IL NUOVO CIMENTO Vol. 30 C, N.

4 Luglio-Agosto 2007
DOI 10.1393/ncc/i2007-10252-8

An inner additional condition for solving Munk-like ocean

circulation models

F. Crisciani(1 )(∗ ) and F. Cavallini(2 )

(1 ) ISMAR-CNR, Istituto di Scienze Marine del CNR - Trieste, Italy
(2 ) OGS, Istituto Nazionale di Oceanografia e di Geofisica Sperimentale - Trieste, Italy

(ricevuto il 22 Agosto 2007; approvato l’1 Dicembre 2007; pubblicato online il 2 Febbraio 2008)

Summary. — The problem of determining the additional boundary conditions, in

the framework of the wind-driven single-gyre ocean circulation with lateral diffusion
of relative vorticity, is explored with reference to the western boundary and in the
context of the classical linear Munk model. As, of necessity, each model solution
satisfies a certain inner additional condition, the latter is used to select special kinds
of partial-slip boundary conditions, one of which assures the dynamic stability of
the linear solution. Moreover, as the boundary conditions are left unaffected by
nonlinearity, the same condition can be applied also to the nonlinear model. In
particular, an open question about the flow energetics, reported in the literature, is
solved by using the present results.
PACS 92.10.Fj – Upper ocean and mixed layer processes.

1. – Introduction

Lateral diffusion of relative vorticity was first introduced by W. Munk [1] in his
archetypal model of steady wind-driven ocean circulation to parameterise vorticity dis-
sipation. Although nowadays this model is considered nothing but a classical result in
the framework of the so-called homogeneous models of the wind-driven ocean circulation,
both Munk’s model and those which followed from it exhibit a still unresolved indeter-
minateness. In fact, the term related to lateral diffusion of relative vorticity raises the
order of the governing differential equation thus demanding additional conditions, besides
those of no mass-flux, to single out a unique model solution. In the standard approach,
all the additional conditions have the form of boundary conditions, such as no-slip (null
tangential velocity) or free-slip (null relative vorticity) or super-slip (null flux of relative
vorticity) or hyper-slip (null flux of total vorticity) or, finally, partial-slip (linear and
homogeneous superposition of the first three conditions) to be applied along the border

(∗ ) E-mail: [email protected]


c Società Italiana di Fisica 347
348 F. CRISCIANI and F. CAVALLINI

of the fluid domain. Each additional boundary condition has its own kinematical mean-
ing and influence on the model output, but no criterion is known to select, according
to a shared physical viewpoint, the “true” or the “best” additional condition (if any)
or, at least, to restrict their set. Here we focus our attention to the western additional
condition and put forward an alternative inner one. Within the β-plane approximation,
we start from the fact that, for every gyre and for every latitude y included in the fluid
domain, a turning longitude ξ necessarily exists at which the meridional current of the
gyre vanishes. In general, ξ is determined by the full model solution. Unlike this, in the
present alternative ξ is fixed a priori and the meridional velocity is consistently forced
to vanish at ξ, thus releasing any additional condition at the western coastline. The
assumption of a traditional sinusoidal wind-stress curl, spanning the whole latitudinal
extension of the gyre, considerably simplifies the mathematics and allows us to transform
the vorticity equation into an ordinary differential equation whose integral depends para-
metrically on ξ. In turn, to obtain physically meaningful solutions, ξ must be restricted
to a suitable longitudinal interval and, for each ξ of this interval, the model solution is
in one-to-one relation with ξ itself. Then, given any model solution parameterised by ξ,
the additional condition at the western boundary can be inferred from it. In particular,
special partial-slip conditions are able to determine every ξ-solution, that is to say, every
physically meaningful solution of the linear model. Moreover, additional conditions of
this kind assure the dynamic nonlinear stability in the energy norm of every solution of
the linear model. It is also possible to find the value of ξ that minimizes the difference
between observed and computed (in the linear context) values of basic quantities in the
western boundary layer; so, from this viewpoint, the “best” additional condition can be
actually derived.
As boundary conditions are left unchanged even if the advection of relative vorticity
is added in the vorticity equation, the same additional condition previously inferred is
still valid also in the nonlinear context. This fact plays a noticeable role to carry out the
energetics of the gyre, which, in general, is sensitive to the kind of additional condition.
Unlike other additional conditions, which are problematic in the computation of the gyre
energy in a closed fluid domain, the special partial-slip conditions that are considered
here successfully lead lateral friction to drain energy, as it is physically expected.

2. – The Munk-like models and the inner condition

We start from the linear, non-dimensional vorticity equation governing the steady
wind-driven circulation (notation as in Pedlosky [2] is understood):

∂ψ
(2.1) = wE (x, y) + ε∇4 ψ
∂x

in the square fluid domain

(2.2) D = [0 ≤ x ≤ 1] × [0 ≤ y ≤ 1]

of the β-plane. The shape of (2.2) means that D is included between two circles of
latitude (i.e. at y = 0 and y = 1) along which the wind-stress curl is zero (see the
following equation (2.4)) and two coastlines in the meridional direction; this geometry
considerably simplifies the subsequent model solution.
INNER CONDITION FOR MUNK 349

The no mass-flux condition across the boundary ∂D is expressed, in terms of the

stream function ψ, by

(2.3) ψ(x, y) = 0 ∀(x, y) ∈ ∂D.

To describe a rudimentary subtropical single gyre included into (2.2), the simple
traditional assumption

(2.4) wE (x, y) = − sin(πy)



is made about the vertical component wE = k̂ · ∇ × τ of the wind-stress curl.
Equation (2.1) is the linear approximation of the fully nonlinear vorticity equation,
which takes into account the inertial behaviour of the flow; therefore, the eddy viscosity
coefficient ε in (2.1) is constrained by the estimate

(2.5) ε = O(10−3 ).

A classic result of geophysical fluid dynamics states that, far enough from the western
boundary and in the linear or weakly nonlinear regime, the transport is adequately
represented by the so-called Sverdrup flow. In the present context, because of (2.4), the
related stream function ψSv is

(2.6) ψSv (x, y) = (1 − x) sin(πy).

Since (2.6) implies

(2.7) ∇2 ψSv = −π 2 ψSv ,

the identity

(2.8) ∇2 ψSv = 0 ∀(x, y) ∈ C

holds true along the zonal and eastern parts of ∂D, say C. An equation of the same form
as (2.8) is assumed for the overall model solution ψ, setting

(2.9) ∇2 ψ = 0 ∀(x, y) ∈ C.

In the latter case, (2.9) constitutes the additional boundary condition (so-called free-slip)
that will be imposed along C for solving the model governed by (2.1) [3].
Finally, unlike all the Munk-like models considered in the literature, we renounce every
a priori additional condition at the western boundary in favour of the inner additional
condition
 
∂ψ
(2.10) = 0, ∀y ∈ [0, 1].
∂x x=ξ

In (2.10), ξ ∈ ]0, 1[ is, for a given single-gyre flow, the unique longitude at which the
350 F. CRISCIANI and F. CAVALLINI

∂ψ
meridional current ∂x necessarily reverses its sign, i.e.

∂ψ ∂ψ
(2.11) > 0 (northward flow) ∀x < ξ and < 0 (southward flow) ∀x < ξ.
∂x ∂x

We stress that (2.10) is nothing but a consequence of Rolle’s theorem applied to ψ in the
interval 0 < x < 1, while the uniqueness of ξ, according to (2.11), is demanded by the
phenomenology of the actual flow patterns.
To summarise, the present model is composed of eq. (2.1), the no mass-flux boundary
condition (2.3) along ∂D, the additional boundary condition (2.9) along C and the inner
additional condition (2.10) for a certain longitude ξ. Note that, because of (2.3), (2.9)
is equivalent to

   
∂2ψ ∂2ψ
(2.12) = =0
∂y 2 y=0 ∂y 2 y=1

and
 
∂2ψ
(2.13) = 0.
∂x2 x=1

The trial stream function

(2.14) ψ(x, y) = Φ(x) sin(πy)

identically satisfies condition (2.12); moreover its substitution into (2.1) leads to the
following ordinary differential equation for the sole Φ (primes and Roman superscripts
denote x-derivatives):

 
(2.15) Φ = −1 + ε ΦIV − 2π 2 Φ + π 4 Φ .

Equation (2.15) is supplemented by the no mass-flux boundary condition at the merid-

ional boundaries coming from (2.3), i.e.

(2.16) Φ(0) = 0 and Φ(1) = 0,

by the inner additional condition (2.10), i.e.

(2.17) Φ (ξ) = 0

and by the free-slip boundary condition (2.13), i.e.

(2.18) Φ (1) = 0.

INNER CONDITION FOR MUNK 351

Fig. 1. – Mid-basin (y = 1/2) meridional velocities Φ (x; ξ) related to the extreme values of
ξ (hatched line is for ξ = 0.144, continuous line is for ξ = 0.315). The marked difference
between these profiles is ascribed only to the different values of the turning longitude in solving
problem (2.15)-(2.18).

3. – The range of validity of the inner condition

To stress that the solution of problem (2.15)-(2.18) depends parametrically on the

turning longitude ξ, we denote its solution with Φ(x; ξ) rather than Φ(x). This problem
(for instance with ε = 10−3 in accordance with (2.5)) can be solved in principle for every
ξ ∈ ]0, 1[; but only the solutions satisfying the double constraint (2.11) are qualitatively
consistent with the phenomenology of the subtropical single gyres. In other words, in-
equalities (2.11) constitute a test of physical validity for Φ(x; ξ), and a numerical analysis
of all the mathematically admissible solutions shows that the test is passed if and only if

(3.1) 0.144 ≤ ξ ≤ 0.315.

Interval (3.1) is closer to the western boundary than to the eastern one, in accordance
with the westward intensification of the meridional flow, which is implicit in the dynamics
governed by (2.1). We expect that, for ε = 10−3 and/or wE (x, y) = − sin(πy), the
range (3.1) slightly changes; but we disregard different choices of ε and wE . We shall
return to this point at the end of sect. 5. In any case, by varying ξ into (3.1), all the
solutions of the above posed model are obtained. In particular, a noticeable feature of
the special solution ψ = Φ(x; 0.315) sin(πy) is

(3.2) Φ (0; 0.315) = 0,

that is to say ψ = Φ(x; 0.315) sin(πy) satisfies the classical no-slip boundary condition
at the western boundary. The plots of Φ (x; 0.315) and Φ (x; 0.144) represent the mid-
latitude meridional velocities in the extreme cases of (3.1) and are depicted in fig. 1.
352 F. CRISCIANI and F. CAVALLINI

Fig. 2. – Upper panel: maximum mid-basin (y = 1/2) northward velocity (4.1) as a function of
ξ. Lower panel: mid-basin (y = 1/2) northward transport (4.2) as a function of ξ.

4. – On the “best” inner condition

The main feature of the simple circulation model governed by (2.1) lies in its ability
to explain the westward intensification of the wind-driven current. However, the compu-
tation of basic dynamic variables in the western boundary layer, such as the northward
transport M or the northward velocity v as well as the width δ of the layer itself, depends
markedly on the additional condition imposed at the western boundary. Therefore, not
every additional condition equally contributes to westward intensification, and we expect
that the same fact happens also if, alternatively, the inner additional condition (2.10) is
used with different values of ξ. Hence, we can single out the turning longitude ξ of (3.1)
that minimizes the difference between the observed and theoretical values taken by M ,
v and δ. The crudeness of the linearity assumption prevents the achievement of a mini-
mization which is quantitatively satisfactory; but, in spite of this, a univocal selection of
the best (in the above said sense) turning longitude can be attained.
The maximum northward current referred to the mid-basin latitude (i.e. for y = 1/2)

(4.1) v(ξ) = max Φ (x; ξ),

0≤x≤ξ
INNER CONDITION FOR MUNK 353

and the northward transport at the same latitude

 ξ
(4.2) M (ξ) = Φ (x; ξ)dx = Φ(ξ; ξ),
0

are evaluated by means of the solution of problem (2.15)-(2.18), while the width of the
western boundary layer is simply

(4.3) δ(ξ) = ξ,

where the turning longitude ξ satisfies (3.1). The plots of (4.1) and (4.2) are reported
in fig. 2. In particular, because of the monotonic decreasing behaviour of v and M , we
have

(4.4) vmax ≡ max v(ξ) = v(0.144) ≈ 36,

ξ
(4.5) Mmax ≡ max M (ξ) = M (0.144) ≈ 2.3,
ξ

while, because of (3.1),

(4.6) min δ(ξ) = 0.144.

ξ

To evaluate the dimensional values of (4.4)-(4.6), we introduce the typical interior velocity
U and the interior horizontal length scale L, where

(4.7) U = O(10−2 m/s),

(4.8) L = O(106 m).

Moreover, the dimensional Sverdrup balance yields the estimate

 
τ  
(4.9) U LH = O = O 107 m3 /s ,
ρβ

where we consider the depth of the motion H = O(103 m), the wind-stress τ =
O(10−1 Pa), the sea water density ρ = O(103 kg/m3 ) and the planetary vorticity gra-
dient β = O(10−11 m−1 s−1 ).
On the whole, eqs. (4.4)-(4.9) show that, for ξ = 0.144, the model yields the maxi-
mum dimensional northward velocity, i.e. U vmax ≈ 0.36 m/s, the maximum dimensional
northward transport, i.e. U LHMmax ≈ 23 Sv and the minimum dimensional boundary
layer width, i.e. Lδmin ≈ 140 km. On the other hand, the orders of magnitude of the
observed values are

(4.10) v∗ = O(1 m/s) > U vmax , M∗ = O(3 Sv) > U LHMmax , δ∗ = O(50 km) < Lδmin ,

so the conclusion is that, even if every solution of the linear model with 0.144 ≤ ξ ≤ 0.315
underestimates both the northward velocity and the northward transport and overesti-
mates the western boundary layer width, the inner additional condition that minimizes
these differences is for

(4.11) ξ = 0.144.
354 F. CRISCIANI and F. CAVALLINI

5. – Connection between the inner and the western boundary conditions

In the range (3.1), the meridional velocity at the western boundary is strictly positive,
namely,

(5.1) Φ (0; ξ) > 0.

Then, in such range, setting

(5.2) Q21 (ξ) = Φ (0; ξ)/Φ (0; ξ)

and

(5.3) Q31 (ξ) = Φ (0; ξ)/Φ (0; ξ),

the identities

(5.4) Q21 (ξ)Φ (0; ξ) − Φ (0; ξ) = 0

and

(5.5) Q31 (ξ)Φ (0; ξ) − Φ (0; ξ) = 0

follow, respectively. Therefore, if the previous problem (2.15)-(2.18) is re-stated by sub-

stituting in place of (2.17) one of the following partial-slip conditions (the notation Φ(x)
is here restored because ξ disappears from the model if (2.17) is absent)

(5.6) λΦ (0) − Φ (0) = 0,

or

(5.7) λΦ (0) − Φ (0) = 0,

then the latter problem admits a unique solution provided that (5.6) can be written in
the form (5.4) or, alternatively, (5.7) can be written in the form (5.5). In case (5.6),
comparison with (5.4) shows that this happens if the equation

(5.8) Q21 (ξ) = λ

in the unknown ξ has a solution in the interval (3.1). In case (5.7), comparison with (5.5)
shows that the same condition involves the equation

(5.9) Q31 (ξ) = λ.

From fig. 3 we ascertain that Q21 (ξ) is a monotonically increasing function of its
argument, such that Q21 (0.144) = −8.57, while Q31 (ξ) is a monotonically decreasing
function and Q31 (0.144) = −4.71, so eqs. (5.8) and (5.9) have, at most, one solution. We
see that the partial-slip condition (5.6) is equivalent to (2.17) provided that λ ≥ −8.57,
while the partial-slip condition (5.7) is equivalent to (2.17) provided that λ ≤ −4.71.
INNER CONDITION FOR MUNK 355

Fig. 3. – Behaviour of the functions Q21 (ξ) (continuous line) and Q31 (ξ) (hatched line), defined
by (5.3) and (5.4), respectively. The upper panel refers to the full domain (3.1) while the lower
panel shows the details in the proximity of ξ = 0.144. These functions are monotonic and
diverge in the limit for ξ → 0.315. The longitude ξ̄ = 0.227, marked in the upper panel, is that
related to the free-slip condition, as shown in sect. 5.

Figure 3 also shows that (5.6) and (5.7) are reciprocally equivalent, in the sense that if
a certain solution is obtained by means of the boundary condition

(5.10) λa Φ (0) − Φ (0) = 0

for a suitable λa , then the same solution follows by substituting (5.10) with the condition

(5.11) λb Φ (0) − Φ (0) = 0,

where λb is such that

(5.12) Q−1 −1
21 (λa ) = Q31 (λb ),

and vice versa. In particular, because of the relations Q−1 −1

21 (−8.57) = Q31 (−4.71) =
0.144, both the boundary conditions

(5.13) 8.57Φ (0) + Φ (0) = 0

and

(5.14) 4.71Φ (0) + Φ (0) = 0

356 F. CRISCIANI and F. CAVALLINI

single out the same solution Φ(x; 0.144), whose derivative is shown in fig. 1.
The free-slip condition

(5.15) ∇2 ψ = 0 at x = 0,

with ψ given by (2.14), takes at the western boundary the form

(5.16) Φ (0) = 0.

¯ is
Equation (5.16) is met by (5.6) with λ = 0 and the related turning longitude, say ξ,
given by

(5.17) ξ¯ = Q−1
21 (0) = 0.227.

The super-slip boundary condition


(5.18) n̂ · ∇(∇2 ψ) = 0, at x = 0

with ψ given by (2.14) and n̂ = −î, takes the form

(5.19) π 2 Φ (0) − Φ (0) = 0.

Comparison of (5.19) with (5.7) leads to the identification

(5.20) λ = π2

but we know that no solution exists if, in (5.7), we set λ > −4.71. The conclusion is
that (5.18) cannot be applied to the present model. However, a previous investigation [4]
suggests that a suitable increase of the eddy viscosity coefficient ε modifies the shape of
Q31 (ξ) in such a way that the equation

(5.21) Q31 (ξ) = π 2

has a solution. For instance, if ε = 10−2 is taken in place of (2.5), then the range (3.1)
changes into

(5.22) 0.248 < ξ < 0.519

and one obtains

(5.23) Q31 (0.248) = 24.3(> π 2 ).

This example shows that condition (5.18) can be applied only in the presence of relatively
high turbulent viscosity, that is, it demands a relatively large boundary layer width, which
makes the related linear model further on far from reality.
INNER CONDITION FOR MUNK 357

6. – Additional conditions and relative vorticity

We reconsider the additional condition (5.6) under the hypothesis Φ (0) > 0, so only
the no-slip condition is excluded in the present discussion. Thus (5.6) implies

1 2
(6.1) (∇ ψ)x=0 > 0 (λ ≥ −8.57),
λ
whence the alternative

(6.2) λ > 0 and (∇2 ψ)x=0 > 0,

or

(6.3) λ < 0 and (∇2 ψ)x=0 < 0

follows.
The phenomenon described by (6.2) and (6.3) is due to a bowl of negative relative
vorticity which is in any case present in the western area of the fluid domain and whose
precise position depends on ξ and induces a consequent change of the sign in the param-
eter λ appearing in the additional boundary condition (5.6). For negative λ, the bowl

Fig. 4. – 3D plot representation of the relative vorticity ∇2 ψ of the model solution for increasing
values of ξ. Note the different positions of the bowl with negative relative vorticity: for instance,
for ξ = 0.15 and ξ = 0.20 (upper panels), the bowl crosses the western boundary so ∇2 ψ < 0 in
x = 0, while, for ξ = 0.25 and ξ = 0.30 (lower panels), the bowl is located on the r.h.s. of the
western boundary, so ∇2 ψ > 0 in x = 0.
358 F. CRISCIANI and F. CAVALLINI

Fig. 5. – Mid-basin (y = 1/2) profiles of the surfaces depicted in fig. 4. For increasing values
of ξ the minimum of (negative) relative vorticity moves eastward and, at the same time, the
relative vorticity at the western boundary increases from a markedly negative value (top left)
to a positive one (bottom right).

crosses the western boundary so, in particular, ∇2 ψ < 0 at the western boundary. On
the other hand, for positive λ, the bowl of negative vorticity is wholly located in the
basin interior; so, necessarily, ∇2 ψ > 0 at the western boundary. Various positions of
the bowl are reported in figs. 4 and 5.

7. – Dynamic stability of the model solutions

The time-dependent version of (2.1) is [2]

∂ 2 ∂Ψ
(7.1) ∇ Ψ+ = wE (x, y) + ε∇4 Ψ.
∂t ∂x

To analyse the stability in the energy norm of the solution (2.14) of problem (2.15),
(2.16), (2.18) and (5.6), the perturbed solution

(7.2) Ψ = ψ(x, y) + ϕ(x, y, t)

is substituted into (7.1) thus yielding the following evolution equation for the distur-
bance ϕ:

∂ 2 ∂ϕ
(7.3) ∇ ϕ+ = ε∇4 ϕ.
∂t ∂x

Linearity of (7.1) allows the decoupling of the basic state from the disturbance, as
eq. (7.3) shows.
The boundary conditions of ϕ are the no mass-flux across ∂D, i.e.

(7.4) ϕ(x, y, t) = 0 ∀(x, y) ∈ ∂D, ∀t,

INNER CONDITION FOR MUNK 359

the free-slip along C, i.e.

(7.5) ∇2 ϕ = 0 ∀(x, y) ∈ C, ∀t

and (5.6), which here becomes

∂ϕ ∂ 2 ϕ
(7.6) λ − = 0 at x = 0, ∀y ∈ [0, 1], ∀t.
∂x ∂x2

To obtain the time growth rate of the energy norm of the perturbation, defined by

 1/2
 
(7.7) E(t) ≡ ∇ϕ · ∇ϕdx dy ,
D

eq. (7.3) is multiplied by ϕ, each product is integrated on the fluid domain D with the
aid of (7.4) and the repeated use of the divergence theorem. The resulting equation is
 
dE 
(7.8) E =ε ∇2 ϕ∇ϕ · n̂ ds − ε (∇2 ϕ)2 dx dy.
dt ∂D D


If the free-slip or no-slip conditions, i.e. ∇2 ψ = 0 or ∇ϕ · n̂ = 0 respectively, are
applied throughout the boundary ∂D, then the circuit integral
 
(7.9) I= ∇2 ϕ∇ϕ · n̂ ds
∂D

is zero and the stability in the energy norm follows from the inequality

dE
(7.10) <0
dt

which results from (7.8). Indeed, because of Poincaré-Wirtinger’s inequality (e.g., [5])

(7.11) − (∇2 ϕ)2 dx dy < −αD E 2 ,
D

where αD = 2π 2 if D is given by (2.2), the r.h.s. of (7.8) can be bounded so that

dE
(7.12) < −εαD E (αD > 0).
dt

Time integration of (7.12) yields

(7.13) E(t) < E(0) exp[−εαD t]

and hence the asymptotic stability in the energy norm is derived.

360 F. CRISCIANI and F. CAVALLINI

In the present model, because of conditions (7.5) and (7.6), integral (7.9) is not zero
but, rather, it takes the form
 1  
∂ 2 ϕ ∂ϕ
(7.14) I=− dy.
0 ∂x2 ∂x x=0

Now, (7.6) implies

 2  
∂ϕ ∂ϕ ∂ 2 ϕ
(7.15) λ =
∂x x=0 ∂x ∂x2 x=0

and the substitution of (7.15) into (7.14) gives

 1  2
∂ϕ
(7.16) I = −λ dy.
0 ∂x x=0

Because of (7.16), eq. (7.8) becomes

 1  2 
1 d  
  2 ∂ϕ
(7.17) ∇ϕ = −ε λ dy + (∇2 ϕ)2 dx dy
2 dt 0 ∂x x=0 D

and we conclude from (7.17) that, if λ ≥ 0, then (7.10) holds true and the basic state is
stable. As (7.17) together with (7.11) imply

1 d  
  2
  2
 
∇ϕ < −ε (∇2 ϕ)2 dx dy < −εαD ∇ϕ ,
2 dt D

also the asymptotic stability immediately follows for non-negative λ.

Recalling moreover that (5.6) presupposes λ ≥ −8.57, we consider separately the
range

(7.18) −8.57 ≤ λ < 0

whose analysis is less immediate. By means of the Fourier expansion (recall (7.4))

(7.19) ϕ= amn (t) sin(mπx) sin(nπy)

m,n≥1

we evaluate
 1  2
∂ϕ π2
(7.20) dy = a2mn (t)m2 ,
0 ∂x x=0 2 m,n

 2 π4  2
(7.21) ∇2 ϕ dx dy = a2mn (t) m2 + n2 .
D 4 m,n
INNER CONDITION FOR MUNK 361

With the aid of (7.20) and (7.21), eq. (7.17) becomes

1 d  
  2 π2 π2  2 2
(7.22) ∇ϕ = −ε a2mn (t) λm2 + m + n2
2 dt 2 m,n
2

and the sign of the l.h.s. of (7.22) can be investigated in terms of the quantities

π2  2 2
(7.23) Cmn ≡ λm2 + m + n2
2

appearing on the r.h.s. of the same equation. Consider

(7.24) C11 = λ + 2π 2 .

First, trivially, C11 > 0 if and only if

(7.25) λ > −2π 2 .

Second, elimination of λ between (7.23) and (7.24) yields

  π2  2 2 π2  2 2
Cmn = C11 − 2π 2 m2 + m + n2 ≥ C11 + m − 1 ≥ C11 ,
2 2

that is

(7.26) Cmn ≥ C11 ∀m ≥ 1, n ≥ 1.

Thus, if (7.25) is true, (7.26) follows and (7.22) implies (7.10). On the whole, a fortiori,
stability holds in the full interval

(7.27) −8.57 < λ < +∞.

We are not able to prove the asymptotic stability of the solution of the linear model
in the interval −8.57 < λ < 0 because of the impossibility to prove that the term

  2 
1
∂ϕ  2
ε|λ| dy − ε ∇2 ϕ dx dy,
0 ∂x x=0 D


appearing on the r.h.s. of (7.17), is less than −ε D (∇2 ϕ)2 dx dy.
In conclusion, every model solution satisfying (2.11) is dynamically stable in the
energy norm (7.7) whatever the turning longitude ξ may be.
We stress that this conclusion is based on (5.6) rather than (5.7), so the former
should be preferred to the latter in spite of their equivalence, in the linear context, as
shown in sect. 5.
362 F. CRISCIANI and F. CAVALLINI

8. – Application to nonlinear Munk models

The method of the inner additional condition relies on the factorisation (2.14) of
the stream function. This possibility is due to the linearity of the governing vorticity
equation (2.1) and it is precluded if the nonlinear version

  ∂ψ
(8.1) RJ ψ, ∇2 ψ + = wE (x, y) + ε∇4 ψ (R > 0)
∂x

of (2.1) is taken into account. However, as boundary conditions (no mass-flux and
additional) are left unchanged even if the advection of relative vorticity is added in
the vorticity equation, then (5.6), written in terms of ψ rather than Φ, is still valid
when (8.1) is considered in place of (2.1). In other words, in the framework of (8.1), the
additional boundary condition at the western boundary that coincides with (5.6) in the
linear counterpart is
 
∂ψ ∂ 2 ψ
(8.2) λ − = 0.
∂x ∂x2 x=0

Further investigations on the solution of model (2.3), (2.9), (8.1) and (8.2) would
demand a numerical approach that goes beyond the scope of this note. In any case,
additional condition (8.2) plays an important role in the time-dependent energetics of
Munk’s models, irrespectively of the linear or nonlinear context. In fact, starting from
the time-dependent version of (8.1), that is

∂ 2   ∂ψ
(8.3) ∇ ψ + RJ ψ, ∇2 ψ + = wE (x, y) + ε∇4 ψ (R ≥ 0),
∂t ∂x

multiplication of (8.3) by ψ, and the subsequent integration on D with the use of (2.3)
and the divergence theorem, gives
   
d 1  
   2 
(8.4) ∇ψ · ∇ψ dx dy = u · τ dx dy − ε ∇2 ψ dx dy + ε ∇2 ψ ∇ψ · n̂ ds,
dt 2 D D D ∂D



where u = k̂ × ∇ψ is the fluid velocity.
Here is an authoritative comment on the last term of (8.4).
In general, the boundary integral of (8.4) is not sign determinant. If the flow satisfies

either the no-slip condition, for which ∇ψ · n̂ vanishes, or the slip condition, for which
∇2 ψ vanishes, the boundary term in the energy equation would also vanish, and the effect
of the horizontal friction would always be a drain of energy. It is certainly a desirable
feature of the dynamics for the lateral mixing definitely to lead to a decrease in energy.
However, if other conditions are used in (8.4), the boundary term does not vanish, and
the effect of lateral friction could actually add energy to the circulation [2]. In spite of
this eventuality, conditions (8.2) and (2.9) lead to the correct decrease in energy, as one
can check by resorting to a computation quite similar to that of sect. 7. In fact, because
of (2.9), we have
  1  
 ∂ψ ∂ 2 ψ
(8.5) ∇2 ψ ∇ψ · n̂ ds = − dy
∂D 0 ∂x ∂x2 x=0
INNER CONDITION FOR MUNK 363

and using (8.2) in (8.5) we obtain

 1    1  2
∂ψ ∂ 2 ψ ∂ψ
(8.6) − dy = −λ dy.
0 ∂x ∂x2 x=0 0 ∂x x=0

Equation (8.6) means that the energy sink of (8.4) can be written as
 
 2 2 
(8.7) −ε ∇ ψ dx dy + ε ∇2 ψ ∇ψ · n̂ ds =
D
   2
∂D
 
1
∂ψ  2 2
= −ε λ dy + ∇ ψ dx dy
0 ∂x x=0 D

and we know, from sect. 7, that, if

(7.25) λ > −2π 2 ,

then
  2 
1
∂ψ  2
(8.8) λ dy + ∇2 ψ dx dy ≥ 0.
0 ∂x x=0 D

Thus, we conclude that the enstrophy integral D (∇2 ψ)2 dx dy of (8.4) compensates,
under hypothesis (7.25), the contribution of the circuit integral (8.5) so that, on the
whole, lateral friction does actually drain energy.

9. – A remark on the forcing

The inner additional condition (2.10) can be applied not only with forcing (2.4), but
also in the more general case in which

(9.1) wE = wn sin(nπy), n ∈ N+
n

whatever the Fourier coefficients wn may be. In fact, in the linear context, the model
solution of the flow forced by (9.1) takes the form

(9.2) ψ= Φn (x) sin(nπy)

n

and condition (2.10) becomes

(9.3) Φn (ξ) sin(nπy) = 0.

n

1
The orthogonality relation 0
sin(nπy) sin(mπy)dy = 0 for n = m, applied to (9.3), yields

(9.4) Φn (ξ) = 0 ∀n ∈ N +

and (9.4) means that each Fourier coefficient Φn (x) of the full solution (9.2) is demanded
to satisfy the same inner condition (2.17).
364 F. CRISCIANI and F. CAVALLINI

10. – Conclusion

The results of the present investigation, concerning with Munk-like wind-driven ocean
circulation models, are summarised in the following list:
1) First of all, we stress the resort to the concept of inner condition in solving prob-
lem (2.1), (2.3), (2.10), (2.12), (2.13): this concept is rather unusual in the theory of
partial differential equations, but it is somehow similar to the idea underlying the Nico-
letti problem of ordinary differential equations (e.g., [6]).
2) In the framework of problem (2.15)-(2.18), all the physically admissible model so-
lutions are obtained by varying the turning longitude ξ, appearing in the inner additional
condition (2.17), in the range (3.1). This means that, whatever the additional condition
at the western boundary may be applied in place of (2.17), the related model solution
necessarily belongs to the above set of solutions.
3) Whatever the choice of ξ in (3.1) may be, the same model solution can be obtained
by substituting to (2.17) a suitable one-parameter partial-slip additional condition of the
kind (5.6) or (5.7).
4) Every additional condition of the kind (5.6) assures the dynamic stability, in the
energy norm, of the related model solution.
5) The same class of partial-slip conditions (5.6) leads to a self-consistent time-
dependent energy equation, which is valid also in the nonlinear context, thus positively
answering an until now open question in the framework of the homogeneous models of
the wind-driven ocean circulation.
∗ ∗ ∗
The authors gratefully acknowledge useful discussions on the subject of the present
note with Profs. J. Pedlosky (WHOI), A. Bellen and P. Omari (Dept. of Mathe-
matics and Informatics, University of Trieste).

REFERENCES

[1] Munk W. H., J. Meteorol., 7 (1950) 79.

[2] Pedlosky J., Ocean Circulation Theory (Springer) 1996.
[3] Crisciani F., Cavallini F. and Mosetti R., Nuovo Cimento C, 30 (2007) 1-10.
[4] Carnevale G. F., Cavallini F. and Crisciani F., J. Phys. Oceanogr., 31 (2001) 2489.
[5] Straughan B., The Energy Method, Stability and Nonlinear Convection (Springer-Verlag)
1992.
[6] Piccinini L. C., Stampacchia G. and Vidossich G., Ordinary Differential Equations in
n (Springer Verlag, New-York) 1984.