Download as pdf or txt
Download as pdf or txt
You are on page 1of 16

Theoret. Comput.

Fluid Dynamics (1998) 10: 459-474


Theoretical and Computational
Fluid Dynamics
© Springer-Verlag 1998

Boundary Vorticity Dynamics Since Lighthill’s 1963 Article:


Review and Development1

J.Z. Wu and J.M. Wu


The University of Tennessee Space Institute,
Tullahoma, TN 37388, U.S.A.

Communicated by M.Y. Hussaini

Received 14 November 1996 and accepted 14 March 1997

Abstract. In his well-known 1963 article: “Introduction. Boundary Layer Theory,” Lighthill explained for
the first time how to describe quantitatively the vorticity creation rate from a solid surface and how this
creation rate is dominated by the tangent pressure gradient. This was the first cornerstone of boundary
vorticity dynamics, which has now been developed to a complete theory on the vorticity creation from solid
or fluid boundaries and the reaction of the created vorticity to the boundaries. In this paper we present a
general formulation of boundary vorticity dynamics, briefly review the theoretical progress for Newtonian
fluid since 1963, examine the effect of variable viscosity and turbulence on the vorticity creation, and, of
many applications, exemplify the use of the theory in aerodynamic diagnosis and optimization.

1. Introduction

In his well-known article “Introduction. Boundary Layer Theory,” Lighthill (1963, referred to as L63 hence-
forth) claims that “although momentum considerations suffice to explain the local behaviour in a boundary
layer, vorticity considerations are needed to place the boundary layer correctly in the flow as a whole. It will
also be shown (surprisingly, perhaps) that they illuminate the detailed development of the boundary layer
. . . just as clearly as do momentum considerations . . . .” Therefore, Lighthill has in fact placed boundary
layer theory, including flow separation, in the realm of vorticity dynamics as a whole. The central ideas of
L63 has been further explored by Lighthill himself on several occasions, especially in his book on theoretical
fluid dynamics (Lighthill, 1986).
The present article addresses one of the wide-range issues pioneered by L63: the theory on vorticity
dynamics on boundaries. It has long been recognized that the most important source of vorticity is a solid
surface via viscosity and no-slip condition. Lamb (1932, Art. 328) has stated that “vortex-motion cannot
originate in the interior of a viscous liquid, but must be diffused inwards from the boundary.” He has even
noticed that (Art. 328a) in a two-dimensional viscous flow the change of circulation is given by
I
DΓ ∂ω
=ν ds,
Dt ∂n
where ν ∂ω/∂n is the vorticity diffusion flux across the circuit. However, the concept of vorticity creation
from the boundary had not become a theory until L63, which sets the first cornerstone of vorticity dynamics

1 This work was supported in part by ONR Grant W00014-96-0412 with Dr. E.P. Rood as program monitor.

459
460 J.Z. Wu and J.M. Wu

on boundaries, or boundary vorticity dynamics for short. After all, L63 explicitly applies the above flux,
or, more generally, σ ≡ νn · ∇ω, to a solid wall (n is its unit normal outward from the fluid) and iden-
tifies σ as the vorticity source strength (per unit area in per unit time). More crucially, L63 illustrates
how to determine σ by considering a two-dimensional flow over a stationary wall ∂B due to a pressure
gradient:
∂ω 1 ∂p
σ=ν = on ∂B. (1)
∂n ρ ∂x
Therefore, L63 is able to describe quantitatively the vorticity-creation process from the wall. As we now
understand, this boundary vorticity dynamics concerns vorticity creation from a boundary and the reaction
of the created vorticity to the boundary. It is the key to a full understanding of various vorticity–boundary
interactions.
Following L63, many works have been done to generalize (1) to three-dimensional flows over arbitrarily
curved moving walls, to compressible flows, to free-surface vorticity dynamics, and to the reaction of
the created vorticity to boundary surfaces. The theory has been used to analyze many flow phenomena
near the boundary, laminar or turbulent, to develop vorticity-based numerical methods, to acoustically
generated vortical waves, to aerodynamic diagnostic and optimization, and to vorticity-creation controls.
For a comprehensive review of these topics see Wu and Wu (1996).
This paper consists of a concise review of the fundamental formulation of boundary vorticity dynamics
and some new relevant results. Section 2 presents a general formulation of boundary vorticity dynamics
for an arbitrary continuum, which improves the earlier formulation by an unambiguous determination of
the measure of the boundary vorticity-creation rate (Section 2.1). It is also shown that, for the vorticity
right on a rigid boundary, the elegant formula due to Caswell (1967) is a special application of a general
triple decomposition of the strain-rate tensor. Section 3 first reviews the basic results for Newtonian fluid
with constant shear viscosity, and then analyzes the effect of variable shear viscosity due to strong heat
flux. Section 4 presents some further development, including the effect of turbulence , the temporal-spatial
variation of the boundary vorticity-creation rate, and a new example of applying the theory to aerodynamic
optimization. Conclusions are made in Section 5.

2. General Formulation

A general theoretical framework of boundary vorticity dynamics for any continuum has been given in Wu
and Wu (1996). However, it is worth restating here in a concise form, with an improved careful argument.
Let T be the stress tensor. For the divergence of T, i.e., the surface force vector, we make a Stokes–
Helmholtz decomposition.
∇ · T = −∇ϕ − ∇ × A, ∇ · A = 0,
such that ϕ and A characterize the two basic dynamic processes: the longitudinal compressing/expanding
process and the transverse shearing process, respectively. Then the Cauchy motion equations reads
Du 1
a≡ = f − (∇ϕ + ∇ + A). (2)
Dt ρ
We shall see that the entire boundary vorticity dynamics and its coupling with other dynamic processes rest
on this longitudinal-transverse decomposition.

2.1. The Measure of Boundary Vorticity-Creation Rate


For a general continuum, the proper definition of the vorticity-creation rate at the boundary is much less
apparent than that of the interior vorticity sources. What we know for sure is that the correct measure of
this rate must be closely related to the transverse process, and hence has its root at the vector potential A.
In the context of viscous incompressible flow, Lyman (1990) proposed an alternative to L63’s definition of
this boundary source. His argument was based on the integrated vorticity transport equation, which by (2)
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 461

reads
Z Z µ ¶ Z
D 1 1
ω dV = ω · ∇u + ∇ρ × ∇ϕ + ∇ × f dV − n × (∇ × A) dS.
Dt D D ρ2 ∂D ρ
This equation would tempt one to choose −n×(∇×A)/ρ as the measure of the boundary vorticity-creation
rate, as Lyman proposed. However, just like the identities
Z Z Z Z
1 1
∇ × f dV = n × f dS and 2
∇ρ × ∇ϕ dV = n × ∇ϕ dS
D ∂D D ρ ∂D ρ
do not mean that the nonconservative force and (generalized) baroclinicity are boundary vorticity sources,
defining a local creation rate on a boundary based on a closed-surface integral is easily misleading. Wu and
Wu (1993, 1996) have argued that Lyman’s proposal is not appropriate.
We now examine how a correct measure of the vorticity-creation rate at the boundary naturally follows in
the present general case. To this end we have to consider the local behavior of vorticity on an open surface.
The proper argument is nothing but a generalized version of Lamb (1932) based on the circulation argument.
It is well known that the vorticity creation inside the fluid or on the boundary is faithfully reflected by a
nonpreserving circulation along any closed material loop C (the Kelvin circulation formula):
I
D ΓC
= a · dx.
Dt C
Thus, by (2), I I I
D ΓC 1 1
= f · dx − ∇ϕ · dx − (dx × ∇) · A. (3)
Dt C C ρ C ρ
Here, the vorticity sources due to the nonconservative body force and generalized baroclinicity are easily
identified; they vanish if f and ∇ϕ/ρ have scalar potentials.
Now, to analyze the effect of the solenoidal vector potential A, consider a material tube with its side wall
along the A-direction and having inward unit normal n̂. It does not matter if the A-tube is not a vortex
tube. Then choose a loop C encompassing the tube, such that its normal is n̂ and its binormal b is along the
tube, i.e., A = Ab, see Figure 1. Let the tangent unit vector of C be t and dx = t ds, so that (t, n̂, b) form a
right-hand orthonormal basis. Thus, because A · n̂ = 0 on C,
µ ¶
1 ∂A ∂A ∂A ∂ n̂
(dx × ∇) · A = b · − n̂ · =b· +A .
ds ∂ n̂ ∂b ∂ n̂ ∂b
We can always make the tube sufficiently thin, so that the curvature of the A-lines measured by −b·(∂ n̂/∂b)
is much smaller than that of C, and hence the above second term can be ignored. Therefore, denoting by
n = −n̂ the outward unit normal, there is
I I
1 1
− (dx × ∇) · A = (n · ∇A) · b ds.
C ρ C ρ

Moreover, conceive a continuous shrinking of C to a point on a surface S bounded by C, such that b remains
the unit normal of S. This allows us to write Z
ΓC = ω · b dS,
S

Figure 1. A local (t, n̂, b) basis on a loop C encompassing a tube generated from the solenoidal A-field.
462 J.Z. Wu and J.M. Wu

and hence (3) becomes


Z I I I
D 1 1
ω · b dS = f · dx − ∇ϕ · dx + (n · ∇A) · b ds.
Dt S C C ρ C ρ
Inspecting the structure of the integrand on the left and that in the last term on the right, it is now clear
that the latter precisely measures how a local flux of A across C, the boundary of S, causes a change of
vorticity over S. Note that unlike interior sources, if C is entirely inside the fluid, the A-flux would merely
transport some vorticity across C but no new vorticity is created. This A-flux becomes a vorticity source
only at those points of C that are on the boundary ∂D. Therefore, we are naturally led to the local definition
of the vectorial source of vorticity at the boundary:
1 1
σ= n · ∇A ≡ [(n × ∇) × A − n × (∇ × A)] on ∂D. (4)
ρ ρ
This definition extends L63’s to the general case. We stress that in the above argument no closed surface
was used and no place can be found for Lyman’s (1990) alternate measure −n × (∇ × A)/ρ.

2.2. General Formulas of Boundary Vorticity and Its Flux


After the correct definition of boundary vorticity source σ is clarified, the general formula of σ is easily
obtained by applying the tangent components of (2) to the boundary ∂D:
1
σ = n × (a − f ) + [n × ∇ϕ + (n × ∇) × A] on ∂D. (5)
ρ
The right-hand side involves quantities on ∂D only. Although the definition of σ is solely in terms of A,
this source is physically determined by the boundary values of all terms in (2) as a result of force balance on
∂D.2 It contains the effects of boundary acceleration a, any body force f (even conservative), the tangent
gradient of ϕ, and certain tangent derivatives of A that appear in three dimensions only. Thus, we may write

σ = σa + σf + σϕ + σA. (6)

Note that σ A is missing from Lyman’s (1990) proposed definition of σ. Using suffix π to denote the tangent
components of a vector including the operator ∇, σ A can be cast to different forms by the following vector
identities:

(n × ∇) × A = n · [(∇A)T − (∇ · A)I] (7a)


= −n(∇π · A) + (∇π An + Aπ · K), (7b)

where T means transpose, I is the unit tensor, and K ≡ −∇π n is the surface curvature tensor.
In addition to the boundary vorticity flux, the vorticity right on the boundary is also a main concern of
boundary vorticity dynamics. However, their behaviors are very different. To examine the latter, we start from
a kinematic relation among the strain-rate tensor S, dilatation ϑ = ∇ · u, vorticity ω, and the surface-strain
rate tensor (Dinshington, 1965)

B = ϑI − (∇u)T with ∇ · B = 0, (8a)

of which the name is from the fact that for any fluid-surface element dS = n dS there is
D
dS = dS · B. (8b)
Dt
Evidently, these quantities are related by an intrinsic triple decomposition

S = ϑI + Ω − B, (9)
2 The normal component of (2) on the boundary is a source of the ϕ field, which is beyond our concern here.
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 463

where Ωij = εijk ωk /2 is the antisymmetric spin tensor. In particular, on any fluid surface with unit normal
n, either inside the flow or on any boundary ∂D including interface, there is a general formula
1 D
n · S = ϑn + 12 ω × n − dS. (10a)
dS Dt
This formula clearly tells how n·S (for a Newtonian fluid it is proportional to viscous surface stress) consists
of three basic processes. It is especially useful on a flexible boundary or interface (see the next section).
On an arbitrarily moving rigid wall ∂B with angular velocity W and the adherence condition imposed, we
simply have n · B = W × n, so that the last term of (10a) can be absorbed into the second term by replacing
ω by the relative vorticity ω r ≡ ω − 2W :
n · S = ϑn + 12 ω r × n on ∂B. (10b)
We now cast (10b) to a more symmetric form. Consider a nonrotating wall first, where ∇π u = 0 and
n · B = 0, the latter implying that n · (∇u)T = ϑn due to (8a). Thus, there is
∇u = n(n · ∇u) = 2n(n · S) − ϑnn,
and hence
(∇u)T = 2(n · S)n − ϑnn.
Therefore, substituting (10b) with ω r = ω into the sum of ∇u and (∇u)T , we immediately obtain the
well-known formula due to Caswell (1967):
2S = 2ϑnn + (ω × n)n + n(ω + n) on translating ∂B, (11)
which is now a special application of (9).
For an arbitrary moving wall, a simple consideration of reference-frame transformation indicates that
in (11) it is merely necessary to replace ω by ω r (Wu and Wu, 1993). This extension can also be directly
derived by repeating the above procedure, as long as we notice that in this case (8a) gives
n · (∇u)T = ϑn − W × n;
and, from the general rigid-motion formula u = U 0 + W × x,
∇π u = −n × (n × ∇u) = −n × I(W · n) − (W × n)n,
in which n × I is an antisymmetric tensor and does not contribute to ∇π u + (∇π u)T .
The Caswell formula has some unique conveniences. For example, the eigenvectors of S on ∂B can
be very easily inferred from (11). In particular, for an incompressible flow the stretching and shrinking
eigenvectors always have an inclination of 45° and 135° to the local flow direction, respectively, while the
intermediate eigenvector is always along the direction of ω r with neither stretching nor shrinking (Wu and
Wu, 1993).

2.3. Total Force on a Closed Boundary in Terms of Vorticity Flux


An interesting fact, that the boundary vorticity flux σ (more precisely, ρσ) alone is responsible for the total
force and moment acting on a closed boundary, was discovered by Wu (1987). Thus, the action of a boundary
to vorticity creation and the reaction of created vorticity to the boundary is intrinsically unified into a whole,
which further indicates the far-reaching impact of L63. The general formulation of this reaction is as follows.
The Stokes–Helmholtz decomposition of ∇ · T implies a corresponding decomposition of the normal and
tangent stresses on a closed boundary ∂D, which gives the total force acting on ∂D:
Z Z
F =− n · T dS = (ϕn + n × A) dS. (12)
∂D ∂D

We remark that (12) remains the same if we add any divergenceless tensor to T. Thus, as we shall see later,
locally some more stress may appear, but the integrated effect remains the same.
464 J.Z. Wu and J.M. Wu

The above remark is merely one example that a part of the integrand in a closed-surface integral can be
removed without affecting the result. The force and moment formulas in terms of σ is nothing but a further
use of this elementary mathematical fact. Physically, a constant ϕ–A-field does not cause a force at all even
though it is responsible for local stresses. For total force, what really matters is only the tangent variation
of ϕ and A over ∂D. Thus, it should be more reasonable to express F directly in terms of these tangent
variations. This is achieved by a pair of integral identities based on the generalized Stokes theorem: for a
closed surface ∂D in a d-dimensional space,
Z Z
1
ϕn dS = − x × (n × ∇ϕ) dS, d = 2, 3, (13a)
∂D d ∂D
Z Z
n × A dS = − x × [(n × ∇) × A] dS, d = 3 only. (13b)
∂D ∂D

Compare the right-hand side of (13a,b) with (5), it immediately follows that if we use the notation of (6),
i.e.,
1 1
σ ϕ ≡ n × ∇ϕ, σ A ≡ (n × ∇) × A, (14a,b)
ρ ρ
then in three dimensions (14) is cast to
Z
F =− ρx × ( 12 σ ϕ + σ A ) dS. (15)
∂D

A similar formula for total moment exists based on the same mathematics (Wu and Wu, 1993, 1996) and is
not repeated here.
The advantage of (15) over the traditional formula (12) is twofold. Mathematically, a great part of the
integrand of the latter, which makes no contribution to the total force, is automatically canceled. The integrand
of (15) is therefore highly localized on only a few key areas of ∂D where σ ϕ and σ A have strong peaks. In
this sense we say that (15) is irreducible.3
Physically, (15) establishes a direct connection of the vorticity-creation rate and the total force. For an
incompressible flow, it has the same physical background as the well-known formula
Z Z
ρ D ρ D
F =− x × ω dV + x × (n × u) dS, (16)
d − 1 Dt D d − 1 Dt ∂D
but (15) explores one step deeper. In fact, for an incompressible flow (15) can be derived from (16) by
shifting D/Dt into the integral and then tracing where the vorticity in (16) comes from (Wu, 1987).

3. Newtonian Fluid: Solid Wall and Free Surface

3.1. Constant Shear Viscosity


For a Newtonian fluid with µ, λ as the first and second viscosities, from (9) we have an intrinsic triple
decomposition of the stress tensor:
T = −ΠI + 2µΩ − 2µB, Π = p − (λ + 2µ)ϑ.

Thus, for constant µ, the divergenceless B disappears from the motion equation and we have a natural
Stokes–Helmholtz decomposition (Truesdell, 1954)
∇ · T = −∇Π − µ∇ × ω.

3 Of course the integrand of (15) could be further projected onto normal and tangent directions, and (13a,b) applied again. This

would lead to a more compact integrand and the procedure could be repeated. However, we see no practical advantage nor new physical
implication of doing so.
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 465

Evidently, (4) reduces to Lighthill’s definition, and (5) becomes

1 µ
σ = n × (a − f ) + n × ∇Π + ν(n × ∇) × ω on ∂D, ν= . (17)
ρ ρ

It took three decades to go from L63’s equation (1) to (17). Morton (1984) added the contribution of wall
acceleration a; Wu (1986) studied the three-dimensional effect ν(n × ∇) × ω (which was already known
to Lighthill in 1963 but no detail was given); Wu et al. (1988b) and Wu and Wu (1993) extended the result
to compressible flows; and, finally, Hornung (1989, 1990) studied the general form of σ independent of
constitutive structure, which motivated the formulation of Section 2.
Similar progress took place for vorticity creation from a free surface. Lugt (1987) was the first to notice
that one a two-dimensional steady free surface S there exists an analogy of (1):

∂H p
σ=− on S, H = 12 |u|2 + + gz.
∂s ρ

Wu (1995) gave a general incompressible theory of boundary vorticity dynamics on a fluid–fluid interface
including free surface, to which (17) equally applies. The dominating effect of free-surface acceleration a on
σ was emphasized by Rood (1994a,b; 1995) for unsteady surface motion. Indeed, in contrast to a solid wall,
where the tangent pressure gradient plays a key role in vorticity creation, on a free surface with negligible
surface tension this gradient is only O(Re−1 ) due to a weak boundary layer (Moore, 1963), and surface
acceleration becomes crucial.
Wu (1995) also showed that the tangent vorticity ω π right on the free surface is solely balanced by the
tangent components of the surface stress due to its strain, −2µn · B. Since by (8a) n · B has the same form
of (7a,b), there is
ω π = −2n × (∇π un + uπ · K). (18)
This equation extends an early result of Longuet-Higgens (1953), that ω = 2U K on two-dimensional steady
free surface, to the general case. This is the only example we have met in which the “B part” of T makes a
crucial contribution to the stress balance on boundaries (another example appears in Section 4.1 but has a
weaker effect).
Various approximations of free-surface vorticity dynamics, including Lundgren’s (1988) boundary layer
theory and vortex sheet dynamics, were discussed by Wu (1995). At large Reynolds numbers, the fundamental
difference between a solid (rigid or flexible) bounary and interface or free surface is that the created vorticity
from the latter is much weaker; from (18) there is |ω π | = O(1). Thus, the last term of (17) can be ignored.
Moreover, most terms in (17) are actually dominated by inviscid effect, which has nothing to do with σ.
This is why the inviscid surface-wave theory has been so successful. Therefore, to estimate σ correctly, first
apply the tangential component of the Euler equation to the free surface, and subtract in from (17). Then it
is found that the viscous correction is merely O(Re−1/2 ), and as Re → ∞ there is σ → 0.
Finally, we mention an interesting alternative expression of σ in an incompressible flow. Project the Lamb
vector l = ω × u onto the solenoidal space (with and without curl) and its orthogonal complement (curl-free
with divergence), and denote the resulting transverse and longitudinal parts of l by l⊥ and lk , respectively.
The momentum equation is then accordingly split into

∂u
+ ν∇ × ω = −l⊥ , ∇h0 = −lk , (19a,b)
∂t
in which h0 is the stagnation enthalpy. Applying the tangential components of (19a) to a stationary solid
boundary ∂B, where l = 0 but generically l⊥ 6= 0, gives

σ = n × l⊥ + ν(n × ∇) × ω on ∂B. (20)

Therefore, the vorticity-creation rate is directly related to the boundary value of the transverse Lamb vector.
Equation (20) is obviously equivalent to the three-dimensional version of (1), since on a stationary wall we
simply have l⊥ = −lk = ∇p/ρ.
466 J.Z. Wu and J.M. Wu

3.2. Variable Shear Viscosity: Effect of Strong Heat Flux


To complete the theory for a Newtonian fluid, it remains to include the effect of the variable shear viscosity
µ. In a highly compressible flow over a solid boundary ∂B, the heat flux q = −κ∇T near ∂B may be strong
enough to cause an appreciable variation of µ. Then by virtue of (9) the Navier–Stokes equation becomes
∇ · T = −∇Π − ∇ × (µω) + ρηq · B, (21)
in which B is defined by (8a) and we have introduced a dimensional scalar function
P rSµ d log µ
η(ρ, T ) ≡ 2 with Sµ ≡ = O(1), (22a)
cp ρT d log T
where P r is the Prandtl number. For a perfect gas with the speed of sound c, there is
γ−1 γ−1
η=2 P rSµ = 2 P rSµ . (22b)
γp ρc2
The last term of (21) is proportional to the viscous resistance of isothermal surfaces to their deformation
(Wu and Wu, 1996), which makes the natural Stokes–Helmholtz decomposition of ∇ · T impossible.
Now, the difficulty caused by variable viscosity due to strong heat flux can be examined in two contexts.
On one hand, if it is wished to obtain a very general σ-based force formula for a closed flexible boundary
or interface with variable µ, then as a part of the surface stresses, the effect of ρηq · B must be reflected in
the potentials ϕ and A of (12) to (15). To this end a decomposition
ρηq · B = −∇ϕq − ∇ × Aq
could be formally assumed, such that in (2) there is
ϕ = p − (λ + 2µ)ϑ + ϕq , A = µω + Aq .
The expense of reaching this generality is that ϕq and Aq have to be computed via solving Poisson equations,
and that the definition of σ has to be modified to
1
σ= n · ∇(µω + Aq ).
ρ
Fortunately, this seldom happens in practice. Form (12) to (15) it is clear that the σ-based total force or
moment only needs n · T on the boundary rather than ∇ · T in the flow. Moreover, a strongly varying µ
may occur only on a solid wall, where as indicated by (10b) no difficulty exists. With A = µω r , the force
formula (15) equally applies.
On the other hand, when we use (5) to compute the vorticity-creation rate σ at a local wall point, we need
ρ−1 n × ∇ · T on ∂D, including the effect of ηq · B. However, in this case we can treat it as an extra force
due to viscous-thermodynamic interaction, and only its value on the wall is required. The vague inclusion of
a vector potential Aq in the definition of σ can thereby be avoided. Wu and Wu (1993) made some attempt
along this line, but their result can be greatly simplified. We show this below for a nonrotating wall ∂B.
Identity (7a) holds if n is replaced by an arbitrary vector, say q. Thus, from (8a) we have
n × (q · B) = (n × q) · ∇u − [(q × ∇)u] · n.
Since ∇π u = 0 on a nonrotating wall, only the normal gradient
∂u
= nϑ + ω × n
∂n
does not vanish. Therefore, we simply obtain n × (q · B) = (n × q)ϑ on ∂B and (17) is generalized to
1
σ = n · ∇(µω)
ρ
1 1
= n × (a − f − ηϑq) + n × ∇Π + (n × ∇) × (µω) on ∂B. (23)
ρ ρ
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 467

On an adiabatic wall n and q are orthogonal, and the effect of heat flux on σ should appear. In contrast, on
an isothermal wall the extra term vanishes. So it does if the wall is stationary and flow is steady, since then
ϑ = 0 (also, Π = p) on ∂B due to the continuity equation and adherence condition. For unsteady flow over
a stationary wall, from (22b) there is
γ − 1 ∂ρ
ηϑ = −2P rSµ on ∂B.
c2 ∂t
Suppose above the wall the flow goes across an unsteady viscous shock layer. The strong inverse pressure
gradient creates a new vorticity with direction opposite to the existing one in the boundary layer via n × ∇Π
(in which ϑ < 0) in (23), and may cause the flow to separate. Now, since η > 0 for gases and q is toward
upstream, the term −ηϑn × q will somewhat offset the effect of the inverse pressure gradient. It would
be of interest to see in what situation ηϑq is strong enough for this effect to be of some use in controlling
shock-induced unsteady flow separation.

4. Some Further Developments

The above two sections cover the fundamental theory of boundary vorticity dynamics for laminar flows
established since L63. In this section we make some further developments. First we extend the theory to
turbulent flow, including considering the effect of the mean fluctuating vorticity field and the instantaneous
variation of the boundary vorticity flux. Then we give a new convincing example to reemphasize the gread
potential of applying the theory to fluid-dynamic diagnosis and optimization.

4.1. Effect of Turbulence


The first laboratory measurement of σ beneath a turbulent boundary layer was recently made by
Andreopoulos and Agui (1996).4 They found a strong time fluctuation of σ due to that of pressure. A
strong fluctuation of σ on a 45° swept delta wing with varying angles of attack was also observed by
Andreopoulos (1996, private communication) and coworker, whose typical rms flux σ + (in terms of inner
scale) is shown in Figure 2. Figure 3 gives the fluctuation of the pressure coefficient for comparison. At
the center line (z/s = 0) the rms σ + becomes strong for α > 15°, while at the location of the main vortex

Figure 2. The rms boundary vorticity flux on a 45° swept delta wing at different angles of attack. x/c = 0.7. Scaled to the friction
velocity and kinematic viscosity. From Andreopoulos (1996, submitted for publication).

4 Actually, the authors obtained the contribution of fluctuating tangent pressure gradient to the vorticity flux. The tangent derivatives

of boundary vorticity, or, equivalently, of skin friction which contributes to the last term of (17), is hard to measure (Andreopoulos,
1996, private communication). Nevertheless, the major part of the vorticity flux below a two-dimensional turbulent boundary layer is
due to the pressure gradient.
468 J.Z. Wu and J.M. Wu

Figure 3. The rms wall pressure-coefficient distribution on a 45° swept delta wing at different angles of attack. x/c = 0.7. From
Andreopoulos (1996, submitted for publication).

(z/s = 0.5) it is strong at small α but then reduces. The same level of fluctuating ∂ωx+ /∂y + and ∂ωz+ /∂y +
implies a 45° inclination of the created fluctuating vorticity with the x axis, as confirmed by their pdf
measurement. It is interesting that the mean σ is only one-fifth to one-tenth of its rms value. Progress of this
type could shed new light on our understanding of the formation of near-wall turbulent coherent structures.
We now ask whether, in turbulent flows, some extra term due to the Reynolds stress may appear in the
mean value of σ. If the flow is incompressible, on a rigid boundary with known wall acceleration, (17) is
linearized and hence there is no extra term. This can also be seen from (20): since l⊥ = ∇p/p on a stationary
wall, the mean fluctuating l⊥ must vanish as does the mean fluctuating pressure. Therefore, as confirmed
by Andreopoulos and Agui (1996), when there is no mean pressure gradient the mean σ is zero. However,
if there is a strong heat flux such that µ is variable, some weak extra terms may be caused by ηϑq and µω.
On an incompressible free surface with ρ = 1, however, the situation is different due to the nonlinearity
of surface acceleration. The simplest way of seeing this is taking the ensemble average of (19a), which leads
to a turbulent force
f turb = −∇ · (uu) = −l − ∇K, K = 12 |u|2 , (24)
indicating that the turbulent force is dominated by the mean fluctuating Lamb vector. (In (24)–(26) lowercase
and uppercase letters denote fluctuating and mean quantities, respectively.) Now, with this turbulent force in
the mean momentum equation, when applying the equation to the free surface there must be a nonnegligible
mean fluctuating vorticity-creation rate n × (l + ∇K).
This being the case, it is of interest to look briefly at the behavior of l, which depends on the level of
turbulence modeling. At the level of second-order closure, instead of the transport equation of a full Reynolds
stress tensor, we now only need to observe the transport equation of l. This equation was derived by Wu et
al. (1997):
∂l X3
− ν∇2 l = Qk , (25)
∂t k=1

where Qk are sources of l that need to be modeled:

† †
Q1 = −∇ · (uL⊥ + L⊥ u), L† ≡ ω × U + Ω × u, (26a)
Q2 = −∇ · (ul⊥ + l⊥ u), (26b)
Q3 = −2νω ,i × u,i . (26c)

It is remarkable that l is governed by a simple linear diffusion equation, not advected nor stretched by
the mean flow; and, in source terms the troublesome pressure-strain correlation in the full Reynolds stress
equation no longer appears.
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 469

4.2. Spatial-Temporal Variation of Boundary Vorticity Flux


Motivated by the observed strong spatial-temporal fluctuation of σ in a turbulent flow, Andreopoulos and
Agui (1996) derived a transport equation for σ and measured its terms in a two-dimensional turbulent
boundary layer. Their equation, however, was not applied to the wall and the σ therein simply represents
the diffusion rate of vorticity across surfaces with constant heights from the wall. Once the surfaces fall
onto the wall, as we show below, the equation can be simplified and thereby reveals physical mechanisms
responsible for the σ-fluctuation. This is especially important for wall-bounded turbulence.
Consider an incompressible flow with ρ = 1 over an arbitrarily curved and nonrotating solid boundary
∂B with n · ω = 0 thereon. Let n be continued into the interior of the fluid. Applying the operator ν ∂/∂n
to the vorticity transport equation yields (Andreopoulos and Agui, 1996)
∂σ ∂u ∂S
+ u · ∇σ − ν∇2 σ = −ν · ∇ω + ν · ω + S · σ. (27)
∂t ∂n ∂n
By further applying (27) to ∂B and denoting skin friction by τ , we find (for the derivation see the Appendix)
µ ¶

− ν∇2 σ = (τ · ∇π )ω − (ω · ∇π )τ + τ (∇π · ω) on ∂B. (28)
∂t
In principle, ∂ 2 σ/∂n2 can be cast to tangent derivatives of p, τ , or ω (Wu et al., 1988a), so that all terms
in (28) involve quantities on ∂B only. We do not go into detail here.
To make the result more appealing, we use the identity
∇π × (ω × τ ) = (τ · ∇π )ω − (ω · ∇π )τ + ω(∇π · τ ) − τ (∇π · ω),
where, since ω × τ = νω 2 n (see (A.1) of the Appendix) and ∇π × n = 0,
∇π × (ω × τ ) = −n × ∇π (νω 2 ) = −n × ∇(νω 2 ).
Therefore, (28) becomes
µ ¶

− ν∇ σ = −n × ∇(νω 2 ) − ω(∇π · τ ) + 2τ (∇π · ω)
2
on ∂B. (29)
∂t
Each term in (29) has a clear physical meaning. The first term on the right-hand side of (29) is the tangent
gradient of enstrophy. The other two terms are appreciable only in the neighborhood of nodal and spiral
points, respectively, of the two-dimensional field τ = νn × ω (but not right at these points where τ and
ω vanish). In particular, as indicated by (A.7) of the Appendix, the spiral points are the source of normal
vorticity.
In a turbulent boundary layer, some observations can be made based on (29). First, the local ragions with
a high gradient of enstrophy on the wall, as well as the local regions near critical points, could strengthen
the spatial-temporal fluctuation of σ, which may in turn cause an even stronger fluctuation in the near-wall
vorticity field as has been observed in experiments.
Second, after taking an ensemble average of (29) and subtracting the mean part, there is
n × ∇(νω 2 ) = −ω(∇π · τ ) + 2τ (∇π · ω), (30)
where ω and τ now denote the fluctuating part. Thus, the high gradient of fluctuating enstrophy on the wall
is statistically associated with the occurence of critical points of the fluctuating τ -field.
Third, after using uτ and ν to recast various quantities to the inner scale, we have
∂ω +
σ+ = , (31a)
∂n+
then (29) becomes
µ ¶

−∇ +2
σ + = −n × ∇+ ω +2 − ω + (∇+π · τ + ) + 2τ + (∇+π · ω + ), (31b)
∂t+
where we have rescaled ∇π to make all rescaled quantities O(1). Therefore, at least along one tangent
direction (usually the transverse direction) the wavelength of σ-fluctuation is of the order of ν/uτ .
470 J.Z. Wu and J.M. Wu

4.3. Diagnosis and Optimization of Aerodynamic Configuration

The most valuable potential application of the irreducible force formula (15) is in the field of aerodynamic
diagnosis and optimization, since one can now concentrate on those few key local regions of the boundary
that have strong peaks of σ ϕ and σ A . Note that the boundary ∂D can also be taken as proper control surfaces.
In this case (13a,b) lead to a set of force-component formulas solely in terms of vorticity-relevant quantities
on a wake plane. These formulas naturally distinguish different physical constituents of the force, and have
been used to diagnose the sources of the lift and drag of a delta wing at a moderate angle of attack, based on
measured wake-plane data (J.M. Wu et al., 1996). However, the wake-plane data is an accumulated effect of
upstream flow, which still cannot pinpoint the key areas of the body surface ∂B that are most responsible for
the total force. Moreover, in diagnosis and optimization based on computational fluid dynamics, to obtain
wake-plane data first the whole flow field including the (p, τ ) distribution on ∂B, which already allows for
a more thorough diagnosis based on (15), must be solved.
To illustrate the use of (15), consider airfoil design in an incompressible flow. Since in two dimensions
(say, the x–z plane) (13b) is not available, we only estimate the pressure force F p on the airfoil. Then,
as a slight modification of (1), there is ρσp = ∂p/∂s, where ds is the arc element of the airfoil. The net
contribution of a unit surface element to the lift and drag due to pressure is

dFpz dFpx
= −ρxσp , = ρzσp . (32a,b)
ds ds

The origin of x can be arbitrarily chosen; for convenience assume it is on the chord line and for x < 0 we
have σp > 0 on the upper surface, and vice versa (Figure 4(a)). Then it is clear that to increase lift the peak
σp regions should be driven on the upper surface, toward the leading and trailing edges; but to reduce the
pressure drag they should be driven on the lower surface toward the origin, as long as no early separation is
caused. Intuitively, such a modification immediately favors an airfoil with a wide flat midportion on top and
heavily loaded at the rear (Figure 4(b)). This is exactly the supercritical-type that also has a good low-speed
lift/drag ratio.
With this qualitative picture in mind, we look at the σp distribution on an NACA-0012 airfoil at a low
angle of attack α (Figure 5). Our first impression is that the integrand of (15) is indeed highly localized.
Then, due to the strong suction peak near the leading edge, we see a waste in vorticity creation signified
by the unfavorable contribution to the lift of σp downstream of the suction peak. Thus, NACA 0012 is not
optimally designed.5 In contrast, the NACA 64410 airfoil, for example, does not have such a waste for
α = 0°–2°: Figure 6 shows that, due to a wide and mild suction on the upper surface, its σp at α = 0°
remains favorable except very close to the trailing edge. If this favorable σp can be enhanced, therefore,
an even better shape could result. It is surprising that a simple play with σp has led to an airfoil (Figure 7) with

Figure 4. The sign and location of vorticity flux due to pressure on airfoils: (a) a traditional airfoil, (b) an improved airfoil with higher
lift/drag ratio at low speed.

5 Wu and Wu (1996) reported that starting from such an airfoil and without using advanced optimization methods, only a small

improvement of lift was gained by modifying the σp distription.


Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 471

Figure 5. NACA 0012 airfoil: distribution of pressure and vortic- Figure 6. NACA 64410 airfoil: distribution of pressure and vor-
ity flux at α = 6°. ticity flux at α = 0°.

double lift at α = 0°. A full Navier–Stokes computation showed that the lift/drag ratio of the new airfoil
increases by 53% at α = 0° and by 46% at α = 2°.
While this example is still preliminary as no other practical constraint was imposed and no advanced
optimization method was incorporated, it nevertheless reveals the gread potential of σ-based diagnosis and
optimization. Further study on this subject is being pursued.

5. Conclusions

Since L63 identified vorticity source strength σ on the boundary and gave its first formula, a unified general
theory on determining σ and the total force and moment acting on the boundary in terms of σ has been
developed. It is applicable to rigid or flexible boundaries and interface or free surface, for both incompressible
and highly compressible viscous flows. In this paper we show that the theory can be best formulated based on
a general Stokes–Helmholtz decomposition of the surface force (per unit volume) ∇· T, and a corresponding
triple decomposition of the strain-rate tensor. Following Lamb, an argument is given that generalizes L63’s
definition of σ to arbitrary continuum. The theory for Newtonian fluid is reviewed, and the effect of variable
shear viscosity due to strong heat flux is clarified.
The general theory can be easily extended to turbulent flow. We found that the effect of turbulence on the
mean vorticity creation is important on a free surface. On the other hand, the spatial-temporal fluctuation of
σ on a solid wall is dominated by the tangent gradient of enstrophy on the wall as well as the critical-point
behavior of the skin-friction field.
472 J.Z. Wu and J.M. Wu

Figure 7. Improved airfoil based on NACA 64410: distribution of


pressure and vorticity flux at α = 0°.

Since L63 boundary vorticity dynamics have been applied to many problems of practical interest. New
evidence is presented to exemplify the great potential of σ-based force formula in aerodynamic optimization.

Acknowledgments

We are very grateful to Mr. Tian-Gui Zheng for computing the results of airfoil optimization, and to Professor
Yiannis Andreopoulos for providing us with his unpublished figures and valuable discussion.

Appendix. Derivation of (28)

First, since
∂u
τ = −ν = νn × ω on ∂B (A.1)
∂n
is the wall skin-friction, there is
∂u
−ν · ∇ω = (τ · ∇π )ω. (A.2)
∂n
Second, writing S = Ω + ∇uT gives
∂ ∂ ∂ui
ν Sij = 12 εijk σk + ν .
∂n ∂xj ∂n
Boundary Vorticity Dynamics Since Lighthill’s 1963 Article: Review and Development 473

Hence, because ω · n = 0,

∂S
ν · ω = 12 ω × σ − (ω · ∇π )τ on ∂B. (A.3)
∂n
Third, by using Caswell’s formula (11) there is

S · σ = 12 (ω × n)σn − 12 n(ω × σ) · n on ∂B. (A.4)

Substituting (A.2), (A.3), and (A.4) into (27) yields, on ∂B,


µ ¶

− ν∇2 σ = (τ · ∇π )ω − (ω · ∇π )τ + 12 (ω × σ)π + 12 (ω × n)σn . (A.5)
∂t

However, by (6) there is (Wu and Wu, 1993, 1996)

∂ω
σ≡ν = n × ∇p + νω · K + nσn on ∂B, (A.6)
∂n
where
σn = n · σ = −ν∇π · ω on ∂B. (A.7)
Thus, from (A.6) it follows that

ω × σ = nω · ∇p + νω × (ω · K) + (ω × n)σn ,

in which the first two terms are along the normal direction. Hence, by using (A.7) and (A.1), in (A.5) we
have
1
2 (ω × σ)π + 12 (ω × n)σn = (ω × n)σn = τ (∇π · ω),
from which (28) follows.

References

Andreopoulos, J., and Agui, J.H. (1996). Wall-vorticity flux dynamics in a two-dimensional turbulent boundary layer. J. Fluid Mech.
309, 45–84.
Caswell, B. (1967). Kinematics and stress on a surface at rest. Arch. Rational. Mech. Anal. 26, 385–399.
Dinshington, R.H. (1965). Rate of surface-strain tensor. Amer. J. Phys. 33, 827–831.
Hornung, H. (1989). Vorticity generation and transport. Australian Fluid Mechanics Conference, Melbourne, 1989.
Hornung, H. (1990). Vorticity generation and transport. Proc. 1990 AMS–SIAM Summer Seminar on Vortex Dynamics and Vortex
Methods, Seattle, WA, 1990.
Lamb, H. (1932). Hydrodynamics. Dover, New York.
Lighthill, M.J. (1963). Introduction: Boundary layer theory. In: Laminar Boundary Theory (L. Rosenhead, ed.). Oxford University Press,
Oxford, pp. 46–113.
Lighthill, M.J. (1986). An Informed Introduction to Theoretical Fluid Dynamics. Oxford University Press, Oxford.
Longuet-Higgens, M.S. (1953). Mass transport in water waves. Philos. Trans. Roy Soc. London Ser. A 245, 535–581.
Lugt, H.J. (1987). Local flow properties at a viscous free surface. Phys. Fluids 30, 3647–3652.
Lundgren, T.S. (1988). A free vortex method with weak viscous effect. In: Mathematic Aspects of Vortex Dynamics (R.E. Caflisch, ed.).
SIAM, Philadelphia, PA, pp. 68–79.
Lyman, F.A. (1990). Vorticity production at a solid boundary. Appl. Mech. Rev. 43, 157.
Moore, D.W. (1963). The boundary layer on a spherical gas bubble. J. Fluid Mech. 16, 161–176.
Morton, B.R. (1984). The generation and decay of vorticity. Geophys. Astrophys. Fluid Dynamics 28, 277–308.
Rood, E.P. (1994a). Interpreting vortex interactions with a free surface. J. Fluids Engrg. 116, 91–93.
Rood, E.P. (1994b). Myths, meth, and physics of free-surface vorticity, Proc. 12th U.S. National Congress on Applied Mechanics, Seattle,
WA, 1994.
Rood, E.P. (1995). Vorticity interactions with a free surface. In: Fluid Vortices (S.I. Green ed.). Kluwer, Dordrecht, pp. 687–730.
Truesdell, C. (1954). The Kinematics of Vorticity. Indiana University Press, Bloomington, IN.
Wu, J.M., Ondrusek, B., and Wu, J.Z. (1996). Exact force diagnostics of vehicles based on wake-plan data. AIAA Paper 96-0559.
474 J.Z. Wu and J.M. Wu

Wu, J.Z. (1986). The generation of vorticity by body surfaces and its dissipation. Acta Aerodyn. Sinica 4, 168–176.
Wu, J.Z. (1987). The force on moving bodies by vorticity field. Acta Aerodyn. Sinica 5, 22–30.
Wu, J.Z. (1995). A theory of three-dimensional interfacial vorticity dynamics. Phys. Fluids 7, 2375–2395.
Wu, J.Z., and Wu, J.M. (1993). Interaction between a solid surface and a viscous compressible flow field. J. Fluid Mech. 254, 183–211.
Wu, J.Z., and Wu, J.M. (1996). Vorticity dynamics on boundaries. Adv. in Appl. Mech. 32, 119–275.
Wu, J.Z., Gu, J.W., and Wu, J.M. (1988a). Steady three-dimensional fluid-particle separation from arbitrary smooth surface and formation
of free vortex layers. Z. Flugwiss. Weltraumforsch. 12, 89–98.
Wu, J.Z., Wu, J.M., and Wu, C.J. (1988b). A viscous compressible theory on the interaction between moving bodies and flow field in
the (ω , ϑ) framework. Fluid Dynamics Res. 3, 203–208.
Wu, J.Z., Lu, X.Y., Zhou, Y., and Fan, M. (1997). Transport of Lamb vector in turbulence. Submitted for publication.

You might also like