1
$\begingroup$

Let's suppose that we have two orthonormal basis, $V=\{|v_i\rangle \}$, $i=1,...n$ (ket notation) and $U=\{|u_j\rangle\}$, $j=1,..,m$, with $m<n$ (i.e. $\dim U< \dim V$), where $|u_j\rangle=\sum_{i=1}^{n}c_i^j |v_i\rangle$. Given that $[A]_V$ is the matrix representation of some transformation $A$ in the basis $V$, under which conditions $[A]_U$ is the matrix representation of the same transformation $A$ in the basis $U$?

$$[A]_{U}=\left[ \begin{array}{c} \langle u_1| \\ \vdots\\ \langle u_m| \end{array} \right] [A]_{V}\left[ \begin{array}{c} |u_1\rangle & \cdots & |u_m\rangle| \end{array} \right]$$

$\endgroup$
  • 1
    $\begingroup$ Are you assuming matrix A is not full-rank? $\endgroup$ – DBPriGuy Nov 11 '16 at 13:07
  • 1
    $\begingroup$ This doesn't make sense without additional assumptions. $A$ represents a linear mapping from an $n$-dimensional space to itself. The set $K$ only spans a two-dimensional subspace of that bigger space, and you can only restrict the mapping $A$ to that subspace if it happens to be invariant under $A$ (that is, if it the case that you whenever you apply $A$ to a vector in the subspace you get something which is still in the subspace). $\endgroup$ – Hans Lundmark Nov 11 '16 at 13:09
  • $\begingroup$ I have changed the original question. I don't know if it now makes sense. $\endgroup$ – AndyK Nov 11 '16 at 18:25
  • $\begingroup$ The written equation is always true, you are just defining the matrix $A_U$. Do you mean to ask when there exists a matrix $A_U$ such that $A_V = \text{[...something involving $A_U$...]}$? $\endgroup$ – Rahul Nov 11 '16 at 18:56
  • $\begingroup$ Ok, I added an answer which may clarify my question. $\endgroup$ – AndyK Nov 12 '16 at 16:04
2
+100
$\begingroup$

A linear transformation $A$ acts on a vector space $X$. If the $X$ is finite dimensional, then every basis of $X$ has the same number of vectors (which is equal to $\dim X$). Hence, your question doesn't make sense from the start: you cannot have two (orthonormal) bases of $X$ with different numbers of basis vectors.

$\endgroup$
  • $\begingroup$ What about an incomplete basis of $X$? Please see my answer below. $\endgroup$ – AndyK Nov 14 '16 at 19:42
  • 1
    $\begingroup$ What's an incomplete basis? I assume you mean an orthonormal set that's not spanning (which is not a basis). Then, the matrix of $[A]_U$ is $m \times m$ but the matrix $[A]_V$ is $n \times n$. If $n \neq m$, how can you expect these matrices to be "the same"? $\endgroup$ – Jon Warneke Nov 14 '16 at 20:08
0
$\begingroup$

There $\nexists$ a linear transformation between two Vector Spaces with different dimensions. If there was, it would be a homomorphism and two Vector Spaces $M, N$ are homomorphic $\iff \dim M = \dim N$.

$\endgroup$
  • 1
    $\begingroup$ This is not true. Consider the projection transformation $T : \mathbb R^2 \to \mathbb R$ given by $T(x, y) = x$. Then $T$ is a linear transformation (as you can check) between vector spaces of different dimensions. $\endgroup$ – Jon Warneke Nov 14 '16 at 16:47
0
$\begingroup$

I figured out the answer to my badly posed question, so please allow me to post it.

For simplicity I use $n=3$ and $m=2$. The basis vectors $|v_i\rangle$ are orthonormal and the new basis vectors $|u_j\rangle$ are defined as linear combintions of $|v_i\rangle$: $$|u_j\rangle=\sum_{i=1}^{3}c_i^j|v_i\rangle$$ In order them to be orthonormal too, the condition is: if $c_i^1\neq 0$ then $c_i^2 = 0$, $\forall i$. Let's suppose, without loss of generality, that \begin{align} |u_1\rangle &=c_1|v_1\rangle+c_2|v_2\rangle=\left[ \begin{array}{c} c_1 \\ c_2\\ 0 \end{array} \right]_V\tag{1}\\ |u_2\rangle &=c_3|v_3\rangle=\left[ \begin{array}{c} 0 \\ 0\\ c_3 \end{array} \right]_V \tag{2} \end{align}

Next, consider some operator $A$, with the following matrix representation in the basis $V$: $$[A]_{V}=\left[ \begin{array}{ccc} A_{11} & A_{12} & A_{13} \\ A_{21} & A_{22} & A_{23}\\ A_{31} & A_{32} & A_{33} \end{array} \right]_V=\sum_{i,j=1}^{3}A_{ij}|v_i\rangle\langle v_j|$$ where $A_{i,j}\equiv \langle v_i|A|v_j \rangle$, $i,j=1,2,3$.

Let's suppose that we can write: $$[A]_{U}=\left[ \begin{array}{cc} A'_{11} & A'_{12} \\ A'_{21} & A'_{22} \end{array} \right]_U=\sum_{i,j=1}^{2}A'_{ij}|u_i\rangle\langle u_j|$$ where $A'_{i,j}\equiv \langle u_i|A|u_j \rangle$, $i,j=1,2$. From $(1)$ and (2) we can express $|u_i\rangle \langle u_j|$ in terms of $|v_k\rangle \langle v_{\ell}|$, $k,\ell=1,2,3$. Therefore \begin{align} [A]_{V} &=A'_{11}\left(|c_1|^2|v_1\rangle \langle v_1|+c_1 c_2^*|v_1\rangle \langle v_2|+c_2 c_1^*|v_2\rangle \langle v_1|+|c_2|^2|v_2\rangle \langle v_2|\right)\\ &+A'_{12}\left(c_1 c_3^*|v_1\rangle \langle v_3|+c_2 c_3^*|v_2\rangle \langle v_3|\right)\\ &+A'_{21}\left(c_3 c_1^*|v_3\rangle \langle v_1|+c_3 c_2^*|v_3\rangle \langle v_2|\right)\\ &+A'_{22}|c_3|^2|v_3\rangle \langle v_3|\\ \end{align} and we get the following conditions: \begin{align} \frac{A_{11}}{|c_1|^2}&=\frac{A_{12}}{c_1 c_2^*}=\frac{A_{21}}{c_2 c_1^*}=\frac{A_{22}}{|c_2|^2}=A'_{11} \\ \frac{A_{13}}{c_1 c_3^*}&=\frac{A_{23}}{c_2 c_3^*}=A'_{12} \\ \frac{A_{31}}{c_3 c_1^*}&=\frac{A_{32}}{c_3 c_2^*}=A'_{21}\\ \frac{A_{33}}{|c_3|^2}&=A'_{22} \end{align} If they are all satisfied, we can express $A$ in the basis $U$: $$ [A]_U=\left[ \begin{array}{c} \langle u_1| \\ \langle u_2| \end{array} \right]_V [A]_V \hspace{0.2cm}\left[ \begin{array}{cc} |u_1\rangle & |u_2\rangle \end{array} \right]_V=\left[ \begin{array}{cc} A'_{11} & A'_{12} \\ A'_{21} & A'_{22} \end{array} \right]_U$$ This of course can be generalized to any $n$ and $m$, $n>m$.

Edit: I spotted the mistake in my answer: the vectors $|u_i\rangle$, $i=1,2$ of course do not form another basis of the $3\times 3$ vector space $V_A$ on which the operator $A$ acts, $A:V_A\to V_A$. They form an incomplete orthogonal set, $S\subset V_A$ (which is a plane in $V_A$). If the mentioned conditions are all satisfied, this just means that $A$ projects all the vector of $V_A$ to $S$, $A:V_A\to S$. But this doesn't mean that we can write $A$ as $2\times 2$ matrix because this would restrict its action to $S$, i.e. $A:S\to S$, which is not true for $A$, because by definition it acts on whole space $V_A$.

$\endgroup$

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.