4

Let $p$ be an odd prime number and $a$, $b$ and $c$ integers such that $p$ does not divide $a$, and does not divide $D=b^2-4ac$, and such that $$ax^2+bx+c\equiv0\pmod{p},$$ is solvable. Prove that in this case $ay^2 + by + c \equiv 0 \mod p^2$ is solvable too.

Well, the only thing that I was able to do is get $D$ from quadratic equation:

\begin{eqnarray*} ax^2 + bx + c &=& a(x^2 + bx/a) + c\\ & =& a(x^2 + 2\cdot b/2a + (b/2a)^2) - a(b/2a)^2 + c\\ &=&a(x + b/2a)^2 - \left(\frac{b^2 - 4ac}{4a}\right)\\ &=& a(x + b/2a)^2 - \left(\frac{D}{4a}\right) \end{eqnarray*} Since $D$ is not divisible by $p$, $D/4a$ is not divisible by $p$ too. So in order to make this equation true remainders of both $\frac{D}{4a}$ and $a(x + b/2a)^2$ should be equal to each other. Let us say that it is $r_1$.

Okay, then let $a(x + b/2a)^2$ be $p\cdot k + r_1$ and $D/4a = p \cdot m + r_1$. But can't move on with this proof. I tried several paths but all of them didn't lead me anywhere.

Your help will be much appreciated!

Servaes
  • 63,261
  • 7
  • 75
  • 163
math-traveler
  • 983
  • 3
  • 9
  • I would fix a solution $x$ modulo $p$, and try to solve for $k$ when substituting $y=kp+x$. – Berci Sep 27 '20 at 15:13

3 Answers3

2

Let $p$ be an odd prime number and $a$, $b$ and $c$ integers such that $p\nmid a$ and $p\nmid D$, where $D=b^2-4ac$. Let $x$ be an integer such that $$ax^2+bx+c\equiv0\pmod{p},\tag{1}$$ so $ax^2+bx+c=mp$ for some integer $m$.

Exercise: Show that $p$ does not divide $2ax+b$, because $p\nmid a$ and $p\nmid D$ and $p$ is odd.

If $2ax+b\equiv0\pmod{b}$ then because $p$ is odd and $p\nmid a$ we have $$x\equiv-b(2a)^{-1}\pmod{p}.$$ Plugging this into $(1)$ yields \begin{eqnarray*} 0&\equiv&a(-b(2a)^{-1})^2+b(-b(2a)^{-1})+c\\ &\equiv&(2a)^{-2}(ab^2-2ab^2+4a^2c)\\ &\equiv&(2a)^{-2}(-aD), \end{eqnarray*} contradicting the fact that $p\nmid a$ and $p\nmid D$ and $p$ is odd.

Now if $y$ is an integer such that $$ay^2+by+c\equiv0\pmod{p}^2,\tag{2}$$ then reducing mod $p$ shows that $y$ is a solution to $(1)$. So it makes sense to suppose that $y=x+kp$ for some integer $k$. Then \begin{eqnarray*} ay^2+by+c&=&a(x+kp)^2+b(x+kp)+c\\ &=&(ax^2+bx+c)+(2ax+b)kp+ak^2p^2\\ &=&((2ax+b)k+m)p+ak^2p^2 \end{eqnarray*} We see that $y$ satisfies $(2)$ if and only if $$(2ax+b)k+m\equiv0\pmod{p}.$$ Now because $2ax+b\not\equiv0\pmod{p}$ we are be done by taking $$k\equiv-m(2ax+b)^{-1}\pmod{p}.$$

Servaes
  • 63,261
  • 7
  • 75
  • 163
  • To be honest, I didn't come up with proof of fact that $2ax + b \not\equiv 0 \mod p$. Can you explain why? – math-traveler Sep 27 '20 at 21:20
  • @math-traveler This is a special case of Hensel's Lemma (Newton's method), as I explain in my answer. It's more powerful (and simpler!) to view it from this more general standpoint. – Bill Dubuque Sep 27 '20 at 21:50
  • @BillDubuque It is really awesome! But is there a bit simpler way? – math-traveler Sep 27 '20 at 22:13
  • @math-traveler I answered under my answer. – Bill Dubuque Sep 27 '20 at 22:33
  • @math-traveler I have added an explanation; simply plug in the corresponding value for $x$ into the quadratic to see that it contradicts the assumptions. – Servaes Sep 28 '20 at 07:15
  • If there are any mathematical reasons why you are downvoting my answer(s) it would be inteeesting to know (constructive critiques are always welcome). I find the downvote(s) quite puzzling. – Bill Dubuque Sep 28 '20 at 07:44
  • @BillDubuque I have provided you with a reason. I've explained this a few times before (though the last time is quite a while ago), and I won't repeat it at every answer I downvote for this reason. – Servaes Sep 28 '20 at 08:39
  • Anyone care to explain the downvote? – Servaes Sep 30 '20 at 07:37
2

Conceptually: $\bmod p\!:\, $ the discriminant $\,D\not\equiv 0\Rightarrow\, r\,$ is not a double root, so $\,\color{#c00}{f'(r)\not\equiv 0},\,$ so Newton's method (Hensel's Lemma) improves (lifts) a root approximation $\!\bmod p\,$ to $\!\bmod p^2,\,$ i..e

completing the square: $\,4af(x) = (2ax+b)^2 -\, D,\ \ \,D = b^2-4ac$

Therefore, $\bmod p\!:\ f(r)\equiv 0\,\Rightarrow\, (2ar\!+b)^2\equiv D\not\equiv 0\,\Rightarrow\,2ar+b\not\equiv 0$

i.e. $\,\color{#c00}{f'(r)\not\equiv 0}\,$ so Newton's method applies - lifting $\,r\,$ to a root $\bmod p^2$ as below (for $\,k,n\!=\!1)$.

Hensel's Lemma $ $ Suppose that $\,1\le n\le k\,$ are both integers and the polynomial $ \:f(x) \in \mathbb Z[x].\,$ If $\, f(r)\equiv 0\pmod{p^k}\,$ and $\,\color{#c00}{p\nmid f'(r)},\, $ then the root $\,r\,$ uniquely lifts to a root $\,\hat r \pmod{p^{k+n}},\,$ i.e. there exists $\,\hat r \equiv r\pmod{p^k}\,$ with $\,f(\hat r)\equiv 0\pmod{p^{k+n}}$ and $\,\hat r\,$ is unique $\bmod {p^{k+n}},\,$ viz.

$\qquad \begin{align}\bmod p^{k+n}\!:\ \ f(\hat r)\equiv 0\iff\hat r\,&\equiv\ r-p^k\left(\dfrac{f(r)/p^k}{\color{#c00}{f'(r)}}\bmod p^n \right)\\[.2em] & \equiv\ r \ -\ f(r) \left( \dfrac{1}{\color{#c00}{f'(r)}}\bmod p^n\right)\end{align}$

Proof $\ $ By Taylor, $ \ 0\equiv f(\hat r)\equiv f(r\!+qp^k) \equiv f(r) + f'(r)\, qp^k \,\pmod{\!p^{k+n}}\,$

$ \qquad \iff p^{k+n}\mid p^k\,(f(r)/p^k + f'(r)q)\iff p^n\mid f(r)/p^k + f'(r)\, q$

By $ \: \color{#c00}{p\nmid f'(r)}\:$ this has a unique solution $\ q \equiv -(f(r)/p^k)/ \color{#c00}{f'(r)}\,\pmod {p^n}$

Bill Dubuque
  • 272,048
  • @math-traveler wrote: "is there a bit simpler way?" It needs only the first two terms of a Taylor series $,f(r+t) = f(r) + f'(r), t \pmod{!t^2},$ for a polynomial $,f(t),,$ which is easily proved, e.g. by the Binomial Theorem, or as here or here. In your case you need it only for a quadratic $f(t) = at^2+bt+c,$ which is even simpler, e.g. simply verify by calculation that $,f(r+t) = f(r) + (2ar!+!b), t + 2r t^2 \equiv f(r)+f'(r),t\pmod{!t^2}.,$ [$,t=kp,$ in the proof]. – Bill Dubuque Sep 27 '20 at 23:39
  • As usual your answer is fine for a reader who already knows and understands the answer, i.e. someone who doesn't need it. It is nearly incomprehensible to anyone who's actually looking for an answer. – Servaes Sep 28 '20 at 08:38
  • @Servaes If you tell me what you think is "incomprehensible" then I'll be happy to elaborate. If you read it closely you will learn that it uses only very simple algebra and number theory. Another answer here uses much heavier machiner but you didn't downvote that - which surely also satisfies your critique. Hmm.... – Bill Dubuque Sep 28 '20 at 08:40
  • As suggested by OP's attempts in the question body, and their follow-up question to my answer, every single claim you make. Also, your first sentence seems incomplete; after "i.e. completing the square:" some expression follows, but you make no connection to the previous statements. I don't care to go into more detail, I have done so before (admittedly a while ago) to no avail. – Servaes Sep 28 '20 at 08:43
  • @Servaes Not helpful without much further detai. You donwvote answers which have "incomplete sentences"? The mind boggles... – Bill Dubuque Sep 28 '20 at 08:44
  • As a general piece of advice, I would suggest to consider why the question you are answering is being asked. If someone is even remotely comfortable relating derivatives at a point (mod $p$) to discriminants and to lifting roots in a single sentence, they certainly would not have asked such a question. – Servaes Sep 28 '20 at 08:46
  • Your last comment is very insincere. – Servaes Sep 28 '20 at 08:48
  • @Servaes The proof is entirely self-contained (with the first comment) and comprehensible even to a bright high-school student. It seems you have not read closely. The links are for perspective - the answer does not depend on them. Don't let the big words scare you - it's really quite simple. – Bill Dubuque Sep 28 '20 at 08:49
  • OP has difficulty with fairly basic algebraic manipulations. Your answer takes doing that for granted at every step, and much more. – Servaes Sep 28 '20 at 08:52
  • @Servaes I see no evidence of that. And even if so, they can always ask questions if anything is not clear. Remember, also, that answers are not only for the OP, but for all readers, and I'm sure we have plenty of readers who can benefit from this answer even if the OP cannot. – Bill Dubuque Sep 28 '20 at 08:56
  • @Servaes And you still didn't answer why you didn't downvote the other answer - which is a more extreme example of the things you critique. Axe to grind? – Bill Dubuque Sep 28 '20 at 09:00
  • We have had this discussion before, and I'm sure it will lead nowhere as before, so I will not bother to reply any further. You may have a point about the other answer though; I hadn't bothered to read it because at a glance it looked like it would be a lot of work. That is, it probably deserve a downvote but I couldn't be bothered (yet) to check whether it does. – Servaes Sep 28 '20 at 09:02
  • 1
    @Servaes I don't recall any such discussion. I think you should be more open-minded about those who teach differently than you do. For example, if I followed your logic I would downvote your answer because I think it is so low-level and non-conceptual that it leaves the OP little hope of grasping the essence of the matter. Yes, maybe they can follow the proof step-by-step to verify the proof - like a theorem proving machine would - but that doesn't lead to a true understanding of the essence of the matter. But I haven't voted (yet), because I have not had the time to read it carefully. – Bill Dubuque Sep 28 '20 at 09:06
  • @Servaes Some of the my best learning experiences were sparked by remarks from teachers that were slightly beyond the horizon of my knowledge level - which intrigued me and motiivated me to learn more about various topics (my teacher Gian-Carlo Rota was quite fond of such). Some of my posts aim to do similarly (not only for the OP). Of course that is much more difficult here than in a classroom - due to lacking much more context. But that doesn't mean that we should give up all hope of doing so. – Bill Dubuque Sep 28 '20 at 09:13
  • @ΑΘΩ You don't appear to me as a nagging hag at all. And I appreciate that you asked. I have given you some feedback/thoughts in the comments below your answer. – Servaes Sep 30 '20 at 08:27
  • 2
    The advantage of this method is that it straightforwardly constructs a solution. Also it isn't necessary to know the theory behind it - just using Newton's method adapted to modulo $p$ works. The thing can be found by taking a solution mod $p$ and trying to find a solution mod $p^2$. I don't see a problem with this answer which gives an illuminating insight. – Mark Bennet Sep 30 '20 at 13:35
  • If $p=2$, $f(r)\equiv 0 \pmod{2^k}\implies \hat r\equiv r+2^{k-1} \left ( \frac{f(r)}{2^k} \mod 2\right) \pmod{2^{k+1}}$. Or $\hat r\equiv r+\frac 12 (f(r) \mod 2^{k+1}) \pmod{2^{k+1}}$. However, 1) the fractions could look like modular fractions but they are not; 2) there are two incongruent solutions when $f(r)/2^k$ is odd. Your first line of the lifting formula contains a regular fraction as well as a modular fraction but fortunately the second line saved it (also since $p$ is odd it gives a unique solution). How would you present the case when $p=2$ in a similar, less confusing way? Thanks – Neat Math Feb 21 '22 at 14:20
  • @NeatMath Could you please give example(s) for specific $f(x)$ to make precise what you mean. – Bill Dubuque Feb 21 '22 at 19:41
  • https://math.stackexchange.com/questions/4387962/solve-polynomial-congruence-fx-equiv-a-pmod2n Thank you. – Neat Math Feb 21 '22 at 22:39
0

I happened to be playing with Hensel's lemma a year or so ago and discovered this by accident and typed it up at the time. I might as well paste it here now since maybe it gives a slightly different perspective. Everywhere $|\cdot|$ appears it refers to the p-adic absolute value. I formulated the problem only slightly.

For $p>2$, if we have $f(x)=ax^2+bx+c$ with p-adic integer coefficients with $|a|=1$, then if an $x$ exists such that $|f(x)|<|b^2-4ac|$ then there is a solution to $x$ in $\mathbb{Q}_p$.

Proof:

We start with Hensel's lifting criteria,

$$|f(x)|<|f'(x)|^2$$

Now expand and rewrite

$$|f'(x)^2| = |4ax+4abx+b^2| = |4a|\left|ax^2+bx+c +\frac{b^2-4ac}{4a}\right|$$

Our conditions $p>2$ and $|a|=1$ allow us to write $|4a|=1$ and so now we have,

$$|f(x)|< \left|f(x) +\frac{b^2-4ac}{4a}\right|$$

The ultrametric inequality can now be used in its strong form to force an equality because one is strictly larger than the other,

$$\left|\frac{b^2-4ac}{4a}\right| = \left|f(x)+\frac{b^2-4ac}{4a} - f(x)\right| = \max\left(\left|f(x)+\frac{b^2-4ac}{4a}\right| , |f(x)| \right) = \left|f(x) + \frac{b^2-4ac}{4a}\right| $$

That means our original condition is equivalent to,

$$|f(x)|< \left|\frac{b^2-4ac}{4a}\right|$$

Again $|4a|=1$ and so for quadratics that satisfy those conditions we have the equivalent case to check for solutions.

$$|f(x)|<|b^2-4ac|$$

Merosity
  • 2,489
  • 1
  • 8
  • 16