← Exercise library← All exercises for this theme
Theme 2 — Hilbert Spaces and Dirac Notation

Functions of a matrix

Keywords: functional calculus · spectral decomposition · square root · inverse · exponential · minimal polynomial · spectral projectors · time evolution

Exercise 1 : Functions of a matrix

For a Hermitian matrix with spectral decomposition A=∑kakPkA=\sum_ka_kP_k, where the PkP_k are the orthogonal projectors onto the eigenspaces, define f(A)=∑kf(ak)Pkf(A)=\sum_kf(a_k)P_k for every function ff defined on the spectrum. Consider again the matrix

A=(211121112)=1+J,\begin{aligned} A=\begin{pmatrix}2&1&1\\1&2&1\\1&1&2\end{pmatrix}=\mathbf{1}+J, \end{aligned}

where JJ is the matrix all of whose entries are 11, which satisfies J2=3JJ^2=3J. Assume that AA has eigenvalue 44, with projector P4=13JP_4=\frac13J, and eigenvalue 11, with projector P1=1−13JP_1=\mathbf{1}-\frac13J.

Question 1
Verify that, for a polynomial ff, the definition of f(A)f(A) agrees with direct calculation. Treat the case f(x)=x2f(x)=x^2.

Solution
The projectors satisfy Pk2=PkP_k^2=P_k and PkPl=0P_kP_l=0 for k≠lk\neq l. Thus A2=(∑kakPk)2=∑kak2PkA^2=\bigl(\sum_ka_kP_k\bigr)^2=\sum_ka_k^2P_k, and by induction An=∑kaknPkA^n=\sum_ka_k^nP_k; by linearity, this extends to every polynomial. Here, A2=16P4+P1=163J+1−13J=1+5JA^2=16P_4+P_1=\frac{16}{3}J+\mathbf{1}-\frac13J=\mathbf{1}+5J. Direct calculation gives (1+J)2=1+2J+J2=1+5J(\mathbf{1}+J)^2=\mathbf{1}+2J+J^2=\mathbf{1}+5J: the two methods agree.

Question 2
Calculate A−1A^{-1} and A\sqrt A, where the square root is defined by the function f(x)=xf(x)=\sqrt x on the spectrum, which is positive. Verify the results.

Solution
Since the eigenvalues are non-zero, AA is invertible, and

A−1=14P4+P1=112J+1−13J=1−14J.A^{-1}=\frac14P_4+P_1=\frac{1}{12}J+\mathbf{1}-\frac13J=\mathbf{1}-\frac14J .

Verification: (1+J)(1−14J)=1+J−14J−14J2=1+34J−34J=1(\mathbf{1}+J)(\mathbf{1}-\frac14J)=\mathbf{1}+J-\frac14J-\frac14J^2=\mathbf{1}+\frac34J-\frac34J=\mathbf{1}. Similarly,

A=2P4+P1=23J+1−13J=1+13J,\sqrt A=2P_4+P_1=\frac23J+\mathbf{1}-\frac13J=\mathbf{1}+\frac13J,

and (1+13J)2=1+23J+19J2=1+23J+13J=1+J=A(\mathbf{1}+\frac13J)^2=\mathbf{1}+\frac23J+\frac19J^2=\mathbf{1}+\frac23J+\frac13J=\mathbf{1}+J=A. This is the unique square root of AA whose eigenvalues are positive; others exist, for example −A-\sqrt A.

Question 3
Show that (A−4 1)(A−1)=0(A-4\,\mathbf{1})(A-\mathbf{1})=0. Deduce that the spectral projectors are polynomials in AA, and recover the expression for A−1A^{-1}.

Solution
We have (A−4 1)(A−1)=(J−3 1)J=J2−3J=0(A-4\,\mathbf{1})(A-\mathbf{1})=(J-3\,\mathbf{1})J=J^2-3J=0. This can also be seen from the spectral decomposition: (A−4)(A−1)=∑k(ak−4)(ak−1)Pk(A-4)(A-1)=\sum_k(a_k-4)(a_k-1)P_k, and each factor vanishes on the spectrum. Now consider the Lagrange polynomials L4(x)=x−14−1L_4(x)=\frac{x-1}{4-1} and L1(x)=x−41−4L_1(x)=\frac{x-4}{1-4}, which equal 11 at one eigenvalue and 00 at the other. We have L4(A)=∑kL4(ak)Pk=P4L_4(A)=\sum_kL_4(a_k)P_k=P_4, hence

P4=A−13=13J,P1=4 1−A3=1−13J,P_4=\frac{A-\mathbf{1}}{3}=\frac13J,\qquad P_1=\frac{4\,\mathbf{1}-A}{3}=\mathbf{1}-\frac13J,

in agreement with the statement. Finally, expanding the relation gives A2−5A+4 1=0A^2-5A+4\,\mathbf{1}=0, or A(5 1−A)=4 1A(5\,\mathbf{1}-A)=4\,\mathbf{1}, hence A−1=5 1−A4=4 1−J4=1−14JA^{-1}=\frac{5\,\mathbf{1}-A}{4}=\frac{4\,\mathbf{1}-J}{4}=\mathbf{1}-\frac14J.

Question 4
Consider a three-state quantum system with Hamiltonian H^=ℏωA\hat H=\hbar\omega A, prepared at time t=0t=0 in the state ∣e1⟩=(1,0,0)\ket{e_1}=(1,0,0). Calculate the evolution operator e−iH^t/ℏe^{-i\hat Ht/\hbar}, then the probability of finding the system in the state ∣e1⟩\ket{e_1} at time tt.

Solution
With f(x)=e−iωtxf(x)=e^{-i\omega tx}, we obtain

e−iH^t/ℏ=e−4iωtP4+e−iωtP1=e−4iωt 13J+e−iωt(1−13J).e^{-i\hat Ht/\hbar}=e^{-4i\omega t}P_4+e^{-i\omega t}P_1=e^{-4i\omega t}\,\frac13J+e^{-i\omega t}\Bigl(\mathbf{1}-\frac13J\Bigr).

Since ⟨e1∣J∣e1⟩=1\bra{e_1}J\ket{e_1}=1 and ⟨e1∣1∣e1⟩=1\bra{e_1}\mathbf{1}\ket{e_1}=1, the amplitude for remaining in the initial state is

⟨e1∣e−iH^t/ℏ∣e1⟩=13e−4iωt+23e−iωt=e−iωt3(e−3iωt+2),\bra{e_1}e^{-i\hat Ht/\hbar}\ket{e_1}=\frac13e^{-4i\omega t}+\frac23e^{-i\omega t}=\frac{e^{-i\omega t}}{3}\Bigl(e^{-3i\omega t}+2\Bigr),

and the probability is

P11(t)=19∣e−3iωt+2∣2=19(5+4cos⁡3ωt).P_{11}(t)=\frac19\bigl|e^{-3i\omega t}+2\bigr|^2=\frac19\bigl(5+4\cos3\omega t\bigr).

It oscillates between 11 and 19\frac19, at angular frequency 3ω3\omega, equal to the difference between the two eigenvalues of H^\hat H divided by ℏ\hbar: this is a beat between the two energy levels. For example, this model describes a particle that can move from one site to either of the other two sites of a triangle.