Mailman 3 October 2016 - Kwant-discuss

barrier transport
by kamal ghosh Oct. 23, 2016

Oct. 23, 2016

Dear All, I have tried with a kwant code to see the transmission through barrier imposed by a qwr between the source and drain. The wire is constructed and its bands are formed successfully; but the plot of transmission through the constant barrier is not obtained. In the long run, I would like to use the standardised varying barrier potential and see the transport. Please help me overcome this difficulty so that I may proceed further. Kind regards. K.K.Ghosh PS. code is attached herewith.

2 1

P-wave superconductivity in Kwant
by antoniolrm＠usp.br Oct. 23, 2016

Oct. 23, 2016

Hello everyone, I am trying to implement Kitaev model in Kwant via the Hamiltonian: H = -\mu \sum_{n=1}^{N} c_{n}^{\dagger}c_{n} - t \sum_{n=1}^{N-1}(c_{n+1}^{\dagger}c_n + h.c.) + \Delta \sum_{n=1}^{N-1}(c_{n}c_{n+1} + h.c.) The first two terms are simple, but I am stucked in the superconducting p-wave pairing term. I've tried to implement it by doing sys[((lat_e(x+1, 0), lat_h(x, 0)) for x in range(0,L-1))] = Delta which seems somehow correct once I got the zero modes at the edges for the Kitaev limit \mu=0 and t=\Delta. But once I vary \mu I lost one of the edge states, and also it does not close the gap, which makes me belive that I am dealing with a trivial phase. Does anyone ever done it? Can you see something wrong with my approach? Cheers, Antônio. The complete code is the following: ==================================================== from cmath import exp import numpy as np import kwant # For eigenvalue computation import scipy.sparse.linalg as sla # For plotting from matplotlib import pyplot def make_system(a=1.0, W=1, L=50, mu=1.5, Delta=1.0, t=1.0): # Start with an empty tight-binding system and two square lattices, # corresponding to electron and hole degree of freedom lat_e = kwant.lattice.square(a=1.0, name='e') lat_h = kwant.lattice.square(a=1.0, name='h') sys = kwant.Builder() #### Define the scattering region. #### sys[(lat_e(x, y) for x in range(L) for y in range(W))] = - mu sys[(lat_h(x, y) for x in range(L) for y in range(W))] = mu # hoppings for both electrons and holes sys[lat_e.neighbors()] = -t sys[lat_h.neighbors()] = t # Superconducting order parameter enters as hopping between # electrons and holes sys[((lat_e(x+1, y), lat_h(x, y)) for x in range(0,L-1) for y in range(W))] = Delta #### Define the leads. #### # Symmetry for the left leads. sym_left = kwant.TranslationalSymmetry((-a, 0)) # left electron lead lead0 = kwant.Builder(sym_left) lead0[(lat_e(0, j) for j in range(W))] = - mu lead0[lat_e.neighbors()] = t lead0[lat_h.neighbors()] = -t # left hole lead lead1 = kwant.Builder(sym_left) lead1[(lat_h(0, j) for j in range(W))] = mu lead1[lat_e.neighbors()] = t lead1[lat_h.neighbors()] = -t # Then the lead to the right # this one is superconducting and thus is comprised of electrons # AND holes sym_right = kwant.TranslationalSymmetry((a, 0)) lead2 = kwant.Builder(sym_right) lead2 += lead0 lead2 += lead1 lead2[((lat_e(1, j), lat_h(0, j)) for j in range(W))] = Delta #### Attach the leads and return the system. #### sys.attach_lead(lead2) sys.attach_lead(lead2.reversed()) return sys def plot_wave_function(sys, L=100): # Calculate the wave functions in the system. ham_mat = sys.hamiltonian_submatrix(sparse=True) evecs = sla.eigsh(ham_mat, k=50, which='SM')[1] # Plot the probability density of the 1st eigenmode. kwant.plotter.map(sys, np.abs(evecs[:, 0])**2, colorbar=False, a=1.0, oversampling=2) def main(): sys = make_system() # Check that the system looks as intended. kwant.plot(sys) # Finalize the system. sys = sys.finalized() plot_wave_function(sys) if __name__ == '__main__': main() =================================================================

1 0

about adding B field
by Lee Kin Oct. 20, 2016

Oct. 20, 2016

Hi there, I start to use kwant not long ago. I am still following the instruction to learn the basic usage of the code. I have a simple code in hand to handle the square lattices with some given potential of certain form. I am thinking to add the constant magnetic field along z direction. In my text, it is said that such field could be described by vector potential such that the potential could be modified as alpha*(x**2 + y**2) in which alpha is proportional to the magnitude of the field. I add this to each lattice site but then it diverges in solution. I saw other material online and it said the magnetic field could be added by modifying the hopping phase. I don't quite follow that. Any suggestion is welcomed. Thanks.

2 1

Re: [Kwant] a step tripping up Kwant
by Maurer, Leon Oct. 17, 2016

Oct. 17, 2016

Hi Abbout, Thanks for the help. -Leon From: Abbout Adel <abbout.adel(a)gmail.com<mailto:abbout.adel@gmail.com>> Date: Monday, October 17, 2016 at 1:59 PM To: "Leon Maurer (lmaurer)" <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> Cc: "kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>" <kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>> Subject: [EXTERNAL] Re: [Kwant] a step tripping up Kwant Hi Leon, The upper and bottom limits of the conduction band are obtained from the relation of dispersion. E=V-2t *cos(k) (# in your case V=2t for left lead and V0 for the right lead ). For more details, you can look for example to : http://www-personal.umich.edu/~sunkai/teaching/Fall_2014/Chapter6.pdf Regards, Adel On Mon, Oct 17, 2016 at 10:34 PM, Maurer, Leon <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> wrote: Hi Abbout, Is the upper bound on the energies in the leads documented somewhere? I guess it’s implicit in Sec. 2.4 of the tutorial, and now that you mention it, it makes sense given the periodic lattice. -Leon From: Abbout Adel <abbout.adel(a)gmail.com<mailto:abbout.adel@gmail.com>> Date: Monday, October 17, 2016 at 11:58 AM To: "Leon Maurer (lmaurer)" <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> Cc: "kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>" <kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>> Subject: [EXTERNAL] Re: [Kwant] a step tripping up Kwant Dear Leon, the value of the parameter 't' in your program is around 16. this means that the conduction band for the left lead is band_l=[0, 4 t ]=[0,64] and the conduction band for the right lead is band_r=[V0, V0+4 t]=[100, 164] as you can notice there is no energy which conducts in both leads. In order to have a non zero transmission you need to use a value of V0<4 t. The result you are calling "exact" is valid for a continuous model: on a lattice, the dispersion relation is not quadratic. To compare, your result with the continuous limit, you need to choose V0<< 2t. For a "non uniform finite differences", you can look at the article [1] Hope that this helps. Adel [1]: http://scitation.aip.org/content/aip/journal/jap/68/8/10.1063/1.346245 On Mon, Oct 17, 2016 at 7:10 PM, Maurer, Leon <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> wrote: Hello everyone, I’ve been playing around with Kwant and come across some situations where the transmission between two leads is identically equal to zero when I wouldn’t expect that result. I’ve come up with a simple working example: a 1D step function with a step height that’s large relative to the lattice spacing (code below), altho this problem seems to sometimes crop up in other, somewhat less extreme situations. I understand that the numerical result should become less accurate as the step-height-to-lattice-spacing ratio increases, but why does the transmission become identically equal to zero at some point? Are there well-defined conditions for when this happens? Is there some way to know that the transmission is zero because of numerical issues rather than the underlying physics? (Ultimately, I’m interested in modeling some systems where the potential mostly varies gradually but has a few small regions with abrupt changes in potential. Moving to a finer mesh (smaller lattice constant) everywhere is cost-prohibitive. Having some tool to easily refine the mesh in a region would be very useful.) Thanks. -Leon (Below code taken from jupyter notebook.) # In[1]: get_ipython().magic('load_ext autoreload') get_ipython().magic('autoreload 2') from numpy import * import matplotlib.pyplot as plt get_ipython().magic('matplotlib inline') import tqdm import kwant # In[2]: m0 = 9.10938215e-31 # Electron mass, [kg] hbar = 1.054571726e-34 # hbar in [J] [s] q = 1.602176565e-19 # Elementary charge, [C] mt = 0.19 ml = 0.92 m = mt*m0 # In[3]: V0 = 100 # step height x = linspace(0,100,30) # thirty grid points U = zeros_like(x) U[len(x)//2:] = V0 plt.plot(x,U) # In[4]: a = x[1]-x[0] # grid spacing [nm] t = hbar**2/(2.*m*(a*1e-9)**2)/q*1e3 #hopping parameter [meV] lat = kwant.lattice.chain(a) # Set up the transport simulation on a 1D latice sys = kwant.Builder() # initialize the transport simulation for i in range(len(U)): # populate based on the potential landscape sys[lat(i)]=U[i]+2*t sys[lat.neighbors()] = -t # set the finite-difference hopping parameters leftLead = kwant.Builder(kwant.TranslationalSymmetry((-a,))) # the lead to the left leftLead[lat(0)] = 2*t + U[0] leftLead[lat.neighbors()] = -t sys.attach_lead(leftLead) # attach it rightLead = kwant.Builder(kwant.TranslationalSymmetry((a,))) # the lead to the right rightLead[lat(0)] = 2*t + U[-1] rightLead[lat.neighbors()] = -t sys.attach_lead(rightLead) # attach it sys = sys.finalized() # In[5]: def plot_conductance(sys, energies): # Compute transmission numerically data = [] for energy in tqdm.tqdm(energies,leave=True): smatrix = kwant.smatrix(sys, energy) data.append(smatrix.transmission(1, 0)) # Compute exact conductance k1 = sqrt(2*m*energies/hbar**2) k2 = sqrt(2*m*(energies - V0)/hbar**2) T = 4*k1*k2/(k1+k2)**2 T[energies <= V0] = 0 plt.figure() plt.plot(energies, data, energies, T) plt.legend(('numerical','exact'), loc=4) plt.xlabel("energy [V0]") plt.ylabel("Transmission") plt.show() return data, T # In[6]: stepNumerical, stepExact = plot_conductance(sys,linspace(1e-9,2*V0,201)) # In[7]: print(stepNumerical) -- Abbout Adel -- Abbout Adel

1 0

Re: [Kwant] a step tripping up Kwant
by Maurer, Leon Oct. 17, 2016

Oct. 17, 2016

Hi Abbout, Is the upper bound on the energies in the leads documented somewhere? I guess it’s implicit in Sec. 2.4 of the tutorial, and now that you mention it, it makes sense given the periodic lattice. -Leon From: Abbout Adel <abbout.adel(a)gmail.com<mailto:abbout.adel@gmail.com>> Date: Monday, October 17, 2016 at 11:58 AM To: "Leon Maurer (lmaurer)" <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> Cc: "kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>" <kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>> Subject: [EXTERNAL] Re: [Kwant] a step tripping up Kwant Dear Leon, the value of the parameter 't' in your program is around 16. this means that the conduction band for the left lead is band_l=[0, 4 t ]=[0,64] and the conduction band for the right lead is band_r=[V0, V0+4 t]=[100, 164] as you can notice there is no energy which conducts in both leads. In order to have a non zero transmission you need to use a value of V0<4 t. The result you are calling "exact" is valid for a continuous model: on a lattice, the dispersion relation is not quadratic. To compare, your result with the continuous limit, you need to choose V0<< 2t. For a "non uniform finite differences", you can look at the article [1] Hope that this helps. Adel [1]: http://scitation.aip.org/content/aip/journal/jap/68/8/10.1063/1.346245 On Mon, Oct 17, 2016 at 7:10 PM, Maurer, Leon <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> wrote: Hello everyone, I’ve been playing around with Kwant and come across some situations where the transmission between two leads is identically equal to zero when I wouldn’t expect that result. I’ve come up with a simple working example: a 1D step function with a step height that’s large relative to the lattice spacing (code below), altho this problem seems to sometimes crop up in other, somewhat less extreme situations. I understand that the numerical result should become less accurate as the step-height-to-lattice-spacing ratio increases, but why does the transmission become identically equal to zero at some point? Are there well-defined conditions for when this happens? Is there some way to know that the transmission is zero because of numerical issues rather than the underlying physics? (Ultimately, I’m interested in modeling some systems where the potential mostly varies gradually but has a few small regions with abrupt changes in potential. Moving to a finer mesh (smaller lattice constant) everywhere is cost-prohibitive. Having some tool to easily refine the mesh in a region would be very useful.) Thanks. -Leon (Below code taken from jupyter notebook.) # In[1]: get_ipython().magic('load_ext autoreload') get_ipython().magic('autoreload 2') from numpy import * import matplotlib.pyplot as plt get_ipython().magic('matplotlib inline') import tqdm import kwant # In[2]: m0 = 9.10938215e-31 # Electron mass, [kg] hbar = 1.054571726e-34 # hbar in [J] [s] q = 1.602176565e-19 # Elementary charge, [C] mt = 0.19 ml = 0.92 m = mt*m0 # In[3]: V0 = 100 # step height x = linspace(0,100,30) # thirty grid points U = zeros_like(x) U[len(x)//2:] = V0 plt.plot(x,U) # In[4]: a = x[1]-x[0] # grid spacing [nm] t = hbar**2/(2.*m*(a*1e-9)**2)/q*1e3 #hopping parameter [meV] lat = kwant.lattice.chain(a) # Set up the transport simulation on a 1D latice sys = kwant.Builder() # initialize the transport simulation for i in range(len(U)): # populate based on the potential landscape sys[lat(i)]=U[i]+2*t sys[lat.neighbors()] = -t # set the finite-difference hopping parameters leftLead = kwant.Builder(kwant.TranslationalSymmetry((-a,))) # the lead to the left leftLead[lat(0)] = 2*t + U[0] leftLead[lat.neighbors()] = -t sys.attach_lead(leftLead) # attach it rightLead = kwant.Builder(kwant.TranslationalSymmetry((a,))) # the lead to the right rightLead[lat(0)] = 2*t + U[-1] rightLead[lat.neighbors()] = -t sys.attach_lead(rightLead) # attach it sys = sys.finalized() # In[5]: def plot_conductance(sys, energies): # Compute transmission numerically data = [] for energy in tqdm.tqdm(energies,leave=True): smatrix = kwant.smatrix(sys, energy) data.append(smatrix.transmission(1, 0)) # Compute exact conductance k1 = sqrt(2*m*energies/hbar**2) k2 = sqrt(2*m*(energies - V0)/hbar**2) T = 4*k1*k2/(k1+k2)**2 T[energies <= V0] = 0 plt.figure() plt.plot(energies, data, energies, T) plt.legend(('numerical','exact'), loc=4) plt.xlabel("energy [V0]") plt.ylabel("Transmission") plt.show() return data, T # In[6]: stepNumerical, stepExact = plot_conductance(sys,linspace(1e-9,2*V0,201)) # In[7]: print(stepNumerical) -- Abbout Adel

2 1

a step tripping up Kwant
by Maurer, Leon Oct. 17, 2016

Oct. 17, 2016

Hello everyone, I’ve been playing around with Kwant and come across some situations where the transmission between two leads is identically equal to zero when I wouldn’t expect that result. I’ve come up with a simple working example: a 1D step function with a step height that’s large relative to the lattice spacing (code below), altho this problem seems to sometimes crop up in other, somewhat less extreme situations. I understand that the numerical result should become less accurate as the step-height-to-lattice-spacing ratio increases, but why does the transmission become identically equal to zero at some point? Are there well-defined conditions for when this happens? Is there some way to know that the transmission is zero because of numerical issues rather than the underlying physics? (Ultimately, I’m interested in modeling some systems where the potential mostly varies gradually but has a few small regions with abrupt changes in potential. Moving to a finer mesh (smaller lattice constant) everywhere is cost-prohibitive. Having some tool to easily refine the mesh in a region would be very useful.) Thanks. -Leon (Below code taken from jupyter notebook.) # In[1]: get_ipython().magic('load_ext autoreload') get_ipython().magic('autoreload 2') from numpy import * import matplotlib.pyplot as plt get_ipython().magic('matplotlib inline') import tqdm import kwant # In[2]: m0 = 9.10938215e-31 # Electron mass, [kg] hbar = 1.054571726e-34 # hbar in [J] [s] q = 1.602176565e-19 # Elementary charge, [C] mt = 0.19 ml = 0.92 m = mt*m0 # In[3]: V0 = 100 # step height x = linspace(0,100,30) # thirty grid points U = zeros_like(x) U[len(x)//2:] = V0 plt.plot(x,U) # In[4]: a = x[1]-x[0] # grid spacing [nm] t = hbar**2/(2.*m*(a*1e-9)**2)/q*1e3 #hopping parameter [meV] lat = kwant.lattice.chain(a) # Set up the transport simulation on a 1D latice sys = kwant.Builder() # initialize the transport simulation for i in range(len(U)): # populate based on the potential landscape sys[lat(i)]=U[i]+2*t sys[lat.neighbors()] = -t # set the finite-difference hopping parameters leftLead = kwant.Builder(kwant.TranslationalSymmetry((-a,))) # the lead to the left leftLead[lat(0)] = 2*t + U[0] leftLead[lat.neighbors()] = -t sys.attach_lead(leftLead) # attach it rightLead = kwant.Builder(kwant.TranslationalSymmetry((a,))) # the lead to the right rightLead[lat(0)] = 2*t + U[-1] rightLead[lat.neighbors()] = -t sys.attach_lead(rightLead) # attach it sys = sys.finalized() # In[5]: def plot_conductance(sys, energies): # Compute transmission numerically data = [] for energy in tqdm.tqdm(energies,leave=True): smatrix = kwant.smatrix(sys, energy) data.append(smatrix.transmission(1, 0)) # Compute exact conductance k1 = sqrt(2*m*energies/hbar**2) k2 = sqrt(2*m*(energies - V0)/hbar**2) T = 4*k1*k2/(k1+k2)**2 T[energies <= V0] = 0 plt.figure() plt.plot(energies, data, energies, T) plt.legend(('numerical','exact'), loc=4) plt.xlabel("energy [V0]") plt.ylabel("Transmission") plt.show() return data, T # In[6]: stepNumerical, stepExact = plot_conductance(sys,linspace(1e-9,2*V0,201)) # In[7]: print(stepNumerical)

2 1

Kwant Open course
by T.C. Wu Oct. 14, 2016

Oct. 14, 2016

Dear Kwant developers, I recently visited http://topocondmat.org/w7_defects/ti_majoranas.html in the Kwant open course. However, I got a bit confused about what libraries / function should I install or introduce in order to run the code in the input? It seems that some functions / classes /libraries are used without specification. Thanks. Best, Johnny

1 0

Re: [Kwant] more MUMPS problems
by Maurer, Leon Oct. 13, 2016

Oct. 13, 2016

It turns out that the solution was quite simple: I was using Intel’s compiler, so I need to change “gfortran” to “ifcore” in build.conf. That combined with Michael’s advice got things working. -Leon From: Kwant-discuss <kwant-discuss-bounces(a)kwant-project.org<mailto:kwant-discuss-bounces@kwant-project.org>> on behalf of "Leon Maurer (lmaurer)" <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>> Date: Friday, October 7, 2016 at 9:10 AM To: "kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>" <kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>> Subject: Re: [Kwant] [EXTERNAL] Re: more MUMPS problems Hello Michael, I’ve tried adding the new libraries, and now I have a slightly different error: ImportError: /ascldap/users/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/linalg/_mumps.cpython-35m-x86_64-linux-gnu.so: undefined symbol: for_close Previously, the undefined symbol was “__kmpc_ok_to_fork”. I guess that’s progress! Any more advice? -Leon From: Kwant-discuss <kwant-discuss-bounces(a)kwant-project.org<mailto:kwant-discuss-bounces@kwant-project.org>> on behalf of Michael Wimmer <M.T.Wimmer(a)tudelft.nl<mailto:M.T.Wimmer@tudelft.nl>> Date: Friday, October 7, 2016 at 12:01 AM To: "kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>" <kwant-discuss(a)kwant-project.org<mailto:kwant-discuss@kwant-project.org>> Subject: [EXTERNAL] Re: [Kwant] more MUMPS problems Dear Leon, Try the following: In the libraries line with the mkl limbs, add iomp5 pthread The errors that you see refer to that the openmp library is missing. Alternatively you could try to use anaconda (search previous emails on this list for details) Best, Michael Von meinem iPad gesendet Am 06.10.2016 um 23:12 schrieb Maurer, Leon <lmaurer(a)sandia.gov<mailto:lmaurer@sandia.gov>>: Hello everyone, I’m in the unfortunate position of having to compile Kwant (and all of its dependencies) from scratch in my home folder, and I’m having trouble getting Kwant to play well with MUMPS. After quite a bit of futzing, I’ve got everything to compile happily (the Kwant build recognizes the "User-configured MUMPS”), but Kwant still won’t work with MUMPS at runtime: ############################# In [1]: import kwant /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/solvers/default.py:18: RuntimeWarning: MUMPS is not available, SciPy built-in solver will be used as a fallback. Performance can be very poor in this case. "Performance can be very poor in this case.", RuntimeWarning) In [2]: from kwant.linalg import mumps --------------------------------------------------------------------------- ImportError Traceback (most recent call last) <ipython-input-2-2962310cb52c> in <module>() ----> 1 from kwant.linalg import mumps /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/linalg/mumps.py in <module>() 15 import numpy as np 16 import scipy.sparse ---> 17 from . import _mumps 18 from .fortran_helpers import prepare_for_fortran 19 ImportError: /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/linalg/_mumps.cpython-35m-x86_64-linux-gnu.so: undefined symbol: __kmpc_ok_to_fork ############################# For what it’s worth, I’ve put my entire build process below. It’s probably too much information, but the short version is that I’m using Intel’s complier on Linux. The Kwant-specific stuff is near the end. The earlier stuff is compiling Python, SCOTCH, METIS, and MUMPS. Any advice? Thanks. -Leon # folders to hold everything cd mkdir local mkdir build cd build # install python and packages wget https://www.python.org/ftp/python/3.5.2/Python-3.5.2.tar.xz tar xf Python-3.5.2.tar.xz cd Python-3.5.2 ./configure --prefix=~/local make make install export PATH=~/local/bin:$PATH cd .. # install important python packages pip3 install scipy tinyarray matplotlib cython # get SCOTCH, METIS, MUMPS wget http://gforge.inria.fr/frs/download.php/file/34618/scotch_6.0.4.tar.gz wget http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/metis-5.1.0.tar.gz wget http://graal.ens-lyon.fr/MUMPS/MUMPS_5.0.2.tar.gz tar xf scotch_6.0.4.tar.gz tar xf metis-5.1.0.tar.gz tar xf MUMPS_5.0.2.tar.gz # install METIS cd metis-5.1.0 make config shared=1 prefix=~/local/ make make install cd .. # install SCOTCH cd scotch_6.0.4/src/ cp Make.inc/Makefile.inc.x86-64_pc_linux2.icc Makefile.inc # need to tweak the makefile nano Makefile.inc Add "-fPIC” to the line that starts with "CFLAGS" make make esmumps make install prefix=~/local # make install won't install esmumps... copy files manually cd ../.. cp scotch_6.0.4/lib/libesmumps.a ~/local/lib/ cp scotch_6.0.4/include/esmumps.h ~/local/include/ # install MUMPS cd MUMPS_5.0.2 cp Make.inc/Makefile.INTEL.SEQ Makefile.inc # need to make a bunch of changes to Makefile.inc nano Makefile.inc 1) change "#SCOTCHDIR = ${HOME}/scotch_6.0" to "SCOTCHDIR = ${HOME}/local/lib" 2) uncomment the line "LSCOTCH = -L$(SCOTCHDIR)/lib -lesmumps -lscotch -lscotcherr" 3) change the line "#LMETISDIR = /local/metis/" to "LMETISDIR = ${HOME}/local/lib" 4) uncomment the line "LMETIS = -L$(LMETISDIR) -lmetis" 5) change the line "ORDERINGSF = -Dpord" to "ORDERINGSF = -Dpord -Dscotch -Dmetis" 6) change "MKLROOT=/opt/intel/mkl/lib/intel64" to "MKLROOT=/opt/intel-12.1/mkl/lib/intel64" 7) add the "-fPIC" compiler flag to the lines that begin "OPTF", "OPTL", and "OPTC" # about damn time! make z # there's no automatic installation... cp include/* ~/local/include/ cp lib/* ~/local/lib/ cp libseq/libmpiseq.a ~/local/lib/ cp libseq/*.h ~/local/include/ cd .. # now for kwant git clone --branch v1.2.2 http://gitlab.kwant-project.org/kwant/kwant.git cd kwant # need to make a new configuration file nano build.conf #### START OF FILE [lapack] libraries = mkl_intel_lp64 mkl_sequential mkl_core mkl_def library_dirs = /opt/intel-12.1/mkl/lib/intel64 extra_link_args = -Wl,-rpath=/opt/intel-12.1/mkl/lib/intel64 [mumps] libraries = zmumps mumps_common pord metis esmumps scotch scotcherr mpiseq gfortran library_dirs = /home/lmaurer/local/lib/ include_dirs = /home/lmaurer/local/include #### END OF FILE # finally make and install kwant python3 setup.py build python3 setup.py install # need to do the following or kwant won't be able to find libmetis.so export LD_LIBRARY_PATH=~/local/lib/:$LD_LIBRARY_PATH For reference, here are my libs and headers $ ls local/lib/ libesmumps.a libmpiseq.a libpord.a libscotch.a libscotcherrexit.a libzmumps.a python3.5 libmetis.so libmumps_common.a libpython3.5m.a libscotcherr.a libscotchmetis.a pkgconfig $ ls local/include/ cmumps_c.h cmumps_struc.h dmumps_root.h elapse.h metis.h mpi.h mumps_c_types.h scotchf.h smumps_c.h smumps_struc.h zmumps_root.h cmumps_root.h dmumps_c.h dmumps_struc.h esmumps.h mpif.h mumps_compat.h python3.5m scotch.h smumps_root.h zmumps_c.h zmumps_struc.h

2 1

more MUMPS problems
by Maurer, Leon Oct. 7, 2016

Oct. 7, 2016

Hello everyone, I’m in the unfortunate position of having to compile Kwant (and all of its dependencies) from scratch in my home folder, and I’m having trouble getting Kwant to play well with MUMPS. After quite a bit of futzing, I’ve got everything to compile happily (the Kwant build recognizes the "User-configured MUMPS”), but Kwant still won’t work with MUMPS at runtime: ############################# In [1]: import kwant /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/solvers/default.py:18: RuntimeWarning: MUMPS is not available, SciPy built-in solver will be used as a fallback. Performance can be very poor in this case. "Performance can be very poor in this case.", RuntimeWarning) In [2]: from kwant.linalg import mumps --------------------------------------------------------------------------- ImportError Traceback (most recent call last) <ipython-input-2-2962310cb52c> in <module>() ----> 1 from kwant.linalg import mumps /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/linalg/mumps.py in <module>() 15 import numpy as np 16 import scipy.sparse ---> 17 from . import _mumps 18 from .fortran_helpers import prepare_for_fortran 19 ImportError: /home/lmaurer/local/lib/python3.5/site-packages/kwant-1.2.2-py3.5-linux-x86_64.egg/kwant/linalg/_mumps.cpython-35m-x86_64-linux-gnu.so: undefined symbol: __kmpc_ok_to_fork ############################# For what it’s worth, I’ve put my entire build process below. It’s probably too much information, but the short version is that I’m using Intel’s complier on Linux. The Kwant-specific stuff is near the end. The earlier stuff is compiling Python, SCOTCH, METIS, and MUMPS. Any advice? Thanks. -Leon # folders to hold everything cd mkdir local mkdir build cd build # install python and packages wget https://www.python.org/ftp/python/3.5.2/Python-3.5.2.tar.xz tar xf Python-3.5.2.tar.xz cd Python-3.5.2 ./configure --prefix=~/local make make install export PATH=~/local/bin:$PATH cd .. # install important python packages pip3 install scipy tinyarray matplotlib cython # get SCOTCH, METIS, MUMPS wget http://gforge.inria.fr/frs/download.php/file/34618/scotch_6.0.4.tar.gz wget http://glaros.dtc.umn.edu/gkhome/fetch/sw/metis/metis-5.1.0.tar.gz wget http://graal.ens-lyon.fr/MUMPS/MUMPS_5.0.2.tar.gz tar xf scotch_6.0.4.tar.gz tar xf metis-5.1.0.tar.gz tar xf MUMPS_5.0.2.tar.gz # install METIS cd metis-5.1.0 make config shared=1 prefix=~/local/ make make install cd .. # install SCOTCH cd scotch_6.0.4/src/ cp Make.inc/Makefile.inc.x86-64_pc_linux2.icc Makefile.inc # need to tweak the makefile nano Makefile.inc Add "-fPIC” to the line that starts with "CFLAGS" make make esmumps make install prefix=~/local # make install won't install esmumps... copy files manually cd ../.. cp scotch_6.0.4/lib/libesmumps.a ~/local/lib/ cp scotch_6.0.4/include/esmumps.h ~/local/include/ # install MUMPS cd MUMPS_5.0.2 cp Make.inc/Makefile.INTEL.SEQ Makefile.inc # need to make a bunch of changes to Makefile.inc nano Makefile.inc 1) change "#SCOTCHDIR = ${HOME}/scotch_6.0" to "SCOTCHDIR = ${HOME}/local/lib" 2) uncomment the line "LSCOTCH = -L$(SCOTCHDIR)/lib -lesmumps -lscotch -lscotcherr" 3) change the line "#LMETISDIR = /local/metis/" to "LMETISDIR = ${HOME}/local/lib" 4) uncomment the line "LMETIS = -L$(LMETISDIR) -lmetis" 5) change the line "ORDERINGSF = -Dpord" to "ORDERINGSF = -Dpord -Dscotch -Dmetis" 6) change "MKLROOT=/opt/intel/mkl/lib/intel64" to "MKLROOT=/opt/intel-12.1/mkl/lib/intel64" 7) add the "-fPIC" compiler flag to the lines that begin "OPTF", "OPTL", and "OPTC" # about damn time! make z # there's no automatic installation... cp include/* ~/local/include/ cp lib/* ~/local/lib/ cp libseq/libmpiseq.a ~/local/lib/ cp libseq/*.h ~/local/include/ cd .. # now for kwant git clone --branch v1.2.2 http://gitlab.kwant-project.org/kwant/kwant.git cd kwant # need to make a new configuration file nano build.conf #### START OF FILE [lapack] libraries = mkl_intel_lp64 mkl_sequential mkl_core mkl_def library_dirs = /opt/intel-12.1/mkl/lib/intel64 extra_link_args = -Wl,-rpath=/opt/intel-12.1/mkl/lib/intel64 [mumps] libraries = zmumps mumps_common pord metis esmumps scotch scotcherr mpiseq gfortran library_dirs = /home/lmaurer/local/lib/ include_dirs = /home/lmaurer/local/include #### END OF FILE # finally make and install kwant python3 setup.py build python3 setup.py install # need to do the following or kwant won't be able to find libmetis.so export LD_LIBRARY_PATH=~/local/lib/:$LD_LIBRARY_PATH For reference, here are my libs and headers $ ls local/lib/ libesmumps.a libmpiseq.a libpord.a libscotch.a libscotcherrexit.a libzmumps.a python3.5 libmetis.so libmumps_common.a libpython3.5m.a libscotcherr.a libscotchmetis.a pkgconfig $ ls local/include/ cmumps_c.h cmumps_struc.h dmumps_root.h elapse.h metis.h mpi.h mumps_c_types.h scotchf.h smumps_c.h smumps_struc.h zmumps_root.h cmumps_root.h dmumps_c.h dmumps_struc.h esmumps.h mpif.h mumps_compat.h python3.5m scotch.h smumps_root.h zmumps_c.h zmumps_struc.h

2 2