Linear Algebra Done Right Solutions Manual Sheldon Axler 7 July 1997 © 1997 Sheldon AxderThis solutions manual provides answers to all the exercises in my book Linear Algebra Done Right. It is distributed (without charge) only to in- structors who are using Linear Algebra Done Right as a textbook. If you are such an instructor, please contact me (preferably via e-mail) to obtain the appropriate version of the solutions manual, This version of the solu- tions manual is intended to accompany the second edition (1997) of Linear Algebra Done Right. ‘This solutions manual is copyrighted by me, but instructors using Linear Algebra Done Right as a textbook have permission to modify and/or abridge this solutions manual and to distribute copies only to students enrolled in their courses (if students are charged for copies, the price may not exceed the cost of materials and reproduction). If you want to modify and/or abridge this solutions manual, you can request (via e-mail) the IATEX file from me (or from Springer if I am no longer around). Some instructors may not want to distribute any of these solutions to students; other instructors may want to save class time by distributing to students solutions to some or even many of the exercises. Obviously students should not see the solution to an exercise until they have attempted the exercise themselves. Please do not distribute solutions needlessly, especially if there is a danger that many solutions will get passed down to next year’s class. I will not distribute this solutions manual to students, so if neither you nor colleagues in your department make copies for students, you can be Teasonably confident that your students are working the exercises without benefit of these solutions. This solutions manual has not been subjected to the same amount of scrutiny as the book, so errors are more likely. I would be grateful for information about any errors that you notice. If you know nicer solutions to any of the exercises than the solutions given here, please let me know so that I can improve future versions of this solutions manual. Please check my web site for errata and other information about Linear Algebra Done Right. I welcome comments about either the book or the solutions manual. Have fun! Sheldon Axder Mathematics Department San Francisco State University San Francisco, CA 94132, USA e-mail: axler@math.sfsu.edu www home page: http: //math.sfsu.edu/axlerCHAPTER 1 Vector Spaces Suppose a and 6 are real numbers, not both 0. Find real numbers ¢ and d such that I/(a+ bi) =c4di. SOLUTION: Multiplying both the numerator and the denominator of the left side of the equation above by a — bi gives Sat ek ceai. a+ Thus we must have a b c= ae od d= TR because a and b are not both 0, we are not dividing by 0. Comment: Note that these formulas for c and d are derived under the assumption that a+ bi has a multiplicative inverse. However, we can forget about the derivation and verify (using the definition of complex multiplica- tion) that . a bo (+0 (atp- ape) =) which shows that every nonzero complex number does indeed have a multi- plicative inverse.CHAPTER 1. Vector Spaces Show that -1+4 Vai 2 is a cube root of 1 (meaning that its cube equals 1). SoLUTION: Using the definition of complex multiplication, we have ayy Thus i ay CSE) " Prove that —(—v) = u for every v € V. Souution: Let v € V. By the definition of additive inverse, we have v+(-v) =0. The additive inverse of —v, which by definition is —(—v), is the unique vector that when added to —v gives 0. The equation above shows that v has this property. Thus —(—v) = v. COMMENT: Using 1.6 twice leads to another proof that —(—v) = v. How- ever, the proof given above uses only the additive structure of V, whereas a proof using 1.6 also uses the multiplicative structure. Prove that if a € F,v € V, and av = 0, then a =0 or v= 0. SOLUTION: Suppose that a € F, v € V, and av=0. We want to prove that a = 0 or v= 0. Ifa = 0, then we are done. So suppose that a # 0. Multiplying both sides of the equation above by 1/a givesCwAPTER 1. Vector Spaces The associative property shows that the left side of the equation above equals 1v, which equals v. The right side of the equation above equals 0 (by 1.6). Thus v = 0, completing the proof. For each of the following subsets of F°, determine whether it is a subspace of F*: (a) {(v1, 22,23) € F9: 2 + 202 + 323 = 0}; (bd) {(1, 22,23) € F9: 2) + 2a +323 = 4}; (©) {(21, 22,23) € FS: eyz9¢3 = 0); (4) {(x1, 22,23) € F9: 2 = 523}. Sotution: (a) Let U = {(x1, 22,23) € F?: 21 + 2x2 + 323 = 0}. To show that U is a subspace of F%, first note that (0,0,0) € U, so U is nonempty, Next, suppose that (x1, 22,23) € U and (y1,42,43) € U. Then 21 + 2x2 + 323 =0 ti + 240 + 8y = 0. Adding these equations, we have (x1 + y1) + 2(zo + ye) + 3(z3 + ys) = 0, which means that (x1 + 91,22 + y2,23 +43) € U. Thus U is closed under addition. Next, suppose that (21,20,23) € U anda é F. Then TZ + 222 + 323 = 0. Multiplying this equation by a, we have (ax) + 2(az2) + 3(az3) = 0, which means that (az1,az2, a3) € U. Thus U is closed under scalar multi- plication. Because U is a nonempty subset of F3 that is closed under addition and scalar multiplication, U is a subspace of F°,CuapTER 1. Vector Spaces (b) Let U = {(21, 22,23) € F® : 2, + 2x9 +3a3 = 4}. Then (4,0,0) € U but 0(4,0,0), which equals (0,0,0), is not in U. Thus U is not closed under scalar multiplication. Thus U is not a subspace of F°. (c) Let U = {(e1, 22,23) € F° : eyz9z3 = 0}. Then (1,1,0) € U and (0,0, 1) € U, but the sum of these two vectors, which equals (1,1, 1), is not in U, Thus U is not closed under addition. Thus U is not a subspace of F3, (d) Let U = {(21, 22,23) € FS: 2 = 5x3}. To show that U is a subspace of F%, first note that (0,0,0) € U, so U is nonempty. Next, suppose that (z1,22,23) €U and (y1,y2,43) € U. Then 2, = 5a w= 5y3- Adding these equations, we have 21 + = 5(x3 + ys), which means that (x, + 1,22 + y2,23 +43) € U. Thus U is closed under addition. Next, suppose that (x1, 22,23) €U and a € F. Then 2 = 623. Multiplying this equation by a, we have ax, = 5(az3), which means that (az1,ax2,a23) € U. Thus U is closed under scalar multi- plication. Because U is a nonempty subset of F% that is closed under addition and scalar multiplication, U is a subspace of F°.§ CHAPTER 1. Vector Spaces 6. Give an example of a nonempty subset U of R? such that U is closed under addition and under taking additive inverses (meaning —u € U whenever u€U), but U is not a subspace of R?. Sotution: Let U = {(m,n) : mand n are integers}. Then clearly U is closed under addition and under taking additive inverses. However, (1,1) € U but 3(1,1), which equals (}, 4), is not in U, so U is not closed under scalar multiplication. Thus U is not a subspace of R?. Of course there are also many other examples. 7. — Give an example of a nonempty subset U of R? such that U is closed under scalar multiplication, but U is not a subspace of R?. Soxution: Let U be the union of the two coordinate axes in R?. More precisely, let U ={(z,0): ze R}U{(0,y):y ER}. Then clearly U is closed under scalar multiplication. However, (1,0) and (0,1) are in U but their sum, which equals (1,1) is not in U, so U is not closed under addition. Thus U is not a subspace of R?. Of course there are also many other examples. 8. Prove that the intersection of any collection of subspaces of V is a subspace of V. SoLutIon: Suppose (Ua}acr is a collection of subspaces of V; here I is an arbitrary index set. We need to prove that (\,cr Ua, which equals the set of vectors that are in Ug for every a € I’, is a subspace of V. The additive identity 0 is in U, for every a € I (because each Ug is a subspace of V). Thus 0 € Maer Ua. In particular, (|r Ua is a nonempty subset of V. Suppose u,v € MaerUa- Then u,v. € Ua for every a € T. Thus utv€ Ug for every a € T' (because each Ug is a subspace of V). Thus utv€ [ger Ua. Thus (ger Va is closed under addition, Suppose u € (\,ep Ua and a € F. Then u € Ug for every a € T. Thus au € Ug for every a € T (because each Us is a subspace of V). Thus au € (er Ua. Thus (),er Ua is closed under scalar multiplication. Because (\,cr Ua is a nonempty subset of V that is closed under addition and scalar multiplication, (\,¢r Ua is a subspace of V. Comment: For many students, the hardest part of this exercise is un- derstanding the meaning of an arbitrary intersection of sets. Instructors who10. coe CHAPTER 1. Vector Spaces do not want to deal with this issue should change the exercise to “Prove that the intersection of any finite collection of subspaces of V is a subspace of V.” Many students will then prove that the intersection of two subspaces of V is a subspace of V and use induction to get the result for finite collections of subspaces. Prove that the union of two subspaces of V is.a subspace of V if and only if one of the subspaces is contained in the other. SOLUTION: Suppose U and W are subspaces of V such that UUW is a subspace of V. We will use proof by contradiction to show that U C W or W CU. Suppose that our desired result is false. Then U ¢ W and W ¢U. This means that there exists u € U such that u ¢ W and there exists w ¢ W such that w ¢ U. Because u and w are both in UUW, which is a subspace of V, we can conclude that u-+weUUW. Thusu+we€U orut+wew. First consider the possibility that ut+-w € U. In this case w, which equals (u+w) + (—u), would be in the sum of two elements of U and hence we would have w € U, contradicting our assumption that w ¢ U. Now consider the possibility that u+tw € W. In this case u, which equals (u+w) + (—w), would be in the sum of two elements of W and hence we would have u € W, contradicting our assumption that u ¢ W. The two paragraphs above show that u+ w ¢ U andu+w ¢ W, con- tradicting the final sentence of the first paragraph of this solution. This contradiction completes our proof that UC W or WCU. The other direction of this exercise is trivial: if we have two subspaces of V, one of which is contained in the other, then the union of these two subspaces equals the larger of them, which is a subspace of V. Suppose that U is a subspace of V. What is U + U? Souution: By definition, U+U = {u+v:u,v € U}. Clearly U CU+U because if u € U, then u equals u + 0, which expresses u as a sum of two elements of U. Conversely, U-+U C U because the sum of two elements of U is an element of U (because U is a subspace of V). Conclusion: U+U =U. Is the operation of addition on the subspaces of V commutative? Associa tive? (In other words, if Uy, U2, Us are subspaces of V, is U; +U2 = U2 +i? Is (Uy + U2) + Us = Uy + (U2 + Us)?) SoLuTion: Suppose U;, U2, Us are subspaces of V. A typical element of U; +Us is a vector of the form u; +u2, where uy € Uy and uz € U2. Because addition of vectors is commutative, wu + ug equals13. 14. GuapTER 1. Vector Spaces uu +1, which is a typical element of Uz +Uy. Thus Uy +U2 = Uz +0). In other words, the operation of addition on the subspaces of V is commutative. A typical element of (U; + U2) +Us is a vector of the form (u; +u2)-+u3, where ui € Uj, up € Up, and us © Us. Because addition of vectors is associative, (uw, + tz) + ug equals u; + (ug + ug), which is a typical element of U; + (Uz + U3). Thus (U; + U2) + U3 = Uy + (U2 + U3). In other words, the operation of addition on the subspaces of V is associative. Does the operation of addition on the subspaces of V have an additive identity? Which subspaces have additive inverses? SoLurion: The subspace {0} is an additive identity for the operation of addition on the subspaces of V. More precisely, if U is a subspace of V, then U + {0} = {0} +U =U. For a subspace U of V to have an additive inverse, there would have to be another subspace W of V such that U + W = {0}. Because both U and W are contained in U + W, this is possible only if U = W = {0}. Thus {0} is the only subspace of V that has an additive inverse. Prove or give a counterexample: if Ui, U2,W are subspaces of V such that U, +W=U2+W, then U, = Uy SoLuTIoN: To construct @ counterexample for the assertion above, choose V to be any nonzero vector space. Let U; = {0}, U2 = V, and W =V. Then U; + W and U2 + W are both equal to V, but U; # Up. Of course there are also many other examples. Suppose U is the subspace of P(F) consisting of all polynomials p of the form (2) = a2? +628, where a,6 € F. Find a subspace W of P(F) such that P(F) =U @ W. SouuTion: Let W be the set of all polynomials (with coefficients in F) whose z?-coefficient and z°-coefficient both equal 0. Then every polynomial in P(P) can be written uniquely in the form p+q, where p € U andq ¢ W. Thus P(F) =U @W.15. CHAPTER 1. Vector Spaces Comment: There are other possible choices for W that give a correct solution to this exercise, but the choice for W made above is certainly the most natural one. Prove or give a counterexample: if U1,U2,W are subspaces of V such that V=U,0W and V=lhow, then Uy = U2. SOLUTION: To construct a counterexample for the assertion above, let V =F, let Ui = {(z,0) : x € Fh, let Up = {(0,y) : y € F}, and let W = {(z,z):2€F}. Then F=U0W and FP =2@W, as is easy to verify, but U # Up. Of course there are also many other examples.CHAPTER 2 Finite-Dimensional Vector Spaces Prove that if (v1,...,tn) spans V, then so does the list (U1 = 02,02 = 035. -+5Un—1 — Uns Un) obtained by subtracting from each vector (except the last one) the following vector. SoLuTION: Suppose (vj,.-., Un) spans V. Let v € V. To show that v € span(u1 — v2, 02 — vg,---; Un-1 — Un) Un), We need to find a1,...,¢n €F such that v= ay (vy — v2) + a2(v9 — vy) + ..-@n—1(Yn—1 — Up) + anda Rearranging terms of the equation above, we see that we need to find @,...,@_ € F such that (@) oe Because (v1,...,0n) spans V, there exist bj,...,b, € F such that 101 + (a2 — a1)v2 + (a3 — ay)u3 + +++ + (an — Oni) Up. (b) v= byuy + bate + b303 +--+ bata. Comparing equations (a) and (b), we see that (a) will be satisfied if we choose a; to equal 6; and then choose a2 to equal by + a; and then choose ag to equal b3 + a2, and so on.10 CHAPTER 2._Finite-Dimensional Vector Spaces Prove that if (v,..., un) is linearly independent in V, then so is the list (21 = 92, 02 — U5) ---,Un—-1 — Un Yn) obtained by subtracting from each vector (except the last one) the following vector. SOLUTION: Suppose (v1,...,Un) is linearly independent in V. To prove that the list displayed above is linearly independent, suppose a1,...,an € F are such that a4 (v1 — v2) + a2(v2 — 03) + +++ + @n-1(Un-1 — Un) + Gntn = 0. Rearranging terms, the equation above can be rewritten as a0, + (a2 — a1)¥2 + (a3 — ag)ug + +++ + (@n — Qn—1) Un = 0. Because (v;,...,Un) is linearly independent, the equation above implies that a=0 a2—a,=0 a3—a,=0 Qn — On = 0. ‘The first equation above tells us that a; = 0. That information, combined with the second equation, tells us that a2 = 0. That information, combined with the third equation, tells us that as = 0. Continue in this fashion, getting a, = = ay = 0. Thus (v1) — v2,¥2 — U3,...,Un—1 — Uns Un) is. linearly independent. Suppose (v1,.-.,Un) is linearly independent in V and w € V. Prove that if (v1 + w,.-.,Un +w) is linearly dependent, then w € span(u),...,Un)- SoLuTION: Suppose (v) + w,.-.,Un +w) is linearly dependent. Then there exist scalars @},...,@,, not all 0, such that y(t; +W) +--+ + n(dn + w) Rearranging this equation, we have 10, +--+ + Gntp = —(a; + +++ +an)w.ey CHAPTER 2. Finite-Dimensional Vector Spaces If a, ++++ + ay were 0, then the equation above would contradict the linear independence of (v1,...,0n). Thus a1 +++ +n #0. Hence we can divide both sides of the equation above by —(a; + +--+ an), showing that w € span(v1,..., tn): Suppose m is a positive integer. Is the set consisting of 0 and all polynomials with coefficients in F and with degree equal to m a subspace of P(F)? SOLUTION: The set consisting of 0 and all polynomials with coefficients in F and with degree equal to m is not a subspace of P(F) because it is not closed under addition. Specifically, the sum of two polynomials of degree m may be a polynomial with degree less than m. For example, suppose m = 2. Then 7 + 4z + 52? and 1 + 2z — 52? are both polynomials of degree 2 but their sum, which equals 8 + 62, is a polynomial of degree 1. Prove that F® is infinite dimensional. SoLuTion: For each positive integer m, let em be the clement of F° whose m'® coordinate equals 1 and whose other coordinates equal 0: m = (0,.++50)1,0,..+). T m* coordinate Then (e},...,€m) is a linearly independent list of vectors in F™, as is easy to verify. This implies, by the marginal comment attached to 2.6, that F° is infinite dimensional. Prove that the real vector space consisting of all continuous real-valued functions on the interval {0, 1] is infinite dimensional. SoLuTION: Let V denote the real vector space of all continuous real- valued functions on the interval [0,1]. For each positive integer m, the list (1,z,.-.,2) is linearly independent in V (because if a9,...,am € R are such that, 09 +47 +++ + Omz™ =0 for every x € [0,1], then the polynomial above has infinitely many roots and hence all its coefficients must equal 0). This implies, by the marginal comment attached to 2.6, that V is infinite dimensional.12 CHAPTER 2. Finite-Dimensional Vector Spaces Prove that V is infinite dimensional if and only if there is a sequence 1, 02,-.. of vectors in V such that (v1,...,U,) is linearly independent for every positive integer n. SoLuTion: First suppose that V is infinite dimensional. Choose v to be any nonzero vector in V. Choose v2,03,... by the following induc- tive process: suppose that v1,...,Un—1 have been chosen; choose any vector vn € V such that vn ¢ span(v1,...,Un—1)—because V is not finite dimen- sional, span(v,..., 0-1) cannot equal V so choosing v, in this fashion is possible. The linear dependence lemma (2.4) implies that (u1,...,Un) is linearly independent for every positive integer n, as desired. Conversely, suppose there is a sequence v1,02,... of vectors in V such that (v1,...,0,) is linearly independent for every positive integer n. This implies, by the marginal comment attached to 2.6, that V is infinite dimen- sional. Let U be the subspace of R® defined by U = {(x1, 22,23, 24,25) € R® : 21 = 322 and 23 = 724}. Find a basis of U. SoLution: Obviously U = {(8r2, 22, 724,24, 25) : T2,24,75 € R}. From this representation of U, we see casily that ((3, 1,0, 0, 0), (0,0, 7, 1,0), (0,0, 0, 0, 1) is a basis of U. Of course there are also other possible choices of bases of U. Prove or disprove: there exists a basis (pp, P1, P2, P3) of P3(F) such that none of the polynomials pp, pi, p2, ps has degree 2. SoLuTION: Define po, p1,P2, Pa € Pa(F) by po(z) = 1, pi(z) =z, pr(2) = 2? +2°, ps(z) = 2°.13 10. iL. 13. CHAPTER 2. Finite Dimensional Vector Spaces None of the polynomials po, 71, p2,p3 has degree 2, but (po, p1,P2,p3) is a basis of Ps(F), as is easy to verify. Of course there are also other possible choices of bases of P3(F) without using polynomials of degree 2. Suppose that V is finite dimensional, with dimV =n. Prove that there exist one-dimensional subspaces U;,...,Uy of V such that V=U6---@Un. SoLution: Let (v1,...,0n) be a basis of V. For each j, let U; equal span(vj); in other words, U; = {avj: a € F}. Because (v1,..-,0n) is a basis of V, each vector in V can be written uniquely in the form YU ++ +AnUn, Suppose that V is finite dimensional and U is a subspace of V such that dimU = dimV. Prove that U = V. SOLUTION: Let (u1,...,tq) be @ basis of U. Thus n = dimU, and by hypothesis we also have n = dimV. Thus (uj,...,tn) is a linearly independent (because it is a basis of U) list of vectors in V with length dimV. From 2.17, we see that (u,...,tn) is a basis of V. In particular every vector in V is a linear combination of (u,...,tm). Because each uj € U, this implies that U = V. Suppose that Po, Pi,..-sPm are polynomials in Pya(F) such that p,(2) = 0 for cach j. Prove that (pp, P1,---, Pm) is not linearly independent in Pm(F). SOLUTION: Because p;(2) = 0 for each j, the constant polynomial 1 is not in span(pp,---,Pm)- Thus (Po, .-.,Pm) is not a basis of Pm(F). Because )+++3Pm) is a list of length m +1 and Pm(F) has dimension m + 1, this implies (by 2.17) that (po,...,Pm) is not linearly independent. Suppose U and W are subspaces of R® such that dimU = 3, dimW = 5, and U + W = R®. Prove that UNW = {0}. SoLuTion: We know (from 2.18) that dim(U + W) = dimU + dimW - dim(U NW).14 14. 15. CHAPTER 2. Finite-Dimensional Vector Spaces Because dim(U + W) = 8, dimU = 3, and dimW = 5, this implies that dim(U NW) = 0. Thus UN W = {0}. Suppose that U and W are both five-dimensional subspaces of R9. Prove that UW # {0}. SOLUTION: Using 2.18 we have 9>dim(U + W) = dimU +dimW -dim(UNW) =10-dim(UNW). Thus dim(U'N W) > 1. In particular, UN W # {0}. You might guess, by analogy with the formula for the number of elements in the union of three subsets of a finite set, that if U,,U2,Us are subspaces of a finite-dimensional vector space, then dim(, + Uz + Us) = dim U, + dim U2 + dimU; — dim(U; 9 U2) — dim(U; N Ug) — dim(Uz N Us) + dim(U, 1U, Us). Prove this or give a counterexample. SoLuTioN: To give a counterexample, let V = R?, and let UW = {(z,0): 2 € R}, U2 = {(0,y):y € R}, Us = {(z,2):2€ R}. Then Uy + Uz + U3 = R?, so dim(U; + Uz + U3) = 2. However, dim U; = dimU2 = dim U3 = 1 and dim(U, N U2) = dim(U; N Us) = dim(U2 N U3) = dim(U, N U2 N Us) = 0. Thus in this case our guess would reduce to the formula 2 = 3, which obviously is false. OF course there are also many other examples.16. 17. Cnaprer 2. Finite-Dimensional Vector Spaces Prove that if V is finite dimensional and Uj,...,Um are subspaces of V, then dim(0i +-+++ Um) {Tu : u € U}. To prove the inclusion in the other direction, suppose v € V. Then there exist w € nullT and wu! € U such that =uty. Applying T to both sides of this equation, we have Tv = Tw + Tul = Tu. Thus Tv € {Tu:u € U}. Because v was an arbitrary vector in V (and thus Tv is an arbitrary vector in rangeT), this implies that21 10. rangeT C {Tu:u € U}. Thus rangeT = {Tu: u € U}, as desired. Prove that if T is a linear map from F“ to F? such that null T = {(1, 22,23, 24) € F4 : 2) = 5a and x3 = 724}, then T is surjective. SoLuTION: Suppose T € £( F4,F*) is such that nullT is as above. Then ((5,1,0,0), (0,0,7, 1) is a basis of nullT’, and hence dim nullT = 2. From 3.4 we have dim range T = dim F4 — dimnullT =4-2 =2. Because range T is a two-dimensional subspace of R?, we have rangeT = R2, In other words, T is surjective. Prove that there does not exist a linear map from F® to F? whose null space equals {(c1, 22,29, 24,25) € FS: xy = Say and 23 = 14 = 5}. SOLUTION: Suppose U is the subspace of F® displayed above. Then ((3,1,0, 0,0), (0,0, 1,1, 1)) is a basis of U, and hence dimU = 2. IfT € £(F°, F*) then from 3.4 we have dim null T = dim F® — dim rangeT = 5 —dimrangeT 23 > dimU, where the first inequality holds because range’ C F?. The inequality above shows that if T € £(F*, F2), then null T’ #U, as desired. Prove that if there exists a linear map on V whose null space and range are both finite dimensional, then V is finite dimensional. SOLUTION: Suppose there exists a linear map T from V into some vec- tor space such that null 7’ and rangeT are both finite dimensional. Thus22 Cuaprer 3. Linear Maps there exist vectors ui,...,tm € V and wy,...,wWn € rangeT such that (u1,...5Um) spans nullT and (wi,...,wW,) spans rangeT. Because each w; € rangeT, there exists vj € V such that w; = Tv;. Suppose v € V. Then Tu € rangeT, so there exist b1,...,bn € F such that Tv = bw, + +--+ bp, = bTu +--+ +bpT on = T (by, +...byPn)- The equation above implies that T(u — bv: — --- — bata) = 0. In other words, v — b)u; — ---— bata € nullT. Thus there exist a1,...,am € F such that uv — by — +++ — avy = ay +--+ + Omtm. The equation above can be rewritten as US Qty $+ + Amt + 810; +++ + Pada The equation above shows that an arbitrary vector v € V is a linear com- bination of (u1,...,tm,1,--+)Un)- In other words, (u1,..-,tUm;V1,--+)Un) spans V. Thus V is finite dimensional. Comment: The hypothesis of 3.4 is that V is finite dimensional (which is what we are trying to prove in this exercise), so 3.4 cannot be used in this exercise. 12. Suppose that V and W are both finite dimensional. Prove that there exists a surjective linear map from V onto W if and only if dimW < dimV. SoLuTioN: First suppose that there exists a surjective linear map T’ from V onto W. Then dim W = dimrangeT = dimV —dimnullT dimV —dimW. SoLuTion: First suppose that there exists T € C(V,W) such that null T =U. Then dimU = dimnullT =dimV —dimrangeT >dimV —dimW, where the second equality comes from 3.4, To prove the other direction, now suppose that dim U > dim V —dimW. Let (u1,... stm) be @ basis of U. Extend to a basis (u1,..., tems 1,+++s0n) of V. Let (wi,..., wp) be a basis of W. For aj,...,0msb1,-.-,5n € F define Taye +++ + amt + b101 +++ + bntn) by T(ayuy + +++ + Qmtm + 6101 +--+ + bad) = byw +++ + batn Because dim W > dim V—dimU, we have p > n and so wn on the right side of the equation above makes sense. Clearly T € £(V,W) and null T =U. Suppose that W is finite dimensional and T € L(V,W). Prove that T is injective if and only if there exists S € £(W,V) such that ST is the identity map on V. SoLution: First suppose that T is injective. Define S’: rangeT — V by S'(T») U; because T is injective, each element of rangeT can be represented in the form Tv in only one way, so T is well defined. As can be easily checked, S’ isa linear map on rangeT.. By Exercise 3 of this chapter, 5’ can be extended toa linear map S € L(W,V). Ifu € V, then (ST)v = S(Tv) = S'(Tv) =v. ‘Thus ST is the identity map on V, as desired.24 15. 16. Cuapter 3. Linear Maps To prove the implication in the other direction, now suppose that there exists S € £(W,V) such that ST is the identity map on V. If u,v € V are such that Tu = Tv, then u=(ST)(u) = S(Tu) = 5(Tv) = (ST)v =0 and hence u = v. Thus T’ is injective, as desired. Suppose that V is finite dimensional and T © L(V,W). Prove that T is surjective if and only if there exists S € L(W, V) such that TS is the identity map on W. SotuTion: First suppose that T’ is surjective. Thus W, which equals rangeT, is finite dimensional (by 3.4). Let (wi,...,tm) be a basis of W. Because T is surjective, for each j there exists vj € V such that w; = Tj. Define S € L(W,V) by Slaw, +++ + OmWm) = av, +++ + Omtm- Then (TS)(ayw, + +++ + @mwm) = T(ain +--+ + amim) ++ amTom Thus T'S is the identity map on W. ‘To prove the implication in the other direction, now suppose that there exists S € £(W,V) such that TS is the identity map on W. If w € W, then w= T(Sw), and hence w € rangeT. Thus rangeT = W. In other words, T is surjective, as desired. Suppose that U and V are finite-dimensional vector spaces and that S € L£(V,W), T € L(U,V). Prove that dim null ST < dim null S + dim null T. SOLUTION: Define a linear map I’: nullST + V by Tu = Tu. If u € null ST, then S(Tu) = 0, which means that Tu € nullS. In other words, rangeT” C null $. Now dim null ST = dimnull T’ + dim range T” =I and (P-'s“)(ST) = T'S"! Sy TT30 CuapTer 3. Linear Maps Thus T-1S-' satisfies the properties required for an inverse of ST. Thus ST is invertible and (ST)“! = T-!$—1, 23. Suppose that V is finite dimensional and $,T € L(V). Prove that ST = I if and only if TS = I. SOLUTION: First suppose that sT=I. Because J is invertible, the previous exercise implies that S and T are both invertible. Multiply both sides of the equation above by T~! on the right, getting S=T". Now multiply both sides of the equation above by T on the left, getting TS=I, as desired. ‘To prove the implication in the other direction, simply reverse the roles of S and T in the direction we have already proved, showing that if TS = I, then ST =I. 24. Suppose that V is finite dimensional and T € L(V). Prove that T is a scalar multiple of the identity if and only if ST = T'S for every S € L(V). Souution: First suppose that T = al for some a € F. Let S € L(V). Then ST = S(al) =aS = (als =TS. To prove the implication in the other direction, suppose now that ST = TS for all S € L(V). We begin by proving that (v, Tv) is linearly dependent for every v € V. To do this, fix v € V, and suppose that (v, Tv) is linearly independent. Then (v,T'v) can be extended to a basis (v, Tv, u1,...,Un) of V. Define 5 € L(V) by S(av + bTv + cits + --- + entin) = bv.31 25. Cuaprer 3, Linear Maps Thus S(T'v) = v and Sv = 0. Thus the equation S(Tv) = T(Sv) becomes the equation v = 0, a contradiction because (v,Tv) was assumed to be linearly independent. This contradiction shows that (v, Tv) is linearly de- pendent for every v € V. This implies that for each v € V\ {0}, there exists ay € F such that Tu =ayv. To show that T is a scalar multiple of the identity, we must show that a, is independent of v. To do this, suppose v,w € V \ {0}. We want to show that a, = ay. First consider the case where (v,w) is linearly dependent. Then there exists 6 € F such that w = bv. We have Tw = T(bv) =0tTv = b(ayv) =ayw, ay’ which shows that ay = ay, as desired. Finally, consider the case where (u,v) is linearly independent. We have Qysu(v + w) = T(v + w) Tu+Tw = ay +auw, which implies that (Qu+w — @v)0 + (Guiw — au)w = 0. Because (v,w) is linearly independent, this implies that a,4w = ay and Qy4y = dy, SO again we have ay = dy, as desired. Prove that if V is finite dimensional with dim V > 1, then the set of non- invertible operators on V is not a subspace of L(V). SOLUTION: Suppose that V is finite dimensional with dimV > 1. Let n=dimV and let (v1,...,0n) be a basis of V. Define 5,T € L(V) by S(aiu, +-+++aatn) = air and32 26. Cuaprer 3. Linear Maps Tay + +++ + @nva) = Q2t2 +--+ + OnUp- jective because Suz = 0 (this is where we use the hypothesis that dim V > 1), and T is not injective because Tv, = 0. Thus both S and T are not invertible. However, $ +T equals I, which is invertible. Thus the set of noninvertible operators on V is not closed under addition, and hence it is not a subspace of L(V). Comment: If dimV = 1, then the set of noninvertible operators on V equals {0}, which is a subspace of £(V). Suppose n is a positive integer and aj; € F for i,j = 1,...,n. Prove that the following are equivalent: (a) The trivial solution 2) = --- = tq = 0 is the only solution to the homogeneous system of equations Saree =0 m= anete = 0. k (b) For every c1,...,¢n € F, there exists a solution to the system of equations aren =e m= Vance = cn- Note that here we have the same number of equations as variables. Souution: Define T € £(F") by (Saetes sangre). a a Then (a) above is the assertion that T is injective, and (b) above is the assertion that T is surjective. By 3.21, these two assertions are equivalent. T(a1,.--s2n)CHAPTER 4 Polynomials Suppose m and n are positive integers with m F™+! by Tp = (p(41),---,P(2m4t))- We need to prove that T is injective (which implies that at most one polyno- mial p satisfies the condition required by the exercise) and surjective (which implies that at least one polynomial p satisfies the condition required by the exercise). Clearly T is a linear map. If p € nullT, then Plz) = P(Zm+1) = 0, which means that p is a polynomial of degree m with at least m +1 distinct roots, which means that p = 0 (by 4.3). Thus p is injective, as desired. 3334 CHaprer 4, Polynomials Now dim range T = dim P,,(F) — dim null =(m+1)-0 =dimF™*, where the first equality comes from 3.4 and the second equality holds because null T= {0}. The last equality above implies that rangeT = F™!. Thus T is surjective, as desired. ComMENT: Surjectivity of T can also be proved by using an explicit construction. But linear algebra, specifically 3.4, gives us surjectivity easily once we get injectivity. 3. Prove that if p,q € P(F), with p # 0, then there exist unique polynomials s,r € P(F) such that =sptr and degr < degp. In other words, add a uniqueness statement to the division algorithm (4.5). SOLUTION: Suppose p,q € P(F), with p # 0. We know from the division algorithm (4.5) that there exist s,r € P(F), with degr < degp, such that q=sp+r. To prove that s and r are unique, suppose that 3,7 are in P(F), with degF < deg p and q= pti. Subtracting the last two equations are rearranging, we have (§-s)p=r The right side of the equation above is a polynomial whose degree is less than deg p. If § were not equal to s, then the left side of the equation above would be a polynomial whose degree is at least degp. Thus we must have 5 = s, which, from the equation above, implies that # = r. Thus the choices of s and r were indeed unique.35 CHAPTER 4. Polynomials Suppose p € P(C) has degree m. Prove that p has m distinct roots if and only if p and its derivative p’ have no roots in common. SOLUTION: First suppose that p has m distinct roots. Because p has degree m, this implies that p can be written in the form D(z) = e( — a) ---(2- Am): where Aj,...,Am are distinct. To prove that p and p/ have no roots in common, we must show that p'(;) # 0 for each j. To do this, fix j. The expression above for p shows that we can write p in the form P(z) = (2 — As)a(z), where q is a polynomial such that 9(A;) # 0. Differentiating both sides of this equation, we have B(2) = (2 —dj)q'(z) + (2). Thus P'(As) = as) #0, as desired. To prove the other direction, we will proved the contrapositive, meaning that we will prove that if p has less than m distinct roots, then p and p/ have at least one root in common. To do this, suppose that p has less than m distinct roots. Then for some root A of p, we can write p in the form P(z) = (2— A)" a2), where n > 2 and g is a polynomial. Differentiating both sides of this equa- tion, we have B(z) = (2 - d)"a'(2) + nz — A)"14(2). Thus p/(X) = 0, and so A is a common root of p and 7’, as desired. Prove that every polynomial with odd degree and real coefficients has a real root. SOLUTION: Suppose that p is a polynomial with odd degree and real coefficients. By 4.14, p is a constant times the product of factors of the form36 CHAPTER 4. Polynomials z—Xand/or 2? +ax+ 8, where ,a, € R. Not all the factors can be of the form x? +0x-+£, because otherwise p would have even degree. Thus at least one factor must be of the form x — d. Any such is a real root of p. Comment: Here is another proof, using calculus but not using 4.14. Suppose p is a polynomial with odd degree m. We can write p in the form P(z) = a9 tat +++ +am2™, where a9,...,0m @ Rand am # 0. Replacing p with —p if necessary, we can assume that am > 0. Now ay es P(x) = a) eet PEE ban). = This implies that -lim ple) =-co and Jim p(x) = ‘The intermediate value theorem now implies that there is a real number \ such that p(A) = 0. In other words, p has a real root.CHAPTER 5 Eigenvalues and Eigenvectors Suppose T € L(V). Prove that if Ui,...,Um are subspaces of V invariant under T, then U; + --- + Um is invariant under T. SoLurion: Suppose Uj,...,Um are subspaces of V invariant under T. Consider a vector u € Ui +---+Um. There exist uy € Uj,...,um € Um such that usute tim Applying T to both sides of this equation, we get Tu=Tuy +--+ +Tum. Because each Uj is invariant under T, we have Tu: € Uj,..., Tum € Um Thus the equation above shows that Tu € U; +---+Ujn, which implies that U;, +-+++Um is invariant under T. Suppose T € £(V). Prove that the intersection of any collection of subspaces of V invariant under T is invariant under T. SoLuTION: Suppose {Ua}aer is a collection of subspaces of V invariant under T; here T’ is an arbitrary index set. We need to prove that (aepUa, which equals the set of vectors that are in Ug for every a € I, is invariant under T. To do this, suppose u € (|,¢p Ua. Then u € Ug for every ae TL. Thus Tu € Ug for every « € T’ (because every Ua is invariant under T). Thus Tu € Maer Ua, which implies that cp Ua is invariant under T. Prove or give a counterexample: if U is a subspace of V that is invariant under every operator on V, then U = {0} or U =V. 3738 CHAPTER 5. Eigenvalues and Eigenvectors SOLUTION: We will prove that if U is a subspace of V that is invariant under every operator on V, then U = {0} or U = V. Actually we will prove the (logically equivalent) contrapositive, meaning that we will prove that if U is a subspace of V such that U # {0} and U # V, then there exists T € L(V) such that U is not invariant under T. To do this, suppose U is a subspace of V such that U # {0} and U # V. Choose u € U \ {0} (this is possible because U # {0}) and w € V\ U (this is possible because U # V). Extend the list (u), which is linearly independent because u # 0, to a basis (u,v1,..-,0,) of V. Define T € L(V) by T(au + 610) +...bntn) = aw. Thus Tu = w. Because u € U but w ¢ U, this shows that U is not invariant under T, as desired. Suppose that S,T € £(V) are such that ST = TS. Prove that null(T — A) is invariant under S$ for every \ € F. SoLuTion: Fix \€ F. Suppose v € null(T — AF). Then (T - Al)(Sv) = TSv— Sv = STv—dSv = 3(Tv — dv) =0. Thus Sv € null(T — MI). Hence null(T — AZ) is invariant under S. Define T € £(F?) by T(w,z) = (z,w). Find all eigenvalues and eigenvectors of T. SoLUTION: Suppose A is an eigenvalue of T. For this particular operator, the eigenvalue-eigenvector equation T(w,z) = A(w,z) becomes the system of equations z=dhw w=rz. Substituting the value for z from the first equation into the second equation gives w = \’w. Thus 1 = .? (we can ignore the possibility that w = 039 Ciapren 5. Eigenvalues and Eigenvectors because if w = 0, then the first equation above implies that z = 0). Thus A=1or A= ~—1. The set of eigenvectors corresponding to the eigenvalue 1 is {(w,w): we F}; ‘The set of eigenvectors corresponding to the eigenvalue —1 is {(w, -w) : w € F}. Define T € £(F*) by T(z, 22, 23) = (222,0, 523). Find all eigenvalues and eigenvectors of T. Soxurion: Suppose \ is an eigenvalue of T. For this particular operator, the eigenvalue-eigenvector equation T(z1, 22, 23) = A(z1, 22, 23) becomes the system of equations 22 = Ay O= Azz 52s = dey. If \ # 0, then the second equation implies that 22 = 0, and the first equation then implies that 2; = 0. Because an eigenvalue must have a nonzero eigenvector, there must be a solution to the system above with z3 #0. The third equation then shows that \ = 5. In other words, 5 is the only nonzero eigenvalue of T. The set of eigenvectors corresponding to the eigenvalue 5 is {(0,0, z3) : z3 € F}. If = 0, the first and third equations above show that 22 = 0 and z3 = 0. With these values for 22, 23, the equations above are satisfied for all values of 2. Thus 0 is an eigenvalue of T. The set of eigenvectors corresponding to the eigenvalue 0 is {(z1,0,0): 21 € F}. Suppose n is a positive integer and T € C(F*) is defined by T(e1,...5 20 = (Git + tne Bete + aA); in other words, T is the operator whose matrix (with respect to the standard basis) consists of all 1’s. Find all eigenvalues and eigenvectors of T.40 CHAPTER 5. Eigenvalues and Eigenvectors SOLUTION: Suppose ) is an eigenvalue of T. For this particular oper- ator, the eigenvalue-eigenvector equation Tz = Ax becomes the system of equations bet aty = Ae Ip = tne Thus Hence cither 4 = 0 or Consider first the possibility that 4 = 0. In this case all the equations in the eigenvector-eigenvalue system of equations above become the equation 21 +++++2 = 0. Thus we see that 0 is an eigenvalue of T and that the corresponding set of eigenvectors equals {(z1,---,2n) € FP ay + +2 = 0}. Now consider the possibility that 2; = --- = zp; let ¢ denote the com- mon value of z},...,2n. In this case all the equations in the eigenvector- eigenvalue system of equations above become the equation nt = At. Hence A must equal n (an eigenvalue must have a nonzero eigenvector, so we can take t #0). Thus we see that n is an eigenvalue of T and that the corresponding set of eigenvectors equals {(21,-..,2n) €F%: 21 = +++ = ap}. Because the eigenvector-eigenvalue system of equations above implies that A = 0 or 21 = +++ = aq, we see that T has no eigenvalues other than 0 and n. Find all eigenvalues and eigenvectors of the backward shift operator T € L(F™) defined by T (21, 22 23) --+) = (22,23)-++)e SouuTION: Suppose A is an eigenvalue of T. For this particular oper- ator, the cigenvalue-eigenvector equation Tz = Az becomes the system of equations41 10. Cuaprer 5. Eigenvalues and Eigenvectors z= Ay zy = Azz 24 = Azg From this we see that we can choose z) arbitrarily and then solve for the other coordinates: mary 23 = Az = Ny 2 = zg = Bay ‘Thus each d € F is an eigenvalue of T and the set of corresponding eigen- vectors is {(w, dw, ?w, Bw...):w © FP. Suppose T € £(V) and dimrangeT = k. Prove that T' has at most k+1 distinct eigenvalues. SoLuTION: Let A1,...,Am be the distinct eigenvalues of T, and let U1, .++;Um be corresponding nonzero eigenvectors. If \; # 0, then L(vj/Aj) = v;- Because at most one of A1,..., Am equals 0, this implies that at least m—1 of the vectors t1,...,Um are in range T. These vectors are linearly independent (by 5.6), which implies that m—1 Ilull?. ‘Taking square roots gives ||u|| < ||u + aul], as desired. To prove the implication in the other direction, now suppose that |[ul| < lu + aul for all a € F. Squaring this inequality, we get lull? < [fe + antl? (u + av, u + av) = (u,u) + (u,av) + (av, u) + (av, av) = [hul?? + a(u,v) + au, 0) + [aol = [lull? + 2Rea(u, v) + [a)*lfol? for alla € F. Thus —2Rea(u,v) < [a)*llu|? for alla € F. In particular, we can let a equal —t{u, v) for ¢ > 0. Substitut- ing this value for a into the inequality above gives 2t|(u, v) |? < Pl(u, v)/? loll? for all £ > 0. Divide both sides of the inequality above by t, getting51 CHAPTER 6,_Inner-Product Spaces (u,v)? < €|(u, v) loll? for all t > 0. If v = 0, then (u,v) = 0, as desired. If v # 0, set ¢ equal to 1/|lv|? in the inequality above, getting 2{(u,v)P? < |(u,»)/?, which implies that (u,v) = 0, as desired. 3. Prove that (Rem) < (9) fa) Ma IMS for all real numbers a1,...,@, and ,...,n+ SOLUTION: Suppose @1,...,@n,b1,...,6n € R. Using the usual inner product on R", we have (San)? = (Stv5a5)/VI)° j=l j=) = ((a1, V2a2,..., Vian), (b1 b2/V2,. S (Nar, V2a2, ..., Van) |? Ili, be/ V2, = os a?) (39%H) where the inequality above comes from the Cauchy-Schwarz inequality. bal VO) bal Va)? 4. Suppose u,v € V are such that lull =3, llutol=4, llu—vll=6. What number must |ju|| equal? SoLUTION: From the parallelogram equality, we have ju + vf? + [lu — uff? — 2[lull? llol? = [lu + vf? + flu — ll? — 2jlull _ 16+36-18 ~ 2 =17.52 Cxaprer 6._Inner-Product Spaces ‘Thus [ll = VIF. 5. Prove or disprove: there is an inner product on R? such that the associated norm is given by @1,22)Il = eal + eal for all (21,22) € R?. SOLUTION: We will show that there does not exist an inner product on R? such that the associated norm is given by the formula above by showing that the parallelogram equality is violated. Let u=(3,2) and v=(1,3). ‘Then utv=(4,5) and u—v=(2,-1). Using the formula above, we then have [lu + ol)? + [lu — ol]? and 2(l\ul?? + lloll?) = 2(25 + 16) = 82. ‘Thus the parallelogram equality fails, as desired. 6. Prove that if V is a real inner-product space, then 2 ty — yl? (u,v) — Webel = tu = olf 4 for all u,v € V. SoLuTION: Suppose V is a real inner-product space and u,v € V. Then ) U+v,Uut) —(u-v,u tu vl? = fu vl? _ 4 4 = llull? +2(e, v) + llol? P = 2(u, v) + lvl?) 4(u,v) 4 = (u,v),53 CHAPTER 6._Inner-Product Spaces as desired. Prove that if V is a complex inner-product space, then — Wet vl? = [lu = ol)? + flee + toll? — flu — ivf? eee eeeeaeaereree”e (we for all u,v € V. SOLUTION: Suppose V is a complex inner-product space and u,v € V. ‘Then [lu + vl}? = (ut vu to) = lull? + (u,v) + (oyu) + [lol and lu — vl]? =—(u- 0, u- 0) = —[lul? + (4,0) + (vu) = [lol?? and illu + iv||? = i(u + iv, u + iv) ull? + cu, 0) — cv, ue) + allel? and [lu — eof? = i(u —iv,u— iv) ull? + (u,v) — (vu) v2. Adding the four equations, we have [lu + vl)? — l]u — vl? + du + dul? — illu — iv]? = 4(u, v), as desired. A norm on a vector space U is a function || ||: U — [0, 00) such that |u|] = 0 if and only if u = 0, {lowl| = |alllull for all a € F and all u € U, and lu + afl < [lull + [lvl] for all u,v € U. Prove that a norm satisfying the parallelogram equality comes from an inner product (in other words, show that if || || is a norm on U satisfying the parallelogram equality, then there is an inner product (, ) on U such that ||u|| = (u, u)!/? for all u€ U).CHAPTER 6._Inner-Product Spaces CoMMENT: This is among the hardest exercises in the book. Instructors may want to simplify this exercise slightly by allowing students to consider only the case where F = R. SOLUTION: Suppose that U is a vector space and || || is a norm on U satisfying the parallelogram equality, We want to find an inner product ( , ) on U such that ||ul| = (u,u)'/? for all we U. First consider the case where F = R. For'u,v € U, define (u,v) by _ let ol? = lhe- ol? (u,v) This definition is motivated by Exercise 6 of this chapter, which gives a formula for the inner product in terms of norms. For u € U we have — uta? - - 4 — ||2ul?? — jo)? (uu) =u? = [lul??. Thus |lul] = (u,u)'/2, as desired. However, we still must show that (, ) satisfies the properties required of an inner product. Because (u, u) = ||u||? (as shown above), we have (u,u) > 0 for all ue U, with equality if and only if u = 0; these properties follow from the properties of a norm. Thus ( , ) satisfies the positivity and definiteness properties required of an inner product. To prove that (, ) is additive in the first slot, let u,v,w € U. Then w) — (v,w)) = lle tu + wh? — fu +o — wll? — fut wl? + [lu — wll? — [lo + wll? + fv — wo? 4((u +0, w) — = [fut ut wl)? + (lu — wl? + [lv — wl?) = [lento — wll? — (lu + wl]? + [lv + w?), where the first equality comes from the definition of { , ). In the last equality above, the parentheses indicate groupings to which we will apply55 Ctaprer 6._Inner-Product Spaces the parallelogram equality, which asserts that the sum of the norms squared of two vectors x,y € U can be computed from the formula 2 ey? fot? + tt = tHE Meal? Applying the parallelogram equality to the two terms in parentheses above (take z = u—w,y = uv — w for the first sum in parentheses, then x = u+w,y =v +w for the second term in parentheses) gives 4((u + v, w) — (u,w) — (v, w)) owl? | ual? = jut ot uies Metal? bea eae wpe ete + Qwl2 juve llu+v— wf 2 3 — 2Qwi|? = (ut vt ule + [ule + Meee 20 2 = (et wl? + fol) — eee tl Applying the parallelogram equality to the two terms in parentheses above (take s = u+vu+w,y = w for the first sum in parentheses, then z = u+v-—w,y = w for the second term in parentheses) gives 4((u + v, w) — (u,w) — (v,w)) e+ oe awl? | lle sel let 2 2w||? fu+ull? — ljutu—2wil? — llut+v+ 2w|? 2 2 2 (u+v,w) = (u,w) + (u,w), completing the proof that ( , ) is additive in the first slot. To prove that ( , ) is homogeneous in the first slot, let u,v € U. If nis a positive integer, then56 Guaprer 6. Inner-Product Spaces (nu,v) = (ut--++4u,v) times = (u,v) +++ + (u,v) na times =n(u,v), where the second equality comes from additivity in the first slot, which we have already verified. Replacing u with u/n in the equality above gives (u,v) = n(u/n,v), which implies that (2,2) = 2. Let m be another positive integer, and replace u with mu in the equality above, getting (Baye) = Arm, 2) 2 tue), where the second equality holds because we have already shown that (, ) is homogeneous in the first slot with respect to positive integers. We have now shown that ( , ) is homogencous in the first slot with respect to positive rational numbers. From the definition of (, ), we have =u + vl)? = [|= - of)? 4 (-u, 0) = et ol? = luo? 4 = (u,v). Combining this with the result from the previous paragraph, we can now conclude that { , ) is homogeneous in the first slot with respect to all rational numbers. Now suppose that \ € R. There exists a sequence m1,r2,... of rational numbers such that limn—oo Tn = A. Thus57. CHAPTER 6._Inner-Product Spaces u,v) Jim rales) Jig ramo) = tm rau tl? = fm, 4 In the next paragraph we will show that limy-soo ||rau + vl] = ||Au+ ol] and limyeo Ita — vl] = ||Au— |]. Combining this with the last equation above we can conclude that 2 5 vy = Batol? — pu— al 4 Au, = (au, 2), which will complete the proof that (, ) is homogeneous in the first slot. If2,y €U, then Izil = liy+@—y)Il S liv + le — al and thus Ilzll - llvll < lz - ull. Interchanging the roles of z and y, we get livll = ll < tle — yll. Because | |[zl| — |[yll| equals |x| — [lyll or llyll — [lx], we can now conclude that lilt — ltvlll < lz - yl. With ¢=rnu+v and y= Au +», this inequality gives [llrau + vl] — w+ 9]l] < [row — dull = Irn — Alllull- Because limp_oo Tn = A, this shows that lim, llrau + v]| = [Au + vf]. Replacing v with —v, we have58. CHAPTER 6._Inner-Product Spaces Lim frat — ofl = [Au lh. The last two equations are the promised ingredients that were needed for the proof that (, ) is homogeneous in the first slot. Finally, we must show that (u,v) = (v,u) (recall that we are considering the case where F = R). This last step is easy: [lu + ull? — fie — oll? (u,v) = = e+ ul? = lle = al? = (v,u). This completes the proof that (, ) is an inner product when F = R. Whew! Now consider the case where F = C. For u,v € U, define (u,v) by 2 (uy) = Het ol? = the This definition is motivated by Exercise 7 of this chapter, which gives a formula for the inner product in terms of norms. For u € U we have (2 + [fu + iv|]?é — flex — du)? 7 - (lee wf? = ff — ull? + Ufc + deulf?6 — fer — deal?s {u, u) _ (aul? +)+4 = Allul? + 2Ifull?é — 2lful?é = ul? Thus |lul| = (u,u)"/?, as desired. However, we still must show that (, ) satisfies the properties required of an inner product. Because (u, u) = ||ul|? (as shown above), we have (u,u) > 0 for all u€ U, with equality if and only if u = 0; these properties follow from the properties of a norm. Thus ( , ) satisfies the positivity and definiteness properties required of an inner product. For convenience, let’s define {, )x by (uo = eb = =o59 CHAPTER 6. _Inner-Product Spaces Here the subscript R. reminds us that (, )r was the inner product we defined when considering the case F = R. Now we are assuming that F = C, but (, )p is still well defined. Note that (u,v) = (u,o)R + (uy iv)pi. We have already proved that (, )r is additive in the first slot, and now we use that information. Let u,v, w €U. Then (utv,w) = (utv,w)r + (u+v,iw)Ri = (u,w)r + (v,w)R + (u, dw)Ri + (v, dw) Ri = ((4u)R + (u, iw)Ri) + ((v,w)R + (v, tw) Ri) = (u,w) + (v,w). Thus (, ) is additive in the first slot. To prove that (, ) is homogeneous in the first slot, let u,v €U. EAE R, then (u,v) = Qu, 2dr + Qu, de)Ri = Au, t)p + Au, dv)Ri = Alu, »), where we have used the homogeneity of (, )p in the first slot: The last equation above shows that (, ) is homogeneous in the first slot with respect to all real numbers. We must still extend this result to complex numbers. Note that fiuo) = Heat vl? = lu — ol? +l + iv — li(u+v)[Pi- leu — |? ; [liu + év)|? + [liu — tv)? = He + oll?é— [fu — 74 ~ lf + dv? + Ile — ll? 4 = i(u,v). Combining this result with additivity and homogeneity with respect to real numbers, we get that ((a + bi)u, v) = (a +bi)(u, v)60. Cuaprer 6. Inner-Product Spaces for all a,b € R. In other words, ( , ) is homogeneous in the first slot with respect to all complex numbers. Finally, we must show that (u,v) = {v,u). This last step is easy: lle oll? = flu ~ vl]? + llu + to] 4 _ etl? =o wl + 1 tru — tu (u,v) = ) (iu + v)|Pé This completes the proof that { , ) is an inner product when F = C, 9. Suppose n is a positive integer. Prove that ( 1_ sing sin2c sinnz cosx cos2r emt) Vin SR TRE eR is an orthonormal list of vectors in C[—1, n], the vector space of continuous real-valued functions on [—z, 7] with inner product (f.0)= [” se)ale) dx. Comment: ‘This orthonormal list is often used for modeling periodic phenomena such as tides. SoLuTION: First we need to show that each element of the list above has norm 1. This follows easily from the following formulas: os 2jt - sin 2j¢ zg OS focinsey dt Gi jt + sin 2jt f (cos jt)? dt = Next we need to show that any two distinct elements of the list above are orthogonal. This follows easily from the following formulas, valid when i#k61 10. ul. CHAPTER 6._Inner-Product Spaces f (sin j¢)(sin kt) dt = jsin(j — k)t + ksin(j — k)t — jsin(j + k)t+ ksinGg + k)t 245 — k)G +k) | (sin jt)(cos kt) dt = jcos(j — k)t + keos(j — k)t + j cos(j + k)t — koos(j + k)t 2(k — jG +k) {(cosit)coske at = jsin(j — k)t+ ksin(g 25 — k)G +k) f (sin jt)(cos j¢) dt = - ei, On P,(R), consider the inner product given by toa) = [ola Apply the Gram-Schmidt procedure to the basis (1,2,22) to produce an orthonofmal basis of P2(R). Soturion: Applying the Gram-Schmidt procedure to (1, 2, 2”) produces (using elementary calculus and some arithmetic) the following orthonormal basis of P2(R): (1, V8(-1 + 22), VB(1 — 6x + 627). What happens if the Gram-Schmidt procedure is applied to a list of vectors that is not linearly independent? SOLUTION: Suppose (v,...,Um) is a linearly dependent list of vectors in V. If u = 0, then at the first step of applying the Gram-Schmidt procedure to (v1,..., 0m) we will be dividing by 0 when trying to set e; = v;/||v1/|. If v, # 0, then by the linear dependence lemma (2.4), some vj is in span(u1,...,0j-1); here we choose j to be the smallest positive integer with this property. If we apply the Gram-Schmidt procedure to produce (e1,-..,e;-1) at the end of step j, then span(vj,...,0;-1) = span(ey,...,ej-1).62 CHAPTER 6._Inner-Product Spaces Thus vj € span(ex,...,€;-1). By 6.17, this implies that uj = (vj er)er +++ + (vj, 65-1) ej-1- Thus the Gram-Schmidt formula 6.23 for e; includes a division by 0, which is not allowed. 12. Suppose V is a real inner-product space and (v,...,Um) is a linearly inde pendent list of vectors in V. Prove that there exist exactly 2” orthonormal lists (€1,---, em) of vectors in V such that span(vy,...,0;) = span(es,...,¢;) for all j € {1,...,m}. So.uTion: For j = 1, the condition above states that span(v,) = span(e;). Because there are only two vectors in span(v) with norm 1 (these two vectors are v;/||vi|| and —v,/||v;||), we have only these two choices for e1. Now suppose that j > 1 and that an orthonormal list (e1,...,e;-1) has been chosen such that span(v1,...,¥j-1) = span(e1,...,e;-1). The Gram-Schmidt procedure produces e; € V such that (e1,...,e;) is an orthonormal list and span(vj,..., 0) = span(e1,..., ej). Suppose e;/ € V is another vector with these properties, meaning that (e1,---,€-1,6;’) is an orthonormal list and span(uj,..., vj) = span(e1,...,€;-1,€;'). ‘The last two equations show that span(ey,...,¢;-1,¢;') = span(e1,...,e;). In particular, ej’ € span(e1,...,e;), which implies that ej = (ej ,er)er +--+ + (e;,e;)e3 = (es esej, where the first equality comes from 6.17 (with span(ey,..., ej) replacing V and j replacing n) and the second equality holds because (e1,...,¢;-1¢;")63 CuapTer 6._Inner-Product Spaces 13, 14. is an orthonormal list. Taking norms of both sides of the last equation, and recalling that e; and ¢;' both have norm 1, we sce that |(e’,¢;)| = 1. Thus (e;',e;) = 1 or (e;',e;) = —1. Hence the last equation above implies that e;' = ej or e;' = ~e;. We have shown that there are exactly two possible choices for each ej. As j ranges from 1 to m, this gives us exactly 2" possible choices for (€15-++5€m)- Suppose (€1,..-,€m) is an orthonormal list of vectors in V. Let v € V. Prove that llvll? = Qu, er)? +--+ 1(0, em)? if and only if v € span(ey,...,€m)- SOLUTION: Extend (¢1,...,¢m) to an orthonormal basis (e1,...,€n) of V. Then u= (v,er)e, +--- + (0, en)en and Utell? = [tv ex)? +--+ + [Ct end? see 6.17. From the last equation, we see that lull? = [{v,en)[? +--+ + Kv, em)? if and only if (v,em41) = this happens if and only if = (v,en) = 0. From the first equation above, v= (v,erer +--+ (v,em)em: which happens if and only if v € span(e1,...,¢m)- Find an orthonormal basis of P2(R) (with inner product as in Exercise 10) such that the differentiation operator (the operator that takes p to p’) on P2(R) has an upper-triangular matrix with respect to this basis, SoLUTION: Because 1/ = 0, 2/ = 1, and (x2)! = 2z, the differentiation operator on P2(R.) has an upper-triangular matrix with respect to the basis (1,z,2). However, (1,2, 2?) is not an orthonormal basis, But, as can be seen from the proof of 6.27, if the Gram-Schmidt procedure is applied to this64 15. 16. 17. CHAPTER 6._Inner-Product Spaces basis, we will get an orthonormal basis with respect to which the differenti- ation operator has an upper-triangular matrix. As we saw in Exercise 10 of this chapter, the Gram-Schmidt procedure applied to (1,7, x”) gives (1, V3(-1 + 22), V8(1 — 62 + 62”). which is our desired orthonormal basis of P2(R). Suppose U is a subspace of V. Prove that dim U+ = dimV — dimU. SOLUTION: From 6.29, we know that vsueut. Thus by Exercise 17 in Chapter 2, we have dim V = dimU + dimu+, which implies that dimU+ = dimV — dimU. Suppose U is a subspace of V. Prove that U4 = {0} if and only if U = V. SOLUTION: From 6.29, we know that v=veut. This clearly implies that U4 = {0} if and only if U = V. Prove that if P € L(V) is such that P? = P and every vector in null P is orthogonal to every vector in range P, then P is an orthogonal projection. SoLUTION: Suppose P € L(V) is such that P? = P and every vector in null P is orthogonal to every vector in range P. Let U = rangeP. We will show that P equals the orthogonal projection Py. To do this, suppose véV. Then v= Put+(v— Pv). Clearly Pu € range P = U. Also, P(v — Pv) = Pu — P?v = 0, which means that v — Pu € null P. Thus v — Pv is orthogonal to every vector in U. In other words, v— Pu € U+. Thus the equation above writes v as the sum of a vector in U and a vector in U+. In this decomposition, the vector in U equals, by definition, Pyv. Hence Pu = Pyv, as desired.65 18. 19. CHAPTER 6._Inner-Product Spaces Prove that if P € £(V) is such that P? = P and Poll < llell for every uv € V, then P is an orthogonal projection. SovuTion: Suppose u € range P and w € null P. If we can show that (u,w) = 0, then by the previous exercise we can conclude that P is an orthogonal projection. Because u € range P, there exists u’ € V such that u= Pu. Applying P to both sides of this equation, we have Pu= Pu! =P =u Because w € null P, this implies that P(utaw)=u for every a € F. Thus llul? = [P(e + aw) |? B50: — 10722? + 6325) ete Now compute Pyv using 6.35 (with m = 6), getting _ 105(1485- 1530? +04) — 315(1155~ 1257? +24) Puv= ano = 4x8 * 693(945 — 105n? + x4 + Brt0 Finally, 6.36 and the discussion following 6.42 show that the function above is the one we seek. 24. Find a polynomial g € P2(R) such that Lf (5) = [ wa)ala) ax ‘Oo for every p € P2(R). SOLUTION: We will need an orthonormal basis of P(R), where the inner product of two polynomials in P2(R) is defined to be the integral from 0 to 1 of the product of the two polynomials. An orthonormal basis of P2(R) was already computed in Exercise 10 of this chapter. Specifically, let ex(z) =1 €2(z) = V3(-1 + 2x) e3(x) = VB(1 — 6 + 62”).70 25. 26. Cuarren 6. {nner-Product Spaces ‘Then (¢1, €2,€3) is an orthonormal basis of P2(R). Define a linear functional y on P2(R) by #0) = 2(5). We seek q € P2(R) such that o(p) = (p,q) for every p € P2(R). By the formula given in the proof of 6.45, we have a= pler)er + pler)er + vles)es. Evaluate the right side of the equation above to get 3 2 a(z) = 3+ 15x — 1527. Find a polynomial ¢ € P2(R) such that 1 : ff wa\oosne) de =f n)ale) de 0 ‘O. for every p € P2(R). SOLUTION: Define a linear functional y on P2(R) by ote) = [°s(2)e00n) de, We seek g € P,(R) such that ¢(p) = (p,q) for every p € P2(R), where the inner product on P2(R) is defined as in the previous exercise. Letting €1,€2,¢3 be as in the previous exercise, but using our new definition of y, we again have 9 = yler)er + oler)en + oles)es. Evaluate the right side of the equation above to get 12 — 2dr (2) =a Fix a vector v € V and define T € £(V,F) by Tu = (u,v). For a € F, find a formula for T*a, Souution: Because T € L(V,F), we know that T* € L(F,V). Fix a€F. Then Ta is the unique vector in V such that7 CHAPTER 6._Inner-Product Spaces (a) (Tu, a) = (u,T*a) for all ti € U. The inner product on the right is the inner product in V, but the inner product on the left is the usual inner product on F: the product of the entry in the first slot with the complex conjugate of the entry in the second slot. Thus (Tu, a) = (Tu)a = (u,v) (b) = (u, av). Comparing (a) with (b) gives (u,T*a) = (u,av) for all u€U. Thus T*a = av. 27. Suppose n is a positive integer. Define T € £L(F") by T(2iy.++12n) = (0, 21,..-, Zn-1). Find a formula for T*(z1,...2n)- Souution: Fix (21,.-.,2n) € F". Then for every (w1,...,Wn) € F", we have ((w1,- ++ Wn), Tiss 2n)) = (Pwr, +, Wn); (2102s Zn) = (0, w1,---sWn-1); (21)---12n)) = wie +--+ wie = (ws .++,Wn)s (225+++)2n50))- Thus T "(21,0455 2n) = (22)--+5 2n,0)- 28. Suppose T’ € £(V) and A € F. Prove that A is an eigenvalue of T if and only if is an eigenvalue of T*. SOLUTION: We have72 Cuarrer 6._tnner-Product Spaces 2 is an not eigenvalue of T’ <=> T — AI is invertible <=> S(T - AI) =(T-ANS =I for some S$ € L(V) <=> (T -AM)*S* = S*(T— XI) = for some S € L(V) <=> (T — AM)" is invertible <=> T* — Nis invertible <=> J is not an eigenvalue of T*. Thus A is an eigenvalue of T' if and only if ) is an eigenvalue of T*. 29. Suppose T € L(V) and U is a subspace of V. Prove that U is invariant under T if and only if U- is invariant under T*. Souution: First suppose that U is invariant under T. To prove that U?+ is invariant under T*, let v € Ut. We need to show that T*v € U+. But for every u € U (because if u € U, then Tu € U and hence Tu is orthogonal to v, an element of U+). Thus T*v € U+, and hence U* is invariant under T*, as desired. ‘To prove the other direction, now suppose that U+ is invariant under T*. ‘Then by the first direction, we know that (U+)+ is invariant under (T*)*. But (U+)+ = U (by 6.33) and (7*)* = T, so U is invariant under T, completing the proof. 30. Suppose T € £(V,W). Prove that (a) Tis injective if and only if T* is surjective; (b) Tis surjective if and only if T* is injective. SoLUTION: First we prove (a): T is injective <=> nullT = {0} <=> (rangeT*)* = {0} <> rangeT* = W <=> T" is surjective ,73 3. 32. CHAPTER 6._Inner-Product Spaces where the second line comes from 6.46(c). Now that (a) has been proved, (b) follows immediately by replacing T with T” in (a). Prove that dim null T* = dim null T + dim W — dimV and dim rangeT* = dimrangeT for every T € L(V,W). Sotution: Let T € £(V,W). Then dim null T* = dim(rangeT)+ = dim W —dimrangeT = dimnull T + dim W — dim V, where the first equality comes from 6.46(a), the second equality comes from Exercise 15 of this chapter, and the third equality comes from 3.4. This proves the first equality that we seek. To prove the second equality, note that dim rangeT* = dim W — dim null T* im V — dim null T = dimrangeT, where the first and third equalities come from 3.4 and the second equality comes from the first part of this exercise. This proves the second equality that we seek. Suppose A is an m-by-n matrix of real numbers. Prove that the dimension of the span of the columns of A (in R™) equals the dimension of the span of the rows of A (in R®). Souution: Let T € £(R",R™) be such that the matrix of T (with respect to the standard bases) equals A. Then rangeT’ equals the span of the columns of A. Thus4 CHAPTER 6. _Inner-Product Spaces dimension of the span of the columns of A = dimrangeT dimrangeT* = dimension of the span of the columns of M(T*) dimension of the span of the columns of the transpose of A = dimension of the span of the rows of A, where the second equality comes from the previous exercise and the fourth equality comes from 6.47.CHAPTER 7 Operators on Inner-Product Spaces Make (IR) into an inner-product space by defining 1 toa = f ale)ateyas. Define T € £(P2(R)) by T(a9 + 12 +4922) = az. (a) Show that Tis not self-adjoint. (b) The matrix of T with respect to the basis (1, x, x?) is 000 010]. 000 This matrix equals its conjugate transpose, even though T'is not self- adjoint. Explain why this is not a contradiction. SoLuTion: (a): Note that {T1,2) = (0,2) =0 but (1,72) = (1,2) =i => 1576 CHAPTER 7. Operators on Inner-Product Spaces ‘Thus (T1,z) # (1,Tz), which shows that T is not self-adjoint. (b): The result stating that the matrix of T* is the conjugate transpose of the matrix of T has as a hypothesis that we are working with orthonormal bases (see 6.47). Because (1, x, x”) is not an orthonormal basis of Po(R), we cannot compute the matrix of T* with respect to this basis by taking the conjugate transpose of the matrix of T. Prove or give a counterexample: the product of any two self-adjoint opera- tors on a finite-dimensional inner-product space is self-adjoint. Soturion: Let 5,1 € L(F?) be the operators whose matrices (with respect to the standard basis) are given by 10 ms)= [5 2 o1 wa sae=[° 3]. Each of these matrices obviously equals its conjugate transpose, and hence S,T are self-adjoint. Now M(ST) = M(S)M(T) = [: |: Because M(ST) does not equal its conjugate transpose, ST is not self- adjoint. Thus we have an example of two self-adjoint operators whose prod- uct is not self-adjoint. Of course there are also many other examples. ComMENT: Suppose $,T’ € L(V) are self-adjoint. Then ST is self- adjoint if and only if ST = TS (as is casy to see). (a) Show that if V is a real inner-product space, then the set of self-adjoint. operators on V is a subspace of L(V). (b) Show that if V is a complex inner-product space, then the set of self- adjoint operators on V is not a subspace of L(V). SovuTion: (a): Suppose V is a real inner-product space. Obviously the zero operator is self-adjoint. Furthermore, if S,T € L(V) are self-adjoint, then (S+T)'=S°+T" =S+T,7 Cuaprer 7. Operators on Inner-Product Spaces and thus $+-T is self-adjoint. Finally, if T € £(V) is self-adjoint anda €R, then (ar)* = eT" =a, and thus aT is self-adjoint. We have shown that the set of self-adjoint operators on V contains the zero operator and that it is closed under addition and scalar multiplication. Thus the set of self-adjoint operators on V is a subspace of L(V). (b): Suppose now that V is a complex vector space. The identity operator 1 is self-adjoint, but (if)* = -iI so iI is not self-adjoint. Thus the set of self-adjoint operators on V is not closed under scalar multiplication and hence it is not a subspace of L(V). Suppose P € L(V) is such that P? = P. Prove that P is an orthogonal projection if and only if P is self-adjoint. Sotution: First suppose that P is an orthogonal projection. Thus there is a subspace U of V such that P = Py. Suppose v1, v2 € V. Write Vp HUt+wi, V2 = U2 + We, where uy, up € U and wi, w2 € U+ (see 6.29). Now (Por, v2! (uz, uz + we) = (un, u2) + (un, we) = (u, U2) = (ur, U2) + (wi, u2) = (ur + wi, v2) = (u, Pg). Thus P = P*, and hence P is self-adjoint. ‘To prove the implication in the other direction, now suppose that P is self-adjoint. Let » € V. Because P(v— Pu) = Pu — P*v = 0, we have v— Pv null P = (range P*)+ = (range P)+, where the first equality comes from 6.46(c). Writing v Put(v— Pv),78 CuapTER 7. Operators on Inner-Product Spaces we have Pu € range P and (v — Pv) € (range P)'. Thus Pu = Prange pv. Because this holds for all v € V, we have P = Prange P, which shows that P is an orthogonal projection. 5. Show that if dimV > 2, then the set of normal operators on V is not a subspace of £(V). SoLuTION: Suppose dimV > 2. Let (c1,...,e,) be an orthonormal basis of V. Define 5,T € L(V) by Saver +--+ nen) = a2er — aier and T(aye1 + +++ +Gnen) = arer + a1e2. A simple calculation verifies that S*(aie1 +++ + anen) = —a2€1 + aye. From this formula, another simple calculation shows that SS* = S*S. Yet another simple calculation shows that T' is self-adjoint. Thus both S and T are normal. However, 5 +T is given by the formula (S4T)(a1e1 + +++ + @nén) = aver. A simple calculation verifies that (S+T)*(a1e1 + +++ + nen) = 2arer. A final simple calculation shows that ($+ 7T)(S+T)* 4 (§+T)*(S+T). In other words, S + T is not normal. Thus the set of normal operators on V is not closed under addition and hence is not a subspace of £(V). 6. Prove that if T € L(V) is normal, then rangeT = rangeT*. So.ution; Suppose T is normal. Then rangeT = (null T*)*79 CHaprer 7. Operators on Inner-Product Spaces where the first equality comes from 6.46(d), the second equality comes from 7.6 (see especially the marginal comment at 7.6), and the third equality comes from 6.46(b). 7. Prove that if T € £(V) is normal, then nullT* =nullT and rangeT* = rangeT for every positive integer k. SoLUTION: Suppose T’ € £(V) is normal and that k is a positive integer. Obviously we can assume that k > 2. First we will prove that null T* = null T. If v € nullT, then Ty = T*-(Tv) =T19 =0, and so v € null T*, Thus null 7 null T*. To prove an inclusion in the other direction, suppose now that v € null T*, Then (Tey, Ty) = (TTT 1p, TH 10) = (T°T*y, TH) = (0, Tv) =0, where the second equality holds because T*T = TT*. The last equality above implies that T*T*-!v = 0. Thus 0 = (T*T*'y, TE) = (Tey, Phy), Hence T*-1y = 0. In other words, v € null T*~!. The same argument, with k replaced with k — 1, shows that v € null T*-?. Repeat this process until reaching the conclusion that v € nullT. This shows that null T* C nullT, completing that proof that null 7* = null T. Now we will show that rangeT* = rangeT. If v € rangeT*, then there exists u € V such that v = T*u = T(T*")u, which implies that v € rangeT. Thus range T* C rangeT. Note that80 10. CHAPTER 7. Operators on tnner-Product Spaces dim rangeT* = dim V — dim null T* = dimV — dimnullT = dimrangeT, where the first and third equalities come from 3.4 and the second equality comes from the first part of this exercise. Because rangeT* and rangeT have the same dimension and one of them is contained in the other, these two subspaces of V must be equal, completing the proof. Prove that there does not exist a self-adjoint operator T’ € £(R3) such that T(1,2,3) = (0,0,0) and T(2, 5,7) = (2,5, 7). SOLUTION: Suppose T € £(R§) is such that T(1,2,3) = (0,0,0) and T(2,5,7) = (2,5,7). Obviously (1,2,3) is an eigenvector of T with eigen- value 0 and (2,5,7) is an eigenvector of T with eigenvalue 1. If T were self-adjoint, then eigenvectors corresponding to distinct eigenvalues would be orthogonal (see 7.8). Because (1,2,3) and (2,5,7) are not orthogonal, T cannot be self-adjoint. Prove that a normal operator on’ a complex inner-product space is self- adjoint if and only if all its eigenvalues are real. COMMENT: This exercise strengthens the analogy (for normal operators) between self-adjoint operators and real numbers. SoLuTion: Suppose V is a complex inner product space and T € £(V) is normal. If T is self-adjoint, then by 7.1 alll its eigenvalues are real. Conversely, suppose that all the eigenvalues of T’ are real. By the com- plex spectral theorem (7.9), there is an orthonormal basis (e1,...,€n) of V consisting of eigenvectors of T. Thus there exist real numbers \1,...,An such that Te; = Aye; for j = 1,...,2. The matrix of T with respect to the basis (€,...,€n) is the diagonal matrix with A1,..., An on the diagonal. This matrix equals its conjugate transpose. Thus T = T*. In other words, T is self-adjoint, as desired. Suppose V is a complex inner-product space and T € L(V) is a normal operator such that T° = TS. Prove that T is self-adjoint and T? = T. SoLUTION: By the complex spectral theorem (7.9), there is an orthonor- mal basis (¢1,...,€n) of V consisting of eigenvectors of T. Let Au,..-,An be the corresponding eigenvalues. Thus81 ll. 12. CHAPTER 7._Operators on Inner-Product Spaces Te; = dye} for j = 1,...,n. Applying T repeatedly to both sides of the equation above, AjSe; and Te; = dj8e;. Thus A,° = Aj8, which implies that Xj equals 0 or 1. In particular, all the eigenvalues of Tare real. This implies (by the previous exercise) that T is self-adjoint. Applying T to both sides of the equation above, we get Te; = djPe4 = Ayes =Te;, where the second equality holds because A; equals 0 or 1. Because T? and T agree on a basis, they must be equal. Suppose V is a complex inner-product space. Prove that every normal operator on V has a square root. (An operator S € L(V) is called a square root of T € L(V) if S? =T.) SoLuTION: Suppose T’ € L(V) is normal. By the complex spectral theorem (7.9), there is an orthonormal basis (¢1,...,€n) of V consisting of eigenvectors of T. Thus there exist complex numbers j,-.., An such that Aje; for j = 1,...,n. Define S to be the operator on V such that AwMe; for j +n; here A;'/? denotes a complex square root of A; (every nonzero complex number has two square roots—it does not matter which one is chosen). Then, as is easy to verify, S? = T. Thus S is ‘a square root of T. Give an example of a real inner-product space V and T € £(V) and real numbers a, 8 with a? < 48 such that T? + aT + AI is not invertible. COMMENT: This exercise shows that the hypothesis that T is self-adjoint is needed in 7.11, even for real vector spaces. Sotution: Let T € £(R?) be the counterclockwise rotation on R2; so T(z,y) = (—y,z) for (z,y) € R?. Thus T? = —/. Taking a = 0 and B = we have a? < 48 and P+oT+Hl=T? 41 =0. In particular, T? + aT + BI is not invertible.82 14, CHAPTER 7. Operators on Inner-Product Spaces Prove or give a counterexample: every self-adjoint operator on V has a cube root. (An operator $ € £(V) is called a cube root of T € L(V) if S3 =T.) SouuTION: Suppose T € L(V) is self-adjoint. By the spectral theorem (7.13), there is an orthonormal basis (¢1,...,€n) of V consisting of eigen- vectors of T. The corresponding eigenvalues must be real (by 7.1). Thus there exist real numbers Ay,...,An such that Te; = je; for j = 1,...,n. Define S' to be the operator on V such that Se; = \;/Se; for j = 1,...,n. Then, as is easy to verify, S? = T. Thus S is a cube root of T, completing the proof that every self-adjoint operator on V has a cube root. Suppose T € £(V) is self-adjoint, A € F, and « > 0. Prove that if there exists v € V such that [Jul] = 1 and |Tv Av] |Tv — dv]? = Or =A) (render +++ + An — A) (vs endenll? = [Ar — AP I(v, 1)? +--+ Ln — AP T(ey en)? > (min{[Ar — Al?,...]4n — AP?}) (dvs ex)/? + - =min{|A1 — A[?,...5 [An — AP?} + [(v,en)/?) Thus € > |A; — Al for some j. In other words, there is an eigenvalue whose distance from ) is less than ¢, as desired.83 15. 16, 17. 18. Carrer 7. Operators on Inner-Product Spaces Suppose U is a finite-dimensional real vector space and T € £(U). Prove that U_has a basis consisting of eigenvectors of T if and only if there is an inner product on U that makes T into a self-adjoint operator. SoLuTION: First suppose that U has a basis (e1,..., en) of eigenvectors of T. Because (é1,...,¢n) is a basis of U, every element of U can be uniquely written as a linear combination of (e1,...,@n). Thus we can define an inner product on U by (aye + +++ + Open, beer + +++ + Bnen) = a1by ++++ + Onda. It is easy to verify that this is indeed an inner product on U and that (e1,..-,€n) is on orthonormal basis of U with respect to this inner product. Because each e; is an eigenvector of T, the operator T has a diagonal matrix with respect to the orthonormal basis (e1,...,€n)- Thus T is self-adjoint. Conversely, now suppose that there is an inner product on U that makes T into a self-adjoint operator. Then by the spectral theorem (7.13), U has a basis consisting of eigenvectors of T. Give an example of an operator T on an inner product space such that T has an invariant subspace whose orthogonal complement is not invariant under T. Comment: This exercise shows that 7.18 can fail without the hypothesis that T is normal. Sotution: Define T € £(F*) by T(w,z) = (z,0). Then T(w, 0) = (0,0) for all w € F. Thus the subspace U defined by U = {(w,0) : w € F} is invariant under T. However, U4 = {(0,z) : € F}, which is not invariant under T’ because (0,1) € U+ but T(0, 1) = (1,0) ¢ U+. Of course there are also many other examples. Prove that the sum of any two positive operators on V is positive. SoLuTION: Suppose S and T are positive operators on V. Because S and T are self-adjoint, so is $ + T. Furthermore, ((S + T)u, vu) = (Sv, v) + (Tv, v) 20. Thus S +T is a positive operator, as desired. Prove that if T € £(V) is positive, then so is T* for every positive integer k.84 CHAPTER 7. Operators on Inner-Product Spaces SoLution: Suppose T € L(V) is positive and k is a positive integer. Then T* is self-adjoint (because T is self-adjoint). First consider the case where k is an even integer. Then we can write k = 2m for some positive integer m. Now (Tv, v) = (17,0) = (Tu, T™0) 20 for every v € V, where the second equality holds because T is self-adjoint. The inequality above shows that T* is positive, as desired. Now consider the case where k is an odd integer. Then we can write k& = 2m +1 for some nonnegative integer m. Now (Tey, v) = (T?™*1y, v) = (T(I™v), Tv) 20 for every v € V, where the second equality holds because T is self-adjoint and the inequality holds because T is positive. The inequality above shows that T* is positive, as desired. Suppose that T is a positive operator on V. Prove that T is invertible if and only if (Tv,v) >0 for every v € V \ {0}. SoLuTION: First suppose that T is invertible. By 7.27, there exists an operator S € L(V) such that T = $*S. Suppose v € V \ {0}. Then Su £0 because otherwise we would have Tv = S*Sv = 0, which would contradict the invertibility of T. Now (Tv, v) = (S*Sv, v) = (Su, Sv) >0, as desired. Now suppose that (Tv, v) > 0 for every v € V \ {0}. In particular, this means that Tv # 0 for every v € V \ {0}. Thus T is injective, and hence T is invertible (see 3.21), as desired.85 20. 21. 22. 23. CHAPTER 7. Operators on Inner-Product Spaces Prove or disprove: the identity operator on F* has infinitely many self- adjoint square roots. Sotution: For each ¢ € [-1,1], the operator whose matrix (with respect to the standard basis) equals t vi-#® vI-@ +t | is self-adjoint and a square root of the identity operator, as can be verified by squaring the matrix above. Thus the identity operator has infinitely many self-adjoint square roots. Prove or give a counterexample: if $ € £(V) and there exists an orthonormal basis (€1,...,€n) of V such that || Se,|| = 1 for each e;, then S is an isometry. SOLUTION: Define S € £(F*) by S(w, 2) = (w + 2,0). With the usual inner product on F?, the standard basis ((1,0), (0, 1)) is an orthonormal basis of F?. Note that ||$(1,0)|| = [|$(0,1)|| = 1. However, $ is not an isometry because ||5(1,—1)|| = 0. Of course there are also many other examples. Prove that if $ € £(R*) is an isometry, then there exists a nonzero vector z € RS such that S?z =z. SoLution: Suppose S € £(R*) is an isometry. Then there is a basis of R§ with respect to which S has a block diagonal matrix, where each block on the diagonal is a 1-by-1 matrix containing 1 or —1 or is a 2-by-2 matrix (see 7.38). Because R® has odd dimension, at least one of these blocks must be a 1-by-1 matrix. In other words, either 1 or —1 must be an eigenvalue of S. Thus there is a nonzero vector z € R® such that Sz = Ax, where A= 41. Hence S?x = $(Sz) = S(Az) = ASz = Na Define T € £(F*) by T (21, 22, 23) = (23, 221,322). Find (explicitly) an isometry S € £(F*) such that T = SVT*T.86, 24. CHAPTER 7. Operators on Inner-Product Spaces SOLUTION: With respect to the standard basis of F?, we have ool M(T)=|2 00}. 030 Thus 020 M(T")=|0 0 3]. 100 Computing the product M(T*)M(T), which equals M(T"T), we get 400 M(r'T)=|0 9 0]. oot From the matrix above, we see that (T*T)(z1, 22,23) = (42, 9z2, 23). Thus VTT (21, 22, 23) = (221,322, 29). Hence if we define S € L(F%) by S(21, 22, 23) = (23,21, 22), then S is an isometry and T = SVT*T. Suppose T € L(V), S € £(V) is an isometry, and R € L(V) is a positive operator such that T = SR. Prove that R= /T*T. ComMENT: This exercise shows that if we write T as the product of an isometry and a positive operator (as in the polar decomposition), then the positive operator must equal VT*T. SOLUTION: Taking adjoints of both sides of the equation T = SR, we have T= RS* = RS*, where the last equation holds because R is positive (and hence self-adjoint). Multiplying together our formulas for T* and T,, we get T°T = RS*SR =F,87 25, 26. 27. CHAPTER 7. Operators on Inner-Product Spaces where the last equation holds because S is an isometry (and hence S*S = I by 7.36). The equation above asserts that R is a square root of T*T; because Ris positive, this implies that R = VT"T. Suppose T' € L(V). Prove that T is invertible if and only if there exists a unique isometry S € £(V) such that T= SYT*T. SOLUTION: First suppose that T is invertible. The polar decomposition (7.41) states that there exists an isometry S € £(V) such that T=SVT"T. Because T is invertible, this implies that VT°T is invertible (see Exercise 22 in Chapter 3). Thus the equation above implies that S = T(VT*T)'. Because S$ must be given by this formula, we see that there is a unique operator $ € £(V) such that T= SVT*T, as desired. Now suppose that there exists a unique isometry S € C(V) such that T = SVT*T. This means that the linear map Sp in the proof of the polar decomposition (7.41) must be 0 because otherwise we could replace Sp with —S2 and get another choice for S. But range S2 equals (range), and hence (range T) = {0}. This implies that rangeT = V, which implies that T is invertible (by 3.21), as desired. Prove that if T € L(V) is self-adjoint, then the singular values of T equal the absolute values of the eigenvalues of T (repeated appropriately). SoLuTion: Suppose T € L(V) is self-adjoint. There exists an orthonor- mal basis (e1,...,¢n) of V consisting of eigenvectors of T. Thus Tes = dye; for each j, where \1,...,n € R are the eigenvalues of T. Thus T'Te; = Te; = (A;)?e5 for each j. The equation above implies that VT*Te; = |)jl Thus the singular values of T are |Ai|,..-,[Anl, a8 desired. for each j. Prove or give a counterexample: if T € £(V), then the singular values of T? equal the squares of the singular values of T. SOLUTION: Define T' € £(F*) by88 28. 29. 30. CHAPTER 7. Operators on Inner-Product Spaces T (21,22) = (22,0). Then T*T(21,z2) = (0,22) and hence VT*T(21,22) = (0,22). Thus the eigenvalues of VI*T are 0,1. Hence the singular values of T are 0,1. However, T? = 0, so the singular values of J? are 0,0. Thus for this operator T’, the singular values of T? do not equal the squares of the singular values of T. Of course there are also many other examples. Suppose T € £(V). Prove that T is invertible if and only if 0 is not a singular value of T. Souution: If § € £(V) and ST = TS = I, then taking adjoints we get T*S* = S*T* = I. Thus if T is invertible, then so is T*. Now T is invertible <=> T and 7" are invertible <= T'T is invertible <=> VI'TVT'T is invertible <=> VI'T is invertible <= 0 is not an eigenvalue of VI*T <= 0 is not a singular value of T, where the second and fourth equivalences follow from Exercise 22 in Chap- ter 2. Suppose T € L(V). Prove that dimrangeT equals the number of nonzero singular values of T. SOLUTION: By the singular value decomposition (7.46), there exist or- thonormal bases (uj,..-,tn) and (wi,..., Wn) of V such that Tv = 5, (v, u)w +-+- + Sa(v, Un) Wn for every v € V, where 51,..., 5, are the singular values of T. For each j, we have Tu; = sjw;. Thus each w; corresponding to a nonzero s; is in rangeT. The equation above also shows that the w;’s corresponding to nonzero s;’s span rangeT. Thus dimrangeT equals the number of nonzero singular values of T. Suppose S € £(V). Prove that $ is an isometry if and only if all the singular values of S equal 1.89 Cuaprer 7. Operators on Inner-Product Spaces SOLUTION: We have Sis an isometry => S*S =I <> VEE =I <= all the eigenvalues of V5*S equal 1 <=> all the singular values of 5 equal 1, where the first equivalence comes from 7.36 and the third equivalence comes from the spectral theorem (7.9 or 7.13) applied to the self-adjoint operator VS*S. 31. Suppose T, 72 € L(V). Prove that T; and Tp have the same singular values if and only if there exist isometries $1, S2 € £(V) such that Ty = SiTSo. Souurion: First suppose that 7, and 72 have the same singular values S1,-++;Sn+ By the singular-value decomposition (7.46), there exist orthonor- mal bases (€1,...,¢n); (fis---+fn)s (¢hs---.€n)s fty---s Fa) of V such that Ty = si(v,er)fr +--+ + 5n(0, en) far Tau = 81(v,€4) fi +--+ + sults en) fn for every v € V. Define 51, 52 € L(V) by Si(aif, +++ + anfa) = ari +--+ + onfas S2(aye1 + +++ + Qnen) = aye) ++-> + ane. Then WSi(enf{ +--+ an fa)I? = laa fi +++ + an fall? ++ lanl? “tanfal?, and thus Sj is an isometry. Similarly, S2 is an isometry. This implies that Sa° = S27" (see 7.36). In particular, Sp*el = e;. Now for v € V we have Ta(S2v) s1(S2v,e\) ft +++ + in(Savsen) fa = sv, S2°eh)fi +++ + Sav, S2°en) fa = silver) fi to-- + Salven) Sn-90 32. CHAPTER 7. Operators on Inner-Product Spaces Thus Si (TaSav) = s1(v,e1) Sift +++ + Snr en) Sife = s1(vye1) fi t+++ + $n (ven) fn =Tyv for every v € V. Hence S1TS2 = Th, as desired. To prove the implication in the other direction, now suppose that there exist isometries 51,52 € L(V) such that 7, = S;T2S2. Using 7.36, we have Ty"Ty = Sq"T2" S151 T2S2 = S271 T*T,Sp. This implies that T,*T, and T)*T> have the same eigenvalues (and that the corresponding spaces of eigenvectors have the same dimensions). Thus T} and Ts have the same singular values. Suppose T' € £(V) has singular-value decomposition given by Tu = 81(v,e1) fi +++ + $n(v,€n) fr for every v € V, where s1,..., Sq are the singular values of T and (e1,.-.,€n) and (f1,..., fn) are orthonormal bases of V. (a) Prove that Try = si(v, filer +--+ + Sa(v, faden for every u € V. (b) Prove that if T is invertible, then Ayer 5.4 Wi faden st Sr Tvs for every v EV. Sotution: (a): Fix v € V. Then (w,T*v) = (Tw, v) = (s1(w,e1) fi +--- + Sn (w, en) fas 0) = 51(w, €1)(fi,d) +--+ + Sn(w, en) (fas) = (w, s1(v, filer +°-* + 5n(¥, fn)en)go 33. CHAPTER 7. Operators on Inner-Product Spaces for all w € V. This implies that Tv = 51(v, filer t++++ Sn(v, fndens as desired. (b): Suppose T is invertible. Let v € V and let w — Wfler 5.4 Wfaden, 1 Sn none of the singular values s1,..., 8, equals 0 (see Exercise 28 of this chap- ter), so this makes sense. Now ip = WATE |, (vs fa)Ten 81 Bn _Wuhdorh 5, (ufndonfn 1 en " (v, fifi to + (0 fa) fn =v. Thus w = T~!v, as desired. Suppose T' € L(V). Let § denote the smallest singular value of T,, and let s denote the largest singular value of T’. Prove that sllull < Tell < sllv|] for every v € V. Souurion: Let v ¢ V. By the singular value decomposition (7.46), there exist orthonormal bases (11,..., tn) and (w1,...,Wn) of V such that Tv = 51(v,u)up +++ + 5n(v, Un) wn, where s1,...,5n are the singular values of T. Because (u1,...,tq) and (w1,---,t,) are both orthonormal bases of V, we have H lle? = (vu)? + --- + Kv, ttn) ?) S$ s17(vyt)[? +++ + Sn? l(v, Un)? = |ITeI??,92 34. CHAPTER 7. Operators on Inner-Product Spaces giving the first desired inequality. Also, De]? = s17[{v,an))? +--+ + sn |(o, en)? S$ 8*(\(v,un)P ++ + [(, tn) ?) = sifull?, giving the second desired inequality. Suppose T’, T” € L(V). Let s! denote the largest singular value of 7°, let s” denote the largest singular value of 7", and let s denote the largest singular value of T’ + 7”. Prove that s < s' +s". SoLuTion: Let T= 7’ +7”. Because s is a singular value of T, we know that s is an eigenvalue of V7*T. Thus there exists a vector u € V such that |jul] = 1 and VT*Tv = sv. Now s = ||sol] =||VFTo = [Toll = |Tv+T"ol| S [Pell + (ITI) Ss'llul| + slo] =s' +s" where the third line above comes from 7.42 and the sixth line above comes from the previous exercise.CHAPTER 8 Operators on Complex Vector Spaces Define T € £(C*) by T(w, z) = (z,0). Find all generalized cigenvectors of T. SoLution: Suppose ) is an eigenvalue of T. For this particular operator, the eigenvalue-eigenvector equation T(w,z) = X(w,z) becomes the system of equations z=hw O=drz. If A £0, then the second equation implies that z = 0, and the first equation then implies that w = 0. Because an eigenvalue must have a nonzero eigen- vector, this shows that 0 is the only possible eigenvalue of T. For \ = 0, the equations above show that z must equal 0, but w can be arbitrary. Thus 0 is indeed an eigenvalue of T,, and the set of eigenvectors corresponding to this eigenvalue is {(w,0): w € F}. Note that T? = 0. Thus every vector in C? is a generalized eigenvector of T (corresponding to the eigenvalue 0). 9394 Cuaprer 8. Operators on Complex Vector Spaces Define T € £(C?) by T(w,z) = (-z,w). Find all generalized eigenvectors of T. SOLUTION: On page 78 of the textbook we saw that the eigenvalues of T are i and —i. Note that T? = -I. ‘The set of generalized eigenvectors of T corresponding to the eigenvalue i equals null(T’ — iZ)? (by 8.7). To compute this, note that (T - il)? =T? -2T -1 =-21 - 2T = -2i(T — il). Thus nul(T — iI)? equals the set of eigenvectors of T corresponding to the eigenvalue i. On page 78 of the textbook we noted that this equals {(a, -ia) : a € C}. The set of generalized eigenvectors of T’ corresponding to the cigen- value —é equals null(T'+ iI)? (by 8.7). To compute this, note that (T+ =T? +217 -1 =-20+2iT = 2(T +i). Thus null(T + i)? equals the set of eigenvectors of T’ corresponding to the eigenvalue —i. On page 78 of the textbook we noted that this equals {(a, ia) :a € Ch. Suppose T € £(V), mis a positive integer, and v € V is such that T™—'v #0 but T™v = 0. Prove that (v, Tv, Tv, ..., 7710) is linearly independent. SOLUTION: Suppose a9,a},02,...,@m-1 € F are such that gu + ay Tv + a2T7v +--+ + @m—iT™'v = 0. Because T™v = 0, if we apply T”~! to both sides of the equation above, we get agT™-1v = 0. Because T™~'y # 0, this implies that a9 = 0. Thus the equation above can be rewritten as95 CHAPTER 8. Operators on Complex Vector Spaces, ay Tv + agT?u +++ + a_—1T™ 1v = 0. Applying T™-? to both sides of this equation, we get a,T™~!v = 0. Thus a; = 0. Continuing in this fashion, we have a9 = a) = a2 = ++» = am—1 = 0, which means that (v,Tv,T?v,...,27”-1v) is linearly independent. Suppose T € £(C) is defined by T(21, 22,23) = (22, 23,0). Prove that T has No square root. More precisely, prove that there does not exist S € £(C*) such that S$? = T. Souution: Note that T? = 0. Suppose there exists S € £(C3) such that S? = T. Then S* = T? = 0, so S is nilpotent. By 8.8, this implies that 5° =0. Thus T= = 53% =0. But T?(z1, 22, z3) = (23,0,0), 80 T? is not the 0 operator, contradicting the equation above. This contradiction shows that our supposition that there exists S € £(C3) such that S? = T must have been false. Suppose S,T' € L(V). Prove that if ST is nilpotent, then T'S is nilpotent. So.ution: Suppose ST is nilpotent. Thus there exists a positive integer nsuch that (ST)" = 0. Now (T)"*! = (TS)(TS)...(TS) = T(ST)(ST)...(ST)S (ST)"S = (T)(0)(S) =0, and thus TS is nilpotent. Suppose N € L(V) is nilpotent. Prove (without using 8.26) that 0 is the only eigenvalue of N. SOLUTION: There is a positive integer m such that N™ = 0. This implies that N is not injective, so 0 is an eigenvalue of N. Conversely, suppose ) is an cigcnvalue of N. Then there exists a nonzero vector v € V such that96 CtiaPTER 8. Operators on Complex Vector Spaces dv = Nv. Repeatedly applying N to both sides of this equation shows that A™y = N™y =0. Thus A = 0, as desired. Suppose V is an inner-product space. Prove that if N € L(V) is self-adjoint and nilpotent, then N =0. SowvTion: Suppose N € £(V) is self-adjoint and nilpotent. Because N is self-adjoint, there is an orthonormal basis (e1,...,¢n) of V consisting of eigenvectors of N (by the spectral theorem). Because N is nilpotent, 0 is the only eigenvalue of N (see Exercise 6 of this chapter). Thus the eigenvalue corresponding to each ej must equal 0. In other words, Ne; = 0 for each j. Because (€1,...,€n) is a basis of V, this implies that N = 0. Suppose N € L(V) is such that nullN@=¥-1 4 null N@™Y, Prove that N is nilpotent and that dim null NF = j for every integer j with 0 dimV, a contradic- tion because a subspace of V cannot have dimension larger than dim V. ‘Thus the dimension increases by exactly one at each step. In other words, dimnullN3 = j for every integer j with 0 < j < dimV. In particular, taking j = dimV, we have dimnullN@™V = dimV. This means that null VéimV = V, Thus N4i™V — 0, and so N is nilpotent. Suppose T € L(V) and m is a nonnegative integer such that rangeT™ = rangeT™*). Prove that range T* = range T™ for all k > m.97 10. ll. CHAPTER 8. Operators on Complex Vector Spaces SOLUTION: Suppose u € rangeT™*!. Thus there exists a vector v in V (the domain of T) such that u = T"™1y, Now T™v is in rangeT™, which by our hypothesis equals rangeT™+!. Thus there exists w € V such that T™y =T™+ly. Putting all this together, we have wa=T™ly =T(I™v) =T(T™w) =T™ Hy, Thus u € range T™+?, Because u was an arbitrary vector in rangeT™ +1, we have shown that range T™+! C rangeT™+2, We also have an easy inclusion in the other direction, so we conclude that rangeT™+! = rangeT™+?, In the paragraph above, we showed that rangeT™ = rangeT™+ implies rangeT™+! = rangeT™+?, Apply that result, with m replaced with m +1, to conclude that rangeT™+? = range T™+3. Continuing in this fashion, we see that rangeT™ = rangeT™*+! = rangeT™+? = ..., as desired. Prove or give a counterexample: if T € C(V), then V =nullT @ rangeT. SoLuTION: Define T € £(F*) by T(w, z) = (z,0). Thus null T’ = rangeT = {(w,0) : w € F}, which clearly implies that F? is not the direct sum of null T’ and range. Of course there are also many other examples. Prove that if T € £(V), then V =nullT" @ rangeT®, where n = dimV. Souution: Let T € L(V). First we show that V =nullT* + rangeT™.98 12. 13. CHAPTER 8. Operators on Complex Vector Spaces To do this, let v€ V. Then v=(v-T"u) +T%u for any vector u € V. Obviously Tu € rangeT*. Thus looking at the equation above, we see that we need to show that there exists u € V such that v—7*u € nullT™. In other words, we want a vector u € V such that T?(v — Tu) = 0, which is equivalent to T'y = Tu. But T*v € rangeT”, and rangeT™ = rangeT?" (by 8.9), so T"v € rangeT?". Thus there indeed exists u € V such that T"v = T?"u, completing our proof v € null7”4rangeT”, Because v was an arbitrary vector in V, this implies that V = nullT™ + range 7. For any linear map (and in particular for T*), the dimension of the domain equals the sum of the dimensions of the null space and range (by 3.4). In other words, dim V = dim nullT” + dim range T*. This equation, along with the equation V = null 7” +rangeT™, implies that V = null” @ range T™ (by 2.19). Suppose V is a complex vector space, N € L(V), and 0 is the only eigenvalue of N. Prove that 1V is nilpotent. Give an example to show that this is not necessarily true on a real vector space. SOLUTION: Because 0 is the only eigenvalue of N, 8.23(a) implies that every vector in V is a generalized eigenvector of T corresponding to the eigenvalue 0. This implies that N is nilpotent. Define T € £(R®) by T(z,u,2) = (—y2,0). Then 0 is an eigenvalue of T because T(0, 0,1) = (0,0,0). As can be ver- ified from the definition of eigenvalue, T has no other eigenvalues (which must be in R, because Tis an operator on a real vector space). However, (x,y,z) = (y,—2,0). In particular, T? # 0. Thus T is not nilpotent. Of course there are also many other examples. Suppose that V is a complex vector space with dimV =n and T € L(V) is such that null T?~? # null T™“!. Prove that T has at most two distinct eigenvalues.99 14. 15. CHAPTER 8. Operators on Complex Vector Spaces SOLUTION: Because null 7"? # null7™1, we see that dimnullT! is at least 1 more than dimnullT~! for j = 1,...,»—1 (by 8.5). Thus dim null T"~! > n—1. In particular, 0 is an eigenvalue of T with multiplicity at least rn — 1. Because the sum of the multiplicities of all the eigenvalues of T equals n (by 8.18), this implies that T’ can have at most one additional eigenvalue. Give an example of an operator on C4 whose characteristic polynomial equals (z — 7)?(z — 8). Sotution: Define T € £(C4) by T (21, 22, 23, 24) = (7z1, 722, 823, 824). Then null(T — 77) is the two-dimensional subspace {(z1, 22,0,0) : 21,22 € C} and null(T — 82) is the two-dimensional subspace {(0,0, 2g, za) : 23,24 € C}. ‘Thus 7 is an eigenvalue of T with multiplicity at least 2 and 8 is an eigenvalue of T with multiplicity at least 2. Because 2+2 = 4 = dim C‘, there can be no other eigenvalues of T and the eigenvalues 7 and 8 must have multiplicity 2 (by 8.18). Thus the characteristic polynomial of T equals (z — 7)?(z — 8)?. Of course there are also many other examples. Suppose V is a complex vector space. Suppose T’ € £(V) is such that 5 and 6 are eigenvalues of T and that T has no other eigenvalues. Prove that (r-sn\(r—6nr 0, where n = dimV. Sotution: Because 5 and 6 are eigenvalues of T and T has no other eigenvalues, the characteristic polynomial of T must be of the form (2-5)" (2-6), where 1 < di and 1 < da. Because dj +d2 = n, we must also have dj 1, then the degree of p, would be less than the degree of p, contradicting the definition of minimal polynomial. Thus we can conclude that m = 1, as desired. CoMMENT: The proof given above works on both real and complex vector spaces. If V is a complex vector space, then this exercise can be done by using the complex spectral theorem (7.9) and Exercise 23 of this chapter. Suppose T € £(V) and v € V. Let p be the monic polynomial of smallest degree such that p(T)v = 0. Prove that p divides the minimal polynomial of T. SoLution: Let q denote the minimal polynomial of T. By the division algorithm (4.5), there exist polynomials s,r € P(F) such that q=sp+r and degr < deg p. Thus a(L)u = 9(T)p(T)w + r(T)v. Because q(T) = 0 and p(T)v = 0, the equation above shows that r(T)v = 0. This implies that r = 0 (otherwise we could multiply r by a scalar to get a monic polynomial with degree smaller than deg p that when applied to T gives an operator having v in its null space, which would contract the definition of p as the monic polynomial of smallest degree with this property). Using the information that r = 0, rewrite the formula above for q as a= sp. Thus p divides g, the minimal polynomial of T. Give an example of an operator on C4 whose characteristic and minimal polynomials both equal 2(z — 1)?(z — 3). SoLuTion: Define T € £(C‘) by T(wr, we, ws, wa) = (0, we + wy, ws, 304).107 27. CHAPTER 8. Operators on Complex Vector Spaces An easy computation shows that T(T — I)*(T — 32) = 0. Thus the minimal polynomial of T is a divisor of z(z — 1)*(z — 3) (by 8.34). Note that 0 is an eigenvalue of T because T(1,0,0,0) = (0,0,0,0) and 1 is an eigenvalue of T' because T(0, 1,0,0) = (0,1, 0,0) and 3 is an eigenvalue of T because T(0,0,0,1) = (0,0, 0,3). Thus 0, 1, and 3 must be roots of the minimal polynomial of T (by 8.36). ‘The only monic polynomials that divide 2(z—1)?(z—3) and have 0,1,3 as roots are 2(2—1)(z—8) and 2(z—1)*(z—3). Because T(T-1)(T—3) ¥ 0, as is easy to check, this implies that z(z—1)?(z— 3) is the minimal polynomial of T, as desired. Because T is an operator on a four-dimensional complex vector space and the minimal polynomial of T has degree 4, the characteristic polynomial of T (which is a monic polynomial of degree 4 that is divisible by the minimal polynomial of T) must equal the minimal polynomial of T. Of course there are also many other examples. Give an example of an operator on C* whose characteristic polynomial equals 2(z—1)?(z—3) and whose minimal polynomial equals 2(z—1)(z—3). SowuTion: Define T € £(C*) by T (wi, wa, ws, wa) = (0, we, ws, 34). An easy computation shows that T(T—I)(T—3I) = 0. Thus the minimal polynomial of T is a divisor of z(z — 1)(z— 3) (by 8.34). Note that 0 is an eigenvalue of T because T(1,0,0,0) = (0,0,0,0) and 1 is an eigenvalue of T because T(0, 1,0,0) = (0,1,0,0) and 3 is an eigenvalue of T because T(0, 0,0, 1) = (0,0,0,3). Thus 0, 1, and 3 must be roots of the minimal polynomial of T (by 8.36). ‘The only monic polynomial that is a divisor of 2(z — 1)(z — 3) and has 0,1,3 as roots is 2(z — 1)(z — 3). Thus z(z — 1)(z — 3) is the minimal polynomial of T,, as desired. Note that every vector in {(0, we, w3,0) : w2,w3 € C) is an eigenvector of T corresponding to the eigenvalue 1. Thus the eigenvalue 1 of T has multiplicity at least 2. The eigenvalues 0 and 3 of T have multiplicity at least 1. Because T is an operator on a four-dimensional complex vector space, the sum of the multiplicities of all the eigenvalues equals 4 (by 8.18). Thus the each use of the phrase “at least” in the previous paragraph can be replaced by “equal” because if any of the eigenvalues had larger multiplicity, the sum of the multiplicities of all the eigenvalues would exceed 4.108 CuapTer 8. Operators on Complex Vector Spaces Because 0 and 3 are eigenvalues of T’ with multiplicity 1 and 1 is an eigenvalue of T with multiplicity 2 and T has no other eigenvalues (the mul- tiplicities of the eigenvalues mentioned already sum to 4), the characteristic polynomial of T equals 2(z — 1)2(z — 3), as desired. Of course there are also many other examples. 28. Suppose ag,...,an-1 € C. Find the minimal and characteristic polynomials of the operator on C” whose matrix (with respect to the standard basis) is 0 ag 10 —a 1c. a 0 -an—2 1 ana Comment: This exercise shows that every monic polynomial is the characteristic polynomial of some operator. SoLuTion: Suppose that 7 € £(C") has matrix as above with respect to the standard basis (e1,...,€n) of C®. Thus Te; =e2 Te, = Ten = e3 Te, = Ten = en Tre, = Ten = —age; — ay€2 — +++ — Qn-1en- Thus (e1, Ter, Te1,.--,T* es) = (€1,€2;€3)---€n): In particular, (e1,Te1,Te1,...,T7"~*e1) is linearly independent. Thus if p is a monic polynomial with degree less than n, then p(T)e: # 0. Thus the minimal polynomial of T must have degree n. Writing Te, as a linear combination of (e:,Te1,T%e1,...,T"-1e1) is possible in only one way: Te, = —ae1 — a, Tey — +» ~Qn—1T”*e. Thus setting109 29. 30. CUuAPTER 8. Operators on Complex Vector Spaces P(z) = 09 + a1z +21 $F an-12"-) $2", we see that p is the only monic polynomial of degree n such that p(T)e, = 0. Hence p must equal the minimal polynomial of T. Because T is an operator on an n-dimensional complex vector space and the minimal polynomial of T has degree n, the characteristic polynomial of T (which is a monic polynomial of degree n that is divisible by the min- imal polynomial of T) must equal the minimal polynomial of T. Thus the characteristic polynomial of T’ also equals p. Suppose N € L(V) is nilpotent. Prove that the minimal polynomial of N is 2™+1, where m is the length of the longest consecutive string of 1’s that appears on the line directly above the diagonal in the matrix of N with respect to any Jordan basis for N. SOLUTION: Suppose (v),...,Un) is a Jordan basis for N and that m equals the length of the longest consecutive string of 1’s that appears on the line directly above the diagonal of M(N,(vi,...,tn)). The diagonal of M(N, (v1,-.-;0,)) contains only 0’s (by Exercise 6 of this chapter). Thus MU(N, (u1,-.5 } is a block diagonal matrix whose blocks have the form and the largest such block in an (m+ 1)-by-(m-+1) matrix. For each v;, we see that N™+ly; = 0. Because N™*+! equals 0 on a basis of V, we conclude that N™+! = 0, Thus the minimal polynomial of N must divide 2+ (by 8.34) and hence must be of the form z* for some k 1, then we could Ict U denote the span of the basis vectors corre- sponding to A; and let W denote the span of the basis vectors corresponding to Us,...,Um; we would have V = U@W, where U and W would be proper subspaces of V invariant under T. This contradiction shows that m = 1. Thus the matrix of T with respect to our Jordan basis is just Ay. In other words, there is a basis (1,...,0simv) of V such that the matrix of T— AI with respect to this basis is 01 0 1 0 0 ‘The previous problem now implies that the minimal polynomial of T— J equals z4imV This clearly implies that the minimal polynomial of T equals (z—-A)%™Y, as desired. To prove the implication in the other direction, suppose that the minimal polynomial of T equals (z — \)*". Suppose that there exist two proper subspaces Uj, Uz of V, each invariant under T, such that V =U; @ U2. Let p; denote the minimal polynomial of T|u, and pz denote the minimal polynomial of T|y,. If 1 € Ui, then (pim)(T)u = pa(T)pi (Ta =0. Similarly, if uz € Ua, thenlu 31. CHAPTER 8. Operators on Complex Vector Spaces (pipe)(T)ue = pi (T)pe2(T)u2 =0. Because every vector in V can be written in the form uj +ug, where w € U; and uz € Ue, the equations above imply that (pip2)(T) = 0. Thus the minimal polynomial of T, which equals (z — )*™V, is a divisor of pip. (by 8.34). The degree of the monic polynomial pp) equals the degree of p; plus the degree of pz, which is less than or equal to dim U; + dim U2, which equals dim V. Because (z — \)#™ is a divisor of pip2, this implies that (2-ajaav, pi(z)p2(z) This implies that Pilz) = (z= A)8™4% and pale) = (z— d)8MU, Let m = max{dim Uj, dim U2}. Let p(z) = (2 — d)™. The equations above show that p(T)|y, = 0 and p(T)|v, = 0. Thus p(T) = 0. Because the degree of pis less than dim V, this contradicts our hypothesis that (z—A)*™ is the minimal polynomial of T. This contradiction means that our assumption that there exists a direct sum decomposition of V into two proper subspaces invariant under T must have been false, as desired. Suppose T € £(V) and (v1,...,0q) is a basis of V that is a Jordan ba- sis for T. Describe the matrix of T’ with respect to the basis (Une ++ s U1) obtained by reversing the order of the v's. SoLuTion: The matrix of T with respect to (v1,...,0n) is a block di- agonal matrix A 0 0 Am where each A; is an upper-triangular matrix of the form 41 0 Aj =2 CHAPTER 8. Operators on Complex Vector Spaces Temporarily fix j, and let (u1,..., ug) be the part of (vj,..., Un) correspond- ing to the block A,. Thus Tur = Ayu Tug =u + Ajua Tus = tz + Ajus Tey = up + Aju. Write the equations above in the form Tug = jun + un—1 Tug—y = Ajue—1 + Up-2 Tug = Aju + uy Tuy = Aj. Thus the matrix of Tlepan(ur,..ux) With respect to (us,...,%2) is 4 0 1: 0 1» In other words, B; is obtained from A; by reflection across the diagonal, so in B; the 1’s lie below the diagonal instead of above it. Now we see that the matrix of T with respect to (vp, ---,01) is the block diagonal matrix Bm 0 0 ByCHAPTER 9 Operators on Real Vector Spaces Prove that 1 is an eigenvalue of every square matrix with the property that the sum of the entries in each row equals 1. SOLUTION: Suppose that A is an n-by-n matrix such that the sum of the entries in each row of A equals 1. Let x be the n-by-1 matrix all of whose entries equal 1. Then from the definition of matrix multiplication we see that the entry in row j, column 1 of Ax equals that sum of the entries in row j of A, which equals 1. Thus Az = x, which implies that 1 is an eigenvalue of A. Consider a 2-by-2 matrix of real numbers ac 4=[§ §]. Prove that A has an eigenvalue (in R) if and only if (a-d)? + 4be > 0. Soturion: A number A € R is an eigenvalue of A if and only if there exist numbers x,y € R, not both 0, such that [e slls]-a[3]. The left side of this equation equals [ 5 7 3 ], so the equation above is equivalent to the system of equations 1134 CHAPTER 9. Operators on Real Vector Spaces (a-A)x + cy=0 bx + (d-A)y = 0. It is easy to see that this system of cquations has a solution other than z= y=0 if and only if (a- A)(d—- A) = be, which is equivalent to the equation —(a+d)A+ (ad — be) = 0. There is a real number 2 satisfying the equation above if and only if (a +d)? — 4(ad — be) > 0. The left side of the inequality above equals (a — d)? + dbc, and thus we conclude that A has a real eigenvalue if and only if (a— d)? + 4be > 0. 3. Suppose A is a block diagonal matrix A, 0 A= oe ’ 0 An where each A; is a square matrix. Prove that the set of eigenvalues of A equals the union of the eigenvalues of A1,...,Am- SoLuTion: Suppose that A; has size nj-by-nj. Let a where each x; is an n,-by-1 matrix. Then Ain, Ar= i Amt ‘Thus the equation Ax = Az is equivalent to the system of equations115 CHAPTER 9. Operators on Real Vector Spaces Ayx, = Ax, Am@m = Atm. Suppose that is an eigenvalue of A. Then there is a nonzero vector x satisfying the system of equations above. Because x is nonzero, there exists k such that 2}, is nonzero. Because Ayz, = Azx, this implies that \ is an eigenvalue of Ay. Thus is in the union of the eigenvalues of Aj,..., A, as desired. To prove the implication in the other direction, suppose now that 2 is in the union of the eigenvalues of Aj,...,Am- Then there exists & such that A is an eigenvalue of Ay. Thus there exists a nonzero ng-by-1 vector x, such that Ayr, = Arg. For j # k, define 2; to be the nj-by-1 matrix all of whose entries equal 0, and define « to be the matrix determined by 21,...,2m as above. Then is nonzero (because zz is nonzero) and Az = Az, which shows that A is an eigenvalue of A, as desired. Suppose A is a block upper-triangular matrix A * A= “ 0 Am where each A; is a square matrix. Prove that the set of eigenvalues of A equals the union of the eigenvalues of A1,...,Am- Comment: Clearly Exercise 4 is a stronger statement than Exercise 3. Even so, students may want to do Exercise 3 first because it is easier than Exercise 4. SOLUTION: We will prove that 0 is an eigenvalue of A if and only if 0 is an eigenvalue of at least one of the A,’s. This is all we need to do, because for arbitrary A € F, we can replace A with A—XJ and each Ax with Ay —AJ, concluding that is an eigenvalue of A if and only if \ is an eigenvalue of at least one of the A,’s. This last statement implies that the set of eigenvalues of A equals the union of the eigenvalues of Ai,...,Am- Suppose that A has size n-by-n and each A; has size nj-by-nj. We can write a typical n-by-1 matrix x in the form116 Cuapter 9. Operators on Real Vector Spaces where each 2j is an nj-by-1 matrix. The product Az can be computed by multiplying together the block matrices of A and given above, using the same formula as one would use when multiplying matrices of numbers (this follows from the definition of matrix multiplication). First suppose that 0 is an eigenvalue of A. Thus there exists a nonzero n-by-I matrix 2 such that Ax = 0. Write z in the form above, and let k be the largest. index with nonzero 2; thus we can write (If k = m, then the 0’s shown above at the tail of x do not appear.) If we break Az into blocks of the same size as was done for z, then the k** block of Az will equal Axzg; this follows from the block upper-triangular form of A and the 0’s that appear in z after the k" block. But Az = 0, so the kt® block of Ax equals 0, so Az, = 0. Because x, # 0, this implies that 0 is an eigenvalue of Ag, as desired. To prove the implication in the other direction, suppose now that 0 is an eigenvalue of some A,. This means that the operator on Mat(n, 1,F) (the vector space of n-by-1 matrices) that sends z, € Mat(n,1,F) to Axx is not injective. Thus this operator is not surjective (by 3.21). Thus the operator on Mat(n; + ---+n,,1,F) that sends 2 Tk to AL * z 0 A | | 2x is not surjective (because the last block in the product above will be Apze, which cannot be an arbitrary ng-by-1 matrix). Again using 3.21, this means that the last operator is not injective. In other words, there exists a nonzero vector7 CHAPTER 9. Operators on Real Vector Spaces Ty € Mat(n +--+ +n, 1,F) Tk such that. AL *)fa 0 Ak Tk Adjoining an appropriate number of 0’s, this implies that A, + zy Ak Zk =0. Arti 0 0 Am 0 In other words, 0 is an eigenvalue of A, as desired. Suppose V is a real vector space and T' € L(V). Suppose a, € R are such that T? + aT + BI =0. Prove that T has an eigenvalue if and only if a? > 4B, SotuTion: First suppose that T has an eigenvalue \ ¢ R. Thus there exists a nonzero vector v € V such that T'v = Av. Applying T to both sides of the last equation, we get T?v = \?v. Thus 0=(1? +aT +Al)v =v +adv+ pu = (7 +ad+ A). Because v # 0, the last equation implies that YP +orA+ fH =0, which implies (recall that A, a, and are all real) that a? > 46, as desired. To prove the implication in the opposite direction, suppose now that a? > 46. Then the polynomial 2? +02 + B has two real roots, which means that we can writeCHAPTER 9. Operators on Real Vector Spaces 2? 40+ 6 =(z—11)(x—12) for some 71,72 € R. Thus O=T?+oT+ fl = (T-rI)(T - rel). In particular, (T —riI)(T—raZ) is not injective, which implies that at least one of T — J and T — oJ is not injective. In other words, at least one of r1,r2 must be an eigenvalue of T. Thus T has an eigenvalue, as desired. Suppose V is a real inner-product space and T € £(V). Prove that there is an orthonormal basis of V with respect to which T has a block upper- triangular matrix AL * 0 An where each A; is a 1-by-1 matrix or a 2-by-2 matrix with no eigenvalues. SOLUTION: We know that there is a basis (v1,-..,U,) of V with respect to which the matrix of T' has the block upper-triangular form above, where each A; is a 1-by-1 matrix or a 2-by-2 matrix with no eigenvalues (see 9.4). Apply the Gram-Schmidt procedure to (v1,...,0n), getting an orthonor- mal basis (€1,...,€n) of V such that span(w1,..-,0j) = span(e1,...,€3) for j = 1,...,n (see 6.20). This condition on the spans implies that the matrix of T with respect to (e),...,€n) is also a block upper-triangular of the form above, where each A; is a I-by-1 matrix or a 2-by-2 matrix (these A;'s may differ from the Aj’s corresponding to the matrix of T’ with respect to(v1,..-,0,)). All that remains is to show that, if necessary, we can modify our orthonormal basis so that none of the 2-by-2 blocks on the diagonal have eigenvalues. Suppose that A; is a 2-by-2 block on the diagonal of the matrix of T with respect to (€1,...,@,) and that A; has an eigenvalue A. Thus there exist z,y € R, not both 0, such that a[i]-[5]:ug CHAPTER 9. Operators on Real Vector Spaces Let ex,€.41 denote the basis vectors corresponding to Aj. Then, as is easy to verify, T (rex + yersi) = ut A(wex + erst) for some u € span(er,...,¢4-1). Let rex + Yes [zen + versal and choose fx41 € V such that (ft, fi+1) is orthonormal and span(frs fis) = span(ek, €x+1)- Then the matrix of T’ with respect to the orthonormal basis (ety. ++ sets Ses feta Cet21---1€n) will be a block upper-triangular matrix with the same entries on the diagonal as previously, except that A; will be replaced by the 2-by-2 matrix [0 :]. In other words, where we had A; on the diagonal, we can now think of having two 1-by-1 matrices on the diagonal (and we still have a block upper- triangular matrix because of the 0 in the lower-left entry of the matrix above). Repeating, when necessary, the procedure described above, we obtain an orthonormal basis of V with respect to which T has a block upper-triangular matrix A * 0 Am where each A; is a 1-by-1 matrix or a 2-by-2 matrix with no eigenvalues. Prove that if T € £(V) and j is a positive integer such that j < dim V, then T has an invariant subspace whose dimension equals j — 1 or j. SOLUTION: Suppose T' € L(V) and j is a positive integer such that i 2m, which clearly implies that m < (dimV)/2. Because S™ = 0, this implies that 54m V/2 — 9, as desired. Prove that if T € C(R%) and 5,7 are eigenvalues of T, then T’ has no eigenpairs. SOLUTION: Suppose that T € £(R¥) and 5,7 are eigenvalues of T. Of course each of these two eigenvalues must have multiplicity at least 1. By 9.17, the sum of the multiplicities of all the eigenvalues of T' plus twice the sum of the multiplicities of all the eigenpairs of T equals 3. Because the sum of the multiplicities of all the eigenvalues of T is at least 2, there is no room for an eigenpair, which would add at least 2 more to the sum (because we take twice the sum of the multiplicities of all the eigenpairs). Thus T’ has no cigenpairs, Suppose V is a real vector space with dim V = n and T € L(V) is such that null T*~? # null T™“}, Prove that T has at most two distinct. eigenvalues and that T has no eigen- pairs. SOLUTION: Because nullT*~? # nullT"-!, we see that dim null Td is at least 1 more than dimnullT#! for 7 = «2-1 (by 8.5). Thus14, CHAPTER 9._Operators on Real Vector Spaces dimnullT"-! > n—1. In particular, 0 is an eigenvalue of T with mul- tiplicity at least n — 1. Because the sum of the multiplicities of all the eigenvalues of T' plus the sum of twice the multiplicities of all the eigenpairs of T equals n (by 9.17), this implies that T’ can have no eigenpairs and at most one additional eigenvalue. Suppose V is a vector space with dimension 2 and T'€ C(V). Prove that if [é 4] is the matrix of T’ with respect to some basis of V, then the characteristic polynomial of T’ equals (z — a)(z — d) — be. COMMENT: As usual unless otherwise specified, here V may be a real or complex vector space. Soution: Let q(z) = (z — a)(z — d) — be. If (v1, v2) is the basis of V with respect to which T has the matrix above, then Tu, = av, +bv, and Tv = cv, + duo. From these equations you can easily verify that ¢(T)v, = 0 and q(T')v2 = 0. Because q(T) is 0 on a basis of V, we conclude that (T) = 0. Because q is a monic polynomial of degree 2 and q(T) = 0, we conclude that the minimal polynomial of T' has degree 1 or 2. Suppose first that the minimal polynomial of T’ has degree 2. Because the minimal polynomial of T is a divisor of q (by 8.34), and because a monic polynomial of degree 2 can be a divisor of another monic polynomial of degree 2 only if the two polynomials are equal, we conclude that q is the minimal polynomial of T’. The Cayley-Hamilton theorem now implies that q is a divisor of the characteristic polynomial of T’, which is also a monic polynomial of degree 2. This implies that q is the characteristic polynomial of T, as desired. Now consider the only remaining possible case, which is that the minimal polynomial of T has degree 1, meaning that it equals z — d for some \ € F. This implies that T = XZ, which implies that the characteristic polynomial of T equals (z—)?. Because T’ = MI, we must have a = d= Xandb =c=0. Thus ¢(z) = (z — A)?. In particular, g is the characteristic polynomial of T, as desired. COMMENT: Note that we did not need to find the eigenvalues of T to do this exercise.123 15. CHAPTER 9. Operators on Real Vector Spaces Suppose V is a real inner-product space and $ € £(V) is an isometry. Prove that if (a, 8) is an eigenpair of S, then = 1. SoLution: There is a basis of V with respect to which S has a block diagonal matrix, where each block on the diagonal is a 1-by-1 matrix or a 2-by-2 matrix of the form cos@ —sin@ sin@ cos@ |’ with @ € (0, 1) (see 7.38). The characteristic polynomial of the matrix above is (x — cos6)? + sin? 6, which equals x? — 2(cos@)x + 1. If (a, f) is an cigenpair of S, then dim null(S? + a8 + BI)%™Y > 0, which implies that x? + ax + is the characteristic polynomial of a 2-by-2 matrix of the form displayed above (see 9.9). Thus = 1.CHAPTER 10 Trace and Determinant Suppose that T € L(V) and (vj,...,n) is a basis of V. Prove that M(T,(u1,.-.,0q)) is invertible if and only if T is invertible. So.uTion: First suppose that M(T) is an invertible matrix (because the only basis is sight is (v1,...,Un), we can leave the basis out of the notation). ‘Thus there exists an n-by-n matrix B such that M(T)B = BM(T) = I. There exists an operator S € £(V) such that M(S) = B (sec 3.19). Thus the equation above becomes M(T)M(S) = M(S)M(T) = which we can rewrite as M(TS) =-M(ST) = M(1), which implies that TS = ST =1. Thus T is invertible, as desired, with inverse 3. To prove the implication in the other direction, suppose now that T is invertible. Thus there exists S € £(V) such that TS=ST=I. This implies that 124CHAPTER 10. Trace and Determinant M(TS) = M(ST) = M(1), which implies that M(T)M(S) = M(S)M(T) = I. Thus M(T) is invertible, as desired, with inverse M(S). Prove that if A and B are square matrices of the same size and AB = I, then BA =I. So.ution: Suppose that A and B are n-by-n matrices and AB = I. There exist 5,1’ € £(F") such that M(S)=A and M(T) =B; here we are using the standard basis of F* (the existence of $,7 € £(F") satisfying the equations above follows from 3.19). Because AB = I, we have M(S)M(Z) = I, which implies that M(ST) = M(I), which implies that ST =I, which implies that T'S = I (by Exercise 23 in Chapter 3). Thus BA=M(T)M(S) = M(TS) =M(1) al. Suppose T' € £(V) has the same matrix with respect to every basis of V. Prove that Tis a scalar multiple of the identity operator. SoLUTION: We begin by proving that (v, Tv) is linearly dependent for every v € V. To do this, fix v € V, and suppose that (v,T'v) is linearly independent. Then (v,Tv) can be extended to a basis (u,v, u1,-.+;tn) of V. The fitst column of the matrix of T' with respect to this basis is 0 1 0 0 Clearly (2v,T'v,1,...,tn) is also a basis of V. The first column of the matrix of T’ with respect to this basis is126 CHAPTER 10. Trace and Determinant on 0 Thus T has different matrices with respect tothe two bases we have consid- ered. This contradiction shows that (v, Tv) is linearly dependent for every v €V. This implies that for every vector in V is an eigenvector of T. This implies that T' is a scalar multiple of the identity operator (by Exercise 12 in Chapter 5). Suppose that (u,...,tn) and (v1,...,n) are bases of V. Let T € L(V) be the operator such that Tux = uy for k = 1,...,n. Prove that M(T, (u1y---49m)) = M (I, (ur, --+5%tn)s (Uy-+ Un). Souution: Fix k. Write Up = QV, ++++ +GnUn, where a1,...,@n € F. Because Tr, = uz, the k* column of the matrix M(T,(v1,...;0,)) consists of the numbers a},...,dn. Because Iuz = up, the kt column of M(I,(u1,---stn)(v1---1n)) also consists of the num- bers a1,...,@n- Because M(T,(v1,.-.,0m)) and M(T, (u1,---;tn); (01) ...5%n)) have the same columns, these two matrices must be equal. Prove that if B is a square matrix with complex entries, then there exists an invertible square matrix A with complex entries such that A~!BA is an upper-triangular matrix. SOLUTION: Suppose B is an n-by-n matrix with complex entries. Let (e1,--.,€n) denote the standard basis of O". There exists T € £(C*) such that M(T,(e1,..-,én)) = B (see 3.19). There is a basis (v1,..., Un) of V such that M(T, (v1,...,0n)) is an upper- triangular matrix (see 5.13). Let A = M((u1,...,0n);(e1,---,€n))- Then Ais invertible (by 10.2) and A“BA=A™M(T,(e1,.--5€n))A = M(T, (Ui,.--,0n))sCHAPTER 10. Trace and Determinant where the second equality comes from 10.3. Thus A~'BA is an upper triangular matrix. Give an example of a real vector space V and T € £(V) such that trace(T?) < 0. Souution: Define T € £(R?) by T(z,u) = (-v,2). ‘Then T? = —I, so traco(T?) = -2. Suppose V is a real vector space, T € £(V), and V has a basis consisting of eigenvectors of T’. Prove that trace(T?) > 0. SOLUTION: Let (v1,...,n) be a basis of V consisting of eigenvectors of T. Thus there exist A1,...,An € R such that Toy = jv; for each j. Clearly the matrix of T? with respect to the basis (u1,...,n) is the diagonal matrix a? 0 0 an? Thus traceT? = 4y? +++++An? > 0. Suppose V is an inner-product space and v,w € L(V). Define T € £(V) by Tu = (u,v)w. Find a formula for trace T. SOLUTION: First suppose that v # 0. Extend (re) to an orthonormal basis (j4y,¢1,---€n) of V. Note that for each j, we have T’e; = 0 (because (ej,v) = 0). The trace of T’ equals the sum of the diagonal entries in the matrix of T with respect to the basis (747,€1,---€n). Thus v *Ioll = Car vw, br trace = (CTT) Tag) + (Pesvea) +--+ (Ten, ea) = (w,v). If v = 0, then T = 0 and so traceT = 0 = (w,v). Thus we have the formula128 10. Il. CHAPTER 10. Trace and Determinant traceT = (w,v) regardless of whether or not v = 0. Prove that if P € L(V) satisfies P? = P, then traceP is a nonnegative integer. SoLution: Suppose that T’ € L(V) satisfies P? = P. Let (u1,...,tm) be a basis of range P and let (v1,...,Un) be a basis of null P. Then (tay +++ ths ty Un) is a basis of V (this holds because V = range’ @ nullT’; see Exercise 21 in Chapter 5). For each uj we have Pus = uj and for each v we have Pux = 0. Thus the matrix of P with respect to the basis above of V is a diagonal matrix whose diagonal contains m 1's followed by n 0's. Thus trace P = m, which is a nonnegative integer, as desired. In fact, we have shown that trace P = dim range P. Prove that if V is an inner-product space and T € C(V), then trace T* = traceT. SoLuTion: Suppose that V is an inner-product space and T € L(V). Let (€1,...,€n) be an orthonormal basis of V. The trace of any operator on V equals the sum of the diagonal entries on the matrix of the operator with respect to this basis. Thus trace T* = (T*e1,¢1) +--+ + (Ten; €n) = (e1, Te) + +++ + (en; Ten) (Tey, €1) + +++ + (Tens en) Te1,€1) + +--+ (Ten, en) = traceT. H Suppose V is an inner-product space. Prove that if T’€ L(V) is a positive operator and trace T’ = 0, then T = 0. So.ution: Suppose T € L(V) is a positive operator and traceT = 0. There exists an operator S € L(V) such that T = S*S (by 7.27). Let (e1,--.,¢n) be an orthonormal basis of V. Then129 12. 13. CHAPTER 10. Trace and Determinant 0=traceT = (Te, €1) +--+ + (Ten, en) = (S*Se1,e1) +--+ + (S*Sen,en) = [[Ser|? + +--+ [|Senl]?. The equation above implies that Se; = 0 for each j. Because S is 0 on a basis of V, we have S = 0. Because T = S*S, this implies that T = 0, Suppose T € £(C%) is the operator whose matrix is 51-12 -21 60 -40 -28]. 57 -68 1 Someone tells you (accurately) that —48 and 24 are eigenvalues of T. With- out using a computer or writing anything down, find the third eigenvalue of T. Sovution: The sum of the eigenvalues of T equals the sum of the diagonal terms of the matrix above (both quantities equal traceT). The sum of the diagonal terms of the matrix above equals 12. The sum of two of the eigenvalues of T’, —48 and 24, equals —24. Because the sum of all three eigenvalues of T' must equal 12, the third eigenvalue of T must be 36. Prove or give a counterexample: if T € £(V) and c € F, then trace(cT) = ctraceT. SOLUTION: Suppose J’ € L(V) and c € F. To prove that trace(eT) = etraceT, consider a basis of V. Then trace T’ equals the sum of the diagonal terms of the matrix of T with respect to this basis. The matrix of cT, with respect to the same basis, equals c times the matrix of T. Thus the sum of the diagonal terms of the matrix of cI equals c times the sum of the diagonal terms of the matrix of T. In other words, trace(cT) = ctraceT. Prove or give a counterexample: if S,T € £(V), then trace(ST) = (trace S)(traceT). Sowution: Define $,T € £(F?) by S(x,y) = T(z,y) = (-y,2). Then with respect to the standard bases the matrix of S (which of course equals the matrix of T) is130 15. 16. CuapTer 10. Trace and Determinant 0-1 1 04}° Thus trace S = traceT = 0. However, ST = —I, so trace ST = —2. Thus for this choice of S and T, we have trace(ST) # (trace S)(traceT). Of course there are also many other examples. Suppose T € L(V). Prove that if trace(ST) = 0 for all S € L(V), then T=0. SOLUTION: Suppose that trace(ST) = 0 for all S € £(V). Then trace(T'S) = 0 for all S € L(V) (by 10.12). Suppose that there exists v € V such that Tv # 0. Then (Tv) can be extended to a basis (Tv, u1,..., Un) of V. Define S € L(V) by S(aTv + buy +--+ + bntin) = av. Thus (Tv) = v and Su; = 0 for each j. Hence (TS)(Tv) = T(S(Tv)) =Tv and (TS)(uj) = 0 for each j. This implies that with respect to the basis (Tv, u1,...,tn), the matrix of TS consists of all 0’s except for a 1 in the upper-left corner. Thus trace(TS) = 1. This contradiction shows that our assumption that T'v # 0 must have been false. Thus Tu = 0 for every v € V, which means that T = 0. Suppose V is an inner-product space and T' € L(V). Prove that if (e,-..,€n) is an orthonormal basis of V, then trace(I*T) = {|Tei ||? +--- + |lTenll?. Conclude that the right side of the equation above is independent of which orthonormal basis (e1,...,€n) is chosen for V. SoLuTION: Suppose that (e1,...,€n) is an orthonormal basis of V. Then traceT*T = (T*Te,,e1) +--+ + (T*T en, en) = (Te,, Tey) +--+ + (Ten, Ten) = [[Tel? +--+ [IPen|?. Because traceT*T does not depend upon the choice of a basis of V, the formula above shows that ||T; ||? + --- + ||Ten||? is independent of the orthonormal basis (e1,...,€n)-131 17. CHAPTER 10. Trace and Determinant. Suppose V is a complex inner-product space and T € £(V). Let At,...,An be the eigenvalues of T, repeated according to multiplicity. Suppose Qi ee Qin Ont ve Onn is the matrix of T with respect to some orthonormal basis of V. Prove that IMP +e-+ PnP SOD Nagel. kel j=l SoLuTIon: Suppose that (e1,...,en) is the orthonormal basis with re- spect to which T has the matrix above. Thus for each k, we have Teg = 1,461 + +++ + Onpens which implies that Wen? = larg? +--+ + lanl. Thus Perl? +--+ (Peal? => >> laze? kei By the previous exercise, the left side of this equation equals trace(T*T). This reduces the exercise at hand to proving that [Ai ?? + +++ + [Aal? < trace(T*T). There is an orthonormal basis (f;,..., fn) with respect to which T has an upper-triangular matrix (by 6.28). The diagonal entries of the matrix of T with respect to (fi,-.-, fn) are precisely Ai,...,An (by 8.10), where we can relabel the eigenvalues of T’ so that they appear in the order A1,..., An along the diagonal. In other words, ST * M(P, (fiy-+-1 fn) = From the matrix above, we see that132 18. CHAPTER 10. Trace and Determinant Akl? < Tell? for each k. Thus [Aa]? +++ nl? SIDA? +--+ Fall? = trace(T"T), as desired (here the last equality comes from the previous exercise). Suppose V is an inner-product space. Prove that (S,T) = trace(ST*) defines an inner product on £(V). SoLuTION: Suppose that (-,-) is defined as above and R,S,T' € L(V). Then (T,T) = trace(TT*), and thus (T,T) > 0 (by the formula given in Exercise 16 of this chapter, with T replaced with T*). Because TT” is a positive operator (see 7.27), we also see that (T,T) = 0 if and only if T =0 (by Exercise 11 of this chapter). Now (R+ 8,7) = trace((R + $)T*) = trace(RT* + ST") = trace(RT*) + trace(ST*) = (R,T) + (5,7), where the third equality comes from 10.12. For c € F, we have (cS,T) = trace(cST") = ctrace(ST*) =c(S,T), where the second equality comes from Exercise 13 of this chapter. Finally, (S,T) = trace(ST*) trace((ST*)*) trace(TS*) GT),133 CHAPTER 10, Trace and Determinant where the second equality comes from Exercise 10 of this chapter. We have shown that (-,-) satisfies all the properties required.of an inner product. COMMENT: Suppose (e1,...,¢n) is an orthonormal basis of V and A On Gna ++. Onn is the matrix of T with respect to this basis. Then (T,T) = trace(TT") = trace(T"T) = LV basal, katja where the second equality comes from 10.12 and the third equality comes ‘from Exercise 16 of this chapter. Thus the | norm on £(V) induced by (:,+) is the same as the standard norm on F"* (here we are identifying each operator with its matrix, which has n? entries). Because norms determine the inner product (see Exercises 6 and 7 in Chapter 6), this means that the inner product (-,-) is the same as the standard inner product of F™ (again using the identification via matrices with respect to the orthonormal basis (e1,---s€n))- 19. Suppose V is an inner-product space and T’ € L(V). Prove that if (T*eI] < [ITI for every u € V, then T is normal. Sotution: Suppose that ||T*v|| < ||Tul] for every v € V. Suppose u € V with |lul| = 1. Extend (u) to an orthonormal basis (u,€1,...,€n) of V. Then trace(TT*) = ||T*ul|? + [[T°ex[? +--+ + [IT*en||? S$ l[Tull? + Teil? +--+ + [IZenll? trace(T*T) = trace(TT"),134 20. 21. CHAPTER 10. Trace and Determinant where the first and third lines come from Exercise 16 of this chapter and the last line comes from 10.12, Because the first and last lines above are equal, we must have equality throughout. Thus ||T*ul] = ||Tull. This clearly implies that IT" (au)|| = [I7(@u) for every a € F, Because every vector in V can be written in the form au for some a € F and some u € V with |jul| = 1, this implies that |[T*vl] = ||Tu|] for every v € V. This implies that T is normal (by 7.6). COMMENT: This exercise fails on infinite-dimensional inner-product spaces, leading to what are called hyponormal operators, which have a well- developed theory. Prove or give a counterexample: if T € £(V) and c € F, then det(cT) = climY det T. Sotution: Let n= dim V. If A is an n-by-n matrix, then det(cA) = c" det A for every c € F (this follows immediately from the definition 10.25). Now suppose that T € L(V). Because det T equals the determinant of the matrix of T with respect to any basis (see 10.33), the equation above implies that for every c € F we have det(cT) = det M(cT) = det(cM(T)) = c"det M(T) = c" det T, as desired. Prove or give a counterexample: if S,T € £(V), then det(S + T) = det S +detT. Soution: Define $,T € £(F?) by S(z,u) Then S$ and T are both not invertible. Thus det S = detT = 0 (by 10.14). However, $+ T =I and det J = 1, so det(S + T) # det S + detT. (2,0) and T(z,y) = (0,y).135 22. CHAPTER 10. Trace and Determinant Of course there are also many other examples. Suppose A is a block upper-triangular matrix Ai * A= oF , 0 Am where each A; along the diagonal is a square matrix. Prove that det A = (det Ai)... (det Am). SoLuTion: First consider the case where m = 2, so we can write A in the form Be a=[0 6]: where B is an n-by-n matrix bia bin B : i bn Pan and C is a p-by-p matrix CLL ++) Clp c i i Cpl +++ Gp Let a; denote the entry in row j, column k of A. Note that aj, =Oifj>n and k < n (this follows from the block upper-triangular form of A). Because Ais an (n + p)-by-(n + p) matrix, to compute det A we need to consider a typical permutation (m1,...,Mnyp) € perm(n + p). If any of m),..., Tn is greater than n, then Om; ,L+++Omarp.ntp = O- Thus in computing det A we need only consider permutations in which the first n coordinates all come from {1,...,n}, which means that the last p coordinates all come from {n + 1, n+p}. We can break any such per- mutation (mj,...,™a4p) into two pieces: a permutation of {1,...,n} and (relabeling n+ 1,...,n +p as 1,...,p to correspond to the labeling of the136 CHAPTER 10. Trace and Determinant entries of C) a permutation of {1,...,p}. Clearly the sign of (m1,..., n+p) will equal the product of the signs of these two permutations. Putting all this together, we have det A= > (sign(m1,..-,7n+p))@m,1 +++ @marpin-tp (my,...Tin¢p )Eperm(n+p) [(signGi,-.-,in)) (sign (a, ---s bp) + Bju,1 Bin nCky.d ++ Ckpe] = SS (Sign (yt,---15n)) bina + Bian” Givin) €perm a SS Gign(e,--- sp) ce. += eke ‘p)€perm p = (det B)(det C), completing the proof when m = 2. Suppose now that m > 2 and A . A= : 0 An Writing AL * B= . , 0 Am-1 we have B «x a-[0 an] Thus det A = (det B) (det Am) = (det Aj)... (det Am—1)(det Am),137 23. 24. CuapTer 10. Trace and Determinant where the first equality holds by the m = 2 case proved above and the second equality comes from induction on m (meaning that we can assume the desired result holds when m is replaced with m — 1). Suppose A is an n-by-n matrix with real entries. Let § € £(C") denote the operator on CG" whose matrix equals A, and let T € £(R") denote the operator on R" whose matrix equals A. Prove that trace S = traceT and det 5 = det T. Sotution: The formulas defining the trace and determinant of a matrix do not depend upon whether we think of the matrix entries as real or complex numbers. We have trace A = trace and traceA = traceT’ (by 10.11). Thus trace S' = trace T. Similarly, we have det A = det $ and det A = detT (by 10.33). Thus det § = det T. Suppose V is an inner-product space and T' € £(V). Prove that det T* = det T. Use this to prove that |det T| = det VT°T, giving a different proof than was given in 10.37. Sovurion: Let n=dimV. If \¢ F, then ((T—AZ)")* = (T*— 0)", which implies that dim null(T — AJ)" = dim null(T* — 42)", where we have used Exercise 31 in Chapter 7. The equation above shows that the eigenvalues of T* are precisely the complex conjugates of the eigenvalues of T, with the same multiplicities. If F = C, the determinant equals the product of the eigenvalues, counting multiplicity, so we have det T* = det’ (so far just on complex vector spaces). Now suppose F = R, so we must consitier eigenpairs. If a, 8 € R, then dim null(T? + oT + 61)" = dimnull((T*)? + aT" + BI)", again by Exercise 31 in Chapter 7. Thus T and T* have the same eigenpairs with the same multiplicities. From the paragraph above, T and T* have the same eigenvalues with the same multiplicities. Because the determinant equals the product of the eigenvalues (counting multiplicity) times the prod- uct of the second coordinates of the eigenpairs (counting multiplicity), this implies that det T* = det T = detT.138 CHAPTER 10. Trace and Determinant At this point, we know that det T* = det T regardless of whether F = C or F=R. Thus (det VI"T)? = (det VI*T) (det VT*T) = det(T*T) = (det T*)(detT) = (det T)(det T) = |det T/. ‘Taking square roots (and recalling that the positive operator VI*T has a nonnegative determinant), we have det /I'*T = |det T|, as desired. 25. Let a,b, be positive numbers. Find the volume of the ellipsoid oy {@wz)eR?: St eta

You might also like

• Her Body and Other Parties: Stories

  

  From Everand

  Her Body and Other Parties: Stories

  Carmen Maria Machado

  Rating: 4 out of 5 stars

  4/5 (822)
• A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  

  From Everand

  A Heartbreaking Work Of Staggering Genius: A Memoir Based on a True Story

  Dave Eggers

  Rating: 3.5 out of 5 stars

  3.5/5 (231)
• Fear: Trump in the White House

  

  From Everand

  Fear: Trump in the White House

  Bob Woodward

  Rating: 3.5 out of 5 stars

  3.5/5 (738)
• • Magazines
  • Podcasts
  • Sheet music
• Rise of ISIS: A Threat We Can't Ignore

  

  From Everand

  Rise of ISIS: A Threat We Can't Ignore

  Jay Sekulow

  Rating: 3.5 out of 5 stars

  3.5/5 (137)
• Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  

  From Everand

  Devil in the Grove: Thurgood Marshall, the Groveland Boys, and the Dawn of a New America

  Gilbert King

  Rating: 4.5 out of 5 stars

  4.5/5 (266)
• On Fire: The (Burning) Case for a Green New Deal

  

  From Everand

  On Fire: The (Burning) Case for a Green New Deal

  Naomi Klein

  Rating: 4 out of 5 stars

  4/5 (74)
• Team of Rivals: The Political Genius of Abraham Lincoln

  

  From Everand

  Team of Rivals: The Political Genius of Abraham Lincoln

  Doris Kearns Goodwin

  Rating: 4.5 out of 5 stars

  4.5/5 (234)
• The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  

  From Everand

  The Sympathizer: A Novel (Pulitzer Prize for Fiction)

  Viet Thanh Nguyen

  Rating: 4.5 out of 5 stars

  4.5/5 (122)
• Brooklyn: A Novel

  

  From Everand

  Brooklyn: A Novel

  Colm Tóibín

  Rating: 3.5 out of 5 stars

  3.5/5 (1947)
• The Yellow House: A Memoir (2019 National Book Award Winner)

  

  From Everand

  The Yellow House: A Memoir (2019 National Book Award Winner)

  Sarah M. Broom

  Rating: 4 out of 5 stars

  4/5 (98)
• Sing, Unburied, Sing: A Novel

  

  From Everand

  Sing, Unburied, Sing: A Novel

  Jesmyn Ward

  Rating: 4 out of 5 stars

  4/5 (1104)
• A Man Called Ove: A Novel

  

  From Everand

  A Man Called Ove: A Novel

  Fredrik Backman

  Rating: 4.5 out of 5 stars

  4.5/5 (4770)
• Yes Please

  

  From Everand

  Yes Please

  Amy Poehler

  Rating: 4 out of 5 stars

  4/5 (1898)
• Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  

  From Everand

  Hidden Figures: The American Dream and the Untold Story of the Black Women Mathematicians Who Helped Win the Space Race

  Margot Lee Shetterly

  Rating: 4 out of 5 stars

  4/5 (897)
• A Tree Grows in Brooklyn

  

  From Everand

  A Tree Grows in Brooklyn

  Betty Smith

  Rating: 4.5 out of 5 stars

  4.5/5 (1929)
• The World Is Flat 3.0: A Brief History of the Twenty-first Century

  

  From Everand

  The World Is Flat 3.0: A Brief History of the Twenty-first Century

  Thomas L. Friedman

  Rating: 3.5 out of 5 stars

  3.5/5 (2259)
• The Perks of Being a Wallflower

  

  From Everand

  The Perks of Being a Wallflower

  Stephen Chbosky

  Rating: 4.5 out of 5 stars

  4.5/5 (2104)
• The Little Book of Hygge: Danish Secrets to Happy Living

  

  From Everand

  The Little Book of Hygge: Danish Secrets to Happy Living

  Meik Wiking

  Rating: 3.5 out of 5 stars

  3.5/5 (401)
• The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  

  From Everand

  The Hard Thing About Hard Things: Building a Business When There Are No Easy Answers

  Ben Horowitz

  Rating: 4.5 out of 5 stars

  4.5/5 (348)
• The Light Between Oceans: A Novel

  

  From Everand

  The Light Between Oceans: A Novel

  M.L. Stedman

  Rating: 4.5 out of 5 stars

  4.5/5 (789)
• Shoe Dog: A Memoir by the Creator of Nike

  

  From Everand

  Shoe Dog: A Memoir by the Creator of Nike

  Phil Knight

  Rating: 4.5 out of 5 stars

  4.5/5 (540)
• The Unwinding: An Inner History of the New America

  

  From Everand

  The Unwinding: An Inner History of the New America

  George Packer

  Rating: 4 out of 5 stars

  4/5 (45)
• John Adams

  

  From Everand

  John Adams

  David McCullough

  Rating: 4.5 out of 5 stars

  4.5/5 (2409)
• The Emperor of All Maladies: A Biography of Cancer

  

  From Everand

  The Emperor of All Maladies: A Biography of Cancer

  Siddhartha Mukherjee

  Rating: 4.5 out of 5 stars

  4.5/5 (271)
• Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  

  From Everand

  Elon Musk: Tesla, SpaceX, and the Quest for a Fantastic Future

  Ashlee Vance

  Rating: 4.5 out of 5 stars

  4.5/5 (474)
• Little Women

  

  From Everand

  Little Women

  Louisa May Alcott

  Rating: 4 out of 5 stars

  4/5 (104)
• Bad Feminist: Essays

  

  From Everand

  Bad Feminist: Essays

  Roxane Gay

  Rating: 4 out of 5 stars

  4/5 (1018)
• The Outsider: A Novel

  

  From Everand

  The Outsider: A Novel

  Stephen King

  Rating: 4 out of 5 stars

  4/5 (1850)
• The Woman in Cabin 10

  

  From Everand

  The Woman in Cabin 10

  Ruth Ware

  Rating: 3.5 out of 5 stars

  3.5/5 (2522)
• Wolf Hall: A Novel

  

  From Everand

  Wolf Hall: A Novel

  Hilary Mantel

  Rating: 4 out of 5 stars

  4/5 (3811)
• Angela's Ashes: A Memoir

  

  From Everand

  Angela's Ashes: A Memoir

  Frank McCourt

  Rating: 4.5 out of 5 stars

  4.5/5 (441)
• The Constant Gardener: A Novel

  

  From Everand

  The Constant Gardener: A Novel

  John le Carré

  Rating: 3.5 out of 5 stars

  3.5/5 (104)
• Grit: The Power of Passion and Perseverance

  

  From Everand

  Grit: The Power of Passion and Perseverance

  Angela Duckworth

  Rating: 4 out of 5 stars

  4/5 (590)
• Manhattan Beach: A Novel

  

  From Everand

  Manhattan Beach: A Novel

  Jennifer Egan

  Rating: 3.5 out of 5 stars

  3.5/5 (792)
• The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  

  From Everand

  The Subtle Art of Not Giving a F*ck: A Counterintuitive Approach to Living a Good Life

  Mark Manson

  Rating: 4 out of 5 stars

  4/5 (5814)
• Principles: Life and Work

  

  From Everand

  Principles: Life and Work

  Ray Dalio

  Rating: 4 out of 5 stars

  4/5 (609)
• The Art of Racing in the Rain: A Novel

  

  From Everand

  The Art of Racing in the Rain: A Novel

  Garth Stein

  Rating: 4 out of 5 stars

  4/5 (4203)
• The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  

  From Everand

  The Gifts of Imperfection: Let Go of Who You Think You're Supposed to Be and Embrace Who You Are

  Brene Brown

  Rating: 4 out of 5 stars

  4/5 (1092)
• Never Split the Difference: Negotiating As If Your Life Depended On It

  

  From Everand

  Never Split the Difference: Negotiating As If Your Life Depended On It

  Chris Voss

  Rating: 4.5 out of 5 stars

  4.5/5 (844)
• The Glass Castle: A Memoir

  

  From Everand

  The Glass Castle: A Memoir

  Jeannette Walls

  Rating: 4.5 out of 5 stars

  4.5/5 (1717)
• Steve Jobs

  

  From Everand

  Steve Jobs

  Walter Isaacson

  Rating: 4.5 out of 5 stars

  4.5/5 (807)