Energy Efficient Building Climate Control using

Stochastic Model Predictive Control and Weather Predictions

Frauke Oldewurtel, Alessandra Parisio, Colin N. Jones, Manfred Morari
Dimitrios Gyalistras, Markus Gwerder, Vanessa Stauch, Beat Lehmann, Katharina Wirth

Abstract— One of the most critical challenges facing society and simulation environment for building climate control
today is climate change and thus the need to realize massive developed within the OptiControl1 project, which focuses on
energy savings. Since buildings account for about 40% of global the development of predictive control strategies for building
final energy use, energy efficient building climate control can
have an important contribution. In this paper we develop and climate control. A bilinear model is used for both simulation
analyze a Stochastic Model Predictive Control (SMPC) strategy and control. The crucial part of the control problem is how to
for building climate control that takes into account weather deal with the inherent uncertainty due to weather predictions.
predictions to increase energy efficiency while respecting con- The following controllers are assessed:
straints resulting from desired occupant comfort. We investigate
a bilinear model under stochastic uncertainty with probabilistic, • Rule Based Control (RBC): Current practice. The con-
time varying constraints. We report on the assessment of this trol inputs are defined with simple rules: “if condition
control strategy in a large-scale simulation study where the then action”.
control performance with different building variants and under • MPC: Two different MPC schemes are considered.
different weather conditions is studied. For selected cases the
SMPC approach is analyzed in detail and shown to significantly The first strategy follows common practice, which is
outperform current control practice. to simply neglect the uncertainty in the problem and
is therefore termed Certainty Equivalence (CE). The
I. INTRODUCTION second strategy takes into account the uncertainty in the
A. Integrated Room Automation controller directly and solves a stochastic MPC (SMPC)
In building climate control heating, ventilation, and air- problem. For this, we follow the approach introduced in
conditioning (HVAC) systems are employed to keep room [11].
temperature within a predefined range, the so-called comfort • Performance Bound (PB): Optimal control action given
range. In this paper we focus on Integrated Room Automa- perfect knowledge of future weather. This is an ultimate
tion, where the building system consists of an HVAC-system, bound on the performance of any controller, and thus
an automated lighting system, and a blind positioning system used as a theoretical benchmark.
[4], [9]. The control task is to keep the room temperature
as well as CO2 and illuminance levels within a predefined C. Outline
comfort range, which can be fulfilled with a set of different In Section II the modeling is described in detail. This is
actuators. The actuators differ in terms of response time divided into two parts, the building modeling and the weather
and effectiveness, in their dependence on weather conditions uncertainty modeling. In Section III the control strategies
(e.g. cooling tower or blinds), and in energy costs. The are presented. First, the RBC strategy is explained, then the
goal is to optimally choose the actuator settings depending MPC problem is posed. This can be solved by neglecting the
on future weather conditions in order to fulfill the comfort uncertainty as in CE or by directly taking it into account as in
requirements and minimize energy costs. SMPC. Both approaches are presented in detail. Finally, the
B. Assessment of Control Strategies PB is introduced. Section IV introduces the concept of the
controller assessment and describes the setup of the large-
Aiming at investigating how much energy can be saved
scale simulation study. The simulation results are presented
with advanced control techniques we compare Model Pre-
in Section V.
dictive Control (MPC) strategies taking into account wea-
ther predictions with current best-practice control. For this
assessment we use BACLab, a MATLAB-based modeling D. Notation

Frauke Oldewurtel, Colin N. Jones and Manfred Morari are with the
The real number set is denoted by R, the set of non-
Automatic Control Laboratory, Department of Electrical Engineering, Swiss negative integers by N (N+ := N\{0}). For matrices A
Federal Institute of Technology in Zurich (ETHZ), Switzerland. and B of equal dimension inequalities A{<, ≤, >, ≥}B hold
{oldewurtel,jones,morari}@control.ee.ethz.ch component-wise. The expectation of a stochastic variable w
A. Parisio is with the Department of Engineering, Universitá degli
Studi del Sannio, Benevento, Italy; M. Gwerder is with Siemens Building given the observation τ is denoted by E[w|τ ]. The probability
Technologies, Zug, Switzerland; D. Gyalistras is with the Systems Ecology of an event ρ is denoted by P[ρ].
Group, ETH Zurich, Switzerland; V. Stauch is with MeteoSwiss, Zurich,
Switzerland; B. Lehmann and K. Wirth are with Building Technologies
Lab., EMPA, Dübendorf, Switzerland. 1 www.opticontrol.ethz.ch
 II. M ODELING
A. Building Model
For computing the building-wide energy use it is common
practice to sum the energy uses of single rooms or building
zones [4]. We follow this approach and focus on the dyna-
mics of a single room. We first explain the building thermal
dynamics in detail and then the different actuators.
Remark 1: Illuminance and CO2 concentration were mode-
led by instantaneous responses since the time constants invol-
ved were much smaller than the hourly time step employed Fig. 1. Heat transmission between nodes. The modeling is based on the
for our modeling and simulations and modelling details of description of heat transmission between nodes (left) that are representing
these are ommitted for brevity. The interested reader can find the temperatures at different locations in the building (right).
the details on this in [10].
The principle of the thermal dynamics modeling can easily • Cooling: evaporative cooling (wet cooling tower) /
be described with a small example as given in Figure 1. The mechanical ventilation / chilled ceiling / TABS
room can be thought of as network of first-order systems, • Ventilation: with/without mechanical ventilation (inclu-
where the nodes are the states x and these are representing ding energy recovery); with/without natural night-time
the room temperature or the temperatures in the walls, floor ventilation
or ceiling. Then the heat transfer rate is given by The delivered heating or cooling power, the used air change
rates as well as lighting and blind positioning correspond to
dQ the control inputs u. The control task consists of finding
= Kie · (ϑe − ϑi )
dt the optimal combination of inputs that differed in their
dQ dϑi (1) weather dependence, dynamical effects and energy use. For
⇒ · = Kie · (ϑe − ϑi ),
dϑi dt example mechanical ventilation, which provides the room
|{z}
Ci with fresh air to guarantee indoor air quality, can be used
both for cooling and for heating, depending on the weather
where t denotes the time, ϑi and ϑe are the temperatures conditions. But heating can also be done with radiators,
in layers i and e respectively, Q is thermal energy, and Ci which are independent of weather conditions. TABS can be
denotes the thermal capacitance of layer i. The total heat used for heating and cooling but are much slower compared
transmission coefficient Kie is computed as to ventilation or radiators etc. Further details on the experi-
1 1 1 mental setup can be found in [9].
= + , (2)
Kie Ki Ke Assumption 1: The room dynamics are described as
where the heat transmission coefficients Ki and Ke depend
xk+1 = Axk + Buk + ...
on the materials of i and e as well as on the cross sectional
m
area of the heat transmission. For each node, i.e. state, X (3)
... + Bv vk + [(Bvu,i vk + Bxu,i xk )uk,i ]
such a differential equation as in (1) is formulated. Control
i=1
actions were introduced by assuming that selected resistances
were variable. For example, solar heat gains and luminous where xk ∈ Rn is the state, uk ∈ Rm is the input, and
flux through the windows were assumed to vary linearly vk ∈ Rp is the weather input at time step k, and the matrices
with blinds position, i.e. the corresponding resistances were A, B, Bv , Bvu,i , and Bxu,i are of appropriate sizes. The
multiplied with an input u ∈ [0, 1]. This leads to a bilinear sampling time is 1 hour.
model, i.e. bilinear in state and input as well as in disturbance The overall building model was validated by building experts
and input. A detailed description of the building model can [10] and its dynamic response compared to simulations with
be found in [10]. TRNSYS 3 , which is a well known simulation software for
Concerning the actuators, we investigated several variants buildings and HVAC systems. It was found that the model
of building systems in integrated room automation. All captures sufficiently well the relevant behavior of a building
system variants had the basic actuators blind positioning and [10].
electric lighting. They employed different combinations of
B. Weather Uncertainty Model
the following subsystems:
The weather predictions were given by archived forecasts
• Heating: radiators / mechanical ventilation / floor hea-
of the COSMO-7 numerical weather prediction model ope-
ting / TABS2
rated by MeteoSwiss. The data comprised the outside air
2 TABS = Thermally activated building system, i.e. the building mass temperature, the wetbulb temperature and the incoming solar
is incorporated as thermal storage for heating and cooling purposes and
activated by a tube-system located in the slabs. 3 http://sel.me.wisc.edu/trnsys/
radiation. COSMO-7 delivers hourly predictions for the next A. Rule Based Control
three days with an update cycle of 12 hours [6]. The major The standard strategy in current practice and used by,
challenge from a control point of view with using numerical amongst others, Siemens Building Technologies is rule-based
weather predictions lies in their inherent uncertainty due to control [5]. As the name indicates, RBC determines all
the stochastic nature of atmospheric processes, the imperfect control inputs based on a series of rules of the kind “if
knowledge of the weather models initial conditions as well condition then action”. As a benchmark we used here RBC-
as modeling errors. The actual disturbance acting on the 5 as defined in [7]. This is the currently best RBC controller
building can be decomposed as known to us that assumes hourly blind movement as the other
control strategies considered in this study.
vk = v̄k + ṽk , (4)
B. Model Predictive Control Approach
where v̄k is the COSMO-7 weather prediction and ṽk is the
For the MPC approach, we employ the model of (6).
prediction error at each time step k. In order to improve
the estimation of future disturbances acting on the building Remark 2: We substitute (4) in (6) and use (5) to extend
the following autoregressive model driven by Gaussian noise the model, such that the resulting model depends linearly on
was identified based on the archived weather predictions and the Gaussian disturbance w.
corresponding in-situ measurements Consider the prediction horizon N ∈ N+ and define
T T
x := xT ∈ R(N +1)n
 
ṽk+1 = F ṽk + wk , (5) 0 , . . . , xN
T
u := u0 , . . . , uT
 T
∈ RN m

where F ∈ Rp×p and wk ∈ Rp . N −1
T
w := w0T , . . . , wN T
∈ RN p

Assumption 2: The disturbance wk follows a Gaussian dis- −1
 T T T
∈ RN p

tribution, wk ∼ N (w¯k , ΣΣT ), ∀k. w̄ := w̄0 , . . . , w̄N −1
Testing the randomness of residuals showed that the good-
and prediction dynamics matrices A, B and E such that
ness of fit was satisfying for all investigated cases, i.e.
autocorrelation coefficients for the the residuals did not differ x = Ax0 + Bu + Ew .
significantly from zero.
We assume that all inputs u can vary between zero and an
The model in (5) is used twofold: first, for augmentation of upper bound (full operation), which defines hard constraints
the controller model such that it accounted for the effects of on the inputs. The room temperature is constrained to lie
the uncertain weather predictions on the building’s dynamics, within upper and lower bounds. Motivated by European buil-
and second for continuous correction of the COSMO-7 ding standards, e.g. [14], we do not require constraints to be
weather predictions based on hourly weather measurements satisfied at all times, but only with a predefined probability,
with a standard Kalman filter. which is formulated with so-called chance constraints.
C. Overall Model Assumption 3: The constraints on inputs u and states x
over the prediction horizon N are
The dynamic behavior of the building is nonlinear; in this
case bilinear between inputs, states and weather parameters. Su ≤ s
Non-linearities in the dynamic equations of an MPC problem (7)
∧ Gx ≤ g,
will generally result in a non-convex optimization, which can
be extremely difficult to solve. The approach that we take where S ∈ Rq×mN , s ∈ Rq , G ∈ Rr×n(N +1) , and g ∈
here is a form of Sequential Linear Programming (SLP) for Rr and the state constraints are applied according to the
solving nonlinear problems in which we iteratively linearize definition in the standards as chance constraints
the non-convex constraints around the current solution, solve P[Gi x ≤ gi ] ≥ 1 − αi , ∀i = 1, ..., r, (8)
the optimization problem and repeat until a convergence
condition is met [3]. To keep formulations simple, we will where 1 − αi with αi ∈ [0, 1] denotes the probability level.
assume for the remainder of the paper that we do the This means that we formulate the chance constraints on
linearization at each hourly time step k, which results in the states as individual chance constraints, i.e. each row i
the new input matrix Bu,k and formulate the problem for has to be individually fulfilled with the probability 1 − αi .
the linear system of the form For some initial state x0 the control objective is to mini-
mize energy usage.
xk+1 = Axk + Bu uk + Bv vk . (6)
Assumption 4: A linear cost function V : R → R
III. C ONTROL S TRATEGIES N −1
X
In this section we present the different control strategies V (x0 , u0 , ..., uN −1 ) := cT · uk (9)
that are considered in our assessment. These are RBC, which k=0
is current practice, MPC, which takes into account weather is assumed, where x0 ∈ Rn is the initial state and c ∈ Rm is
predictions, and PB, which is a theoretical benchmark. the cost of the different actuators, i.e. a scaling factor, that
considers the non-renewable primary energy usage of each In matrix form this leads to
actuator, see [9].
u =
Mw + h (12)
The optimal control input u over the prediction horizon
 
0 ··· ··· 0
N is determined by solving MPC Problem 1.  M1,0 0 ··· 0
M :=  (13)
 
Problem 1: .. .. .. .. 
 . . . .
u∗ (x0 )M P C := arg min E[cT u|x0 ] MN −1,0 ··· MN −1,N −2 0
u
s.t. P[Gi (Ax0 + Bu + Ew) ≤ gi ] ≥ 1 − αi ∈ RN m×N p
(10) T
h := hT T
∈ RN m .

∀i = 1, ..., r 0 , . . . , hN −1

Su ≤ s Remark 4: The formulation (12) sets the inputs to be affine

where c ∈ R Nm
denotes the cost vector for the whole functions of normally distributed random variables having
horizon. Problem 1 is a dynamic programming problem, it an unbounded value range. Input constraints can thus not
is however not clear how to solve it since it depends on the be guaranteed for all possible outcomes of the disturbance,
disturbance w which follows a Gaussian distribution. There which renders infeasible optimization problems, unless M =
are two principal ways to proceed: 1) The standard procedure 0. A possible approach to addressing this issue is to relax the
is to assume w = w̄ and solve a deterministic MPC hard input constraints and restrict the constraint satisfaction
problem. This problem is known as Certainty Equivalence. to subsets with prescribed probability levels. We thus define
2) We follow the approach introduced in [11] and solve the chance constraints also on the inputs, not only on the states,
stochastic MPC problem approximately. as follows

1) Certainty Equivalence: With the assumption that w = P[Sj u ≤ sj ] ≥ 1 − αu,j , ∀j = 1, ..., q. (14)
w̄, Problem 1 simplifies to Since the constraints on inputs are hard constraints, it is
Problem 2: desirable to impose a higher probability of satisfaction on
u∗ (x0 )CE := arg min cT u input constraints. We denote this probability level by 1−αu,j
u for constraint j.
s.t. G(Ax0 + Bu + Ew̄) ≤ g (11)
As a second step we do a deterministic reformulation of
Su ≤ s the chance constraints on the states and on the inputs. With
This is now a deterministic problem known as nominal MPC the affine disturbance feedback, the chance constraints on the
problem and is readily solvable. states are now of the form
Remark 3: Note that the state constraints are only guaran- P[Gi (Ax0 + BMw + Bh + Ew) ≤ gi ] ≥ 1 − αi . (15)
teed to be satisfied for w = w̄. Thus, other outcomes of w
are likely to violate the constraints. Note that the functions describing the constraints in (15) are
bi-affine in the decision variables and the disturbances. It is
2) Stochastic Model Predictive Control: Following the well-known that if the disturbance is normally distributed,
assumption that the disturbance w is normally distributed, the functions are bi-affine in the decision variables and the
we get a stochastic MPC problem which is not readily solva- disturbances are considered in the constraints, then indivi-
ble. We reformulate and approximate the stochastic control dual chance constraints can be equivalently formulated as
problem in two steps. First, we approximate the dynamic deterministic second order cone constraints [13] as follows
programming problem and second, we define a convex,
deterministic reformulation of the probabilistic constraints in Φ−1 (1 − αi )kGi (BM + E)k2 ≤ gi − Gi (Ax0 + Bh) (16)
order to cast the SMPC Problem 1 as a convex and tractable where Φ is the standard Gaussian cumulative probability
optimization problem, which can be solved at each time step. function. The inequalities (16) are second order cone cons-
For approximating the dynamic programming problem we traints that are convex in the decision variables M and h.
employ affine disturbance feedback, which has shown good We obtain the following convex deterministic second-order
performance in robust MPC problems [1], [2]. cone program (SOCP).
With affine disturbance feedback, the control inputs are
parameterized as affine functions of the disturbance sequence Problem 3:
as follows (M∗ (x0 ), h∗ (x0 )) := arg min cT (Mw̄ + h)
Pi−1 (M,h)
ui = hi + j=0 Mi,j wj
s.t.
hi ∈ R m
Φ−1 (1 − αi )kGi (BM + E)k2 ≤ gi − Gi (Ax0 + Bh)
Mi,j ∈ Rm×p ∀j ∈ Nk0
Φ−1 (1 − αu,j )kSj Mk2 ≤ sj − Sj h
−1
∀(i, j) ∈ N × NN 0 . By doing so, the optimization in
MPC Problem 1 can be solved in a computationally efficient Remark 5: Please note that the expected value of the linear
fashion using convex optimization methods. cost is only affected by the mean, not the covariance.
C. Performance Bound • Weather conditions: We used weather data from four
PB is not a controller that can be implemented in reality; locations (Lugano, Marseille, Zurich, Vienna) being
it is a concept. PB is defined as optimal control with perfect representative for different climatic regions within Eu-
weather and internal gains predictions and thus gives an rope. All weather predictions and observations were
ultimate bound on what any controller can achieve. historical data of 2007.

Remark 6: In order to compute the PB, we solve an MPC V. RESULTS

problem, but with perfect weather predictions and a very A. Theoretical Potential Analysis
long prediction horizon (6 days). Evaluated is the annual total (all automated subsystems)
IV. CONTROLLER ASSESSMENT CONCEPT AND non-renewable Primary Energy (NRPE) usage and the annual
SIMULATION SETUP amount of thermal comfort violations (integral of room
temperature above or below comfort range limits). Here we
A. Controller Assessment Concept
report comparison results for the found 1228 cases where
The aim is to estimate the potential of using MPC and the amounts of violations by RBC are < 300 Kh/a. Figure
weather predictions in building climate control. For this 3 shows the joint cumulative distribution function of the
purpose the simulation study was carried out in two steps, theoretical energy savings potential (as additional energy use
which is shown in Figure 2: in % of PB) and the amount of comfort violations in Kh/a.
It can be seen that more than a half of the considered cases
show an additional energy use of more than 40 %. Thus, for
many cases there is a significant savings potential, which can
potentially be exploited by MPC.

Fig. 2. Controller assessment concept. First the theoretical potential

is assessed (comparison of RBC and PB), then the practical potential is Amount of violations [Kh/a]

assessed (comparison of RBC and MPC).

Additional energy use in % of PB

1) Theoretical potential: The first step consists of the Fig. 3. Joint cumulative distribution function of a particular additional
comparison of RBC and PB. This is done because there energy use with RBC in % of PB and a particular amount of violations in
Kh/a.
is only hope for a significant improvement, if the gap
between RBC and PB is large. This investigation is
done in a systematic large-scale factorial simulation B. Practical Potential Analysis
study for a broad range of cases representing different With real weather predictions, it can happen that cons-
buildings and different weather conditions as described traints are being violated. Therefore, controller performance
below. For further details see [7], [8]. is assessed in terms of both energy usage and constraint
2) Practical potential: In this investigation we compare violation.
the performance of RBC and MPC strategies only
for selected cases from the theoretical potential study.
Further details can be found in [12].
B. Simulation Setup
For the potential assessment there were on total 1228 cases
considered. The variants were done for the HVAC system,
the building itself and its requirements, and the weather
Typical violation level
conditions. The different variants are listed here:
• HVAC system: Considered are five building system
variants (cf. Section II. A).
• Building: The factors vary in building standard (Passive Fig. 4. Comparison of SMPC and RBC.
House/Swiss Average), construction type (light/heavy),
window area fraction (low/high), internal gains level Figure 4 depicts the result of the comparison of SMPC
(occupancy plus appliances; low/high; also associated and RBC for the selected set of experiments: MPC has
CO2 -production), facade orientation (N or S for normal always clearly less energy use than RBC and in four of six
offices, and S+E or S+W for corner offices). cases smaller amounts of violations. This indicates that the
additional energy use with RBC can be reduced significantly
with MPC.
Figures 5 and 6 show the resulting room temperature
profile throughout the whole year for one of the selected
cases with the RBC and SMPC respectively. It can be seen
that SMPC has smaller and less frequent violations than
RBC. Furthermore, the diurnal temperature variations are
much smaller with SMPC, which is a much more favourable
behavior for the room comfort.
Fig. 7. Tradeoff for SMPC and CE between energy use and violation and
comparison with PB and RBC.

and which low/high-cost energy sources are needed to keep

the room temperature in the required comfort levels. SMPC
was shown to outperform both rule-based control (RBC) as
well as a predictive non-stochastic controller (CE). Further
benefits or SMPC are easy tunability with a single tuning
Time step [h] parameter describing the level of constraint violation as well
Fig. 5. Room temperature profile of RBC for one year.
as comparatively small diurnal temperature variations.
VII. ACKNOWLEDGMENTS
Swisselectric Research, CCEM-CH and Siemens Building
Technologies are gratefully acknowledged for their financial
support of the OptiControl project.
R EFERENCES
[1] A. Ben-Tal and A. Goryashko and E. Guslitzer and A. Nemirovski,
“Adjustable robust solutions of uncertain linear programs”, in Mathe-
matical Programming, vol. 99(2), 2004, pp. 351-376.
Time step [h] [2] P. J. Goulart and E. C. Kerrigan and J. M. Maciejowski, “Optimization
Fig. 6. Room temperature profile of SMPC for one year. over state feedback policies for robust control with constraints”,
Automatica, vol. 42(4), 2006, pp. 523-533.
[3] R. E. Griffith and R. A. Steward, “A nonlinear programming technique
Usage of CE for the six selected cases generally yield for the optimization of continuous processing systems”, Journal of
much more violations than allowed in the standards (results Management Science, vol. 7, 1961, pp. 379-392.
[4] M. Gwerder, J. Toedtli, “Predictive control for integrated room auto-
not shown). One can however tune CE by assuming a mation”, CLIMA 2005, Lausanne, 2005.
tighter comfort band for the controller, which results in less [5] M. Gwerder, J. Toedtli, D. Gyalistras, “Ruled-based control strategies”,
violations and more energy use. Thus, for different comfort in [6].
[6] D. Gyalistras, M. Gwerder (Eds.), “Use of weather and occupancy
band widths one gets a tradeoff curve between energy use and forecasts for optimal building climate control (OptiControl): Two
violations. Similarly, SMPC can be tuned. This is however years progress report”, Technical report, ETH Zurich, Switzerland and
much easier, since there exists a natural tuning knob, the Siemens Building Technologies Division, Siemens Switzerland Ltd.,
Zug, Switzerland, 2009.
control parameter α, that describes the probability level of [7] D. Gyalistras et al., “Performance bounds and potential assessment”,
constraint violation. in [6].
Figure 7 shows the tuning curves of SMPC and CE for one [8] D. Gyalistras, K. Wirth, B. Lehmann, “Analysis of savings potentials
and peak electricity demand”, in [6].
month (January 2007) as well as the corresponding results [9] B. Lehmann, K. Wirth, V. Dorer et al., “Control problem and experi-
of PB and RBC. It can be seen that PB shows no violations mental setup”, in [6].
and the smallest energy usage as expected. Further can be [10] B. Lehmann, K. Wirth, S. Carl, et al., “Modeling of buildings and
building systems”, in [6].
seen that both CE and SMPC controllers can achieve a better [11] F. Oldewurtel, C. N. Jones, M. Morari, “A Tractable Approximation
tradeoff between energy use and probability of constraint of Chance Constrained Stochastic MPC based on Affine Disturbance
violations than RBC by moving along the tuning curve. The Feedback”, in Proc. 47th IEEE Conf. on Decision and Control,
Cancun, Mexico, 2008 pp. 4731-4736.
closer to the origin a tuning curve lies, the better the control [12] F. Oldewurtel et al., “Analysis of model predictive control strategies”,
performance. Thus, SMPC performs clearly better than both, in [6].
CE as well as RBC. [13] A. Prékopa, “Stochastic programming”, Kluwer Academic Publishers,
Dordrecht, The Netherlands, 1995.
[14] SIA Standard 382/1, “Lüftings- und klimaanlagen - allgemeine grund-
VI. C ONCLUSIONS lagen und anforderungen”, 2006.
A Stochastic Model Predictive Control (SMPC) strategy [15] V. Stauch, F. Schubiger, P. Steiner, “Local weather forecasts and
observations”, in [6].
was applied to building climate control. The controller makes
use of weather predictions to compute how much energy