Download as pdf or txt
Download as pdf or txt
You are on page 1of 4

SPATIAL (TWO-POINT) CORRELATIONS

The concepts correlation functions, Taylor scales and integral scales


extend to (3D) space as well — consider say two points in space at
a distance r apart.

Consider measurements of velocity fluctuations along a line:


ρ(u(x), u(x + r)) = f (r)
a function of r alone if turbulence is homogeneous in x1 direction.

“Integral length scales” (a measure of size of large eddies) defined in


a manner analogous to integral time scales:
ˆ ∞
ρ(r) dr.
0
However the spatial separation between any 2 points in space is a
vector. In general the 3D “covariance tensor” is
Rij (r) = ⟨ui(x, t) uj (x + r, t)⟩
which is a function of the separation vector r that has 3 coordinate
components. For example, consider the two-point correlation
ραα (rβ ) = ⟨uα (x, t) uα (x + reβ , t)⟩/⟨u2α ⟩
(no sum over Greek subscripts). If α = β this is called a longitudinal
correlation. If α ̸= β it is a transverse correlation.

In isotropic turbulence, there are three statistically equivalent choices


of longitudinal correlations, and 6 for transverse.

“Structure functions” refer to moments of the relative velocity be-


tween two points in space.
1
SPECTRAL DESCRIPTION OF TURBULENCE

Can we decompose a turbulent “signal” into contributions from dif-


ferent scale sizes? How much energy is present at what scales?

WAVENUMBERS AND SPECTRA

Fourier decomposition of a fluctuating signal: e.g.


X
u(x) = û(k) exp(ιkx)
k

where û(k) is called a Fourier coefficient, ι = −1, and k is called
a wavenumber.

Note a complete waveform of each exp(ιkx) occurs over a period of


2π/k. Thus a wavenumber is like the reciprocal of a wavelength.
Low wavenumber ⇐⇒ large scales
High wavenumber ⇐⇒ small scales

̸ 0 over a wide range


If we have a wide range of scales, then û(k) =
of k included through the summation sign. We also refer to each
nonzero value of k as a “Fourier mode”.

Spectrum

Each Fourier coefficient is a complex number, with an amplitude


whose square is ûû∗ (where asterisk denotes complex conjugate).

Each Fourier mode for the velocity has a certain amount of “energy”
represented by 12 ûiû∗i . In most turbulent flows we can expect that
this “modal energy” is larger at lower k, smaller at higher k.
2
The energy of the turbulent signal can be considered as the sum of
the energies of a collection of Fourier modes which may represent con-
tributions from different wavenumbers (in space, as written above)
or frequencies (in time, see Appendix D of Pope).

A function that describes the contributions from different wavenum-


bers or frequencies to the energy of the signal is called a spectrum.

3D wavenumbers

Since turbulence is 3D, we need to account for dependence in all


three directions: i.e. a Fourier decomposition in 3D.

X
u(x) = û(k) exp(ιk · x)
k
where the “wavenumber vector” is defined by
k = (kx, ky , kz ) or k = (k1, k2, k3)
and, of course, x = (x1, x2, x3). In tensor notation we may write
X
ui(x) = ûi(k) exp(ιkj xj )
k

This relation gives velocity at (physical) space locations in terms of


Fourier coefficients. It can be inverted to give:
 3 X
1
ûi(k) = ui(x) exp(−ιkj xj )
2π x

The tasks of going from u(x) to û(k) or vice versa are referred to as
“taking a Fourier transform” (either forward or backward).
3
P
Note: the sign here implies summing over a finite number of
Fourier modes, which are all periodic in space. This describes tur-
bulence in a 3D periodic domain, with periodic boundary conditions
in each direction. This is convenient in DNS., but if the summation
is replaced by an integral then the periodic BC is not required.

Spatial correlations and spectrum tensor

A turbulent signal that has more content at low wavenumbers (dom-


inated by large scales) is likely to have a two-point correlation that
falls off more slowly with r, hence giving larger integral length scales.

The Fourier decomposition of Rij (r) is


˚
Rij (r) = exp(ιk · r) ϕij (k) dk
where ϕij (k) is called the energy spectrum tensor (see T&L Sec.
8.1, Pope Sec. 6.5.1, Ex. 6.18), and the triple integrals cover all
wavenumber modes in 3D.
 3 ˚
1
ϕij (r) = exp(ιk · r) Rij (x) dx

For the choice r = 0, we recover the single-point Reynolds stress
tensor ⟨uiuj ⟩ as ˚
Rij (0) = ϕij (k) dk
Likewise, by taking half the trace,
˚
1 1
2 ⟨ui ui ⟩ = 2 ϕii(k) dk

This suggests the function 12 ϕii(k) may be considered as a “spectral


density”. We will next consider how its functional form may depend
on the wavenumber magnitude |k|.
4

You might also like