Flora Tsourtou & Vlasis G. Mavrantzas. HPC for Innovation: when Science meets Industry, May, Barcelona, Spain

Size: px

Start display at page:

Download "Flora Tsourtou & Vlasis G. Mavrantzas. HPC for Innovation: when Science meets Industry, May, Barcelona, Spain"

Arron Kennedy
5 years ago
Views:

1 Optimized, Large-scale Monte Carlo and Molecular Dynamics Algorithms for modelling the self-organization of two classes of materials: semifluorinated alkanes & semiconducting polymers based on thiophenes Flora Tsourtou & Vlasis G. Mavrantzas HPC for Innovation: when Science meets Industry, May, Barcelona, Spain 1

2 Outline 1. Soft Matter Materials a. Semifluorinated Alkanes b. Thiophene Semiconducting polymers 2. Atomistic Molecular Simulations a. Molecular Dynamics b. Monte Carlo 3. Applications a. Simulation details b. High performance issues c. Results 4. Conclusions & Ongoing Plans 5. Acknowledgments 2

Soft Matter Materials Semifluorinated Alkanes Soft matter materials Apart

designof new materials with tailored properties applications of these

Aggregation and wormlike surface micelles [2] 1 st Example Semifluorinated

F14H20 on substrate [1] F(CF 2 )m(ch 2 )nh or FmHn [1] Brinatowski et al.

3 Soft Matter Materials Semifluorinated Alkanes Soft matter materials Apart from chemistry and microstructure, morphology plays a key factor in the designof new materials with tailored properties applications of these materials in industrial practice. crystal mesophase amorphous 24nm 2. Aggregation and wormlike surface micelles [2] 1 st Example Semifluorinated Alkanes (SFAs): 1. Crystal packing of F12Hn [1] 3. F14H20 on substrate [1] F(CF 2 )m(ch 2 )nh or FmHn [1] Brinatowski et al., Adv. Colloid Interface Sci., 138 (2008) 63. [2]de Viguerie et al., Langmuir, 27 (2011) Applications of SFAs: Biomedical uses: as components of artificial blood, in drug delivery, for ophthalmologic uses and in lung surfactant replacement preparations. 3

PT R=C6 P3HT (10T) Typical microstructure

thin films [6] (P3HT) 16 Å Unsubstituted

Mol. El., 6 (1990) 228. [5]Sander et al.

4 Soft Matter Materials Thiophene Semiconducting polymers 2 nd Example: Thiophene semiconducting polymers Substituted alkyl-thiophene polymers R=H PT R=C6 P3HT (10T) Typical microstructure of rr-p3ht Molecular Ordering on P3HT thin films [6] (P3HT) 16 Å Unsubstituted short thiophene polymers Crystalline polymorphism [3] Mesophases [4] v PM picture of α-6t Molecular Ordering on substrate [5] Applications of Thiophene semiconducting polymers: In opticoelectronic devices, such as light-emitting diodes (OLEDs), field effect transistors (FETs) or solar cells. [3]Yang et al., Chem. Mater., 22 (2008) [4]Taliani et al., J. Mol. El., 6 (1990) 228. [5]Sander et al., J. Chem. Phys., 141 (2014) [6]Lim et al., Materials Today, 13 (2010) 14. 4

5 Atomistic Molecular Simulations Molecular Dynamics method Simulation and detailed characterization of soft matter materials with rich self-assembly behavior? SIMULATIONS ON ATOMISTIC LEVEL!!! MOLECULAR DYNAMICS METHOD: i i ( ) ( ) r t=0 v t=0 Drawbacks: Potential function U NVT statistical ensemble Integration equations of motion pη pi q η = i = Q m η i N 2 pi p i = Fi ( = ru) p dnk i η = BT m classical trajectory file r = r ( t) & It is possible to simulate today only pieces of materials of size of up to (100nm) 3 and for times up to 5μs. This is not enough to capture morphology of nanostructured polymer systems. i= 1 i i v i i = v ( t), i= 1,2,..., N i Dynamic & structural properties 5

6 Atomistic Molecular Simulations ATOMISTIC Monte Carlo method MC is a stochastic method relying on transition probabilities between different states of the simulated system. Initial state Random Selection of r i Random Selection of a move r i, U(r i ) r i, U(r i ) Final state Accept the final state ρ ' NVT Ur ( i) Ur ( i ) acc ~ exp kt B random number MC possesses the potential to overcome the long relaxation time problem. MC offers complete information about the static properties (e.g., morphology) but not for dynamic properties. Artificial, clever moves increase dramatically the rate of sampling new microstates. 6

7 Atomistic Molecular Simulations Simple Monte Carlo moves for chain-like systems End-Mer Rotation Reptation [7] Concerted Rotation [9,10] Internal Flip Configurational Bias (CB)[8] [7] Vacatello et al., J. Chem. Phys., 73 (1980) 548. [8]Siepmann and Frenkel, Mol. Phys., 75 (1992) 59. [9]Dodd et al., Mol. Phys., 78 (1993) 961. [10] Pant and Theodorou, Macromolecules, 28 (1995)

1st Application: SFAs F12H12-Experimental data Two phase

[11,12,13] [13] V (cm 3 /g) II Isotropic liquid [11,12]

B, 112 (2008) 6542. [12] Lee et al., J. Phys. Chem.

8 1st Application: SFAs F12H12-Experimental data Two phase transitions (bilayer & monolayer lamellar phases).[11,12,13] [13] V (cm 3 /g) II Isotropic liquid [11,12] I Temperature (K) [11] Nunez et al., J. Phys. Chem. B, 112 (2008) [12] Lee et al., J. Phys. Chem. B, 113 (2009) 2. [13] Russell et al., Macromolecules, 19 (1986)

9 1st Application with a united atom model: SFAs-Simulation details System studied Number of F12H12 chains Number of atomistic units Volume (Å 3 ) Monte Carlo simulations a MC algorithm was designed, with following set of MC moves: , x 172 x 172 Conditions Temperature range : 600K - 300Κ Pressure : 100 atm up to 600 atm End-mer rotation Molecular Dynamics simulations Simulation code: LAMMPS MPI version Simulation time : ns CPU cores: Reptate 9

10 1st Application: SFAs Monte Carlo moves Flip Conrot (f) Segment-exchange move Configuration Bias 10

1st Application: SFAs Where is the need for

Volume Fluctuation : displacement of all

implementation [14] Parallelism of the two

Every atomistic unit was assigned to a thread.

11 1st Application: SFAs Where is the need for optimization? Volume Fluctuation : displacement of all atomistic units subroutine CPU utilization % Verlet list LJ energy 6.82 Poor performance of MC algorithm MC and GPU implementation [14] Parallelism of the two most-cpu demanding MC subroutines. Every atomistic unit was assigned to a thread. Thousands of atomistic units: more than one blocks 2-D grid of blocks of threads in CUDA [14] Tsourtou et al., Chem. Eng. Sci., 121 (2015)

1st Application: SFAs Where is the need for optimization? Reduction operation: find the total sum of an array Two most-cpu demanding subroutines System CPU time (s) GPU-time (s) Speedup 750-chain 0.

12 1st Application: SFAs Where is the need for optimization? Reduction operation: find the total sum of an array Two most-cpu demanding subroutines System CPU time (s) GPU-time (s) Speedup 750-chain chain Atomic add operation: calculate the total sum of LJ energy Total MC algorithm System CPU time (s) GPU-time (s) Speedup 750-chains chains The computational problem was solved! 12

13 1st Application: SFAs Results A spontaneous phase transition occurs from an isotropic phase to a smectic phase. [14] 300K (a) 6000-chain system [14] Tsourtou et al., Chem. Eng. Sci., 121 (2015)

14 1st Application: SFAs Results Does any experimental structure match the smectic phase predicted by our simulations? [13] (b) (d) [13] Russell et al., Macromolecules, 19 (1986)

2nd Application: α-unsubstituted Oligothiophenes Literature data α-unsubstituted Oligothiophenes nts with n=2-8 Herringbone crystal structure High melting points (T m > 450K for n>3) high degree of

15 2nd Application: α-unsubstituted Oligothiophenes Literature data α-unsubstituted Oligothiophenes nts with n=2-8 Herringbone crystal structure High melting points (T m > 450K for n>3) high degree of molecular and crystalline ordering (key advantage over their parent polymers) Applications: TFTs and OLEDs α-4t/ht α-6t/ht α-8t α-6t improved mobilities for an organic semiconductor (0.15 cm 2 /Vs in single crystal & cm 2 /Vs in thin film) Polymorphic crystalline phases T m = 587 K [15] XRD, DSC and POM techniques: a mesophase exists, when 6T crystal heated above 580K : nematic or smectic? [4,16] Only one computational work: Three phase transitions [17] All-atom MD simulations Home made force field Transition T (K) Crystal Smectic 580 Smectic Nematic 610 Nematic Isotropic 670 [15] Fichou et al., Adv. Mater., 9 (1997) 75. [4] Taliani et al., J. Mol. El., 6 (1990) 225. [16] Destri et al., Adv. Mater.,5 (1993) 43. [17] Pizzirusso et al., J. Mater. Chem., 21 (2011)

16 2nd Application: α-unsubstituted Oligothiophenes Simulation Details MD Force Field A simplified united-atom model was developed. U = U + U + U It relies on Dreiding force field [18] in combination with the rescaling of Lennard-Jones (LJ) parameters. It does not include electrostatic terms. + U tot streching bending dihedral LJ MC Force Field A stiffer united-atom model was developed. U = Uint + Uint + U tot er ring er ring LJ bending dihedral Fixed bond lengths & intra-ring angles (bending and dihedral). All atomistic units on a thiophene ring lie on the same plane. System studied: α-unsubstituted 6Τ Number of α- 6T chains Number of atomistic units Molecular Dynamics simulations Simulation code: LAMMPS MPI version Simulation time : ns CPU cores: Volume (Å 3 ) , x 97 x , x 192 x 192 Conditions Statistical ensemble: NPT Temperature range : 800K - 570Κ Pressure : 1 atm [18] Mayo et al., J. Phys. Chem., 94 (1990)

17 2nd Application: α-unsubstituted Oligothiophenes MC simulations MC algorithm was designed, with the following set of MC moves: Chain-Translation Reptate End-ring rotation Chain-Rotation Flip Conrot 17

2nd Application: α-unsubstituted Oligothiophenes MD results Order Parameter 1.2 1.0 0.8 0.6 0.4 0.

[17] MD, 1120-chain system 1.40 MD 1120-chain system density (gcm -3 ) 1.35 1.30 1.25 1.20 1.15 1.

90 550 600 650 700 750 800 T (K) α-6τ: Three phase transitions occur: nematic (640K), smectic

18 2nd Application: α-unsubstituted Oligothiophenes MD results Order Parameter T(K) 1.50 Pizzirusso et al. [17] 1.45 Pizzirusso et al. [17] MD, 1120-chain system 1.40 MD 1120-chain system density (gcm -3 ) T (K) α-6τ: Three phase transitions occur: nematic (640K), smectic (620K), crystal (590K). Our model also captures the melting point of 6T (587K)! [15] [15] Fichou et al., Adv. Mater., 9 (1997) 75. [17] Pizzirusso et al., J. Mater. Chem., 21 (2011)

2nd Application: α-unsubstituted Oligothiophenes Validity of MC algorithm Phantom Chain-tests and comparison with analytical distributions of angles. P(θ) 0.10 0.09 0.08 0.07 0.06 0.05 0.04 0.03 0.

19 2nd Application: α-unsubstituted Oligothiophenes Validity of MC algorithm Phantom Chain-tests and comparison with analytical distributions of angles. P(θ) θ 2,1 θ 2,2 θ 3,1 θ 3,2 theoretical θ (deg) P(φ) phantom-chain test theoretical φ (deg) Identical distribution of angles!!! Correct implementation of MC moves. P(φ) phantom-chain test theoretical φ (deg) 19

2nd Application: α-unsubstituted Oligothiophenes MPI & Preliminary MC runs MC and MPI implementation: Parallelism of the two most- CPU demanding MC subroutines (Verlet neighboring list, LJ energy of

20 2nd Application: α-unsubstituted Oligothiophenes MPI & Preliminary MC runs MC and MPI implementation: Parallelism of the two most- CPU demanding MC subroutines (Verlet neighboring list, LJ energy of volfluc move). Speedup chain system number of CPU cores Order Parameter 140-chain system at 1 atm and various Temperatures MC-620K T (K) Pizzirusso et al. MD simulation MC simulation 20

21 2nd Application: α-unsubstituted Oligothiophenes Comparison of methods 1120-chain system at 800K and 1 atm property MC method MD method <R ete2 > (Å 2 ) 413 ± ± 1 ρ (gcm -3 ) 0.91 ± ± MD simulation MC simulation MD simulation MC simulation P(φ) φ SCCS (degr ees) g(r) intra S-S distance (Å) g(r) inter S-S MD simulation MC simulation distance (Å) 21

Conclusions & Ongoing Plans Conclusions 1. MC algorithm can be used to simulate chain self-organization and morphology in soft matter materials. 2.

22 Conclusions & Ongoing Plans Conclusions 1. MC algorithm can be used to simulate chain self-organization and morphology in soft matter materials. 2. Parallel programming (GPU or MPI) is mandatory to proceed with productions runs. We accomplished this for SFAs. We still have to do this for oligothiophenes. Ongoing-Future Plans 1. GPU implementation in case of oligothiopene Monte Carlo algorithm. 2. Strategy is being extended to 2.1 α-nts with n=4,5,7,8, 2.2 P3HT semiconducting polymers & 2.3 Polypeptides, so as to predict morphology over several length scales. 3. Optimization of the united force field used in simulation for SFAs. Vacuum MC simulation of L-alanine 22

Acknowledgments Dr. Orestis Alexiadis (SFAs) Dr. Vasileios Kolonias (GPU implementation) Dr.

23 Acknowledgments Dr. Orestis Alexiadis (SFAs) Dr. Vasileios Kolonias (GPU implementation) Dr. Stavros Peroukidis (α-unsubstituted oligothiophenes) We acknowledge PRACE for awarding us with access to the resource unit SZEGED, based in Hungary at NIIF. We also acknowledge GRNET for awarding us with access to the resource unit in the National HPC facility - ARIS, based in Athens, Greece. Financial support from the Project (code 1010 D655) titled: General method for the simulation of self-organization in nanostructured polymeric systems (GENESIS) is gratefully acknowledged. This project is part of the Action ARISTEIA - Education and Lifelong Learning Programme and is supported both by the European Union (European Social Fund-ESF) and national funds. 23

24 THANK YOU! 24

Supporting Information

Supporting Information Projection of atomistic simulation data for the dynamics of entangled polymers onto the tube theory: Calculation of the segment survival probability function and comparison with modern tube models Pavlos