
2
κ(s, t) as a function of its arc-length s. Second, we have
introduced the equations for rods in the limit of small
oscillations, which can seem a rather restrictive assump-
tion. The purpose is merely to avoid unessential compu-
tational difficulties in the presentation. In fact, we did
take these nonlinearities into account in the numerical
simulations presented below. Third, the small amplitude
oscillations of a rod are classically studied in terms of
the transverse displacement y(s, t). Here, the important
variable which is connected to the failure of the rod in
flexion is the curvature κ(s, t), which was therefore cho-
sen as the unknown in Eq. (1).
On Eq. (1), we impose clamping conditions at s = L:
κ
,s
2
(L, t) = 0, κ
,s
3
(L, t) = 0, and free boundary condi-
tions at s = 0: κ(0, t) = 0, κ
,s
(0, t) = 0. These four
boundary conditions in s associated with the two initial
conditions κ(s, 0) = κ
0
and κ
,t
(s, 0) = 0 (uniform curva-
ture κ
0
, no initial velocity) warrant, in principle, a unique
solution κ(s, t) to Eq. (1).
A key remark must be made here, which is at the heart
of the rich dynamics of the released rod. These initial
and boundary conditions are inconsistent: the curvature
κ(0, t) at the free end has to be κ
0
6= 0 at initial time
t = 0, while the free end condition requires that it van-
ishes at any time t > 0. This inconsistency can be under-
stood easily: the initial configuration with uniform cur-
vature κ
0
violates the constitutive relation of the rod (the
curvature is proportional to the internal moment, even in
the dynamic theory of rods) and must therefore vanish
near a free end. This is a typical boundary layer situ-
ation: the assumptions underlying the derivation of the
Kirchhoff equations break down in some domain where
the solution is sought (here, at small times and in the
vicinity of the free end) — an example of a boundary
layer arising for similar reasons in a static problem is
the eversion of an elastic ball [14]. A detailed analysis
of this boundary layer will be presented in a separated
paper [15]. Here, it is sufficient to remark that in order
to solve this boundary layer, one has to restore in one
way or another the small thickness r of the rod into the
equations — for instance by taking into account the finite
time needed for the initial crack to propagate through a
cross-section of the rod, leading to a decrease of M
0
over
a small but finite timescale T
s
= r/c ∼ 1 µs for spaghetti,
where c is the typical speed of propagation of the trans-
verse dynamic crack. Being based on the length-scale r,
this boundary layer characterizes the rod dynamics over
typical times T
s
and in a region of size ∼ r around the
free end. The ratio of this timescale to the ‘macroscopic’
timescale introduced before reads T/T
s
= 2 (L/r)
2
, that
is the square of the (large) aspect ratio of the rod. Since
L/r ∼ 250 for spaghetti pasta, there are three to four
orders of magnitude between T
s
and T . As long as one is
not interested in describing the dynamics of the rod over
timescales as short as T
s
, one can disregard the details
of this boundary layer.
The inconsistency in the boundary and initial condi-
tions of Eq. (1) is therefore solved by noticing that the
initial curvature κ(L, t) relaxes from its initial value, κ
0
,
to zero over the short timescale T
s
≪ T . This simple
remark has two crucial consequences. First the relax-
ation of κ(L, t), being very abrupt, generates a burst of
flexural waves which are strong enough to break the rod,
as we explain below. Second, the separation of scales
T
s
≪ T can be utilized to derive an analytic solution
to our problem in the so-called intermediate asymptotic
regime
T
s
t T (2)
which we study here. Solutions of the regularized prob-
lem are indeed described in the limit t T
s
by self-
similar solutions [16]. Owing to the obvious scaling
s ∼ L
p
t/T, we seek a solution of Eq. (1) in the form
κ(s, t) = κ
0
u(ξ), where the self-similarity variable is
ξ = (s/L)/
p
t/T = s/
p
(γ t). Note that we have fac-
tored out the initial curvature κ
0
, as we discuss the small
amplitude limit of the Kirchhoff equations [17]. The
boundary conditions for u(ξ) are derived from those for
κ: u(0) = 0, u
0
(0) = 0 and u(+∞) → 1. Substitut-
ing this self-similar form of κ(s, t) into Eq. (1) yields the
following equation for the self-similar solution u(ξ):
4 u
0000
(ξ) + ξ
2
u
00
(ξ) + 3 ξ u
0
(ξ) = 0 (3)
Imposing that u(ξ) tends toward a constant for ξ → +∞
implies that u
00
(0) = 0, as shown with the help of an
integral of motion. This last condition, in addition to
the previous ones, yields a unique self-similar solution to
Eq. (3):
κ(s, t) = 2κ
0
S
1
√
2π
s
√
γ t
, (4)
where we have introduced the Fresnel sine integral,
S(x) =
R
x
0
sin(
π
2
y
2
)dy, arising in diffraction theory.
Equation (4) does not describe a progressive wave with
constant velocity, s ∼ c t, but instead a self-similar so-
lution s ∼
√
γ t. This reflects the dispersive nature of
Eq. (1).
Bent rods that are suddenly released at one end are
all described in the intermediate regime (2) by the same
universal solution (4) independently of the material prop-
erties, of the details of the initial release or breaking (as
long as they take place over a short timescale T
s
≪ T )
and even of the boundary conditions imposed at the other
end s = L, which have not been used to derive Eq. (4).
This universal solution is plotted in Fig. 2, bottom, along
with a numerical solution of the Kirchhoff equations —
including nonlinearities omitted in Eq. (1). This numeri-
cal solution features, as expected, the self-similar regime
for T
s
t T in which a burst of flexural waves emit-
ted from the released end s = 0 travels along the rod
with a square root time dependence. The self-similar
solution (4) accurately describes the rod dynamics un-
til reflections on the clamped end s = L take place, for