Sophie

Sophie

distrib > Mageia > 7 > armv7hl > by-pkgid > d96c571a2c0bcc32a63f4cd5f968e345 > files > 1026

xmds-2.2.3-2.mga7.noarch.rpm

<simulation xmds-version="2">
  <name> nonlinear_SE </name>
  
  <features>
    <benchmark />
    <auto_vectorise />
    <globals>
      <![CDATA[
        real N = 10.0; // number of atoms
        real g = 1.0;  // nonlinear coupling
      ]]>
    </globals>
  </features>
  
  <geometry>
    <propagation_dimension> t </propagation_dimension>
    <transverse_dimensions>
      <dimension name="x" lattice="512"
                 domain="(-7, 7)" />
    </transverse_dimensions>
  </geometry>

  <vector name="potential" type="real">
    <components> V </components>
    <initialisation>
      <![CDATA[
        V = 0.5*x*x;
      ]]>
    </initialisation>
  </vector>
  
  <!-- Although the density is real, we need to Fourier transform it, so it must be complex -->
  <computed_vector name="density" type="complex">
  	<components>dens</components>
	<evaluation>
		<dependencies>wavefunction</dependencies>
		<![CDATA[
			dens = mod2(psi);
		]]>
	</evaluation>
  </computed_vector>
  
  <computed_vector name="electrostatic" type="complex">
  	<components> V_E </components>
	<evaluation>
		<!-- 
		We compute the electrostatic self-potential in fourier space where the convolution
		is simply a local multiplication.  We've taken the analytical fourier transform of 1/r in 3D
		here, but we've needed to truncate the upper limit of the integral to keep it finite.
		This is simply because if the system is assumed to be periodic, then it will be everywhere in space,
		not just the place we think of it.  Truncating the integral corresponds to not considering the
		electrostatic potential due to the other copies (which we obviously don't want to include).
		
		If the BEC is kept to the inner half of the domain, the correct truncation of the integral will be
		half the width of the domain (the minimum separation between copies).
		-->
		<dependencies basis="kx">density</dependencies>
		<![CDATA[
				  // The analytic term is (1/(kx*kx) * (1 - cos(kx * R))) where R is the upper limit on the
				  // Fourier transform.  This has finite limit near kx = 0 analytically, but the computer
				  // won't know that when it tries to divide by zero.
    real factor = 0.0;
    if (kx == 0.0)
      factor = 0.125 * (_max_x - _min_x) * (_max_x - _min_x);
    else
      factor = (1.0 - cos(kx * 0.5 * (_max_x - _min_x))) / (kx * kx);

    V_E = +1e-1 * dens * factor;
    ]]>
	</evaluation>
  </computed_vector>
  
  <vector name="wavefunction" type="complex">
    <components> psi </components>
    <initialisation>
      <![CDATA[
        psi = sqrt(N) * pow(M_PI, -0.25) * exp(-x*x/2);
      ]]>
    </initialisation>
  </vector>

  <sequence>
    <integrate algorithm="ARK45" interval="6.28" 
                                 tolerance="1e-5">
      <samples> 50 </samples>
      <operators>

        <operator kind="ip">
          <operator_names> T </operator_names>
          <![CDATA[
            T = -i * 0.5 * kx * kx;
          ]]>
        </operator>

        <integration_vectors>
          wavefunction
        </integration_vectors> 
        <dependencies> potential electrostatic </dependencies>
        <![CDATA[
          dpsi_dt = T[psi] - i * (V + V_E + g * mod2(psi)) * psi;
        ]]>
      </operators>
    </integrate>
  </sequence>

  <output format="hdf5">
    <sampling_group initial_sample="yes" basis="x">
      <dependencies> wavefunction </dependencies>
      <moments> psireal psiimag </moments>
      <![CDATA[
        psireal = Re(psi);
        psiimag = Im(psi);
      ]]>
    </sampling_group>
  </output>
</simulation>