Solve $ (X+A)(X+A)'=mathrmdiag(I,0)+AA' $

The name of the pictureThe name of the pictureThe name of the pictureClash Royale CLAN TAG#URR8PPP











up vote
1
down vote

favorite












Let $A$ be a given $m times n$ matrix with $m geq n$ and Rank$(A)=n$. Does there always exist an $m times n$ matrix $X$ (with Rank$(X)=n$) that solves,
$$
(X+A)(X+A)'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]+AA'?
$$



Simplifying, I get
$$
XA'+AX'+ XX'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]
$$



but not sure where to go from here.



I was trying to investigate the case where $m=n$ but didn't make progress.



I suspect the answer is yes (but this is pure speculation from the case when all objects are scalars). This reminds me of Sylvester equations https://en.wikipedia.org/wiki/Sylvester_equation and pseudo inverses https://en.wikipedia.org/wiki/Moore%E2%80%93Penrose_pseudoinverse










share|cite|improve this question























  • If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
    – D. Thomine
    Nov 20 '15 at 14:40














up vote
1
down vote

favorite












Let $A$ be a given $m times n$ matrix with $m geq n$ and Rank$(A)=n$. Does there always exist an $m times n$ matrix $X$ (with Rank$(X)=n$) that solves,
$$
(X+A)(X+A)'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]+AA'?
$$



Simplifying, I get
$$
XA'+AX'+ XX'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]
$$



but not sure where to go from here.



I was trying to investigate the case where $m=n$ but didn't make progress.



I suspect the answer is yes (but this is pure speculation from the case when all objects are scalars). This reminds me of Sylvester equations https://en.wikipedia.org/wiki/Sylvester_equation and pseudo inverses https://en.wikipedia.org/wiki/Moore%E2%80%93Penrose_pseudoinverse










share|cite|improve this question























  • If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
    – D. Thomine
    Nov 20 '15 at 14:40












up vote
1
down vote

favorite









up vote
1
down vote

favorite











Let $A$ be a given $m times n$ matrix with $m geq n$ and Rank$(A)=n$. Does there always exist an $m times n$ matrix $X$ (with Rank$(X)=n$) that solves,
$$
(X+A)(X+A)'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]+AA'?
$$



Simplifying, I get
$$
XA'+AX'+ XX'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]
$$



but not sure where to go from here.



I was trying to investigate the case where $m=n$ but didn't make progress.



I suspect the answer is yes (but this is pure speculation from the case when all objects are scalars). This reminds me of Sylvester equations https://en.wikipedia.org/wiki/Sylvester_equation and pseudo inverses https://en.wikipedia.org/wiki/Moore%E2%80%93Penrose_pseudoinverse










share|cite|improve this question















Let $A$ be a given $m times n$ matrix with $m geq n$ and Rank$(A)=n$. Does there always exist an $m times n$ matrix $X$ (with Rank$(X)=n$) that solves,
$$
(X+A)(X+A)'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]+AA'?
$$



Simplifying, I get
$$
XA'+AX'+ XX'=left[beginarrayccI_n & 0_n times (m-n) \ 0_(m-n) times n & 0_(m-n) times (m-n)\ endarrayright]
$$



but not sure where to go from here.



I was trying to investigate the case where $m=n$ but didn't make progress.



I suspect the answer is yes (but this is pure speculation from the case when all objects are scalars). This reminds me of Sylvester equations https://en.wikipedia.org/wiki/Sylvester_equation and pseudo inverses https://en.wikipedia.org/wiki/Moore%E2%80%93Penrose_pseudoinverse







matrices matrix-equations






share|cite|improve this question















share|cite|improve this question













share|cite|improve this question




share|cite|improve this question








edited Aug 31 at 5:44









Did

243k23209444




243k23209444










asked Nov 20 '15 at 13:50









user103828

1,2831031




1,2831031











  • If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
    – D. Thomine
    Nov 20 '15 at 14:40
















  • If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
    – D. Thomine
    Nov 20 '15 at 14:40















If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
– D. Thomine
Nov 20 '15 at 14:40




If $m > n$, then the LHS has rank at most $n$, but I see no reason why the RHS should be rank $n$ (depending on the circumstances, I think it could very well be rank $m$). A counter-example would be given by $A' = (0,1)$. I don't have an opinion on the $n=m$ case, though.
– D. Thomine
Nov 20 '15 at 14:40










1 Answer
1






active

oldest

votes

















up vote
1
down vote



accepted










Let $Y=X+A$; then $YY^T=diag(I_n,0_m-n)+AA^T=B$.



Since $rank(YY^T)leq n$, if we want that a solution exists, then necessarily $rank(B)leq n$.



EDIT 1. Since $AA^T$ is symmetric $geq 0$, $rank(B)geq n$ and we assume that $rank(B)=n$. Since $rank(AA^T)=n$, the spd symmetric matrix $AA^T$ is in the form $diag(U_n,0_m-n)$, where $U$ is a pd symmetric matrix. Finally we may assume that $B=diag(b_1,cdots,b_n,0_m-n)$ where $b_i>0$. We can choose $Y=beginpmatrixdiag(pmsqrtb_i)\0_m-n,nendpmatrix$.



EDIT 2. Answer to user. Necessarily, for any $y$, $[0_n,y^T]B[0_n,y]=0$; then $AA^T$ is in the form $beginpmatrixU_n&V\V^T&0_m-nendpmatrix$. Thus if $A=[P,Q]^T$,then $QQ^T=0$, that implies that $Q=0$. Finally $V$ is also a zero matrix and we are done.



Conclusion. If $A$ is not in the form $[P,0]^T$, then $0$ solution; otherwise the solutions are $Y=[F,0]^T$ with $FF^T=I_n+PP^T=Wdiag(b_i)W^T$ where $Win O(n)$. Thus there are at least $2^n$ solutions: $X=beginpmatrixWdiag(pmsqrtb_i)W^T\0_m-n,nendpmatrix-A$.






share|cite|improve this answer






















  • What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
    – user103828
    Nov 21 '15 at 11:03











  • @ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
    – loup blanc
    Nov 21 '15 at 15:59










  • Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
    – user103828
    Nov 22 '15 at 18:39










Your Answer




StackExchange.ifUsing("editor", function ()
return StackExchange.using("mathjaxEditing", function ()
StackExchange.MarkdownEditor.creationCallbacks.add(function (editor, postfix)
StackExchange.mathjaxEditing.prepareWmdForMathJax(editor, postfix, [["$", "$"], ["\\(","\\)"]]);
);
);
, "mathjax-editing");

StackExchange.ready(function()
var channelOptions =
tags: "".split(" "),
id: "69"
;
initTagRenderer("".split(" "), "".split(" "), channelOptions);

StackExchange.using("externalEditor", function()
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled)
StackExchange.using("snippets", function()
createEditor();
);

else
createEditor();

);

function createEditor()
StackExchange.prepareEditor(
heartbeatType: 'answer',
convertImagesToLinks: true,
noModals: false,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
noCode: true, onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
);



);













 

draft saved


draft discarded


















StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f1538425%2fsolve-xaxa-mathrmdiagi-0aa%23new-answer', 'question_page');

);

Post as a guest






























1 Answer
1






active

oldest

votes








1 Answer
1






active

oldest

votes









active

oldest

votes






active

oldest

votes








up vote
1
down vote



accepted










Let $Y=X+A$; then $YY^T=diag(I_n,0_m-n)+AA^T=B$.



Since $rank(YY^T)leq n$, if we want that a solution exists, then necessarily $rank(B)leq n$.



EDIT 1. Since $AA^T$ is symmetric $geq 0$, $rank(B)geq n$ and we assume that $rank(B)=n$. Since $rank(AA^T)=n$, the spd symmetric matrix $AA^T$ is in the form $diag(U_n,0_m-n)$, where $U$ is a pd symmetric matrix. Finally we may assume that $B=diag(b_1,cdots,b_n,0_m-n)$ where $b_i>0$. We can choose $Y=beginpmatrixdiag(pmsqrtb_i)\0_m-n,nendpmatrix$.



EDIT 2. Answer to user. Necessarily, for any $y$, $[0_n,y^T]B[0_n,y]=0$; then $AA^T$ is in the form $beginpmatrixU_n&V\V^T&0_m-nendpmatrix$. Thus if $A=[P,Q]^T$,then $QQ^T=0$, that implies that $Q=0$. Finally $V$ is also a zero matrix and we are done.



Conclusion. If $A$ is not in the form $[P,0]^T$, then $0$ solution; otherwise the solutions are $Y=[F,0]^T$ with $FF^T=I_n+PP^T=Wdiag(b_i)W^T$ where $Win O(n)$. Thus there are at least $2^n$ solutions: $X=beginpmatrixWdiag(pmsqrtb_i)W^T\0_m-n,nendpmatrix-A$.






share|cite|improve this answer






















  • What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
    – user103828
    Nov 21 '15 at 11:03











  • @ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
    – loup blanc
    Nov 21 '15 at 15:59










  • Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
    – user103828
    Nov 22 '15 at 18:39














up vote
1
down vote



accepted










Let $Y=X+A$; then $YY^T=diag(I_n,0_m-n)+AA^T=B$.



Since $rank(YY^T)leq n$, if we want that a solution exists, then necessarily $rank(B)leq n$.



EDIT 1. Since $AA^T$ is symmetric $geq 0$, $rank(B)geq n$ and we assume that $rank(B)=n$. Since $rank(AA^T)=n$, the spd symmetric matrix $AA^T$ is in the form $diag(U_n,0_m-n)$, where $U$ is a pd symmetric matrix. Finally we may assume that $B=diag(b_1,cdots,b_n,0_m-n)$ where $b_i>0$. We can choose $Y=beginpmatrixdiag(pmsqrtb_i)\0_m-n,nendpmatrix$.



EDIT 2. Answer to user. Necessarily, for any $y$, $[0_n,y^T]B[0_n,y]=0$; then $AA^T$ is in the form $beginpmatrixU_n&V\V^T&0_m-nendpmatrix$. Thus if $A=[P,Q]^T$,then $QQ^T=0$, that implies that $Q=0$. Finally $V$ is also a zero matrix and we are done.



Conclusion. If $A$ is not in the form $[P,0]^T$, then $0$ solution; otherwise the solutions are $Y=[F,0]^T$ with $FF^T=I_n+PP^T=Wdiag(b_i)W^T$ where $Win O(n)$. Thus there are at least $2^n$ solutions: $X=beginpmatrixWdiag(pmsqrtb_i)W^T\0_m-n,nendpmatrix-A$.






share|cite|improve this answer






















  • What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
    – user103828
    Nov 21 '15 at 11:03











  • @ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
    – loup blanc
    Nov 21 '15 at 15:59










  • Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
    – user103828
    Nov 22 '15 at 18:39












up vote
1
down vote



accepted







up vote
1
down vote



accepted






Let $Y=X+A$; then $YY^T=diag(I_n,0_m-n)+AA^T=B$.



Since $rank(YY^T)leq n$, if we want that a solution exists, then necessarily $rank(B)leq n$.



EDIT 1. Since $AA^T$ is symmetric $geq 0$, $rank(B)geq n$ and we assume that $rank(B)=n$. Since $rank(AA^T)=n$, the spd symmetric matrix $AA^T$ is in the form $diag(U_n,0_m-n)$, where $U$ is a pd symmetric matrix. Finally we may assume that $B=diag(b_1,cdots,b_n,0_m-n)$ where $b_i>0$. We can choose $Y=beginpmatrixdiag(pmsqrtb_i)\0_m-n,nendpmatrix$.



EDIT 2. Answer to user. Necessarily, for any $y$, $[0_n,y^T]B[0_n,y]=0$; then $AA^T$ is in the form $beginpmatrixU_n&V\V^T&0_m-nendpmatrix$. Thus if $A=[P,Q]^T$,then $QQ^T=0$, that implies that $Q=0$. Finally $V$ is also a zero matrix and we are done.



Conclusion. If $A$ is not in the form $[P,0]^T$, then $0$ solution; otherwise the solutions are $Y=[F,0]^T$ with $FF^T=I_n+PP^T=Wdiag(b_i)W^T$ where $Win O(n)$. Thus there are at least $2^n$ solutions: $X=beginpmatrixWdiag(pmsqrtb_i)W^T\0_m-n,nendpmatrix-A$.






share|cite|improve this answer














Let $Y=X+A$; then $YY^T=diag(I_n,0_m-n)+AA^T=B$.



Since $rank(YY^T)leq n$, if we want that a solution exists, then necessarily $rank(B)leq n$.



EDIT 1. Since $AA^T$ is symmetric $geq 0$, $rank(B)geq n$ and we assume that $rank(B)=n$. Since $rank(AA^T)=n$, the spd symmetric matrix $AA^T$ is in the form $diag(U_n,0_m-n)$, where $U$ is a pd symmetric matrix. Finally we may assume that $B=diag(b_1,cdots,b_n,0_m-n)$ where $b_i>0$. We can choose $Y=beginpmatrixdiag(pmsqrtb_i)\0_m-n,nendpmatrix$.



EDIT 2. Answer to user. Necessarily, for any $y$, $[0_n,y^T]B[0_n,y]=0$; then $AA^T$ is in the form $beginpmatrixU_n&V\V^T&0_m-nendpmatrix$. Thus if $A=[P,Q]^T$,then $QQ^T=0$, that implies that $Q=0$. Finally $V$ is also a zero matrix and we are done.



Conclusion. If $A$ is not in the form $[P,0]^T$, then $0$ solution; otherwise the solutions are $Y=[F,0]^T$ with $FF^T=I_n+PP^T=Wdiag(b_i)W^T$ where $Win O(n)$. Thus there are at least $2^n$ solutions: $X=beginpmatrixWdiag(pmsqrtb_i)W^T\0_m-n,nendpmatrix-A$.







share|cite|improve this answer














share|cite|improve this answer



share|cite|improve this answer








edited Nov 23 '15 at 0:15

























answered Nov 20 '15 at 17:43









loup blanc

20.6k21649




20.6k21649











  • What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
    – user103828
    Nov 21 '15 at 11:03











  • @ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
    – loup blanc
    Nov 21 '15 at 15:59










  • Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
    – user103828
    Nov 22 '15 at 18:39
















  • What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
    – user103828
    Nov 21 '15 at 11:03











  • @ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
    – loup blanc
    Nov 21 '15 at 15:59










  • Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
    – user103828
    Nov 22 '15 at 18:39















What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
– user103828
Nov 21 '15 at 11:03





What about @D.Thomine comment? with $A=(0,1)'$ in which case $YY'=diag(1,0)$ but $diag(1,0)+B=(1,0;1,0)$.... I think the issue is that $AA^T$ can also be $diag(0_m-n,U_n)$. So maybe the statement holds as long as $AA^T=diag(U_n,0_m-n)$?
– user103828
Nov 21 '15 at 11:03













@ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
– loup blanc
Nov 21 '15 at 15:59




@ user103828 , you are wrong; indeed the Thomine's example has no solutions (as he wanted) because $B=I_2$. See my second edit.
– loup blanc
Nov 21 '15 at 15:59












Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
– user103828
Nov 22 '15 at 18:39




Ok. I see. Then, I don't quite understand your solution. Does your solution prove that there is always an $A$ that solves the equation?
– user103828
Nov 22 '15 at 18:39

















 

draft saved


draft discarded















































 


draft saved


draft discarded














StackExchange.ready(
function ()
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f1538425%2fsolve-xaxa-mathrmdiagi-0aa%23new-answer', 'question_page');

);

Post as a guest













































































這個網誌中的熱門文章

How to combine Bézier curves to a surface?

Mutual Information Always Non-negative

Why am i infinitely getting the same tweet with the Twitter Search API?