In this seetion we develop a method for finding the inverse of a matrix if it exists ‘Touse this method we do not have to find out first whether A~! exists, We start to find A~'; if in the course of the computation we hit a certain situation, then wwe know that A~! does not exist. Otherwise. we proceed to the end and obtain A+, This method requires that elementary row operations of types L,I, and ILL (ee Section 2.1) be performed on A. We clarify these notions by starting with the following definition: PPTTSSY WI Ac 0x 7 elementary matrix of type 1, type HL or type HI is a matrix obtained Bea Theorem 2.5 Gite! from the identity matrix J, by performing a single elementary row or elementary ‘column operation of type T, type TI, or type TI, respectively ‘The following are elementary matrices: 001 10 0 &=(0 10], m=/o -2 oO}, 100 oo 1 103 ]} eft] oo. Matrix £; is of type I—we interchanged the first and third rows of /y; Ey is of type II—we multiplied the second row of 1; by (—2); Es is of type III —we added twice the second row oF 13 to the first row of 3; and £; is of type I —we added three times the first column of Js to the third column of Zs, 7 Let A be an me x n matrix, and let an elementary row (column) operation of type T, type Il, or type TIT be performed on A to yield matrix B. Let E be the elementary matrix obtained from Jy, (7,) by performing the same elementary row (column) ‘operation as was performed on A. Then B = EA(B = AE), Proof Exercise | . ‘Theorem 2.5 say's that an elementary row operation on A can be achieved by premultiplying A (mult plying A on the ieft) by the corresponding elementary ma- trix E; an elementary column operation on A can be obtained by postmultiplying A (multiplying A on the right) by the corresponding elementary matrix. Let i ged A=[-123 4 3012 and let B= Acapy¢p)-ory3then 5 3 OWS B=|-1 23 4 300 1 2 We can realy vesily drat B= BA, .Theorem 2.6 Theorem 2.7 Theorem 2.7 Lemma 2.1° If A and B are m xm matrices, then A isrow (column) equivalent to B ifand only if there exist elementary matrices £, a..... i such that B= Epi 1-- EBA (B= AB, Ey-- Ex By). Proof We prove only the theorem for row equivalence. If A is row equivalent to B, then B results from A by a sequence of elementary cow operations, This implies that there exist elementary matrices Ey, Fs,..., &, such that B= By By. --- B2E yA. Conversely, if B= E, By, --- E,E)A, where the E, areelementary matrives, then B results from A by a sequence of elementary row operations, which implies that A is row equivalent to B. . An clementary matrix £ is nonsingular, and its inverse is an elementary matrix of the same type, Proof Bxerise 6 . Thus an elementary row operation ean be “undone” by another elementary row operation of the same type We now obiain an algorithm for finding A if it exists; first, we prove the following lemma: If A and B are m>n matrices, then A isrov (coluran) equivalent to B ifand only if there exist elementary matrices E), B2,.... Bx such that B= ExEi 1+ E2BiA (BS ABLEy Eis). Proof ‘We prove only the theorem for row equivalence. If A is row equivalent to B, then B results from A by a sequence of elementary row operations. This implies that there exist elementary matrices E,, E>,..., Ey such that B = Ey By EE, Conversely, if B= E,E,_ -” E>£ A, where the E, are elementary mati then B results from A by a sequence of elementary row operations, which implies that A is row equivalent to B, . An elementary matrix & is nonsingular, and its inverse isan elementary matrix of the same type. Proof Exercise 6. . Thus an clementary row operation can be “undone” by another elementary row operation of the same type. We now obtain an algorithm for finding A" if it exis {olloving lemma: st, we prove the Let A he an nn matrix and let the homogeneous system Ax = 0 have only the trivial solution x = 0. Then A is row equivalent to J,. (That is. the reduced row echelon form of A is [,,) Proof Let B be a matrix in reduced row echelon form that is row equivalent to A. Then the homogeneous systems Ax = 0 and Bx = 0 are equivalent, and thus Bx = 0 also has only the trivial solution. Tt is clear that ifr isthe numher of nonzero rows of B, then the homogeneous system BX = 0 is equivalent to the homogeneous system whose coefficient mairix consists ofthe nonzero rows of B and is therefore 1 x n, Since this last homogeneous system has only the trivial solution, we con- clude from Theorem 2.4 that r > n, Since B ism x n,r fi — ‘We now perform elementary row operations that uansform [A { 5] 10 [5 [A]: we consider [A | Js] as 23 x 6 matsix, and whatever we do te a sow of A we also do to the corresponding row of Js, In place of using elementary matrices directly We arrange our computations, using elementary row operations as followsTheorem 2.10 Ea Apply Sey trees 14 2 3 sou 11 4 Apply jes > re o 2 3 oo 4 toa 0 Apply Irs > o 1 oo i 2 a Apply — fest rs > wy and 52 Se eee oo 4 1 1 0 Aprly —Ies br ey o 1 0 oo 18 1 0 0 x 0 ° $3 oo 1 o =} Hence ‘We can readily verify that AA . The question that srises at this point is how to tell when A is singular, The answer is that A is singular if and only if A is row equivalent to matrix &, baying at feast one row that consists entirely of zeros, We now prove this result Ann xn matrix A is singular if and only if A is row equivalent to a mawix that has a row of zeros. (That is, the reduced row echelon form of A has a row of zer0s.) Proof First, let A be row equivalent to a matix B thet has a row consisting entirely of zeros. From Exercise 46 of Section 1.5, it follows that B is singular Now B= ByEy1--B\A, where Ey, F2,.++, Ex are elementary matrices, If A is nonsingular, then B is ronsingular, a contradiction. ‘Thus 4 is singular. Conversely, if A is singular, then A is not row equivalent to J), by Corullary 2.2. Thus A is row equivalent toa matrix B 4 J,,, which is in reduced row echelon form, From Exercise 9 of Section 2.1, it follows that & must have a row of zeros, 7 ‘This means that in orcer to find A~!, we do not have to determine, in advance, whether it exists. We merely start to calculate A~*; if at any point in the com= ulation we find a ma‘tix B that is row equivalent to A and has a raw of zeros, then A~ does not oxist, That is, we transform the partitioned matrix [A | J, | to reduced row echelon form, obtaining [C | D]. 11 C = J,,then D = A-!. If © # Jy, then © has a 70w of zeros and We conelice that A 1s singular. 4 7 3 Let > oe Lt To find A~!, we proceed as follows:A a 12 fo BB deinterm 1-2 010 52 oo 1 [eet ee een 0-4 4a to 52 3) oot TS S71 oO aetna. 4 ala ap op pis ot a 3 R 10 o 0 0 at Ais pee Ancora 2-3 aa), 0 0 the last matrix under A. Since B has-a row of zeros, we stop and conclude that A is a singular matrix .