3
$\begingroup$

Let $$z(t)=\begin{pmatrix}x(t) \\ y(t)\end{pmatrix}~,~ A=\begin{pmatrix}0 & -\beta \\ \alpha & -(\alpha + \beta)\end{pmatrix}$$ satisfies the following system of linear differential inequalities $$z'(t) \leq A\,z(t)$$ where $\alpha, \beta > 0$ are fixed constants, and $x(t), y(t) \in [0,\infty)$, I am wondering if there is a vectorized version of Gronwall lemma which allows me to deduce that $$z(t)\leq \mathrm{e}^{At}\,z(0).$$ It seems hard to find a literature which addresses precisely my question. Any help (including referring to precise literatures) will be greatly appreciated!


Side note: by writing $u \leq v$ for $u,v \in \mathbb{R}^2_+$ I mean that each component of $u$ is less than or equal to the corresponding component of $v$. Also (and unfortunately), the off-diagonal entries of $A$ are not all non-negative, which means that the discussion mentioned in this post may not be applicable...

$\endgroup$
7
  • $\begingroup$ Dear Whitepiller, I have already tried hard and I still haven't found a precise literature in support of my naive guess/claim $\endgroup$
    – Fei Cao
    Commented May 26, 2023 at 22:27
  • 1
    $\begingroup$ That was a Yugioh reference lol $\endgroup$
    – Brian
    Commented May 26, 2023 at 22:33
  • 1
    $\begingroup$ It seems you forgot to mention what equation $z$ solves. $\endgroup$ Commented May 26, 2023 at 22:36
  • $\begingroup$ @ChristianRemling Hello, the $2$-dimensional $z(t)$ satisfies the system of linear differential inequalities (instead of solving the corresponding system of linear ODEs), I am sorry that I forgot to mention it in the original post. $\endgroup$
    – Fei Cao
    Commented May 27, 2023 at 0:11
  • 1
    $\begingroup$ @HopefulWhitepiller lol I am not YuGi, Judai, Yusei, Yuma, Yuya, or Yusaku (which means that I do not always "win the game"). $\endgroup$
    – Fei Cao
    Commented May 27, 2023 at 0:14

1 Answer 1

1
$\begingroup$

Counter-example. Let $\left(z(t)\right)_{t\geq 0}$ be the solution to

$$\dot{z}(t)=A z(t)-\varepsilon \mathbf{e}_2,$$

with $\mathbf{e}_2:=\left[0 \,\,\,\,1\right]^{\top}$ and initial condition $z(0)$. Then, $\left(z(t)\right)_{t\geq 0}$ obeys the inequality $\dot{z}(t)\leq Az(t)$ for all $t$.

Let $\left(x(t)\right)_{t\geq 0}$ be the solution to $\dot{x}(t)=Ax(t)$ with initial condition $x(0)=z(0)$. Then, $x(t)=e^{At}z(0)$ for all $t\geq 0$.

Now, remark that,

$$ \dot{z}_1(0)=\left[Az(0)\right]_1=\left[Ax(0)\right]_1=\dot{x}_1(0)$$

and

$$ \ddot{z}_1(0)=-\beta \dot{z}_2(0)>-\beta \dot{x}_2(0)=\ddot{x}_1(0),$$

since $\dot{z}_2(0)=\left[Az(0)\right]_2-\varepsilon < \left[Az(0)\right]_2 =\left[Ax(0)\right]_2 = \dot{x}_2(0).$ In other words, $w(t)\overset{\Delta}= z_1(t)-x_1(t)$ is analytic with $w(0)=0$, $\dot{w}(0)=0$ and $\ddot{w}(0)>0$. Therefore, $w(t)>0$ for all $t\in\left(0,\delta\right)$, for some $\delta>0$ small enough. That is,

$$z_1(t)>\left[e^{At}z(0)\right]_1$$

for all $t\in\left(0,\delta\right)$ which breaks Grönwall's.


For the record. If the two off-diagonal entries of $A$ are negative (with arbitrary diagonal entries), then we can show that Grönwall's inequality holds w.r.t. the partial order $\leq_a$ defined as $x\leq_a y \Leftrightarrow (x_1 \geq y_1 \wedge x_2 \leq y_2)$ or to the partial order $\leq_b$ defined as $x\leq_b y \Leftrightarrow (x_1 \leq y_1 \wedge x_2 \geq y_2)$. In other words, for this regime, if $\dot{z}(t)\leq_i Az(t)$ then, $z(t)\leq_i e^{At} z(0)$ for $i\in \left\{a,b\right\}$.

This is done simply via comparing $\left(z(t)\right)$ and $\left(x(t)\right)$ (where $x(t)$ is the solution to the linear system with $x(0)=z(0)$) at the hitting moment $T$ for the possible cases: i) $z_1(T)=x_1(T)$ and $z_2(T)<x_2(T)$; ii) $z_1(T)>x_1(T)$ and $z_2(T)=x_2(T)$; or iii) $z_1(T)=x_1(T)$ and $z_2(T)=x_2(T)$. In the latter case you need to resort to the higher order derivatives as $\dot{z}_1(T)=\dot{x}_1(T)$ and $\dot{z}_2(T)=\dot{x}_{2}(T)$.

$\endgroup$
2
  • $\begingroup$ Thank you Professor. It seems that the vectorized Gronwall inequality can only be valid when all the off-diagonal entries of the matrix $A$ are non-negative... $\endgroup$
    – Fei Cao
    Commented May 28, 2023 at 20:44
  • $\begingroup$ @MathandYuGiOhlover, the counter-example above shows that Gronwall's inequality does not necessarily hold with respect to the partial order ≤ (induced by the nonnegative orthant). I believe that if the off-diagonal entries are negative, then Gronwall's inequality holds w.r.t. another partial order as in my updated answer. $\endgroup$ Commented May 29, 2023 at 9:30

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .