examples/sequence: discrepancy of fidelity between MQNS and SeQUeNCe

[yoursunny]

There are two variants of scalability_randomtopo_run.py running supposedly the same scenario in MQNS and SeQUeNCe.
However, their outputs are significantly different, especially in the fidelity values.

# mqns commit 51853ae0eea02b0deee734a68879f3e954cb9f92
python ./examples/scalability_randomtopo_run.py --seed 1
python ./examples/sequence/scalability_randomtopo_run.py --seed 1

Output files: 10-16-20-1.zip

simulator	src-dst	throughput	fidelity
MQNS	n10-n2	572	0.5568304175444247
MQNS	n8-n15	550	0.5441014033030881
SeQUeNCe	n10-n2	449	0.9310950000000102
SeQUeNCe	n8-n15	464	0.9310950000000108

[yoursunny]

These are the outputs for --seed 2: 10-16-20-2.zip

simulator	route	throughput	fidelity
MQNS	n8-n4-n7	603	0.5494161632252202
MQNS	n11-n3-n1-n2	281	0.43090034067972044
SeQUeNCe	n8-n4-n7	449	0.9310950000000102
SeQUeNCe	n11-n3-n1-n2	151	0.8756948474999999

The fidelity values from SeQUeNCe seems to match the calculation from updated_fidelity function:

• initial fidelity: 0.99
• swap once: 0.99 * 0.99 * 0.95 = 0.9310949999999999
• swap twice: 0.9310949999999999 * 0.99 * 0.95 = 0.8756948474999998

class EntanglementSwappingA(EntanglementProtocol):
    def updated_fidelity(self, f1: float, f2: float) -> float:
        return f1 * f2 * self.degradation
# degradation defaults to 0.95

It seems that SeQUeNCe does fidelity *= 0.95 for memory dephasing, regardless of how long the qubit has stayed in the memory.

[yoursunny]

Prompted by recent runs using #129 code, I found significant discrepancy in throughput too.
100-64-80.zip is the result from 100 qubit pairs, 64 nodes, 80 edges.
Files in the archive include:

• 100-64-80-SEED.json: MQNS result
• 100-64-80-SEED.sequence.json: SeQUeNCe result
• 100-64-80-SEED.sequence-topo.json: topology in SeQUeNCe format

SEED=200 throughput:

request	MQNS	SeQUeNCe
n1-n52	0	5000
n10-n61	6360	5645
n17-n33	1736	918
n29-n4	0	2
n39-n44	3264	1534
n64-n26	3221	1979
SUM	14852	15079

SEED=201 throughput:

request	MQNS	SeQUeNCe
n17-n59	0	0
n33-n14	153	0
n46-n15	3198	1
n48-n3	1751	0
n49-n25	0	0
n60-n34	0	0
SUM	5103	1

It is especially concerning that for SEED=201, the SeQUeNCe simulation produced very few end-to-end entanglements.

[amar-ox]

These simulations were initially meant to show the execution time difference compared to Sequence on larger networks.
As there are many things that may not match between the two simulators, we should provide more specific and controlled simulation (with know requests, swapping policy, etc.)