Figure 1.
Schedule of hardware-native two-qubit gates: (a) CX gate on qubit pair (0,1) of ibm_cairo, and (b) echoed cross-resonance (ECR) gate on qubit pair (0,1) of ibm_cusco. The system cycle time () is 2/9 ns ≈ 0.22 ns in ibm_cairo, while it is 0.50 ns in ibm_cusco. represents the drive channel acting on qubit i, and is the control channel for a corresponding qubit pair driving the control qubit c at the frequency of the target qubit t.
Figure 1.
Schedule of hardware-native two-qubit gates: (a) CX gate on qubit pair (0,1) of ibm_cairo, and (b) echoed cross-resonance (ECR) gate on qubit pair (0,1) of ibm_cusco. The system cycle time () is 2/9 ns ≈ 0.22 ns in ibm_cairo, while it is 0.50 ns in ibm_cusco. represents the drive channel acting on qubit i, and is the control channel for a corresponding qubit pair driving the control qubit c at the frequency of the target qubit t.
Figure 2.
Two types of dynamical decoupling (DD) sequences: (a) Carr–Purcell–Meiboom–Gill (CPMG) and (b) XY4. The delay time t represents the idle time of the qubit minus the duration of the corresponding X or Y pulses.
Figure 2.
Two types of dynamical decoupling (DD) sequences: (a) Carr–Purcell–Meiboom–Gill (CPMG) and (b) XY4. The delay time t represents the idle time of the qubit minus the duration of the corresponding X or Y pulses.
Figure 3.
Two implementations of a three-qubit QAOA: (a) CX implementation and (b) CZ implementation, where both are decomposed and optimized using Qiskit’s transpiler with optimization level 3. Highlighted yellow boxes represent CPMG sequences.
Figure 3.
Two implementations of a three-qubit QAOA: (a) CX implementation and (b) CZ implementation, where both are decomposed and optimized using Qiskit’s transpiler with optimization level 3. Highlighted yellow boxes represent CPMG sequences.
Figure 4.
Impact of circuit fidelity on algorithm performance and DD effectiveness: (a) normalized approximation ratio (NAR), (b) normalized success probability (NSP), (c) improvement in NAR after applying DD (), and (d) improvement in NSP after applying DD (). Higher values of , , , and indicate better performance on actual quantum devices. Values exceeding unity indicate that the performance achieved on the IBM quantum hardware surpasses the results obtained from the noise-free simulation. Positive values of and demonstrate improvements due to DD. The CPMG sequence is used for all data points. Each line in the graph represents a linear fit of the data. The reported and are 85.55% and 66.8%, respectively.
Figure 4.
Impact of circuit fidelity on algorithm performance and DD effectiveness: (a) normalized approximation ratio (NAR), (b) normalized success probability (NSP), (c) improvement in NAR after applying DD (), and (d) improvement in NSP after applying DD (). Higher values of , , , and indicate better performance on actual quantum devices. Values exceeding unity indicate that the performance achieved on the IBM quantum hardware surpasses the results obtained from the noise-free simulation. Positive values of and demonstrate improvements due to DD. The CPMG sequence is used for all data points. Each line in the graph represents a linear fit of the data. The reported and are 85.55% and 66.8%, respectively.
Figure 5.
Influence of logarithmic transformation of circuit schedule duration (
) on algorithm performance and DD effectiveness with the same datasets as in
Figure 4: (
a) NAR, (
b) NSP, (
c)
, and (
d)
.
Figure 5.
Influence of logarithmic transformation of circuit schedule duration (
) on algorithm performance and DD effectiveness with the same datasets as in
Figure 4: (
a) NAR, (
b) NSP, (
c)
, and (
d)
.
Figure 6.
Influence of circuit fidelity and on algorithm performance and DD effectiveness: (a) , (b) , (c) , and (d) .
Figure 6.
Influence of circuit fidelity and on algorithm performance and DD effectiveness: (a) , (b) , (c) , and (d) .
Figure 7.
Comparison of two hardware-native gate sets: {CX, ID, RZ, SX, X} and {ECR, ID, RZ, SX, X}, denoted as CX and ECR gate sets, respectively. Results obtained using four 27-qubit quantum processing units (QPUs) ibmq_mumbai, ibmq_kolkata, ibm_cairo, and ibmq_ehningen for the CX gate set, and four 127-qubit QPUs ibm_kyoto, ibm_cusco, ibm_brisbane, and ibm_sherbrooke for the ECR gate set: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit to the corresponding data.
Figure 7.
Comparison of two hardware-native gate sets: {CX, ID, RZ, SX, X} and {ECR, ID, RZ, SX, X}, denoted as CX and ECR gate sets, respectively. Results obtained using four 27-qubit quantum processing units (QPUs) ibmq_mumbai, ibmq_kolkata, ibm_cairo, and ibmq_ehningen for the CX gate set, and four 127-qubit QPUs ibm_kyoto, ibm_cusco, ibm_brisbane, and ibm_sherbrooke for the ECR gate set: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit to the corresponding data.
Figure 8.
Comparison of CX and CZ implementations of QAOA across four 27-qubit IBM QPUs ibmq_mumbai, ibmq_kolkata, ibm_cairo, and ibmq_ehningen: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit of the data.
Figure 8.
Comparison of CX and CZ implementations of QAOA across four 27-qubit IBM QPUs ibmq_mumbai, ibmq_kolkata, ibm_cairo, and ibmq_ehningen: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit of the data.
Figure 9.
Comparison of CPMG and XY4 sequences across seven IBM QPUs ibmq_mumbai, ibmq_kolkata, ibm_cairo, ibmq_ehningen, ibm_kyoto, ibm_cusco, and ibm_brisbane: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . Each line represents a linear fit of the data.
Figure 9.
Comparison of CPMG and XY4 sequences across seven IBM QPUs ibmq_mumbai, ibmq_kolkata, ibm_cairo, ibmq_ehningen, ibm_kyoto, ibm_cusco, and ibm_brisbane: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . Each line represents a linear fit of the data.
Figure 10.
Comparison of optimization level 1 (Opt1) and optimization level 3 (Opt3) across five IBM QPUs ibmq_kolkata, ibm_cairo, ibmq_ehningen, ibm_cusco, and ibm_kyoto: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit of the data.
Figure 10.
Comparison of optimization level 1 (Opt1) and optimization level 3 (Opt3) across five IBM QPUs ibmq_kolkata, ibm_cairo, ibmq_ehningen, ibm_cusco, and ibm_kyoto: (a) NAR, (b) NSP, (c) , (d) , (e) circuit fidelity, and (f) . The CPMG sequence is used for all data points. Each line represents a linear fit of the data.
Table 1.
Noise-free simulation results of the quantum approximate optimization algorithm (QAOA) for portfolio optimization using Qiskit’s Qasm simulator with 30,000 shots.
Table 1.
Noise-free simulation results of the quantum approximate optimization algorithm (QAOA) for portfolio optimization using Qiskit’s Qasm simulator with 30,000 shots.
Number of Qubits | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12 |
---|
Approximation ratio | 0.9751 | 0.4342 | 0.3776 | 0.3734 | 0.3589 | 0.3144 | 0.2806 | 0.3241 | 0.2933 | 0.3161 |
Success probability | 0.9747 | 0.1536 | 0.1131 | 0.0422 | 0.0227 | 0.0124 | 0.0065 | 0.0057 | 0.0018 | 0.0006 |
Table 2.
Parameters derived from the analysis of
Figure 4.
Table 2.
Parameters derived from the analysis of
Figure 4.
Metric | Mean | Fit Function | Correlation Coefficient | p-Value |
---|
| 0.610 | | | 0 |
| 0.687 | | | 0 |
| 0.556 | | | 0 |
| 0.631 | | | 0 |
| 0.077 | | | 0.00075 |
| 0.075 | | | 0.66997 |
Table 3.
Parameters derived from the analysis of
Figure 5.
Table 3.
Parameters derived from the analysis of
Figure 5.
Metric | Fit Function | Correlation Coefficient | p-Value |
---|
| | | 0 |
| | | 0 |
| | | 0 |
| | | 0 |
| | | 0 |
| | | 0.14281 |
Table 4.
Parameters derived from the analysis of
Figure 7.
Table 4.
Parameters derived from the analysis of
Figure 7.
Metric | Mean | Fit Function | Correlation Coefficient | p-Value |
---|
| 0.548 | | | 0 |
| 0.656 | | | 0 |
| 0.712 | | | 0 |
| 0.771 | | | 0 |
| 0.464 | | | 0 |
| 0.582 | | | 0.00092 |
| 0.618 | | | 0.00009 |
| 0.716 | | | 0.00044 |
| 0.108 | | | 0.09252 |
| 0.059 | | | 0.30429 |
| 0.118 | | | 0.08407 |
| 0.098 | | | 0.50688 |
Circuit fidelity (CX) | 0.539 | | | 0 |
Circuit fidelity (ECR) | 0.594 | | | 0 |
(CX) | 10.703 | | | 0 |
(ECR) | 10.577 | | | 0 |
Table 5.
Parameters derived from the analysis of
Figure 8.
Table 5.
Parameters derived from the analysis of
Figure 8.
Metric | Mean | Fit Function | Correlation Coefficient | p-Value |
---|
| 0.548 | | | 0 |
| 0.656 | | | 0 |
| 0.638 | | | 0 |
| 0.691 | | | 0 |
| 0.464 | | | 0 |
| 0.582 | | | 0.00092 |
| 0.635 | | | 0.00208 |
| 0.653 | | | 0.00006 |
| 0.108 | | | 0.09252 |
| 0.053 | | | 0.50206 |
| 0.118 | | | 0.08407 |
| 0.018 | | | 0.21130 |
Circuit fidelity (CX) | 0.539 | | | 0 |
Circuit fidelity (CZ) | 0.527 | | | 0 |
(CX) | 10.703 | | | 0 |
(CZ) | 10.799 | | | 0 |
Table 6.
Parameters derived from the analysis of
Figure 9.
Table 6.
Parameters derived from the analysis of
Figure 9.
Metric | Mean | Fit Function | Correlation Coefficient | p-Value |
---|
| 0.610 | | | 0 |
| 0.699 | | | 0 |
| 0.6 | | | 0 |
| 0.697 | | | 0 |
| 0.525 | | | 0 |
| 0.629 | | | 0.00001 |
| 0.509 | | | 0 |
| 0.648 | | | 0.00148 |
| 0.09 | | | 0.11748 |
| 0.097 | | | 0.05502 |
| 0.104 | | | 0.23425 |
| 0.139 | | | 0.33818 |
Circuit fidelity (CPMG) | 0.555 | | | 0 |
Circuit fidelity (XY4) | 0.55 | | | 0 |
(CPMG) | 10.589 | | | 0 |
(XY4) | 10.586 | | | 0 |
Table 7.
Parameters derived from the analysis of
Figure 10.
Table 7.
Parameters derived from the analysis of
Figure 10.
Metric | Mean | Fit Function | Correlation Coefficient | p-Value |
---|
| 0.554 | | | 0 |
| 0.657 | | | 0 |
| 0.618 | | | 0 |
| 0.688 | | | 0 |
| 0.509 | | | 0 |
| 0.589 | | | 0 |
| 0.536 | | | 0.00001 |
| 0.606 | | | 0.00056 |
| 0.103 | | | 0.12785 |
| 0.069 | | | 0.38513 |
| 0.08 | | | 0.65396 |
| 0.07 | | | 0.34466 |
Circuit fidelity (Opt1) | 0.417 | | | 0 |
Circuit fidelity (Opt3) | 0.559 | | | 0 |
(Opt1) | 10.540 | | | 0 |
(Opt3) | 10.475 | | | 0 |
Table 8.
Impact of hardware and algorithm factors on DD effectiveness and robustness.
Table 8.
Impact of hardware and algorithm factors on DD effectiveness and robustness.
Factor | Mean | Slope Coefficient | EMSR |
---|
| | | | | | | |
---|
Circuit fidelity | 0.610 | 0.077 | 0.556 | 0.075 | | | 85.55% | 66.80% |
Schedule duration | | |
CX gate set | 0.548 | | 0.464 | 0.118 | | | | |
ECR gate set | | 0.059 | 0.618 | 0.098 | | | | |
CX implementation | 0.548 | | 0.464 | 0.118 | | | | |
CZ implementation | 0.638 | 0.053 | | 0.018 | | | | |
CPMG sequence | 0.610 | 0.090 | 0.525 | 0.104 | | | | |
XY4 sequence | 0.600 | 0.097 | 0.509 | | | | | |
Opt1 | 0.554 | 0.103 | 0.509 | 0.080 | | | | |
Opt3 | 0.618 | 0.069 | 0.536 | 0.070 | | | | |