Find the largest possible value of n , such that there exists n real numbers x 1 , . . . , x n which satisfy ∣ x i − x j ∣ > 1 0 0 1 ( 1 + x i x j ) for all values of i = j .
This section requires Javascript.
You are seeing this because something didn't load right. We suggest you, (a) try
refreshing the page, (b) enabling javascript if it is disabled on your browser and,
finally, (c)
loading the
non-javascript version of this page
. We're sorry about the hassle.
Ok, at the beginning, you have a typo: It's 1 0 0 1 ( 1 + x i x j ) but on the whole, nice solution.
And p ( x ) is increasing .
Log in to reply
Yup thanks for pointing it out. And also, this problem seems to me as more of an algebraic than a number theory one...
Sorry if this is a dumb question but I am unable to see the link from ∣ tan ( α i − α j ) ∣ = ∣ 1 + x i x j x i − x j ∣ to 0 < α i − α j < n − 1 π .
If you could explain it again, that would be awesome. Thank you.
Log in to reply
Yup sure! Sorry that I was unclear.
So basically the 0 < α i − α j < n − 1 π does not follow from ∣ tan ( α i − α j ∣ = … It in fact follows from the bound on the α i namely α i ∈ ( − 2 π , 2 π ) . Because if we consider the trigonometric circle, as in the following picture:
there exists at least one pair angles such that 0 < α i − α j < n − 1 π by pigeonhole .
Log in to reply
Hmm, still a little confused on how n − 1 π appeared, especially the sudden appearance of n . Really sorry for making you explain again.
Log in to reply
@Jing Hao Pang – Ok, so here let's talk more about the motivation. Basically, once we did the trigonometric substitution , which was considerably apparent (in fact I first saw it from the second approach which led me to the greatly simplified first approach), we find that we need ∣ tan ( α i − α j ) ∣ < 1 0 0 1 . But this does not help in solving the problem. We need to find n . So we want something like ∣ tan ( α i − α j ) ∣ = tan ( n × k ) < 1 0 0 1 . We don't really need n × k , but the general idea is to involve n . But remember we have an interval for α i − α j . The standard treatment for intervals is to consider pigeonhole which gives us nice bounds for our disposal. I drew a "trigonometric number line", counted the differences between each α 1 − α 2 , … . Then, to formalise the argument, I turned to the trigonometric circle.
Is this clearer? (:
Log in to reply
@Anqi Li – I get it now with your explanation, as well as a document I found on pigeonhole which showed an example using this very exact trig substitution. Never knew something that seems so trivial could be used in such a way.
Something I've noticed: You mentioned above that ∣ tan ( α i − α j ) ∣ < 1 0 0 1 once trigonometric substitution is done. However, in the original solution, looking at ( ∗ ) , ∣ 1 + x i x j x i − x j ∣ > 1 0 0 1 , and ∣ tan ( α i − α j ) ∣ = ∣ 1 + x i x j x i − x j ∣ . Therefore, shouldn't it be ∣ tan ( α i − α j ) ∣ > 1 0 0 1 rather than ∣ tan ( α i − α j ) ∣ < 1 0 0 1 ?
Thanks for the detailed explanations so far, really appreciate it. :)
Log in to reply
@Jing Hao Pang – Haha nope.
Ok I think I screwed up in my solution. Like I don't think I brought across what I meant and there was a typo in the generalisation. ARGH!
Anyways, I wanted to solve n − 1 π ≤ arctan 1 0 0 1 holds for n ≥ 3 1 6 . In other words, this is counterexample. So we have n = 3 1 5 .
1 0.0000000 80 1.009193137 159 -109.2745084 238 -0.9729170 2 0.0100000 81 1.0295836 160 -52.2110931 239 -0.9536389 3 0.0200020 82 1.0503984 161 -34.2951963 240 -0.9347250 4 0.0300080 83 1.0716550 162 -25.5297265 241 -0.9161614 5 0.0400200 84 1.0933722 163 -20.3296280 242 -0.8979349 6 0.0500400 85 1.1155695 164 -16.8866375 243 -0.8800328 7 0.0600701 86 1.1382677 165 -14.4384660 244 -0.8624430 8 0.0701122 87 1.1614885 166 -12.6080561 245 -0.8451540 9 0.0801684 88 1.1852551 167 -11.1875265 246 -0.8281548 10 0.0902408 89 1.2095919 168 -10.0528601 247 -0.8114349 11 0.1003313 90 1.2345246 169 -9.1254876 248 -0.7949841 12 0.1104421 91 1.2600806 170 -8.3532159 249 -0.7787928 13 0.1205753 92 1.2862889 171 -7.7000169 250 -0.7628518 14 0.1307329 93 1.3131802 172 -7.1402188 251 -0.7471521 15 0.1409171 94 1.3407871 173 -6.6550348 252 -0.7316853 16 0.1511301 95 1.3691444 174 -6.2303995 253 -0.7164432 17 0.1613740 96 1.3982890 175 -5.8555739 254 -0.7014180 18 0.1716510 97 1.4282602 176 -5.5222164 255 -0.6866020 19 0.1819633 98 1.4591000 177 -5.2237497 256 -0.6719881 20 0.1923133 99 1.4908530 178 -4.9549172 257 -0.6575693 21 0.2027031 100 1.5235671 179 -4.7114679 258 -0.6433389 22 0.2131351 101 1.5572936 180 -4.4899264 259 -0.6292905 23 0.2236117 102 1.5920870 181 -4.2874242 260 -0.6154177 24 0.2341353 103 1.6280063 182 -4.1015724 261 -0.6017146 25 0.2447082 104 1.6651145 183 -3.9303656 262 -0.5881755 26 0.2553330 105 1.7034793 184 -3.7721080 263 -0.5747947 27 0.2660123 106 1.7431740 185 -3.6253557 264 -0.5615668 28 0.2767484 107 1.7842770 186 -3.4888717 265 -0.5484867 29 0.2875442 108 1.8268735 187 -3.3615901 266 -0.5355493 30 0.2984023 109 1.8710553 188 -3.2425876 267 -0.5227497 31 0.3093253 110 1.9169220 189 -3.1310602 268 -0.5100833 32 0.3203161 111 1.9645815 190 -3.0263048 269 -0.4975454 33 0.3313776 112 2.0141511 191 -2.9277036 270 -0.4851316 34 0.3425126 113 2.0657586 192 -2.8347116 271 -0.4728377 35 0.3537241 114 2.1195433 193 -2.7468464 272 -0.4606596 36 0.3650153 115 2.1756573 194 -2.6636793 273 -0.4485931 37 0.3763892 116 2.2342673 195 -2.5848277 274 -0.4366344 38 0.3878490 117 2.2955561 196 -2.5099499 275 -0.4247796 39 0.3993980 118 2.3597249 197 -2.4387387 276 -0.4130252 40 0.4110397 119 2.4269953 198 -2.3709182 277 -0.4013674 41 0.4227775 120 2.4976123 199 -2.3062392 278 -0.3898029 42 0.4346150 121 2.5718470 200 -2.2444762 279 -0.3783282 43 0.4465558 122 2.6500010 201 -2.1854249 280 -0.3669399 44 0.4586037 123 2.7324099 202 -2.1288994 281 -0.3556350 45 0.4707626 124 2.8194488 203 -2.0747304 282 -0.3444101 46 0.4830366 125 2.9115381 204 -2.0227635 283 -0.3332623 47 0.4954297 126 3.0091507 205 -1.9728573 284 -0.3221886 48 0.5079462 127 3.1128201 206 -1.9248821 285 -0.3111860 49 0.5205905 128 3.2231510 207 -1.8787190 286 -0.3002516 50 0.5333672 129 3.3408311 208 -1.8342584 287 -0.2893828 51 0.5462809 130 3.4666458 209 -1.7913995 288 -0.2785766 52 0.5593364 131 3.6014970 210 -1.7500492 289 -0.2678305 53 0.5725388 132 3.7464243 211 -1.7101212 290 -0.2571418 54 0.5858933 133 3.9026336 212 -1.6715359 291 -0.2465079 55 0.5994052 134 4.0715305 213 -1.6342194 292 -0.2359263 56 0.6130800 135 4.2547645 214 -1.5981028 293 -0.2253946 57 0.6269235 136 4.4542838 215 -1.5631225 294 -0.2149102 58 0.6409417 137 4.6724060 216 -1.5292190 295 -0.2044708 59 0.6551408 138 4.9119104 217 -1.4963367 296 -0.1940739 60 0.6695272 139 5.1761587 218 -1.4644240 297 -0.1837174 61 0.6841074 140 5.4692561 219 -1.4334325 298 -0.1733988 62 0.6988886 141 5.7962689 220 -1.4033169 299 -0.1631160 63 0.7138778 142 6.1635232 221 -1.3740348 300 -0.1528666 64 0.7290826 143 6.5790228 222 -1.3455465 301 -0.1426486 65 0.7445107 144 7.0530442 223 -1.3178147 302 -0.1324596 66 0.7601702 145 7.5990054 224 -1.2908043 303 -0.1222976 67 0.7760697 146 8.2347657 225 -1.2644823 304 -0.1121605 68 0.7922178 147 8.9846289 226 -1.2388177 305 -0.1020460 69 0.8086239 148 9.8825383 227 -1.2137812 306 -0.0919522 70 0.8252974 149 10.9773823 228 -1.1893451 307 -0.0818769 71 0.8422485 150 12.3422368 229 -1.1654835 308 -0.0718181 72 0.8594875 151 14.0914351 230 -1.1421717 309 -0.0617737 73 0.8770254 152 16.4144694 231 -1.1193863 310 -0.0517417 74 0.8948737 153 19.6498955 232 -1.0971055 311 -0.0417202 75 0.9130443 154 24.4677908 233 -1.0753082 312 -0.0317069 76 0.9315498 155 32.4070896 234 -1.0539748 313 -0.0217001 77 0.9504032 156 47.9593044 235 -1.0330863 314 -0.0116975 78 0.9696185 157 92.1765243 236 -1.0126250 315 -0.0016973 79 0.9892101 158 1178.3320865 237 -0.9925739 316 0.0083025
by using Excel sheet and input this formula : B1=0 then B2= B1 + (0.01+(B1)^2/100) / (1-- B1/100)
then copy B2 and paste in B3 to B400 you will find that at B316 come inside previous range.
For ease of reading and writing, x i and x j will be replaced by a and b .
First, we notice that if the condition holds for some a , b , the condition also holds for − a , − b . We can also easily verify that all the numbers are distinct, or else ∣ a − b ∣ = 0 and a ⋅ b ≥ 0 → 1 0 0 1 ( 1 + a ⋅ b ) > 0 .
For now, let's only look at positive reals. Because the numbers are all distinct, we can assume that a > b in the given inequality, and we can get rid of the absolute value. This assumption also lets us get a strict bound on the value of b given a . Specifically, we solve the inequality for b : ∣ a − b ∣ > 1 0 0 1 ( 1 + a ⋅ b ) → a − b > 1 0 0 1 ( 1 + a ⋅ b ) → 1 0 0 a − 1 > a ⋅ b + 1 0 0 b → b < a + 1 0 0 1 0 0 a − 1 . Because a > b > 0 , we don't need to worry about a zero denominator or any strange sign switches while manipulating the inequality.
We can quickly verify that this inequality is strictly decreasing. This means that if we begin with with a specific a 0 value, solve for b 0 , and use this value for a 1 , the bounds on b 1 will be tighter than those on b 0 . This also tells us that we don't need to check if every pair of numbers works. We just need the ones that are closest to each other to satisfy the inequality.
The next step is to see how many positive values we can have. By increasing the largest value, x 1 , we can bring the second highest value, x 2 , arbitrarily close to 1 0 0 , x 3 arbitrarily close to 1 0 0 + 1 0 0 1 0 0 ( 1 0 0 ) − 1 = 4 9 . 9 9 5 , etc. This means that the upper bound on x k = f ( x k − 1 ) , where f ( x ) = x + 1 0 0 1 0 0 x − 1 . We can now calculate what these upper bounds are.
To simplify the process, let's assume that f ( x 1 ) = x 2 = 1 0 0 . Clearly, no value can actually make this happen, but we can get as close as we'd like. (From this point on, we will work under the assumption that we can bring x i arbitrarily close to its upper bound and will treat x i as being equal to its upper bound.) Repeatedly applying f gives us x 3 = 4 9 . 9 9 5 , x 4 = 3 3 . 3 2 4 4 , . . . , x 1 5 8 ≈ . 0 0 0 8 5 , x 1 5 9 ≈ − . 0 0 9 1 5 . Because x 1 5 9 < 0 , we can stop here, since we're still only looking at positive reals. There seem to be a maximum of 1 5 8 positive reals that, in pairs, satisfy the given inequality, and because all the pairings of their opposites also satisfy the inequality, the answer should be 2 ∗ 1 5 8 = 3 1 6 .
However, we're overlooking something: if we include x 1 5 8 and − x 1 5 8 = x 3 1 6 , then choosing these two as a and b will not satisfy the inequality because ∣ x 1 5 8 − x 3 1 6 ∣ = 2 ⋅ x 1 5 8 < 1 0 0 1 ( 1 − x 1 5 8 2 ) . This is not the case for x 1 5 7 and − x 1 5 7 = x 3 1 5 , which also implies that it's not the case for any other x k , 1 ≤ k ≤ 1 5 6 because as x k increases, 1 − x k 2 decreases and 2 ⋅ x k increases. Therefore, we can just take out x 1 5 8 and x 3 1 6 and be sure that every other pairing satisfies the inequality.
We can easily verify that any number already in the list can be paired with 0 to form a pair that satisfies the given inequality, since all the numbers have magnitude at least . 0 1 , which means ∣ x k − 0 ∣ > 1 0 0 1 ( 1 + 0 ⋅ x k ) . Therefore, we can add 0 into our list of reals.
We can now see that there are a maximum of 2 ⋅ 1 5 7 + 1 = 3 1 5 reals that pairwise satisfy the given inequality, and so n = 3 1 5 .
Consider a i = arctan x i . Thus we have that tan a i = x i . Now we note that ∣ tan ( a i − a j ) ∣ > 1 / 1 0 0 . Thus ∣ a i − a j ∣ > a r c t a n ( 1 / 1 0 0 ) . It is simple then to note that if there are more than 315 x's, then we can simply consider the intervals [karctan(1/100),(k+1)arctan(1/100)) for k = 0,1,...,314. At least one of these intervals will contain 2 a_i's (since we limit the range of arctan from 0 to pi) and thus we attain a contradiction.
In our arrangement, for all x i , define y i = tan − 1 x i , therefore, for some x i > x j
∣ x i − x j ∣ > 1 0 0 1 ( 1 + x i x i ) ⇒ tan ( y i − y j ) > 1 0 0 1 ⇒ y i − y j > tan − 1 1 0 0 1
Since y i ∈ ( − 2 π , 2 π ) for all i , so if we divide the interval ( − 2 π , 2 π ) into atmost ⌈ tan − 1 1 0 0 1 1 8 0 ⌉ parts, each y i belonging to each part, we have a valid arrangement. This yields the maximum n ≥ 3 1 5 .
If we take one more x i or equivalently, y i , by PHP, it falls into one of the already occupied partitions & the arrangement becomes invalid, hence giving the maximum 3 1 5 .
I used 1 8 0 for division since I calculated tan − 1 1 0 0 1 in degrees.
Problem Loading...
Note Loading...
Set Loading...
Let us rewrite the inequality , motivated by the fact that isolating variables is sometimes quite a good technique, and for reasons that will soon be apparent. However, before we do anything, I claim that we only need to consider 1 + x i x j > 0 . This is motivated by the absolute value as absolute values are unpredictable and we should consider positive and negative cases separately. Indeed, notice that if 1 + x i x j = 0 then clearly we have x i = x j so ∣ x i − x j ∣ > 0 = 1 0 0 1 ( 1 + x i x j ) so we are trivially done. And if 1 + x i x j < 0 then ∣ x i − x j ∣ > 0 > 1 0 0 1 ( 1 + x i x j ) .
With the above observations in mind, we rewrite our inequality as:
∣ 1 + x i x j x i − x j ∣ > 1 0 0 1 ( ∗ ) .
For those who are familiar with the sum and difference formula for tangents , in this new form, it should definitely ring a bell in your head. Indeed, let us now use a substitution of x i = tan α i . We can bound α i by α i ∈ ( − 2 π , 2 π ) . So now, with the substitution, rewrite ( ∗ ) again to the following:
∣ tan ( α i − α j ) ∣ = ∣ 1 + tan α i tan α j tan α i − tan α j ∣ = ∣ 1 + x i x j x i − x j ∣ ( ∗ ∗ )
Here, we see why bounding α i is of such utmost importance. Because, now we can safely conclude that there will exist at least one angle 0 < α i − α j < n − 1 π . So ( ∗ ∗ ) holds iff tan n − 1 π < 1 0 0 1 ↔ n − 1 π < arctan 1 0 0 1 . One easily verifies that this holds for n < 3 1 6 .
So now we give a construction n = 3 1 5 . Let us choose an arbitrarily small but non-negative c . We will attempt to make tan ( α i − α j ) > tan n − 1 π − c = tan 3 1 4 π − c > 1 0 0 1 . This motivates us to write α l = n − 1 ( l − 1 ) ( π − c ) . However, we remember that we need to fit this in the correct range , and also remember that we are considering absolute value so we can also have tan ( α i − α j ) < 0 . Now, finally, we can our construction for l = 0 , 1 , … , n as α l = − 2 π + 2 c + n − 1 ( l − 1 ) ( π − c ) .
In conclusion, the answer is 3 1 5 .
Or, if this is too convoluted, one might attempt a more straight-forward, constructive albeit tedious process. We conclude as above that we only have to consider positive reals. The motivating principle here is to first "order" the elements. So here, let x i > x j to remove the absolute values. For instance, we can again rewrite the inequality as x i > 1 0 0 − x j 1 0 0 x j + 1 . Now, notice first that in this other form of ( 1 0 0 − x j ) x i > 1 0 0 x j + 1 , we must have x j < 1 0 0 or otherwise we would violate the positive reals condition.
It is quite easy to verify that over ( − ∞ , 1 0 0 ) , p ( x ) = 1 0 0 − k 1 0 0 k + 1 is increasing. For notation sake, let y i + 1 = f ( y i ) and y 1 = x 1 < 1 0 0 . To simplify the procedure just let y 1 → − ∞ and see how many iterations of the function is needed to exceed 1 0 0 .
Again, I had to use the substitution of tangents to simplify the work. By letting 1 0 0 = tan r and y i = − tan s (the negative is because of the signs in our rewritten inequality) we can perform some algebraic manipulation to get:
1 + 2 π − r 2 π + r < n < 2 + 2 π − r 2 π + r
Solving gives: n = 3 1 5 as above.
Remark : Using both methods we can generalise that suppose we need ∣ x i − x j ∣ > m 1 ( 1 + x i x j ) , then we have:
Method 1 : n − 1 π < arctan m 1
Method 2 : n = ⌊ 2 + 2 π − arctan m 2 π + arctan m ⌋