MIT OpenCourseWare http://ocw.mit.edu 5.80 Small-Molecule Spectroscopy and Dynamics Fall 2008 For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms. 5.80 Lecture #1 Fall, 2008 Page 1 Lecture #1: Matrices are Useful in Spectroscopic Theory For next time, read handout on the Van Vleck transformation and look at the notes on coupled harmonic oscillators. Outline * Correspondences bra operator * * row column N × 1 column vector ⎛ ⎞ ψi → i = ⎜ ⎟ ⎝ ⎠ ket * ψi → i = 1 × N →O O row vector N × N square matrix Solution of Schrödinger Equation corresponds to solving a determinantal secular equation for the {Ek} ψ ; The {Ek} are the H kk in the special “diagonal” representation of H * The {ψk} are obtained from the basis states {φk} as columns of the unitary matrix T from the similarity transformation T†HφT = Hψ that “diagonalizes” Hφ; * Matrix representation of arbitrary f(Q). Usual procedure to obtain a fit model. a exact H differential operator a complete set of basis functions Approximate H perturbation Fit model theory an infinite * finite matrix simplified e.g. Born- expressed Van Vleck a finite matrix matrix Oppenheimer in terms of transformation expressed in combine OR and normal structure ↓ linearly terms of * algebraic mode dependent parameters introduces microscopic formulas separations e.g. fij force many small model molecular expressed in of variables constants parameters constants terms of terms effective not necessarily molecular the same even constants if they have same name Dunham constants ωe, ωexe exact H truncate H 5.80 Lecture #1 Fall, 2008 Page 2 Matrices are useful because: * they display all necessary information; * can be “read” and simplified by perturbation theory “order sorting” via * labor saving tricks for avoiding the evaluation of unnecessary integrals, such as formulas in the Harmonic Oscillator basis set. and P for all matrix elements of Q No integrals actually evaluated. No functions actually looked at. H ij ; E − E oj o i Quantum Mechanics Operators follow the rules of matrix multiplication. B ) ψ dτ = e.g. ∫ ψ*i ( A ∑ j ( ∫ ψ Aψ dτ ) ( ∫ ψ Bψ dτ ) k * i k * k j completeness of {ψ} = ∑ A ik Bkj = (AB) ij k This is very useful because we can generate many matrices by simple operations on one matrix. E.g. Q = R – Re V(Q) = ∑ cn Qn n n matrix of Q (Qn) = (Q)n Q × Q × … Q so instead of evaluating Q1, Q2, Q3, etc. we just evaluate Q and derive all the rest by matrix operations. ↑ formulas, not integrals! ⎡* shortcuts to selection rules There is even some diagrammatic insight ⎢* calculations of a specific element of Qn ⎣ 5.80 Lecture #1 Fall, 2008 0 0 0 Q= 0 0 0 Page 3 0 0 0 0 0 0 0 (Q 2 ) = 0 0 0 0 0 0 v = ±1 non-zero matrix elements of Q 0 0 v = 0, ±2 selection rules At the end of this lecture we will see that we are not restricted to integer powers of Q. etc. Suppose we have a convenient and complete (orthonormal) basis set {i}. ) can be expanded in terms of the {}. Any arbitrary function (including an eigenfunction of H wavefunction picture N k = a i can be k i *i k d = a ki i=1 matrix picture *k = U U is a transformation that converts {} into {}. 0 0 k k = 1 1 i = i 0 0 k = 0…1…0 N 1 * i i = 0…1…0 column matrix 5.80 Lecture #1 Fall, 2008 Page 4 N We know ψk = ∑ a ki φ i i=1 We want ψ = Uφ where U is N × N matrix that transforms φ into ψ. By this we mean ⎛ 0 ⎞ ⎛ U k1 ⎞ ⎛ 0 ⎞ ⎛ 0 ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ 0 ⎟ ⎜ U k 2 ⎟ ⎜ ⎟ ψk = ⎜1 ⎟ = ∑ U ki φ i = ⎜ ⎟ + ⎜ ⎟ + … ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ i ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ 0 0 0 U ⎝ ⎠ ψ ⎝ ⎠ φ ⎝ ⎠ φ ⎝ kN ⎠ φ k-th row of U ⎛ 0 ⎞ ⎛ U k1 ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜1 ⎟ = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ 0 U ⎝ ⎠ ψ ⎝ kN ⎠ φ U ki = a ki = ∫ φ*iψk dτ We can go in the opposite direction U–1ψ = φ U −ij1 = U*ji and show that U–1 = U† φi is the i-th row of U–1 or the i-th column of U*. 5.80 Lecture #1 Fall, 2008 Page 5 Derivation of Secular Determinant Now for every operator there is a matrix representation. ⎛ O11 O1N ⎞ ⎟ → O = ⎜ O ⎟ ⎜ ⎜ ⎟ ⎝ O N1 O NN ⎠ φ φ dτ O ij = ∫ φ*i O j Schrödinger Equation in ψ eigenbasis picture ψ = E ψ H k k k mixing coefficient ∞ ψk = ∑ a ki φ i basis function - convenient usually defined as eigenfunctions of a part called H ° of H °φ = E °φ H i i i i=1 Matrix notation: H φij = ∫ φ dτ = φ*i H j φ iH j φ 5.80 Lecture #1 Fall, 2008 Page 6 φ φ ⎞ ⎛ k ⎞ ⎛ H11 ⎛ a1k ⎞ H1N a1 ⎜ φ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ H 21 ⎟ ⎜ ⎟ = E ⎜ ⎟ k⎜ ⎜ ⎟ ⎜ ⎟ ⎟ ⎜ φ ⎟ ⎜ k ⎟ ⎜ k ⎟ ⎝ H N1 ⎠ φ ⎝ a N ⎠ φ ⎝ a N ⎠ φ convenient representation ⎛ φ ⎜∑ H1i ⎜ i ⎜ ⎜∑ H 2 i LHS = ⎜ i ⎜ ⎜ ⎜ ⎜∑ H Ni ⎝ i initially unknown mixing coefficients initially unknown energies ⎞ a ki ⎟ ⎟ ⎛ a1k ⎞ ⎟ ⎜ ⎟ a ki ⎟ ⎜ ⎟ ⎟ = Ek ⎜ ⎟ ⎟ ⎜ k ⎟ ⎟ ⎝ a N ⎠ ⎟ a ki ⎟ ⎠ move everything into one column matrix ⎛ ⎞ k k ⎞ ⎜ ∑ ( H1i a i − δ1i E k a i ) ⎟ ⎛ k ⎜ i ⎟ ⎜∑ a i ( H1i − δ1i E k ) ⎟ ⎛ 0 ⎞ ⎟ ⎜ 0 ⎟ ⎜ ⎟ ⎜ i k k ⎜ ⎟ = ⎜ ⎟ ⎜∑ ( H 2i a i − δ 2i E k a i ) ⎟ = etc. ⎜ ⎟ ⎜ ⎟ ⎜ i ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ etc. ⎜ ⎟ ⎝ 0 ⎠ ⎜⎜ ⎠ ⎟⎟ ⎝ ⎝ ⎠ A system of N linear homogeneous equations in N unknowns ( a ki i = 1…N ) . A nontrivial solution exists if the determinant of coefficients of the unknown {a ki } is zero. H11 − E k 0 = H 21 H12 H 22 − E k H1N unit matrix φ = |H –1E k| HNN − E k 5.80 Lecture #1 Fall, 2008 Page 7 This is the secular determinant. Must solve for special values of Ek which satisfy the requirement 0 = |Hφ – 1Ek|. . {E } k = 1, 2, … N These are eigenvalues of H k COMPUTERS! 1. start with complete set {φ} 2. compute all Hφ matrix elements 3. “diagonalize” Hφ time required ∝ N3 4. solve for {a ki } : one set of N coefficients for each of the N eigenvalues. How to “diagonalize” Hφ? Seek a similarity transformation T. ψ ⎛ H11 ⎞ ⎜ ⎟ ψ H 22 ⎜ ⎟ −1 φ φ T H T=H = ⎜ ⎟ ⎜ ⎟ ψ H NN ⎝ ⎠ 0 0 H iiψ ≡ E k are the eigenvalues we seek. Computer generates T iteratively (Jacobi Rotations, see handout) Special properties: * all Quantum Mechanical operators (that correspond to real observables) are “Hermitian” (real eigenvalues) which means that their matrix representations have the property O = O† Oij = O*ji * T is “unitary” T†T = 1 (i.e., T† = T–1) 5.80 Lecture #1 Fall, 2008 Now I’ll show that T contains the information we seek about the {a ki } . ⎛ 0 ⎞ ⎛ 0 ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ψ⎜ ⎟ † φ φ T H T = H where H 1 = E k ⎜1 ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎝ 0 ⎠ ψ ⎝ 0 ⎠ ψ ψk left multiply by T T†HφTψk = Ekψk TT† = 1 TT †H φ Tψk = TE k ψk H φ ( Tψk ) = E k ( Tψk ) an eigenvector of Hφ an eigenvalue equation ⎛ 0 ⎞ ⎛ T1k ⎞ ⎛ a1k ⎞ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ Tψk = T ⎜1 ⎟ = ⎜ ⎟ = ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ ⎟ ⎜ k ⎟ ⎝ 0 ⎠ ψ ⎝ TNk ⎠ψ ⎝ a N ⎠ φ kth column of T The columns of T are the eigenvectors of H in the φ representation. φ = Tψ ψ = T†φ Page 8 5.80 Lecture #1 Fall, 2008 Page 9 Special example of finding matrix representation of any general function of a matrix. Suppose we want Q3/2 * Generate Qφ in the convenient φ basis * diagonalize Qφ [not the representation that diagonalizes H] θ ⎛ Q11 0 0 ⎞ ⎜ ⎟ T †Qφ T = Qθ = ⎜ 0 0 ⎟ ⎜ ⎟ 0 QθNN ⎠ ⎝ 0 * * ( Qθ ) 3/2 ⎛ θ 3/2 0 ⎜( Q11 ) = ⎜ 0 ⎜ ⎜ 0 0 ⎝ ⎞ ⎟ ⎟ ⎟ 3/2 θ ( QNN ) ⎠⎟ 0 0 transform back to φ basis T(Qθ)3/2T† = [Q3/2]φ The only reason why this is not as wonderful and general as it seems is that it is impossible to diagonalize an infinite matrix and all Harmonic Oscillator basis sets are infinite. Still, truncation at a finite (large) dimension gives accurate results for the lowest few eigenstates. Accuracy can be tested by doing calculation twice, once for N × N and once for (N + 1) × (N + 1) and looking for stability of results for the lowest levels of interest.