Refactor formula for getting costs of QROAM (#943)

qualtran/bloqs/chemistry/black_boxes.py

@@ -32,7 +32,39 @@
     from qualtran.resource_counting import BloqCountT, SympySymbolAllocator
 
 
-def get_qroam_cost(
+def qroam_cost(x, data_size: int, bitsize: int, adjoint: bool = False):
+    # See appendix B of https://arxiv.org/pdf/1902.02134
+    if adjoint:
+        return data_size / x + x
+    else:
+        return data_size / x + bitsize * (x - 1)
+
+
+def qroam_cost_dirty(x, data_size: int, bitsize: int, adjoint: bool = False):
+    # See appendix A of https://arxiv.org/pdf/1902.02134
+    if adjoint:
+        return data_size / x + x
+    else:
+        return 2 * (data_size / x - 1) + 4 * bitsize * (x - 1)
+
+
+def get_optimal_log_block_size_clean_ancilla(
+    data_size: int, bitsize: int, adjoint: bool = False, qroam_block_size: Optional[int] = None
+) -> int:
+    if qroam_block_size is None:
+        if adjoint:
+            log_blk = 0.5 * np.log2(data_size)
+            qroam_block_size = 2**log_blk
+        else:
+            log_blk = 0.5 * np.log2(data_size / bitsize)
+            assert log_blk >= 0
+            qroam_block_size = 2**log_blk
+    k = np.log2(qroam_block_size)
+    k_int = np.array([np.floor(k), np.ceil(k)])
+    return int(k_int[np.argmin(qroam_cost(2**k_int, data_size, bitsize, adjoint))])
+
+
+def get_qroam_cost_clean_ancilla(
     data_size: int, bitsize: int, adjoint: bool = False, qroam_block_size: Optional[int] = None
 ) -> int:
     """This gives the optimal k and minimum cost for a QROM over L values of size M.
@@ -51,22 +83,8 @@ def get_qroam_cost(
     """
     if qroam_block_size == 1:
         return data_size - 1
-    if adjoint:
-        if qroam_block_size is None:
-            log_blk = 0.5 * np.log2(data_size)
-            qroam_block_size = 2**log_blk
-        value = lambda x: data_size / x + x
-    else:
-        if qroam_block_size is None:
-            log_blk = 0.5 * np.log2(data_size / bitsize)
-            assert log_blk >= 0
-            qroam_block_size = 2**log_blk
-        value = lambda x: data_size / x + bitsize * (x - 1)
-    k = np.log2(qroam_block_size)
-    k_int = np.array([np.floor(k), np.ceil(k)])
-    k_opt = k_int[np.argmin(value(2**k_int))]
-    val_opt = np.ceil(value(2**k_opt))
-    return int(val_opt)
+    k_opt = get_optimal_log_block_size_clean_ancilla(data_size, bitsize, adjoint, qroam_block_size)
+    return int(np.ceil(qroam_cost(2**k_opt, data_size, bitsize, adjoint)))
 
 
 @frozen
@@ -96,7 +114,7 @@ def signature(self) -> Signature:
         return Signature.build(sel=(self.data_size - 1).bit_length(), trg=self.target_bitsize)
 
     def build_call_graph(self, ssa: 'SympySymbolAllocator') -> Set['BloqCountT']:
-        cost = get_qroam_cost(
+        cost = get_qroam_cost_clean_ancilla(
             self.data_size,
             self.target_bitsize,
             adjoint=self.is_adjoint,

qualtran/bloqs/chemistry/black_boxes_test.py

@@ -14,10 +14,10 @@
 import pytest
 from openfermion.resource_estimates.utils import QI, QR
 
-from qualtran.bloqs.chemistry.black_boxes import get_qroam_cost
+from qualtran.bloqs.chemistry.black_boxes import get_qroam_cost_clean_ancilla
 
 
 @pytest.mark.parametrize("data_size, bitsize", ((100, 10), (100, 3), (1_000, 13), (1_000_000, 20)))
 def test_qroam_factors(data_size, bitsize):
-    assert get_qroam_cost(data_size, bitsize) == QR(data_size, bitsize)[-1]
-    assert get_qroam_cost(data_size, bitsize, adjoint=True) == QI(data_size)[-1]
+    assert get_qroam_cost_clean_ancilla(data_size, bitsize) == QR(data_size, bitsize)[-1]
+    assert get_qroam_cost_clean_ancilla(data_size, bitsize, adjoint=True) == QI(data_size)[-1]

qualtran/bloqs/chemistry/chem_tutorials.py

@@ -103,6 +103,13 @@ def plot_linear_log_log(
     y_vals = np.exp(intr) * x_vals**slope
     if label is None:
         label = ''
-    ax.loglog(xs, ys, marker='o', ls='None', label=rf'{label} $N^{{{slope:3.2f}}}$', color=color)
+    ax.loglog(
+        xs,
+        ys,
+        marker='o',
+        ls='None',
+        label=rf'{label} ${{{np.exp(intr):3.1f}}}N^{{{slope:3.2f}}}$',
+        color=color,
+    )
     ax.loglog(x_vals, y_vals, marker='None', linestyle='--', color=color)
     ax.legend()