Minor-Embedding#
To solve an arbitrarily posed binary quadratic problem directly on a D-Wave
system requires mapping, called minor embedding, to the QPU Topology
of the system’s quantum processing unit (QPU). This preprocessing can be done
by a composed sampler consisting of the
DWaveSampler()
and a composite that performs
minor-embedding. (This step is handled automatically by
LeapHybridSampler()
and
dwave-hybrid reference samplers.)
For example, a simple two-variable bqm,
might be embedded to two connected qubits, such as 633
and 3603
on an Advantage
system:
In the Advantage Pegasus topology, most qubits are connected to fifteen
other qubits, so other valid minor-embeddings might be 633
for \(s_0\)
and 634
for \(s_1\) or 4628
for \(s_0\) and 1749
for
\(s_1\).
Chains#
Larger problems often require chains because the QPU topology is not fully connected. For example, a \(K_5\) graph is not native to the Pegasus topology, meaning that the following five-variable bqm with a clique (fully-connected) \(K_5\) graph cannot be represented by five qubits on an Advantage QPU.
>>> import dimod
>>> bqm = dimod.generators.doped(1, 5)
>>> bqm.add_linear_from({v: 1 for v in bqm.variables})
Instead, at least one variable is represented by a chain of physical qubits, as in this minor-embedding:
The minor-embedding above was derived from the hueristic used by
EmbeddingComposite()
on the working graph of
an Advantage selected by DWaveSampler()
:
>>> from dwave.system import DWaveSampler, EmbeddingComposite, FixedEmbeddingComposite
>>> sampler = EmbeddingComposite(DWaveSampler())
Other qubits might have been chosen; for example,
>>> sampler = FixedEmbeddingComposite(DWaveSampler(),
... embedding={0: [4408, 2437], 1: [4333], 2: [4348], 3: [2497], 4: [2512]})
intentionally sets the embedding shown below to represent this same \(K_5\) graph:
Chain Strength#
For a chain of qubits to represent a variable, all its constituent qubits must return the same value for a sample. This is accomplished by setting a strong coupling to the edges connecting these qubits. That is, for the qubits in a chain to be likely to return identical values, the coupling strength for their connecting edges must be strong compared to the coupling with other qubits that influence non-identical outcomes.
The \(K_5\) BQM has ten ground states (best solutions). These are shown below—solved by brute-force stepping through all possible configurations of values for the variables—with ground-state energy -3.0:
>>> print(dimod.ExactSolver().sample(bqm).lowest())
0 1 2 3 4 energy num_oc.
0 +1 +1 -1 -1 -1 -3.0 1
1 -1 +1 +1 -1 -1 -3.0 1
2 +1 -1 +1 -1 -1 -3.0 1
3 -1 -1 +1 +1 -1 -3.0 1
4 -1 +1 -1 +1 -1 -3.0 1
5 +1 -1 -1 +1 -1 -3.0 1
6 -1 -1 -1 +1 +1 -3.0 1
7 -1 -1 +1 -1 +1 -3.0 1
8 -1 +1 -1 -1 +1 -3.0 1
9 +1 -1 -1 -1 +1 -3.0 1
['SPIN', 10 rows, 10 samples, 5 variables]
These solutions are states in which two variables are assigned one value and the remaining three variables the complementary value; for example two are \(-1\) and three \(+1\) as shown below.
A typical submission of the problem to a quantum computer, using the same minor-embedding as the previous subsection and the default chain strength, returned all the ground states, with these solutions constituting over 90% of the returned samples.
>>> sampleset = sampler.sample(bqm, num_reads=1000)
>>> print(sampleset.lowest())
0 1 2 3 4 energy num_oc. chain_.
0 -1 +1 +1 -1 -1 -3.0 69 0.0
1 +1 -1 -1 +1 -1 -3.0 115 0.0
2 +1 -1 -1 -1 +1 -3.0 95 0.0
3 +1 -1 +1 -1 -1 -3.0 84 0.0
4 -1 -1 -1 +1 +1 -3.0 116 0.0
5 -1 -1 +1 +1 -1 -3.0 99 0.0
6 -1 -1 +1 -1 +1 -3.0 91 0.0
7 +1 +1 -1 -1 -1 -3.0 71 0.0
8 -1 +1 -1 +1 -1 -3.0 98 0.0
9 -1 +1 -1 -1 +1 -3.0 95 0.0
10 +1 -1 -1 +1 -1 -3.0 1 0.2
['SPIN', 11 rows, 934 samples, 5 variables]
The default chain strength is set by the
uniform_torque_compensation()
function:
>>> print(round(sampleset.info['embedding_context']['chain_strength'], 3))
2.828
Resubmitting with a much lower chain strength produced less satisfactory results (only ~10% of returned samples are ground states).
>>> sampleset = sampler.sample(bqm, num_reads=1000, chain_strength=1)
>>> print(sampleset.lowest())
0 1 2 3 4 energy num_oc. chain_.
0 -1 +1 +1 -1 -1 -3.0 12 0.0
1 +1 -1 -1 +1 -1 -3.0 13 0.0
2 +1 -1 -1 -1 +1 -3.0 13 0.0
3 +1 -1 +1 -1 -1 -3.0 16 0.0
4 -1 -1 -1 +1 +1 -3.0 6 0.0
5 -1 -1 +1 +1 -1 -3.0 10 0.0
6 -1 -1 +1 -1 +1 -3.0 17 0.0
7 +1 +1 -1 -1 -1 -3.0 17 0.0
8 -1 +1 -1 +1 -1 -3.0 10 0.0
9 -1 +1 -1 -1 +1 -3.0 7 0.0
['SPIN', 10 rows, 121 samples, 5 variables]
Many of the remaining ~90% of returned samples have “broken chains”, meaning qubits of a chain did not have identical values due to insufficiently strong coupling compared to quadratic coefficients of interactions (of the variable the chain represented) with other variables.
For information on handling embedding and chains, see the following documentation:
Boolean AND Gate, Multiple-Gate Circuit, and Using the Problem Inspector examples
Show through some simple examples how to embed and set chain strength.
minorminer tool
Is the hueristic used by common Ocean embedding composites.
problem inspector tool
Visualizes embeddings.
dwave-system Composites section
Provides embedding composites
dwave-system Embedding section
Describes chain-related functionality.