tensorcircuit-nightly 1.0.2.dev20250108__py3-none-any.whl → 1.4.0.dev20251103__py3-none-any.whl

tests/test_circuit.py

@@ -1,1699 +0,0 @@
-# pylint: disable=invalid-name
-
-import os
-import sys
-from functools import partial
-
-import numpy as np
-import opt_einsum as oem
-import pytest
-from pytest_lazyfixture import lazy_fixture as lf
-
-# see https://stackoverflow.com/questions/56307329/how-can-i-parametrize-tests-to-run-with-different-fixtures-in-pytest
-
-thisfile = os.path.abspath(__file__)
-modulepath = os.path.dirname(os.path.dirname(thisfile))
-
-sys.path.insert(0, modulepath)
-import tensorcircuit as tc
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_wavefunction(backend):
-    qc = tc.Circuit(2)
-    qc.unitary(
-        0,
-        1,
-        unitary=tc.gates.Gate(np.arange(16).reshape(2, 2, 2, 2).astype(np.complex64)),
-    )
-    assert np.real(qc.wavefunction()[2]) == 8
-    qc = tc.Circuit(2)
-    qc.unitary(
-        1,
-        0,
-        unitary=tc.gates.Gate(np.arange(16).reshape(2, 2, 2, 2).astype(np.complex64)),
-    )
-    qc.wavefunction()
-    assert np.real(qc.wavefunction()[2]) == 4
-    qc = tc.Circuit(2)
-    qc.unitary(
-        0, unitary=tc.gates.Gate(np.arange(4).reshape(2, 2).astype(np.complex64))
-    )
-    qc.wavefunction()
-    assert np.real(qc.wavefunction()[2]) == 2
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_basics(backend):
-    c = tc.Circuit(2)
-    c.x(0)
-    np.testing.assert_allclose(tc.backend.numpy(c.amplitude("10")), np.array(1.0))
-    c.CNOT(0, 1)
-    np.testing.assert_allclose(tc.backend.numpy(c.amplitude("11")), np.array(1.0))
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_measure(backend):
-    c = tc.Circuit(3)
-    c.H(0)
-    c.h(1)
-    c.toffoli(0, 1, 2)
-    assert c.measure(2)[0] in [0, 1]
-
-
-def test_gates_in_circuit():
-    c = tc.Circuit(2, inputs=np.eye(2**2))
-    c.iswap(0, 1)
-    ans = tc.gates.iswap_gate().tensor.reshape([4, 4])
-    np.testing.assert_allclose(c.state().reshape([4, 4]), ans, atol=1e-5)
-
-
-def test_control_vgate():
-    c = tc.Circuit(2)
-    c.x(1)
-    c.crx(1, 0, theta=0.3)
-    np.testing.assert_allclose(
-        c.expectation([tc.gates._z_matrix, 0]), 0.95533645, atol=1e-5
-    )
-
-
-def test_adjoint_gate_circuit():
-    c = tc.Circuit(1)
-    c.X(0)
-    c.SD(0)
-    np.testing.assert_allclose(c.state(), np.array([0.0, -1.0j]))
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_jittable_measure(backend):
-    @partial(tc.backend.jit, static_argnums=(2, 3))
-    def f(param, key, n=6, nlayers=3):
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for j in range(nlayers):
-            for i in range(n - 1):
-                c.exp1(i, i + 1, theta=param[2 * j, i], hamiltonian=tc.gates._zz_matrix)
-            for i in range(n):
-                c.rx(i, theta=param[2 * j + 1, i])
-        return c.measure_jit(0, 1, 2, with_prob=True)
-
-    if tc.backend.name == "tensorflow":
-        import tensorflow as tf
-
-        print(f(tc.backend.ones([6, 6]), None))
-        print(f(tc.backend.ones([6, 6]), None))
-        print(f(tc.backend.ones([6, 6]), tf.random.Generator.from_seed(23)))
-        print(f(tc.backend.ones([6, 6]), tf.random.Generator.from_seed(24)))
-    elif tc.backend.name == "jax":
-        import jax
-
-        print(f(tc.backend.ones([6, 6]), jax.random.PRNGKey(23)))
-        print(f(tc.backend.ones([6, 6]), jax.random.PRNGKey(24)))
-
-    # As seen here, though I have tried the best, the random API is still not that consistent under jit
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_jittable_depolarizing(backend):
-    @tc.backend.jit
-    def f1(key):
-        n = 5
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for i in range(n):
-            c.cnot(i, (i + 1) % n)
-        for i in range(n):
-            c.unitary_kraus(
-                [
-                    tc.gates._x_matrix,
-                    tc.gates._y_matrix,
-                    tc.gates._z_matrix,
-                    tc.gates._i_matrix,
-                ],
-                i,
-                prob=[0.2, 0.2, 0.2, 0.4],
-            )
-        for i in range(n):
-            c.cz(i, (i + 1) % n)
-        return c.wavefunction()
-
-    @tc.backend.jit
-    def f2(key):
-        n = 5
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for i in range(n):
-            c.cnot(i, (i + 1) % n)
-        for i in range(n):
-            c.unitary_kraus(
-                tc.channels.depolarizingchannel(0.2, 0.2, 0.2),
-                i,
-            )
-        for i in range(n):
-            c.X(i)
-        return c.wavefunction()
-
-    @tc.backend.jit
-    def f3(key):
-        n = 5
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for i in range(n):
-            c.cnot(i, (i + 1) % n)
-        for i in range(n):
-            c.depolarizing(i, px=0.2, py=0.2, pz=0.2)
-        for i in range(n):
-            c.X(i)
-        return c.wavefunction()
-
-    @tc.backend.jit
-    def f4(key):
-        n = 5
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for i in range(n):
-            c.cnot(i, (i + 1) % n)
-        for i in range(n):
-            c.depolarizing2(i, px=0.2, py=0.2, pz=0.2)
-        for i in range(n):
-            c.X(i)
-        return c.wavefunction()
-
-    @tc.backend.jit
-    def f5(key):
-        n = 5
-        if key is not None:
-            tc.backend.set_random_state(key)
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for i in range(n):
-            c.cnot(i, (i + 1) % n)
-        for i in range(n):
-            c.unitary_kraus2(
-                tc.channels.depolarizingchannel(0.2, 0.2, 0.2),
-                i,
-            )
-        for i in range(n):
-            c.X(i)
-        return c.wavefunction()
-
-    for f in [f1, f2, f3, f4, f5]:
-        if tc.backend.name == "tensorflow":
-            import tensorflow as tf
-
-            np.testing.assert_allclose(tc.backend.norm(f(None)), 1.0, atol=1e-4)
-            np.testing.assert_allclose(
-                tc.backend.norm(f(tf.random.Generator.from_seed(23))), 1.0, atol=1e-4
-            )
-            np.testing.assert_allclose(
-                tc.backend.norm(f(tf.random.Generator.from_seed(24))), 1.0, atol=1e-4
-            )
-
-        elif tc.backend.name == "jax":
-            import jax
-
-            np.testing.assert_allclose(
-                tc.backend.norm(f(jax.random.PRNGKey(23))), 1.0, atol=1e-4
-            )
-            np.testing.assert_allclose(
-                tc.backend.norm(f(jax.random.PRNGKey(24))), 1.0, atol=1e-4
-            )
-
-
-@pytest.mark.parametrize("backend", [lf("jaxb")])  # too slow for np
-def test_large_scale_sample(backend):
-    L = 30
-    c = tc.Circuit(L)
-    c.h(0)
-    c.cnot([i for i in range(L - 1)], [i + 1 for i in range(L - 1)])
-    results = c.sample(
-        allow_state=False, batch=1024, format="count_dict_bin", jittable=False
-    )
-    assert (
-        results["0" * L] / results["1" * L] < 1.2
-        and results["0" * L] / results["1" * L] > 0.8
-    )
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_expectation(backend):
-    c = tc.Circuit(2)
-    c.H(0)
-    np.testing.assert_allclose(
-        tc.backend.numpy(c.expectation((tc.gates.z(), [0]))), 0, atol=1e-7
-    )
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb"), lf("cpb")])
-def test_exp1(backend):
-    @partial(tc.backend.jit, jit_compile=True)
-    def sf():
-        c = tc.Circuit(2)
-        xx = np.array(
-            [[0, 0, 0, 1], [0, 0, 1, 0], [0, 1, 0, 0], [1, 0, 0, 0]], dtype=np.complex64
-        )
-        c.exp1(0, 1, unitary=xx, theta=tc.num_to_tensor(0.2))
-        s = c.state()
-        return s
-
-    @tc.backend.jit
-    def s1f():
-        c = tc.Circuit(2)
-        xx = np.array(
-            [[0, 0, 0, 1], [0, 0, 1, 0], [0, 1, 0, 0], [1, 0, 0, 0]], dtype=np.complex64
-        )
-        c.exp(0, 1, unitary=xx, theta=tc.num_to_tensor(0.2))
-        s1 = c.state()
-        return s1
-
-    s = tc.backend.numpy(sf())
-    s1 = tc.backend.numpy(s1f())
-    np.testing.assert_allclose(s, s1, atol=1e-4)
-
-
-def test_complex128(highp, tfb):
-    c = tc.Circuit(2)
-    c.H(1)
-    c.rx(0, theta=tc.gates.num_to_tensor(1j))
-    c.wavefunction()
-    np.testing.assert_allclose(c.expectation((tc.gates.z(), [1])), 0)
-
-
-# def test_qcode():
-#     qcode = """
-# 4
-# x 0
-# cnot 0 1
-# r 2 theta 1.0 alpha 1.57
-# """
-#     c = tc.Circuit.from_qcode(qcode)
-#     assert c.measure(1)[0] == "1"
-#     assert c.to_qcode() == qcode[1:]
-
-
-def universal_ad():
-    @tc.backend.jit
-    def forward(theta):
-        c = tc.Circuit(2)
-        c.R(0, theta=theta, alpha=0.5, phi=0.8)
-        return tc.backend.real(c.expectation((tc.gates.z(), [0])))
-
-    gg = tc.backend.grad(forward)
-    vg = tc.backend.value_and_grad(forward)
-    gg = tc.backend.jit(gg)
-    vg = tc.backend.jit(vg)
-    theta = tc.gates.num_to_tensor(1.0)
-    grad1 = gg(theta)
-    v2, grad2 = vg(theta)
-    assert grad1 == grad2
-    return v2, grad2
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_ad(backend):
-    # this amazingly shows how to code once and run in very different AD-ML engines
-    print(universal_ad())
-
-
-def test_single_qubit():
-    c = tc.Circuit(1)
-    c.H(0)
-    w = c.state()[0]
-    np.testing.assert_allclose(w, np.array([1, 1]) / np.sqrt(2), atol=1e-4)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_expectation_between_two_states(backend):
-    zp = np.array([1.0, 0.0])
-    zd = np.array([0.0, 1.0])
-    assert tc.expectation((tc.gates.y(), [0]), ket=zp, bra=zd) == 1j
-
-    c = tc.Circuit(3)
-    c.H(0)
-    c.ry(1, theta=tc.num_to_tensor(0.8))
-    c.cnot(1, 2)
-
-    state = c.wavefunction()
-    x1z2 = [(tc.gates.x(), [0]), (tc.gates.z(), [1])]
-    e1 = c.expectation(*x1z2)
-    e2 = tc.expectation(*x1z2, ket=state, bra=state, normalization=True)
-    np.testing.assert_allclose(tc.backend.numpy(e2), tc.backend.numpy(e1))
-
-    c = tc.Circuit(3)
-    c.H(0)
-    c.ry(1, theta=tc.num_to_tensor(0.8 + 0.7j))
-    c.cnot(1, 2)
-
-    state = c.wavefunction()
-    x1z2 = [(tc.gates.x(), [0]), (tc.gates.z(), [1])]
-    e1 = c.expectation(*x1z2) / tc.backend.norm(state) ** 2
-    e2 = tc.expectation(*x1z2, ket=state, normalization=True)
-    np.testing.assert_allclose(tc.backend.numpy(e2), tc.backend.numpy(e1))
-
-    c = tc.Circuit(2)
-    c.X(1)
-    s1 = c.state()
-    c2 = tc.Circuit(2)
-    c2.X(0)
-    s2 = c2.state()
-    c3 = tc.Circuit(2)
-    c3.H(1)
-    s3 = c3.state()
-    x1x2 = [(tc.gates.x(), [0]), (tc.gates.x(), [1])]
-    e = tc.expectation(*x1x2, ket=s1, bra=s2)
-    np.testing.assert_allclose(tc.backend.numpy(e), 1.0)
-    e2 = tc.expectation(*x1x2, ket=s3, bra=s2)
-    np.testing.assert_allclose(tc.backend.numpy(e2), 1.0 / np.sqrt(2))
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb"), lf("cpb")])
-def test_any_inputs_state(backend):
-    c = tc.Circuit(2, inputs=tc.array_to_tensor(np.array([0.0, 0.0, 0.0, 1.0])))
-    c.X(0)
-    z0 = c.expectation((tc.gates.z(), [0]))
-    assert z0 == 1.0
-    c = tc.Circuit(2, inputs=tc.array_to_tensor(np.array([0.0, 0.0, 1.0, 0.0])))
-    c.X(0)
-    z0 = c.expectation((tc.gates.z(), [0]))
-    assert z0 == 1.0
-    c = tc.Circuit(2, inputs=tc.array_to_tensor(np.array([1.0, 0.0, 0.0, 0.0])))
-    c.X(0)
-    z0 = c.expectation((tc.gates.z(), [0]))
-    assert z0 == -1.0
-    c = tc.Circuit(
-        2,
-        inputs=tc.array_to_tensor(np.array([1 / np.sqrt(2), 0.0, 1 / np.sqrt(2), 0.0])),
-    )
-    c.X(0)
-    z0 = c.expectation((tc.gates.z(), [0]))
-    np.testing.assert_allclose(tc.backend.numpy(z0), 0.0, rtol=1e-4, atol=1e-4)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("cpb")])
-def test_postselection(backend):
-    c = tc.Circuit(3)
-    c.H(1)
-    c.H(2)
-    c.mid_measurement(1, 1)
-    c.mid_measurement(2, 1)
-    s = c.wavefunction()
-    np.testing.assert_allclose(tc.backend.numpy(s[3]).real, 0.5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_unitary(backend):
-    c = tc.Circuit(2, inputs=np.eye(4))
-    c.X(0)
-    c.Y(1)
-    answer = tc.backend.numpy(np.kron(tc.gates.x().tensor, tc.gates.y().tensor))
-    np.testing.assert_allclose(
-        tc.backend.numpy(c.wavefunction().reshape([4, 4])), answer, atol=1e-4
-    )
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("cpb")])
-def test_expectation_ps(backend):
-    c = tc.Circuit(2)
-    c.X(0)
-    r = c.expectation_ps(z=[0, 1])
-    np.testing.assert_allclose(tc.backend.numpy(r), -1, atol=1e-5)
-
-    c = tc.Circuit(2)
-    c.H(0)
-    r = c.expectation_ps(z=[1], x=[0])
-    np.testing.assert_allclose(tc.backend.numpy(r), 1, atol=1e-5)
-    r1 = c.expectation_ps(ps=[1, 3])
-    np.testing.assert_allclose(tc.backend.numpy(r1), 1, atol=1e-5)
-    r1 = c.expectation_ps(z=[1, 2], ps=[1, 3])
-    np.testing.assert_allclose(tc.backend.numpy(r1), 1, atol=1e-5)
-
-
-def test_probability():
-    for c_cls in [tc.Circuit, tc.DMCircuit]:
-        c = c_cls(2)
-        c.h(0)
-        c.h(1)
-        np.testing.assert_allclose(
-            c.probability(), np.array([1, 1, 1, 1]) / 4, atol=1e-5
-        )
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_dqas_type_circuit(backend):
-    eye = tc.gates.i().tensor
-    x = tc.gates.x().tensor
-    y = tc.gates.y().tensor
-    z = tc.gates.z().tensor
-
-    def f(params, structures):
-        paramsc = tc.backend.cast(params, dtype="complex64")
-        structuresc = tc.backend.softmax(structures, axis=-1)
-        structuresc = tc.backend.cast(structuresc, dtype="complex64")
-        c = tc.Circuit(5)
-        for i in range(5):
-            c.H(i)
-        for j in range(2):
-            for i in range(4):
-                c.cz(i, i + 1)
-            for i in range(5):
-                c.any(
-                    i,
-                    unitary=structuresc[i, j, 0]
-                    * (
-                        tc.backend.cos(paramsc[i, j, 0]) * eye
-                        + tc.backend.sin(paramsc[i, j, 0]) * x
-                    )
-                    + structuresc[i, j, 1]
-                    * (
-                        tc.backend.cos(paramsc[i, j, 1]) * eye
-                        + tc.backend.sin(paramsc[i, j, 1]) * y
-                    )
-                    + structuresc[i, j, 2]
-                    * (
-                        tc.backend.cos(paramsc[i, j, 2]) * eye
-                        + tc.backend.sin(paramsc[i, j, 2]) * z
-                    ),
-                )
-        return tc.backend.real(c.expectation([tc.gates.z(), (2,)]))
-
-    structures = tc.array_to_tensor(
-        np.random.normal(size=[16, 5, 2, 3]), dtype="float32"
-    )
-    params = tc.array_to_tensor(np.random.normal(size=[5, 2, 3]), dtype="float32")
-
-    vf = tc.backend.vmap(f, vectorized_argnums=(1,))
-
-    np.testing.assert_allclose(vf(params, structures).shape, [16])
-
-    vvag = tc.backend.vvag(f, argnums=0, vectorized_argnums=1)
-
-    vvag = tc.backend.jit(vvag)
-
-    value, grad = vvag(params, structures)
-
-    np.testing.assert_allclose(value.shape, [16])
-    np.testing.assert_allclose(grad.shape, [5, 2, 3])
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_mixed_measurement_circuit(backend):
-    n = 4
-
-    def f(params, structures):
-        structuresc = tc.backend.cast(structures, dtype="complex64")
-        c = tc.Circuit(n)
-        for i in range(n):
-            c.H(i)
-        for j in range(2):
-            for i in range(n):
-                c.cnot(i, (i + 1) % n)
-            for i in range(n):
-                c.rz(i, theta=params[j, i])
-        obs = []
-        for i in range(n):
-            obs.append(
-                [
-                    tc.gates.Gate(
-                        sum(
-                            [
-                                structuresc[i, k] * g.tensor
-                                for k, g in enumerate(tc.gates.pauli_gates)
-                            ]
-                        )
-                    ),
-                    (i,),
-                ]
-            )
-        loss = c.expectation(*obs, reuse=False)
-        return tc.backend.real(loss)
-
-    # measure X0 to X3
-
-    structures = tc.backend.cast(tc.backend.eye(n), "int32")
-    structures = tc.backend.onehot(structures, num=4)
-
-    f_vvag = tc.backend.jit(tc.backend.vvag(f, vectorized_argnums=1, argnums=0))
-    v, g = f_vvag(tc.backend.ones([2, n], dtype="float32"), structures)
-    np.testing.assert_allclose(
-        v,
-        np.array(
-            [
-                0.157729,
-                0.157729,
-                0.157728,
-                0.085221,
-            ]
-        ),
-        atol=1e-5,
-    )
-    np.testing.assert_allclose(
-        g[0],
-        np.array([-0.378372, -0.624019, -0.491295, -0.378372]),
-        atol=1e-5,
-    )
-
-
-def test_circuit_add_demo():
-    # to be refactored for better API
-    c = tc.Circuit(2)
-    c.x(0)
-    c2 = tc.Circuit(2, mps_inputs=c.quvector())
-    c2.X(0)
-    answer = np.array([1.0, 0, 0, 0])
-    np.testing.assert_allclose(c2.wavefunction(), answer, atol=1e-4)
-    c3 = tc.Circuit(2)
-    c3.X(0)
-    c3.replace_mps_inputs(c.quvector())
-    np.testing.assert_allclose(c3.wavefunction(), answer, atol=1e-4)
-
-
-def test_circuit_replace_inputs():
-    n = 3
-    c = tc.Circuit(n, inputs=np.zeros([2**n]))
-    for i in range(n):
-        c.H(i)
-    evenstate = np.ones([2**n])
-    evenstate /= np.linalg.norm(evenstate)
-    c.replace_inputs(evenstate)
-    for i in range(n):
-        np.testing.assert_allclose(c.expectation_ps(z=[i]), 1.0, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_circuit_matrix(backend):
-    c = tc.Circuit(2)
-    c.x(1)
-    c.cnot(0, 1)
-    np.testing.assert_allclose(c.matrix()[3], np.array([0.0, 0.0, 0.0, 1.0]), atol=1e-5)
-    np.testing.assert_allclose(c.state(), np.array([0, 1.0, 0, 0]), atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_circuit_split(backend):
-    n = 4
-
-    def f(param, max_singular_values=None, max_truncation_err=None, fixed_choice=None):
-        if (max_singular_values is None) and (max_truncation_err is None):
-            split = None
-        else:
-            split = {
-                "max_singular_values": max_singular_values,
-                "max_truncation_err": max_truncation_err,
-                "fixed_choice": fixed_choice,
-            }
-        c = tc.Circuit(
-            n,
-            split=split,
-        )
-        for i in range(n):
-            c.H(i)
-        for j in range(2):
-            for i in range(n - 1):
-                c.exp1(i, i + 1, theta=param[2 * j, i], hermitian=tc.gates._zz_matrix)
-            for i in range(n):
-                c.rx(i, theta=param[2 * j + 1, i])
-        loss = c.expectation(
-            (
-                tc.gates.z(),
-                [1],
-            ),
-            (
-                tc.gates.z(),
-                [2],
-            ),
-        )
-        return tc.backend.real(loss)
-
-    s1 = f(tc.backend.ones([4, n]))
-    s2 = f(tc.backend.ones([4, n]), max_truncation_err=1e-5)
-    s3 = f(tc.backend.ones([4, n]), max_singular_values=2, fixed_choice=1)
-
-    np.testing.assert_allclose(s1, s2, atol=1e-5)
-    np.testing.assert_allclose(s1, s3, atol=1e-5)
-
-    f_jit = tc.backend.jit(f, static_argnums=(1, 2, 3))
-
-    s1 = f_jit(tc.backend.ones([4, n]))
-    # s2 = f_jit(tc.backend.ones([4, n]), max_truncation_err=1e-5) # doesn't work now
-    # this cannot be done anyway, since variable size tensor network will fail opt einsum
-    s3 = f_jit(tc.backend.ones([4, n]), max_singular_values=2, fixed_choice=1)
-
-    # np.testing.assert_allclose(s1, s2, atol=1e-5)
-    np.testing.assert_allclose(s1, s3, atol=1e-5)
-
-    f_vg = tc.backend.jit(
-        tc.backend.value_and_grad(f, argnums=0), static_argnums=(1, 2, 3)
-    )
-
-    s1, g1 = f_vg(tc.backend.ones([4, n]))
-    s3, g3 = f_vg(tc.backend.ones([4, n]), max_singular_values=2, fixed_choice=1)
-
-    np.testing.assert_allclose(s1, s3, atol=1e-5)
-    # DONE(@refraction-ray): nan on jax backend?
-    # i see, complex value SVD is not supported on jax for now :)
-    # I shall further customize complex SVD, finally it has applications
-
-    # tf 2.6.2 also doesn't support complex valued SVD AD, weird...
-    # if tc.backend.name == "tensorflow":
-    np.testing.assert_allclose(g1, g3, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_gate_split(backend):
-    n = 4
-
-    def f(param, max_singular_values=None, max_truncation_err=None, fixed_choice=None):
-        if (max_singular_values is None) and (max_truncation_err is None):
-            split = None
-        else:
-            split = {
-                "max_singular_values": max_singular_values,
-                "max_truncation_err": max_truncation_err,
-                "fixed_choice": fixed_choice,
-            }
-        c = tc.Circuit(
-            n,
-        )
-        for i in range(n):
-            c.H(i)
-        for j in range(2):
-            for i in range(n - 1):
-                c.exp1(
-                    i,
-                    i + 1,
-                    theta=param[2 * j, i],
-                    unitary=tc.gates._zz_matrix,
-                    split=split,
-                )
-            for i in range(n):
-                c.rx(i, theta=param[2 * j + 1, i])
-        loss = c.expectation(
-            (
-                tc.gates.x(),
-                [1],
-            ),
-        )
-        return tc.backend.real(loss)
-
-    s1 = f(tc.backend.ones([4, n]))
-    s2 = f(tc.backend.ones([4, n]), max_truncation_err=1e-5)
-    s3 = f(tc.backend.ones([4, n]), max_singular_values=2, fixed_choice=1)
-
-    np.testing.assert_allclose(s1, s2, atol=1e-5)
-    np.testing.assert_allclose(s1, s3, atol=1e-5)
-
-
-def test_toqir():
-    split = {
-        "max_singular_values": 2,
-        "fixed_choice": 1,
-    }
-    c = tc.Circuit(3)
-    c.H(0)
-    c.rx(1, theta=tc.array_to_tensor(0.7))
-    c.exp1(
-        0, 1, unitary=tc.gates._zz_matrix, theta=tc.array_to_tensor(-0.2), split=split
-    )
-    z1 = c.expectation((tc.gates.z(), [1]))
-    qirs = c.to_qir()
-    c = tc.Circuit.from_qir(qirs, circuit_params={"nqubits": 3})
-    assert len(c._nodes) == 7
-    z2 = c.expectation((tc.gates.z(), [1]))
-    np.testing.assert_allclose(z1, z2, atol=1e-5)
-    c.append_from_qir(qirs)
-    z3 = c.expectation((tc.gates.z(), [1]))
-    assert len(c._nodes) == 11
-    np.testing.assert_allclose(z3, 0.202728, atol=1e-5)
-
-
-def test_vis_tex():
-    c = tc.Circuit(3)
-    for i in range(3):
-        c.H(i)
-    for i in range(3):
-        c.any(i, (i + 1) % 3, unitary=tc.backend.ones([4, 4]), name="hihi")
-    c.any(2, unitary=tc.backend.ones([2, 2]), name="invisible")
-    c.cz(1, 2)
-    c.any(1, 0, 2, unitary=tc.backend.ones([8, 8]), name="ccha")
-    c.z(2)
-    c.cnot(0, 1)
-    c.cz(2, 1)
-
-    print(c.vis_tex(init=["0", "1", ""], measure=["x", "y", "z"]))
-
-
-def test_debug_contract():
-    n = 10
-    d = 4
-    try:
-        import cotengra  # pylint: disable=unused-import
-
-    except ImportError:
-        pytest.skip("cotengra is not installed")
-
-    @tc.set_function_contractor(
-        "custom_stateful",
-        optimizer=oem.RandomGreedy,
-        max_time=10,
-        max_repeats=64,
-        minimize="size",
-        debug_level=2,
-        contraction_info=True,
-    )
-    def small_tn():
-        param = tc.backend.ones([2 * d, n])
-        c = tc.Circuit(n)
-        c = tc.templates.blocks.example_block(c, param, nlayers=d)
-        return c.state()
-
-    np.testing.assert_allclose(small_tn(), np.zeros([2**n]), atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_teleportation(backend):
-    key = tc.backend.get_random_state(42)
-
-    @tc.backend.jit
-    def f(key):
-        tc.backend.set_random_state(key)
-        c = tc.Circuit(2)
-        c.H(0)
-        r = c.cond_measurement(0)
-        c.conditional_gate(r, [tc.gates.i(), tc.gates.x()], 1)
-        return r, c.expectation([tc.gates.z(), [1]])
-
-    keys = []
-    for _ in range(6):
-        key, subkey = tc.backend.random_split(key)
-        keys.append(subkey)
-    rs = [f(k) for k in keys]
-    for r, e in rs:
-        if tc.backend.numpy(r) > 0.5:
-            np.testing.assert_allclose(e, -1, atol=1e-5)
-        else:
-            np.testing.assert_allclose(e, 1, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_append_circuit(backend):
-    c = tc.Circuit(2)
-    c.cnot(0, 1)
-    c1 = tc.Circuit(2)
-    c1.x(0)
-    c.append(c1)
-    np.testing.assert_allclose(c.expectation_ps(z=[1]), 1.0)
-
-    c = tc.Circuit(2)
-    c.cnot(0, 1)
-    c1 = tc.Circuit(2)
-    c1.x(0)
-    c.prepend(c1)
-    np.testing.assert_allclose(c.expectation_ps(z=[1]), -1.0)
-
-    c = tc.Circuit(2)
-    c1 = tc.Circuit(1)
-    c1.x(0)
-    c.append(c1, [1])
-    np.testing.assert_allclose(c.state(), [0, 1, 0, 0])
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_apply_mpo_gate(backend):
-    gate = tc.gates.multicontrol_gate(tc.gates._x_matrix, ctrl=[1, 0])
-    ans = np.array(
-        [
-            [1.0, 0, 0, 0, 0, 0, 0, 0],
-            [0, 1.0, 0, 0, 0, 0, 0, 0],
-            [0, 0, 1.0, 0, 0, 0, 0, 0],
-            [0, 0, 0, 1.0, 0, 0, 0, 0],
-            [0, 0, 0, 0, 0, 1.0, 0, 0],
-            [0, 0, 0, 0, 1.0, 0, 0, 0],
-            [0, 0, 0, 0, 0, 0, 1.0, 0],
-            [0, 0, 0, 0, 0, 0, 0, 1.0],
-        ]
-    )
-    c = tc.Circuit(3)
-    c.X(0)
-    c.mpo(0, 1, 2, mpo=gate.copy())
-    np.testing.assert_allclose(c.expectation([tc.gates.z(), [2]]), -1, atol=1e-5)
-    c = tc.Circuit(3)
-    c.X(1)
-    c.mpo(0, 1, 2, mpo=gate.copy())
-    np.testing.assert_allclose(c.expectation([tc.gates.z(), [2]]), 1, atol=1e-5)
-    np.testing.assert_allclose(gate.eval_matrix(), ans, atol=1e-5)
-
-
-def test_apply_multicontrol_gate():
-    c = tc.Circuit(3)
-    c.X(2)
-    c.multicontrol(0, 2, 1, ctrl=[0, 1], unitary=tc.gates._x_matrix)
-    np.testing.assert_allclose(c.expectation([tc.gates.z(), [1]]), -1, atol=1e-5)
-    c = tc.Circuit(3)
-    c.X(0)
-    c.multicontrol(0, 2, 1, ctrl=[0, 1], unitary=tc.gates._x_matrix)
-    np.testing.assert_allclose(c.expectation([tc.gates.z(), [1]]), 1, atol=1e-5)
-    c = tc.Circuit(4)
-    c.X(0)
-    c.X(2)
-    c.multicontrol(0, 1, 2, 3, ctrl=[1, 0], unitary=tc.gates.swap())
-    np.testing.assert_allclose(c.expectation([tc.gates.z(), [3]]), -1, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_circuit_quoperator(backend):
-    c = tc.Circuit(3)
-    c.x(0)
-    c.cnot(0, 1)
-    c.cz(1, 2)
-    c.y(2)
-    c.exp1(0, 2, theta=1.0, unitary=tc.gates._xx_matrix)
-    c.H(1)
-    c.multicontrol(0, 2, 1, ctrl=[1, 0], unitary=tc.gates.z())
-    qo = c.quoperator()
-    np.testing.assert_allclose(qo.eval_matrix(), c.matrix(), atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_qir2cirq(backend):
-    try:
-        import cirq
-    except ImportError:
-        pytest.skip("cirq is not installed")
-    n = 6
-    c = tc.Circuit(n)
-    for i in range(n):
-        c.H(i)
-    zz = np.array([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
-    for i in range(n):
-        c.exp(
-            i,
-            (i + 1) % n,
-            theta=tc.array_to_tensor(np.random.uniform()),
-            unitary=tc.array_to_tensor(zz),
-            name="zz",
-        )
-    c.exp1(
-        1, 3, theta=tc.array_to_tensor(0.0j), unitary=tc.array_to_tensor(zz), name="zz"
-    )
-    c.fredkin(0, 1, 2)
-    c.cswap(1, 2, 3)
-    c.ccnot(1, 2, 3)
-    c.cx(2, 3)
-    c.swap(0, 1)
-    c.iswap(0, 1)
-    c.iswap(1, 3, theta=-1.9)
-    c.toffoli(0, 1, 2)
-    c.s(1)
-    c.t(1)
-    c.sd(1)
-    c.td(1)
-    c.x(2)
-    c.y(2)
-    c.z(2)
-    c.wroot(3)
-    c.cnot(0, 1)
-    c.cy(0, 1)
-    c.cz(0, 1)
-    c.oy(4, 3)
-    c.oz(4, 3)
-    c.ox(4, 3)
-    c.oy(4, 3)
-    c.oz(4, 3)
-    c.ox(3, 4)
-    c.phase(2, theta=0.3)
-    c.cphase(1, 0, theta=-1.2)
-    c.rxx(0, 2, theta=0.9)
-    c.ryy(1, 4, theta=-2.0)
-    c.rzz(1, 3, theta=0.5)
-    c.u(2, theta=0, lbd=4.6, phi=-0.3)
-    c.cu(4, 1, theta=1.2)
-    c.rx(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.r(5, theta=tc.array_to_tensor(np.random.uniform()))
-    c.cr(
-        1,
-        2,
-        theta=tc.array_to_tensor(np.random.uniform()),
-        alpha=tc.array_to_tensor(np.random.uniform()),
-        phi=tc.array_to_tensor(np.random.uniform()),
-    )
-    c.ry(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.rz(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.crz(2, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.crx(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.cry(1, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.orx(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.ory(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.orz(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-
-    c.any(1, 3, unitary=tc.array_to_tensor(np.reshape(zz, [2, 2, 2, 2])))
-
-    gate = tc.gates.multicontrol_gate(
-        tc.array_to_tensor(tc.gates._x_matrix), ctrl=[1, 0]
-    )
-    c.mpo(0, 1, 2, mpo=gate.copy())
-    c.multicontrol(
-        0,
-        2,
-        4,
-        1,
-        5,
-        ctrl=[0, 1, 0],
-        unitary=tc.array_to_tensor(tc.gates._zz_matrix),
-        name="zz",
-    )
-    tc_unitary = c.matrix()
-    tc_unitary = np.reshape(tc_unitary, [2**n, 2**n])
-
-    cirq = c.to_cirq()
-    cirq_unitary = cirq.unitary()
-    cirq_unitary = np.reshape(cirq_unitary, [2**n, 2**n])
-
-    np.testing.assert_allclose(tc_unitary, cirq_unitary, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_qir2qiskit(backend):
-    try:
-        import qiskit.quantum_info as qi
-
-        from tensorcircuit.translation import perm_matrix
-    except ImportError:
-        pytest.skip("qiskit is not installed")
-
-    n = 6
-    c = tc.Circuit(n, inputs=tc.array_to_tensor(np.eye(2**n)))
-
-    for i in range(n):
-        c.H(i)
-    zz = np.array([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
-    for i in range(n):
-        c.exp(
-            i,
-            (i + 1) % n,
-            theta=tc.array_to_tensor(np.random.uniform()),
-            unitary=tc.array_to_tensor(zz),
-            name="zz",
-        )
-    c.exp1(
-        1, 3, theta=tc.array_to_tensor(0.0j), unitary=tc.array_to_tensor(zz), name="zz"
-    )
-    c.fredkin(1, 2, 3)
-    c.cswap(1, 2, 3)
-    c.ccnot(1, 2, 3)
-    c.cx(2, 3)
-    c.swap(0, 1)
-    c.iswap(0, 1)
-    c.iswap(1, 3, theta=-1.9)
-    c.toffoli(0, 1, 2)
-    c.s(1)
-    c.t(1)
-    c.sd(1)
-    c.td(1)
-    c.x(2)
-    c.y(2)
-    c.z(2)
-    c.wroot(3)
-    c.cnot(0, 1)
-    c.cy(0, 1)
-    c.cz(0, 1)
-    c.oy(4, 3)
-    c.oz(4, 3)
-    c.ox(4, 3)
-    c.oy(4, 3)
-    c.oz(4, 3)
-    c.ox(3, 4)
-    c.phase(2, theta=0.3)
-    c.cphase(1, 0, theta=-1.2)
-    c.rxx(0, 2, theta=0.9)
-    c.ryy(1, 4, theta=-2.0)
-    c.rzz(1, 3, theta=0.5)
-    c.u(2, theta=0, lbd=4.6, phi=-0.3)
-    c.cu(4, 1, theta=1.2)
-    c.rx(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.r(5, theta=tc.array_to_tensor(np.random.uniform()))
-    c.cr(
-        1,
-        2,
-        theta=tc.array_to_tensor(np.random.uniform()),
-        alpha=tc.array_to_tensor(np.random.uniform()),
-        phi=tc.array_to_tensor(np.random.uniform()),
-    )
-    c.ry(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.rz(1, theta=tc.array_to_tensor(np.random.uniform()))
-    c.crz(2, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.crx(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.cry(1, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.orx(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.ory(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-    c.orz(5, 3, theta=tc.array_to_tensor(np.random.uniform()))
-
-    c.any(1, 3, unitary=tc.array_to_tensor(np.reshape(zz, [2, 2, 2, 2])))
-    gate = tc.gates.multicontrol_gate(
-        tc.array_to_tensor(tc.gates._x_matrix), ctrl=[1, 0]
-    )
-    c.mpo(0, 1, 2, mpo=gate.copy())
-    c.multicontrol(
-        0,
-        2,
-        4,
-        1,
-        5,
-        ctrl=[0, 1, 0],
-        unitary=tc.array_to_tensor(tc.gates._zz_matrix),
-        name="zz",
-    )
-    tc_unitary = c.wavefunction()
-    tc_unitary = np.reshape(tc_unitary, [2**n, 2**n])
-
-    qisc = c.to_qiskit()
-    qis_unitary = qi.Operator(qisc)
-    qis_unitary = np.reshape(qis_unitary, [2**n, 2**n])
-
-    p_mat = perm_matrix(n)
-    np.testing.assert_allclose(p_mat @ tc_unitary @ p_mat, qis_unitary, atol=1e-5)
-
-
-def test_qiskit2tc():
-    try:
-        import qiskit.quantum_info as qi
-        from qiskit import QuantumCircuit
-        from qiskit.circuit.library import HamiltonianGate
-        from qiskit.circuit.library.standard_gates import MCXGate, SwapGate
-
-        from tensorcircuit.translation import perm_matrix
-    except ImportError:
-        pytest.skip("qiskit is not installed")
-    n = 6
-    qisc = QuantumCircuit(n)
-    for i in range(n):
-        qisc.h(i)
-    zz = np.array([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
-    exp_op = qi.Operator(zz)
-    for i in range(n):
-        gate = HamiltonianGate(exp_op, time=np.random.uniform())
-        qisc.append(gate, [i, (i + 1) % n])
-    qisc.fredkin(1, 2, 3)
-    qisc.cswap(1, 2, 3)
-    qisc.swap(0, 1)
-    qisc.iswap(0, 1)
-    qisc.toffoli(0, 1, 2)
-    # test Instructions
-    qisc2 = QuantumCircuit(1)
-    qisc2.h(0)
-    qisc.compose(qisc2, qubits=[1], inplace=True, wrap=True)
-    qisc.barrier(0, 1, 2)
-    qisc.s(1)
-    qisc.t(1)
-    qisc.sdg(2)
-    qisc.tdg(2)
-    qisc.x(3)
-    qisc.y(3)
-    qisc.z(3)
-    qisc.cu(
-        5.868768495722669, 2.24809352294186, 3.59102783505607, 2.0223650288392, 1, 3
-    )
-    qisc.cnot(0, 1)
-    qisc.cy(0, 1)
-    qisc.cz(0, 1, ctrl_state=0)
-    qisc.cy(0, 1, ctrl_state=0)
-    qisc.cx(0, 1, ctrl_state=0)
-    qisc.rxx(0.3, 1, 2)
-    qisc.rzz(-0.8, 2, 0)
-    qisc.u(0.3, 0.9, -1.2, 2)
-    qisc.rx(np.random.uniform(), 1)
-    qisc.ry(np.random.uniform(), 2)
-    qisc.rz(np.random.uniform(), 3)
-    qisc.crz(np.random.uniform(), 2, 3)
-    qisc.crz(np.random.uniform(), 2, 3)
-    qisc.crz(np.random.uniform(), 2, 3)
-    qisc.crz(np.random.uniform(), 2, 3, ctrl_state=0)
-    qisc.crz(np.random.uniform(), 2, 3, ctrl_state=0)
-    qisc.crz(np.random.uniform(), 2, 3, ctrl_state=0)
-    qisc.r(np.random.uniform(), np.random.uniform(), 1)
-    qisc.unitary(exp_op, [1, 3])
-    mcx_g = MCXGate(3, ctrl_state="010")
-    qisc.append(mcx_g, [0, 1, 2, 3])
-    qisc.ccx(0, 1, 2, ctrl_state="01")
-    CCCRX = SwapGate().control(2, ctrl_state="01")
-    qisc.append(CCCRX, [0, 1, 2, 3])
-
-    c = tc.Circuit.from_qiskit(qisc, n, np.eye(2**n))
-    tc_unitary = c.wavefunction()
-    tc_unitary = np.reshape(tc_unitary, [2**n, 2**n])
-    qis_unitary = qi.Operator(qisc)
-    qis_unitary = np.reshape(qis_unitary, [2**n, 2**n])
-    p_mat = perm_matrix(n)
-    np.testing.assert_allclose(p_mat @ tc_unitary @ p_mat, qis_unitary, atol=1e-5)
-    qisc_from_tc = c.to_qiskit(enable_instruction=True)
-    qis_unitary2 = qi.Operator(qisc_from_tc)
-    qis_unitary2 = np.reshape(qis_unitary2, [2**n, 2**n])
-    np.testing.assert_allclose(qis_unitary2, qis_unitary, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("torchb")])
-def test_qiskit2tc_parameterized(backend):
-    try:
-        from qiskit.circuit import Parameter, ParameterVector, QuantumCircuit
-        from qiskit.circuit.library import TwoLocal
-        from qiskit.quantum_info import Operator
-        from qiskit_nature.second_q.circuit.library import UCCSD
-        from qiskit_nature.second_q.mappers import ParityMapper
-    except ImportError:
-        pytest.skip("qiskit or qiskit-nature is not installed")
-    from tensorcircuit.translation import perm_matrix
-
-    mapper = ParityMapper()
-    ansatz1 = UCCSD(2, [1, 1], mapper)
-    ansatz2 = TwoLocal(2, rotation_blocks="ry", entanglement_blocks="cz")
-    ansatz3 = QuantumCircuit(1)
-    ansatz3_param = Parameter("θ")
-    ansatz3.rx(ansatz3_param, 0)
-    ansatz4 = QuantumCircuit(1)
-    ansatz4_param = ParameterVector("φ", 3)
-    ansatz4.rx(2.0 * ansatz4_param[0] + 5.0, 0)
-    ansatz4.ry(ansatz4_param[0] * ansatz4_param[1] + ansatz4_param[2], 0)
-    ansatz4.rz(
-        np.exp(np.sin(ansatz4_param[0]))
-        + np.abs(ansatz4_param[1]) / np.arctan(ansatz4_param[2]),
-        0,
-    )
-    ansatz_list = [ansatz1, ansatz2, ansatz3, ansatz4]
-    for ansatz in ansatz_list:
-        n = ansatz.num_qubits
-        if ansatz in [ansatz1, ansatz2, ansatz4]:
-            params = np.random.rand(ansatz.num_parameters)
-        else:
-            params = {ansatz3_param: 0.618}
-        qisc = ansatz.assign_parameters(params)
-        qiskit_unitary = Operator(qisc)
-        qiskit_unitary = np.reshape(qiskit_unitary, [2**n, 2**n])
-
-        # test jit
-        @tc.backend.jit
-        def get_unitary(params):
-            return tc.Circuit.from_qiskit(
-                ansatz, inputs=np.eye(2**n), binding_params=params
-            ).state()
-
-        tc_unitary = get_unitary(params)
-        tc_unitary = np.reshape(tc_unitary, [2**n, 2**n])
-        p_mat = tc.array_to_tensor(perm_matrix(n))
-        np.testing.assert_allclose(
-            p_mat @ tc_unitary @ p_mat, qiskit_unitary, atol=1e-5
-        )
-
-        # test grad
-        def cost_fn(params):
-            return tc.backend.real(tc.backend.sum(get_unitary(params)))
-
-        if ansatz in [ansatz1, ansatz2, ansatz4]:
-            grad = tc.backend.grad(cost_fn)(tc.backend.convert_to_tensor(params))
-            assert tc.backend.sum(tc.num_to_tensor(np.isnan(grad))) == 0
-        else:
-            # tf only supports tf tensor as input
-            grad = tc.backend.grad(cost_fn)(
-                {ansatz3_param: tc.backend.convert_to_tensor(0.618)}
-            )
-            assert not np.isnan(grad[ansatz3_param])
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_qiskit_vs_tc_intialization(backend):
-    try:
-        import qiskit.quantum_info as qi
-        from qiskit import QuantumCircuit
-    except ImportError:
-        pytest.skip("qiskit is not installed")
-
-    n = 3
-
-    qis_c = QuantumCircuit(n)
-    qis_c.h(0)
-    qis_c.cnot(0, 1)
-    qis_c.y(2)
-    state = qi.Statevector(qis_c)
-    qis_c = QuantumCircuit(n)
-    qis_c.initialize(state)
-    qis_c.cnot(1, 2)
-    c = tc.Circuit.from_qiskit(qis_c)
-    c2 = tc.Circuit(n)
-    c2.h(0)
-    c2.cnot(0, 1)
-    c2.y(2)
-    c2.cnot(1, 2)
-    np.testing.assert_allclose(c.state(), c2.state(), atol=1e-8)
-    np.testing.assert_allclose(
-        qi.Statevector(c.to_qiskit(enable_inputs=True)),
-        qi.Statevector(qis_c),
-        atol=1e-8,
-    )
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_batch_sample(backend):
-    c = tc.Circuit(3)
-    c.H(0)
-    c.cnot(0, 1)
-    print(c.sample())
-    print(c.sample(batch=8))
-    print(c.sample(random_generator=tc.backend.get_random_state(42)))
-    print(c.sample(allow_state=True))
-    print(c.sample(batch=8, allow_state=True))
-    print(
-        c.sample(
-            batch=8, allow_state=True, random_generator=tc.backend.get_random_state(42)
-        )
-    )
-    print(
-        c.sample(
-            batch=8,
-            allow_state=True,
-            status=np.random.uniform(size=[8]),
-            format="sample_bin",
-        )
-    )
-
-
-def test_expectation_y_bug():
-    c = tc.Circuit(1, inputs=1 / np.sqrt(2) * np.array([-1, 1.0j]))
-    m = c.expectation_ps(y=[0])
-    np.testing.assert_allclose(m, -1, atol=1e-5)
-
-
-def test_lightcone_expectation():
-    def construct_c(pbc=True):
-        n = 4
-        ns = n
-        if pbc is False:
-            ns -= 1
-        c = tc.Circuit(n)
-        for j in range(2):
-            for i in range(n):
-                c.rx(i, theta=0.2, name="rx" + str(j) + "-" + str(i))
-            for i in range(ns):
-                c.cnot(i, (i + 1) % n, name="cnot" + str(j) + "-" + str(i))
-        return c
-
-    for b in [True, False]:
-        c = construct_c(b)
-        m1 = c.expectation_ps(z=[0], enable_lightcone=True)
-        m2 = c.expectation_ps(z=[0])
-        np.testing.assert_allclose(m1, m2, atol=1e-5)
-        nodes = c.expectation_before([tc.gates.z(), 0], reuse=False)
-        l1 = len(nodes)
-        nodes = tc.simplify._full_light_cone_cancel(nodes)
-        l2 = len(nodes)
-        if b is False:
-            assert l1 == 37 and l2 == 25
-        else:
-            assert l1 == 41 and l2 == 41
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_circuit_inverse(backend):
-    inputs = np.random.uniform(size=[8])
-    inputs /= np.linalg.norm(inputs)
-    c = tc.Circuit(3, inputs=inputs)
-    c.H(1)
-    c.rx(0, theta=0.5)
-    c.cnot(1, 2)
-    c.rzz(0, 2, theta=-0.8)
-    c1 = c.inverse()
-    c.append(c1)
-    np.testing.assert_allclose(c.state(), inputs, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_circuit_inverse_2(backend):
-    inputs = np.random.uniform(size=[8])
-    inputs /= np.linalg.norm(inputs)
-    c = tc.Circuit(3, inputs=inputs)
-    c.iswap(0, 1)
-    c.iswap(1, 0, theta=0.6)
-    c.rxx(1, 2, theta=-0.2)
-    c.cu(0, 1, lbd=2.0, theta=-0.7)
-    c.r(2, alpha=0.3)
-    c.sd(2)
-    c.cx(1, 2)
-    c.unitary(0, unitary=tc.gates._x_matrix)
-    c1 = c.inverse()
-    c.append(c1)
-    print(c.draw())
-    np.testing.assert_allclose(c.state(), inputs, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_jittable_amplitude(backend):
-    # @tc.backend.jit
-    def amp(s):
-        c = tc.Circuit(3)
-        c.H(0)
-        c.cnot(0, 1)
-        c.swap(1, 2)
-        return c.amplitude(s)
-
-    np.testing.assert_allclose(
-        amp(tc.array_to_tensor([0, 1, 1], dtype="float32")), 0, atol=1e-5
-    )
-    np.testing.assert_allclose(
-        amp(tc.array_to_tensor([0, 0, 0], dtype="float32")), 1 / np.sqrt(2), atol=1e-5
-    )
-
-
-def test_draw_cond_measure():
-    c = tc.Circuit(2)
-    c.H(0)
-    c.cond_measure(0)
-    c.cnot(0, 1)
-    print("")
-    print(c.draw())
-
-
-def test_minus_index():
-    c = tc.Circuit(3)
-    c.H(-2)
-    c.H(0)
-    np.testing.assert_allclose(tc.backend.real(c.expectation_ps(x=[0])), 1, atol=1e-5)
-    np.testing.assert_allclose(tc.backend.real(c.expectation_ps(x=[1])), 1, atol=1e-5)
-    np.testing.assert_allclose(tc.backend.real(c.expectation_ps(x=[-1])), 0, atol=1e-5)
-    np.testing.assert_allclose(tc.backend.real(c.expectation_ps(z=[-2])), 0, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_sexpps(backend):
-    c = tc.Circuit(1, inputs=1 / np.sqrt(2) * np.array([1.0, 1.0j]))
-    y = c.sample_expectation_ps(y=[0])
-    ye = c.expectation_ps(y=[0])
-    np.testing.assert_allclose(y, 1.0, atol=1e-5)
-    np.testing.assert_allclose(ye, 1.0, atol=1e-5)
-
-    c = tc.Circuit(4)
-    c.H(0)
-    c.cnot(0, 1)
-    c.rx(1, theta=0.3)
-    c.rz(2, theta=-1.2)
-    c.ccnot(2, 3, 1)
-    c.rzz(0, 3, theta=0.5)
-    c.ry(3, theta=2.2)
-    c.s(1)
-    c.td(2)
-    y = c.sample_expectation_ps(x=[1], y=[0], z=[2, 3])
-    ye = c.expectation_ps(x=[1], y=[0], z=[2, 3])
-    np.testing.assert_allclose(ye, y, atol=1e-5)
-    y2 = c.sample_expectation_ps(x=[1], y=[0], z=[2, 3], shots=81920)
-    assert np.abs(y2 - y) < 0.01
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_sample_format(backend):
-    c = tc.Circuit(2)
-    c.H(0)
-    c.cnot(0, 1)
-    key = tc.backend.get_random_state(42)
-    for allow_state in [False, True]:
-        print("allow_state: ", allow_state)
-        for batch in [None, 1, 3]:
-            print("  batch: ", batch)
-            for format_ in [
-                None,
-                "sample_int",
-                "sample_bin",
-                "count_vector",
-                "count_tuple",
-                "count_dict_bin",
-                "count_dict_int",
-            ]:
-                print("    format: ", format_)
-                print(
-                    "      ",
-                    c.sample(
-                        batch=batch,
-                        allow_state=allow_state,
-                        format_=format_,
-                        random_generator=key,
-                    ),
-                )
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_channel_auto_register(backend, highp):
-    c = tc.Circuit(2)
-    c.H(0)
-    c.reset(0, status=0.8)
-    s = c.state()
-    np.testing.assert_allclose(s[0], 1.0, atol=1e-9)
-
-
-@pytest.mark.parametrize("backend", [lf("npb"), lf("tfb"), lf("jaxb")])
-def test_circuit_to_json(backend):
-    c = tc.Circuit(3)
-    c.h(0)
-    c.CNOT(1, 2)
-    c.rxx(0, 2, theta=0.3)
-    c.crx(0, 1, theta=-0.8)
-    c.r(1, theta=tc.backend.ones([]), alpha=0.2)
-    c.toffoli(0, 2, 1)
-    c.ccnot(0, 1, 2)
-    c.multicontrol(1, 2, 0, ctrl=[0, 1], unitary=tc.gates._x_matrix)
-    s = c.to_json()
-    c2 = tc.Circuit.from_json(s)
-    print(c2.draw())
-    np.testing.assert_allclose(c.state(), c2.state(), atol=1e-5)
-
-
-def test_gate_count():
-    c = tc.Circuit(3)
-    c.x(0)
-    c.h(0)
-    c.rx(1, theta=-0.2)
-    c.h(2)
-    c.multicontrol(0, 1, 2, ctrl=[0, 1], unitary=tc.gates._x_matrix)
-    c.toffoli(0, 2, 1)
-    c.ccnot(1, 2, 0)
-    c.ccx(1, 2, 0)
-    assert c.gate_count() == 8
-    assert c.gate_count("h") == 2
-    assert c.gate_count(["ccnot"]) == 3
-    assert c.gate_count(["rx", "multicontrol"]) == 2
-    assert c.gate_count_by_condition(lambda qir: qir["index"] == (0,)) == 2
-    assert c.gate_count_by_condition(lambda qir: qir["mpo"]) == 1
-    print(c.gate_summary())
-    # {'x': 1, 'h': 2, 'rx': 1, 'multicontrol': 1, 'toffoli': 3}
-
-
-def test_to_openqasm(tmp_path):
-    c = tc.Circuit(3)
-    c.H(0)
-    c.rz(2, theta=0.2)
-    c.cnot(2, 1)
-    c.rzz(0, 1, theta=-1.0)
-    c.ccx(1, 2, 0)
-    c.u(2, theta=0.5, lbd=1.3)
-    c.measure_instruction(1)
-    print(c.to_openqasm(formatted=True))
-    s = c.to_openqasm()
-    c1 = tc.Circuit.from_openqasm(s)
-    print(c1.draw())
-    np.testing.assert_allclose(c.state(), c1.state())
-    c.to_openqasm_file(os.path.join(tmp_path, "test.qasm"))
-    c2 = tc.Circuit.from_openqasm_file(os.path.join(tmp_path, "test.qasm"))
-    np.testing.assert_allclose(c.state(), c2.state())
-
-
-def test_from_qasm_keep_measure_order():
-    qasm_str = """OPENQASM 2.0;
-include "qelib1.inc";
-qreg q[2];
-creg c[2];
-h q[0];
-measure q[1] -> c[1];
-measure q[0] -> c[0];"""
-    c = tc.Circuit.from_openqasm(qasm_str)
-    c.to_openqasm().split("\n")[-2][-3] == "1"
-    c = tc.Circuit.from_openqasm(qasm_str, keep_measure_order=True)
-    c.to_openqasm().split("\n")[-2][-3] == "0"
-
-
-def test_initial_mapping():
-    c = tc.Circuit(3)
-    c.cnot(0, 1)
-    c.h(1)
-    c.rx(1, theta=0.5)
-    c.cz(2, 1)
-    c.measure_instruction(2)
-
-    c1 = c.initial_mapping({0: 1, 1: 2, 2: 0})
-    print(c1.draw())
-
-    np.testing.assert_allclose(
-        c.expectation_ps(z=[1]), c1.expectation_ps(z=[2]), atol=1e-5
-    )
-    assert c1._extra_qir[0]["index"][0] == 0
-
-    c2 = c1.initial_mapping({1: 0, 2: 1, 0: 2})
-    np.testing.assert_allclose(
-        c.expectation_ps(z=[1]), c2.expectation_ps(z=[1]), atol=1e-5
-    )
-
-    c3 = c.initial_mapping({0: 2, 1: 7, 2: 0}, n=9)
-    np.testing.assert_allclose(
-        c.expectation_ps(z=[1]), c3.expectation_ps(z=[7]), atol=1e-5
-    )
-    print(c3.draw())
-
-
-def test_get_positional_logical_mapping():
-    c = tc.Circuit(3)
-    c.cx(0, 1)
-    c.cz(1, 2)
-    c.h(1)
-    assert c.get_positional_logical_mapping() == {0: 0, 1: 1, 2: 2}
-    c = tc.Circuit(3)
-    c.cx(0, 1)
-    c.h(1)
-    c.measure_instruction(2)
-    c.measure_instruction(0)
-    assert c.get_positional_logical_mapping() == {0: 2, 1: 0}
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb")])
-def test_inverse_jit(backend):
-    K = tc.backend
-
-    def simple_ansatz(param):
-        c = tc.Circuit(3)
-        for i in range(3):
-            c.cx(i, (i + 1) % 3)
-            c.rzz(i, (i + 1) % 3, theta=param[i])
-        c1 = c.inverse()
-        c2 = tc.Circuit(3)
-        c2.x(1)
-        c1.append(c2)
-        return tc.backend.real(c1.expectation_ps(z=[1]))
-
-    v_ansatz = K.jit(K.vvag(simple_ansatz))
-    vs, gs = v_ansatz(K.ones([2, 3], dtype="float32"))
-    assert K.shape_tuple(gs) == (2, 3)
-    np.testing.assert_allclose(K.numpy(vs), -1.0 * K.ones([2]), atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("npb")])
-def test_fancy_circuit_indexing(backend):
-    c = tc.Circuit(4)
-    c.cx([0, 1], [-1, -2])
-    c.h(range(c._nqubits))
-    c.rz([0], theta=0.2)
-    c.rx([1, 2], theta=[0.3, 0.5])
-    c.rzz([2, 3], [0, 1], theta=tc.backend.ones([2]))
-    c.rxx([2, 0, 1], [0, 1, 2], theta=tc.backend.ones([1]))
-    assert c.gate_count("h") == 4
-    assert c.gate_count("rzz") == 2
-    assert c.gate_count("rxx") == 3
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("npb")])
-def test_general_kraus(backend):
-    c = tc.Circuit(2)
-    c.h([0, 1])
-    p = 0.5
-    status = [0.3, 0.8]
-    rs = []
-    for i in range(2):
-        rs.append(
-            c.general_kraus(
-                [
-                    np.sqrt(p) * np.array([[1.0, 0], [0, 0]]),
-                    np.sqrt(p) * np.array([[0, 0], [0, 1.0]]),
-                    np.sqrt(1 - p) * np.eye(2),
-                ],
-                i,
-                status=status[i],
-            )
-        )
-    np.testing.assert_allclose(rs[0], 1)
-    np.testing.assert_allclose(rs[1], 2)
-    np.testing.assert_allclose(c.expectation_ps(z=[0]), -1, atol=1e-5)
-    np.testing.assert_allclose(c.expectation_ps(z=[1]), 0, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("npb")])
-def test_general_kraus_with_prob(backend):
-    c = tc.Circuit(2)
-    c.h([0, 1])
-    p = 0.5
-    status = [0.3, 0.8]
-    rs = []
-    for i in range(2):
-        rs.append(
-            c.general_kraus(
-                [
-                    np.sqrt(p) * np.array([[1.0, 0], [0, 0]]),
-                    np.sqrt(p) * np.array([[0, 0], [0, 1.0]]),
-                    np.sqrt(1 - p) * np.eye(2),
-                ],
-                i,
-                status=status[i],
-                with_prob=True,
-            )
-        )
-    np.testing.assert_allclose(rs[0][0], 1)
-    np.testing.assert_allclose(rs[1][0], 2)
-    np.testing.assert_allclose(c.expectation_ps(z=[0]), -1, atol=1e-5)
-    np.testing.assert_allclose(c.expectation_ps(z=[1]), 0, atol=1e-5)
-    np.testing.assert_allclose(rs[0][1], [0.25, 0.25, 0.5], atol=1e-5)
-    np.testing.assert_allclose(rs[1][1], [0.25, 0.25, 0.5], atol=1e-5)
-    np.testing.assert_allclose(tc.backend.norm(c.state()), 1, atol=1e-5)
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("npb")])
-def test_circuit_copy(backend):
-    c = tc.Circuit(2)
-    c.h(0)
-    c1 = c.copy()
-    c.rz(0, theta=0.1)
-    assert c1.gate_count() == 1
-
-
-@pytest.mark.parametrize("backend", [lf("tfb"), lf("jaxb"), lf("npb")])
-def test_projected_subsystem(backend):
-    c = tc.Circuit(3)
-    c.h(0)
-    c.cnot(0, 1)
-    c.rx(1, theta=0.9)
-    c.cy(1, 2)
-    s = c.projected_subsystem(tc.backend.convert_to_tensor(np.array([1, 1, 1.0])), (0,))
-    np.testing.assert_allclose(s, np.array([0.43496, 0.900447j]), atol=1e-5)
-    s = c.projected_subsystem(
-        tc.backend.convert_to_tensor(np.array([0, 0, 0.0])), (0, 2)
-    )
-    np.testing.assert_allclose(s[0], 0.900447, atol=1e-5)
-
-    c = tc.DMCircuit(3)
-    c.h(0)
-    c.cnot(0, 1)
-    c.rx(1, theta=0.9)
-    c.cy(1, 2)
-    s = c.projected_subsystem(tc.backend.convert_to_tensor(np.array([1, 1, 1.0])), (0,))
-    assert tc.backend.shape_tuple(s) == (2, 2)
-    np.testing.assert_allclose(s[1, 1], 0.8108051, atol=1e-5)
-    s = c.projected_subsystem(
-        tc.backend.convert_to_tensor(np.array([1, 1, 1.0])), (1, 2)
-    )
-    assert tc.backend.shape_tuple(s) == (4, 4)
-    np.testing.assert_allclose(s[3, 3], 0.8108051, atol=1e-5)