Elastic scattering on a quantum computer

Muhammad Yusf,∗ Ling Gan,† Cameron Moffat,‡ and Gautam Rupak§

Department of Physics & Astronomy and HPC2 Center for Computational Sciences,
Mississippi State University, Mississippi State, MS 39762, USA

Scattering probes the internal structure of quantum systems. We calculate the two-particle elastic
scattering phase shift for a short-ranged interaction on a quantum computer. Short-ranged inter-
actions with a large scattering length or shallow bound state describe a universality class that is of
interest in atomic, condensed matter, nuclear, and particle physics. The phase shift is calculated
by relating the ground state energy of the interacting particles in a harmonic trap. The relaxation
method is used as the variational quantum eigensolver for the ground state calculation. Schmidt
decomposition is used to reduce quantum circuits nominally requiring tens of qubits to 2-qubit cir-
cuits, thus reducing the noise in quantum measurements. Calculations in multi-particle systems with
many-body interactions would benefit from this reduction of qubits in noisy quantum processors.
arXiv:2406.09231v1 [nucl-th] 13 Jun 2024

Keywords: Quantum computing, variational quantum eigensolver, phase shift, scattering, Schmidt decom-
position

I. INTRODUCTION putation in evaluating dynamical properties. Further,

Monte Carlo calculations in Euclidean time suffer from
Scattering—elastic and inelastic—is a central tool for the fermionic sign problem [10–13] which becomes severe
probing the internal structure of a quantum system. The for large systems. This is another area where QCs are
discovery of the atomic electronic structure in Ruther- expected to be advantageous to classical computations.
ford’s 1911 gold-foil scattering experiment and the par- However, it is known that QC on the Noisy Intermediate-
ton structure of nucleons in SLAC-MIT deep inelastic Scale Quantum (NISQ) era computers has challenges.
scattering experiments [1–3] of the late 1960s are well- These include efficient qubitization of the fermionic and
known examples of that. Scattering experiments describe bosonic degrees of freedom [14, 15], and the extraction of
the time evolution of quantum systems subjected to a meaningful signals from noisy measurements. Even if we
Hamiltonian associated with the interactions between the had access to an ideal universal quantum computer, QC
scattering objects. On the other hand, nuclear many- of scattering cross sections is not straightforward [16, 17].
body calculations on classical computers are typically In a recent work [18], a general algorithm for calculat-
done in Euclidean time using Monte Carlo simulations. ing inelastic reactions a(b, γ)c, a(b, c)d where a, b, c, d are
Static properties such as the energy levels of atomic nu- atomic nuclei or particles, and γ a photon was presented.
clei can be calculated from first principles using micro- The frequency of the Rabi oscillations between the for-
scopic nuclear interactions, see Ref. [4] and references ward and backward reactions that would invariably re-
therein. However, working in Euclidean time, the extrac- sult from the unitary time evolutions were related to the
tion of real-time evolution of quantum systems becomes reaction amplitude. The algorithm was demonstrated
difficult. Currently, Lüschur’s method [5], the so-called for a single particle in one spatial dimension. Extension
Busch formula [6] and the spherical-wall method [7, 8] to higher dimensions and with more particles would be
are used to extract elastic scattering phase shifts in clas- needed to demonstrate a quantum advantage.
sical calculations. A general method for classical calcula- In this work, we calculate the two-particle elastic scat-
tions of inelastic scattering amplitude for nuclear many- tering phase shift for a short-ranged interaction. It de-
body systems in Euclidean-time has not been developed scribes a universal class of interactions of relevance in
yet. In relativistic field theory, the Maiaini-Testa theo- atomic, condensed matter, nuclear, and particle physics.
rem [9] precludes calculations of inelastic cross sections At momenta that are small compared to the inverse range
away from kinematical thresholds from Euclidean-time of the interaction, the two-particle interaction can be
correlation functions. treated as zero-ranged in the leading order approxima-
In contrast to classical computations, quantum com- tion. When the strength of the interaction is such that
putations (QCs) are done in Minkowski-time as unitary the two-particle scattering length is much larger than the
evolution of quantum states. Thus, there is an expecta- range of the interaction (a resonating system), the low
tion that it will provide an advantage over classical com- momentum phase shift becomes universal, independent
of the details of the interaction. This happens naturally,
for example, for two-nucleon and atomic 4 He-4 He s-wave
∗ scattering [19, 20]. Atomic systems near a Feshbach res-
[email protected]
† [email protected] onance can be tuned to be in this universal regime as
‡ [email protected] well [21, 22].
§ [email protected] The phase shift calculation in 1-spatial dimension uses
 2

the Busch formula that measures the energy shift of parti- Schrödinger equation where we have in a simple harmonic
cles in a harmonic trap due to the interaction. The classi- trap with frequency ω
cal calculation of the three-particle system in a harmonic
trap was treated in Ref. [23] that has been extended to 
1 d2 1 d2 1 1
many-body calculations as well [24]. Experimental real- − 2 − 2 + m1 ω 2 x21 + m2 ω 2 x22
2m1 dx1 2m2 dx2 2 2
ization of such a quantum system in a harmonic trap has
been possible [25, 26]. +c0 δ(x1 − x2 )] Ψ(x1 , x2 ) = Etotal Ψ(x1 , x2 ) , (3)
In section II, we present a derivation of the Busch
formula relating the relative two-particle center-of-mass for two particles with masses m1 and m2 at locations
(cm) energy to the scattering phase shift. The lattice x1 and x2 , respectively. In terms of cm X = (m1 x1 +
Hamiltonian for the energy calculation is presented in m2 x2 )/(m1 + m2 ), relative x = x1 − x2 coordinates, we
section III. The eigenenergy calculation of an eigenstate write for the cm motion
through quantum phase estimation is difficult since that
1 d2
 
would involve eigenstate preparation. Adiabatic time 1 2 2 1
− 2
+ M ω X ψn (X) = (n + )ωψn (X) ,
evolution is a physically motivated approach for state 2M dX 2 2
preparation. However, it is computationally expensive (4)
to obtain accurate results, even in a small system such
as a particle in a box, requiring many small time steps which describes a particle of mass M = m1 + m2 moving
with each time step incurring errors in NISQ machines in a harmonic trap of frequency ω with energy eigen-
significant enough to render the calculation impractical. function ψn (X). The energy spectrum of the cm motion
Thus, the energy calculation is performed using a vari- is the usual simple harmonic oscillator (SHO) energies
ational quantum eigensolver (VQE) where a trial wave (n + 1/2)ω with n = 0, 1, · · · .
function in the general two-particle coordinate system The relative motion is governed by the Schrödinger
is iteratively evolved for an upper bound on the ground equation
state energy. We use the relaxation method [27, 28] for
the iterations that are done classically. Results are pre- 1 d2
 
1 2 2
sented in section IV. Calculations using both ideal simu- − + µω x + c0 δ(x) ϕ(x) = Eϕ(x) , (5)
2µ dx2 2
lations and a physical quantum processor are presented.
The conclusions are in section V. that is solved exactly in a SHO basis by inserting in
Eq. (5)
II. PHASE SHIFT IN 1-D ∞
X
ϕ(x) = ϕn (x) , (6)
The 1-D Schrödinger equation for two particles inter- n=0
acting with a short-ranged force in s-wave is
where ϕn (x) are the energy egienfunctions of the SHO
1 d2 with mass µ and frequency ω. A self-consistent treat-
− ϕ(x) + c0 δ(x)ϕ(x) = Eϕ(x) , (1)
2µ dx2 ment [6] leads to the two-particle Green’s function rela-
tion
where x is the relative coordinate, µ the reduced mass,
E the relative energy. c0 > 0 corresponds to a repulsive ∞
X ϕm (0)ϕ⋆m (0) 1
potential with an outgoing scattering√solution ϕ(x > 0) = =− . (7)
E m − E c0
t exp(ipx) where momentum p = 2µE. For incident n=0
wave ϕ(x < 0) = exp(ipx), the complex transmission
amplitude is The 1/c0 term on the right hand side above is related
to the phase shift in Eq. (2). The left hand side can
p̂ be solved exactly using the known eigenfunctions ϕn (x)
t= eiδ(p) , (2)
1 + p̂2 resulting in the master equation

where the phase shift δ(p) = arctan(−1/p̂) with dimen- 1 p2

p2 1 p2 Γ( 4 − 4µω )
sionless p̂ = p/(µc0 ). p cot δ = − = √ , (8)
µc0 2 µω Γ( 3 − p2
The normalized bound state solution is ϕB (x) = 4 4µω )
√
γ exp(−γ|x|) for c0 < 0 where the binding energy
B = γ 2 /(2µ) with γ = −µc0 . Busch et al. [6] defines a where only the even n = 0, 2, · · · , eigenfunctions ϕn (0)
scattering length a0 = −1/(µc0 ) that gives γ = 1/a0 co- contribute at the origin. p cot δ has poles at energies
inciding with the 3-D effective range expansion for short- E = (2n + 1/2)ω and zeros at energies E = (2n + 3/2)ω,
ranged interactions [29]. respectively, for n = 0, 1, · · · . Several limits can be de-
The Busch formula [6] relating energy shift to scat- rived from the above that can serve as checks on the
tering phase shift is derived from the general coordinate numerical results.
 3

A. Weak coupling limit the energy curves go through E = (2n + 1/2)ω for n = 0
1, · · · . We use m1 = 940 MeV = m2 and ω = 10 MeV. At
In the limit |c0 | ≪ 1, one can use 1st order perturba- any coupling strength, numerically solving for E allows
tion theory to calculate the energy shift: one to calculate the scattering phase shift using Eq. (8).
The energy calculation is described in the next section.
∆E2n = ⟨ϕ2n |c0 δ(x)|ϕ2n ⟩ = c0 |ϕ2n (0)|2
r   r
µω n − 1/2 µω (2n − 1)!!
= c0 , = c0 , (9) 5
π n π 2n n!
that matches exactly with the weak coupling expansion of

E/ω
Eq. (8) with E ≈ (2n + 1/2)ω
+ ∆E2n for small ∆E2n ≪ 0
ω. We used the notation np for combinatorial n choose
p. Thepweak coupling limit is more precisely defined as
c0 ≪ ω/µ. -5

-1.0 -0.5 0.0 0.5 1.0

B. Strong coupling limit c0

We can provide a straightforward verification of the

p FIG. 1. The (blue) circles, (red) squares, (green) diamonds
strong coupling limit |c0 | ≫ ω/µ as a 1st order pertur-
and (purple) triangles are exact numerical calculations of the
bation in the case of an attractive interaction: relative energy E for various couplings c0 in the continuum
1 µω 2 and infinite volume. The colored solid curves are just to guide
∆E = ⟨ϕB | µω 2 x2 |ϕB ⟩ = , (10) the eye. The (black) dashed horizontal lines identify the poles
2 4γ 2 in Eq. (12) at E = (2n + 3/2)ω.
where the normalized bound state wave function
√
⟨x|ϕB ⟩ = γ exp(−γ|x|) with binding energy B =
γ 2 /(2µ). Directly from Eq. (8), expanding in ω ≪ E
III. VARIATIONAL QUANTUM EIGENSOLVER
r "   2 #
√ E 1 1/2 2ω
− µc0 ≈ 2 µω − 1− − , The phase shifts can be calculated once the relative
2ω 8 2 E
eigenenergies E at a fixed coupling c0 are known. As
µω 2 we work in general coordinates, the identification of a
⇒E + B = ∆E ≈ , (11)
4γ 2 generic eigenstate gets difficult since one has to add the
cm energy Ecm = (n + 1/2)ω to the relative energies
in agreement with the perturbative result where we used shown in Fig. 1. However, the ground state can always
γ = −µc0 . be identified as the lowest energy state. Thus we calcu-
late the total energy Etotal of the ground state and then
obtain the relative energy E = Etotal − ω/2 by subtract-
C. Energy shifts
ing the lowest cm energy. To get phase shifts at different
momenta, we simply vary the SHO frequency ω with a
The qualitative nature of the energy shift for nonzero larger value resulting in a more tightly bound wave func-
c0 can be understood from Eq. (8), written in an alternate tion and a higher momentum phase shift.
form The discretized Hamiltonian, in a box with L sites, in
r
ω Γ( 34 − 2ω
E
) the 2nd-quantization language is written as
−c0 = 2 1 E
, (12)
µ Γ( 4 − 2ω ) 2 L−2
X X †
Ĥ = −ĥ [ψ̂σ,l+1 ψ̂σ,l + h. c.]
that has zeros at E/ω = 2n + 1/2 and poles at E/ω = σ=1 l=0
2n + 3/2 for non-negative integer n. As the RHS goes 2 L−1
through zero (where c0 = 0), it changes sign while ap- 1 X X L−1 2 †
+ m̂ω̂ 2 [l − ] ψ̂σ,l ψ̂σ,l
proaching the poles (where c0 = ±∞) on either side of the 2 σ=1
2
l=0
zeros. One finds [6], for repulsive interactions c0 > 0, so- † †
lutions between the zeros and the poles for 12 < E/ω < 32 , + c0 (ψ̂2,l ψ̂2,l )(ψ̂1,l ψ̂1,l ) , (13)
5 7
2 < E/ω < 2 , · · · . For attractive interactions c0 < 0, the where the three terms are the kinetic energy, single-
solutions originate at the poles and terminate at the zeros particle harmonic potential and two-particle interactions,
for E/ω < 21 , 32 < E/ω < 25 , · · · , see Fig. 1. respectively. Dimensionful quantities in units of the lat-
The energy shifts in the presence of the two-body in- tice spacing b are indicated with a hat. The physical ex-
teraction are numerically calculated in Fig. 1. At c0 = 0 tent of the box is defined from −(L − 1)b/2 to (L − 1)b/2
 4

for odd integer L. We define the hopping parameter particles, one of each kind.
ĥ = 1/(2m̂) with m1 = m = m2 . The diagonal term
†
2ĥψ̂σ,l ψ̂σ,l in the kinetic operator at each lattice site
l = 0, 1, · · · , L − 1 for particle type σ = 1, 2 is not shown. A. Energy expectation and Schmidt decomposition
VQE is implemented through the relaxation
method [27, 28]. The differential operator in the The computation of Êtotal involves expectation value
non-relativistic Hamiltonian is a Laplacian whose solu- ⟨Ĥ⟩ calculations on a quantum computer. In a box with
tions are harmonic functions. These have the property L lattice sites, we require 2L qubits to represent a two-
that the function at the center of a region is given by particle state. The fermionic degrees of freedom in the
the average of the function on the boundary of the Hamiltonian, Eq. (13), are represented by Pauli matri-
region. This is the underlying principle of the relaxation ces using the Jordan-Wigner transformation [31]. The
method. Given some trial wave function in 1st quantized kinetic term in lattice units is
form Ψ(x1 , x2 ) we write an updated wave function [27]
on the lattice from Eq. (3) L−2
d = 4ĥ − ĥ
X
KE [Xl+1 Xl + Yl+1 Yl
ˆ
4ĥΨ 2
(new) i1 ,i2 l=0
Ψ̂i1 ,i2 = ,
4ĥ + [V̂i1 ,i2 − Êtotal ]Ψ̂i1 ,i2 ] + Xl+1+L Xl+L + Yl+1+L Yl+L ] , (15)
1
V̂i1 ,i2 = m̂ω̂ 2 (i21 + i22 ) + c0 δi1 ,i2 , where we put the particles of type-1 (the 1st particle) on
2
Ψi +1,i2 + Ψi1 −1,i2 + Ψi1 ,i2 +1 + Ψi1 ,i2 −1 the lowest L qubits, and the particles of type-2 (the 2nd
ˆ
Ψ = 1 ,
i1 ,i2 particle) on the highest L qubits. The two-particle coor-
4
(14) dinate (i, j) for particles type-1 and type-2, respectively,
is mapped to the computational basis |2j+L + 2i ⟩. This
where the updated wave function is proportional to the is done so that hopping of the particles is to the nearest
average of the nearest neighbors. Ψi1 ,i2 is the two- sites to avoid any Jordan strings. We adopt the nota-
particle wave function on the lattice with particle 1 at tion that only the non-identity operations in terms of the
site i1 and particle 2 at site i2 . The relaxation method Pauli matrices are shown explicitly. The diagonal term
†
is the classical component of the VQE algorithm once 2ĥψ̂σ,l ψ̂σ,l contributes an energy of 4h at each lattice site
the energy Êtotal is calculated on the quantum simula- shifting the overall energy by 4h that we do not calculate
tor. In the Gauss-Siedel method outlined in Ref. [27], explicitly on the quantum computer. The harmonic trap
only a single quantum computation of ⟨H⟩ with the trial contribution to the potential energy in Eq. (13) uses the
wave function is needed at the first step in the iterations, same qubit ordering of the two types of particles as the
with the rest of the updates done classically. We also KE
d term, and gives in lattice units
implemented the Jacobi method where each iteration re-
quires a quantum computation of ⟨H⟩ with the updated L−1
c SHO = 1 m̂ω̂ 2 L−1 2
] [1l − Zl ]
X
wave function. It is computationally expensive as each PE [l −
iteration requires a QC. On noisy simulators, we need 4 2
l=0
around 30 iterations to get a ground state estimate that L−1
1 L−1 2
] [1l+L − Zl+L ]
X
is within a few percent of the exact value. We do not use + m̂ω̂ 2 [l −
the Jacobi method for our final result. 4 2
l=0
The particular implementation of VQE we employ is m̂ω̂ 2 L(L2 − 1)
different from typical applications where the trial wave =
24
function ansatz is parameterized with unitary rotation 2L−1
angles that are varied to minimize the ground state en- m̂ω̂ 2 X L−1 2
− [mod(l, L) − ] Zl . (16)
ergy calculation. We find optimization with initial angles 4 2
l=0
chosen to reproduce the trial functions used in section IV
gives results with errors of about 20% for L = 5 even on The two particle interaction at each lattice site l =
an ideal simulator [30]. The error is found to be larger as 0, · · · , L − 1 is written as
L increases. We expect the errors to be larger on physical
devices due to the large number of two-qubit entangling
gates required in such an approach. The states parame-
†
4(ψ̂2,l †
ψ̂2,l )(ψ̂1,l ψ̂1,l ) → 12l+1 12l − 12l+1 Z2l
terized with unitary rotations often include superposition − Z2l+1 12l + Z2l+1 Z2l , (17)
of states other than those with a single particle of each
type as intended. Instead, in our calculations, we treat where we put particle of type-1 at the even numbered
the wave function amplitudes in position space as the qubits and particle of type-2 at the odd numbered qubits
parameters that are varied according to Eq. (14) for en- for convenience—coordinate (i, j) is mapped to compu-
ergy minimization. The evolved states in the relaxation tational basis |22j+1 + 22i ⟩—that would become evident
method remain within the subspace of states with two when we talk about Schmidt decomposition later. This
 5

gives for the two-particle interaction potential in lattice in the Hilbert space B is trivial:
units
⟨ψ|Z2L−1 ⊗ 1B |ψ⟩ = Tr(ρA Z2L−1 )
L−1
c0 c0 X X
PE
c int. = L+ [−Z2l − Z2l+1 + Z2l+1 Z2l ] [ρA ]aa′ ≡ [ρ]ab;a′ b . (22)
4 4 b
l=0
2L−1 L−1
c0 c0 X c0 X The calculation of the reduced density matrix ρA is
= L− Zl + Z2l+1 Z2l . (18)
4 4 4 simplified in the Schmidt decomposition. In the bipartite
l=0 l=0
state
We combine the potential energy contributions for con- X
venience into single Zl and double Z2l+1 Zl terms as |ψ⟩ = Mab |a, b⟩ , (23)
a,b
2 2
c Z = m̂ω̂ L(L − 1)
PE the probability amplitude Mab is interpreted as a ma-
24
2L−1  trix with row-column indices a-b. Then singular value
m̂ω̂ 2 X

L − 1 2 c0 decomposition (SVD) gives
− [mod(l, L) − ] − Zl ,
4 2 4
l=0
M = U SV † . (24)
L−1
c0 c0 X
PE
c ZZ = L+ Z2l+1 Z2l . (19) In our calculations we have size A < B, and U is of size
4 4
l=0 A × A and V † is of size A × B with a diagonal matrix
Saa = λa of size A×A. The Schmidt decomposition P of the
The Hamiltonian is given by the sum KE
d + PEc Z + PE
c ZZ .
state |ψ⟩ is in |i⟩
terms of a new set of bases A = a Uai |a⟩
The calculation of the energy expectation value ⟨Ĥ⟩ and |i⟩B = b Vbi⋆ |b⟩ such that
P
is reduced in VQE to the expectation values of Pauli
matrices for some trial state r≤A
X
22L −1
|ψ⟩ = λi |i⟩A ⊗ |i⟩B , (25)
X i=1
|ψ⟩ = ψk |k⟩ , (20)
k=0 where r is the Schmidt rank identifying the number of
non-zero singular values. One notes that the same num-
written in the computational basis |k⟩ for 2L qubits.
ber of bases from Hilbert space A and B are needed.
PConsider the measurement of ⟨Z2L−1 ⟩ =
⋆ ′ Even though the Hilbert space of B is much larger than
k,k ′ ψk ′ ψ k ⟨k |Z 2L−1 |k⟩ for the SHO PE term. A
A in our calculations, Schmidt rank r ≤ A contains all
little deliberation leads one to perform the quantum
the entanglement information between the two Hilbert
calculation in the reduced Hilbert space of only the
spaces. The reduced density matrix can now be written
highest (2L − 1)th qubit after tracing over the lower
by tracing over the subspace B as
qubits where the expectation value is simply one. Specif-
ically, we bipartate the Hilbert space into tensor product r≤A r≤A
A ⊗ B of dimensions A and B, respectively. Here, A is
X . X 2
ρA = B ⟨i|ψ⟩⟨ψ|i⟩B = λi |i⟩AA ⟨i| , (26)
a 2-dimensional space represented by the highest qubit i=1 i=1
q2L−1 and B a 22L−1 -dimensional space represented by
the lowest qubits q2L−2 · · · q0 with orthonormal bases which gives the required expectation value for a generic
|a⟩ and |b⟩, respectively. Then one writes [32], including operator
explicit identity operator in the lowest 2L − 2 qubits r≤A
⟨ψ|OA ⊗ 1B |ψ⟩ = Tr(ρA OA ) =
X
λ2i A ⟨i|OA |i⟩A . (27)
⟨ψ|Z2L−1 ⊗ 1B |ψ⟩ = ⟨ψ|Z2L−1 ⊗ 1B |a, b⟩⟨a, b|ψ⟩
X
i=1
a,b
In the quantum circuit, we prepare a superposition state
⟨a, b|ψ⟩⟨ψ|a′ b′ ⟩⟨a′ b′ |Z2L−1 ⊗ 1B |a, b⟩
XX
=
defined on the reduced space A:
a,b a′ ,b′
r≤A
X
= [ρ]ab;a′ b′ [Z2L−1 ]a′ a δb′ b X
a,a′ ,b,b′
|ψA ⟩ = λi |i⟩A , (28)
X i=1
= [ρ]ab;a′ b [Z2L−1 ]a′ a , (21)
a,a′ ,b which results in the required expectation values.
Single Pauli expectations such as ⟨Z2L−1 ⟩ are done in
where we define the tensor product state |a, b⟩ ≡ |a⟩ ⊗ |b⟩ the Hilbert space of size A = 2, represented by a single
and the density matrix ρ ≡ |ψ⟩⟨ψ| using standard nota- qubit. To construct the matrix M in Eq. (23) for Schmidt
tions. The relation in Eq. (21) is simplified by marginal- decomposition, we simply divide the qubit state |ψ⟩ in
izing over the index b where the expectation value of 1B two equal parts, and these become the two rows of the
 6

matrix M . For generic measurement ⟨Zl ⟩ with l ̸= 2L−1, 1, the full circuit with 10-qubits for L = 5 is compared
we simply re-index our site labels for the wave function with the results from measurements with 1- and 2-qubit
to put the relevant ql as the highest qubit, and then the circuits that result after SVD. We write the generic 2-
same procedure applies. qubit circuits using a single entangling gate as explained
For two Pauli expectations such as ⟨Z2L−1 Z2L−2 ⟩ in in appendix A. The number of measurements (shots) is
Eq. (17), we divide the state |ψ⟩ into 4 equal parts, which as indicated, and should
√ result in very small statistical
form the 4 rows of the matrix M in Eq. (23) for Schmidt errors scaling as 1/ shots. One can see that the SVD
decomposition. It is evident, we want to put the Pauli circuit measurement is more accurate than the full cir-
matrices whose expectation value we wish to calculate cuit measurement on QPU, as expected. For Runs 1
at the highest two qubits. This explains the ordering of and 4, we did not specify any particular initial layout
the two-particle qubits we used for the kinetic and the for the physical qubits, and let qiskit select the necessary
interaction terms in Eqs. (15), (17). connected qubits to implement the appropriate circuits.
We find that using the Schmidt decomposition to trace Runs 2 and 3 use the same parameter values but are run
out the spectator space B that only involves the iden- on different physical qubits as indicated in Table II. We
tity operator 1B , gives more accurate results in the noisy selected the qubits that had low posted readout errors at
QCs. The Schmidt decomposition can be optimized fur- runtime. The results are noticeably different. Coinciden-
ther by noting that one could perform SVD on the smaller tally, combining the 1-qubit measurements from Run 2
square matrix M M † of size A × A such that with the 2-qubit measurements from Run 3 gives nearly
perfect results in this particular instance.
M M † = U S2U † , (29) Looking more closely at the discrepancy between the
which avoids storing and manipulating large matrices Run 2 and Run 3 measurements in Table II, one no-
with at least one of the dimensions as large as 22L−1 . tices that the individual expectation value measurements
We only need S, U to construct the reduced state |ψA ⟩. of ∆KE, ∆PEZZ and ∆PEZ are similar, differing no
more than 10% between the runs. However, there are
large cancellations between the various terms in the to-
IV. RESULTS tal energy calculation. This coincidentally gives Etotal =
22.5 MeV when using ∆PEZ from Run 2 with ∆KE and
For the numerical work, we continue from subsec- ∆PEZZ from Run 3 in close agreement with the exact
tion II C with particle masses m1 = 940 MeV = m2 mo- result 22.33 MeV for the trial wave function, see Table I.
tivated by the proton-neutron masses. The range of the Nevertheless, we use the noisy results as listed in Table I
nuclear potential is set by the Compton wavelength of in the relaxation method for the ground state calculation
the pion, and it is roughly around a fermi. At distances to evaluate how well the relaxation method performs in
much larger than the pion Compton wavelength the nu- calculating the ground state energy.
clear interaction can be treated as point-like. Thus, to In Fig. 2 we show the phase shift calculated using VQE
make the numerical work physically motivated, we take with the Gauss-Siedel relaxation method on an ideal sim-
a lattice spacing b ∼ 1 fm and Lb ∼ 10 fm such that we ulator and QPU measurements from Table I. We per-
can calculate the two-particle scattering phase shifts at formed 200 iterations on a classical computer though
momenta Lb 1
≪ p ≪ 1b that approximates the continuum around 30 iterations would have been sufficient to get
and infinite volume results. a variational ground state energy within a few percent of
We performed several QCs using the ideal simulator the exact result. The lowest three momentum data points
on qiskit [33] and the IBMQ-Brisbane (ibmq_brisbane) were calculated at lattice spacing b = 1.25 fm, lattice size
quantum processor unit (QPU) at different values of box L = 17 where the SHO oscillator frequency was varied
size L, lattice spacing b, and oscillator frequency ω. The from ω = 3 MeV to 5 MeV in 1 MeV interval. The Hilbert
two-particle interaction strength c0 was also varied in space for this lattice size is 234 ≈ 2 × 1010 dimensional,
the simulator but for the QPU calculations, we limited yet, we calculate the expectation values using at most 22
the measurements to a single, conveniently chosen, non- dimensional mixed state after Schmidt decomposition.
p
perturbative c0 = 0.2 ≫ πω/(4µ) value to fully utilize The next 4 data points, in Fig. 2, were at b = 1.0 fm,
the resources available in the IBMQ Open Plan. L = 15 and from ω = 6 MeV to 9 MeV in 1 MeV interval.
In Table I, the theory parameters L, b, c0 , ω are listed The highest momentum point was at b = 0.8 fm, L = 13
to identify the specific “runs” of QCs. The trial wave and ω = 10 MeV. The lattice parameter b, L choices
function was chosen to be the normalized ground state give phase shift quite comparable to the continuum, in-
of two free particles in a box that is non-zero only at finite volume black dot-dashed curve. The QPU results
most 7 lattice sites in the middle of the harmonic trap. are on top of the ideal simulator results. The Gauss-
We get similar results using the ground state of two non- Siedel relaxation method is very efficient in reproducing
interacting particles in a harmonic trap. Using a trial the ground state wave function and energy. This leads
wave function with a compact support makes the SVD to a perfect match to the ideal VQE results.
calculation more accurate and gives less noisy results The small discrepancy between the analytical and nu-
when tested on the noisy qiskit simulator [33]. In Run merical results in Fig. 2 results from lattice artifacts
 7

TABLE I. On IBMQ-Brisbane, the single Pauli measurements were done with 10000 shots and the double Pauli measurements
were done with 20000 shots, with optimization level=2. The last four columns are energy measurements relative to the exact
energy.
Run L b (fm) ω (MeV) c0 Exact (MeV) Full ideal Full QPU SVD ideal SVD QPU
1 5 1.5 5 0.2 21.33 1.003 1.873 0.9599 1.009
2 13 0.8 10 0.2 22.33 — — 0.9905 1.459
3 13 0.8 10 0.2 22.33 — — 0.9905 1.346
4 15 1.0 6 0.2 15.52 — — 0.9723 1.324

c0 L mω 2 b2 L(L2 −1)
TABLE II. Expectation components ∆KE = KE − 4h, ∆PEZZ = PEZZ − 4b
, ∆PEZ = PEZ − 24
.

run L qubits ∆KE (MeV) KE (MeV) ∆PEZZ (MeV) PEZZ (MeV) ∆PEZ (MeV) PEZ (MeV) Etotal (MeV)
2 13 [8,9] -111 18.8 118 278 -405 -264 32.6
3 13 [84,85] -116 13.5 113 273 -397 -257 30.1

-40 noise in the quantum measurements renders the calcu-

lation inaccurate. A hybrid approach for two-nucleon
VQE ideal
-50 VQE IBMQ scattering where the spin degrees of freedom are evolved
on a quantum computer and the spatial wave function is
evolved on a classical computer was presented in Ref. [35].
-60 They calculated the trajectories for two-nucleon scatter-
ing though a phase shift was not evaluated for us to com-
δ

-70 pare with.

During the writing of this manuscript, we discovered
a recent work by Wang et al. [36] that also proposes us-
-80
ing a harmonic trap for phase shift calculations. In their
work, they use the rodeo algorithm [37] to calculate the
-90 energy eigen values for the phase shift determinations.
0 20 40 60 80 100 120
p (MeV ) Two example calculations using a spherical-well poten-
tial and a Woods-Saxon potential were performed on an
ideal qiskit simulator. QC on a physical device was not
FIG. 2. Two-body phase shift data calculated on the qasm- considered [36].
simulator as explained on the text. These are compared with
the analytical continuum, infinite volume (black) dot-dashed
curve. V. CONCLUSIONS

We calculated the two-particle scattering phase shift

which can be removed using smaller lattice spacings and for a short-ranged interaction on a quantum computer.
larger boxes. Larger boxes increase the number of cir- The calculation used the Busch formula [6] to relate the
cuits. We chose box sizes and lattice spacings in accor- relative energy of two interacting particles in a harmonic
dance with the available computational resources while trap to the phase shift. As the exact form of the Busch
keeping the errors within a few percent. formula is known, the challenge is to calculate the ener-
Before concluding, we comment on a few recent works gies for the various eigenstates of the system.
on scattering on quantum computers. The work by The ground state energy was used for the phase shift
Sharma et al. [34] calculates the two-particle scattering calculations. It was obtained using a VQE where the
phase shift as well. They work in relative coordinates probability amplitudes of the trial wave function in co-
where a hard spherical-wall boundary condition is im- ordinate space were used as the variational parameters.
posed, and extract the phase shift from the excited state For two particles, this required L2 amplitudes on a lattice
energies. The excited states above the bound ground with L sites. In the variational calculations, we used the
state are the discrete lattice spectrum of the continuum relaxation method [27, 28] which requires initial inputs
states. Their calculations were done in an abstract en- of the trial wave function and the corresponding energy
ergy eigenbasis such that n qubits are required to calcu- that was calculated on a quantum computer.
late phase shifts at (n − 1) momenta from the (n − 1) ex- The Hamiltonian for the energy expectation values
cited states above the ground state. The analytic phase was defined in a 22L dimensional Hilbert space in 2nd
shift is reproduced for n ≤ 4 but for larger qubits the quantization—a qubit per lattice site for each type of
 8

particles. For the two-particle system, the VQE requires ACKNOWLEDGMENTS

expectation value measurements of at most two Pauli ma-
trices at a time that we reduced to measurements on
The authors benefited from discussions with Paulo Be-
two-qubit circuits instead of on 2L-qubit circuits using
daque, Thomas Cohen, and Ratna Khadka. We thank
Schmidt decomposition. The two-qubit circuits needed
Dean Lee for his insightful comments on the manuscript.
only a single CNOT gate thus reducing the noise on NISQ
This work was partially supported by grants DOE DE-
hardware measurements compared to the same calcula-
SC00211, DOE DE-SC0024286 and NSF PHY-2209184.
tion on 2L qubits that didn’t use the Schmidt decomposi-
Part of the work was completed by GR at the Mary-
tion. We found the variational calculations to be robust
land Center for Fundamental Physics, Univ. of Mary-
against noisy measurements. For the system we studied,
land, College Park during his sabbatical. We acknowl-
not only the ground state energy but also the ground
edge the use of IBM Quantum services for this work. In
state wave function was obtained accurately.
this paper we used ibmq_brisbane.
Generalization to many-body systems with A particles
would require A × L qubits on a lattice with L sites.
In the relaxation method, one would need LA ampli-
tudes even though the Hilbert space grows as 2AL . A
Appendix A: 2-qubit state
rough estimate of the memory requirement for the clas-
sical component of the hybrid calculation would restrict
us to about 10 particles in one spatial dimension or a Consider a generic 2-qubit state
few particles in three spatial dimensions with L ∼ 10
on a laptop. However, these are not hard limits as one |ψ⟩ = α0 |00⟩ + α1 |01⟩ + α2 |10⟩ + α3 |11⟩ . (A1)
can work sequentially with the probability amplitudes in
Eq. (14) from numbers stored on a hard drive. Thus, the
relaxation method might remain practical for larger sys- If |α0 |2 + |α1 |2 = 0 or |α2 |2 + |α3 |2 = 0, then
tems. Schmidt decomposition can be applied to reduce we have product states |1⟩ ⊗ (α2 |0⟩ + α3 |1⟩) or |0⟩ ⊗
the number of qubits on the quantum measurements. A (α0 |0⟩ + α1 |1⟩), respectively. These product states
n-body interaction would be reduced to measurements on can be prepared with 1-qubit gates. In particular
n-qubit circuits instead of the full (A × L)-qubit circuit. the state a|0⟩ + b|1⟩ is created with unitary rotation
The reduction in qubit utilization was shown to reduce Rz [phase(b)]Ry [2 cos−1 (a)]Rz [− phase(b)] for a > 0 with-
the noise in measurements. This would be beneficial in out loss of generality. The generic case not covered by
many-body calculations. There is usually a hierarchy either of the two situations above is prepared as follows.
where the lower-body interactions dominate at low den- We write the 2-qubit state after Schmidt decomposi-
sity. For example, in nuclear systems, chiral symmetry tion as
limits the interactions to at most 4-nucleon interactions
at the leading order approximation [38]. α0 |00⟩ + α1 |01⟩ + α2 |10⟩ + α3 |11⟩
Future work should involve calculations in higher di-
mensions. The current calculation involving a delta func- = (u ⊗ v ⋆ )(λ1 |00⟩ + λ2 |11⟩) , (A2)
tion should be extended to include derivatives of delta
function which would bring in the effective range cor- where the unitary matrices u and v † are the left and right
rection to the phase shift. Moreover, few particles in matrices used in the SVD. The entangled state λ1 |00⟩ +
higher dimensions should be investigated to explore the λ2 |11⟩ is created from (λ1 |0⟩ + λ2 |1⟩) ⊗ |0⟩ with a single
effectiveness of the relaxation method given that conven- CNOT gate. The (λ1 |0⟩ + λ2 |1⟩) state with λi ≥ 0 is
tional VQE with unitary rotations was found to have created with a single Ry (2 cos−1 λ1 ) gate. The unitary
insufficient accuracy for the two-particle system in one operations u, v ∗ are implemented with single-qubit gates
dimension. to obtain the desired 2-qubit state.

[1] J. I. Friedman, Rev. Mod. Phys. 63, 615 (1991). Physics A 424, 47 (1984).
[2] H. W. Kendall, Rev. Mod. Phys. 63, 597 (1991). [8] B. Borasoy, E. Epelbaum, H. Krebs, D. Lee, and U.-G.
[3] R. E. Taylor, Rev. Mod. Phys. 63, 573 (1991). Meißner, Eur. Phys. J. A34, 185 (2007).
[4] B.-N. Lu, N. Li, S. Elhatisari, D. Lee, E. Epelbaum, [9] L. Maiani and M. Testa, Physics Letters B 245, 585
and U.-G. Meißner, Phys. Lett. B 797, 134863 (2019), (1990).
arXiv:1812.10928 [nucl-th]. [10] E. Y. Loh, J. E. Gubernatis, R. T. Scalettar, S. R. White,
[5] M. Lüscher, Nucl. Phys. B 354, 531 (1991). D. J. Scalapino, and R. L. Sugar, Phys. Rev. B 41, 9301
[6] T. Busch, B.-G. Englert, K. Rzażewski, and M. Wilkens, (1990).
Found. Phys. 28, 549 (1998). [11] G. Aarts, Journal of Physics: Conference Series 706,
[7] J. Carlson, V. Pandharipande, and R. Wiringa, Nuclear 022004 (2016).
 9

[12] J. Carlson, S. Gandolfi, F. Pederiva, S. C. Pieper, R. Schi- Phys. Rev. A 76, 063613 (2007), arXiv:0705.4335 [cond-
avilla, K. E. Schmidt, and R. B. Wiringa, Rev. Mod. mat.other].
Phys. 87, 1067 (2015). [25] M. Köhl, H. Moritz, T. Stöferle, K. Günter, and
[13] S. Elhatisari, E. Epelbaum, H. Krebs, T. A. Lähde, T. Esslinger, Phys. Rev. Lett. 94, 080403 (2005).
D. Lee, N. Li, B.-n. Lu, U.-G. Meißner, and G. Rupak, [26] T. Stöferle, H. Moritz, K. Günter, M. Köhl, and
Phys. Rev. Lett. 119, 222505 (2017). T. Esslinger, Phys. Rev. Lett. 96, 030401 (2006).
[14] B. Bauer, S. Bravyi, M. Motta, and G. K.-L. Chan, [27] D. V. Schroeder, American Journal of Physics 85,
Chemical Reviews 120, 12685 (2020), pMID: 33090772. 698 (2017), https://pubs.aip.org/aapt/ajp/article-
[15] A. Alexandru, P. F. Bedaque, A. Carosso, M. J. Cervia, pdf/85/9/698/13126437/698 1 online.pdf.
and A. Sheng, Phys. Rev. D 107, 034503 (2023), [28] J. D. Jackson, Classical Electrodynamics, 3rd ed. (John
arXiv:2209.00098 [hep-lat]. Wiley & Sons, Inc., Hoboken, 1998) , sections 1.12 and
[16] S. P. Jordan, K. S. M. Lee, and J. Preskill, Science 336, 1.13.
1130 (2012), arXiv:1111.3633 [quant-ph]. [29] H. A. Bethe, Phys. Rev. 76, 38 (1949).
[17] S. P. Jordan, K. S. M. Lee, and J. Preskill, (2014), [30] C. Moffat, (2024), master’s thesis research.
arXiv:1404.7115 [hep-th]. [31] P. Jordan and E. Wigner, Z. Phys. 47, 631 (1928).
[18] P. F. Bedaque, R. Khadka, G. Rupak, and M. Yusf, [32] D. H. McIntyre, Quantum Mechanics: A paradigms ap-
(2022), arXiv:2209.09962 [nucl-th]. proach (Pearson, Boston, 2011).
[19] R. E. Grisenti, W. Schöllkopf, J. P. Toennies, G. C. [33] Qiskit contributors, “Qiskit: An open-source framework
Hegerfeldt, T. Köhler, and M. Stoll, Phys. Rev. Lett. for quantum computing,” (2023).
85, 2284 (2000). [34] S. Sharma, T. Papenbrock, and L. Platter, (2023),
[20] R. B. Wiringa, V. G. J. Stoks, and R. Schiavilla, Phys. arXiv:2311.09298 [nucl-th].
Rev. C 51, 38 (1995). [35] F. Turro et al., Phys. Rev. A 108, 032417 (2023).
[21] K. M. O’Hara, S. L. Hemmer, M. E. Gehm, S. R. [36] P. Wang, W. Du, W. Zuo, and J. P. Vary, (2024),
Granade, and J. E. Thomas, Science 298, 2179 (2002). arXiv:2401.17138 [nucl-th].
[22] C. Regal, C. Ticknor, J. Bohn, and D. Jin, Nature (Lon- [37] K. Choi, D. Lee, J. Bonitati, Z. Qian, and J. Watkins,
don) 424, 47 (2003). Phys. Rev. Lett. 127, 040505 (2021).
[23] T. Luu and A. Schwenk, Phys. Rev. Lett. 98, 103202 [38] C. J. Yang, A. Ekström, C. Forssén, G. Hagen, G. Ru-
(2007), arXiv:cond-mat/0606069. pak, and U. van Kolck, Eur. Phys. J. A 59, 233 (2023),
[24] I. Stetcu, B. R. Barrett, U. van Kolck, and J. P. Vary, arXiv:2109.13303 [nucl-th].