Megasecond: Difference between revisions

Revision as of 18:24, 17 December 2013

Divide-and-conquer eigenvalue algorithms are a class of eigenvalue algorithms for Hermitian or real symmetric matrices that have recently (circa 1990s) become competitive in terms of stability and efficiency with more traditional algorithms such as the QR algorithm. The basic concept behind these algorithms is the divide-and-conquer approach from computer science. An eigenvalue problem is divided into two problems of roughly half the size, each of these are solved recursively, and the eigenvalues of the original problem are computed from the results of these smaller problems.

Here we present the simplest version of a divide-and-conquer algorithm, similar to the one originally proposed by Cuppen in 1981. Many details that lie outside the scope of this article will be omitted; however, without considering these details, the algorithm is not fully stable.

Background

As with most eigenvalue algorithms for Hermitian matrices, divide-and-conquer begins with a reduction to tridiagonal form. For an $m \times m$ matrix, the standard method for this, via Householder reflections, takes $\frac{4}{3} m^{3}$ flops, or $\frac{8}{3} m^{3}$ if eigenvectors are needed as well. There are other algorithms, such as the Arnoldi iteration, which may do better for certain classes of matrices; we will not consider this further here.

In certain cases, it is possible to deflate an eigenvalue problem into smaller problems. Consider a block diagonal matrix

T = [\begin{matrix} T_{1} & 0 \\ 0 & T_{2} \end{matrix}] .

The eigenvalues and eigenvectors of $T$ are simply those of $T_{1}$ and $T_{2}$ , and it will almost always be faster to solve these two smaller problems than to solve the original problem all at once. This technique can be used to improve the efficiency of many eigenvalue algorithms, but it has special significance to divide-and-conquer.

For the rest of this article, we will assume the input to the divide-and-conquer algorithm is an $m \times m$ real symmetric tridiagonal matrix $T$ . Although the algorithm can be modified for Hermitian matrices, we do not give the details here.

Divide

The divide part of the divide-and-conquer algorithm comes from the realization that a tridiagonal matrix is "almost" block diagonal.

File:Almost block diagonal.png

The size of submatrix $T_{1}$ we will call $n \times n$ , and then $T_{2}$ is $(m - n) \times (m - n)$ . Note that the remark about $T$ being almost block diagonal is true regardless of how $n$ is chosen (i.e., there are many ways to so decompose the matrix). However, it makes sense, from an efficiency standpoint, to choose $n \approx m / 2$ .

We write $T$ as a block diagonal matrix, plus a rank-1 correction:

File:Block diagonal plus correction.png

The only difference between $T_{1}$ and ${\hat{T}}_{1}$ is that the lower right entry $t_{n n}$ in ${\hat{T}}_{1}$ has been replaced with $t_{n n} - β$ and similarly, in ${\hat{T}}_{2}$ the top left entry $t_{n + 1, n + 1}$ has been replaced with $t_{n + 1, n + 1} - β$ .

The remainder of the divide step is to solve for the eigenvalues (and if desired the eigenvectors) of ${\hat{T}}_{1}$ and ${\hat{T}}_{2}$ , that is to find the diagonalizations ${\hat{T}}_{1} = Q_{1} D_{1} Q_{1}^{T}$ and ${\hat{T}}_{2} = Q_{2} D_{2} Q_{2}^{T}$ . This can be accomplished with recursive calls to the divide-and-conquer algorithm, although practical implementations often switch to the QR algorithm for small enough submatrices.

Conquer

The conquer part of the algorithm is the unintuitive part. Given the diagonalizations of the submatrices, calculated above, how do we find the diagonalization of the original matrix?

First, define $z^{T} = (q_{1}^{T}, q_{2}^{T})$ , where $q_{1}^{T}$ is the last row of $Q_{1}$ and $q_{2}^{T}$ is the first row of $Q_{2}$ . It is now elementary to show that

T = [\begin{matrix} Q_{1} \\ Q_{2} \end{matrix}] ([\begin{matrix} D_{1} \\ D_{2} \end{matrix}] + β z z^{T}) [\begin{matrix} Q_{1}^{T} \\ Q_{2}^{T} \end{matrix}]

The remaining task has been reduced to finding the eigenvalues of a diagonal matrix plus a rank-one correction. Before showing how to do this, let us simplify the notation. We are looking for the eigenvalues of the matrix $D + w w^{T}$ , where $D$ is diagonal with distinct entries and $w$ is any vector with nonzero entries.

If w_i is zero, ( $e_{i}$ ,d_i) is an eigenpair of $D + w w^{T}$ since $(D + w w^{T}) e_{i} = D e_{i} = d_{i} e_{i}$ .

If $λ$ is an eigenvalue, we have:

(D + w w^{T}) q = λ q

where $q$ is the corresponding eigenvector. Now

(D - λ I) q + w (w^{T} q) = 0

q + (D - λ I)^{- 1} w (w^{T} q) = 0

w^{T} q + w^{T} (D - λ I)^{- 1} w (w^{T} q) = 0

Keep in mind that $w^{T} q$ is a nonzero scalar. Neither $w$ nor $q$ are zero. If $w^{T} q$ were to be zero, $q$ would be an eigenvector of $D$ by $(D + w w^{T}) q = λ q$ . If that were the case, $q$ would contain only one nonzero position since $D$ is distinct diagonal and thus the inner product $w^{T} q$ can not be zero after all. Therefore, we have:

1 + w^{T} (D - λ I)^{- 1} w = 0

or written as a scalar equation,

1 + \sum_{j = 1}^{m} \frac{w_{j}^{2}}{d_{j} - λ} = 0.

This equation is known as the secular equation. The problem has therefore been reduced to finding the roots of the rational function defined by the left-hand side of this equation.

All general eigenvalue algorithms must be iterative, and the divide-and-conquer algorithm is no different. Solving the nonlinear secular equation requires an iterative technique, such as the Newton–Raphson method. However, each root can be found in O(1) iterations, each of which requires $Θ (m)$ flops (for an $m$ -degree rational function), making the cost of the iterative part of this algorithm $Θ (m^{2})$ .

Analysis

As is common for divide and conquer algorithms, we will use the Master theorem to analyze the running time. Remember that above we stated we choose $n \approx m / 2$ . We can write the recurrence relation:

T (m) = 2 \times T (\frac{m}{2}) + Θ (m^{2})

In the notation of the Master theorem, $a = b = 2$ and thus $\log_{b} a = 1$ . Clearly, $Θ (m^{2}) = Ω (m^{1})$ , so we have

T (m) = Θ (m^{2})

Remember that above we pointed out that reducing a Hermitian matrix to tridiagonal form takes $\frac{4}{3} m^{3}$ flops. This dwarfs the running time of the divide-and-conquer part, and at this point it is not clear what advantage the divide-and-conquer algorithm offers over the QR algorithm (which also takes $Θ (m^{2})$ flops for tridiagonal matrices).

The advantage of divide-and-conquer comes when eigenvectors are needed as well. If this is the case, reduction to tridiagonal form takes $\frac{8}{3} m^{3}$ , but the second part of the algorithm takes $Θ (m^{3})$ as well. For the QR algorithm with a reasonable target precision, this is $\approx 6 m^{3}$ , whereas for divide-and-conquer it is $\approx \frac{4}{3} m^{3}$ . The reason for this improvement is that in divide-and-conquer, the $Θ (m^{3})$ part of the algorithm (multiplying $Q$ matrices) is separate from the iteration, whereas in QR, this must occur in every iterative step. Adding the $\frac{8}{3} m^{3}$ flops for the reduction, the total improvement is from $\approx 9 m^{3}$ to $\approx 4 m^{3}$ flops.

Practical use of the divide-and-conquer algorithm has shown that in most realistic eigenvalue problems, the algorithm actually does better than this. The reason is that very often the matrices $Q$ and the vectors $z$ tend to be numerically sparse, meaning that they have many entries with values smaller than the floating point precision, allowing for numerical deflation, i.e. breaking the problem into uncoupled subproblems.

Variants and implementation

The algorithm presented here is the simplest version. In many practical implementations, more complicated rank-1 corrections are used to guarantee stability; some variants even use rank-2 corrections.Potter or Ceramic Artist Truman Bedell from Rexton, has interests which include ceramics, best property developers in singapore developers in singapore and scrabble. Was especially enthused after visiting Alejandro de Humboldt National Park.

There exist specialized root-finding techniques for rational functions that may do better than the Newton-Raphson method in terms of both performance and stability. These can be used to improve the iterative part of the divide-and-conquer algorithm.

The divide-and-conquer algorithm is readily parallelized, and linear algebra computing packages such as LAPACK contain high-quality parallel implementations.

References

Many property agents need to declare for the PIC grant in Singapore. However, not all of them know find out how to do the correct process for getting this PIC scheme from the IRAS. There are a number of steps that you need to do before your software can be approved.

Naturally, you will have to pay a safety deposit and that is usually one month rent for annually of the settlement. That is the place your good religion deposit will likely be taken into account and will kind part or all of your security deposit. Anticipate to have a proportionate amount deducted out of your deposit if something is discovered to be damaged if you move out. It's best to you'll want to test the inventory drawn up by the owner, which can detail all objects in the property and their condition. If you happen to fail to notice any harm not already mentioned within the inventory before transferring in, you danger having to pay for it yourself.

In case you are in search of an actual estate or Singapore property agent on-line, you simply should belief your intuition. It's because you do not know which agent is nice and which agent will not be. Carry out research on several brokers by looking out the internet. As soon as if you end up positive that a selected agent is dependable and reliable, you can choose to utilize his partnerise in finding you a home in Singapore. Most of the time, a property agent is taken into account to be good if he or she locations the contact data on his website. This may mean that the agent does not mind you calling them and asking them any questions relating to new properties in singapore in Singapore. After chatting with them you too can see them in their office after taking an appointment.

Have handed an trade examination i.e Widespread Examination for House Brokers (CEHA) or Actual Property Agency (REA) examination, or equal; Exclusive brokers are extra keen to share listing information thus making certain the widest doable coverage inside the real estate community via Multiple Listings and Networking. Accepting a severe provide is simpler since your agent is totally conscious of all advertising activity related with your property. This reduces your having to check with a number of agents for some other offers. Price control is easily achieved. Paint work in good restore-discuss with your Property Marketing consultant if main works are still to be done. Softening in residential property prices proceed, led by 2.8 per cent decline within the index for Remainder of Central Region

Once you place down the one per cent choice price to carry down a non-public property, it's important to accept its situation as it is whenever you move in – faulty air-con, choked rest room and all. Get round this by asking your agent to incorporate a ultimate inspection clause within the possibility-to-buy letter. HDB flat patrons routinely take pleasure in this security net. "There's a ultimate inspection of the property two days before the completion of all HDB transactions. If the air-con is defective, you can request the seller to repair it," says Kelvin.

15.6.1 As the agent is an intermediary, generally, as soon as the principal and third party are introduced right into a contractual relationship, the agent drops out of the image, subject to any problems with remuneration or indemnification that he could have against the principal, and extra exceptionally, against the third occasion. Generally, agents are entitled to be indemnified for all liabilities reasonably incurred within the execution of the brokers´ authority.

To achieve the very best outcomes, you must be always updated on market situations, including past transaction information and reliable projections. You could review and examine comparable homes that are currently available in the market, especially these which have been sold or not bought up to now six months. You'll be able to see a pattern of such report by clicking here It's essential to defend yourself in opposition to unscrupulous patrons. They are often very skilled in using highly unethical and manipulative techniques to try and lure you into a lure. That you must also protect your self, your loved ones, and personal belongings as you'll be serving many strangers in your home. Sign a listing itemizing of all of the objects provided by the proprietor, together with their situation. HSR Prime Recruiter 2010.

Template:Numerical linear algebra

@@ Line 1: / Line 1: @@
-== Polo Ralph Lauren Femme Et Aussie John McBeath ==
+'''Divide-and-conquer eigenvalue algorithms''' are a class of [[eigenvalue algorithm]]s for [[Hermitian matrix|Hermitian]] or [[real number|real]] [[Symmetric matrix|symmetric matrices]] that have recently (circa 1990s) become competitive in terms of [[Numerical stability|stability]] and [[Computational complexity theory|efficiency]] with more traditional algorithms such as the [[QR algorithm]].  The basic concept behind these algorithms is the [[Divide and conquer algorithm|divide-and-conquer]] approach from [[computer science]].  An [[eigenvalue]] problem is divided into two problems of roughly half the size, each of these are solved [[Recursion|recursively]], and the eigenvalues of the original problem are computed from the results of these smaller problems.
-Market Street.) Voir Boston L après discussion T Wharf.. Un vieux de plusieurs siècles statue grotesque d'un vieil homme est à moitié enterré quelques mètres. Pommes poussent sur de petits arbres à feuilles caduques. Il suffit de télécharger votre conception et le contrôle en ligne. <br><br>Et Aussie John McBeath, courir de nouveaux rêves de faire des affaires dans l'ancien paradis hippie de Goa, craint à quelle vitesse cela peut arriver qu'il se querelle avec l'Inde maladroit bureaucratie et rencontre l'étrange et souvent inquiétant, tout en essayant de sauver une relation. <br><br>[15] vagues plus en plus de colons vietnamiens, dont le royaume du Cambodge ne pouvait pas empêcher, car il a été affaibli par la guerre avec la Thaïlande, lentement . Un geste audacieux, mais avec beaucoup de grands [http://www.allogateaux.fr/albumpiecesmontees/slides/public.html Polo Ralph Lauren Femme] blocs de bureau à proximité d'écureuil rouge pourrait être juste ce que les besoins de la région.. <br><br>Tout cela serait faire est de mettre les voisins en guerre avec un autre. [http://www.allogateaux.fr/piecesmontees/album/gallery.html Ray Ban Clubmaster] Alors maintenant, il continue à tomber, à la fois en raison de la crise en Europe et qui sait quoi d'autre. Miaulements. Aujourd'hui, il vit à Los Angeles et est plus susceptible d'être trouvé surfer sur les vagues de [http://www.stadt-land.ch/jcarousel/images/banner.asp Ray Ban Suisse] coincé à l'intérieur d'une salle de sport: "Je suis la personne la plus malsain dans le monde.". <br><br>Les élèves entendent parler de comment la pollution peut être dévastateur pour la nature. Les gens qui me connaissent déjà auront eu un sentiment croissant d'appréhension tout de leurs propres comme ils lisent ceci. En outre, un corps robuste et de plus en plus de recherches montrent clairement que la vitamine D est absolument essentiel pour la bonne santé et la prévention des maladies. <br><br>Vivre des filles sexy sur la came répondre à vos demandes et directions vous met dans le contrôle et permet à vos fantasmes se déchaînent comme les filles des réponses et des discussions avec vous, répond à vos demandes, et réalise quoi que ce soit, elle est prête à faire ce que vous lui diriger. <br><br>Entièrement développé, le réseau de chauffage par conséquent d'économiser environ 300 000 chaque année. Vouloir laisser [les femmes dans le public] connaître OK si elles avaient des avortements. Développé documents de meilleures pratiques pour les [http://www.societelibre-eure.org/config/comment.php Ceinture Louis Vuitton Pas Cher] médias sociaux, le renforcement des liens et l'optimisation de conversion. <br><br>Ils ont fait leur argent et surtout gardé presque entièrement avec les détaillants nationaux solvables Les gens de D 'accord n'a pas couper les coins ronds, mais négocié offres intelligentes avant de se passer à la construction. Instagram de commencer à partager des données utilisateur avec Facebook.<ul>
+Here we present the simplest version of a divide-and-conquer algorithm, similar to the one originally proposed by Cuppen in 1981.  Many details that lie outside the scope of this article will be omitted; however, without considering these details, the algorithm is not fully stable.
-   <li>[http://oscarbernie.com/boards/profile.php?id=38413 http://oscarbernie.com/boards/profile.php?id=38413]</li>
-   <li>[http://www.thomassankara.net/spip.php?article605&date=2008-05/ http://www.thomassankara.net/spip.php?article605&date=2008-05/]</li>
-   <li>[http://enseignement-lsf.com/spip.php?article64#forum18089638 http://enseignement-lsf.com/spip.php?article64#forum18089638]</li>
-   <li>[http://scottalanciolek.com/Main_Page http://scottalanciolek.com/Main_Page]</li>
-   <li>[http://enseignement-lsf.com/spip.php?article64#forum18069912 http://enseignement-lsf.com/spip.php?article64#forum18069912]</li>
- </ul>
-== Nike Air Max Note de la courbe dans le bas du dos / lombaire ==
+==Background==
+As with most eigenvalue algorithms for Hermitian matrices, divide-and-conquer begins with a reduction to [[Tridiagonal matrix|tridiagonal]] form.  For an <math>m \times m</math> matrix, the standard method for this, via [[Householder reflection]]s, takes <math>\frac{4}{3}m^{3}</math> [[flops]], or <math>\frac{8}{3}m^{3}</math> if [[eigenvector]]s are needed as well.  There are other algorithms, such as the [[Arnoldi iteration]], which may do better for certain classes of matrices; we will not consider this further here.
-Si vous ne voulez pas aller trouver un de ces stands d'information, alors la plupart des hôtels de la région auront également une belle sélection de informations touristiques et Guide de Voyage brochures. Ils varient en taille de moins d'un quart de pouce à plusieurs pouces de diamètre. <br><br>Le mécanisme derrière vertige positionnel bénin ou VPPB est peu de débris à l'intérieur de l'oreille interne. En déverrouillant ces obstacles, les professeurs comme M. Nous offrons les services de gestion à des tarifs très raisonnables. Rappelez-vous vos concurrents sont littéralement à [http://www.stierlihof.ch/hofladen/weine/define.asp Nike Air Max] portée de clic, riches manchettes de mots clés si forts et contenu formidable aideront à convertir les visiteurs en acheteurs. <br><br>Nous testons en interne jusqu'à ce que nous nous sentons toutes les fonctionnalités du site ont été conservés, aucune donnée n'a été perdue, et le nouveau site est suffisamment robuste pour aller vivre, je l'espère un peu de temps en Novembre. Sur l'icône, vous devriez voir le mot "file d'attente" avec une barre en dessous. <br><br>Samedi, j'ai rempli une carte de commentaires et j'ai appelé le centre de loisirs dès lundi matin pour parler au directeur de la piscine. Il a déménagé à Saint-Pétersbourg et est devenu un révolutionnaire professionnel. Une excellente approche pour en apprendre davantage sur l'optimisation de la page off est tout simplement en allant à autant d'ateliers et des séminaires sur le marketing des moteurs de recherche que possible. <br><br>INVESTISSEURS: Deutsche Börse et NYSE Euronext ont fait beaucoup de l'efficacité à être [http://www.elektrosmets.be/printable/assets/header.asp Hollister Anvers] arrachées [http://www.jubla-so.ch/scripts/images/simple.asp Vibram Suisse] de la combinaison de leurs marchés, et la transparence améliorée à la disposition des chiens de garde du marché. Si vous postez ici, ne postez pas là. Initialement concentrée au centre-ville le long de Pennsylvania Avenue, Chinatown propose plusieurs blocs au nord pour faire place à l'achèvement du complexe de bureaux Triangle fédéral dans les années 1930. <br><br>Après la planification, la recherche et la conception, l'ensemble du projet a pris quatre semaines. C'est une bonne idée de commencer avec "sait quoi, pourquoi ne pas [http://www.opzrekem.be/scripts/search/editor.asp Michael Kors Bruxelles] aller trouver (sa maman / papa / tuteur) et nous pouvons tous parler de la raison pour laquelle X n'est pas ici". <br><br>Sérieusement, la plupart des amateurs de bricolage et les constructeurs achètent seulement UPE détail, cartes mères et cartes graphiques. Je viens de formation, debout à la plate-forme en bois, flottant sur la rivière du démon Village. Ils connaissent nos NOM. <br><br>99% sûr que je ne suis pas enceinte''. La région où j'habite a une petite localité de adjecent peuplée par une trentaine de personnes bizarres (femmes includung et childern) vivant dans environ neuf maisons. (Note de la courbe dans le bas du dos / lombaire) Le haut du corps devrait soulever juste assez pour arriver à ce point.<ul>
+In certain cases, it is possible to ''deflate'' an eigenvalue problem into smaller problems.  Consider a [[block diagonal matrix]]
+:<math>T = \begin{bmatrix} T_{1} & 0 \\ 0 & T_{2}\end{bmatrix}.</math>
-   <li>[http://www.polyhb.com/bbs/showtopic-500806.aspx http://www.polyhb.com/bbs/showtopic-500806.aspx]</li>
+The eigenvalues and eigenvectors of <math>T</math> are simply those of <math>T_{1}</math> and <math>T_{2}</math>, and it will almost always be faster to solve these two smaller problems than to solve the original problem all at once.  This technique can be used to improve the efficiency of many eigenvalue algorithms, but it has special significance to divide-and-conquer.
-   <li>[http://team-impressive.de/index.php?site=news_comments&newsID=148 http://team-impressive.de/index.php?site=news_comments&newsID=148]</li>
-   <li>[http://www.bissogaleto.it/e107_plugins/forum/forum_viewtopic.php?72881.last http://www.bissogaleto.it/e107_plugins/forum/forum_viewtopic.php?72881.last]</li>
-   <li>[http://nntpdirect.com/public-nntp-server/profile.php?id=319706 http://nntpdirect.com/public-nntp-server/profile.php?id=319706]</li>
-   <li>[http://studysupport.biz/ http://studysupport.biz/]</li>
-  </ul>
-== Michael Kors Sac  à la fin de la saison de Coupe du monde ==
+For the rest of this article, we will assume the input to the divide-and-conquer algorithm is an <math>m \times m</math> real symmetric tridiagonal matrix <math>T</math>.  Although the algorithm can be modified for Hermitian matrices, we do not give the details here.
-Chaque jour, de nombreux sites hébergés sur l'Internet. En Février, Wendy K. Les autres équipes qui sont là-bas, les frais moins de coûts de Voyage, ils devront payer trop.. Il pas un guide parfait par tous les moyens, mais il leur permet de saisir l'essentiel le meilleur équipement à l'échelle dans le jeu à un stade précoce. <br><br>J'apprécie générer mes visiteurs se sentent important et unique. Balises capitaine d'urgence peuvent être activées manuellement ou automatiquement soit par être frappé ou de frapper l'eau. La phase publique de l'œuvre a été lancé dans chacune des zones du projet sept en mai 2009 et s'est poursuivie jusqu'en 2012. <br><br>Un comité de la conférence convoquée pour marteler les différences entre le Sénat et la Chambre des budgets, puis les 140 législateurs se prononcera sur le budget. Tout d'abord, cette approche nécessite une interface utilisateur pour capturer, de créer et d'administrer des tables physiques réels dans l'entrepôt de données. <br><br>D'tre trangle. Elle a également conduit à des tentatives moins bien médiatisés à la fabrication PageRank (et Matt a répondu à cette question ainsi en disant que le système n'est pas aussi simple que cela). Vos informations de connexion sera envoyer par la poste. <br><br>Nous rencontrons frères Anya qui sont heureux que les gens la voient pour son talent plus que sa beauté. L'événement intitulé «Creative transformations de Trauma" comprendra des conférences et des [http://www.opzrekem.be/scripts/search/editor.asp Michael Kors Sac] spectacles d'artistes de renom qui ont de même choisi d'explorer les expériences traumatisantes de leur travail. <br><br>Les gens qui participent d'un concours de référencement vont essayer de [http://www.allogateaux.fr/piecesmontees/album/gallery.html Ray Ban Aviator] gagner par la construction d'un grand nombre de liens et le réseau de comparaître en haut moteur de recherche. Cependant, certaines applications Web sont trop gros pour être exécuté directement à partir du navigateur, et exigent des clients du côté de l'utilisateur de tenir une partie ou la plupart des informations, graphiques et autres données. <br><br>Peut-être mon deuxième contrainte est de vérifier les choses beaucoup, et j'aime vérifier les choses par trois. Finalement, à la fin de la saison de Coupe du monde, j'ai [http://www.allogateaux.fr/albumpiecesmontees/slides/public.html Polo Ralph Lauren Pas Cher] fait exactement $ 0 de sorte que si il ya [http://www.canige-constancia.org/database/upload.asp Nike Soldes] un moyen d'augmenter le trafic, je grattage ou le bradage du domaine sur flippa. <br><br>Mais revenons à l'objet de votre journal. Le groupe a décidé que le risque l'emporte sur l'intérêt et nous étions sur le point de tourner autour quand l'un des explorateurs repéré une chouette rayée. Pourquoi vos clients vous quittent avant même d'avoir started.How à lancer un site web avec un SystemHow de gestion de contenu CMS pour démarrer une WebsiteTestimonialsI recommande vivement Web Pro Aide! Lori m'a aidé à faire le plus de mon site Blogger..<ul>
+==Divide==
-   <li>[http://www.zs58tc.com/news/html/?73405.html http://www.zs58tc.com/news/html/?73405.html]</li>
-   <li>[http://enseignement-lsf.com/spip.php?article64#forum18086220 http://enseignement-lsf.com/spip.php?article64#forum18086220]</li>
-   <li>[http://pedagogie-differenciee.eu/spip.php?page=auteur&id_auteur=1&lang=fr/ http://pedagogie-differenciee.eu/spip.php?page=auteur&id_auteur=1&lang=fr/]</li>
-   <li>[http://www.ovnprod.com/spip.php?article24/ http://www.ovnprod.com/spip.php?article24/]</li>
-   <li>[http://forum.isped.u-bordeaux2.fr/Alzheimer.aspx?g=posts&t=301079 http://forum.isped.u-bordeaux2.fr/Alzheimer.aspx?g=posts&t=301079]</li>
- </ul>
-== Nike A Talon S/2013 / 354 ==
+The ''divide'' part of the divide-and-conquer algorithm comes from the realization that a tridiagonal matrix is "almost" block diagonal.
+<!-- For original TeX, see image description page -->
+:[[Image:Almost block diagonal.png]]
-Ils peuvent se produire à plusieurs endroits dans le tractus gastro-intestinal mais sont plus fréquentes dans le côlon. J'ai eu un client qui avait une belle gros bouton rouge et il était très sensible, mais un poussais, parce qu'il était rouge. Si vous cherchez un bon site pour écrire pour, ce qu'il est. <br><br>Son chose la plus favori-à-dire ces jours-ci est "uh oh", et que vous se dit beaucoup ces jours-ci. Si vous êtes juste un joueur régulier de vous voulez un repère qui a une durée de vie plus longue parce que vous ne voulez pas garder l'argent de poche sur votre queue. <br><br>Pour les utilisateurs de première fois, je vous recommande d'assister aux réunions. Il renforce également le système immunitaire par le traitement de presque toutes les maladies important associé avec le coeur et les poumons. Morton a déclaré que le service d'incendie de Cle Elum est financé de façon adéquate, que si les électeurs approuvent don TMT un service d'incendie percevoir le département sera en danger d'être arrêté. <br><br>Même sentiments Lane évolué comme elle a enregistré ces femmes histoires. Le Sultanat du Gujarat est resté indépendant jusqu'en 1576, quand l'empereur moghol Akbar le Grand a conquis et annexé à l'Empire moghol. Pas de sélection d'un outil, de [http://www.aldes-international.com/inettuts/international.html Nike A Talon] changer quelques articles autour de la main, sauver, exporter, ajout, de donner aux gens le nouveau nom de fichier, etc HTML / CSS est construit pour le prototypage itératif rapide tandis que Photoshop ISN est maladroit. <br><br>J'avais complètement oublié sur le site, et a décidé de revenir en arrière. Conseils d'un médecin pour perdre du [http://www.tourisme-in-france.com/database/file.asp Ceinture Gucci Pas Cher] poids sans l'aide et les ressources intensive n'est pas efficace. Greasemonkey affiche une boîte de dialogue intitulée "script d'installation [http://www.reuzepas.be/include/class.asp Hollister Bruxelles] de l'utilisateur," l'affichage du nom du script que vous vous apprêtez à installer (Butler), une brève description de ce que fait le script, et une liste de pages inclus et exclus. <br><br>Cette rébellion a conduit à l'entrée dans la Confédération canadienne du Manitoba comme [http://www.portobelloleuven.be/test/files.asp Nike Blazer Bruxelles] cinquième province du Canada en 1870. La seule réponse était une réponse e par la poste de la Colline en indiquant: sont un groupe de contribuables de Richmond Hill concernés par la façon dont ce conseil municipal gère le budget de la ville et comment ils traitent les résidents à la recherche de réponses. <br><br>L'augmentation de la consommation conduit à une augmentation dans le nombre total de calories brûlées dans votre séance d'exercice. Le Conseil de sécurité devrait adopter une déclaration présidentielle de demain (12 Décembre) sur la région du Sahel suite à une séance d'information par l'Envoyé spécial du Secrétaire général pour le Sahel, Romano Prodi, sur la mise en œuvre de la stratégie intégrée de l'ONU pour le Sahel (S/2013 / 354).<ul>
+The size of submatrix <math>T_{1}</math> we will call <math>n \times n</math>, and then <math>T_{2}</math> is <math>(m - n) \times (m - n)</math>.  Note that the remark about <math>T</math> being almost block diagonal is true regardless of how <math>n</math> is chosen (i.e., there are many ways to so decompose the matrix).  However, it makes sense, from an efficiency standpoint, to choose <math>n \approx m/2</math>.
-   <li>[http://202.109.115.218:8080/read.php?tid=5407338 http://202.109.115.218:8080/read.php?tid=5407338]</li>
+We write <math>T</math> as a block diagonal matrix, plus a [[Rank (linear algebra)|rank-1]] correction:
+<!-- For original TeX, see image description page -->
-   <li>[http://www.histoirepassion.eu/spip.php?article7/ http://www.histoirepassion.eu/spip.php?article7/]</li>
+:[[Image:Block diagonal plus correction.png]]
-   <li>[http://lab.nqnwebs.com/lavoz_bak/spip.php?article5818/ http://lab.nqnwebs.com/lavoz_bak/spip.php?article5818/]</li>
+The only difference between <math>T_{1}</math> and <math>\hat{T}_{1}</math> is that the lower right entry <math>t_{nn}</math> in <math>\hat{T}_{1}</math> has been replaced with <math>t_{nn} - \beta</math> and similarly, in <math>\hat{T}_{2}</math> the top left entry <math>t_{n+1,n+1}</math> has been replaced with <math>t_{n+1,n+1} - \beta</math>.
-   <li>[http://enseignement-lsf.com/spip.php?article65#forum18456790 http://enseignement-lsf.com/spip.php?article65#forum18456790]</li>
+The remainder of the divide step is to solve for the eigenvalues (and if desired the eigenvectors) of <math>\hat{T}_{1}</math> and <math>\hat{T}_{2}</math>, that is to find the [[diagonalizable matrix|diagonalization]]s <math>\hat{T}_{1} = Q_{1} D_{1} Q_{1}^{T}</math> and <math>\hat{T}_{2} = Q_{2} D_{2} Q_{2}^{T}</math>.  This can be accomplished with recursive calls to the divide-and-conquer algorithm, although practical implementations often switch to the QR algorithm for small enough submatrices.
-   <li>[http://passerelle.ethiopie.free.fr/spip.php?article448/ http://passerelle.ethiopie.free.fr/spip.php?article448/]</li>
+==Conquer==
-  </ul>
+The ''conquer'' part of the algorithm is the unintuitive part.  Given the diagonalizations of the submatrices, calculated above, how do we find the diagonalization of the original matrix?
+First, define <math>z^{T} = (q_{1}^{T},q_{2}^{T})</math>, where <math>q_{1}^{T}</math> is the last row of <math>Q_{1}</math> and <math>q_{2}^{T}</math> is the first row of <math>Q_{2}</math>.  It is now elementary to show that
+:<math>T = \begin{bmatrix} Q_{1} & \\ & Q_{2} \end{bmatrix} \left( \begin{bmatrix} D_{1} & \\ & D_{2} \end{bmatrix} + \beta z z^{T} \right) \begin{bmatrix} Q_{1}^{T} & \\ & Q_{2}^{T} \end{bmatrix}</math>
+The remaining task has been reduced to finding the eigenvalues of a diagonal matrix plus a rank-one correction.  Before showing how to do this, let us simplify the notation.  We are looking for the eigenvalues of the matrix <math>D + w w^{T}</math>, where <math>D</math> is diagonal with distinct entries and <math>w</math> is any vector with nonzero entries.
+If w<sub>i</sub> is zero, (<math>e_i</math>,d<sub>i</sub>) is an eigenpair of <math>D + w w^{T}</math> since
+<math>(D + w w^{T})e_i = De_i = d_i e_i</math>.
+If <math>\lambda</math> is an eigenvalue, we have:
+:<math>(D + w w^{T})q = \lambda q</math>
+where <math>q</math> is the corresponding eigenvector.  Now
+:<math>(D - \lambda I)q + w(w^{T}q) = 0</math>
+:<math>q + (D - \lambda I)^{-1} w(w^{T}q) = 0</math>
+:<math>w^{T}q + w^{T}(D - \lambda I)^{-1} w(w^{T}q) = 0</math>
+Keep in mind that <math>w^{T}q</math> is a nonzero scalar. Neither <math>w</math> nor <math>q</math> are zero. If <math>w^{T}q</math> were to be zero, <math>q</math> would be an eigenvector of <math>D</math> by <math>(D + w w^{T})q = \lambda q</math>. If that were the case, <math>q</math> would contain only one nonzero position since <math>D</math> is distinct diagonal and thus the inner product <math>w^{T}q</math> can not be zero after all. Therefore, we have:
+:<math>1 + w^{T}(D - \lambda I)^{-1} w = 0</math>
+or written as a scalar equation,
+:<math>1 + \sum_{j=1}^{m} \frac{w_{j}^{2}}{d_{j} - \lambda} = 0.</math>
+This equation is known as the ''secular equation''. The problem has therefore been reduced to finding the roots of the [[rational function]] defined by the left-hand side of this equation.
+All general eigenvalue algorithms must be iterative, and the divide-and-conquer algorithm is no different.  Solving the [[nonlinear]] secular equation requires an iterative technique, such as the [[Newton's method|Newton–Raphson method]].  However, each root can be found in [[Big O notation|O]](1) iterations, each of which requires <math>\Theta(m)</math> flops (for an <math>m</math>-degree rational function), making the cost of the iterative part of this algorithm <math>\Theta(m^{2})</math>.
+==Analysis==
+As is common for divide and conquer algorithms, we will use the [[Master theorem]] to analyze the running time.  Remember that above we stated we choose <math>n \approx m/2</math>.  We can write the [[recurrence relation]]:
+:<math>T(m) = 2 \times T\left(\frac{m}{2}\right) + \Theta(m^{2})</math>
+In the notation of the Master theorem, <math>a = b = 2</math> and thus <math>\log_{b} a = 1</math>.  Clearly, <math>\Theta(m^{2}) = \Omega(m^{1})</math>, so we have
+:<math>T(m) = \Theta(m^{2})</math>
+Remember that above we pointed out that reducing a Hermitian matrix to tridiagonal form takes <math>\frac{4}{3}m^{3}</math> flops.  This dwarfs the running time of the divide-and-conquer part, and at this point it is not clear what advantage the divide-and-conquer algorithm offers over the QR algorithm (which also takes <math>\Theta(m^{2})</math> flops for tridiagonal matrices).
+The advantage of divide-and-conquer comes when eigenvectors are needed as well.  If this is the case, reduction to tridiagonal form takes <math>\frac{8}{3}m^{3}</math>, but the second part of the algorithm takes <math>\Theta(m^{3})</math> as well.  For the QR algorithm with a reasonable target precision, this is <math>\approx 6 m^{3}</math>, whereas for divide-and-conquer it is <math>\approx \frac{4}{3}m^{3}</math>.  The reason for this improvement is that in divide-and-conquer, the <math>\Theta(m^{3})</math> part of the algorithm (multiplying <math>Q</math> matrices) is separate from the iteration, whereas in QR, this must occur in every iterative step.  Adding the <math>\frac{8}{3}m^{3}</math> flops for the reduction, the total improvement is from <math>\approx 9 m^{3}</math> to <math>\approx 4 m^{3}</math> flops.
+Practical use of the divide-and-conquer algorithm has shown that in most realistic eigenvalue problems, the algorithm actually does better than this.  The reason is that very often the matrices <math>Q</math> and the vectors <math>z</math> tend to be ''numerically sparse'', meaning that they have many entries with values smaller than the [[floating point]] precision, allowing for ''numerical deflation'', i.e. breaking the problem into uncoupled subproblems.
+==Variants and implementation==
+The algorithm presented here is the simplest version.  In many practical implementations, more complicated rank-1 corrections are used to guarantee stability; some variants even use rank-2 corrections.{{Citation needed|date=September 2011}}
+There exist specialized root-finding techniques for rational functions that may do better than the Newton-Raphson method in terms of both performance and stability.  These can be used to improve the iterative part of the divide-and-conquer algorithm.
+The divide-and-conquer algorithm is readily [[Parallel algorithm|parallelized]], and [[linear algebra]] computing packages such as [[LAPACK]] contain high-quality parallel implementations.
+==References==
+*{{citation
+ | last = Demmel | first = James W. | authorlink = James Demmel
+ | mr = 1463942
+ | isbn = 0-89871-389-7
+ | location = Philadelphia, PA
+ | publisher = [[Society for Industrial and Applied Mathematics]]
+ | title = Applied Numerical Linear Algebra
+  | year = 1997}}.
+{{Numerical linear algebra}}
+[[Category:Numerical linear algebra]]

Megasecond: Difference between revisions

76.31.235.64 (talk)

Revision as of 18:24, 17 December 2013

Contents

Background

Divide

Conquer

Analysis

Variants and implementation

References

Navigation menu

Megasecond: Difference between revisions

76.31.235.64 (talk)

Revision as of 18:24, 17 December 2013

Background

Divide

Conquer

Analysis

Variants and implementation

References

Navigation menu

Search