Scribd
Upload
14 views5 pages

Spectral Methods in Fluid Dynamics

The document discusses spectral methods in the context of fluid dynamics, highlighting their application in high-resolution simulations of turbulence. An erratum is noted regarding the reproduction of certain figures and pages in non-final form, with corrections provided. The authors explore various polynomial approximations and their error estimates in Sobolev spaces, emphasizing the importance of these methods in numerical analysis and boundary-value problems.

Uploaded by

nyemb
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
14 views5 pages

Spectral Methods in Fluid Dynamics

The document discusses spectral methods in the context of fluid dynamics, highlighting their application in high-resolution simulations of turbulence. An erratum is noted regarding the reproduction of certain figures and pages in non-final form, with corrections provided. The authors explore various polynomial approximations and their error estimates in Sobolev spaces, emphasizing the importance of these methods in numerical analysis and boundary-value problems.

Uploaded by

nyemb
Copyright
© All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd

See discussions, stats, and author profiles for this publication at: [Link]

net/publication/45627396

Spectral Methods

Article · January 2006


Source: OAI

CITATIONS READS
25 4,911

4 authors, including:

Mohammed Yousuff Hussaini Alfio Quarteroni


Florida State University Politecnico di Milano
474 PUBLICATIONS 26,486 CITATIONS 1,026 PUBLICATIONS 45,606 CITATIONS

SEE PROFILE SEE PROFILE

All content following this page was uploaded by Alfio Quarteroni on 10 April 2014.

The user has requested enhancement of the downloaded file.


Erratum

Spectral Methods
Fundamentals in Single Domains

C. Canuto · M.Y. Hussaini · A. Quarteroni · T.A. Zang

© Springer-Verlag Berlin Heidelberg 2006

Due to a technical error the caption of Figure 1.6 on page 29 and


the content of pages 311 and 312 were reproduced in non-final form.
Please find the corrected pages below. On pages 311 and 312 the
changes are highlighted in red.
1.3 Three-Dimensional Applications in Fluids: A Look Ahead 29

many results obtained from their high-resolution simulations was convincing


evidence that the scaled energy spectrum (where the wavenumber is scaled by
the inverse of the Kolmogorov length scale η = (ν 3 /¯
)1/4 , with ν the viscosity
and ¯ the average dissipation rate) is not the classical Kolmogorov result of
k−5/3 , but rather k−m with m  5/3 − 0.10.

Fig. 1.6. Direct numerical simulation of incompressible isotropic turbulence on


a 20483 grid by Y. Kaneda and T. Ishihara (2006): High-Resolution Direct Numer-
ical Simulation of Turbulence. Journal of Turbulence 7(20), 1–17. The figure shows
the regions of intense vorticity in a subdomain with 1/4 the length in each coordi-
nate direction of the full domain [Reprinted with kind permission by the authors
and the publisher Taylor & Francis Ltd., [Link]

Rogallo (1977) developed a transformation that permits Fourier spectral


methods to be used for homogeneous turbulence flows, such as flows with
uniform shear. Blaisdell, Mansour and Reynolds (1993) used the extension of
this transformation to the compressible case to simulate compressible, homoge-
5.7 Other Polynomial Approximations 311
⎛ ⎞1/2
m
vHwm (R + ) =⎝ v (j) 2L2 (R w +)
⎠ .
j=0

A related family of weighted Sobolev spaces is useful, namely,


 
m
Hw;α (R+ ) = v ∈ L2w (R+ ) | (1 + x)α/2 v ∈ Hwm (R+ ) , m≥0,
(5.7.3)

equipped with the natural norm vHw;α m (R ) = (1 + x)


+
α/2
vHwm (R + ) .
For each u ∈ Lw (R+ ), let PN u ∈ PN be the truncation of its Laguerre
2

series, i.e., the orthogonal projection of u upon PN with respect to the inner
product of L2w (R+ ):

(u − PN u) φ e−x dx = 0 for all φ ∈ PN .
R+

The following error estimate holds for any m ≥ 0 and 0 ≤ k ≤ m:


m
u − PN uHwk (R + ) ≤ CN k− 2 uHw;m
m (R ) .
+
(5.7.4)

For the orthogonal projection PN1 upon PN in the norm of Hw1 (R+ ), the
following estimate holds for m ≥ 1, 1 ≤ k ≤ m:

u − PN1 uHwk (R + ) ≤ CN k+ 2 − 2 uHw;m−1


1 m
m (R + ) ; (5.7.5)

the same result holds for the projection PN1,0 upon P0N (Guo and Shen (2000)).
Concerning interpolation, let us consider the N + 1 Gauss-Radau points
xj , j = 0, . . . , N , where x0 = 0 and xj , for j = 1, . . . , N , are the zeros

of lN +1 (x), the derivative of the (N + 1)-th Laguerre polynomial. For each
continuous function u on R+ , let IN u ∈ PN be the interpolant of u at the
points xj . Then, for any integer m ≥ 1, 0 ≤ k ≤ m and 0 <  < 1, one has
1 m
u − IN uHwk (R + ) ≤ C N k+ 2 +− 2 uHw;m
m (R )
+
(5.7.6)

(see Xu and Guo (2002), where additional approximation results can be


found). The result stems from the error analysis given by Mastroianni and
Monegato (1997) in the family of norms (r ≥ 0 real)
∞ 1/2

vHw;∗
r (R ) =
+
(1 + k) r
v̂k2 ,
k=0

(0)
where v̂k = (v, lk )L2w (R + ) are the Laguerre coefficients of v. For such norms,
one has vHw;∗
r (R ) ≤ cvH r (R ) for any integer r. Examples of applica-
+ w;r +
tions to spectral Laguerre discretizations of boundary-value problems in R+
are provided in the above references. Usually, an appropriate change of
312 5. Global Approximation Results

unknown function is needed to cast the differential problem into the correct
functional setting based on Laguerre-weighted Sobolev spaces.
Hermite approximations can be studied in a similar manner. The basic
weighted space L2w (R) involves the norm
 1/2
v 2 (x)e−x dx
2
vL2w (R ) = .
R

The Sobolev spaces Hwm (R) are defined as above, with respect to this norm.
The L2w -orthogonal projection operator PN upon PN satisfies the estimate

u − PN uHwk (R ) ≤ CN 2 − 2 uHwm (R )
k m
(5.7.7)

for all m ≥ 0 and 0 ≤ k ≤ m (Guo (1999)). Interestingly, all Hw -orthogonal


projection operators PN upon PN , for  ≥ 0, coincide with PN , due to prop-
erty (2.6.12) of Hermite polynomials. For the interpolation operator IN at
the Hermite-Gauss nodes in R, Guo and Xu (2000) proved the estimate

u − IN uHwk (R ) ≤ CN 3 + 2 − 2 uHwm (R ) ,
1 k m
(5.7.8)

for m ≥ 1 and 0 ≤ k ≤ m.
When dealing with the unbounded intervals R+ and R, an alternative to
polynomials as approximating functions is given by functions that are the
product of a polynomial times the natural weight for the interval. Thus, one
uses the Laguerre functions ψ(x) = φ(x)e−x in R+ or the Hermite functions
ψ(x) = φ(x)e−x in R, where φ is any polynomial in PN . The behavior at
2

infinity of the function to be approximated may suggest such a choice. We


refer, e.g., to Funaro and Kavian (1990) and to Guo and Shen (2003) for the
corresponding approximation results and for applications.

5.8 Approximation in Cartesian-Product Domains

We shall now extend to several space dimensions some of the approximation


results we presented in the previous sections for a single spatial variable. The
three expansions of Fourier, Legendre and Chebyshev will be considered.
However, we will only be concerned with those Sobolev-type norms that are
most frequently applied to the convergence analysis of spectral methods.

5.8.1 Fourier Approximations

Let us consider the domain Ω = (0, 2π)d in Rd , for d = 2 or 3, and denote an


element of Rd by x = (x1 , . . . , xd ). The space L2 (Ω), as well as the Sobolev
spaces Hpm (Ω) of periodic functions, are defined in Appendix A (see (A.9.h)

View publication stats

Common questions

Powered by AI

Weighted Laguerre and Hermite polynomials are crucial for spectral methods in unbounded domains because they naturally incorporate the weight functions that match the behavior of the solutions at infinity, optimizing convergence and computational stability. These polynomials allow for effective representation and approximation of functions that decay at infinity, thereby enhancing numerical accuracy in such contexts .

Convergence properties in spectral methods differ significantly between periodic and non-periodic domains due to the nature of boundary conditions and the associated function spaces. In periodic domains, spectral expansions like Fourier naturally ensure continuity and smoothness across boundaries, leading to rapid convergence. However, in non-periodic domains, achieving such convergence is more complex, requiring weighted polynomials and transformations to accommodate boundary discontinuities .

Orthogonal projection operators improve spectral method approximations by providing a means to project functions onto polynomial spaces that respect the inner product norms specific to weighted Sobolev spaces. These operators help minimize the error between the actual function and its approximation by ensuring orthogonality, which contributes to achieving optimal convergence rates in these spaces .

Weighted Sobolev spaces allow for refined error estimates by incorporating weight functions that adapt to the behavior of the solutions at infinity or other boundary singularities. This results in error estimates that can better capture the asymptotic behavior of the solutions, leading to improved approximation in unbounded domains. For example, error estimates in Laguerre or Hermite polynomial approximations are directly related to these spaces, which accommodate the decay properties of certain functions .

Fourier, Legendre, and Chebyshev expansions face challenges such as handling complex boundary conditions and geometric intricacies in multi-dimensional spaces. Ensuring convergence and computational efficiency becomes more demanding as the dimensionality increases, as does managing the orthogonality properties across higher dimensional spaces. Additionally, these expansions must be adeptly applied to ensure smoothness and continuity in such contexts .

In high-resolution simulations, the classical Kolmogorov spectrum, which follows a power law of k^-5/3, is found to be modified to k^-m, with m slightly less than 5/3. This deviation results from the high precision and resolution of spectral simulations, which can capture more detailed aspects of the energy cascade and distribution across scales .

Gauss-Radau points are effective for interpolation in Laguerre polynomial approximations because they include the endpoint of the interval, thus providing accurate approximations near the boundary, which is critical for functions defined over semi-infinite intervals. Moreover, these points align with the zeros of the derivative of the higher-order Laguerre polynomial, optimizing the distribution of interpolation error across the domain .

The approximation power of polynomial functions, when combined with their natural weight, arises from their ability to model the asymptotic behavior of functions over unbounded domains efficiently. This product formulation leverages the intrinsic decay properties enforced by the weight functions, resulting in enhanced accuracy and stability in approximating functions that proliferate over infinite domains .

Rogallo's transformation allows Fourier spectral methods to address homogeneous turbulence flows including those with uniform shear by modifying the system to suit periodic computational domains. It permits the manipulation of turbulence dynamics through transformations that adapt periodic boundary conditions to homogeneous turbulence conditions, thus facilitating accurate simulations and analyses of turbulence phenomena .

Spectral methods offer superior accuracy for smooth problems due to their capability of achieving exponential convergence rates when the solution is smooth. This is in contrast to traditional finite difference methods, which typically have polynomial convergence rates. Additionally, spectral methods can efficiently handle complex geometries and high-dimensional problems by leveraging transformations, such as Fourier and Legendre expansions, and are thus particularly beneficial in simulations of turbulence .

You might also like