Hilbert-Schmidt basis for many qubits - reference Every density matrix of $n$ qubits can be written in the following way
$$\hat{\rho}=\frac{1}{2^n}\sum_{i_1,i_2,\ldots,i_n=0}^3 t_{i_1i_2\ldots i_n} \hat{\sigma}_{i_1}\otimes\hat{\sigma}_{i_2}\otimes\ldots\otimes\hat{\sigma}_{i_n},$$
where $-1 \leq t_{i_1i_2\ldots i_n} \leq 1$ are real numbers and $\{\hat{\sigma}_0,\hat{\sigma}_1,\hat{\sigma}_2,\hat{\sigma}_3\}$ are the Pauli matrices. In particular for one particle ($n=1$) it is the Bloch representation.
Such representation is used e.g. in a work by Horodecki arXiv:quant-ph/9607007 (they apply $n=2$ to investigate the entanglement of two qubit systems). It is called decomposition in the Hilbert-Schmidt basis.
The question is if there is any good reference for such representation for qubits - either introducing it for quantum applications or a review paper?
I am especially interested in the constrains on $t_{i_1i_2\ldots i_n}$.
 A: I use this decomposition all the time, but I have never read a paper solely devoted to the topic.  From my experience a complete characterization of the constraints on $t_{i_{1}, t_{2},..t_{n}}$ is tricky, and so if you want to be sure $\rho$ is physical you should calculate the density matrix and its eigenvalues.  
However, there are a lot of necessary conditions that have a useful form in this decomposition.  For example, for a positive unit-trace Hermitian operator $\rho$ is follows that
$|t_{i_{1}, i_{2},.. i_{n}}| \leq 1$
$tr ( \rho^{2} ) =\frac{1}{2^{n}} \sum_{i_{1}, i_{2},.. i_{n}} t_{i_{1}, i_{2},.. i_{n}}^{2} \leq 1 $
The above condition tells us that if we think of $t$ as a vector in a real vector space, then the physical states live within the unit sphere.  This is a bit like the Bloch sphere for 1 qubit but for many qubits we have some other constraints that take the form of hyperplanes.  For every $\vert \psi \rangle$ expressed in the same form
$\vert \psi \rangle \langle \psi \vert = \frac{1}{2^{n}} \sum_{i_{1},i_{2},... i_{n}} Q_{i_{1},i_{2},... i_{n}}\sigma_{i1} \otimes \sigma_{i2}... \sigma_{in}$ we require that
$\langle \psi \vert \rho \vert \psi \rangle \geq 0 $ and so
$\sum Q_{i_{1},i_{2},... i_{n}}t_{i_{1},i_{2},... i_{n}}\geq 0$
which defines a hyperplane.
The problem is you have a hyperplane for every $\psi$ so that requiring $t$ to satisfy every inequality one of the infinite hyperplanes is impossible to check by brute force.  If you want sufficient conditions for positivity of $\rho$ I suspect you have to calculate eigenvalues.
A: Claudio Altafini studies precisely this subject, in Tensor of coherences parameterization of multiqubit density operators for entanglement characterization and some follow-ups.
A: A good starting point, I have checked just chapter 4 but there is more, is
R. R. Puri, Mathematical Methods of Quantum Optics, Springer (2001) (see here).
