Quantizing a two-dimensional Fermion Oscillator

[strangerep]

In the chapter Dirac Field, P&S put anti-commutation
relations to the fields only in the quantization. They start
the chapter with the classical Dirac field, and there is no
mentioning about anti-commuting Grassmann numbers in that
context yet. They let the reader assume that the classical
Dirac field is \psi\in\mathbb{C}^4 and it is
only the quantized field that anti-commutes.

Actually, in that chapter they start by *attempting* to
quantize the classical field, and find that it doesn't work.
Then they assume anti-commutation instead. This doesn't
really correspond to any process of "quantization" unless
you take the classical variables as anti-commuting in the
first place.

I'm still trying to keep hopes up for the
possibility, that the anti-commuting classical variables
would belong only to the path integral quantization, because
at the moment it seems the only way this could start making
sense.

Quantization does not make rigorous sense. The passage from
classical to quantum is ill-defined guesswork. It's better
to think of the quantum theory and then see that a limit
as \hbar\to 0 gives a sensible classical theory.

I used to call the P&S Introduction to the QFT a
"bible of QFT", because the proofs are left as matter of"faith".

Huh? I've always been able to follow their proofs. But
nobody claims quantization is a "proof". So even though it's
a dark art, you don't have to "believe" in it without
evidence. The "proofs" are in whether it works
experimentally.

 

[strangerep]

[...] and the equations of motion are
<br /> \dot{x}\; =\; \frac{\partial H}{\partial p_x}\; +\;<br /> u_1\frac{\partial\phi_1}{\partial p_x}\;+\; u_2<br /> \frac{\partial\phi_2}{\partial p_x} \; =\; u_1,<br />
[...]

I note that you don't have much time, but I think you need
to re-study the Wiki page pen-in-hand. (I.e.,
http://en.wikipedia.org/wiki/Dirac_bracket). Don't just
skim-read it.

The point of the Dirac bracket is that, at the end of the
procedure, you can continue to use the original eqns
of motion (no u's), provided you use the Dirac bracket in
place of the Poisson bracket. The Dirac bracket respects the
constraints, unlike the Poisson bracket.

<br /> \{\phi_1, \phi_2\}_{PB}<br /> ~=~ \partial_y \phi_1 \partial_{p_y} \phi_2<br /> - \partial_{p_x} \phi_1 \partial_x \phi_2<br /> ~=~ -2 ~.<br />

I don't understand how one could see from this what are second class
constraints.

Again, study the Wiki page when you're not rushed for time.

Any phase-space function f(q,p) is called "first class" if its Poisson
bracket with all of the constraints weakly vanishes, that is,
\{f, \phi_j\}_{PB} \approx 0, \forall j. So a constraint
\phi_i is called first class if its PB with all
the other constraints vanishes weakly (i.e: becomes 0 when you
set all the \phi's to 0). Since the PB above is -2,
it doesn't vanish weakly, hence the constraints themselves
are "second class" in this case.

[...] Am I now correct to say that

<br /> u_2+x = 0,\quad u_1-y = 0<br />

are the secondary constraints?

No. Look at points 1-4 in the "Consistency conditions" section
of the Wiki page. From point 3, "secondary constraints" do not
involve the u_k.

Rather, the above correspond to Wiki's point 4 (equations that
help determine the u_k).

But in your case, you need not muck around with the
u_k too much. You can just jump from the PB
of constraints to the matrix M_{ab}, which is
the crucial thing needed to write down the Dirac brackets.
That's what I did in my earlier post.

About 2/3 of math in #46 & #47 still ahead...

Probably better to study all of it thoroughly, together with
Wiki, before attempting a reply.

 

[jostpuur]

No. Look at points 1-4 in the "Consistency conditions" section
of the Wiki page. From point 3, "secondary constraints" do not
involve the u_k.

Ok, I made a mistake. I have the Dirac's lecture notes now, and tried to read it from there. He talks about different kind of equations, and then says that one of those are called secondary constraints, and I simply made a mistake when interpreting about what kind he was talking.

 

[jostpuur]

Any phase-space function f(q,p) is called "first class" if its Poisson
bracket with all of the constraints weakly vanishes, that is,
\{f, \phi_j\}_{PB} \approx 0, \forall j. So a constraint
\phi_i is called first class if its PB with all
the other constraints vanishes weakly (i.e: becomes 0 when you
set all the \phi's to 0). Since the PB above is -2,
it doesn't vanish weakly, hence the constraints themselves
are "second class" in this case.

So functions being first class or second class are different thing from primary and secondary constraints?

A second try:

<br /> p_x - y = 0,\quad p_y + x = 0<br />

are the primary constraints, and

<br /> u_2+x = 0,\quad u_1-y = 0<br />

are consistency conditions involving u's with no better name?

 

[strangerep]

So functions being first class or second class are different thing from primary and secondary constraints?

Yes.

<br /> p_x - y = 0,\quad p_y + x = 0<br />
are the primary constraints,

Yes.

and
<br /> u_2+x = 0,\quad u_1-y = 0<br />
are consistency conditions involving u's with no better name?

Yes.

 

[jostpuur]

More thoughts on multiplying:

It would not make sense to say that the nature of \mathbb{R}^3 is such, that the product \mathbb{R}^3\times\mathbb{R}^3\to\mathbb{R}^3 is given by the cross product (x_1,x_2)\mapsto x_1\times x_2. We can define what ever products on \mathbb{R}^3 we want, and different products could have different applications, all correct for different things. If you don't know what you want to calculate, then none of the products would be correct.

Similarly, it doesn't make sense to say that the Nature is of such kind, that the product of the classical Dirac field is given by the anti-commuting Grassmann product. I could even define my own product (E_1,E_2)\mapsto E_1 E_2 for the electric field, with no difficulty! So the real question is, that for what purpose do we want the anti-commuting Grassmann product?

Actually, in that chapter they start by *attempting* to quantize the classical field, and find that it doesn't work.

I would have been surprised if the same operators would have worked for the Dirac field, that worked for the Klein-Gordon field, since the Dirac field has so different Lagrange's function. It would not have been proper quantizing. You don't quantize the one dimensional infinite square well by stealing operators from harmonic oscillator either!

Then they assume anti-commutation instead.

For the operators. There is no clear mentioning about anti-commuting classical variables in this context.

This doesn't really correspond to any process of "quantization" unless you take the classical variables as anti-commuting in the first place.

I'm not arguing against this, and not believing either. I must know what would happen to the operators if classical variables were not anti-commuting.

 

[jostpuur]

By the standard prescription, we can quantize the theory by using
the original Hamiltonian \widehat H := \hat x^2 + \hat y^2,
together with the commutation relations:
<br /> [\hat x,\hat y] = [\hat x,\hat p_x]<br /> = [\hat y,\hat p_y] = [\hat p_x,\hat p_y]<br /> = \frac{i\hbar}{2} ~~~~~~(CR)<br />

On page 34 Dirac says

We further impose certain supplementary conditions on the wave function, namely:
<br /> \phi_j\psi=0.<br />

I suppose the motivation behind this is, that this way the classical limit will respect the original constraints.

It is so easy to write

<br /> (\hat{p}_x - \hat{y})\psi = 0<br />

<br /> (\hat{p}_y + \hat{x})\psi = 0,<br />

but can one do with these? Would the next step be to solve some explicit representations for these operators? It seems a difficult task, with so strange commutation relations between them.

 

[jostpuur]

Or then it is not so difficult. For example

<br /> \hat{x} = x + i\hbar\partial_y<br />

<br /> \hat{y} = y + \frac{i\hbar}{2}\partial_x<br />

<br /> \hat{p}_x = -\frac{1}{2}y - i\hbar\partial_x<br />

<br /> \hat{p}_y = -x - i\hbar\partial_y<br />

have these commutation relations, but this is not the only possible choice.

 

[jostpuur]

hmhmhmhmh... it would be a Schrödinger's equation

<br /> i\hbar\partial_t \psi \;=\; \big(x^2 \;+\; y^2 \;+\; i\hbar(2x\partial_y \;+\; \frac{1}{2}y\partial_x) \;-\; \hbar^2(\frac{1}{4}\partial^2_x \;+ \;\partial_y^2)\big)\psi<br />

with a supplementary condition

<br /> (y+i\hbar\partial_x)\psi = 0<br />

then?

 

[strangerep]

It would not make sense to say that the nature of \mathbb{R}^3 is such, that the product \mathbb{R}^3\times\mathbb{R}^3\to\mathbb{R}^3 is given by the cross product (x_1,x_2)\mapsto x_1\times x_2. We can define what ever products on \mathbb{R}^3 we want, and different products could have different applications, all correct for different things. If you don't know what you want to calculate, then none of the products would be correct.

Right. \mathbb{R}^3 is just a representation space which can carry various
algebras. The usual cross product of vectors corresponds to the Lie algebra o(3). The
fundamental thing in any model of physical phenomena is the abstract algebra
underlying it. One can then construct concrete representations of this algebra on various
representation spaces.

The confusing thing about \mathbb{R}^3 and o(3) is that there's an
isomorphism between them, so one tends to think of them as the same thing. But
that temptation should be resisted. First choose the abstract algebra, then decide
what representation space is most convenient for calculations.

Similarly, it doesn't make sense to say that the Nature is of such kind, that the product of the classical Dirac field is given by the anti-commuting Grassmann product. I could even define my own product (E_1,E_2)\mapsto E_1 E_2 for the electric field, with no difficulty!

In general, one must show that the algebra is closed. In the simple case above, it means
that all such products must be in the original algebra, which is easy enough for the
simple commutative algebra above. But if one writes down a non-commuting algebra,
one must show that [A,B] is also in the original algebra, i.e., that if [A,B]=C
then C is in the original algebra. That's part of the definition of a Lie algebra,
i.e., for any A,B in the algebra, the commutator [A,B] is equal to
a linear combination of the basis elements of the algebra.

So the real question is, that for what purpose do we want the anti-commuting Grassmann product?

Because any theory of electrons must be wrong unless the Pauli exclusion principle
is in there somewhere. That means we need an algebra such that A^2 = B^2 = 0,
etc, etc. Now, given a collection of algebra elements that all square to zero, we can
take a linear combinations of these, e.g., E=A+B and to get E^2 = 0
we must have AB + BA = 0. I.e., if we want the Pauli exclusion principle, together
with symmetry transformations that mix the algebra elements while continuing to respect
the Pauli principle, it is simpler just to start from a Grassman algebra where AB + BA = 0
and then A^2 = 0 becomes a special case.

I would have been surprised if the same operators would have worked
for the Dirac field, that worked for the Klein-Gordon field, since the
Dirac field has so different Lagrange's function. It would not have
been proper quantizing. You don't quantize the one dimensional
infinite square well by stealing operators from harmonic oscillator
either!

They're not using "the same operators that worked for the K-G field".
They're (attempting to) use the same prescription based on a
correspondence between Poisson brackets of functions on phase space,
and commutators of operators on Hilbert space. They find
that commutators don't work, and resort to anti-commutators. So in the
step between classical phase space and Hilbert space, they've
implicitly introduced a Grassman algebra even though they don't use
that name until much later in the book. The crucial point is that
the anti-commutativity is introduced before the correct Hilbert space
is constructed.

I must know what would happen to the operators if classical
variables were not anti-commuting.

You get a theory of electrons without the Pauli exclusion principle,
and without strictly positive energy. Such a theory is wrong.

 

[strangerep]

[...] this way the classical limit will respect the original constraints.

Wait,... let's go back to what I said in my previous post about algebras.
In (advanced) classical mechanics one works with functions over phase space,
e.g. f(p,q), g(p,q), etc. The Lagrangian action is such a function, and its
extremum gives the classical equation of motion through phase space.
The Hamiltonian is another such function.

The Hamiltonian formulation of such dynamics gives rise to the Poisson
bracket because we want any transformation of phase space functions to
leave the form of the Hamilton equations unchanged. Such transformations
form a group (a symplectic group) whose Lie algebra is expressed by the
Poisson bracket. I.e., we have an infinite-dimensional Lie algebra, consisting
of the set of functions f(p,q), g(p,q), etc, etc, all of whose Poisson brackets with
each other yield a function which is itself in the set. That's the important
thing - the product expressed by the Poisson bracket must close on the algebra.

For well-behaved cases (where the Poisson brackets close on the algebra),
quantization can then proceed by taking this Lie algebra and representing
it via operators on Hilbert space. For the ill-behaved cases with constraints,
the Poisson brackets don't close on the algebra, so we cannot yet perform
this quantization step. See below.

It is so easy to write

<br /> (\hat{p}_x - \hat{y})\psi = 0<br />

<br /> (\hat{p}_y + \hat{x})\psi = 0,<br />

but can one do with these? Would the next step be to solve some explicit representations for these operators?

No. We need a valid Lie algebra first. There's no point
trying to find a representation for an ill-defined algebra.

Suppose we have two functions f(p,q) and g(p,q) which satisfy the equations
of motion, and also respect the constraints. The crucial point is that
it is not automatic that h(p,q) := \{f,g\}_{PB} will
also satisfy the constraints. If h(p,q) doesn't satisfy the constraints,
we do not have a closed algebra, and therefore it's useless. We need
a closed Lie algebra. That's the whole point behind modifying
the Poisson bracket into the Dirac-Bergmann bracket. A function
b(p,q) := \{f,g\}_{DB}~~ does satisfy the constraints
and therefore gives a closed algebra which we can proceed to
represent sensibly on a Hilbert space.

 

[jostpuur]

I'm getting down to simpler questions: So classical Dirac field is not a map \mathbb{R}^3\to\mathbb{C}^4, but instead a map \mathbb{R}^3\to X, where X is some Grassmann algebra. Now... what is X? Is it a set? If it is, is there a definition for it, so that I could understand what it is? There exists lot of different Grassmann algebras, so that the information that X is a Grassmann algebra alone does not yet answer my question.

 

[strangerep]

[...]: So classical Dirac field is not a map \mathbb{R}^3\to\mathbb{C}^4, but instead a map \mathbb{R}^3\to X, where X is some Grassmann algebra. Now... what is X? Is it a set?

Not sure I understand the question. Any algebra is a set -- together with various operations
that map elements of the set amongst themselves.

If it is, is there a definition for it, so that I could understand what it is? There exists lot of different Grassmann algebras, so that the information that X is a Grassmann algebra alone does not yet answer my question.

Again, the algebra is just as described in Peskin & Schroeder section 9.5, especially
pp299-301. On p299, think of their \theta,\eta as corresponding to basis elements
(spin-up, and spin-down, say). Taking linear combinations of these basis elements (i.e., multiply
them by complex scalars, e.g., \lambda := A\theta + B\eta, where A,B are complex),
is enough to represent (massless) neutrinos. Let's call the space of all these combinations
"V". To get a massive Dirac field, one must recognize that taking the complex
conjugate of the above results in an inequivalent algebra \bar V(since they're not
related by a similarity transformation -- you can't get to \bar\lambda via a transformation
like \omega\lambda\omega^{-1}). The Dirac field is then just a direct sum of these
two inequivalent algebras. This is related to the stuff on p300 of P&S where
they introduce complex Grassman numbers, starting just before eq(9.65).

 

[jostpuur]

On p299, think of their \theta,\eta as corresponding to basis elements
(spin-up, and spin-down, say). Taking linear combinations of these basis elements (i.e., multiply
them by complex scalars, e.g., \lambda := A\theta + B\eta, where A,B are complex),
is enough to represent (massless) neutrinos.

I didn't understand \theta,\eta were supposed to be considered as some fixed basis elements. I though they are some arbitrary variables \theta,\eta\in X belonging to some set (and I'm now trying to figure out what the set X is). However, when they write expressions like

<br /> \int d\theta\; f(\theta)<br />

it sure doesn't look like \theta is some basis element. It looks like a variable that goes through some domain of different values. I mean, if the \theta is some fixed element, then the integral is as absurd as

<br /> \int d4\; f(4)<br />

 

[jostpuur]

Here, how to make given numbers grassmann, I gave a construction that makes the set \mathbb{R} anti-commuting. Is that construction completely disconnected from the Grassmann algebras we actually need in physics?

 

[strangerep]

I didn't understand \theta,\eta were supposed to be considered as some fixed basis elements. I though they are some arbitrary variables \theta,\eta\in X belonging to some set (and I'm now trying to figure out what the set X is).

Look at P&S pp301-302. Take eq(9.71):

<br /> \psi(x) ~=~ \sum_i \psi_i \phi_i(x) ~.<br />

Which are the "basis" elements? The \psi_i or the \phi_i(x)?
The answer depends which space you're focussing on --
the Grassmann values or the spacetime manifold. But what really
matters is the Grassmann-valued field on the LHS.

However, when they write expressions like

<br /> \int d\theta\; f(\theta)<br />

it sure doesn't look like \theta is some basis element. It looks like
a variable that goes through some domain of different values. [...]

The purpose of these Grassmann integrals is to define
functional integrals for fermionic fields. (See P&S's unnumbered eqn at the
top of page 302.)

Here, how to make given numbers grassmann, I gave a construction that makes the set R
anti-commuting. Is that construction completely disconnected from the Grassmann algebras we actually need in physics?

I didn't have time to follow your construction carefully, so I'll just say that
what really matters are the abstract algebraic rules, not how you represent them.

 

[jostpuur]

I didn't have time to follow your construction carefully, so I'll just say that
what really matters are the abstract algebraic rules, not how you represent them.

Unfortunately the mere knowledge of anti-commutation does not fix the construction up to any reasonable isomorphism, as my example in the linear algebra sub-forum shows, because it probably isn't anything that we need with the fermions now. Actually my construction was not an algebra according to the definition of algebra in mathematics... I should have noticed it... but it did have anti-commuting numbers at least!

We can define one three dimensional algebra like this. Set multiplications of the basis elements to be

(1,0,0)(1,0,0)=0
(1,0,0)(0,1,0)=(0,0,1)
(1,0,0)(0,0,1)=0
(0,1,0)(1,0,0)=-(0,0,1)
(0,1,0)(0,1,0)=0
(0,1,0)(0,0,1)=0
(0,0,1)(1,0,0)=0
(0,0,1)(0,1,0)=0
(0,0,1)(0,0,1)=0

and we get a bilinear mapping \cdot:\mathbb{R}^3\times\mathbb{R}^3\to\mathbb{R}^3, which makes (\mathbb{R}^3,\cdot) an algebra. If we then notate

<br /> \theta:=(1,0,0),\quad \eta:=(0,1,0),<br />

we can start calculating according to the rules

<br /> \theta\eta = -\eta\theta,\quad \alpha(\theta\eta)=(\alpha\theta)\eta\quad\alpha\in\mathbb{R}<br />

and so on...

Is this the kind of thing we need with fermions?

 

[strangerep]

Unfortunately the mere knowledge of anti-commutation does not
fix the construction up to any reasonable isomorphism, [...]

I think you mean "representation" rather than "construction". (You're devising a
concrete representation of an abstract algebra.) If one representation has different
properties than another, then some other algebraic item(s) have been introduced
somewhere.

[...]Is this the kind of thing we need with fermions?

Most people seem to get by ok using canonical anti-commutations relations
(or abstract Grassman algebras) directly. I still don't really know where
you're trying to go with all this.

BTW, "exterior algebras" are a well-known case of Grassman algebras.
The Wiki page for the latter even redirects to the former.

 

[jostpuur]

Most people seem to get by ok using canonical anti-commutations relations
(or abstract Grassman algebras) directly. I still don't really know where
you're trying to go with all this.

I am only trying to understand what P&S are talking about, and I'm still not fully convinced that it is like

<br /> \theta:=(1,0,0),\quad\eta:=(0,1,0)<br />

in my previous post, because it seems extremely strange to use notation

<br /> \int d\theta\;f(\theta) = \int d(1,0,0)\; f(1,0,0)<br />

for anything.

In fact now it would make tons of sense to define integrals like

<br /> \int\limits_{\gamma} d\gamma\;f(\gamma) = \lim_{N\to\infty}\sum_{k=0}^N (\gamma(t_{k+1}) - \gamma(t_k))f(\gamma(t_k))<br />

where \gamma:[a,b]\to\mathbb{R}^3 is some path, and where we use the Grassmann multiplication

<br /> \big(\gamma(t_{k+1}) - \gamma(t_k),\; f(\gamma(t_k))\big)\mapsto (\gamma(t_{k+1}) - \gamma(t_k))f(\gamma(t_k)).<br />

For example with

<br /> f(x_1,x_2,x_3) = (0,x_1^2,0)<br />

and

<br /> \gamma(t) = (t,0,0),\quad 0\leq t\leq L<br />

the integral would be

<br /> \int\limits_{\gamma} d\gamma\; f(\gamma) = (0,0,\frac{1}{3}L^3).<br />

I'm sure this is one good definition for the Grassmann integration, but I cannot know if this is the kind that we are supposed to have.

 

[jostpuur]

I think you mean "representation" rather than "construction". (You're devising a
concrete representation of an abstract algebra.)

I was careful to use word "construction", because the thing I defined in the linear algebra subforum was not an algebra. It was something else, but had something anti-commuting.

 

[strangerep]

I am only trying to understand what P&S are talking about, [...]

Have you tried Zee? I found P&S ch9 quite poor at explaining the essence of path
integrals the first time I read it. Especially the generating functional Z(J) and what it
is used for. Zee explains it more clearly and directly. After that, the more extensive
treatment in P&S started to become more understandable.

[...]I'm sure this is one good definition for the Grassmann integration[...]

Your definition of Grassman integration seems wrong to me (though again I don't
have time to fully deconstruct the details).
If f(\theta) is a constant, the integral is zero in standard Grassman
calculus, but yours looks like it would give some other value.

 

[jostpuur]

My construction where

<br /> \theta=(1,0,0),\quad\eta=(0,1,0),\quad\theta\eta=(0,0,1)<br />

was wrong. In P&S f is said to be a function of a Grassmann variable \theta. It is not possible for the theta to be a fixed basis vector.

Okey, I still don't know what the Grassmann algebra is.

If I denote G the construction I gave in linear algebra subforum (basically G=\mathbb{R}^2 with some additional information), perhaps

<br /> \bigoplus_{g\in G} \mathbb{C}<br />

could be a correct kind of algebra...

Have you tried Zee?

No. At some point I probably will, but it is always a labor to get new books. Library is usually empty of the most popular ones.

 

[strangerep]

Okey, I still don't know what the Grassmann algebra is.

Let A,B,C,... denotes ordinary complex numbers.

Then a 1-dimensional Grassman algebra consists of a single Grassman
variable \theta, its complex multiples A\theta,
and a 0 element, (so far it's a boring 1D vector space over \mathbb{C}),
and the multiplication rules \theta^2 = 0; A\theta = \theta A.

The most general function f(\theta) of a single
Grassman variable is A + B\theta (because higher order
terms like \theta^2 are all 0.

A 2-dimensional Grassman algebra consists of a two Grassman
variables \theta,\eta, their complex linear combinations,
A\theta + B\eta, a 0 element, (so far it's a 2D vector space
over \mathbb{C}), with the same multiplication
rules as above for \theta,\eta separately, but also
\theta\eta + \eta\theta = 0.

The most general function f(\theta,\eta) of a two
Grassman variables is A + B\theta + C\eta + D\theta\eta
(because any higher order terms are either 0 or reduce to a lower
order term).

And so on for higher-dimensional Grassman algebras.

That's about all there is to it.

Integral calculus over a Grassman algebra proceeds partly by analogy
with ordinary integration. In particular, d\theta is
required to be the same as d(\theta+\alpha) (where
\alpha is a constant Grassman number). This leads to
the rules shown in P&S at the top of p300 -- eqs 9.63 and 9.64.

 

[jostpuur]

Let A,B,C,... denotes ordinary complex numbers.

Then a 1-dimensional Grassman algebra consists of a single Grassman
variable \theta, its complex multiples A\theta,
and a 0 element, (so far it's a boring 1D vector space over \mathbb{C}),
and the multiplication rules \theta^2 = 0; A\theta = \theta A.

The most general function f(\theta) of a single
Grassman variable is A + B\theta (because higher order
terms like \theta^2 are all 0.

Could \theta as well be called Grassmann constant? If it is called variable, it sounds like \theta could have different values.

Also, if A and B are complex numbers, and I was given a quantity A+4B, I would not emphasize A and B being constants, and calling this expression the function of 4, like f(4)=A+4B.

 

[jostpuur]

Or is it like this: \theta can have different values, and there exists a Grassmann algebra for each fixed \theta?

 

[strangerep]

Could \theta as well be called Grassmann constant?

No.

If it is called variable, it sounds like \theta could have different values.

Consider a function f(x) where x is real. You wouldn't call "x" constant, even though any specific
value of x you plug into f(x) _is_ constant. \theta is an element of a 1-dimensional
vector space. Besides \theta, this vector space contains 0 and any complex multiple of
\theta, e.g: C\theta.

if A and B are complex numbers, and I was given a quantity A+4B, I would not
emphasize A and B being constants, and calling this expression the function of 4, like f(4)=A+4B.

All the symbols occurring "A+4B" are from the same vector space, i.e., \mathbb C,
so this is not the same thing as A+B\theta[/tex].

 

[jostpuur]

No.Consider a function f(x) where x is real. You wouldn't call "x" constant, even though any specific
value of x you plug into f(x) _is_ constant.

Ok. But then we need more precise definition for the set of allowed values of \theta. It is not my intention to only complain about lack of rigor, but I honestly haven't even got very good intuitive picture about this set either. I think I have now my own definition/construction ready for this, so that it seems to make sense, and I'm not sure that this claim:

\theta is an element of a 1-dimensional
vector space. Besides \theta, this vector space contains 0 and any complex multiple of
\theta, e.g: C\theta.

is fully right. For each fixed \theta we have a vector space V=\{C\theta\;|\;C\in\mathbb{C}\}, but I don't see how this could be the same set from which \theta was originally chosen.

Here's my way to get Grassmann algebra, where Grassmann variables would be as similar to the real numbers as possible:

First we define a multiplication on the \mathbb{R}^2 like it was done in my post in linear algebra subforum. That means, \mathbb{R}^2\times\mathbb{R}^2\to\mathbb{R}^2,

For all x\in\mathbb{R}, (x,0)(x,0)=(0,0).

If 0&lt;x&lt;x&#039;, then (x,0)(x&#039;,0)=(0,xx&#039;) and (x&#039;,0)(x,0)=(0,-xx&#039;).

If x&lt;0&lt;x&#039; or x&lt;x&#039;&lt;0 just put the signs naturally.

Finally for all (x,y),(x&#039;,y&#039;)\in\mathbb{R}^2 put (x,y)(x&#039;,y&#039;)=(x,0)(x&#039;,0)

Now the \mathbb{R} has been naturally (IMO naturally, perhaps somebody has something more natural...) extended to smallest possible set so that it has a nontrivial anti-commuting product.

At this point one should notice that it is not a good idea to define scalar multiplication \mathbb{R}\times\mathbb{R}^2\to\mathbb{R}^2 like (\lambda,(x,y))\mapsto (\lambda x,\lambda y), because the axiom (\lambda \theta)\eta=\theta(\lambda \eta) would not be satisfied.

However a set

<br /> G=\bigoplus_{(x,y)\in\mathbb{R}^2}\mathbb{C}<br />

becomes a well defined vector space, whose members are finite sums

<br /> \lambda_1(x_1,y_1)+\cdots+\lambda_n(x_n,y_n).<br />

It has a natural multiplication rule G\times G\to G, which can be defined recursively from

<br /> (\lambda_1(x_1,y_1) + \lambda_2(x_2,y_2))(\lambda_3(x_3,y_3) + \lambda_4(x_4,y_4)) = \lambda_1\lambda_3 (x_1,y_1)(x_3,y_3) + \lambda_1\lambda_4 (x_1,y_1)(x_4,y_4) + \lambda_2\lambda_3 (x_2,y_2)(x_3,y_3) + \lambda_2\lambda_4 (x_2,y_2)(x_4,y_4),<br />

where we use the previously defined multiplication on \mathbb{R}^2.

To my eye it seems that this G is now a well defined algebra and has the desired properties: If one chooses a member \theta\in G, one gets a vector space \{C\theta\;|\;C\in\mathbb{C}\}\subset G, and if one chooses two members \theta,\eta\in G, then the identity \theta\eta = -\eta\theta is always true.

 

[jostpuur]

Now I thought about this more, and my construction doesn't yet make sense. The identity \theta\eta=-\eta\theta would be true only if there is a scalar multiplication (-1)(x,y)=(-x,-y), which wasn't there originally. It could be made it too complicated because I was still thinking about my earlier construction attempt...

Then a 1-dimensional Grassman algebra consists of a single Grassman
variable \theta, its complex multiples A\theta,
and a 0 element, (so far it's a boring 1D vector space over \mathbb{C}),
and the multiplication rules \theta^2 = 0; A\theta = \theta A.

More ideas!:

I think this one dimensional Grassmann algebra can be considered as the set \mathbb{R}\times\{0,1\} (with (0,0) and (0,1) identified as the common origin 0), with multiplication rules

<br /> (x,0)(x&#039;,0)=(xx&#039;,0)<br />
<br /> (x,0)(x&#039;,1)=(xx&#039;,1)<br />
<br /> (x,1)(x&#039;,0)=(xx&#039;,1)<br />
<br /> (x,1)(x&#039;,1)=0<br />

Here \mathbb{R}\times\{0\} are like ordinary numbers, and \mathbb{R}\times\{1\} are the Grassmann numbers. One could emphasize it with Greek letters \theta=(\theta,1).

A 2-dimensional Grassman algebra consists of a two Grassman
variables \theta,\eta, their complex linear combinations,
A\theta + B\eta, a 0 element, (so far it's a 2D vector space
over \mathbb{C}), with the same multiplication
rules as above for \theta,\eta separately, but also
\theta\eta + \eta\theta = 0.

This would be a set \mathbb{R}\times\{0,1,2,3\} with multiplication rules

<br /> (x,0)(x&#039;,k) = (xx&#039;,k),\quad k\in\{0,1,2,3\}<br />
<br /> (x,1)(x&#039;,1) = 0<br />
<br /> (x,1)(x&#039;,2) = (xx&#039;, 3)<br />
<br /> (x,2)(x&#039;,1) = (-xx&#039;,3)<br />
<br /> (x,2)(x&#039;,2) = 0<br />
<br /> (x,k)(x&#039;,3) = 0 = (x,3)(x&#039;,k),\quad k\in\{1,2,3\}<br />

hmhmhmhmh?

Argh! But now I forgot that these are not vector spaces... Why cannot I just read the definition from somewhere...

btw. I think that if you try to define two dimensional Grassmann algebra like that, it inevitably becomes a three dimensional, because there are members like

<br /> x\theta + y\eta + z\theta\eta<br />

 

[jostpuur]

strangerep, I'm not saying that there would be anything wrong with your explanation, but it must be missing something. When the Grassmann algebra is defined like this:

Then a 1-dimensional Grassman algebra consists of a single Grassman
variable \theta, its complex multiples A\theta,
and a 0 element, (so far it's a boring 1D vector space over \mathbb{C}),
and the multiplication rules \theta^2 = 0; A\theta = \theta A.

The most general function f(\theta) of a single
Grassman variable is A + B\theta (because higher order
terms like \theta^2 are all 0.

It is already assumed, that we know from what set the \theta\in ? is.

Consider a function f(x) where x is real. You wouldn't call "x" constant, even though any specific value of x you plug into f(x) _is_ constant. \theta is an element of a 1-dimensional vector space. Besides \theta, this vector space contains 0 and any complex multiple of \theta, e.g: C\theta.

Once \theta exists, we get a vector space V_{\theta}:=\{c\theta\;|\;c\in\mathbb{C}\}, and it is true that \theta\in V_{\theta}, but you cannot use this vector space to define what \theta is, because the \theta is already needed in the definition of this vector space.

This is important. At the moment I couldn't tell for example if a phrase "Let \theta=4..." would be absurd or not. Are they numbers that anti-commute like 3\cdot4 = -4\cdot 3? Is the multiplication some map \mathbb{R}\times\mathbb{R}\to \mathbb{R}, or \mathbb{R}\times\mathbb{R}\to X, or X\times X\to X, where X is something?

 

[strangerep]

Why cannot I just read the definition from somewhere...

You have, but you also have a persistent mental block against it that is beyond my
skill to dislodge.

Once \theta exists, we get a vector space V_{\theta}:=\{c\theta\;|\;c\in\mathbb{C}\}, and it is true that \theta\in V_{\theta}, but you cannot use this vector space to define what \theta is,
because the \theta is already needed in the definition of this vector space.

\theta is an abstract mathematical entity such that \theta^2=0.
There really is nothing more to it than that.

This is all a bit like asking what i is. For some students initially, the answer
that "i is an abstract mathematical entity such that i^2 = -1" is
unsatisfying, and they try to express i in terms of something else they
already understand, thus missing the essential point that i was originally
invented because that's not possible.

 

[jostpuur]

The biggest difference between i and \theta is that i is just a constant, where as \theta is a variable which can have different values.

If I substitute \theta=3 and on the other hand \theta=4, will the product of these two Grassmann numbers be zero, or will it anti-commute non-trivially: Like 3\cdot 4= 0 = 4\cdot 3, or 3\cdot 4 = - 4\cdot 3 \neq 0?

Did I already do something wrong when I substituted 3 and 4? If so, is there something else whose substitution would be more allowed?

 

[jostpuur]

This is all a bit like asking what i is. For some students initially, the answer
that "i is an abstract mathematical entity such that i^2 = -1" is
unsatisfying, and they try to express i in terms of something else they
already understand, thus missing the essential point that i was originally
invented because that's not possible.

IMO you cannot get satisfying intuitive picture of complex numbers unless you see at least one construction for them. The famous one is of course the one where we set \mathbb{R}=\mathbb{R}\times\{0\}, i=(0,1)\in\mathbb{R}^2, and let \mathbb{R}\cup \{i\} generate the field \mathbb{C}.

Another one is where we identify all real numbers x\in\mathbb{R} with diagonal matrices

<br /> \left[\begin{array}{cc}<br /> x &amp; 0 \\ 0 &amp; x \\<br /> \end{array}\right]<br />

We can then set

<br /> i = \left[\begin{array}{cc}<br /> 0 &amp; 1 \\ -1 &amp; 0 \\<br /> \end{array}\right]<br />

and we get the complex numbers again.

\theta is an abstract mathematical entity such that \theta^2=0.
There really is nothing more to it than that.

There must be more. If that is all, I could set

<br /> \theta = \left[\begin{array}{cc}<br /> 0 &amp; 1 \\ 0 &amp; 0 \\<br /> \end{array}\right]<br />

and be happy. The biggest difference between this matrix, and the \theta we want to have, is that this matrix is not a variable that could have different values, but \theta is supposed to be a variable.


...


btw would it be fine to set

<br /> \theta = \left[\begin{array}{cc}<br /> 0 &amp; \theta \\ 0 &amp; 0 \\<br /> \end{array}\right]?<br />

 

[strangerep]

IMO you cannot get satisfying intuitive picture of complex numbers unless you see at least one construction for them. The famous one is of course the one where we set \mathbb{R}=\mathbb{R}\times\{0\}, i=(0,1)\in\mathbb{R}^2,
and let \mathbb{R}\cup \{i\} generate the field \mathbb{C}.

OK, I think I see the source of some of the confusion. Let's do a reboot, and
change the notation a bit to be more explicit...

Begin with a (fixed) nilpotent entity \Upsilon whose only properties
are that it commutes with the complex numbers, and \Upsilon^2 = 0.
Also, 0\Upsilon = \Upsilon 0 = 0. Then let \mathbb{A} := \mathbb{C}\cup \{\Upsilon\}
generate an algebra. I'll call the set of numbers \mathbb{U} := \{z \Upsilon : z \in \mathbb{C}\}
the nilpotent numbers.

I can now consider a nilpotent variable \theta \in \mathbb{U}.
Similarly, I can consider a more general variable a \in \mathbb{A}.
I can also consider functions f(\theta) : \mathbb{U} \to \mathbb{A}.

More generally, I can consider two separate copies of \mathbb{U}, called
\mathbb{U}_1, \mathbb{U}_2, say. I can then impose the condition
that elements of each copy anticommute with each other. I.e., if
\theta \in \mathbb{U}_1, ~\eta \in \mathbb{U}_2, then
\theta\eta + \eta\theta = 0. In this way, one builds up multidimensional
Grassman algebras.

 

[jostpuur]

Okey, thanks for patience I see this started getting frustrating, but I pressed on because confusion was genuine.

So my construction in the post #67 was otherwise correct, expect that it was a mistake to define

<br /> \theta:=(1,0,0),\quad \eta:=(0,1,0).<br />

Instead the notation \theta should have been preserved for all members of \langle(1,0,0)\rangle (the vector space spanned by the unit vector (1,0,0)), and similarly with \eta.

 

[jostpuur]

I would have been surprised if the same operators would have worked
for the Dirac field, that worked for the Klein-Gordon field, since the
Dirac field has so different Lagrange's function. It would not have
been proper quantizing. You don't quantize the one dimensional
infinite square well by stealing operators from harmonic oscillator
either!

They're not using "the same operators that worked for the K-G field".
They're (attempting to) use the same prescription based on a
correspondence between Poisson brackets of functions on phase space,
and commutators of operators on Hilbert space. They find
that commutators don't work, and resort to anti-commutators. So in the
step between classical phase space and Hilbert space, they've
implicitly introduced a Grassman algebra even though they don't use
that name until much later in the book. The crucial point is that
the anti-commutativity is introduced before the correct Hilbert space
is constructed.

I must know what would happen to the operators if classical
variables were not anti-commuting.

If we did not let the Dirac's field be composed of anti-commuting numbers, then wouldn't the canonical way of quantizising it be the quantization as constrained system, because that is what the Dirac's field is. It has constraints between the canonical momenta field and the \psi configuration. P&S are not talking about any constraints in their "first quantization attempt", but only try the quantization as harmonic oscillator.

 

[jostpuur]

You created this thread and gave it the title "fermion oscillator", yet you don't seem to know the difference between Fermi and Bose dynamics.
You wrote an incorrect Bosonic Lagrangian and asked us to help you quantize that wrong Lagrangian! You also asked us to obtain information from the wrong Bosonic Lagrangian and use that information to explain Fermion oscillator! These requests of yours are certainly meaningless!

I see my original question wasn't logical, but I have some excuse for this. My first encounter with the Dirac field was with the book by Peskin & Schroeder. They could have honestly said that they are going to postulate anti-commuting operators, but instead they preferred motivating the quantization somehow. Basically they introduce a classical Dirac field described by Lagrangian

<br /> \mathcal{L}_{\textrm{Dirac}} = \overline{\psi}(i\gamma^{\mu}\partial_{\mu} - m)\psi,<br />

which is an example of a system where the canonical momenta is constrained with the generalized coordinates according to

<br /> \pi =i\psi^{\dagger},<br />

and then explain that because this system cannot be quantizised the same way as harmonic oscillators can be, therefore the quantization of system described by \mathcal{L}_{\textrm{Dirac}} must involve anti-commuting operators. This is where I got the idea, that a constraint between canonical momenta and generalized coordinate leads into fermionic system, then devised the simplest example of a similar constrained system,

<br /> L=\dot{x}y - x\dot{y} - x^2 - y^2<br />

which has the constraint

<br /> (p_x,p_y) = (y,-x),<br />

and came here to ask that how does this give a fermionic system, and caused lot of confusion. Was that a my mistake? I'm not sure. It's fine if you think so. My opinion is that the explanation by Peskin & Schroeder sucks incredibly.

Your equations of motion represent a 2-D oscillator, your Lagrangian does not represent any system!

Here you are making a mistake. The Lagrangian I wrote is an example of a constrained system.

 

[strangerep]

[...] Peskin & Schroeder [...] could have honestly said that they are going to postulate anti-commuting operators, but instead they preferred motivating the quantization somehow.[...]

I think you are too harsh on P&S. In sect 3.5, bottom of p52, they have a section
"How Not to Quantize the Dirac Field". Then over on pp55-56 they show that
anti-commutation relations resolve various problems. The last two paragraphs
on p56 do indeed talk about postulating anti-commutation relations, but
they done so in the context of a larger discussion about why this is a good thing.

For the purposes of P&S's book, introducing the full machinery of constrained
Dirac-Bergman quantization would have consumed several chapters by itself,
and does not really belong in an "Introduction" to QFT.

 

[jostpuur]

For the purposes of P&S's book, introducing the full machinery of constrained
Dirac-Bergman quantization would have consumed several chapters by itself,
and does not really belong in an "Introduction" to QFT.

I wouldn't have expected them to explain the quantization of constrained systems, but their presentation left me in belief that it is the constraint between momenta and generalized coordinates, that forces us into anti-commuting brackets, and at the same time I was left fully unaware that there even existed some other theory about quantization with constraints. Assuming I'm now right when I think that the constraint never had anything to do with the anti-commuting brackets, I suppose it's: end fine, all fine?

I would be curious to know if I'm the only one who's had similar mislead thoughts with the Dirac field.

 

[jostpuur]

Some trivial remarks concerning a quantization of a zero dimensional system:

If we were given a task of quantisizing a system whose coordinate space is zero dimensional point, a natural way to approach this using already known concepts would be to consider a one dimensional infinitely deep well of width L, and study it on the limit L\to 0, because on this limit the one dimensional coordinate space becomes zero dimensional. All the energy levels diverge on the limit L\to 0,

<br /> E_n = \frac{\hbar^2\pi^2n^2}{2mL^2} \to \infty,<br />

however, the divergence of the ground state is not a problem, because we can always normalize the energy so that the ground state remains as the origo in the energy space. The truly important remark is that the energy difference between the ground state and all the excitation states diverge,

<br /> E_n - E_1 = \frac{\hbar^2\pi^2}{2mL^2}(n^2 - 1)\to\infty,\quad\quad n&gt;1,<br />

thus we can conclude that when the potential well is pushed zero dimensional, all the excitation states become unattainable with finite energies. My final conclusion of all this would be, that the zero dimensional one point system is quantized so that it has only one energy level, and thus very trivial dynamics.

A more interesting application of zero dimensional system:

We start with a one dimensional system with a following potential

<br /> V(x)=\left\{\begin{array}{ll}<br /> \infty,\quad&amp; x &lt; 0\\<br /> 0,\quad&amp; 0&lt;x&lt;L\\<br /> \infty,\quad&amp; L&lt;x&lt;M\\<br /> 0,\quad &amp;M&lt;x&lt;M + \sqrt{1 + \alpha L^2}L\\<br /> \infty,\quad &amp; M + \sqrt{1 + \alpha L^2}L &lt; x\\<br /> \end{array}\right.<br />

where \alpha&gt;0 is some constant. So basically the system consists of two disconnected wells. Other one has width L, and the other one \sqrt{1 + \alpha L^2}L. On the limit L\to 0 the excitation states of each well vanish again, but now it turns out that the difference between the ground states of the each well remains finite.

<br /> E_{\textrm{zero right}} - E_{\textrm{zero left}} \;=\; \frac{\hbar^2\pi^2}{2m}\Big(\frac{1}{1 + \alpha L^2}\; -\; 1\Big)\frac{1}{L^2}\; =\; \frac{\hbar^2\pi^2}{2m} \frac{\alpha}{1 + \alpha L^2} \;\to\; \frac{\hbar^2\pi^2\alpha}{2m}<br />

Now the behavior of the quantized system on the limit L\to 0 is that it has precisely two energy levels, which can be thought of as the particle occupying either one the zero dimensional points \{0\} or \{M\}, which together compose the coordinate space.

If on the other hand, I was given the knowledge that a behavior of some quantum system is such that it has two energy levels, and I was then given the task of coming up with a suitable classical coordinate space and a Lagrangian that produces this two level behavior, this is what I would give. System consisting of two points, or alternatively a limit definition starting with a more traditional one dimensional system. Would this be frowned upon? To me this looks simple and understandable, but would more professional theoreticians prefer devising some Grassmann algebra explanation for the asked two energy level system? How different would it be from the naive construction given by me here, in the end?