Name               : Kanchan Sarkar
Birthday            : 20.02.1985
City of birth       : Suri, India

Sex                    : Male

Blood Group    : A +ve
Marital Status  : Single
Nationality       : Indian
Religion           : Hindu

Current City     :Kolkata, India

 

 

 

Research Summery 

I am doing my PhD work under the supervision of Prof. S. P. Bhattacharyya at the Department of Physical Chemistry, Indian Association for the Cultivation of Science. One of the fundamental strategies of our research is to bring together the vectors of theory and practice. We have systematically explored the theoretical and conceptual underpinnings of the general optimization problems and different Soft-computing tools to identify knowledge gaps in existing literature. We have focused specifically on computing the properties of atoms and molecules using the principles of quantum mechanics and approximation methods. Solving the Schrodinger equation governing the properties of molecular system is extremely complex. Schrodinger Equation can only be solved exactly for simple systems like the rigid rotor, harmonic oscillator, particle in a box, hydrogen atom, etc. Majority of the quantum chemical problems (i.e. many electron atoms or molecules), however can not be solved exactly even with the use of super-computers and require some simplifying approximations -- use either semi-empirical or DFT methods as a reasonable compromise between model accuracy and computational cost. The cost of the calculation increases in nonlinear fashion with the accuracy of the calculation and the size of the system. It grows in a nonlinear manner, such as quadratically O(N$^2$), cubically O(N$^3$), quartically O(N$^4$), and greater. So, problems of this class are typically ill defined, difficult to model, and usually have large solution spaces. Environments of these problems are usually dynamic. Precise models are impractical, too expensive, or non-existent. Soft computing methods have a special appeal in this context. Motivation for working on SC methods was the need to cope with the complexity of existing mathematical and computational models of such chemical and physical systems. The key aspect for moving from Hard to Soft Computing is the observation that the computational effort required by conventional approaches often makes the problem almost unfeasible in many cases, while the enormity of the effect is a cost paid to gain a precision that in many applications is not really needed or, at least, can be relaxed without a significant effect on the quality of the solution. As such, we have developed a robust set of Soft-computing methodologies with a view to improving their computational power, and using these methodologies as means of constructing new approaches for solving some classical and quantum chemical problems. Due to the cross-disciplinary nature, the research had to be overlapping multi-directional. Specifically, we have addressed the following issues:

 

Development of Single String Based Computationally Intelligent Evolutionary Global Optimizer for Geometry Optimization: Despite their importance and the efforts invested on population based global optimizers so far, the status of such algorithms for cluster geometry optimization is still open to question. Population set based strategies that work on a set of sample points (n) from the feasible region, should at least be n-fold more efficient in locating the fitness-maximal string than a single string based optimizer. The linear scalability, however, is not guaranteed for most of the fitness landscapes. Therefore it is worthwhile to think of a single string based evolutionary optimizer for fitness maximization. That means it must be a mutation only process. The mutation process defined on floating point string has two flexible parameters viz mutation probability ($P_m$) and mutation intensity ($\Delta m$). We have proposed an adaptive `mutation only' heuristic in which both $P_m$ and $\Delta m$ are adaptively determined by the algorithm on the basis of continuous evaluation of the performance of past explorations guiding the search to the global optimum. Moreover, we insist on strong elitism in that a mutation is accepted only when it does not lower the fitness. The resulting algorithm has been called Completely Adaptive Random Mutation Hill Climbing (CARMHC) method. We demonstrate its working principles and workability with reference to determining the global minimum energy configurations of 3-D Coulomb balls. Structural properties, energetics, distribution of the normal modes are analyzed as functions of confinement strength and the number of charges in the system. Certain magic numbers are identified. The convergence behavior of the algorithm compares favorably with those of other existing global optimizer. To further test the efficacy of the CARMHC on more complex energy landscapes, we have applied the algorithm for optimizing the geometries of some (MgO)$_n$ nanostructures, described by Coulomb – Born-Mayer potential and finding global minimum of some Lennard-Jones Clusters. The algorithm is a computationally intelligent, cost-effective procedure endowed with the capability of automatically escaping from local minima and handling much larger and more complex systems. Our results also show that self-adaptation was found to be highly beneficial in reducing redundancy when compared with other algorithms. Achieving the best performance in problems of geometry optimization in strongly coupled finite clusters of particles interacting through long range forces via a single string based approach with adaptive parameter settings (CARMHC), indicates that a population based search may not be the best way of handling these class of problems. Since ab initio calculations are very costly, the CARMHC algorithm in conjunction with good semiempirical potentials can be used to generate the initial structures for ab initio quantum chemical refinements. Coulomb balls show a new kind of structural order in complex plasmas. The particles are arranged in nested shells. The observed structural order of Coulomb balls is similar to the ordering found in laser-cooled ion clusters in Paul and Penning traps.

 

Coulomb Balls under Anisotropic Confinement : The single string based evolutionary algorithm `CARMHC' has been used to investigate the effects of confinement anisotropy on small n-component mixture (n=1-3) of charged particles. We have studied the influence of size of the clusters and confinement anisotropy on the structural phase transition process and have gained new insight in to the phase behavior in such systems. The structures of these Coulomb clusters and their normal mode spectrum are analyzed as functions of the charge type present and asymmetric confinement strength. The clusters show structural phase transitions (both 1D to 2D and 2D to 3D and vice versa) at very low asymmetric confinement strength along a particular direction. For larger clusters these transition become continuous. Like mono-dispersed clusters, in bi- and tri-dispersed Coulomb clusters, these transitions are very much dependent on the number of ions present in the system and independent of the type of the charges of the system. For mixed clusters there are certain compositions that are relatively more stable. A much closer examination of the process is needed to understand the nature of the transition completely.

 

Blending Determinism with Soft-Computing:  In strongly convex potential energy surface, deterministic methods outperform EC. It can be said that those deterministic methods are specifically designed to take advantage of the convexity of such surfaces. However, the potential energy surface posed in quantum chemical systems are often multimodal and generally deterministic methods rapidly converge to nearby local optimum. Moreover, EC offers a significant advantage when the objective function is not differentiable, smooth or continuous. Calculations of eigenstates associated with discrete energy levels leads to diagonalization problem of the Hamiltonian matrix, which is a NP-hard problem. EC usually involve high computational cost and suffers from the low convergence rate at the end of the search. Essentially efforts must be made in order to improve the numerical behavior of EC. A logical step to do so is to hybridize the determinism with EC to decrease computational cost and increase numerical performance of the pure stochastic search. Thus the global convergence furnished by the EC can be blended with the precision of the deterministic formulation. We have introduced a new method of calculating electronic structure and molecular geometry of fairly long polythiophene (PT) chains by blending deterministic method based on a trial single particle density matrix $[P^0 (R)]$ for the guessed structural parameters $(R)$, which is allowed to evolve under a unitary transformation generated by the Hamiltonian $H(R)$, with previously developed evolutionary computing technique (CARMHC). The scale $(\lambda)$ of the transformation is optimized by making energy $[E(\lambda)]$ stationary with respect to $\lambda$. The resulting method has been called $\lambda$ - optimized Directed Random Mutation Hill Climbing algorithm (DRMHC). Thus the technique we have developed is electron-density based, and makes use of a non-deterministic search procedure, augmented with a deterministic bias. The algorithm is further utilized to calculate the electronic structures of a series of polaron-doped, bipolaron-doped and electron-doped PT chains. All the dopings significantly lower the transportation band gap. Two electron doping has almost the same effect as one bipolaron doping. Distortion energy of formation for two polarons and one bipolaron are the same, but the decrease in ionization energy is much more pronounced in the case of bipolarons than for two polarons. So one bipolaron is thermodynamically more stable than two polarons in these systems despite the coulomb repulsion between two like charges. So essentially it is not the spin-carrying polaron pairs, but the spinless bipolaron which is held responsible for band gap lowering in doped PT systems. We observed that bipolaron doped PT structures are stable, have low band gap, but may not be metallic, even at the saturation level of doping.

 

Designing Clues for Low Band Gap Organic Materials : The same density matrix-based random mutation hill climbing algorithm is invoked under the constraint that the band gap measured by HOMO-LUMO energy difference is also a minimum in each case. The minimization of energy of a polythiophene or polyselenophene oligomer under the zero band gap constraint is a multiobjective optimization problem and is not easy-to-handle by traditional means. So to test the efficiency of density matrix based $\lambda$ - optimized Directed Random Mutation Hill Climbing algorithm, we pose the inverse problem theoretically. We propose to probe whether imposition of zero band gap condition is theoretically associated with the emergence of uniquely identifiable structural changes in the systems under consideration, and if these structures match with the hole doped structures. This is ensured by constructing and minimizing an objective function that is a weighted sum of $(E-E_L)^2$ and $(\epsilon_{lumo}-\epsilon_{homo})^2$, $E_L$ being an updatable lower bound to the energy. For the accepted mutations the mutated system Hamiltonian is diagonalized, and the one-electron density matrix and the band gap are used to redefine the objective function which is sought to be further minimized by `mutation'. The Pareto fronts are constructed and the optimal weightage of the two constraints are determined and the electronic molecular structures at the optimally constrained level are analyzed. We find that the zero band gap constraint automatically leads to the emergence of electron deficient quinonoid structures embedded in the aromatic chain. It turns out that hole doping also creates such quinonoid structures leading to the transport band gap reduction. The zero band gap constraint or any specific property constrained optimization by the proposed method can therefore be used as a designing tool for molecule based materials. The method can in principle, be used in conjunction with ab-initio framework to electronic structure calculations leading to designer molecule.

 

A hybrid PSO-CARMHC Algorithm: Exploring Fuzzy Adaptation : The CARMHC global optimizer is good for locating the minimum energy structures of strongly coupled finite ionic clusters, systems dominated by long-range forces. The CARMHC algorithm has also successfully located all the lowest known minima up to LJ$_{50}$ from random starting points. However, multiple runs are often required when relatively large clusters ($n > 27$) are encountered. The elapsed number of generations in computing the minimum energy structures increases with increasing complexity of the PES. So we have focused on the possible means of hybridization of CARMHC algorithm with Particle Swarm Optimization (PSO) with a view to enhancing the capability of the search heuristic. A fuzzy logic based adaptive search for optimal parameters of the search is explored. Fuzzy logic and Evolutionary Computing are the two most powerful weapons in the arsenal of Soft Computing to deal with highly complex physical and chemical problems. Each of them excels in solving a certain types of problem. However, there are certain classes of problems in which neither of the two can optimally perform alone. Almost any control system can be replaced with a fuzzy logic based control system. This may be overkill in many places; however, it simplifies the design of many more complicated cases. So fuzzy logic is not the answer to everything, it must be used when appropriate to provide better control. Evolutionary computing has very high success rate over a wide range of problems, but cannot be applied to all problems with efficiency equal or better than the domain-specific optimization techniques. Thus it is not a panacea for optimization problems -- "no free lunch theorem". Obviously the evaluation function does give information to the EC, but that information does not give a direction for adaptive change for each bit-string evaluated, but rather it tells one just how well each bit-string performed when evaluated. The result passed back to the EC does not give the EC any insight. It merely passes back a scalar number, which when compared to other scalar numbers, forms a ranking of the bit strings. Besides, in general EC maintains a diverse population of strings, which results in a tremendously high-cost overhead. The creation of effective artificial intelligent system to bypass the slowness of the evolutionary process towards the end of the search using different tools of Soft-computing is a very important area of research. An attractive way to attain this goal is the combination between fuzzy logic controller and evolutionary optimization methods. We have developed fuzzy particle swarm optimization, in which inertia weight is adaptively adjusted in an intelligent way according to the control information translated from the fuzzy logic controller during the search process. So the developed search meta-heuristic consists of two stages, the creation of new individuals and their evaluation through PSO and aesthetic judgment of the velocity by fuzzy logic from the available information to move in the search space. The great majority of the best known hybrid algorithms of cluster geometry optimization, are the hybridization of a global search and a local search methods and usually is more effective to solve the large scale problem than the global search method or local search methods working alone. This led us to experiment with the hybridization of previously developed CARMRHC algorithm with PSO. To be specific we intend to hybridize the fast converging nature and broader exploration of PSO (with fuzzy control) in the beginning of a search, refined pinpointing search abilities of CARMHC and the directive nature of steepest descent method. We demonstrated its working principles on some benchmark Lennard-Jones cluster optimization problems, which are notoriously well known to present difficult search landscapes.

 

Fuzzy Adaptive RMHC: Thereafter we thought of fuzzy controlling of a single string based evolutionary optimizer for fitness maximization to further reduce the computational cost. Although this approach works well with smaller clusters (both single component and bi-metallic) interacting through many-body Gupta potential, required more insight to deal with larger clusters. So the immediate plan is further development of that single string based strategy through fuzzy control to handle energy landscape with high dimension, with a view to reducing the computational expenses to a great extent.

 

The future plans include the inclusion of the Soft Probability to cover all the types of discrete, continuous, and conditional probabilities with the aim of developing intelligent Soft-computing techniques. We also intend to work with the integration of Neural and Evolutionary Computing to provide the search heuristic with self-learning and self-tuning capabilities. The ultimate goal will be to develop a new AI-based networking technology where the network may be modeled and built as an intelligent adaptive autonomous system and the nodes subscribing to the network act as intelligent agents with distributed control capabilities, optimization goals and mechanisms, and apply neural-fuzzy-evolutionary computing to intelligently capture the uncertainty and chaos of the dynamical environment in order to offer quality solutions at low cost.

 

 

 

Publications

 

1) Kanchan Sarkar, R. Sharma, S. P. Bhattacharyya, Blending determinism with evolutionary com-puting: Applications to the calculation of the molecular electronic structure of polythiophene. Journal of Chemical Theory and Computation 6, 3, 718–726 (2010).

 

2) Kanchan Sarkar, N. K. Datta, M. Ghosh, Frequency dependent linear and non-linear response prop-erties of electron impurity doped quantum dots: Influence of impurity location. Physica E: Low-dimensionalSystems and Nanostructures 42, 5, 1659 – 1666 (2010).

 

3) Kanchan Sarkar, N. K. Datta, M. Ghosh, Dynamics of Electron Impurity Doped Quantum Dots in thepresence of Time-Varying Fields: Influence of Impurity Location. Physica E: Low-dimensional Systems andNanostructures 43, 1, 345 – 353 (2010).

 

4) Kanchan Sarkar, N. K. Datta, M. Ghosh, Interplay between Size and Impurity Position of DopedQuantum Dot. Superlattices and Microstructures 50, 1, 69 – 79 (2011).

 

5) Kanchan Sarkar, R. Sharma, S. P. Bhattacharyya, Designing Clues for Low Band Gap OrganicMaterials : A Constrained Variational Approach. International Journal of Quantum Chemistry 112, 6, 1547–1558 (2012).

 

6) Kanchan Sarkar, S. P. Bhattacharyya, Computationally efficient algorithm in cluster geometry op-timization, in AIP Conference Proceedings/ Volume 1512/ B. Soft Condensed Matter Including BiologicalSystems, edited by A. K. Chauhan, C. Murli, and S. C. Gadkari, pages 162–163, AIP, 2013.

 

7) Kanchan Sarkar, S. P. Bhattacharyya, Exploring New Computing Paradigms in Theoretical Chem-istry. Journal of Indian Chemical Society, 90, 879-889, July 2013 (Invited, Based on Professor J. N. MukherjeeMemorial Lecture delivered in ’Acharya Prafulla Chandra Ray Memorial Symposium on Chemistry and In-dustry (2012)’ organized by the Indian Chemical Society held at Kolkata on August 02, 2012)

 

8) Kanchan Sarkar, S. P. Bhattacharyya, Adaptive Mutation-driven Search for Global Minima in 3-DCoulomb Clusters: A new Method with Preliminary Applications. Proceedings of the Second InternationalConference on Soft Computing for Problem Solving (SocProS), to be published (October 15, 2013) in Advancesin Intelligent Systems and Computing Series, Vol. 236 of Springer.

 

9) Kanchan Sarkar, S. P. Bhattacharyya, Single String Based Global Optimizer for Geometry Optimiza-tion in Strongly Coupled Finite Clusters: an Adaptive Mutation-Driven Strategy. The Journal of Chemical Physics 139 , 7, 074106 (2013).

Works in Progress

 

1) Kanchan Sarkar, S. P. Bhattacharyya, Fuzzy Adaptive Control in RMHC Driven Search for Global Minima in Selected Metal Clusters. (submitted: CTTC).

 

2) Kanchan Sarkar, S. P. Bhattacharyya, A hybrid PSO-CARMHC Algorithm for Global Optimization: Exploring Fuzzy Adaptation. (to be communicated : JCP).

 

3) Kanchan Sarkar, S. P. Bhattacharyya, Coulomb Clusters under Anisotropic Confinement : Structureand Energetics of one, two and three component clusters. (to be communicated : PRB).

 

4) Kanchan Sarkar, S. P. Bhattacharyya, Evolutionary Algorithms in Computing Electronic Structure of Atoms, Molecules and Clusters. (to be communicated : ).

 

Conferences Attained

 

⇒ National Symposium on Quantum Chemistry, Soft Computation & Optimization (April 04-05, 2008) atIndian Association for the Cultivation of Science (IACS), Kolkata, India.

⇒ Changing Paradigms in Theoretical and Computational Chemistry: From Atoms to Molecular Clusters.(December 18-20, 2009), University of Pune, India.

⇒ Recent Trends in Atomic and Molecular Physics Research (February 13, 2010) at Ramakrishna MissionVivekananda University, Belur Math, India.

⇒ Recent Developments & Trends in Computational Chemistry (March 12-13, 2010) at Department ofChemistry, NEHU, Shillong, India.

⇒ International Conference on Physics of Novel and Emerging Materials (November 15-17, 2011) at IndianAssociation for the Cultivation of Science (IACS), Kolkata, India.

⇒ International Symposium on Chemistry and Complexity (December 6-8, 2011) at Indian Association forthe Cultivation of Science (IACS), Kolkata, India.

⇒ 57th DAE Solid State Physics Symposium DAE-SSPS 2012, (December 3 - 7, 2012), at Indian Instituteof Technology Bombay, Mumbai, India.

⇒ 2nd International conference on soft computing for problem solving (December 28-30, 2012) at JKLakshmipat University, Jaipur, India.

⇒ International Symposium on Molecular Organization and Complexity : A Chemical Perspective (IS-MOC), Feb 6-8, 2013, Department of Chemistry, University of Calcutta.

⇒ An International Conference on: Electronic Structure and Dynamics of Molecules and Clusters (ESDMC-2013), (February 17-20, 2013), IACS, Kolkata, India.

 

Awards/Honours

 

1) Rank in the West Bengal Joint Entrance examination (Medical) : 713

2) Awarded Junior Research Fellowship & Senior Research Fellowship by Council of

    Scientific and Industrial Research, New Delhi, India  (2007 & 2009).

3) Best Project in Indian Statistical Institute (2009) (CSC&MIU)

4) Poster Prize on the Conference "Changing Paradigms in Theoretical and

    Computational Chemistry: From Atoms to Molecular Clusters". (December 18-20,  

    2009), at University of Pune.

 

 

 

Advance Courses Completed

 

1)   Advance Quantum Mechanics (IACS)

2)   Group Theory in Quantum Mechanics (IACS)

3)   Computer Languages : C, Matlab, Fortran77 & 95, Latex (ISI)

4)   Pattern Recognition (ISI)

5)   Data Mining (ISI)

6)   Digital Image Processing (ISI)

7)   Neural Networks (ISI)

8)   Fuzzy Sets (ISI)

9)   Rough Sets (ISI)

10) Evolutionary Computation (ISI)

Project Completed

 

1) Data Mining Optimization in an Image Database

2) Image Processing on different band of Images of Calcutta

3) Simulation of a Single Layer Perceptron

4) Determination of Transition Temperature in an Ising Model System by using

     Genetic Algorithm

5) Branch and Bound Algorithm for feature selection

 

 

 

Project Investigated

 

 

1) Determination of Minimum Potential Energy Configuration of 'N' Point Charges 

    on the Surface of a Sphere using Genetic Algorithm.

    By Swarnendu Bhattacharyya

          Siksha Bhavana, Visva Bharati University (Central University)

 

2) Exploring Structures of Confined Coulomb Clusters by a Directed Random    

     Mutation Hill Climbing Method.

     By Subrahmanyam Sappati

           School of Chemistry, University of Hyderabad (Central University)

 

3) Dynamic Optimization by Deepanjan Sarkar, IIT, Chennai

Extracurricular Activity

Maintaining Open Software, playing |cricket, chess, Karom, cards, Badminton|  travelling and Photography.

KEY I.T. SKILLS

Word, Excel, Access, PowerPoint, Adobe Photoshop

Software Packages Handled

Gaussian, Gamess, Mopac

Operating System Handled

Windows (XP, Vista, 7), Unix (Fedora, Debian, Ubuntu, Redhat Enterprise, Solaris, Macintosh, Centos, Rocks, Scientific Linux, SUSE Linux)

Programming Languages

Fortran (77,90), C, MATLAB, Latex