$z$ is a vector in $R^n$. Given the following functions:

$c_i(z): R^n\to R$

$c(z) = (c_1(z),\dots,c_m(z))^T$

$A(z)^T = [\nabla c_i (z)]_{i\in [m]}$, where $[m] = \{1,2,\dots,m\}$

And $A(z)^T$ has full column rank.

We construct a matrix $Z$, whose columns are a basis of $Nul\ A(z)$.

Please prove $$ \left[\begin{array}{c}{A\left(z\right)} \\\\{Z^{T}}\end{array}\right] $$ is invertible (nonsingular).

Let me know if you have any puzzles about the problem itself.

  • $\begingroup$ Have you ever seen a proof of the rank-nullity theorem? $\endgroup$ – Arnaud Mortier Nov 6 '19 at 10:36
  • $\begingroup$ Yes, I have looked up the proof. $\endgroup$ – Ben Nov 6 '19 at 10:38
  • $\begingroup$ Well then :-) I'll let you answer your own question and collect all the upvotes. $\endgroup$ – Arnaud Mortier Nov 6 '19 at 10:40
  • $\begingroup$ But what I get is only $rank(A(z)) = m$, $rank(Z) = n-m$ $\endgroup$ – Ben Nov 6 '19 at 10:43
  • $\begingroup$ This is why I asked you to look at the proof, not at the theorem. $\endgroup$ – Arnaud Mortier Nov 6 '19 at 10:45

For any subspace $V$ of $\Bbb R^n$, you have a decomposition $$\Bbb R^n=V\oplus V^\perp$$ In the present case, consider $V$ to be the row space of $A(z)$. Then by definition $V^\perp=\operatorname{Ker}A(z)$.

By construction, the rows of $A(z)$ are a basis of $V$ and the rows of $Z^T$ are a basis of $V^\perp$. Since $\Bbb R^n=V\oplus V^\perp$, it follows that the rows of the full matrix form a basis of $\Bbb R^n$, so that the matrix is non-singular.

  • $\begingroup$ What does $𝑉^⊥$ mean? $\endgroup$ – Ben Nov 6 '19 at 11:26
  • $\begingroup$ @Ben It means the orthogonal subspace to $V$, the set of all vectors whose scalar product with any element of $V$ is zero. $\endgroup$ – Arnaud Mortier Nov 6 '19 at 11:27
  • $\begingroup$ Why do we have, for any subspace 𝑉 of $ℝ_n$, we have a decomposition? What theorem does this rely on? $\endgroup$ – Ben Nov 6 '19 at 11:45
  • $\begingroup$ @Ben It is well-known, you will find it in any textbook. You first show that by Gaussian elimination the dimension of $V^\perp$ is $n-\dim V$, and then you show separately that $V$ and $V^\perp$ are in direct sum. $\endgroup$ – Arnaud Mortier Nov 6 '19 at 11:59
  • $\begingroup$ All right. Thanks. $\endgroup$ – Ben Nov 6 '19 at 12:01

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.