I need to show the two following results. If true, it must be a simple proof but I do not seem to be able to make it work. Thank you in advance.

Consider a continuous symmetric bilinear form $B$ on a real Hilbert space $H$. Let $D$ be a closed subspace of that Hilbert space over which the form is coercive, i.e, there exists $\alpha >0 \in \mathbb{R}$ such that $B(d,d)\geq \alpha \|d\|^2$ for all $d\in D$. Take the new closed subspace of $H$ generated by $D$ and a one dimensional subspace. Assume $B$ is definite positive on the new subspace. Then I need to show that $B$ is also coercive on the new subspace.

It seems like the logical thing to do is to use the Lax-Milgram theorem but I am not exactly sure how to do this.


Here is a proof from scratch, in particular, without the Lax-Milgram.

Lemma. Let $\widetilde H$ be a real vector space with positive definite inner product (completeness not assumed). Suppose that $K$ is a subspace of $\widetilde H$ which is complete. Then any element $v\in \widetilde H$ can be written as $v_1+v_2$ where $v_1\in K$ and $v_2\in K^\perp$.

Proof. Let $v_1$ be the nearest element of $K$ to $v$; the standard proof of the existence of $v_1$ in a Hilbert space applies here as well, because it uses only the completeness of $K$ (e.g., Theorem I.2.5 in A Course in Functional Analysis by Conway). Let $v_2=v-v_1$. If there is $u\in K$ such that $\langle u,v_2\rangle>0$, then $$\|v-(v_1+tu)\|^2=\|v_2- tu\|^2 = \|v_2\|^2 -2t\langle u,v_2\rangle +O(t^2)<\|v_2\|^2$$ for small $t>0$, contradicting the fact that $v_1$ is the nearest element of $K$ to $v$. $\Box$

Apply the lemma with $K=D$ and $\widetilde H$ being the space generated by $D$ and another vector; the inner product is given by $B$. (Note that $D$ is complete with respect to $B$-inner product, because of coercitivity). The lemma provides a nonzero vector $v\in \widetilde H$ such that $v$ is $B$-orthogonal to $D$. Every element of $\widetilde H$ can be written as $d+tv$ for some $d\in D$ and $t\in \mathbb R$. We have $$ B(d+tv,d+tv) = B(d,d)+t^2B(v,v) \tag1$$ and $$\|d+tv\|^2 \le 2 \|d\|^2 +2t^2\|v\|^2\tag2$$ Let $\gamma$ be the smallest of the numbers $\alpha/2$ and $\frac12 \|v\|^2/ B(v,v)$. Then $$ B(d+tv,d+tv) \ge 2\gamma\|d\|^2 + 2 \gamma t^2\|v\|^2\ge \gamma \|d+tv\|^2 \tag3$$ as required. $\Box$

  • $\begingroup$ And now that I have opened my eyes, I realize that the proof by contradiction is actually shorter than the direct one. $\endgroup$ – Julien May 16 '13 at 23:50
  • $\begingroup$ Julien, si tu as une preuve plus simple, laisse-moi le savoir stp! $\endgroup$ – Stephane May 17 '13 at 14:43

Here is the answer to my own question. Please feel free to comment. Any suggestion is welcome:

Let $\widetilde{H}$ be the closed subspace generated by $D$ and the one-dimensional subspace of $H$. Since $B$ is positive definite, there exists a one-to-one bounded symmetric operator $L$ on $\widetilde{H}$ such that $B(u,v)=\langle Lu,\,v\rangle$. Let $y\in \widetilde{H}$ such that $\langle Ld,\,y\rangle=0$ for all $d\in D$. Then, any element $u$ of $\widetilde{H}$ has a unique representation $u=ay+d$ for some $a\in \mathbb{R}$ and $d\in D$. Let $\widetilde{L}$ be the restriction of $L$ on $D$, i.e., if $v\in \widetilde{H}$ and $Lv=a'y+d'$ for some $a'\in \mathbb{R}$ and $d'\in D$, then $\widetilde{L}v\equiv d'$. The operator $\widetilde{L}$ is such that $B(d_1,\,d_2)=\langle \widetilde{L}d_1,\,d_2\rangle$ for all $d_1,d_2 \in D$. Since $B$ is coercive on $D$, $\widetilde{L}$ is an isomorphism on $D$ by the Lax-Milgram theorem. Thus the image of $D$ under $L$ is a closed subspace of codimension 1 in $\widetilde{H}$. Since $L$ is one-to-one, it follows that $L$ is an isomorphism on $\widetilde{H}$ and thus $B$ is coercive on $\widetilde{H}$.

  • $\begingroup$ Looks okay to me, but how do you justify the existence of $y$ such that $\langle Ld,y\rangle=0$ for all $d\in D$? In my version of the proof (just posted) I made this a lemma. $\endgroup$ – 75064 May 16 '13 at 22:56
  • $\begingroup$ Dear 75064, thank you. Yes, the lemma is really needed. $\endgroup$ – Stephane May 17 '13 at 14:44

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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