Beruflich Dokumente
Kultur Dokumente
Vidvuds Ozolins
Department of Materials Science & Engineering
University of California, Los Angeles
DMR-1106024
Supported by:
UCLA Laboratory for the Quantum
Prediction of Advanced Materials
Co-conspirators
Gus Hart
(BYU Provo)
Rogjie Lai
(UC Irvine)
Lance Nelson
(BYU Idaho)
Russ Caflisch
(UCLA)
Fei Zhou
(LLNL)
Stan Osher
(UCLA)
Weston Nielson
(UCLA)
Outline
Premise:
Most physics models are approximately sparse
(i.e., have a few terms) in some basis
Questions:
How do we pick the basis?
How do we determine which terms to pick?
How should we generate fitting data?
How do we systematically improve accuracy?
Nb layer
Oxygen
50 meV
Cu3Au
CuAu
All these structures are based on the facecentered cubic (FCC) lattice.
CuAu3
Cluster Expansion
Rewrite as an expansion in clusters of lattice sites:
E = E0 + J1c +
Jf
if
+ ...
ADVANTAGES:
Very fast can be used in Monte Carlo simulations
Works for any structure based on a given lattice
10 000
5-bodies
4-bodies
1000
triplets
100
10.
pairs
1.
0.5
1.0
1.5
2.0
cluster radius
2.5
3.0
Compressive sensing CE
Clusters
Structures
. . . .
1
B 0
B
@ .
.
1
1
.
.
...
...
...
...
...
...
...
S
...
if
10
B
CB
CB
AB
B
B
B
B
B
B
@
?
Jnn
Jnnn
Jtrip
.
.
.
.
.
1 0
C B
C B
C= B
C B
B
C B
C @
C
C
C
C
A
J CS = arg min J 1 : J E 2
J
L. J. Nelson, F. Zhou, G.L.W. Hart, and V. Ozolins, Phys. Rev. B 87, 035125 (2013)
E(
E(
E(
)
)
)
.
.
.
1
C
C
C
C
C
C
A
x1
a11 x1 + a12 x2 = b1
n
x 1 = xi
i=1
x * = argmin x: Ax=b x
x1
a11 x1 + a12 x2 = b1
x2=
x
i=1
2
i
Theorem
Probability to find the correct S-sparse solution is
P >1
if the number of data points satisfies
M C S log ( N )
2
Coherence
1N
Sparsity
Size of
basis set
E. Cands, J. Romberg, and T. Tao, Comm. Pure Appl. Math., vol. 59, no. 8, pp. 12071223 (2006).
Triplets
Cluster radius
L. J. Nelson, F. Zhou, G.L.W. Hart, and V. Ozolins, Phys. Rev. B 87, 035125 (2013).
Quadruplets
Triplets
Cluster radius
L. J. Nelson, F. Zhou, G.L.W. Hart, and V. Ozolins, Phys. Rev. B 87, 035125 (2013).
Quadruplets
Discrete Opt.
Compressive sensing
Training set size
L. J. Nelson, F. Zhou, G.L.W. Hart, and V. Ozolins, Phys. Rev. B 87, 035125 (2013).
Performance of CSCE
L. J. Nelson, G. L. W. Hart, S. Reese, F. Zhou, and V. Ozolins, Physical Review B 88, 155105 (2013).
10
0
-10
-20
-30
-30
-20
-10
0
Direct energy HkcalmolL
L. J. Nelson, F. Zhou, G.L.W. Hart, and V. Ozolins, Phys. Rev. B 87, 035125 (2013).
10
thermopower
S
ZT =
T
total
2
hot
Power factor
cold
Thermopower S = V/T
(direct gaps at L point) result from the occurrence of the Pb s band below the top
of the valence band, setting up coupling and level repulsion at the L point.
From Wei & Zunger, Physical Review B 55, 1360513610 (1997)
Group VI
I-V-VI2
AgSbTe2, NaSbSe2,
M. Nielsen, V. Ozolins, and J. P. Heremans, Energy & Environ. Sci. 6, 570-578 (2013).
M. Nielsen, V. Ozolins, and J. P. Heremans, Energy & Environ. Sci. 6, 570-578 (2013).
0
AuAsTe2
CuBiSe2
AuBiTe2
CuAsTe2
AgAsTe2
AgSbSe2
CuBiTe2
AuSbTe2
NaSbS2
KAsSe2
AgBiS2
CuSbTe2
AgBiSe2
AgBiTe2
AgSbTe2
NaAsSe2
CsSbSe2
RbSbSe2
NaSbSe2
KSbSe2
CsBiS2
RbBiS2
NaAsTe2
NaBiS2
KBiS2
NaSbTe2
KAsTe2
NaBiSe2
RbAsTe2
CsAsTe2
RbBiSe2
CsBiSe2
NaBiTe2
KBiSe2
RbSbTe2
KSbTe2
CsBiTe2
RbBiTe2
KBiTe2
CsSbTe2
Gruneisen parameter
M. Nielsen, V. Ozolins, and J. P. Heremans, Energy & Environ. Sci. 6, 570-578 (2013).
M. Nielsen, V. Ozolins, and J. P. Heremans, Energy & Environ. Sci. 6, 570-578 (2013).
Cu
Sb
S
X. Lu, D. T. Morelli, Y. Xia, F. Zhou, V. Ozolins, H. Chi, and C. Uher. Advanced Energy Materials (2013)
!
!
!1 cm
!!!!!!!!!!!!!!!!!!!!!!!!!!
Figure of Merit
1
1.2
a)
Figure of Merit zT
Figure of Merit zT
1
0.8
0.6
0.4
0.2
0.8
0.6
0.4
f =1 for Zn (2+)
fb)=2 for Fe (3+)
0.2
0
0
300
400
500
600
Temperature (K)
700
800
0.2
0.4
0.6
0.8
Lu, Morelli, Xia, Zhou, Ozolins, Chi, Zhou, and Uher, Advanced Energy Materials (2013).
Cu(1) (3-fold)
Lu, Morelli, Xia, Zhou, Ozolins, Chi, Zhou, and Uher, Advanced Energy Materials (2013).
DE HmeVL
30
20
10
0
-10
-0.4
-0.2
0.0
uCu HL
0.2
0.4
Lu, Morelli, Xia, Zhou, Ozolins, Chi, Zhou, and Uher, Advanced Energy Materials (2013).
CS lattice dynamics
Taylor expansion of the total energy in
terms of the atomic displacements
ua=Ra-R0a:
1
0 + a ua +
2
1
ab ua ub +
6
abc ua ub uc
Forces:
Fa =
@ /@a =
1
( a + ab ub +
2
abc ub uc
+ )
abua ub , aabua2 ub ,
abc ua ub uc , aabc ua2 ub uc ,
abcd ua ub uc ud ,
Calculate forces Fa
Fa =
Displace atoms ua
1
@ /@a
Fa = @( a/@a
+ =ab u(b +
a+
2
1
uc + abc
)ub uc
abc
ab uubb+
2
Compressive sensing
We need to solve an underdetermined linear system
of equations:
1 u1 u1 u1 a
b
b c
ab
2
2 2
1 ub ub uc
abc
CS
1
Fa
= F 2
a
1
= arg min 1 +
+ F
2
2
Symmetry constraints
Commutativity of derivatives
abc
acb
= ...
(aa) =
(ab)
Pair ASR
b6=a
X
a
I ({a, b, c })
=0
Multibodies
Dynamical properties of solids, eds. G.K. Horton & A.A. Maradudin (1974)
Symmetry constraints
Space group symmetry
s)
I (
s)
IJ (
J ()
Transformation matrix:
s)
IJ (
i1
j(1)
u1
F3
F2
in
j(n)
cos
= sin
Dynamical properties of solids, eds. G.K. Horton & A.A. Maradudin (1974)
sin
cos
0
0
0
1
where
CS
1
2
= arg min 1 + + F 2
S
= BT BS BP S.
0.10
0.05
100
200
300
400
500
600
-0.05
Independent FCs by
order, then by distance
Solution is sparse
Force prediction error = 3%
15
FCSLD @eVD
-0.10
20
10
5
0
-5
-10
u/r0=25%
-15
-15 -10 -5
10
FDFT @eVD
Applications
15
20
Performance of CSLD
Applications
L1 regularization of DFT
Conventional Kohn-Sham problem:
N
E = minN j H j
{ i }i=1
j =1
s.t. i j = ij
1 N
E = minN j H j + j s.t. i j = ij ,
1
{ i }i=1 j =1
j =1
1 (x) dx, x R d
This method converges to spatially localized wave f-ns with compact
support. O(N) becomes possible.
V. Ozolins, R. Lai, R. E. Caflisch, and S. Osher, Compressed modes , PNAS (2013)
Results
Results
Dependence on
1
0.6
1
0.6
3
0.4
4
5
0.2
4
5
0.2
0
0
0
3
0.4
10
15
20
25
30
35
40
45
50
10
15
20
25
30
35
40
45
50
= 50
= 30
1
0.6
0.6
2
3
3
0.4
4
5
0.2
4
5
0.2
0
0
0
0.4
10
15
20
25
30
35
40
45
50
10
15
20
25
30
35
40
45
50
= 300
= 50
1
0.6
0.6
3
0.4
0.2
5
0.2
0
0
3
0.4
0
5
10
15
20
25
30
35
40
45
50
= 500
10
15
20
25
30
35
40
45
50
= 5000
Fig. 4.
Computation results of CMs with dierent values of . The first column: the first 5 CMs of the 1D free-electron model. The second column: the first 5 CMs of the
1D Kronig-Penny model.
Fig. 5.
Comparisons of the first 50 eigenvalues of the 1D free electron model (the first row) and
Potential free
KronigPenney
Relative Error
0.08
0.06
0.04
0.02
0
10
50
100
200
300
400
500
M = 50, = 10
0.1
Potential free
KronigPenney
Relative Error
0.08
0.06
0.04
0.02
0
50
60
70
80
90
100
128
Fig. 6.
compressed mod
Schr
odinger oper
( 1 , , M ) of
KP model and 1D
values ( 1 , ,
ators. Figure 5
small value =
can clearly see t
increasing numb
relative error E =
we speculated in
to zero as ! 1
in the top panel
verge to zero as
is illustrated in t
Moreover, th
also work on dom
ample, Figure. 7
CMs of the free = 30. All the
for 2D cases. In
extended to irre
which will be inv
In conclusion
Fig. 7. The first 25 CMs of freel-electron case on a 2D domain [0, 10]2 with
d basis sets that span
= 30. Each CM is color-coded by its height function.
tor, for instance, the
oncept of plane waves
h multi-resolution caV. Ozolins,
ssed in Ref.
[20]. R. Lai, R. E. Caflisch, and S. Osher, Compressed modes , PNAS (2013)
ated (i.e., O(N )) simACKNOWLEDGMENTS. V.O. gratefully acknowledges financial support from the
National Science Foundation under Award Number DMR-1106024 and use of comal theory (DFT) elec-
Set V=0 to derive general compactly supported basis for the Laplacian:
CPW
1
(x) = arg min + 1 s.t. (x) (x jw) = 0 j
2
V. Ozolins, R. Lai, R. E. Caflisch, and S. Osher, Compressed plane waves , UCLA CAM Report (2013)
Introduction
4
0
2
0
5
2
0
10
20
30
40
50
60
70
80
90
100
10
20
30
40
50
5
10
20
30
40
50
60
70
80
90
100
10
20
30
40
50
80
90
100
60
70
80
90
100
5
0
70
60
10
20
30
40
50
60
70
80
90
100
10
20
30
40
50
60
70
80
90
100
6
1
V. Ozolins, R. Lai, R. E. Caflisch, and S. Osher, Compressed plane waves , UCLA CAM Report (2013)
Introduction
0.8
0.6
0.4
0.2
6i=1|i(G)|2
|i(G)|2
0.8
0.6
0.4
0.2
0
0
0.5
1.5
2.5
3
Wave vector G
3.5
4.5
0.5
1.5
2.5
3
Wave vector G
3.5
4.5
Figure: Spectral density distribution of CPWs. Top: the spectral density distribution of
1
, 2 , 3 , 4 , 5 , 6 . Bottom: The total spectral density distribution of the first four modes.
V. Ozolins, R. Lai, R. E. Caflisch, and S. Osher, Compressed plane waves , UCLA CAM Report (2013)
V. Ozolins & R. Lai & R. Caflisch & S. Osher
Summary
CS-based sparse physics is based on rigorous
math, not intuition
Accuracy can be improved systematically
Constraints (symmetry and translational
invariance) can be straightforwardly incorporated
CS automatically picks out important terms
A simple prescription for gathering data