Пример #1
0
class GaussianMPATest(bettertest.TestCase):
    """Mulliken Population Analysis test"""
    def setUp(self):
        self.data, self.logfile = getfile(Gaussian, "basicGaussian03", "dvb_sp.out")
        self.analysis = MPA(self.data)
        self.analysis.logger.setLevel(0)
        self.analysis.calculate()
    def testsum(self):
        """Do the Mulliken charges sum up to the total formal charge?"""
        formalcharge = sum(self.data.atomnos) - self.data.charge
        totalpopulation = sum(self.analysis.fragcharges)
        self.assertInside(totalpopulation, formalcharge, 0.001)
Пример #2
0
class GaussianMPATest(bettertest.TestCase):
    """Mulliken Population Analysis test"""
    def setUp(self):
        self.data, self.logfile = getfile(Gaussian, "basicGaussian03",
                                          "dvb_sp.out")
        self.analysis = MPA(self.data)
        self.analysis.logger.setLevel(0)
        self.analysis.calculate()

    def testsum(self):
        """Do the Mulliken charges sum up to the total formal charge?"""
        formalcharge = sum(self.data.atomnos) - self.data.charge
        totalpopulation = sum(self.analysis.fragcharges)
        self.assertInside(totalpopulation, formalcharge, 0.001)
Пример #3
0
class GaussianMPATest(unittest.TestCase):
    """Mulliken Population Analysis test"""

    def setUp(self):
        self.data, self.logfile = getdatafile(Gaussian, "basicGaussian03", ["dvb_un_sp.out"])
        self.analysis = MPA(self.data)
        self.analysis.logger.setLevel(0)
        self.analysis.calculate()

    def testsumcharges(self):
        """Do the Mulliken charges sum up to the total formal charge?"""
        formalcharge = sum(self.data.atomnos) - self.data.charge
        totalpopulation = sum(self.analysis.fragcharges)
        self.assertAlmostEqual(totalpopulation, formalcharge, delta=1.0e-3)

    def testsumspins(self):
        """Do the Mulliken spins sum up to the total formal spin?"""
        formalspin = self.data.homos[0] - self.data.homos[1]
        totalspin = sum(self.analysis.fragspins)
        self.assertAlmostEqual(totalspin, formalspin, delta=1.0e-3)
Пример #4
0
class GaussianMPATest(unittest.TestCase):
    """Mulliken Population Analysis test"""
    def setUp(self):
        self.data, self.logfile = getdatafile(Gaussian, "basicGaussian03",
                                              ["dvb_un_sp.out"])
        self.analysis = MPA(self.data)
        self.analysis.logger.setLevel(0)
        self.analysis.calculate()

    def testsumcharges(self):
        """Do the Mulliken charges sum up to the total formal charge?"""
        formalcharge = sum(self.data.atomnos) - self.data.charge
        totalpopulation = sum(self.analysis.fragcharges)
        self.assertAlmostEqual(totalpopulation, formalcharge, delta=1.0e-3)

    def testsumspins(self):
        """Do the Mulliken spins sum up to the total formal spin?"""
        formalspin = self.data.homos[0] - self.data.homos[1]
        totalspin = sum(self.analysis.fragspins)
        self.assertAlmostEqual(totalspin, formalspin, delta=1.0e-3)
Пример #5
0
 def setUp(self):
     self.data, self.logfile = getdatafile(Gaussian, "basicGaussian03", ["dvb_un_sp.out"])
     self.analysis = MPA(self.data)
     self.analysis.logger.setLevel(0)
     self.analysis.calculate()
Пример #6
0
 def setUp(self):
     self.data, self.logfile = getdatafile(Gaussian, "basicGaussian03",
                                           ["dvb_un_sp.out"])
     self.analysis = MPA(self.data)
     self.analysis.logger.setLevel(0)
     self.analysis.calculate()
Пример #7
0
def WriteCovalencyReport(fileName,
                         calculationType='LPA',
                         element='Fe',
                         homoLumoRange=15,
                         printThreshold=5):
    filename = fileName
    data = ccread(filename)
    if calculationType == 'LPA':
        m = LPA(data)  #This is for Lowdin Analysis.
    elif calculationType == 'MPA':
        m = MPA(data)  #Mulliken Population Analysis
    elif calculationType == 'csquared':
        m = CSPA(data)  #C-squared Analysis
    else:
        Exception('Calculation type not supported.')
    print(element)
    #Section: Regular Expression Variables
    regexZ2 = re.compile(element + '.*DZ2')
    regexXZ = re.compile(element + '.*DXZ')
    regexYZ = re.compile(element + '.*DYZ')
    regexX2Y2 = re.compile(element + '.*DX2Y2')
    regexXY = re.compile(element + '.*DXY')
    regexPX = re.compile(element + '.*PX')
    regexPY = re.compile(element + '.*PY')
    regexPZ = re.compile(element + '.*PZ')
    regexS = re.compile(element + '.*S')

    #Section: Finding Atomic Orbitals
    Z2, matches1 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexZ2, string)])
    XZ, matches2 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexXZ, string)])
    YZ, matches3 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexYZ, string)])
    X2Y2, matches4 = zip(*[(idx, string)
                           for idx, string in enumerate(data.aonames)
                           if re.match(regexX2Y2, string)])
    XY, matches5 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexXY, string)])
    PX, matches6 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexPX, string)])
    PY, matches7 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexPY, string)])
    PZ, matches8 = zip(*[(idx, string)
                         for idx, string in enumerate(data.aonames)
                         if re.match(regexPZ, string)])
    S, matches9 = zip(*[(idx, string)
                        for idx, string in enumerate(data.aonames)
                        if re.match(regexS, string)])
    #OxoOrbs, matches = zip(*[(idx, string) for idx, string in enumerate(data.aonames) if  re.match('O', string)])#For oxo orbitals
    otherOrbs, matches = zip(*[(idx, string)
                               for idx, string in enumerate(data.aonames)
                               if not (re.match(element, string))])

    #Section: Population Calculation
    m.calculate([S, Z2, XZ, YZ, X2Y2, XY, PX, PY, PZ, otherOrbs])
    Spin = 0  #Alpha orbitals
    h**o = data.homos[Spin]
    character = m.fragresults[Spin].T[:] * 100
    indices = np.arange(0, len(data.aonames), 1)
    orbital = [data.aonames[i][-1] for i in np.argmax(data.mocoeffs[Spin], 1)]
    aos = [
        re.split(('_'), str(data.aonames[i]))
        for i in np.argmax(data.mocoeffs[Spin]**2, 1)
    ]
    AO = np.array(aos).T
    dominantAtom = np.array([
        data.aonames[i].split('_')[0]
        for i in np.argmax(data.mocoeffs[Spin], 1)
    ]).T
    dominantAO = np.array([
        data.aonames[i].split('_')[1]
        for i in np.argmax(data.mocoeffs[Spin], 1)
    ]).T
    energies = data.moenergies[Spin]
    energyArray = np.column_stack(
        (indices, energies, dominantAtom, dominantAO))

    #Section: Building Alpha Dataframe.
    alp = pd.DataFrame(energyArray,
                       columns=[
                           'Orbital Number', 'Energy (eV)', 'Dominant Atom',
                           'Dominant AO'
                       ])
    h**o = data.homos[0]
    alp.loc[0:h**o + 1, 'Occupancy'] = 1
    alp.loc[h**o + 1:, 'Occupancy'] = 0
    alp['Spin'] = 'Alpha'
    alp[element + '_S'] = character[0]
    alp[element + '_DZ2'] = character[1]
    alp[element + '_DXZ'] = character[2]
    alp[element + '_DYZ'] = character[3]
    alp[element + '_DX2Y2'] = character[4]
    alp[element + '_DXY'] = character[5]
    alp[element + '_PX'] = character[6]
    alp[element + '_PY'] = character[7]
    alp[element + '_PZ'] = character[8]
    alp['Other_Character'] = character[9]
    alp['Total_Metal_Character'] = np.sum(character[0:6], axis=0)
    alp['Total_p_Character'] = np.sum(character[6:8], axis=0)
    outDF = alp.loc[(alp.index > h**o - homoLumoRange)
                    & (alp.index < h**o + homoLumoRange) &
                    (alp['Total_Metal_Character'] > printThreshold)].round(3)

    #Section: Final Alpha Output
    outDFalp = outDF[[
        'Spin', 'Orbital Number', 'Energy (eV)', 'Occupancy', element + '_S',
        element + '_DXY', element + '_DXZ', element + '_DYZ',
        element + '_DX2Y2', element + '_DZ2', element + '_PX', element + '_PY',
        element + '_PZ', 'Other_Character', 'Total_Metal_Character',
        'Total_p_Character'
    ]]
    SumAlpha = outDFalp[[
        'Spin', element + '_S', element + '_DXY', element + '_DXZ',
        element + '_DYZ', element + '_DX2Y2', element + '_DZ2',
        element + '_PX', element + '_PY', element + '_PZ', 'Other_Character',
        'Total_Metal_Character', 'Total_p_Character'
    ]].loc[outDFalp['Occupancy'] == 0].sum(axis=0)

    Spin = 1  #beta orbitals
    h**o = data.homos[Spin]
    character = m.fragresults[Spin].T[:] * 100
    indices = np.arange(0, len(data.aonames), 1)
    orbital = [data.aonames[i][-1] for i in np.argmax(data.mocoeffs[Spin], 1)]
    aos = [
        re.split(('_'), str(data.aonames[i]))
        for i in np.argmax(data.mocoeffs[Spin]**2, 1)
    ]
    AO = np.array(aos).T
    dominantAtom = np.array([
        data.aonames[i].split('_')[0]
        for i in np.argmax(data.mocoeffs[Spin], 1)
    ]).T
    dominantAO = np.array([
        data.aonames[i].split('_')[1]
        for i in np.argmax(data.mocoeffs[Spin], 1)
    ]).T
    energies = data.moenergies[Spin]
    energyArray = np.column_stack(
        (indices, energies, dominantAtom, dominantAO))

    #Section: Building Beta Dataframe.
    bet = pd.DataFrame(energyArray,
                       columns=[
                           'Orbital Number', 'Energy (eV)', 'Dominant Atom',
                           'Dominant AO'
                       ])
    bet.loc[0:h**o + 1, 'Occupancy'] = 1
    bet.loc[h**o + 1:, 'Occupancy'] = 0
    bet['Spin'] = "Beta"
    bet[element + '_S'] = character[0]
    bet[element + '_DZ2'] = character[1]
    bet[element + '_DXZ'] = character[2]
    bet[element + '_DYZ'] = character[3]
    bet[element + '_DX2Y2'] = character[4]
    bet[element + '_DXY'] = character[5]
    bet[element + '_PX'] = character[6]
    bet[element + '_PY'] = character[7]
    bet[element + '_PZ'] = character[8]
    #bet['O_Character']=character[8]
    bet['Other_Character'] = character[9]
    bet['Total_Metal_Character'] = np.sum(character[0:6], axis=0)
    bet['Total_p_Character'] = np.sum(character[6:8], axis=0)
    outDF = bet.loc[(bet.index > h**o - homoLumoRange)
                    & (bet.index < h**o + homoLumoRange) &
                    (bet['Total_Metal_Character'] > printThreshold)].round(3)
    #Section: Final Beta Output
    outDFbet = outDF[[
        'Spin', 'Orbital Number', 'Energy (eV)', 'Occupancy', element + '_S',
        element + '_DXY', element + '_DXZ', element + '_DYZ',
        element + '_DX2Y2', element + '_DZ2', element + '_PX', element + '_PY',
        element + '_PZ', 'Other_Character', 'Total_Metal_Character',
        'Total_p_Character'
    ]]
    SumBeta = outDFbet[[
        'Spin', element + '_S', element + '_DXY', element + '_DXZ',
        element + '_DYZ', element + '_DX2Y2', element + '_DZ2',
        element + '_PX', element + '_PY', element + '_PZ', 'Other_Character',
        'Total_Metal_Character', 'Total_p_Character'
    ]].loc[outDFbet['Occupancy'] == 0].sum(axis=0)

    #Section: Write Final Ouput File
    TotalDF = outDFalp.append(outDFbet)
    SumTotal = (SumAlpha + SumBeta)[1:]
    f = open(fileName.split('.')[0] + '.' + calculationType + '.report', 'w')
    f.write('Alpha Hole Character Sum:\n' + str(SumAlpha[1:])[:-13] +
            '\nBeta Hole Character Sum:\n' + str(SumBeta[1:])[:-13] +
            '\nTotal Hole Sum:\n' + str(SumTotal[1:])[:-13])
    TotalDF.to_csv(f, sep='\t')
    return (TotalDF)
    f.close()