fs1132429 / ieee-quantum-hackathon

Travelling Salesman Problem

An IEEE quantum hackathon project. This project includes a qiskit run-time routine to solve the travelling salesman problem with 4 nodes using the built-in VQE program. Our report pdf has been submitted to arxiv.org and we have submitted a PDF version in this repository. We will update the link to the arxiv report when it is published.

Overview of Solution

Throughout our attempt to build this routine by following tutorials and creating our own functions, we encountered many errors and made some discoveries. We successfully used the built-in VQEProgram Object from qiskit_optimization.runtime to solve the travelling salesman problem with 3 nodes. This is seen in the 3-node-runtime-vqe.ipnyb. We then tried to do the same with 4 nodes for the graph. This worked successfully when we used the local VQE object from qiskit.algorithms as seen in local-vqe.ipnyb.

Issues with Qiskit Runtime API and 4 nodes

The final test, our ultimate objective, was to run the tsp solver solution with 4 nodes through the Qiskit Runtime API. This is where we encountered issues in running the api. When running it through the VQEProgram Object from qiskit_optimization.runtime in local Runtime, the program threw errors mentioning that the amount of max retires were exceeded. This happened due to the high amount of iterations required to run the algorithm with 4 nodes. The iterations could not be reduced due to algorithm limitations. This error is seen in errors-found.ipnyb.

To workaround this, we tested a direct call to the qiskit api on the following server: https://runtime-us-east.quantum-computing.ibm.com/openapi/ . These api calls required JSON format to work. This worked with the built-in "sample-program" but when testing with VQE, the paramaters were different. The VQE call requires an 'ansatz' paramater in the form of a Quantum Circuit and an 'operator' in the form of a PauliSumOp, however, these objects are not JSON serializable as seen in errors-found.ipnyb. If there is a solution, we didn't find much documentation related to direct API calls for the vqe program.

Instructions to Test Solution

To test this solution, simply download or clone the repository, upload the Jupyter notebooks to IBM Quantum Lab and run each notebook cell by cell to see the processing live. You can also use Conda Navigator or another Jupyter Notebook interface to run the notebooks as long as you have the required libraries installed (ie; qiskit, numpy, etc).

Next Steps + Final Comments

As young students, new to the industry, we believe we have found some next steps for the Qiskit Runtime API cloud service for improvement and support for applications. We did not receive help from any mentors or guidance. We completed this project by looking at documentation and doing research by ourselves. Going forward, we would like to build a program which we can upload to the API for others to use as a TSP solution. That way, users only need to input cities/locations required + other constraints, without having to worry about building hamiltonians or algorithms. Once public access has been released for uploading custom programs, we plan to work on this program, bringing quantum computing one step closer to the public, accessible, applicable stage. The learning curve we went through tihs to build such a program and understand it in such a short time has been very developmental in our journeys.