The inerter: a retrospective

by
Malcolm C. Smith
Department of Engineering
University of Cambridge
Cambridge CB2 1PZ, U.K.
Abstract
The paper provides an introduction and overview of the inerter concept and device.
Careful attention is given to the distinction between the inerter as an ideal modelling
element and devices that approximate the ideal behaviour. The background is given
to the formal definition of the inerter as a mechanical one-port with terminal forces
proportional to the relative acceleration between them. Four major methods of
construction are described and modelled. The discussion focuses particularly on:
the notion of terminals; the distinction between a device and an effect; sign reversals;
back-driving in geared systems; the conceptual aspects of the modelling step for
inerter embodiments; the problem of reverse engineering to discover a purpose. The
paper includes an analysis and discussion of the rotational inerter. A brief review of
the ideas of passive network synthesis that led to the inerter concept are provided.
A discussion and analysis is given on several examples of integrated mechanical
devices. The article concludes with an imaginary dialogue between the author and
an interlocutor on the understanding and purpose of the inerter.

1 Introduction
The “inerter” is a mechanical device with two terminals such that the equal and opposite
force on the terminals is proportional to the relative acceleration between them [1, 2].
Since its introduction in 2001 the inerter has become established commercially in suspen-
sion systems for high-performance motor vehicles as the third passive element alongside
the spring and damper [3]. Practical inerters may be constructed in a variety of ways
with currently the main types being: ballscrew [2], rack and pinion [2, 1], gear pump [2],
fluid-inerter [4, 5]. The range of potential application areas continues to expand beyond
automotive suspensions [6] to include: vibration absorption [1, 7] and vibration mounts
[8, 9], building suspensions [10, 11, 12, 13, 14, 15, 16, 17], vibration in cables [18, 19], con-
trol of motorcycle steering oscillations [20, 21], railway vehicle suspensions, [22, 23, 24, 25],
shimmy in aircraft landing gear [26, 27], passive walking for bipedal robots [28].
Despite the increasing prominence of the inerter in theory and applications there are
frequent misunderstandings of the concept. These arise principally due to the fact that

1
the inerter is both a device and an ideal modelling element. Whilst the same is true for
springs, dampers, capacitors, resistors etc, a proper understanding of the inerter does
seem to require a clearer appreciation of the distinction between the two. In the following
we will review the concept from first principles and provide a commentary on aspects that
are sometimes misunderstood. These include:

1. the notion of terminals;

2. approximate nature of the defining law;

3. distinction between mass and inertance of the device;

4. distinction between an effect and a device;

5. reductive versus holistic understanding of the purpose of inerters.

2 Mechanical one-ports and inerter definition

Figure 1 shows a schematic of an ideal two-terminal mechanical element (a one-port). The
two terminals are the connection points to other elements and have absolute displacements
x1 and x2 in some inertial frame of reference. The forces F at the terminals are equal and
opposite. Device laws are relations between the through-variable F and the across-variable
x = x2 − x1 .

F Mechanical F
one-port

x2 x1

Figure 1: Mechanical one-port element with sign convention in which a positive force F is
compressive and ẋ = ẋ2 − ẋ1 being positive corresponds to the terminals moving towards
each other.

We begin with the formal definition.

Definition 1 The ideal inerter is a mechanical one-port with the property that the equal
and opposite forces at the terminals are proportional to the relative acceleration between
them, in particular, in the notation of Fig. 1

F (t) = b (ẍ2 (t) − ẍ1 (t)) (1)

2
where b > 0 is the inertance and has units of kilograms.

The precise wording of the definition is crucial. It is essential that it concerns relative
acceleration, which must be measured between two points—the terminals (connection
points)—of the element. The terminals are also the application points for the forces. As
shown in Fig. 1 it is explicit that the forces are equal and opposite and colinear with the
line joining the terminals. The defining property is a proportionality between force and
relative acceleration. Yet neither the force nor the relative acceleration is an input or
output, just as the voltage across and current through a resistor are neither inputs nor
outputs of the device.
Definition 1 is not a law of nature or mechanics such as Newton’s second law which
relates the force on a mass to its absolute acceleration in an inertial frame of reference.
It is the property that a practical device must satisfy to count as an inerter. It must be
emphasised that the definition can only ever be satisfied approximately by real devices.
Even if we assume we can achieve zero losses through friction etc, and any components
used are ideal, it is still an approximation. This can be deduced directly from Definition 1.
Consider the situation that each terminal is subjected to an equal acceleration. From (1),
F (t) = 0. More fundamentally, since the forces at the terminals are equal and opposite,
there is no net force on the device. Hence there can be no net acceleration of the device,
which is a contradiction, unless it has zero mass. Since any practical device will have
some mass, it is evident that equation (1) will always be an approximation. A similar
assumption is inherent in the ideal modelling laws for springs and dampers. Indeed
Definition 1 should be viewed as the definition of an ideal modelling element.
It should be noted that the mass of an inerter device is different to, and typically
much smaller than, its inertance b. We will see in Sections 4 and 5 that (1) is a useful
description of a practical device if it has small mass compared to the objects that it is
connected to.

3 The mass element in mechanical modelling

There are two standard analogies between electrical and mechanical networks which are
power preserving. The older of the two is the force-voltage (velocity-current) analogy,
which has its roots in the notion of electromotive force. For network analysis and syn-
thesis the later force-current (velocity-voltage) analogy [29, 30, 31] is preferable because
it preserves topological structure—series (resp. parallel) connections in one domain are
of the same type in the other. Furthermore, terminals map to terminals and electrical

3
ground maps to mechanical ground (which is a fixed point or a point with constant ve-
locity in an inertial frame of reference). Figure 2 shows the traditional correspondences
between circuit elements in the force-current analogy.

Mechanical Electrical
F F i i
v2 v1
v2 v1 spring inductor
di
dF
dt
=k(v2 −v1 ) = 1 (v −v1 )
dt L 2

F i i
v2 v1
v2 v1 =0 mass capacitor
dv2 d(v2 −v1 )
F =m dt
i=C dt

F F i i
v2 v1
v2 v1 damper resistor
F =c(v2 −v1 ) 1
i= R (v2 −v1 )

Figure 2: Standard mapping of circuit elements in the force-current analogy where stiff-
ness k, mass m, damping c, inductance L, capacitance C, resistance R are positive con-
stants.

Figure 2 is based on the correspondence between through-variables T (Force F or

current i) and across-variables A (velocity v and voltage v). For the damper and resistor
T and A are proportional, for the spring and inductor dtd T and A are proportional, and
for the mass and capacitor T and dtd A are proportional. In this sense the correspondence
seems complete. But Figure 2 highlights the fact that five of the six elements are two-
terminal devices, while the sixth—the mass element—is not. From the point of view of
network synthesis the mass element is exceptional in that it has only one “non-grounded”
terminal. The symbol for the mass element shows a notional connection to ground through
the dotted line, but this is nothing other than an indication that Newton’s second law
relates the applied force to the absolute acceleration in the inertial frame of reference.
Aside from the issue of the terminals there is a further sense in which the mass el-
ement is exceptional. The other five elements in Figure 2 are all “engineering devices”,
namely they are contrivances that must be manufactured to achieve a property. The prop-
erty is only satisfied approximately because of deviations (parasitics) such as hysteresis,
temperature dependence (springs), fluid compressibility, non-laminar flow, cavitation, seal
friction (dampers), series inductance, parallel capacitance, temperature dependence, ther-
mal noise (resistors), Q-factor, dielectric absorption, leakage (capacitors), Q-factor, eddy

4
currents, hysteresis (inductors) etc. In addition there are practical upper limits on the
through and across variables to prevent damage to the devices.
In contrast to the other five elements in Figure 2 mass is not an engineering device
that must be manufactured. Its governing law, Newton’s second law, is presumed to be
exact in the classical theory and is not subject to hard limits on the magnitude of the
force or acceleration. Departures from this law are not of the same nature as for the other
five elements, e.g. relativisitc effects, or if a body is non-rigid. In the former case the
deviation from the classical law is negligible for most engineering purposes. In the latter
case, the fundamental law still holds with respect to the centre of mass.
Consideration of Figure 2 invites the question as to whether a device satisfying Defi-
nition 1 can be constructed. As emphasised in [1] any such devices would need to satisfy
certain practical conditions, e.g. the device should be capable of having a small mass,
independent of the required value of inertance; and the device should have a finite linear
travel which is specifiable independently of the inertance. In subsequent sections we will
discuss various methods of realisation which can achieve these requirements.

4 The ballscrew inerter

flywheel nut threaded rod

F2 F1

x2 bearing support housing x1

Figure 3: Ballscrew inerter.

Consider the two-terminal device depicted in Fig. 3 consisting of a threaded rod, nut,
and a flywheel rigidly connected to the nut which may rotate within the housing. The
housing is attached to terminal one and the threaded rod to terminal two, and they are
constrained not to rotate relative to each other. Let m be the mass of the flywheel, γ
the radius of gyration of the flywheel and p the pitch of the screw. Let the rod have
mass m2 and the rest of the device including nut, bearings, housing and flywheel have
mass m1 > m. Let the longitudinal reaction force between the rod and nut assembly be

5
denoted by F0 , the torque applied by the rod to the nut by T0 , and the angular rotation of
the flywheel with respect to the housing by θ. For modelling purposes we assume that the
rod and housing are constrained not to rotate within the frame of reference and we begin
by allowing different forces F1 and F2 at the two terminals. We may write the dynamic
equations of the device as follows:

−F1 = m1 ẍ1 + F0 , (2)

F2 = m2 ẍ2 − F0 , (3)
p
ẋ2 − ẋ1 = θ̇ , (4)
2π
2π
F0 = − T0 , (5)
p
T0 = mγ 2 θ̈, (6)

where (2)-(3) are Newton’s 2nd law applied to the housing and the rod, (4)-(5) are the
ideal transformer equations for the ballscrew-nut linear-to-rotational transducer, and (6)
is Newton’s 2nd law in rotational form applied to the flywheel. Eliminating T0 , θ and F0
leads to the pair of equations:

F1 = b(ẍ2 − ẍ1 ) − m1 ẍ1 , (7)

F2 = b(ẍ2 − ẍ1 ) + m2 ẍ2 (8)

where  2
2πγ
b= m.
p
The two terminal forces F1 and F2 contain a common term proportional to the relative
acceleration as well as an additional term proportional to the absolute acceleration of the
terminal. Clearly if these latter terms were significant then the device of Fig. 3 would
not be a good approximation of an ideal inerter. On the other hand there is the scaling
factor 2πγ/p which could be made large. For example, in a vehicle suspension application,
values of γ = 25 mm and p = 4 mm are quite practical, which makes (2πγ/p)2 ≈ 1542. If
m is a substantial proportion of m1 and m2 < m1 then the second terms in (7)-(8) may
reasonably be neglected in most situations. Hence, the device as depicted in Fig. 3 can
be a good approximation of an ideal inerter for suitable choices of m, m1 , m2 , γ and p,
namely
F1 = F2 = F = b(ẍ2 − ẍ1 )
holds approximately for F defined in Fig. 1.

6
 It is worth pointing out that (7)–(8) may be viewed as an ideal inerter connected
between a mass m1 at terminal 1 and a mass m2 at terminal 2. If other masses are to be
connected to the two terminals in an application then these additional terms serve just
to increase these masses, probably by a negligible amount.
We remark that use of an inerter in a suspension system will generally involve many
reversals of sign in the relative velocity ẋ2 − ẋ1 . Although ballscrews were not conceived
originally for such rapid and frequent reversals, they have proved more than capable
of operating in such a manner in their now standard use as a suspension element in
motorsport. (See Section 9 for further discussion.)
Penske Racing Shocks has led the commercial development and supply of inerters
since 2008. Figure 4 shows one version of Penske’s Formula One ballscrew inerter. The
construction is similar to the schematic of Figure 3. There is no internal keying against
rotation within the device. The terminal attachments for both the rod and the housing
employ a clevis mount which keys the rod and housing separately against rotation. In
this example the ballscrew is lubricated with grease and operates without seals.

7
 Figure 4: Penske ballscrew inerter.

5 Rack and pinion inerter

Consider the two-terminal device depicted in Fig. 5 consisting of a rack connected to
terminal 2 which is constrained to slide within the housing along the line to terminal 1
which is connected to the housing. The rack pinion and gear are constrained to rotate
within the housing with the angle of rotation θ0 satisfying rθ0 = x2 − x1 where r is the
radius of the rack pinion. The flywheel and pinion are also constrained to rotate within
the housing with the angle of rotation θ satisfying θ = −αθ0 where α is the gearing ratio.
Let the housing have mass m3 , the rack have mass m2 , the rack pinion and gear have
mass m0 and radius of gyration γ0 , and the flywheel and pinion have mass m and radius
of gyration γ. Assuming infinitesimal displacements δx1 , δx2 , δθ0 and δθ, and applying

8
 gear pinions flywheel

F2 F1

x2 rack housing x1

Figure 5: Rack and pinion inerter.

d’Alembert’s principle [32] gives:

0 = (−F1 − (m0 + m3 + m)ẍ1 )δx1 + (F2 − m2 ẍ2 )δx2 − m0 γ02 θ̈0 δθ0 − mγ 2 θ̈δθ
= (−F1 − (m0 + m3 + m)ẍ1 )δx1 + (F2 − m2 ẍ2 )δx2
m0 γ02 + mγ 2 α2
− (ẍ2 − ẍ1 )(δx2 − δx1 ).
r2
Since δx1 and δx2 are independent we obtain:

F1 = b(ẍ2 − ẍ1 ) − m1 ẍ1 , (9)

F2 = b(ẍ2 − ẍ1 ) + m2 ẍ2 , (10)

where b = (m0 γ02 + mγ 2 α2 )/r2 and m1 = m0 + m3 + m. The form of (9)–(10) is similar

to (7)–(8) and the same considerations apply. Values of γ/r ≈ 3 and α ≈ 6 are quite
practical, which provides a multiplier in excess of 300 to the flywheel mass m in the
second term of b. A further gearing stage is possible, say with ratio β, which provides a
third dominant term in b of the form m(γ/r)2 α2 β 2 . A simple meccano demonstrator was
constructed with m ≈ 0.1 kg, γ/r ≈ 2, α = β = 5 to give an inertance b ≈ 250 kg. With
such parameters the second terms in (9)–(10) can reasonably be neglected (or added as
small contributions to any masses connected directly to terminals 1 and 2) to provide a
practical implementation of an ideal inerter.
We conclude the section by recalling the circuit symbol that is commonly used to
represent the inerter which is similar in proportion to the capacitor but is reminiscent of
a flywheel (Figure 6).

9
 F F

x2 x1

Figure 6: Inerter circuit symbol.

6 Devices and effects

Any solid body subject to a stress gives rise to a strain. If the stress is small enough the
stress is proportional to the strain. This elastic behaviour is a “stiffness effect” of the
material and is the basis from which most springs are made. The stiffness effect of a solid
does not create a spring in full generality. Attachment points (terminals) are needed in
order that a spring as described in Figure 2 can transmit forces F of either sign (both
compressive and extensive). It is true that if a spring is pre-loaded so it always acts in
compression the connections to other elements may be simplified. Nevertheless, for the
purposes of the discussion here, we make a distinction between elasticity or compliance
as an effect and a spring as a device with two attachment points.
The same considerations apply to damping. As an effect, damping is associated with
the conversion of a high grade energy to heat through, for example, friction, viscosity
or internal deformation. In a damper, such effects are exploited to produce a device law
with respect to its terminals. The viscous damper, consisting of a piston moving in a
cylinder which forces hydraulic fluid through an orifice, requires careful design to produce
the desired relationship between force and relative velocity. The terminals of a damper
allow the necessary sign reversals in the terminal forces and the relative velocity between
them.
The inerter is an energy storage device, like the spring, but unlike the damper which
is a dissipator. If we denote v(t) = ẋ2 (t) − ẋ1 (t) then the energy delivered to the inerter
between times t1 and t2 is given by
Z t2 Z t2
E = F (t)v(t)dt = bv̇(t)v(t)dt
t1 t1
1
b v(t2 )2 − v(t1 )2


=
2
hence (1/2)bv(t)2 is the stored energy in the inerter at time t. In the ballscrew and rack
and pinion inerters the stored energy of the device is the energy stored in the rotating
flywheel. The effect associated with the inerter device is “inertia”. The way the effect

10
is harnessed to provide a device according to Definition 1 is less straightforward than
for the damper and spring. In both inerter types discussed so far there is, in addition
to the flywheel, a linear-to-rotational transducer, some form of gearing and necessary
bearing supports to connect with the terminals. The presence of a mechanical advantage
is necessary to create an inertance that is much larger than the mass of the device, though
this is not sufficient to indicate the presence of an inerter. For example, a car descending a
hill using engine braking is not an example of an inerter (see Section 11). We will discuss
a further example in Section 8 to highlight the difference between the inerter device and
the effect which it exploits.

7 Simulated mass
If one terminal of an inerter is connected to a fixed point (mechanical ground) as in
Figure 7 the remaining terminal behaves as a “simulated mass”, namely an applied force
satisfies the relationship F = b ẍ where b is the inertance. Although this is not a normal
use for the inerter (the main purpose of which is to have two independently movable
terminals) there are possible applications, as noted in [2, 1]. For example, if a spring-
damper support or absorber is required to be tested before final installation on a mass
which is extremely large (e.g. a large building) then the use of an inerter to simulate the
mass as in Figure 7 may be practical.

Figure 7: Inerter with one terminal grounded.

The arrangement of Figure 7 also serves to further illustrate the “effect” associated
with an inerter device. If a force F is applied to the non-grounded terminal then a
“resistance to acceleration” is experienced that is identical to a mass element of mass b.
It is this characteristic that is instantly identifiable in any inerter embodiment. Though
the effect alone doesn’t prove the existence of an inerter.

11
8 Mousetrap, cafetière and spinning top
An instructive demonstration using household items of the three passive elements (spring,
damper and inerter) can be found in a SkySports feature on racing car suspensions by
Ted Kravitz and Pat Symonds [33]. A rotational spring in the form of a torsion bar is
illustrated by a mousetrap. The proportionality of the torque with angle is clearly seen,
though the spring is pre-loaded to act in compression, so a genuine pair of terminals is
not in evidence. These are visible in the anti-roll bar graphic at the connection to the
rockers shown in the first part of the feature.
The damper is represented by a cafetière filled with oil. As the oil is squeezed through
the holes in the piston the resistance to motion increases with velocity in a manner that
is very similar to a viscous damper. Terminals can be easily imagined attached to the
handle and the bottom of the pot.
The item chosen to represent the inerter is a spinning top, which in general may be
defined as a body with a pointed tip on an axis about which it may be spun. There are
various methods to impart spin to a top, e.g. with the fingers or with a string or with
an external device. The top shown in [33] is a type with a built-in pump-action handle.
Attached to the handle is a rod with a spiral groove which may be pushed in and out of
the top along its axis of rotation. When there is downward pressure on the handle the
rod engages to spin the top, but disengages otherwise to allow free rotation.
In the SkySports feature Pat Symonds is careful to say that the Spinning Top “repre-
sents an inerter”. The property of the inerter that the top illustrates is the inertia effect.
There is a resistance to acceleration when there is downward pressure on the handle. The
effect is produced in a similar manner to the ballscrew inerter. Otherwise, the top behaves
very differently. There is no resistance to acceleration when the handle is pulled upwards.
Also, the pointed tip is quite unlike a terminal. The purpose of the tip is to allow steady
rotation about an axis which may vary, as in precession and nutation.
One can further observe, in contrast to the spinning top, that the angular momentum
in the inerter flywheel is inevitably stored only for a short time since in normal use the
device will reverse and give back its internal energy before the limits of travel are reached.
The coriolis force that is essential in the behaviour of a spinning top is only present for
short periods of time and is incidental to the function of the inerter. Counter-rotating
flywheels were proposed in [2] to eliminate any gyroscopic effects, though this has not
proved necessary in practical applications so far since the effect is small and transitory.

12
9 Sign reversals
A practical inerter has internal end stops to prevent the device falling apart. If the device
is driven into the end stops then it will ‘bounce’ back after hitting the spring buffer
with a more or less equal velocity, providing this velocity is not too large, otherwise the
device may be damaged. In a suspension system, such as a spring-damper-inerter parallel
arrangement, the design of the system ensures that the movement of the suspension will
normally be within the limits of travel of the suspension strut. The suspension will have
its own bump stops to deal with large inputs which take the suspension system out of
its normal limits. These bump stops will typically engage before the inerter reaches its
limit of travel. In such an application, in normal operation, an inerter undergoes repeated
reversals in the sign of v(t) = ẋ2 − ẋ1 while maintaining its defining law between force and
relative acceleration (1). This contrasts with many geared systems containing a clutch
element which disengages a load when the drive reverses. The spinning top of Section 8
is a simple example of this type where the response is different when the sign of v(t)
(the velocity of the handle relative to the tip) reverses. In contrast, it is a fundamental
requirement of the inerter that sign reversals do not affect its defining law. In both
methods to construct inerters presented so far, the ball screw inerter (Section 4) and the
rack and pinion inerter (Section 5), the flywheel angular velocity changes sign whenever
v(t) changes sign.

10 Back-driving
There is a further important design requirement for inerters which is related to the notion
of back-driving in geared systems. Consider a complete cycle for an inerter device in which
v(t) = sin(t) and F (t) = bv̇(t) = b cos(t). The velocity v(t) is positive for the first half
of the cycle and negative for the second half, while the force F (t) is positive only for the
first and last quarter. Each of the possible combinations of sign of v(t) and F (t) occurs
in the 4 quarter cycles. These correspond to forward-driving and back-driving within
the gearing system according to whether the two signs are the same or different. This
is illustrated in Figure 8 for a pair of spur gears. The primary means to transfer torque
between the two shafts is through the normal reaction force Rn at tooth contact. But the
frictional force Rt due to sliding must also be accounted for. The sign of Rt depends on
the sense of rotation. The sense shown in the figure corresponds to anticlockwise rotation
of the smaller gear (forward driving) and it is clear that the torque on the large gear can
be balanced by Rn and Rt even for large pressure angles ψ. For back driving the sign of
Rt reverses and this may no longer be possible, namely the resultant of Rn and Rt may

13
 base T1
circles

Rt
ψ
Rn

T2

Figure 8: Normal and tangential forces, Rn and Rt , at tooth contact in a pair of spur
gears with pressure angle ψ and torques T1 and T2 applied to the smaller and larger gear.

not be able to generate a torque on the large gear of the required sign. Thus the gears
may no longer back drive if ψ is too large.
Similar considerations apply to a screw and nut. Because of the rolling contact in
ball screws the frictional forces are greatly reduced so that back-driving is possible if the
pitch angle of the screw is not too small. It should be noted that forward driving and
back driving is determined according to the sign of F (t)v(t). For both the ball screw
inerter (Section 4) and the rack and pinion inerter (Section 5) back driving corresponds
to F (t)v(t) > 0, when energy is being supplied to the device, and forward driving to
F (t)v(t) < 0, when energy is being released.

11 When does a device count as an inerter?

A rather abstract answer to the question of the section title is: when equation (1) is
satisfied to a sufficient degree of approximation by the device. The realisations discussed
in Section 4 and 5 have become sufficiently well known that instances of a ball screw
connected to a flywheel, or a rack and pinion connected to a flywheel through gearing,
are now readily identified as inerters. But one must be careful: the mere presence of a

14
flywheel, ballscrew etc does not guarantee the presence of an inerter. Summarising the
discussion of Sections 6–10 it is useful to state the following necessary conditions.

1. The parameters (gearing ratio, pitch, masses etc) must be selected in order that
deviations from (1) are small enough. Typically, the inertance is significantly larger
than the mass of the device, and the latter is sufficiently small compared to other
masses to which the device is to be connected.

2. There must be a pair of terminals which are capable of transmitting forces F of

positive and negative sign.

3. The device must be capable of reversals of the relative velocity v while maintaining
the law (1).

4. The gearing must allow back driving.

A car descending a hill using engine braking is not an example of an inerter: there is an
“inertia effect” but 2. and 3. above fail. The spinning top is not an inerter since it fails
2. and 3. (at least).

12 Modelling and reverse-engineering

Assuming that the conditions of Section 11 are all satisfied for a ballscrew or rack and
pinion device, it is important to note that the use of (1) to describe the device’s terminal
behaviour is a modelling step. Without background knowledge on the inerter concept it
would be perfectly natural to model each component separately, e.g. to include each of
equations (2)–(6) individually in the system. A resulting simulation of the device, or a
system incorporating the device, would generally be indistinguishable from one that used
the modelling approximation. For the purpose of evaluating the performance of a system
that is given the modelling step provides perhaps only a small advantage of simplification.
It is in the design and system integration that the modelling step is decisive in allowing
the manner in which the systems is controlled or compensated to be put in a general
synthesis context, and with background theory to explain what is possible to build and
what is not.
Prior to wide dissemination of [1] the modelling step referred to in the previous para-
graph was not generally understood. There is a famous illustration of this fact from the
world of Formula One racing when a drawing of an inerter came into the hands of the
Renault Formula One team. A ruling from the FIA World Motor Sport Council on a

15
possible contravention of the sporting code suggested that the “drawing did not reveal to
Renault enough about the system for the championship to have been affected” [34]. In
truth reverse-engineering to discover a purpose is not straightforward.

13 Mass dampers versus J-dampers

The inerter was raced in Formula One for the first time at the Spanish Grand Prix in
2005 by Kimi Räikkönen after a secret development programme at McLaren Racing Ltd
in collaboration with the University of Cambridge. McLaren’s use of the inerter (code-
named the J-damper) eventually became public knowledge after a scoop by the Autosport
magazine (see [3], [34]). From an engineering perspective the use of the inerter in Formula
One racing provided a perfect illustration of the contrast between the 2-terminal nature
of the inerter and other more classical methods of vibration absorption.
At a similar time the Renault Formula One team was developing the mass damper
approach to improving “mechanical grip” in a racing car.3 In essence the approach is to
mount a “mass on a spring” to the sprung mass of the vehicle. The idea is a classical one
which dates to the early part of the 20th Century, even though the use is somewhat non-
standard in the Formula One context. Classically the mass damper (or tuned vibration
absorber) acts as a mechanical “notch filter” to prevent vibrations being transmitted
at a single frequency. Its purpose in Formula One was to reduce the mean tyre load
variations in a stochastic uneven road profile. The inerter achieves a similar improvement
in mechanical grip spread over a range of frequencies. Figure 9 shows the contrasting
circuit arrangements in which the mass has a single attachment point on the sprung
mass, whereas the inerter acts between the sprung and unsprung masses providing an
equal and opposite force in proportion to the relative acceleration.
The distinct nature of the two approaches is further illustrated by the fact that For-
mula Technical specifications have outlawed mass dampers, whereas the inerter in an
arrangement such as that shown in Figure 9 has remained legal.
It is useful to mention that the inerter also provides an alternative to the tuned mass
damper in its classical use as a “notch filter”. This is described in [1, Section III] (vibration
absorption). To insulate a body from a sinusoidal oscillation of the support at a fixed
frequency, a zero of transmission is needed in the transfer function from the support
displacement to the displacement of the body. In contrast to the classical solution of a
mass on a spring mounted on the body, it is shown that an inerter in parallel with a spring
incorporated into the support may achieve the desired effect with certain advantages.

16
 m

ms ms

mu mu

(a) (b)

Figure 9: (a) Mass damper (tuned vibration absorber) attached to the sprung mass
of a quarter car vehicle model in contrast to (b) an inerter acting in parallel with the
conventional spring-damper suspension elements.

14 Gear pump inerter

A further example of a method to construct an inerter using a gear pump/motor, described
in [2, 35], is illustrated in Figure 10. The device consists of a piston with through-rod
and a cylinder with hydraulic fluid on both sides of the piston. Each end of the cylinder
is connected by pipes to the inlet and outlet of the gear pump. Either of the gears in the
motor may be connected to a flywheel (not shown), either directly or after further gearing.
Alternatively, each of the gear wheels may be connected to flywheels which then would
be counter-rotating. The device is provided with two spring buffers, which are effective
when the limit of travel is reached but otherwise play no role in the modelling below. The
device operates by means of the piston displacing hydraulic fluid from the main cylinder
along the connecting pipes to cause a rotation of the gear pump, and vice versa.

17
 x2 spring buffer cylinder piston x1

F1
F2

rod
pipe

gear
pump/motor
gear

Figure 10: Gear pump inerter.

Let A be the cross-sectional area of the main cylinder minus the rod and let V be the
volumetric displacement of the gear pump per radian. Suppose that a flywheel of mass m
and radius of gyration γ is attached to the gear pump. Let p1 (resp. p2 ) be the pressure
in the hydraulic fluid on the terminal 1 (resp. terminal 2) side of the piston. Then it can
be seen from Newton’s second law that

−F1 + (p1 − p2 )A = m1 ẍ1 , (11)

F2 + (p2 − p1 )A = m2 ẍ2 , (12)

where m2 is the mass of the piston and m1 is the mass of the device without the piston.
Equating the flow rate of fluid displaced by the piston with the fluid flow rate into the
gear pump gives the equation
V θ̇ = A(ẋ2 − ẋ1 ) (13)
where θ is the angular rotation of the upper gear. If the volumetric efficiency of the gear
pump is 100%, namely there are no energy losses, we find by equating the power to drive
the piston with the power to rotate the pump that:

−mγ 2 θ̈ θ̇ = A(p2 − p1 ) (ẋ2 − ẋ1 ). (14)

Using (13) to eliminate θ in (14) and substituting for p2 − p1 in (11), (12) gives

F1 = b(ẍ2 − ẍ1 ) − m1 ẍ1 , (15)

F2 = b(ẍ2 − ẍ1 ) + m2 ẍ2 , (16)

18
where b = mγ 2 (A/V )2 . These equations take the same form as for the ballscrew and rack
and pinion inerter. Evidently the device as modelled above can be a good approximation
of an ideal inerter if b is sufficiently large compared to m1 and m2 . For practical devices,
additional effects may be included in the modelling to take account of other possible
departures from an ideal modelling law, e.g. viscous damping, and series compliance due
to compression of the fluid. (See [35] for a detailed discussion of these issues.)

15 Fluid inerter
A method to construct inerters which is entirely fluid based (the so-called “fluid inerter”)
is described in [4, 5, 36, 37]. The basic concept is illustrated in Figure 11. Again there
is a piston with through-rod within a cylinder with hydraulic fluid on both sides of the
piston and a (helical) tube connecting the two sides of the piston. The movement of the
piston relative to the cylinder causes fluid to flow along the tube.

x2 inlet piston x1

F1
F2

cylinder
rod

helical tube inlet

Figure 11: Fluid inerter.

Let A1 (resp. A2 ) be the cross-sectional area of the main cylinder minus the rod (resp.
the helical tube). Suppose the tube has length ` and the density of the fluid is ρ. Again
let p1 (resp. p2 ) be the pressure in the hydraulic fluid on the terminal 1 (resp. terminal 2)
side of the piston. Then from Newton’s second law
−F1 + (p1 − p2 )A1 = m1 ẍ1 , (17)
F2 + (p2 − p1 )A1 = m2 ẍ2 , (18)
where m2 is the mass of the piston and m1 is the mass of the device without the piston.
In the next modelling steps we will assume the fluid is inviscid and incompressible, and

19
denote by u the mean velocity of the fluid in the helical tube. Equating flow rates we
find A1 (ẋ2 − ẋ1 ) = A2 u. Considering the bulk motion of the fluid in the helical tube and
applying Newton’s second law we obtain

(p1 − p2 )A2 = ρ`A2 u̇. (19)

After substituting in (17), (18) we find

F1 = b(ẍ2 − ẍ1 ) − m1 ẍ1 , (20)

F2 = b(ẍ2 − ẍ1 ) + m2 ẍ2 , (21)

where b = ρ`A21 /A2 . Once again, the device as modelled by (20), (21) can be a good
approximation of an ideal inerter if b is sufficiently large compared to m1 and m2 .
Of course, real fluids are neither incompressible nor inviscid, and it cannot be assumed
that the fluid flow is properly modelled by its bulk motion. A study of these additional
effects is given in [5, 37]. It should be noted that such effects may sometimes be ex-
ploited to provide an integrated device with a circuit model comprising inerters, springs
and nonlinear dampers [37]. (See Section 18 for a more general discussion of integrated
devices.)

16 Rotational inerter
Rotational mechanical networks are analogous both to electrical networks and transla-
tional mechanical network through the notion of through and across variables ([38], [29]).
The terminals of a rotational network or element are a pair of colinear shafts which may be
freely and independently rotated, and the laws of network elements express a relationship
between the equal and opposite torque T applied to the shafts and their relative displace-
ment θ2 − θ1 , velocity or acceleration (see Figure 12). In [2] a definition was proposed of
an ideal rotational inerter as follows.

T θ2 θ1 T
Rotational
mechanical
one-port

Figure 12: Rotational mechanical one-port element with equal and opposite torques T
and rotations θ1 and θ2 about two colinear, rotational terminals.

20
Definition 2 The ideal (rotational) inerter is a rotational mechanical one-port with the
property that the equal and opposite torques T (t) at the terminals are proportional to the
relative angular acceleration between them, in particular,
 
T (t) = b θ̈2 (t) − θ̈1 (t) (22)

where b > 0 is the inertance and has units of kg m2 .

To be useful in applications the inertance b should greatly exceed the moment of inertia
of the device. A method to realise a rotational inerter using epicyclic gears was described
in [2]. A similar construction is shown in Figure 13. Let θ denote the absolute rotation of
the flywheel. Treating the gear box as ideal with ratio n and neglecting all inertias except
that of the flywheel with inertia J gives

T1 = T2 = nJ θ̈,
θ = n(θ2 − θ1 )

from which (22) is satisfied with T = T1 = T2 and b = n2 J.

T2 θ2 θ1 T1

flywheel housing gearbox

Figure 13: Rotational inerter embodiment

It is instructive to carry out a detailed modelling analysis of the construction in [2]

which makes use of a compound epicyclic gear train. An example is shown in Figure 14
in which there are two coaxial shafts (the terminals) with angular velocities and applied
torques ω2 , T2 and ω1 , −T1 respectively (note the opposite sign convention for the torque
at terminal one), gears of radii r1 and r4 , and an intermediate gear wheel with gears of
radii r3 and r2 fixed to a spindle A-B rotating with angular velocity ω4 which is attached
to a planet carrier which rotates about the main axis with angular velocity ω3 . The
gearing imposes the following constraints on the angular velocities:

ω2 r1 = ω3 R − ω4 r3 ,
ω1 r4 = ω3 R − ω4 r2 ,

21
 r3 r2 ω4
A B

T2 , ω2 ω3 −T1 , ω1
r1 r4

Figure 14: Compound epicyclic gear train

where R = r1 + r3 = r2 + r4 . Eliminating ω4 , noting that R(r2 − r3 ) = r1 r2 − r3 r4 , leads

to the identity
ω2 = γω1 + (1 − γ)ω3 (23)
where γ = r3 r4 /r1 r2 . We can also derive the following expression for ω4 in terms of ω1
and ω2 :
γR R
ω4 = − ω1 + ω2 . (24)
r3 (1 − γ) r2 (1 − γ)

− JR3 ω̇3

T2 −J2 ω̇2
J4 ω̇4
r1

−T1 −J1 ω̇1

r4 r1
r4
r2

r3

Figure 15: Compound epicyclic gear train (axial view)

Suppose that the moments of inertia of the gears attached to terminals 1 and 2 are J1
and J2 respectively, that the moment of inertia of the planet carrier about the principal
axis (including the intermediate gears) is J3 , and that the total moment of inertia of the

22
intermediate gears about their own axis is J4 . The axial view in Figure 15 shows the
real and inertial forces and torques on the intermediate gear from tooth contact and the
planet carrier assembly (all forces summed and shown as acting at one intermediate gear).
Equilibrium conditions give the following two equations:
−T1 − J1 ω̇1 T2 − J2 ω̇2 J3
+ − ω̇3 = 0, (25)
r4 r1 R
(−T1 − J1 ω̇1 )r2 (T2 − J2 ω̇2 )r3
+ + J4 ω̇4 = 0. (26)
r4 r1
We can verify that ! −1 !
1
r4
1
r1
R −γ rγ3
r2 r3 =
r4 r1
1−γ 1 − r12
and hence
! !
−T1 − J1 ω̇1 J3 −γ
= (−γ ω̇1 + ω̇2 )
T2 − J2 ω̇2 (1 − γ)2 1
!
2 γ 
J4 R r3
γ 1
− − ω̇1 + ω̇2
(1 − γ)2 − r12 r3 r2
!
ω̇1
= (M1 + M2 ) (27)
ω̇2
where
! !
J3 γ 2 −γ J4 γ 2 α2 −γαβ
M1 = , M2 = ,
(1 − γ)2 −γ 1 (1 − γ)2 −γαβ β2
and α = R/r3 , β = R/r2 .
Writing λ = r1 /r3 then α = 1 + λ, β = 1 + γλ and γα − β = γ − 1. Hence from (27)
 
γ(J3 + αJ4 )
T1 = b(ω̇2 − ω̇1 ) + − J1 ω̇1 , (28)
1−γ
 
J3 + βJ4
T2 = b(ω̇2 − ω̇1 ) + + J2 ω̇2 , (29)
1−γ
where b = γ(J3 +αβJ4 )/(1−γ)2 . By taking γ close to 1 the inertance b may assume a large
value. The equations also have additional terms as in (9)–(10) and (15)-(16) but with the
important difference of a term proportional to 1/(1 − γ). This will not grow as quickly
as the inertance as γ → 1 but may not necessarily be assumed small as in previous cases.
This term could be viewed as addition of some inertia to terminal 2 and corresponding
subtraction from terminal 1, when γ > 1, and vice versa if γ < 1. It is possible that this
term could be exploited to advantage in some mechanical control applications.

23
17 Synthesis of mechanical networks
The proposal in [1] to define a new ideal modelling element arose from the following
question: what is the most general mechanical impedance that can be realised passively?
The answer to the corresponding question in the electrical domain was well-known from
classical circuit theory which we now describe.

i
Z(s) Electrical
v
one-port

Figure 16: Electrical network with two external terminals, terminal voltage v, terminal
current i and driving-point impedance Z(s).

For a two-terminal electrical element or network the impedance (or more fully, the
driving-point impedance) is defined by Z(s) = v̂(s)/î(s) where v(t) is the voltage across
the terminals and i(t) is the current through the network, as shown in Figure 16, and ˆ
denotes the Laplace transform. A network is defined to be passive [39, p. 26],[40, p. 21] if
for all admissible v, i which are square integrable on (−∞, T ],
Z T
v(t)i(t) dt ≥ 0. (30)
−∞

The quantity on the left-hand side of (30) has the interpretation of the total energy
delivered to the network up to time T . Thus, a passive network cannot deliver energy to
the environment.
A function Z(s) is defined to be positive-real if Z(s) is real for real s and one of the
following two equivalent conditions is satisfied:

1. Z(s) is analytic and Z(s) + Z(s)∗ ≥ 0 in Re(s) > 0.

2. Z(s) is analytic in Re(s) > 0, Z(jω) + Z(jω)∗ ≥ 0 for all ω at which Z(jω) is finite,
and any poles of Z(s) on the imaginary axis or at infinity are simple and have a
positive residue.

It can then be shown [39, Chs. 4,5], [40, Theorem 2.7.1,2] that a one-port electrical
network with impedance Z(s) is passive if and only if Z(s) is positive-real. The same
statement holds for the admittance Y (s) = Z(s)−1 .

24
 In the mechanical domain the impedance of an element or network is defined as Z(s) =
v̂(s)/F̂ (s) with F (t) and v(t) = ẋ in the notation of Fig. 1, and the admittance is defined
by Y (s) = Z(s)−1 . Using the force-current analogy (see Section 3) we deduce the following
result.

Theorem 1 A one-port mechanical network is passive if and only if its impedance Z(s),
or equivalently its admittance Y (s), is positive-real.

The above theorem sets the boundary of what is possible to build as a passive me-
chanical device. But it doesn’t state that any such impedance can be physically realised.
Fortunately, this question has been answered in the electrical domain by the Bott-Duffin
theorem [41], which states that any positive real impedance may be realised by a circuit
comprised of resistors, capacitors and inductors only. The correspondences of Figure 2
with the mass element replaced by an inerter, and using the force-current analogy, allows
the result to be translated over directly, which we now state.

Theorem 2 Any positive real function may be realised as the driving-point impedance or
admittance of a mechanical network comprising springs, dampers and inerters.

The Bott-Duffin theorem is a famous and celebrated result in circuit theory, and its
importance increases with its potential use for mechanical networks. Despite it being a
classical result there remained unresolved questions about the construction when research
on the topic petered out in the 1960s. These questions relate to whether the construction
is the simplest possible among circuits that may realise a given mechanical impedance.
Recently it has been shown that the Bott-Duffin realisation is the simplest among series-
parallel circuits for some positive-real functions [42] and that a similar fact holds for
some classically known simplifications of the Bott-Duffin procedure to bridge circuits
[43]. Simplicity of realisation is an important matter in the mechanical domain and there
has been much research on this topic over the last decade. The reader is referred to [44]
and the references therein for further discussion.

18 Integrated mechanical devices

Mechanical network elements are often packaged into single units for reasons of space and
mechanical simplicity. For a more general one-port mechanical impedance/admittance
the external terminals will be in evidence as the external attachment points, as for a
stand alone network element. Internal nodes or terminals may or may not be explicitly

25
identifiable in certain realisations, and indeed there may be advantages of mechanical
simplicity to exploit an effect internally without requiring explicit internal nodes. A few
examples will serve to illustrate this point.

18.1 Parallel spring-damper

gas floating piston

hydraulic fluid
orifice hydraulic fluid

F2 F1

x2 coil spring housing piston x1

Figure 17: Telescopic hydraulic damper with floating piston and coil spring acting in
parallel.

Figure 17 shows a coil spring mounted between the housing and the end of the damper
rod of a telescopic hydraulic damper. Terminal 2 of the damper is rigidly connected to
one of the terminals of the spring, and the other terminal is connected to the housing
to which terminal 1 is rigidly connected. In this example, both terminals of the spring
and damper are clearly identifiable, and the integrated device may easily be dismantled
to exhibit the two devices separately.
Now consider Figure 17 in which the compressibility of the hydraulic fluid is taken
account of. Let us consider the case where the coil spring is removed and the piston orifice
is sealed. Movement of the piston relative to the housing causes an expansion/compression
of the fluid on either side of the piston and will result in a proportionality F2 = F1 = k1 x
to a first linear approximation. This force arises as a pressure difference across the main
piston, as does the damper force, and therefore these forces are notionally equal. In the
linear domain the device is consistent with the circuit diagram of Figure 18 where k is the
stiffness of the coil spring and c is the (linear) damper rate, though the node/terminal
between the damper and series spring is not identifiable as a physical attachment point.

26
 k

k1 c

Figure 18: Equivalent network for the device of Figure 17 showing the coil spring k,
damper c and series compliance k1 .

18.2 Parallel damper-inerter

gas floating piston

hydraulic fluid
damper rod orifice threaded rod
flywheel
F2 F1

x2 housing ball nut piston bearing x1

Figure 19: Telescopic hydraulic damper with floating piston and parallel ballscrew inerter.

Next we consider an integrated damper and inerter acting in parallel realised as in Fig-
ure 19 following [45]. The damper rod and piston with orifice(s) operates as a conventional
hydraulic damper which provides, in a linear approximation, an equal and opposite force
at the terminals in proportion to a relative velocity between them. The damper rod
and housing are keyed against any relative rotation. Within the piston is the nut of a
ballscrew/nut assembly. The threaded rod is arranged to rotate within the damper rod
and telescopes within it so that a relative movement between the damper rod and the
housing causes the threaded rod to rotate. The equal and opposite axial force on the
threaded rod is transmitted to the damper rod by the nut and to the housing by a bear-
ing assembly. A flywheel is attached to the threaded rod to provide an inertial force when
the rod rotates. The integrated device produces terminal forces which are approximately

27
given by
F1 = F2 = b (ẍ2 (t) − ẍ1 (t)) + c (ẋ2 (t) − ẋ1 (t))
and hence the device functions as an integrated parallel damper and inerter as depicted
in Figure 20. A coil spring may easily be added as in Figure 17 to realise an integrated
parallel spring-damper-inerter.

Figure 20: Equivalent network for the device of Figure 19 showing an inerter b in parallel
with a damper c.

Figure 21 shows Penske’s hybrid damper-inerter according to Figure 19 and [45]. Fly
weights attached to the end of the threaded rod may be adjusted by removal of the end
cap to vary the inertance. The damper is of standard telescopic type with an external
floating piston in a pressurised chamber similar to Figure 19. The ballscrew and nut
operate in the hydraulic fluid in a single sealed unit. See [53], [45] for further details on
this device.

28
Figure 21: Penske hybrid damper-inerter.

29
18.3 Inerter in parallel with a mechanical admittance

generator with
threaded rod flywheel coupling passive load
rod

F2 Ri F1
ZL
G

x2 housing ball nut bearing input shaft x1

Figure 22: Ballscrew inerter with parallel network.

Following [46], [47] and [48] we consider an integrated device as shown in Figure 22
in which a general passive admittance, implemented by a motor/generator (e.g. a DC
permanent magnet machine) and a passive electrical network, acts in parallel with an
inerter. In Figure 22, if the coupling to the generator is removed, the device takes the
form of a ballscrew inerter as in Section 4, with the only difference being that the threaded
rod rotates when there is relative movement of the terminals, rather than the nut as in
Figure 3. With the coupling in place, the input shaft to the generator rotates with the
threaded rod and flywheel. If the rod and ball nut have mass m2 , the rest of the device
has mass m1 and p is the pitch of the screw then equations (2)–(5) hold with F0 , T0
and θ defined as in Section 4. Let the rotational elements including the flywheel have a
combined inertia of Jm , and denote by Te the electrical torque produced by the generator.
Then
Jm θ̈ = T0 − Te . (31)
If the rotational motion of the generator induces a voltage Vs and associated armature
current I then these are related to the torque and angular velocity as follows:
Vs = kE θ̇, (32)
Te = kT I, (33)
where kE and kT are the voltage and torque constants, respectively, and kE = kT in SI
units. It is assumed that the generator has an armature resistance Ri and an electrical
load with complex impedance ZL (s) is connected across the terminals of the generator.
Thus
V̂s
= Ri + ZL (s) (34)
Iˆ
30
If the masses m1 and m2 are neglected then from (2)–(5) and (31)–(34) we find that
F1 = F2 = F and the admittance of the device of Figure 22 takes the form
 2 
kE2

2π
Y (s) = Jm s + .
p Ri + ZL (s)
Hence the device can be represented as the circuit of Figure 23 consisting of an inerter b
in parallel with a mechanical admittance Y1 (s) where
 2
2π
b = Jm ,
p
 2
2π kE2
Y1 (s) = .
p Ri + ZL (s)

Y1 (s)

Figure 23: Equivalent network for the device of Figure 19 showing an inerter b in parallel
with a mechanical admittance Y1 (s).

18.4 Rotational mechanical compensator

T ω2 ω4 ω3 ω1 T

T3

oil housing gearbox

Figure 24: Schematic of the physical model of the steering compensator.

A passive mechanical network incorporating inerters was proposed in [20], [21] as a steering
compensator for the control of wobble and weave instabilities in motorcycles. A schematic

31
of a physical model of the steering compensator [49] is shown in Fig. 24. The model
consists of a rotational n : 1 gearbox, a cylinder of inertia J1 connected rigidly to the
output shaft, and a second cylinder with inertia J2 which may rotate independently of
J1 . The assembly is contained in a housing filled with oil intended to provide a shear film
damper between the cylinders with linear damping with coefficient γ by means of a small
clearance δ between the two cylindrical surfaces. Additionally, there is a smaller damping
with coefficient γ1 , acting directly on the flywheel due to the clearance between the outer
cylinder and the housing. We can write down the modelling equations of the compensator
as follows, neglecting the inertia of the housing and rotating elements connected to the
input shaft of the gearbox:

T3 = n−1 T,
ω3 = n(ω2 − ω1 ),
T3 = J1 ω˙3 + γ (ω3 − ω4 ) ,
J2 ω˙4 = γ (ω3 − ω4 ) − γ1 ω4 .

We can verify that the admittance function of the compensator is given by

γ2
 
T̂ 2
= n sJ1 + γ − (35)
ω̂ sJ2 + γ + γ1
 
 2
−1  2 −1 !−1
γγ1 γ γ J2 s
= n2 sJ1 + + + . (36)
γ + γ1 γ + γ1 (γ + γ1 )2

where ω = ω2 − ω1 . The network corresponding to this linear admittance function is

shown in Fig. 25 in which
 2
2 γ
b=n J2 , b1 = n2 J1
γ + γ1
γ2 γγ1
c = n2 , c 1 = n2 .
γ + γ1 γ + γ1
and an additional series spring of stiffness k is included to represent the compliance of
the housing (see [49]).

32
 b
c

b1
k

c1

Figure 25: Equivalent network for the device of Figure 24 showing the series connection
of an inerter b and damper c in parallel with an inerter b1 and a damper c1 , all in series
with a spring k.

18.5 Internal nodes

It has already been noted that the device of Figure 17 does not display an explicit in-
ternal node if the series compliance is modelled. Namely, an effect (fluid compressibility)
produces a behaviour consistent with the circuit diagram of Figure 18.
In the device of Figure 19 the two terminals of the inerter are the nut and the bearing
support which connect rigidly to the external terminals. Hence each of the component
parts have identifiable terminals which coincide with the external terminals.
In the device of Figure 22 the two terminals of the inerter connect directly to the
external terminals. The mechanical torque from the generator adds directly to the inertial
load of the flywheel and hence the reactions transmit to the external terminals exactly as
for a stand alone inerter.
In the device of Figure 24 the parallel inerter acts between the external terminals
exactly as in the rotational inerter of Figure 13. The sheer film damper torque adds
directly to the inertial load of the rotational inertia J1 as in the device of Figure 22 and is
equal to the inertial load of the rotational inertia J2 plus an additional damping term. The
latter acts between the housing and the rotational inertia J2 . The mechanical admittance
takes the form of (35) and it is only after algebraic manipulation that the form (36) is
obtained, which corresponds to the circuit diagram of Figure 25. Thus, there is not a
direct correspondence between the internal nodes in Figure 25 and identifiable features in
Figure 24.

33
19 Epilogue: the purpose of inerters
Stand alone mechanical components typically have an identifiable purpose: springs pro-
vide compliant support, dampers dissipate energy etc. But what about the inerter? Its
origin was the completion of an analogy with the electrical capacitor. Fundamental the-
ory (see Section 17) then shows how to construct the most general passive mechanical
impedances which may be physically realised. The door is opened to improved dynamical
control of mechanical systems by purely passive means. Wherever springs and dampers
are employed, more general passive mechanical networks incorporating inerters could be
considered.
This line of thinking is characteristic of electrical engineering (from where the fun-
damental theory is drawn) and control systems (which are often designed to optimise
performance measures over some class of controllers). In electrical networks, components
are combined together to synthesise an impedance. A given impedance may be equiva-
lently realised by a number of different networks. Thus, it is a matter of debate whether
one should seek a “purpose” in each individual element of a network, or even if it is
possible to do so. The issue may usefully be illustrated by an imaginary interview of the
author (though not so different from some actual interviews).

Q. Thank you for your lecture, Professor Smith. So, please tell us, what does
the inerter actually do?
A. It does exactly what it says on the tin: it produces an equal and opposite
force at the terminals which is proportional to the relative acceleration be-
tween them.
Q. That doesn’t tell me what it does. Please express it in a different way.
A. Well, it’s an energy storage element; energy is stored whenever there is
relative movement between the terminals and it is given up when there is no
relative movement.
Q. But the spring is also an energy storage element and I know what a spring
does.
A. Yes indeed. The spring stores energy in proportion to a relative displace-
ment of the terminals from an unloaded state, either in compression or ex-
tension. The inerter stores energy when there is a relative motion between
the terminals, either moving towards or away from each other. The amount
of energy stored in the spring is proportional to the square of the load. The
amount of energy stored in the inerter is proportional to the square of the

34
relative velocity between its terminals. Force and relative velocity are dual
variables. Thus, the inerter is the mechanical dual of the spring, just as the
capacitor is the electrical dual of the inductor.
Q. This is not the sort of answer I am looking for. If I take a damper I know
that it takes energy out of a system. It dampens things down, and I know
exactly how it does it. I want you to give me a similar explanation of the
inerter.
A. I think I can explain things exactly as you want. But first I must step
back a bit. When electrical engineers talk about impedance they are usually
describing something that is frequency dependent, i.e. the notion of a complex
impedance. It is not just a proportionality between voltage and current, like
Ohm’s law, but there is a frequency dependence as well. It is the same for
mechanical elements. The damper on its own produces a direct proportional-
ity between force and velocity. For the spring, it is a proportionality between
force and the integral of velocity. The spring lags the damper by 90 degrees.1
Electrical engineers know that you need another element that would lead by
90 degrees. That’s the inerter.
Q. You are just giving me more of the same without getting to the point.
What is the use of having something that leads by 90 degrees?
A. At one level you can say that it expands the range of possibilities for con-
trol with elements that don’t require a power source. You would expect to get
better performance, once the system is optimised, at the expense of using an
extra component. Beyond that, there is a deep theoretical result which says
that any complex impedance that is realisable passively can be synthesised by
a circuit comprising elements of the three types (spring, damper and inerter)
only.2 So it is the most you can do in the linear domain before going active.
Q. What I am still struggling with is what the inerter is actually doing when
it is connected up as you describe. What effect is it having? When one has
an inerter in one’s hand one notices a resistance to changes in velocity. There
is an inertia effect. Isn’t the inerter nothing other than an inertial damper?
A. No. In the first place “inertial damper” is not a precisely defined term. It
has a rough meaning that something is damped out by inertia. At its sim-
plest level it might mean the use of a flywheel on a rotating shaft to reduce
fluctuations in angular velocity. But there is no understanding that it is a
two-terminal mechanical device with the defining property of the inerter.

35
Q. Yet your emphasis on the terminals and defining law of the inerter contrasts
with how one tries to understand other circuit elements. One has a feeling
that something is missing.
A. In regard to other circuit elements it is instructive to consider the analo-
gous electrical component: the capacitor. What does the capacitor actually
do? Well, it stores electrical energy. Or, current flows in and out of its ter-
minals in proportion to the rate of change of potential difference across them.
Capacitors have a multitude of applications: high-pass and low-pass filters,
oscillators, rectifiers, temporary power supply, sensing, power factor correc-
tion and so on. Each of these is an example of what a capacitor can do, but
it is misleading to say that any one of them explains what a capacitor does.
Q. Please give some examples of the practical use of inerters.
A. The first practical use of an inerter was to improve the “mechanical grip” in
a racing car,3 namely, it reduces the tyre normal load fluctuations to uneven-
ness in the road surface, which for complex reasons allows the driver to go a
bit faster without coming off the track. A slightly different way to deploy it in
the same application is to improve the damping ratio of key suspension modes,
which allows the static stiffness to be increased, which allows more aerody-
namic downforce to be used, which in turn allows faster cornering speed. The
interesting thing about this second use is that it improves “damping” (by
working in combination with the dampers and springs) in a way that is not
possible with springs and dampers alone.4
Q. Isn’t the inerter just adding “virtual” mass to the vehicle to allow the sus-
pension to work better without increasing the actual mass?
A. That’s not correct. Firstly, one would need to be clear if the inerter is
adding to the sprung mass (which would be an advantage), the unsprung
mass (which would be a disadvantage), or both (for which there is no clear
advantage).5 In fact it is doing none of the above. The easiest way to see
this is to consider the simple system of a pair of terminals between which an
inerter is connected. Compare that to the two terminals being point masses
with nothing connected between them. In the second case a force applied to
one terminal has no effect on the other, in contrast to the behaviour in the
first case. Thus, an inerter connected between two independently moveable
terminals is fundamentally different from any effect that can be produced by
point masses alone, and hence is different from any notion of “virtual” mass.
Q. Returning to the overall effect of the inerter on the vehicle. In essence, it

36
is making the car go faster, isn’t it?
A. Yes, but that’s not the best way to put it. It allows, by passive means,
better control. Hence it permits the driver to go round corners faster without
coming off the track, if the driver has the necessary skill. That is what is
required in this domain of application. The inerter can also be used for other
things in a car suspension, e.g. to improve ride comfort. It is also used in drag
car racing to keep the tyre in contact with the ground in transient situations.
Q. Are these the only applications of inerters?
A. By no means. We have looked at the problem of track wear for railway
trains. There the goal is to reduce the static yaw stiffness in the train suspen-
sion, but that causes instability of the motion of the wheelsets and reduces
the maximum speed. With inerters in the lateral suspension we have found
that we can improve the stability and maximum speed. So it is the opposite
of the approach in racing cars, namely we are trying to reduce stiffness while
preventing instabilities from arising.
Q. So, it is a passive mechanical component to improve the behaviour of ve-
hicles?
A. Not just vehicles in fact. There is a lot of work now to study its application
to building suspensions, e.g. to make them more resistant to earthquakes. I
recently heard of an application to improve the vibration suppression of an
isolation system for Michelangelo’s Rondanini Pietà.6
Q. Finally, we are making progress. At least we have established that the
inerter is a suspension component.
A. No, that is false. The examples of use I have given are all suspension sys-
tems of some type. But those are not the only applications of the inerter. For
example, it can be applied to the control of steering instabilities in motorcy-
cles.
Q. Let me try one last time. I want to understand this thing you call an
inerter. I want to taste it, touch it, feel it. I want to perceive its very essence.
Tell me what it does in a very simple way.
A. Okay, here’s my best shot. (Pause). The inerter is a mechanical network
element which can be combined with other network elements to produce an
overall effect.7
Q. (Angrily). What effect?
A. It depends what effect you want.

37
 Q. I give up. (Aside to the audience.) Will someone tell this fellow to answer
the question?

Notes
1. This treats the velocity as input and the force as output, so strictly the interpretation
is based on the complex admittance rather than the impedance.

2. See Theorem 2 in Section 17.

3. The term “Mechanical grip” is used in motorsport to describe the ability of the sus-
pension to keep the tyre in contact with the road. In terms of analytical performance
measures it maps most closely to J3 in [50] which is the rms tyre load variation for a stan-
dard Brownian motion road excitation. Mechanical grip contrasts with “aerodynamical
grip” which results from the increased normal load provided by the downforce of the wings.

4. See [1, Section IV.E] (suspension strut design example).

5. Vehicle performance metrics may typically be improved if the ratio mu /ms of unsprung
mass mu to sprung mass ms can be reduced. See [50] and [51].

6. See [9].

7. The reply is perhaps not as obtuse as the interlocutor thinks. It expresses the idea of
classical electrical circuit synthesis that circuits are built from a standard set of compo-
nents together with interconnection rules, and that for a given allowed set of components
one seeks to characterise the precise set of impedances that may be realised. Classical
circuit theory answered this question in a number of important cases: one-ports with
RLC components with or without transformers, multiports with RLC plus transformers
with or without gyrators (see [39], [40]). The inerter allows a corresponding answer in the
mechanical domain for one-ports comprising springs, dampers and inerters (Section 17).
The “overall effect” can be interpreted abstractly as the set of impedances generated by
the class, or as the value of a performance measure optimised over that class, or as a par-
ticular outcome such as a zero of transmission at a given frequency to eliminate the effect
of vibrations. The reply also hints at similar ideas in the field of control systems that
place emphasis on the notion of interconnection, e.g. the behavioural theory of Willems
[52].

38
Acknowledgements
The author is grateful to Tim Hughes, Jason Jiang, Alessandro Morelli, Rodolphe Sepul-
chre and the anonymous reviewers for helpful comments on the paper.

References
[1] Malcolm C Smith. Synthesis of mechanical networks: The inerter. IEEE Transactions
on automatic control, 47(10):1648–1662, 2002.

[2] Malcolm Clive Smith. Force-controlling mechanical device, July 4 2001 (priority
date). US Patent 7,316,303.

[3] Michael ZQ Chen, Christos Papageorgiou, Frank Scheibe, Fu-Cheng Wang, and Mal-
colm C Smith. The missing mechanical circuit element. IEEE Circuits and Systems
Magazine, 9(1):10–26, 2009.

[4] AR Glover, MC Smith, NE Houghton, and PJG Long. Force-controlling hydraulic

device. International Patent Application No: PCT/GB2010/001491, 2009.

[5] SJ Swift, Malcolm C Smith, AR Glover, Christakis Papageorgiou, B Gartner, and

Neil E Houghton. Design and modelling of a fluid inerter. International Journal of
Control, 86(11):2035–2051, 2013.

[6] Malcolm C Smith and Fu-Cheng Wang. Performance benefits in passive vehicle
suspensions employing inerters. Vehicle system dynamics, 42(4):235–257, 2004.

[7] Sara Ying Zhang, Jason Zheng Jiang, and Simon A Neild. Passive vibration control:
a structure–immittance approach. Proceedings of the Royal Society A: Mathematical,
Physical and Engineering Sciences, 473(2201):20170011, 2017.

[8] Yu-Chuan Chen, Sheng-Yao Wu, and Fu-Cheng Wang. Vibration control of a three-
leg optical table by mechatronic inerter networks. In 2014 Proceedings of the SICE
Annual Conference (SICE), pages 426–431. IEEE, 2014.

[9] A Siami, HR Karimi, A Cigada, E Zappa, and E Sabbioni. Parameter optimization

of an inerter-based isolator for passive vibration control of michelangelo’s rondanini
pietà. Mechanical Systems and Signal Processing, 98:667–683, 2018.

39
[10] Fu-Cheng Wang, Min-Feng Hong, and Cheng-Wei Chen. Building suspensions with
inerters. Proceedings of the Institution of Mechanical Engineers, Part C: Journal of
Mechanical Engineering Science, 224(8):1605–1616, 2010.

[11] Laurentiu Marian and Agathoklis Giaralis. Optimal design of a novel tuned mass-
damper–inerter (tmdi) passive vibration control configuration for stochastically
support-excited structural systems. Probabilistic Engineering Mechanics, 38:156–164,
2014.

[12] IF Lazar, SA Neild, and DJ Wagg. Using an inerter-based device for structural
vibration suppression. Earthquake Engineering & Structural Dynamics, 43(8):1129–
1147, 2014.

[13] Kaoru Yamamoto and MC Smith. Design of passive interconnections in tall buildings
subject to earthquake disturbances to suppress inter-storey drifts. In Journal of
Physics: Conference Series, volume 744, page 012063. IOP Publishing, 2016.

[14] D Pietrosanti, M De Angelis, and M Basili. Optimal design and performance evalu-
ation of systems with tuned mass damper inerter (tmdi). Earthquake Engineering &
Structural Dynamics, 46(8):1367–1388, 2017.

[15] Sara Ying Zhang, Jason Zheng Jiang, and Simon Neild. Optimal configurations for
a linear vibration suppression device in a multi-storey building. Structural Control
and Health Monitoring, 24(3):e1887, 2017.

[16] Dario De Domenico and Giuseppe Ricciardi. An enhanced base isolation system
equipped with optimal tuned mass damper inerter (tmdi). Earthquake Engineering
& Structural Dynamics, 47(5):1169–1192, 2018.

[17] Yinlong Hu, Michael ZQ Chen, and Malcolm C Smith. Natural frequency assignment
for mass-chain systems with inerters. Mechanical Systems and Signal Processing,
108:126–139, 2018.

[18] IF Lazar, SA Neild, and DJ Wagg. Vibration suppression of cables using tuned
inerter dampers. Engineering Structures, 122:62–71, 2016.

[19] Jiannan Luo, Jason Zheng Jiang, and John HG Macdonald. Cable vibration
suppression with inerter-based absorbers. Journal of Engineering Mechanics,
145(2):04018134, 2018.

40
[20] Simos Evangelou, David JN Limebeer, Robin S Sharp, and Malcolm C Smith. Control
of motorcycle steering instabilities. IEEE Control Systems Magazine, 26(5):78–88,
2006.

[21] Simos Evangelou, David JN Limebeer, Robin S Sharp, and Malcolm C Smith. Me-
chanical steering compensators for high-performance motorcycles. Journal of Applied
Mechanics, 74(2):332–346, 2007.

[22] Fu-Cheng Wang, Min-Kai Liao, Bo-Huai Liao, Wei-Jiun Su, and Hsiang-An Chan.
The performance improvements of train suspension systems with mechanical net-
works employing inerters. Vehicle System Dynamics, 47(7):805–830, 2009.

[23] Fu-Cheng Wang and Min-Kai Liao. The lateral stability of train suspension systems
employing inerters. Vehicle System Dynamics, 48(5):619–643, 2010.

[24] Jason Zheng Jiang, Alejandra Z Matamoros-Sanchez, Roger M Goodall, and Mal-
colm C Smith. Passive suspensions incorporating inerters for railway vehicles. Vehicle
System Dynamics, 50(sup1):263–276, 2012.

[25] Jason Zheng Jiang, Alejandra Z Matamoros-Sanchez, Argyrios Zolotas, Roger M

Goodall, and Malcolm C Smith. Passive suspensions for ride quality improvement
of two-axle railway vehicles. Proceedings of the Institution of Mechanical Engineers,
Part F: Journal of Rail and Rapid Transit, 229(3):315–329, 2015.

[26] Dong Xin, Liu Yuance, and ZQ Chen Michael. Application of inerter to aircraft
landing gear suspension. In 2015 34th Chinese Control Conference (CCC), pages
2066–2071. IEEE, 2015.

[27] Yuan Li, Jason Zheng Jiang, and Simon Neild. Inerter-based configurations for main-
landing-gear shimmy suppression. Journal of Aircraft, 54(2):684–693, 2016.

[28] Yuta Hanazawa, Hiroyuki Suda, and Masaki Yamakita. Analysis and experiment of
flat-footed passive dynamic walker with ankle inerter. In 2011 IEEE International
Conference on Robotics and Biomimetics, pages 86–91. IEEE, 2011.

[29] Floyd A Firestone. A new analogy between mechanical and electrical systems. The
Journal of the Acoustical Society of America, 4(3):249–267, 1933.

[30] Walter Hähnle. Die darstellung elektromechanischer gebilde durch rein elektrische
schaltbilder. In Wissenschaftliche Veröffentlichungen aus dem Siemens-Konzern,
pages 1–23. Springer, 1932.

41
[31] Georges Darrieus. Les modèles mécaniques en électrotechnique, leur application aux
problèmes de stabilité. Bulletin de la Société Française des électriciens, 9:794–809,
1929.

[32] Herbert Goldstein. Classical Mechanics, 2nd edition. Addison-Wesley, 1980.

[33] T Kravitz and P. Symonds. Front suspension, November 2017. SkySports Video,
www.skysports.com/watch/video/sports/f1/11092922/ted-and-pat-8211-suspension.

[34] Malcolm C Smith. Inerters. Princeton University Press, 2015. Chapter V.3, The
Princeton companion to applied mathematics, N.J. Higham (ed.).

[35] Fu-Cheng Wang, Min-Feng Hong, and Tz-Chien Lin. Designing and testing a hy-
draulic inerter. Proceedings of the Institution of Mechanical Engineers, Part C: Jour-
nal of Mechanical Engineering Science, 225(1):66–72, 2011.

[36] Bill J Gartner and Malcolm C Smith. Damping and inertial hydraulic device, Febru-
ary 14 2013. US Patent App. 13/577,234.

[37] Xiaofu Liu, Jason Zheng Jiang, Branislav Titurus, and Andrew Harrison. Model
identification methodology for fluid-based inerters. Mechanical Systems and Signal
Processing, 106:479–494, 2018.

[38] Jesse Lowen Shearer, Arthur T Murphy, and Herbert H Richardson. Introduction to
system dynamics, volume 7017. Addison-Wesley, 1967.

[39] Robert W Newcomb. Linear multiport synthesis. McGraw-Hill, 1966.

[40] B.D.O. Anderson and S. Vongpanitlerd. Network analysis and synthesis. NJ,
Prentice-Hall, 1973.

[41] R Bott and R J Duffin. Impedance synthesis without use of transformers. Journal
of Applied Physics, 20(8):816–816, 1949.

[42] Timothy H Hughes and Malcolm C Smith. On the minimality and uniqueness of
the bott–duffin realization procedure. IEEE Transactions on Automatic Control,
59(7):1858–1873, 2014.

[43] Timothy H Hughes. Why rlc realizations of certain impedances need many more
energy storage elements than expected. IEEE Transactions on Automatic Control,
62(9):4333–4346, 2017.

42
[44] Alessandro Morelli and Malcolm Clive Smith. Passive Network Synthesis: An Ap-
proach to Classification. SIAM, 2019.

[45] Bill J Gartner. Shock absorber with inertance, May 10 2016. US Patent 9,334,914.

[46] Fu-Cheng Wang and Hsiang-An Chan. Mechatronic suspension design and its ap-
plications to vehicle suspension control. In 2008 47th IEEE Conference on Decision
and Control, pages 3769–3774. IEEE, 2008.

[47] Fu-Cheng Wang and Hsiang-An Chan. Vehicle suspensions with a mechatronic net-
work strut. Vehicle System Dynamics, 49(5):811–830, 2011.

[48] L Pires, Malcolm C Smith, Neil E Houghton, and RA McMahon. Design trade-offs
for energy regeneration and control in vehicle suspensions. International Journal of
Control, 86(11):2022–2034, 2013.

[49] Jason Z Jiang, Malcolm C Smith, and Neil E Houghton. Experimental testing and
modelling of a mechanical steering compensator. In 2008 3rd International Sym-
posium on Communications, Control and Signal Processing, pages 249–254. IEEE,
2008.

[50] Frank Scheibe and Malcolm C Smith. Analytical solutions for optimal ride comfort
and tyre grip for passive vehicle suspensions. Vehicle System Dynamics, 47(10):1229–
1252, 2009.

[51] Malcolm C Smith and Stuart J Swift. Design of passive vehicle suspensions for
maximal least damping ratio. Vehicle System Dynamics, 54(5):568–584, 2016.

[52] Jan C Willems. The behavioral approach to open and interconnected systems. IEEE
Control Systems Magazine, 27(6):46–99, 2007.

[53] Simon McBeath. Shocks to the system. Racecar Engineering, 21(11):51–54, 2011.

43