Basic TTN simulation

pytket/extensions/cutensornet/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/backends/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/backends/cutensornet_backend.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/general.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tensor_network_convert.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/__init__.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/cut_rKaHyPar_sea20.ini

@@ -0,0 +1,64 @@
+# general
+mode=recursive
+objective=cut
+seed=-1
+cmaxnet=-1
+vcycles=0
+# main -> preprocessing -> min hash sparsifier
+p-use-sparsifier=true
+p-sparsifier-min-median-he-size=28
+p-sparsifier-max-hyperedge-size=1200
+p-sparsifier-max-cluster-size=10
+p-sparsifier-min-cluster-size=2
+p-sparsifier-num-hash-func=5
+p-sparsifier-combined-num-hash-func=100
+# main -> preprocessing -> community detection
+p-detect-communities=true
+p-detect-communities-in-ip=false
+p-reuse-communities=false
+p-max-louvain-pass-iterations=100
+p-min-eps-improvement=0.0001
+p-louvain-edge-weight=hybrid
+p-large-he-threshold=1000
+# main -> preprocessing -> large he removal
+p-smallest-maxnet-threshold=50000
+p-maxnet-removal-factor=0.01
+# main -> coarsening
+c-type=heavy_lazy
+c-s=3.25
+c-t=160
+# main -> coarsening -> rating
+c-rating-score=heavy_edge
+c-rating-use-communities=true
+c-rating-heavy_node_penalty=multiplicative
+c-rating-acceptance-criterion=best
+c-fixed-vertex-acceptance-criterion=free_vertex_only
+# main -> initial partitioning
+i-mode=direct
+i-technique=flat
+# initial partitioning -> initial partitioning
+i-algo=pool
+i-runs=20
+# initial partitioning -> bin packing
+i-bp-algorithm=worst_fit
+i-bp-heuristic-prepacking=false
+i-bp-early-restart=true
+i-bp-late-restart=true
+# initial partitioning -> local search
+i-r-type=twoway_fm
+i-r-runs=-1
+i-r-fm-stop=simple
+i-r-fm-stop-i=50
+# main -> local search
+r-type=twoway_fm_hyperflow_cutter
+r-runs=-1
+r-fm-stop=adaptive_opt
+r-fm-stop-alpha=1
+r-fm-stop-i=350
+# local_search -> flow scheduling and heuristics
+r-flow-execution-policy=exponential
+# local_search -> hyperflowcutter configuration
+r-hfc-size-constraint=mf-style
+r-hfc-scaling=16
+r-hfc-distance-based-piercing=true
+r-hfc-mbc=true

pytket/extensions/cutensornet/tnstate/general.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/mps.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/mps_gate.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/mps_mpo.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/simulation.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.
@@ -14,10 +14,12 @@
 from typing import Optional
 from enum import Enum
 
+from pathlib import Path
 from collections import defaultdict  # type: ignore
 import numpy as np  # type: ignore
 
-from networkx import Graph, community  # type: ignore
+import networkx as nx  # type: ignore
+import kahypar  # type: ignore
 
 from pytket.circuit import Circuit, Command, Qubit
 from pytket.transform import Transform
@@ -194,23 +196,21 @@ def _get_qubit_partition(
                 )
 
     # Create the connectivity graph in NetworkX
-    connectivity_graph = Graph()
+    connectivity_graph = nx.Graph()
     connectivity_graph.add_nodes_from(circuit.qubits)
     for (u, v), weight in edge_weights.items():
         connectivity_graph.add_edge(u, v, weight=weight)
 
     # Apply balanced bisections until each qubit group is small enough
-    partition = {0: set(circuit.qubits)}
+    partition = {0: circuit.qubits}
     stop_bisec = False  # Do at least one bisection (TTN reqs >1 leaf nodes)
 
     while not stop_bisec:
         old_partition = partition.copy()
         for key, group in old_partition.items():
             # Apply the balanced bisection on this group
-            (groupA, groupB) = community.kernighan_lin_bisection(
+            (groupA, groupB) = _apply_kahypar_bisection(
                 connectivity_graph.subgraph(group),
-                max_iter=2 * len(group),  # Iteractions scaling with number of nodes
-                weight="weight",
             )
             # Groups A and B are on the same subtree (key separated by +1)
             partition[2 * key] = groupA
@@ -223,6 +223,68 @@ def _get_qubit_partition(
     return qubit_partition
 
 
+def _apply_kahypar_bisection(
+    graph: nx.Graph,
+) -> tuple[list[Qubit], list[Qubit]]:
+    """Use KaHyPar to obtain a bisection of the graph.
+
+    Returns:
+        Two lists, each containing the vertices in either group of the bisection.
+    """
+    vertices = list(graph.nodes)
+    edges = list(graph.edges)
+    weight_dict = nx.get_edge_attributes(graph, "weight")
+    qubit_dict = {q: i for i, q in enumerate(vertices)}
+
+    num_vertices = len(vertices)
+    num_edges = len(edges)
+    k = 2  # Number of groups in the partition
+    epsilon = 0.03  # Imbalance tolerance
+
+    # Special case where the graph has no edges; KaHyPar cannot deal with it
+    if num_edges == 0:
+        # Just split the list of vertices in half
+        return (vertices[: num_vertices // 2], vertices[num_vertices // 2 :])
+
+    # KaHyPar expects the list of edges to be provided as a continuous set of vertices
+    # ``edge_stream`` where ``edge_indices`` indicates where each new edge begins
+    # (the latter is necessary because KaHyPar can accept hyperedges)
+    edge_stream = [qubit_dict[vertex] for edge in edges for vertex in edge]
+    edge_indices = [0] + [2 * (i + 1) for i in range(num_edges)]
+    edge_weights = [weight_dict[edge] for edge in edges]
+    vertex_weights = [1 for _ in range(num_vertices)]
+
+    hypergraph = kahypar.Hypergraph(
+        num_vertices,
+        num_edges,
+        edge_indices,
+        edge_stream,
+        k,
+        edge_weights,
+        vertex_weights,
+    )
+
+    # Set up the configuration for KaHyPar
+    context = kahypar.Context()
+    context.setK(k)
+    context.setEpsilon(epsilon)
+    context.suppressOutput(True)
+
+    # Load the default configuration file provided by the KaHyPar devs
+    ini_file = str(Path(__file__).parent / "cut_rKaHyPar_sea20.ini")
+    context.loadINIconfiguration(ini_file)
+
+    # Run the partitioner
+    kahypar.partition(hypergraph, context)
+    partition_dict = {i: hypergraph.blockID(i) for i in range(hypergraph.numNodes())}
+
+    # Obtain the two groups of qubits from ``partition_dict``
+    groupA = [vertices[i] for i, block in partition_dict.items() if block == 0]
+    groupB = [vertices[i] for i, block in partition_dict.items() if block == 1]
+
+    return (groupA, groupB)
+
+
 def _get_sorted_gates(
     circuit: Circuit,
     algorithm: SimulationAlgorithm,

pytket/extensions/cutensornet/tnstate/ttn.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

pytket/extensions/cutensornet/tnstate/ttn_gate.py

@@ -1,4 +1,4 @@
-# Copyright 2019-2023 Quantinuum
+# Copyright 2019-2024 Quantinuum
 #
 # Licensed under the Apache License, Version 2.0 (the "License");
 # you may not use this file except in compliance with the License.

setup.py

@@ -42,7 +42,7 @@
     license="Apache 2",
     packages=find_namespace_packages(include=["pytket.*"]),
     include_package_data=True,
-    install_requires=["pytket ~= 1.11"],
+    install_requires=["pytket ~= 1.11", "kahypar ~= 1.3.5"],
     classifiers=[
         "Environment :: Console",
         "Programming Language :: Python :: 3.9",

tests/test_tnstate.py

@@ -487,7 +487,7 @@ def test_circ_approx_explicit_ttn(circuit: Circuit) -> None:
         ttn_gate = simulate(libhandle, circuit, SimulationAlgorithm.TTNxGate, cfg)
         for g in circuit.get_commands():
             ttn_gate.apply_gate(g)
-        assert np.isclose(ttn_gate.get_fidelity(), 0.62, atol=1e-2)
+        assert np.isclose(ttn_gate.get_fidelity(), 0.64, atol=1e-2)
         assert ttn_gate.is_valid()
         assert np.isclose(ttn_gate.vdot(ttn_gate), 1.0, atol=cfg._atol)