Iterative Quantum Phase

Estimation: Moving Beyond

Traditional QPE
Kaelyn Ferris, PhD
Qiskit Researcher & Community Coordinator
Introduction

- At this point you’ve learned a lot about basic quantum algorithms and have begun to start
considering the effects of noise.

- Many of these algorithms (Grover’s , Shor’s, QPE, etc.) require long depth circuits with many
nonlocal operators.

- Here we’ll discuss a technique to do phase estimation which is more achievable on today’s
quantum computers. Using a single auxiliary qubit to store the phase

3
 QPE Review

4
QPE Review
Why is the phase important to estimate in the first place?

5
QPE Review
Why is the phase important to estimate in the first place?

In physics, often we are looking for the eigenvalue, 𝜆, of an operator 𝑈.

Knowing information like this allows us to characterize a simulated
physical system

6
QPE Review
Why is the phase important to estimate in the first place?

In physics, often we are looking for the eigenvalue, 𝜆, of an operator 𝑈.

Knowing information like this allows us to characterize a simulated
physical system

But what form does this have generally?

7
QPE Review
Why is the phase important to estimate in the first place?

In physics, often we are looking for the eigenvalue, 𝜆, of an operator 𝑈.

Knowing information like this allows us to characterize a simulated
physical system

But what form does this have generally?

For some eigenstate, |𝜆⟩, of 𝑈

𝑈 𝜆 = 𝜆|𝜆⟩
⟨𝜆|𝑈 ! 𝑈 𝜆 = 𝜆∗ 𝜆⟨𝜆|𝜆⟩
⟨𝜆|𝑈 ! = ⟨𝜆|𝜆∗ = |𝜆|# = 1

8
QPE Review
Why is the phase important to estimate in the first place?

In physics, often we are looking for the eigenvalue, 𝜆, of an operator 𝑈.

Knowing information like this allows us to characterize a simulated
physical system

But what form does this have generally?

For some eigenstate, |𝜆⟩, of 𝑈

𝑈 𝜆 = 𝜆|𝜆⟩
𝜆 = 𝑒 !"#$
⟨𝜆|𝑈 ! 𝑈 𝜆 = 𝜆∗ 𝜆⟨𝜆|𝜆⟩
⟨𝜆|𝑈 ! = ⟨𝜆|𝜆∗ = |𝜆|# = 1

9
 QPE Review: Phase Kickback

Control

|0⟩ 0 ; Output State → 0 |𝜆⟩

|𝜆⟩ 𝑈 |1⟩; Output State → 𝑈 1 𝜆

= 𝑒 !"#$ 1 |𝜆⟩
Target

However, this is difficult to

extract. We should try
something different

10
 QPE Review: Phase Kickback
Apply a Hadamard to Output state becomes:
the control qubit
0 + 𝑒 !"#$ |1⟩
Control |𝜆⟩
2
|0⟩ 𝐻

|𝜆⟩ 𝑈 Now the control has a relative

phase between basis states
determined by 𝜃
Target

Recall also how we store the phase in binary:

$#%& 𝜃% 𝜃% 𝜃&
𝑒 ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2

11
 QPE Review

Recall QPE
algorithm

12
 QPE Review These gates will
require a lot of
Requires QFT
(also very expensive!)
SWAPs

We have 𝑡
qubits to store
the phase

13
 Iterative Phase Estimation

14
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.

15
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.
𝜃% 𝜃" 𝜃&
Recall: 𝑒 $#%& ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2

16
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.
𝜃% 𝜃" 𝜃&
Recall: 𝑒 $#%& ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2

|0⟩ 𝐻

!
|𝜓⟩ 𝑈!

' !
0 + 𝑒 $#%&# |1⟩ |𝜓⟩
#

17
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.
𝜃% 𝜃" 𝜃&
Recall: 𝑒 $#%& ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2
"
𝑘 = 1; 𝑒 "!#$! = 𝑒 "!#%.$# $$ …$%
|0⟩ 𝐻
#
𝑘 = 2; 𝑒 "!#$! = 𝑒 "!#$# 𝑒 "!#%.$$ …$%
! $
|𝜓⟩ 𝑈! 𝑘 = 3; 𝑒 "!#$! = 𝑒 "!#$# 𝑒 "!#$$ 𝑒 "!#%.$& …$%

' !
0 + 𝑒 $#%&# |1⟩ |𝜓⟩
#

18
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.
𝜃% 𝜃" 𝜃&
Recall: 𝑒 $#%& ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2
"
𝑘 = 1; 𝑒 "!#$! = 𝑒 "!#%.$# $$ …$%
|0⟩ 𝐻
#
𝑘 = 2; 𝑒 "!#$! = 𝑒 "!#$# 𝑒 "!#%.$$ …$%
! $
|𝜓⟩ 𝑈! 𝑘 = 3; 𝑒 "!#$! = 𝑒 "!#$# 𝑒 "!#$$ 𝑒 "!#%.$& …$%

' !
0 + 𝑒 $#%&# |1⟩ |𝜓⟩
#

19
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.
𝜃% 𝜃" 𝜃&
Recall: 𝑒 $#%& ; 𝜃 ∈ (0,1] 𝜃= + + ⋯ + & = 0. 𝜃%𝜃" … 𝜃&
2 4 2
"
𝑡 = 1; 𝑒 "!#$! = 𝑒 "!#%.$# $$ …$%
|0⟩ 𝐻
#
𝑡 = 2; 𝑒 "!#$! = 𝑒 "!#%.$$ …$%
! $
|𝜓⟩ 𝑈! 𝑡 = 3; 𝑒 "!#$! = 𝑒 "!#%.$& …$%
⋮
𝑐 %'#
𝑘 = 𝑚 − 1; 𝑒 "!#$! = 𝑒 "!#%.$%

' !
0 + 𝑒 $#%&# |1⟩ |𝜓⟩
#

20
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.

What is the first step?

|0⟩ 𝐻 𝐻

%'#
|𝜓⟩ 𝑈!
𝜃* = 0; +
𝑐
𝜃* = 1; |−⟩
'
0 + 𝑒 $#%(.&" |1⟩ |𝜓⟩
#

21
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.

The second step?

|0⟩ 𝐻

|𝜓⟩ 𝑈!
%'$

'
0 + 𝑒 $#%(.&"#$&" |1⟩ 𝜓
#

22
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.

Here, we’ll isolate 𝜃!"# with a

The second step? classically controlled rotation

|0⟩ 𝐻 𝑅( )!#$%
*

|𝜓⟩ 𝑈!
%'$

& &
' $#% "#$ $#% +"
0 + 𝑒 # 𝑒 |1⟩ 𝜓
#
23
 Iterative PE

Instead of needing so many auxiliary qubits, let’s use one and iterate the estimation of the phase.

The second step?

|0⟩ 𝐻 𝑅( )!#$%
𝐻
*

|𝜓⟩ 𝑈!
%'$

𝜃*,' = 0; +
𝑐
𝜃*,' = 1; |−⟩
&
' $#% "#$
0 + 𝑒 # |1⟩ 𝜓
#

24
 Iterative PE

Further Steps

"$%&!"#
|0⟩ 𝐻 𝑅( )!#$% ⋯ ⋯ 𝑅# '
!!

|𝜓⟩ 𝑈!
%'!

'
0 + 𝑒 $#%(.&"#!%$…&" |1⟩ 𝜓
#

25
 Iterative PE
Strategy: Iterate this circuit 𝑚 times by either resetting the
aux qubit or use dynamic circuits until the bitstring θ
= 0. 𝜃( 𝜃! … 𝜃+ is found.
Further Steps

"$%&!"#
|0⟩ 𝐻 𝑅( )!#$% ⋯ ⋯ 𝑅# ' 𝐻
!!

|𝜓⟩ 𝑈!
%'!

𝜃*,01' = 0; +
𝑐
𝜃*,01' = 1; |−⟩
'
0 + 𝑒 $#%&"#!%$⁄# |1⟩ 𝜓
#

26
 Estimating Observables

27
 Estimating Observables

The usage of auxiliary qubits to store phase We can use this technique of phase kickback to
information ends up being a useful tool! directly measure expectations values of a pair of
observables

𝑋 + 𝑖𝑌 = 〈𝑈 , 𝑉⟩
+

𝑉 or 𝑈 , can be any unitary operator,

including the identity: 𝐼
𝜓 𝑉 𝑈

29
 Estimating Observables
But how can this be the case?

We know that: 𝑋 + 𝑖𝑌 = 𝐻𝑍𝐻 + ⟨𝑍𝐻𝑍𝐻⟩

And: 0 𝑈 𝜓 + 1 𝑉|𝜓⟩
Ψ =
2
𝐻𝑍𝐻 = ⟨Ψ|𝐻𝑍𝐻|Ψ⟩ 𝑍𝐻𝑍𝐻 = ⟨Ψ|𝑍𝐻𝑍𝐻|Ψ⟩
= (! ⟨0|𝑈 , 𝜓 + ⟨1|𝑉 , 𝜓 𝐻𝑍𝐻 0 𝑈 𝜓 + 1 𝑉|𝜓⟩ = (! ⟨0|𝑈 , 𝜓 + ⟨1|𝑉 , 𝜓 𝑍𝐻𝑍𝐻 0 𝑈 𝜓 + 1 𝑉|𝜓⟩

= (! 𝜓 𝑈 , 𝑉 𝜓 + (! 𝜓 𝑉 , 𝑈 𝜓 = (! 𝜓 𝑈 , 𝑉 𝜓 − (! 𝜓 𝑉 , 𝑈 𝜓

𝑋 + 𝑖𝑌 = 𝜓 𝑈 ) 𝑉 𝜓 = ⟨𝑈 ) 𝑉⟩

30
 Estimating Observables
We can also utilize time evolution to
characterize physical systems

𝑋 + 𝑖𝑌 = 𝑒 !"# 𝐵$𝑒 %!"# 𝐴

+ 𝒞23 (𝑡)

𝜓 𝐴 𝑈(𝑡) 𝐵

31
 Review

- “Traditional” QPE is too expensive to achieve given the current hardware available.
Both in terms of the circuit depth and the required number of qubits.

- Using an auxiliary qubit to measure individual bits in the estimation string is much
more achievable.

- This allows us to estimate the phase by instead running more circuits which
is generally cheaper to do

- This technique of utilizing an auxiliary qubit to store and measure phase

information can be extended to characterize simulations of physical systems. Very
useful indeed!

32
Thank you