Critical Load of Elastic Columns by the Energy Method
The static method struggles with variable stiffness or high-order stability equations. The energy method, based on the stationary-potential principle, expresses the deflection as a generalized-coordinate series, reducing an infinite-DOF problem to an approximate finite-DOF algebraic problem — the Ritz method. Key worked examples: Ex. 11-5 (cantilever — four trial functions), Ex. 11-6 (frame), Ex. 11-7 (self-weight + concentrated load).
Click any item in the left Contents to jump · Click the green "Next" · Each animation has a ▶ button to advance stages; ↻ at the end for replay.
Why the energy method
MotivationWhen the static method (§11-3) meets a column with variable cross-section or a distributed axial load, the deflection ODE becomes a variable-coefficient equation — usually without an analytical solution — or the stability equation's order grows too high to expand. In such cases, the energy method is highly effective.
Static method Static Method
- Based on the deflection ODE $EIy'' = -M$;
- Analytical solutions possible for constant stiffness with simple boundaries;
- Becomes extremely hard for variable stiffness / complex geometry.
Energy method Energy Method · Ritz
- Based on the stationary-potential principle $\delta E_{\mathrm P} = 0$;
- Approximate with a generalized-coordinate series $y(x) = \sum a_i \varphi_i(x)$;
- Reduces to a finite-DOF algebraic eigenvalue problem — handles arbitrary stiffness distributions.
- How to write the total potential during buckling $E_{\mathrm P} = U + U_{\mathrm P}$ — strain energy $U$ and load potential $U_{\mathrm P}$;
- How to choose an appropriate displacement function — accuracy of the approximation depends on this choice.
Strain energy & load potential
Strain / Load EnergyAs a column bifurcates from straight to bent equilibrium, the axial force stays constant. The bending strain energy is
Substituting $EIy'' = -M$ gives the form in terms of $y''$:
Vertical shortening Δ
During the transition from straight to bent, the loading point drops vertically. For any small segment $\mathrm d x$ at slope $\theta$ (see Animation 11.4.2-a):
Integrating over the full length:
Load potential U_P
The external load $F_{\mathrm P}$ does positive work on the displacement $\Delta$; the potential is therefore negative:
With multiple concentrated loads $F_{\mathrm Pi}$ each moving $\Delta_{i}$:
Total potential · Ritz foundation
Combining $E_{\mathrm P} = U + U_{\mathrm P}$:
The stationary condition $\partial E_{\mathrm P}/\partial a_i = 0$ gives homogeneous algebraic equations in the generalized coordinates $a_i$.
Displacement series · Ritz method
Ritz MethodAssume the buckling displacement can be written as a linear combination of basis functions with generalized coordinates:
Now $y(x)$ is fully determined by the $n$ coordinates $a_i$, reducing the infinite-DOF problem to the n-DOF stability problem of §11-2.
Guidelines for choosing the trial function
- Must satisfy the displacement boundary conditions (geometric constraints); satisfying force boundary conditions also helps but is not required;
- Should be close to the real buckled shape — the closer, the better;
- Adding more terms should improve the estimate; if $n+1$ terms give essentially the same answer as $n$, the estimate is near exact;
- Power series or trigonometric series are common (see table below).
Table 11-1 · Common series satisfying the displacement BCs
| Boundary | Trigonometric | Polynomial |
|---|---|---|
| Pinned-pinned | $y = \sum a_i \sin \dfrac{i\pi x}{l}$ | $y = \sum a_i\, x^{i}(l-x)^{i}$ |
| Fixed-free (cantilever) | $y = \sum a_i \left(1 - \cos \dfrac{(2i-1)\pi x}{2l}\right)$ | $y = \sum a_i\, x^{i+1}$ (drop linear term) |
| Fixed-fixed | $y = \sum a_i \left(1 - \cos \dfrac{2i\pi x}{l}\right)$ | $y = \sum a_i\, x^{i+1}(l-x)^{i+1}$ |
| Fixed-pinned | $y = \sum a_i\, x^{i+1}(l - x)$, etc. | |
Substituting the chosen series into $E_{\mathrm P} = U + U_{\mathrm P}$ makes $E_{\mathrm P}$ a quadratic form in $a_1, \ldots, a_n$. From $\partial E_{\mathrm P}/\partial a_i = 0$ we obtain $n$ homogeneous algebraic equations; the non-trivial condition $D = 0$ gives the stability equation. This procedure is the Ritz method.
Example 11-5 · Cantilever column — four trial functions
Worked Example · Ex. 11-5Using the prismatic cantilever of Animation 11.4.4-a, try four different trial functions to compute the critical load and analyze the results. Displacement boundary conditions: $y = 0,\ y' = 0$ at $x = 0$.
(1) First term of a cosine series · $y = a_{1}(1 - \cos \pi x/2l)$
Derivatives: $y' = (\pi a_{1}/2l)\sin(\pi x/2l)$, $y'' = (\pi^{2} a_{1}/4 l^{2})\cos(\pi x/2l)$.
From $\mathrm d E_{\mathrm P}/\mathrm d a_{1} = 0$ with $a_{1} \ne 0$:
(2) Cubic polynomial · $y = a_{1}(x^{2} - x^{3}/3l)$
Generic form of the deflection under a tip transverse force. Derivatives:
Stationary condition gives
(3) Two coordinates · $y = a_{1} x^{2} + a_{2} x^{3}$
$y' = 2 a_{1} x + 3 a_{2} x^{2}$, $y'' = 2 a_{1} + 6 a_{2} x$. Thus
Let $\alpha = F_{\mathrm P}\, l^{2}/EI$; the stationary conditions give
Setting the determinant to zero: $3\alpha^{2} - 104\alpha + 240 = 0$, smallest root $\alpha = 2.486$, so
(4) Quadratic · $y = a_{1} x^{2}$ (a warning)
Here $y'' = 2 a_{1}$ is constant — the curvature is constant, far from the real cantilever's curvature (which varies as $\cos$).
Applying the stationary condition gives
Accuracy comparison
| Trial function | Coords | $F_{\mathrm{Pcr}}$ | Error |
|---|---|---|---|
| (1) $a_1(1 - \cos\pi x/2l)$ | 1 | $\pi^{2}EI/4l^{2} \approx 2.467\,EI/l^{2}$ | Exact |
| (2) $a_1(x^{2} - x^{3}/3l)$ | 1 | $2.500\,EI/l^{2}$ | +1.32% |
| (3) $a_1 x^{2} + a_2 x^{3}$ | 2 | $2.486\,EI/l^{2}$ | +0.75% |
| (4) $a_1 x^{2}$ | 1 | $3.000\,EI/l^{2}$ | +21.59% |
Example 11-6 · frame by the energy method
Worked Example · Ex. 11-6Recompute Ex. 11-3 (portal frame) of §11-3 by the energy method — columns of equal $EI$, beam of $EA = \infty$, column tops loaded by $F_{\mathrm P}$, sway buckling mode. The static method gave $F_{\mathrm{Pcr}} = 2.706\, EI/l^{2}$; compare.
① Trial function (CD column)
During sway buckling (see Animation 11.4.5-a), the strain energy comes from bending of the compressed $CD$ and the unloaded $EF$; the load potential is due to the displacements at $B$ and $D$.
Take for $CD$ the deflection shape generated by a transverse force at D (same form as Ex. 11-5 variant (2)):
② Strain energy · EF via its lateral stiffness
The strain energy of $EF$ can be found by computing the external work on $F$'s sway $\Delta$. From the chosen trial function $\Delta = \tfrac{2}{3}\, l^{2}\, a_{1}$. $EF$'s lateral stiffness is $k = 3 EI/l^{3}$, so
Total strain energy of both columns:
③ Load potential · two contributions
At buckling $AB$ undergoes only rigid-body rotation of angle $\Delta/l$; thus the potential of $F_{\mathrm P}$ at $B$ is
The potential of $F_{\mathrm P}$ acting on $CD$ follows Eq. (11-17) (same form as Ex. 11-5 variant (2)):
④ Total potential · Stationary condition · Critical load
Total potential:
From $\mathrm d E_{\mathrm P}/\mathrm d a_{1} = 0$:
With $a_{1} \ne 0$ we solve
For the same frame problem, the static method (§11-3) requires solving a transcendental stability equation; the energy method (this section) is purely algebraic. Although the energy result is 0.78% higher, the computational effort is dramatically lower — the advantage grows with problem complexity.
Accuracy · The energy method always over-estimates
Accuracy$F_{\mathrm{Pcr}}$ from the true buckled shape is the exact value; $F_{\mathrm{Pcr}}$ from any assumed shape is generally higher than the exact value. The reason: the assumed shape is only a subset of all possible buckled configurations — effectively imposing additional constraints, which raises the system's resistance to buckling.
Improving accuracy
- Choose a trial function closer to the real shape — the cantilever's cosine first term is exact;
- Add more terms — if $n+1$ terms give essentially the same answer as $n$, the estimate is near exact;
- The trial function must satisfy the displacement BCs; satisfying force BCs is a bonus;
- Ritz gives an upper-bound estimate of $F_{\mathrm{Pcr}}$ — unsafe in design; apply a safety factor.
For variant (4) $y = a_{1} x^{2}$, if we write $M = F_{\mathrm P}(a_{1} l^{2} - y)$ and use $U = \int M^{2}/(2EI)\,\mathrm d x$ (Eq. 11-14) instead of the $y''$ form (Eq. 11-15), we get $F_{\mathrm{Pcr}} = 2.5\, EI/l^{2}$, error only 1.32% (same as variant (2))!
Reason: for simple trial functions, the error in the second derivative is usually much larger than the error in the displacement itself. Expressing the moment $M$ in terms of $y$ and using Eq. (11-14) (the $M$ form) usually gives much higher accuracy than using Eq. (11-15) (the $y''$ form). The same phenomenon appears in finite-element analysis: displacement accuracy typically exceeds stress accuracy (computed from displacement derivatives).
Example 11-7 · · self-weight + concentrated load
Worked Example · Ex. 11-7Animation 11.4.7-a: a pinned-pinned prismatic column $AB$ under a uniformly distributed self-weight $q$ and an axial concentrated load $q\, l$ at point $C$ (at $2l/3$ from the bottom end). Use the energy method to find the critical self-weight intensity $q_{\mathrm{cr}}$.
① Trial function (satisfying pinned-pinned BCs)
Satisfying $y = 0$ at $x = 0$ and $x = l$ (see Animation 11.4.7-a), take
Derivatives:
② Strain energy
③ Potential of the distributed self-weight
From the figure, the axial shortening of an elemental length $\mathrm d x$ due to slope $\theta$ is $\tfrac{1}{2}(y')^{2}\,\mathrm d x$. Under the distributed self-weight $q$, the weight carried above the segment at $x$ is $q(l - x)$, so
Integrating over $[0, l]$:
④ Potential of the concentrated load $q l$ at $C$
The concentrated force $q l$ only travels through the axial shortening of the segment $[0, 2l/3]$ above it:
⑤ Total potential · Stationary condition · Critical load
From $\mathrm d E_{\mathrm P}/\mathrm d a = 0$ with $a \ne 0$:
- The energy method, based on the stationary-potential principle, converts the infinite-DOF problem to $n$-DOF algebra via $y(x) = \sum a_i \varphi_i(x)$;
- Strain energy $U = \tfrac{1}{2}\!\int\! EI(y'')^{2}\,\mathrm d x$; load potential $U_{\mathrm P} = -\tfrac{F_{\mathrm P}}{2}\!\int\!(y')^{2}\,\mathrm d x$;
- Ex. 11-5 shows: the closer the trial function is to the real shape, the higher the accuracy (best 0%, worst 21.6%);
- Ex. 11-6 frame: static method gives $2.706\, EI/l^{2}$; energy method gives $2.727\, EI/l^{2}$ — just 0.78% higher;
- Ex. 11-7 shows: the energy method handles distributed loads + multiple point loads with ease, whereas the static method would be very difficult;
- Ritz gives an upper bound — the better the trial function, the smaller the error;
- §11-5 extends the energy method to compound columns (laced / battened).