection point of the ray AP with the boundary of
t1n −1 . Let p(P), p( P ) , u ( P ) be the vectors of P0 (U0, resp.) attached to P ( P , resp.),
q( P ) , the projection of the vector p( P ) from the vector u ( P ) onto E1n −1 (i.e., as before,
the intersection of with the half-plane spanned by u ( P ) , p( P ) , and the vector u ( P ) that
is diametrically opposite to u ( P ) ), and let q(P) be the vector that is attached to P and
parallel to q( P ) . The vector u(P) to be defined shall now be the vector of the two-
dimensional angle between 0 and π that is spanned by p(P) and q(P), this angle being
divided up such that the angle ratio ∢ {p(P), u(P)}: ∢ {u(P), q(P)} is equal to the
product of the angle ratio ∢ { p( P ) , u ( P ) }: ∢ { u ( P ) , q( P ) } and the line segment ratio
AP : AP ; at A itself, one shall u(A) = p(A). Moreover, the field U0 that is defined on the
entire boundary of t0n satisfies all requirements: It is continuous, everywhere directed into
the interior, and has a single coincidence point A with P0 .
Case α is therefore dealt with, and we then go on to case β, by assuming that P0 is
analytic on the boundary of t0n . Let Kn be a solid ball that lies completely in the interior
of t0n . There then exists a positive angle δ such that every angle is greater than δ whose
 Vector fields in n-dimension manifolds 17

vertex and one side belong to the boundary of t0n , while the other side contains a point of
Kn. We divide the boundary simplexes t2n −1 , …, tnn+−11 into sub-simplexes sρn −1 that are
small enough that that fluctuation of P0 at each individual is smaller than δ; then, if one
vector of P0 that belongs to a point of sρn −1 points to a point of Kn then all P0-vectors of
sρn −1 point to the interior of t0n . The rays that are established by the vectors of P0 that are
attached to the (n − 2)-dimensional boundary simplexes sσn − 2 of sρn −1 define a finite
number of a analytic, (n – 2)-dimensional hypersurface pieces; thus, there are certain
points in Kn that do not lie on any hypersurface; let C be one such point. If we then
define U0 on the sσn − 2 by the demand that the vectors u(P) point to C then no coincidence
point with P0 is present there. We encounter this definition for those sρn −1 in which no
ray that belongs to P0 points to a point of Kn; in the remaining sρn −1 all vectors u(P) point
to the interior of t0n , and the same is true for the vectors of U0 that are already attached to
its boundary. Thus, in each one of them, by the procedure with which we treated the
simplex in case α we can construct vectors u(P) that go inward from them and which are
continuously linked to the vectors of U0 that are already present on the boundary of sρn −1 ,
and coincide with the field P0 at precisely one point in the interior of sρn −1 .
Thus, case β is also dealt with, the validity of the lemma formulated at the end of the
previous paragraph is shown, and Theorem I is proved completely.

§ 5.

Fixed points of small transformations and singularities of

continuous vector fields in closed manifolds

We now make some applications of Theorem I and restrict ourselves exclusively to

the case in which Cn = Mn is a closed manifold (with or without boundary).
Let each point P of Mn be associated with a neighborhood U(P) such that it will be
entirely represented in each Eµnn that is the image of the simplex neighborhood of the
simplexes that contain P when one establishes a definite “distinguished neighborhood
representation” An of Mn – in the terminology of § 2; this condition is certainly fulfilled
for sufficiently small neighborhoods U(P). Now, let f be a single-valued and continuous
map of Mn onto a point set that belongs to Mn and is “small” enough that P, as well as the
image f(P), belongs to the neighborhood U(P); moreover, f has no fixed points on the
boundary, in the event that Mn has a boundary. Then f is a “neighborhood
transformation” relative to An and generates a complex-continuous vector field whose
singularities, which, since they are interior points of Mn, we may assume appear only in
 Vector fields in n-dimension manifolds 18

20
the interior of Tµnn ), are identical with the fixed points of f in position and index. From
Theorem I, it then follows that:

Theorem II. The sum of the indices of the fixed points of a sufficiently small
transformation of the closed manifold Mn into itself is, assuming that at most finitely
many fixed points appear, equal to the Euler characteristic of Mn multiplied by (−1) n.

From this, one obtains:

Theorem IIa. Any sufficiently small transformation of a manifold with an Euler

characteristic that is different from 0 into itself possesses at least one fixed point.

We now pose the question of whether there are then arbitrarily small transformations
with at most finitely many fixed points in any Mn. One recognizes that this question can
be answered affirmatively as follows – always while employing the notation of § 2: Let
T1n , …, Tαnn be the simplexes of An and E1n , …, Eαnn , the elements that represent the
simplex neighborhoods of Tαnn . On the boundary of T1n , one defines a continuous field of
non-vanishing vectors whose endpoints belong to E1n , and which also determines the
lengths; by means of the affine maps that exist between the subsets of the various Eµnn
they correspond to vectors on certain boundary simplexes of certain of T1n , …, Tαnn . We
attach these vectors to the points that they belong to in such a way that now a subset of
the of the boundary simplexes of T1n , …, Tαnn possess vectors. We now attach a field of
vectors to the entire boundary of T2n whose endpoints lie in E2n and which possibly
includes ones that are attached to certain boundary simplexes of T2n ; that this attachment
of vectors is always possible was shown in the paper “Abbildingsklassen n-dimensionaler
Mannigfaltigkeiten” 6) (§ 5.2, 5.3). We then proceed for n = 3, 4, …, αn until the
boundaries of all Tµnn are completely possessed with vectors. We then choose a point Pµ n
in the interior of each Tµnn and associate each point P of Tµnn that is different from it with
that vector PP′ that is parallel to the vector of that boundary point P of Tµnn onto which
the P will be projected from Pµ n , and whose length behaves in relation to the stated
boundary vector as the line segment Pµ n P does in relation to the line segment Pµ n P ; we
associate the point P itself with a vanishing vector. In this way, a vector field with the
singularities Pµ n is defined. By the prescription that each point shall go to that point of
the vector PP′ that divides the line segment PP′ in the ratio t : 1 – t, a neighborhood

20
) To each representation An of Mn there is a representation that is homeomorphic to it, in the sense of
combinatorial topology − i.e., one that comes about by the decomposition and combination of simplexes –
in which finitely many prescribed interior points of Mn will be represented by interior points of the n-
dimensional simplexes.
 Vector fields in n-dimension manifolds 19

transformation ft is defined for each t between 0 and 1. The family of ft converges

uniformly to the identity when t approaches 0; each of these maps has the points Pµ n , and
only these points, as fixed points.
Thus, there are arbitrarily small transformations of Mn with finitely many fixed
points. We infer a consequence from this: If M 1n is a manifold that is homeomorphic Mn
– i.e., one that can be mapped onto Mn in a one-to-one and continuous way – then a
transformation with finitely many fixed points can be constructed in Mn such that each
point moves so slightly from its starting point that this map is a neighborhood
transformation, not only relative to a representation An of Mn, but also relative to a
representation A1n of M 1n . Now, since the index of a fixed point is a topological
invariant of the transformation in question 4), this yields, on the grounds of Theorem II,
the following well-known:

Theorem III. Homeomorphic manifolds have the same Euler characteristics.

This theorem is one of the classical and simplest theorems of combinatorial topology
5), in which one regards two manifolds as homeomorphic when their representations
possess subdivisions that are isomorphic to each other (cf., § 1). The proof carried out
above is valid for topology in the broader sense in which one already designates two
manifolds as being homeomorphic when they can be mapped onto each other in a one-to-
one and continuous way. Theorem III has also been proved by Alexander 21) from this
general viewpoint.
We now pursue further the question raised above of the existence of arbitrarily small
transformations with finitely many fixed points: Is it possible to give an arbitrarily small
transformation that possesses fixed points with prescribed indices q1, …, qm at the
prescribed interior locations Q1 , …, Qm (m ≥ 0), with only the condition that its sum be
equal to the characteristic of Mn multiplied by (−1) n? This is, in fact, always possible 22).
The points P1 , …, Pα n , Q1, …, Qm may then be included in an element F that belongs to
Mn 23), and in it, a further element F1 can be given that includes the stated points in the
interior. We now choose – with the notation above − t sufficiently small that that the
image of F1 under ft lies completely in F. Let F′ be a topological image of F that belongs
to ordinary space, F1′ , the image of F1 in it, and let P1′ , …, Pα′n , Q1′ , …, Qm′ be the
images of P1 , …, Pα n , Q1, …, Qm , resp. The map ft corresponds to a map f t′ of F1′ onto
a subset of F′; its fixed points are P1′ , …, Pα′n , the associated indices are, due to their
topological invariance, the same as the corresponding indices under the map ft . The
vectors that point from the boundary points of F1′ to the image points under the map f t′

21
) J. W. Alexander II, A proof of the invariance of certain constants of Analysis Situs, Trans. of the
Am. Math. Soc. 16 (1915). – There, the invariance of the Betti numbers was proved for the topology in the
broader sense. Since the Euler characteristic is expressible through the Betti numbers (cf., e.g., Tietze, loc.
cit.) Theorem III is thus proved; cf., also H. Kneser, loc. cit., footnote 2 on pp. 12.
22
) We assume that n ≥ 2.
23
) See the paper cited in footnote 6) of § 2.
 Vector fields in n-dimension manifolds 20

thus define a map of the boundary of F1′ onto the direction sphere whose degree is (−1)n ⋅
c. On the grounds of the solubility 22) of a “boundary-value problem for vector
distributions” (see the paper on mapping classes cited above in 6), § 5.4), we can, since
m
one also has ∑
µ
qµ = (−1)n c, extend these boundary vectors to a continuous vector field
=1

that is defined in all of F1′ in such a way that its vectors vanish at the Qµ′ (µ = 1, …, m),
and only there, and that the singularities of the direction field at these points possess the
indices qµ . Above, all, we can choose the vectors of this field to be so small that their
endpoints all lie in the interior of F′. By the prescription that each point of F1′ shall go to
the endpoint of the vector that is attached to it, F1′ will be mapped to a subset of F′ in
such a way that this map g′ agrees with f t′ on the boundary and has fixed points at the
Qµ′ with the indices qµ , but is fixed-point free at the remaining points. The map g′
corresponds to an analogous map g in F1 . If we now replace ft with g in the interior of
F1, while we leave ft unchanged in the exterior and on the boundary of F1, then we have
constructed a map with the desired properties. We have thus proved:

Theorem IV. If Q1 , …, Qm (m ≥ 0) are arbitrary interior points if the manifold Mn

and q1, …, qm are arbitrary whole numbers whose sum is equal to the characteristic of
Mn multiplied by (−1)n then there are arbitrarily small transformations of Mn into itself
that possess fixed points at the Qµ (µ = 1, …, m) with the indices qµ , but are fixed-point-
free at the remaining ones 22).

A special case of this theorem is:

Theorem IVa. Any manifold whose characteristic is 0 admits arbitrarily small fixed-
point-free transformations into itself.

Since the characteristic is 0 for any boundaryless closed manifold of odd dimension,
one has, in particular:

Theorem IVb. Any closed, boundaryless manifold of odd dimension admits

arbitrarily small fixed-point-free transformations into itself.

We now consider vector field that are continuous in the ordinary sense: In a
neighborhood U(P) of each point P of Mn , let a Cartesian coordinate system on a set be
distinguished in such a way that the coordinates of any two coordinates systems
(belonging to the same or different points) go over to each other on a common piece by
continuously differentiable transformations; boundary manifolds of Mn shall be
continuously differentiable in these coordinate systems. In order for the examination of
the indices of such vector fields to lead directly back to the consideration of our complex-
continuous vector fields, we must possess a representation of Mn in which the boundaries
of each individual simplex Tµnn also belong to a planar space relative to one of the
distinguished coordinate systems of Mn . The existence of such a representation is, in
 Vector fields in n-dimension manifolds 21

itself, self-explanatory. We restrict ourselves, in order to avoid the difficulty thus hinted
at, to the special case in which Mn is a Riemannian manifold; i.e., at each point, a
symmetric matrix (gik) (i, k = 1, …, n) is given that depends continuously on the point
relative to any distinguished coordinate system, and whose associated quadratic form
n

∑g
i , k =1
ik dxi dxk = ds2 is positive definite, and its values do not change when one goes from

one distinguished coordinate system to another one. In any such Riemannian manifold,
each sufficiently small vector now corresponds to a displacement of the point to which it
is attached, and each sufficiently small displacement to a vector at the point in question.
With that, it follows from Theorems II and IV:

Theorem V. The sum of the indices of a vector field in a Riemannian manifold is

equal to the characteristic multiplied by (−1)n; one can always 22) construct a vector field
with prescribed singularities and indices, as long as their sum is equal to the stated
number. A singularity-free vector exists when and only when the characteristic is 0; in
particular, such a vector field can be attached to any boundaryless, closed manifold of
odd dimension.

Among the Riemannian manifolds that are thus treated are included, e.g., the ones
that are embedded in the (n + k)-dimensional Euclidian space (k ≥ 0) in a differentiable
way. Thus, the case k = 0 includes the submanifolds of space that are bounded by finitely
many continuously differentiable (n–1)-dimensional closed, boundaryless hypersurfaces.
Moreover, it includes the Clifford-Klein manifolds, as well as many others in which a
Riemannian metric may be defined. As an example, let, perhaps, the complex projective
space Zk be mentioned; i.e., the totality of all ratios z0 : … : zk of complex, not all
vanishing numbers. In it, a metric may be defined 24) with the line element:

k k

1
∑ zi ⋅ zi
i =0
∑ z ⋅ dz
i =0
i i
2
ds = 2 k k
.
 k 
 ∑ zi zi  ∑ dz ⋅ z ∑ dz ⋅ dz
i i i i
 i=0  i =0 i =0

We linger for a moment on the case of submanifolds of n-dimensional space that

bounded by closed hypersurfaces; let Mn be bounded by closed, boundaryless
hypersurface Mn−1. The vectors of a field of the type considered belonging to all such Mn
are then, everywhere on Mn−1, directed either into the interior of Mn or tangentially to
Mn−1 . The map of Mn−1 that these vectors provides has degree (−1)n ⋅ c on the direction
sphere, when c is again the characteristic of Mn. The map that is diametrically opposite to
this map, which will be mediated by a field of vectors that are nowhere directed into the

24
) In the paper cited in 4) of § 5, I have given, in a simple manner, an arbitrarily small transformation (a
vector field, resp.) in Zk with the index sum k + 1, and, in addition, in a somewhat circumstantial way,
showed that the characteristic has the value k+ 1; this determination of the characteristic is, moreover,
superfluous, in the basis of Theorem V.
 Vector fields in n-dimension manifolds 22

interior of Mn thus has the degree (−1)n ⋅ (−1)n ⋅ c = c. This degree is the “curvatura
integra” of Mn 4). With that, we have proved:

Theorem VI. The curvatura integra of a continuously differentiable Jordan

hypersurface that lies in n-dimensional space and is bounded by an n-dimensional
manifold is equal to the characteristic of the bounding manifold.

I have previously only proved this theorem for the special case that the bounding
manifold is an element. Furthermore, I obtained in the stated place: The 2k-dimensional
closed, not necessarily Jordan, continuously differentiable hypersurface m of the (2k + 1)-
dimensional Euclidian space is a “model” for the two-sided, closed, boundaryless
manifolds M2k; its curvatura integra C(m) is then a topological invariant of M2k, and the
index sum of the singularities of each vector field tangential to m is, assuming that only
finitely many singularities are present, equal to 2C(m). From this, it follows, moreover:

Theorem VII. The curvatura integra of a closed, not necessarily Jordan,

continuously differentiable hypersurface in (2k + 1)-dimensional space that is a model of
a two-sided, closed, boundaryless manifold M2k is equal to one half the characteristic of
M2k.

From this, one further deduces (cf., the many papers cited earlier), since the curvatura
integra is always a whole number:

Theorem VIII. A closed, boundaryless, two-sided manifold Mn with odd

characteristic possesses no continuously differentiable hypersurface in (n + 1)-
dimensional Euclidian space as a model, not even when one allows self-intersections.

The simplest example of such an Mn is the four-dimensional manifold that is defined

to be the ‘complex projective plane” Z2 (cf., footnote 24).
An analogue of Theorem VIII is the fact that a 2k-dimensional, closed manifold M2k
that defines the complete boundary of a closed M2k+1 always has an odd characteristic,
namely, twice the characteristic of M2k+1 25). An M2k with odd characteristic − thus, e.g.,
Z2 – can therefore never be embedded in a simply-connected, not necessarily
homeomorphic to ordinary space, closed (2k + 1)-dimensional space R2k+1 – at least, not
in the sense of combinatorial topology, i.e., such that it will be represented by a subset of
the boundary complex of a representation of R2k+1 – since they will then define the
boundary of each of the two subsets into which they must divide R2k+1 26).

(Received on 11 August 1925).

25
) This follows from the fact that the boundaryless (2k+1)-dimensional manifold that comes about
under the identification of corresponding boundary points of two exemplars of M2k+1 possesses the
characteristic 0; cf., Dyck, Beiträge zur Analysis Situs II, Math. Ann. 37 (1890).
26
) H. Kneser, Ein topologischer Zerlegungssatz, Koninkl. Akad. v. Wetenschapen te Amsterdam Proc.
27, Sept. 1924.
 Appendix

I will draw your attention to the fact that the concept of “complex-continuous vector
field,” upon the use of which the results of the paper above rest essentially, is not defined
sufficiently clearly and has given rise to misunderstanding. I thus formulate this
definition again, but more thoroughly than before:
Let An be a reduced, affine representation of the complex Cn. An association B of
vectors v(P) with the points P of A is called a Cn (relative to An) “complex-continuous
vector field” when the following conditions are fulfilled:
A. B is single-valued and continuous in the interior and on the boundary of each
individual Tµnn [µn = 1, …, βn], except for at most finitely many points that lie in the
interior.
B. Let P0 be a boundary point on Tµnn , and let T0n [1 ≤ k ≤ n] be any boundary simplex
0

– not necessarily one of lowest dimension – that P0 belongs to. Let Tρn − k [ρ = 1, …, r] be
the boundary simplexes of the other Tµnn that are to be considered to be identical with
0

T0n − k in Cn, Pρ , the points of Tρn − k that corresponding to P0 , and (Wkn ) ρ [ρ = 1, …, r],
the k-fold angle whose vertex is the planar space Eρn − k to which Tρn − k belongs.
Then one of the following two cases enters in:
I. (Main case) Of the r + 1 vectors v(Pρ), precisely one of them points into the
interior of its (Wkn ) ρ , while all others are directed to the exterior of its (Wkn ) ρ .
II. (Boundary case) One of the vectors v(Pρ) – say, v(P0) − belongs to the boundary
of its (Wkn )0 . Thus, among the vectors v* that correspond to the vector v(P0) under the
affine and transitive association of the vectors that exists between the boundary spaces,
there can be one or more that likewise belongs to B. However, one does not need for all
of these v* to belong to B – in contrast to the special case of the vector field on a
manifold that was continuous in the ordinary sense that was given at the end of § 2. All
remaining vectors v(Pρ) that are not vectors of v* point to the exteriors of their (Wkn ) ρ .

(Received on 26 May 1926)