Skip to main content
added 3 characters in body
Source Link
Iosif Pinelis
  • 128k
  • 8
  • 107
  • 229

The answer is no. Indeed, let $n=1$, $T=1$, and $\Lambda(t,u,v)\equiv v^2$, so that \begin{equation} I(x)=\int_0^1 x'(t)^2\,dt. \end{equation} Let $x_0:=0$ and, for each real $b\ge1$ and all $t\in[0,1]$, \begin{equation} y_b(t):=e^{-bt}. \end{equation} Then \begin{equation} \|y_b-x_0\|=\|y_b\|=\int_0^1 |y_b(t)|\,dt+\int_0^1 |y_b'(t)|\,dt\le1/b+1\le2, \end{equation} so that $y_b$ is in the ball of radius $2$ centered at $x_0$.

However, \begin{equation} I(y_b)-I(x_0)=I(y_b)=\int_0^1 y_b'(t)^2\,dt\sim b/2\to\infty \end{equation} as $b\to\infty$. So, the functional $I$ is not locally Lipschitz.

The answer is no. Indeed, let $n=1$, $T=1$, and $\Lambda(t,u,v)\equiv v^2$, so that \begin{equation} I(x)=\int_0^1 x'(t)^2\,dt. \end{equation} Let $x_0:=0$ and, for each real $b\ge1$ and all $t\in[0,1]$, \begin{equation} y_b(t):=e^{-bt}. \end{equation} Then \begin{equation} \|y_b-x_0\|=\|y_b\|=\int_0^1 |y_b(t)|\,dt+\int_0^1 |y_b'(t)|\,dt\le1/b+1\le2, \end{equation} so that $y_b$ in the ball of radius $2$ centered at $x_0$.

However, \begin{equation} I(y_b)-I(x_0)=I(y_b)=\int_0^1 y_b'(t)^2\,dt\sim b/2\to\infty \end{equation} as $b\to\infty$. So, the functional $I$ is not locally Lipschitz.

The answer is no. Indeed, let $n=1$, $T=1$, and $\Lambda(t,u,v)\equiv v^2$, so that \begin{equation} I(x)=\int_0^1 x'(t)^2\,dt. \end{equation} Let $x_0:=0$ and, for each real $b\ge1$ and all $t\in[0,1]$, \begin{equation} y_b(t):=e^{-bt}. \end{equation} Then \begin{equation} \|y_b-x_0\|=\|y_b\|=\int_0^1 |y_b(t)|\,dt+\int_0^1 |y_b'(t)|\,dt\le1/b+1\le2, \end{equation} so that $y_b$ is in the ball of radius $2$ centered at $x_0$.

However, \begin{equation} I(y_b)-I(x_0)=I(y_b)=\int_0^1 y_b'(t)^2\,dt\sim b/2\to\infty \end{equation} as $b\to\infty$. So, the functional $I$ is not locally Lipschitz.

deleted 582 characters in body
Source Link
Iosif Pinelis
  • 128k
  • 8
  • 107
  • 229

ThisThe answer is quite straightforward: For the derivativeno. Indeed, let $I'(x)$$n=1$, $T=1$, and any $h\in C^1[0,T]$ we have$\Lambda(t,u,v)\equiv v^2$, so that $$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt, \tag{1}$$\begin{equation} I(x)=\int_0^1 x'(t)^2\,dt. \end{equation} whence for allLet $x$ in any bounded neighborhood$x_0:=0$ and, for each real $U$ of$b\ge1$ and all $x_0$ we have$t\in[0,1]$, $$|I'(x)(h)|\le M\int_0^T \|(0,h(t),h'(t))\|\,dt \le M\Big(\int_0^T \|h(t)\|\,dt+\int_0^T \|h'(t)\|\,dt\Big) =M\|h\|,$$\begin{equation} y_b(t):=e^{-bt}. \end{equation} whereThen $$M:=\sup_{x\in U}\|(\nabla\Lambda)(t,x(t),x'(t))\|<\infty,$$\begin{equation} \|y_b-x_0\|=\|y_b\|=\int_0^1 |y_b(t)|\,dt+\int_0^1 |y_b'(t)|\,dt\le1/b+1\le2, \end{equation} sinceso that $\Lambda$ is continuously differentiable.

So,$y_b$ in the derivativeball of radius $I'$ is locally bounded and hence$2$ centered at $I$ is locally Lipschitz$x_0$.


 

Details on $I'$, in response to a comment by the OP: If $I'$ is understood in the Gateaux sense (which is enough for the "locally Lipschitz" condition on $I$) However, then (1) follows from the continuous differentiability of \begin{equation} I(y_b)-I(x_0)=I(y_b)=\int_0^1 y_b'(t)^2\,dt\sim b/2\to\infty \end{equation} as $\Lambda$ by the dominated convergence theorem$b\to\infty$. Working just a bit harderSo, one can show that $I'$ exists in the functional Fréchet sense as well$I$ is not locally Lipschitz.

This is quite straightforward: For the derivative $I'(x)$ and any $h\in C^1[0,T]$ we have $$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt, \tag{1}$$ whence for all $x$ in any bounded neighborhood $U$ of $x_0$ we have $$|I'(x)(h)|\le M\int_0^T \|(0,h(t),h'(t))\|\,dt \le M\Big(\int_0^T \|h(t)\|\,dt+\int_0^T \|h'(t)\|\,dt\Big) =M\|h\|,$$ where $$M:=\sup_{x\in U}\|(\nabla\Lambda)(t,x(t),x'(t))\|<\infty,$$ since $\Lambda$ is continuously differentiable.

So, the derivative $I'$ is locally bounded and hence $I$ is locally Lipschitz.


 

Details on $I'$, in response to a comment by the OP: If $I'$ is understood in the Gateaux sense (which is enough for the "locally Lipschitz" condition on $I$), then (1) follows from the continuous differentiability of $\Lambda$ by the dominated convergence theorem. Working just a bit harder, one can show that $I'$ exists in the Fréchet sense as well.

The answer is no. Indeed, let $n=1$, $T=1$, and $\Lambda(t,u,v)\equiv v^2$, so that \begin{equation} I(x)=\int_0^1 x'(t)^2\,dt. \end{equation} Let $x_0:=0$ and, for each real $b\ge1$ and all $t\in[0,1]$, \begin{equation} y_b(t):=e^{-bt}. \end{equation} Then \begin{equation} \|y_b-x_0\|=\|y_b\|=\int_0^1 |y_b(t)|\,dt+\int_0^1 |y_b'(t)|\,dt\le1/b+1\le2, \end{equation} so that $y_b$ in the ball of radius $2$ centered at $x_0$.

However, \begin{equation} I(y_b)-I(x_0)=I(y_b)=\int_0^1 y_b'(t)^2\,dt\sim b/2\to\infty \end{equation} as $b\to\infty$. So, the functional $I$ is not locally Lipschitz.

added 634 characters in body
Source Link
Iosif Pinelis
  • 128k
  • 8
  • 107
  • 229

This is quite straightforward: For the derivative $I'(x)$ and any $h\in C^1[0,T]$ we have $$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt,$$$$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt, \tag{1}$$ whence for all $x$ in any bounded neighborhood $U$ of $x_0$ we have $$|I'(x)(h)|\le M\int_0^T \|(0,h(t),h'(t))\|\,dt \le M\Big(\int_0^T \|h(t)\|\,dt+\int_0^T \|h'(t)\|\,dt\Big) =M\|h\|,$$ where $$M:=\sup_{x\in U}\|(\nabla\Lambda)(t,x(t),x'(t))\|<\infty,$$ since $\Lambda$ is continuously differentiable.

So, the derivative $I'$ is locally bounded and hence $I$ is locally Lipschitz.


Details on $I'$, in response to a comment by the OP: If $I'$ is understood in the Gateaux sense (which is enough for the "locally Lipschitz" condition on $I$), then (1) follows from the continuous differentiability of $\Lambda$ by the dominated convergence theorem. Working just a bit harder, one can show that $I'$ exists in the Fréchet sense as well.

This is quite straightforward: For the derivative $I'(x)$ and any $h\in C^1[0,T]$ we have $$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt,$$ whence for all $x$ in any bounded neighborhood $U$ of $x_0$ we have $$|I'(x)(h)|\le M\int_0^T \|(0,h(t),h'(t))\|\,dt \le M\Big(\int_0^T \|h(t)\|\,dt+\int_0^T \|h'(t)\|\,dt\Big) =M\|h\|,$$ where $$M:=\sup_{x\in U}\|(\nabla\Lambda)(t,x(t),x'(t))\|<\infty,$$ since $\Lambda$ is continuously differentiable.

So, the derivative $I'$ is locally bounded and hence $I$ is locally Lipschitz.

This is quite straightforward: For the derivative $I'(x)$ and any $h\in C^1[0,T]$ we have $$I'(x)(h)=\int_0^T (\nabla\Lambda)(t,x(t),x'(t))\cdot(0,h(t),h'(t)) \, dt, \tag{1}$$ whence for all $x$ in any bounded neighborhood $U$ of $x_0$ we have $$|I'(x)(h)|\le M\int_0^T \|(0,h(t),h'(t))\|\,dt \le M\Big(\int_0^T \|h(t)\|\,dt+\int_0^T \|h'(t)\|\,dt\Big) =M\|h\|,$$ where $$M:=\sup_{x\in U}\|(\nabla\Lambda)(t,x(t),x'(t))\|<\infty,$$ since $\Lambda$ is continuously differentiable.

So, the derivative $I'$ is locally bounded and hence $I$ is locally Lipschitz.


Details on $I'$, in response to a comment by the OP: If $I'$ is understood in the Gateaux sense (which is enough for the "locally Lipschitz" condition on $I$), then (1) follows from the continuous differentiability of $\Lambda$ by the dominated convergence theorem. Working just a bit harder, one can show that $I'$ exists in the Fréchet sense as well.

Source Link
Iosif Pinelis
  • 128k
  • 8
  • 107
  • 229
Loading