Weighted Residual Method—Use of

a Single Continuous Trial Function

1
Let us Consider a general problem of engineering analysis described
in the form of a differential equation (to be valid within a
praticular domain Ω), while satisfying the prescried boundary
conditions on the boundary Γ. Our scheme of finding
approximate solution to differential equations Consists of the
following steps:

 Assume a guess (or trial) solution to the problem. For example,

for a one-dimensional problem, we may Choose a trial
solution as
ƒ (x) = c0 + c1x + c2x2 + …

 In general, the function so assumed will satisfy neither the

differential equation within the domain (Ω) nor the boundary
Conditions (on Γ). By substituting the assumed function in the
differential equation and the boundary Conditions of the
problem, find the error in satisfying these (we will GALL these
“domain residual” and “boundary residual”).
2
  Determine the unknown parameters (c0, c1, c2, …) in the
assumed trial function in such a way as to make these
residuals as low as possible.
In the process, if we can make the domain and boundary residuals
identical to zero everywhere, we will get the exact solution to the
problem itself. In general, we expect to get a reasonably accurate
solution to the problem at hand. In the context of the finite element
method, we will limit our discussion to trial solutions that satisfy
the applicable boundary conditions and hence, only domain residual
remains. The choice of trial solutions that implicitly satisfy the
differential equation but not the boundary Conditions (thus resulting
in nonzero boundary residual) leads to the “boundary element method”.
As a detailed discussion on the boundary element method is beyond
the scope of this text, the interested reader may refer to standard
texts for details of this technique.
We will use the following simple example to illustrate the
above method of finding approximate solutions to differential equations
3
Example 2.1. Consider a uniform rod Subjected to a uniform axial load as
illustrated in Figure 2.1. It Can be readily shown that the deformation of the
bar is governed by the differential equation

q0

Fig. 2.1 Rod subjected to axial load.

4
Let us now find an approximate solution to this problem using the
method just discussed.

Step 1: Assume a trial or guess solution. Let

u(x) û(x) = c0 + c1x + c2x2

where the Constants c0, c1, c2 are yet to be determined. In

order to satisfy the first boundary condition that û(0) = 0, we
have c0 = 0. To satisfy the secone boundary Condition, we
have c1 = −2c2L. Thus we now have, for our trial solution,

û(x) = c2(x2 − 2Lx)

SINGE the trial solution contains only one free parameter c2, it
is often referred to as a “one- parameter solution”.

5
Step 2: Find the domain residual. Substituting in the governing differential
equation

Step 3: Minimise the residual. Singe there is one residual to

be minimised and one parameter to be determined, we CAN
readily solve for the undetermined Co-effigient by setting the
residual to zero, i.e., Rd = 0, yielding

6
Thus our final solution is

For this simple example, singe we could make the residual

identically zero everywhere, our final solution tallies with the
exact solution

7
Example 2.2. The governing equation for a fully developed
steady laminar flow of a Newtonian Viscous fluid on an
Inclined flat surface (see Figure 2.2) is given by

L y
x v(x)

Gravity
g

Fig. S.S Laminar flow on an inclined surface.

8
 Where

The boundary Conditions are given by

Let us find the velocity distribution v(x) using the weighted residual
9
method
Step 1: Assume a trial solution. Let

Hence,

From the boundary Conditions, c1 = 0, c0 = −c2L2.

Therefore,

10
Step 2: Find the domain residual

Step 3: Minimise the residual. Rd is a constant and can therefore be

set to zero. Hence,

Therefore,

It is readily veriƒied that our solution matches the exact solution as

we can make the domain residual identically zero.
11
Example 2.3. Consider the problem of a cantilever beam under
uniformly distributed load q0 as shown in Figure 2.3. The governing
differential equation is given by

and the boundary Conditions are given by

12
where the first two boundary Conditions enforce zero displacement and
slope at the fixed end and the last two Conditions prescribe zero bending
moment and shear force at the free end.

q0

x
L

Fig. 2.3 Cantilever beam under load.

13
Step 1: Assume a trial or guess solution. We observe that it is not easy to
select a trial function that satisfies all the boundary Conditions. Let us
Choose vˆ(x) = c0x + c1x + c2x2 + c3x3 + c4x4.
From the boundary Conditions that vˆ(0) = 0 and d vˆ/dx (0) = 0,
we have c0 = 0 = c1. In order to satisfy the boundary Conditions at x =
L, we should have

Substituting and rearranging the terms in the trial solution, we get

We thus observe that finding trial solution functions that satisfy

all the boundary conditions Could, in general, be cumbersome

14
 Step 2: Find the domain residual. Substituting in the differential
equation, we get the domain residual as

Step 3: Minimise the residual. SINGE there is one residual to be minimised

and one parameter to be determined, we can readily solve for the
undetermined coefficient c4 by setting the residual to zero,
i.e., Rd = 0, yielding thereby

Thus our trial solution is

which Can be readily verified to be the exact solution itself. This is

to be expected sINGE we were able to make the residual identically
zero within the entire domain.
15
Example 2.4. Let us Consider the example of a simply supported beam
q0

under uniformly distributed load as shown in Figure 2.4. The governing

differential equation and the boundary conditions are given by
q0

x
L

16
 Step 1: Assume a trial solution. We can use the approach
of Example 2.3 and find a polynomial trial solution
satisfying all boundary Conditions. However, in view of the
special boundary Conditions of simple supports, we can make
the process simple by choosing trigonometric functions. Let

This one-parameter trial solution satisfies all boundary

Conditions.

Step 2: Find the domain residual. Substituting the trial solution

vˆ(x) in the governing differential equation, the domain residual
is obtained as

17
Step 3: Minimise the residual. We observe that, unlike in the previous
examples, the domain residual is now varying from point to point within
the domain (0 < x < L). SINGE we have only one coefficients to be
determined, we can set the residual zero only at any one point of
our choice within the domain if we follow the approach of previous
examples. This technique is GALLED the point collocation technique,
wherein we set the residual (in general, a function of x) to zero at
Chosen points within the domain—the number of points being
equal to the number of coefficients in the trial function that need to
be determined. We will illustrate this procedure now. In this
procedure, however, there is a danger that the residual might be unduly
large at some other points within the domain. We may thus want
to “minimise” the residual in an overall sense over the entire
domain rather than setting it identically zero at only few selected
points. This will be discussed in the next section.

18
Solution by point collocation. Let us make Rd = 0 at x = L/4, i.e.

yielding there by

and the resulting trial solution is

We may repeat this process by setting the domain residual

zero at other axial locations. Figure 2.5 compares the solutions
vˆ(x) obtained in this manner against the classical exact

19
solution. Figure 2.6 shows a plot of the domain residual in each case. In
this way, we are able to generate different approximate solutions to the
problem but each of these solutions deviates appreciably from the exact
solution. We can improve our trial solution by adding one more term to
the Sine series. SINGE the problem at hand is symmetric about x = L/2,
we modify our trial solution as follows:
20
Fig. 2.6 Plot of residual-simply supported beam (Example 2.4).

21
With this two-term trial function, we obtain the domain residual as

SINGE we have two constants to be determined, we can set the

residual zero at two selected points. Let us set

i.e.,

22
Solving these equations, we obtain

Figure 2.7 compares the solution vˆ(x) obtained in this manner

against the classical exact solution. Figure 2.8 shows a plot of
the domain residual. We observe that this approximates the
solution perhaps better than the earlier one-parameter
solution, but it still deviates appreciably from the exact
solution.

23
Fig. 2.7 Simply supported beam deflection—Two-parameter solution.

24
Fig. 2.8 Domain residual—Two-parameter solution Rd(L/4 and L/3) = 0.
25
 Rather than trying to set the residual exactly zero at a few selected points
and having no control over the residual at all the other points in the
domain, we will now try to minimise the residual in an overall sense.
This technique is called the weighted residual technique.

Solution by weighted residual technique. For the problem at hand,

we will formulate this technique as

where Wi(x) are appropriately Chosen weighting functions, helping us to

achieve the task of minimising the residual over the entire domain. We
choose as many weighting functions as necessary to generate the required
number of equations for the solution of the undetermined coefficients in
the trial function. For example, for a two-term trial solution, we would take
two different weighting functions to generate two equations necessary to
solve for the two coefficients. While the choice of weighting functions 26 W(x)
is entirely arbitrary (as long as they are nonzero and integrable), Galerkin
(1917) introduced the idea of letting W(x) to be same as the trial
functions themselves. Thus, in this example, we take the weighting
function to be sin (π x/L), sin (3 π x/L), etc. Let us illustrate the procedure
of the Galerkin weighted residual technique for the one-term trial
function, i.e., let

The domain residual

By the Galerkin procedure, we set

27
 We have chosen only one weighting function because there is
only one Constant c1 to be determined. Thus we have

yielding c1 = 0.013071q0L4/EI. Therefore,

Figure 2.9 compares the solution vˆ(x) obtained in this manner

against the classical exact solution. It is observed that the one-
term solution is significantly improved by the Galerkin

28
 Normalised x-LOGATION
Fig. 2.9 Strength of material (SOM) and Galerkin WR solutions—Simply
supported beam (Example S.4).

29
weighted residual minimisation process. The error in the
deflection at the mid-span is just 0.38%. Figure 2.10 shows
a plot of the domain residual obtained by the Galerkin
process.
We have thus discussed a very powerful tool for
determining approximate solutions to boundary value
problems, given the governing differential equation and the
boundary conditions. This method requires us to choose trial
solutions that satisfy the boundary Conditions of the
problem. We then generate sufficient number of equations
to solve for the undetermined parameters by setting the
weighted integral of the residual to zero.

30
Fig. 2.10 Domain residual obtained by Galerkin method (Example S.4)
31
We can, in principle, choose any nonzero, integrable weighting
functions to generate the necessary equations. It is easily realised
that if we choose the weighting functions, to be Dirac− Delta
functions, i.e., W(x) = (x − xp), the weighted residual process will
reduce to the point collocation technique discussed earlier, with the
collocation points being x = xp. In the Galerkin method, we choose
the weighting functions to be the trial functions them selves.
To be precise, if the trial solution is assumed to be

where Ø(x) and Ni(x) are fully known functions of x, pre-

selected such that the boundary conditions are satisfied by ƒ
(x), then we choose the weighting functions to be Wi(x) =
6ƒ/6ci = Ni(x).

32
Note:
It is to be observed that this method is not limited
to problems of Structural Mechanics alone but can
be used to solve any problem given the
appropriate differential equation and the
boundary conditions.

33
Example 2.5. Consider the problem of a long cylinder of
radius R with uniformly distributed heat source q0. The
governing differential equation is given by

and the boundary Conditions are given by

T(R) = Tw, the wall temperature

34
The second boundary Condition implies that heat generated = heat lost.
We wish to determine the temperature distribution T as a function
of radial location, r.
Step 1: Assume a trial solution. We observe that it may be
Convenient to choose the trial solution in terms of (r − R) rather
than r.
Let us choose the trial solution as

From the boundary Conditions, we find that c0 = Tw and c1 =

−(q0R/2k). Thus we have

In our notation introduced above, we can write T = (r) + c2N(r)

where N(r) = (r − R)2 will be our weighting function W(r).
35
Solving, we get c2 = Therefore, our approzimation to the
temperature distribution

Thus we observe that our procedure can be systematically applied to

any problem, given the differential equation and the boundary
Conditions.
We will now illustrate the Galerkin Weighted Residual procedure
on a two-term trial function through the following example.

36
Step 2: Find the domain residual. The domain residual

Step 3: Minimise the residual. We observe that the residual is

actually minimised over the whole volume of the domain and in
radial coordinates, an elemental volume is given by (r dr d θ dz).
Considering the fact that temperature does not vary in the
circumferential direction (θ) or the axial direction (z) for this problem,
explicit integration in these two d direction leads to the Constant (2 π
θL), and the weighted residual statement is given by

37
Reference: A Text Book of Finite Element Analysis, P Seshu, PHI

38