Hölder continuity for operators
Clash Royale CLAN TAG#URR8PPP
up vote
8
down vote
favorite
Let $x,y$ be positive real numbers then
$$|sqrtx-sqrty|=dfracsqrtx+sqrty=sqrtcdot dfracsqrtsqrtx+sqrtyleq 1cdot |x-y|^frac12$$
we obtain $1/2$-Hölder continuity for the square-root.
I would like to know if $x,y$ are positive Hilbert-Schmidt operators. Does it follow then that for some $C>0$
$$leftlVert sqrtx-sqrty rightrVert_HS le C leftlVert x-yrightrVert_HS^frac12.$$
Sounds natural, but on the other hand, it is less obvious to me how this should follow.
One remark however is that if it would hold for finite-rank operators, then a density argument yields the claim.
fa.functional-analysis real-analysis ap.analysis-of-pdes ca.classical-analysis-and-odes operator-theory
add a comment |Â
up vote
8
down vote
favorite
Let $x,y$ be positive real numbers then
$$|sqrtx-sqrty|=dfracsqrtx+sqrty=sqrtcdot dfracsqrtsqrtx+sqrtyleq 1cdot |x-y|^frac12$$
we obtain $1/2$-Hölder continuity for the square-root.
I would like to know if $x,y$ are positive Hilbert-Schmidt operators. Does it follow then that for some $C>0$
$$leftlVert sqrtx-sqrty rightrVert_HS le C leftlVert x-yrightrVert_HS^frac12.$$
Sounds natural, but on the other hand, it is less obvious to me how this should follow.
One remark however is that if it would hold for finite-rank operators, then a density argument yields the claim.
fa.functional-analysis real-analysis ap.analysis-of-pdes ca.classical-analysis-and-odes operator-theory
add a comment |Â
up vote
8
down vote
favorite
up vote
8
down vote
favorite
Let $x,y$ be positive real numbers then
$$|sqrtx-sqrty|=dfracsqrtx+sqrty=sqrtcdot dfracsqrtsqrtx+sqrtyleq 1cdot |x-y|^frac12$$
we obtain $1/2$-Hölder continuity for the square-root.
I would like to know if $x,y$ are positive Hilbert-Schmidt operators. Does it follow then that for some $C>0$
$$leftlVert sqrtx-sqrty rightrVert_HS le C leftlVert x-yrightrVert_HS^frac12.$$
Sounds natural, but on the other hand, it is less obvious to me how this should follow.
One remark however is that if it would hold for finite-rank operators, then a density argument yields the claim.
fa.functional-analysis real-analysis ap.analysis-of-pdes ca.classical-analysis-and-odes operator-theory
Let $x,y$ be positive real numbers then
$$|sqrtx-sqrty|=dfracsqrtx+sqrty=sqrtcdot dfracsqrtsqrtx+sqrtyleq 1cdot |x-y|^frac12$$
we obtain $1/2$-Hölder continuity for the square-root.
I would like to know if $x,y$ are positive Hilbert-Schmidt operators. Does it follow then that for some $C>0$
$$leftlVert sqrtx-sqrty rightrVert_HS le C leftlVert x-yrightrVert_HS^frac12.$$
Sounds natural, but on the other hand, it is less obvious to me how this should follow.
One remark however is that if it would hold for finite-rank operators, then a density argument yields the claim.
fa.functional-analysis real-analysis ap.analysis-of-pdes ca.classical-analysis-and-odes operator-theory
fa.functional-analysis real-analysis ap.analysis-of-pdes ca.classical-analysis-and-odes operator-theory
edited Aug 11 at 9:31
asked Aug 11 at 9:04
Hilbért
953
953
add a comment |Â
add a comment |Â
2 Answers
2
active
oldest
votes
up vote
13
down vote
accepted
Your proposed inequality doesn't even work on diagonal matrices. Let $xinmathbb M_n$ be diagonal with entries $x_igeq 0$. Then $x^1/2$ is diagonal with entries $sqrtx_i$. Thus
$$ |x^1/2|_HS^2 = sum_i x_i, qquad
|x|_HS = Big( sum_i x_i^2 Big)^1/2. $$
Taking e.g. $x_i=1/sqrt n$ we obtain $n/sqrt n leq C^2$ so $Cgeq n^1/4$.
The Powers-Stormer inequality (see section 4 of https://projecteuclid.org/euclid.cmp/1103842028) says that
$$ |x^1/2 - y^1/2|_HS^2 leq |x - y|_1 $$
for all $x,y$ positive and Hilbert Schmidt. Here $|cdot|_1$ is the trace-class norm (which takes the value $infty$ if $x-y$ is not trace-class).
In finite-dimensions, we have that $|x|_1 leq n^1/2 |x|_HS$ (reduce to the diagonal case and apply Cauchy-Schwarz) and so we see that $C=n^1/4$ works in general.
add a comment |Â
up vote
11
down vote
For the Hilbert-Schmidt norm, the inequality $|X^1/2-Y^1/2|_2 le C|X-Y|_2^1/2$ is false in general. Consider for that the case of $ntimes n$ positive definite matrices. Put e.g., $Y=0$, then asking for a $C>0$ amounts to requiring
beginequation*
texttr,X le C'sqrttexttr,X^2
endequation*
which in general cannot hold for a constant independent of the dimension of $X$.
However, for all positive operators $X$ and $Y$, and for any unitarily invariant norm $|!|!| cdot |!|!|$
beginequation*
|!|!|X^p - Y^p|!|!| le |!|!| |X-Y|^p |!|!|,quad 0 le p le 1.
endequation*
Thus, in particular for the operator norm, this inequality implies $p$-Hölder continuity (with $C=1$).
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
add a comment |Â
2 Answers
2
active
oldest
votes
2 Answers
2
active
oldest
votes
active
oldest
votes
active
oldest
votes
up vote
13
down vote
accepted
Your proposed inequality doesn't even work on diagonal matrices. Let $xinmathbb M_n$ be diagonal with entries $x_igeq 0$. Then $x^1/2$ is diagonal with entries $sqrtx_i$. Thus
$$ |x^1/2|_HS^2 = sum_i x_i, qquad
|x|_HS = Big( sum_i x_i^2 Big)^1/2. $$
Taking e.g. $x_i=1/sqrt n$ we obtain $n/sqrt n leq C^2$ so $Cgeq n^1/4$.
The Powers-Stormer inequality (see section 4 of https://projecteuclid.org/euclid.cmp/1103842028) says that
$$ |x^1/2 - y^1/2|_HS^2 leq |x - y|_1 $$
for all $x,y$ positive and Hilbert Schmidt. Here $|cdot|_1$ is the trace-class norm (which takes the value $infty$ if $x-y$ is not trace-class).
In finite-dimensions, we have that $|x|_1 leq n^1/2 |x|_HS$ (reduce to the diagonal case and apply Cauchy-Schwarz) and so we see that $C=n^1/4$ works in general.
add a comment |Â
up vote
13
down vote
accepted
Your proposed inequality doesn't even work on diagonal matrices. Let $xinmathbb M_n$ be diagonal with entries $x_igeq 0$. Then $x^1/2$ is diagonal with entries $sqrtx_i$. Thus
$$ |x^1/2|_HS^2 = sum_i x_i, qquad
|x|_HS = Big( sum_i x_i^2 Big)^1/2. $$
Taking e.g. $x_i=1/sqrt n$ we obtain $n/sqrt n leq C^2$ so $Cgeq n^1/4$.
The Powers-Stormer inequality (see section 4 of https://projecteuclid.org/euclid.cmp/1103842028) says that
$$ |x^1/2 - y^1/2|_HS^2 leq |x - y|_1 $$
for all $x,y$ positive and Hilbert Schmidt. Here $|cdot|_1$ is the trace-class norm (which takes the value $infty$ if $x-y$ is not trace-class).
In finite-dimensions, we have that $|x|_1 leq n^1/2 |x|_HS$ (reduce to the diagonal case and apply Cauchy-Schwarz) and so we see that $C=n^1/4$ works in general.
add a comment |Â
up vote
13
down vote
accepted
up vote
13
down vote
accepted
Your proposed inequality doesn't even work on diagonal matrices. Let $xinmathbb M_n$ be diagonal with entries $x_igeq 0$. Then $x^1/2$ is diagonal with entries $sqrtx_i$. Thus
$$ |x^1/2|_HS^2 = sum_i x_i, qquad
|x|_HS = Big( sum_i x_i^2 Big)^1/2. $$
Taking e.g. $x_i=1/sqrt n$ we obtain $n/sqrt n leq C^2$ so $Cgeq n^1/4$.
The Powers-Stormer inequality (see section 4 of https://projecteuclid.org/euclid.cmp/1103842028) says that
$$ |x^1/2 - y^1/2|_HS^2 leq |x - y|_1 $$
for all $x,y$ positive and Hilbert Schmidt. Here $|cdot|_1$ is the trace-class norm (which takes the value $infty$ if $x-y$ is not trace-class).
In finite-dimensions, we have that $|x|_1 leq n^1/2 |x|_HS$ (reduce to the diagonal case and apply Cauchy-Schwarz) and so we see that $C=n^1/4$ works in general.
Your proposed inequality doesn't even work on diagonal matrices. Let $xinmathbb M_n$ be diagonal with entries $x_igeq 0$. Then $x^1/2$ is diagonal with entries $sqrtx_i$. Thus
$$ |x^1/2|_HS^2 = sum_i x_i, qquad
|x|_HS = Big( sum_i x_i^2 Big)^1/2. $$
Taking e.g. $x_i=1/sqrt n$ we obtain $n/sqrt n leq C^2$ so $Cgeq n^1/4$.
The Powers-Stormer inequality (see section 4 of https://projecteuclid.org/euclid.cmp/1103842028) says that
$$ |x^1/2 - y^1/2|_HS^2 leq |x - y|_1 $$
for all $x,y$ positive and Hilbert Schmidt. Here $|cdot|_1$ is the trace-class norm (which takes the value $infty$ if $x-y$ is not trace-class).
In finite-dimensions, we have that $|x|_1 leq n^1/2 |x|_HS$ (reduce to the diagonal case and apply Cauchy-Schwarz) and so we see that $C=n^1/4$ works in general.
answered Aug 11 at 13:32
Matthew Daws
8,42222255
8,42222255
add a comment |Â
add a comment |Â
up vote
11
down vote
For the Hilbert-Schmidt norm, the inequality $|X^1/2-Y^1/2|_2 le C|X-Y|_2^1/2$ is false in general. Consider for that the case of $ntimes n$ positive definite matrices. Put e.g., $Y=0$, then asking for a $C>0$ amounts to requiring
beginequation*
texttr,X le C'sqrttexttr,X^2
endequation*
which in general cannot hold for a constant independent of the dimension of $X$.
However, for all positive operators $X$ and $Y$, and for any unitarily invariant norm $|!|!| cdot |!|!|$
beginequation*
|!|!|X^p - Y^p|!|!| le |!|!| |X-Y|^p |!|!|,quad 0 le p le 1.
endequation*
Thus, in particular for the operator norm, this inequality implies $p$-Hölder continuity (with $C=1$).
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
add a comment |Â
up vote
11
down vote
For the Hilbert-Schmidt norm, the inequality $|X^1/2-Y^1/2|_2 le C|X-Y|_2^1/2$ is false in general. Consider for that the case of $ntimes n$ positive definite matrices. Put e.g., $Y=0$, then asking for a $C>0$ amounts to requiring
beginequation*
texttr,X le C'sqrttexttr,X^2
endequation*
which in general cannot hold for a constant independent of the dimension of $X$.
However, for all positive operators $X$ and $Y$, and for any unitarily invariant norm $|!|!| cdot |!|!|$
beginequation*
|!|!|X^p - Y^p|!|!| le |!|!| |X-Y|^p |!|!|,quad 0 le p le 1.
endequation*
Thus, in particular for the operator norm, this inequality implies $p$-Hölder continuity (with $C=1$).
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
add a comment |Â
up vote
11
down vote
up vote
11
down vote
For the Hilbert-Schmidt norm, the inequality $|X^1/2-Y^1/2|_2 le C|X-Y|_2^1/2$ is false in general. Consider for that the case of $ntimes n$ positive definite matrices. Put e.g., $Y=0$, then asking for a $C>0$ amounts to requiring
beginequation*
texttr,X le C'sqrttexttr,X^2
endequation*
which in general cannot hold for a constant independent of the dimension of $X$.
However, for all positive operators $X$ and $Y$, and for any unitarily invariant norm $|!|!| cdot |!|!|$
beginequation*
|!|!|X^p - Y^p|!|!| le |!|!| |X-Y|^p |!|!|,quad 0 le p le 1.
endequation*
Thus, in particular for the operator norm, this inequality implies $p$-Hölder continuity (with $C=1$).
For the Hilbert-Schmidt norm, the inequality $|X^1/2-Y^1/2|_2 le C|X-Y|_2^1/2$ is false in general. Consider for that the case of $ntimes n$ positive definite matrices. Put e.g., $Y=0$, then asking for a $C>0$ amounts to requiring
beginequation*
texttr,X le C'sqrttexttr,X^2
endequation*
which in general cannot hold for a constant independent of the dimension of $X$.
However, for all positive operators $X$ and $Y$, and for any unitarily invariant norm $|!|!| cdot |!|!|$
beginequation*
|!|!|X^p - Y^p|!|!| le |!|!| |X-Y|^p |!|!|,quad 0 le p le 1.
endequation*
Thus, in particular for the operator norm, this inequality implies $p$-Hölder continuity (with $C=1$).
answered Aug 11 at 13:34
Suvrit
23.4k659119
23.4k659119
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
add a comment |Â
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
3
3
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
Seems like both Matthew and I were typing at the same time; the counterexample is exactly based on the same idea too!
â Suvrit
Aug 11 at 13:36
add a comment |Â
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmathoverflow.net%2fquestions%2f308048%2fh%25c3%25b6lder-continuity-for-operators%23new-answer', 'question_page');
);
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Sign up or log in
StackExchange.ready(function ()
StackExchange.helpers.onClickDraftSave('#login-link');
);
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password