Notes 8.370/18.

435 Fall 2022

Lecture 3 Prof. Peter Shor
Last time, we started talking about quantum mechanics. We mentioned the princi-
ple that: Isolated quantum systems evolve unitarily. That is, for an isolated system (one
that does not interact with its environment), there is a unitary matrix Ut such that if we
let the system alone for time t, | Ψt y “ Ut | Ψ0 y, where the initial state of the system is
| Ψ0 y and ehe final state of the system is | Ψt y.
Unitary matrices U are the complex analog of rotation matrices, also called orthog-
onal matrices; they take unit length complex vectors to unit length complex vectors. A
matrix is unitary if and only if U : U “ I. (An equivalent condition is U U : “ I.) Here
U : is the complex conjugate transpose of U (also called the Hermitian transpose). We
mentioned three specific unitary matrices last time, the Pauli matrices:
ˆ ˙ ˆ ˙ ˆ ˙
0 1 0 ´i 1 0
σx “ , σy “ , σz “ .
1 0 i 0 0 ´1

We discussed the representation of spin- 21 states on the Bloch sphere, and we looked at
the actions of the Pauli matrices on the Bloch sphere, these being 180˝ rotations around
the x-, y-, and z-axes, respectively. .
Today, we’ll be talking about how unitary matrices arise in quantum mechanics,
and then talk more about the Bloch sphere and rotations on the Bloch sphere.
So suppose you want to build a quantum computer, and you want to implement
a unitary matrix U . What do you do? There’s no magic incantation that takes the
state directly from | ψy to U | ψy. In fact, quantum unitary evolution can only changes
quantum states continuously, and not in discrete jumps.
To explain how to implement a unitary gate, first I need to say something about
quantum mechanics. Quantum mechanics assumes that isolated systems evolve ac-
cording to Schödinger’s equation,
d
i~ | ψptqy “ H | ψptqy .
dt
Here ~ is a very small physical constant, | ψptqy is the quantum state at time t, and H
is a Hamiltonian—a Hermitian1 matrix, to be exact, which can be associated with a
Hermitian quadratic form that takes the quantum state as input and outputs its energy.
If we assume that H is constant, we can solve this equuation:

| ψptqy “ e´iHt{~ | ψp0qy

Let’s assume that the eigenvectors of H are | ξ1 y, | ξ2 y, . . ., | ξd y, with λi be the

eigenvalue corresponding
ř to | ξi y. What the above equation means is that if the intial
state | ψp0qy is i αi | ξi y, then
ÿ
| ψptqy “ e´λi t{~ αi | ξi y .
i
1A Hermitian matrix is one that satisfies M “ M : .

1
We’re not actually going to be using Schrödinger’s equation until much later in the
course. I’m introducing it now to give you some idea as to how you might implement
unitary gates in practice, and to motivate the next thing I’ll be talking about, which is
three one-qubit gates which are rotations of the Bloch sphere by an angle θ. These are

Rx pθq “ e´iθσx {2 ,
Ry pθq “ e´iθσy {2 ,
Rz pθq “ e´iθσz {2 ,

What are these rotations? We have Rz pθq “ e´iθσz {2 . To exponentiate a diagonal

matrix, we can simply exponentiate each of the elements along the diagonal This gives
ˆ ´iθ{2 ˙ ˆ ˙
´iθσz {2 e 0 ´iθ{2 1 0
Rz pθq “ e “ “e .
0 eiθ{2 0 eiθ

You can check that if θ “ π{2, this is a rotation of a π{2 angle around the z-axis
(multiplied by a global phase) and if θ “ π, this is a rotation of π around the z-axis
(again multiplied by a global phase). So this gives us rotations around the z axis of an
arbitrary angle.
To compute Ry pθq, we could diagonalize it to get D “ U : Ry pθqU , and then ex-
ponentiate D to get Ry pθq “ U eD U : . We will compute it using a different method, to
show how this method works. What we do is use a Taylor expansion.
ˆ ˙2 ˆ ˙3 ˆ ˙4 ˆ ˙5
θ 1 θ 1 θ 1 θ 1 θ
e´iθσy {2 “ I ´i σy ´ σy2 `i σy3 ` σy4 ´i σy5 `. . .
2 2 2 3! 2 4! 2 5! 2
Since σy2 “ I, we have
ˆ ˙2 ˆ ˙3 ˆ ˙4
θ 1 θ 1 θ 1 θ
e´iθσy {2 “ I ´ i σy ´ I `i σy ` I ...
2 2 2 3! 2 4! 2
˜ ˆ ˙2 ˆ ˙4 ¸ ˜ ˆ ˙3 ˆ ˙5 ¸
1 θ 1 θ θ 1 θ 1 θ
“I 1´ ` ´ . . . ´ iσy ´ ` ´ ...
2 2 4! 2 2 3! 2 5! 2
θ θ
“ I cos ´ iσy sin
2 2
ˆ ˙
0 ´1
But ´iσy “ , so we see that
1 0
ˆ ˙
cos θ2 ´ sin θ2
e´iθσy {2 “ ,
sin θ2 cos θ2
which we see is a rotation in the x ´ z plane. Looking at its action on the states of the
Bloch sphere, it rotates the Bloch sphere around the y-axis by an angle of θ.
We can similarly see that
ˆ ˙
cos θ2 ´i sin θ2
e´iθσx {2 “ ,
´i sin θ2 cos θ2

2
Now that we have Ry pθq and Rz pθq, we can perform any one-qubit unitary by applying

Rz pθ3 qRy pθ2 qRz pθ1 q.

To see this, let us visualize what this does to the Bloch sphere. We first rotate the Bloch
sphere by an angle of θ1 around the north pole. The second rotation moves the north
pole down to an arbitrary longitude. Finally, by applying Rz pθ3 q, we can move the
north pole to end up at an arbitrary latitude. Combined, these give an arbitrary rotation
of the Bloch sphere. See the figure

Figure 1: Performing an arbitrary rotation by using three rotations, around the z-axis,
the y-axis, and the z-axis.

We now explain one way to find the quantum state | φp y corresponding to a point p
on the Bloch sphere. There are other ways to do this, which lead to simpler expressions;
we may revisit this question later in the term and explain them.
Suppose we have a point p “ ppx , py , pz q on the Bloch sphere pi P R. Since it’s on
a unit sphere, we must have p2x ` p2y ` p2z “ 1. Let us consider the 2 ˆ 2 matrix

Mp “ p x σ x ` p y σ y ` p z σ z ,

Since σx , σy , σz are Hermitian matrices, Mp must also be Hermitian, that is Mp “ Mp: .

Now,

Mp2 “ pp2x ` p2y ` p2z qI ` 2px py pσx σy ` σy σx q ` 2py pz pσy σy ` σz σy q ` 2px pz pσx σz ` σz σx q “ I.

where the last equality follows from the facts that the vector p has length 1 and σa σb “
´σb σz if a ‰ b.
We can also show that

TrMp “ px Trσx ` py Trσy ` pz Trσz “ 0,

since the Pauii matrices all have trace 0. Since Mp2 “ I, its eigenvalues have to be
˘1. And since its trace is 0, one of its eigenvalues has to be 1 and the other has to
be ´1. Now, we will let the eigenvector with eigenvalue `1 be the corresponding
quantum state | ψp y to point p on the unit sphere. We already know this holds for the

3
unit vectors on the x, y, and z axes. Note also that since Mp is a Hermitian matrix its
two eigenvectors are orthogonal, and since Mp “ ´M´p , the ´1 eigenvector of Mp is
the point antipodal to | φp y. ´ ¯
Let’s illustrate this by an example. Let p “ ?1 , 0, ?1 . Then
2 2
ˆ ˙
1 1 1
Mp “ ? .
2 1 ´1

Now, Mp has eigenvalues 1 and ´1, and its eigenvectors are

a ? a ? a ? a ?
2` 2 2´ 2 2´ 2 2` 2
| 0y ` | 1y and ´ | 0y ` | 1y ,
2 2 2 2
where the first eigenvector has eigenvalue `1 and the second ´1.
These vectors are in fact just
π π π π
cos | 0y ` sin | 1y and ´ sin | ´y ` cos | 1y .
8 8 8 8
This makes sence, because the point p “ ?12 p1, 0, 1q is halfway between p “ p1, 0, 0q
and p “ p0, 0, 1q, and these points correspond to cos π4 | 0y ` sin π4 | 1y and cos 0 | 0y `
sin 0 | 1y, respectively.