A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

HARALD SCHMID

Abstract. This paper deals with the question of how to calculate the volume of a body in R3 when it is
cut into slices perpendicular to a given curve. The answer is provided by a formula that can be considered
as a generalized version of the second Pappus-Guldin theorem. It turns out that the computation becomes
very simple if the curve passes directly through the centroids of the perpendicular cross-sections. In this
arXiv:2411.04488v1 [math.MG] 7 Nov 2024

context, the question arises whether a curve with this centroid property exists. We investigate this problem
for a convex body K by using the volume distance and certain features of the so-called floating bodies of
K. As an example, we further determine the non-trivial centroid curves of a triaxial ellipsoid, and finally
we apply our results to derive a rather simple formula for determining the centroid of a bent rod.

1. Introduction
According to Fubini’s theorem, if K ⊂ R3 is a Lebesgue-measurable set which is in Cartesian coordinates
bounded by x = a and x = b, then for each x ∈ [a, b] the slice D(x) ⊂ R2 perpendicular to the x-axis
is Lebesgue-measurable, where its area A(x) := λ(D(x)) is a Lebesgue-integrable function on [a, b], and
Rb
the volume vol(K) = λ(K) of K can be determined by the simple formula vol(K) = a A(x) dx. In the
following, we analyze under which conditions this formula remains valid if we cut the body K into slices
perpendicular to a curved path γ. Our considerations will result in a generalized version of the (second)
Pappus-Guldin centroid theorem. This well-known result from calculus claims thatR for a set D ⊂ R×(0, ∞)
in the (x, y)-plane with Lebesgue-measure A := λ(D) > 0 and distance y = A1 D y dx dy of its centroid
from the x-axis, the volume of the solid generated by rotating D around the x-axis is V = A · 2πy.
A number of generalizations have already been given for this theorem, cf. [2], [3] or [4], and there
even exist higher dimensional versions (see e.g. [12]). As an example, we take a look at the generalized
Pappus-Guldin formula by Goodman & Goodman [2, Theorem 1 and Corollary]. It assumes a sufficiently
smooth curve γ : [0, L] −→ R3 , which is parametrized by arc-length s, and a bounded closed region D with
smooth boundary contained in the plane perpendicular to the curve at γ(0) such that γ(0) is the centroid
of D. The region D can be transported along γ, and this motion perpendicular to γ defines a solid K. If
this body is not self-intersecting,
R then the volume of K is given by vol(K) = λ(D) · L. In other words,
we get vol(K) = γ A(s) ds, where A ≡ λ(D) is a constant function in this case. A slightly more general
version is provided by [13, Theorem 1], where the region D is uniformly scaled by some factor g(s) during
transportation, so that A(s) = g(s)2 λ(D) applies. In Section 2 we show that this formula is also valid for
more general solids where the cross-sections Γ(s) perpendicular to γ are not congruent or similar, but γ(s)
is still the centroid of Γ(s). We refer to a curve with this property as a centroid curve of K. In Section
3 we focus on a convex body K and address the problem of finding a centroid curve that passes through
a given interior point p0 ∈ int(K). For this purpose, we use a result from hydrostatics, Dupin’s theorem,
which deals with the cross-sections of the so-called floating bodies of K. In Section 4 we calculate the

University of Applied Sciences Amberg-Weiden, Amberg, Germany

E-mail address: h.schmid@oth-aw.de.
2020 Mathematics Subject Classification. 51M25, 52A15, 53A15, 74K10.
Key words and phrases. Pappus-Guldin theorem, centroid curve, floating body, volume distance, elastic rod.
1
2 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

centroid curves of a triaxial ellipsoid as an example, and in Section 5 we illustrate how a centroid curve
can be used to determine the centroid of an elastic rod.

2. Volume calculation using perpendicular cross-sections

Let γ : I −→ R3 be a regular curve on some open interval I ⊂ R parametrized by arc-length s, i.e.
T (s) := γ ′ (s) is the unit tangent vector at γ(s) for all s ∈ I. Moreover, let N : I −→ R3 be a unit normal
field along γ such that N (s) ⊥ T (s) and |N (s)| = 1 for all s ∈ I. Here and in the following we assume that
T , N are at least continuously differentiable vector functions on I. The pair (γ, N ) is called a ribbon (or
strip). If we introduce the binormal field B(s) := T (s) × N (s), s ∈ I, then the orthogonal moving frame
F := (T, N, B) satisfies the differential system
 
0 −κn κg
F ′ (s) = F (s)  κn 0 −τg  ,
−κg τg 0
where κn and κg denote the normal and geodesic curvature of the ribbon, respectively, and τg is its
(geodesic) torsion. We further assume that Ω ⊂ I × R2 and W ⊂ R3 are open sets such that
Φ : Ω −→ W, (s, u, v) 7−→ γ(s) + uN (s) + vB(s), (1)
is an orientation-preserving C1 -diffeomorphism. Since
∂Φ
= γ ′ + uN ′ + vB ′ = (1 − uκn + vκg )T − vτg N + uτg B
∂s
∂Φ ∂Φ
and ∂u × ∂v = N × B = T , the Jacobian determinant of Φ is given by
 
∂Φ ∂Φ ∂Φ
det JΦ = , × = 1 − uκn + vκg > 0,
∂s ∂u ∂v
where ⟨ · , · ⟩ denotes the standard scalar product in R3 . Finally, suppose that K ⊂ W is a measurable set.
For fixed s ∈ I, let H(s) be the plane orthogonal to T (s) passing through γ(s), and we set Γ(s) := K ∩H(s).
The area of this perpendicular cross-section is A(s) := λ(Γ(s)), and it coincides with the area of the slice
D(s) := Φ−1 (Γ(s)), as Φ is just an Euclidean transformation for fixed s ∈ I. Furthermore, (u, v) are the
local coordinates of a point on H(s) relative to the axes spanned by N (s) and B(s) with origin at γ(s).
Now, according to the change-of-variable formula and Fubini’s theorem, the volume of K is given by
Z Z Z  ZZ 
vol(K) = 1 dλ = | det JΦ | dλ = 1 − uκn (s) + vκg (s) du dv ds.
K Φ−1 (K) I D(s)

Note that in the case A(s) > 0,

1 1
ZZ ZZ
u(s) := u du dv and v(s) := v du dv
A(s) D(s) A(s) D(s)

are the local coordinates of the centroid of the cross-section Γ(s) in the plane H(s). If we set u(s) = v(s) = 0
for the slices with A(s) = 0, then
Z
vol(K) = A(s) − A(s)(u(s)κn (s) − v(s)κg (s)) ds.
γ

In the case where the perpendicular cross-sections Γ(s) are congruent for all s ∈ I, this result corresponds
to the generalized Pappus-Guldin formula given by Flanders in [4]. The above considerations give rise to
the following slightly more general version of the second Pappus-Guldin theorem.
 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 3

Theorem 1. Let γ : I −→ R3 be a regular curve on some open interval I ⊂ R with unit tangent vector
T (s) = γ ′ (s), s ∈ I, and let (T (s), N (s), B(s)) be an orthogonal moving frame along γ with normal and
geodesic curvature κn (s) and κg (s), respectively. Further, suppose that Ω ⊂ I × R2 and W ⊂ R3 are open
sets such that (1) is an orientation-preserving C1 -diffeomorphism, and let K ⊂ W be a measurable set. If
A(s) denotes the area of the cross-section Γ(s) at γ(s) orthogonal to T (s) and if u(s), v(s) for A(s) > 0
are the coordinates of its centroid in the local (N (s), B(s)) system, then
!
u(s) κg (s)
Z
vol(K) = A(s) 1 − ds. (2)
γ v(s) κn (s)

As a diffeomorphism, the mapping Φ given by (1) is injective, and this prevents the perpendicular cross
sections Γ(s) from overlapping, so that in particular the body K ⊂ W = Φ(Ω) is not self-intersecting. A
necessary condition for Φ to be an orientation-preserving C1 -diffeomorphism is det JΦ = 1 − uκn + vκg > 0
or
u κg (s)
< 1 for all (s, u, v) ∈ Ω. (3)
v κn (s)
To give a first application, formula (2) provides a way of deforming a solid K without changing its volume:
If we move the slices Γ(s) perpendicular to T (s) in the direction of the vector κg (s)N (s) + κn (s)B(s),
then the determinant in (2) and hence also the volume do not change. Moreover, if the centroids of the
perpendicular cross-sections satisfy
! !
u(s) κg (s)
= µ(s) · , s ∈ I,
v(s) κn (s)
R
with some scalar function µ : I −→ R, then (2) reduces to the simple formula vol(K) = γ A(s) ds, which
we were looking for.
In the following, we consider the special case µ ≡ 0 in more detail. In this situation the centroids of the
perpendicular cross-sections are located right on the curve.

Definition 2. Let K ⊂ R3 be a measurable set. A regular curve γ : I −→ R on some open interval

I ⊂ R with |γ ′ (s)| = 1 for all s ∈ I is called centroid curve of K, if the centroid of the plane cross-section
orthogonal to γ ′ (s) passing through γ(s) coincides with γ(s) for all s ∈ I.

Within the framework of Theorem 1, this means u(s) = v(s) ≡ 0 for all s ∈ I, and we get the following
result:

Corollary 3. If γ : I −→ R3 is a centroid curve of the measurable set K ⊂ W and N : I −→ R3 is a unit

normal field along γ such that (1) is an orientation-preserving C1 -diffeomorphism, then
Z
vol(K) = A(s) ds, (4)
γ

where A(s) is the area of the perpendicular cross-section at γ(s), s ∈ I.

Provided that such a centroid curve exists, there is a second type of volume preserving deformations: If
the plane cross-sections passing through γ(s) perpendicular to γ ′ (s) are deformed in such a way that their
area values A(s) do not change and γ(s) remains their centroid, then the volume of the body is not altered
(see Fig. 1). This is especially the case when we simply rotate the slices Γ(s) perpendicular to the curve
around their centroids γ(s).
4 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

Figure 1. Two toroidal bodies with the same volume. A circle of the same size forms
a centroid curve in both solids, where the vertical cross-sections on the left-hand side are
circular disks with varying radii, while on the right-hand side they are rotating ellipses
having the same area as the circular disks.

3. On the existence of centroid curves in a convex body

We will now address the following problem: What are the centroid curves of a given solid K in R3 and
how can they be found? The more complicated the shape of the body, the more difficult it is to answer
this question. In addition, centroid curves are certainly not unique. For example, if K is simply a ball
with center m ∈ R3 and radius r > 0, then each line γ(s) = m + s · n, s ∈ R, with arbitrary unit normal
vector n ∈ S2 is a centroid curve, where in this case the perpendicular cross-sections are circular disks or
empty sets.
Because of the difficulties we might expect with a more complicated body shape, we restrict our con-
siderations to a convex body, i.e. in this section we assume that K ⊂ R3 is a compact convex set with
nonempty interior int(K), where its boundary ∂K is a smooth surface with positive Gaussian curvature
everywhere. Under these conditions on K, the centroids of the plane cross-sections passing through a given
point p0 ∈ int(K) depend continuously on the unit normal vectors n of the cutting planes, and hence there
exists at least one such plane H0 with the property that p0 is the centroid of the cross-section K ∩ H0 ; this
is a consequence of the “hairy ball theorem” (cf. [6, §2, Sec. 8]). If we assume uniform mass density, then
the centroid coincides with the barycenter, and therefore such a cross-section is also called barycentric cut.
A normal vector n0 ∈ S2 of such a plane H0 would then be the tangent vector of a centroid curve passing
through p0 , and the question remains, if we can extend this line element to a path which actually forms a
centroid curve of K.
At first we will prove that if p ∈ int(K) is sufficiently close to the boundary ∂K, then there is exactly
one such plane which yields a barycentric cut, and its unit normal vector depends smoothly on p. To this
end, let H(n, p) be the plane perpendicular to the vector n ∈ S2 passing through the point p ∈ int(K), and
we denote by V (n, p) the volume of the body which is cut off from K by H(n, p) with n pointing outwards
from this segment, i.e. V (n, p) = λ({x ∈ K : ⟨x − p, n⟩ ≤ 0}). According to [14, Proposition 2.1 and
Lemma 2.4] we get
∂V ∂V
= A(n, p)(p − c(n, p)) and = A(n, p)n,
∂n ∂p
where A(n, p) is the area of the cross-section K ∩ H(n, p), and c(n, p) denotes its centroid. Moreover, for
q ∈ ∂K let η(q) be the affine normal vector pointing to the interior of K. From [14, Proposition 2.3] it
follows that there is a half-neighborhood
Q = {q + tη(q) : q ∈ ∂K, 0 < t < τ (q)}
 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 5

of ∂K with some smooth function τ (q) such that for each p ∈ Q there exists a unique vector n = n(p) ∈ S2
that minimizes V (n, p) and depends smoothly on p. In particular, we have ∂V ∂n (n(p), p) = 0 for all p ∈ Q,
which implies p = c(n(p), p), so that p is the centroid of the cross-section K ∩ H(n(p), p). Furthermore,
v(p) := min{V (n, p) : n ∈ S2 } = V (n(p), p) is known as the volume distance from p to ∂K; it is a smooth
function satisfying Dv(p) = A(n(p), p)n(p) for all p ∈ Q (cf. [14, Lemma 2.4]). Now, for a given point
p0 ∈ int(K), the existence and uniqueness theorem yields that the differential system
γ ′ (s) = n(γ(s)), γ(0) = p0 ,
has exactly one solution γ : (a, b) −→ Q in a neighborhood of s = 0. All in all, we get the following result:
Theorem 4. Let K ⊂ R3 be a convex body with smooth boundary ∂K. There exists a half-neighborhood
Q ⊂ int(K) of ∂K such that for each p0 ∈ Q there is a smooth centroid curve γ : (a, b) −→ Q, a < 0 < b,
with γ(0) = p0 .
In order to apply the volume formula (4), we need to have a centroid curve that can be continued to the
boundary ∂K. Since the half-neighborhood Q in Theorem 4 is difficult to quantify with the method given
in the proof of [14, Proposition 2.3], we are looking for an alternate approach to determine the centroid
curve passing through p0 .
For a convex body K ⊂ R3 and some 0 < δ ≤ vol(K)/2, a convex body K[δ] ⊂ int(K) with the property
that each of its supporting planes cuts off a segment of constant volume δ from K is called floating body of K
to the parameter δ (see [11, Definition 7.1]). This term goes back to Archimedes’ principle in hydrostatics:
If K has the specific weight δ/ vol(K), then K[δ] is the part of K that never submerges below the water
surface for any position of the body. In certain cases, it may happen that such a floating body K[δ] does not
exist for any δ > 0 in contrast to the convex floating body Kδ due to Schütt & Werner, see [8]. However, if
∂K is at least twice continuously differentiable with positive Gaussian curvature everywhere and minimal
principal radius of curvature ρ > 0, then K possesses a strictly convex floating body K[δ] = Kδ for all
0 < δ < σ := 2πρ3 /3, cf. [11, Theorem 10.10], and ∂K[δ] is again of class C2 . In addition, if K is centrally-
symmetric, then K[δ] even exists for all 0 < δ < vol(K)/2 (see [9, Theorem 3]). Floating bodies have some
remarkable properties. Every supporting plane H of K[δ] with inner unit normal vector n ∈ S2 intersects
K[δ] in exactly one point, namely the centroid c[δ] (n) of K ∩ H. This result can be traced back to Dupin
(cf. [1, Ch. XXXI, Sec. 651]). Moreover, as Leichtweiß has shown in the proof of [11, Theorem 10.10],
c[δ] : S2 −→ R3 is a twice differentiable function, and the image of this map is exactly ∂K[δ] . Hence, ∂K[δ]
is formed by the centroids of all cross-sections K ∩ H, where H cuts off a section of volume δ from K, and
for this reason K[δ] is also known as Dupin’s floating body. In the following, we will use this relationship
between the floating bodies and the centroids of the cross-sections to extend a local centroid curve as far
as possible.
Theorem 5. Assume that K ⊂ R3 is a convex body with boundary ∂K of class C2 , and let ρ > 0 be the
minimal principal radius of curvature of ∂K. Moreover, let σ := 2πρ3 /3 and R := int(K) \ K[σ] . For each
p0 ∈ R there exists a unique centroid curve γ : (a, b) −→ R passing through p0 , where γ(s) ∈ ∂K[δ(s)] and
γ ′ (s) ⊥ ∂K[δ(s)] for all s ∈ (a, b) with some strictly increasing function δ = δ(s) satisfying lims→a δ(s) = 0
and lims→b δ(s) = σ.
Proof. First of all, we prove that for a given point p0 ∈ R there is a value δ0 ∈ (0, σ) and a vector n0 ∈ S2
such that p0 ∈ ∂K[δ0 ] and n0 ⊥ ∂K[δ0 ] hold. Since p0 ̸∈ K[σ] can be strongly separated from the convex
set K[σ] (cf. [15, Theorem 1.3.4]), we can find a point p ∈ ∂K[σ] and an inner unit normal vector n, such
that p0 lies in the outer half-space of the supporting plane H(n, p) of K[σ] . Let q be the (unique) point
on ∂K with inner normal vector n. If we denote by h0 and h the distances of p0 and p from H(n, q),
respectively, then 0 < h0 < h. Moreover, as V (n, q + zn) is a strictly increasing function for z ∈ [0, h]
with V (n, q) = 0 and V (n, q + hn) = V (n, p) = σ, it follows that V (n, p0 ) = V (n, q + h0 n) < σ. If we set
6 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

δ0 := min{V (u, p0 ) : u ∈ S2 }, then 0 < δ0 < σ, and there exists a minimizing n0 ∈ S2 with V (n0 , p0 ) = δ0 .
Now, from [14, Proposition 2.1] it follows that p0 = c(n0 , p0 ), so that p0 = c[δ0 ] (n0 ) ∈ ∂K[δ0 ] and n0 ⊥ ∂K[δ0 ]
applies. To prove that δ0 and n0 are uniquely determined by these properties, let us assume that there
is another unit vector n1 ̸= n0 and some δ1 ∈ (0, σ) satisfying p0 ∈ ∂K[δ1 ] and n1 ⊥ ∂K[δ1 ] . Since δ0 is
the (minimal) volume distance from p0 to ∂K and n0 is the uniquely determined inner unit normal vector
to ∂K[δ0 ] at p0 , we conclude δ1 > δ0 . Furthermore, as V (n0 , p0 + zn1 ) is strictly increasing with z, there
exists some z1 > 0 such that V (n0 , p1 ) = δ1 with p1 := p0 + z1 n1 . It follows that K[δ1 ] = Kδ1 lies in the
half-space {x ∈ R3 : ⟨x − p1 , n1 ⟩ ≥ 0}, cf. [8, p. 276]. However, for the point p0 , that also belongs to K[δ1 ]
by assumption, we get ⟨p0 − p1 , n1 ⟩ = −z1 < 0, which is a contradiction.
By applying an Euclidean transformation, we can assume that p0 = (0, 0, 0) and n0 = (0, 0, 1)T . In a
sufficiently small neighborhood U ⊂ S2 of n0 = (0, 0, 1)T , each vector n ∈ U can be represented by local
coordinates (ξ1 , ξ2 ) in a neighborhood of (0, 0). Furthermore,
1 − ξ12
     
ξ1 0
1 1 p
n= ξ2  , x1 = p −ξ1 ξ2  , x2 = p  1 − ξ12 − ξ22 
    
2 2

p 1 − ξ1 p 1 − ξ1
1 − ξ12 − ξ22 −ξ1 1 − ξ12 − ξ22 −ξ2
provides an orthonormal basis of R3 , where the vectors n and x1 , x2 depend continuously differentiable on
(ξ1 , ξ2 ). Let s = s(ξ1 , ξ2 ) be the point on ∂K[δ0 ] with normal vector n, and p = p(ξ1 , ξ2 , z) be the centroid
of the cross section K ∩ H(n, s + zn) for z ∈ (−ε, ε) with some small ε > 0, cf. Fig. 2. According to [5,
eqs. (9b), (11) and (15)],
 ∂s1
∂ξk (0, 0)

∂s1 ∂s1
!
∂s  ∂s2 ∂ξ1 (0, 0) ∂ξ2 (0, 0)
(0, 0) =  ∂ξk (0, 0) (k = 1, 2), det ∂s2 > 0. (5)

∂ξk ∂s2
∂ξ1 (0, 0) ∂ξ2 (0, 0)
0
Now, if we introduce polar coordinates on the plane

∂K[δ] ∂K[δ0 ] ∂K

n
H(
n, s
+z
n) n
p
z n0
s
p0

Figure 2. In this sectional view, p0 and s are centroids of plane cross-sections tangential
to the floating body K[δ0 ] . The hatched segments cut off from K by these cross-sections
have the same volume δ0 . Moreover, p is the centroid of the plane cross-section H(n, p)
tangential to ∂K[δ] and perpendicular to n, where H(n, p) and ∂K enclose the gray shaded
segment with volume δ.

H(n, s + zn) = {s + zn + λ1 x1 + λ2 x2 : (λ1 , λ2 ) ∈ R2 }

 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 7

with pole at s + zn, then the boundary of K ∩ H(n, s + zn) is given by r(φ) = r(φ; ξ1 , ξ2 , z) with φ ∈ [0, 2π[,
and it depends continuously differentiable on (ξ1 , ξ2 , z). Furthermore, the centroid of K ∩ H(n, s + zn) is
located at
p(ξ1 , ξ2 , z) = s(ξ1 , ξ2 ) + z · n(ξ1 , ξ2 ) + c1 (ξ1 , ξ2 , z) · x1 (ξ1 , ξ2 ) + c2 (ξ1 , ξ2 , z) · x2 (ξ1 , ξ2 ),
where
R 2π 1 3
R 2π 1 3
0 3 r(φ) cos φ dφ 0 3 r(φ) sin φ dφ
c1 (ξ1 , ξ2 , z) = R 2π and c2 (ξ1 , ξ2 , z) = R 2π .
1 1
0 2
r(φ)2 dφ 0 2
r(φ)2 dφ
Hence, p also depends continuously differentiable on ξ = (ξ1 , ξ2 , z) in a neighborhood of (0, 0, 0). Note that
c1 (ξ1 , ξ2 , 0) = c2 (ξ1 , ξ2 , 0) ≡ 0 for all n ∈ U , since s(ξ1 , ξ2 ) is the centroid of K ∩ H(n, s + 0 · n). Therefore,
we have ∂p ∂z (ξ1 , ξ2 , 0) = n(ξ1 , ξ2 ), and due to (5), the Jacobian determinant of p at (0, 0, 0) is given by
 ∂s1 ∂s2
∂ξ1 (0, 0) ∂ξ1 (0, 0) 0

∂p
(0, 0, 0) = det  ∂s
∂ξ1 (0, 0)
∂s2
∂ξ2 (0, 0) 0
> 0.
 2 
∂ξ
0 0 1
From the inverse function theorem it follows that ξ = (ξ1 , ξ2 , z) is a continuously differentiable function of
p in a neighborhood Q ⊂ R of p0 , and hence also n = n(p) depends continuously differentiable on p ∈ Q.
Moreover, p is the centroid of the cross-section K ∩ H(n, s + zn), which cuts off a segment with volume
Z z Z 2π
1 2
δ = δ(ξ1 , ξ2 , z) = δ0 + 2 r(φ; ξ1 , ξ2 , ζ) dφ dζ
0 0

from K. Therefore, p ∈ ∂K[δ] and n ⊥ ∂K[δ] , where in turn δ = δ(p) depends continuously differentiable
on p ∈ Q. Moreover, the derivative of δ at p0 in the direction of n0 is given by
Z 2π
∂δ 1 2
(0, 0, 0) = A(n0 , p0 ) := 2 r(φ; 0, 0, 0) dφ > 0,
∂z 0

which is the area of the cross-section K ∩ H(n0 , p0 ). We can apply this reasoning to other points as well,
∂δ
so that ∂n (p) = A(n, p) > 0 holds for all p ∈ Q. Now, by the existence and uniqueness theorem, the initial
value problem
γ ′ (s) = n(γ(s)), γ(0) = p0 , (6)
has a local solution on some interval [α, β] with α < 0 < β, where in addition δ(γ(s)) for s ∈ [α, β] is a
continuously differentiable function satisfying
d ∂δ
δ(γ(s)) = Dδ(γ(s)) · n(γ(s)) = (γ(s)) = A(γ ′ (s), γ(s)) > 0.
ds ∂n
If we shorten δ(γ(s)) to δ(s), then we get γ(s) ∈ ∂K[δ(s)] and γ ′ (s) ⊥ ∂K[δ(s)] , where 0 < δ(α) < δ(β) < σ.
Now we may apply the above considerations once again to the points γ(α), γ(β) ∈ R. This way the solution
of (6) can be extended to a maximum interval (a, b) for which lims→a δ(s) = 0 and lims→b δ(s) = σ holds,
which finally provides a centroid curve with the properties stated in Theorem 5. □

At last, we will examine how the centroid curve in Theorem 5 behaves if it is continued to the boundary
of K. It turns out that it approaches ∂K from the interior of K in a perpendicular direction.
Proposition 6. Let K ⊂ R3 be a convex body with boundary ∂K of class C3 and γ : (a, b) −→ R be a
centroid curve as given in Theorem 5. If q0 = lims→a γ(s) exists, then also n0 := lims→0 γ ′ (s), and we
have n0 ⊥ ∂K.
8 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

Proof. For each n ∈ S2 , let q(n) be the point on ∂K with inner normal vector n, and let η(n) be the affine
normal vector of ∂K at q(n) pointing towards the interior of K. Following the proof of [7, Satz 1], we obtain
that the centroid of the cross-section K ∩ H(n, q + zn) takes the form
p(n, z) = q(n) + z · η(n) + o(n, z),
where o(n, z)/z → 0 uniformly on S2 for z → 0 (see [7, footnote 7]). Since η is a continuous function of n
on the compact set S2 , we get |p(n, z) − q(n)| ≤ C1 z for all n ∈ S2 with some constant C1 > 0. Moreover,
if ∆δ (n) denotes the height of the segment with volume δ cut off from K starting √ at the point 2q(n) in the
direction n, then there exists another constant C2 > 0 such that ∆δ (n) ≤ C2 · δ for all n ∈ S according
to [5, Hilfssatz 2]. Now, as p = γ(s) is the centroid of the cross-section K ∩ H(n(s), q(n(s)) + z(s)n(s))
with n(s) := γ ′ (s) and z(s) := ∆δ(s) (n), we receive
p
|q(n(s)) − q0 | ≤ |γ(s) − q0 | + |γ(s) − q(n(s))| ≤ |γ(s) − q0 | + C1 C2 δ(s), s ∈ (a, b).
As q0 = lims→a γ(s) and lims→a δ(s) = 0, we obtain that lims→a q(n(s)) = q0 and, in particular, q0 ∈ ∂K.
If n0 ∈ S2 is the inner normal vector of ∂K at q0 , then the continuity of the inverse Gauss map implies
lims→a n(s) = n0 , □

4. Example: The centroid curves of an ellipsoid

For the unit ball B := {(x, y, z) ∈ R3 : x2 + y 2 + z 2 ≤ 1}, the centroid curves are just the straight lines
passing the origin O, and the floating bodies of B are balls centered at O. More precisely, if Br denotes
the ball with radius r, then the tangent plane H(n, p) at the point p ∈ ∂Br with inner unit normal vector
n = −1/r · p cuts off a spherical cap from B with the volume
δ(r) = π
3 (1 − r)2 (2 + r), r ∈ (0, 1). (7)
The sphere ∂Br can be parametrized as usual by
 
r sin θ cos φ
p =  r sin θ sin φ  , 0 ≤ θ ≤ π, 0 ≤ φ < 2π
r cos θ
and the points on ∂Br in turn are the centroids of the cross-sections B∩H(n, p) with |p| = r and n = −1/r·p.
As a more complicated example, we now determine the centroid curves of a triaxial ellipsoid. Of course,
the semi-axes of the ellipsoid are centroid lines. However, there are also non-trivial centroid curves that do
not extend along the coordinate axes.
Proposition 7. Let K be an ellipsoid with semi-axes a, b, c > 0 centered at the origin. The centroid curves
of K passing a point p0 = (x0 , y0 , z0 ) ∈ int(K) with x0 > 0 are given by
 x a2 /b2  x a2 /c2
y(x) = y0 · , z(x) = z0 · , x ∈ (0, x0 + ε) (8)
x0 x0
with some ε > 0.

Proof. The ellipsoid is the image of the unit ball B under the linear mapping T : p 7−→ diag(a, b, c) · p. This
is an affine transformation, and therefore the floating bodies K[abcδ] = T (B[δ] ) are homothetic ellipsoids
parametrized by
   
x a sin θ cos φ
p = y  = r(δ)  b sin θ sin φ  , 0 ≤ θ ≤ π, 0 ≤ φ < 2π,
z c cos θ
 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 9

Γ(s)

γ p0

x0 y

Figure 3. The centroid curve γ passing through p0 in an ellipsoid with semi-axes a = 1,

b = 0.625, c = 0.5. The centroid curve is given by y(x) = −0.3 (x/0.8)2.56 , z(x) =
0.18 (x/0.8)4 for x ∈ [0, 0.8], and it has been continued symmetrically to the origin. The
gray-shaded ellipse is the cross-section Γ(s) perpendicular to the centroid curve at p0 .

where r(δ) is the inverse function to (7). Because of

 bc 
a x
   
bc sin θ cos φ bc sin θ cos φ
∂p ∂p
× = r(δ)2 sin θ ac sin θ sin φ , ac
we get that n := − ac sin θ sin φ = −  b y 
 
∂θ ∂φ
ab cos θ ab cos θ ab
z
c

is an inner normal vector to ∂K[abcδ] at p = (x, y, z). Hence, due to (6), a centroid curve γ : (a, b) −→ K
passing a given point p0 = (x0 , y0 , z0 ) ∈ int(K) satisfies the differential equation
 bc
a x(s)
 ′      
x (s) x(0) x0
y ′ (s) = γ ′ (s) = −1
 ac y(s) , y(0) =  y0  .
 
b
|n(γ(s))|

z ′ (s) ab
z(s) z(0) z0
c

It follows that
y ′ (s) a2 x′ (s) z ′ (s) a2 x′ (s)
= 2 and = 2 ,
y(s) b x(s) z(s) c x(s)
which implies y(s) = C1 x(s)α1 , z(s) = C1 x(s)α2 with some constants C1 , C2 ∈ R and α1 := a2 /b2 ,
α2 := a2 /c2 . As this curve passes through p0 , we obtain C1 = y0 · x0−α1 and C2 = z0 · x−α
0
2
, which yields
(8). Fig. 3 gives an example of such a centroid curve. □

5. Application: The centroid of an elastic rod

With the assumptions on K from Theorem 1 and assuming that γ is a centroid curve of K, we now
investigate how the centroid c(K) of the body K is related to the centroid c(γ) of the curve γ. Here we
consider the special case where the perpendicular cross-sections Γ(s) are axisymmetric, either to the axis
through γ(s) in the direction of N (s) or of B(s) for all s ∈ I. The points x ∈ K are given by x = Φ(s, u, v)
10 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

for (s, u, v) ∈ Φ−1 (K), and thus the change-of-variable theorem implies
1 1
Z Z
c(K) = x dλ = | det JΦ | · Φ dλ
vol(K) K vol(K) Φ−1 (K)
 
1
Z ZZ
= (1 − uκn (s) + vκg (s))(γ(s) + uN (s) + vB(s)) du dv ds.
vol(K) I D(s)
RR RR
As γ(s) is the centroid of the slice Γ(s), we have D(s) u du dv = D(s) v du dv = 0. Hence,
ZZ
(1 − uκn (s) + vκg (s))(γ(s) + uN (s) + vB(s)) du dv
D(s)
= A(s)γ(s) − Iv (s)κn (s)N (s) + Iu (s)κg (s)B(s) + Iuv (s)(κg (s)N (s) − κn (s)B(s)),
where ZZ ZZ ZZ
Iu (s) := v 2 du dv, Iv (s) := u2 du dv, Iuv (s) := uv du dv
D(s) D(s) D(s)
are the so-called second moments of area for the cross-section Γ(s) with respect the centroidal axes. Since
D(s) is symmetric to the u-axis or v-axis according to our additional assumption on Γ(s), we obtain that
the product moment Iuv (s) vanishes, and therefore
1
Z
c(K) = A(s)γ(s) − Iv (s)κn (s)N (s) + Iu (s)κg (s)B(s) ds. (9)
vol(K) γ
The calculation of c(K) can be considerably simplified if K is generated by a “natural motion” of a
measurable set D0 ⊂ R2 along a geodesic ribbon (γ, N ). More precisely, we make the following assumptions:
(a) γ : [0, L] −→ R3 is a regular curve parametrized by arc-length s, and (γ, N ) is a ribbon with κg ≡ 0
on [0, L].
(b) D0 ⊂ R2 is a Lebesgue-measurable set in the (u, v)-plane with area A = λ(D0 ) > 0 and second area
moment Iv . Moreover, D0 is symmetric either to the u-axis or to the v-axis, and its centroid is located
at the origin (0, 0).
(c) D0 ⊂ G := {(u, v) ∈ R2 : |uµ| < 1}, where µ := max{|κn (s)| : s ∈ [0, L]}, and Φ : (0, L) × G −→ R3
with Φ(s, u, v) := γ(s) + uN (s) + vB(s) is an injective mapping.
In the following we will determine the centroid of the body
K := {γ(s) + uN (s) + vB(s) : s ∈ (0, L), (u, v) × D0 }. (10)
Let us briefly discuss what type of body is described by (10). If D0 is a measurable set in the (u, v)-plane
with centroid at the origin, then [0, L] × D0 corresponds to a prism-shaped body generated by a motion of
D0 perpendicular to the (u, v)-plane from s = 0 to s = L. Such a solid can be considered as an elastic rod
with profile D0 and length L which extends along the s-axis in its initial (undistorted) state, see Fig. 4. If
we assume that this rod is bent in such a way that [0, L] is mapped to the curve γ and the position of the
cross-sections Γ(s) relative to the geodesic ribbon (γ, N ) with κg ≡ 0 do not change, then (10) describes a
bent rod. Its perpendicular cross-sections Γ(s) are congruent to D0 , and their centroids are located on γ.
This means, that the so-called “axial curve” of the bent rod K coincides with the centroid curve γ, cf. [10]
or [16, Section 2]. Finally, assumption (c) prevents the rod from being bent too much and, in particular,
from intersecting itself.
Theorem 8. If the conditions (a) – (c) are satisfied, then the volume of the body (10) is given by vol(K) =
AL, and its centroid is located at
Iv
c(K) = c(γ) + · (γ ′ (0) − γ ′ (L)), (11)
AL
where c(γ) is the centroid of the curve γ.
 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 11

u s
L

D0

N (s)
B(s)

Γ(s)

Figure 4. An elastic rod, above in the initial state and below in bent form with centroid
curve γ and unit normal field N .

Proof. The map Φ defined by (1) with I = (0, L), Ω = (0, L) × G and W = Φ(Ω) is bijective and
continuously differentiable. Since κg ≡ 0 and uκn (s) < 1 for all s ∈ I and (u, v) ∈ G, it follows that
det JΦ = 1 − uκn > 0 on I × G, and hence Φ is an orientation-preserving C1 -diffeomorphism, where
K = Φ(I × D0 ) ⊂ W . Corollary 3 with constant A(s) ≡ A implies vol(K) = A · L, and (9) with constant
second area moment Iv (s) ≡ Iv yields
1 1 Iv
Z Z Z
c(K) = A γ(s) − Iv κn (s)N (s) ds = γ(s) ds − κn (s)N (s) ds.
AL γ L γ AL γ
Here, c(γ) = L1 γ γ(s) ds is the centroid of the curve γ, and κn (s)N (s) = T ′ (s) implies γ κn (s)N (s) ds =
R R

T (L) − T (0), which completes the proof. □

Remark 9. If D0 ⊂ R2 is bounded, then Theorem 8 remains valid if we replace (10) by

K = {γ(s) + uN (s) + vB(s) : (s, u, v) ∈ [0, L] × D0 }
since the planar frontal surfaces {0} × Γ(0) and {L} × Γ(L) have measure zero. If, in addition, γ is a
closed curve, then γ ′ (0) = γ ′ (L), and (11) reduces to c(K) = c(γ). Finally, if 0 < κn (s) ≤ µ holds for all
s ∈ [0, L], then we can substitute the condition on D0 in (c) by D0 ⊂ G := {(u, v) ∈ R2 : uµ < 1}, because
in this case det JΦ = 1 − uκn (s) > 0 still remains valid for all (s, u, v) ∈ I × G.
As an example, we consider the segment of a body of revolution K generated by a bounded, measurable
set B0 ⊂ (0, c) × R in the (x, z)-plane with area A = λ(B0 ) > 0, which is rotated around the z-axis by
an angle α (in radians) relative to the positive x-axis (cf. [18, Figure 1]). Here, we additionally assume
that the centroid of B0 is located on the x-axis with distance r ∈ (0, c) from the z-axis, and that B0 is
axisymmetric either to the line z = 0 or to the axis x = r. The centroid curve of this solid, parametrized
12 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

by arc-length, is the (planar) circular arc

   
cos(s/r) sin α
r
γ(s) = r  sin(s/r)  , s ∈ [0, rα] with centroid c(γ) = 1 − cos α and length L = rα.
α
0 0
In conjunction with the principal normal vector N (s) = γ ′′ (s), we obtain a geodesic ribbon (γ, N ) with
κg = τg ≡ 0 and κn (s) = 1/r for all s ∈ [0, rα], i.e. µ = 1/r. In local coordinates u = r − x and v = z,
the centroid of the generating axisymmetric figure D0 := {(u, v) : (r − u, v) ∈ B0 } with area measure A
is located at the origin (0, 0), where D0 ⊂ G := (−∞, r) × R = {(u, v) ∈ R2 : uµ < 1}, and the second
moment of area Iv of D0 coincides the second area moment Iz of B0 with respect to its centroidal axis in
z-direction. Now, from (11) it follows that
       
sin α sin α 2 sin α sin α
r Iz Ar + Iz  Iz
c(K) = 1 − cos α + · 1 − cos α = · 1 − cos α = · 1 − cos α ,
α Arα Arα Arα
0 0 0 0
where Iz := Ar2 + Iz is the second moment of area of B0 with respect to the axis of rotation according to
the parallel axis theorem (also known as Steiner’s theorem). If we take into account Irz = 0 due to the
symmetry of B0 , then this result agrees with the formula for the centroid coordinates given in [18, eqs.
(8), (9) and (12)].

6. Conclusion
Once we have found aRcentroid curve for a solid K, it is quite simple to calculate its volume by means
of the formula vol(K) = γ A(s) ds. However, the example with the ellipsoid shows that the calculation of
a centroid curve for a certain body passing through a specific point is not that easy. On the other hand,
there are also situations in which a centroid curve of a body is already known, for example when it emerges
from another body with a straight centroid line by means of a deformation, as in the case of the bent rod.
A centroid curve can also be used to determine other geometric quantities like the barycenter, and it is
helpful in examining how these quantities behave when the body is further deformed.
In the present paper, we have shown that in a convex body there is always a (uniquely determined) local
segment of a centroid curve passing through a given interior point provided that this point is sufficiently
close to the boundary. Nevertheless, there are still some open issues, e.g. whether such a local centroid
curve can be continued to a path that passes through the entire body and extends from boundary to
boundary, especially in the case of a non-convex solid. One of the problems that arise when examining the
existence of a global centroid curve is already encountered with convex bodies: For certain inner points p
of K there are different planes H passing through p so that p is the centroid of K ∩ H, cf. [17, Theorem
1.10]. Thus, as with the center of a sphere or an ellipsoid, there are points in a convex body that will be
passed by more than one centroid curves.

Appendix: Calculation of the surface area

The Pappus-Guldin formula for calculating the volume of a body of revolution is often referred to as
the second centroid theorem. In fact, there is yet another formula due to Pappus and Guldin, the so-called
first centroid theorem, that deals with the surface of a body of revolution. This begs the question: Is there
a formula like (2) also for the surface area of K? This question has already been raised in [2] for solids with
congruent cross-sections, and in general it must be negated. However, we can provide a lower estimate for
area(∂K) in terms of the perimeters and centroids of the cross-sections ∂Γ(s) perpendicular to γ. For this
purpose, we assume that the boundary ∂K of K is given by
p(s, t) = γ(s) + u(s, t) N (s) + v(s, t) B(s), (s, t) ∈ [a, b] × [0, c]
 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM 13

with some continuously differentiable functions u = u(s, t) and v = v(s, t). We denote by ps and pt the
partial derivatives with respect to s and t. From
ps = T + uN ′ + vB ′ + us N + vs B = (1 − uκn + vκg )T + (us − τg v)N + (vs + τg u)B
and pt = ut N + vt B it follows that
ps × pt = (1 − uκn + vκg )ut (T × N ) + (1 − uκn + vκg )vt (T × B)
+ (us − τg v)vt (N × B) + (vs + τg u)ut (B × N )
= (1 − uκn + vκg )ut B − (1 − uκn + vκg )vt N + ((us − τg v)vt − (vs + τg u)ut )T,
where we have used T × N = B, B × T = N and N × B = T . Since T , N , B are orthonormal vectors, we
obtain
|ps × pt |2 = (1 − uκn + vκg )2 (u2t + vt2 ) + ((us − τg v)vt − (vs + τg u)ut )2 .
If we assume (3) once again, then 1 − uκn + vκg > 0 implies
Z bZ c Z bZ c q
area(∂K) = |ps × pt | dt ds ≥ (1 − uκn + vκg ) u2t + vt2 dt ds.
a 0 a 0
For fixed s ∈ [a, b], the length of the boundary curve ∂Γ(s) of the perpendicular cross-section Γ(s) is given
by Z cq
L(s) := u2t + vt2 dt,
0
and in the case L(s) ̸= 0,
c c
1 1
Z q Z q
u(s) := u u2t + vt2 dt, v(s) := v u2t + vt2 dt
L(s) 0 L(s) 0

are the coordinates of its centroid with respect to the local (N (s), B(s)) system at γ(s). Therefore we have
!
u(s) κg (s)
Z
area(∂K) ≥ L(s) 1 − ds. (12)
γ v(s) κn (s)
In (12) equality applies if and only if (us − τg v)vt − (vs + τg u)ut ≡ 0. This is the case, for example, if the
perpendicular cross-sections Γ(s) along a ribbon (γ, N ) with some planar R curve γ do not change, as then
τg ≡ 0 and us = vs ≡ 0. Another special case in which area(∂K) = γ L(s) ds holds can be found in [3,
Theorem 2].

Acknowledgment
The author would like to thank Heinrich Kammerdiener from the University of Applied Sciences Amberg-
Weiden for some fruitful discussions on the theory of bent rods.

References
[1] P. Appell, Traité de mécanique rationnelle, Tome 3: Équilibre et mouvement des milieux continus,
Gauthier-Villars, 1952.
[2] A. W. Goodman, G. Goodman, Generalizations of the Theorems of Pappus, The American Mathe-
matical Monthly 76 (4) (1969), 355–366. doi:10.1080/00029890.1969.12000217.
[3] L. E. Pursell, More Generalizations of a Theorem of Pappus, The American Mathematical Monthly
77 (9), 961–965. doi:10.2307/2318111.
[4] H. Flanders, A Further Comment on Pappus, The American Mathematical Monthly 77 (9) (1970),
965–968. doi:10.1080/00029890.1970.11992639.
14 A GENERALIZATION OF THE SECOND PAPPUS-GULDIN THEOREM

[5] K. Leichtweiß, Über eine Formel Blaschkes zur Affinoberfläche, Studia Scientiarum Mathematicarum
Hungarica 21 (1986), 453–474.
[6] T. Bonnesen, W. Fenchel, Theory of Convex Bodies, BCS Associates, Moscow, Idaho USA, 1987.
[7] K. Leichtweiß, Über eine geometrische Deutung des Affinnormalenvektors einseitig gekrümmter Hy-
perflächen, Archiv der Mathematik 53 (1989), 613–621.
[8] C. Schütt, E. Werner, The convex floating body, Mathematica Scandinavica 66 (1990), 275–290.
doi:10.7146/math.scand.a-12311.
[9] M. Meyer, S. Reisner, A geometric property of the boundary of symmetric convex bodies and convexity
of flotation surfaces, Geometriae Dedicata 37 (1991), 327–337. doi:10.1007/BF00181409.
[10] E.H. Dill, Kirchhoff’s Theory of Rods. Archive for History of Exact Sciences 44 (1992), 1–23. doi:
10.1007/BF00379680.
[11] K. Leichtweiß, Affine Geometry of Convex Bodies, Johann Ambrosius Barth Verlag, 1998.
[12] M. C. Domingo-Juan, V. Miquel, Pappus type theorems for motions along a submanifold, Differential
Geometry and its Applications 21 (2) (2004), 229–251. doi:10.1016/j.difgeo.2004.05.005.
[13] X. Gual-Arnau, V. Miquel, Pappus-Guldin theorems for weighted motions, Bulletin of the Belgian
Mathematical Society - Simon Stevin 13 (1) (2006), 123–137. doi:10.36045/bbms/1148059338.
[14] M. Craizer, R. C. Teixeira, Volume distance to hypersurfaces: Asymptotic behavior of its hessian,
Differential Geometry and its Applications 31 (4) (2013), 510–516. doi:10.1016/j.difgeo.2013.
05.002.
[15] R. Schneider, Convex Bodies: The Brunn–Minkowski Theory, 2nd ed., Cambridge University Press,
2013.
doi:10.1017/CBO9781139003858.
[16] M. Lembo, On nonlinear deformations of nonlocal elastic rods, International Journal of Solids and
Structures 90 (2016), 215–227. doi:10.1016/j.ijsolstr.2016.02.034.
[17] Z. Patáková, M. Tancer, U. Wagner, Barycentric Cuts Through a Convex Body, Discrete & Compu-
tational Geometry 68 (2022), 1133–1154. doi:10.1007/s00454-021-00364-7.
[18] T. J. Cloete, Extensions to the theorems of Pappus to determine the centroids of solids and surfaces
of revolution, International Journal of Mechanical Engineering Education 51 (4) (2023), 227–332.