Re: r24280 - /branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py -- June 24, 2014

Wow, this is impressive!  And if you come up with a multi-rank
replacement for the square_matrix_power() function, you'll be able to
finally eliminate all the looping and make this numeric model hugely
faster.  Although we can't test it directly, I would not be surprised
if this would then be by far the fastest implementation in the field.
Though we don't have all of the tricks implemented for the numeric
models - off-resonance effects, the correction factor Andy talked
about, etc. - this will still be huge.

Regards,

Edward



On 24 June 2014 14:58,  <tlinnet@xxxxxxxxxxxxx> wrote:

Author: tlinnet
Date: Tue Jun 24 14:58:07 2014
New Revision: 24280

URL: http://svn.gna.org/viewcvs/relax?rev=24280&view=rev
Log:
Speeded up model NS CPMG 2site star, by moving the forming of the 
propagator matrix out of the for loops, and preform it.

Task #7807 (https://gna.org/task/index.php?7807): Speed-up of dispersion 
models for Clustered analysis.

Modified:
    branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py

Modified: branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py
URL: 
http://svn.gna.org/viewcvs/relax/branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py?rev=24280&r1=24279&r2=24280&view=diff
==============================================================================
--- branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py       
(original)
+++ branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py       Tue 
Jun 24 14:58:07 2014
@@ -57,7 +57,7 @@
 """

 # Python module imports.
-from numpy import add, array, conj, dot, fabs, float64, isfinite, log, 
min, multiply, sum
+from numpy import add, array, conj, dot, einsum, fabs, float64, isfinite, 
log, min, multiply, sum
 from numpy.ma import fix_invalid, masked_where

 # relax module imports.
@@ -221,8 +221,15 @@
     # The matrix R that contains all the contributions to the evolution, 
i.e. relaxation, exchange and chemical shift evolution.
     R_mat, cR2_mat, Rr_mat, Rex_mat, RCS_mat = rcpmg_star_rankN(R2A=r20a, 
R2B=r20b, dw=dw, k_AB=k_AB, k_BA=k_BA, tcp=tcp)

+    # The the essential evolution matrix.
+    # This matrix is a propagator that will evolve the magnetization with 
the matrix R for a delay tcp.
     eR_mat = matrix_exponential_rank_NE_NS_NM_NO_ND_x_x(R_mat)
     ecR2_mat = matrix_exponential_rank_NE_NS_NM_NO_ND_x_x(cR2_mat)
+
+    # Preform the matrix.
+    # This is the propagator for an element of [delay tcp; 180 deg pulse; 
2 times delay tcp; 180 deg pulse; delay tau], i.e. for 2 times tau-180-tau.
+    prop_2_mat = evolution_matrix_mat = einsum('...ij,...jk', eR_mat, 
ecR2_mat)
+    prop_2_mat = evolution_matrix_mat = einsum('...ij,...jk', prop_2_mat, 
eR_mat)

     # Loop over the spins
     for si in range(NS):
@@ -236,16 +243,11 @@
                 # Extract the values from the higher dimensional arrays.
                 power_si_mi_di = int(power[0, si, mi, 0, di])

-                # This matrix is a propagator that will evolve the 
magnetization with the matrix R for a delay tcp.
-                eR_tcp = eR_mat[0, si, mi, 0, di]
-                ecR2_tcp = ecR2_mat[0, si, mi, 0, di]
-
                 # This is the propagator for an element of [delay tcp; 180 
deg pulse; 2 times delay tcp; 180 deg pulse; delay tau], i.e. for 2 times 
tau-180-tau.
-                prop_2 = dot(eR_tcp, ecR2_tcp)
-                prop_2 = dot(prop_2, eR_tcp)
+                prop_2_i = prop_2_mat[0, si, mi, 0, di]

                 # Now create the total propagator that will evolve the 
magnetization under the CPMG train, i.e. it applies the above 
tau-180-tau-tau-180-tau so many times as required for the CPMG frequency 
under consideration.
-                prop_total = square_matrix_power(prop_2, power_si_mi_di)
+                prop_total = square_matrix_power(prop_2_i, power_si_mi_di)

                 # Now we apply the above propagator to the initial 
magnetization vector - resulting in the magnetization that remains after 
the full CPMG pulse train.  It is called M of t (t is the time after the 
CPMG train).
                 Moft = dot(prop_total, M0)


_______________________________________________
relax (http://www.nmr-relax.com)

This is the relax-commits mailing list
relax-commits@xxxxxxx

To unsubscribe from this list, get a password
reminder, or change your subscription options,
visit the list information page at
https://mail.gna.org/listinfo/relax-commits

Re: r24280 - /branches/disp_spin_speed/lib/dispersion/ns_cpmg_2site_star.py

Header

Content

Related Messages