# Singular value decomposition

{{ safesubst:#invoke:Unsubst||$N=Use dmy dates |date=__DATE__ |$B= }}

In linear algebra, the **singular value decomposition** (**SVD**) is a factorization of a real or complex matrix, with many useful applications in signal processing and statistics.

Formally, the singular value decomposition of an *m* × *n* real or complex matrix **M** is a factorization of the form **M** = **UΣV**^{∗}, where **U** is an *m* × *m* real or complex unitary matrix, **Σ** is an *m* × *n* rectangular diagonal matrix with non-negative real numbers on the diagonal, and **V**^{∗} (the conjugate transpose of **V**, or simply the transpose of **V** if **V** is real) is an *n* × *n* real or complex unitary matrix. The diagonal entries **Σ**_{i,i} of **Σ** are known as the **singular values** of **M**. The Template:Mvar columns of **U** and the Template:Mvar columns of **V** are called the **left-singular vectors** and **right-singular vectors** of **M**, respectively.

The singular value decomposition and the eigendecomposition are closely related. Namely:

- The left-singular vectors of
**M**are eigenvectors of**MM**^{∗}. - The right-singular vectors of
**M**are eigenvectors of**M**^{∗}**M**. - The non-zero singular values of
**M**(found on the diagonal entries of**Σ**) are the square roots of the non-zero eigenvalues of both**M**^{∗}**M**and**MM**^{∗}.

- The left-singular vectors of

Applications that employ the SVD include computing the pseudoinverse, least squares fitting of data, matrix approximation, and determining the rank, range and null space of a matrix.

## Contents

- 1 Statement of the theorem
- 2 Intuitive interpretations
- 3 Example
- 4 Singular values, singular vectors, and their relation to the SVD
- 5 Applications of the SVD
- 6 Relation to eigenvalue decomposition
- 7 Existence
- 8 Geometric meaning
- 9 Calculating the SVD
- 10 Reduced SVDs
- 11 Norms
- 12 Tensor SVD
- 13 Bounded operators on Hilbert spaces
- 14 History
- 15 See also
- 16 Notes
- 17 References

## Statement of the theorem

Suppose **M** is a *m* × *n* matrix whose entries come from the field Template:Mvar, which is either the field of real numbers or the field of complex numbers. Then there exists a factorization of the form

where **U** is an *m* × *m* unitary matrix over Template:Mvar (orthogonal matrix if *K* = **R**), **Σ** is a *m* × *n* diagonal matrix with non-negative real numbers on the diagonal, and the *n* × *n* unitary matrix **V**^{∗} denotes the conjugate transpose of the *n* × *n* unitary matrix **V**. Such a factorization is called a singular value decomposition of **M**.

The diagonal entries Template:Mvar of **Σ** are known as the **singular values** of **M**. A common convention is to list the singular values in descending order. In this case, the diagonal matrix **Σ** is uniquely determined by **M** (though the matrices **U** and **V** are not).

## Intuitive interpretations

### Rotation, scaling

In the special, yet common case when **M** is an *m* × *m* real square matrix with positive determinant, **U**, **V**^{∗}, and **Σ** are real *m* × *m* matrices as well, **Σ** can be regarded as a scaling matrix, and **U**, **V**^{∗} can be viewed as rotation matrices. Thus the expression **UΣV**^{∗} can be intuitively interpreted as a composition (or sequence) of three geometrical transformations: a rotation, a scaling, and another rotation. For instance, the figure above explains how a shear matrix can be described as such a sequence.

### Singular values as semiaxes of an ellipse or ellipsoid

As shown in the figure, the singular values can be interpreted as the semiaxes of an ellipse in 2D. This concept can be generalized to Template:Mvar-dimensional Euclidean space, with the singular values of any *n* × *n* square matrix being viewed as the semiaxes of an Template:Mvar-dimensional ellipsoid. See below for further details.

### The columns of *U* and *V* are orthonormal bases

Since **U** and **V**^{∗} are unitary, the columns of each of them form a set of orthonormal vectors, which can be regarded as basis vectors. By the definition of a unitary matrix, the same is true for their conjugate transposes **U**^{∗} and **V**. In short, the columns of **U**, **U**^{∗}, **V**, and **V**^{∗} are orthonormal bases.

## Example

Consider the 4 × 5 matrix

A singular value decomposition of this matrix is given by **UΣV**^{∗}

Notice **Σ** is zero outside of the diagonal and one diagonal element is zero. Furthermore, because the matrices **U** and **V**^{∗} are unitary, multiplying by their respective conjugate transposes yields identity matrices, as shown below. In this case, because **U** and **V**^{∗} are real valued, they each are an orthogonal matrix.

This particular singular value decomposition is not unique. Choosing such that

is also a valid singular value decomposition.

## Singular values, singular vectors, and their relation to the SVD

A non-negative real number Template:Mvar is a **singular value** for **M** if and only if there exist unit-length vectors *u* in *K ^{m}* and

*v*in

*K*such that

^{n}The vectors *u* and *v* are called **left-singular** and **right-singular vectors** for Template:Mvar, respectively.

In any singular value decomposition

the diagonal entries of **Σ** are equal to the singular values of **M**. The columns of **U** and **V** are, respectively, left- and right-singular vectors for the corresponding singular values. Consequently, the above theorem implies that:

- An
*m*×*n*matrix**M**has at most*p*= min(*m*,*n*) distinct singular values. - It is always possible to find an orthogonal basis
**U**for Template:Mvar consisting of left-singular vectors of**M**. - It is always possible to find an orthogonal basis
**V**for Template:Mvar consisting of right-singular vectors of**M**.

A singular value for which we can find two left (or right) singular vectors that are linearly independent is called *degenerate*.

Non-degenerate singular values always have unique left- and right-singular vectors, up to multiplication by a unit-phase factor *e*^{iφ} (for the real case up to sign). Consequently, if all singular values of **M** are non-degenerate and non-zero, then its singular value decomposition is unique, up to multiplication of a column of **U** by a unit-phase factor and simultaneous multiplication of the corresponding column of **V** by the same unit-phase factor.

Degenerate singular values, by definition, have non-unique singular vectors. Furthermore, if *u*_{1} and *u*_{2} are two left-singular vectors which both correspond to the singular value σ, then any normalized linear combination of the two vectors is also a left-singular vector corresponding to the singular value σ. The similar statement is true for right-singular vectors. Consequently, if **M** has degenerate singular values, then its singular value decomposition is not unique.

## Applications of the SVD

### Pseudoinverse

The singular value decomposition can be used for computing the pseudoinverse of a matrix. Indeed, the pseudoinverse of the matrix **M** with singular value decomposition **M** = **UΣV**^{∗} is

where **Σ**^{+} is the pseudoinverse of **Σ**, which is formed by replacing every non-zero diagonal entry by its reciprocal and transposing the resulting matrix. The pseudoinverse is one way to solve linear least squares problems.

### Solving homogeneous linear equations

A set of homogeneous linear equations can be written as **Ax** = **0** for a matrix **A** and vector **x**. A typical situation is that **A** is known and a non-zero **x** is to be determined which satisfies the equation. Such an **x** belongs to **A**'s null space and is sometimes called a (right) null vector of **A**. The vector **x** can be characterized as a right-singular vector corresponding to a singular value of **A** that is zero. This observation means that if **A** is a square matrix and has no vanishing singular value, the equation has no non-zero **x** as a solution. It also means that if there are several vanishing singular values, any linear combination of the corresponding right-singular vectors is a valid solution. Analogously to the definition of a (right) null vector, a non-zero **x** satisfying **x**^{∗}**A** = **0**, with **x**^{∗} denoting the conjugate transpose of **x**, is called a left null vector of **A**.

### Total least squares minimization

A total least squares problem refers to determining the vector **x** which minimizes the 2-norm of a vector **Ax** under the constraint Template:!!**x**Template:!! = 1. The solution turns out to be the right-singular vector of **A** corresponding to the smallest singular value.

### Range, null space and rank

Another application of the SVD is that it provides an explicit representation of the range and null space of a matrix **M**. The right-singular vectors corresponding to vanishing singular values of **M** span the null space of **M**. E.g., the null space is spanned by the last two columns of **V** in the above example. The left-singular vectors corresponding to the non-zero singular values of **M** span the range of **M**. As a consequence, the rank of **M** equals the number of non-zero singular values which is the same as the number of non-zero diagonal elements in **Σ**.

In numerical linear algebra the singular values can be used to determine the *effective rank* of a matrix, as rounding error may lead to small but non-zero singular values in a rank deficient matrix.

### Low-rank matrix approximation

Some practical applications need to solve the problem of approximating a matrix **M** with another matrix , said truncated, which has a specific rank Template:Mvar. In the case that the approximation is based on minimizing the Frobenius norm of the difference between **M** and under the constraint that it turns out that the solution is given by the SVD of **M**, namely

where is the same matrix as **Σ** except that it contains only the Template:Mvar largest singular values (the other singular values are replaced by zero). This is known as the **Eckart–Young theorem**, as it was proved by those two authors in 1936 (although it was later found to have been known to earlier authors; see Template:Harvnb).

### Separable models

The SVD can be thought of as decomposing a matrix into a weighted, ordered sum of separable matrices. By separable, we mean that a matrix **A** can be written as an outer product of two vectors **A** = **u** ⊗ **v**, or, in coordinates, . Specifically, the matrix **M** can be decomposed as:

Here **U**_{i} and **V**_{i} are the Template:Mvar-th columns of the corresponding SVD matrices, Template:Mvar are the ordered singular values, and each **A**_{i} is separable. The SVD can be used to find the decomposition of an image processing filter into separable horizontal and vertical filters. Note that the number of non-zero Template:Mvar is exactly the rank of the matrix.

Separable models often arise in biological systems, and the SVD decomposition is useful to analyze such systems. For example, some visual area V1 simple cells' receptive fields can be well described^{[1]} by a Gabor filter in the space domain multiplied by a modulation function in the time domain. Thus, given a linear filter evaluated through, for example, reverse correlation, one can rearrange the two spatial dimensions into one dimension, thus yielding a two-dimensional filter (space, time) which can be decomposed through SVD. The first column of **U** in the SVD decomposition is then a Gabor while the first column of **V** represents the time modulation (or vice versa). One may then define an index of separability,

which is the fraction of the power in the matrix M which is accounted for by the first separable matrix in the decomposition.^{[2]}

### Nearest orthogonal matrix

It is possible to use the SVD of a square matrix **A** to determine the orthogonal matrix **O** closest to **A**. The closeness of fit is measured by the Frobenius norm of **O** − **A**. The solution is the product **UV**^{∗}.^{[3]} This intuitively makes sense because an orthogonal matrix would have the decomposition **UIV**^{∗} where **I** is the identity matrix, so that if **A** = **UΣV**^{∗} then the product **A** = **UV**^{∗} amounts to replacing the singular values with ones.

A similar problem, with interesting applications in shape analysis, is the orthogonal Procrustes problem, which consists of finding an orthogonal matrix **O** which most closely maps **A** to **B**. Specifically,

where denotes the Frobenius norm.

This problem is equivalent to finding the nearest orthogonal matrix to a given matrix **M** = **A**^{T}**B**.

### The Kabsch algorithm

The Kabsch algorithm (called Wahba's problem in other fields) uses SVD to compute the optimal rotation (with respect to least-squares minimization) that will align a set of points with a corresponding set of points. It is used, among other applications, to compare the structures of molecules.

### Signal processing

The SVD and pseudoinverse have been successfully applied to signal processing and big data, e.g., in genomic signal processing.^{[4]}^{[5]}^{[6]}^{[7]}

### Other examples

The SVD is also applied extensively to the study of linear inverse problems, and is useful in the analysis of regularization methods such as that of Tikhonov. It is widely used in statistics where it is related to principal component analysis and to Correspondence analysis, and in signal processing and pattern recognition. It is also used in output-only modal analysis, where the non-scaled mode shapes can be determined from the singular vectors. Yet another usage is latent semantic indexing in natural language text processing.

The SVD also plays a crucial role in the field of quantum information, in a form often referred to as the Schmidt decomposition. Through it, states of two quantum systems are naturally decomposed, providing a necessary and sufficient condition for them to be entangled: if the rank of the **Σ** matrix is larger than one.

One application of SVD to rather large matrices is in numerical weather prediction, where Lanczos methods are used to estimate the most linearly quickly growing few perturbations to the central numerical weather prediction over a given initial forward time period; i.e., the singular vectors corresponding to the largest singular values of the linearized propagator for the global weather over that time interval. The output singular vectors in this case are entire weather systems. These perturbations are then run through the full nonlinear model to generate an ensemble forecast, giving a handle on some of the uncertainty that should be allowed for around the current central prediction.

Another application of SVD for daily life is that point in perspective view can be unprojected in a photo using the calculated SVD matrix, this application leads to measuring length (a.k.a. the distance of two unprojected points in perspective photo) by marking out the 4 corner points of known-size object in a single photo. PRuler is a demo to implement this application by taking a photo of a regular credit card.

SVD has also been applied to reduced order modelling. The aim of reduced order modelling is to reduce the number of degrees of freedom in a complex system which is to be modelled. SVD was coupled with radial basis functions to interpolate solutions to three-dimensional unsteady flow problems.^{[8]}

SVD has also been applied in inverse problem theory. An important application is constructing computational models of oil reservoirs.^{[9]}

Singular value decomposition is used in recommender systems to predict people's item ratings.^{[10]} Distributed algorithms have been developed for the purpose of calculating the SVD on clusters of commodity machines.^{[11]}

Low-rank SVD has been applied for hotspot detection from spatiotemporal data with application to disease outbreak detection
.^{[12]} A combination of SVD and higher-order SVD also has been applied for real time event detection from complex data streams (multivariate data with space and time dimensions) in Disease surveillance.^{[13]}

## Relation to eigenvalue decomposition

The singular value decomposition is very general in the sense that it can be applied to any *m* × *n* matrix whereas eigenvalue decomposition can only be applied to certain classes of square matrices. Nevertheless, the two decompositions are related.

Given an SVD of **M**, as described above, the following two relations hold:

The right-hand sides of these relations describe the eigenvalue decompositions of the left-hand sides. Consequently:

- The columns of
**V**(right-singular vectors) are eigenvectors of**M**^{∗}**M**. - The columns of
**U**(left-singular vectors) are eigenvectors of**MM**^{∗}. - The non-zero elements of
**Σ**(non-zero singular values) are the square roots of the non-zero eigenvalues of**M**^{∗}**M**or**MM**^{∗}.

- The columns of

In the special case that **M** is a normal matrix, which by definition must be square, the spectral theorem says that it can be unitarily diagonalized using a basis of eigenvectors, so that it can be written **M** = **UDU**^{∗} for a unitary matrix **U** and a diagonal matrix **D**. When **M** is also positive semi-definite, the decomposition **M** = **UDU**^{∗} is also a singular value decomposition.

However, the eigenvalue decomposition and the singular value decomposition differ for all other matrices **M**: the eigenvalue decomposition is **M** = **UDU**^{−1} where **U** is not necessarily unitary and **D** is not necessarily positive semi-definite, while the SVD is **M** = **UΣV**^{∗} where **Σ** is a diagonal positive semi-definite, and **U** and **V** are unitary matrices that are not necessarily related except through the matrix **M**.

## Existence

An eigenvalue Template:Mvar of a matrix **M** is characterized by the algebraic relation **M***u* = *λu*. When **M** is Hermitian, a variational characterization is also available. Let **M** be a real *n* × *n* symmetric matrix. Define

By the extreme value theorem, this continuous function attains a maximum at some *u* when restricted to the closed unit sphere {||*x*|| ≤ 1}. By the Lagrange multipliers theorem, *u* necessarily satisfies

where the nabla symbol, ∇, is the del operator.

A short calculation shows the above leads to **M***u* = *λu* (symmetry of **M** is needed here). Therefore Template:Mvar is the largest eigenvalue of **M**. The same calculation performed on the orthogonal complement of *u* gives the next largest eigenvalue and so on. The complex Hermitian case is similar; there *f*(*x*) = *x* M x* is a real-valued function of 2*n* real variables.

Singular values are similar in that they can be described algebraically or from variational principles. Although, unlike the eigenvalue case, Hermiticity, or symmetry, of **M** is no longer required.

This section gives these two arguments for existence of singular value decomposition.

### Based on the spectral theorem

Let **M** be an *m* × *n* complex matrix. Since **M**^{∗}**M** is positive semi-definite and Hermitian, by the spectral theorem, there exists a unitary *n* × *n* matrix **V** such that

where **D** is diagonal and positive definite. Partition **V** appropriately so we can write

Therefore:

The second equation implies **MV**_{2} = **0**. Also, since **V** is unitary:

where the subscripts on the identity matrices are there to keep in mind that they are of different dimensions. Define

Then

We see that this is almost the desired result, except that **U**_{1} and **V**_{1} are not unitary in general since they might not be square . However, we do know that for **U**_{1}, the number of rows is greater than the number of columns since the dimensions of **D** is no greater than *m* and *n*. To finish the argument, we simply have to "fill it out". For example, we can choose **U**_{2} such that the following matrix is unitary:

For **V**_{1} we already have **V**_{2} to make it unitary. Now, define

where extra zero rows are added **or removed** to make the number of zero rows equal the number of columns of **U**_{2}. Then

which is the desired result:

Notice the argument could begin with diagonalizing **MM**^{∗} rather than **M**^{∗}**M** (This shows directly that **MM**^{∗} and **M**^{∗}**M** have the same non-zero eigenvalues).

### Based on variational characterization

The singular values can also be characterized as the maxima of **u**^{T}**Mv**, considered as a function of **u** and **v**, over particular subspaces. The singular vectors are the values of **u** and **v** where these maxima are attained.

Let **M** denote an *m* × *n* matrix with real entries. Let *S*^{m−1} and *S*^{n−1} denote the sets of unit 2-norm vectors in **R**^{m} and **R**^{n} respectively. Define the function

Consider the function Template:Mvar restricted to *S*^{m−1} × *S*^{n−1}. Since both *S*^{m−1} and *S*^{n−1} are compact sets, their product is also compact. Furthermore, since Template:Mvar is continuous, it attains a largest value for at least one pair of vectors **u** ∈ *S*^{m−1} and **v** ∈ *S*^{n−1}. This largest value is denoted *σ*_{1} and the corresponding vectors are denoted **u**_{1} and **v**_{1}. Since *σ*_{1} is the largest value of *σ*(**u**, **v**) it must be non-negative. If it were negative, changing the sign of either **u**_{1} or **v**_{1} would make it positive and therefore larger.

**Statement.****u**_{1},**v**_{1}are left and right-singular vectors of**M**with corresponding singular value*σ*_{1}.

**Proof:** Similar to the eigenvalues case, by assumption the two vectors satisfy the Lagrange multiplier equation:

After some algebra, this becomes

Multiplying the first equation from left by and the second equation from left by and taking Template:!!**u**Template:!! = Template:!!**v**Template:!! = 1 into account gives

Plugging this into the pair of equations above, we have

This proves the statement.

More singular vectors and singular values can be found by maximizing *σ*(**u**, **v**) over normalized **u**, **v** which are orthogonal to **u**_{1} and **v**_{1}, respectively.

The passage from real to complex is similar to the eigenvalue case.

## Geometric meaning

Because **U** and **V** are unitary, we know that the columns **U**_{1}, ..., **U**_{m} of **U** yield an orthonormal basis of Template:Mvar and the columns **V**_{1}, ..., **V**_{n} of **V** yield an orthonormal basis of Template:Mvar (with respect to the standard scalar products on these spaces).

has a particularly simple description with respect to these orthonormal bases: we have

where Template:Mvar is the Template:Mvar-th diagonal entry of **Σ**, and *T*(**V**_{i}) = 0 for *i* > min(*m*,*n*).

The geometric content of the SVD theorem can thus be summarized as follows: for every linear map *T* : *K ^{n}* →

*K*one can find orthonormal bases of Template:Mvar and Template:Mvar such that Template:Mvar maps the Template:Mvar-th basis vector of Template:Mvar to a non-negative multiple of the Template:Mvar-th basis vector of Template:Mvar, and sends the left-over basis vectors to zero. With respect to these bases, the map Template:Mvar is therefore represented by a diagonal matrix with non-negative real diagonal entries.

^{m}To get a more visual flavour of singular values and SVD decomposition — at least when working on real vector spaces — consider the sphere Template:Mvar of radius one in **R**^{n}. The linear map Template:Mvar maps this sphere onto an ellipsoid in **R**^{m}. Non-zero singular values are simply the lengths of the semi-axes of this ellipsoid. Especially when *n* = *m*, and all the singular values are distinct and non-zero, the SVD of the linear map Template:Mvar can be easily analysed as a succession of three consecutive moves: consider the ellipsoid *T*(*S*) and specifically its axes; then consider the directions in **R**^{n} sent by Template:Mvar onto these axes. These directions happen to be mutually orthogonal. Apply first an isometry **V**^{∗} sending these directions to the coordinate axes of **R**^{n}. On a second move, apply an endomorphism **D** diagonalized along the coordinate axes and stretching or shrinking in each direction, using the semi-axes lengths of *T*(*S*) as stretching coefficients. The composition **D** ∘ **V**^{∗} then sends the unit-sphere onto an ellipsoid isometric to *T*(*S*). To define the third and last move **U**, apply an isometry to this ellipsoid so as to carry it over *T*(*S*). As can be easily checked, the composition **U** ∘ **D** ∘ **V**^{∗} coincides with Template:Mvar.

## Calculating the SVD

### Numerical approach

The SVD of a matrix **M** is typically computed by a two-step procedure. In the first step, the matrix is reduced to a bidiagonal matrix. This takes O(*mn*^{2}) floating-point operations (flops), assuming that *m* ≥ *n*. The second step is to compute the SVD of the bidiagonal matrix. This step can only be done with an iterative method (as with eigenvalue algorithms). However, in practice it suffices to compute the SVD up to a certain precision, like the machine epsilon. If this precision is considered constant, then the second step takes O(*n*) iterations, each costing O(*n*) flops. Thus, the first step is more expensive, and the overall cost is O(*mn*^{2}) flops Template:Harv.

The first step can be done using Householder reflections for a cost of 4*mn*^{2} − 4*n*^{3}/3 flops, assuming that only the singular values are needed and not the singular vectors. If *m* is much larger than *n* then it is advantageous to first reduce the matrix *M* to a triangular matrix with the QR decomposition and then use Householder reflections to further reduce the matrix to bidiagonal form; the combined cost is 2*mn*^{2} + 2*n*^{3} flops Template:Harv.

The second step can be done by a variant of the QR algorithm for the computation of eigenvalues, which was first described by Template:Harvtxt. The LAPACK subroutine DBDSQR^{[14]} implements this iterative method, with some modifications to cover the case where the singular values are very small Template:Harv. Together with a first step using Householder reflections and, if appropriate, QR decomposition, this forms the DGESVD^{[15]} routine for the computation of the singular value decomposition.

The same algorithm is implemented in the GNU Scientific Library (GSL). The GSL also offers an alternative method, which uses a one-sided Jacobi orthogonalization in step 2 Template:Harv. This method computes the SVD of the bidiagonal matrix by solving a sequence of 2 × 2 SVD problems, similar to how the Jacobi eigenvalue algorithm solves a sequence of 2 × 2 eigenvalue methods Template:Harv. Yet another method for step 2 uses the idea of divide-and-conquer eigenvalue algorithms Template:Harv.

There is an alternative way which is not explicitly using the eigenvalue decomposition.^{[16]} Usually the singular value problem of a matrix **M** is converted into an equivalent symmetric eigenvalue problem such as **M M**^{*}, **M**^{*}**M**, or

The approaches using eigenvalue decompositions are based on QR algorithm which is well-developed to be stable and fast.
Note that the singular values are not complex and right- and left- singular vectors are not required to form any similarity transformation. Alternating QR decomposition and LQ decomposition can be claimed to use iteratively to find the real diagonal matrix with Hermittian matrices. QR decomposition gives **M** ⇒ **Q** **R** and LQ decomposition of **R** gives **R** ⇒ **L** **P**^{*}. Thus, at every iteration, we have **M** ⇒ **Q** **L** **P**^{*}, update **M** ⇐ **L** and repeat the orthogonalizations.
Eventually, QR decomposition and LQ decomposition iteratively provide unitary matrices for left- and right- singular matrices, respectively.
This approach does not come with any acceleration method such as spectral shifts and deflation as in QR algorithm. It is because the shift method is not easily defined without using similarity transformation. But it is very simple to implement where the speed does not matter. Also it give us a good interpretation that only orthogonal/unitary transformations can obtain SVD as the QR algorithm can calculate the eigenvalue decomposition.

### Analytic result of 2 × 2 SVD

The singular values of a 2 × 2 matrix can be found analytically. Let the matrix be

where are complex numbers that parameterize the matrix, **I** is the identity matrix, and denote the Pauli matrices. Then its two singular values are given by

## Reduced SVDs

In applications it is quite unusual for the full SVD, including a full unitary decomposition of the null-space of the matrix, to be required. Instead, it is often sufficient (as well as faster, and more economical for storage) to compute a reduced version of the SVD. The following can be distinguished for an *m*×*n* matrix *M* of rank *r*:

### Thin SVD

Only the *n* column vectors of *U* corresponding to the row vectors of *V** are calculated. The remaining column vectors of *U* are not calculated. This is significantly quicker and more economical than the full SVD if *n*≪*m*. The matrix *U*_{n} is thus *m*×*n*, Σ_{n} is *n*×*n* diagonal, and *V* is *n*×*n*.

The first stage in the calculation of a thin SVD will usually be a QR decomposition of *M*, which can make for a significantly quicker calculation if *n*≪*m*.

### Compact SVD

Only the *r* column vectors of *U* and *r* row vectors of *V** corresponding to the non-zero singular values Σ_{r} are calculated. The remaining vectors of *U* and *V** are not calculated. This is quicker and more economical than the thin SVD if *r*≪*n*. The matrix *U*_{r} is thus *m*×*r*, Σ_{r} is *r*×*r* diagonal, and *V*_{r}* is *r*×*n*.

### Truncated SVD

Only the *t* column vectors of *U* and *t* row vectors of *V** corresponding to the *t* largest singular values Σ_{t} are calculated. The rest of the matrix is discarded. This can be much quicker and more economical than the compact SVD if *t*≪*r*. The matrix *U*_{t} is thus *m*×*t*, Σ_{t} is *t*×*t* diagonal, and *V*_{t}* is *t*×*n*.

Of course the truncated SVD is no longer an exact decomposition of the original matrix *M*, but as discussed above, the approximate matrix is in a very useful sense the closest approximation to *M* that can be achieved by a matrix of rank *t*.

## Norms

### Ky Fan norms

The sum of the *k* largest singular values of *M* is a matrix norm, the Ky Fan *k*-norm of *M*.

The first of the Ky Fan norms, the Ky Fan 1-norm is the same as the operator norm of *M* as a linear operator with respect to the Euclidean norms of *K*^{m} and *K*^{n}. In other words, the Ky Fan 1-norm is the operator norm induced by the standard *l*^{2} Euclidean inner product. For this reason, it is also called the operator 2-norm. One can easily verify the relationship between the Ky Fan 1-norm and singular values. It is true in general, for a bounded operator *M* on (possibly infinite-dimensional) Hilbert spaces

But, in the matrix case, (*M* M*)^{½} is a normal matrix, so ||*M* M*||^{½} is the largest eigenvalue of (*M* M*)^{½}, i.e. the largest singular value of *M*.

The last of the Ky Fan norms, the sum of all singular values, is the trace norm (also known as the 'nuclear norm'), defined by ||*M*|| = Tr[(*M* M*)^{½}] (the eigenvalues of *M* M* are the squares of the singular values).

### Hilbert–Schmidt norm{{safesubst:#invoke:anchor|main}}

The singular values are related to another norm on the space of operators. Consider the Hilbert–Schmidt inner product on the *n* × *n* matrices, defined by

So the induced norm is

Since trace is invariant under unitary equivalence, this shows

where Template:Mvar are the singular values of **M**. This is called the **Frobenius norm**, **Schatten 2-norm**, or **Hilbert–Schmidt norm** of **M**. Direct calculation shows that the Frobenius norm of **M** = (*m _{ij}*) coincides with:

## Tensor SVD

The problem of finding a low rank approximation to a tensor is ill-posed. In other words, a best possible solution does not exist, but instead a sequence of better and better approximations that converge to infinitely large matrices. In spite of this, there are several ways to attempt decomposition.

Two types of tensor decompositions exist, which generalise SVD to multi-way arrays. One of them decomposes a tensor into a sum of rank-1 tensors, see Candecomp-PARAFAC (CP) algorithm. The CP algorithm should not be confused with a rank-*R* decomposition but, for a given *N*, it decomposes a tensor into a sum of *N* rank-1 tensors that optimally fit the original tensor. The second type of decomposition computes the orthonormal subspaces associated with the different axes or modes of a tensor (orthonormal row space, column space, fiber space, etc.). This decomposition is referred to in the literature as the Tucker3/TuckerM, *M*-mode SVD, multilinear SVD and sometimes referred to as a higher-order SVD (HOSVD). In addition, multilinear principal component analysis in multilinear subspace learning involves the same mathematical operations as Tucker decomposition, being used in a different context of dimensionality reduction.

## Bounded operators on Hilbert spaces

The factorization **M** = **UΣV**^{∗} can be extended to a bounded operator *M* on a separable Hilbert space *H*. Namely, for any bounded operator *M*, there exist a partial isometry *U*, a unitary *V*, a measure space (*X*, *μ*), and a non-negative measurable *f* such that

where is the multiplication by *f* on *L*^{2}(*X*, *μ*).

This can be shown by mimicking the linear algebraic argument for the matricial case above. *VT _{f} V** is the unique positive square root of

*M*M*, as given by the Borel functional calculus for self adjoint operators. The reason why

*U*need not be unitary is because, unlike the finite-dimensional case, given an isometry

*U*

_{1}with nontrivial kernel, a suitable

*U*

_{2}may not be found such that

is a unitary operator.

As for matrices, the singular value factorization is equivalent to the polar decomposition for operators: we can simply write

and notice that *U V** is still a partial isometry while *VT _{f} V** is positive.

### Singular values and compact operators

To extend notion of singular values and left/right-singular vectors to the operator case, one needs to restrict to compact operators. It is a general fact that compact operators on Banach spaces have only discrete spectrum. This is also true for compact operators on Hilbert spaces, since Hilbert spaces are a special case of Banach spaces. If Template:Mvar is compact, every non-zero Template:Mvar in its spectrum is an eigenvalue. Furthermore, a compact self adjoint operator can be diagonalized by its eigenvectors. If **M** is compact, so is **M**^{∗}**M**. Applying the diagonalization result, the unitary image of its positive square root Template:Mvar has a set of orthonormal eigenvectors {*e _{i}*} corresponding to strictly positive eigenvalues {

*σ*}. For any

_{i}*ψ*∈

*H*,

where the series converges in the norm topology on Template:Mvar. Notice how this resembles the expression from the finite-dimensional case. Template:Mvar are called the singular values of **M**. {**U***e _{i}*} (resp. {

**V**

*e*} ) can be considered the left-singular (resp. right-singular) vectors of

_{i}**M**.

Compact operators on a Hilbert space are the closure of finite-rank operators in the uniform operator topology. The above series expression gives an explicit such representation. An immediate consequence of this is:

**Theorem.****M**is compact if and only if**M**^{∗}**M**is compact.

## History

The singular value decomposition was originally developed by differential geometers, who wished to determine whether a real bilinear form could be made equal to another by independent orthogonal transformations of the two spaces it acts on. Eugenio Beltrami and Camille Jordan discovered independently, in 1873 and 1874 respectively, that the singular values of the bilinear forms, represented as a matrix, form a complete set of invariants for bilinear forms under orthogonal substitutions. James Joseph Sylvester also arrived at the singular value decomposition for real square matrices in 1889, apparently independently of both Beltrami and Jordan. Sylvester called the singular values the *canonical multipliers* of the matrix *A*. The fourth mathematician to discover the singular value decomposition independently is Autonne in 1915, who arrived at it via the polar decomposition. The first proof of the singular value decomposition for rectangular and complex matrices seems to be by Carl Eckart and Gale Young in 1936;^{[17]} they saw it as a generalization of the principal axisTemplate:Disambiguation needed transformation for Hermitian matrices.

In 1907, Erhard Schmidt defined an analog of singular values for integral operators (which are compact, under some weak technical assumptions); it seems he was unaware of the parallel work on singular values of finite matrices. This theory was further developed by Émile Picard in 1910, who is the first to call the numbers *singular values* (or in French, *valeurs singulières*).

Practical methods for computing the SVD date back to Kogbetliantz in 1954, 1955 and Hestenes in 1958.^{[18]} resembling closely the Jacobi eigenvalue algorithm, which uses plane rotations or Givens rotations. However, these were replaced by the method of Gene Golub and William Kahan published in 1965,^{[19]} which uses Householder transformations or reflections.
In 1970, Golub and Christian Reinsch^{[20]} published a variant of the Golub/Kahan algorithm that is still the one most-used today.

## See also

## Notes

- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ The Singular Value Decomposition in Symmetric (Lowdin) Orthogonalization and Data Compression
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ S. Walton, O. Hassan, K. Morgan, Reduced order modelling for unsteady fluid flow using proper orthogonal decomposition and radial basis functions, Applied Mathematical Modelling, http://www.sciencedirect.com/science/article/pii/S0307904X13002771
- ↑ Gharib Shirangi, M., History matching production data and uncertainty assessment with an efficient TSVD parameterization algorithm, Journal of Petroleum Science and Engineering, http://www.sciencedirect.com/science/article/pii/S0920410513003227
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ Netlib.org
- ↑ Netlib.org
- ↑ mathworks.co.kr/matlabcentral/fileexchange/12674-simple-svd
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}
- ↑ {{#invoke:Citation/CS1|citation |CitationClass=journal }}

## References

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:Citation/CS1|citation

|CitationClass=journal }}

- {{#invoke:Citation/CS1|citation

|CitationClass=journal }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- Halldor, Bjornsson and Venegas, Silvia A. (1997). "A manual for EOF and SVD analyses of climate data". McGill University, CCGCR Report No. 97-1, Montréal, Québec, 52pp.
- {{#invoke:Citation/CS1|citation

|CitationClass=journal }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:Citation/CS1|citation

|CitationClass=journal }}

- {{#invoke:citation/CS1|citation

|CitationClass=book }}

- {{#invoke:citation/CS1|citation

|CitationClass=citation }}