J. math. fluid mech.

11 (2009) 459–463
1422-6928/09/030459-5 Journal of Mathematical
c 2008 Birkhäuser Verlag, Basel Fluid Mechanics
DOI 10.1007/s00021-008-0269-y

Floating Bodies in Neutral Equilibrium

Robert Finn and Mattie Sloss

Abstract. In his paper preceding in this issue, Finn proved that if the contact angle γ of a
convex body B with a given liquid is π/2, and if B can be made to float in “neutral equilibrium”
in the liquid in any orientation, then B is a metric ball. The present work extends that result,
with an independent proof, to any contact angle in the range 0 < γ < π. Our result is equivalent
to the general geometric theorem that if for every orientation of a plane, it can be translated to
meet a given strictly convex body B in a fixed angle γ within the above range, then B is a metric
ball.

Mathematics Subject Classification (2000). 76B45, 52A15.

Keywords. Capillarity, contact angle, floating criteria, convex bodies, neutral equilibrium.

The concept of “neutral equilibrium” for floating bodies of general shape, subject
to capillary surface forces in an unbounded fluid, was introduced in [1] by analogy
with the two-dimensional case studied in [2], for which the exterior fluid surface
is necessarily rectilinear and horizontal. By extension, [1] defines the neutral
equilibrium fluid surface interfaces in the (physical) three dimensional case to be
those that are horizontal planes. Such configurations are a relatively small subset
of those that actually occur, but are not an insignificant set and are encountered
in various situations.
For the particular contact angle γ = π/2, [1] contains a proof that if a smooth
strictly convex closed body B can be made to float in neutral mechanical equilib-
rium in any orientation, then B is a metric ball. We present here an independent
proof of that result, which applies for any contact angle in the range 0 < γ < π.
We must of course point out that the original proof in [1] proceeds through an
intermediate step yielding rotational symmetry under more limited hypotheses.
That is a matter of independent interest that we will not pursue here. We note
that it is not an accident that the end points in the range for γ are excluded; the
reader can verify easily that the result fails in those cases.
For background material and orientation we refer the reader to [1]. We as-
sume that B ∈ C (2) in local coordinates and that B is convex, in the strict sense
that the contact set of each support plane with B is a single point, see, e.g., [3]
460 R. Finn and M. Sloss JMFM

Fig. 1. Floating body B; triple interface C; contact angle γ.

Fig. 2. Gaussian image Z of C on unit sphere, at latitudinal height cos γ.

Definitions 7.5, 7.6. We do not require non-vanishing of the Hessian determinant.

Corresponding to prescribed contact angle γ, 0 < γ < π, we assert:

Theorem. If in every spatial orientation of B there is a height h of its centroid

relative to the fluid surface level, at which B will float in neutral equilibrium with
contact angle γ, then B is a metric ball.

To prove the assertion, we examine B in one of the considered neutral equi-

librium configurations, partially immersed in a horizontal bath of liquid whose
surface it meets in a (convex) closed curve C (triple interface) at the constant
angle γ, see Figure 1. C lies on the horizontal plane Π of the liquid surface.
Since the oriented support planes are equivalent to the Gauss image, we find
that the Gauss map of B to the unit sphere is biunique. We adopt a
coordinate
system (x, y, z) with (x, y) in the plane of the liquid, and denote by n = nx , ny , nz
the unit exterior normal to B on C. We then have nz = cos γ; and we see that
the Gauss map of C is biunique to the latitude circle Z of height cos γ on the
unit sphere centered at the origin, see Figure 2. Since all of Z is covered in this
mapping, no points on B distinct from the level curve C map onto Z.
A theorem of Joachimsthal ([4], see also [5] p. 276 for the case encountered
here) states that if two surfaces intersect with constant angle in a curve C, and
if C is a curvature line on one of the surfaces, then it is a curvature line on both
of them. In the present case, we choose one of the surfaces to be the horizontal
Vol. 11 (2009) Floating Bodies in Neutral Equilibrium 461

plane Π. Since every curve on a plane is a curvature line, and the angle γ is
constant, we conclude that C is a curvature line on B.
Let P be a point of C, nP the normal to B at P . Rotate B about nP through
an angle τ that is not an integer multiple of π/2. Following the rotation, B will
again meet Π at P in the same angle γ. Since, as we have shown, that angle can
occur at only one level, and we know by hypothesis that for any orientation of B
there exists at least one level curve meeting Π at angle γ, we conclude that the
new level curve Cτ through P meets Π on its entire traverse in angle γ and is again
a curvature line on B. But Cτ meets C at P in an angle differing from π/2; since
curvature lines at a point are orthogonal except at umbilics, we find that P is an
umbilical point of B. Since P is arbitrary on C, all points of C are umbilic. That is
again the case for all the level curves created by varying τ . Letting τ vary from 0
to 2π we sweep out all of B, and we thus conclude that B is totally umbilic. Hence
B is a metric ball, cf. [6], p. 147. We are done.

Discussion

It may at first seem surprising that a hypothesis of strict convexity is needed for
a theorem that one might initially expect to hold in great generality. One sees
however from the example of the annular domain between two concentric spheres,
with contact angle π/2 and the inner ball half filled with fluid, that significant hy-
potheses beyond local regularity must be introduced. The example is of course an
isolated case, dependent heavily on symmetry, and (perhaps most significantly) in-
volving a domain whose boundary has multiple components. A deeper explanation
may perhaps lie in the circumstance that for non-convex domains in general, the
hypothesis of neutral equilibrium in every orientation cannot in any configuration
other than the one just cited be satisfied, so that the set of relevant configurations
is essentially empty. This is an assertion of independent interest for which we do
not have a formal proof, however we offer a heuristic justification based on the two
dimensional case for which B is a planar set bounded by a single smooth closed
curve Γ, for which the points of vanishing curvature are isolated. We assert that
if there is a single point of vanishing curvature on Γ, then there cannot exist a
family S of line segments with endpoints on Γ, meeting Γ in the angle π/2 at both
ends, and whose endpoints map Γ in a smooth biunique way onto itself. Since
every such Γ must be in part convex relative to neighboring points of B, we obtain
as consequence that convexity in the strict sense of non-vanishing curvature over
the entire curve is the only possibility for which the geometric requirement can be
fulfilled.
The supposed family S would provide a mapping y(x(s)) from Γ → Γ, along
which (y − x)ẏ = (y − x)ẋ = 0. Thus, (|y − x|2 )· = 0, and we see that the length
d of the segments is constant over Γ.
Consider a point p ∈ Γ at which the curvature vector k 6= 0. The normals to
462 R. Finn and M. Sloss JMFM

Fig. 3. The mapping y(x(s)), indicating convergence to infinity of focal points at an in-
flection. If the mapping is non-singular, the two curves cannot join.

Γ have a focal point (center of curvature) on the normal through p, at a distance

rp = 1/|k| from p, in the curvature direction along that normal. The same focal
point appears for the image q of p. If we consider rp to be positive or negative,
according to whether k is directed into or out of B, then rp + rq = d.
If rp > 0 at every p ∈ Γ, then B is (strictly) convex and a curve of constant
width, see [1]. Otherwise there would be a non-null subarc Γ+ ∈ Γ, at which k is
directed into B and at the end points of which rp becomes +∞. (It is not a priori
excluded that both endpoints coincide.) Since d is constant, one must have rq
becomes −∞. That tells us first that the image of a point of vanishing curvature
again has that property, so that the endpoints of Γ+ do not coincide, and then
that the image of Γ+ near q is curving in the opposite sense with respect to B, see
Figure 3. This tells us also that sufficiently close to these points no focal points
appear between the two curves.
Were a focal point to appear on the image of Γ+ , there would have to be a
first such point p+ , starting from an end point of Γ+ , at which that happens. We
may suppose that p+ corresponds to the parameter value s = 0. At a focal point,
there holds dy/ds = 0, thus we find y ≈ 21 k0 s2 near p+ , with k0 = d2 y/ds2 6= 0
since k is non-vanishing interior to Γ+ . This means that the image has a cusp at
p+ , contradicting the assumed smoothness of Γ. Thus no focal points occur in the
closed region swept out by the normals to Γ+ of length d. The image appears as a
curve roughly similar to Γ+ , obtained by shifting that curve a fixed distance along
its normals, leading to a non-singular shape distortion, as indicated in Figure 3.
The end points of Γ+ are points of vanishing curvature, which by hypothesis
are isolated. As these points are crossed, we will by the same reasoning again find
Vol. 11 (2009) Floating Bodies in Neutral Equilibrium 463

that the senses of curvature with respect to B of Γ and of its image are opposite,
and we can continue the reasoning into a new maximal subset of non-vanishing
curvature. After a finite number n of such steps, the end point of the nth step must
join with the initial point of the first step. We obtain an annular channel of fixed
width measured along segments that are simultaneously normal to both sides of
the channel, and indeed domains of this character of quite general shape are easily
constructed. The domain that we obtain contradicts however our hypothesis of B
being bounded by a single component, thus establishing our assertion.
Thus it would seem that under modest assumptions, the hypotheses needed
for the “metric ball” theorem cannot be fulfilled in the case of non-convexity,
in the sense that the set of domains to be examined outside the set of convex
domains degenerates to the null set. We emphasize of course that our reasoning as
presented is for a lower dimensional configuration; the extent to which it applies
to the physical case of our theorem is uncertain.
We note that in the example discussed, strict convexity is imposed by the
underlying hypothesis, but even so, the hypothesis falls short of yielding circular
domains, permitting arbitrary curves of constant width. Thus in this respect a
stronger result is obtained in the physical case we consider here.

Acknowledgments. The former author thanks the Max-Planck-Institut für Math-

ematik in den Naturwissenschaften, in Leipzig, for its hospitality during comple-
tion of the work. The latter author was supported by a VPUE undergraduate
education grant at Stanford University, administered by the Mathematics Depart-
ment.

References
[1] R. Finn, Floating Bodies Subject to Capillary Attractions, J. Math. Fluid Mech., DOI
10.1007/s00021-008-0268-z.
[2] E. Raphaël, J.-M. di Meglio, M. Berger and E. Calabi, Convex Particles at Interfaces,
J. Phys. I France 2 (1992), 571–579.
[3] F. A. Valentine, Convex Sets, McGraw-Hill, Inc., New York, 1964.
[4] F. Joachimsthal, Demonstrationes theorematum ad superficies curvas spectantium, J.
reine angew. Math. 30 (1846), 347–350.
[5] E. Kreyszig, Differential Geometry, University of Toronto Press, 1959.
[6] M. P. do Carmo, Differential Geometry of Curves and Surfaces, Prentice-Hall, Inc. En-
glewood Cliffs, New Jersey, 1976.

R. Finn and M. Sloss

Mathematics Department
Stanford University
Stanford, CA 94305-2125
USA
e-mail: [email protected]
[email protected]

(accepted: March 12, 2008; published Online First: July 7, 2008)