Maths Olympiad Prep

Library / /32 of 42

Algebra Difficulty 6.7 National olympiad Prove it Ireland

Suppose f:{0,1}10Rf : \{0,1\}^{10} \to \mathbb{R}, i.e. f(x1,x2,x3,,x10)f(x_1, x_2, x_3, \dots, x_{10}) is defined whenever xi{0,1}x_i \in \{0,1\} for each 1i101 \le i \le 10. We are not given these values f(x1,,x10)f(x_1, \dots, x_{10}), but, for every choice of x1,,x10{0,1}x_1, \dots, x_{10} \in \{0,1\}, we know each of the following ten sums of two values of ff:
f(0,x2,x3,,x10)+f(1,x2,x3,,x10)f(x1,0,x3,,x10)+f(x1,1,x3,,x10)f(x1,x2,0,,x10)+f(x1,x2,1,,x10)+f(x1,x2,,x9,0)+f(x1,x2,,x9,1) \begin{aligned} f(0, x_2, x_3, \dots, x_{10}) &+ f(1, x_2, x_3, \dots, x_{10}) \\ f(x_1, 0, x_3, \dots, x_{10}) &+ f(x_1, 1, x_3, \dots, x_{10}) \\ f(x_1, x_2, 0, \dots, x_{10}) &+ f(x_1, x_2, 1, \dots, x_{10}) \\ & \qquad + \dots \\ f(x_1, x_2, \dots, x_9, 0) &+ f(x_1, x_2, \dots, x_9, 1) \end{aligned}
Show that knowing these sums does not allow us to compute the values of f(x1,,x10)f(x_1, \dots, x_{10}).
Write down one or more other sums of values of ff at two distinct points such that if we know these sums as well as the above sums, then we can compute all values of ff. You should justify your answer and use as few additional sums as possible.

Solution

We consider the natural generalisation with 1010 replaced everywhere by nn for some n>1n > 1, nNn \in \mathbb{N}. It is convenient to use vector notation, writing f(x)f(\underline{x}) in place of f(x1,,xn)f(x_1, \dots, x_n). To show that the given sums are insufficient to compute all ff-values, it suffices to note that if we write s(x)=i=1nxis(\underline{x}) = \sum_{i=1}^n x_i and then redefine ff by adding (1)s(x)(-1)^{s(\underline{x})} to f(x)f(\underline{x}), then all the listed sums remain the same.
We will prove that, the value of the additional sum
f(0,0,0,0,,0)+f(1,1,0,0,,0) f(0, 0, 0, 0, \dots, 0) + f(1, 1, 0, 0, \dots, 0)
suffices to compute the values of ff.
It can be shown in a similar way that any other additional sum of the form f(a)+f(b)f(\underline{a}) + f(\underline{b}), for some fixed a,b\underline{a}, \underline{b}, where s(a)s(\underline{a}) and s(b)s(\underline{b}) have the same parity, or equivalently where a\underline{a} and b\underline{b} differ in an even number of positions, is sufficient.

Solution 1. We know all sums of the form S:=f(x)+f(y)S := f(\underline{x}) + f(\underline{y}) for vectors x,y\underline{x}, \underline{y} that differ in a single position. We also know T:=f(y)+f(z)T := f(\underline{y}) + f(\underline{z}), whenever y\underline{y} and z\underline{z} differ in a single position, and so we know ST=f(x)f(z)S - T = f(\underline{x}) - f(\underline{z}). Thus, we know such differences whenever x\underline{x} and z\underline{z} differ in exactly two positions. By repeating the process of taking differences, we can compute f(u)f(v)f(\underline{u}) - f(\underline{v}) whenever u\underline{u} and v\underline{v} differ in any even number of positions. In particular, we can compute f(a)f(b)f(\underline{a}) - f(\underline{b}) where a=(0,0,0,0,,0)\underline{a} = (0, 0, 0, 0, \dots, 0) and b=(1,1,0,0,,0)\underline{b} = (1, 1, 0, 0, \dots, 0). Since we also know f(a)+f(b)f(\underline{a}) + f(\underline{b}), we can compute f(a)f(\underline{a}).
Knowing f(a)f(\underline{a}) allows us to compute all values. In fact, if u\underline{u} differs from a\underline{a} in exactly one position, then knowing f(a)+f(u)f(\underline{a}) + f(\underline{u}) allows us to compute f(u)f(\underline{u}). Applying the same argument to f(u)f(\underline{u}), we compute f(v)f(\underline{v}) whenever v\underline{v} differs from u\underline{u} in exactly one position, and so whenever v\underline{v} differs from a\underline{a} in exactly two positions. Iterating this approach, we get all other values of ff.

Solution 2. We prove our claim by induction on the dimension n>1n > 1. Suppose first that n=2n = 2, so we know
A:=f(0,0)+f(1,0),B:=f(1,1)+f(0,1),C:=f(0,0)+f(0,1),D:=f(1,1)+f(1,0). \begin{aligned} A &:= f(0, 0) + f(1, 0), & B &:= f(1, 1) + f(0, 1), \\ C &:= f(0, 0) + f(0, 1), & D &:= f(1, 1) + f(1, 0). \end{aligned}
The additional sum is E=f(0,0)+f(1,1)E = f(0, 0) + f(1, 1). Since AD=f(0,0)f(1,1)A - D = f(0, 0) - f(1, 1), knowing EE allows us to compute
f(0,0)=(E+AD)/2, f(0, 0) = (E + A - D)/2,
and using A,C,EA, C, E we find the remaining three values of ff.

Want a route through all this instead of an archive? The track puts 2,000 problems in a working order, from AMC 10 level to the IMO shortlist.

Source: MathNet, licensed CC-BY-4.0. Statement and solution reproduced as published; topic and difficulty added by this site.