# Expressing a matrix in terms of four basis matrices

I have five matrices and I want to represent one in the basis of other four. So how do I find the value of coefficients?

For example, if the matrices are I1, I2, I3, I4, T and the relation between them is

T = (a*I1) + (b*I2) + (c*I3) + (d*I4)


Now to determine the value of coefficients a, b, c, d, what do I have to do in Mathematica?

I have tried this

Solve[{T - a*I1 - b*I2 - c*I3 - d*I4 == 0}, {a, b, c, d}]


but it is showing this error:

Solve::svars: Equations may not give solutions for all "solve" variables.

• Do you have a space between variable names? Anothe thing: dependent on the dimensions of your matrices you may have an overdetermined system. Commented Oct 3, 2014 at 5:45
• Not knowing anything about your matricies, what happens if you try this on your matricies: NMinimize[ Norm[Flatten[T] - (aFlatten[I1] + bFlatten[I2] + cFlatten[I3] + dFlatten[I4])], {a, b, c, d}, MaxIterations -> 10^4] When I try that on random Real 3x3 matricies I rapidly get very good approximations of the scalars I used to construct T from I1, I2, I3, I4. If this works for you then you can enhance this with additional precision if needed.
– Bill
Commented Oct 3, 2014 at 7:30
• Please, reopen the question, I have an answer. It is a common task to decompose a matrix as a linear combination of basis matrices (e.g. the Pauli matrices). It can be a duplicate, not an unclear question. Commented Oct 3, 2014 at 12:19
• @ybeltukov I was thinking the same thing. Voting to re-open. Commented Oct 3, 2014 at 14:41
• @ybeltukov Reopened on request. In the future you are welcome to flag a post ("other") if you have a good answer ready for a closed question. Commented Oct 3, 2014 at 15:47

# LinearSolve

If you have a proper dimensions I recommend you to use LinearSolve here. Let us take 4 random matrices as a basis (complex matrices for generality)

basis = RandomComplex[1 + I, {4, 2, 2}];
MatrixForm /@ basis


The target matrix

T = RandomReal[1, {2, 2}];
T // MatrixForm


We can treat 2x2 matrices as vectors with 4 elements, Flatten them and apply LinearSolve to find the coefficients

v = LinearSolve[Flatten[basis, {{2, 3}}], Flatten[T]]


Validation:

Norm[v.basis - T]


1.26609*10^-16

# Dot

If your matrices are orthogonal to each other (as flattened vectors) you can simply use the matrix multiplication. An example with the Pauli matrices:

basis = PauliMatrix@Range[0, 3];
MatrixForm /@ basis


The flattened basis

fb = Flatten[basis, {{1}, {2, 3}}]


The basis is orthogonal (but not normalized)

fb.ConjugateTranspose[fb] // MatrixForm
nrm = Diagonal[fb.ConjugateTranspose[fb]];


Coefficients

v = Flatten[T].ConjugateTranspose[fb]/nrm


Validation

Norm[v.basis - T]


0.

You can also solve this for symbolic matrices.

If you want to decompose the matrix m={{a,b},{c,d}} into a sum over the Pauli Matrices you just need to define:

approx = Total@Table[v[i] PauliMatrix[i], {i, 0, 3}];
vars = Table[v[i], {i, 0, 3}];
m = {{a, b}, {c, d}};


The first line gives the sum of the basis matrices with the variable weights.

The seconde line is a fancy way to get all the variables together. Not really necessary in this small example, but useful for larger ones.

The last definition is the target matrix itself.

Now execute:

Solve[m - approx == 0, vars]


and you will get the values for all the 'v's.