Non-Associated Flow

Understanding how materials deform permanently under load is a cornerstone of modern engineering. The theory of plasticity provides a framework for this, but a central question remains: once a material yields, in what direction does it flow? An elegant and historically important answer is the associated flow rule, which posits that the direction of plastic strain is uniquely determined by the same function that defines the material's strength. This creates a beautifully simple and unified model that works well for many metals.

However, this elegant picture shatters when confronted with the behavior of other critical materials, such as soils, rocks, and concrete. For these materials, the associated flow rule makes predictions about volume changes that starkly conflict with experimental reality. This discrepancy reveals a fundamental knowledge gap, forcing us to choose between theoretical simplicity and predictive accuracy. This article addresses this challenge by introducing the concept of non-associated flow, a pragmatic and powerful modification to plasticity theory.

Across the following chapters, we will explore this crucial concept. First, in "Principles and Mechanisms," we will deconstruct the ideal associated flow rule, identify its limitations regarding dilatancy, and introduce the non-associated model as a solution, examining the theoretical trade-offs involved. Subsequently, in "Applications and Interdisciplinary Connections," we will see how this theory is an indispensable tool in fields from geomechanics to metallurgy, and we will uncover the deep consequences it has for computation and structural analysis.

In our journey to understand how materials yield and flow, we now arrive at the very heart of the matter. We have a general idea of a ​​yield surface​​—a boundary in the abstract space of stresses that separates elastic behavior from permanent, plastic deformation. But a crucial question remains: once a material decides to yield, in what direction does it flow? If you push a block of metal, it gets shorter and fatter. But by how much? Is there a universal rule? The answer to this question leads us down a fascinating path, beginning with an idea of remarkable elegance and ending with a pragmatic compromise that reflects the beautiful complexity of the real world.

Imagine you are standing on a rolling hill, and you place a ball on its side. Which way will it roll? It will roll straight down the steepest path. This path, as any mathematician will tell you, is perpendicular, or ​​normal​​, to the contour line at that point. In the early days of plasticity theory, a similar, beautiful idea was proposed: perhaps plastic flow behaves the same way. The "hill" is the landscape of stress, and the "contour lines" are the yield surfaces. The idea, known as the ​​associated flow rule​​, states that the direction of the plastic strain rate is always normal to the yield surface at the current stress state.

This is a wonderfully simple and powerful concept. It means that the single mathematical function that defines the yield boundary, which we call the ​​yield function​​ $f(\boldsymbol{\sigma}, \dots)$, also governs the direction of flow. The direction is simply given by the gradient of the yield function, $\frac{\partial f}{\partial \boldsymbol{\sigma}}$. This is why we call it "associated"—the flow rule is directly associated with the yield function.

This "normality rule" is not just a pretty guess. It is deeply connected to fundamental principles of stability. A postulate by Daniel C. Drucker suggests that stable materials should behave in a certain predictable way. One of the consequences of this postulate is that for a material with a convex yield surface (meaning it doesn't have any inward-curving parts), the normality rule must hold (2711718). This connection between stability and associated flow is profound. It implies that the plastic dissipation—the energy converted into heat during plastic flow—is maximized. In other words, for a given increment of plastic strain, the material "chooses" the stress state on the yield surface that does the most work (2654976). Nature, in this ideal picture, appears to follow a principle of maximum effort.

The elegance is compelling: one function, $f$, defines both the limit of strength and the law of flow. For many materials, especially metals under conditions where pressure doesn't significantly affect their yielding (like in the von Mises yield criterion), this beautiful picture works exceptionally well (2711713). The plastic flow is incompressible, meaning the volume doesn't change, and the theory matches experiments.

However, nature is often more subtle than our most elegant theories. When we turn our attention from ductile metals to other crucial materials like soils, rocks, concrete, and granular aggregates, a serious crack appears in this beautiful edifice. These materials are ​​pressure-dependent​​; their strength increases the more they are squeezed. Think of a pile of sand: it’s much harder to push your finger into it if it's confined and compressed.

Yield criteria like the Mohr-Coulomb or Drucker-Prager models were developed to capture this frictional behavior. They define yield surfaces that are not simple cylinders in stress space (like von Mises) but are more like cones or pyramids, where strength increases with hydrostatic pressure. Now, what happens if we apply the associated flow rule to these yield surfaces?

Let’s visualize this in a simplified 2D stress space with coordinates for pressure, $p$, and shear stress, $q$. For a frictional material, the Mohr-Coulomb yield surface is a sloped line (2559749). The slope of this line is related to the material's internal ​​friction angle​​, $\phi$. If we apply the normality rule, the plastic flow vector must be perpendicular to this line. Because the line is sloped, the normal vector has components in both the shear and pressure directions. A flow in the pressure direction corresponds to a change in volume. The associated flow rule predicts that as the material is sheared, it must also expand in volume. This phenomenon is called ​​dilatancy​​, and it's very real. If you shear a dense bag of sand, it will expand.

Here's the problem: the amount of dilatancy predicted by the associated flow rule is directly coupled to the friction angle $\phi$. The higher the friction, the more the material is predicted to expand. And for most soils and rocks, this prediction is wildly inaccurate—it's often an over-prediction by a very large factor (2711713). The elegant unity of the associated flow rule breaks down when confronted with the experimental data for these common and critically important materials.

When a beautiful theory doesn't fit the facts, we have a choice. We can discard it, or we can see if a clever modification can save it. In this case, physicists and engineers chose the latter. The problem was that a single function, $f$, was being asked to do two jobs: define strength and define flow. And it was failing at the second job.

The solution was to introduce a second, independent function, called the ​​plastic potential​​, $g(\boldsymbol{\sigma}, \dots)$ (2671041). The new rule is this:

• The ​​yield function​​ $f$ still defines the boundary of elasticity. The material yields only when the stress hits the surface where $f=0$.
• The ​​plastic potential​​ $g$ now defines the direction of flow. The plastic strain rate is normal to the level surfaces of $g$, not $f$.

This is called a ​​non-associated flow rule​​ because the flow is no longer associated with the yield function. By choosing $g \neq f$, we can decouple the material's strength from its flow behavior.

Let's return to our picture of the Mohr-Coulomb material in the $p-q$ plane (2559749).

• The yield surface $f=0$ is still a line whose slope is governed by the friction angle $\phi$, which we measure from strength tests.
• We can now introduce a plastic potential $g=0$ that is another line, but its slope is governed by a different angle, the ​​dilation angle​​ $\psi$.

The plastic flow vector is now normal to the $g=0$ line. If we choose a dilation angle $\psi$ that is smaller than the friction angle $\phi$, the new flow vector will be tilted more vertically. It will have a smaller component in the pressure direction, which means less volume expansion. We can choose $\psi$ to match the experimentally observed dilatatancy, all while leaving the friction angle $\phi$ to correctly model the material's strength. We have traded elegance for accuracy. The ratio of the predicted volumetric strain in the non-associated model to the associated one is simply the ratio of their respective slope parameters (2893867). This gives engineers precise control over the model's predictions.

This clever fix is not without its costs. By introducing a second function and breaking the link between yield and flow, we lose some of the beautiful mathematical properties that came with the associated model.

First, we explicitly violate the principle of maximum plastic dissipation (2655028). The material no longer follows the path of "maximum effort." The actual stress state is no longer the one that does the most work for a given plastic strain. This has consequences for macroscopic theories like limit analysis, where the clean correspondence between upper and lower bounds on collapse loads becomes blurred (2654976).

Second, the guarantee of material stability given by Drucker's postulate is lost. Non-associated models can, under certain conditions, predict unstable behavior, such as strain localization, where deformation concentrates in very narrow bands.

Third, a very practical consequence for computational engineers is that the governing equations lose a key property: ​​symmetry​​. When simulating these materials using methods like the finite element method, the internal stiffness matrix of the material model, which relates stress rates to strain rates, becomes non-symmetric (2631371). This means that the numerical algorithms used to solve the equations become more complex, less efficient, and require more memory. It is a significant practical price to pay for a more realistic material description.

With all these losses, a deep question arises: Is a non-associated model even physically legitimate? Does it violate some fundamental law of nature? The ultimate arbiter here is the Second Law of Thermodynamics, which demands that in any irreversible process, the total dissipation (energy lost to heat) cannot be negative.

For a plastic material, this boils down to the condition that the plastic dissipation rate, $\boldsymbol{\sigma} : \dot{\boldsymbol{\varepsilon}}^{p}$, must be non-negative. Let's check this for our non-associated rule $\dot{\boldsymbol{\varepsilon}}^{p} = \dot{\lambda} \frac{\partial g}{\partial \boldsymbol{\sigma}}$, where $\dot{\lambda}$ is a positive scalar representing the rate of plastic flow. The dissipation is $\dot{\lambda} (\boldsymbol{\sigma} : \frac{\partial g}{\partial \boldsymbol{\sigma}})$. To ensure this is non-negative, we need $\boldsymbol{\sigma} : \frac{\partial g}{\partial \boldsymbol{\sigma}} \ge 0$.

It turns out that this condition can be satisfied without requiring $g=f$. A sufficient condition is for the plastic potential $g$ to be a ​​convex function​​ that has its minimum value at the zero-stress state (2711752). This gives us our answer: non-associated flow does not inherently violate the Second Law of Thermodynamics. While it sacrifices the elegance of stability postulates and computational symmetry, it can be formulated in a way that is perfectly consistent with the fundamental laws of energy.

In the end, the story of non-associated flow is a perfect example of the scientific process. We start with a simple, beautiful theory. We test it against reality and find its limits. We then modify it, making it more complex but also more powerful and accurate, sacrificing some elegance in the process. And all the while, we check back with the fundamental laws of physics to make sure our new creation, for all its pragmatic complexity, is still standing on solid ground.

Now that we have grappled with the principles of non-associated flow, you might be wondering, "Is this just a mathematical curiosity?" Far from it. This is where the story gets truly exciting. The distinction between a yield function $f$ and a plastic potential $g$ is not a mere academic subtlety; it is a vital tool that unlocks a more profound and accurate understanding of the world around us. It is the key that separates a simplified cartoon of nature from a model that can predict the behavior of real materials, from the ground beneath our cities to the metallic bones of our most advanced machines. Let's embark on a journey to see where these ideas take us.

The most immediate and fundamental application of non-associated flow is its ability to describe materials as they truly are. Think of a dense pile of sand. To make it "yield," or start to flow, you need to push on it with a certain amount of shear stress, and this ability to resist shear is related to the friction between the grains. This is what the yield function, $f$, describes. But what happens once it starts to flow? The grains, which are tightly packed, have to jostle and roll over one another. To do this, they must push each other apart, and the pile as a whole expands in volume. This phenomenon is called ​​dilatancy​​.

In a simple, "associated" world, one might guess that the amount of a material's expansion (dilatancy) is directly locked to its internal friction. But experiments tell us this is not the case for many materials, including soils, rocks, and concrete. A material's friction angle, which determines when it yields, is often significantly larger than its dilatancy angle, which describes how much it expands as it deforms. An associated flow rule, where the flow direction is dictated by the yield function itself, forces these two properties to be linked. This often leads to a massive overestimation of the predicted volume expansion.

Here is where the non-associated flow rule rides to the rescue. By introducing a separate plastic potential, $g$, we can decouple these two effects. The yield function $f$ continues to define the "stress budget" — how much stress the material can take before it yields. But the plastic potential $g$ gets its own independent parameter to describe the direction of plastic flow—specifically, the amount of volumetric expansion,. By tuning the parameter in $g$ (often called a dilatancy parameter, $\beta$ or $\psi$), we can accurately match the observed volume change without corrupting our description of the material's strength.

But how can we be sure we're not just playing a mathematical game? We can measure it. Imagine taking a metal bar that is sensitive to pressure and pulling on it in a standard uniaxial tension test. As we pull it, it gets longer in the axial direction and thinner in the lateral directions. For plastic deformation, we might naively expect the volume to stay constant. However, by carefully measuring the axial and lateral strains and separating the plastic part from the elastic part (a standard procedure in materials testing), we can calculate the plastic Poisson's ratio. A non-associated flow rule predicts that this ratio will depend directly on the parameter in the plastic potential, $\beta$. If our measurements reveal plastic volume changes, we have caught non-associativity red-handed, in our own laboratory!.

The classic home for non-associated plasticity is ​​geomechanics​​. The behavior of soils, rock, and concrete under load is fundamentally pressure-sensitive and non-associated. When engineers design foundations, tunnels, or dams, their computer models must account for the fact that the friction angle $\phi$ of the soil is not equal to its dilatancy angle $\psi$. Advanced models like the Modified Cam-Clay for clays explicitly incorporate non-associated flow rules to better predict how soil will compact or dilate under different loading conditions, which is crucial for predicting settlement or failure.

But the reach of this concept extends far beyond civil engineering. Consider the field of ​​metallurgy and damage mechanics​​. When a metal component is stretched to its limits, tiny microscopic voids can form and grow within it. The growth of these voids causes the material to swell, an effect that is, in essence, a form of dilatancy. Conversely, under high compressive pressure, these voids can be crushed, causing the material to compact. To model the failure of ductile metals, scientists and engineers use sophisticated models like the Gurson-Tvergaard-Needleman (GTN) model. And at the heart of these models? You guessed it: a non-associated flow rule. The yield surface describes the combination of stress and void content that the metal can withstand, while a separate plastic potential governs the rate at which the voids grow or shrink, ultimately leading to fracture. It is a beautiful example of a unifying principle: the same fundamental idea that describes a landslide can also describe the tearing of a steel plate.

The choice to use a non-associated flow rule is not a simple tweak. It sends deep ripples through our understanding of physics, mathematics, and computation.

First, consider the ​​geometry of motion​​. In the classical theory of plasticity for perfectly plastic materials, there exist "slip-lines," which are natural paths along which the material prefers to deform. For an associated material, a wonderful simplification occurs: the directions of maximum shear stress and the directions of the plastic velocity are one and the same. The stress characteristics and velocity characteristics coincide. Non-associativity breaks this elegant symmetry. The material still yields along the lines of maximum shear (governed by the yield function $f$), but it moves along a different set of characteristic lines (governed by the plastic potential $g$). The map of stress and the map of motion become decoupled, revealing a more complex and subtle choreography of failure.

Second, this choice poses a formidable challenge to ​​computational mechanics​​. Modern engineering relies on the Finite Element Method (FEM) to simulate and design virtually everything. When we implement a material model in an FEM code, a critical part of the process is verifying that the code actually solves the equations we think it's solving. A powerful verification test for a non-associated model involves running a simulation for a single point under a simple, controlled load and checking that the ratio of plastic volume change to plastic shear deformation precisely matches the dilatancy parameter from the plastic potential, $g$—not the friction parameter from the yield function, $f$.

More profoundly, adopting a more realistic non-associated model can cause our simulations to become unstable. The numerical engine of most FEM codes, the Newton-Raphson method, works best when the underlying system of equations has a symmetric and positive-definite structure. Non-associated flow rules destroy this. The resulting "tangent matrix," which guides the numerical solution, becomes non-symmetric. This is not a bug; it is a feature. It is the mathematical signature of a deep physical property. This non-symmetry can lead to a loss of material stability even when the material is still hardening. This is precisely the kind of instability that leads to phenomena like shear bands—narrow zones of intense deformation that are precursors to catastrophic failure. The numerical algorithm's struggle for convergence is the physics telling us that something dramatic is about to happen,. Engineers have developed clever numerical strategies, like line-searches or adaptive regularization, to navigate these treacherous computational waters and successfully simulate the behavior of these complex materials.

Finally, non-associativity forces us to reconsider one of the most elegant and powerful tools in structural engineering: ​​limit analysis​​. The classical theorems of limit analysis allow engineers to calculate rigorous upper and lower bounds on the collapse load of a structure using remarkably simple methods. The upper bound theorem, in particular, relies on a beautiful idea called the principle of maximum plastic dissipation, which states that among all possible stress states, the true stress state is the one that does the most work on the deforming plastic material. However, this principle is only guaranteed to hold for associated materials.

For a non-associated material, the actual dissipated energy is less than this theoretical maximum. As a result, the classical upper bound theorem, which uses this maximum value to estimate the collapse load, can give an answer that is too high. The "upper bound" is no longer a guaranteed safe estimate and may dangerously overestimate the structure's strength. This is a sobering and profound lesson. Even our most beautiful and trusted theorems have their limits, and understanding the subtle physics of non-associated flow is what allows us to see the "Warning: Here Be Dragons" signs on our intellectual maps. It reminds us that our journey to understand nature is a continuous process of refinement, where each new layer of complexity reveals both new challenges and a deeper, more satisfying truth.