528 views

Skip to first unread message

Oct 29, 2006, 9:34:28 PM10/29/06

to

To guarantee f_xy (a,b) = f_yx (a,b), do we require that f_xy, f_yx be

continuous at (a,b) or a stronger requirement ( continuity in some

neighbourhood of (a,b) )?

Oct 30, 2006, 12:49:58 AM10/30/06

to

In article <1162175668....@e64g2000cwd.googlegroups.com>,

In Adams, "Calculus, A Complete Course" (third edition), sec.

13.4, the requirement is that f_{xy} and f_{yx} are

continuous at (a,b), while f, f_x and f_y are continuous in

a neighbourhood of (a,b).

Robert Israel isr...@math.ubc.ca

Department of Mathematics http://www.math.ubc.ca/~israel

University of British Columbia Vancouver, BC, Canada

Oct 30, 2006, 5:57:16 AM10/30/06

to

There is a proof in Advanced Calculus by Angus Taylor (later editions

added an author ) ,p220 which require the first partials only to exist

in a neighborhood of (a,b) (no continuity assumed) and the second

partials exist in a neighborhood and continuous at (a,b) (as you said).

There is a second theorem stated which require instead that the first

partials exist in a neighborhood of (a,b) and both be differentiable (

functions of (x,y)) at the point (a,b) .In this theorem no hypothesis

is required for 2nd Partials.Regards,Stuart M Newberger

Oct 30, 2006, 7:31:44 AM10/30/06

to

smn wrote:

>

> There is a proof in Advanced Calculus by Angus Taylor (later editions

> added an author ) ,p220 which require the first partials only to exist

> in a neighborhood of (a,b) (no continuity assumed) and the second

> partials exist in a neighborhood and continuous at (a,b) (as you said).

>

If second partials exist in a neighbourhood of (a,b) and are continous

at (a,b), this automatically implies that the first partials are

continuous in same neighbourhood of (a,b) - right? So I don't see why

this book states the theorem this way.

>

> There is a second theorem stated which require instead that the first

> partials exist in a neighborhood of (a,b) and both be differentiable (

> functions of (x,y)) at the point (a,b) .In this theorem no hypothesis

> is required for 2nd Partials.Regards,Stuart M Newberger

But first partials being differentiable functions at (a,b) does say a

lot about the 2nd partials.

Oct 30, 2006, 5:35:06 PM10/30/06

to

G Patel wrote:

> smn wrote:

> >

> > There is a proof in Advanced Calculus by Angus Taylor (later editions

> > added an author ) ,p220 which require the first partials only to exist

> > in a neighborhood of (a,b) (no continuity assumed) and the second

> > partials exist in a neighborhood and continuous at (a,b) (as you said).

> >

>

> at (a,b), this automatically implies that the first partials are

> continuous in same neighbourhood of (a,b) - right? So I don't see why

> this book states the theorem this way.

NO! Existence of partial derivatives of a function says nothing about

continuity of the function since partial derivatives only involve

function values on horizontal and vertical lines.

Also in the case at hand the existence only of the mixed 2nd partials

in a neighborhood is hypothesized in the first theorem,nothing needed

about f_xx and f_yy.

> > There is a second theorem stated which require instead that the first

> > partials exist in a neighborhood of (a,b) and both be differentiable (

> > functions of (x,y)) at the point (a,b) .In this theorem no hypothesis

> > is required for 2nd Partials.Regards,Stuart M Newberger

>

> But first partials being differentiable functions at (a,b) does say a

> lot about the 2nd partials.

It implies existence of all the second partials at (a,b) but says

nothing about the existence of any 2nd partials away from (a,b) .The 2

theorems are different,neither is stronger.

Another reference for all this (perhaps better is Apostol ,Tom

-Mathematical Analysis 2nd edition.This is an excellant reference for

alot of things.Regards,smn

Oct 30, 2006, 7:23:18 PM10/30/06

to

G Patel wrote:

>> To guarantee f_xy (a,b) = f_yx (a,b), do we require that

>> f_xy, f_yx be continuous at (a,b) or a stronger requirement

>> ( continuity in some neighbourhood of (a,b) )?

Robert Israel wrote:

> In Adams, "Calculus, A Complete Course" (third edition), sec.

> 13.4, the requirement is that f_{xy} and f_{yx} are

> continuous at (a,b), while f, f_x and f_y are continuous in

> a neighbourhood of (a,b).

In the 1940's the Russian mathematician Georgii Pavlovich

Tolstov [Tolstoff] published several papers on this topic.

One of his examples was a function f(x,y) with continuous

first order partial derivatives everywhere and such that

both mixed second order partial derivatives exist everywhere,

but the two mixed second order partial derivatives differ

on a set of positive measure. Another example was a function

g(x,y) with continuous first order partial derivatives

such that the mixed second order partial derivatives

exist and differ almost everywhere (i.e. on the complement

of a set of measure zero). The difference between the first

and second examples is that in his second example, at least

one of the mixed partial derivatives fails to exist at

some of the points that belong to the exceptional set of

measure zero.

In the positive direction, Tolstov proved that if all the

partial derivatives up to and including order n exist

throughout an open connected set, then: (a) each mixed partial

derivative of order less than n is independent of the order

of differentiation at each point in the domain; (b) each mixed

partial derivative of order n is independent of the order

of differentiation at almost every point in the domain;

(c) each partial derivative of order up to and including n

is a Baire one function. These results hold for functions

having an arbitrary finite number of variables.

I'm sure these results must have been improved in several

different directions by now (pun intended), but I don't

know any of the literature on this topic. The results I

gave for Tolstov are given on p. 209 of the following

paper.

A. N. Kolmogorov, V. Ya. Kozlov, D. E. Men'shov, and I. Ya

Verchenko, "Georgii Pavlovich Tolstov (on his sixtieth

birthday)", Russian Mathematical Surveys 27 #1 (1972), 207-217.

Dave L. Renfro

Oct 30, 2006, 9:43:59 PM10/30/06

to

I had some confusions and your post forced me to clear up most of them.

I was kind of reading the theorem as all 2nd partials (not just mixed)

being continuous at (a,b).

So let me try and sum up the first theorem:

"If first partials of f are defined in some neighbourhood of (a,b) and

mixed second partials of f are continuous at (a,b), then these mixed

second partials are equal at (a,b)"

**Is there a requirement for f to be defined on some neighbourhood of

(a,b)? The version of this theorem I've encountered in school (last

sem.) also inluded this as a hypothesis.

Since f_x and f_y (only one of these needeed I believed) are defined on

some neighbourhood of (a,b), this implies that f is defined on some

neighbourhood of (a,b). Correct? Therefore I can cross the

redundant hypothesis out?

I originally had a gut feeling that this hypothesis was unecessary,

then I think I came up with a proof (I think):

f_x exists in neighbourhood of (a,b) implies that

f( p + h, q ) - f( p , q)

f_x( p,q ) = lim{h->0} ------------------------------------

h

exists for all (p,q) in some nighbourhood of (a,b)

And this limit only exists if the numerator approaches 0, which

requires the subtrahend of numerator, namely f(p,q) , exists [this can

be made more precise].

Thus f exists in some neighbourhood of (a,b).

ACTUALLY, by the same reasoning, can't we even cross out the hypothesis

dealing with f_x, f_y?

Because, f_xy being continuous at (a,b) implies:

lim{(x,y) -> (a,b)} f_xy (x,y) = f_xy (a,b)

which requires that f_xy be defined in a neighbourhood of (a,b). Which

then implies that f_x exists in some neighbourhood of (a,b) by similar

proof as above.

Similarly, continuity of f_yx at (a,b) => f_y exists in some

neighbourhood of (a,b)

So can't the theorem by stripped down to:

"If f has mixed second partials that are continuous at (a,b), then

these partials at (a,b) are equal"

Good?

And, thank you to everyone who has helped me so far {Robert Israel,

smn, Dave Renfro}.

Oct 30, 2006, 11:42:42 PM10/30/06

to

Yes with the classical assumption that lim means limit over a full

neighborhood in R^2 .One can also define limits thru the domain of a

function even if the domain is not a full neighborhood (just that the

point where the limit is taken must be a limit point of this domain) so

authors add the redundent hypothesis for clarity.Regards,smn

Nov 25, 2006, 2:46:07 PM11/25/06

to

Dave L. Renfro wrote (in part):

> I'm sure these results must have been improved in several

> different directions by now (pun intended), but I don't

> know any of the literature on this topic. The results I

> gave for Tolstov are given on p. 209 of the following

> paper.

This was in reference to results I posted about some work

the Russian mathematician G. P. Tolstov did with mixed

partial derivatives in the 1940's.

It turns out that I do have some literature references on

this topic -- a few papers, and copies of Mathematical Reviews

for many more papers. However, essentially all of the papers

and reviews I have seem to deal with another result that

Tolstov proved in his 1949 paper, namely if f(x,y) is separately

continuous in each variable, then f(x,y) is determined by

the values it takes on a dense subset of its domain.

Nonetheless, my records show that an English translation

of Tolstov's paper exists (which I apparently don't have

a copy of).

Tolstov, "On partial derivatives" (Russian), Izvestiya Akad.

Nauk SSSR. Ser. Mat. 13 (1949), 425-446.

[MR 11,167b; Zbl 38.04003]

Tolstov, "On partial derivatives", American Mathematical

Society Translation 1952 (1952), no. 69, 30 pages.

[MR 13,926a]

Tolstov, "On partial derivatives", in "Translations, Series 1,

Volume 10: Functional Analysis and Measure Theory", American

Mathematical Society, 1962. [MR 38 #1985]

Dave L. Renfro

Dec 21, 2006, 12:29:27 PM12/21/06

to

Dave L. Renfro wrote:

[snip G. P. Tolstov references on partial derivative commutativity]

I came across something that I thought would be useful

to archive in this thread.

What follows is from The American Mathematical Monthly

67 #8 (October 1960), 813-814.

----------------------------------------------------------

Discontinuous Function with Partial Derivatives Everywhere

4876 [1959, 921]. Proposed by Naoki Kimura, University of Washington

If a real valued function f(x,y) of two real variables

possesses all of its partial derivatives

( del^(m+n) )f(x,y) / (del x^m)(del y^n)

at every point, is it necessarily continuous?

Solution by John Burr, University of New England, Australia.

The following example shows that the function need not be

continuous. The function f(x,y) = exp(-x^2/y^2 - y^2/x^2)

(xy /= 0), f(0,y) = f(x,0) = 0, is discontinuous at (0,0),

since when t --> 0, f(t,t) --> exp(-2) /= f(0,0). Suppose

that it has been proved that a particular partial derivative

phi(x,y) has the properties (i) phi(0,y) = phi(x,0) = 0,

(ii) if xy /= 0 then phi(x,y) = R(x,y)*f(x,y) where R(x,y)

is a rational [function] with denominator of the form

(x^p)(y^q). Then, by (i), (phi_y)(0,y) = (phi_x)(x,0) = 0;

by (ii) (phi_x)(0,y) = 0 when y /= 0, since when x --> 0,

f(x,y) --> 0 more rapidly than any power of x; similarly

(phi_y)(x,0) = 0. Hence (phi_x)(x,y) and (phi_y)(x,y) both

have the property (i), and it is clear from (ii) that

they also have the property (ii). Since f(x,y) has these

properties, it follows by induction that all the partial

derivatives have them, and that these derivatives exist

at every point.

It may be noted that this function satisfies conditions

more stringent than those prescribed in the problem;

namely the additional conditions that the mixed partial

derivatives all exist, and are independent of the order

in which the differentiations are performed.

Also solved by George Piranian, and the proposer.

Editorial Note. Piranian asks: What are the point

sets E in the xy-plane for which there exists a function

(f(x,y); E) that is discontinuous everywhere on E and

possesses all partial derivatives everywhere in the plane?

For example, can E be everywhere dense? Can it be the

entire plane?

----------------------------------------------------------

Dave L. Renfro

Feb 16, 2015, 10:47:15 AM2/16/15

to

A question: Suppose f : R^2 --> R is a function such that

D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

and

D_1 D_2 f is continuous at (0,0).

Does it follow that (D_2 D_1 f)(0,0) exists and is equal

to (D_1 D_2 f)(0,0) ?

Thanks!

D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

and

D_1 D_2 f is continuous at (0,0).

Does it follow that (D_2 D_1 f)(0,0) exists and is equal

to (D_1 D_2 f)(0,0) ?

Thanks!

Feb 16, 2015, 11:14:50 AM2/16/15

to

adam...@umn.edu wrote:

> A question: Suppose f : R^2 --> R is a function such that

> D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

> and

> D_1 D_2 f is continuous at (0,0).

Don't you need D_2 D_1 f is continuous at (0,0) as well? With that
> A question: Suppose f : R^2 --> R is a function such that

> D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

> and

> D_1 D_2 f is continuous at (0,0).

extra assumption D_1 and D_2 do commute. Consider

double_difference f = (f(x+h,y+h)-f(x+h,y))-(f(x,y+h)-f(x,y))

and compute lim (double_difference f/h^2) as h->0 in two ways to show

that it equals both D_1 D_2 f(x,y) and D_2 D_1 f(x,y).

> Does it follow that (D_2 D_1 f)(0,0) exists and is equal

> to (D_1 D_2 f)(0,0) ?

>

> Thanks!

>

Sorrow in all lands, and grievous omens.

Great anger in the dragon of the hills,

And silent now the earth's green oracles

That will not speak again of innocence.

David Sutton -- Geomancies

Feb 16, 2015, 11:35:13 AM2/16/15

to

Justin Thyme wrote:

> adam...@umn.edu wrote:

>> A question: Suppose f : R^2 --> R is a function such that

>> D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

>> and

>> D_1 D_2 f is continuous at (0,0).

>

> Don't you need D_2 D_1 f is continuous at (0,0) as well? With that

> extra assumption D_1 and D_2 do commute. Consider

>

> double_difference f = (f(x+h,y+h)-f(x+h,y))-(f(x,y+h)-f(x,y))

>

> and compute lim (double_difference f/h^2) as h->0 in two ways to show

> that it equals both D_1 D_2 f(x,y) and D_2 D_1 f(x,y).

>

>> Does it follow that (D_2 D_1 f)(0,0) exists and is equal

>> to (D_1 D_2 f)(0,0) ?

>>

>> Thanks!

Note that "D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2"
> adam...@umn.edu wrote:

>> A question: Suppose f : R^2 --> R is a function such that

>> D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2

>> and

>> D_1 D_2 f is continuous at (0,0).

>

> Don't you need D_2 D_1 f is continuous at (0,0) as well? With that

> extra assumption D_1 and D_2 do commute. Consider

>

> double_difference f = (f(x+h,y+h)-f(x+h,y))-(f(x,y+h)-f(x,y))

>

> and compute lim (double_difference f/h^2) as h->0 in two ways to show

> that it equals both D_1 D_2 f(x,y) and D_2 D_1 f(x,y).

>

>> Does it follow that (D_2 D_1 f)(0,0) exists and is equal

>> to (D_1 D_2 f)(0,0) ?

>>

>> Thanks!

is unnecessarily restrictive, being defined at (0,0) is enough.

Feb 16, 2015, 1:13:56 PM2/16/15

to

On Monday, February 16, 2015 at 10:14:50 AM UTC-6, Justin Thyme wrote:

> Don't you need D_2 D_1 f is continuous at (0,0) as well? With that

> extra assumption D_1 and D_2 do commute. Consider

I simply don't know whether continuity of D_2 D_1 f at (0,0) is needed. That's exactly the question. Can we drop that hypothesis?
> Don't you need D_2 D_1 f is continuous at (0,0) as well? With that

> extra assumption D_1 and D_2 do commute. Consider

Thanks!

Feb 16, 2015, 1:15:47 PM2/16/15

to

On Monday, February 16, 2015 at 10:35:13 AM UTC-6, Justin Thyme wrote:

> Note that "D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2"

> is unnecessarily restrictive, being defined at (0,0) is enough.

Are you assuming that D_2 D_1 f is continuous at (0,0)? I want to drop that, if possible. I do want to assume that D_1 D_2 f is continuous at (0,0), though.
> Note that "D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2"

> is unnecessarily restrictive, being defined at (0,0) is enough.

- Scot

Feb 16, 2015, 2:05:31 PM2/16/15

to

adam...@umn.edu wrote:

> On Monday, February 16, 2015 at 10:35:13 AM UTC-6, Justin Thyme wrote:

>> Note that "D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2"

>> is unnecessarily restrictive, being defined at (0,0) is enough.

>

> Are you assuming that D_2 D_1 f is continuous at (0,0)?

Yes.
> On Monday, February 16, 2015 at 10:35:13 AM UTC-6, Justin Thyme wrote:

>> Note that "D_1 f, D_2 f, D_1 D_2 f are all three defined on all of R^2"

>> is unnecessarily restrictive, being defined at (0,0) is enough.

>

> Are you assuming that D_2 D_1 f is continuous at (0,0)?

> I want to drop that, if possible. I do want to assume that D_1 D_2 f is continuous at (0,0), though.

I know is that if D_1 D_2 f and D_2 D_1 f are both continuous at the

point in question, then D_1 and D_2 commute.

Feb 19, 2015, 10:05:52 AM2/19/15

to

defined in a neighbourhood of (0,0) and D_1 D_2 f is continuous at (0,0).

We need the following:

Fact1: if F' is bounded on [a, b] then F(b) - F(a) = \int_a^b F'(s) ds

Fact2: if g(x) = \int_0^x h(s) ds and g is differentiable at 0,

then lim inf_{s -> 0+} h(s) \leq g'(0) \leq lim sup_{s -> 0+} h(s)

Now:

1) D_1 D_2 f is continuous at (0,0) => D_1 D_2 f is bounded in

a a neighbourhood of (0,0)

2) D_1 D_2 f bounded in a rectangle [-a, a]x[-a, a] =>

(D_2f)(x, y) - (D_2f)(0, y) = \int_0^x (D_1 D_2 f)(s, y)ds

for x, y in [-a, a] =>

(f(x, y) - f(0, y)) - (f(x, 0) - f(0, 0)) =

\int_0^y\int_0^x (D_1 D_2 f)(s, t)dsdt

for x, y in [a, -a] =>

f(x, y) = \int_0^y\int_0^x (D_1 D_2 f)(s, t)dsdt +

f(x, 0) + f(0, y) - f(0, 0)

(Use twice the Fact1).

3) Rearranging 2) we get:

(f(x, y) - f(x, 0)) = \int_0^x\int_0^y(D_1 D_2 f)(s, t)dsdt +

(f(0, y) - f(0, 0))

4)

lim inf_{x -> 0+} \int_0^y (D_1 D_2 f)(x, t)dsdt \leq

(D_1f)(0, y) - (D_1f)(0, 0) \leq

lim sup_{x -> 0+} \int_0^y (D_1 D_2 f)(x, t)dsdt

5) Since D_1 D_2 f is continouos at 0 we have:

lim_{y -> 0} y^{-1}lim inf_{x -> 0+} \int_0^y (D_1 D_2 f)(x, t)dsdt =

lim_{y -> 0} y^{-1}lim sup_{x -> 0+} \int_0^y (D_1 D_2 f)(x, t)dsdt

= (D_1 D_2 f)(0, 0)

6) By pinching theorem

(D_2D_1f)(0, 0) = lim_{y -> 0} y^{-1}((D_1f)(0, y) - (D_1f)(0, 0)) =

(D_1 D_2 f)(0, 0)

--

Waldek Hebisch

heb...@math.uni.wroc.pl

Feb 19, 2015, 4:54:47 PM2/19/15

to

Just a few remarks. Let me know if I make any mistakes.

Fact 1 is a Lebesgue integration version of the Fundamental

Theorem of Calculus. It isn't obvious, but follows from

Th 8.21, p. 179, of Rudin's "Real and Complex Analysis".

I'm sure there are many other references.

Fact 2 looks easy to prove directly from definitions.

A small typo: In Parts 4) and 5), "ds dt" should be "dt".

By Fact 2, I think you can go directly from

(f(x, y) - f(0, y)) - (f(x, 0) - f(0, 0)) =

\int_0^y \int_0^x (D_1 D_2 f)(s,t) ds dt,
for all x, y in [a, -a]

to Part 4). This skips the last half of Part 2) and all of Part 3).

Part 5) looks easy to prove directly from definitions.

I think all the one-sided limsups at 0+

could be replaced by two-sided limsups at 0.

I think all the one-sided liminfs at 0+

could be replaced by two-sided liminfs at 0.

------

Let's now *only* assume that D_1 D_2 f is continuous at (0,0)

and that (D_1 f)(0,0) exists.

By boundedness of D_1 D_2 f near (0,0), by the Mean Value Theorem,

and by the Dominated Convergence Theorem, we see:

For all y sufficiently near 0, the limit, as h --> 0, of

\int_0^y [1/h] [ ((D_2 f)(h,t)) - ((D_2 f)(0,t)) ] dt

exists, and is equal to \int_0^y (D_1 D_2 f)(0,t) dt.

By Fact 1, we see, for all (h,y) sufficiently near (0,0), that

\int_0^y [1/h] [ ((D_2 f)(h,t)) - ((D_2 f)(0,t)) ] dt

is equal to

[1/h] [ f(h,y) - f(0,y) ]

- [1/h] [ f(h,0) - f(0,0) ].

Consequently, for all y sufficiently near 0,

the limit, as h --> 0, of

[1/h] [ f(h,y) - f(0,y) ]

- [1/h] [ f(h,0) - f(0,0) ]

exists, and is equal to \int_0^y (D_1 D_2 f)(0,t) dt.

By definition of D_1, the limit, as h --> 0, of

[1/h] [ f(h,0) - f(0,0) ]

exists, and is equal to (D_1 f)(0,0).

Adding, we find, for all y sufficiently near 0, that

the limit, as h --> 0, of

[1/h] [ f(h,y) - f(0,y) ]

exists and is equal to

[ \int_0^y (D_1 D_2 f)(0,t) dt ] + [ (D_1 f)(0,0) ].
Thus, for all y sufficiently near 0, (D_1 f)(0,y) exists.

Once we know this, we can apply the argument from the previous post

(thanks, again!) to show that

(D_2 D_1 f)(0,0) exists, and is equal to

(D_1 D_2 f)(0,0).

-------

This yields, the following, which is, I think,

from a certain point of view, sharp:

Thm: Let f be a function defined on a subset of R^2.

Suppose (D_1 f)(0,0) exists.

Suppose D_1 D_2 f is continuous at (0,0).

Then (D_2 D_1 f)(0,0) exists, and is equal to

(D_1 D_2 f)(0,0).

Anyone see any mistake?

Is there a reference for this? A simpler proof?

Feb 19, 2015, 11:32:25 PM2/19/15

to

> By Fact 2, I think you can go directly from

> (f(x, y) - f(0, y)) - (f(x, 0) - f(0, 0)) =

> \int_0^y \int_0^x (D_1 D_2 f)(s,t) ds dt,

> for all x, y in [a, -a]

> to Part 4). This skips the last half of Part 2) and all of Part 3).

not write it explicitely but logically necessary...

lim_{h->0} [1/h]\int_0^h g(t) dt = g(0)

when g is continuous at 0 and apply this to

[1/h]((D_1 f)(0, h) - (D_1 f)(0, 0)) = [1/h]\int_0^h (D_1 D_2 f)(0,t) dt

> -------

>

> This yields, the following, which is, I think,

> from a certain point of view, sharp:

>

> Thm: Let f be a function defined on a subset of R^2.

> Suppose (D_1 f)(0,0) exists.

> Suppose D_1 D_2 f is continuous at (0,0).

> Then (D_2 D_1 f)(0,0) exists, and is equal to

> (D_1 D_2 f)(0,0).

>

> Anyone see any mistake?

>

> Is there a reference for this? A simpler proof?

Waldek Hebisch

heb...@math.uni.wroc.pl

Feb 20, 2015, 5:08:43 PM2/20/15

to

> Actually, to finish this argument we need only to know that

> lim_{h->0} [1/h]\int_0^h g(t) dt = g(0)

> when g is continuous at 0 and apply this to

> [1/h]((D_1 f)(0, h) - (D_1 f)(0, 0)) = [1/h]\int_0^h (D_1 D_2 f)(0,t) dt

[1/h]((D_1 f)(0, h) - (D_1 f)(0, 0)) = [1/h]\int_0^h (D_1 D_2 f)(0,t) dt

?
I feel as if your original argument is needed to finish the proof, but it'd be great if there were something simpler. (Perhaps I'm misunderstanding what you're saying -- if so, sorry.)

- Scot

Feb 21, 2015, 11:03:35 AM2/21/15

to

(D_1 f)(0,h)

for h sufficiently near 0, but now I see: The same argument also gives, for small y, a formula for

(D_1 f)(0,y)

and, if, in that formula, you replace y by h, subtract (D_1 f)(0,0) and divide by h, you get

[1/h][((D_1 f)(0,h)) - ((D_1 f)(0,0))] = [1/h]\int_0^h (D_1 D_2 f)(0,t) dt

for h sufficiently near 0. Now just let h-->0.

So this uses some graduate level mathematics, but it does give the nice result:

Existence of (D_1 f)(0,0) and

continuity at (0,0) of (D_1 D_2 f)(0,0)

implies:

(D_2 D_1 f)(0,0) exists and is equal to (D_2 D_1 f)(0,0).

Still wondering about a reference and/or an undergraduate level argument for the same result.

Apr 3, 2015, 3:39:55 PM4/3/15

to

> Still wondering about a reference and/or an undergraduate level argument for the same result.

Commutativity of derivatives (under the assumption that the function is twice differentiable, appropriately defined) appears as Cor (partComm) at the very end of http://www.math.umn.edu/~adams005/MATH4604/differentiability.pdf . This writeup is appropriate for an undergraduate "Advanced Calculus" course.
May 10, 2016, 3:32:03 PM5/10/16

to

Essentially the same result appears in Charles Pugh's book "Real Mathematical Analysis". See Theorem 16 in Chapter 5 on p. 292 of the 2nd edition.

Message has been deleted

Apr 3, 2018, 12:47:08 PM4/3/18

to

On Saturday, April 1, 2017 at 4:14:53 PM UTC-5, adam...@umn.edu wrote:

> > Also solved by George Piranian, and the proposer.

> >

> > Editorial Note. Piranian asks: What are the point

> > sets E in the xy-plane for which there exists a function

> > (f(x,y); E) that is discontinuous everywhere on E and

> > possesses all partial derivatives everywhere in the plane?

> > For example, can E be everywhere dense? Can it be the

> > entire plane?

>

> I have put my thoughts about this online. I would appreciate any comments or corrections (even typographical errors). See "Discussion about infinite directional differentiability" which begins near the bottom of p. 1 of
> > Also solved by George Piranian, and the proposer.

> >

> > Editorial Note. Piranian asks: What are the point

> > sets E in the xy-plane for which there exists a function

> > (f(x,y); E) that is discontinuous everywhere on E and

> > possesses all partial derivatives everywhere in the plane?

> > For example, can E be everywhere dense? Can it be the

> > entire plane?

>

>

> http://www-users.math.umn.edu/~adams005/MATH4604/Archive/MATH4604Spring2017/ClassNotes/notesW10D1.pdf

>

> The proof of "Fact 0.1" therein seems clear to me, although mistakes can always happen. I need to add more detail to some of the comments following that proof, but I think the statements are all correct. Again, comments/corrections are very much appreciated. I'll try to add more detail during the summer.

>

> Thanks.

>

> - Scot Adams

> ad...@math.umn.edu

Reply all

Reply to author

Forward

0 new messages

Search

Clear search

Close search

Google apps

Main menu