4
$\begingroup$

Marsden's Identity states that for every $\tau$ in $\mathbb{R }$:

$$(\cdot -\tau)^{k-1}=\sum_j\Psi_{j,k}(\tau)B_{j,k,t} \, ,$$

with $\Psi_{j,k}=(t_j-\tau)\times...\times(t_{j+k-1}-\tau)$.

Following de Boor's notation we have that $B_{j,k,t}$ stands for the $j-th$ B-spline of order $k$ defined over the knot vector $t$, i.e., $(t_{j+k}-t_j)\cdot [t_j,...,t_{j+k}](\cdot - x)_+^{k-1}$.

Also, define the space spanned by the B-splines as:

$\$_{k,t}:=\{\sum\alpha_jB_{j,k,t} : \alpha_j\in\mathbb{R}\}$

Technically, using Marsden's Identity I should be able to show that $\mathbb{P}_k$, the space of all polynomials of degree $<k$, is contained in $\$_{k,t}$, by putting $\alpha_j=\Psi_{j,k}$. But when I do that, I don't really see how this expression describes all possible polynomials of degree $k-1$.

Doesn't $(\cdot -\tau)^{k-1}$ represents all polynomials of degree $k-1$ where $\tau$ is a root with multiplicity $k-1$?

Am I missing something here?

$\endgroup$
2
  • $\begingroup$ What are $B_{j,k,t}$? Same as the $B_{i,n,t}$ in en.wikipedia.org/wiki/B-spline#Cardinal_B-spline ? $\endgroup$ Dec 7, 2018 at 16:17
  • $\begingroup$ I am using de Boor notation, $B_{j,k,t}$ stands for the $j-th$ B-spline of order $k$ defined over the knot vector $t$. The cardinal spline in the link is actually formed using a different normalization, in this case the definition is :$(t_{j+k}-t_j)\cdot [t_j,...,t_{j+k}](\cdot - x)_+^{k-1}$ $\endgroup$
    – Chaos
    Dec 7, 2018 at 16:22

1 Answer 1

2
$\begingroup$

I found the solution after some research, hence I'll post it here in case anyone have curiosity:

Marsden's Identity states that for all $\tau$ in $\mathbb{R}$ it holds that: $$(\cdot -\tau)^{k-1}=\sum_j\Psi_{j,k}(\tau)B_{j,k} \, ,$$

It's straightforward to show that $((\cdot-\tau_j)^{k-1}:j=1,...,k)$ , $\tau_1<...<\tau_k$, is a basis for the space $\Pi_{<k}$ (the space of polynomials of degree smaller than $k$).

Hence any polynomials of degree $<k$ can be written as a linear combination of the elements in the basis, i.e.:

$$\beta_1(\cdot-\tau_1)^{k-1}+...+\beta_k(\cdot-\tau_k)^{k-1},$$

By Marsden's Identity we have that the latter equals: $$\sum_j\beta_1\Psi_{j,k}(\tau_1)B_{j,k} +...+\sum_j\beta_k\Psi_{j,k}(\tau_k)B_{j,k}$$

Reordering it yelds to: $$\sum_j(\beta_1\Psi_{j,k}(\tau_1)+...+\beta_k\Psi_{j,k}(\tau_k))B_{j,k}$$

Letting $\alpha_j=(\beta_1\Psi_{j,k}(\tau_1)+...+\beta_k\Psi_{j,k}(\tau_k))$ we have shown that any polynomial in $\Pi_{<k}$ is contained in $\$_{j,k}$.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.