I'm trying to better understand the manifold GL+(3,R)/S0(3) which is diffeomorphic to positive definite symmetric matrices. My motivation is to understand U in F = RU where F in GL+(3,R) = deformation gradient, R in S0(3), & U in GL+(3,R)/S0(3) = stretches & shears.

I think that:
(1) GL+(3,R)/SO(3) being diffeomorphic to SYMMETRIC positive definite matrices results from SO(3) not being SYMMETRIC.
(2) Any U in GL+(3,R)/SO(3) can be composed from the set of matrices consisting of all principal shears (identity matrices with 1 off-diag. non-zero entry), all principal stretches (diag. matrices with non-zero diag. entries), & the identity matrix.
(3) This set is not a group for reason given below.

An earlier discussion Shear transformations about shear transformations has been very helpful. It highlighted that:
(a) Any upper/lower triangular matrix with 1's on diag. is a composition of shears.
(b) Any matrix with det.= 1 including (a) but also S0(3) can be decomposed into upper/lower triangular matrices with 1's on diag..
(c) Given (b), shear transformations don't form a group as SO(3) not a shear but can be composed from shear transformations.
(d) Upper (or lower) triangular matrices with 1's on diag. (i.e. composed shears) are a group but obviously not containing ALL shears since lower (or upper) triangular shear matrices missing. And, I'm assuming, composing these two groups NOT a group as per (c).

So, I'm asking:
i) If my thoughts about GL+(3,R)/SO(3) in points (1) & (2) above are correct ?
ii) How, as in (b), can composing an upper & lower triangular matrix (both composed shears) result in a rotation in SO(3) ?

Sorry for such a long post !


1 Answer 1


$GL^+(3,R)/SO(3)$ is the space of 3 dimensional positive definite symmetric matrices because the polar decomposition of $g \in GL^+(3,R)$ is $ g = o p $ , $o \in SO(3)$ and $p$ is positive definite symmetric .

The wikipedia page treats the complex case, but by repeating the computation for the real case you actually get a constructive proof why matrices $p$ are positive definite symmetric.

The answer to your second question (ii) is that almost any orthogonal matrix can be decomposed as a product of a lower triangular and upper triangular matrix (the LU decomposition which is the Bruhat decomposition of the big cell). As an exception you can take a permutation matrix which doesn't allow such a decomposition.

  • $\begingroup$ David, Thanks very much. I'll try wiki computation. I'm still wondering about my point 2 about matrices to generate any U in GL+(3,R)/S0(3). The shear matrices which I've chosen are not symmetric nor, necessarily, are their compositions. Please see matrices below. So, have I chosen the wrong matrices ? Thanks, John. My list of 10 matrices: Principal Shears (6 matrices) [(1,0,0),(0,1,0),(k,0,1)] + 5 more moving "k" to other "O" spots. Principal Stretches (3 matrices) [(a,0,0),(0,1,0),(0,0,1)] + 2 more by moving "a" on diag. Id matrix [(1,0,0),(0,1,0),(0,0,1)] $\endgroup$ Oct 21, 2010 at 18:17

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.