Movatterモバイル変換

[0]ホーム

Jump to content

Dynamic programming

Edit links

From Wikipedia, the free encyclopedia

Problem optimization method

Not to be confused withDynamic programming language orDynamic problem.

**Figure 1.** Finding the shortest path in a graph using optimal substructure; a straight line indicates a single edge; a wavy line indicates a shortest path between the two vertices it connects (among other paths, not shown, sharing the same two vertices); the bold line is the overall shortest path from start to goal.

Dynamic programming is both amathematical optimization method and analgorithmic paradigm. The method was developed byRichard Bellman in the 1950s and has found applications in numerous fields, such asaerospace engineering andeconomics.

In both contexts it refers to simplifying a complicated problem by breaking it down into simpler sub-problems in arecursive manner. While some decision problems cannot be taken apart this way, decisions that span several points in time do often break apart recursively. Likewise, in computer science, if a problem can be solved optimally by breaking it into sub-problems and then recursively finding the optimal solutions to the sub-problems, then it is said to haveoptimal substructure.

If sub-problems can be nested recursively inside larger problems, so that dynamic programming methods are applicable, then there is a relation between the value of the larger problem and the values of the sub-problems.^[1] In the optimization literature this relationship is called theBellman equation.

Overview

[edit]

Mathematical optimization

[edit]

In terms of mathematical optimization, dynamic programming usually refers to simplifying a decision by breaking it down into a sequence of decision steps over time.

This is done by defining a sequence ofvalue functionsV₁,V₂, ...,V_n takingy as an argument representing thestate of the system at timesi from 1 ton.

The definition ofV_n(y) is the value obtained in statey at the last timen.

The valuesV_i at earlier timesi = n −1, n − 2, ..., 2, 1 can be found by working backwards, using arecursive relationship called theBellman equation.

Fori = 2, ..., n,V_i−1 at any statey is calculated fromV_i by maximizing a simple function (usually the sum) of the gain from a decision at timei − 1 and the functionV_i at the new state of the system if this decision is made.

SinceV_i has already been calculated for the needed states, the above operation yieldsV_i−1 for those states.

Finally,V₁ at the initial state of the system is the value of the optimal solution. The optimal values of the decision variables can be recovered, one by one, by tracking back the calculations already performed.

Control theory

[edit]

Incontrol theory, a typical problem is to find an admissible control $\mathbf {u} ^{\ast }$ which causes the system ${\dot {\mathbf {x} }}(t)=\mathbf {g} \left(\mathbf {x} (t),\mathbf {u} (t),t\right)$ to follow an admissible trajectory $\mathbf {x} ^{\ast }$ on a continuous time interval $t_{0}\leq t\leq t_{1}$ that minimizes acost function

J=b\left(\mathbf {x} (t_{1}),t_{1}\right)+\int _{t_{0}}^{t_{1}}f\left(\mathbf {x} (t),\mathbf {u} (t),t\right)\mathrm {d} t

The solution to this problem is an optimal control law or policy $\mathbf {u} ^{\ast }=h(\mathbf {x} (t),t)$ , which produces an optimal trajectory $\mathbf {x} ^{\ast }$ and acost-to-go function $J^{\ast }$ . The latter obeys the fundamental equation of dynamic programming:

-J_{t}^{\ast }=\min _{\mathbf {u} }\left\{f\left(\mathbf {x} (t),\mathbf {u} (t),t\right)+J_{x}^{\ast {\mathsf {T}}}\mathbf {g} \left(\mathbf {x} (t),\mathbf {u} (t),t\right)\right\}

apartial differential equation known as theHamilton–Jacobi–Bellman equation, in which $J_{x}^{\ast }={\frac {\partial J^{\ast }}{\partial \mathbf {x} }}=\left[{\frac {\partial J^{\ast }}{\partial x_{1}}}~~~~{\frac {\partial J^{\ast }}{\partial x_{2}}}~~~~\dots ~~~~{\frac {\partial J^{\ast }}{\partial x_{n}}}\right]^{\mathsf {T}}$ and $J_{t}^{\ast }={\frac {\partial J^{\ast }}{\partial t}}$ . One finds that minimizing $\mathbf {u}$ in terms of $t {\displaystyle t}$ , $\mathbf {x}$ , and the unknown function $J_{x}^{\ast }$ and then substitutes the result into the Hamilton–Jacobi–Bellman equation to get the partial differential equation to be solved with boundary condition $J\left(t_{1}\right)=b\left(\mathbf {x} (t_{1}),t_{1}\right)$ .^[2] In practice, this generally requiresnumerical techniques for some discrete approximation to the exact optimization relationship.

Alternatively, the continuous process can be approximated by a discrete system, which leads to a following recurrence relation analog to the Hamilton–Jacobi–Bellman equation:

J_{k}^{\ast }\left(\mathbf {x} _{n-k}\right)=\min _{\mathbf {u} _{n-k}}\left\{{\hat {f}}\left(\mathbf {x} _{n-k},\mathbf {u} _{n-k}\right)+J_{k-1}^{\ast }\left({\hat {\mathbf {g} }}\left(\mathbf {x} _{n-k},\mathbf {u} _{n-k}\right)\right)\right\}

at the $k {\displaystyle k}$ -th stage of $n {\displaystyle n}$ equally spaced discrete time intervals, and where ${\hat {f}}$ and ${\hat {\mathbf {g} }}$ denote discrete approximations to $f {\displaystyle f}$ and $\mathbf {g}$ . This functional equation is known as theBellman equation, which can be solved for an exact solution of the discrete approximation of the optimization equation.^[3]

Example from economics: Ramsey's problem of optimal saving

[edit]

Computer science

[edit]

There are two key attributes that a problem must have in order for dynamic programming to be applicable:optimal substructure andoverlapping sub-problems. If a problem can be solved by combining optimal solutions tonon-overlapping sub-problems, the strategy is called "divide and conquer" instead.^[1] This is whymerge sort andquick sort are not classified as dynamic programming problems.

Optimal substructure means that the solution to a given optimization problem can be obtained by the combination of optimal solutions to its sub-problems. Such optimal substructures are usually described by means ofrecursion. For example, given a graphG=(V,E), the shortest pathp from a vertexu to a vertexv exhibits optimal substructure: take any intermediate vertexw on this shortest pathp. Ifp is truly the shortest path, then it can be split into sub-pathsp₁ fromu tow andp₂ fromw tov such that these, in turn, are indeed the shortest paths between the corresponding vertices (by the simple cut-and-paste argument described inIntroduction to Algorithms). Hence, one can easily formulate the solution for finding shortest paths in a recursive manner, which is what theBellman–Ford algorithm or theFloyd–Warshall algorithm does.

Overlapping sub-problems means that the space of sub-problems must be small, that is, any recursive algorithm solving the problem should solve the same sub-problems over and over, rather than generating new sub-problems. For example, consider the recursive formulation for generating the Fibonacci sequence:F_i =F_i−1 +F_i−2, with base caseF₁ = F₂ = 1. ThenF₄₃ = F₄₂ + F₄₁, andF₄₂ = F₄₁ + F₄₀. NowF₄₁ is being solved in the recursive sub-trees of bothF₄₃ as well asF₄₂. Even though the total number of sub-problems is actually small (only 43 of them), we end up solving the same problems over and over if we adopt a naive recursive solution such as this. Dynamic programming takes account of this fact and solves each sub-problem only once.

**Figure 2.** The subproblem graph for the Fibonacci sequence. The fact that it is not atree indicates overlapping subproblems.

This can be achieved in either of two ways:^[4]

Top-down approach: This is the direct fall-out of the recursive formulation of any problem. If the solution to any problem can be formulated recursively using the solution to its sub-problems, and if its sub-problems are overlapping, then one can easilymemoize or store the solutions to the sub-problems in a table (often anarray orhashtable in practice). Whenever we attempt to solve a new sub-problem, we first check the table to see if it is already solved. If a solution has been recorded, we can use it directly, otherwise we solve the sub-problem and add its solution to the table.
Bottom-up approach: Once we formulate the solution to a problem recursively as in terms of its sub-problems, we can try reformulating the problem in a bottom-up fashion: try solving the sub-problems first and use their solutions to build-on and arrive at solutions to bigger sub-problems. This is also usually done in a tabular form by iteratively generating solutions to bigger and bigger sub-problems by using the solutions to small sub-problems. For example, if we already know the values ofF₄₁ andF₄₀, we can directly calculate the value ofF₄₂.

Someprogramming languages can automaticallymemoize the result of a function call with a particular set of arguments, in order to speed upcall-by-name evaluation (this mechanism is referred to ascall-by-need). Some languages make it possible portably (e.g.Scheme,Common Lisp,Perl orD). Some languages have automaticmemoization built in, such as tabledProlog andJ, which supports memoization with theM. adverb.^[5] In any case, this is only possible for areferentially transparent function. Memoization is also encountered as an easily accessible design pattern within term-rewrite based languages such asWolfram Language.

Bioinformatics

[edit]

Dynamic programming is widely used in bioinformatics for tasks such assequence alignment,protein folding, RNA structure prediction and protein-DNA binding. The first dynamic programming algorithms for protein-DNA binding were developed in the 1970s independently byCharles DeLisi in the US^[6] and by Georgii Gurskii and Alexander Zasedatelev in theSoviet Union.^[7] Recently these algorithms have become very popular in bioinformatics andcomputational biology, particularly in the studies ofnucleosome positioning andtranscription factor binding.

Examples: computer algorithms

[edit]

Dijkstra's algorithm for the shortest path problem

[edit]

From a dynamic programming point of view,Dijkstra's algorithm for theshortest path problem is a successive approximation scheme that solves the dynamic programming functional equation for the shortest path problem by theReaching method.^[8]^[9]^[10]

In fact, Dijkstra's explanation of the logic behind the algorithm,^[11] namely

Problem 2. Find the path of minimum total length between two given nodes $P {\displaystyle P}$ and $Q {\displaystyle Q}$ .
We use the fact that, if $R {\displaystyle R}$ is a node on the minimal path from $P {\displaystyle P}$ to $Q {\displaystyle Q}$ , knowledge of the latter implies the knowledge of the minimal path from $P {\displaystyle P}$ to $R {\displaystyle R}$ .

is a paraphrasing ofBellman's famousPrinciple of Optimality in the context of theshortest path problem.

Fibonacci sequence

[edit]

Using dynamic programming in the calculation of thenth member of theFibonacci sequence improves its performance greatly. Here is a naïve implementation, based directly on the mathematical definition:

function fib(n)if n <= 1return nreturn fib(n − 1) + fib(n − 2)

Notice that if we call, say,fib(5), we produce a call tree that calls the function on the same value many different times:

fib(5)
fib(4) + fib(3)
(fib(3) + fib(2)) + (fib(2) + fib(1))
((fib(2) + fib(1)) + (fib(1) + fib(0))) + ((fib(1) + fib(0)) + fib(1))
(((fib(1) + fib(0)) + fib(1)) + (fib(1) + fib(0))) + ((fib(1) + fib(0)) + fib(1))

In particular,fib(2) was calculated three times from scratch. In larger examples, many more values offib, orsubproblems, are recalculated, leading to an exponential time algorithm.

Now, suppose we have a simplemap object,m, which maps each value offib that has already been calculated to its result, and we modify our function to use it and update it. The resulting function requires onlyO(n) time instead of exponential time (but requiresO(n) space):

var m :=map(0 → 0, 1 → 1)function fib(n)ifkey nis not inmap m         m[n] := fib(n − 1) + fib(n − 2)return m[n]

This technique of saving values that have already been calculated is calledmemoization; this is the top-down approach, since we first break the problem into subproblems and then calculate and store values.

In thebottom-up approach, we calculate the smaller values offib first, then build larger values from them. This method also uses O(n) time since it contains a loop that repeats n − 1 times, but it only takes constant (O(1)) space, in contrast to the top-down approach which requires O(n) space to store the map.

function fib(n)if n = 0return 0elsevar previousFib := 0, currentFib := 1repeat n − 1times// loop is skipped if n = 1var newFib := previousFib + currentFib            previousFib := currentFib            currentFib  := newFibreturn currentFib

In both examples, we only calculatefib(2) one time, and then use it to calculate bothfib(4) andfib(3), instead of computing it every time either of them is evaluated.

A type of balanced 0–1 matrix

[edit]

This sectiondoes notcite anysources. Please helpimprove this section byadding citations to reliable sources. Unsourced material may be challenged andremoved.(May 2013) (Learn how and when to remove this message)

Consider the problem of assigning values, either zero or one, to the positions of ann ×n matrix, withn even, so that each row and each column contains exactlyn / 2 zeros andn / 2 ones. We ask how many different assignments there are for a givenn. For example, whenn = 4, five possible solutions are

{\begin{bmatrix}0&1&0&1\\1&0&1&0\\0&1&0&1\\1&0&1&0\end{bmatrix}}{\text{ and }}{\begin{bmatrix}0&0&1&1\\0&0&1&1\\1&1&0&0\\1&1&0&0\end{bmatrix}}{\text{ and }}{\begin{bmatrix}1&1&0&0\\0&0&1&1\\1&1&0&0\\0&0&1&1\end{bmatrix}}{\text{ and }}{\begin{bmatrix}1&0&0&1\\0&1&1&0\\0&1&1&0\\1&0&0&1\end{bmatrix}}{\text{ and }}{\begin{bmatrix}1&1&0&0\\1&1&0&0\\0&0&1&1\\0&0&1&1\end{bmatrix}}.

There are at least three possible approaches:brute force,backtracking, and dynamic programming.

Brute force consists of checking all assignments of zeros and ones and counting those that have balanced rows and columns (n / 2 zeros andn / 2 ones). As there are $2^{n^{2}}$ possible assignments and ${\tbinom {n}{n/2}}^{n}$ sensible assignments, this strategy is not practical for arbitrarily large values of $n {\displaystyle n}$ .

Backtracking for this problem consists of choosing some order of the matrix elements and recursively placing ones or zeros, while checking that in every row and column the number of elements that have not been assigned plus the number of ones or zeros are both at leastn / 2. While more sophisticated than brute force, this approach will visit every solution once, making it impractical forn larger than six, since the number of solutions is already116963796250 forn = 8, as we shall see.

Dynamic programming makes it possible to count the number of solutions without visiting them all. Imagine backtracking values for the first row – what information would we require about the remaining rows, in order to be able to accurately count the solutions obtained for each first row value? We considerk ×n boards, where1 ≤k ≤n, whosek rows contain $n/2$ zeros and $n/2$ ones. The functionf to whichmemoization is applied maps vectors ofn pairs of integers to the number of admissible boards (solutions). There is one pair for each column, and its two components indicate respectively the number of zeros and ones that have yet to be placed in that column. We seek the value of $f((n/2,n/2),(n/2,n/2),\ldots (n/2,n/2))$ (n arguments or one vector ofn elements). The process of subproblem creation involves iterating over every one of ${\tbinom {n}{n/2}}$ possible assignments for the top row of the board, and going through every column, subtracting one from the appropriate element of the pair for that column, depending on whether the assignment for the top row contained a zero or a one at that position. If any one of the results is negative, then the assignment is invalid and does not contribute to the set of solutions (recursion stops). Otherwise, we have an assignment for the top row of thek ×n board and recursively compute the number of solutions to the remaining(k − 1) ×n board, adding the numbers of solutions for every admissible assignment of the top row and returning the sum, which is being memoized. The base case is the trivial subproblem, which occurs for a1 ×n board. The number of solutions for this board is either zero or one, depending on whether the vector is a permutation ofn / 2(0, 1) andn / 2(1, 0) pairs or not.

For example, in the first two boards shown above the sequences of vectors would be

((2, 2) (2, 2) (2, 2) (2, 2))       ((2, 2) (2, 2) (2, 2) (2, 2))     k = 4  0      1      0      1              0      0      1      1((1, 2) (2, 1) (1, 2) (2, 1))       ((1, 2) (1, 2) (2, 1) (2, 1))     k = 3  1      0      1      0              0      0      1      1((1, 1) (1, 1) (1, 1) (1, 1))       ((0, 2) (0, 2) (2, 0) (2, 0))     k = 2  0      1      0      1              1      1      0      0((0, 1) (1, 0) (0, 1) (1, 0))       ((0, 1) (0, 1) (1, 0) (1, 0))     k = 1  1      0      1      0              1      1      0      0((0, 0) (0, 0) (0, 0) (0, 0))       ((0, 0) (0, 0), (0, 0) (0, 0))

The number of solutions (sequenceA058527 in theOEIS) is

1, 2, 90,297200,116963796250,6736218287430460752, ...

Links to the MAPLE implementation of the dynamic programming approach may be found among theexternal links.

Checkerboard

[edit]

Consider acheckerboard withn ×n squares and a cost functionc(i, j) which returns a cost associated with square(i,j) (i being the row,j being the column). For instance (on a 5 × 5 checkerboard),

	1	2	3	4	5
5	6	7	4	7	8
4	7	6	1	1	4
3	3	5	7	8	2
2	–	6	7	0	–
1	–	–	5	–	–

Thusc(1, 3) = 5

Let us say there was a checker that could start at any square on the first rank (i.e., row) and you wanted to know the shortest path (the sum of the minimum costs at each visited rank) to get to the last rank; assuming the checker could move only diagonally left forward, diagonally right forward, or straight forward. That is, a checker on(1,3) can move to(2,2),(2,3) or(2,4).

	2	3	4
5
4
3
2	x	x	x
1		o

This problem exhibitsoptimal substructure. That is, the solution to the entire problem relies on solutions to subproblems. Let us define a functionq(i, j) as

q(i,j) = the minimum cost to reach square (i,j).

Starting at rankn and descending to rank1, we compute the value of this function for all the squares at each successive rank. Picking the square that holds the minimum value at each rank gives us the shortest path between rankn and rank1.

The functionq(i, j) is equal to the minimum cost to get to any of the three squares below it (since those are the only squares that can reach it) plusc(i, j). For instance:

	2	3	4
5
4		A
3	B	C	D
2
1

q(A)=\min(q(B),q(C),q(D))+c(A)\,

Now, let us defineq(i, j) in somewhat more general terms:

q(i,j)={\begin{cases}\infty &j<1{\text{ or }}j>n\\c(i,j)&i=1\\\min(q(i-1,j-1),q(i-1,j),q(i-1,j+1))+c(i,j)&{\text{otherwise.}}\end{cases}}

The first line of this equation deals with a board modeled as squares indexed on1 at the lowest bound andn at the highest bound. The second line specifies what happens at the first rank; providing a base case. The third line, the recursion, is the important part. It represents theA,B,C,D terms in the example. From this definition we can derive straightforward recursive code forq(i, j). In the following pseudocode,n is the size of the board,c(i, j) is the cost function, andmin() returns the minimum of a number of values:

function minCost(i, j)if j < 1or j > nreturn infinityelse if i = 1return c(i, j)elsereturnmin( minCost(i-1, j-1), minCost(i-1, j), minCost(i-1, j+1) ) + c(i, j)

This function only computes the path cost, not the actual path. We discuss the actual path below. This, like the Fibonacci-numbers example, is horribly slow because it too exhibits theoverlapping sub-problems attribute. That is, it recomputes the same path costs over and over. However, we can compute it much faster in a bottom-up fashion if we store path costs in a two-dimensional arrayq[i, j] rather than using a function. This avoids recomputation; all the values needed for arrayq[i, j] are computed ahead of time only once. Precomputed values for(i,j) are simply looked up whenever needed.

We also need to know what the actual shortest path is. To do this, we use another arrayp[i, j]; apredecessor array. This array records the path to any squares. The predecessor ofs is modeled as an offset relative to the index (inq[i, j]) of the precomputed path cost ofs. To reconstruct the complete path, we lookup the predecessor ofs, then the predecessor of that square, then the predecessor of that square, and so on recursively, until we reach the starting square. Consider the following pseudocode:

function computeShortestPathArrays()for xfrom 1to n        q[1, x] := c(1, x)for yfrom 1to n        q[y, 0]     := infinity        q[y, n + 1] := infinityfor yfrom 2to nfor xfrom 1to n            m := min(q[y-1, x-1], q[y-1, x], q[y-1, x+1])            q[y, x] := m + c(y, x)if m = q[y-1, x-1]                p[y, x] := -1else if m = q[y-1, x]                p[y, x] :=  0else                p[y, x] :=  1

Now the rest is a simple matter of finding the minimum and printing it.

function computeShortestPath()    computeShortestPathArrays()    minIndex := 1    min := q[n, 1]for ifrom 2to nif q[n, i] < min            minIndex := i            min := q[n, i]    printPath(n, minIndex)

function printPath(y, x)print(x)print("<-")if y = 2print(x + p[y, x])else        printPath(y-1, x + p[y, x])

Sequence alignment

[edit]

Ingenetics,sequence alignment is an important application where dynamic programming is essential.^[12] Typically, the problem consists of transforming one sequence into another using edit operations that replace, insert, or remove an element. Each operation has an associated cost, and the goal is to find thesequence of edits with the lowest total cost.

The problem can be stated naturally as a recursion, a sequence A is optimally edited into a sequence B by either:

inserting the first character of B, and performing an optimal alignment of A and the tail of B
deleting the first character of A, and performing the optimal alignment of the tail of A and B
replacing the first character of A with the first character of B, and performing optimal alignments of the tails of A and B.

The partial alignments can be tabulated in a matrix, where cell (i,j) contains the cost of the optimal alignment of A[1..i] to B[1..j]. The cost in cell (i,j) can be calculated by adding the cost of the relevant operations to the cost of its neighboring cells, and selecting the optimum.

Different variants exist, seeSmith–Waterman algorithm andNeedleman–Wunsch algorithm.

Tower of Hanoi puzzle

[edit]

A model set of the Towers of Hanoi (with 8 disks)

An animated solution of the Tower of Hanoi puzzle forT(4,3)

TheTower of Hanoi orTowers ofHanoi is amathematical game orpuzzle. It consists of three rods, and a number of disks of different sizes which can slide onto any rod. The puzzle starts with the disks in a neat stack in ascending order of size on one rod, the smallest at the top, thus making a conical shape.

The objective of the puzzle is to move the entire stack to another rod, obeying the following rules:

Only one disk may be moved at a time.
Each move consists of taking the upper disk from one of the rods and sliding it onto another rod, on top of the other disks that may already be present on that rod.
No disk may be placed on top of a smaller disk.

The dynamic programming solution consists of solving thefunctional equation

S(n,h,t) = S(n-1,h, not(h,t)) ; S(1,h,t) ; S(n-1,not(h,t),t)

where n denotes the number of disks to be moved, h denotes the home rod, t denotes the target rod, not(h,t) denotes the third rod (neither h nor t), ";" denotes concatenation, and

S(n, h, t) := solution to a problem consisting of n disks that are to be moved from rod h to rod t.

For n=1 the problem is trivial, namely S(1,h,t) = "move a disk from rod h to rod t" (there is only one disk left).

The number of moves required by this solution is 2ⁿ − 1. If the objective is tomaximize the number of moves (without cycling) then the dynamic programmingfunctional equation is slightly more complicated and 3ⁿ − 1 moves are required.^[13]

Egg dropping puzzle

[edit]

A famouspuzzle relates to dropping eggs from a building to determine at which height they start to break. The following is a description involving N=2 eggs and a building with H=36 floors:^[14]

Suppose that we wish to know which stories in a 36-story building are safe to drop eggs from, and which will cause the eggs to break on landing (usingU.S. English terminology, in which the first floor is at ground level). We make a few assumptions:

An egg that survives a fall can be used again.
A broken egg must be discarded.
The effect of a fall is the same for all eggs.
If an egg breaks when dropped, then it would break if dropped from a higher window.
If an egg survives a fall, then it would survive a shorter fall.
It is not ruled out that the first-floor windows break eggs, nor is it ruled out that eggs can survive the 36th-floor windows.

If only one egg is available and we wish to be sure of obtaining the right result, the experiment can be carried out in only one way. Drop the egg from the first-floor window; if it survives, drop it from the second-floor window. Continue upward until it breaks. In the worst case, this method may require 36 droppings. Suppose 2 eggs are available. What is the lowest number of egg-droppings that is guaranteed to work in all cases?

To derive a dynamic programmingfunctional equation for this puzzle, let thestate of the dynamic programming model be a pair s = (n,k), where

n = number of test eggs available,n = 0, 1, 2, 3, ...,N − 1.

k = number of (consecutive) floors yet to be tested,k = 0, 1, 2, ...,H − 1.

For instance,s = (2,6) indicates that two test eggs are available and 6 (consecutive) floors are yet to be tested. The initial state of the process iss = (N,H) whereN denotes the number of test eggs available at the commencement of the experiment. The process terminates either when there are no more test eggs (n = 0) or whenk = 0, whichever occurs first. If termination occurs at states = (0,k) andk > 0, then the test failed.

Now, let

W(n,k) = minimum number of trials required to identify the value of the critical floor under the worst-case scenario given that the process is in states = (n,k).

Then it can be shown that^[15]

W(n,k) = 1 + min{max(W(n − 1,x − 1),W(n,k −x)):x = 1, 2, ...,k }

withW(n,0) = 0 for alln > 0 andW(1,k) =k for all k. It is easy to solve this equation iteratively by systematically increasing the values ofn and k.

Faster DP solution using a different parametrization

[edit]

Notice that the above solution takes $O(nk^{2})$ time with a DP solution. This can be improved to $O(nk\log k)$ time by binary searching on the optimal $x {\displaystyle x}$ in the above recurrence, since $W(n-1,x-1)$ is increasing in $x {\displaystyle x}$ while $W(n,k-x)$ is decreasing in $x {\displaystyle x}$ , thus a local minimum of $\max(W(n-1,x-1),W(n,k-x))$ is a global minimum. Also, by storing the optimal $x {\displaystyle x}$ for each cell in the DP table and referring to its value for the previous cell, the optimal $x {\displaystyle x}$ for each cell can be found in constant time, improving it to $O(nk)$ time. However, there is an even faster solution that involves a different parametrization of the problem:

Let $k {\displaystyle k}$ be the total number of floors such that the eggs break when dropped from the $k {\displaystyle k}$ th floor (The example above is equivalent to taking $k=37$ ).

Let $m {\displaystyle m}$ be the minimum floor from which the egg must be dropped to be broken.

Let $f(t,n)$ be the maximum number of values of $m {\displaystyle m}$ that are distinguishable using $t {\displaystyle t}$ tries and $n {\displaystyle n}$ eggs.

Then $f(t,0)=f(0,n)=1$ for all $t,n\geq 0$ .

Let $a {\displaystyle a}$ be the floor from which the first egg is dropped in the optimal strategy.

If the first egg broke, $m {\displaystyle m}$ is from $1 {\displaystyle 1}$ to $a {\displaystyle a}$ and distinguishable using at most $t-1$ tries and $n-1$ eggs.

If the first egg did not break, $m {\displaystyle m}$ is from $a+1$ to $k {\displaystyle k}$ and distinguishable using $t-1$ tries and $n {\displaystyle n}$ eggs.

Therefore, $f(t,n)=f(t-1,n-1)+f(t-1,n)$ .

Then the problem is equivalent to finding the minimum $x {\displaystyle x}$ such that $f(x,n)\geq k$ .

To do so, we could compute $\{f(t,i):0\leq i\leq n\}$ in order of increasing $t {\displaystyle t}$ , which would take $O(nx)$ time.

Thus, if we separately handle the case of $n=1$ , the algorithm would take $O(n{\sqrt {k}})$ time.

But the recurrence relation can in fact be solved, giving $f(t,n)=\sum _{i=0}^{n}{\binom {t}{i}}$ , which can be computed in $O(n)$ time using the identity ${\binom {t}{i+1}}={\binom {t}{i}}{\frac {t-i}{i+1}}$ for all $i\geq 0$ .

Since $f(t,n)\leq f(t+1,n)$ for all $t\geq 0$ , we can binary search on $t {\displaystyle t}$ to find $x {\displaystyle x}$ , giving an $O(n\log k)$ algorithm.^[16]

Matrix chain multiplication

[edit]

Main article:Matrix chain multiplication

Matrix chain multiplication is a well-known example that demonstrates utility of dynamic programming. For example, engineering applications often have to multiply a chain of matrices. It is not surprising to find matrices of large dimensions, for example 100×100. Therefore, our task is to multiply matrices⁠ $A_{1},A_{2},....A_{n}$ ⁠. Matrix multiplication is not commutative, but is associative; and we can multiply only two matrices at a time. So, we can multiply this chain of matrices in many different ways, for example:

((A₁ ×A₂) ×A₃) × ...A_n

A₁×(((A₂×A₃)× ... ) ×A_n)

(A₁ ×A₂) × (A₃ × ...A_n)

and so on. There are numerous ways to multiply this chain of matrices. They will all produce the same final result, however they will take more or less time to compute, based on which particular matrices are multiplied. If matrix A has dimensions m×n and matrix B has dimensions n×q, then matrix C=A×B will have dimensions m×q, and will require m*n*q scalar multiplications (using a simplisticmatrix multiplication algorithm for purposes of illustration).

For example, let us multiply matrices A, B and C. Let us assume that their dimensions are m×n, n×p, and p×s, respectively. Matrix A×B×C will be of size m×s and can be calculated in two ways shown below:

Ax(B×C) This order of matrix multiplication will require nps + mns scalar multiplications.
(A×B)×C This order of matrix multiplication will require mnp + mps scalar calculations.

Let us assume that m = 10, n = 100, p = 10 and s = 1000. So, the first way to multiply the chain will require 1,000,000 + 1,000,000 calculations. The second way will require only 10,000 + 100,000 calculations. Obviously, the second way is faster, and we should multiply the matrices using that arrangement of parenthesis.

Therefore, our conclusion is that the order of parenthesis matters, and that our task is to find the optimal order of parenthesis.

At this point, we have several choices, one of which is to design a dynamic programming algorithm that will split the problem into overlapping problems and calculate the optimal arrangement of parenthesis. The dynamic programming solution is presented below.

Let's call m[i,j] the minimum number of scalar multiplications needed to multiply a chain of matrices from matrix i to matrix j (i.e. A_i × .... × A_j, i.e. i<=j). We split the chain at some matrix k, such that i <= k < j, and try to find out which combination produces minimum m[i,j].

The formula is:

if i = j, m[i,j]= 0if i < j, m[i,j]= min over all possible values of k(m[i,k]+m[k+1,j] + $p_{i-1}*p_{k}*p_{j}$ )

wherek ranges fromi toj − 1.

⁠ $p_{i-1}$ ⁠ is the row dimension of matrix i,
⁠ $p_{k}$ ⁠ is the column dimension of matrix k,
⁠ $p_{j}$ ⁠ is the column dimension of matrix j.

This formula can be coded as shown below, where input parameter "chain" is the chain of matrices, i.e.⁠ $A_{1},A_{2},...A_{n}$ ⁠:

function OptimalMatrixChainParenthesis(chain)    n = length(chain)for i = 1, n        m[i,i] = 0// Since it takes no calculations to multiply one matrixfor len = 2, nfor i = 1, n - len + 1            j = i + len -1            m[i,j] = infinity// So that the first calculation updatesfor k = i, j-1q = m[i, k] + m[k+1, j] + $p_{i-1}*p_{k}*p_{j}$ if q < m[i, j]// The new order of parentheses is better than what we had                    m[i, j] = q// Update                    s[i, j] = k// Record which k to split on, i.e. where to place the parenthesis

So far, we have calculated values for all possiblem[i,j], the minimum number of calculations to multiply a chain from matrixi to matrixj, and we have recorded the corresponding "split point"s[i,j]. For example, if we are multiplying chainA₁×A₂×A₃×A₄, and it turns out thatm[1, 3] = 100 ands[1, 3] = 2, that means that the optimal placement of parenthesis for matrices 1 to 3 is⁠ $(A_{1}\times A_{2})\times A_{3}$ ⁠ and to multiply those matrices will require 100 scalar calculations.

This algorithm will produce "tables"m[, ] ands[, ] that will have entries for all possible values of i and j. The final solution for the entire chain is m[1, n], with corresponding split at s[1, n]. Unraveling the solution will be recursive, starting from the top and continuing until we reach the base case, i.e. multiplication of single matrices.

Therefore, the next step is to actually split the chain, i.e. to place the parenthesis where they (optimally) belong. For this purpose we could use the following algorithm:

function PrintOptimalParenthesis(s, i, j)if i = j        print "A"ielse        print "("         PrintOptimalParenthesis(s, i, s[i, j])         PrintOptimalParenthesis(s, s[i, j] + 1, j)         print ")"

Of course, this algorithm is not useful for actual multiplication. This algorithm is just a user-friendly way to see what the result looks like.

To actually multiply the matrices using the proper splits, we need the following algorithm:

functionMatrixChainMultiply(chainfrom1ton)// returns the final matrix, i.e. A1×A2×... ×AnOptimalMatrixChainParenthesis(chainfrom1ton)// this will produce s[ . ] and m[ . ] "tables"OptimalMatrixMultiplication(s,chainfrom1ton)// actually multiplyfunctionOptimalMatrixMultiplication(s,i,j)// returns the result of multiplying a chain of matrices from Ai to Aj in optimal wayifi<j// keep on splitting the chain and multiplying the matrices in left and right sidesLeftSide=OptimalMatrixMultiplication(s,i,s[i,j])RightSide=OptimalMatrixMultiplication(s,s[i,j]+1,j)returnMatrixMultiply(LeftSide,RightSide)elseifi=jreturnAi// matrix at position ielseprint"error, i <= j must hold"functionMatrixMultiply(A,B)// function that multiplies two matricesifcolumns(A)=rows(B)fori=1,rows(A)forj=1,columns(B)C[i,j]=0fork=1,columns(A)C[i,j]=C[i,j]+A[i,k]*B[k,j]returnCelseprint"error, incompatible dimensions."

History of the name

[edit]

The termdynamic programming was originally used in the 1940s byRichard Bellman to describe the process of solving problems where one needs to find the best decisions one after another. By 1953, he refined this to the modern meaning, referring specifically to nesting smaller decision problems inside larger decisions,^[17] and the field was thereafter recognized by theIEEE as asystems analysis andengineering topic. Bellman's contribution is remembered in the name of theBellman equation, a central result of dynamic programming which restates an optimization problem inrecursive form.

Bellman explains the reasoning behind the termdynamic programming in his autobiography,Eye of the Hurricane: An Autobiography:

I spent the Fall quarter (of 1950) atRAND. My first task was to find a name for multistage decision processes. An interesting question is, "Where did the name, dynamic programming, come from?" The 1950s were not good years for mathematical research. We had a very interesting gentleman in Washington namedWilson. He was Secretary of Defense, and he actually had a pathological fear and hatred of the word "research". I'm not using the term lightly; I'm using it precisely. His face would suffuse, he would turn red, and he would get violent if people used the term research in his presence. You can imagine how he felt, then, about the term mathematical. The RAND Corporation was employed by the Air Force, and the Air Force had Wilson as its boss, essentially. Hence, I felt I had to do something to shield Wilson and the Air Force from the fact that I was really doing mathematics inside the RAND Corporation. What title, what name, could I choose? In the first place I was interested in planning, in decision making, in thinking. But planning, is not a good word for various reasons. I decided therefore to use the word "programming". I wanted to get across the idea that this was dynamic, this was multistage, this was time-varying. I thought, let's kill two birds with one stone. Let's take a word that has an absolutely precise meaning, namely dynamic, in the classical physical sense. It also has a very interesting property as an adjective, and that is it's impossible to use the word dynamic in a pejorative sense. Try thinking of some combination that will possibly give it a pejorative meaning. It's impossible. Thus, I thought dynamic programming was a good name. It was something not even a Congressman could object to. So I used it as an umbrella for my activities.

— Richard Bellman,Eye of the Hurricane: An Autobiography (1984, page 159)

The worddynamic was chosen by Bellman to capture the time-varying aspect of the problems, and because it sounded impressive.^[12] The wordprogramming referred to the use of the method to find an optimalprogram, in the sense of a military schedule for training or logistics. This usage is the same as that in the phraseslinear programming andmathematical programming, a synonym formathematical optimization.^[18]

The above explanation of the origin of the term may be inaccurate: According to Russell and Norvig, the above story "cannot be strictly true, because his first paper using the term (Bellman, 1952) appeared before Wilson became Secretary of Defense in 1953."^[19] Also,Harold J. Kushner stated in a speech that, "On the other hand, when I asked [Bellman] the same question, he replied that he was trying to upstageDantzig's linear programming by adding dynamic. Perhaps both motivations were true."^[20]

References

[edit]

^^a ^bCormen, T. H.; Leiserson, C. E.; Rivest, R. L.; Stein, C. (2001), Introduction to Algorithms (2nd ed.), MIT Press & McGraw–Hill,ISBN 0-262-03293-7 . pp. 344.
^Kamien, M. I.;Schwartz, N. L. (1991).Dynamic Optimization: The Calculus of Variations and Optimal Control in Economics and Management (Second ed.). New York: Elsevier. p. 261.ISBN 978-0-444-01609-6.
^Kirk, Donald E. (1970).Optimal Control Theory: An Introduction. Englewood Cliffs, NJ: Prentice-Hall. pp. 94–95.ISBN 978-0-13-638098-6.
^"Algorithms by Jeff Erickson".jeffe.cs.illinois.edu. Retrieved2024-12-06.
^"M. Memo".J Vocabulary. J Software. Retrieved28 October 2011.
^Delisi, Charles (July 1974), "Cooperative phenomena in homopolymers: An alternative formulation of the partition function",Biopolymers,13 (7):1511–1512,doi:10.1002/bip.1974.360130719
^Gurskiĭ, G. V.; Zasedatelev, A. S. (September 1978), "Precise relationships for calculating the binding of regulatory proteins and other lattice ligands in double-stranded polynucleotides",Biofizika,23 (5):932–946,PMID 698271
^Sniedovich, M. (2006),"Dijkstra's algorithm revisited: the dynamic programming connexion"(PDF),Journal of Control and Cybernetics,35 (3):599–620.Online version of the paper with interactive computational modules.
^Denardo, E.V. (2003),Dynamic Programming: Models and Applications, Mineola, NY:Dover Publications,ISBN 978-0-486-42810-9
^Sniedovich, M. (2010),Dynamic Programming: Foundations and Principles,Taylor & Francis,ISBN 978-0-8247-4099-3
^Dijkstra, E. W. (December 1959). "A note on two problems in connexion with graphs".Numerische Mathematik.1 (1):269–271.doi:10.1007/BF01386390.
^^a ^bEddy, S. R. (2004). "What is Dynamic Programming?".Nature Biotechnology.22 (7):909–910.doi:10.1038/nbt0704-909.PMID 15229554.S2CID 5352062.
^Moshe Sniedovich (2002), "OR/MS Games: 2. The Towers of Hanoi Problem",INFORMS Transactions on Education,3 (1):34–51,doi:10.1287/ited.3.1.45.
^Konhauser J.D.E., Velleman, D., and Wagon, S. (1996).Which way did the Bicycle Go? Dolciani Mathematical Expositions – No 18.The Mathematical Association of America.
^Sniedovich, Moshe (2003)."OR/MS Games: 4. The Joy of Egg-Dropping in Braunschweig and Hong Kong".INFORMS Transactions on Education.4 (1):48–64.doi:10.1287/ited.4.1.48.
^Dean Connable Wills,Connections between combinatorics of permutations and algorithms and geometry
^Stuart Dreyfus."Richard Bellman on the birth of Dynamical Programming".
^Nocedal, J.; Wright, S. J. (2006).Numerical Optimization. Springer. p. 9.ISBN 9780387303031.
^Russell, S.; Norvig, P. (2009).Artificial Intelligence: A Modern Approach (3rd ed.). Prentice Hall.ISBN 978-0-13-207148-2.
^Kushner, Harold J. (2004-07-01)."Richard E. Bellman Control Heritage Award". Archived fromthe original on 2014-10-19.

External links

[edit]

King, Ian, 2002 (1987), "A Simple Introduction to Dynamic Programming in Macroeconomic Models." An introduction to dynamic programming as an important tool in economic theory.

Optimization:Algorithms,methods, andheuristics

Unconstrained nonlinear

Functions

Gradients

Convergence	Trust region Wolfe conditions
Quasi–Newton	Berndt–Hall–Hall–Hausman Broyden–Fletcher–Goldfarb–Shanno andL-BFGS Davidon–Fletcher–Powell Symmetric rank-one (SR1)
Other methods	Conjugate gradient Gauss–Newton Gradient Mirror Levenberg–Marquardt Powell's dog leg method Truncated Newton

Hessians

Newton's method

Graph of a strictly concave quadratic function with unique maximum. — Optimization computes maxima and minima.

Constrained nonlinear

General	Barrier methods Penalty methods
Differentiable	Augmented Lagrangian methods Sequential quadratic programming Successive linear programming

Convex optimization

Convex
minimization

Linear and
quadratic

Interior point	Affine scaling Ellipsoid algorithm of Khachiyan Projective algorithm of Karmarkar
Basis-exchange	Simplex algorithm of Dantzig Revised simplex algorithm Criss-cross algorithm Principal pivoting algorithm of Lemke Active-set method

Combinatorial

Paradigms

Graph
algorithms

Minimum spanning tree	Borůvka Prim Kruskal

Shortest path	Bellman–Ford SPFA Dijkstra Floyd–Warshall

Network flows

Metaheuristics
Evolutionary algorithm Hill climbing Local search Parallel metaheuristics Simulated annealing Spiral optimization algorithm Tabu search

Software

v t e Parsing algorithms
Top-down	Earley LL Recursive descent Tail recursive
Bottom-up	Precedence Simple Operator Shunting-yard LR Simple Look-ahead Canonical Generalized CYK Recursive ascent Shift-reduce
Mixed, other	Combinator Chart Left corner Statistical
Related topics	PEG Definite clause grammar Deterministic parsing Dynamic programming Memoization Parser generator LALR Parse tree AST Scannerless parsing History of compiler construction Comparison of parser generators Operator-precedence grammar

v t e Data structures andalgorithms
Data structures	Array Associative array Binary search tree Fenwick tree Graph Hash table Heap Linked list Queue Segment tree Stack String Tree Trie
Algorithms andalgorithmic paradigms	Backtracking Binary search Breadth-first search Brute-force search Depth-first search Divide and conquer Dynamic programming Graph traversal Fold Greedy Hash function Minimax Online Randomized Recursion Root-finding Sorting Streaming Sweep line String-searching Topological sorting
List of data structures List of algorithms