diff --git a/corneto/backend/_base.py b/corneto/backend/_base.py
index 98d706e..e4ff64c 100644
--- a/corneto/backend/_base.py
+++ b/corneto/backend/_base.py
@@ -768,6 +768,10 @@ def AcyclicFlow(
     ) -> ProblemDef:
         if not varname:
             varname = VAR_FLOW
+        if isinstance(lb, list):
+            lb = np.array(lb)
+        if isinstance(ub, list):
+            ub = np.array(ub)
         if not isinstance(lb, np.ndarray):
             lb = np.array([lb] * g.num_edges)
         if not isinstance(ub, np.ndarray):
@@ -820,8 +824,8 @@ def AcyclicFlow(
             e_ix = np.array([i for i, (s, t) in e_pos if len(s) > 0 and len(t) > 0])
             edges = [g.get_edge(i) for i in e_ix]
             # Get the index of the source / target vertices of the edge
-            s_idx = np.array([vix[list(s)[0]] for (s, _) in edges])
-            t_idx = np.array([vix[list(t)[0]] for (_, t) in edges])
+            s_idx = np.array([vix[next(iter(s))] for (s, _) in edges])
+            t_idx = np.array([vix[next(iter(t))] for (_, t) in edges])
             # The layer position in a DAG of the target vertex of the edge
             # has to be greater than the source vertex, otherwise Ip (pos flow) has to be 0
             if len(e_ix) > 0:
@@ -837,8 +841,8 @@ def AcyclicFlow(
             e_ix = np.array([i for i, (s, t) in e_neg if len(s) > 0 and len(t) > 0])
             edges = [g.get_edge(i) for i in e_ix]
             # Get the index of the source / target vertices of the edge
-            s_idx = np.array([vix[list(s)[0]] for (s, _) in edges])
-            t_idx = np.array([vix[list(t)[0]] for (_, t) in edges])
+            s_idx = np.array([vix[next(iter(s))] for (s, _) in edges])
+            t_idx = np.array([vix[next(iter(t))] for (_, t) in edges])
             if len(e_ix) > 0:
                 P += L[s_idx] - L[t_idx] >= In[e_ix] + (1 - g.num_vertices) * (
                     1 - In[e_ix]
diff --git a/tests/test_backend.py b/tests/test_backend.py
index a522b51..fd46e3f 100644
--- a/tests/test_backend.py
+++ b/tests/test_backend.py
@@ -399,6 +399,42 @@ def test_acyclic_flow_undirected_edge(backend):
     assert np.allclose(sol, vsol1) or np.allclose(sol, vsol2)
 
 
+def test_feasible_loop(backend):
+    G = Graph()
+    G.add_edge((), 'A')
+    G.add_edge('A', ())
+    G.add_edge('A', 'B')
+    G.add_edge('B', 'A')
+    G.add_edge((), 'B')
+    G.add_edge('B', ())
+    P = backend.Flow(G)
+    P += P.expr.flow[2] >= 1
+    P += P.expr.flow[3] >= 1
+    if isinstance(backend, PicosBackend):
+        P.solve(solver="glpk")
+    else:
+        P.solve()
+    assert np.sum(P.expr.flow.value) >= 2
+
+
+def test_acyclic_unfeasible_loop(backend):
+    G = Graph()
+    G.add_edge((), 'A')
+    G.add_edge('A', ())
+    G.add_edge('A', 'B')
+    G.add_edge('B', 'A')
+    G.add_edge((), 'B')
+    G.add_edge('B', ())
+    P = backend.AcyclicFlow(G)
+    P += P.expr.with_flow[2] == 1
+    P += P.expr.with_flow[3] == 1
+    if isinstance(backend, PicosBackend):
+        P.solve(solver="glpk", primals=False)
+    else:
+        P.solve()
+    assert np.all(P.expr.flow.value == None)
+
+
 @pytest.mark.skip(reason="Only a small subset of solvers support this")
 def test_l2_norm(backend):
     x = np.array([1, 2])