Spectral Methods 143
3 ALGEBRAIC POLYNOMIAL
EXPANSION
When a boundary-value problem with nonperiodic data (of
Dirichlet, Neumann, or mixed type) has to be solved numer-
ically, the trigonometric expansion is no longer adequate to
guarantee high order of accuracy. Then, Jacobi orthogonal
polynomials are used to provide orthogonal bases for the
approximation space.
The finite dimensional space
P
N
is now made of algebraic
polynomials of degrees less than or equal to N.
The historical approach, inspired by the Fourier method,
aimed at expanding the approximate solution with respect
to a basis of orthogonal polynomials
u
N
(x) =
N
k=0
u
N,k
p
k
(x) (9)
where u
N,k
now represent the unknown frequency coeffi-
cients.
The matter of choice were the Chebyshev polynomi-
als, p
k
(x) = T
k
(x) = cos(kθ), θ = cos
−1
(x), −1 ≤ x ≤ 1,
owing to their analogy with trigonometric polynomials.
Since the Chebyshev basis does not necessarily match
the boundary requirement (as T
k
(1) = 1, T
k
(−1) = (−1)
k
,
∀k ≥ 0), one device consists of projecting the equation
residual on the reduced space
P
N−2
, enforcing the bound-
ary conditions afterward. For instance, for a Dirichlet
boundary-value problem like (2), where now = (−1, 1),
and Dirichlet boundary conditions u(−1) = u
−
, u(+1) =
u
+
, the solution (9) is required to satisfy the following
equations:
(Lu
N
− f, T
k
)
ω
= 0, 0 ≤ k ≤ N − 2
u
N
(−1) = u
−
,u
N
(+1) = u
+
(10)
This modal approach was termed the Lanczos–Tau
method. The symbol (u, v)
ω
=
1
−1
uv ω dx is the so-called
weighted scalar product with respect to the Chebyshev
weight function ω(x) = (1 −x
2
)
−1/2
, −1 <x<1. The
weighted scalar product is used, instead of the more tra-
ditional one (·, ·), in order to take advantage (to the highest
possible extent) of the Chebyshev orthogonality,
(T
k
,T
m
)
ω
= 0ifk = m
(T
0
,T
0
)
ω
= π
(T
k
,T
k
)
ω
=
π
2
∀k ≥ 1
When L has constant coefficients, the Lanczos–Tau prob-
lem (10) yields an algebraic system for the frequency coef-
ficients {u
N,k
} with a structured matrix for which efficient
diagonalization algorithms can be devised, a circumstance
that is also featured by the multidimensional problems that
are generated by tensorization.
However, this is not general enough, as this structure
gets lost for a more general kind of differential opera-
tors. A more flexible approach (in analogy with what was
done in the Fourier case) consists of adopting a nodal
representation of u
N
at selected Gauss–Lobatto nodes
x
j
= cos(πj/N), j = 0,...,N, then looking for a stan-
dard Galerkin approximation with integrals replaced by
Gauss–Lobatto integration:
(u, v)
N
=
N
j=0
α
j
u(x
j
)v(x
j
)(11)
where α
j
= (π/N) for j = 1,...,N − 1, α
0
= α
N
=
(π/2N) are the quadrature coefficients.
Should we still consider the baby Dirichlet boundary-
value problem for the operator L introduced in (5), the
corresponding discrete problem would read:
find u
N
∈ P
N
,u
N
(−1) = u
−
,u
N
(1) = u
+
, s.t.
(αu
N
,v
N
)
N
+ (βu
N
,v
N
)
N
+ (γu
N
,v
N
)
N
= (f, v
N
)
N
,
∀v
N
∈ P
0
N
(12)
where now
P
0
N
={v
N
∈ P
N
|v
N
(−1) = v
N
(1) = 0}.This
time, however, the expansion is made in terms of the nodal
Lagrangian basis at Gauss–Lobatto nodes, that is, using
instead of (9)
u
N
(x) =
N
j=0
u
j
ψ
j
(x)
where ψ
j
is the unique algebraic polynomial of degree N
such that ψ
j
(x
i
) = δ
ij
, ∀i, j = 0,...,N.
One may show that
ψ
j
(x) =
−1
N(N +1)
·
(1 − x
2
)
(x − x
j
)
·
L
N
(x)
L
N
(x
j
)
,j= 0,...,N
(13)
The same approximation framework can be set up by
replacing the Chebyshev polynomials with the Legendre
polynomials {L
k
,k= 0, 1,...}, which are orthogonal with
respect to the traditional L
2
-scalar product (otherwise said
with respect to the weight function ω = 1).
The approximate problem still reads like (12); however,
this time the nodes {x
j
} and the coefficients {α
j
} are those
of the (Legendre) Gauss–Lobatto integration.
The approach described above is named G-NI (Galerkin
with Numerical Integration). A similar G-NI approach can
be undertaken in several dimensions. For instance, consider