Wikipedia:Reference desk/Archives/Mathematics/2009 September 8

= September 8 =

set
can the infinite set r = {r1, r2, ... rn, ...} where $$r_{k+1}=\frac{k+3}{k+2+r_k}$$ and $$r_1=\frac{3}{2}$$ be described by some function $$f$$ where $$f(k)=r_k$$? —Preceding unsigned comment added by 92.8.193.46 (talk) 01:43, 8 September 2009 (UTC)

You are looking for a closed formula for the function, as the function itself can simply be defined as the mapping f:k-->rk.Julzes (talk) 01:48, 8 September 2009 (UTC)


 * Defining sk = 1/(rk-1) you get a much nicer recursive relationship, sk+1 = -(k+3)sk - 1, with s1 = 2. You can construct an explicit (but awful) formula for rk from that:
 * $$r_k = 1 + (-1)^{k-1}\left[(k+2)!\left(\frac{1}{3} + \sum_{n=2}^k{\frac{(-1)^n}{(n+2)!}}\right)\right]^{-1}$$
 * Rckrone (talk) 04:04, 8 September 2009 (UTC)


 * Trivially so: you have already provided an inductive definition of the function you desire. &mdash; Carl (CBM · talk) 04:12, 8 September 2009 (UTC)

can all sets of that form where $$r_k=r_k(r_{k-1}, r_{k-2}, ... r_{k-n+1}, r_{k-n})$$ be written as a function of $$k$$ in the same manner provided the first $$n$$ terms are given? —Preceding unsigned comment added by 92.8.193.46 (talk) 11:10, 8 September 2009 (UTC)


 * Clarify what you mean by 'of that form' in the more complex situation.Julzes (talk) 17:58, 8 September 2009 (UTC)


 * You may invent ad hoc notations signifying any solution you need. But the sequence (1, 2, 4, 16, 65536, ...) satisfying r k+1 = 2 r k, r 0 = 1 is not written in terms of elementary functions. Bo Jacoby (talk) 12:38, 8 September 2009 (UTC),

Does it help you to know that
 * $$r_k = 1+\frac{(-1)^{k-1}e}{\Gamma(k+3,-1)} \ ? $$

Where Γ is the incomplete Γ-function and e is Euler's constant? (N.B. the incomplete Γ-function is related to the Γ-function which generalises the factorial function to take a complex variable z with Re(z) > 0.)  Dr Dec  ( Talk )    15:06, 8 September 2009 (UTC)

Can the choice of numbers and the sequence be changed and still have a closed form like that, Declan? It looks like a special case.Julzes (talk) 02:21, 9 September 2009 (UTC)


 * I'm not sure, I just though evaluating Rckrone's expression for $$r_k$$ might help some how. I don't really understand the problem in its totality.  Dr Dec  ( Talk )    15:38, 10 September 2009 (UTC)

The sorgenfrey plane 2
I've done some more thinking to my question above (the sorgenfrey plane). Let A be all the rational points on the diagonal and B be a {r+sqrt2 | r is rational}. Obviously A and B are both dense and disjoint. Now if you recall how the proof of "regular+countable basis -> normal" goes, we can apply the following theorem: if A and B can be covered by two countable collections {Un} and {Vn} respectively, and that the closure of each Un is disjoint from B, and the closure of each Vn is disjoint from A, then A and B can be covered by disjoint open sets. The above A (rationals) and B (rationals+sqrt2) satisfy this property, hence they can be covered by disjoint open sets. I believe my intuition was right: it is possible to separate two disjoint dense sets, provided both are countable. I don't know what's up with the irrationals; there's more to them than meets the eye. The irrationals are a Baire space in the order topology, but rationals aren't, even though both are densely ordered. My solution is: there's something fishy with the irrationals that our intuition fails to capture. Anyone agree?

To Algebraist: I tried using your example but cant see how it'll work for p/q+1/(q^q). The page Liouville number just says |x-p/q|<1/(q^n), for any n. But I guess it's good enough, it must be that there's always a real number that can be approximated extremely closely by a rational p/q with a low denominator q, no matter how we rig the half open square.

To pma: you must have had one those those moments when two totally unrelated things give exactly the same result right? This and other "miracles" as I call them. I don't believe in miracles, so when I see something like this I tend to give it a deeper thought and try to bring more intuition into it. So please, if you accept logic without question that's fine by me, but things like Godel's incompleteness theorem and Skolem's paradox really puts my faith down. Breath of the Dying (talk) 02:18, 8 September 2009 (UTC)
 * I agree, in principle; but let's go back: the main thing was very simple: no matter how one fixes an Euclidean nbd Ir for each rational r, the set of all points that belong to infinitely many Ir,
 * $$\cap_{F\in\mathcal{F}}\cup_{r\in\Q\setminus F}I_r$$
 * is an intersection of countably many dense open subsets of $$\R$$, hence an uncountable set by the Baire's category argument ($$\mathcal{F}$$ being the family of all finite subsets of $$\Q$$). This somehow is counter-intuitive, since it seems it tell us that a very large part of the irrational numbers are stuffed somewhere very very closely to the rationals. For instance, the trascendental number that Algebr. provided, is approximable by rationals as fast as a lightning. On the contrary, one would expect that being so "close" to the rationals would leave very small room. I think in this case one's intuition is just too much attached to finite or smooth analogies. I'm confident that your doubts will soon disappear.  --pma (talk) 06:49, 8 September 2009 (UTC)


 * The intuition that "being so close to the rationals would leave very small room" is actually correct to the extent that the set of irrational numbers with approximation exponent greater than 2 has Lebesgue measure zero. — Emil J. 11:59, 8 September 2009 (UTC)
 * Good remark --so maybe what is counterintuitive is just any second category sets with null Lebesgue measure --pma (talk) 14:11, 8 September 2009 (UTC)

mathematics
what is the solution of (integration(differentiation of dx)) —Preceding unsigned comment added by 122.168.71.208 (talk) 10:30, 8 September 2009 (UTC)


 * Your question as posed does not entirely make sense, but you might find something useful in our article on the fundamental theorem of calculus. Gandalf61 (talk) 10:59, 8 September 2009 (UTC)


 * Your question is not specific enough to be meaningful, but I'd guess that you're taking integration and differentiation to be inverse operations, as in (square(square root of x)). So what do you think the solution to your question is?86.155.186.70 (talk) 11:06, 8 September 2009 (UTC)
 * (ec) If you're asking what the indefinite integral of the derivative of f(x) is, it's f(x). If you're asking what $$\int f(x) dx$$ means, see integral. Where we're calculating the integral over a range (see definite integral), the product of f(x) and dx (see differential) points towards the notion of considering the area under the curve of y=f(x) as the infinite summation of rectangles with height f(x) and width dx (see Riemann sum). In that way, the long-s could be intutitively related to the capital sigma notation. That's the brief overview of the generic notation. If what I've said doesn't help, you'll probably need to come back and be more specific. — Anonymous Dissident  Talk 11:10, 8 September 2009 (UTC)]
 * Isn't $$\int f'(x) \ dx = f(x) + c$$ for some constant c? So the antiderivative of the derivative is unique up to an addative constant.  Dr Dec  ( Talk )    15:13, 8 September 2009 (UTC)
 * I might be wrong here, but is +c really necessary when we're considering the antiderivative of the derivative of a function we already know? +c notes an ambiguity – but if we already know the original function, there is no ambiguity. Speaking from a purely theoretical and objective reference frame, the indefinite integral of the derivative of f(x) is f(x), without the +c – because c is already known and is part of the function, if it exists. Contrast this to the position of a person who is trying to integrate the derivative of a function they do not know. Naturally, I'm only assuming that we do know f(x) here. — Anonymous Dissident  Talk 02:23, 9 September 2009 (UTC)
 * Derivatives don't have to be integrable, in the Lebesgue theory - so it's not always the case that $$\int f'(x) \ dx $$ even makes sense. But in suitably nice circumstances, this does work. Tinfoilcat (talk) 16:42, 8 September 2009 (UTC)


 * A derivative of a differentiable function does not have to be Lebesgue integrable, but it is always Henstock–Kurzweil integrable, and its indefinite integral is indeed the original function up to an additive constant. — Emil J. 17:12, 8 September 2009 (UTC)
 * Although if a function f is only differentiable a.e. then $$\int f'(x) \ dx $$ may still be still meaningful in the Lebesgue sense, but not necessarily equal to the original function unless f is absolutely continuous. Rckrone (talk) 17:22, 8 September 2009 (UTC)

Piecewise functions
I am getting confused on where to shade...
 * $$h(x) := \begin{cases} 3, & \mbox{if } -1 \le x\le1  \\ 4,  & \mbox{if } 1<x\leq4

\\ x, & \mbox{if } 4\leq x. \end{cases} $$ and how do you graph y=x? Accdude92 (talk) (sign) 13:30, 8 September 2009 (UTC)
 * I suggest you start by graphing points, and see if you detect a trend. Ray  Talk 14:10, 8 September 2009 (UTC)
 * (ec) When you have a function outputting a constant, it's as easy as graphing a line like y=10. When the function outputs different constants for different ranges, which line you're drawing differs based on x value. For instance, for "3 if -1≤x≤1", you're graphing y=3 from x = -1 to x = 1. Another way of thinking about it is that you're graphing y=3 for every x value that meets the criterion "x is greater than or equal to -1 and less than or equal to 1". y=x is a line that cuts the origin diagonally – every x value is matched to every y value: (1,1), (2,2), (3,3) and so on. — Anonymous Dissident  Talk 14:19, 8 September 2009 (UTC)

Dihedral group of a polygon
Someone asked me to explain the dihedral group of a polygon to them. This is what I said: Dihedral group is the group of symmetries of a polygon by reflection and rotation. If n is odd then besides the n rotations there are n reflections around the n lines originating from the n vertices and bisecting each interior angle. If n is even then besides the n rotations there are n reflections around the perpendicular bisectors of the n edges. So in either case the dihedral group has order 2n.

What I am not sure is whether in case n is odd the reflections are taken around the angle bisectors or not. Can someone clarify this. I searched on the net but couldn't find anything. Thanks--Shahab (talk) 15:35, 8 September 2009 (UTC)


 * The dihedral group of a regular n sided polygon is generated by the reflections in the perpendicular bisectors of its sides and the bisectors of its interior angles. If n is odd each axis of symmetry connects the mid-point of one side to the opposite vertex. If n is even there are n/2 axes of symmetry connecting the mid-points of opposite sides and n/2 axes of symmetry connecting opposite vertices. In either case there are n axes of symmetry altogether and 2n elements in the symmetry group. Reflecting in one axis of symmetry followed by reflecting in another axis of symmetry produces a rotation through twice the angle between the axes. Gandalf61 (talk) 15:48, 8 September 2009 (UTC)


 * Thanks for the quick response. This information should also be in the Dihedral groups article.--Shahab (talk) 16:09, 8 September 2009 (UTC)

Metric completion of a field
This question should be elementary, but it's giving me the blues. Suppose we have a field, $$\left(K,+,\cdot\right)$$ endowed with a absolute value $$|\cdot|:K\rarr\mathbb{R}_+$$ that satisfies the three properties:


 * AV1: $$\left|x\right|=0 \iff x=0$$
 * AV2: $$\left|x\cdot y\right|=\left|x\right|\cdot\left|y\right|$$
 * AV3: $$\left|x+y\right|\leq\left|x\right|+\left|y\right|$$

Now, suppose $$K$$ is not metrically complete, i.e., there exist sequences $$\left(x_n\right)_{n=1}^\infty$$ that are Cauchy in $$K$$ with no limit in $$K$$. We define $$\overline{K}$$ to be the set of Cauchy sequences in $$K$$ modulo Cauchy sequences that converge to zero. Then, it is natural to define an absolute value, $$\left|\cdot\right|^\star$$ by $$\left|\left(x_n\right)_{n=1}^\infty\right|^\star = \lim_{n\rarr\infty}\left|x_n\right|$$. This is all very nice.

My question: how can I prove that $$\left|\cdot\right|^\star$$ is unique, i.e., that it is the only absolute value that we can define on $$\overline{K}$$ that is well-defined, satisfies the axioms AV1-AV3, and restricts to $$\left|\cdot\right|$$ on $$K$$ (i.e., $$\left|\iota\left(x\right)\right|^\star=\left|x\right|$$, where $$\iota\left(x\right)=\left(x\right)_{n=1}^\infty$$, the constant sequence)?

I think there's something subtle going on here that's eluding me. I've read Complete metric space, but it didn't quite ease my mind. The first paragraph under the header "Completion" asserts this claim, but doesn't say how it's proven. Can someone give me a hint as to what I'm missing? Thanks in advance. -GTBacchus(talk) 15:41, 8 September 2009 (UTC)


 * An absolute value defines a topology, and it is continuous with respect to this topology. Let $$|\cdot|'$$ be another absolute value on $$\overline{K}$$ which satisfies the properties above. If $$x=(x_n)_{n=0}^\infty$$ is a Cauchy sequence, then $$x=\lim_{n\to\infty}x_n$$ in $$\overline{K}$$, thus $$|x|'=\lim_{n\to\infty}|x_n|'=\lim_{n\to\infty}|x_n|=|x|^*$$. Now, the catch with this argument is that I am tacitly assuming that the topology induced by $$|\cdot|'$$ coincides with the natural topology on $$\overline{K}$$, and I don't quite see why this should be true in general. — Emil J. 16:59, 8 September 2009 (UTC)


 * Also, I've taken a look at Complete metric space as you suggest, and I do not see anything related to absolute values there. The generic extension method described there only implies that there exists a unique (uniformly) continuous function $$|\cdot|^*$$ which extends $$|\cdot|$$. It does not assert that $$|\cdot|^*$$ is an absolute value (though this should be easy enough to prove), and more importantly, it does not assert anything about uniqueness wrt the class of not-necessarily-continuous absolute values. Are you sure that the claim you are trying to prove is even true? — Emil J. 17:07, 8 September 2009 (UTC)
 * No, I'm not sure. The claim I really want to prove involves a non-Archimedean absolute value (featuring the strong triangle inequality), but I think it might be true in general. Actually, the assertion is simply that the absolute value extends "naturally" to the completion. Uniqueness would be bonus. This comes from a set of lecture notes, and may very well be an ill-posed question. If it's not true, I suppose there's a counter-example? I've already proven that $$|\cdot|^\star$$ is an absolute value satisfying the same AV1-AV3 axioms. -GTBacchus(talk) 18:51, 8 September 2009 (UTC)


 * (ec) What you need is the universal property of the completion $$\iota:K\to\overline K$$. For any metric space $$K$$ and any uniformly continuous map $$f:K\to M$$ to a complete metric space $$M$$ there exists a unique uniformly continuous $$\overline f:\overline K\to M $$ such that $$f=\iota\circ\overline f$$ (you may say that $$\overline f$$ is the unique u.continuous extension of $$f$$ over $$\overline K$$, if you like to think $$K$$ as a subset of $$\overline K$$).
 * This situation may be described in term of an adjunction, if you like the language of category theory. In any case the universal property allows to extend the absolute value and the algebraic operations from $$K$$ to $$\overline K$$ together with their identities. There are just some small detalis to work out; in case just ask. --pma (talk) 17:07, 8 September 2009 (UTC) PS: as to the unicity matter: precisely, $$\overline f$$ is the unique continuous extension, and it turns out to be uniformly continuous with the same modulus of continuity of $$f$$. --pma (talk) 17:13, 8 September 2009 (UTC)
 * Ok, so it's the unique continuous extension, but there may be other extensions that aren't continuous? This makes sense to me, and reminds me of a function that appears in Counterexamples in Analysis. There's an additive but non-homogeneous function on $$\mathbb{R}$$ that sends each number to the sum of the coefficients of its representation in terms of the basis of $$\mathbb{R}$$ as a vector space over $$\mathbb{Q}$$. That function, as I recall, is additive, but not remotely continuous. Its graph is dense in the plane, again IIRC. -GTBacchus(talk) 18:51, 8 September 2009 (UTC)
 * Sure. So, the whole matter is extremely simple; should something be still unclear to you, we can help you to clarify it. --pma (talk) 19:56, 8 September 2009 (UTC)


 * A good algebraic number theory book should handle this. I recall that Janusz's Algebraic Number Fields had a very readable and careful discussion, and Lang's Algebraic Number Theory probably covers this adequately as well.  Unfortunately I have neither of these books with me and I don't remember the answer!  I'll check Milne (available here) if I remember to do so later.
 * Regarding proving the claim, I got stuck at the same place as Emil, in proving that $$(x_n)_{n=0}^\infty = \lim_{n\to\infty}\iota(x_n)$$ in $$\overline K$$. I think that if you drop requirement 2 of an absolute value, that $$|xy| = |x| |y|$$, then you get a counterexample in $$\mathbb R / \mathbb Q$$ following the plan you mentioned.  I think the best bet for finding a counterexample without dropping requirement 2 is in the completion $$\mathbb C_p$$ of $$\hat {\mathbb Q}_p$$.  Eric.  216.27.191.178 (talk) 00:56, 9 September 2009 (UTC)

Here's a fairly trivial counterexample. Pick K such that Aut($\overline{K}$/K) is nontrivial (for example, we may take K = Q(i) with the usual complex absolute value, so that $\overline{K}$ = C is algebraically closed). Fix a nonidentical field automorphism σ of $\overline{K}$ which is identical on K, and define |x|' = |σ(x)|*. On the one hand, it is easy to see that |·|' is an absolute value extending |·|. On the other hand, since σ is nonidentical, and K is dense in $\overline{K}$, σ is discontinuous, and a fortiori it is not an isometry. Thus, there are x, y such that $$|x-y|^*\ne|\sigma(x)-\sigma(y)|^*=|\sigma(x-y)|^*=|x-y|'$$, i.e., |·|' ≠ |·|*. — Emil J. 10:45, 9 September 2009 (UTC)
 * Interesting. I'm looking at Algebraic Number Theory, by Jürgen Neukirch, page 124, from the section on "Completions". Ahem:
 * How does this jibe with your counterexample, which looks perfectly reasonable to me? (By the way, what's the LaTeX for the angle-shaped "hat"? I'm using \overline in place of it now.) -GTBacchus(talk) 15:49, 9 September 2009 (UTC)
 * How does this jibe with your counterexample, which looks perfectly reasonable to me? (By the way, what's the LaTeX for the angle-shaped "hat"? I'm using \overline in place of it now.) -GTBacchus(talk) 15:49, 9 September 2009 (UTC)


 * What he claims is uniqueness up to an isomorphism (preserving the field structure, absolute value, and elements of K). It's perfectly consistent with the example, in fact, his σ is my σ. A hat is \hat in TeX: $$\hat K$$. — Emil J. 16:09, 9 September 2009 (UTC)
 * Ah, so the isomorphism preserves algebraic relations, but non necessarily the topology, because the absolute value in continuous in one case but not in the other? Is that right? -GTBacchus(talk) 19:11, 9 September 2009 (UTC)

Let me summarize and state the whole thing as I see it: a list of simple facts:
 * 1) Let K be a AV field (field with absolute value, i.e. AV1 AV2 AV3 hold). Then, $$d_K(x,y):=|x-y|$$ makes $$K$$ a metric space, and a topological field (the field operation are continuous).
 * 2) There exists an AV field $$\hat K$$ such that $$K$$ is a sub AV field of $$\hat K$$ (i.e., it is a subfield of $$\hat K$$ and its AV is the restriction of the one of $$\hat K$$); moreover $$\hat K$$ is complete as a metric space and $$K$$ is dense in $$\hat K$$. Any such $$\hat K$$ is called an AV field completion of $$K$$.
 * 3) The AV field $$\hat K$$ may be constructed starting from a metric completion $$M,\hat{d}$$ of the metric space $$K,d_K$$; then the AV and the field operations of $$K$$ extend uniquely to $$M$$ in such a way to make it an AV field $$\hat K$$ whose distance $$d_{\hat K}$$ coincides with the distance $$\hat d$$.
 * 4) If $$\overline K$$ is another AV field completion of K, there exists a unique homeomorphism $$\sigma:\hat K\to \overline K$$ such that $$\sigma_{|K}=id_K$$; moreover, $$\sigma$$ is an isomorphism of AV fields.
 * 5) Remark. (EmilJ example) If K is an AV field and H is a super-field of K, then H may admit several AV that make it a completion of K (essentially due to the fact that if K is not complete there is a lot of trascendence in the completion extension).
 * Note that there is an analogous situation in the context of completion of normed linear spaces, including the remark (that becames much simpler: there is a lot of linear independence passing to the completion.) Do you agree? There are possibly typos here and there  .--pma (talk) 20:13, 9 September 2009 (UTC)
 * Is "homeomorphism" a typo? If not I'm confused about that since Emil J gave a convincincing argument that a map like that would be discontinuous, but maybe there's something I'm not getting about the topology involved here. Rckrone (talk) 23:19, 9 September 2009 (UTC)
 * From Merriam-Webster: "a function that is a one-to-one mapping between sets such that both the function and its inverse are continuous and that in topology exists for geometric figures which can be transformed one into the other by an elastic deformation".Alan (talk) 23:27, 9 September 2009 (UTC)
 * I think the question was not, "what does homeomorphism mean", but rather "is homeomorphism in this context a typo for homomorphism, in this case, of valued fields?" A homomorphism (no 'e') preserves algebraic structure, but is not necessarily continuous. Or, if it is necessarily continuous in the case of valued fields, that's a theorem. I'm also wondering about the continuity issue. -GTBacchus(talk) 05:00, 10 September 2009 (UTC)
 * Not a typo: I wrote "a unique homeomorphism $$\sigma$$" to emphasize that $$\sigma$$ is unique even in the class of all homeomorphism that extend the identity of K forgetting the algebraic structure; I could have even written "a unique continuous $$\sigma$$": this unique continuous element $$\sigma$$ is then automatically an isometry of metric spaces and an isomorphism of AV field (in particular, a homomorphism). On the contrary, if we try to characterize $$\sigma$$ algebrically, we meet difficulties, exactly because (in general) there are a lot of non-continuous automorphisms of $$\hat K$$ that extend the identity of K, as EmilJ's remark showed. The completion of an absolute valued field is primarily a completion of a metric space: the algebraic extension follows the metric extension. There are other mixed topological-algebraic structures, where a topological property is characterized in a purely algebraic way (first example that cames to my mind: a complex-valued homomorphism of a Banach algebra is necessarily continuous); but this is not the case. Still I do not understand why you seem to be unsatisfied with a unicity property stated in terms of the topological structure. Here the natural thing is to think in terms of completion of metric spaces: in this more general and simpler category, completion is unique up to a unique isometry (or if you like, up to a unique homeomorphism that is automatically an isometry). Once this is clear, take it to the context of valued fields: the same $$\sigma$$ will be automatically a valued field isomorphism also. --pma (talk) 06:23, 10 September 2009 (UTC)
 * I'm sorry if I'm being dense about this (no pun), and I thank you for your patience. In your numbered points above, here's my understanding:
 * Start w/ an AV-field K, on which the AV and field operations are continuous.
 * There exists some metric completion of K that is an AV-field (with AV and field ops continuous) that contains K as a dense sub-AV-field.
 * You can get such a completion by completing K as a metric space to obtain K-hat, and then uniquely extending the AV and operations - all of which presumably remain continuous.
 * If there's some K-bar (different from K-hat) that does what K-hat does (completes K metrically, and extends the AV and algebra so they're still continuous), then K-bar and K-hat are equivalent via a unique homeomorphism that respects all the structure.
 * Instead of starting with a metric completion, we could take a super-field, which extends the algebra first, and add topological structure to it afterwards. Then we can get away with a discontinuous AV. Even though it restricts to the old AV on K, and gets along w/ the field operations, it's not a metric completion of K, so point #4 doesn't apply. That's why #3 is a more natural way to do it.
 * Is that all correct? -GTBacchus(talk) 07:29, 10 September 2009 (UTC)
 * I understand where I was confused (thanks pma) and I think it's the same place, so I will give it a shot. Referring to Emil J's example, the map σ as a regular-old field isomorphism from $\overline{K}$ to itself is discontinuous with the topology induced by |.|*.  However as AV fields, $\overline{K}$ with metric |.|* is distinct from $\overline{K}$ with metric |.|'.  Call them $\overline{K}$1 and $\overline{K}$2.  σ as a map between AV fields from $\overline{K}$1 to $\overline{K}$2 is continuous since open sets according to the topology induced by |.|* get mapped to open sets according to the topology induced by |.|' (instead of using |.|* at both ends like before), which can be seen by the fact that |x|* = |σ(x)|'.  In other words σ preserves the AV as the transition is made from the one metric to the other. Rckrone (talk) 08:13, 10 September 2009 (UTC)
 * One way to think about it is that there are different ways to embed the metric completion of K into the superfield $\overline{K}$, two of them being $\overline{K}$1 and $\overline{K}$2, and σ is the AV field isomorphism that relates them. They both contain the same sets of elements, so merely as fields they are identical, but with the added structure of a metric they are distinct. Rckrone (talk) 08:38, 10 September 2009 (UTC)
 * agreed. Also, they have the same set of elements, and they are isomorphic, but not via the identity map.--pma (talk) 09:24, 10 September 2009 (UTC)
 * (ec) Correct. As to point 1, note that the continuity of the structure mappings (sum, product, multiplicative inverse, absolute value) follows as a consequence of the AV field axioms and the choice of the topology on K (the metric topology of the distance dK(x,y):=|x-y|), so it doesn't need to be assumed. In particular, the distance dK makes
 * $$+:K\times K\to K$$ a Lipschitz function;
 * $$\cdot:K\times K\to K$$ a locally Lipschitz function (it's Lip on every disk); also, $$x\to ax$$ is Lipschitz (of constant |a|) from K to K;
 * $$(\cdot)^{-1}:K\setminus\{0\}\to K$$ (multiplicative inverse) a locally Lipschitz function: precisely, Lipschitz on the complement of every nbd of 0 (just by AV2, because |1/x-1/y|=|x-y|/|xy|);
 * $$|\cdot|:K\to\R$$ (the abs value) a Lipschitz function, just by AV3.
 * As to point 3, if M is any completion of K as metric space (hence a plain metric space with no algebraic structure) the above structure maps (+, $$\cdot$$, inverse, $$|\cdot|$$), thanks to the universal property of metric completion, have unique continuous extensions respectively to maps:
 * $$\hat{+}:M\times M\to M$$ ;
 * $$\hat{\cdot}:M\times M\to M$$ ;
 * $$\hat{(\cdot)^{-1}}:M\setminus\{0\}\to M$$ ;
 * $$\hat{|\cdot|}:M\to\R$$ ,
 * that a priori are only continuous maps: but then all the AV field axioms holds for them by continuity, since they hold on the dense subset K. In particular the distance of M is, of course, just $$\hat{|x-y|}$$. Technical detail: notice that, while the continuous extension of the sum and of the AV is an immediate application of the universal property of metric completion, it is a bit less immediate for the product and the inverse, because they are only locally uniformly continuous; the latter is possibly not even defined on a complete domain, M\{0}. But that difficulty is very easily bypassed: first extend uniquely by continuity the product map (respectively, the multiplicative inverse map) in the closed sets of MxM (resp., of M\{0}) where it is Lipschitz, then glue together the extensions using the unicity property, and observe that the result of the gluing is still locally Lip, hence continuous.
 * As to point 4, we can make a stronger statement putting a comma pregnant with meaning in the middle (making it a non-restrictive clause): K-bar and K-hat are equivalent via a unique homeomorphism , which respects all the structure. That is, the identity map on K extends to a unique homeomorphism between them, and, this unique homeomorphism turns out to be an AV field isomorphism. Is that OK? (note: I used everywhere the term AV field just because I'm not sure about the exact term, maybe you know). I think the other people above also agreed with this picture: maybe they may correct or add something. --pma (talk) 09:18, 10 September 2009 (UTC)
 * That's awesome. Thank you so much. Thank you all, who have commented here. I'm, like, smarter now. :) I'll check back for any further notes or corrections here, but I really feel comfortable with what we've just walked through, including the technical points about the product and multiplicative inverse maps not being uniformly continuous on the whole field.
 * Neukirch, in the book I consulted, calls the structure we're talking about a "valued field"; I don't know how standard that terminology is. -GTBacchus(talk) 11:07, 10 September 2009 (UTC)

Poker and math
What problem, if any, do computers have in playing poker? Could a computer playing online poker win against human players? --Mr.K. (talk) 16:12, 8 September 2009 (UTC)
 * The combinatorial work necessary for optimal estimates of the strengths of others hands is hard, probably intractable. Coming up with good heuristics is also difficult. Dealing with the non-mechanistic issues (bluffing, modeling the other players) is also highly non-trivial. But yes, they can, and they do, regularly. My success at Sam Fox Strip Poker was limited (as was, to be honest, my motivation, due to the 320x200 greyscale images). --Stephan Schulz (talk) 16:34, 8 September 2009 (UTC)
 * Limit poker is much easier to do than no-limit, I believe. Have a look at Computer poker players Tinfoilcat (talk) 16:38, 8 September 2009 (UTC)
 * Yeah that's right. It's all about the pot odds (and more importantly the implied pot odds). If you need to put in x% of the pot to call and you expect to make a winning hand more than x% of the time then you've got good poot odds and you should call. (Obviously you might not make you hand and lose, but if you made the same call 1,000 times then you'd expect to make a profit). Like Stephan says: it's the bluffing that is hard to get around. You can make a really big bet so that the pot odds are so bad that the player wouldn't normally call, e.g. if you bet the twice the pot then the pot odds are will be so bad that you'd need to be able to make a winning hand 2 times in 3 (or better) to be able to call. But that's the bluff: I could have rubbush cards, but make a big bet so that the other player doesn't have the pot odds to call.  Dr Dec  ( Talk )    17:13, 8 September 2009 (UTC)
 * Yes, limit is easier for a computer. There are far fewer options to consider (you can fold, call or raise by a fix amount, rather than fold, call or raise by any amount of your choice [as long as it is over the minimum raise]). --Tango (talk) 20:49, 8 September 2009 (UTC)

Hermite polynomials and Hermite Interpolation
Does anyone know if there is a connection between Hermite polynomials and Hermite interpolation? ...Other than the association with Charles Hermite.

When interpolating a polynomial it can help to use a basis other than simple powers of x so I thought the Hermite polynomial might be useful for Hermite interpolation in the same way as the Newton basis polynomials are used in the Divided differences algorithm.

Neither article in Wikipedia states such a relationship but there is a link to an external article on Hermite interpolation from the article on Hermite polynomials... but that could just have been places in error.

Thanks,

Yaris678 (talk) 19:38, 8 September 2009 (UTC)


 * Well, I do not see any connection at all between Hermite polynomials and interpolation problems. --pma (talk) 20:10, 8 September 2009 (UTC)

Derivative of a function between matrices
Hi there refdesk!

I've been completing a couple of old exercises in analysis and I've come upon the following:

Deﬁne f: Mn → Mn by f(A) = A3. Show that f is differentiable everywhere, then ﬁnd its derivative.

Now I've realised I haven't a clue how to differentiate a function between matrices (Mn the set of real n x n matrices), nor even how the derivative is defined. Could anyone point me in the direction of an explanation or give me a very brief rundown of how to solve a question like this, which sounds like it's probably very simple once you know how!

Thanks a lot,

Typeships17 (talk) 20:38, 8 September 2009 (UTC)


 * Matrix calculus might help. --Tango (talk) 20:44, 8 September 2009 (UTC)


 * Doesn't the space of n × n matrices have a manifold structure? The space of n × n matrices would be an n2-dimensional manifold. So write a general matrix as M := (mi,j), and then compute M3 = (mi,j)3. Then you just need to check that each of the entries of M3 is differentiable as a function of (m1,1,…,mi,j,…,mn,n). This is clearly true since each entry with just be a homogeneous polynomial of degree 3.  Dr Dec  ( Talk )    20:50, 8 September 2009 (UTC)
 * Yes, it can be thought of as a family of n2 functions each defined on $$\R^{n^2}$$ and differentiated accordingly. --Tango (talk) 21:33, 8 September 2009 (UTC)
 * Thanks a lot guys, I'll try that link out now. In addition, I was wondering if anyone could explain to me why the derivative $$Df\vert_x(h)$$ of a function between Rn and Rm is orthogonal to x in not-too-technical terms: I can get my head around the concept where m=1 and you can look at level sets, but I'm not sure what geometrical argument you can apply when mapping to more than 1 dimension. Thanks very much, Typeships17 (talk) 23:39, 8 September 2009 (UTC)
 * sorry but it is absolutely not clear what you mean. What's x? What's $$Df\vert_x$$? --pma (talk) 09:48, 9 September 2009 (UTC)
 * I think I know what s/he might be getting at. If f : Rm → Rn is a submersion with x0 a regular point of f and y0 a regular value of f then the tangent space to {x | f(x) = y0} at x0 is given by the kernel of the differential of f evaluated at x0 (i.e. the Jacobian matrix evaluated at x0). In the case when n = 1 then the differential if just the gradient vector of f and its kernel is just the orthogonal space of the gradient vector.  Dr Dec  ( Talk )    22:37, 9 September 2009 (UTC)
 * I've tried to answer what I think the question was talking about. See here. Let me know if I've made any errors.  Dr Dec  ( Talk )    12:10, 10 September 2009 (UTC)

Consider the big laugh expansion: $$(A+H)^3=A^3+(HAA+AHA+AAH)+o(\|H\|)$$: your $$f$$ is differentiable at any $$A$$ and the Fréchet differential is $$Df(A)H=HAA+AHA+AAH$$. This holds for $$A$$ in any Banach algebra of course.--pma (talk) 07:04, 9 September 2009 (UTC)
 * thanks for the correction... ehm, I made an error counting to 3 ;-) --pma (talk) 13:51, 9 September 2009 (UTC)

Is this a hyperbola or...
Draw line segments connecting (a,0) to (0,1), (a-1,0) to (0,2) etc. up to (1,0) to (0,a). Is the curve that is approximated a hyperbola (well, half of one), or is it something else? Donald Hosek (talk) 21:57, 8 September 2009 (UTC)


 * EnvelopeAnim.gif You mean an envelope like this? It certainly looks like a hyperbola to me, although I can't guarantee it. --Tango (talk) 22:08, 8 September 2009 (UTC)
 * As the very article you linked states, it's a parabola, not a hyperbola. Algebraist 22:22, 8 September 2009 (UTC)


 * Don't stop when y gets down to 0; keep going. Likewise keep going in the opposite direction.  If you do that, you'll see that the curve does not have asymptotes.  So it cannot be a hyperbola. Michael Hardy (talk) 02:09, 9 September 2009 (UTC)


 * Indeed, to get an equilateral hyperbola one has to take all line segments whose X and Y intercepts have constant $$product$$, not $$sum$$. BTW, using a ruler, this fact gives a more precise way of drawing a hyperbola, compared with drawing by points. --pma (talk) 15:13, 9 September 2009 (UTC)

(ec) I think you're talking about the envelope of the family of lines given by
 * $$ y = \left(\frac{1+t}{t-a}\right)x + (1+t) \, $$

where, for each value of t with 0 ≤ t ≤ a – 1, we get the equation of a line. Well, let
 * $$ F((x,y),t):= y - \left(\frac{1+t}{t-a}\right)x - (1+t) \ . $$

Then for a fixed t0 the corresponding line is given by
 * $$ \{(x,y) \in \mathbb{R}^2 : F((x,y),t_0) = 0\} . $$

In this case, the envelope is given by solving F = Ft = 0 in terms of x and y. Doing this gives the envelope, or discriminant:
 * $$ \mathcal{D}_F(t) = \left(\frac{(a-t)^2}{a+1},\frac{(1+t)^2}{a+1} \right) . $$

This curve has the property that all of the lines are tangent to it. The question is now: is this curve part of a hyperbola?  Dr Dec  ( Talk )    22:25, 8 September 2009 (UTC)

One way to visualize that it's a parabola rather than a hyperbola without finding the explicit function is to consider what happens when you continue drawing lines past where you stopped: (0, 0) to (0, a+1), (-1, 0) to (0, a+2),... (-n, 0) to (0, a+1+n),... The slope of these lines go through the vertical and approach 1. You can also continue in the other direction to see a similar trend. Rckrone (talk) 22:49, 8 September 2009 (UTC)


 * Yes, that's a(n arc of) parabola. Let's call s>0 and t>0 respectively the X and the Y intercept of a segment in the family. If c is the constant value of s+t in the family of segments, following Declan's approach you'll get
 * $$x^{1/2}+y^{1/2}=c^{1/2}$$;
 * and if you rationalize and change coordinates after a π/4 rotation (i.e. x=(u+v)/2, y=(u-v)/2) you get a normal form of a parabola.
 * BTW, you can generalize this picture a bit. Given any $$\alpha>0$$ and $$ c>0$$, the envelope of the family of all segments within the first quadrant with x and y intercepts satisfying
 * $$s^{\alpha}+t^{\alpha}=c^{\alpha}$$
 * is a curve of equation
 * $$x^\frac{\alpha}{\alpha+1}+y^\frac{\alpha}{\alpha+1}=c^\frac{\alpha}{\alpha+1}$$
 * (yes, $$\alpha=2$$ gives 2/3).
 * You do not need any differential calculus; it's a plain consequence of the Hölder inequality in $$\R^2$$ (and for your particular case $$\alpha=1$$, it's just a Cauchy-Schwarz).--pma (talk) 08:28, 9 September 2009 (UTC)
 * hey!now that's all in the first link, too! ;-) --pma (talk) 13:54, 9 September 2009 (UTC)