Constraints on the coupling at the core-mantle and inner core boundaries inferred from nutation observations

),

The gravitational (tidal) forces from the Moon, the Sun and other planets apply a torque on the elliptical Earth.In response to this torque, the orientation of the Earth's rotation and figure axes varies in time when observed from a frame fixed in space, giving rise to precession/nutation motion.The precession is the large secular part of the motion while nutations are defined as the small periodic variations with periods larger than two days.
The amplitudes of the nutations depend on the Earth's ellipticity, internal density distribution, and rheological properties.For a three-layer Earth comprised of a mantle, fluid outer core and solid inner core, each of these individual regions would respond differently to the tidal torque on account of their different densities and ellipticities.As the outer core is fluid, differential rotations between the three layers can occur.The Earth's nutations are then also dependent on the interaction between the fluid core and mantle at the core-mantle boundary (CMB) and that between the fluid core and inner core at the inner core boundary (ICB).
Moreover, the amplitudes of the forced nutations are amplified by the presence of two normal modes, the 'free core nutation' (FCN) and the 'free inner core nutation' (FICN), whose frequencies are close to those of the tidal forcing.These modes are characterized by a differential rotation between the mantle, the fluid outer core and the solid inner core and mechanical couplings at boundaries affect their natural frequencies.Similarly, the damping of these modes depends on the energy dissipated through this coupling.The observed amplitude and phase of the forced nutations depend thus directly on the frequencies of the normal modes, which in turn depend on the mechanical coupling at the CMB and ICB.
Many of the physical properties at these boundaries on which the couplings depend are not well known.For instance, viscous coupling requires a knowledge of the viscosity of the fluid core close to the boundaries.Measurements are performed in laboratories on liquid iron alloys (e.g.Rutter et al. 2002a;Rutter et al. 2002b) but those experiments are carried on at much lower pressures (∼6 GPa) and temperatures (∼2000 K) than those typical of the outer core.Similarly, electromagnetic (EM) coupling depends on the electrical conductivity of both the fluid core and of the solid side of the boundary (mantle or inner core) as well as on the magnetic field at the boundary.We only have partial knowledge of the magnetic field at the CMB, the short-wavelength being masked by the crustal magnetic field (e.g.Stacey & Davis 2008).The magnetic field at the ICB cannot be observed at all due to the screening effect of the conducting core.
The tidal torque is known very precisely from celestial mechanics (Bretagnon et al. 1998;Roosbeek & Dehant 1998;Souchay et al. 1999).Without a precise knowledge of the key parameters that participate in the coupling at boundaries, it is then difficult to predict accurately the nutational response of the Earth.However, conversely, precise observations of Earth's nutations offer an opportunity to obtain information on these internal quantities for which we have little or no other constraints.
The nutational response of the Earth to the tidal torque has been the subject of many studies, including the model of Mathews et al. (2002) on which our study is based.This model predicts the nutation motion of an Earth model at hydrostatic equilibrium comprised of an elastic solid inner core, an inviscid fluid outer core and an anelastic mantle surrounded by an ocean layer.The gravitational interaction between the internal regions as well as the fluid pressure acting on the elliptical boundaries are included in the model.Other possible coupling mechanisms are not explicitly modelled but are taken into account by means of two complex parameters, referred to as 'coupling constants', K CMB and K ICB , characterizing the coupling at the CMB and ICB, respectively.
The key internal parameters of the nutation model, among them K CMB and K ICB , can thus be determined by an inversion procedure, where one infers knowledge on these parameters from a comparison between the nutation observations and model.A first estimation of K CMB and K ICB was obtained by Mathews et al. (2002).They used a linearized least-squares inversion in the frequency domain, only including the most dominant forced nutations that could be reliably estimated from the observations (Herring et al. 2002).In Koot et al. (2008), we developed a different inversion procedure directly in the time domain, thus avoiding any loss of information from retaining only a few of the forced nutations.Additionally, we used Bayesian inversion method, allowing us to take better into account the nonlinearity of the nutation model.
In this study, we further improve on the inversion procedure developed by Koot et al. (2008).We now include in the model nonperiodic variations in the tidal forcing, the so-called 'Poisson terms', and the freely excited FCN mode.In addition, we use nutation timeseries computed by different Very Long Baseline Interferometry (VLBI) analysis centres.This allows us to determine the accuracy with which we can recover the parameters that have an influence on nutations.
Our analysis is focused on the coupling constants K CMB and K ICB .We find that our retrieved value for K CMB is in agreement with that of Mathews et al. (2002) while K ICB is markedly different.Since K ICB characterizes the coupling at the ICB, this difference implies that previous estimates of physical parameters entering coupling models at the ICB must be revised.The coupling constant K ICB has been previously interpreted in terms of EM coupling at the ICB (Buffett et al. 2002).Mathews & Guo (2005) and Deleplace & Cardin (2006) generalized the model of Buffett et al. (2002) in order to take viscous coupling into account.In this paper, we concentrate our attention on these two coupling mechanisms.We investigate the fluid viscosity and the amplitude and structure of the magnetic field at the ICB that are compatible with our new value of K ICB .
Our paper is organized as follow.We start by presenting a brief description of the nutation model in Section 2. In Sections 3 and 4, we describe the nutation data sets that we use in our study and our inversion strategy, respectively.Results are presented in Sections 5 and 6, the latter also including the interpretation of our coupling constants in terms of physical parameters.We conclude by a summary and a discussion of our results in Section 7.

Forced nutations
From the perspective of an observer rotating with the Earth, the action of the gravitational (tidal) potential ( φ) on the elliptical Earth leads to small deviations from a uniform angular rotation 0 about an axis i 3 aligned with the geometrical figure.This deviation is termed 'wobble'.The perturbed rotation vector is then = 0 (i 3 + m), where m is a non-dimensional vector characterizing the wobble.The perturbation in the rotation vectors of the fluid core ( f ), and solid inner core ( s ) can be similarly defined as f = 0 (i 3 + m + m f ) and s = 0 (i 3 + m + m s ), where m f and m s are the wobbles of the fluid core and the solid inner core, respectively.
A prediction of the response of the Earth (in terms of these wobbles) from a given tidal forcing depends on the Earth's interior properties and on the coupling between the mantle, fluid core and inner core.For our study, we use the model of Mathews et al. (2002), an updated version of the model developed in Mathews et al. (1991a).We only present here a brief description of the model and refer the interested reader to the original articles.The model consists of a system of three dynamic equations relating the equatorial angular momentum of the whole Earth, outer core and inner core, to the equatorial torques acting on them.A fourth kinematic equation describes the orientation of the inner core.As the deviations from the steady rotation state are expected to be very small, the dynamic equations are developed to the first order in the wobbles.
The angular momentum of each internal region is given by the product of its moment of inertia tensor and its instantaneous rotation vector.The former are expressed in terms of mean principal moments of inertia ( A, A f and A s in the equatorial plane, C, C f and C s in the polar direction, respectively for the whole Earth, the fluid core, and the solid inner core) and of small corrections c ij , c f ij and c s ij (respectively for the whole Earth, the outer core and the inner core) due to deformations.These are from three different sources: (1) tidal forces (characterized by the tidal potential φ), (2) departure of the centrifugal force from that of the steady axial rotation (characterized by the wobbles m, m f and m s ) and (3) the time-dependent surface loading from the ocean (characterized by a loading potential φL ).To the first order, the c ij 's can be expressed as (Sasao et al. 1980;Sasao & Wahr 1981;Mathews et al. 1991a): where the tilde stands for the complex combination of the two equatorial components, for example, m = m 1 + im 2 .Eqs (1) introduce 12 so-called 'compliances' or 'generalized Love numbers ' (κ, ξ , ζ , χ , γ , β, δ, η, θ, α, ν and λ), characterizing the deformation of each region in response to a given forcing.For an elastic Earth, these compliances are real numbers.Here, we follow Mathews et al. (2002) and take mantle anelasticity into account, in which case the compliances are complex numbers.Since nutations are located in a very narrow frequency band, the compliances are assumed to be frequency independent.Nutation constraints on CMB and ICB coupling 1281 In addition to the external tidal torque, misalignments between the rotation vector of each region lead to internal torques between them.Inertial and gravitational torques have been computed by Sasao et al. (1980) and Mathews et al. (1991a) to the first order in the perturbations induced by the tidal potential.The torques from other coupling mechanisms, such as electromagnetic, viscous and topographic, are not modelled explicitly in Mathews et al. (2002) but are introduced in a parametrized way.The amplitude of these torques depends on the differential rotation at region boundaries.To first order, the equatorial torque applied on the outer core by the mantle ( ˜ CMB ) and that applied on the inner core by the outer core ( ˜ ICB ) can be written as (Mathews et al. 2002) where mf and ( ms − mf ) characterize the differential rotation between the mantle and outer core, and between the outer and inner cores, respectively.The non-dimensional parameters K CMB and K ICB are constants characterizing the strength of the coupling.They are complex parameters: their norm characterizes the strength of the coupling and the imaginary part the amount of dissipation.
The first-order dynamic equations form a system of coupled linear differential equations in the variables m, mf , ms , and ñs , this latter variable representing the deviation of the figure axis of the inner core from that of the mantle.Solutions for the dynamic variables depend on the forcing from the tidal potential φ.Because the motion of the celestial bodies is nearly periodic, to a very good degree of approximation, the tidal potential can be written as a sum of periodic terms where σ l is the angular frequency in cycles per sidereal day (cpsd), as seen from an Earth-fixed reference frame.
As the dynamic differential equations are linear, they can be solved independently for each term φ0 (σ l ) of the tidal potential (3).The four dynamic variables can also be expressed as a sum of periodic terms, such as and similar expressions for mf (t), ms (t) and ñs (t).Time derivatives are replaced by iσ l 0 and solutions for each coefficient m(σ l ) can be found independently in terms of the coefficient φ0 (σ l ) of the tidal potential.
The dynamic equations form a system of coupled linear algebraic equations written concisely in matrix form as where the vector x = [ m, mf , ms , ns ] T and φ0 (σ l ), φL (σ l ) and ĥ(σ l ) are the three sources of excitation of the wobble motion, all in the frequency domain: the tidal and ocean loading potentials, as well as the changes in the relative angular momentum of the ocean represented by ĥ(σ l ).Among these, the tidal potential is by far the most important.The 4 × 4 matrix M and the four-components vectors y, y L and y h describe the rotational response of the Earth to these three types of excitation.It includes the physical properties of the Earth characterized by the principal moments of inertia of each region, the 12 compliances, and the coupling constants K CMB and K ICB .The complete expression of M and y are given in Mathews et al. (1991a) (and in Mathews et al. (2002) for the inclusion of the coupling constants).The vectors y L and y h are given by: y L (σ ) = [(1 + σ )(τ − χ ), −σ η, −σ λ, 0] T , and y h (σ ) = [−(1 + σ ), 0, 0, 0] T , where τ = a 5 2 0 /(3G A), a is the mean radius of the Earth, and G is the gravitational constant.
The solution of the wobble in the frequency domain is where the notation [. ..] 1 indicates the first component of the vector.Eqs (4) and ( 6) allow us to compute the wobble m(t) for given excitation sources.This wobble is kinematically related to the nutation of the figure axis, that is, the variations in the direction of this axis in space.The latter is expressed in terms of two variables: the 'nutation in obliquity' ( ), defined as the angle between the figure axis and the pole of the ecliptic, and the 'nutation in longitude' ( ψ) corresponding to the motion of the vernal point (the intersection between the equator and the ecliptic) along the ecliptic.This 2-D motion is represented by the complex nutation variable η(t) defined as where 0 = 23 • 26 21.4 is the mean obliquity of the equator at J2000 (i.e. on 2000 January 1).The relation between the wobble m(t) and the nutation η(t), referred to as the Euler kinematic relation, is given by (e.g.Moritz & Mueller 1987;Mathews & Bretagnon 2003) This expression allows us to obtain a prediction of the nutation η(t) based on the wobble solution m(t) of our model.This prediction can be compared to the actual observations and the parameters that enter the nutation model can then be estimated.

Secular terms
Let us first consider the particular case of a periodic wobble with frequency σ l = −1, namely m(t) = m(−1)e −i 0 t .In this case, eq. ( 8) becomes: d η/dt = −i 0 m(−1).As the right-hand side is constant, the solution for η(t) is linear in time.This is the secular part of the motion which, for the component in longitude, corresponds to the precession.This secular term can be written in the form where P and ˙ are constants called the precession and obliquity rates, respectively, and c ψ and c are offset constants characterizing ηsec at the initial time t 0 .Here, we fix t 0 to J2000.The constants P and ˙ depend directly on H ≡ (C − A)/C, and for this reason H is referred to as the 'precession constant'.

Periodic terms
For a wobble m(t) given by eq. ( 4) with σ l = −1, ∀ l , eq. ( 8) gives the associated nutations C 2010 The Authors, GJI, 182, 1279-1294 Journal compilation C 2010 RAS Thus, the nutation associated with a periodic wobble of frequency σ l = −1 and amplitude m(σ l ) is also periodic with a frequency (1 + σ l ) and an amplitude − m(σ l )/(1 + σ l ).The difference of one cycle per day in the angular frequencies of the wobble and the associated nutation results from the different reference frames in which they are defined: the wobble is defined in the terrestrial frame, itself rotating with an angular velocity of one cycle per day with respect to the inertial frame in which the nutation is defined.With eqs (6), ( 10) and ( 11), a prediction for the nutation variable η(t) can be computed.
In order to emphasize the role played by deformations and internal couplings on the nutation motion, the solution is classically given in terms of the nutation ηR that the Earth would have if it were completely rigid.The relation between the rigid-Earth nutation and the tidal potential is given by (Mathews et al. 1991a) where e R is the dynamic ellipticity [i.e.(C − A)/A] of the rigid-Earth model.Rigid-Earth nutations ηR 0 (σ l ) can be computed from the very precise ephemerides of the solar system bodies.Several rigid Earth nutation models are available; here, we use the model REN2000 computed by Souchay et al. (1999).
Written in terms of the rigid-Earth nutation, the prediction for the nutation of the non-rigid Earth becomes where the coefficients TF G (σ l ), TF L (σ l ) and TF h (σ l ) are called 'transfer functions' and are given by TF G (σ l ) accounts for all deformations and internal couplings characterizing the non-rigidity of the Earth.TF L (σ l ) and TF h (σ l ) describe the nutational response of the Earth to the loading potential and relative angular momentum of the ocean, respectively.Writing the nutation motion in the form (13) allows for a separation of two different problems: the modelling of the rigid-Earth nutation, which is a purely astronomical problem that can be solved from celestial mechanics, and the modelling of the transfer functions, a geophysical problem that depends on parameters characterizing the Earth's interior.
The transfer function TF G (σ l ) can be rewritten in the form (Mathews et al. 2002) where H and H R are the precession constants of the Earth and of the rigid-Earth model, respectively.The σ i are four resonance frequencies, corresponding to the frequencies of the free modes allowed by the system (5).These corresponds to the frequencies for which Det[M (σ )] = 0.They are: the Chandler wobble (CW), the FCN, the FICN and the inner core wobble (ICW).Eq. ( 15) shows that the nutational response of the Earth to an external forcing depends on the forcing frequency in a resonant way: the closer is the forcing frequency to that of one of the free modes, the largest is the amplitude of the corresponding nutation.To the first order, the frequencies of the free modes are given, in the terrestrial frame (Mathews et al. 2002) where e, e f and e s are the dynamic ellipticities of the Earth, the outer core and the inner core, respectively, and α 2 is a parameter introduced by Mathews et al. (1991a) in the computation of the inertial and gravitational torque on the inner core.If there was no dissipation inside the Earth, the free mode frequencies would be real.In our model, dissipation occurs through mantle anelasticity and couplings at the CMB and ICB and the free modes have complex frequencies.Two of the free modes have frequencies that are close to diurnal and therefore close to those of the tidal potential, the FCN and FICN.These modes are then of major importance for the nutation motion.
The oceanic effects on nutation are described in eq. ( 13) by φL (σ l ) and ĥ(σ l ).Following Mathews et al. (2002), we only consider the effects from ocean tides and we do not take into account other dynamic effects of the oceanic circulation.The oceanic tidal angular momentum (OTAM) can be separated into two terms, called respectively the 'height' ( Ĥ L ) and 'current' ( Ĥ c ) terms, which are directly related to φL and ĥ by Ĥ L = −Aτ 0 φL and Ĥ c = ĥ.The OTAM can be computed from models of the main ocean tides (see e.g.Chao et al. 1996).For the other frequencies, the OTAM is computed from an interpolation procedure introduced by Wahr & Sasao (1981).Further details on the computation of the OTAM can be found in Koot et al. (2008).

Poisson terms
Although we approximated φ in (3) as a sum of periodic terms, the current precision of nutation observations is such that the effect of small departures from a purely periodic tidal potential can be detected.Very small linear corrections, φ1 (σ l ), called 'Poisson terms', have to be introduced These small φ1 (σ l ) terms are assumed to give rise to a nutation motion of the form Folgueira et al. (2007) showed that, to the first order, η1 (σ l ) is given by where ηR 1 (σ l ) are the Poisson terms of the rigid-earth model, which are directly related to the tidal potential terms φ1 (σ l ).

Free nutations
In addition to their amplification effect on the forced nutations, the free nutations FCN and FICN, excited by other means, participate in the total nutation motion.Their amplitude is expected to be timedependent, depending on the time-history of their excitation source.
While the free nutation arising from the FICN mode has not been observed yet, the FCN mode can actually be observed in the nutation data (e.g.Herring et al. 2002), with an amplitude approximately five orders of magnitude smaller than that of the main forced nutation (less than 0.5 mas over the period 1979-2010).The FCN mode is presumed to be excited, at least partly, by diurnal variations in atmospheric pressure (Sasao & Wahr 1981;Lambert 2006).
Unfortunately, models of the atmospheric angular momentum (AAM), computed from general circulation models, are not precise enough in the diurnal band to allow for a modelling of the excited FCN mode (Lambert 2006).For this reason, we model the free nutation empirically, as a pseudo-periodic term with an amplitude varying in time where F FCN is the frequency of the FCN mode in the celestial frame, that is, Herring et al. (2002), we choose to model the time-varying amplitude a FCN (t) as a piecewise linear function, that is, a function which is linear between given time nodes.The time nodes are chosen from a preliminary analysis to be at 1979, 1984, 1990, 1997, 2001 and 2010.

Corrections to the model
The nutation model is based on classical mechanics and does not take into account relativistic effects such as the geodesic nutation.This effect has been computed by Fukushima (1991) and contribute a change of −0.0304 and −0.0004 mas on the prograde annual and semi-annual nutations, respectively, and 0.0304 and 0.0004 mas on the retrograde annual and semiannual nutations, respectively.We have added these corrections to our model.Additionally, as the equations are developed to the first order, the model does not take into account non-linear effects.Second order effects have been computed by Lambert & Mathews (2006), who showed that these effects almost compensate each other so that the only non-negligible contributions are on the 18.6 yr nutation and are (0.0070 -i 0.0003) mas for the prograde component and (−0.0070 -i 0.0013) mas for the retrograde component.These corrections are also added to our model.

Parameters to be estimated
Among the parameters of the precession/nutation model, some have a large influence on the motion and can thus be estimated from the observations.These parameters are listed in Table 1.Parameters that have a lower influence on nutation are assigned fixed numerical values and are listed in Table 2.

Geophysical parameters
The focus of our study is on the geophysical parameters.Among them, the precession constant H [or equivalently the dynamic ellipticity e which is related to H by H = e/(1 + e)] determines the precession rate and has the largest influence on the precession/nutation motion.Additionally, it also directly influences the nutation transfer function (see eq. 15) so that it affects the amplitude of all of the forced nutations as well.
The most important parameters for the nutation motions are those that determine the frequencies and damping of the three main free modes (σ CW , σ FCN and σ FICN ) and the strengths of the associated resonances (N 1 , N 2 , N 3 in eq.15).The expressions for N 1 and N 2 are (Mathews et al. 2002) the expression of N 3 is not given as it does not introduce other geophysical parameters.Eqs (16a), (21a) and (21b) show the importance of the compliances κ and γ .These compliances are complex numbers and can be written as the sum of their (real) value for an elastic earth model and a small complex correction for mantle anelasticity effects: κ = κ el + κ AE and γ = γ el + γ AE .Following Mathews et al. (2002), we fix the value of the anelastic contributions κ AE and γ AE to that computed from an Earth interior model and we estimate the elastic term, namely κ el and γ el .
Eqs (16b) and (16c) show combinations between K CMB , K ICB , β, ν, e f and e s .Following Mathews et al. (2002), we choose to fix the values of the compliances (except κ el and γ el ) to those computed from an Earth interior model.The compliances for an elastic Earth have been computed by Mathews et al. (1991b) for the PREM model (Dziewonski & Anderson 1981) and are listed in Table 2. Small contributions to the compliances from mantle anelasticity can be computed, assuming no bulk dissipation, from the shear wave quality factor Q μ of the PREM model.We computed these contributions following the procedure described in Mathews et al. (2002) and the numerical values are listed in Table 2. Also, we fix e s to its hydrostatic value as the whole of the inner core is not expected to depart much from hydrostatic equilibrium (e.g.Defraigne   1996).The parameters that are to be estimated are then e f + Re(K CMB ), Im(K CMB ), Re(K ICB ) and Im(K ICB ).Their values are directly dependent on the values adopted for the fixed parameters.For example, if e s departs slightly from its hydrostatic equilibrium value, this departure is absorbed in the parameter Re(K ICB ).Similarly, errors in the computation of the compliances β and ν directly affect the estimate of the coupling constants.
The numerical values of A, A f , A s and e s , as well as the parameters α i are taken as those computed by Mathews et al. (1991b) from the PREM interior model (Dziewonski & Anderson 1981).They are given in Table 2.

Secular terms parameters
The secular component of the motion, given by eq. ( 9), depends on the parameters P, ˙ , c ψ and c .The latter three are new parameters to estimate and are listed in Table 1.The precession rate P can be expressed in terms of the precession constant H in the form where P R (H ) is the precession rate of the rigid Earth, which depends directly on H, and P NR is the contribution to the precession rate due to non-rigid effects and is independent of H.The value of H is close to that estimated from rigid Earth models (H R ) so that P R (H ) can be expanded as Numerical values for H R , P R (H R ) and d P R /d H R from Bretagnon et al. (1998) are listed in Table 2 along with the value of P NR computed by Mathews et al. (2002).

Free nutation parameters
The amplitudes a FCN = {a FCN (t i )} 6 i=1 of the FCN mode at the time of the selected time nodes {t i } 6 i=1 are parameters of the model to be estimated from the observations.

Atmospheric contributions parameters
Variations in the AAM, mainly due to the diurnal cycle in the heating of the atmosphere by the Sun, also induce forced nutations whose amplitudes are typically five orders of magnitude smaller than those driven by the tidal potential.In principle, these contributions could be computed from AAM models (Bizouard et al. 1998) but, as shown by Yseboodt et al. (2002), the results strongly depend on the choice of the atmospheric model.The estimation from AAM models is thus not precise enough to be added to our model.However, atmospheric effects influence mainly the annual prograde nutation (Bizouard et al. 1998); we thus follow Mathews et al. (2002) and incorporate them as a correction to the annual prograde nutation.This correction a ap is an additional parameter of the model to be estimated.

N U TAT I O N D ATA
Nutations are computed from VLBI observations.This technique consists in observing very distant galactic objects (quasars) from an array of Earth based large radio-telescopes.After each observing session, the signals recorded at each telescope are compared and differences between arrival times (delays) are inferred by correlations.These observed delays can be modelled in terms of several parameters such as the geographic location of the telescopes, the position of the quasars, the propagation delays due to the atmosphere, and the orientation of the Earth in space (and thus, in particular, the precession/nutation motion).The delays are then inverted to obtain an estimation of these parameters.
VLBI-delay data are processed by several analysis centres in order to get estimations of these parameters, including the precession/nutation angles of interest in our study.Each centre uses different analysis strategies and softwares to compute the nutation offsets.In this study, we use nutation angle time-series produced by three such centres: the NASA Goddard Space Flight Center (GSFC), the Paris Observatory (OPA) and the Institute of Applied Astronomy (IAA).We choose these centres because they present important differences in their analysis strategies.While both the GSFC and OPA centres use the Calc/Solv software, the IAA centre uses the OCCAM software.One of the characteristics of the OPA solution is that they use a different celestial reference frame that is defined by 247 stable sources selected by Feissel-Vernier et al. (2006)

I N V E R S I O N S T R AT E G Y
The strategy that we use for the inversion of the nutation observations has been described in details in Koot et al. (2008).We summarize here its main features.

Inversion in the time-domain
As the gravitational forcing is mainly periodic, nutation models are often developed in the frequency domain.However, nutation observations are available as time-series.In order to compare the model to the data, there are two possibilities: either (1) to estimate the amplitudes of the most important periodic terms in the timedomain data and compare these amplitudes to the frequency-domain nutation model or (2) to expand the nutation model in the time domain and compare it directly to the time-domain observations.The first method was used by Herring et al. (2002) while in Koot et al. (2008) we developed the second one.
The main advantage of using the time-domain method is that it allows us to use all the available data, which is not the case in the frequency domain where only the amplitudes of some 21 periodic terms can be extracted from the data (Herring et al. 2002).
The time-domain nutation model that we use is composed of the secular (eq.9), periodic (eqs 10, 13), Poisson (eq.18) and free FCN (eq.20) motions, where the symbol θ represents the parameters of the model that are to be estimated.The time-domain model given by ( 24) is more complete than the one used in Koot et al. (2008) where the Poisson terms were not taken into account.Also, in Koot et al. (2008), the time-variable amplitude of the FCN mode was estimated separately from a preliminary analysis and was removed from the data before the final inversion.Here, we estimate it jointly with the other geophysical parameters, a more self-consistent approach.

The Bayesian framework
In a Bayesian inversion framework (e.g.Gregory 2005;Tarantola 2005), knowledge of the model parameters (θ) is inferred from independent probability density functions (pdf) that describe the information provided by the observed data (d), the data-independent prior information on the parameters (π ) and the model predicted data (d * ).
The probability for the parameters given the observed data and the prior information is called the 'posterior' pdf and can be written as where L(d|θ ) represents the likelihood of observing the data d given the parameter values θ .We construct L by assuming that the observed data are related to the model prediction by where e d represents the uncertainty associated with the observed data and is chosen to be a normal distribution of zero mean and variance σ 2 .The model predictions d * are related to the parameters by where M(θ) is the model, also called the 'forward problem', and e M characterizes model uncertainty, independent of data.We choose the pdf representing e M to be a normal distribution of zero mean and variance σ 2 M .In this case, it can be shown (e.g.Gregory 2005) that the likelihood for observing data d given parameter values θ is From the posterior pdf, we can get the pdf for one specific parameter θ i by computing the marginal of the posterior defined by From this marginal, an estimated value for the parameter θ i can be obtained by computing its mean or the value for which the probability is the highest (the maximum a posteriori, MAP, solution).Confidence intervals (CI) corresponding to a given probability can also be estimated from the marginal in order to get an estimation of the error on the parameter.

Posterior used in our study
In our case, the data are two time series: d R = { ψ i sin 0 } N i=1 for the nutation in longitude and d I = { i } N i=1 for the nutation in obliquity, corresponding to the time indices {t i } N i=1 .We assume that the uncertainties in the measurement of each are independent and that they are given by normal distributions of zero mean and standard deviations σ R = {σ ψ i sin 0 } N i=1 and σ I = {σ i } N i=1 , respectively.In this case we have and the likelihood functions L(d R |θ) and L(d I |θ) can both be computed from eq. ( 28), which gives where with η(t i , θ) given by eq. ( 24).
In writing eq. ( 31), we make the hypothesis that the standard deviation characterizing the model uncertainty σ M is the same for nutation in longitude and obliquity.As this parameter is unknown, it is considered as an additional parameter to be estimated.From eq. ( 25), the posterior pdf is then given by where L(d|θ ) is given by eq. ( 31) and π (θ, σ M ) is the prior pdf on the parameters θ and σ M .Detail on the prior that we use are given in Koot et al. (2008).Note that, by our choice of prior, we impose the following constraints on the coupling constants: C 2010 The Authors, GJI, 182, 1279-1294 Journal compilation C 2010 RAS  Im(K CMB ) < 0, Re(K ICB ) > 0 and Im(K ICB ) < 0, in accordance with theoretical models of the couplings (Buffett et al. 2002;Mathews & Guo 2005).Samples of the posterior pdf are obtained from the Metropolis-Hastings algorithm (Metropolis et al. 1953;Hastings 1970).We note finally that the parameter σ M appearing in eq. ( 31) cannot be distinguished from a constant additive correction to the measurement errors σ R i and σ I i .Therefore, our inversion takes into account a possible underestimation of VLBI nutation angles errors, which have been shown by Herring et al. (2002) to be too optimistic.

I N V E R S I O N R E S U LT S
We have carried out individual inversions for each of the three nutation time-series independently, as well as an inversion with all the times series together.As the quality of the data largely improves with time, we have also performed an inversion of the GSFC timeseries restricted to the last 20 yr, that is, from 1990 to 2009.3.

Estimated parameters
The estimated values and the 99.7 per cent CI for all parameters, for each inversion, are presented in Table 3.The time-dependent amplitude of the FCN is given in Table 4 for the GSFC data set and shown in Fig. 1 for the three data sets.As expected, for all the parameters, the estimated values obtained from the joint inversion are practically equal to the mean of the individual inversions.The associated errors of the joint inversion are smaller than for the individual inversions, consistent with the fact that more data were used.
Figs 2 and 3 show the marginal posteriors for Im(K CMB ), Re(K ICB ) and Im(K ICB ).The marginal posteriors for other parameters are similar.Figs 2 and 3 also illustrate that while the errors for the inversions are small, the fact that the pdf's of individual inversions do not completely overlap indicates that these errors are likely underestimated.
In general, our retrieved parameters are in good agreement with the values estimated in Mathews et al. (2002).The only discrepancies are for the estimates of Re(K ICB ), Im(K ICB ) and e f + Re(K CMB ).The differences for the former two and their implication are the focus of Section 6.The discrepancy for e f + Re(K CMB ) does not come from our different inversion strategy as we get the same estimation when inverting the frequency data used by Mathews et al. (2002).It comes probably from a different value of Re( β AE ).
Differences between the results presented here and those in Koot et al. (2008) are from the inclusion of the Poisson terms which alter the amplitude of the long periods nutations.Other differences are from the contribution of mantle anelasticity on all of the compliances (rather than only on κ and γ ).The inclusion of the anelastic contribution to the compliance β affects directly our estimates of e f + Re(K CMB ) and Im(K CMB ) because, as already noted in Section 2.4.1, it is the sum (e f + K CMB − β) that enters the expression of the FCN frequency (16b).

Residuals
The residuals between the data and the model can be computed by the following estimator: ri The estimated residuals are shown on Fig. 4 for the GSFC timeseries.The residuals for the other data sets are very similar.The weighted root-mean-squares (WRMS) of the residuals, defined by  are 0.12 mas for both the ψ sin 0 and the components, for the GSFC data set, and 0.13 mas for the OPA and IAA data sets.These are smaller than the WRMS obtained by Herring et al. (2002).They are also smaller than in Koot et al. (2008) because the Poisson terms are no longer part of the residuals.This is also why the parameter σ M , characterizing model uncertainty, is smaller here than in Koot et al. (2008).The direct consequence of a smaller σ M is smaller errors on the parameters.

Periods and Q of the normal modes
From the samples of the posterior pdf given by eq. ( 32), we can infer the pdf of the complex frequencies (σ ) of the rotational normal modes that are in the frequency band of nutation, namely the FCN and FICN.These frequencies correspond to values of σ for which Det[M (σ )] = 0.The estimations for the corresponding periods T and quality factors Q, defined by: σ + 1 = T −1 [1 − (i/2Q)], are given in Table 5 for the individual inversions.

C O U P L I N G C O N S TA N T S A N D P H Y S I C A L P RO P E RT I E S O F T H E C M B A N D I C B
We now focus our attention on the coupling parameters.At the CMB, as Re(K CMB ) cannot be separated from e f , only Im(K CMB ) can be used to infer information on the coupling.We are thus interested in the parameters Im(K CMB ), Re(K ICB ) and Im(K ICB ).The numerical values of the latter two parameters are significantly different from previous nutation inversions.We explore below the reasons for these differences and investigate the consequence of these new values on the coupling at both the CMB and ICB.

Coupling constants
The estimates of Im(K CMB ), Re(K ICB ) and Im(K ICB ) obtained from the OPA and IAA time-series are in close agreement (see Table 3).The estimates from the GSFC time-series are somewhat different but are still consistent with the two other time-series as the 99.7 per cent CI always have an intersection.The inversion of the GSFC1 time-series gives mean values slightly different from that of the complete GSFC time-series but again the 99.7 per cent CI widely overlap one another.
Our estimate of Im(K CMB ) from the joint inversion is (−1.78 ± 0.02) × 10 −5 .As mentioned above, and shown on Fig. 2, the pdf's of Im(K CMB ) from the different inversions do not overlap one another exactly.The smallest (resp.largest) absolute value of Im(K CMB ), within the 99.7 per cent CI, from all our inversions is Im(K CMB ) = −1.72 × 10 −5 from the IAA time-series [resp.Im(K CMB ) = −1.83× 10 −5 , from GSFC].These upper and lower bounds on Im(K CMB ) are perhaps more representative of the uncertainties than the very small errors obtained in each separate inversions.We use these values in our analysis below.
Although Re(K ICB ) and Im(K ICB ) are significantly different, our estimate of Im(K CMB ) is in agreement with that of Mathews et al. (2002).This can be explained as follows.The parameter K ICB affects directly the complex frequency of the FICN (see eq. 16c).The period of this mode is approximately 1000 d in the celestial frame (Mathews et al. 2002, and Table 5) so that the forced nutations that Nutation constraints on CMB and ICB coupling 1289 are the most affected by resonance are the ones with long periods and in particular the nutation of 18.6 yr.The amplitude of this nutation is known with less precision than the other dominant forced nutations because only 30 yr of data are available.The difference in K ICB comes then mainly from a better account of the long-period terms: all the long-period periodic and Poisson terms are part of our model (not only the 18.6 yr periodic term).A part of the difference in K ICB also originates simply from the 10 yr of additional nutation observations that are part of our inversion.In contrast, Im(K CMB ) is mainly determined from the amplitude of the retrograde annual nutation, the forced nutation with the closest frequency to that of the FCN.Since the amplitude of the retrograde annual nutation could already be well determined with data prior to 2000, our estimate of Im(K CMB ) is not different from that of Mathews et al. (2002).

Coupling model
Our new estimations of the coupling constants have implications for the coupling at the CMB and ICB and in turn, on the physical parameters on which these depend.Our goal here is not to develop new models of coupling.We are merely interested in determining the numerical values of the physical parameters that are consistent with our coupling constants.We restrict our attention to EM and viscous coupling, which have been the focus of the most recent studies.We adopt the viscomagnetic coupling model of Mathews & Guo (2005), in which the coupling constant at one boundary b is given by where a b is the boundary radius, μ 0 is the magnetic permeability of free space, and θ is colatitude.I em b and I vis b include the contributions from EM and viscous forces, respectively, to the boundary torque.They depend on the radial magnetic field at the boundary (B r ), on the electrical conductivities of the fluid core (σ f ) and the solid side of the boundary (σ m for the mantle and σ s for the inner core), on the kinematic viscosity of the fluid core close to the boundary (ν), and on the density of the fluid core.The complete expressions for I em b and I vis b , at the CMB and ICB, can be found in Mathews & Guo (2005).In the limit of zero viscosity, the second term of the integrand in (35) vanishes and the purely EM model of Buffett et al. (2002) is retrieved.
We fix the electrical conductivities of the outer and inner cores to that of iron under core conditions, namely σ f = σ s = 5 × 10 5 S m −1 (Stacey & Anderson 2001).We take the densities of the fluid core at the CMB and ICB as given by PREM (Dziewonski & Anderson 1981), 9903.4 and 12166.3kg m −3 , respectively.
The radial magnetic field at the CMB and ICB that enters I em b is decomposed into a dipolar (B D r ) and a uniform (B ND r ) component, the latter representing an effective contribution to the EM torque from all field components other than the axial dipole.The viscous coupling part depends on the kinematic viscosity of the fluid core.Since it is poorly known, it is considered as an unknown parameter.

Coupling at the CMB
Our estimate of Im(K CMB ) is in agreement with that of Mathews et al. (2002) used in the study of Mathews & Guo (2005).Thus, we agree with the interpretation of Im(K CMB ) in terms of physical parameters presented in this latter study.Nevertheless, for completeness, we do include here an analysis of the coupling at the CMB on the basis of updated values of B D r and B ND r based on the recent model CHAOS-2s (Olsen et al. 2009) constructed from satellite observations of the magnetic field.Also, in our study, we give the errors on the physical parameters that arise from the errors on Im(K CMB ), which was not done by Mathews & Guo (2005).
Assuming a perfectly insulating mantle, the long-wavelength components (spherical harmonic degrees smaller or equal to 13) of the magnetic field at the CMB can be determined from a downward continuation of the magnetic field observed at the Earth's surface.The amplitude of the axial dipole at the CMB is thus well determined though it is currently decreasing at a rate of approximately 150 nT per year (e.g Gubbins et al. 2006).Since this corresponds to a very small change over the time-span of the nutation observations, we simply assume a fixed rms value of B D r = 0.209 mT based on CHAOS-2s (Olsen et al. 2009) evaluated at J2000.As for rms strength of B ND r at the CMB, though its long wavelength contribution is well determined, the short-wavelength contribution cannot be obtained from surface observations because it is masked by the crustal field.The true rms strength of B ND r is thus unknown.However, the power spectrum of the field between degrees 2 and 13 is found to follow a log-linear trend.Assuming this trend continues to higher degrees, an estimate of BND r can be constructed by extrapolation.Based on the mean power spectrum of CHAOS-2s between 2000 and 2008, we get BND r = 0.281 mT.We refer to this value of BND r below as that inferred from magnetic field observations, though this remains an unknown quantity.
We consider first the case of a purely EM coupling.We use B D r = 0.209 mT and treat BND r as an unknown parameter.The other unknown parameter is σ m .These two parameters cannot be both estimated from the single parameter Im(K CMB ).We choose to estimate BND r for different choices of σ m .Following Buffett et al. (2002), we first consider the case of a very high lowermost mantle conductivity taken to be equal to that of the core, σ m = 5 × 10 5 S m −1 .
Fig. 5 shows the real and imaginary parts of K CMB as a function of BND r .The black solid horizontal line corresponds to Im(K CMB ) = −1.78× 10 −5 , our estimate from the joint inversion (see Table 3 6.Since σ m cannot be larger than the electrical conductivity of the fluid core, the retrieved value of BND r = 0.617 mT under the above scenario corresponds to a lower bound on BND r .Smaller values of σ m require larger values of BND r : as reported in Table 6, with σ m = 5×10 4 Sm −1 , BND r must be approximately 0.91 mT.These estimates of BND r are much larger than our above quoted value of 0.281 mT based on magnetic field observations.This has been interpreted to indicate that, if no other coupling than EM participates, the small wavelength component of the radial field at the CMB has much larger amplitude than that expected from an extrapolation of its long wavelength component (Buffett et al. 2002;Mathews et al. 2002;Buffett & Christensen 2007).
As an alternative to the large BND r required in the purely EM coupling scenario, a part of Im(K CMB ) may be explained by viscous coupling.In Fig. 5, we show how adding viscous coupling affects the real and imaginary parts of K CMB and how a particular value of Im(K CMB ) can be explained by different combinations of ν and BND  Fig. 6 shows the relation between ν and BND r in agreement with our Im(K CMB ) estimates.Our curve for σ m = 5 × 10 5 S m −1 is similar to that obtained by Mathews & Guo (2005); we also show the curves of ν for other choices of σ m .As expected, the viscosity required to explain the estimated coupling constant increases as BND r decreases.
In particular, assuming σ m = 5 × 10 5 S m −1 , in order to reconcile BND r with 0.281 mT (the lowest value shown on Fig. 6), the viscosity of the fluid core has to be between 0.023 and 0.028 m 2 s −1 .Smaller σ m require larger values of ν , though the increase is modest: even for σ m reduced by four orders of magnitude, the required increase in ν is less than a factor 2. These values are upper bounds on the viscosity of the fluid core at the CMB and are reported in Table 7.

Coupling at the ICB
Our estimate of K ICB is not in agreement with that of Mathews et al. (2002).Both the real and imaginary parts are outside of the 99.7 per cent CI estimated by these authors.Moreover, for each  or of the relative proportion between the dipolar and uniform component, the ratio Re(K ICB )/ |Im(K ICB )| is always larger than 1.Consequently, our new value of K ICB appears incompatible with a purely EM coupling, at least on the basis of the model developed by Buffett et al. (2002).
It is important to note that there remains a degree of uncertainty in the ratio Re(K ICB )/|Im(K ICB )| that we infer.Since Re(K ICB ) is determined by the resonance of the nutations with the FICN (see eq. 16c), larger values of Re(K ICB ) would be obtained for larger value of the combination α 2 e s + ν el .The numerical values of these three parameters is given in We note finally that it is not possible to explain our estimated value of K ICB by viscous coupling alone.This is because the ratio Re(K ICB )/|Im(K ICB )| does not permit it, regardless of how large the viscosity near the ICB can be.Based on our results, 30 m 2 s −1 corresponds to an approximate upper bound for the kinematic viscosity near the ICB.Values as high as 10 7 m 2 s −1 that have been proposed (e.g.Smylie 1999) are incompatible with nutation observations.

D I S C U S S I O N A N D C O N C L U S I O N
The coupling constant at the CMB that we have inferred from nutation observations, is in agreement with the results previously obtained by Mathews et al. (2002).We find Im(K CMB ) = (−1.78± 0.02) 10 −5 for an inversion that includes three different nutation time-series.The errors correspond to 99.7 per cent confidence interval, though these are likely underestimates of the true uncertainties in Im(K CMB ).The highest and lowest value of Im(K CMB ) within the 99.7 per cent CI that we find from inversions of individual time-series are −1.72 × 10 −5 and −1.83 × 10 −5 , respectively.
The physical properties of the CMB that we derive on the basis of Im(K CMB ) are thus in agreement with of Buffett et al. (2002) and Mathews & Guo (2005).If the coupling is purely of EM nature, the non-dipolar part of the rms radial field at the CMB, BND r , must be at least 0.631 ± 0.004 mT or larger in order to explain the strength of the coupling.This is much larger than BND r = 0.281 mT inferred on the basis of an extrapolation of the long wavelength power spectrum of the field model CHAOS-2s.However, it is not inconceivable that the small wavelength part of the field at the CMB contains more energy than based on such a simple extrapolation (Buffett & Christensen 2007).If this is the case, then BND r = 0.631 mT may be taken as a constraint from nutation observations.If instead one takes BND r = 0.281 mT to represent an upper bound, then an additional coupling mechanism must be present.If this coupling is from viscous forces, the required kinematic viscosity of the fluid near the CMB is between 0.023 and 0.041 m 2 s −1 , depending on the electrical conductivity of the lower mantle and the precise value of Im(K CMB ).These values are upper bounds on the viscosity of the fluid core at the CMB.
Our new values of Re(K ICB ) and Im(K ICB ) have profound implications for the nature of the coupling at the ICB.The previous estimate of these parameters by Mathews et al. (2002) could be explained by EM coupling alone: this required a total rms radial field of 7.17 mT at the ICB.This is a much larger value than the typical magnetic field amplitudes in the core of 2-3 mT expected on the basis of numerical simulations of the geodynamo (Christensen & Aubert 2006), but could be taken as an observational constraint from nutations.However, with our new values of Re(K ICB ) and Im(K ICB ), the coupling at the ICB cannot be explained in terms of EM coupling alone, at least on the basis of existing models, regardless of the amplitude of the field.This is because the ratio Re(K ICB )/|Im(K ICB )| that we find is smaller than 1, whereas the EM coupling model can only accommodate a ratio larger than 1 (see Fig. 7).
An additional coupling mechanism must then be present.If this coupling is of viscous nature, kinematic viscosities between 10 and 30 m 2 s −1 are required, depending on the partition between the dipolar and non-dipolar part of the radial magnetic field at the ICB.These correspond to approximate upper bounds on the kinematic viscosity near the ICB.Higher values are incompatible with nutation observations.Importantly, we note that viscous coupling alone cannot explain our values of Re(K ICB ) and Im(K ICB ): a part of the coupling must still be accomplished by EM forces.Our estimate of the rms of the total radial field is between 6 and 7 mT, a value that remains much larger than that inferred from geodynamo simulations.
The kinematic viscosities quoted above, at both the CMB and ICB, are very large.Typical values inferred from laboratory measurements (Rutter et al. 2002b;Rutter et al. 2002a) and'ab initio' computations (Alfè et al. 2000) suggest that the molecular viscosity of iron at temperatures and pressures corresponding to that of the core should be of the order of 10 −6 m 2 s −1 .Our much larger viscosity estimates must then be taken as representing an effective viscosity ν e caused by turbulent motion transport in the fluid (e.g.Deleplace & Cardin 2006), similar to the eddy viscosity concept adopted when modelling boundary layers in the ocean and atmosphere (Pedlosky 1987).
Such a concept may seem appropriate at first glance.Adopting a simple isotropic turbulence framework, the effective kinematic viscosity should be approximately equal to other diffusivities in the system.The largest molecular diffusivity in the Earth's core is the magnetic diffusivity η = 1/μ 0 σ f ≈ 2 m 2 s −1 , a value approximately compatible with the effective viscosities that we infer.However, turbulence in the Earth's core is expected to be highly anisotropic on account of Coriolis and Lorentz forces (Braginsky & Meytlis 1990).It is then not clear whether taking ν e ≈ η is justified.
The eddy viscosity interpretation also has additional shortcommings.An effective viscosity is a property of the flow, not a property C 2010 The Authors, GJI, 182, 1279-1294 Journal compilation C 2010 RAS of the fluid.The size and turn-over timescale of the eddies that act as an apparent viscosity must be compatible with those expected to be present in the flow.A discussion on these issues is presented in Buffett & Christensen (2007).We shall not repeat this discussion here, but it is worth emphasizing a few important points.The eddy viscosity should scale as ν e ≈ vl, where v and l are typical velocity and eddy size (l is sometimes referred to as a mixing length).An upper bound limit on l is the thickness δ of the effective viscous boundary layer, otherwise the whole concept of eddy transport within the boundary layer falls appart.δ should scale as (ν e / o ) 1/2 based on an Ekman boundary layer thickness (e.g.Greenspan 1968).This leads to l ∼ δ ∼ 300 m when ν e = 10 m 2 s −1 , and requiring v to be of the order of 0.03 m s −1 .This is a velocity two orders of magnitude larger than the typical velocities of the large scale flows in the core.Realistically, l should be much smaller than δ ∼ 300 m for eddies to act like an apparent viscosity, in which case even higher eddy velocities are required.The same exercise with a viscosity typical of those that we retrieve near the CMB, ν e = 0.03 m 2 s −1 , leads to l ∼ δ ∼ 20 m and eddy velocities of at least 0.0015 m s −1 , still an order of magnitude larger than typical core flows.
Whether typical eddy velocities higher than 0.03 m s −1 and mixing lengths smaller than ∼100 m are compatible with the energetics and the force balance in the fluid core is clearly too vast a question to be tackled here.Though these would be the typical values that are required to explain the coupling at boundaries inferred from nutation observations.We thus reach the conclusion that, at the ICB, not only a large radial magnetic field is required to explain the coupling constant, a large fluid viscosity difficult to reconcile with molecular or eddy viscosity values is also required.As these large values for both the magnetic field and fluid viscosity appear to be incompatible with other geophysical constraints, this may indicate that the present-day models of EM and viscous coupling have serious inadequacies.For instance, the assumption that the differential rotation of the fluid core in the characterization of the free modes is adequately represented by a rigid motion may be incorrect.Using a Largrangian approach, Rogister & Valette (2009) has shown that indeed, for both the FCN and FICN, departures from rigid motion may be important.These non-rigid motions need to be taken into account in the coupling between the flow and the boundaries.Alternatively, other unmodelled physical processes may also be at play.For instance, viscous dissipation in the inner core may be important on diurnal timescales and could explain a significant part of K ICB (e.g.Greff-Lefftz et al. 2000).Some of these ideas are currently under investigation.

A C K N O W L E D G M E N T S
Part of the work of L. Koot was performed under a research fellowship of the Belgian National Fund for Scientific Research.L. Koot and A. Rivoldini acknowledge the support of the Belgian Science Policy Office (Action 1 Projects).M. Dumberry is currently supported by a NSERC/CRSNG Discovery grant.The contribution of O. de Viron to this paper is IPGP contribution 3008.We thank T. A. Herring, the editor B. A. Buffett, and an anonymous referee for their useful comments that helped us to improve the paper.

Figure 1 .
Figure 1.Estimated values and 99.7 per cent CI for the time-variable amplitude of the free FCN mode computed from the GSFC, OPA and IAA data sets.

Figure 2 .
Figure 2. Marginal posterior pdf of Im(K CMB ) for the individual inversion of the three time-series (GSFC, OPA and IAA), the joint inversion (Joint) and for the GSFC data truncated to the period 1990-2009.3(GSFC1).The blue box is the 99.7 per cent CI obtained by Mathews et al. (2002) (MHB).

Figure 3 .Figure 4 .
Figure 3. Marginal posterior pdf of Re(K ICB ) and Im(K ICB ) for the individual inversion of the three time-series (GSFC, OPA and IAA), the joint inversion (Joint) and for the GSFC data truncated to the period 1990-2009.3(GSFC1).The blue boxes are the 99.7 per cent CI obtained by Mathews et al. (2002) (MHB).

Figure 5 .
Figure 5. Re(K CMB ) and Im(K CMB ) as a function of BND r at the CMB, for σ m = 5 × 10 5 S m −1 and B D r = 0.209 mT.The results for three different choices of the fluid core viscosity are shown: ν = 0 (orange), ν = 10 −2 m 2 s −1 (purple), and ν = 2.6 × 10 −2 m 2 s −1 (blue).The horizontal lines are the values of Im(K CMB ) estimated from nutation observations and the three black vertical lines are the corresponding values of BND r in the case of the purely EM model.The vertical red line correspond to BND r = 0.281 mT inferred from magnetic field observation.

Figure 6 .
Figure6.Fluid core viscosity and radial uniform magnetic field at the CMB, for different values of the conductivity of the lowermost mantle σ m .The solid lines correspond to Im(K CMB ) obtained from the joint inversion, the dashed (resp.dotted) lines correspond to the lowest (resp.highest) absolute value of Im(K CMB ) allowed by our various inversions.

Table 1 .
Parameters to be estimated from the nutation observations.

Table 2 .
Numerical values of other parameters entering the nutation model.
, while the other centres use the conventional 212 sources that defined the Nutation constraints on CMB and ICB coupling 1285 International Celestial Reference Frame (ICRF).All three nutation time-series cover the time interval from 1979 up to the present.The specific data sets used here are available on the International VLBI Service (IVS) web site: http://ivscc.gsfc.nasa.govandlabelled 'gsf2008a.eops','opa2009b.eops',and 'iaa2007a.eops'.

Table 3 .
Estimated values and errors for the nutation model parameters computed from the GSFC, OPA and IAA nutation time-series, as well as for the joint inversion.

Table 4 .
1990lts for the GSFC time-series are shown for the whole period(1979-2009.3)andalsofor the period1990 -2009.3(labeledGSFC1).3(labeledGSFC1).The errors, corresponding to the 99.7 per cent CI, refer to the last written digits.The last column shows the results obtained byMathews et al. (2002) (MHB).Estimated values and 99.7 per cent CI for the timevariable amplitude of the FCN mode computed from the inversion of the GSFC data set (units: mas).

Table 5 .
Periods and quality factors of the FCN and FICN.The errors correspond to the 99.7 per cent CI.

Table 6 .
The rms of the radial magnetic field at the CMB required to match the estimated coupling constant for purely EM coupling.B D r is fixed to 0.209 mT.

Table 7 .
The fluid core viscosity ν at the CMB required to match the estimated coupling constant when the magnetic field at the CMB is fixed to B D r = 0.209 mT and BND r = 0.281 mT.r

Table 8 .
Estimates of ν, BND The first line gives the estimated value whereas the values in parentheses on the second line correspond to the 99.7 per cent CI.The results are given for K ICB based on the joint inversion of GSFC, OPA and IAA time-series.Magnetic field values are in mT and kinematic viscosities are in m 2 s −1 .