Skip to main content
Version: Latest

Creep or earthquakes

A frictional fault either creeps steadily or locks and fails in repeated earthquakes. Rate-and-state friction decides which through two quantities: the sign of and the characteristic slip distance . This tutorial runs the three examples shipped in examples/earthquakes/rate_and_state_friction/, which differ only in those parameters, for 1000 years each:

CaseOutcome
ep_rsf_creep0.0060.003+0.0031 mmsteady creep
ep_rsf_eq0.0030.015−0.0121 mmthree earthquakes
ep_rsf_creep_largeDc0.0030.015−0.012100 mmslow creep, no earthquake

The same model also shows how global velocity scaling carries one simulation through centuries of loading and seconds of rupture. See rate-and-state friction for the equations and for global velocity scaling.

At a glance
Dimension2D
Buildmake ndims=2
Run timeabout 10 minutes per case on 3 threads
Inputsexamples/earthquakes/rate_and_state_friction/ in the DynEarthSol repository
Checked withDynEarthSol 68cc2a0

Setup​

Model setup

Figure 1. The model, with depth exaggerated twice. The top moves right at the loading rate and the base is fixed. P is the monitor point on the top of the band at x = 50 km.

SettingValue
Domain100 km × 10 km (2D), gravity 10 m/s²
Band4–6 km depth, full width, rate-and-state friction
Hostsabove and below the band; cohesion Pa keeps them elastic
Elastic moduli and density = 50 GPa, = 30 GPa, 2700 kg/m³
Band strengthcohesion 4 MPa, = 30°, no dilation, no weakening
RSFaging law, m/s, , and as in the table above
Loadingtop: m/s (3.2 cm/yr), ; base fixed; sides free
Meshrsf_long_strike_model_low_resolution.poly; about 215 m elements in the band, 300 m in the hosts; 32,091 elements
Mechanicsexplicit, global velocity scaling with , damping option 1

The .poly file divides the band into three regions (material IDs 3, 2, 1 from left to right) so that the central 10 km can be given different friction. In these examples all band materials share the same and .

Input and how to run​

Build dynearthsol2d in the DES root. Then:

cd examples/earthquakes/rate_and_state_friction
python3 run_rsf_examples.py --exe ../../../dynearthsol2d
python3 plot_rsf_time_series.py -o rsf_time_series.png

The runner creates one output folder per case and runs the three inputs one after another. On a laptop each case takes about 10 minutes with three OpenMP threads (OMP_NUM_THREADS=3). You can also run a single case directly:

mkdir -p ep_rsf_eq
OMP_NUM_THREADS=3 ../../../dynearthsol2d ep_rsf_eq.cfg

The settings that matter for the cycle are:

[sim]
max_time_in_yr = 1000
output_time_interval_in_yr = 100 # regular frames
earthquake_output_step_interval = 500 # extra frames during an event
earthquake_start_factor = 5.0 # event: fastest speed > 5 x boundary speed, and > 1e-6 m/s
earthquake_end_factor = 2.0

[control]
use_global_velocity_scaling = true
inertial_scaling = 1e5
damping_option = 1

[mat]
rheology_type = elasto-plastic-rate-state-friction
direct_a = [0.003, 0.003, 0.003, 0.003, 0.003]
evolution_b = [0.015, 0.015, 0.015, 0.015, 0.015]
characteristic_velocity = [1e-9]
characteristic_distance = [1e-3]
state_var_model = 1 # aging law

[monitor]
enabled = yes
step_interval = 10
points_x = [50]
points_y = [-4.0] # z in 2D
points_unit = km
output_velocity = yes
output_stress = yes
output_dynamic_friction = yes
output_state_variable = yes

Output​

Each case writes to its own folder (ep_rsf_eq/, ep_rsf_creep/, ep_rsf_creep_largeDc/):

  • <case>.save.* field frames: one every 100 years, plus one every 500 steps while an event is under way. The earthquake case has 40 frames (about 0.7 GB), the creep cases 11 (about 0.2 GB).
  • <case>.chkpt.* checkpoints for restarting.
  • <case>.info: one line per frame with the step, time (s) and time step (s).
  • monitor_point_0.csv: the history at P every 10 steps, with velocity_x, velocity_z, stress_0–stress_2 (σxx, σzz, σxz), dynamic_friction and state_variable.

plot_rsf_time_series.py plots the speed and displacement at P for all three cases. To read the monitor yourself:

import numpy as np
d = np.genfromtxt('ep_rsf_eq/monitor_point_0.csv', delimiter=',', names=True)
year = 365.2422 * 86400
t = d['time_s'] / year
speed = np.hypot(d['velocity_x'], d['velocity_z'])
shear = np.abs(d['stress_2']) / 1e6 # MPa
dt = np.diff(d['time_s']) / np.diff(d['step'])
print(f'{speed.max():.2f} m/s peak; dt from {dt.min():.2g} to {dt.max():.2g} s')

Result​

Thousand-year histories

Figure 2. Histories at P for the three cases: (a) speed, (b) shear stress σxz, (c) friction coefficient, (d) state variable.

Loading (0–620 years). All three cases are identical. The band is still below its Mohr–Coulomb strength, so it deforms elastically, the friction law has no effect, and shear stress at P rises linearly. P moves at about 0.6 of the loading rate, its share of the elastic shear across the 10 km column.

After yield (from about 620 years) the cases separate:

  • Velocity strengthening (). The band creeps at the loading rate and the shear stress stays near 57 MPa. Any acceleration raises friction and damps itself.
  • Velocity weakening (, = 1 mm). The band fails in three earthquakes, at 632, 791 and 950 years, about 159 years apart. Peak speed at P is 0.7–1.6 m/s. Shear stress at P drops by 11–21 MPa in each event and then reloads.
  • Velocity weakening with large (100 mm). The band yields and then creeps while its stress slowly declines, from 59 to 53 MPa by 1000 years. No event occurs. With a 100 times larger , weakening is spread over 100 times more slip, and the critical stiffness falls by the same factor, which favours stable sliding.

First earthquake

Figure 3. The first earthquake at P, from 400 s before to 300 s after peak speed: (a) speed, (b) shear stress, (c) friction coefficient.

Inside an event. Speed at P rises from about m/s to 1.6 m/s in the last few seconds before the peak, and stays above 1 mm/s for about 12 s. The friction coefficient falls from 0.55 to near zero at peak slip rate, because the state collapses as the band slips many times . It then heals back toward 0.45–0.47 within minutes as slip slows. The shear stress swings below its final value and settles at 35 MPa, a static drop of about 21 MPa.

Global velocity scaling at work. In the earthquake case the time step ranges from s (4.5 days) during loading to s at peak slip, a factor of . Each event takes about 2300–2700 steps; the whole 1000-year run takes 98,000 steps. Without the scaling the run would need about steps at the event time step.

Things to try​

  • Change along the band. Give the central region (material 2) a velocity-weakening and the flanks (materials 1 and 3) a velocity-strengthening one, as in the shear-band tutorial. Events then nucleate in the centre and stop in the flanks.
  • Vary between 1 and 100 mm. Find the value at which the earthquakes stop. Compare the recurrence interval and peak speed on the way.
  • Record seismic moment. Set sim.seismic_moment_calculate_output = yes to write each event's start and end times and moment to seismic_moment_magnitude.txt. The moment is per unit length out of the plane in 2D (see Earthquake output). Each of the three earthquakes gives about N·m/m, that is 5 m of average slip on the 100 km layer; it is followed by a few short entries five to seven orders of magnitude smaller, as the speed crosses the thresholds again while the model rings down.
  • Limit the state change per step. Add control.rsf_slip_rate_projection_option = 1 and control.rsf_dtheta_max = 0.5 and check how much the event changes.

Limits​

  • One point. The monitor records the velocity of P on the top of the band, not the slip rate across it. Use the saved fields to see where an event starts and how far it spreads.
  • Resolution. The band elements (about 215 m) are much larger than the nucleation length of the 1 mm case, about 20 m for the lithostatic normal stress at 5 km depth. Event timing and size here are mesh dependent. In the first event the state also reaches its lower bound of s.
  • Damped inertia. Force damping stays on during events, and the default shear wave-speed cap keeps the scaled density above the physical density. The ruptures are therefore damped, quasi-dynamic events, not fully dynamic ones.
  • Slip-rate measure. These inputs use the default velocity projection (rsf_slip_rate_projection_option = 0), which reads the element velocity rather than the velocity jump across it.