There is an array of ~n~ real numbers. The ~i~-th real number is ~h_i~. It is guaranteed that initially ~h_i~s are distinct.

In each operation, you may choose an arbitrary number of ~h_i~s and set the value of each chosen ~h_i~ to be the average of the chosen numbers. The chosen numbers do not need to be consecutive (i.e. you can choose ~h_1~ and ~h_3~ but not ~h_2~). You may apply the operation for at most ~k~ times.

Compute the maximum possible ~h_1~ in this case.

Input Specification

The first line of the input consists of three positive integers ~n,k,p~ denoting the length of the array, the maximum number of operations, and the required precision.

The next line consists of ~n~ positive integers. The ~i~-th integer denotes ~h_i~. It is guaranteed that ~h_i~s are distinct. ~1 \le h_i \le 10^5~.

Output Specification

Output a line with a real number denoting the maximum ~h_1~ using at most ~k~ operations.

The answer can only contain a non-negative integer part, the decimal point, and the decimal part. The non-negative integer part is required, and it is unnecessary to add signs before the output. If there is a decimal part, the integral part and the decimal part are separated by a decimal point. If there is no decimal part, there shall be no decimal points.

In your output, there can be at most ~2p~ digits after the decimal point. It is recommended that you should keep at least ~p~ digits in the decimal part. It is guaranteed that the absolute difference between the reference answer and the true answer is at most ~10^{-2p}~.

Your output is considered to be correct if and only if the absolute difference between your output and the reference answer is less than ~10^{-p}~.

If the absolute difference between your output and the reference answer is at least ~10^{-p}~ but less than ~10^{-5}~, you may get ~40\%~ of the points of the test case.

Sample Input 1

3 1 3
1 4 3

Sample Output 1

2.666667

Explanation for Sample 1

There are five possibilities since we can use at most one operation (not counting the trivial operation when you just choose one number).

• Performing no operations: ~h_1 = 1~.
• Choosing ~h_1, h_2~: now ~h_1 = h_2 = (1+4)/2 = 5/2~.
• Choosing ~h_1, h_3~: now ~h_1 = h_3 = (1+3)/2 = 2~.
• Choosing ~h_2, h_3~: now ~h_1 = 1~.
• Choosing ~h_1, h_2, h_3~: now ~h_1 = h_2 = h_3 = (1+4+3)/3 = 8/3~.

Sample Input 2

3 2 3
1 4 3

Sample Output 2

3.000000

Explanation for Sample 2

The optimal solution is to apply two operations: first, choose ~h_1~ and ~h_3~. Then choose ~h_1~ and ~h_2~.

Attachment Package

The samples are available here.

Sample 3

See ex_drink3.in and ex_drink3.ans.

Constraints

Test case	~n~	~k~	~p~
1	~\le 2~	~\le 5~	~=5~
2	~\le 4~
3	~\le 4~
4	~\le 10~	~=1~
5		~=10^9~
6		~\le 10~
7
8	~\le 100~	~=1~
9		~=10^9~	~=40~
10		~\le 10^9~
11
12
13	~\le 250~		~=100~
14	~\le 500~		~=200~
15	~\le 700~		~=300~
16
17
18	~\le 2\,500~		~=1\,000~
19	~\le 4\,000~		~=1\,500~
20	~\le 8\,000~		~=3\,000~

For all test cases, it is guaranteed that ~3 \le p \le 3\,000~, ~1 \le n \le 8\,000~, and ~1 \le k \le 10^9~.

Hints

To guarantee precision, we need to keep more than ~p~ digits after the decimal point if possible when performing arithmetic operations. It can be shown that the reference solution to each subtask can guarantee the absolute difference between the output of the solution to the subtask and the reference answer is less than ~10^{-p}~ when each arithmetic operation keeps at most ~\frac{6}{5}p~ digits after the decimal point.

A library for high-precision floating number operations is provided here.

Using the library is optional.

The Decimal library supports high-precision floating point addition, subtraction, multiplication, division, comparisons, and conversions to and from double, integers, and strings. Note: Here, one operand for multiplication and division must be an integer. You cannot multiply a double or a Decimal with another Decimal.

The library defines a constant ~P~, which says the absolute error of each operation is at most ~10^{-P}~. If you are using the C++ version, in the 9-th line of drink_sample.cpp, const int PREC = 2100; defines the constant ~P~. If you are using the C version of the library, the line #define PREC 2100 in the 9-th line of drink_sample.c defines the constant ~P~. If you are using Pascal, 2100 in the 8-th and the 14-th lines of drink_sample.pas is the constant ~P~. If you need to modify ~P~, please change both occurrences to the same value.

The time complexity of any arithmetic operation provided by the library is bounded above by ~\mathcal O(P)~.

The space complexity of any instance of the Decimal class is bounded above by ~\mathcal O(P)~. More precisely, each instance should take at most ~\frac{4P}{9} + 16~ bytes of space.

When calling a function provided by the Decimal class, the following conditions must be satisfied:

• In each intermediate step, a Decimal number has an absolute value of at most ~10^{18}~.
• int/longint parameters have absolute values of at most ~10^9~.
• long long/int64 parameters have absolute values of at most ~10^{18}~.
• A double parameter must be a valid real number and has an absolute value of at most ~10^9~.
• An argument of type string must represent a valid real number. In other words, the string should begin with the sign part (a negative number has a minus sign, and a non-negative number has no signs), then a string consisting of digits representing the integer part, a decimal point, and a string consisting of digits representing the decimal part. The decimal part and the decimal point can be omitted simultaneously. The string should not represent a real number with absolute value greater than ~10^9~.
• You cannot divide by zero.
• When converting a Decimal to a string, the number of digits after the decimal point must be greater than zero. If you are using C, please make sure you've allocated enough space to store the string returned by to_string_d(Decimal, char*, int).

The programs drink_sample.cpp/c/pas are empty programs except the library. You may work directly on the samples, but this is optional.

The programs decimal_test.cpp/c/pas are example programs for the Decimal library. You can use the program to learn more about the interfaces and the implementation of the library, as well as how to use each function of the library.