Develop Reference

Total of all numbers from 1 to N will always be zero

The problem is I have to print all combinations of a sequence of
numbers from 1 to N that will always result to zero. It is allowed
to insert "+" (for adding) and "-" (for subtracting) between each
numbers so that the result will be zero.
//Output
N = 7
1 + 2 - 3 + 4 - 5 - 6 + 7 = 0
1 + 2 - 3 - 4 + 5 + 6 - 7 = 0
1 - 2 + 3 + 4 - 5 + 6 - 7 = 0
1 - 2 - 3 - 4 - 5 + 6 + 7 = 0
So how can I implement this? I am not asking for the actual
codes to do this, just a hint and ideas to solve this will
do. Thank you..

You could also use recursion here. Just remember your current integer, your max integer, your current sum and some kind of history of operations (could also be your final sequence).
In every level you proceed the path in two dirdctions: adding to your sum and substracting from it.
I did a quick implementation in Python, but it should be easy to transfer this to Java or whatever you are using.
def zero_sum(curr, n, seq, sum):
if curr == n and sum == 0:
print(seq)
elif curr < n:
zero_sum(curr + 1, n, seq + " - " + str(curr + 1), sum - (curr + 1))
zero_sum(curr + 1, n, seq + " + " + str(curr + 1), sum + (curr + 1))
zero_sum(1, 7, "1", 1)
Hopefully you get the idea.

The first step is to turn the problem into an entirely regularly formed problem:
n
∑ ±i = -1
i=2
n-2
∑ ±(i+2) = -1
i=0
The term 1 at the start has no prefix +/-. And the walking index better runs from 0 when using a Java array.
So one has n-1 coefficients -1 or +1 for the possible values.
A brute force approach would be to start with the highest values, i = n-2.
The upper/lower bounds for j = 0, ..., i would be ± (i + 1) * (2 + i + 2) / 2, so one can cut the evaluation there - when the till then calculated sum can no longer reach -1.
To represent the coefficients, one could make a new int[n - 1] or simply a new BitSet(n-1).
public void solve(int n) {
int i = n-2;
int sumDone = 0;
BigSet negates = new BitSet(n - 1);
solveRecursively(i, sumDone, negates);
}
private void solveRecursively(int i, int SumDone, BitSet negates) {
if (i < 0) {
if (sumDone == -1) {
System.out.println("Found: " + negates);
}
return;
}
...
}
The interesting, actual (home) work I leave to you. (With BitSet better i = n, ... , 2 by -1 seems simpler though.)

The question here is how much efficiency matters. If you're content to do a brute-force approach, a regression method like the one indicated by holidayfun is a fine way to go, though this will become unwieldy as n gets large.
If performance speed matters, it may be worth doing a bit of math first. The easiest and most rewarding check is whether such a sum is even possible: since the sum of the first n natural numbers is n(n+1)/2, and since you want to divide this into two groups (a "positive" group and a "negative" group) of equal size, you must have that n(n+1)/4 is an integer. Therefore if neither n nor n+1 is divisible by four, stop. You cannot find such a sequence that adds to zero.
This and a few other math tricks might speed up your application significantly, if speed is of the essence. For instance, finding one solution will often help you find others, for large n. For instance, if n=11, then {-11, -10, -7, -5} is one solution. But we could swap the -5 for any combination that adds to 5 that isn't in our set. Thus {-11, -10, -7, -3, -2} is also a solution, and similarly for -7, giving {-11, -10, -5, -4, -3} as a solution (we are not allowed to use -1 because the 1 must be positive). We could continue replacing the -10, the -11, and their components similarly to pick up six more solutions.
This is probably how I'd approach this problem. Use a greedy algorithm to find the "largest" solution (the solution using the largest possible numbers), then keep splitting the components of that solution into successively smaller solutions. It is again fundamentally a recursion problem, but one whose running time decreases with the size of the component under consideration and which at each step generates another solution if a "smaller" solution exists. That being said, if you want every solution then you still have to check non-greedy combinations of your split (otherwise you'd miss solutions like {-7, -4, -3} in your n=7 example). If you only wanted a lot of solutions it would definitely be faster; but to get all of them it may be no better than a brute-force approach.

If I were you I would go for a graph implementation, and DFS algorithm. Imagine you have N nodes that are representing your numbers. Each number is connected to another via an "add" edge, or a "subtract" edge. So you have a fully connected graph. You can start from a node and compute all dfs paths that lead to zero.
For more information about DFS algorithm, you can see the wikipage.
Edit: In order to clarify my solution, the graph you will end up having will be a multigraph, which means that it has more than one edge between nodes. DFS in a multigraph is slightly more complicated, but it is not that hard.

I would suggest a straight forward solution because as you mentioned you are dealing with consecutive integer from 1 to N which are fixed. The only things that vary are the operators in between.
Let's look at your example before we implement a general solution:
For n = 7 you need somehow to produce all possible combinations:
1+2+3+4+5+6+7
1+2+3+4+5+6-7
1+2+3+4+5-6+7
1+2+3+4+5-6-7
...
1-2-3-4-5-6+7
1-2-3-4-5-6-7
If we remove the numbers from above strings/expressions then we'll have:
++++++
+++++-
++++-+
++++--
...
----+-
-----+
------
Which reminds on binary numbers; if we interpret + as 0 and - as 1 the above can be mapped to the binary numbers from 000000 to 111111.
For an input n you'll have n-1 operators inbetween, which means the count of all possible combinations will be 2^n-1.
Putting all the above together something like below can be used to print those which sums are zero:
public static void main(String args[]) throws IOException{
permute(7);
}
public static void permute(int n){
int combinations = (int)Math.pow(2, n-1);
for(int i = 0; i < combinations; i++){
String operators =String.format("%"+(n-1)+"s", Integer.toBinaryString(i)).replace(' ', '0');
int totalSum = 1;
StringBuilder sb = new StringBuilder();
for(int x = 0; x< operators.length(); x++){
sb.append(x+1);
if(operators.charAt(x)=='0'){
sb.append("+");
totalSum = totalSum + (x+2);
}
else{
sb.append("-");
totalSum = totalSum-(x+2);
}
}
sb.append(n);
if(totalSum == 0){
System.out.println(sb.toString() + " = " + totalSum);
}
}
}
Note/Example: String.format("%6s", Integer.toBinaryString(13)).replace(' ', '0') will produce a string with length = 6 from the binary representation of 13 with leading zeros, i.e 001101 instead of 1101 so that we get the required length of the operators.

This is an interesting question. It involves more math than programming because only if you discover the math portion then you may implement an efficient algorithm.
However even before getting into the math we must actually understand what exactly the question is. The question can be rephrased as
Given array [1..n], find all possible two groups (2 subarrays) with equal sum.
So the rules;
sum of [1..n] is n*n(+1)/2
If n*(n+1)/2 is odd then there is no solution.
If your target sum is t then you should not iterate further for lower values than Math.ceil((Math.sqrt(8*t+1)-1)/2) (by solving n from n(n+1)/2 = t equation)
Sorry... I know the question requests Java code but I am not fluent in Java so the code below is in JavaScript. It's good though we can see the results. Also please feel free to edit my answer to include a Java version if you would like to transpile.
So here is the code;
function s2z(n){
function group(t,n){ // (t)arget (n)umber
var e = Math.ceil((Math.sqrt(8*t+1)-1)/2), // don't try after (e)nd
r = [], // (r)esult
d; // (d)ifference
while (n >= e){
d = t-n;
r = d ? r.concat(group(d, d < n ? d : n-1).map(s => s.concat(n)))
: [[n]];
n--;
}
return r;
}
var sum = n*(n+1)/2; // get the sum of series [1..n]
return sum & 1 ? "No solution..!" // if target is odd then no solution
: group(sum/2,n);
}
console.log(JSON.stringify(s2z(7)));
So the result should be [[1,6,7],[2,5,7],[3,4,7],[1,2,4,7],[3,5,6],[1,2,5,6],[1,3,4,6],[2,3,4,5]].
what does this mean..? If you look into that carefuly you will notice that
These are all the possible groups summing up to 14 (half of 28 which is the sum of [1..7].
The first group (at index 0) is complemented by the last group (at index length-1) The second is complemented with the second last and so on...
Now that we have the interim results it's up to us how to display them. This is a secondary and trivial concern. My choice is a simple one as follows.
var arr = [[1,6,7],[2,5,7],[3,4,7],[1,2,4,7],[3,5,6],[1,2,5,6],[1,3,4,6],[2,3,4,5]],
res = arr.reduce((r,s,i,a) => r+s.join("+")+"-"+a[a.length-1-i].join("-")+" = 0 \n","");
console.log(res);
Of course you may put the numbers in an order or might stop halfway preventing the second complements taking positive values while the firsts taking negative values.
This algorithm is not hard tested and i might have overlooked some edges but I believe that this should be a very efficient algorithm. I have calculated up to [1..28] in a very reasonable time resulting 2399784 uniques groups to be paired. The memory is only allocated for the constructed result set despite this is a resursive approach.

It is a good question, but first you must have to try to solve it and show us what you tried so we can help you in the solution, this way you will improve more effectively.
However, the below code is a solution I have write before years, I think the code need improvement but it will help..
public static void main(String[] args) {
String plus = " + ", minus = " - ";
Set<String> operations = new HashSet<>();
operations.add("1" + plus);
operations.add("1" + minus);
// n >= 3
int n = 7;
for (int i = 1; i < n - 1; i++) {
Set<String> newOperation = new HashSet<>();
for (String opt : operations) {
if ((i + 2) == n) {
newOperation.add(opt + (i + 1) + plus + n);
newOperation.add(opt + (i + 1) + minus + n);
} else {
newOperation.add(opt + (i + 1) + plus);
newOperation.add(opt + (i + 1) + minus);
}
}
operations.clear();
operations.addAll(newOperation);
}
evalOperations(operations);
}
private static void evalOperations(Set<String> operations) {
// from JDK1.6, you can use the built-in Javascript engine.
ScriptEngineManager mgr = new ScriptEngineManager();
ScriptEngine engine = mgr.getEngineByName("JavaScript");
try {
for (String opt : operations) {
if ((int) engine.eval(opt) == 0) {
System.out.println(opt + " = 0");
}
}
} catch (ScriptException e) {
e.printStackTrace();
}
}

First, the question is the special case of sum to N.
Second, sum a list to N, could be devided to the first element plus sublist and minus sublist.
Third, if there are only one element in the list, check if n equals the element.
Fourth, make recursion.
Here's the scala implementation, try finishing your java version:
def nSum(nums: List[Int], n: Int, seq: String, res: ListBuffer[String]): Unit =
nums match {
case Nil => if (n == 0) res.append(seq)
case head :: tail => {
nSum(tail, n - head, seq + s" + $head", res)
nSum(tail, n + head, seq + s" - $head", res)
}
}
def zeroSum(nums: List[Int]): List[String] = {
val res = ListBuffer[String]()
nSum(nums.tail, -nums.head, s"${nums.head}", res)
res.map(_ + " = 0").toList
}
val expected = List(
"1 + 2 - 3 + 4 - 5 - 6 + 7 = 0",
"1 + 2 - 3 - 4 + 5 + 6 - 7 = 0",
"1 - 2 + 3 + 4 - 5 + 6 - 7 = 0",
"1 - 2 - 3 - 4 - 5 + 6 + 7 = 0")
assert(expected == zeroSum((1 to 7).toList))

Starting from 1 how far can I count, when I can use any digit a maximum of N times

My problem is as follows; for number N, I need to find out what is the largest value I can count to, when each digit can be used N times.
For example if N = 5, the largest value is 12, since at that point the digit 1 has been used 5 times.
My original approach was to simply iterate through all numbers and keep a tally of how many times each digit has been used so far. This is obviously very inefficient when N is large, so am looking for advice on what would be a smarter (and more efficient) way to achieve this.
public class Counter {
private static Hashtable<Integer, Integer> numbers;
public static void main(String[] args){
Counter c = new Counter();
c.run(9);
}
public Counter() {
numbers = new Hashtable<Integer, Integer>();
numbers.put(0, 0);
numbers.put(1, 0);
numbers.put(2, 0);
numbers.put(3, 0);
numbers.put(4, 0);
numbers.put(5, 0);
numbers.put(6, 0);
numbers.put(7, 0);
numbers.put(8, 0);
numbers.put(9, 0);
}
public static void run(int maxRepeat) {
int keeper = 0;
for(int maxFound = 0; maxFound <= maxRepeat; maxFound++) {
keeper++;
for (int i = 0; i < Integer.toString(keeper).length(); i++) {
int a = Integer.toString(keeper).charAt(i);
//here update the tally for appropriate digit and check if max repeats is reached
}
}
System.out.println(keeper);
}
}

For starters, rather than backing your Counter with a Hashtable, use an int[] instead. When you know exactly how many elements your map has to have, and especially when the keys are numbers, an array is perfect.
That being said, I think the most effective speedup is likely to come from better math, not better algorithms. With some experimentation (or it may be obvious), you'll notice that 1 is always the first digit to be used a given number of times. So given N, if you can find which number is the first to use the digit 1 N+1 times, you know your answer is the number right before that. This would let you solve the problem without actually having to count that high.
Now, let's look at how many 1's are used counting up to various numbers. Throughout this post I will use n to designate a number when we are trying to figure out how many 1's are used to count up to a number, whereas capital N designates how many 1's are used to count up to something.
One digit numbers
Starting with the single-digit numbers:
1: 1
2: 1
...
9: 1
Clearly the number of 1's required to count up to a one-digit number is... 1. Well, actually we forgot one:
0: 0
That will be important later. So we should say this: the number of 1's required to count up to a one-digit number X is X > 0 ? 1 : 0. Let's define a mathematical function f(n) that will represent "number of 1's required to count up to n". Then
f(X) = X > 0 ? 1 : 0
Two-digit numbers
For two-digit numbers, there are two types. For numbers of the form 1X,
10: 2
11: 4
12: 5
...
19: 12
You can think of it like this: counting up to 1X requires a number of 1's equal to
f(9) (from counting up to 9) plus
1 (from 10) plus
X (from the first digits of 11-1X inclusive, if X > 0) plus
however many 1's were required to count up to X
Or mathematically,
f(1X) = f(9) + 1 + X + f(X)
Then there are the two-digit numbers higher than 19:
21: 13
31: 14
...
91: 20
The number of 1's required to count to a two-digit number YX with Y > 1 is
f(19) (from counting up to 19) plus
f(9) * (Y - 2) (from the 1's in numbers 20 through (Y-1)9 inclusive - like if Y = 5, I mean the 1's in 20-49, which come from 21, 31, 41) plus
however many 1's were required to count up to X
Or mathematically, for Y > 1,
f(YX) = f(19) + f(9) * (Y - 2) + f(X)
= f(9) + 1 + 9 + f(9) + f(9) * (Y - 2) + f(X)
= 10 + f(9) * Y + f(X)
Three-digit numbers
Once you get into three-digit numbers, you can kind of extend the pattern. For any three-digit number of the form 1YX (and now Y can be anything), the total count of 1's from counting up to that number will be
f(99) (from counting up to 99) plus
1 (from 100) plus
10 * Y + X (from the first digits of 101-1YX inclusive) plus
however many 1's were required to count up to YX in two-digit numbers
so
f(1YX) = f(99) + 1 + YX + f(YX)
Note the parallel to f(1X). Continuing the logic to more digits, the pattern, for numbers which start with 1, is
f(1[m-digits]) = f(10^m - 1) + 1 + [m-digits] + f([m-digits])
with [m-digits] representing a sequence of digits of length m.
Now, for three-digit numbers ZYX that don't start with 1, i.e. Z > 1, the number of 1's required to count up to them is
f(199) (from counting up to 199) plus
f(99) * (Z - 2) (from the 1's in 200-(Z-1)99 inclusive) plus
however many 1's were required to count up to YX
so
f(ZYX) = f(199) + f(99) * (Z - 2) + f(YX)
= f(99) + 1 + 99 + f(99) + f(99) * (Z - 2) + f(YX)
= 100 + f(99) * Z + f(YX)
And the pattern for numbers that don't start with 1 now seems to be clear:
f(Z[m-digits]) = 10^m + f(10^m - 1) * Z + f([m-digits])
General case
We can combine the last result with the formula for numbers that do start with 1. You should be able to verify that the following formula is equivalent to the appropriate case given above for all digits Z 1-9, and that it does the right thing when Z == 0:
f(Z[m-digits]) = f(10^m - 1) * Z + f([m-digits])
+ (Z > 1) ? 10^m : Z * ([m-digits] + 1)
And for numbers of the form 10^m - 1, like 99, 999, etc. you can directly evaluate the function:
f(10^m - 1) = m * 10^(m-1)
because the digit 1 is going to be used 10^(m-1) times in each of the m digits - for example, when counting up to 999, there will be 100 1's used in the hundreds' place, 100 1's used in the tens' place, and 100 1's used in the ones' place. So this becomes
f(Z[m-digits]) = Z * m * 10^(m-1) + f([m-digits])
+ (Z > 1) ? 10^m : Z * ([m-digits] + 1)
You can tinker with the exact expression, but I think this is pretty close to as good as it gets, for this particular approach anyway. What you have here is a recursion relation that allows you to evaluate f(n), the number of 1's required to count up to n, by stripping off a leading digit at each step. Its time complexity is logarithmic in n.
Implementation
Implementing this function is straightforward given the last formula above. You can technically get away with one base case in the recursion: the empty string, i.e. define f("") to be 0. But it will save you a few calls to also handle single digits as well as numbers of the form 10^m - 1. Here's how I'd do it, omitting a bit of argument validation:
private static Pattern nines = Pattern.compile("9+");
/** Return 10^m for m=0,1,...,18 */
private long pow10(int m) {
// implement with either pow(10, m) or a switch statement
}
public long f(String n) {
int Z = Integer.parseInt(n.substring(0, 1));
int nlen = n.length();
if (nlen == 1) {
return Z > 0 ? 1 : 0;
}
if (nines.matcher(n).matches()) {
return nlen * pow10(nlen - 1);
}
String m_digits = n.substring(1);
int m = nlen - 1;
return Z * m * pow10(m - 1) + f_impl(m_digits)
+ (Z > 1 ? pow10(m) : Z * (Long.parseLong(m_digits) + 1));
}
Inverting
This algorithm solves the inverse of the the question you're asking: that is, it figures out how many times a digit is used counting up to n, whereas you want to know which n you can reach with a given number N of digits (i.e. 1's). So, as I mentioned back in the beginning, you're looking for the first n for which f(n+1) > N.
The most straightforward way to do this is to just start counting up from n = 0 and see when you exceed N.
public long howHigh(long N) {
long n = 0;
while (f(n+1) <= N) { n++; }
return n;
}
But of course that's no better (actually probably worse) than accumulating counts in an array. The whole point of having f is that you don't have to test every number; you can jump up by large intervals until you find an n such that f(n+1) > N, and then narrow down your search using the jumps. A reasonably simple method I'd recommend is exponential search to put an upper bound on the result, followed by a binary search to narrow it down:
public long howHigh(long N) {
long upper = 1;
while (f(upper + 1) <= N) {
upper *= 2;
}
long lower = upper / 2, mid = -1;
while (lower < upper) {
mid = (lower + upper) / 2;
if (f(mid + 1) > N) {
upper = mid;
}
else {
lower = mid + 1;
}
}
return lower;
}
Since the implementation of f from above is O(log(n)) and exponential+binary search is also O(log(n)), the final algorithm should be something like O(log^2(n)), and I think the relation between N and n is linear enough that you could consider it O(log^2(N)) too. If you search in log space and judiciously cache computed values of the function, it might be possible to bring it down to roughly O(log(N)). A variant that might provide a significant speedup is sticking in a round of interpolation search after determining the upper bound, but that's tricky to code properly. Fully optimizing the search algorithm is probably a matter for another question though.

This should be more efficient. Use integer array of size 10 to keep the count of digits.
public static int getMaxNumber(int N) {
int[] counts = new int[10];
int number = 0;
boolean limitReached = false;
while (!limitReached) {
number++;
char[] digits = Integer.toString(number).toCharArray();
for (char digit : digits) {
int count = counts[digit - '0'];
count++;
counts[digit - '0'] = count;
if (count >= N) {
limitReached = true;
}
}
}
return number;
}
UPDATE 1: As #Modus Tollens mentioned initial code has a bug. When N = 3 it returns 11, but there are four 1s between 1 and 11. The fix is to check if limit is breached count[i] > N on given number, previous number should be return. But if for some i count[i] == N for other j count[j] <= N, the actual number should be returned.
Please see corresponding code below:
public static int getMaxNumber(int N) {
int[] counts = new int[10];
int number = 0;
while (true) {
number++;
char[] digits = Integer.toString(number).toCharArray();
boolean limitReached = false;
for (char digit : digits) {
int count = counts[digit - '0'];
count++;
counts[digit - '0'] = count;
if (count == N) {
//we should break loop if some count[i] equals to N
limitReached = true;
} else if (count > N) {
//previous number should be returned immediately
//, if current number gives more unique digits than N
return number - 1;
}
}
if (limitReached) {
return number;
}
}
}
UPDATE 2: As #David Z and #Modus Tollens mentioned, in case if N=13, 30 should be returned, ie, algo stops when N is breached but not reached. If this is initial requirement, the code will be even simpler:
public static int getMaxNumber(int N) {
int[] counts = new int[10];
int number = 0;
while (true) {
number++;
char[] digits = Integer.toString(number).toCharArray();
for (char digit : digits) {
int count = counts[digit - '0'];
count++;
counts[digit - '0'] = count;
if (count > N) {
return number - 1;
}
}
}
}

Given a stock of integers 0-9, what is the last number I can write before I run out of some integer?

As the title says, given a stock of integers 0-9, what is the last number I can write before I run out of some integer?
So if I'm given a stock of, say 10 for every number from 0 to 9, what is the last number I can write before I run out of some number. For example, with a stock of 2 I can write numbers 1 ... 10:
1 2 3 4 5 6 7 8 9 10
at this point my stock for ones is 0, and I cannot write 11.
Also note that if I was given a stock of 3, I could still write only numbers 1 ... 10, because 11 would cost me 2 ones, which would leave my stock for ones at -1.
What I have come up so far:
public class Numbers {
public static int numbers(int stock) {
int[] t = new int[10];
for (int k = 1; ; k++) {
int x = k;
while (x > 0) {
if (t[x % 10] == stock) return k-1;
t[x % 10]++;
x /= 10;
}
}
}
public static void main(String[] args) {
System.out.println(numbers(4));
}
}
With this I can get the correct answer for fairly big stock sizes. With a stock size of 10^6 the code completes in ~2 seconds, and with a stock of 10^7 numbers it takes a whole 27 seconds. This is not good enough, since I'm looking for a solution that can handle stock sizes of as big as 10^16, so I probably need a O(log(n)) solution.
This is a homework like assignment, so I didn't come here without wrestling with this pickle for quite some time. I have failed to come up with anything similiar by googling, and wolfram alpha doesn't recognize any kind of pattern this gives.
What I have concluded so far is that ones will allways run out first. I have no proof, but it is so.
Can anyone come up with any piece of advice? Thanks a lot.
EDIT:
I have come up with and implemented an efficient way of finding the cost of numbers 1...n thanks to btilly's pointers (see his post and comments below. also marked as a solution). I will elaborate this further after I have implemented the binary search for finding the last number you can write with the given stock later today.
EDIT 2: The Solution
I had completely forgotten about this post, so my apologies for not editing in my solution earlier. I won't copy the actual implementation, though.
My code for finding the cost of a number does the following:
First, let us choose a number, e.g. 9999. Now we will get the cost by summing the cost of each tens of digits like so:
9 9 9 9
^ ^ ^ ^
^ ^ ^ roof(9999 / 10^1) * 10^0 = 1000
^ ^ roof(9999 / 10^2) * 10^1 = 1000
^ roof(9999 / 10^3) * 10^2 = 1000
roof(9999 / 10^4) * 10^3 = 1000
Thus the cost for 9999 is 4000.
the same for 256:
2 5 6
^ ^ ^
^ ^ roof(256 / 10^1) * 10^0 = 26
^ roof(256 / 10^2) * 10^1 = 30
roof(256 / 10^3) * 10^2 = 100
Thus the cost for 256 is 156.
Implementing with this idea would make the program work only with numbers that have no digits 1 or 0, which is why further logic is needed. Let's call the method explained above C(n, d), where n is the number for which we are getting the cost for, and d is the d'th digit from n that we are currently working with. Let's also define a method D(n, d) that will return the d'th digit from n. Then we will apply the following logic:
sum = C(n, d)
if D(n, d) is 1:
for each k < d, k >= 0 :
sum -= ( 9 - D(n, k) ) * 10^(k-1);
else if D(n, d) is 0:
sum -= 10^(d-1)
With this the program will calculate the correct cost for a number efficiently. After this we simply apply a binary search for finding the number with the correct cost.

Step 1. Write an efficient function to calculate how much stock needs to be used to write all of the numbers up to N. (Hint: calculate everything that was used to write out the numbers in the last digit with a formula, and then use recursion to calculate everything that was used in the other digits.)
Step 2. Do a binary search to find the last number you can write with your amount of stock.

We can calculate the answer directly. A recursive formula can determine how much stock is needed to get from 1 to numbers that are powers of ten minus 1:
f(n, power, target){
if (power == target)
return 10 * n + power;
else
return f(10 * n + power, power * 10, target);
}
f(0,1,1) = 1 // a stock of 1 is needed for the numbers from 1 to 9
f(0,1,10) = 20 // a stock of 20 is needed for the numbers from 1 to 99
f(0,1,100) = 300 // a stock of 300 is needed for the numbers from 1 to 999
f(0,1,1000) = 4000 // a stock of 4000 is needed for the numbers from 1 to 9999
Where it gets complicated is accounting for the extra 1's needed when our calculation lands after the first multiple of any of the above coefficients; for example, on the second multiple of 10 (11-19) we need an extra 1 for each number.
JavaScript code:
function f(stock){
var cs = [0];
var p = 1;
function makeCoefficients(n,i){
n = 10*n + p;
if (n > stock){
return;
} else {
cs.push(n);
p *= 10;
makeCoefficients(n,i*10);
}
}
makeCoefficients(0,1);
var result = -1;
var numSndMul = 0;
var c;
while (stock > 0){
if (cs.length == 0){
return result;
}
c = cs.pop();
var mul = c + p * numSndMul;
if (stock >= mul){
stock -= mul;
result += p;
numSndMul++;
if (stock == 0){
return result;
}
}
var sndMul = c + p * numSndMul;
if (stock >= sndMul){
stock -= sndMul;
result += p;
numSndMul--;
if (stock == 0){
return result;
}
var numMul = Math.floor(stock / mul);
stock -= numMul * mul;
result += numMul * p;
}
p = Math.floor(p/10);
}
return result;
}
Output:
console.log(f(600));
1180
console.log(f(17654321));
16031415
console.log(f(2147483647));
1633388154

More efficient solution: Project Euler #2: Even Fibonacci Numbers

Problem:
Each new term in the Fibonacci sequence is generated by adding the
previous two terms.
By starting with 1 and 2, the first 10 terms will
be:
1, 2, 3, 5, 8, 13, 21, 34, 55, 89, ...
By considering the terms in the Fibonacci sequence whose values do not
exceed four million, find the sum of the even-valued terms.
My code: (which works fine)
public static void main(String[] agrs){
int prevFirst=0;
int prevSecond=1;
int bound=4_000_000;
int evenSum=0;
boolean exceed=false; //when fib numbers > bound
while(!exceed){
int newFib=prevFirst + prevSecond;
prevFirst = prevSecond;
prevSecond = newFib;
if(newFib > bound){
exceed=true;
break;
}
if(newFib % 2 == 0){
evenSum += newFib;
}
}
System.out.println(evenSum);
}
I'm looking for a more efficient algorithm to do this question. Any hints?

When taking the following rules into account:
even + even = even
even + odd = odd
odd + even = odd
odd + odd = even
The parity of the first Fibonacci numbers is:
o o e o o e o o e ...
Thus basically, you simply need to do steps of three. Which is:
(1,1,2)
(3,5,8)
(13,21,34)
Given (a,b,c) this is (b+c,b+2*c,2*b+3*c).
This means we only need to store the two last numbers, and calculate given (a,b), (a+2*b,2*a+3*b).
Thus (1,2) -> (5,8) -> (21,34) -> ... and always return the last one.
This will work faster than a "filter"-approach because that uses the if-statement which reduces pipelining.
The resulting code is:
int b = 1;
int c = 2, d;
long sum = 0;
while(c < 4000000) {
sum += c;
d = b+(c<<0x01);
c = d+b+c;
b = d;
}
System.out.println(sum);
Or the jdoodle (with benchmarking, takes 5 microseconds with cold start, and on average 50 nanoseconds, based on the average of 1M times). Of course the number of instructions in the loop is larger. But the loop is repeated one third of the times.

You can't improve it much more, any improvement that you'll do will be negligible as well as depended on the OS you're running on.
Example:
Running your code in a loop 1M times on my Mac too 73-75ms (ran it a few times).
Changing the condition:
if(newFib % 2 == 0){
to:
if((newFib & 1) == 0){
and running it again a few times I got 51-54ms.
If you'll run the same thing on a different OS you might (and
probably will) get different results.
even if we'll consider the above as an improvement, divide ~20ms in 1M and the "improvement" that you'll get for a single run is meaningless (~20 nanos).

assuming consecutive Fibonacci numbers
a, b,
c = a + b,
d = a + 2b,
e = 2a + 3b,
f = 3a + 5b,
g = 5a + 8b = a + 4(a + 2b) = a + 4d,
it would seem more efficient to use
ef0 = 0, ef1 = 2, efn = efn-2 + 4 efn-1

as I mentioned in my comment there is really no need to further improvement.
I did some measurements
looped 1000000 times the whole thing
measure time in [ms]
ms / 1000000 = ns
so single pass times [ns] are these:
[176 ns] - exploit that even numbers are every third
[179 ns] - &1 instead of %2
[169 ns] - &1 instead of %2 and eliminated if by -,^,&
[edit1] new code
[105 ns] - exploit that even numbers are every third + derived double iteration of fibonaci
[edit2] new code
[76 ns] - decreased operand count to lower overhead and heap trashing
the last one clearly wins on mine machine (although I would expect that the first one will be best)
all was tested on Win7 x64 AMD A8-5500 3.2GHz
App with no threads 32-bit compiler BDS2006 Trubo C++
1,2 are nicely mentioned in Answers here already so I comment just 3:
s+=a&(-((a^1)&1));
(a^1) negates lovest bit
((a^1)&1) is 1 for even and 0 for odd a
-((a^1)&1)) is -1 for even and 0 for odd a
-1 is 0xFFFFFFFF so anding number by it will not change it
0 is 0x00000000 so anding number by it will be 0
hence no need for if
also instead of xor (^) you can use negation (!) but that is much slower on mine machine
OK here is the code (do not read further if you want to code it your self):
//---------------------------------------------------------------------------
int euler002()
{
// Each new term in the Fibonacci sequence is generated by adding the previous two terms.
// By starting with 1 and 2, the first 10 terms will be: 1, 2, 3, 5, 8, 13, 21, 34, 55, 89, ...
// By considering the terms in the Fibonacci sequence whose values do not exceed four million, find the sum of the even-valued terms.
int a,a0=0,a1=1,s=0,N=4000000;
/*
//1. [176 ns]
a=a0+a1; a0=a1; a1=a; // odd
a=a0+a1; a0=a1; a1=a; // even
for (;a<N;)
{
s+=a;
a=a0+a1; a0=a1; a1=a; // odd
a=a0+a1; a0=a1; a1=a; // odd
a=a0+a1; a0=a1; a1=a; // even
}
//2. [179 ns]
for (;;)
{
a=a0+a1; a0=a1; a1=a;
if (a>=N) break;
if ((a&1)==0) s+=a;
}
//3. [169 ns]
for (;;)
{
a=a0+a1; a0=a1; a1=a;
if (a>=N) break;
s+=a&(-((a^1)&1));
}
//4. [105 ns] // [edit1]
a0+=a1; a1+=a0; a=a1; // 2x
for (;a<N;)
{
s+=a; a0+=a1; a1+=a0; // 2x
a=a0+a1; a0=a1; a1=a; // 1x
}
*/
//5. [76 ns] //[ edit2]
a0+=a1; a1+=a0; // 2x
for (;a1<N;)
{
s+=a1; a0+=a1; a1+=a0; // 2x
a=a0; a0=a1; a1+=a; // 1x
}
return s;
}
//---------------------------------------------------------------------------
[edit1] faster code add
CommuSoft suggested to iterate more then 1 number per iteration of fibonaci to minimize operations.
nice idea but code in his comment does not give correct answers
I tweaked a little mine so here is the result:
[105 ns] - exploit that even numbers are every third + derived double iteration of fibonaci
this is almost twice the speedup of 1. from which it is derived
look for [edit1] in code or look for //4.
[edit2] even faster code add
- just reorder of some variable and operation use for more speed
- [76 ns] decreased operand count to lower overhead and heap trashing

if you check Fibonacci series, for even numbers 2 8 34 144 610 you can see that there is a fantastic relation between even numbers, for example:
34 = 4*8 + 2,
144 = 34*4 + 8,
610 = 144*4 + 34;
this means that next even in Fibonacci can be expressed like below
Even(n)=4*Even(n-1)+E(n-2);
in Java
public static void main(String[] args) {
Scanner in = new Scanner(System.in);
int t = in.nextInt();
for(int a0 = 0; a0 < t; a0++){
long n = in.nextLong();
long a=2;
long b=8;
long c=0;
long sum=10;
while(b<n)
{
sum +=c;
c=b*4+a;
a=b;
b=c;
}
System.out.println(sum);
}
}

F(n) be the nth Fibonnaci number i.e F(n)=F(n-1)+F(n-2)
Lets say that F(n) is even, then
F(n) = F(n-1) + F(n-2) = F(n-2) + F(n-3) + F(n-2)
F(n) = 2F(n-2) + F(n-3)
--This proves the point that every third term is even (if F(n-3) is even, then F(n) must be even too)
F(n) = 2[F(n-3) + F(n-4)] + F(n-3)
= 3F(n-3) + 2F(n-4)
= 3F(n-3) + 2F(n-5) + 2F(n-6)
From eq.1:
F(n-3) = 2F(n-5) + F(n-6)
2F(n-5) = F(n-3) - F(n-6)
F(n) = 3F(n-3) + [F(n-3) - F(n-6)] + 2F(n-6)
= 4F(n-3) + F(n-6)
If the sequence of even numbers consists of every third number (n, n-3, n-6, ...)
Even Fibonacci sequence:
E(k) = 4E(k-1) + E(k-2)
Fib Sequence F= {0,1,1,2,3,5,8.....}
Even Fib Sequence E={0,2,8,.....}
CODE:
public static long findEvenFibSum(long n){
long term1=0;
long term2=2;
long curr=0;
long sum=term1+term2;
while((curr=(4*term2+term1))<=n){
sum+=curr;
term1=term2;
term2=curr;
}
return sum;
}

The answer for project Euler problem 2 is(in Java):
int x = 0;
int y = 1;
int z = x + y;
int sumeven = 0;
while(z < 4000000){
x = y;
y = z;
z = x + y;
if(z % 2 == 0){
sumeven += z; /// OR sumeven = sumeven + z
}
}
System.out.printf("sum of the even-valued terms: %d \n", sumeven);
This is the easiest answer.

How to find the next lower integer (with the same number of 1s)

How to find the next lower binary number for an integer (same number of 1s)? For example: if given input number n = 10 (1010), the function should return 9 (1001), or n = 14 (1110) then return 13 (1101), or n = 22 (10110) then return 21 (10101), n = 25 (11001) then return 22 (10110)... etc.

You can do this.
static int nextLower(int n) {
int bc = Integer.bitCount(n);
for (int i = n - 1; i > 0; i--)
if (Integer.bitCount(i) == bc)
return i;
throw new RuntimeException(n+" is the lowest with a bit count of "+bc);
}
Of course if this is homework you are going to have trouble convincing someone you wrote this ;)

For the sake of clarity, in this answer I will use the term 'cardinality' to indicate the number of 1s in the binary representation of a number.
One (obvious) way is to run a downwards loop, and seek for the first number with the same cardinality as your input (just like Peter Lawrey suggested).
I don't think this is inefficient, because I guess the output number is always pretty close to the input. More precisely, all you have to do is to find the rightmost '10' bit sequence, and change it to '01'. Then replace the right part with a number having all 1s at its left, as many as you can, without breaking the postcondition. This brings us to another solution, which consists in converting the number to a binary string (like user2573153 showed you), performing the replacement (with a regular expression, maybe), and then converting back to int.
A slightly faster version of Peter's algorithm should be the following, which performs on integers the manipulation I proposed you for strings:
static int nextLower(int n) {
int fixPart = 0;
int shiftCount = 0;
while ((n & 3) != 2) {
if (n == 0) {
throw new IllegalArgumentException(
fixPart + " is the lowest number with its cardinality");
}
fixPart |= (n & 1) << shiftCount;
shiftCount += 1;
n /= 2;
}
int fixZeros = shiftCount - Integer.bitCount(fixPart);
return ((n ^ 3) << shiftCount) | (((1 << shiftCount) - 1) & ~((1 << fixZeros) - 1));
}
which is O(log n) rather than O(n), but it's definitely harder to understand, and may also be practically slower, due to its complexity. Anyway, you could only notice a difference if you try with some huge difficult number.
EDIT I tried a little benchmark, and found that this code is 67% faster than Peter Lawrey's when applied consecutively to all numbers from 2 to 100,000,000. I don't think this is enough to justify the increased code complexity.

I like such binary task, so to find next lower number you should find right most 1 followed by 0 and exchange them,. UPDATE: you need to "reorder" the rest part of number with 1s at left and 0s at right
10 1010 ->
9 1001
14 1110 ->
13 1101
25 11001 ->
22 10110
here is sample code:
int originalValue = 25;
int maskToCheck = 2; // in binary 10b
int clearingMask = 1;
int settingMask = 0;
int zeroCount = 0;
while (maskToCheck > 0)
{
if ( (originalValue&(maskToCheck|(maskToCheck>>1))) == maskToCheck ) // we found such
{
int newValue = originalValue&(~maskToCheck); // set 1 with 0
newValue = newValue&(~clearingMask)|(settingMask<<zeroCount); // clear all the rest bits, and set most valuable ones
newValue = newValue|(maskToCheck>>1); // set 0 with 1
System.out.println("for " + originalValue + " we found " + newValue);
break;
}
else
{
if ( (originalValue&(maskToCheck>>1)) > 0) // we have 1 bit in cleared part
settingMask = (settingMask<<1) | 1;
else
zeroCount++;
maskToCheck = maskToCheck<<1; // try next left bits
clearingMask = (clearingMask<<1)|1;
}
}