Symmetric tensor

From Wikipedia, the free encyclopedia
Jump to: navigation, search

In mathematics, a symmetric tensor is a tensor that is invariant under a permutation of its vector arguments:

T(v_1,v_2,\dots,v_r) = T(v_{\sigma 1},v_{\sigma 2},\dots,v_{\sigma r})

for every permutation σ of the symbols {1,2,...,r}. Alternatively, a symmetric tensor of order r represented in coordinates as a quantity with r indices satisfies

T_{i_1i_2\dots i_r} = T_{i_{\sigma 1}i_{\sigma 2}\dots i_{\sigma r}}.

The space of symmetric tensors of order r on a finite-dimensional vector space is naturally isomorphic to the dual of the space of homogeneous polynomials of degree r on V. Over fields of characteristic zero, the graded vector space of all symmetric tensors can be naturally identified with the symmetric algebra on V. A related concept is that of the antisymmetric tensor or alternating form. Symmetric tensors occur widely in engineering, physics and mathematics.

Definition[edit]

Let V be a vector space and

T\in V^{\otimes k}

a tensor of order k. Then T is a symmetric tensor if

\tau_\sigma T = T\,

for the braiding maps associated to every permutation σ on the symbols {1,2,...,k} (or equivalently for every transposition on these symbols).

Given a basis {ei} of V, any symmetric tensor T of rank k can be written as

T = \sum_{i_1,\dots,i_k=1}^N T_{i_1i_2\dots i_k} e^{i_1} \otimes e^{i_2}\otimes\cdots \otimes e^{i_k}

for some unique list of coefficients T_{i_1i_2\dots i_k} (the components of the tensor in the basis) that are symmetric on the indices. That is to say

T_{i_{\sigma 1}i_{\sigma 2}\dots i_{\sigma k}} = T_{i_1i_2\dots i_k}

for every permutation σ.

The space of all symmetric tensors of order k defined on V is often denoted by Sk(V) or Symk(V). It is itself a vector space, and if V has dimension N then the dimension of Symk(V) is the binomial coefficient

\dim\, \operatorname{Sym}^k(V) = {N + k - 1 \choose k}.

We then construct Sym(V) as the direct sum of Symk(V) for k = 0,1,2,…

\operatorname{Sym}(V)= \bigoplus_{k=0}^\infty \operatorname{Sym}^k(V).

Examples[edit]

There are many examples of symmetric tensors. Some include, the metric tensor, g_{\mu\nu}, the Einstein tensor, G_{\mu\nu} and the Ricci tensor, R_{\mu\nu}.

Many material properties and fields used in physics and engineering can be represented as symmetric tensor fields; for example: stress, strain, and anisotropic conductivity. Also, in diffusion MRI one often uses symmetric tensors to describe diffusion in the brain or other parts of the body.

Ellipsoids are examples of algebraic varieties; and so, for general rank, symmetric tensors, in the guise of homogeneous polynomials, are used to define projective varieties, and are often studied as such.

Symmetric part of a tensor[edit]

Suppose V is a vector space over a field of characteristic 0. If TVk is a tensor of order k, then the symmetric part of T is the symmetric tensor defined by

\operatorname{Sym}\, T = \frac{1}{k!}\sum_{\sigma\in\mathfrak{S}_k} \tau_\sigma T,

the summation extending over the symmetric group on k symbols. In terms of a basis, and employing the Einstein summation convention, if

T = T_{i_1i_2\dots i_k}e^{i_1}\otimes e^{i_2}\otimes\cdots \otimes e^{i_k},

then

\operatorname{Sym}\, T = \frac{1}{k!}\sum_{\sigma\in \mathfrak{S}_k} T_{i_{\sigma 1}i_{\sigma 2}\dots i_{\sigma k}} e^{i_1}\otimes e^{i_2}\otimes\cdots \otimes e^{i_k}.

The components of the tensor appearing on the right are often denoted by

T_{(i_1i_2\dots i_k)} = \frac{1}{k!}\sum_{\sigma\in \mathfrak{S}_k} T_{i_{\sigma 1}i_{\sigma 2}\dots i_{\sigma k}}

with parentheses around the indices which have been symmetrized. [Square brackets are used to indicate anti-symmetrization.]

Symmetric product[edit]

If T is a simple tensor, given as a pure tensor product

T=v_1\otimes v_2\otimes\cdots \otimes v_r

then the symmetric part of T is the symmetric product of the factors:

v_1\odot v_2\odot\cdots\odot v_r := \frac{1}{r!}\sum_{\sigma\in\mathfrak{S}_r} v_{\sigma 1}\otimes v_{\sigma 2}\otimes\cdots\otimes v_{\sigma r}.

In general we can turn Sym(V) into an algebra by defining the commutative and associative product '\odot'.[1] Given two tensors T1∈Symk1(V) and T2∈Symk2(V), we use the symmetrization operator to define:

T_1\odot T_2 = \operatorname{Sym}(T_1\otimes T_2)\quad\left(\in\operatorname{Sym}^{k_1+k_2}(V)\right).

It can be verified (as is done by Kostrikin and Manin[1]) that the resulting product is in fact commutative and associative. In some cases the operator is not written at all: T1T2 = T1\odotT2.

In some cases an exponential notation is used:

v^{\odot k} = \underbrace{v \odot v \odot \cdots \odot v}_{k\text{ times}}=\underbrace{v \otimes v \otimes \cdots \otimes v}_{k\text{ times}}=v^{\otimes k}.

Where v is a vector. Again, in some cases the '\odot' is left out:

v^k=\underbrace{v\,v\,\cdots\,v}_{k\text{ times}}=\underbrace{v\odot v\odot\cdots\odot v}_{k\text{ times}}.

Decomposition[edit]

In analogy with the theory of symmetric matrices, a (real) symmetric tensor of order 2 can be "diagonalized". More precisely, for any tensor T ∈ Sym2(V), there are an integer r, non-zero unit vectors v1,...,vr ∈ V and weights λ1,...,λr such that

T = \sum_{i=1}^r \lambda_i \, v_i\otimes v_i.

The minimum number r for which such a decomposition is possible is the (symmetric) rank of T. The vectors appearing in this minimal expression are the principal axes of the tensor, and generally have an important physical meaning. For example, the principal axes of the inertia tensor define the Poinsot's ellipsoid representing the moment of inertia. Also see Sylvester's law of inertia.

For symmetric tensors of arbitrary order k, decompositions

T = \sum_{i=1}^r \lambda_i \, v_i^{\otimes k}

are also possible. The minimum number r for which such a decomposition is possible is the symmetric rank of T.[2] For second-order tensors this corresponds to the rank of the matrix representing the tensor in any basis, and it is well known that the maximum rank is equal to the dimension of the underlying vector space. However, for higher orders this need not hold: the rank can be higher than the number of dimensions in the underlying vector space. The higher-order singular value decomposition of a symmetric tensor is a special decomposition of this form [2] (often called the canonical decomposition.)

See also[edit]

Notes[edit]

  1. ^ a b Kostrikin, Alexei I.; Manin, Iurii Ivanovich (1997). Linear algebra and geometry. Algebra, Logic and Applications 1. Gordon and Breach. pp. 276–279. ISBN 9056990497. 
  2. ^ a b Comon, P.; Golub, G.; Lim, L. H.; Mourrain, B. (2008). "Symmetric Tensors and Symmetric Tensor Rank". SIAM Journal on Matrix Analysis and Applications 30 (3): 1254. doi:10.1137/060661569.  edit

References[edit]

External links[edit]