Fixes bar_with_pessimistic_uncertainty convergence error (#1344)

* A bug introduced in #1324

tests/test_bar.py

@@ -180,6 +180,11 @@ def test_df_from_u_kln_does_not_raise_on_incomplete_convergence():
     assert np.isfinite(df)
     assert np.isnan(ddf)  # returns NaN for uncertainty on incomplete convergence
 
+    bootstrap_df, bootstrap_ddf = bar_with_pessimistic_uncertainty(u_kln, maximum_iterations=1)
+    np.testing.assert_equal(df, bootstrap_df)
+    # With the bootstrapping, we will compute a finite error
+    assert np.isfinite(bootstrap_ddf)
+
 
 def test_pair_overlap_from_ukln():
     # identical distributions

timemachine/fe/bar.py

@@ -179,12 +179,8 @@ def bootstrap_bar(
     * TODO[deboggle] -- upgrade from pymbar3 to pymbar4 and remove this
     * TODO[performance] -- multiprocessing, if needed?
     """
-    u_kn, N_k = ukln_to_ukn(u_kln)
-    mbar = pymbar.MBAR(u_kn, N_k, maximum_iterations=maximum_iterations)
 
-    df, ddf = mbar.getFreeEnergyDifferences()
-    full_bar_result = df[0, 1]
-    full_bar_err = ddf[0, 1]
+    full_bar_result, full_bar_err = df_and_err_from_u_kln(u_kln, maximum_iterations=maximum_iterations)
 
     _, _, n = u_kln.shape
 
@@ -197,7 +193,7 @@ def bootstrap_bar(
         u_kln_sample = rng.choice(u_kln, size=(n,), replace=True, axis=2)
         bar_result = df_from_u_kln(
             u_kln_sample,
-            initial_f_k=mbar.f_k,  # warm start
+            initial_f_k=np.array([0.0, full_bar_result]),  # warm start
             maximum_iterations=maximum_iterations,
         )
         bootstrap_samples.append(bar_result)