Bramble–Hilbert lemma explained

styleu

by a polynomial of order at most

stylem-1

in terms of derivatives of

styleu

of order

stylem

. Both the error of the approximation and the derivatives of

styleu

are measured by

styleLp

norms
on a bounded domain in

styleRn

. This is similar to classical numerical analysis, where, for example, the error of linear interpolation

styleu

can be bounded using the second derivative of

styleu

. However, the Bramble–Hilbert lemma applies in any number of dimensions, not just one dimension, and the approximation error and the derivatives of

styleu

are measured by more general norms involving averages, not just the maximum norm.

Additional assumptions on the domain are needed for the Bramble–Hilbert lemma to hold. Essentially, the boundary of the domain must be "reasonable". For example, domains that have a spike or a slit with zero angle at the tip are excluded. Lipschitz domains are reasonable enough, which includes convex domains and domains with continuously differentiable boundary.

The main use of the Bramble–Hilbert lemma is to prove bounds on the error of interpolation of function

styleu

by an operator that preserves polynomials of order up to

stylem-1

, in terms of the derivatives of

styleu

of order

stylem

. This is an essential step in error estimates for the finite element method. The Bramble–Hilbert lemma is applied there on the domain consisting of one element (or, in some superconvergence results, a small number of elements).

The one-dimensional case

Before stating the lemma in full generality, it is useful to look at some simple special cases. In one dimension and for a function

styleu

that has

stylem

derivatives on interval

style\left(a,b\right)

, the lemma reduces to
inf
v\inPm-1

l\Vertu\left(-v\left(

r\Vert
Lp\left(a,b\right)

\leqC\left(m,k\right)\left(b-a\right)m-kl\Vertu\left(

r\Vert
Lp\left(a,b\right)

foreachintegerk\leqmandextendedrealp\geq1,

where

stylePm-1

is the space of all polynomials of degree at most

stylem-1

and

f(k)

indicatesthe

k

th derivative of a function

f

.

In the case when

stylep=infty

,

stylem=2

,

stylek=0

, and

styleu

is twice differentiable, this means that there exists a polynomial

stylev

of degree one such that for all

stylex\in\left(a,b\right)

,

\left\vertu\left(x\right)-v\left(x\right)\right\vert\leqC\left(b-a\right)2\sup\left(\left\vertu\prime\prime\right\vert.

This inequality also follows from the well-known error estimate for linear interpolation by choosing

stylev

as the linear interpolant of

styleu

.

Statement of the lemma

Suppose

style\Omega

is a bounded domain in

styleRn

,

stylen\geq1

, with boundary

style\partial\Omega

and diameter

styled

.
k(\Omega)
styleW
p
is the Sobolev space of all function

styleu

on

style\Omega

with weak derivatives

styleD\alphau

of order

style\left\vert\alpha\right\vert

up to

stylek

in

styleLp(\Omega)

. Here,

style\alpha=\left(\alpha1,\alpha2,\ldots,\alphan\right)

is a multiindex,

style\left\vert\alpha\right\vert=

style\alpha1+\alpha2+ … +\alphan

and

styleD\alpha

denotes the derivative

style\alpha1

times with respect to

stylex1

,

style\alpha2

times with respect to

stylex2

, and so on. The Sobolev seminorm on
m(\Omega)
styleW
p
consists of the

styleLp

norms of the highest order derivatives,

\left\vertu\right\vert

m(\Omega)
W
p

=\left(\sum\left\vert\left\VertD\alpha

p\right)
u\right\Vert
Lp(\Omega)

1/pif1\leqp<infty

and

\left\vertu\right\vert

m
W(\Omega)
infty

=max\left\vert\left\VertD\alphau\right\Vert

Linfty(\Omega)

stylePk

is the space of all polynomials of order up to

stylek

on

styleRn

. Note that

styleD\alphav=0

for all

stylev\inPm-1

and

style\left\vert\alpha\right\vert=m

, so

style\left\vertu+v\right\vert

m(\Omega)
W
p
has the same value for any

stylev\inPm-1

.

Lemma (Bramble and Hilbert) Under additional assumptions on the domain

style\Omega

, specified below, there exists a constant

styleC=C\left(m,\Omega\right)

independent of

stylep

and

styleu

such that for any

styleu\in

m(\Omega)
W
p
there exists a polynomial

stylev\inPm-1

such that for all

stylek=0,\ldots,m,

\left\vertu-v\right\vert

k(\Omega)
W
p

\leqCdm-k\left\vertu\right\vert

m(\Omega)
W
p

.

The original result

The lemma was proved by Bramble and Hilbert [1] under the assumption that

style\Omega

satisfies the strong cone property; that is, there exists a finite open covering

style\left\{Oi\right\}

of

style\partial\Omega

and corresponding cones

style\{Ci\}

with vertices at the origin such that

stylex+Ci

is contained in

style\Omega

for any

stylex

style\in\Omega\capOi

.

The statement of the lemma here is a simple rewriting of the right-hand inequality stated in Theorem 1 in.[1] The actual statement in [1] is that the norm of the factorspace

m
styleW
p

(\Omega)/Pm-1

is equivalent to the
m
styleW
p

(\Omega)

seminorm. The
m
styleW
p

(\Omega)

norm is not the usual one but the terms are scaled with

styled

so that the right-hand inequality in the equivalence of the seminorms comes out exactly as in the statement here.

In the original result, the choice of the polynomial is not specified, and the value of constant and its dependence on the domain

style\Omega

cannot be determined from the proof.

A constructive form

An alternative result was given by Dupont and Scott [2] under the assumption that the domain

style\Omega

is star-shaped; that is, there exists a ball

styleB

such that for any

stylex\in\Omega

, the closed convex hull of

style\left\{x\right\}\cupB

is a subset of

style\Omega

. Suppose that

style\rhomax

is the supremum of the diameters of such balls. The ratio

style\gamma=d/\rhomax

is called the chunkiness of

style\Omega

.

Then the lemma holds with the constant

styleC=C\left(m,n,\gamma\right)

, that is, the constant depends on the domain

style\Omega

only through its chunkiness

style\gamma

and the dimension of the space

stylen

. In addition,

v

can be chosen as

v=Qmu

, where

styleQmu

is the averaged Taylor polynomial, defined as

Qmu=\intB

mu\left(
T
y

x\right)\psi\left(y\right)dx,

where

m
T
y

u\left(x\right)

m-1
=\sum\limits
k=0

\sum\limits\left\vert

1
\alpha!

D\alphau\left(y\right)\left(x-y\right)\alpha

is the Taylor polynomial of degree at most

stylem-1

of

styleu

centered at

styley

evaluated at

stylex

, and

style\psi\geq0

is a function that has derivatives of all orders, equals to zero outside of

styleB

, and such that

\intB\psidx=1.

Such function

style\psi

always exists.

For more details and a tutorial treatment, see the monograph by Brenner and Scott.[3] The result can be extended to the case when the domain

style\Omega

is the union of a finite number of star-shaped domains, which is slightly more general than the strong cone property, and other polynomial spaces than the space of all polynomials up to a given degree.[2]

Bound on linear functionals

This result follows immediately from the above lemma, and it is also called sometimes the Bramble–Hilbert lemma, for example by Ciarlet.[4] It is essentially Theorem 2 from.[1]

Lemma Suppose that

style\ell

is a continuous linear functional on
m
styleW
p

(\Omega)

and

style\left\Vert\ell\right\Vert

m
W(\Omega
\prime
)
p
its dual norm. Suppose that

style\ell\left(v\right)=0

for all

stylev\inPm-1

. Then there exists a constant

styleC=C\left(\Omega\right)

such that

\left\vert\ell\left(u\right)\right\vert\leqC\left\Vert\ell\right\Vert

m
W
\prime
(\Omega)
p

\left\vertu\right\vert

m
W(\Omega)
p

.

External links

Notes and References

  1. J. H. Bramble and S. R. Hilbert. Estimation of linear functionals on Sobolev spaces with application to Fourier transforms and spline interpolation. SIAM J. Numer. Anal., 7:112–124, 1970.
  2. Todd Dupont and Ridgway Scott. Polynomial approximation of functions in Sobolev spaces. Math. Comp., 34(150):441–463, 1980.
  3. [Susanne Brenner|Susanne C. Brenner]
  4. [Philippe G. Ciarlet]