8
$\begingroup$

Let $a, b, c \in \mathbb{R}^n$ , $p \in [1, +\infty)$, prove that

$$\left( \sum_{1\leq i < j <k \leq n} \left| \det\left(\begin{matrix} a_i & b_i & c_i \\ a_j & b_j & c_j \\ a_k & b_k & c_k \end{matrix}\right)\right|^p \right)^{\frac{1}{p}} \leq c_p \left( \sum_{i=1}^n |a_i|^p \right)^{\frac{1}{p}} \left( \sum_{1\leq j <k \leq n} \left| \det\left(\begin{matrix} b_j & c_j \\ b_k & c_k \end{matrix}\right)\right|^p \right)^{\frac{1}{p}} $$

where $c_p = \max(1, 3^{1-\frac{2}{p}})$.

A 2-dimensional analogue of this problem was discussed here: An inequality related to Lagrange's identity and $L_p$ norm

Remark:

  1. When $p=1$, the proof is straightforward since $$\left| \det\left(\begin{matrix} a_i & b_i & c_i \\ a_j & b_j & c_j \\ a_k & b_k & c_k \end{matrix}\right)\right| \leq |a_i| \left| \det\left(\begin{matrix} b_j & c_j \\ b_k & c_k \end{matrix}\right)\right| + |a_j| \left| \det\left(\begin{matrix} b_i & c_i \\ b_k & c_k \end{matrix}\right)\right| +|a_k| \left| \det\left(\begin{matrix} b_i & c_i\\ b_j & c_j \end{matrix}\right)\right| $$, by Laplace expansion and triangle inequality. Summing up all these inequalities is enough. $p = \infty$ case can be proved in a similar way.

Using Holder's inequality directly on the Laplace expansion gives a weaker bound: $3^{1 - \frac{1}{p}}$

  1. When $p=2$, LHS is the volume of Parallelepiped spanned by three vectors $a,b,c$, while RHS is the norm of $a$ times the area of parallelogram spanned by $b,c$, so the inequality is clearly true. (This fact can be proved by using Cauchy-Binet)

  2. As users @fedja and @mahdi suggested in An inequality related to Lagrange's identity and $L_p$ norm , this problem is closely related to Riesz-Thorin interpolation theorem. However, I find it difficult to apply the theorem directly on my problem.

Thanks!

$\endgroup$
2
  • $\begingroup$ If we could prove that $$ \sum_{1\leq i,j,k\leq n} |a_i c_{jk}-a_jc_{ik}+a_kc_{ij}|^2 \leq \left(\sum_{i=1}^n a_i^2 \right) \left( \sum_{1\leq i,j\leq n}c_{ij}^2 \right) $$ for all $a \in \mathbb R^n$, and all $n\times n$ skew-symmetric matrix $C=[c_{ij}]$, then Riesz-Thorin interpolation theorem implies your mentioned inequality. $\endgroup$
    – Mahdi
    Jul 8, 2018 at 10:03
  • $\begingroup$ Yes I have already proved that too. I am a bit busy these days, and will update a proof later. Thank you Mahdi! $\endgroup$
    – Chen Dan
    Jul 9, 2018 at 1:56

1 Answer 1

1
$\begingroup$

Actually, Riesz-Thorin interpolation theorem proves a generalized result. That is,

Observation. for all $a \in \mathbb C^n$, and all $n\times n$ skew-Hermitian matrix $C$, we have $$ \sum_{1\leq i<j<k\leq n} |a_i c_{jk}-a_jc_{ik}+a_kc_{ij}|^p \leq c^p_p\left(\sum_{i=1}^n a_i^p \right) \left( \sum_{1\leq i<j\leq n}c_{ij}^p \right) $$ where $c_p := \max (1,3^{1-2/p})$ and $1\leq p \leq \infty$.

When $p=1,\infty$, above observation is obvious. Also, above observation holds true when $p=2$ (as OP said). Now, by considering a multilinear function $\Lambda$ defined as $[\Lambda(a,C)]_{i,j,k} = a_i c_{jk}-a_jc_{ik}+a_kc_{ij}$ and interpolating between $1$ and $2$ and then between $2$ and $\infty$, above observation was proved for $1\leq p \leq \infty$ (similar to the answer of a previous question of OP).

If we set $c_{ij} := \det \begin{pmatrix}b_i&c_i\\b_j&c_j\end{pmatrix}$ in the above observation, the mentioned inequality of OP follows.

$\endgroup$

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.