예제 #1
0
파일: swarm.py 프로젝트: AMC-dyn/PyMCE
def poptraj(T1):
    dyn = initdyn()
    geo = initgeom()
    ph = physconst()

    pec, der = abinitio.inp_out(0, 0, geo, T1)  # First ab-initio run

    T1.setpotential_traj(pec)  # taking V(R) from ab-initio
    T1.setderivs_traj(
        der
    )  # derivatives matrix mass-weighted (possibly change that), diagonals are forces and off-d are nacmes
    T1.setmass_traj(
        geo.masses
    )  # mass of every atom in a.u (the dimmension is natoms/nparts)
    T1.setmassall_traj(
        geo.massrk
    )  # mass in every degree of freedom (careful to use it, it can triple the division/multiplication easily)

    amps = np.zeros(T1.nstates, dtype=np.complex128)
    amps[
        dyn.inipes -
        1] = 1.00  # Amplitudes of Ehrenfest trajectories, they should be defined as a=d *exp(im*S)

    T1.setamplitudes_traj(amps)

    phases = np.zeros(
        T1.nstates)  # Phase of the wfn, would be S in the previous equation

    T1.setphases_traj(phases)
    T1.setwidth_traj(dyn.gamma)

    return T1
예제 #2
0
파일: swarm.py 프로젝트: AMC-dyn/PyMCE
def inittraj():
    dyn = initdyn()
    geo = initgeom()
    ph = physconst()
    '''First initialize and populate one trajectory'''

    T1 = initialize_traj.trajectory(geo.natoms, 3, dyn.nstates)
    # qin, pin = Wigner_dist.WignerSampling()

    q = np.zeros(geo.ndf)

    p = np.zeros_like(q)
    with open('initialqp.dat', 'r') as f:
        f.readline()
        for i in range(geo.ndf):
            N, M = f.readline().strip().split()
            q[i] = np.double(float(N.replace('D', 'E'))) / np.sqrt(
                geo.massrk[i])
            p[i] = np.double(float(M.replace('D', 'E'))) * np.sqrt(
                geo.massrk[i])

    T1.setposition_traj(q + geo.rkinit)
    T1.setmomentum_traj(p)

    pec, der = abinitio.inp_out(0, 0, geo, T1)  # First ab-initio run
    print(pec)
    T1.setpotential_traj(pec)  # taking V(R) from ab-initio
    T1.setderivs_traj(
        der
    )  # derivatives matrix mass-weighted (possibly change that), diagonals are forces and off-d are nacmes
    T1.setmass_traj(
        geo.masses
    )  # mass of every atom in a.u (the dimmension is natoms/nparts)
    T1.setmassall_traj(
        geo.massrk
    )  # mass in every degree of freedom (careful to use it, it can triple the division/multiplication easily)

    amps = np.zeros(T1.nstates, dtype=np.complex128)
    amps[
        dyn.inipes -
        1] = 1.00  # Amplitudes of Ehrenfest trajectories, they should be defined as a=d *exp(im*S)

    T1.setamplitudes_traj(amps)

    phases = np.zeros(
        T1.nstates)  # Phase of the wfn, would be S in the previous equation

    T1.setphases_traj(phases)
    T1.setwidth_traj(dyn._gamma)

    return T1
예제 #3
0
def velocityverlet(T, timestep, NN):
    geo = initgeom()
    ii = complex(0, 1.00)
    magnus_slice = 20
    nst = T.nstates
    M = T.getmassall_traj()
    print(M)
    R0 = T.getposition_traj()
    P0 = T.getmomentum_traj()
    V0 = T.getvelocity_traj()
    FO = T.get_traj_force()
    E0 = T.getpotential_traj()
    A0 = T.getamplitude_traj()

    HE_0 = np.zeros((nst, nst), dtype=np.complex128)
    for n1 in range(nst):
        HE_0[n1, n1] = T.getpotential_traj_i(n1)
        for n2 in range(n1 + 1, nst):
            HE_0[n1, n2] = -ii * coupdotvel(T, n1, n2)
            HE_0[n2, n1] = -HE_0[n1, n2]

    nslice = magnus_slice

    Ab = A0
    F0 = 0.0
    for i in range(1, nslice + 1):
        dt = timestep / nslice
        A1 = np.matmul(magnus_2(-ii * HE_0, -ii * HE_0, dt), Ab)
        Ab = A1

        F0 += T.get_traj_force() / nslice

    T.stateAmpE = A0
    es0 = np.zeros(nst)
    fs0 = np.zeros((T.ndim, nst))
    cs0 = np.zeros((T.ndim, nst, nst))

    for i in range(nst):
        es0[i] = T.getpotential_traj_i(i)
        fs0[:, i] = T.getforce_traj(i)
        for j in range(nst):
            cs0[:, i, j] = T.getcoupling_traj(i, j)
    T.phase += timestep / 2.0 * T.phasedot()

    R1 = R0 + timestep * V0 + timestep**2 / 2.0 * F0 / M

    T.setposition_traj(R1)

    pes, der = ab.inp_out(NN, 0, geo, T)
    print('coupling before: ', T.getcoupling_traj(0, 1))
    T.setderivs_traj(der)
    print('coupling after: ', T.getcoupling_traj(0, 1))
    F1 = T.get_traj_force()
    T.setpotential_traj(pes)
    es1 = np.zeros(nst)
    fs1 = np.zeros((T.ndim, nst))
    cs1 = np.zeros((T.ndim, nst, nst))

    for i in range(nst):
        es1[i] = T.getpotential_traj_i(i)
        fs1[:, i] = T.getforce_traj(i)
        for j in range(nst):
            cs1[:, i, j] = T.getcoupling_traj(i, j)

    HE_1 = np.zeros_like(HE_0, dtype=np.complex128)

    for n1 in range(nst):
        HE_1[n1, n1] = T.getpotential_traj_i(n1)
        for n2 in range(n1 + 1, nst):
            HE_1[n1, n2] = -ii * coupdotvel(T, n1, n2)
            HE_1[n2, n1] = -HE_1[n1, n2]

    nslice = magnus_slice
    Ab = A0
    F1 = 0.0
    for n in range(1, nslice + 1):
        dt = timestep / nslice

        f_b = (n - 0.5) / np.double(float(nslice))
        HE_b = (1.0 - f_b) * HE_0 + f_b * HE_1
        esb = (1.0 - f_b) * es0 + f_b * es1
        fsb = (1.0 - f_b) * fs0 + f_b * fs1
        csb = (1.0 - f_b) * cs0 + f_b * cs1
        A1 = np.matmul(magnus_2(-ii * HE_b, -ii * HE_b, dt), Ab)
        Ab = A1
        T.stateAmpE = Ab
        T.HE = HE_1
        fb = ip.compforce(T, A1, fsb, esb, csb)
        F1 += fb * 1.00 / nslice

    P1 = P0 + timestep * F1

    T.setamplitudes_traj(Ab)

    T.setmomentum_traj(P1)
    T.phase += timestep / 2.0 * T.phasedot()

    T.setphases_traj(T.phase)
    T.setoldpos_traj(R0)
    T.setoldmom_traj(P0)
    return T
예제 #4
0
파일: swarm.py 프로젝트: AMC-dyn/PyMCE
def createswarm(ntraj, npart, ndim, numstates):
    geo = geometry.initgeom()
    trajs = []
    Tinit = inittraj()
    Tinit.settrajid_traj(0)
    # trajs.append(Tinit)
    print(Tinit.PotEn)
    X = Tinit.getposition_traj()
    print(X)
    P = Tinit.getmomentum_traj()
    sigma = 0.1
    p_norm = np.sqrt(np.sum(P**2))
    for i in range(ntraj):
        T = initialize_traj.trajectory(npart=npart,
                                       ndim=ndim,
                                       numstates=numstates)
        n1 = np.random.rand()
        dx = sigma * (2 * n1 - 1.0)
        n2 = np.random.rand()
        dp = sigma / p_norm * (2 * n2 - 1)

        T.setposition_traj(X + dx)

        T.setmomentum_traj(P + dp)
        T.setamplitudes_traj(Tinit.getamplitude_traj())
        T.setphases_traj(Tinit.getphase_traj())
        T.setwidth_traj(Tinit.getwidth_traj())
        print('widths:', T.getwidth_traj())
        T.settrajid_traj(i)
        pec, der = abinitio.inp_out(0, 0, geo, T)  # First ab-initio run
        T.setpotential_traj(pec)  # taking V(R) from ab-initio

        T.setderivs_traj(
            der
        )  # derivatives matrix mass-weighted (possibly change that), diagonals are forces and off-d are nacmes
        T.setmass_traj(
            geo.masses
        )  # mass of every atom in a.u (the dimmension is natoms/nparts)
        T.setmassall_traj(
            geo.massrk
        )  # mass in every degree of freedom (careful to use it, it can triple the division/multiplication easily)
        trajs.append(T)

    B = bundle.bundle(ntraj, npart, ndim, numstates)
    B.setTraj_bundle(trajs)
    norm = B.get_calc_set_norm()

    for i in range(ntraj):
        B.Traj[i].amp = B.Traj[i].amp / cmath.sqrt(norm)

    B.setamps_bundle(B.getamps_bundle() / cmath.sqrt(norm))

    print('Amplitudes before normalizing: ', B.getamps_bundle())
    B = buildhs.buildsandh(B)

    Sif = np.zeros(ntraj, dtype=np.complex128)
    for i in range(ntraj):
        Sif[i] = ov.overlap_trajs(Tinit, B.Traj[i])

    print('Overlaps: ', Sif)
    print('Sinv :', B.Sinv)

    Sif_2 = np.matmul(np.conj(B.Sinv), Sif)

    print('Overlaps*Sinv: ', Sif_2)
    B.setamps_bundle(Sif_2)

    norm = B.get_calc_set_norm()

    print('norm: ', norm)

    B.setamps_bundle(B.getamps_bundle() / np.sqrt(norm))

    print('Final bundle amplitudes: ', B.getamps_bundle())
    print('Norm Final bundle: ', np.linalg.norm(B.getamps_bundle()))
    print('Norm Final bundle branch: ', B.get_calc_set_norm())
    return B
예제 #5
0
def velocityverlet(T, timestep, NN, calc1, phasewf):
    ab2 = ab_par()
    geo = initgeom()
    geo2 = singlepart()
    ii = np.complex128(0 + 1.00j)
    magnus_slice = 20
    nst = T.nstates
    M = T.getmassall_traj()
    R0 = T.getposition_traj()
    P0 = T.getmomentum_traj()
    V0 = T.getvelocity_traj()
    A0 = T.getamplitude_traj()
    phase = T.getphase_traj()
    HE_0 = np.zeros((nst, nst), dtype=np.complex128)
    for n1 in range(nst):
        HE_0[n1, n1] = T.getpotential_traj_i(n1)
        for n2 in range(n1 + 1, nst):
            HE_0[n1, n2] = np.complex128(-ii * coupdotvel(T, n1, n2))
            HE_0[n2, n1] = -HE_0[n1, n2]

    nslice = magnus_slice

    Ab = A0
    F0 = 0.0
    for i in range(0, nslice):
        dt = timestep / np.double(nslice)
        if T.nstates > 1:
            A1 = np.matmul(magnus_2(-ii * HE_0, -ii * HE_0, dt),
                           Ab,
                           dtype=np.complex128)
        else:
            A1 = magnus_2(-ii * HE_0, -ii * HE_0, dt) * Ab

        Ab = A1
        T.setamplitudes_traj(A1)
        F0 += T.get_traj_force() / nslice

    T.setamplitudes_traj(A0)
    es0 = np.zeros(nst)
    fs0 = np.zeros((T.ndim, nst))
    cs0 = np.zeros((T.ndim, nst, nst))

    for i in range(nst):
        es0[i] = T.getpotential_traj_i(i)
        fs0[:, i] = T.getforce_traj(i)
        for j in range(nst):
            cs0[:, i, j] = T.getcoupling_traj(i, j)
    phase += timestep / 2.0 * T.phasedot()

    R1 = R0 + timestep * V0 + timestep**2.0 / 2.00 * F0 / M

    P1 = P0 + timestep * F0
    oldcis = T.getcivecs()
    T.setoldpos_traj(R0)
    T.setoldmom_traj(P0)
    T_try = copy(T)
    T.setposition_traj(R1)
    T.setmomentum_traj(P1)

    if not calc1:
        pes, der, cis, configs = ab.inp_out(NN, 0, geo, T)
    else:
        pes = np.sum(0.5 * geo2.K * T.getposition_traj()**2)
        der = np.zeros(3)
        cis = 0
        configs = 0
        for i in range(3):
            der[i] = -geo2.K * T.getposition_traj()[i]

    T.setderivs_traj(der)
    T.setpotential_traj(pes)
    T.setcivecs(cis)
    T.setconfigs(configs)
    phasewf = T.getphasewf()
    ovs = np.zeros((T.nstates, T.nstates))

    ov1, ov2 = ovwf(T_try, T)
    print('wfoverlaps', ov1, ov2)
    for n1 in range(T.nstates):
        for n2 in range(T.nstates):
            ovs[n1, n2] = (np.dot(T.getcivecs()[:, n1], oldcis[:, n2]))
    print(ovs[0, 0])
    print(ovs[1, 1])
    print(abs(ovs[0, 0]) + abs(ovs[1, 1]), abs(ovs[0, 1]) + abs(ovs[1, 0]))
    # if abs(ovs[0, 0]) + abs(ovs[1, 1]) < abs(ovs[0, 1]) + abs(ovs[1, 0]):
    #     print('Trying to reduce timestep')
    #     T.setposition_traj(T.getoldpos_traj())
    #     T.setmomentum_traj(T.getoldmom_traj())
    #     os.system('cp /home/AndresMoreno/wfu/002.molpro.wfu /home/AndresMoreno/wfu/003.molpro.wfu')
    #     os.system('cp 002.molpro.wfu 003.molpro.wfu')
    #     pes, der, cis = ab.inp_out(NN, 0, geo, T)
    #     T.setderivs_traj(der)
    #     T.setpotential_traj(pes)
    #     T.setcivecs(cis)
    #     print('momentum stored: ', T.getmomentum_traj()[0])
    #     print(T.getkineticlass() + T.getpotential_traj())
    #
    #     timestep = timestep / 2.00
    #     for ts in range(2):
    #         T = velocityverlet(T, timestep, 120+ts, calc1, phasewf)
    #     print('returning to the main routine')
    #     return T

    print(T.getcivecs()[:, 1])
    print(oldcis[:, 1])
    for i in range(T.nstates):
        ovs = np.dot(T.getcivecs()[:, i], oldcis[:, i])
        print(ovs)
        if abs(ovs) > 0.9:
            phasewf = phasewf * np.dot(
                T.getcivecs()[:, i], oldcis[:, i]) / np.abs(
                    np.dot(T.getcivecs()[:, i], oldcis[:, i]))
        else:
            print('STEP TO CHECK; CI OVERLAP IS WRONG')
            changingindex = np.ones(np.size(T_try.configs)).astype(int)
            megamatrix1, M1, syms1 = readingmolden(T_try.getfilecalc())
            megamatrix2, M2, syms2 = readingmolden(T.getfilecalc())

            syms1 = (syms1[ab2.closed_orb:] - ab2.closed_orb).astype(int)
            syms2 = (syms2[ab2.closed_orb:] - ab2.closed_orb).astype(int)

            for nc in range(np.size(T_try.configs)):
                cfg = T.configs[nc]
                newcfg = ''
                for nstring in range(len(cfg)):
                    newcfg += cfg[syms2[nstring]]
                index1 = T_try.configs.index(newcfg)
                changingindex[nc] = int(index1)
                print(index1)
            newCIs = T.getcivecs()[changingindex, :]
            print(oldcis[:, i])
            print(newCIs[:, i])
            print('changed CIvectors')

            ovs_2 = np.dot(newCIs[:, i], oldcis[:, i])
            print('newoverlap: ', ovs_2)

    print('up to here')
    derivs = np.zeros((T.ndim, T.nstates, T.nstates))
    for n1 in range(T.nstates):
        for n2 in range(T.nstates):
            if n1 != n2:
                derivs[:, n1, n2] = phasewf * T.getcoupling_traj(n1, n2)
            else:
                derivs[:, n1, n1] = T.getforce_traj(n1)
    T.setphasewf(phasewf)
    T.setderivs_traj(derivs)
    es1 = np.zeros(nst)
    fs1 = np.zeros((T.ndim, nst))
    cs1 = np.zeros((T.ndim, nst, nst))

    for i in range(nst):
        es1[i] = T.getpotential_traj_i(i)
        fs1[:, i] = T.getforce_traj(i)
        for j in range(nst):
            cs1[:, i, j] = T.getcoupling_traj(i, j)

    HE_1 = np.zeros_like(HE_0, dtype=np.complex128)

    for n1 in range(nst):
        HE_1[n1, n1] = T.getpotential_traj_i(n1)
        for n2 in range(n1 + 1, nst):
            HE_1[n1, n2] = np.complex128(-ii * coupdotvel(T, n1, n2))
            HE_1[n2, n1] = -HE_1[n1, n2]

    nslice = magnus_slice
    Ab = A0
    F1 = 0.0
    for n in range(1, nslice + 1):
        dt = timestep / np.double(float(nslice))

        f_b = (n - 0.5) / np.double(float(nslice))
        HE_b = (1.0 - f_b) * HE_0 + f_b * HE_1
        esb = (1.0 - f_b) * es0 + f_b * es1
        fsb = (1.0 - f_b) * fs0 + f_b * fs1
        csb = (1.0 - f_b) * cs0 + f_b * cs1
        if T.nstates > 1:
            A1 = np.matmul(magnus_2(-ii * HE_b, -ii * HE_b, dt),
                           Ab,
                           dtype=np.complex128)
        else:
            A1 = magnus_2(-ii * HE_b, -ii * HE_b, dt) * Ab

        Ab = A1

        T.setamplitudes_traj(A1)
        T.HE = HE_1
        fb = T.compforce(A1, fsb, esb, csb)
        F1 += fb / np.double(float(nslice))

    P1 = P0 + timestep * F1
    T.setmomentum_traj(P1)
    phase += timestep / 2.0 * T.phasedot()
    ICycle = np.floor(phase / (2.0000 * np.pi))
    phase = phase - 2.000 * np.pi * ICycle
    T.setphase_traj(phase)
    T.setoldpos_traj(R0)
    T.setoldmom_traj(P0)
    return T
예제 #6
0
def rkf45(T, dt, time, nstep):
    geo = initgeom()
    d = T.getd_traj()
    q = T.getposition_traj()
    p = T.getmomentum_traj()
    s = T.getphases_traj()
    A0 = T.getamplitude_traj()
    ii = np.complex128(0 + 1j)
    nstates = T.nstates
    masses = T.getmassall_traj()

    time = np.linspace(0, dt, 10)
    # Coefficients used to compute the independent variable argument of f

    a2 = 2.500000000000000e-01  # 1/4
    a3 = 3.750000000000000e-01  # 3/8
    a4 = 9.230769230769231e-01  # 12/13
    a5 = 1.000000000000000e+00  # 1
    a6 = 5.000000000000000e-01  # 1/2

    # Coefficients used to compute the dependent variable argument of f

    b21 = 2.500000000000000e-01  # 1/4
    b31 = 9.375000000000000e-02  # 3/32
    b32 = 2.812500000000000e-01  # 9/32
    b41 = 8.793809740555303e-01  # 1932/2197
    b42 = -3.277196176604461e+00  # -7200/2197
    b43 = 3.320892125625853e+00  # 7296/2197
    b51 = 2.032407407407407e+00  # 439/216
    b52 = -8.000000000000000e+00  # -8
    b53 = 7.173489278752436e+00  # 3680/513
    b54 = -2.058966861598441e-01  # -845/4104
    b61 = -2.962962962962963e-01  # -8/27
    b62 = 2.000000000000000e+00  # 2
    b63 = -1.381676413255361e+00  # -3544/2565
    b64 = 4.529727095516569e-01  # 1859/4104
    b65 = -2.750000000000000e-01  # -11/40

    # Coefficients used to compute local truncation error estimate.  These
    # come from subtracting a 4th order RK estimate from a 5th order RK
    # estimate.

    r1 = 2.777777777777778e-03  # 1/360
    r3 = -2.994152046783626e-02  # -128/4275
    r4 = -2.919989367357789e-02  # -2197/75240
    r5 = 2.000000000000000e-02  # 1/50
    r6 = 3.636363636363636e-02  # 2/55

    # Coefficients used to compute 4th order RK estimate

    c1 = 1.157407407407407e-01  # 25/216
    c3 = 5.489278752436647e-01  # 1408/2565
    c4 = 5.353313840155945e-01  # 2197/4104
    c5 = -2.000000000000000e-01  # -1/5

    n = len(time)

    a = 0.0
    b = 2.5
    t = a
    hmax = 1.25
    h = hmax
    hmin = 0.01
    tol = 1e-7
    tol2 = 0.000001
    T2 = T
    while t < b:
        T2 = T
        energy0 = T.getkineticlass() + T.getpotential_traj()
        print(h, energy0)
        if t + h > b:
            h = b - t

        qk1 = h * qdot(T)
        pk1 = h * pdot(T)
        dk1 = h * ddot(T)
        sk1 = h * sdot(T)

        T2.setposition_traj(q + qk1)
        T2.setmomentum_traj(p + pk1)
        T2.setd_traj(d + dk1)
        T2.setphases_traj(s + sk1)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        qk2 = h * (qdot(T) + b21 * qk1)
        pk2 = h * (pdot(T) + b21 * pk1)
        dk2 = h * (ddot(T) + b21 * dk1)
        sk2 = h * (sdot(T) + b21 * sk1)

        T2.setposition_traj(q + qk2)
        T2.setmomentum_traj(p + pk2)
        T2.setd_traj(d + dk2)
        T2.setphases_traj(s + sk2)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        qk3 = h * (qdot(T) + b31 * qk1 + b32 * qk2)
        pk3 = h * (pdot(T) + b31 * pk1 + b32 * pk2)
        dk3 = h * (ddot(T) + b31 * dk1 + b32 * dk2)
        sk3 = h * (sdot(T) + b31 * sk1 + b32 * sk2)

        T2.setposition_traj(q + qk3)
        T2.setmomentum_traj(p + pk3)
        T2.setd_traj(d + dk3)
        T2.setphases_traj(s + sk3)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        qk4 = h * (qdot(T) + b41 * qk1 + b42 * qk2 + b43 * qk3)
        pk4 = h * (pdot(T) + b41 * pk1 + b42 * pk2 + b43 * pk3)
        dk4 = h * (ddot(T) + b41 * dk1 + b42 * dk2 + b43 * dk3)
        sk4 = h * (sdot(T) + b41 * sk1 + b42 * sk2 + b43 * sk3)

        T2.setposition_traj(q + qk4)
        T2.setmomentum_traj(p + pk4)
        T2.setd_traj(d + dk4)
        T2.setphases_traj(s + sk4)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        qk5 = h * (qdot(T) + b51 * qk1 + b52 * qk2 + b53 * qk3 + b54 * qk4)
        pk5 = h * (pdot(T) + b51 * pk1 + b52 * pk2 + b53 * pk3 + b54 * pk4)
        dk5 = h * (ddot(T) + b51 * dk1 + b52 * dk2 + b53 * dk3 + b54 * dk4)
        sk5 = h * (sdot(T) + b51 * sk1 + b52 * sk2 + b53 * sk3 + b54 * sk4)

        T2.setposition_traj(q + qk5)
        T2.setmomentum_traj(p + pk5)
        T2.setd_traj(d + dk5)
        T2.setphases_traj(s + sk5)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        qk6 = h * (qdot(T) + b61 * qk1 + b62 * qk2 + b63 * qk3 + b64 * qk4 +
                   b65 * qk5)
        pk6 = h * (pdot(T) + b61 * pk1 + b62 * pk2 + b63 * pk3 + b64 * pk4 +
                   b65 * pk5)
        dk6 = h * (ddot(T) + b61 * dk1 + b62 * dk2 + b63 * dk3 + b64 * dk4 +
                   b65 * dk5)
        sk6 = h * (sdot(T) + b61 * sk1 + b62 * sk2 + b63 * sk3 + b64 * sk4 +
                   b65 * sk5)

        T2.setposition_traj(q + qk6)
        T2.setmomentum_traj(p + pk6)
        T2.setd_traj(d + dk6)
        T2.setphases_traj(s + sk6)
        T2.setamplitudes_traj(
            T2.getd_traj() *
            np.exp(T2.getphases_traj() * 1j, dtype=np.complex128))

        epot, derivs = ab.inp_out(nstep, 0, geo, T2)
        T.setderivs_traj(derivs)
        T.setpotential_traj(epot)

        rq = abs(r1 * qk1 + r3 * qk3 + r4 * qk4 + r5 * qk5 + r6 * qk6) / h
        rp = abs(r1 * qk1 + r3 * qk3 + r4 * qk4 + r5 * qk5 + r6 * qk6) / h

        energy2 = T2.getkineticlass() + T2.getpotential_traj()

        if len(np.shape(rq)) > 0:
            rq = max(rq)
            rp = max(rp)
        if abs(rp) <= tol2:
            t = t + h
            q = q + c1 * qk1 + c3 * qk3 + c4 * qk4 + c5 * qk5
            p = p + c1 * pk1 + c3 * pk3 + c4 * pk4 + c5 * pk5
            d = d + c1 * dk1 + c3 * dk3 + c4 * dk4 + c5 * dk5
            s = s + c1 * sk1 + c3 * sk3 + c4 * sk4 + c5 * sk5

            T.setposition_traj(q)
            T.setmomentum_traj(p)
            T.setamplitudes_traj(d * np.exp(s * 1j, dtype=np.complex128))

        h = h * min(max(0.84 * (tol2 / rp)**0.25, 0.1), 4.0)

        if h > hmax:
            h = hmax
        elif h < hmin:
            print("Error: stepsize should be smaller than %e." % hmin)
            break

    print(h, q[0])

    return T
예제 #7
0
def rk_method_4(T, dt, time, nstep):
    geo = initgeom()
    d = T.getd_traj()
    q = T.getposition_traj()
    p = T.getmomentum_traj()
    s = T.getphases_traj()
    a = T.getamplitude_traj()
    ii = np.complex128(0 + 1j)
    nslice = 20

    nstates = T.nstates
    epot = T.PotEn
    ndf = T.ndim
    grad = np.zeros((nstates, ndf))
    nacs = np.zeros((nstates, nstates, ndf))
    masses = T.getmassall_traj()
    for i in range(nstates):
        grad[i, :] = T.getforce_traj(i)
        for j in range(nstates):
            nacs[i, j, :] = T.getcoupling_traj(i, j)
    '''first define the total energy and the d norm'''
    dnorm = 0.00

    for i in range(nstates):
        dnorm = dnorm + np.abs(np.conj(d[i]) * d[i])

    epot_av = 0.000
    for i in range(nstates):
        epot_av += np.abs(np.conj(d[i]) * d[i]) * epot[i]

    ekin = np.sum(0.5 * p**2 / masses)

    tot_en = ekin + epot_av

    print('Total Energy:', tot_en)

    qk1 = dt * p / masses
    pk1 = dt * T.get_traj_force()

    d_dot = np.zeros(nstates, dtype=np.complex128)
    for i in range(nstates):
        for j in range(nstates):
            if i != j:
                nac1d = 0.0
                for n in range(ndf):
                    nac1d += p[n] / masses[n] * T.getcoupling_traj(j, i)[n]
                d_dot[i] = d_dot[i] - np.complex128(nac1d) * d[j] * np.exp(
                    1.0j * (s[j] - s[i]), dtype=np.complex128)
    dk1 = dt * d_dot

    s_dot = np.zeros(nstates, dtype=np.complex128)  # Derivative of S (action)
    tmp = 0.0
    for i in range(ndf):
        tmp += 0.5 * (p[i] / masses[i] * p[i] - q[i] * T.get_traj_force()[i])
    for j in range(nstates):
        s_dot[j] = tmp - (0.5 * np.dot(p, p / masses) + epot[j])

    sk1 = dt * s_dot

    T.setposition_traj(q + 0.5 * qk1)
    T.setmomentum_traj(p + 0.5 * pk1)
    T.setd_traj(d + 0.5 * dk1)
    T.setphases_traj(s + 0.5 * sk1)

    amps = T.getd_traj() * np.exp(1j * T.getphases_traj(), dtype=np.complex128)
    T.setamplitudes_traj(amps)

    epot, derivs = ab.inp_out(nstep, 0, geo, T)
    T.setderivs_traj(derivs)
    T.setpotential_traj(epot)

    p2 = T.getmomentum_traj()
    q2 = T.getposition_traj()
    # a2 = T.getamplitude_traj()
    d2 = T.getd_traj()
    s2 = T.getphases_traj()

    qk2 = dt * p2 / masses
    pk2 = dt * T.get_traj_force()

    d_dot = np.zeros(nstates, dtype=np.complex128)
    for i in range(nstates):
        for j in range(nstates):
            if i != j:
                nac1d = 0.0
                for n in range(ndf):
                    nac1d += p2[n] / masses[n] * T.getcoupling_traj(j, i)[n]
                d_dot[i] = d_dot[i] - np.complex128(nac1d) * d2[j] * np.exp(
                    1.0j * (s2[j] - s2[i]), dtype=np.complex128)
    dk2 = dt * d_dot

    s_dot = np.zeros(nstates, dtype=np.complex128)  # Derivative of S (action)
    tmp = 0.0
    for i in range(ndf):
        tmp += 0.5 * (p2[i] / masses[i] * p2[i] -
                      q2[i] * T.get_traj_force()[i])
    for j in range(nstates):
        s_dot[j] = tmp - (0.5 * np.dot(p2, p2 / masses) + epot[j])

    sk2 = dt * s_dot

    T.setposition_traj(q + 0.5 * qk2)
    T.setmomentum_traj(p + 0.5 * pk2)
    T.setd_traj(d + 0.5 * dk2)
    T.setphases_traj(s + 0.5 * sk2)

    amps = T.getd_traj() * np.exp(1j * T.getphases_traj(), dtype=np.complex128)
    T.setamplitudes_traj(amps)

    epot, derivs = ab.inp_out(nstep, 0, geo, T)
    T.setderivs_traj(derivs)
    T.setpotential_traj(epot)

    p3 = T.getmomentum_traj()
    q3 = T.getposition_traj()
    d3 = T.getd_traj()
    s3 = T.getphases_traj()

    qk3 = dt * p3 / masses
    pk3 = dt * T.get_traj_force()

    d_dot = np.zeros(nstates, dtype=np.complex128)
    for i in range(nstates):
        for j in range(nstates):
            if i != j:
                nac1d = 0.0
                for n in range(ndf):
                    nac1d += p3[n] / masses[n] * T.getcoupling_traj(j, i)[n]
                d_dot[i] = d_dot[i] - np.complex128(nac1d) * d3[j] * np.exp(
                    1.0j * (s3[j] - s3[i]), dtype=np.complex128)
    dk3 = dt * d_dot

    s_dot = np.zeros(nstates, dtype=np.complex128)  # Derivative of S (action)
    tmp = 0.0
    for i in range(ndf):
        tmp += 0.5 * (p3[i] / masses[i] * p3[i] -
                      q3[i] * T.get_traj_force()[i])
    for j in range(nstates):
        s_dot[j] = tmp - (0.5 * np.dot(p3, p3 / masses) + epot[j])

    sk3 = dt * s_dot

    T.setposition_traj(q + qk3)
    T.setmomentum_traj(p + pk3)
    T.setd_traj(d + dk3)
    T.setphases_traj(s + sk3)

    amps = T.getd_traj() * np.exp(1j * T.getphases_traj(), dtype=np.complex128)
    T.setamplitudes_traj(amps)
    epot, derivs = ab.inp_out(nstep, 0, geo, T)
    T.setderivs_traj(derivs)
    T.setpotential_traj(epot)

    p4 = T.getmomentum_traj()
    q4 = T.getposition_traj()
    d4 = T.getd_traj()
    s4 = T.getphases_traj()

    qk4 = dt * p4 / masses
    pk4 = dt * T.get_traj_force()

    d_dot = np.zeros(nstates, dtype=np.complex128)
    for i in range(nstates):
        for j in range(nstates):
            if i != j:
                nac1d = 0.0
                for n in range(ndf):
                    nac1d += p4[n] / masses[n] * T.getcoupling_traj(j, i)[n]
                d_dot[i] = d_dot[i] - np.complex128(nac1d) * d4[j] * np.exp(
                    1.0j * (s4[j] - s4[i]), dtype=np.complex128)
    dk4 = dt * d_dot

    s_dot = np.zeros(nstates, dtype=np.complex128)  # Derivative of S (action)
    tmp = 0.0
    for i in range(ndf):
        tmp += 0.5 * (p4[i] / masses[i] * p4[i] -
                      q4[i] * T.get_traj_force()[i])
    for j in range(nstates):
        s_dot[j] = tmp - (0.5 * np.dot(p4, p4 / masses) + epot[j])

    sk4 = dt * s_dot

    T.setposition_traj(q + (qk1 + 2.000 * qk2 + 2.00 * qk3 + qk4) / 6.00)
    T.setmomentum_traj(p + (pk1 + 2.000 * pk2 + 2.00 * pk3 + pk4) / 6.00)
    T.setd_traj(d + (dk1 + 2.000 * dk2 + 2.00 * dk3 + dk4) / 6.00)
    T.setphases_traj(s + (sk1 + 2.000 * sk2 + 2.00 * sk3 + sk4) / 6.00)

    amps = T.getd_traj() * np.exp(1j * T.getphases_traj(), dtype=np.complex128)
    T.setamplitudes_traj(amps)

    return T
예제 #8
0
T1 = initialize_traj.trajectory(geo.natoms, 3, dyn.nstates)
# qin, pin = Wigner_dist.WignerSampling()
q = np.zeros(geo.ndf)
p = np.zeros_like(q)
with open('initialqp_2.dat', 'r') as f:
    f.readline()
    for i in range(geo.ndf):
        N, M = f.readline().strip().split()
        q[i] = np.double(float(N.replace('D', 'E')))
        p[i] = np.double(float(M.replace('D', 'E')))

T1.setposition_traj(q)
T1.setmomentum_traj(p)

pec, der, cis, configs = abinitio.inp_out(0, 0, geo, T1)  # First ab-initio run

T1.setpotential_traj(pec)  # taking V(R) from ab-initio
T1.setderivs_traj(
    der
)  # derivatives matrix not mass-weighted (possibly change that), diagonals are forces and off-d are nacmes
T1.setmass_traj(
    geo.masses)  # mass of every atom in a.u (the dimmension is natoms/nparts)
T1.setmassall_traj(
    geo.massrk
)  # mass in every degree of freedom (careful to use it, it can triple the division/multiplication easily)
T1.setcivecs(cis)
T1.setconfigs(configs)
amps = np.zeros(T1.nstates, dtype=np.complex128)
amps[dyn.inipes - 1] = np.complex128(
    1.00 + 0.00j