Handle undetectable errors

README.md

@@ -147,7 +147,7 @@ print(num_errors)  # prints 8
 
 To decode instead with correlated matching, set `enable_correlations=True` both when configuiing the `pymatching.Matching` object:
 ```python
-matching_corr = pymatching.Matching.from_detector_error_model(dem, enable_correlations=True)
+matching_corr = pymatching.Matching.from_detector_error_model(model, enable_correlations=True)
 ```
 
 as well as when decoding:

src/pymatching/matching.py

@@ -1296,14 +1296,23 @@ def from_detector_error_model(
         return m
 
     @staticmethod
-    def from_detector_error_model_file(dem_path: Union[str, Path]) -> 'pymatching.Matching':
+    def from_detector_error_model_file(
+        dem_path: Union[str, Path],
+        *,
+        enable_correlations: bool = False
+    ) -> 'pymatching.Matching':
         """
         Construct a `pymatching.Matching` by loading from a stim DetectorErrorModel file path.
 
         Parameters
         ----------
         dem_path : str
             The path of the detector error model file
+        enable_correlations : bool, optional
+            If `enable_correlations==True`, the detector error model is converted into an internal
+            representation that allows correlated matching to be used. Note that you must set
+            `enable_correlations=True` here in order to use `enable_correlations=True` when decoding.
+            By default, False.
 
         Returns
         -------
@@ -1314,7 +1323,10 @@ def from_detector_error_model_file(dem_path: Union[str, Path]) -> 'pymatching.Ma
         if isinstance(dem_path, Path):
             dem_path = str(dem_path)
         m = Matching()
-        m._matching_graph = _cpp_pm.detector_error_model_file_to_matching_graph(dem_path)
+        m._matching_graph = _cpp_pm.detector_error_model_file_to_matching_graph(
+            dem_path,
+            enable_correlations=enable_correlations
+        )
         return m
 
     @staticmethod
@@ -1371,7 +1383,11 @@ def from_stim_circuit(circuit: 'stim.Circuit', *, enable_correlations=False) ->
         return m
 
     @staticmethod
-    def from_stim_circuit_file(stim_circuit_path: Union[str, Path]) -> 'pymatching.Matching':
+    def from_stim_circuit_file(
+        stim_circuit_path: Union[str, Path],
+        *,
+        enable_correlations: bool = False
+    ) -> 'pymatching.Matching':
         """
         Construct a `pymatching.Matching` by loading from a stim circuit file path.
 
@@ -1386,11 +1402,19 @@ def from_stim_circuit_file(stim_circuit_path: Union[str, Path]) -> 'pymatching.M
             A `pymatching.Matching` object representing the graphlike error mechanisms in the stim circuit
             in the file `stim_circuit_path`, with any hyperedge error mechanisms decomposed into graphlike error
             mechanisms. Parallel edges are merged using `merge_strategy="independent"`.
+        enable_correlations : bool, optional
+            If `enable_correlations==True`, the stim circuit's detector error model is converted into an internal
+            representation that allows correlated matching to be used. Note that you must set
+            `enable_correlations=True` here in order to use `enable_correlations=True` when decoding.
+            By default, False.
         """
         if isinstance(stim_circuit_path, Path):
             stim_circuit_path = str(stim_circuit_path)
         m = Matching()
-        m._matching_graph = _cpp_pm.stim_circuit_file_to_matching_graph(stim_circuit_path)
+        m._matching_graph = _cpp_pm.stim_circuit_file_to_matching_graph(
+            stim_circuit_path,
+            enable_correlations=enable_correlations
+        )
         return m
 
     def _load_from_detector_error_model(self, model: 'stim.DetectorErrorModel', *, enable_correlations: bool = False) -> None:

src/pymatching/sparse_blossom/driver/user_graph.h

@@ -249,7 +249,7 @@ struct DecomposedDemError {
     /// The probability of this error occurring.
     double probability;
     /// Effects of the error.
-    stim::FixedCapVector<UserEdge, 8> components;
+    std::vector<UserEdge> components;
 
     bool operator==(const DecomposedDemError& other) const;
     bool operator!=(const DecomposedDemError& other) const;
@@ -283,6 +283,7 @@ void iter_dem_instructions_include_correlations(
         component->node1 = SIZE_MAX;
         component->node2 = SIZE_MAX;
         size_t num_component_detectors = 0;
+        bool instruction_contains_separator = false;
         for (auto& target : instruction.target_data) {
             // Decompose error
             if (target.is_relative_detector_id()) {
@@ -309,30 +310,46 @@ void iter_dem_instructions_include_correlations(
             } else if (target.is_observable_id()) {
                 component->observable_indices.push_back(target.val());
             } else if (target.is_separator()) {
-                // If the previous error in the decomposition had 3 or more components, we ignore it.
-                if (component->node1 == SIZE_MAX) {
+                instruction_contains_separator = true;
+                // If the previous error in the decomposition had 3 or more detectors, we throw an exception.
+                if (num_component_detectors > 2) {
                     throw std::invalid_argument(
                         "Encountered a decomposed error instruction with a hyperedge component (3 or more detectors). "
                         "This is not supported.");
-                } else if (p > 0) {
+                } else if (num_component_detectors == 0) {
+                    throw std::invalid_argument(
+                        "Encountered a decomposed error instruction with an undetectable component (0 detectors). "
+                        "This is not supported.");
+                } else if (num_component_detectors > 0) {
+                    // If the previous error in the decomposition had 1 or 2 detectors, we handle it
                     handle_dem_error(p, {component->node1, component->node2}, component->observable_indices);
+                    decomposed_err.components.push_back({});
+                    component = &decomposed_err.components.back();
+                    component->node1 = SIZE_MAX;
+                    component->node2 = SIZE_MAX;
+                    num_component_detectors = 0;
                 }
-                decomposed_err.components.push_back({});
-                component = &decomposed_err.components.back();
-                component->node1 = SIZE_MAX;
-                component->node2 = SIZE_MAX;
-                num_component_detectors = 0;
             }
         }
-        // If the final error in the decomposition had 3 or more components, we ignore it.
-        if (component->node1 == SIZE_MAX) {
+
+        if (num_component_detectors > 2) {
             // Undecomposed hyperedges are not supported
             throw std::invalid_argument(
                 "Encountered an undecomposed error instruction with 3 or mode detectors. "
                 "This is not supported when using `enable_correlations=True`. "
                 "Did you forget to set `decompose_errors=True` when "
                 "converting the stim circuit to a detector error model?");
-        } else if (p > 0) {
+        } else if (num_component_detectors == 0) {
+            if (instruction_contains_separator) {
+                throw std::invalid_argument(
+                    "Encountered a decomposed error instruction with an undetectable component (0 detectors). "
+                    "This is not supported.");
+            } else {
+                // Ignore errors that are undetectable, provided they are not a component of a decomposed error
+                return;
+            }
+
+        } else if (num_component_detectors > 0) {
             if (component->node2 == SIZE_MAX) {
                 handle_dem_error(p, {component->node1}, component->observable_indices);
             } else {

src/pymatching/sparse_blossom/driver/user_graph.test.cc

@@ -308,7 +308,8 @@ TEST(IterDemInstructionsTest, ThreeDetectorErrorThrowsInvalidArgument) {
     stim::DetectorErrorModel dem("error(0.1) D0 D1 D2");
     TestHandler handler;
     std::map<std::pair<size_t, size_t>, std::map<std::pair<size_t, size_t>, double>> joint_probabilities;
-    ASSERT_THROW(pm::iter_dem_instructions_include_correlations(dem, handler, joint_probabilities), std::invalid_argument);
+    ASSERT_THROW(
+        pm::iter_dem_instructions_include_correlations(dem, handler, joint_probabilities), std::invalid_argument);
 }
 
 // Test a decomposed error instruction. The handler should be called for each component.
@@ -353,13 +354,30 @@ TEST(IterDemInstructionsTest, DecomposedErrorWithHyperedgeThrows) {
         pm::iter_dem_instructions_include_correlations(dem, handler, joint_probabilities), std::invalid_argument);
 }
 
+// Test that a decomposed error with an undetectable component throws an exception.
+TEST(IterDemInstructionsTest, DecomposedErrorWithUndetectableErrorThrows) {
+    stim::DetectorErrorModel dem("error(0.15) L0 ^ D2 D4 ^ D5 D6 L2");
+    TestHandler handler;
+    std::map<std::pair<size_t, size_t>, std::map<std::pair<size_t, size_t>, double>> joint_probabilities;
+
+    // Assert that the function throws std::invalid_argument when processing the DEM.
+    ASSERT_THROW(
+        pm::iter_dem_instructions_include_correlations(dem, handler, joint_probabilities), std::invalid_argument);
+
+    stim::DetectorErrorModel dem2("error(0.15) D2 D4 ^ D5 D6 L2 ^ L1");
+    // Assert that the function throws std::invalid_argument when processing the DEM.
+    ASSERT_THROW(
+        pm::iter_dem_instructions_include_correlations(dem2, handler, joint_probabilities), std::invalid_argument);
+}
+
 // Test a complex DEM with multiple instruction types and edge cases combined.
 TEST(IterDemInstructionsTest, CombinedComplexDem) {
     stim::DetectorErrorModel dem(R"DEM(
         error(0.1) D0            # Instruction 1: Simple
+        error(0.3) L0            # Instruction 2: Undetectable error, ignored
         error(0.2) D1 D2 L0      # Instruction 2: Two detectors, one observable
-        error(0.0) D7            # Instruction 4: Zero probability, ignored
-        error(0.4) D8 ^ D9 L1    # Instruction 5: Decomposed
+        error(0.0) D7            # Instruction 3: Zero probability, ignored
+        error(0.4) D8 ^ D9 L1    # Instruction 4: Decomposed
     )DEM");
     TestHandler handler;
     std::map<std::pair<size_t, size_t>, std::map<std::pair<size_t, size_t>, double>> joint_probabilities;
@@ -395,6 +413,13 @@ double bernoulli_xor(double p1, double p2) {
     return p1 * (1 - p2) + p2 * (1 - p1);
 }
 
+TEST(IterDemInstructionsTest, MoreThanEightComponents) {
+    stim::DetectorErrorModel dem("error(0.1) D0 ^ D1 ^ D2 ^ D3 ^ D4 ^ D5 ^ D6 ^ D7 ^ D8");
+    TestHandler handler;
+    std::map<std::pair<size_t, size_t>, std::map<std::pair<size_t, size_t>, double>> joint_probabilities;
+    pm::iter_dem_instructions_include_correlations(dem, handler, joint_probabilities);
+}
+
 // Tests that multiple error instructions on the same edge correctly combine their probabilities.
 TEST(IterDemInstructionsTest, MultipleErrorsOnSameEdgeCombine) {
     stim::DetectorErrorModel dem("error(0.1) D0 D1\n error(0.2) D0 D1");

tests/matching/decode_test.py

@@ -387,22 +387,32 @@ def test_load_from_circuit_with_correlations():
         predictions, weights = m.decode_batch(shots=shots, return_weights=True, enable_correlations=True)
 
 
-def test_use_correlations_with_uncorrelated_dem_load_raises_value_error():
+def test_use_correlations_with_uncorrelated_dem_load_raises_value_error(tmp_path):
     stim = pytest.importorskip("stim")
+    d = 3
+    p = 0.001
     circuit = stim.Circuit.generated(
         code_task="surface_code:rotated_memory_x",
-        distance=3,
-        rounds=3,
-        after_clifford_depolarization=0.001
+        distance=d,
+        rounds=d,
+        after_clifford_depolarization=p
     )
+    dem = circuit.detector_error_model(decompose_errors=True)
     shots = circuit.compile_detector_sampler().sample(shots=10)
     matching_1 = pymatching.Matching(circuit, enable_correlations=False)
     matching_2 = pymatching.Matching.from_stim_circuit(circuit=circuit, enable_correlations=False)
     matching_3 = pymatching.Matching.from_detector_error_model(
-        model=circuit.detector_error_model(decompose_errors=True),
+        model=dem,
         enable_correlations=False
     )
-    for m in (matching_1, matching_2, matching_3):
+    fn = f"surface_code_x_d{d}_r{d}_p{p}"
+    stim_file = tmp_path / f"{fn}.stim"
+    circuit.to_file(stim_file)
+    matching_4 = pymatching.Matching.from_stim_circuit_file(stim_file, enable_correlations=False)
+    dem_file = tmp_path / f"{fn}.dem"
+    dem.to_file(dem_file)
+    matching_5 = pymatching.Matching.from_detector_error_model_file(dem_file, enable_correlations=False)
+    for m in (matching_1, matching_2, matching_3, matching_4, matching_5):
         with pytest.raises(ValueError):
             m.decode_batch(shots=shots, return_weights=True, enable_correlations=True)
         with pytest.raises(ValueError):
@@ -413,16 +423,22 @@ def test_use_correlations_with_uncorrelated_dem_load_raises_value_error():
             m.decode(shots[0], enable_correlations=True)
 
 
-def test_use_correlations_without_decompose_errors_raises_value_error():
+def test_use_correlations_without_decompose_errors_raises_value_error(tmp_path):
     stim = pytest.importorskip("stim")
+    d = 3
+    p = 0.001
     circuit = stim.Circuit.generated(
         code_task="surface_code:rotated_memory_x",
-        distance=3,
-        rounds=3,
-        after_clifford_depolarization=0.001
+        distance=d,
+        rounds=d,
+        after_clifford_depolarization=p
     )
     dem = circuit.detector_error_model(decompose_errors=False)
+    dem_file = tmp_path / "surface_code.dem"
+    dem.to_file(dem_file)
     with pytest.raises(ValueError):
         pymatching.Matching.from_detector_error_model(dem, enable_correlations=True)
     with pytest.raises(ValueError):
         pymatching.Matching(dem, enable_correlations=True)
+    with pytest.raises(ValueError):
+        pymatching.Matching.from_detector_error_model_file(dem_file, enable_correlations=True)