vdot implementation

CHANGELOG.md

@@ -8,6 +8,8 @@
 - [#846](https://github.com/helmholtz-analytics/heat/pull/846) Fixed an issue in `min`, `max` where DNDarrays with empty processes can't be computed.
 
 ## Feature Additions
+### Linear Algebra
+- [#842](https://github.com/helmholtz-analytics/heat/pull/842) New feature: `vdot`
 
 ### DNDarray
 - [#856](https://github.com/helmholtz-analytics/heat/pull/856) New `DNDarray` method `__torch_proxy__`

heat/core/linalg/basics.py

@@ -34,6 +34,7 @@
     "transpose",
     "tril",
     "triu",
+    "vdot",
     "vecdot",
     "vector_norm",
 ]
@@ -1917,6 +1918,42 @@ def triu(m: DNDarray, k: int = 0) -> DNDarray:
 DNDarray.triu.__doc__ = triu.__doc__
 
 
+def vdot(x1: DNDarray, x2: DNDarray) -> DNDarray:
+    """
+    Computes the dot product of two vectors. Higher-dimensional arrays will be flattened.
+
+    Parameters
+    ----------
+    x1 : DNDarray
+        first input array. If it's complex, it's complex conjugate will be used.
+    x2 : DNDarray
+        second input array.
+
+    Raises
+    ------
+    ValueError
+        If the number of elements is inconsistent.
+
+    See Also
+    --------
+    dot
+        Return the dot product without using the complex conjugate.
+
+    Examples
+    --------
+    >>> a = ht.array([1+1j, 2+2j])
+    >>> b = ht.array([1+2j, 3+4j])
+    >>> ht.vdot(a,b)
+    DNDarray([(17+3j)], dtype=ht.complex64, device=cpu:0, split=None)
+    >>> ht.vdot(b,a)
+    DNDarray([(17-3j)], dtype=ht.complex64, device=cpu:0, split=None)
+    """
+    x1 = manipulations.flatten(x1)
+    x2 = manipulations.flatten(x2)
+
+    return arithmetics.sum(arithmetics.multiply(complex_math.conjugate(x1), x2))
+
+
 def vecdot(
     x1: DNDarray, x2: DNDarray, axis: Optional[int] = None, keepdim: Optional[bool] = None
 ) -> DNDarray:

heat/core/linalg/tests/test_basics.py

@@ -1767,6 +1767,21 @@ def test_triu(self):
         if result.comm.rank == result.shape[0] - 1:
             self.assertTrue(result.larray[0, -1] == 1)
 
+    def test_vdot(self):
+        a = ht.array([[1 + 1j, 2 + 2j], [3 + 3j, 4 + 4j]], split=0)
+        b = ht.array([[1 + 2j, 3 + 4j], [5 + 6j, 7 + 8j]], split=0)
+
+        vdot = ht.vdot(a, b)
+        self.assertEqual(vdot.dtype, a.dtype)
+        self.assertEqual(vdot.split, None)
+        self.assertTrue(ht.equal(vdot, ht.array([110 + 10j])))
+
+        vdot = ht.vdot(b, a)
+        self.assertTrue(ht.equal(vdot, ht.array([110 - 10j])))
+
+        with self.assertRaises(ValueError):
+            ht.vdot(ht.array([1, 2, 3]), ht.array([[1, 2], [3, 4]]))
+
     def test_vecdot(self):
         a = ht.array([1, 1, 1])
         b = ht.array([1, 2, 3])

heat/core/manipulations.py

@@ -803,6 +803,8 @@ def flatten(a: DNDarray) -> DNDarray:
     >>> ht.flatten(a)
     DNDarray([1, 2, 3, 4, 5, 6, 7, 8], dtype=ht.int64, device=cpu:0, split=None)
     """
+    sanitation.sanitize_in(a)
+
     if a.split is None:
         return factories.array(
             torch.flatten(a.larray), dtype=a.dtype, is_split=None, device=a.device, comm=a.comm