added test for ESMDA vs. SIES with one iteration

src/iterative_ensemble_smoother/esmda.py

@@ -35,6 +35,7 @@
     inversion_subspace,
     normalize_alpha,
 )
+from iterative_ensemble_smoother.utils import sample_mvnormal
 
 
 class BaseESMDA(ABC):
@@ -115,18 +116,13 @@ def perturb_observations(
         # a zero cented normal means that y := L @ z, where z ~ norm(0, 1).
         # Therefore, scaling C_D by alpha is equivalent to scaling L with sqrt(alpha).
 
-        D: npt.NDArray[np.double]
-
         # Two cases, depending on whether C_D was given as 1D or 2D array
-        if self.C_D.ndim == 2:
-            D = self.observations.reshape(-1, 1) + np.sqrt(
-                alpha
-            ) * self.C_D_L @ self.rng.normal(size=size)
-        else:
-            D = self.observations.reshape(-1, 1) + np.sqrt(alpha) * self.rng.normal(
-                size=size
-            ) * self.C_D_L.reshape(-1, 1)
-        assert D.shape == size
+        D: npt.NDArray[np.double]
+        # TODO: what if we have cov = [1, 2, 3] and we mask?
+        # we must pick out correct indices. 'size' is not enough...
+        D = self.observations.reshape(-1, 1) + np.sqrt(alpha) * sample_mvnormal(
+            C_dd_cholesky=self.C_D_L, rng=self.rng, size=size[1]
+        )
 
         return D
 

tests/test_esmda.py

@@ -25,9 +25,53 @@
 
 from iterative_ensemble_smoother.esmda import ESMDA
 from iterative_ensemble_smoother.esmda_inversion import empirical_cross_covariance
+from iterative_ensemble_smoother.sies import SIES
 
 
 class TestESMDA:
+    @pytest.mark.parametrize("num_inputs", [10, 25, 50])
+    @pytest.mark.parametrize("num_outputs", [5, 25, 50])
+    @pytest.mark.parametrize("sies_inversion", ["direct", "subspace_exact"])
+    @pytest.mark.parametrize("seed", list(range(1)))
+    def test_that_ESMDA_and_SIES_produce_same_result_with_one_step(
+        self, seed, sies_inversion, num_outputs, num_inputs
+    ):
+        """With a single step (alpha=1), ESMDA = SIES = ES.
+
+        When num_inputs < num_ensemble - 1, then Section (2.4.3) in the SIES
+        paper triggers and the result is not identical.
+        """
+        rng = np.random.default_rng(seed)
+
+        num_ensemble = 10
+        alpha = 1
+
+        # Create problem instance
+        X = rng.normal(size=(num_inputs, num_ensemble))
+        Y = rng.normal(size=(num_outputs, num_ensemble))
+        covariance = np.exp(rng.normal(size=num_outputs))
+        observations = rng.normal(size=num_outputs, loc=1)
+
+        # Create ESMDA instance and perform one iteration
+        esmda = ESMDA(covariance, observations, alpha=alpha, seed=0, inversion="exact")
+        X_ESMDA = np.copy(X)
+        for _ in range(esmda.num_assimilations()):
+            X_ESMDA = esmda.assimilate(X_ESMDA, Y)
+
+        # Create SIES instance and perform one iteration
+        sies = SIES(
+            parameters=X,
+            covariance=covariance,
+            observations=observations,
+            inversion=sies_inversion,
+            truncation=1.0,
+            seed=0,
+        )
+
+        X_SIES = sies.sies_iteration(responses=Y, step_length=1)
+
+        assert np.allclose(X_ESMDA, X_SIES)
+
     @pytest.mark.parametrize("seed", list(range(10)))
     @pytest.mark.parametrize("inversion", ["exact", "subspace"])
     def test_that_diagonal_covariance_gives_same_answer_as_dense(self, seed, inversion):
@@ -450,6 +494,6 @@ def g(X):
         args=[
             __file__,
             "-v",
-            "-k test_row_by_row_assimilation",
+            "-k test_that_ESMDA_and_SIES_produce_same_result_with_one_step",
         ]
     )