Cutting Stock Problem

• Problem of cutting an unlimited number of pieces of material (paper rolls,

for instance) of length l
to produce
ni pieces of length li , i = 1, 2, …, I.

• The objective is to minimize the number of pieces of material to meet the

demands.

• Note:
minimize number of pieces ≡ minimize total waste
• Cutting pattern Pj (j = 1, 2, …, NJ) corresponds to a specific way of cutting
a piece of material:

aij = number of pieces of length li produced with cutting pattern Pj

where aij ≥ 0 and integer, i = 1, 2, …, I

I
and ∑a l
i =1
ij i ≤l
• Mathematical Model
NJ
(CSP) min ∑x
j =1
j

NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, integer j = 1, ⋯ , NJ

where xj represents the number of pieces of material cut with the pattern j
 NJ
(CSP) min ∑x
j =1
j

NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, integer j = 1, ⋯ , NJ

• Integer programming problem difficult to solve.

• Drop the constraints that the xj are required to be integer
• When the demands ni are large, little is lost with respect to optimality by
rounding the solution obtained by solving the relaxed problem (CSP)

NJ
(CSP) min ∑x j =1
j

NJ
Subject to ∑a j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, j = 1, ⋯ , NJ
• The number NJ of patterns may be very large.
For problems where the piece of material have length l = 800 cm., and
where there are demands for 40 different lengths li from 20 cm. to 80 cm.,
the number of admissible patterns can easily reach 10 to 100 millions.

• Interest to solve this problem using column generation.

{ }
• At each iteration, a subset of patterns P1 , P2 , ⋯ , Pj are available to
specify a restricted problem of (CSP)
 NJ
(CSP) min ∑x
j =1
j

NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, j = 1, ⋯ , NJ

Restricted Problem of (CSP) :

NJ
min ∑x
j =1
j

NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, j = 1, ⋯ , j
xj =0 j = j + 1, ⋯ , NJ
 Restricted Problem of (CSP) :
j
NJ
min ∑x
j =1
j

j
NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I πi

x j ≥ 0, j = 1, ⋯ , j
xj =0 j = j + 1, ⋯ , NJ
Solve the restricted problem.
Denote x an optimal solution.
T
Denote also π = π 1 , π 2 ,⋯ , π I  the vector of optimal simplex multipliers.

Since x is optimal, the relative cost of each variable x is non negative , j = 1, ⋯ , j

j
I

c = 1− ∑π a ≥ 0
j i
ij
j = 1, ⋯ , j
i =1
Now x is feasible for CSP( )
NJ
(CSP ) min ∑x
j =1
j

NJ
Subject to ∑a
j =1
ij x j ≥ ni i = 1, ⋯ , I

x j ≥ 0, j = 1, ⋯ , NJ

( )
To verify if x is optimal for CSP , we have to verify that
the relative cost is non negative for each variable:
I
c j = 1 − ∑ π i aij ≥ 0 j = 1,⋯ , NJ
i =1

or if
 I

min 1 − ∑ π i aij  ≥ 0.
j =1,⋯, NJ
 i =1 
But recall that the cutting patterns Pj are not known for j = j + 1,⋯ , NJ .
In order to be able to verify if

{ I
min 1 − ∑ π i aij ≥ 0,
j =1,⋯, NJ i =1
}
recall that any cutting pattern P is characterized by

ai the number of pieces of length li generated, i = 1,⋯ , I

where
ai ≥ 0, integer i = 1,⋯ , I
I
∑ ai li ≤ l.
i =1
 j =1,⋯, NJ  
 i =1
recall that any cutting pattern P is characterized by
ai the number of pieces of length li generated, i = 1,⋯, I
where
ai ≥ 0, integer i = 1,⋯, I
I

∑a l
i −1
i i ≤ l.

Hence
 I

min 1 − ∑ π i a ij 
j =1,⋯ , NJ
 i =1 
is equivalent to determine an optimal solution a T =  a1 , a 2, ⋯ , a I  of
 

 = 1 + min −
I   I 
∑ ∑
I
min 1 − ∑ π i a i   π i a i  = 1 − max  π i a i 
 i =1   i =1   i =1 
I
Subject to ∑al
i =1
i i ≤l

a i ≥ 0, integer i = 1, ⋯ , I .
Hence
 I

min 1 − ∑ π i a ij 
j =1,⋯ , NJ
 i =1 
is equivalent to determine an optimal solution a T =  a1 , a 2, ⋯ , a I  of
 I
  I   I 
min 1 − ∑ π i a i  = 1 + min − ∑ ∑
π i a i  = 1 − max  π i a i 
 i =1   i =1   i =1 
I
Subject to ∑al
i =1
i i ≤l

a i ≥ 0, integer i = 1, ⋯ , I .
The problem is then equivalent to solve the knapsack problem
 I 
(KP) max ∑ π i ai 
 i =1 
I
Subject to ∑a l
i =1
i i ≤l

ai ≥ 0, integer i = 1,⋯ , I .
 T
=  a1 , a2 , ⋯ , a I  be an optimal solution of ( KP ) ,
* * * *
Let a
*
and denote P the associated cutting pattern.

I
1 − ∑ π i ai ≥ 0,
*
If
i =1

STOP the procedure because x is an optimal solution of (CSP).

I
1 − ∑ π i ai < 0,
*
If
i =1

defined a new restricted problem of (CSP) by

j +1 *
including a new apttern P = P (i.e., replacing
the constraint x j +1 = 0 by x j +1 ≥ 0).