Molecular Control Engineering Nonlinear Control at the Nanoscale Raj Chakrabarti PSE Seminar Feb 8, 2013 What is Molecular Control Engineering? Control engineering: Manipulation of system dynamics through nonequilibrium modeling and optimization. Inputs and outputs are macroscopic variables. Molecular control engineering: Control of chemical phenomena through microscopic inputs and chemical physics modeling. Adapts to changes in the laws of Nature at these length and time scales. Aims Reaching ultimate limits on product selectivity Reaching ultimate limits on sustainability Emulation of and improvement upon Nature’s strategies Approaches to Molecular Design and Control Quantum Control of Chemical Reaction Dynamics Control of Biochemical Reaction Networks Molecular Design femtoseconds, angstroms picoseconds, nanometers milliseconds, micrometers Parallel Parking and Nonlinear Control Stepping on gas not enough: can’t move directly in direction of interest Must change directions repeatedly Left, Forward + Right, Reverse enough in most situations Tight spots: Move perpendicular to curb through sequences composed of Left, Forward + Left, Reverse + Right, Forward + Right, Reverse Vector Fields 8. Finalize these Control with Linear Vector Fields Lie Brackets and Directions of Motion From classical control to the coherent control of chemical processes FMO photosynthetic protein complex transports solar energy with ~100% efficiency Phase coherent oscillations in excitonic transport: exploit wave interference Biology exploits changes in the laws of nature in control strategy: can we? Coherent Control versus Catalysis Potential Energy Surface with two competing reaction channels Saddle points separate products from reactants Dynamically reshape the wavepacket traveling on the PES to maximize the probability of a transition into the desired product channel probability density time interatomic distance C. Brif, R. Chakrabarti and H. Rabitz, New J. Physics, 2010. C. Brif, R. Chakrabarti and H. Rabitz, Control of Quantum Phenomena. Advances in Chemical Physics, 2011. Femtosecond Quantum Control Laser Setup 2011: An NSF funded quantum control experiment collaboration between Purdue’s Andy Weiner (a founder of fs pulse shaping) and Chakrabarti Group Prospects and Challenges for Quantum Control Engineering Coherent Control of State Transitions in Atomic Rubidium Bilinear and Affine Control Engineering R. Chakrabarti, R. Wu and H. Rabitz, Quantum Multiobservable Control. Phys. Rev. A, 2008. Few-Parameter Control of Quantum Dynamics Conventional strategies based on excitation with resonant frequencies fails to achieve maximal population transfer to desired channels Selectivity is poor; more directions of motion are needed to avoid undesired states Optimal Control of Quantum Dynamics Shaped laser pulse generates all directions necessary for steering system toward target state Exploits wave-particle duality to achieve maximal selectivity, like coherent control of photosynthesis Understanding Interferences U I (t ) IN i ( ) n i V (t) dt ( I 0 t t 0 0 c (t ) j | 1 ji t i t n 1 0 t 0 i ) 2 t t V (t)V (t) dtdt 0 0 I I VI (t )VI (t ) VI (t n ) dt n dt Need to introduce V_ VI (t ) dt | i Remove the lambdas i c (t ) j | ( ) n ji n t t 1 0 0 n VI (t ) VI (t n ) dt n dt | i We don’t show the interm we for consistency w belo 9. Finalize these Quantum Interferences and Quantum Steering T N T t2 N N T t3 t2 0 l 1 0 0 j 1 k 1 0 0 0 U ba (T ) vba (t1 )dt1 vbl (t2 ) vla (t1 )dt1dt2 vbj (t3 ) v jk (t2 ) vka (t1 )dt1dt2dt3 6 5 4 3 2 1 | c (T ) c (T ) | 1 ba 2 ba 2 * 1 2 c (T ) c (T ) cba (T ) cba (T ) 1 1 2 | cba (T ) |2 2 Re[cba (T )cba (T )*] | cb2a (T ) |2 1 ba 2 ba Interference 6 5 4 3 2 1 • Mechanism identification techniques have been devised to efficiently extract important constructive and destructive interferences V. Bhutoria, A. Koswara and R. Chakrabarti, Quantum Gate Control Mechanism Identification, in preparation Control of Molecular Dynamics HCl CO 1 E1 diag exp N Ej kT exp j 1 kT Mixed state density matrix: (0) Pure state: (0) | (0) (0) | diag (1, 0, Expectation value of observable: F (U (T )) Tr (U (T ) (0)U † (T )O) Cost functional: J ( (·)) F (U (·) (T )) , E , exp N kT , 0) R. Chakrabarti, R. Wu and H. Rabitz, Quantum Pareto Optimal Control. Phys. Rev. A, 2008. Quantum System Learning Control: Critical Topology J i Tr O(T ), (0) (t ) 0, (t ) O(T ), (0) 0 2 N 2 2J H (t , t ') l (t ) l (t ') (t ) (t ) l 1 R. Wu, R. Chakrabarti and H. Rabitz, Critical Topology for Optimization on the Symplectic Group. J Opt. Theory, 2009 R. Chakrabarti and H. Rabitz, Quantum Control Landscapes, Int. Rev. Phys. Chem., 2007 K.W. Moore, R. Chakrabarti, G. Riviello and H. Rabitz, Search Complexity and Resource Scaling for the Quantum Optimal Control of Unitary Transformations. Phys. Rev. A, 2011. Quantum Robust Control R. Chakrabarti and A. Ghosh. Optimal State Estimation of Controllable Quantum Dynamical Systems. Phys. Rev. A, 2011. Improving quantum control robustness Check sign, fix in From Quantum Control to Bionetwork Control • Nature has also devised remarkable catalysts through molecular design / evolution • Maximizing kcat/Km of a given enzyme does not always maximize the fitness of a network of enzymes and substrates • More generally, modulate enzyme activities in real time to achieve maximal fitness or selectivity of chemical products The Polymerase Chain Reaction: An example of bionetwork control Nobel Prize in Chemistry 1994; one of the most cited papers in Science (12757 citations in Science alone) Produce millions of DNA molecules starting from one (geometric growth) Used every day in every Biochemistry and Molecular Biology lab ( Diagnosis, Genome Sequencing, Gene Expression, etc.) Generality of biomolecular amplification: propagation of molecular information - a key feature of living, replicating systems Single Strand – Primer Duplex Extension D S1 S2 k1m , k2m DNA Melting DNA Melting Again S1 P1 S1P1 k11 ,k21 k1 ,k2 S2 P2 S2 P2 2 2 Primer Annealing ke ,k e SP E E.SP k n , k n E.SP N [ E.SP.N ] kcat E.D1 k n , k n E.D1 N [ E.D1.N ] kcat E.D2 . E.DN E DNA kcat ' k11t ,k21 t S1 S2 DNA 3/19/2016 School of Chemical Engineering, Purdue University 27 The DNA Amplification Control Problem and Cancer Diagnostics Wild Type DNA Mutated DNA Can’t maximize concentration of target DNA sequence by maximizing any individual kinetic parameter Analogy between a) exiting a tight parking spot b) maximizing the concentration of one DNA sequence in the presence of single nucleotide polymorphisms PCR Temperature Control Model Sequence-dependent annealing DNA targets Cycling protocol Sequence-dependent Model Development f r S1 S2 D k ,k Reaction Equilibrium Information G k f / kr K exp RT Relaxation Time ΔG – From Nearest Neighbor Model Similar to the Time constant in Process Control 1 kr k f CS 1eq CS 2eq τ – Relaxation time (Theoretical/Experimental) Solve above equations to obtain rate constants K. Marimuthu and R. Chakrabarti, Sequence-Dependent Modeling of DNA Hybridization Kinetics: Deterministic and Stochastic Theory, in preparation Sequence-dependent rate constant prediction σ – Nucleation constant for resistance to form the first base pair The forward rate constant is a fixed parameter Estimate σ, forward rate constant offline based on our experimental data Compute and hence kf, kr for a given DNA sequence using S. Moorthy, K. Marimuthu and R. Chakrabarti, in preparation Variation of rate constants Flow representation of standard PCR cycling (t1 , , t30 )( p) fu3 ct30 f u2 bt30 cycle 30 u1 fu1 at30 fu3 ct1 fu2 bt1 ( p) fu1 at1 cycle 1 [0.00, 3340, 0.00, 3340, 0.00, 0.00, 0.00]T u2 [30.0, 5.95, 30.0, 5.95, 0.04, 0.62, 53.5]T u3 [0.00, 19.0, 0.00, 19.0, 0.01, 0.90, 275]T u U {u1 , u2 , u3} Choose times t1 , , t30 : Lie brackets, analogy to parallel parking From standard to generalized PCR cycling Accessibility May mention reachable set here rather than above move / send to backup R ( p) tukk tu11 ( p), k 1, u1 , 6. Decide what to sho , uk U , t1 , , tk 0 u1 L span{[ f u1 , ,[ f uk 1 , f uk ]]} L span{[hik , ,[hi2 , hi1 ]]}, hi { fu1 , L span{[him , L span{[him , ,[hi2 , hi1 ]]}, hi { fu1 , Specify controls in finite set u2 u3 u , f um } , f uk } ,[hi2 , hi1 ]]}, hi { f ( y ), g ( y )} May show affine extension state equations in u,f,g format T dU T ( s ) [ (t )] T [ 0 ds Then transition to full OCT – for nonlinear problem, application of vector fields in arbitrary com PCR gradient, mentioning PMP and definition of \phi(t) (can then indicate below that gradien Project flow w Gramian in terms of \phi(t) – for comments on model-free learning control of Optimal Control of DNA Amplification Min CDNA t f C T (t ) st max DNA 2 dx f x, u dt x CS1 , CS2 ,.....CE .D1 .....CDNA Tr For N nucleotide template – 2N + 13 state equations Typically N ~ 103 R. Chakrabarti et al. Optimal Control of Evolutionary Dynamics, Phys. Rev. Lett., 2008 K. Marimuthu and R. Chakrabarti, Optimally Controlled DNA amplification, in preparation Optimal control of PCR 95 90 85 Temperature in Deg C 80 75 70 65 60 55 50 45 -10 10 30 50 70 Time in Seconds 90 110 130 150 Optimal control of PCR 95 90 85 Temperature in Deg C 80 Minimal time control? Apply Lagrange cost 75 70 T L 65 dt 0 60 55 50 45 0 20 40 Annealing Time = 10 s 60 80 100 Time in Seconds Annealing time = 12 s 120 140 Annealing time = 15 s Optimal control of PCR 1 90 0.9 85 0.8 80 0.7 75 0.6 70 0.5 65 0.4 60 0.3 55 0.2 50 0.1 45 0 0 20 40 Annealing Time = 10 s 60 80 100 Time in Seconds Annealing time = 12 s 120 140 Annealing time = 15 s Efficiency Temperature in Deg C 95 Optimal control of PCR 1 90 0.9 85 0.8 80 0.7 75 0.6 70 Competitive problems? 0.5 65 Check rank of Gramian 0.4 60 0.3 55 0.2 50 0.1 45 0 0 20 40 Annealing Time = 10 s 60 80 100 Time in Seconds Annealing time = 12 s 120 140 Annealing time = 15 s Efficiency Temperature in Deg C 95 Optimal control of PCR 95 90 85 Cycle 1 Temperature in Deg C 80 Cycle 2 75 70 Geometric growth: after 15 cycles, DNA concentrations are 65 60 red – 4×10-10 M blue – 8×10-9 M green – 2×10-8 M 55 50 45 0 20 40 Annealing Time = 10 s 60 80 100 Time in Seconds Annealing time = 12 s 120 140 Annealing time = 15 s Technology Development for Control of Molecular Amplification Next steps: application of nonlinear programming dynamic optimization strategies for longer sequences, competitive problems Future work: robust control, real-time feedback control using parameter distributions we obtain from experiments Summary • Can reach ultimate limits in sustainable and selective chemical engineering through advanced dynamical control strategies at the nanoscale • Requires balance of systems strategies and chemical physics • New approaches to the integration of computational and experimental design are being developed Reviews of our work Quantum control R. Chakrabarti and H. Rabitz, “Quantum Control Landscapes”, Int. Rev. Phys. Chem., 2007 C. Brif, R. Chakrabarti and H. Rabitz, “Control of Quantum Phenomena” New Journal of Physics, 2010; Advances in Chemical Physics, 2011 R. Chakrabarti and H. Rabitz, Quantum Control and Quantum Estimation Theory, Invited Book, Taylor and Francis, in preparation. Bionetwork Control and Biomolecular Design “Progress in Computational Protein Design”, Curr. Opin. Biotech., 2007 “Do-it-yourself-enzymes”, Nature Chem. Biol., 2008 R. Chakrabarti in PCR Technology: Current Innovations, CRC Press, 2003. Media Coverage of Evolutionary Control Theory: The Scientist, 2008. Princeton U Press Releases Pathway Examples • 6 level system, Pif transition 1 4 6 5 – (i) Amplitude of 2nd order pathway via state 2: 4 3 2 T t2 2( 2) U 41 v42 (t2 )v21 (t1 )dt1dt2 1 (i) 0 0 6 – (ii) Transition amplitude for 3rd order pathway 1 2 5 5 4 4 3 T t3 t 2 3( 2 , 5 ) U 41 v45 (t3 )v52 (t2 )v21 (t1 )dt1dt2 dt3 0 0 0 2 (ii) 1 Interference Identification VI ( s) H I exp(i s) H(t,s) U (T ) ı b | n T 0 t2 0 VI (tn ) VI (t1 )dt1 U ba (T , s) U ban exp(in s) n 1 Uba (T , ) Uba (T , s) exp(i s) ds Uba(T) Decode dU I (t , s ) iVI ( s)U I (t , s) dt n ba Normal Dynamics Encode H(t) Encoded Dynamics dtn | a Uba(T,s) Linear Programming Formulation: Observable Max Quantum observable maximization: R† R diag{ 1 , ,1 ; ; r , p1 S †S diag{1 , J (U ) |U ij |2 i j i, j pr , 1 ; q1 , r }, ; s , , s } qs J (U ) Tr[( R†US ) ( R†US )] Tr (U U †) N |U j 1 ij |2 1; i 1, ,N ij |2 1; ,N N | U i 1 j 1, Translation to linear programming: J (U ) J (x) cT x x(i 1) N j |Uij |2 Me c(i 1) N j i j Ax b bi 1; i 1, , 2N 1 K. Moore, R. Chakrabarti, G. Riviello and H. Rabitz, Search Complexity and Resource Scaling for Quantum Control of Unitary Transformations. Phys. Rev. A, 2010 The analogy to the “assignment problem” Maximum weighted bipartite matching (assignment prob): Given N agents and N tasks Any agent can be assigned to perform any task, incurring some cost depending on assignment Goal: perform all tasks by assigning exactly one agent to each task so as to maximize/minimize total cost N N max cij xij , i 1 j 1 N x i 1 ij 1, N x j 1 ij 1, xij 0, cij Foundation for Quantum System Learning Control. II: Geometry of Search Space 5. Maximum weighted bip • Maximum weighted bipartite matching of prob):Would need to menti indicate the two examples \gamma_i,\lambda_j slide, then show projected one matrix G_thick, indicat ( s, t ) compatibility cond’n, and in iTr T† ( s )F ( T ( s )) (t ) start from points within pol t vertex (do not need to draw T d T ( s ) [ (t )] T [ (t )] dt F ( T ( s)) 0 ds Replace w dxT ( s ) 1 M F ( xT ( s )) • Birkhoff polytope: ds • flows start from points within polytopes and proceed toM: in on p optimal vertex R. Chakrabarti and R.B. Wu, Riemannian Geometry of the Quantum Observable Control Problem, 2013, in preparation. R. Chakrabarti, Notions of Local Controllability and Optimal Feedforward Control for Quantum Systems. J. Physics A: Mathematical and Theoretical, 2011. Quantum Estimation Sequence-dependent rate constant prediction dx Ax dt Negative reciprocal of the maximum Eigenvalue is the Relaxation time. Kinetic rate constant control: general formulation Kinetic rate constant control m dx ui gi ( x ) •general formulation dt of i 1 rate constant control x [ x1, , xn ]T ; xi 0,i 1, , n u [u1 , , um ]T ; ui 0, i 1, , m •temperature control formulation u2 um 1 k 2u1 k mu1m 1 T (t ) Ea ,1 u (t ) R ln k1 3. Use beamer for now? Finalize Ea ,i 1 i Ea1 Decide wheth not essential