A classic problem from the most important math class.

แชร์
ฝัง
  • เผยแพร่เมื่อ 1 ธ.ค. 2024

ความคิดเห็น • 112

  • @agbenfante
    @agbenfante 2 ปีที่แล้ว +61

    “Or as I like to call it, Linear Algebra” 😂😂😂

    • @rheticus5198
      @rheticus5198 2 ปีที่แล้ว +8

      And "Advanced" Calculus, or as MP likes to call it, Calculus.

    • @SAHarry
      @SAHarry 2 ปีที่แล้ว +8

      You already know his exams are fucking hard just by this statement

  • @wagsman9999
    @wagsman9999 2 ปีที่แล้ว +76

    I heard a math professor once say: "You can never get enough linear algebra."

  • @MihaiNicaMath
    @MihaiNicaMath 2 ปีที่แล้ว +5

    This problem is also a great example of the power of using invariant subspaces. Here is an alternate proof that uses invariant subspaces:
    Let E be the eigenspace of A for some eignevalue lambda. You showed that v \in E => Bv \in E, i.e. B is E invariant. But then B can be thought of as an operator acting on the vector space E. (i.e. B restricted to E is a totally valid operator) Since we are working over the complex numbers, this operator must have eigenvectors which are in E. But everything in E is eigenvectors for A, they must be eignevectors for both! (The polynomial approach used in the video is essentially the proof that every linear operator has eigenvectors) I believe with a bit more thought, this could tell you somethign about how the dimensions of the eigenspaces of A and B must be compatible. (For example if E is k dimensional, then the eigenvectors of B restricted to E must have algebraic multiplicy that sum to k).

  • @abrahammekonnen
    @abrahammekonnen 2 ปีที่แล้ว +8

    A good linear algebra video is always appreciated. Thank you and I look forward to your next video.

  • @chuckaway6580
    @chuckaway6580 2 ปีที่แล้ว +7

    A lot of people in the comments seem to be confusing this result with the (somewhat famous in quantum mechanics) result that *hermitian* linear transformations are simultaneously *unitarily* diagonalizable iff they commute. By contrast, this video gives a different result which says *any* pair of commuting linear transformations shares at least 1 eigenvector (but, importantly, may still fail to be simultaneously diagonlizable. Indeed, neither of the matrices need be diagonlizable in the first place). No assumption on the linear transformations is required other than that the field is algebraically closed and the vector space is finite dimensional.

    • @APaleDot
      @APaleDot 2 ปีที่แล้ว

      But by the stronger result, any two matrices that commute _never_ fail to be simultaneously diagonalizable, right? Am I missing something?

    • @chuckaway6580
      @chuckaway6580 2 ปีที่แล้ว

      @@APaleDot What if the matrices aren't diagonalizable in the first place?

    • @APaleDot
      @APaleDot 2 ปีที่แล้ว

      @@chuckaway6580
      Well if "two matrices are simultaneously diagonalizable if and _only_ if they commute" is true, then two matrices which commute _must_ be simultaneously diagonalizable, and therefore they must also both be individually diagonalizable. If they are not individually diagonalizable, then they won't ever commute, apparently.

    • @chuckaway6580
      @chuckaway6580 2 ปีที่แล้ว +2

      @@APaleDot yes, but the statement "two matrices are simultaneously diagonalizable iff they commute" isn't true. What is true, as michael shows, is that they share at least one eigenvector. But we can't say more than that without some additional hypotheses on the matrices (such as being hermitian).

  • @Vladimir_Pavlov
    @Vladimir_Pavlov 2 ปีที่แล้ว +16

    The topic has applications in quantum mechanics.
    For example, that there are identical (up to a factor depending on other variables) eigenfunctions of the operators of the square of the orbital moment and its projection on any axis. Which means the existence of a quantum state in which these quantities can have simultaneously certain values determined by the orbital and magnetic quantum numbers.

    • @depressedguy9467
      @depressedguy9467 2 ปีที่แล้ว

      Yes my physics mate told this application and I got this new theorem in the list

  • @andreaswinter
    @andreaswinter 2 ปีที่แล้ว +15

    I love your videos of olympiad and similar problems! Great work, all the way to explaining basic mathematical facts. Here I think you made yourself more work than necessary: the vectors p(B)v, for arbitrary polynomials p, form a linear subspace V of C^n, which is spanned by v, Bv, B^2v, etc, and thus is non-empty. At the same time it consists entirely of lambda-eigenvectors of A, and is stable under B, i.e. B maps V to V. The restriction of B to this subspace, say of dimension k=dim V is equivalent to a kxk-matrix; denote the restriction of B to V by B'. Thus B' has at least an eigenvector w in V (something you took for granted for A, using already the fact that we're over an algebraically closed field): Bw = B'w = mu.w, and as noted before Aw = lambda.w.

    • @GeekProdigyGuy
      @GeekProdigyGuy ปีที่แล้ว

      What is the restriction of B to V? I think this requires significantly more work for change of basis.

  • @bentoomey15
    @bentoomey15 2 ปีที่แล้ว +6

    Another great video. Today, for some reason, I finally thought, "gotta be a climber." Looked you up, and sure enough, a really good one!

  • @cyrillechevallier7835
    @cyrillechevallier7835 2 ปีที่แล้ว +3

    NOT THE RIGTH WAY TO DO IT !!!
    A has at least one eigenvalue (we’re in C), let’s call it m with x as one of its eigenvector,
    ABx =BAx = mBx so S = the sub-eigenspace of m for A is stable by B, so B is an endomorphism of S
    => because we’re in C, like at the start, there is r, and y in S (it’s important) such that By = ry, but by definition y is in S so :
    Ay = my
    y is what u want

  • @taraszablotskyi1479
    @taraszablotskyi1479 2 ปีที่แล้ว +1

    13:10 you can prove that each term of the factorization commute with the others and put the factor that gives zero(miu m) to be the first one that multiply V so V would be a eigenvector of B. This would prove that each eigenvector of A is an eigenvector of B and vice versa.

  • @AnCar88
    @AnCar88 2 ปีที่แล้ว +22

    This is a nice solution. Still there is a bit of discussion that was omitted for the case that Bv is actually 0. Of course one could pick a different v.
    In fact, if there exists some eigenvector of A such that Bv=0, the problem is much easier. Namely, in this case, due to commutativity, BAv=0.
    If Av=0 as well (namely the eigenvalue of A associated to v was 0), then we have that Av=Bv=0 so v is the common eigenvector corresponding to the 0 eigenvalue for both matrices.
    If Av
    eq 0, then, since v was chosen as an eigenvector of A, let the associated eigenvalue be a
    eq 0, i.e., Av=av. But then BAv=0 implies Bav=0, so Bv=0 and thus v is an eigenvector of B as well (with eigenvalue 0).

    • @skvortsovalexey
      @skvortsovalexey 2 ปีที่แล้ว +10

      If Bv is 0, then v is eigenvector of B with eigenvalue = 0, so v is shared eigenvector of A and B ( Bv = 0*v ).

    • @AnCar88
      @AnCar88 2 ปีที่แล้ว +2

      @@skvortsovalexey Oh, yes that's even more straightforward then what i said.

  • @synaestheziac
    @synaestheziac 2 ปีที่แล้ว +6

    7:16 shirt transformation. But is it linear?

  • @l.a.s8274
    @l.a.s8274 2 ปีที่แล้ว +3

    If E is an eigenspace of A, then BE is contained in E. So B restricts to E. Call this restriction B*. B* has an eigenvector by the algebraic closure of C, which is a common eigenvector of B and A. Am i missing something ?

    • @bob53135
      @bob53135 2 ปีที่แล้ว +2

      That's how I solved it. Seems quick and efficient :)

  • @tracyh5751
    @tracyh5751 2 ปีที่แล้ว

    And thus the center of GL(n, CC) is the subgroup of the scalar matrices.

  • @washieman2445
    @washieman2445 2 ปีที่แล้ว +3

    At 12:30, why must one of the matrices be 0? There can exist nonzero matrices that multiply to zero.

    • @ngoriyasjil2085
      @ngoriyasjil2085 2 ปีที่แล้ว +5

      This confused me as well a little bit, but it's not actually about one of the matrices being zero, it's about the *product* being zero at that point.
      Since the whole expression is zero, either the left-most matrix takes a non-zero vector to zero, or the vector was already zero by that point, so we can ask the same question of the second matrix, and so on. At some point we must have a non-zero vector being taken to zero by a (B-µI) matrix
      I hope that helps!

  • @amidhmi5243
    @amidhmi5243 2 ปีที่แล้ว +2

    If A is the identity matrix, its eigenvectors are the unit base vectors (1,0,0,0,...), (0,1,0,0,...),...
    Does it mean any matrix will have at least a unit base vector as eigenvector?

    • @Alex_Deam
      @Alex_Deam 2 ปีที่แล้ว +6

      Not quite, but good question. Remember that if two eigenvectors have the same eigenvalue, then you can take any linear combination of those eigenvectors to get another eigenvector. For the identity matrix, with unit eigenvectors, all the eigenvalues equal 1. Which means you can take any linear combination to get another eigenvector i.e. literally every vector is an eigenvector of the identity matrix, as you'd expect given it's the identity! So the eigenspace for I_n is just the whole space, C_n. So naturally every matrix will share eigenvectors with the identity, given that any eigenvector must be in C_n. Those shared eigenvectors just won't necessarily be the unit basis vectors themselves.

  • @user-dpscriberz
    @user-dpscriberz 2 ปีที่แล้ว +23

    Not first but still earlier than Good Place to Stop

    • @moonlightcocktail
      @moonlightcocktail 2 ปีที่แล้ว +4

      Also earlier than Good Place to Stop

    • @Asterism_Desmos
      @Asterism_Desmos 2 ปีที่แล้ว +4

      Also earlier than good place to stop

    • @bjornfeuerbacher5514
      @bjornfeuerbacher5514 2 ปีที่แล้ว

      Good Place to Stop is really late today. :/

    • @zalut_sky
      @zalut_sky 2 ปีที่แล้ว

      STILL earlier than Good Place to Stop 😆

  • @hassanalihusseini1717
    @hassanalihusseini1717 2 ปีที่แล้ว

    Thank you for this lecture!

  • @cmilkau
    @cmilkau 2 ปีที่แล้ว

    But then A - λI, B - μI (λ, μ the corresponding Eigenvalues) operating on a complement of span v (v being the common Eigenvector) also commute and hence also share at least one Eigenvector and so on ... so in this finite case, shouldn't they share a common diagonalization basis? What happens in infinite dimensions?
    EDIT: this works if and only if A,B have a diagonalization (otherwise, in the finite case, at some point no complement of span v is invariant under A,B). E.g. the shift operator only has one Eigenvector (up to scalar multiples).

  • @RickyisSwan
    @RickyisSwan 2 ปีที่แล้ว +6

    Thumbnail equation is very easy to solve. It’s a pop group.

  • @replicaacliper
    @replicaacliper 2 ปีที่แล้ว +18

    This is a very important result in quantum mechanics

    • @HershO.
      @HershO. 2 ปีที่แล้ว +1

      So cool to me how physics and math go hand in hand almost everywhere.

    • @masonskiekonto590
      @masonskiekonto590 2 ปีที่แล้ว +1

      @@HershO. Well, that's because the most efficient way of modeling our reality is through maths.

  • @depressedguy9467
    @depressedguy9467 2 ปีที่แล้ว

    Well also normal operators have same eigen vectors

  • @davidblauyoutube
    @davidblauyoutube 2 ปีที่แล้ว +2

    Nice result! And nice costume change between Act 2 and Act 3. 😂

  • @PsyKosh
    @PsyKosh 2 ปีที่แล้ว

    This may be a dumb question, but is there an analogous thing for infinite dimensional vector spaces like function spaces or such? That is, what conditions would be needed to be fulfilled so that it'd apply to those? This proof obviously isn't going to work in those cases, so..

  • @Nova206
    @Nova206 2 ปีที่แล้ว +2

    It is a bit strange as a German when words beginning with eigen are also used in english language. Because eigen is a german word and it wont be translated. In the most cases it would be translated as own. For Example eigenvector in english is Eigenvektor in german with the same spelling, but the german Eigentum is Ownership.

    • @videolome
      @videolome 2 ปีที่แล้ว +3

      There was a time that Germany dominated the math world. There are many things that we still use. For instance, the notation Z for the set of integers.
      But then something happened…

    • @riadsouissi
      @riadsouissi 2 ปีที่แล้ว

      French uses the word "propre" for eigen which means own.

    • @arrotino9501
      @arrotino9501 2 ปีที่แล้ว +1

      We use “auto” in italian (autovettori, autovalori), autovectors could probably work in english, or selfvectors

    • @axelnils
      @axelnils 2 ปีที่แล้ว

      Egenvektor in swedish

    • @marcoantonioircide740
      @marcoantonioircide740 2 ปีที่แล้ว

      @@arrotino9501 Piccola curiosità storica: i termini italiani hanno ricalcato praticamente sempre il variare della terminologia inglese per quanto riguarda questo argomento. Sono stati usati, in ordine: radici o valori latenti, radici proprie o caratteristiche e, in fine, autovalori

  • @TheDannyAwesome
    @TheDannyAwesome ปีที่แล้ว

    Does the converse hold true? I noticed that there was no assumption on lambda, so we could immediately strengthen this to, corresponding to each eigenvalue there is a shared eigenvector. In this case would the converse hold true? And if still no, is there any known further strengthening of the proposition in the video to make it a double implication?

  • @KosmasNikopoulos
    @KosmasNikopoulos 2 ปีที่แล้ว +1

    More linear algebra problems!!!

  • @vidiot5533
    @vidiot5533 2 ปีที่แล้ว

    linear algebra is so abstract that it can be hard to follow. introducing nondescriptive constants mutliplied by some variables just to transform them into other nondescriptive constants multiplied by a different set of variables just to prove that a redefinition of those variables equals some function that, rewritten with yet another set of variables proves that the share this feature.... got it lol. this coming from someone who understands how differential equations and fourier transforms work

  • @kenanwood6916
    @kenanwood6916 2 ปีที่แล้ว +3

    Isn't your factorization of p(z) predicated upon a_n being nonzero, so that it actually has n roots? What happens if a_n = 0?

    • @alexismiller2349
      @alexismiller2349 2 ปีที่แล้ว +3

      Good point, he should've put a polynomial of degree

    • @nimennacnamme6328
      @nimennacnamme6328 2 ปีที่แล้ว

      I just had the same problem :D

    • @NotoriousSRG
      @NotoriousSRG 2 ปีที่แล้ว

      What happens is Godzilla eats the square root of 2.

    • @Ligatmarping
      @Ligatmarping 2 ปีที่แล้ว

      No problem, if you can get the v, Bv, B^2v, etc to be l.d. before n, just use that a_k instead of a_n. If you need the n+1 vectors, it's because a_n is nonzero (else you would have a 0 combination before). I speak spanish so sorry if something is badly written.

  • @pierrickbramberger4206
    @pierrickbramberger4206 ปีที่แล้ว

    Since an eigenspace of A is B-invariant you can simply restrict the domain of B to an eigenspace of A and find an eigenvector of B there!?

  • @BenfanichAbderrahmane
    @BenfanichAbderrahmane 2 ปีที่แล้ว +1

    U have to distinguish the case when Bv=0

    • @adriengui2646
      @adriengui2646 2 ปีที่แล้ว +3

      If Bv = 0 you have Bv = 0.v so v is the shared eigenvector of A and B

  • @friedrichwaterson3185
    @friedrichwaterson3185 2 ปีที่แล้ว

    So coool thank you

  • @depressedguy9467
    @depressedguy9467 2 ปีที่แล้ว

    One of my physics mate told this to me

  • @pandabearguy1
    @pandabearguy1 2 ปีที่แล้ว +1

    If AB*BA = BA*AB, then A+B = B + A and AB-BA=0.
    My fields medal awaits

  • @ttrss
    @ttrss 2 ปีที่แล้ว

    How do we know lambda and mu are distinct?

  • @alexismiller2349
    @alexismiller2349 2 ปีที่แล้ว

    Btw I don't see the annotation of the calculus analysis video popping up at the end of the video

    • @yoav613
      @yoav613 2 ปีที่แล้ว

      That's why he always says:" it should be in the screen.." and not it is in the screen..

    • @alexismiller2349
      @alexismiller2349 2 ปีที่แล้ว

      @@yoav613 you're right he didn't say anything wrong

  • @christopherellis2663
    @christopherellis2663 2 ปีที่แล้ว

    [(A+B)/2]²-[(A-B)/2]²= AB=BA

  • @muhammadmahdidacosta5188
    @muhammadmahdidacosta5188 2 ปีที่แล้ว +3

    Why is linear algebra the most important math class?

    • @synaestheziac
      @synaestheziac 2 ปีที่แล้ว +3

      It’s definitely the most important for computer science and quantum mechanics applications, but I’m not sure if that’s Michael’s reason for saying that

    • @leif_p
      @leif_p 2 ปีที่แล้ว +4

      I've also seen applications in mechanical eng., chemical eng., electrical eng., (pretty much all engineering fields, really). I think it's so widely applicable because most tractable problems are just systems of linear equations, and if you have an intractable problem, you very often can get away with discretizing it and solving the linear approximation on a computer.

    • @synaestheziac
      @synaestheziac 2 ปีที่แล้ว

      @@leif_p yeah exactly

  • @michaelduffy7469
    @michaelduffy7469 2 ปีที่แล้ว +1

    Is the converse true, ie matrices that share an eigenvalue commute?

    • @APaleDot
      @APaleDot 2 ปีที่แล้ว +2

      No, and it's pretty easy to construct a counter-example. Consider two transformations in 3D space: the matrix A rotates vectors around the x-axis, the matrix B scales vectors along the y-axis. Both of these transformations leave all vectors lying along the x-axis unchanged, therefore Ax = λx and Bx = λx (and λ=1). But it's pretty clear to see that rotating and then scaling will give a different result than scaling then rotating.
      For instance, if you have a vector along the y-axis and rotate it around the x-axis 90°, it now lies on the z-axis and the scaling doesn't affect its length at all. But if you scale first, its length is different.

    • @bob53135
      @bob53135 2 ปีที่แล้ว

      No. Not even if they share an eigenvector, even with the same eigenvalue: counter-example is (1 2; 0 -1) and (1 -2; 0 -1). Those both have (1;0) as eigenvector for eigenvalue 1, but they don't commute.

  • @vcaro12
    @vcaro12 2 ปีที่แล้ว +1

    Wouldn't this reasoning apply to all eigenvectors? Why does the proposition claims only "at least one"?

    • @immakiku
      @immakiku 2 ปีที่แล้ว +5

      You don’t know that the procedure doesn’t find the same w for other eigenvectors

    • @skvortsovalexey
      @skvortsovalexey 2 ปีที่แล้ว +1

      @@immakiku but it finds different vectors for different lambdas. So there is as much different vectors as different eigenvalues of matrix A (or B) exist.

    • @bjornfeuerbacher5514
      @bjornfeuerbacher5514 2 ปีที่แล้ว +1

      @@skvortsovalexey And why should one suppose that there are different lambdas? A could have only one eigenvalue lambda (with the algebraic multplicity n).

    • @skvortsovalexey
      @skvortsovalexey 2 ปีที่แล้ว +3

      @@bjornfeuerbacher5514 We should not suppose. But if different lambdas (of A or B) exist, then...

    • @immakiku
      @immakiku 2 ปีที่แล้ว

      @@skvortsovalexey (edit - yes I think it’s true that A and B share all eigenvectors), but at least this proof doesn’t seem sufficient still. You can have second set of different eigval and eigvec, but how do you know from this proof the w you find isn’t just a rescaled version of the w you found earlier?

  • @chrissquarefan86
    @chrissquarefan86 2 ปีที่แล้ว

    12:18 what if (B-m_n * I)v is 0? Then there's no cutoff point. This case needs to be discussed.

    • @enderyu
      @enderyu 2 ปีที่แล้ว +3

      Then q(B) = 1 and the shared eingevector is the original vector v

  • @egoreremeev9969
    @egoreremeev9969 2 ปีที่แล้ว

    Doesn't these two matrices have all of their eigenvectors shared (if they are diagonalizable)?

    • @chuckaway6580
      @chuckaway6580 2 ปีที่แล้ว

      Yes, but that is a fundamentally different result with stronger hypotheses and thus a stronger conclusion.

    • @egoreremeev9969
      @egoreremeev9969 2 ปีที่แล้ว

      @@chuckaway6580 what about generalized vectors, what about Jordan form? would they share be simultaneously convertible to Jordan form?

  • @eyezuel5307
    @eyezuel5307 2 ปีที่แล้ว

    I don't even know what an eigenvector is 😔

  • @glacagnina
    @glacagnina ปีที่แล้ว

    cool! it reminds me of Krylov subspaces, something I have not heard about since 20 years plus :-)

  • @rogorix4991
    @rogorix4991 2 ปีที่แล้ว +1

    Completely off topic but I had a math exam this afternoon had not looked at any L.A. on ytb for months and what you know it shows this in my recommandations , like wow the odds !
    Well what may be true is I am on a Google account althought not having any direct contacts to people in my class, I think google linked phone numbers to Instagram and WhatsApp or smth were some people of my class have perhaps looked up such a video after the test and then It decided to suggest it.
    Anyways this is very freaky, I refuse to believe its a coincidence greetings from Switzerland anyways.
    Like legit a whole problem on
    [A,B]->AB-BA

  • @cmilkau
    @cmilkau 2 ปีที่แล้ว

    if Av = λv, then ABv = λBv (stumbled across this)

  • @travisporco
    @travisporco 2 ปีที่แล้ว

    this is nice

  • @daboffey
    @daboffey 2 ปีที่แล้ว

    One point: in producing p(z) as a product, you assumed that a = a sub n was non-zero, which is not necessarily true.

  • @JosBergervoet
    @JosBergervoet 2 ปีที่แล้ว

    Michael, you should state more explicitly (at this low level) that "each matrix has at least one eigenvector" th-cam.com/video/0elP0J91gew/w-d-xo.html
    People who don't know that already would perhaps not get that message from what you say in the video...

  • @cazadoroculto3219
    @cazadoroculto3219 2 ปีที่แล้ว

    And, In the Reals..?🤔

  • @jimiwills
    @jimiwills 2 ปีที่แล้ว

    Nice

  • @icew0lf98
    @icew0lf98 2 ปีที่แล้ว

    nice

  • @Thankmel8r
    @Thankmel8r 2 ปีที่แล้ว

    If AB=BA, then BA=AB

    • @АндрейДенькевич
      @АндрейДенькевич 2 ปีที่แล้ว

      Your logic is commutative. Non commutative logic has curvature caused by hidden '0 ' or '1' of higher dimension logic. It's my own as layman(programmer) opinion.
      '0' is externity, '1' is internity.
      For example "if a=b then b:=a+1". In this line "b:=a+1" causes this curvature.
      In our case higher dimension logic is computer. It's "0" is processor(energy), it's "1" is memory(space). "b:=a+1" goes from "0" to "1" and see curved "b" because
      while going through "0" "b" was incremented. "a=b" goes from "1" to "0" and see "b" not incremented.
      Curvature is positive because caused by "0".

  • @trueriver1950
    @trueriver1950 2 ปีที่แล้ว

    It's this just an excuse to celebrate ABBA?
    (any younger folk might need to ask a boomer about who ABBA were)

  • @dXter76
    @dXter76 2 ปีที่แล้ว

    linear algebra - where Mewtwo is just a root of a polynomial

  • @anamikagudia2506
    @anamikagudia2506 2 ปีที่แล้ว +2

    First

  • @ranani20
    @ranani20 2 ปีที่แล้ว +2

    first

  • @diogeneslaertius3365
    @diogeneslaertius3365 2 ปีที่แล้ว +1

    15:05 we do not have Aw = lambda w because f(z) is not q(z). The proof is obviously correct except for that last transition which is weird and probably incorrect.

    • @Alex_Deam
      @Alex_Deam 2 ปีที่แล้ว +2

      It says "for all f(z)" i.e. f(z) is any generic polynomial, remember from the proof of the earlier lemma. q(z) is an example of an f(z) and therefore fits the rule and is correct.

  • @NotoriousSRG
    @NotoriousSRG 2 ปีที่แล้ว

    You changed shirts mid video lol