𝑟𝑎𝑛𝑘(𝐴) = 𝑁𝑜. 𝑜𝑓 𝑛𝑜𝑛 𝑧𝑒𝑟𝑜 𝑟𝑜𝑤𝑠 𝑖𝑛 𝑒𝑐ℎ𝑒𝑙𝑜𝑛 𝑓𝑜𝑟𝑚 𝑜𝑓 𝐴 𝑑𝑒𝑡(𝐸) is not necessarily 1, where 𝐸 is an elementary matrix. (It is -1 for elementary matrix formed by exchanging two rows of 𝐼) Every vector space has zero vector in it. 𝐶[𝑎, 𝑏] is notation for set of all function which are continuous in [𝑎, 𝑏]. Dimension of 0 vector space is 0 and its basis is null set. Zero vector cannot belong to a L.I. set. Subset of L.I. set is L.I. set and superset of L.D. subset is L.D. If a set 𝐵 is basis of vector space 𝑉(𝐹) then: o 𝐵 is L.I. o 𝐿(𝐵) = 𝑉(𝐹) o 𝐵 is maximal L.I. set, number of elements in 𝐵 is dimension of 𝑉(𝐹). o Every 𝑢 ∈ 𝑉(𝐹) is a unique linear combination of vectors in 𝐵. Any finite L.I. set in 𝑉(𝐹) is a part of a basis. If 𝑉(𝐹) has no finite basis we say its infinite dimensional. Ordered basis helps us to write “co-ordinate” vector of any 𝑢 ∈ 𝑉(𝐹). 𝑟𝑜𝑤 𝑟𝑎𝑛𝑘(𝐴) = dim(𝑟𝑜𝑤 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐴) = 𝑐𝑜𝑙𝑢𝑚𝑛 𝑟𝑎𝑛𝑘(𝐴) = dim(𝑐𝑜𝑙𝑢𝑚𝑛 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐴) = 𝑠𝑖𝑚𝑝𝑙𝑦 𝑟𝑎𝑛𝑘(𝐴) If 𝐴 and 𝐵 are row equivalent matrices (i.e., can be converted into each other by elementary row operations) then 𝑟𝑜𝑤 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐴 is same as 𝑟𝑜𝑤 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐵. To find out which columns of 𝐴 are forming basis for 𝑐𝑜𝑙𝑢𝑚𝑛 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐴 (i.e., which rows constitute L.I. set), we find which columns have the leading variable in 𝑟𝑟𝑒𝑓(𝐴). Likewise, for finding which rows of 𝐴 are forming basis for 𝑟𝑜𝑤 𝑠𝑝𝑎𝑐𝑒 𝑜𝑓 𝐴 we find which columns have the leading variable in 𝑟𝑟𝑒𝑓(𝐴 ). If 𝑊 (= 𝐿(𝑆 )) and 𝑊 (= 𝐿(𝑆 )) are subspaces of 𝑉(𝐹) then 𝑊 + 𝑊 (= 𝐿(𝑆 ∪ 𝑆 )) is also a subspace. 𝑊 + 𝑊 is called direct sum if 𝑊 ∩ 𝑊 = {0}. If 𝑉 = 𝑊 + 𝑊 then 𝑉 is direct sum of 𝑊 and 𝑊 if and only if every 𝑣 ∈ 𝑉 can be written as 𝑣 = 𝑤 + 𝑤 for a unique 𝑤 ∈ 𝑊 and a unique 𝑤 ∈ 𝑊 (which means that if this condition is satisfied then 𝑊 ∩ 𝑊 = {0} and vice-versa). For any linear transformation 𝑇 , 𝑇(0) = 0. Any linear transformation 𝑇: 𝑅 → 𝑅 is given by: 𝑇(𝑥, 𝑦) = (𝑎𝑥 + 𝑏𝑦, 𝑐𝑥 + 𝑑𝑦) for some choices of 𝑎, 𝑏, 𝑐, 𝑑 ∈ 𝑅. A linear transformation 𝑇: 𝑉 → 𝑉 is also called a linear operator. If a linear transformation 𝑇: 𝑉 → 𝑊 is bijective then the inverse map is also a linear transformation. Such 𝑇 is called isomorphism. And 𝑉and 𝑊 are called isomorphic if such a 𝑇 exists. 𝑘𝑒𝑟(𝑇) = {𝑣 ∈ 𝑉 ∶ 𝑇(𝑣) = 0} and 𝑇(𝑉) = {𝑇(𝑣): 𝑣 ∈ 𝑉} For a linear operator over a finite dimensional vector space, 𝑇: 𝑉 → 𝑉, 𝑇 𝑖𝑠 𝑖𝑛𝑗𝑒𝑐𝑡𝑖𝑣𝑒 ⇔ 𝑇 𝑖𝑠 𝑠𝑢𝑟𝑗𝑒𝑐𝑡𝑖𝑣𝑒 Two matrices 𝐴 and 𝐵 are called conjugate/similar if ∃ invertible 𝑃 s.t. 𝐵 = 𝑃 𝐴𝑃. Trace and determinant of conjugate matrices are same. Suppose 𝑇: 𝑉 → 𝑊 and 𝑇 : 𝑊 → 𝑈 and 𝐵, 𝐵 , 𝐵 is basis of 𝑉, 𝑊, 𝑈 respectively. Then, [𝑇 . 𝑇] = [𝑇 ] . [𝑇] . By definition, an eigenvector cannot be zero vector. For an invertible operator 𝑇, zero cannot be an eigenvalue. An if 𝜆 is an eigenvalue of 𝑇 then 𝜆 is an eigenvalue of 𝑇 . For linear operator 𝑇: 𝑉 → 𝑉, det (𝑋𝐼 − 𝑇) is characteristic polynomial. Degree of this polynomial is dim(𝑉). These three statements are equivalent: o 𝜆 is an eigenvalue of 𝑇. o The operator 𝜆𝐼 − 𝑇: 𝑉 → 𝑉 is not injective (as its kernel is not zero space) o det (𝜆𝐼 − 𝑇)=0 All above properties related to 𝑇 have their matrix form too, where 𝑇 is replaced by [𝑇] and vector 𝑣 by column matrix [𝑣] . Or, in general with square matrix 𝐴 and column matrix 𝑥. For eg., (𝜆𝐼 − 𝐴)𝑥 = 0 says that 𝑥 is eigenspace (i.e., is an eigenvector) of 𝐴 associated to the eigenvalue 𝜆. Subspaces 𝑊 , 𝑊 , … , 𝑊 ∈ 𝑉 are called independent if: 𝑤 + 𝑤 + ⋯ 𝑤 = 0 for 𝑤 ∈ 𝑊 ⇒ 𝑤 = 0 ∀ 𝑖 = 1,2, … , 𝑛. Eigenspaces (and thus eigenvectors) corresponding to distinct eigenvalues are independent. A linear operator 𝑇: 𝑉 → 𝑉 is called diagonalizable if ∃ a basis 𝐵 s.t. [𝑇] is diagonal matrix. Likewise, a matrix 𝐴 is called diagonalizable if it is conjugate/similar to a diagonal matrix. Cayley-Hamilton Theorem: Every matrix satisfies its characteristic equation. Applications of Cayley-Hamilton Theorem: o To find inverse of matrix. o 𝑃(0) = (−1) det(𝐴), so 𝑃(0) ≠ 0 ⇒ 𝐴 is invertible. A vector space with inner product defined on it, is called inner product space. Inner product on a vector space is a map 𝑉 × 𝑉 → 𝐹 with the following properties: o < 𝑢|𝑢 > ≥ 0 (𝑒𝑞𝑢𝑎𝑙𝑖𝑡𝑦 ℎ𝑜𝑙𝑑𝑠 𝑤ℎ𝑒𝑛 𝑢 = 0) o < 𝑢|𝑣 > = < 𝑣|𝑢 > o < 𝛼𝑢 + 𝛽𝑣 | 𝑤 > = 𝛼 < 𝑢|𝑤 > +𝛽 < 𝑣|𝑤 >, but o < 𝑤 | 𝛼𝑢 + 𝛽𝑣 > = 𝛼 < 𝑤|𝑢 > +𝛽̅ < 𝑤|𝑣 > o |𝑣| = < 𝑣|𝑣 > For functions < 𝑓|𝑔 > = ∫ 𝑓(𝑡) 𝑔(𝑡) 𝑑𝑡 , < 𝑓|𝑓 > ≥ 0, and < 𝑓|𝑓 >= 0 ⇒ 𝑓 = 0 if 𝑓 is continuous. Zero vector is orthogonal to every vector. Every orthogonal set of non- zero vectors is L.I. All inner products in 𝑅 are characterized as < 𝑥|𝑦 > = 𝑎𝑥 𝑦 + 𝑏(𝑥 𝑦 + 𝑥 𝑦 ) + 𝑑𝑥 𝑦 , for some choice of 𝑎, 𝑏, 𝑑 ∈ 𝑅 𝑠. 𝑡. 𝑎 > 0 , 𝑎𝑑 − 𝑏 > 0. Here, 𝑥 = (𝑥 , 𝑥 ) and 𝑦 = (𝑦 , 𝑦 ). Let 𝑉 is an i.p.s and we have L.I. set 𝑆 ∈ 𝑉 of 𝑛 vectors in 𝑉, then we can always construct an orthogonal/orthonormal L.I. set 𝑆 ∈ 𝑉 of 𝑛 vectors such that 𝐿(𝑆 ) = 𝐿(𝑆 ). ~ Gram-Schmidth orthogonalization process. If an i.p.s 𝑉 has an orthogonal basis set {𝑞 , 𝑞 , … , 𝑞 } and we write | some 𝑣 ∈ 𝑉 as 𝑣 = ∑ 𝛼 𝑞 , then 𝛼 = , so finding coefficients | is easier if we have orthogonal basis. Advantage of orthogonal basis For an i.p.s 𝑉, 𝑊 ⊆ 𝑉, and some 𝑣 ∈ 𝑉, 𝑤 ∈ 𝑊 is best approximation of 𝑣 by a vector in 𝑊 (i.e., 𝑤 is the nearest vector to 𝑣 in 𝑊), if |𝑣 − 𝑤 | ≤ |𝑣 − 𝑤| ∀ 𝑤 ∈ 𝑊.